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
 351 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 352 {
 353         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 354                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 355         else
 356                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 357 }
 358
 359 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 360 {
 361         kvm_make_request(KVM_REQ_EVENT, vcpu);
 362         vcpu->arch.nmi_pending = 1;
 363 }
 364 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 365
 366 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 367 {
 368         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 369 }
 370 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 371
 372 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 373 {
 374         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 375 }
 376 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 377
 378 /*
 379  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 380  * a #GP and return false.
 381  */
 382 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 383 {
 384         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 385                 return true;
 386         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 387         return false;
 388 }
 389 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 390
 391 /*
 392  * This function will be used to read from the physical memory of the currently
 393  * running guest. The difference to kvm_read_guest_page is that this function
 394  * can read from guest physical or from the guest's guest physical memory.
 395  */
 396 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 397                             gfn_t ngfn, void *data, int offset, int len,
 398                             u32 access)
 399 {
 400         gfn_t real_gfn;
 401         gpa_t ngpa;
 402
 403         ngpa     = gfn_to_gpa(ngfn);
 404         real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
 405         if (real_gfn == UNMAPPED_GVA)
 406                 return -EFAULT;
 407
 408         real_gfn = gpa_to_gfn(real_gfn);
 409
 410         return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
 411 }
 412 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 413
 414 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 415                                void *data, int offset, int len, u32 access)
 416 {
 417         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 418                                        data, offset, len, access);
 419 }
 420
 421 /*
 422  * Load the pae pdptrs.  Return true is they are all valid.
 423  */
 424 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 425 {
 426         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 427         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 428         int i;
 429         int ret;
 430         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 431
 432         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 433                                       offset * sizeof(u64), sizeof(pdpte),
 434                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 435         if (ret < 0) {
 436                 ret = 0;
 437                 goto out;
 438         }
 439         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 440                 if (is_present_gpte(pdpte[i]) &&
 441                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 442                         ret = 0;
 443                         goto out;
 444                 }
 445         }
 446         ret = 1;
 447
 448         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 449         __set_bit(VCPU_EXREG_PDPTR,
 450                   (unsigned long *)&vcpu->arch.regs_avail);
 451         __set_bit(VCPU_EXREG_PDPTR,
 452                   (unsigned long *)&vcpu->arch.regs_dirty);
 453 out:
 454
 455         return ret;
 456 }
 457 EXPORT_SYMBOL_GPL(load_pdptrs);
 458
 459 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 460 {
 461         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 462         bool changed = true;
 463         int offset;
 464         gfn_t gfn;
 465         int r;
 466
 467         if (is_long_mode(vcpu) || !is_pae(vcpu))
 468                 return false;
 469
 470         if (!test_bit(VCPU_EXREG_PDPTR,
 471                       (unsigned long *)&vcpu->arch.regs_avail))
 472                 return true;
 473
 474         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 475         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 476         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 477                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 478         if (r < 0)
 479                 goto out;
 480         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 481 out:
 482
 483         return changed;
 484 }
 485
 486 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 487 {
 488         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 489         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 490                                     X86_CR0_CD | X86_CR0_NW;
 491
 492         cr0 |= X86_CR0_ET;
 493
 494 #ifdef CONFIG_X86_64
 495         if (cr0 & 0xffffffff00000000UL)
 496                 return 1;
 497 #endif
 498
 499         cr0 &= ~CR0_RESERVED_BITS;
 500
 501         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 502                 return 1;
 503
 504         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 505                 return 1;
 506
 507         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 508 #ifdef CONFIG_X86_64
 509                 if ((vcpu->arch.efer & EFER_LME)) {
 510                         int cs_db, cs_l;
 511
 512                         if (!is_pae(vcpu))
 513                                 return 1;
 514                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 515                         if (cs_l)
 516                                 return 1;
 517                 } else
 518 #endif
 519                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 520                                                  kvm_read_cr3(vcpu)))
 521                         return 1;
 522         }
 523
 524         kvm_x86_ops->set_cr0(vcpu, cr0);
 525
 526         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 527                 kvm_clear_async_pf_completion_queue(vcpu);
 528                 kvm_async_pf_hash_reset(vcpu);
 529         }
 530
 531         if ((cr0 ^ old_cr0) & update_bits)
 532                 kvm_mmu_reset_context(vcpu);
 533         return 0;
 534 }
 535 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 536
 537 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 538 {
 539         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 540 }
 541 EXPORT_SYMBOL_GPL(kvm_lmsw);
 542
 543 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 544 {
 545         u64 xcr0;
 546
 547         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 548         if (index != XCR_XFEATURE_ENABLED_MASK)
 549                 return 1;
 550         xcr0 = xcr;
 551         if (kvm_x86_ops->get_cpl(vcpu) != 0)
 552                 return 1;
 553         if (!(xcr0 & XSTATE_FP))
 554                 return 1;
 555         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
 556                 return 1;
 557         if (xcr0 & ~host_xcr0)
 558                 return 1;
 559         vcpu->arch.xcr0 = xcr0;
 560         vcpu->guest_xcr0_loaded = 0;
 561         return 0;
 562 }
 563
 564 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 565 {
 566         if (__kvm_set_xcr(vcpu, index, xcr)) {
 567                 kvm_inject_gp(vcpu, 0);
 568                 return 1;
 569         }
 570         return 0;
 571 }
 572 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 573
 574 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
 575 {
 576         struct kvm_cpuid_entry2 *best;
 577
 578         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 579         return best && (best->ecx & bit(X86_FEATURE_XSAVE));
 580 }
 581
 582 static void update_cpuid(struct kvm_vcpu *vcpu)
 583 {
 584         struct kvm_cpuid_entry2 *best;
 585
 586         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 587         if (!best)
 588                 return;
 589
 590         /* Update OSXSAVE bit */
 591         if (cpu_has_xsave && best->function == 0x1) {
 592                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
 593                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
 594                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
 595         }
 596 }
 597
 598 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 599 {
 600         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 601         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 602
 603         if (cr4 & CR4_RESERVED_BITS)
 604                 return 1;
 605
 606         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 607                 return 1;
 608
 609         if (is_long_mode(vcpu)) {
 610                 if (!(cr4 & X86_CR4_PAE))
 611                         return 1;
 612         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 613                    && ((cr4 ^ old_cr4) & pdptr_bits)
 614                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 615                                    kvm_read_cr3(vcpu)))
 616                 return 1;
 617
 618         if (cr4 & X86_CR4_VMXE)
 619                 return 1;
 620
 621         kvm_x86_ops->set_cr4(vcpu, cr4);
 622
 623         if ((cr4 ^ old_cr4) & pdptr_bits)
 624                 kvm_mmu_reset_context(vcpu);
 625
 626         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 627                 update_cpuid(vcpu);
 628
 629         return 0;
 630 }
 631 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 632
 633 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 634 {
 635         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 636                 kvm_mmu_sync_roots(vcpu);
 637                 kvm_mmu_flush_tlb(vcpu);
 638                 return 0;
 639         }
 640
 641         if (is_long_mode(vcpu)) {
 642                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 643                         return 1;
 644         } else {
 645                 if (is_pae(vcpu)) {
 646                         if (cr3 & CR3_PAE_RESERVED_BITS)
 647                                 return 1;
 648                         if (is_paging(vcpu) &&
 649                             !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 650                                 return 1;
 651                 }
 652                 /*
 653                  * We don't check reserved bits in nonpae mode, because
 654                  * this isn't enforced, and VMware depends on this.
 655                  */
 656         }
 657
 658         /*
 659          * Does the new cr3 value map to physical memory? (Note, we
 660          * catch an invalid cr3 even in real-mode, because it would
 661          * cause trouble later on when we turn on paging anyway.)
 662          *
 663          * A real CPU would silently accept an invalid cr3 and would
 664          * attempt to use it - with largely undefined (and often hard
 665          * to debug) behavior on the guest side.
 666          */
 667         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 668                 return 1;
 669         vcpu->arch.cr3 = cr3;
 670         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 671         vcpu->arch.mmu.new_cr3(vcpu);
 672         return 0;
 673 }
 674 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 675
 676 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 677 {
 678         if (cr8 & CR8_RESERVED_BITS)
 679                 return 1;
 680         if (irqchip_in_kernel(vcpu->kvm))
 681                 kvm_lapic_set_tpr(vcpu, cr8);
 682         else
 683                 vcpu->arch.cr8 = cr8;
 684         return 0;
 685 }
 686 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 687
 688 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 689 {
 690         if (irqchip_in_kernel(vcpu->kvm))
 691                 return kvm_lapic_get_cr8(vcpu);
 692         else
 693                 return vcpu->arch.cr8;
 694 }
 695 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 696
 697 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 698 {
 699         switch (dr) {
 700         case 0 ... 3:
 701                 vcpu->arch.db[dr] = val;
 702                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 703                         vcpu->arch.eff_db[dr] = val;
 704                 break;
 705         case 4:
 706                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 707                         return 1; /* #UD */
 708                 /* fall through */
 709         case 6:
 710                 if (val & 0xffffffff00000000ULL)
 711                         return -1; /* #GP */
 712                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 713                 break;
 714         case 5:
 715                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 716                         return 1; /* #UD */
 717                 /* fall through */
 718         default: /* 7 */
 719                 if (val & 0xffffffff00000000ULL)
 720                         return -1; /* #GP */
 721                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 722                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 723                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
 724                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
 725                 }
 726                 break;
 727         }
 728
 729         return 0;
 730 }
 731
 732 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 733 {
 734         int res;
 735
 736         res = __kvm_set_dr(vcpu, dr, val);
 737         if (res > 0)
 738                 kvm_queue_exception(vcpu, UD_VECTOR);
 739         else if (res < 0)
 740                 kvm_inject_gp(vcpu, 0);
 741
 742         return res;
 743 }
 744 EXPORT_SYMBOL_GPL(kvm_set_dr);
 745
 746 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 747 {
 748         switch (dr) {
 749         case 0 ... 3:
 750                 *val = vcpu->arch.db[dr];
 751                 break;
 752         case 4:
 753                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 754                         return 1;
 755                 /* fall through */
 756         case 6:
 757                 *val = vcpu->arch.dr6;
 758                 break;
 759         case 5:
 760                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 761                         return 1;
 762                 /* fall through */
 763         default: /* 7 */
 764                 *val = vcpu->arch.dr7;
 765                 break;
 766         }
 767
 768         return 0;
 769 }
 770
 771 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 772 {
 773         if (_kvm_get_dr(vcpu, dr, val)) {
 774                 kvm_queue_exception(vcpu, UD_VECTOR);
 775                 return 1;
 776         }
 777         return 0;
 778 }
 779 EXPORT_SYMBOL_GPL(kvm_get_dr);
 780
 781 /*
 782  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 783  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 784  *
 785  * This list is modified at module load time to reflect the
 786  * capabilities of the host cpu. This capabilities test skips MSRs that are
 787  * kvm-specific. Those are put in the beginning of the list.
 788  */
 789
 790 #define KVM_SAVE_MSRS_BEGIN     8
 791 static u32 msrs_to_save[] = {
 792         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 793         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 794         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 795         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
 796         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 797         MSR_STAR,
 798 #ifdef CONFIG_X86_64
 799         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 800 #endif
 801         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 802 };
 803
 804 static unsigned num_msrs_to_save;
 805
 806 static u32 emulated_msrs[] = {
 807         MSR_IA32_MISC_ENABLE,
 808         MSR_IA32_MCG_STATUS,
 809         MSR_IA32_MCG_CTL,
 810 };
 811
 812 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 813 {
 814         u64 old_efer = vcpu->arch.efer;
 815
 816         if (efer & efer_reserved_bits)
 817                 return 1;
 818
 819         if (is_paging(vcpu)
 820             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 821                 return 1;
 822
 823         if (efer & EFER_FFXSR) {
 824                 struct kvm_cpuid_entry2 *feat;
 825
 826                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 827                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 828                         return 1;
 829         }
 830
 831         if (efer & EFER_SVME) {
 832                 struct kvm_cpuid_entry2 *feat;
 833
 834                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 835                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 836                         return 1;
 837         }
 838
 839         efer &= ~EFER_LMA;
 840         efer |= vcpu->arch.efer & EFER_LMA;
 841
 842         kvm_x86_ops->set_efer(vcpu, efer);
 843
 844         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 845
 846         /* Update reserved bits */
 847         if ((efer ^ old_efer) & EFER_NX)
 848                 kvm_mmu_reset_context(vcpu);
 849
 850         return 0;
 851 }
 852
 853 void kvm_enable_efer_bits(u64 mask)
 854 {
 855        efer_reserved_bits &= ~mask;
 856 }
 857 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 858
 859
 860 /*
 861  * Writes msr value into into the appropriate "register".
 862  * Returns 0 on success, non-0 otherwise.
 863  * Assumes vcpu_load() was already called.
 864  */
 865 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 866 {
 867         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 868 }
 869
 870 /*
 871  * Adapt set_msr() to msr_io()'s calling convention
 872  */
 873 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 874 {
 875         return kvm_set_msr(vcpu, index, *data);
 876 }
 877
 878 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 879 {
 880         int version;
 881         int r;
 882         struct pvclock_wall_clock wc;
 883         struct timespec boot;
 884
 885         if (!wall_clock)
 886                 return;
 887
 888         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
 889         if (r)
 890                 return;
 891
 892         if (version & 1)
 893                 ++version;  /* first time write, random junk */
 894
 895         ++version;
 896
 897         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 898
 899         /*
 900          * The guest calculates current wall clock time by adding
 901          * system time (updated by kvm_guest_time_update below) to the
 902          * wall clock specified here.  guest system time equals host
 903          * system time for us, thus we must fill in host boot time here.
 904          */
 905         getboottime(&boot);
 906
 907         wc.sec = boot.tv_sec;
 908         wc.nsec = boot.tv_nsec;
 909         wc.version = version;
 910
 911         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 912
 913         version++;
 914         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 915 }
 916
 917 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 918 {
 919         uint32_t quotient, remainder;
 920
 921         /* Don't try to replace with do_div(), this one calculates
 922          * "(dividend << 32) / divisor" */
 923         __asm__ ( "divl %4"
 924                   : "=a" (quotient), "=d" (remainder)
 925                   : "0" (0), "1" (dividend), "r" (divisor) );
 926         return quotient;
 927 }
 928
 929 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
 930                                s8 *pshift, u32 *pmultiplier)
 931 {
 932         uint64_t scaled64;
 933         int32_t  shift = 0;
 934         uint64_t tps64;
 935         uint32_t tps32;
 936
 937         tps64 = base_khz * 1000LL;
 938         scaled64 = scaled_khz * 1000LL;
 939         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
 940                 tps64 >>= 1;
 941                 shift--;
 942         }
 943
 944         tps32 = (uint32_t)tps64;
 945         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
 946                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
 947                         scaled64 >>= 1;
 948                 else
 949                         tps32 <<= 1;
 950                 shift++;
 951         }
 952
 953         *pshift = shift;
 954         *pmultiplier = div_frac(scaled64, tps32);
 955
 956         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
 957                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
 958 }
 959
 960 static inline u64 get_kernel_ns(void)
 961 {
 962         struct timespec ts;
 963
 964         WARN_ON(preemptible());
 965         ktime_get_ts(&ts);
 966         monotonic_to_bootbased(&ts);
 967         return timespec_to_ns(&ts);
 968 }
 969
 970 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 971 unsigned long max_tsc_khz;
 972
 973 static inline int kvm_tsc_changes_freq(void)
 974 {
 975         int cpu = get_cpu();
 976         int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
 977                   cpufreq_quick_get(cpu) != 0;
 978         put_cpu();
 979         return ret;
 980 }
 981
 982 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
 983 {
 984         if (vcpu->arch.virtual_tsc_khz)
 985                 return vcpu->arch.virtual_tsc_khz;
 986         else
 987                 return __this_cpu_read(cpu_tsc_khz);
 988 }
 989
 990 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
 991 {
 992         u64 ret;
 993
 994         WARN_ON(preemptible());
 995         if (kvm_tsc_changes_freq())
 996                 printk_once(KERN_WARNING
 997                  "kvm: unreliable cycle conversion on adjustable rate TSC\n");
 998         ret = nsec * vcpu_tsc_khz(vcpu);
 999         do_div(ret, USEC_PER_SEC);
1000         return ret;
1001 }
1002
1003 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1004 {
1005         /* Compute a scale to convert nanoseconds in TSC cycles */
1006         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1007                            &vcpu->arch.tsc_catchup_shift,
1008                            &vcpu->arch.tsc_catchup_mult);
1009 }
1010
1011 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1012 {
1013         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1014                                       vcpu->arch.tsc_catchup_mult,
1015                                       vcpu->arch.tsc_catchup_shift);
1016         tsc += vcpu->arch.last_tsc_write;
1017         return tsc;
1018 }
1019
1020 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1021 {
1022         struct kvm *kvm = vcpu->kvm;
1023         u64 offset, ns, elapsed;
1024         unsigned long flags;
1025         s64 sdiff;
1026
1027         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1028         offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1029         ns = get_kernel_ns();
1030         elapsed = ns - kvm->arch.last_tsc_nsec;
1031         sdiff = data - kvm->arch.last_tsc_write;
1032         if (sdiff < 0)
1033                 sdiff = -sdiff;
1034
1035         /*
1036          * Special case: close write to TSC within 5 seconds of
1037          * another CPU is interpreted as an attempt to synchronize
1038          * The 5 seconds is to accommodate host load / swapping as
1039          * well as any reset of TSC during the boot process.
1040          *
1041          * In that case, for a reliable TSC, we can match TSC offsets,
1042          * or make a best guest using elapsed value.
1043          */
1044         if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1045             elapsed < 5ULL * NSEC_PER_SEC) {
1046                 if (!check_tsc_unstable()) {
1047                         offset = kvm->arch.last_tsc_offset;
1048                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1049                 } else {
1050                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1051                         offset += delta;
1052                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1053                 }
1054                 ns = kvm->arch.last_tsc_nsec;
1055         }
1056         kvm->arch.last_tsc_nsec = ns;
1057         kvm->arch.last_tsc_write = data;
1058         kvm->arch.last_tsc_offset = offset;
1059         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1060         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1061
1062         /* Reset of TSC must disable overshoot protection below */
1063         vcpu->arch.hv_clock.tsc_timestamp = 0;
1064         vcpu->arch.last_tsc_write = data;
1065         vcpu->arch.last_tsc_nsec = ns;
1066 }
1067 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1068
1069 static int kvm_guest_time_update(struct kvm_vcpu *v)
1070 {
1071         unsigned long flags;
1072         struct kvm_vcpu_arch *vcpu = &v->arch;
1073         void *shared_kaddr;
1074         unsigned long this_tsc_khz;
1075         s64 kernel_ns, max_kernel_ns;
1076         u64 tsc_timestamp;
1077
1078         /* Keep irq disabled to prevent changes to the clock */
1079         local_irq_save(flags);
1080         kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1081         kernel_ns = get_kernel_ns();
1082         this_tsc_khz = vcpu_tsc_khz(v);
1083         if (unlikely(this_tsc_khz == 0)) {
1084                 local_irq_restore(flags);
1085                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1086                 return 1;
1087         }
1088
1089         /*
1090          * We may have to catch up the TSC to match elapsed wall clock
1091          * time for two reasons, even if kvmclock is used.
1092          *   1) CPU could have been running below the maximum TSC rate
1093          *   2) Broken TSC compensation resets the base at each VCPU
1094          *      entry to avoid unknown leaps of TSC even when running
1095          *      again on the same CPU.  This may cause apparent elapsed
1096          *      time to disappear, and the guest to stand still or run
1097          *      very slowly.
1098          */
1099         if (vcpu->tsc_catchup) {
1100                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1101                 if (tsc > tsc_timestamp) {
1102                         kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1103                         tsc_timestamp = tsc;
1104                 }
1105         }
1106
1107         local_irq_restore(flags);
1108
1109         if (!vcpu->time_page)
1110                 return 0;
1111
1112         /*
1113          * Time as measured by the TSC may go backwards when resetting the base
1114          * tsc_timestamp.  The reason for this is that the TSC resolution is
1115          * higher than the resolution of the other clock scales.  Thus, many
1116          * possible measurments of the TSC correspond to one measurement of any
1117          * other clock, and so a spread of values is possible.  This is not a
1118          * problem for the computation of the nanosecond clock; with TSC rates
1119          * around 1GHZ, there can only be a few cycles which correspond to one
1120          * nanosecond value, and any path through this code will inevitably
1121          * take longer than that.  However, with the kernel_ns value itself,
1122          * the precision may be much lower, down to HZ granularity.  If the
1123          * first sampling of TSC against kernel_ns ends in the low part of the
1124          * range, and the second in the high end of the range, we can get:
1125          *
1126          * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1127          *
1128          * As the sampling errors potentially range in the thousands of cycles,
1129          * it is possible such a time value has already been observed by the
1130          * guest.  To protect against this, we must compute the system time as
1131          * observed by the guest and ensure the new system time is greater.
1132          */
1133         max_kernel_ns = 0;
1134         if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1135                 max_kernel_ns = vcpu->last_guest_tsc -
1136                                 vcpu->hv_clock.tsc_timestamp;
1137                 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1138                                     vcpu->hv_clock.tsc_to_system_mul,
1139                                     vcpu->hv_clock.tsc_shift);
1140                 max_kernel_ns += vcpu->last_kernel_ns;
1141         }
1142
1143         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1144                 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1145                                    &vcpu->hv_clock.tsc_shift,
1146                                    &vcpu->hv_clock.tsc_to_system_mul);
1147                 vcpu->hw_tsc_khz = this_tsc_khz;
1148         }
1149
1150         if (max_kernel_ns > kernel_ns)
1151                 kernel_ns = max_kernel_ns;
1152
1153         /* With all the info we got, fill in the values */
1154         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1155         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1156         vcpu->last_kernel_ns = kernel_ns;
1157         vcpu->last_guest_tsc = tsc_timestamp;
1158         vcpu->hv_clock.flags = 0;
1159
1160         /*
1161          * The interface expects us to write an even number signaling that the
1162          * update is finished. Since the guest won't see the intermediate
1163          * state, we just increase by 2 at the end.
1164          */
1165         vcpu->hv_clock.version += 2;
1166
1167         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1168
1169         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1170                sizeof(vcpu->hv_clock));
1171
1172         kunmap_atomic(shared_kaddr, KM_USER0);
1173
1174         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1175         return 0;
1176 }
1177
1178 static bool msr_mtrr_valid(unsigned msr)
1179 {
1180         switch (msr) {
1181         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1182         case MSR_MTRRfix64K_00000:
1183         case MSR_MTRRfix16K_80000:
1184         case MSR_MTRRfix16K_A0000:
1185         case MSR_MTRRfix4K_C0000:
1186         case MSR_MTRRfix4K_C8000:
1187         case MSR_MTRRfix4K_D0000:
1188         case MSR_MTRRfix4K_D8000:
1189         case MSR_MTRRfix4K_E0000:
1190         case MSR_MTRRfix4K_E8000:
1191         case MSR_MTRRfix4K_F0000:
1192         case MSR_MTRRfix4K_F8000:
1193         case MSR_MTRRdefType:
1194         case MSR_IA32_CR_PAT:
1195                 return true;
1196         case 0x2f8:
1197                 return true;
1198         }
1199         return false;
1200 }
1201
1202 static bool valid_pat_type(unsigned t)
1203 {
1204         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1205 }
1206
1207 static bool valid_mtrr_type(unsigned t)
1208 {
1209         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1210 }
1211
1212 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1213 {
1214         int i;
1215
1216         if (!msr_mtrr_valid(msr))
1217                 return false;
1218
1219         if (msr == MSR_IA32_CR_PAT) {
1220                 for (i = 0; i < 8; i++)
1221                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
1222                                 return false;
1223                 return true;
1224         } else if (msr == MSR_MTRRdefType) {
1225                 if (data & ~0xcff)
1226                         return false;
1227                 return valid_mtrr_type(data & 0xff);
1228         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1229                 for (i = 0; i < 8 ; i++)
1230                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1231                                 return false;
1232                 return true;
1233         }
1234
1235         /* variable MTRRs */
1236         return valid_mtrr_type(data & 0xff);
1237 }
1238
1239 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1240 {
1241         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1242
1243         if (!mtrr_valid(vcpu, msr, data))
1244                 return 1;
1245
1246         if (msr == MSR_MTRRdefType) {
1247                 vcpu->arch.mtrr_state.def_type = data;
1248                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1249         } else if (msr == MSR_MTRRfix64K_00000)
1250                 p[0] = data;
1251         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1252                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1253         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1254                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1255         else if (msr == MSR_IA32_CR_PAT)
1256                 vcpu->arch.pat = data;
1257         else {  /* Variable MTRRs */
1258                 int idx, is_mtrr_mask;
1259                 u64 *pt;
1260
1261                 idx = (msr - 0x200) / 2;
1262                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1263                 if (!is_mtrr_mask)
1264                         pt =
1265                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1266                 else
1267                         pt =
1268                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1269                 *pt = data;
1270         }
1271
1272         kvm_mmu_reset_context(vcpu);
1273         return 0;
1274 }
1275
1276 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1277 {
1278         u64 mcg_cap = vcpu->arch.mcg_cap;
1279         unsigned bank_num = mcg_cap & 0xff;
1280
1281         switch (msr) {
1282         case MSR_IA32_MCG_STATUS:
1283                 vcpu->arch.mcg_status = data;
1284                 break;
1285         case MSR_IA32_MCG_CTL:
1286                 if (!(mcg_cap & MCG_CTL_P))
1287                         return 1;
1288                 if (data != 0 && data != ~(u64)0)
1289                         return -1;
1290                 vcpu->arch.mcg_ctl = data;
1291                 break;
1292         default:
1293                 if (msr >= MSR_IA32_MC0_CTL &&
1294                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1295                         u32 offset = msr - MSR_IA32_MC0_CTL;
1296                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1297                          * some Linux kernels though clear bit 10 in bank 4 to
1298                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1299                          * this to avoid an uncatched #GP in the guest
1300                          */
1301                         if ((offset & 0x3) == 0 &&
1302                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1303                                 return -1;
1304                         vcpu->arch.mce_banks[offset] = data;
1305                         break;
1306                 }
1307                 return 1;
1308         }
1309         return 0;
1310 }
1311
1312 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1313 {
1314         struct kvm *kvm = vcpu->kvm;
1315         int lm = is_long_mode(vcpu);
1316         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1317                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1318         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1319                 : kvm->arch.xen_hvm_config.blob_size_32;
1320         u32 page_num = data & ~PAGE_MASK;
1321         u64 page_addr = data & PAGE_MASK;
1322         u8 *page;
1323         int r;
1324
1325         r = -E2BIG;
1326         if (page_num >= blob_size)
1327                 goto out;
1328         r = -ENOMEM;
1329         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1330         if (!page)
1331                 goto out;
1332         r = -EFAULT;
1333         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1334                 goto out_free;
1335         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1336                 goto out_free;
1337         r = 0;
1338 out_free:
1339         kfree(page);
1340 out:
1341         return r;
1342 }
1343
1344 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1345 {
1346         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1347 }
1348
1349 static bool kvm_hv_msr_partition_wide(u32 msr)
1350 {
1351         bool r = false;
1352         switch (msr) {
1353         case HV_X64_MSR_GUEST_OS_ID:
1354         case HV_X64_MSR_HYPERCALL:
1355                 r = true;
1356                 break;
1357         }
1358
1359         return r;
1360 }
1361
1362 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1363 {
1364         struct kvm *kvm = vcpu->kvm;
1365
1366         switch (msr) {
1367         case HV_X64_MSR_GUEST_OS_ID:
1368                 kvm->arch.hv_guest_os_id = data;
1369                 /* setting guest os id to zero disables hypercall page */
1370                 if (!kvm->arch.hv_guest_os_id)
1371                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1372                 break;
1373         case HV_X64_MSR_HYPERCALL: {
1374                 u64 gfn;
1375                 unsigned long addr;
1376                 u8 instructions[4];
1377
1378                 /* if guest os id is not set hypercall should remain disabled */
1379                 if (!kvm->arch.hv_guest_os_id)
1380                         break;
1381                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1382                         kvm->arch.hv_hypercall = data;
1383                         break;
1384                 }
1385                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1386                 addr = gfn_to_hva(kvm, gfn);
1387                 if (kvm_is_error_hva(addr))
1388                         return 1;
1389                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1390                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1391                 if (copy_to_user((void __user *)addr, instructions, 4))
1392                         return 1;
1393                 kvm->arch.hv_hypercall = data;
1394                 break;
1395         }
1396         default:
1397                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1398                           "data 0x%llx\n", msr, data);
1399                 return 1;
1400         }
1401         return 0;
1402 }
1403
1404 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1405 {
1406         switch (msr) {
1407         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1408                 unsigned long addr;
1409
1410                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1411                         vcpu->arch.hv_vapic = data;
1412                         break;
1413                 }
1414                 addr = gfn_to_hva(vcpu->kvm, data >>
1415                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1416                 if (kvm_is_error_hva(addr))
1417                         return 1;
1418                 if (clear_user((void __user *)addr, PAGE_SIZE))
1419                         return 1;
1420                 vcpu->arch.hv_vapic = data;
1421                 break;
1422         }
1423         case HV_X64_MSR_EOI:
1424                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1425         case HV_X64_MSR_ICR:
1426                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1427         case HV_X64_MSR_TPR:
1428                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1429         default:
1430                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1431                           "data 0x%llx\n", msr, data);
1432                 return 1;
1433         }
1434
1435         return 0;
1436 }
1437
1438 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1439 {
1440         gpa_t gpa = data & ~0x3f;
1441
1442         /* Bits 2:5 are resrved, Should be zero */
1443         if (data & 0x3c)
1444                 return 1;
1445
1446         vcpu->arch.apf.msr_val = data;
1447
1448         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1449                 kvm_clear_async_pf_completion_queue(vcpu);
1450                 kvm_async_pf_hash_reset(vcpu);
1451                 return 0;
1452         }
1453
1454         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1455                 return 1;
1456
1457         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1458         kvm_async_pf_wakeup_all(vcpu);
1459         return 0;
1460 }
1461
1462 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1463 {
1464         if (vcpu->arch.time_page) {
1465                 kvm_release_page_dirty(vcpu->arch.time_page);
1466                 vcpu->arch.time_page = NULL;
1467         }
1468 }
1469
1470 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1471 {
1472         switch (msr) {
1473         case MSR_EFER:
1474                 return set_efer(vcpu, data);
1475         case MSR_K7_HWCR:
1476                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1477                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1478                 if (data != 0) {
1479                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1480                                 data);
1481                         return 1;
1482                 }
1483                 break;
1484         case MSR_FAM10H_MMIO_CONF_BASE:
1485                 if (data != 0) {
1486                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1487                                 "0x%llx\n", data);
1488                         return 1;
1489                 }
1490                 break;
1491         case MSR_AMD64_NB_CFG:
1492                 break;
1493         case MSR_IA32_DEBUGCTLMSR:
1494                 if (!data) {
1495                         /* We support the non-activated case already */
1496                         break;
1497                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1498                         /* Values other than LBR and BTF are vendor-specific,
1499                            thus reserved and should throw a #GP */
1500                         return 1;
1501                 }
1502                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1503                         __func__, data);
1504                 break;
1505         case MSR_IA32_UCODE_REV:
1506         case MSR_IA32_UCODE_WRITE:
1507         case MSR_VM_HSAVE_PA:
1508         case MSR_AMD64_PATCH_LOADER:
1509                 break;
1510         case 0x200 ... 0x2ff:
1511                 return set_msr_mtrr(vcpu, msr, data);
1512         case MSR_IA32_APICBASE:
1513                 kvm_set_apic_base(vcpu, data);
1514                 break;
1515         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1516                 return kvm_x2apic_msr_write(vcpu, msr, data);
1517         case MSR_IA32_MISC_ENABLE:
1518                 vcpu->arch.ia32_misc_enable_msr = data;
1519                 break;
1520         case MSR_KVM_WALL_CLOCK_NEW:
1521         case MSR_KVM_WALL_CLOCK:
1522                 vcpu->kvm->arch.wall_clock = data;
1523                 kvm_write_wall_clock(vcpu->kvm, data);
1524                 break;
1525         case MSR_KVM_SYSTEM_TIME_NEW:
1526         case MSR_KVM_SYSTEM_TIME: {
1527                 kvmclock_reset(vcpu);
1528
1529                 vcpu->arch.time = data;
1530                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1531
1532                 /* we verify if the enable bit is set... */
1533                 if (!(data & 1))
1534                         break;
1535
1536                 /* ...but clean it before doing the actual write */
1537                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1538
1539                 vcpu->arch.time_page =
1540                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1541
1542                 if (is_error_page(vcpu->arch.time_page)) {
1543                         kvm_release_page_clean(vcpu->arch.time_page);
1544                         vcpu->arch.time_page = NULL;
1545                 }
1546                 break;
1547         }
1548         case MSR_KVM_ASYNC_PF_EN:
1549                 if (kvm_pv_enable_async_pf(vcpu, data))
1550                         return 1;
1551                 break;
1552         case MSR_IA32_MCG_CTL:
1553         case MSR_IA32_MCG_STATUS:
1554         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1555                 return set_msr_mce(vcpu, msr, data);
1556
1557         /* Performance counters are not protected by a CPUID bit,
1558          * so we should check all of them in the generic path for the sake of
1559          * cross vendor migration.
1560          * Writing a zero into the event select MSRs disables them,
1561          * which we perfectly emulate ;-). Any other value should be at least
1562          * reported, some guests depend on them.
1563          */
1564         case MSR_P6_EVNTSEL0:
1565         case MSR_P6_EVNTSEL1:
1566         case MSR_K7_EVNTSEL0:
1567         case MSR_K7_EVNTSEL1:
1568         case MSR_K7_EVNTSEL2:
1569         case MSR_K7_EVNTSEL3:
1570                 if (data != 0)
1571                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1572                                 "0x%x data 0x%llx\n", msr, data);
1573                 break;
1574         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1575          * so we ignore writes to make it happy.
1576          */
1577         case MSR_P6_PERFCTR0:
1578         case MSR_P6_PERFCTR1:
1579         case MSR_K7_PERFCTR0:
1580         case MSR_K7_PERFCTR1:
1581         case MSR_K7_PERFCTR2:
1582         case MSR_K7_PERFCTR3:
1583                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1584                         "0x%x data 0x%llx\n", msr, data);
1585                 break;
1586         case MSR_K7_CLK_CTL:
1587                 /*
1588                  * Ignore all writes to this no longer documented MSR.
1589                  * Writes are only relevant for old K7 processors,
1590                  * all pre-dating SVM, but a recommended workaround from
1591                  * AMD for these chips. It is possible to speicify the
1592                  * affected processor models on the command line, hence
1593                  * the need to ignore the workaround.
1594                  */
1595                 break;
1596         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1597                 if (kvm_hv_msr_partition_wide(msr)) {
1598                         int r;
1599                         mutex_lock(&vcpu->kvm->lock);
1600                         r = set_msr_hyperv_pw(vcpu, msr, data);
1601                         mutex_unlock(&vcpu->kvm->lock);
1602                         return r;
1603                 } else
1604                         return set_msr_hyperv(vcpu, msr, data);
1605                 break;
1606         case MSR_IA32_BBL_CR_CTL3:
1607                 /* Drop writes to this legacy MSR -- see rdmsr
1608                  * counterpart for further detail.
1609                  */
1610                 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1611                 break;
1612         default:
1613                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1614                         return xen_hvm_config(vcpu, data);
1615                 if (!ignore_msrs) {
1616                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1617                                 msr, data);
1618                         return 1;
1619                 } else {
1620                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1621                                 msr, data);
1622                         break;
1623                 }
1624         }
1625         return 0;
1626 }
1627 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1628
1629
1630 /*
1631  * Reads an msr value (of 'msr_index') into 'pdata'.
1632  * Returns 0 on success, non-0 otherwise.
1633  * Assumes vcpu_load() was already called.
1634  */
1635 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1636 {
1637         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1638 }
1639
1640 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1641 {
1642         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1643
1644         if (!msr_mtrr_valid(msr))
1645                 return 1;
1646
1647         if (msr == MSR_MTRRdefType)
1648                 *pdata = vcpu->arch.mtrr_state.def_type +
1649                          (vcpu->arch.mtrr_state.enabled << 10);
1650         else if (msr == MSR_MTRRfix64K_00000)
1651                 *pdata = p[0];
1652         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1653                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1654         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1655                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1656         else if (msr == MSR_IA32_CR_PAT)
1657                 *pdata = vcpu->arch.pat;
1658         else {  /* Variable MTRRs */
1659                 int idx, is_mtrr_mask;
1660                 u64 *pt;
1661
1662                 idx = (msr - 0x200) / 2;
1663                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1664                 if (!is_mtrr_mask)
1665                         pt =
1666                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1667                 else
1668                         pt =
1669                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1670                 *pdata = *pt;
1671         }
1672
1673         return 0;
1674 }
1675
1676 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1677 {
1678         u64 data;
1679         u64 mcg_cap = vcpu->arch.mcg_cap;
1680         unsigned bank_num = mcg_cap & 0xff;
1681
1682         switch (msr) {
1683         case MSR_IA32_P5_MC_ADDR:
1684         case MSR_IA32_P5_MC_TYPE:
1685                 data = 0;
1686                 break;
1687         case MSR_IA32_MCG_CAP:
1688                 data = vcpu->arch.mcg_cap;
1689                 break;
1690         case MSR_IA32_MCG_CTL:
1691                 if (!(mcg_cap & MCG_CTL_P))
1692                         return 1;
1693                 data = vcpu->arch.mcg_ctl;
1694                 break;
1695         case MSR_IA32_MCG_STATUS:
1696                 data = vcpu->arch.mcg_status;
1697                 break;
1698         default:
1699                 if (msr >= MSR_IA32_MC0_CTL &&
1700                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1701                         u32 offset = msr - MSR_IA32_MC0_CTL;
1702                         data = vcpu->arch.mce_banks[offset];
1703                         break;
1704                 }
1705                 return 1;
1706         }
1707         *pdata = data;
1708         return 0;
1709 }
1710
1711 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1712 {
1713         u64 data = 0;
1714         struct kvm *kvm = vcpu->kvm;
1715
1716         switch (msr) {
1717         case HV_X64_MSR_GUEST_OS_ID:
1718                 data = kvm->arch.hv_guest_os_id;
1719                 break;
1720         case HV_X64_MSR_HYPERCALL:
1721                 data = kvm->arch.hv_hypercall;
1722                 break;
1723         default:
1724                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1725                 return 1;
1726         }
1727
1728         *pdata = data;
1729         return 0;
1730 }
1731
1732 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1733 {
1734         u64 data = 0;
1735
1736         switch (msr) {
1737         case HV_X64_MSR_VP_INDEX: {
1738                 int r;
1739                 struct kvm_vcpu *v;
1740                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1741                         if (v == vcpu)
1742                                 data = r;
1743                 break;
1744         }
1745         case HV_X64_MSR_EOI:
1746                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1747         case HV_X64_MSR_ICR:
1748                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1749         case HV_X64_MSR_TPR:
1750                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1751         default:
1752                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1753                 return 1;
1754         }
1755         *pdata = data;
1756         return 0;
1757 }
1758
1759 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1760 {
1761         u64 data;
1762
1763         switch (msr) {
1764         case MSR_IA32_PLATFORM_ID:
1765         case MSR_IA32_UCODE_REV:
1766         case MSR_IA32_EBL_CR_POWERON:
1767         case MSR_IA32_DEBUGCTLMSR:
1768         case MSR_IA32_LASTBRANCHFROMIP:
1769         case MSR_IA32_LASTBRANCHTOIP:
1770         case MSR_IA32_LASTINTFROMIP:
1771         case MSR_IA32_LASTINTTOIP:
1772         case MSR_K8_SYSCFG:
1773         case MSR_K7_HWCR:
1774         case MSR_VM_HSAVE_PA:
1775         case MSR_P6_PERFCTR0:
1776         case MSR_P6_PERFCTR1:
1777         case MSR_P6_EVNTSEL0:
1778         case MSR_P6_EVNTSEL1:
1779         case MSR_K7_EVNTSEL0:
1780         case MSR_K7_PERFCTR0:
1781         case MSR_K8_INT_PENDING_MSG:
1782         case MSR_AMD64_NB_CFG:
1783         case MSR_FAM10H_MMIO_CONF_BASE:
1784                 data = 0;
1785                 break;
1786         case MSR_MTRRcap:
1787                 data = 0x500 | KVM_NR_VAR_MTRR;
1788                 break;
1789         case 0x200 ... 0x2ff:
1790                 return get_msr_mtrr(vcpu, msr, pdata);
1791         case 0xcd: /* fsb frequency */
1792                 data = 3;
1793                 break;
1794                 /*
1795                  * MSR_EBC_FREQUENCY_ID
1796                  * Conservative value valid for even the basic CPU models.
1797                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1798                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1799                  * and 266MHz for model 3, or 4. Set Core Clock
1800                  * Frequency to System Bus Frequency Ratio to 1 (bits
1801                  * 31:24) even though these are only valid for CPU
1802                  * models > 2, however guests may end up dividing or
1803                  * multiplying by zero otherwise.
1804                  */
1805         case MSR_EBC_FREQUENCY_ID:
1806                 data = 1 << 24;
1807                 break;
1808         case MSR_IA32_APICBASE:
1809                 data = kvm_get_apic_base(vcpu);
1810                 break;
1811         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1812                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1813                 break;
1814         case MSR_IA32_MISC_ENABLE:
1815                 data = vcpu->arch.ia32_misc_enable_msr;
1816                 break;
1817         case MSR_IA32_PERF_STATUS:
1818                 /* TSC increment by tick */
1819                 data = 1000ULL;
1820                 /* CPU multiplier */
1821                 data |= (((uint64_t)4ULL) << 40);
1822                 break;
1823         case MSR_EFER:
1824                 data = vcpu->arch.efer;
1825                 break;
1826         case MSR_KVM_WALL_CLOCK:
1827         case MSR_KVM_WALL_CLOCK_NEW:
1828                 data = vcpu->kvm->arch.wall_clock;
1829                 break;
1830         case MSR_KVM_SYSTEM_TIME:
1831         case MSR_KVM_SYSTEM_TIME_NEW:
1832                 data = vcpu->arch.time;
1833                 break;
1834         case MSR_KVM_ASYNC_PF_EN:
1835                 data = vcpu->arch.apf.msr_val;
1836                 break;
1837         case MSR_IA32_P5_MC_ADDR:
1838         case MSR_IA32_P5_MC_TYPE:
1839         case MSR_IA32_MCG_CAP:
1840         case MSR_IA32_MCG_CTL:
1841         case MSR_IA32_MCG_STATUS:
1842         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1843                 return get_msr_mce(vcpu, msr, pdata);
1844         case MSR_K7_CLK_CTL:
1845                 /*
1846                  * Provide expected ramp-up count for K7. All other
1847                  * are set to zero, indicating minimum divisors for
1848                  * every field.
1849                  *
1850                  * This prevents guest kernels on AMD host with CPU
1851                  * type 6, model 8 and higher from exploding due to
1852                  * the rdmsr failing.
1853                  */
1854                 data = 0x20000000;
1855                 break;
1856         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1857                 if (kvm_hv_msr_partition_wide(msr)) {
1858                         int r;
1859                         mutex_lock(&vcpu->kvm->lock);
1860                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1861                         mutex_unlock(&vcpu->kvm->lock);
1862                         return r;
1863                 } else
1864                         return get_msr_hyperv(vcpu, msr, pdata);
1865                 break;
1866         case MSR_IA32_BBL_CR_CTL3:
1867                 /* This legacy MSR exists but isn't fully documented in current
1868                  * silicon.  It is however accessed by winxp in very narrow
1869                  * scenarios where it sets bit #19, itself documented as
1870                  * a "reserved" bit.  Best effort attempt to source coherent
1871                  * read data here should the balance of the register be
1872                  * interpreted by the guest:
1873                  *
1874                  * L2 cache control register 3: 64GB range, 256KB size,
1875                  * enabled, latency 0x1, configured
1876                  */
1877                 data = 0xbe702111;
1878                 break;
1879         default:
1880                 if (!ignore_msrs) {
1881                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1882                         return 1;
1883                 } else {
1884                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1885                         data = 0;
1886                 }
1887                 break;
1888         }
1889         *pdata = data;
1890         return 0;
1891 }
1892 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1893
1894 /*
1895  * Read or write a bunch of msrs. All parameters are kernel addresses.
1896  *
1897  * @return number of msrs set successfully.
1898  */
1899 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1900                     struct kvm_msr_entry *entries,
1901                     int (*do_msr)(struct kvm_vcpu *vcpu,
1902                                   unsigned index, u64 *data))
1903 {
1904         int i, idx;
1905
1906         idx = srcu_read_lock(&vcpu->kvm->srcu);
1907         for (i = 0; i < msrs->nmsrs; ++i)
1908                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1909                         break;
1910         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1911
1912         return i;
1913 }
1914
1915 /*
1916  * Read or write a bunch of msrs. Parameters are user addresses.
1917  *
1918  * @return number of msrs set successfully.
1919  */
1920 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1921                   int (*do_msr)(struct kvm_vcpu *vcpu,
1922                                 unsigned index, u64 *data),
1923                   int writeback)
1924 {
1925         struct kvm_msrs msrs;
1926         struct kvm_msr_entry *entries;
1927         int r, n;
1928         unsigned size;
1929
1930         r = -EFAULT;
1931         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1932                 goto out;
1933
1934         r = -E2BIG;
1935         if (msrs.nmsrs >= MAX_IO_MSRS)
1936                 goto out;
1937
1938         r = -ENOMEM;
1939         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1940         entries = kmalloc(size, GFP_KERNEL);
1941         if (!entries)
1942                 goto out;
1943
1944         r = -EFAULT;
1945         if (copy_from_user(entries, user_msrs->entries, size))
1946                 goto out_free;
1947
1948         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1949         if (r < 0)
1950                 goto out_free;
1951
1952         r = -EFAULT;
1953         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1954                 goto out_free;
1955
1956         r = n;
1957
1958 out_free:
1959         kfree(entries);
1960 out:
1961         return r;
1962 }
1963
1964 int kvm_dev_ioctl_check_extension(long ext)
1965 {
1966         int r;
1967
1968         switch (ext) {
1969         case KVM_CAP_IRQCHIP:
1970         case KVM_CAP_HLT:
1971         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1972         case KVM_CAP_SET_TSS_ADDR:
1973         case KVM_CAP_EXT_CPUID:
1974         case KVM_CAP_CLOCKSOURCE:
1975         case KVM_CAP_PIT:
1976         case KVM_CAP_NOP_IO_DELAY:
1977         case KVM_CAP_MP_STATE:
1978         case KVM_CAP_SYNC_MMU:
1979         case KVM_CAP_USER_NMI:
1980         case KVM_CAP_REINJECT_CONTROL:
1981         case KVM_CAP_IRQ_INJECT_STATUS:
1982         case KVM_CAP_ASSIGN_DEV_IRQ:
1983         case KVM_CAP_IRQFD:
1984         case KVM_CAP_IOEVENTFD:
1985         case KVM_CAP_PIT2:
1986         case KVM_CAP_PIT_STATE2:
1987         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1988         case KVM_CAP_XEN_HVM:
1989         case KVM_CAP_ADJUST_CLOCK:
1990         case KVM_CAP_VCPU_EVENTS:
1991         case KVM_CAP_HYPERV:
1992         case KVM_CAP_HYPERV_VAPIC:
1993         case KVM_CAP_HYPERV_SPIN:
1994         case KVM_CAP_PCI_SEGMENT:
1995         case KVM_CAP_DEBUGREGS:
1996         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1997         case KVM_CAP_XSAVE:
1998         case KVM_CAP_ASYNC_PF:
1999         case KVM_CAP_GET_TSC_KHZ:
2000                 r = 1;
2001                 break;
2002         case KVM_CAP_COALESCED_MMIO:
2003                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2004                 break;
2005         case KVM_CAP_VAPIC:
2006                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2007                 break;
2008         case KVM_CAP_NR_VCPUS:
2009                 r = KVM_MAX_VCPUS;
2010                 break;
2011         case KVM_CAP_NR_MEMSLOTS:
2012                 r = KVM_MEMORY_SLOTS;
2013                 break;
2014         case KVM_CAP_PV_MMU:    /* obsolete */
2015                 r = 0;
2016                 break;
2017         case KVM_CAP_IOMMU:
2018                 r = iommu_found();
2019                 break;
2020         case KVM_CAP_MCE:
2021                 r = KVM_MAX_MCE_BANKS;
2022                 break;
2023         case KVM_CAP_XCRS:
2024                 r = cpu_has_xsave;
2025                 break;
2026         case KVM_CAP_TSC_CONTROL:
2027                 r = kvm_has_tsc_control;
2028                 break;
2029         default:
2030                 r = 0;
2031                 break;
2032         }
2033         return r;
2034
2035 }
2036
2037 long kvm_arch_dev_ioctl(struct file *filp,
2038                         unsigned int ioctl, unsigned long arg)
2039 {
2040         void __user *argp = (void __user *)arg;
2041         long r;
2042
2043         switch (ioctl) {
2044         case KVM_GET_MSR_INDEX_LIST: {
2045                 struct kvm_msr_list __user *user_msr_list = argp;
2046                 struct kvm_msr_list msr_list;
2047                 unsigned n;
2048
2049                 r = -EFAULT;
2050                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2051                         goto out;
2052                 n = msr_list.nmsrs;
2053                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2054                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2055                         goto out;
2056                 r = -E2BIG;
2057                 if (n < msr_list.nmsrs)
2058                         goto out;
2059                 r = -EFAULT;
2060                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2061                                  num_msrs_to_save * sizeof(u32)))
2062                         goto out;
2063                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2064                                  &emulated_msrs,
2065                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2066                         goto out;
2067                 r = 0;
2068                 break;
2069         }
2070         case KVM_GET_SUPPORTED_CPUID: {
2071                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2072                 struct kvm_cpuid2 cpuid;
2073
2074                 r = -EFAULT;
2075                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2076                         goto out;
2077                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2078                                                       cpuid_arg->entries);
2079                 if (r)
2080                         goto out;
2081
2082                 r = -EFAULT;
2083                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2084                         goto out;
2085                 r = 0;
2086                 break;
2087         }
2088         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2089                 u64 mce_cap;
2090
2091                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2092                 r = -EFAULT;
2093                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2094                         goto out;
2095                 r = 0;
2096                 break;
2097         }
2098         default:
2099                 r = -EINVAL;
2100         }
2101 out:
2102         return r;
2103 }
2104
2105 static void wbinvd_ipi(void *garbage)
2106 {
2107         wbinvd();
2108 }
2109
2110 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2111 {
2112         return vcpu->kvm->arch.iommu_domain &&
2113                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2114 }
2115
2116 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2117 {
2118         /* Address WBINVD may be executed by guest */
2119         if (need_emulate_wbinvd(vcpu)) {
2120                 if (kvm_x86_ops->has_wbinvd_exit())
2121                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2122                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2123                         smp_call_function_single(vcpu->cpu,
2124                                         wbinvd_ipi, NULL, 1);
2125         }
2126
2127         kvm_x86_ops->vcpu_load(vcpu, cpu);
2128         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2129                 /* Make sure TSC doesn't go backwards */
2130                 s64 tsc_delta;
2131                 u64 tsc;
2132
2133                 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2134                 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2135                              tsc - vcpu->arch.last_guest_tsc;
2136
2137                 if (tsc_delta < 0)
2138                         mark_tsc_unstable("KVM discovered backwards TSC");
2139                 if (check_tsc_unstable()) {
2140                         kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2141                         vcpu->arch.tsc_catchup = 1;
2142                 }
2143                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2144                 if (vcpu->cpu != cpu)
2145                         kvm_migrate_timers(vcpu);
2146                 vcpu->cpu = cpu;
2147         }
2148 }
2149
2150 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2151 {
2152         kvm_x86_ops->vcpu_put(vcpu);
2153         kvm_put_guest_fpu(vcpu);
2154         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2155 }
2156
2157 static int is_efer_nx(void)
2158 {
2159         unsigned long long efer = 0;
2160
2161         rdmsrl_safe(MSR_EFER, &efer);
2162         return efer & EFER_NX;
2163 }
2164
2165 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2166 {
2167         int i;
2168         struct kvm_cpuid_entry2 *e, *entry;
2169
2170         entry = NULL;
2171         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2172                 e = &vcpu->arch.cpuid_entries[i];
2173                 if (e->function == 0x80000001) {
2174                         entry = e;
2175                         break;
2176                 }
2177         }
2178         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2179                 entry->edx &= ~(1 << 20);
2180                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2181         }
2182 }
2183
2184 /* when an old userspace process fills a new kernel module */
2185 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2186                                     struct kvm_cpuid *cpuid,
2187                                     struct kvm_cpuid_entry __user *entries)
2188 {
2189         int r, i;
2190         struct kvm_cpuid_entry *cpuid_entries;
2191
2192         r = -E2BIG;
2193         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2194                 goto out;
2195         r = -ENOMEM;
2196         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2197         if (!cpuid_entries)
2198                 goto out;
2199         r = -EFAULT;
2200         if (copy_from_user(cpuid_entries, entries,
2201                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2202                 goto out_free;
2203         for (i = 0; i < cpuid->nent; i++) {
2204                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2205                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2206                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2207                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2208                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2209                 vcpu->arch.cpuid_entries[i].index = 0;
2210                 vcpu->arch.cpuid_entries[i].flags = 0;
2211                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2212                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2213                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2214         }
2215         vcpu->arch.cpuid_nent = cpuid->nent;
2216         cpuid_fix_nx_cap(vcpu);
2217         r = 0;
2218         kvm_apic_set_version(vcpu);
2219         kvm_x86_ops->cpuid_update(vcpu);
2220         update_cpuid(vcpu);
2221
2222 out_free:
2223         vfree(cpuid_entries);
2224 out:
2225         return r;
2226 }
2227
2228 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2229                                      struct kvm_cpuid2 *cpuid,
2230                                      struct kvm_cpuid_entry2 __user *entries)
2231 {
2232         int r;
2233
2234         r = -E2BIG;
2235         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2236                 goto out;
2237         r = -EFAULT;
2238         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2239                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2240                 goto out;
2241         vcpu->arch.cpuid_nent = cpuid->nent;
2242         kvm_apic_set_version(vcpu);
2243         kvm_x86_ops->cpuid_update(vcpu);
2244         update_cpuid(vcpu);
2245         return 0;
2246
2247 out:
2248         return r;
2249 }
2250
2251 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2252                                      struct kvm_cpuid2 *cpuid,
2253                                      struct kvm_cpuid_entry2 __user *entries)
2254 {
2255         int r;
2256
2257         r = -E2BIG;
2258         if (cpuid->nent < vcpu->arch.cpuid_nent)
2259                 goto out;
2260         r = -EFAULT;
2261         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2262                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2263                 goto out;
2264         return 0;
2265
2266 out:
2267         cpuid->nent = vcpu->arch.cpuid_nent;
2268         return r;
2269 }
2270
2271 static void cpuid_mask(u32 *word, int wordnum)
2272 {
2273         *word &= boot_cpu_data.x86_capability[wordnum];
2274 }
2275
2276 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2277                            u32 index)
2278 {
2279         entry->function = function;
2280         entry->index = index;
2281         cpuid_count(entry->function, entry->index,
2282                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2283         entry->flags = 0;
2284 }
2285
2286 #define F(x) bit(X86_FEATURE_##x)
2287
2288 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2289                          u32 index, int *nent, int maxnent)
2290 {
2291         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2292 #ifdef CONFIG_X86_64
2293         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2294                                 ? F(GBPAGES) : 0;
2295         unsigned f_lm = F(LM);
2296 #else
2297         unsigned f_gbpages = 0;
2298         unsigned f_lm = 0;
2299 #endif
2300         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2301
2302         /* cpuid 1.edx */
2303         const u32 kvm_supported_word0_x86_features =
2304                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2305                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2306                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2307                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2308                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2309                 0 /* Reserved, DS, ACPI */ | F(MMX) |
2310                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2311                 0 /* HTT, TM, Reserved, PBE */;
2312         /* cpuid 0x80000001.edx */
2313         const u32 kvm_supported_word1_x86_features =
2314                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2315                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2316                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2317                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2318                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2319                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2320                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2321                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2322         /* cpuid 1.ecx */
2323         const u32 kvm_supported_word4_x86_features =
2324                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2325                 0 /* DS-CPL, VMX, SMX, EST */ |
2326                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2327                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2328                 0 /* Reserved, DCA */ | F(XMM4_1) |
2329                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2330                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2331                 F(F16C);
2332         /* cpuid 0x80000001.ecx */
2333         const u32 kvm_supported_word6_x86_features =
2334                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2335                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2336                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2337                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2338
2339         /* all calls to cpuid_count() should be made on the same cpu */
2340         get_cpu();
2341         do_cpuid_1_ent(entry, function, index);
2342         ++*nent;
2343
2344         switch (function) {
2345         case 0:
2346                 entry->eax = min(entry->eax, (u32)0xd);
2347                 break;
2348         case 1:
2349                 entry->edx &= kvm_supported_word0_x86_features;
2350                 cpuid_mask(&entry->edx, 0);
2351                 entry->ecx &= kvm_supported_word4_x86_features;
2352                 cpuid_mask(&entry->ecx, 4);
2353                 /* we support x2apic emulation even if host does not support
2354                  * it since we emulate x2apic in software */
2355                 entry->ecx |= F(X2APIC);
2356                 break;
2357         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2358          * may return different values. This forces us to get_cpu() before
2359          * issuing the first command, and also to emulate this annoying behavior
2360          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2361         case 2: {
2362                 int t, times = entry->eax & 0xff;
2363
2364                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2365                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2366                 for (t = 1; t < times && *nent < maxnent; ++t) {
2367                         do_cpuid_1_ent(&entry[t], function, 0);
2368                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2369                         ++*nent;
2370                 }
2371                 break;
2372         }
2373         /* function 4 and 0xb have additional index. */
2374         case 4: {
2375                 int i, cache_type;
2376
2377                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2378                 /* read more entries until cache_type is zero */
2379                 for (i = 1; *nent < maxnent; ++i) {
2380                         cache_type = entry[i - 1].eax & 0x1f;
2381                         if (!cache_type)
2382                                 break;
2383                         do_cpuid_1_ent(&entry[i], function, i);
2384                         entry[i].flags |=
2385                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2386                         ++*nent;
2387                 }
2388                 break;
2389         }
2390         case 0xb: {
2391                 int i, level_type;
2392
2393                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2394                 /* read more entries until level_type is zero */
2395                 for (i = 1; *nent < maxnent; ++i) {
2396                         level_type = entry[i - 1].ecx & 0xff00;
2397                         if (!level_type)
2398                                 break;
2399                         do_cpuid_1_ent(&entry[i], function, i);
2400                         entry[i].flags |=
2401                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2402                         ++*nent;
2403                 }
2404                 break;
2405         }
2406         case 0xd: {
2407                 int i;
2408
2409                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2410                 for (i = 1; *nent < maxnent && i < 64; ++i) {
2411                         if (entry[i].eax == 0)
2412                                 continue;
2413                         do_cpuid_1_ent(&entry[i], function, i);
2414                         entry[i].flags |=
2415                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2416                         ++*nent;
2417                 }
2418                 break;
2419         }
2420         case KVM_CPUID_SIGNATURE: {
2421                 char signature[12] = "KVMKVMKVM\0\0";
2422                 u32 *sigptr = (u32 *)signature;
2423                 entry->eax = 0;
2424                 entry->ebx = sigptr[0];
2425                 entry->ecx = sigptr[1];
2426                 entry->edx = sigptr[2];
2427                 break;
2428         }
2429         case KVM_CPUID_FEATURES:
2430                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2431                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2432                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2433                              (1 << KVM_FEATURE_ASYNC_PF) |
2434                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2435                 entry->ebx = 0;
2436                 entry->ecx = 0;
2437                 entry->edx = 0;
2438                 break;
2439         case 0x80000000:
2440                 entry->eax = min(entry->eax, 0x8000001a);
2441                 break;
2442         case 0x80000001:
2443                 entry->edx &= kvm_supported_word1_x86_features;
2444                 cpuid_mask(&entry->edx, 1);
2445                 entry->ecx &= kvm_supported_word6_x86_features;
2446                 cpuid_mask(&entry->ecx, 6);
2447                 break;
2448         }
2449
2450         kvm_x86_ops->set_supported_cpuid(function, entry);
2451
2452         put_cpu();
2453 }
2454
2455 #undef F
2456
2457 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2458                                      struct kvm_cpuid_entry2 __user *entries)
2459 {
2460         struct kvm_cpuid_entry2 *cpuid_entries;
2461         int limit, nent = 0, r = -E2BIG;
2462         u32 func;
2463
2464         if (cpuid->nent < 1)
2465                 goto out;
2466         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2467                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2468         r = -ENOMEM;
2469         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2470         if (!cpuid_entries)
2471                 goto out;
2472
2473         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2474         limit = cpuid_entries[0].eax;
2475         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2476                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2477                              &nent, cpuid->nent);
2478         r = -E2BIG;
2479         if (nent >= cpuid->nent)
2480                 goto out_free;
2481
2482         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2483         limit = cpuid_entries[nent - 1].eax;
2484         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2485                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2486                              &nent, cpuid->nent);
2487
2488
2489
2490         r = -E2BIG;
2491         if (nent >= cpuid->nent)
2492                 goto out_free;
2493
2494         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2495                      cpuid->nent);
2496
2497         r = -E2BIG;
2498         if (nent >= cpuid->nent)
2499                 goto out_free;
2500
2501         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2502                      cpuid->nent);
2503
2504         r = -E2BIG;
2505         if (nent >= cpuid->nent)
2506                 goto out_free;
2507
2508         r = -EFAULT;
2509         if (copy_to_user(entries, cpuid_entries,
2510                          nent * sizeof(struct kvm_cpuid_entry2)))
2511                 goto out_free;
2512         cpuid->nent = nent;
2513         r = 0;
2514
2515 out_free:
2516         vfree(cpuid_entries);
2517 out:
2518         return r;
2519 }
2520
2521 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2522                                     struct kvm_lapic_state *s)
2523 {
2524         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2525
2526         return 0;
2527 }
2528
2529 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2530                                     struct kvm_lapic_state *s)
2531 {
2532         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2533         kvm_apic_post_state_restore(vcpu);
2534         update_cr8_intercept(vcpu);
2535
2536         return 0;
2537 }
2538
2539 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2540                                     struct kvm_interrupt *irq)
2541 {
2542         if (irq->irq < 0 || irq->irq >= 256)
2543                 return -EINVAL;
2544         if (irqchip_in_kernel(vcpu->kvm))
2545                 return -ENXIO;
2546
2547         kvm_queue_interrupt(vcpu, irq->irq, false);
2548         kvm_make_request(KVM_REQ_EVENT, vcpu);
2549
2550         return 0;
2551 }
2552
2553 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2554 {
2555         kvm_inject_nmi(vcpu);
2556
2557         return 0;
2558 }
2559
2560 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2561                                            struct kvm_tpr_access_ctl *tac)
2562 {
2563         if (tac->flags)
2564                 return -EINVAL;
2565         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2566         return 0;
2567 }
2568
2569 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2570                                         u64 mcg_cap)
2571 {
2572         int r;
2573         unsigned bank_num = mcg_cap & 0xff, bank;
2574
2575         r = -EINVAL;
2576         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2577                 goto out;
2578         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2579                 goto out;
2580         r = 0;
2581         vcpu->arch.mcg_cap = mcg_cap;
2582         /* Init IA32_MCG_CTL to all 1s */
2583         if (mcg_cap & MCG_CTL_P)
2584                 vcpu->arch.mcg_ctl = ~(u64)0;
2585         /* Init IA32_MCi_CTL to all 1s */
2586         for (bank = 0; bank < bank_num; bank++)
2587                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2588 out:
2589         return r;
2590 }
2591
2592 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2593                                       struct kvm_x86_mce *mce)
2594 {
2595         u64 mcg_cap = vcpu->arch.mcg_cap;
2596         unsigned bank_num = mcg_cap & 0xff;
2597         u64 *banks = vcpu->arch.mce_banks;
2598
2599         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2600                 return -EINVAL;
2601         /*
2602          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2603          * reporting is disabled
2604          */
2605         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2606             vcpu->arch.mcg_ctl != ~(u64)0)
2607                 return 0;
2608         banks += 4 * mce->bank;
2609         /*
2610          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2611          * reporting is disabled for the bank
2612          */
2613         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2614                 return 0;
2615         if (mce->status & MCI_STATUS_UC) {
2616                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2617                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2618                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2619                         return 0;
2620                 }
2621                 if (banks[1] & MCI_STATUS_VAL)
2622                         mce->status |= MCI_STATUS_OVER;
2623                 banks[2] = mce->addr;
2624                 banks[3] = mce->misc;
2625                 vcpu->arch.mcg_status = mce->mcg_status;
2626                 banks[1] = mce->status;
2627                 kvm_queue_exception(vcpu, MC_VECTOR);
2628         } else if (!(banks[1] & MCI_STATUS_VAL)
2629                    || !(banks[1] & MCI_STATUS_UC)) {
2630                 if (banks[1] & MCI_STATUS_VAL)
2631                         mce->status |= MCI_STATUS_OVER;
2632                 banks[2] = mce->addr;
2633                 banks[3] = mce->misc;
2634                 banks[1] = mce->status;
2635         } else
2636                 banks[1] |= MCI_STATUS_OVER;
2637         return 0;
2638 }
2639
2640 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2641                                                struct kvm_vcpu_events *events)
2642 {
2643         events->exception.injected =
2644                 vcpu->arch.exception.pending &&
2645                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2646         events->exception.nr = vcpu->arch.exception.nr;
2647         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2648         events->exception.pad = 0;
2649         events->exception.error_code = vcpu->arch.exception.error_code;
2650
2651         events->interrupt.injected =
2652                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2653         events->interrupt.nr = vcpu->arch.interrupt.nr;
2654         events->interrupt.soft = 0;
2655         events->interrupt.shadow =
2656                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2657                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2658
2659         events->nmi.injected = vcpu->arch.nmi_injected;
2660         events->nmi.pending = vcpu->arch.nmi_pending;
2661         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2662         events->nmi.pad = 0;
2663
2664         events->sipi_vector = vcpu->arch.sipi_vector;
2665
2666         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2667                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2668                          | KVM_VCPUEVENT_VALID_SHADOW);
2669         memset(&events->reserved, 0, sizeof(events->reserved));
2670 }
2671
2672 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2673                                               struct kvm_vcpu_events *events)
2674 {
2675         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2676                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2677                               | KVM_VCPUEVENT_VALID_SHADOW))
2678                 return -EINVAL;
2679
2680         vcpu->arch.exception.pending = events->exception.injected;
2681         vcpu->arch.exception.nr = events->exception.nr;
2682         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2683         vcpu->arch.exception.error_code = events->exception.error_code;
2684
2685         vcpu->arch.interrupt.pending = events->interrupt.injected;
2686         vcpu->arch.interrupt.nr = events->interrupt.nr;
2687         vcpu->arch.interrupt.soft = events->interrupt.soft;
2688         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2689                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2690                                                   events->interrupt.shadow);
2691
2692         vcpu->arch.nmi_injected = events->nmi.injected;
2693         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2694                 vcpu->arch.nmi_pending = events->nmi.pending;
2695         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2696
2697         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2698                 vcpu->arch.sipi_vector = events->sipi_vector;
2699
2700         kvm_make_request(KVM_REQ_EVENT, vcpu);
2701
2702         return 0;
2703 }
2704
2705 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2706                                              struct kvm_debugregs *dbgregs)
2707 {
2708         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2709         dbgregs->dr6 = vcpu->arch.dr6;
2710         dbgregs->dr7 = vcpu->arch.dr7;
2711         dbgregs->flags = 0;
2712         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2713 }
2714
2715 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2716                                             struct kvm_debugregs *dbgregs)
2717 {
2718         if (dbgregs->flags)
2719                 return -EINVAL;
2720
2721         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2722         vcpu->arch.dr6 = dbgregs->dr6;
2723         vcpu->arch.dr7 = dbgregs->dr7;
2724
2725         return 0;
2726 }
2727
2728 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2729                                          struct kvm_xsave *guest_xsave)
2730 {
2731         if (cpu_has_xsave)
2732                 memcpy(guest_xsave->region,
2733                         &vcpu->arch.guest_fpu.state->xsave,
2734                         xstate_size);
2735         else {
2736                 memcpy(guest_xsave->region,
2737                         &vcpu->arch.guest_fpu.state->fxsave,
2738                         sizeof(struct i387_fxsave_struct));
2739                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2740                         XSTATE_FPSSE;
2741         }
2742 }
2743
2744 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2745                                         struct kvm_xsave *guest_xsave)
2746 {
2747         u64 xstate_bv =
2748                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2749
2750         if (cpu_has_xsave)
2751                 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2752                         guest_xsave->region, xstate_size);
2753         else {
2754                 if (xstate_bv & ~XSTATE_FPSSE)
2755                         return -EINVAL;
2756                 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2757                         guest_xsave->region, sizeof(struct i387_fxsave_struct));
2758         }
2759         return 0;
2760 }
2761
2762 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2763                                         struct kvm_xcrs *guest_xcrs)
2764 {
2765         if (!cpu_has_xsave) {
2766                 guest_xcrs->nr_xcrs = 0;
2767                 return;
2768         }
2769
2770         guest_xcrs->nr_xcrs = 1;
2771         guest_xcrs->flags = 0;
2772         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2773         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2774 }
2775
2776 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2777                                        struct kvm_xcrs *guest_xcrs)
2778 {
2779         int i, r = 0;
2780
2781         if (!cpu_has_xsave)
2782                 return -EINVAL;
2783
2784         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2785                 return -EINVAL;
2786
2787         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2788                 /* Only support XCR0 currently */
2789                 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2790                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2791                                 guest_xcrs->xcrs[0].value);
2792                         break;
2793                 }
2794         if (r)
2795                 r = -EINVAL;
2796         return r;
2797 }
2798
2799 long kvm_arch_vcpu_ioctl(struct file *filp,
2800                          unsigned int ioctl, unsigned long arg)
2801 {
2802         struct kvm_vcpu *vcpu = filp->private_data;
2803         void __user *argp = (void __user *)arg;
2804         int r;
2805         union {
2806                 struct kvm_lapic_state *lapic;
2807                 struct kvm_xsave *xsave;
2808                 struct kvm_xcrs *xcrs;
2809                 void *buffer;
2810         } u;
2811
2812         u.buffer = NULL;
2813         switch (ioctl) {
2814         case KVM_GET_LAPIC: {
2815                 r = -EINVAL;
2816                 if (!vcpu->arch.apic)
2817                         goto out;
2818                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2819
2820                 r = -ENOMEM;
2821                 if (!u.lapic)
2822                         goto out;
2823                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2824                 if (r)
2825                         goto out;
2826                 r = -EFAULT;
2827                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2828                         goto out;
2829                 r = 0;
2830                 break;
2831         }
2832         case KVM_SET_LAPIC: {
2833                 r = -EINVAL;
2834                 if (!vcpu->arch.apic)
2835                         goto out;
2836                 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2837                 r = -ENOMEM;
2838                 if (!u.lapic)
2839                         goto out;
2840                 r = -EFAULT;
2841                 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2842                         goto out;
2843                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2844                 if (r)
2845                         goto out;
2846                 r = 0;
2847                 break;
2848         }
2849         case KVM_INTERRUPT: {
2850                 struct kvm_interrupt irq;
2851
2852                 r = -EFAULT;
2853                 if (copy_from_user(&irq, argp, sizeof irq))
2854                         goto out;
2855                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2856                 if (r)
2857                         goto out;
2858                 r = 0;
2859                 break;
2860         }
2861         case KVM_NMI: {
2862                 r = kvm_vcpu_ioctl_nmi(vcpu);
2863                 if (r)
2864                         goto out;
2865                 r = 0;
2866                 break;
2867         }
2868         case KVM_SET_CPUID: {
2869                 struct kvm_cpuid __user *cpuid_arg = argp;
2870                 struct kvm_cpuid cpuid;
2871
2872                 r = -EFAULT;
2873                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2874                         goto out;
2875                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2876                 if (r)
2877                         goto out;
2878                 break;
2879         }
2880         case KVM_SET_CPUID2: {
2881                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2882                 struct kvm_cpuid2 cpuid;
2883
2884                 r = -EFAULT;
2885                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2886                         goto out;
2887                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2888                                               cpuid_arg->entries);
2889                 if (r)
2890                         goto out;
2891                 break;
2892         }
2893         case KVM_GET_CPUID2: {
2894                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2895                 struct kvm_cpuid2 cpuid;
2896
2897                 r = -EFAULT;
2898                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2899                         goto out;
2900                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2901                                               cpuid_arg->entries);
2902                 if (r)
2903                         goto out;
2904                 r = -EFAULT;
2905                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2906                         goto out;
2907                 r = 0;
2908                 break;
2909         }
2910         case KVM_GET_MSRS:
2911                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2912                 break;
2913         case KVM_SET_MSRS:
2914                 r = msr_io(vcpu, argp, do_set_msr, 0);
2915                 break;
2916         case KVM_TPR_ACCESS_REPORTING: {
2917                 struct kvm_tpr_access_ctl tac;
2918
2919                 r = -EFAULT;
2920                 if (copy_from_user(&tac, argp, sizeof tac))
2921                         goto out;
2922                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2923                 if (r)
2924                         goto out;
2925                 r = -EFAULT;
2926                 if (copy_to_user(argp, &tac, sizeof tac))
2927                         goto out;
2928                 r = 0;
2929                 break;
2930         };
2931         case KVM_SET_VAPIC_ADDR: {
2932                 struct kvm_vapic_addr va;
2933
2934                 r = -EINVAL;
2935                 if (!irqchip_in_kernel(vcpu->kvm))
2936                         goto out;
2937                 r = -EFAULT;
2938                 if (copy_from_user(&va, argp, sizeof va))
2939                         goto out;
2940                 r = 0;
2941                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2942                 break;
2943         }
2944         case KVM_X86_SETUP_MCE: {
2945                 u64 mcg_cap;
2946
2947                 r = -EFAULT;
2948                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2949                         goto out;
2950                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2951                 break;
2952         }
2953         case KVM_X86_SET_MCE: {
2954                 struct kvm_x86_mce mce;
2955
2956                 r = -EFAULT;
2957                 if (copy_from_user(&mce, argp, sizeof mce))
2958                         goto out;
2959                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2960                 break;
2961         }
2962         case KVM_GET_VCPU_EVENTS: {
2963                 struct kvm_vcpu_events events;
2964
2965                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2966
2967                 r = -EFAULT;
2968                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2969                         break;
2970                 r = 0;
2971                 break;
2972         }
2973         case KVM_SET_VCPU_EVENTS: {
2974                 struct kvm_vcpu_events events;
2975
2976                 r = -EFAULT;
2977                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2978                         break;
2979
2980                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2981                 break;
2982         }
2983         case KVM_GET_DEBUGREGS: {
2984                 struct kvm_debugregs dbgregs;
2985
2986                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2987
2988                 r = -EFAULT;
2989                 if (copy_to_user(argp, &dbgregs,
2990                                  sizeof(struct kvm_debugregs)))
2991                         break;
2992                 r = 0;
2993                 break;
2994         }
2995         case KVM_SET_DEBUGREGS: {
2996                 struct kvm_debugregs dbgregs;
2997
2998                 r = -EFAULT;
2999                 if (copy_from_user(&dbgregs, argp,
3000                                    sizeof(struct kvm_debugregs)))
3001                         break;
3002
3003                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3004                 break;
3005         }
3006         case KVM_GET_XSAVE: {
3007                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3008                 r = -ENOMEM;
3009                 if (!u.xsave)
3010                         break;
3011
3012                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3013
3014                 r = -EFAULT;
3015                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3016                         break;
3017                 r = 0;
3018                 break;
3019         }
3020         case KVM_SET_XSAVE: {
3021                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3022                 r = -ENOMEM;
3023                 if (!u.xsave)
3024                         break;
3025
3026                 r = -EFAULT;
3027                 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3028                         break;
3029
3030                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3031                 break;
3032         }
3033         case KVM_GET_XCRS: {
3034                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3035                 r = -ENOMEM;
3036                 if (!u.xcrs)
3037                         break;
3038
3039                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3040
3041                 r = -EFAULT;
3042                 if (copy_to_user(argp, u.xcrs,
3043                                  sizeof(struct kvm_xcrs)))
3044                         break;
3045                 r = 0;
3046                 break;
3047         }
3048         case KVM_SET_XCRS: {
3049                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3050                 r = -ENOMEM;
3051                 if (!u.xcrs)
3052                         break;
3053
3054                 r = -EFAULT;
3055                 if (copy_from_user(u.xcrs, argp,
3056                                    sizeof(struct kvm_xcrs)))
3057                         break;
3058
3059                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3060                 break;
3061         }
3062         case KVM_SET_TSC_KHZ: {
3063                 u32 user_tsc_khz;
3064
3065                 r = -EINVAL;
3066                 if (!kvm_has_tsc_control)
3067                         break;
3068
3069                 user_tsc_khz = (u32)arg;
3070
3071                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3072                         goto out;
3073
3074                 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3075
3076                 r = 0;
3077                 goto out;
3078         }
3079         case KVM_GET_TSC_KHZ: {
3080                 r = -EIO;
3081                 if (check_tsc_unstable())
3082                         goto out;
3083
3084                 r = vcpu_tsc_khz(vcpu);
3085
3086                 goto out;
3087         }
3088         default:
3089                 r = -EINVAL;
3090         }
3091 out:
3092         kfree(u.buffer);
3093         return r;
3094 }
3095
3096 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3097 {
3098         int ret;
3099
3100         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3101                 return -1;
3102         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3103         return ret;
3104 }
3105
3106 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3107                                               u64 ident_addr)
3108 {
3109         kvm->arch.ept_identity_map_addr = ident_addr;
3110         return 0;
3111 }
3112
3113 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3114                                           u32 kvm_nr_mmu_pages)
3115 {
3116         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3117                 return -EINVAL;
3118
3119         mutex_lock(&kvm->slots_lock);
3120         spin_lock(&kvm->mmu_lock);
3121
3122         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3123         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3124
3125         spin_unlock(&kvm->mmu_lock);
3126         mutex_unlock(&kvm->slots_lock);
3127         return 0;
3128 }
3129
3130 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3131 {
3132         return kvm->arch.n_max_mmu_pages;
3133 }
3134
3135 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3136 {
3137         int r;
3138
3139         r = 0;
3140         switch (chip->chip_id) {
3141         case KVM_IRQCHIP_PIC_MASTER:
3142                 memcpy(&chip->chip.pic,
3143                         &pic_irqchip(kvm)->pics[0],
3144                         sizeof(struct kvm_pic_state));
3145                 break;
3146         case KVM_IRQCHIP_PIC_SLAVE:
3147                 memcpy(&chip->chip.pic,
3148                         &pic_irqchip(kvm)->pics[1],
3149                         sizeof(struct kvm_pic_state));
3150                 break;
3151         case KVM_IRQCHIP_IOAPIC:
3152                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3153                 break;
3154         default:
3155                 r = -EINVAL;
3156                 break;
3157         }
3158         return r;
3159 }
3160
3161 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3162 {
3163         int r;
3164
3165         r = 0;
3166         switch (chip->chip_id) {
3167         case KVM_IRQCHIP_PIC_MASTER:
3168                 spin_lock(&pic_irqchip(kvm)->lock);
3169                 memcpy(&pic_irqchip(kvm)->pics[0],
3170                         &chip->chip.pic,
3171                         sizeof(struct kvm_pic_state));
3172                 spin_unlock(&pic_irqchip(kvm)->lock);
3173                 break;
3174         case KVM_IRQCHIP_PIC_SLAVE:
3175                 spin_lock(&pic_irqchip(kvm)->lock);
3176                 memcpy(&pic_irqchip(kvm)->pics[1],
3177                         &chip->chip.pic,
3178                         sizeof(struct kvm_pic_state));
3179                 spin_unlock(&pic_irqchip(kvm)->lock);
3180                 break;
3181         case KVM_IRQCHIP_IOAPIC:
3182                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3183                 break;
3184         default:
3185                 r = -EINVAL;
3186                 break;
3187         }
3188         kvm_pic_update_irq(pic_irqchip(kvm));
3189         return r;
3190 }
3191
3192 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3193 {
3194         int r = 0;
3195
3196         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3197         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3198         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3199         return r;
3200 }
3201
3202 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3203 {
3204         int r = 0;
3205
3206         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3207         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3208         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3209         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3210         return r;
3211 }
3212
3213 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3214 {
3215         int r = 0;
3216
3217         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3218         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3219                 sizeof(ps->channels));
3220         ps->flags = kvm->arch.vpit->pit_state.flags;
3221         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3222         memset(&ps->reserved, 0, sizeof(ps->reserved));
3223         return r;
3224 }
3225
3226 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3227 {
3228         int r = 0, start = 0;
3229         u32 prev_legacy, cur_legacy;
3230         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3231         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3232         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3233         if (!prev_legacy && cur_legacy)
3234                 start = 1;
3235         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3236                sizeof(kvm->arch.vpit->pit_state.channels));
3237         kvm->arch.vpit->pit_state.flags = ps->flags;
3238         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3239         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3240         return r;
3241 }
3242
3243 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3244                                  struct kvm_reinject_control *control)
3245 {
3246         if (!kvm->arch.vpit)
3247                 return -ENXIO;
3248         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3249         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3250         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3251         return 0;
3252 }
3253
3254 /*
3255  * Get (and clear) the dirty memory log for a memory slot.
3256  */
3257 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3258                                       struct kvm_dirty_log *log)
3259 {
3260         int r, i;
3261         struct kvm_memory_slot *memslot;
3262         unsigned long n;
3263         unsigned long is_dirty = 0;
3264
3265         mutex_lock(&kvm->slots_lock);
3266
3267         r = -EINVAL;
3268         if (log->slot >= KVM_MEMORY_SLOTS)
3269                 goto out;
3270
3271         memslot = &kvm->memslots->memslots[log->slot];
3272         r = -ENOENT;
3273         if (!memslot->dirty_bitmap)
3274                 goto out;
3275
3276         n = kvm_dirty_bitmap_bytes(memslot);
3277
3278         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3279                 is_dirty = memslot->dirty_bitmap[i];
3280
3281         /* If nothing is dirty, don't bother messing with page tables. */
3282         if (is_dirty) {
3283                 struct kvm_memslots *slots, *old_slots;
3284                 unsigned long *dirty_bitmap;
3285
3286                 dirty_bitmap = memslot->dirty_bitmap_head;
3287                 if (memslot->dirty_bitmap == dirty_bitmap)
3288                         dirty_bitmap += n / sizeof(long);
3289                 memset(dirty_bitmap, 0, n);
3290
3291                 r = -ENOMEM;
3292                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3293                 if (!slots)
3294                         goto out;
3295                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3296                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3297                 slots->generation++;
3298
3299                 old_slots = kvm->memslots;
3300                 rcu_assign_pointer(kvm->memslots, slots);
3301                 synchronize_srcu_expedited(&kvm->srcu);
3302                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3303                 kfree(old_slots);
3304
3305                 spin_lock(&kvm->mmu_lock);
3306                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3307                 spin_unlock(&kvm->mmu_lock);
3308
3309                 r = -EFAULT;
3310                 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3311                         goto out;
3312         } else {
3313                 r = -EFAULT;
3314                 if (clear_user(log->dirty_bitmap, n))
3315                         goto out;
3316         }
3317
3318         r = 0;
3319 out:
3320         mutex_unlock(&kvm->slots_lock);
3321         return r;
3322 }
3323
3324 long kvm_arch_vm_ioctl(struct file *filp,
3325                        unsigned int ioctl, unsigned long arg)
3326 {
3327         struct kvm *kvm = filp->private_data;
3328         void __user *argp = (void __user *)arg;
3329         int r = -ENOTTY;
3330         /*
3331          * This union makes it completely explicit to gcc-3.x
3332          * that these two variables' stack usage should be
3333          * combined, not added together.
3334          */
3335         union {
3336                 struct kvm_pit_state ps;
3337                 struct kvm_pit_state2 ps2;
3338                 struct kvm_pit_config pit_config;
3339         } u;
3340
3341         switch (ioctl) {
3342         case KVM_SET_TSS_ADDR:
3343                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3344                 if (r < 0)
3345                         goto out;
3346                 break;
3347         case KVM_SET_IDENTITY_MAP_ADDR: {
3348                 u64 ident_addr;
3349
3350                 r = -EFAULT;
3351                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3352                         goto out;
3353                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3354                 if (r < 0)
3355                         goto out;
3356                 break;
3357         }
3358         case KVM_SET_NR_MMU_PAGES:
3359                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3360                 if (r)
3361                         goto out;
3362                 break;
3363         case KVM_GET_NR_MMU_PAGES:
3364                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3365                 break;
3366         case KVM_CREATE_IRQCHIP: {
3367                 struct kvm_pic *vpic;
3368
3369                 mutex_lock(&kvm->lock);
3370                 r = -EEXIST;
3371                 if (kvm->arch.vpic)
3372                         goto create_irqchip_unlock;
3373                 r = -ENOMEM;
3374                 vpic = kvm_create_pic(kvm);
3375                 if (vpic) {
3376                         r = kvm_ioapic_init(kvm);
3377                         if (r) {
3378                                 mutex_lock(&kvm->slots_lock);
3379                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3380                                                           &vpic->dev);
3381                                 mutex_unlock(&kvm->slots_lock);
3382                                 kfree(vpic);
3383                                 goto create_irqchip_unlock;
3384                         }
3385                 } else
3386                         goto create_irqchip_unlock;
3387                 smp_wmb();
3388                 kvm->arch.vpic = vpic;
3389                 smp_wmb();
3390                 r = kvm_setup_default_irq_routing(kvm);
3391                 if (r) {
3392                         mutex_lock(&kvm->slots_lock);
3393                         mutex_lock(&kvm->irq_lock);
3394                         kvm_ioapic_destroy(kvm);
3395                         kvm_destroy_pic(kvm);
3396                         mutex_unlock(&kvm->irq_lock);
3397                         mutex_unlock(&kvm->slots_lock);
3398                 }
3399         create_irqchip_unlock:
3400                 mutex_unlock(&kvm->lock);
3401                 break;
3402         }
3403         case KVM_CREATE_PIT:
3404                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3405                 goto create_pit;
3406         case KVM_CREATE_PIT2:
3407                 r = -EFAULT;
3408                 if (copy_from_user(&u.pit_config, argp,
3409                                    sizeof(struct kvm_pit_config)))
3410                         goto out;
3411         create_pit:
3412                 mutex_lock(&kvm->slots_lock);
3413                 r = -EEXIST;
3414                 if (kvm->arch.vpit)
3415                         goto create_pit_unlock;
3416                 r = -ENOMEM;
3417                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3418                 if (kvm->arch.vpit)
3419                         r = 0;
3420         create_pit_unlock:
3421                 mutex_unlock(&kvm->slots_lock);
3422                 break;
3423         case KVM_IRQ_LINE_STATUS:
3424         case KVM_IRQ_LINE: {
3425                 struct kvm_irq_level irq_event;
3426
3427                 r = -EFAULT;
3428                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3429                         goto out;
3430                 r = -ENXIO;
3431                 if (irqchip_in_kernel(kvm)) {
3432                         __s32 status;
3433                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3434                                         irq_event.irq, irq_event.level);
3435                         if (ioctl == KVM_IRQ_LINE_STATUS) {
3436                                 r = -EFAULT;
3437                                 irq_event.status = status;
3438                                 if (copy_to_user(argp, &irq_event,
3439                                                         sizeof irq_event))
3440                                         goto out;
3441                         }
3442                         r = 0;
3443                 }
3444                 break;
3445         }
3446         case KVM_GET_IRQCHIP: {
3447                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3448                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3449
3450                 r = -ENOMEM;
3451                 if (!chip)
3452                         goto out;
3453                 r = -EFAULT;
3454                 if (copy_from_user(chip, argp, sizeof *chip))
3455                         goto get_irqchip_out;
3456                 r = -ENXIO;
3457                 if (!irqchip_in_kernel(kvm))
3458                         goto get_irqchip_out;
3459                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3460                 if (r)
3461                         goto get_irqchip_out;
3462                 r = -EFAULT;
3463                 if (copy_to_user(argp, chip, sizeof *chip))
3464                         goto get_irqchip_out;
3465                 r = 0;
3466         get_irqchip_out:
3467                 kfree(chip);
3468                 if (r)
3469                         goto out;
3470                 break;
3471         }
3472         case KVM_SET_IRQCHIP: {
3473                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3474                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3475
3476                 r = -ENOMEM;
3477                 if (!chip)
3478                         goto out;
3479                 r = -EFAULT;
3480                 if (copy_from_user(chip, argp, sizeof *chip))
3481                         goto set_irqchip_out;
3482                 r = -ENXIO;
3483                 if (!irqchip_in_kernel(kvm))
3484                         goto set_irqchip_out;
3485                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3486                 if (r)
3487                         goto set_irqchip_out;
3488                 r = 0;
3489         set_irqchip_out:
3490                 kfree(chip);
3491                 if (r)
3492                         goto out;
3493                 break;
3494         }
3495         case KVM_GET_PIT: {
3496                 r = -EFAULT;
3497                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3498                         goto out;
3499                 r = -ENXIO;
3500                 if (!kvm->arch.vpit)
3501                         goto out;
3502                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3503                 if (r)
3504                         goto out;
3505                 r = -EFAULT;
3506                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3507                         goto out;
3508                 r = 0;
3509                 break;
3510         }
3511         case KVM_SET_PIT: {
3512                 r = -EFAULT;
3513                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3514                         goto out;
3515                 r = -ENXIO;
3516                 if (!kvm->arch.vpit)
3517                         goto out;
3518                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3519                 if (r)
3520                         goto out;
3521                 r = 0;
3522                 break;
3523         }
3524         case KVM_GET_PIT2: {
3525                 r = -ENXIO;
3526                 if (!kvm->arch.vpit)
3527                         goto out;
3528                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3529                 if (r)
3530                         goto out;
3531                 r = -EFAULT;
3532                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3533                         goto out;
3534                 r = 0;
3535                 break;
3536         }
3537         case KVM_SET_PIT2: {
3538                 r = -EFAULT;
3539                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3540                         goto out;
3541                 r = -ENXIO;
3542                 if (!kvm->arch.vpit)
3543                         goto out;
3544                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3545                 if (r)
3546                         goto out;
3547                 r = 0;
3548                 break;
3549         }
3550         case KVM_REINJECT_CONTROL: {
3551                 struct kvm_reinject_control control;
3552                 r =  -EFAULT;
3553                 if (copy_from_user(&control, argp, sizeof(control)))
3554                         goto out;
3555                 r = kvm_vm_ioctl_reinject(kvm, &control);
3556                 if (r)
3557                         goto out;
3558                 r = 0;
3559                 break;
3560         }
3561         case KVM_XEN_HVM_CONFIG: {
3562                 r = -EFAULT;
3563                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3564                                    sizeof(struct kvm_xen_hvm_config)))
3565                         goto out;
3566                 r = -EINVAL;
3567                 if (kvm->arch.xen_hvm_config.flags)
3568                         goto out;
3569                 r = 0;
3570                 break;
3571         }
3572         case KVM_SET_CLOCK: {
3573                 struct kvm_clock_data user_ns;
3574                 u64 now_ns;
3575                 s64 delta;
3576
3577                 r = -EFAULT;
3578                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3579                         goto out;
3580
3581                 r = -EINVAL;
3582                 if (user_ns.flags)
3583                         goto out;
3584
3585                 r = 0;
3586                 local_irq_disable();
3587                 now_ns = get_kernel_ns();
3588                 delta = user_ns.clock - now_ns;
3589                 local_irq_enable();
3590                 kvm->arch.kvmclock_offset = delta;
3591                 break;
3592         }
3593         case KVM_GET_CLOCK: {
3594                 struct kvm_clock_data user_ns;
3595                 u64 now_ns;
3596
3597                 local_irq_disable();
3598                 now_ns = get_kernel_ns();
3599                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3600                 local_irq_enable();
3601                 user_ns.flags = 0;
3602                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3603
3604                 r = -EFAULT;
3605                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3606                         goto out;
3607                 r = 0;
3608                 break;
3609         }
3610
3611         default:
3612                 ;
3613         }
3614 out:
3615         return r;
3616 }
3617
3618 static void kvm_init_msr_list(void)
3619 {
3620         u32 dummy[2];
3621         unsigned i, j;
3622
3623         /* skip the first msrs in the list. KVM-specific */
3624         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3625                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3626                         continue;
3627                 if (j < i)
3628                         msrs_to_save[j] = msrs_to_save[i];
3629                 j++;
3630         }
3631         num_msrs_to_save = j;
3632 }
3633
3634 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3635                            const void *v)
3636 {
3637         int handled = 0;
3638         int n;
3639
3640         do {
3641                 n = min(len, 8);
3642                 if (!(vcpu->arch.apic &&
3643                       !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3644                     && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3645                         break;
3646                 handled += n;
3647                 addr += n;
3648                 len -= n;
3649                 v += n;
3650         } while (len);
3651
3652         return handled;
3653 }
3654
3655 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3656 {
3657         int handled = 0;
3658         int n;
3659
3660         do {
3661                 n = min(len, 8);
3662                 if (!(vcpu->arch.apic &&
3663                       !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3664                     && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3665                         break;
3666                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3667                 handled += n;
3668                 addr += n;
3669                 len -= n;
3670                 v += n;
3671         } while (len);
3672
3673         return handled;
3674 }
3675
3676 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3677                         struct kvm_segment *var, int seg)
3678 {
3679         kvm_x86_ops->set_segment(vcpu, var, seg);
3680 }
3681
3682 void kvm_get_segment(struct kvm_vcpu *vcpu,
3683                      struct kvm_segment *var, int seg)
3684 {
3685         kvm_x86_ops->get_segment(vcpu, var, seg);
3686 }
3687
3688 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3689 {
3690         return gpa;
3691 }
3692
3693 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3694 {
3695         gpa_t t_gpa;
3696         struct x86_exception exception;
3697
3698         BUG_ON(!mmu_is_nested(vcpu));
3699
3700         /* NPT walks are always user-walks */
3701         access |= PFERR_USER_MASK;
3702         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3703
3704         return t_gpa;
3705 }
3706
3707 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3708                               struct x86_exception *exception)
3709 {
3710         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3711         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3712 }
3713
3714  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3715                                 struct x86_exception *exception)
3716 {
3717         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3718         access |= PFERR_FETCH_MASK;
3719         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3720 }
3721
3722 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3723                                struct x86_exception *exception)
3724 {
3725         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3726         access |= PFERR_WRITE_MASK;
3727         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3728 }
3729
3730 /* uses this to access any guest's mapped memory without checking CPL */
3731 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3732                                 struct x86_exception *exception)
3733 {
3734         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3735 }
3736
3737 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3738                                       struct kvm_vcpu *vcpu, u32 access,
3739                                       struct x86_exception *exception)
3740 {
3741         void *data = val;
3742         int r = X86EMUL_CONTINUE;
3743
3744         while (bytes) {
3745                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3746                                                             exception);
3747                 unsigned offset = addr & (PAGE_SIZE-1);
3748                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3749                 int ret;
3750
3751                 if (gpa == UNMAPPED_GVA)
3752                         return X86EMUL_PROPAGATE_FAULT;
3753                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3754                 if (ret < 0) {
3755                         r = X86EMUL_IO_NEEDED;
3756                         goto out;
3757                 }
3758
3759                 bytes -= toread;
3760                 data += toread;
3761                 addr += toread;
3762         }
3763 out:
3764         return r;
3765 }
3766
3767 /* used for instruction fetching */
3768 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3769                                 gva_t addr, void *val, unsigned int bytes,
3770                                 struct x86_exception *exception)
3771 {
3772         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3773         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3774
3775         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3776                                           access | PFERR_FETCH_MASK,
3777                                           exception);
3778 }
3779
3780 static int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3781                                gva_t addr, void *val, unsigned int bytes,
3782                                struct x86_exception *exception)
3783 {
3784         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3785         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3786
3787         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3788                                           exception);
3789 }
3790
3791 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3792                                       gva_t addr, void *val, unsigned int bytes,
3793                                       struct x86_exception *exception)
3794 {
3795         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3796         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3797 }
3798
3799 static int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3800                                        gva_t addr, void *val,
3801                                        unsigned int bytes,
3802                                        struct x86_exception *exception)
3803 {
3804         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3805         void *data = val;
3806         int r = X86EMUL_CONTINUE;
3807
3808         while (bytes) {
3809                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3810                                                              PFERR_WRITE_MASK,
3811                                                              exception);
3812                 unsigned offset = addr & (PAGE_SIZE-1);
3813                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3814                 int ret;
3815
3816                 if (gpa == UNMAPPED_GVA)
3817                         return X86EMUL_PROPAGATE_FAULT;
3818                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3819                 if (ret < 0) {
3820                         r = X86EMUL_IO_NEEDED;
3821                         goto out;
3822                 }
3823
3824                 bytes -= towrite;
3825                 data += towrite;
3826                 addr += towrite;
3827         }
3828 out:
3829         return r;
3830 }
3831
3832 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3833                                   unsigned long addr,
3834                                   void *val,
3835                                   unsigned int bytes,
3836                                   struct x86_exception *exception)
3837 {
3838         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3839         gpa_t                 gpa;
3840         int handled;
3841
3842         if (vcpu->mmio_read_completed) {
3843                 memcpy(val, vcpu->mmio_data, bytes);
3844                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3845                                vcpu->mmio_phys_addr, *(u64 *)val);
3846                 vcpu->mmio_read_completed = 0;
3847                 return X86EMUL_CONTINUE;
3848         }
3849
3850         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3851
3852         if (gpa == UNMAPPED_GVA)
3853                 return X86EMUL_PROPAGATE_FAULT;
3854
3855         /* For APIC access vmexit */
3856         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3857                 goto mmio;
3858
3859         if (kvm_read_guest_virt(ctxt, addr, val, bytes, exception)
3860             == X86EMUL_CONTINUE)
3861                 return X86EMUL_CONTINUE;
3862
3863 mmio:
3864         /*
3865          * Is this MMIO handled locally?
3866          */
3867         handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3868
3869         if (handled == bytes)
3870                 return X86EMUL_CONTINUE;
3871
3872         gpa += handled;
3873         bytes -= handled;
3874         val += handled;
3875
3876         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3877
3878         vcpu->mmio_needed = 1;
3879         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3880         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3881         vcpu->mmio_size = bytes;
3882         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3883         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3884         vcpu->mmio_index = 0;
3885
3886         return X86EMUL_IO_NEEDED;
3887 }
3888
3889 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3890                         const void *val, int bytes)
3891 {
3892         int ret;
3893
3894         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3895         if (ret < 0)
3896                 return 0;
3897         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3898         return 1;
3899 }
3900
3901 static int emulator_write_emulated_onepage(unsigned long addr,
3902                                            const void *val,
3903                                            unsigned int bytes,
3904                                            struct x86_exception *exception,
3905                                            struct kvm_vcpu *vcpu)
3906 {
3907         gpa_t                 gpa;
3908         int handled;
3909
3910         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3911
3912         if (gpa == UNMAPPED_GVA)
3913                 return X86EMUL_PROPAGATE_FAULT;
3914
3915         /* For APIC access vmexit */
3916         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3917                 goto mmio;
3918
3919         if (emulator_write_phys(vcpu, gpa, val, bytes))
3920                 return X86EMUL_CONTINUE;
3921
3922 mmio:
3923         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3924         /*
3925          * Is this MMIO handled locally?
3926          */
3927         handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3928         if (handled == bytes)
3929                 return X86EMUL_CONTINUE;
3930
3931         gpa += handled;
3932         bytes -= handled;
3933         val += handled;
3934
3935         vcpu->mmio_needed = 1;
3936         memcpy(vcpu->mmio_data, val, bytes);
3937         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3938         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3939         vcpu->mmio_size = bytes;
3940         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3941         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3942         memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3943         vcpu->mmio_index = 0;
3944
3945         return X86EMUL_CONTINUE;
3946 }
3947
3948 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3949                             unsigned long addr,
3950                             const void *val,
3951                             unsigned int bytes,
3952                             struct x86_exception *exception)
3953 {
3954         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3955
3956         /* Crossing a page boundary? */
3957         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3958                 int rc, now;
3959
3960                 now = -addr & ~PAGE_MASK;
3961                 rc = emulator_write_emulated_onepage(addr, val, now, exception,
3962                                                      vcpu);
3963                 if (rc != X86EMUL_CONTINUE)
3964                         return rc;
3965                 addr += now;
3966                 val += now;
3967                 bytes -= now;
3968         }
3969         return emulator_write_emulated_onepage(addr, val, bytes, exception,
3970                                                vcpu);
3971 }
3972
3973 #define CMPXCHG_TYPE(t, ptr, old, new) \
3974         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3975
3976 #ifdef CONFIG_X86_64
3977 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3978 #else
3979 #  define CMPXCHG64(ptr, old, new) \
3980         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3981 #endif
3982
3983 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
3984                                      unsigned long addr,
3985                                      const void *old,
3986                                      const void *new,
3987                                      unsigned int bytes,
3988                                      struct x86_exception *exception)
3989 {
3990         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3991         gpa_t gpa;
3992         struct page *page;
3993         char *kaddr;
3994         bool exchanged;
3995
3996         /* guests cmpxchg8b have to be emulated atomically */
3997         if (bytes > 8 || (bytes & (bytes - 1)))
3998                 goto emul_write;
3999
4000         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4001
4002         if (gpa == UNMAPPED_GVA ||
4003             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4004                 goto emul_write;
4005
4006         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4007                 goto emul_write;
4008
4009         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4010         if (is_error_page(page)) {
4011                 kvm_release_page_clean(page);
4012                 goto emul_write;
4013         }
4014
4015         kaddr = kmap_atomic(page, KM_USER0);
4016         kaddr += offset_in_page(gpa);
4017         switch (bytes) {
4018         case 1:
4019                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4020                 break;
4021         case 2:
4022                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4023                 break;
4024         case 4:
4025                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4026                 break;
4027         case 8:
4028                 exchanged = CMPXCHG64(kaddr, old, new);
4029                 break;
4030         default:
4031                 BUG();
4032         }
4033         kunmap_atomic(kaddr, KM_USER0);
4034         kvm_release_page_dirty(page);
4035
4036         if (!exchanged)
4037                 return X86EMUL_CMPXCHG_FAILED;
4038
4039         kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4040
4041         return X86EMUL_CONTINUE;
4042
4043 emul_write:
4044         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4045
4046         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4047 }
4048
4049 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4050 {
4051         /* TODO: String I/O for in kernel device */
4052         int r;
4053
4054         if (vcpu->arch.pio.in)
4055                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4056                                     vcpu->arch.pio.size, pd);
4057         else
4058                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4059                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4060                                      pd);
4061         return r;
4062 }
4063
4064
4065 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4066                                     int size, unsigned short port, void *val,
4067                                     unsigned int count)
4068 {
4069         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4070
4071         if (vcpu->arch.pio.count)
4072                 goto data_avail;
4073
4074         trace_kvm_pio(0, port, size, count);
4075
4076         vcpu->arch.pio.port = port;
4077         vcpu->arch.pio.in = 1;
4078         vcpu->arch.pio.count  = count;
4079         vcpu->arch.pio.size = size;
4080
4081         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4082         data_avail:
4083                 memcpy(val, vcpu->arch.pio_data, size * count);
4084                 vcpu->arch.pio.count = 0;
4085                 return 1;
4086         }
4087
4088         vcpu->run->exit_reason = KVM_EXIT_IO;
4089         vcpu->run->io.direction = KVM_EXIT_IO_IN;
4090         vcpu->run->io.size = size;
4091         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4092         vcpu->run->io.count = count;
4093         vcpu->run->io.port = port;
4094
4095         return 0;
4096 }
4097
4098 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4099                                      int size, unsigned short port,
4100                                      const void *val, unsigned int count)
4101 {
4102         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4103
4104         trace_kvm_pio(1, port, size, count);
4105
4106         vcpu->arch.pio.port = port;
4107         vcpu->arch.pio.in = 0;
4108         vcpu->arch.pio.count = count;
4109         vcpu->arch.pio.size = size;
4110
4111         memcpy(vcpu->arch.pio_data, val, size * count);
4112
4113         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4114                 vcpu->arch.pio.count = 0;
4115                 return 1;
4116         }
4117
4118         vcpu->run->exit_reason = KVM_EXIT_IO;
4119         vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4120         vcpu->run->io.size = size;
4121         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4122         vcpu->run->io.count = count;
4123         vcpu->run->io.port = port;
4124
4125         return 0;
4126 }
4127
4128 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4129 {
4130         return kvm_x86_ops->get_segment_base(vcpu, seg);
4131 }
4132
4133 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4134 {
4135         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4136 }
4137
4138 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4139 {
4140         if (!need_emulate_wbinvd(vcpu))
4141                 return X86EMUL_CONTINUE;
4142
4143         if (kvm_x86_ops->has_wbinvd_exit()) {
4144                 int cpu = get_cpu();
4145
4146                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4147                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4148                                 wbinvd_ipi, NULL, 1);
4149                 put_cpu();
4150                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4151         } else
4152                 wbinvd();
4153         return X86EMUL_CONTINUE;
4154 }
4155 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4156
4157 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4158 {
4159         kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4160 }
4161
4162 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4163 {
4164         return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4165 }
4166
4167 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4168 {
4169
4170         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4171 }
4172
4173 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4174 {
4175         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4176 }
4177
4178 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4179 {
4180         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4181         unsigned long value;
4182
4183         switch (cr) {
4184         case 0:
4185                 value = kvm_read_cr0(vcpu);
4186                 break;
4187         case 2:
4188                 value = vcpu->arch.cr2;
4189                 break;
4190         case 3:
4191                 value = kvm_read_cr3(vcpu);
4192                 break;
4193         case 4:
4194                 value = kvm_read_cr4(vcpu);
4195                 break;
4196         case 8:
4197                 value = kvm_get_cr8(vcpu);
4198                 break;
4199         default:
4200                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4201                 return 0;
4202         }
4203
4204         return value;
4205 }
4206
4207 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4208 {
4209         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4210         int res = 0;
4211
4212         switch (cr) {
4213         case 0:
4214                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4215                 break;
4216         case 2:
4217                 vcpu->arch.cr2 = val;
4218                 break;
4219         case 3:
4220                 res = kvm_set_cr3(vcpu, val);
4221                 break;
4222         case 4:
4223                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4224                 break;
4225         case 8:
4226                 res = kvm_set_cr8(vcpu, val);
4227                 break;
4228         default:
4229                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4230                 res = -1;
4231         }
4232
4233         return res;
4234 }
4235
4236 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4237 {
4238         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4239 }
4240
4241 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4242 {
4243         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4244 }
4245
4246 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4247 {
4248         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4249 }
4250
4251 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4252 {
4253         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4254 }
4255
4256 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4257 {
4258         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4259 }
4260
4261 static unsigned long emulator_get_cached_segment_base(
4262         struct x86_emulate_ctxt *ctxt, int seg)
4263 {
4264         return get_segment_base(emul_to_vcpu(ctxt), seg);
4265 }
4266
4267 static bool emulator_get_cached_descriptor(struct x86_emulate_ctxt *ctxt,
4268                                            struct desc_struct *desc, u32 *base3,
4269                                            int seg)
4270 {
4271         struct kvm_segment var;
4272
4273         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4274
4275         if (var.unusable)
4276                 return false;
4277
4278         if (var.g)
4279                 var.limit >>= 12;
4280         set_desc_limit(desc, var.limit);
4281         set_desc_base(desc, (unsigned long)var.base);
4282 #ifdef CONFIG_X86_64
4283         if (base3)
4284                 *base3 = var.base >> 32;
4285 #endif
4286         desc->type = var.type;
4287         desc->s = var.s;
4288         desc->dpl = var.dpl;
4289         desc->p = var.present;
4290         desc->avl = var.avl;
4291         desc->l = var.l;
4292         desc->d = var.db;
4293         desc->g = var.g;
4294
4295         return true;
4296 }
4297
4298 static void emulator_set_cached_descriptor(struct x86_emulate_ctxt *ctxt,
4299                                            struct desc_struct *desc, u32 base3,
4300                                            int seg)
4301 {
4302         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4303         struct kvm_segment var;
4304
4305         /* needed to preserve selector */
4306         kvm_get_segment(vcpu, &var, seg);
4307
4308         var.base = get_desc_base(desc);
4309 #ifdef CONFIG_X86_64
4310         var.base |= ((u64)base3) << 32;
4311 #endif
4312         var.limit = get_desc_limit(desc);
4313         if (desc->g)
4314                 var.limit = (var.limit << 12) | 0xfff;
4315         var.type = desc->type;
4316         var.present = desc->p;
4317         var.dpl = desc->dpl;
4318         var.db = desc->d;
4319         var.s = desc->s;
4320         var.l = desc->l;
4321         var.g = desc->g;
4322         var.avl = desc->avl;
4323         var.present = desc->p;
4324         var.unusable = !var.present;
4325         var.padding = 0;
4326
4327         kvm_set_segment(vcpu, &var, seg);
4328         return;
4329 }
4330
4331 static u16 emulator_get_segment_selector(struct x86_emulate_ctxt *ctxt, int seg)
4332 {
4333         struct kvm_segment kvm_seg;
4334
4335         kvm_get_segment(emul_to_vcpu(ctxt), &kvm_seg, seg);
4336         return kvm_seg.selector;
4337 }
4338
4339 static void emulator_set_segment_selector(struct x86_emulate_ctxt *ctxt,
4340                                           u16 sel, int seg)
4341 {
4342         struct kvm_segment kvm_seg;
4343
4344         kvm_get_segment(emul_to_vcpu(ctxt), &kvm_seg, seg);
4345         kvm_seg.selector = sel;
4346         kvm_set_segment(emul_to_vcpu(ctxt), &kvm_seg, seg);
4347 }
4348
4349 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4350                             u32 msr_index, u64 *pdata)
4351 {
4352         return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4353 }
4354
4355 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4356                             u32 msr_index, u64 data)
4357 {
4358         return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4359 }
4360
4361 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4362 {
4363         emul_to_vcpu(ctxt)->arch.halt_request = 1;
4364 }
4365
4366 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4367 {
4368         preempt_disable();
4369         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4370         /*
4371          * CR0.TS may reference the host fpu state, not the guest fpu state,
4372          * so it may be clear at this point.
4373          */
4374         clts();
4375 }
4376
4377 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4378 {
4379         preempt_enable();
4380 }
4381
4382 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4383                               struct x86_instruction_info *info,
4384                               enum x86_intercept_stage stage)
4385 {
4386         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4387 }
4388
4389 static struct x86_emulate_ops emulate_ops = {
4390         .read_std            = kvm_read_guest_virt_system,
4391         .write_std           = kvm_write_guest_virt_system,
4392         .fetch               = kvm_fetch_guest_virt,
4393         .read_emulated       = emulator_read_emulated,
4394         .write_emulated      = emulator_write_emulated,
4395         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
4396         .invlpg              = emulator_invlpg,
4397         .pio_in_emulated     = emulator_pio_in_emulated,
4398         .pio_out_emulated    = emulator_pio_out_emulated,
4399         .get_cached_descriptor = emulator_get_cached_descriptor,
4400         .set_cached_descriptor = emulator_set_cached_descriptor,
4401         .get_segment_selector = emulator_get_segment_selector,
4402         .set_segment_selector = emulator_set_segment_selector,
4403         .get_cached_segment_base = emulator_get_cached_segment_base,
4404         .get_gdt             = emulator_get_gdt,
4405         .get_idt             = emulator_get_idt,
4406         .set_gdt             = emulator_set_gdt,
4407         .set_idt             = emulator_set_idt,
4408         .get_cr              = emulator_get_cr,
4409         .set_cr              = emulator_set_cr,
4410         .cpl                 = emulator_get_cpl,
4411         .get_dr              = emulator_get_dr,
4412         .set_dr              = emulator_set_dr,
4413         .set_msr             = emulator_set_msr,
4414         .get_msr             = emulator_get_msr,
4415         .halt                = emulator_halt,
4416         .wbinvd              = emulator_wbinvd,
4417         .fix_hypercall       = emulator_fix_hypercall,
4418         .get_fpu             = emulator_get_fpu,
4419         .put_fpu             = emulator_put_fpu,
4420         .intercept           = emulator_intercept,
4421 };
4422
4423 static void cache_all_regs(struct kvm_vcpu *vcpu)
4424 {
4425         kvm_register_read(vcpu, VCPU_REGS_RAX);
4426         kvm_register_read(vcpu, VCPU_REGS_RSP);
4427         kvm_register_read(vcpu, VCPU_REGS_RIP);
4428         vcpu->arch.regs_dirty = ~0;
4429 }
4430
4431 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4432 {
4433         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4434         /*
4435          * an sti; sti; sequence only disable interrupts for the first
4436          * instruction. So, if the last instruction, be it emulated or
4437          * not, left the system with the INT_STI flag enabled, it
4438          * means that the last instruction is an sti. We should not
4439          * leave the flag on in this case. The same goes for mov ss
4440          */
4441         if (!(int_shadow & mask))
4442                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4443 }
4444
4445 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4446 {
4447         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4448         if (ctxt->exception.vector == PF_VECTOR)
4449                 kvm_propagate_fault(vcpu, &ctxt->exception);
4450         else if (ctxt->exception.error_code_valid)
4451                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4452                                       ctxt->exception.error_code);
4453         else
4454                 kvm_queue_exception(vcpu, ctxt->exception.vector);
4455 }
4456
4457 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4458 {
4459         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4460         int cs_db, cs_l;
4461
4462         cache_all_regs(vcpu);
4463
4464         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4465
4466         vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
4467         vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4468         vcpu->arch.emulate_ctxt.mode =
4469                 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4470                 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4471                 ? X86EMUL_MODE_VM86 : cs_l
4472                 ? X86EMUL_MODE_PROT64 : cs_db
4473                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4474         vcpu->arch.emulate_ctxt.guest_mode = is_guest_mode(vcpu);
4475         memset(c, 0, sizeof(struct decode_cache));
4476         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4477         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4478 }
4479
4480 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4481 {
4482         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4483         int ret;
4484
4485         init_emulate_ctxt(vcpu);
4486
4487         vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4488         vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4489         vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip +
4490                                                                  inc_eip;
4491         ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4492
4493         if (ret != X86EMUL_CONTINUE)
4494                 return EMULATE_FAIL;
4495
4496         vcpu->arch.emulate_ctxt.eip = c->eip;
4497         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4498         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4499         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4500
4501         if (irq == NMI_VECTOR)
4502                 vcpu->arch.nmi_pending = false;
4503         else
4504                 vcpu->arch.interrupt.pending = false;
4505
4506         return EMULATE_DONE;
4507 }
4508 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4509
4510 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4511 {
4512         int r = EMULATE_DONE;
4513
4514         ++vcpu->stat.insn_emulation_fail;
4515         trace_kvm_emulate_insn_failed(vcpu);
4516         if (!is_guest_mode(vcpu)) {
4517                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4518                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4519                 vcpu->run->internal.ndata = 0;
4520                 r = EMULATE_FAIL;
4521         }
4522         kvm_queue_exception(vcpu, UD_VECTOR);
4523
4524         return r;
4525 }
4526
4527 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4528 {
4529         gpa_t gpa;
4530
4531         if (tdp_enabled)
4532                 return false;
4533
4534         /*
4535          * if emulation was due to access to shadowed page table
4536          * and it failed try to unshadow page and re-entetr the
4537          * guest to let CPU execute the instruction.
4538          */
4539         if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4540                 return true;
4541
4542         gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4543
4544         if (gpa == UNMAPPED_GVA)
4545                 return true; /* let cpu generate fault */
4546
4547         if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4548                 return true;
4549
4550         return false;
4551 }
4552
4553 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4554                             unsigned long cr2,
4555                             int emulation_type,
4556                             void *insn,
4557                             int insn_len)
4558 {
4559         int r;
4560         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4561         bool writeback = true;
4562
4563         kvm_clear_exception_queue(vcpu);
4564
4565         /*
4566          * TODO: fix emulate.c to use guest_read/write_register
4567          * instead of direct ->regs accesses, can save hundred cycles
4568          * on Intel for instructions that don't read/change RSP, for
4569          * for example.
4570          */
4571         cache_all_regs(vcpu);
4572
4573         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4574                 init_emulate_ctxt(vcpu);
4575                 vcpu->arch.emulate_ctxt.interruptibility = 0;
4576                 vcpu->arch.emulate_ctxt.have_exception = false;
4577                 vcpu->arch.emulate_ctxt.perm_ok = false;
4578
4579                 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4580                         = emulation_type & EMULTYPE_TRAP_UD;
4581
4582                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4583
4584                 trace_kvm_emulate_insn_start(vcpu);
4585                 ++vcpu->stat.insn_emulation;
4586                 if (r)  {
4587                         if (emulation_type & EMULTYPE_TRAP_UD)
4588                                 return EMULATE_FAIL;
4589                         if (reexecute_instruction(vcpu, cr2))
4590                                 return EMULATE_DONE;
4591                         if (emulation_type & EMULTYPE_SKIP)
4592                                 return EMULATE_FAIL;
4593                         return handle_emulation_failure(vcpu);
4594                 }
4595         }
4596
4597         if (emulation_type & EMULTYPE_SKIP) {
4598                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4599                 return EMULATE_DONE;
4600         }
4601
4602         /* this is needed for vmware backdoor interface to work since it
4603            changes registers values  during IO operation */
4604         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4605                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4606                 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4607         }
4608
4609 restart:
4610         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4611
4612         if (r == EMULATION_INTERCEPTED)
4613                 return EMULATE_DONE;
4614
4615         if (r == EMULATION_FAILED) {
4616                 if (reexecute_instruction(vcpu, cr2))
4617                         return EMULATE_DONE;
4618
4619                 return handle_emulation_failure(vcpu);
4620         }
4621
4622         if (vcpu->arch.emulate_ctxt.have_exception) {
4623                 inject_emulated_exception(vcpu);
4624                 r = EMULATE_DONE;
4625         } else if (vcpu->arch.pio.count) {
4626                 if (!vcpu->arch.pio.in)
4627                         vcpu->arch.pio.count = 0;
4628                 else
4629                         writeback = false;
4630                 r = EMULATE_DO_MMIO;
4631         } else if (vcpu->mmio_needed) {
4632                 if (!vcpu->mmio_is_write)
4633                         writeback = false;
4634                 r = EMULATE_DO_MMIO;
4635         } else if (r == EMULATION_RESTART)
4636                 goto restart;
4637         else
4638                 r = EMULATE_DONE;
4639
4640         if (writeback) {
4641                 toggle_interruptibility(vcpu,
4642                                 vcpu->arch.emulate_ctxt.interruptibility);
4643                 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4644                 kvm_make_request(KVM_REQ_EVENT, vcpu);
4645                 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4646                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4647                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4648         } else
4649                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4650
4651         return r;
4652 }
4653 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4654
4655 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4656 {
4657         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4658         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4659                                             size, port, &val, 1);
4660         /* do not return to emulator after return from userspace */
4661         vcpu->arch.pio.count = 0;
4662         return ret;
4663 }
4664 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4665
4666 static void tsc_bad(void *info)
4667 {
4668         __this_cpu_write(cpu_tsc_khz, 0);
4669 }
4670
4671 static void tsc_khz_changed(void *data)
4672 {
4673         struct cpufreq_freqs *freq = data;
4674         unsigned long khz = 0;
4675
4676         if (data)
4677                 khz = freq->new;
4678         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4679                 khz = cpufreq_quick_get(raw_smp_processor_id());
4680         if (!khz)
4681                 khz = tsc_khz;
4682         __this_cpu_write(cpu_tsc_khz, khz);
4683 }
4684
4685 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4686                                      void *data)
4687 {
4688         struct cpufreq_freqs *freq = data;
4689         struct kvm *kvm;
4690         struct kvm_vcpu *vcpu;
4691         int i, send_ipi = 0;
4692
4693         /*
4694          * We allow guests to temporarily run on slowing clocks,
4695          * provided we notify them after, or to run on accelerating
4696          * clocks, provided we notify them before.  Thus time never
4697          * goes backwards.
4698          *
4699          * However, we have a problem.  We can't atomically update
4700          * the frequency of a given CPU from this function; it is
4701          * merely a notifier, which can be called from any CPU.
4702          * Changing the TSC frequency at arbitrary points in time
4703          * requires a recomputation of local variables related to
4704          * the TSC for each VCPU.  We must flag these local variables
4705          * to be updated and be sure the update takes place with the
4706          * new frequency before any guests proceed.
4707          *
4708          * Unfortunately, the combination of hotplug CPU and frequency
4709          * change creates an intractable locking scenario; the order
4710          * of when these callouts happen is undefined with respect to
4711          * CPU hotplug, and they can race with each other.  As such,
4712          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4713          * undefined; you can actually have a CPU frequency change take
4714          * place in between the computation of X and the setting of the
4715          * variable.  To protect against this problem, all updates of
4716          * the per_cpu tsc_khz variable are done in an interrupt
4717          * protected IPI, and all callers wishing to update the value
4718          * must wait for a synchronous IPI to complete (which is trivial
4719          * if the caller is on the CPU already).  This establishes the
4720          * necessary total order on variable updates.
4721          *
4722          * Note that because a guest time update may take place
4723          * anytime after the setting of the VCPU's request bit, the
4724          * correct TSC value must be set before the request.  However,
4725          * to ensure the update actually makes it to any guest which
4726          * starts running in hardware virtualization between the set
4727          * and the acquisition of the spinlock, we must also ping the
4728          * CPU after setting the request bit.
4729          *
4730          */
4731
4732         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4733                 return 0;
4734         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4735                 return 0;
4736
4737         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4738
4739         raw_spin_lock(&kvm_lock);
4740         list_for_each_entry(kvm, &vm_list, vm_list) {
4741                 kvm_for_each_vcpu(i, vcpu, kvm) {
4742                         if (vcpu->cpu != freq->cpu)
4743                                 continue;
4744                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4745                         if (vcpu->cpu != smp_processor_id())
4746                                 send_ipi = 1;
4747                 }
4748         }
4749         raw_spin_unlock(&kvm_lock);
4750
4751         if (freq->old < freq->new && send_ipi) {
4752                 /*
4753                  * We upscale the frequency.  Must make the guest
4754                  * doesn't see old kvmclock values while running with
4755                  * the new frequency, otherwise we risk the guest sees
4756                  * time go backwards.
4757                  *
4758                  * In case we update the frequency for another cpu
4759                  * (which might be in guest context) send an interrupt
4760                  * to kick the cpu out of guest context.  Next time
4761                  * guest context is entered kvmclock will be updated,
4762                  * so the guest will not see stale values.
4763                  */
4764                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4765         }
4766         return 0;
4767 }
4768
4769 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4770         .notifier_call  = kvmclock_cpufreq_notifier
4771 };
4772
4773 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4774                                         unsigned long action, void *hcpu)
4775 {
4776         unsigned int cpu = (unsigned long)hcpu;
4777
4778         switch (action) {
4779                 case CPU_ONLINE:
4780                 case CPU_DOWN_FAILED:
4781                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4782                         break;
4783                 case CPU_DOWN_PREPARE:
4784                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
4785                         break;
4786         }
4787         return NOTIFY_OK;
4788 }
4789
4790 static struct notifier_block kvmclock_cpu_notifier_block = {
4791         .notifier_call  = kvmclock_cpu_notifier,
4792         .priority = -INT_MAX
4793 };
4794
4795 static void kvm_timer_init(void)
4796 {
4797         int cpu;
4798
4799         max_tsc_khz = tsc_khz;
4800         register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4801         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4802 #ifdef CONFIG_CPU_FREQ
4803                 struct cpufreq_policy policy;
4804                 memset(&policy, 0, sizeof(policy));
4805                 cpu = get_cpu();
4806                 cpufreq_get_policy(&policy, cpu);
4807                 if (policy.cpuinfo.max_freq)
4808                         max_tsc_khz = policy.cpuinfo.max_freq;
4809                 put_cpu();
4810 #endif
4811                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4812                                           CPUFREQ_TRANSITION_NOTIFIER);
4813         }
4814         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4815         for_each_online_cpu(cpu)
4816                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4817 }
4818
4819 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4820
4821 static int kvm_is_in_guest(void)
4822 {
4823         return percpu_read(current_vcpu) != NULL;
4824 }
4825
4826 static int kvm_is_user_mode(void)
4827 {
4828         int user_mode = 3;
4829
4830         if (percpu_read(current_vcpu))
4831                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4832
4833         return user_mode != 0;
4834 }
4835
4836 static unsigned long kvm_get_guest_ip(void)
4837 {
4838         unsigned long ip = 0;
4839
4840         if (percpu_read(current_vcpu))
4841                 ip = kvm_rip_read(percpu_read(current_vcpu));
4842
4843         return ip;
4844 }
4845
4846 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4847         .is_in_guest            = kvm_is_in_guest,
4848         .is_user_mode           = kvm_is_user_mode,
4849         .get_guest_ip           = kvm_get_guest_ip,
4850 };
4851
4852 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4853 {
4854         percpu_write(current_vcpu, vcpu);
4855 }
4856 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4857
4858 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4859 {
4860         percpu_write(current_vcpu, NULL);
4861 }
4862 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4863
4864 int kvm_arch_init(void *opaque)
4865 {
4866         int r;
4867         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4868
4869         if (kvm_x86_ops) {
4870                 printk(KERN_ERR "kvm: already loaded the other module\n");
4871                 r = -EEXIST;
4872                 goto out;
4873         }
4874
4875         if (!ops->cpu_has_kvm_support()) {
4876                 printk(KERN_ERR "kvm: no hardware support\n");
4877                 r = -EOPNOTSUPP;
4878                 goto out;
4879         }
4880         if (ops->disabled_by_bios()) {
4881                 printk(KERN_ERR "kvm: disabled by bios\n");
4882                 r = -EOPNOTSUPP;
4883                 goto out;
4884         }
4885
4886         r = kvm_mmu_module_init();
4887         if (r)
4888                 goto out;
4889
4890         kvm_init_msr_list();
4891
4892         kvm_x86_ops = ops;
4893         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4894         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4895                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
4896
4897         kvm_timer_init();
4898
4899         perf_register_guest_info_callbacks(&kvm_guest_cbs);
4900
4901         if (cpu_has_xsave)
4902                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4903
4904         return 0;
4905
4906 out:
4907         return r;
4908 }
4909
4910 void kvm_arch_exit(void)
4911 {
4912         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4913
4914         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4915                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4916                                             CPUFREQ_TRANSITION_NOTIFIER);
4917         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4918         kvm_x86_ops = NULL;
4919         kvm_mmu_module_exit();
4920 }
4921
4922 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4923 {
4924         ++vcpu->stat.halt_exits;
4925         if (irqchip_in_kernel(vcpu->kvm)) {
4926                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4927                 return 1;
4928         } else {
4929                 vcpu->run->exit_reason = KVM_EXIT_HLT;
4930                 return 0;
4931         }
4932 }
4933 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4934
4935 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4936                            unsigned long a1)
4937 {
4938         if (is_long_mode(vcpu))
4939                 return a0;
4940         else
4941                 return a0 | ((gpa_t)a1 << 32);
4942 }
4943
4944 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4945 {
4946         u64 param, ingpa, outgpa, ret;
4947         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4948         bool fast, longmode;
4949         int cs_db, cs_l;
4950
4951         /*
4952          * hypercall generates UD from non zero cpl and real mode
4953          * per HYPER-V spec
4954          */
4955         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4956                 kvm_queue_exception(vcpu, UD_VECTOR);
4957                 return 0;
4958         }
4959
4960         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4961         longmode = is_long_mode(vcpu) && cs_l == 1;
4962
4963         if (!longmode) {
4964                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4965                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4966                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4967                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4968                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4969                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4970         }
4971 #ifdef CONFIG_X86_64
4972         else {
4973                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4974                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4975                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4976         }
4977 #endif
4978
4979         code = param & 0xffff;
4980         fast = (param >> 16) & 0x1;
4981         rep_cnt = (param >> 32) & 0xfff;
4982         rep_idx = (param >> 48) & 0xfff;
4983
4984         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4985
4986         switch (code) {
4987         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4988                 kvm_vcpu_on_spin(vcpu);
4989                 break;
4990         default:
4991                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4992                 break;
4993         }
4994
4995         ret = res | (((u64)rep_done & 0xfff) << 32);
4996         if (longmode) {
4997                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4998         } else {
4999                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5000                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5001         }
5002
5003         return 1;
5004 }
5005
5006 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5007 {
5008         unsigned long nr, a0, a1, a2, a3, ret;
5009         int r = 1;
5010
5011         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5012                 return kvm_hv_hypercall(vcpu);
5013
5014         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5015         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5016         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5017         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5018         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5019
5020         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5021
5022         if (!is_long_mode(vcpu)) {
5023                 nr &= 0xFFFFFFFF;
5024                 a0 &= 0xFFFFFFFF;
5025                 a1 &= 0xFFFFFFFF;
5026                 a2 &= 0xFFFFFFFF;
5027                 a3 &= 0xFFFFFFFF;
5028         }
5029
5030         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5031                 ret = -KVM_EPERM;
5032                 goto out;
5033         }
5034
5035         switch (nr) {
5036         case KVM_HC_VAPIC_POLL_IRQ:
5037                 ret = 0;
5038                 break;
5039         case KVM_HC_MMU_OP:
5040                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5041                 break;
5042         default:
5043                 ret = -KVM_ENOSYS;
5044                 break;
5045         }
5046 out:
5047         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5048         ++vcpu->stat.hypercalls;
5049         return r;
5050 }
5051 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5052
5053 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5054 {
5055         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5056         char instruction[3];
5057         unsigned long rip = kvm_rip_read(vcpu);
5058
5059         /*
5060          * Blow out the MMU to ensure that no other VCPU has an active mapping
5061          * to ensure that the updated hypercall appears atomically across all
5062          * VCPUs.
5063          */
5064         kvm_mmu_zap_all(vcpu->kvm);
5065
5066         kvm_x86_ops->patch_hypercall(vcpu, instruction);
5067
5068         return emulator_write_emulated(&vcpu->arch.emulate_ctxt,
5069                                        rip, instruction, 3, NULL);
5070 }
5071
5072 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5073 {
5074         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5075         int j, nent = vcpu->arch.cpuid_nent;
5076
5077         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5078         /* when no next entry is found, the current entry[i] is reselected */
5079         for (j = i + 1; ; j = (j + 1) % nent) {
5080                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5081                 if (ej->function == e->function) {
5082                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5083                         return j;
5084                 }
5085         }
5086         return 0; /* silence gcc, even though control never reaches here */
5087 }
5088
5089 /* find an entry with matching function, matching index (if needed), and that
5090  * should be read next (if it's stateful) */
5091 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5092         u32 function, u32 index)
5093 {
5094         if (e->function != function)
5095                 return 0;
5096         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5097                 return 0;
5098         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5099             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5100                 return 0;
5101         return 1;
5102 }
5103
5104 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5105                                               u32 function, u32 index)
5106 {
5107         int i;
5108         struct kvm_cpuid_entry2 *best = NULL;
5109
5110         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5111                 struct kvm_cpuid_entry2 *e;
5112
5113                 e = &vcpu->arch.cpuid_entries[i];
5114                 if (is_matching_cpuid_entry(e, function, index)) {
5115                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5116                                 move_to_next_stateful_cpuid_entry(vcpu, i);
5117                         best = e;
5118                         break;
5119                 }
5120         }
5121         return best;
5122 }
5123 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5124
5125 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5126 {
5127         struct kvm_cpuid_entry2 *best;
5128
5129         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5130         if (!best || best->eax < 0x80000008)
5131                 goto not_found;
5132         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5133         if (best)
5134                 return best->eax & 0xff;
5135 not_found:
5136         return 36;
5137 }
5138
5139 /*
5140  * If no match is found, check whether we exceed the vCPU's limit
5141  * and return the content of the highest valid _standard_ leaf instead.
5142  * This is to satisfy the CPUID specification.
5143  */
5144 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5145                                                   u32 function, u32 index)
5146 {
5147         struct kvm_cpuid_entry2 *maxlevel;
5148
5149         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5150         if (!maxlevel || maxlevel->eax >= function)
5151                 return NULL;
5152         if (function & 0x80000000) {
5153                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5154                 if (!maxlevel)
5155                         return NULL;
5156         }
5157         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5158 }
5159
5160 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5161 {
5162         u32 function, index;
5163         struct kvm_cpuid_entry2 *best;
5164
5165         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5166         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5167         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5168         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5169         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5170         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5171         best = kvm_find_cpuid_entry(vcpu, function, index);
5172
5173         if (!best)
5174                 best = check_cpuid_limit(vcpu, function, index);
5175
5176         if (best) {
5177                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5178                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5179                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5180                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5181         }
5182         kvm_x86_ops->skip_emulated_instruction(vcpu);
5183         trace_kvm_cpuid(function,
5184                         kvm_register_read(vcpu, VCPU_REGS_RAX),
5185                         kvm_register_read(vcpu, VCPU_REGS_RBX),
5186                         kvm_register_read(vcpu, VCPU_REGS_RCX),
5187                         kvm_register_read(vcpu, VCPU_REGS_RDX));
5188 }
5189 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5190
5191 /*
5192  * Check if userspace requested an interrupt window, and that the
5193  * interrupt window is open.
5194  *
5195  * No need to exit to userspace if we already have an interrupt queued.
5196  */
5197 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5198 {
5199         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5200                 vcpu->run->request_interrupt_window &&
5201                 kvm_arch_interrupt_allowed(vcpu));
5202 }
5203
5204 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5205 {
5206         struct kvm_run *kvm_run = vcpu->run;
5207
5208         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5209         kvm_run->cr8 = kvm_get_cr8(vcpu);
5210         kvm_run->apic_base = kvm_get_apic_base(vcpu);
5211         if (irqchip_in_kernel(vcpu->kvm))
5212                 kvm_run->ready_for_interrupt_injection = 1;
5213         else
5214                 kvm_run->ready_for_interrupt_injection =
5215                         kvm_arch_interrupt_allowed(vcpu) &&
5216                         !kvm_cpu_has_interrupt(vcpu) &&
5217                         !kvm_event_needs_reinjection(vcpu);
5218 }
5219
5220 static void vapic_enter(struct kvm_vcpu *vcpu)
5221 {
5222         struct kvm_lapic *apic = vcpu->arch.apic;
5223         struct page *page;
5224
5225         if (!apic || !apic->vapic_addr)
5226                 return;
5227
5228         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5229
5230         vcpu->arch.apic->vapic_page = page;
5231 }
5232
5233 static void vapic_exit(struct kvm_vcpu *vcpu)
5234 {
5235         struct kvm_lapic *apic = vcpu->arch.apic;
5236         int idx;
5237
5238         if (!apic || !apic->vapic_addr)
5239                 return;
5240
5241         idx = srcu_read_lock(&vcpu->kvm->srcu);
5242         kvm_release_page_dirty(apic->vapic_page);
5243         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5244         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5245 }
5246
5247 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5248 {
5249         int max_irr, tpr;
5250
5251         if (!kvm_x86_ops->update_cr8_intercept)
5252                 return;
5253
5254         if (!vcpu->arch.apic)
5255                 return;
5256
5257         if (!vcpu->arch.apic->vapic_addr)
5258                 max_irr = kvm_lapic_find_highest_irr(vcpu);
5259         else
5260                 max_irr = -1;
5261
5262         if (max_irr != -1)
5263                 max_irr >>= 4;
5264
5265         tpr = kvm_lapic_get_cr8(vcpu);
5266
5267         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5268 }
5269
5270 static void inject_pending_event(struct kvm_vcpu *vcpu)
5271 {
5272         /* try to reinject previous events if any */
5273         if (vcpu->arch.exception.pending) {
5274                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5275                                         vcpu->arch.exception.has_error_code,
5276                                         vcpu->arch.exception.error_code);
5277                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5278                                           vcpu->arch.exception.has_error_code,
5279                                           vcpu->arch.exception.error_code,
5280                                           vcpu->arch.exception.reinject);
5281                 return;
5282         }
5283
5284         if (vcpu->arch.nmi_injected) {
5285                 kvm_x86_ops->set_nmi(vcpu);
5286                 return;
5287         }
5288
5289         if (vcpu->arch.interrupt.pending) {
5290                 kvm_x86_ops->set_irq(vcpu);
5291                 return;
5292         }
5293
5294         /* try to inject new event if pending */
5295         if (vcpu->arch.nmi_pending) {
5296                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5297                         vcpu->arch.nmi_pending = false;
5298                         vcpu->arch.nmi_injected = true;
5299                         kvm_x86_ops->set_nmi(vcpu);
5300                 }
5301         } else if (kvm_cpu_has_interrupt(vcpu)) {
5302                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5303                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5304                                             false);
5305                         kvm_x86_ops->set_irq(vcpu);
5306                 }
5307         }
5308 }
5309
5310 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5311 {
5312         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5313                         !vcpu->guest_xcr0_loaded) {
5314                 /* kvm_set_xcr() also depends on this */
5315                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5316                 vcpu->guest_xcr0_loaded = 1;
5317         }
5318 }
5319
5320 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5321 {
5322         if (vcpu->guest_xcr0_loaded) {
5323                 if (vcpu->arch.xcr0 != host_xcr0)
5324                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5325                 vcpu->guest_xcr0_loaded = 0;
5326         }
5327 }
5328
5329 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5330 {
5331         int r;
5332         bool nmi_pending;
5333         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5334                 vcpu->run->request_interrupt_window;
5335
5336         if (vcpu->requests) {
5337                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5338                         kvm_mmu_unload(vcpu);
5339                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5340                         __kvm_migrate_timers(vcpu);
5341                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5342                         r = kvm_guest_time_update(vcpu);
5343                         if (unlikely(r))
5344                                 goto out;
5345                 }
5346                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5347                         kvm_mmu_sync_roots(vcpu);
5348                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5349                         kvm_x86_ops->tlb_flush(vcpu);
5350                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5351                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5352                         r = 0;
5353                         goto out;
5354                 }
5355                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5356                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5357                         r = 0;
5358                         goto out;
5359                 }
5360                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5361                         vcpu->fpu_active = 0;
5362                         kvm_x86_ops->fpu_deactivate(vcpu);
5363                 }
5364                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5365                         /* Page is swapped out. Do synthetic halt */
5366                         vcpu->arch.apf.halted = true;
5367                         r = 1;
5368                         goto out;
5369                 }
5370         }
5371
5372         r = kvm_mmu_reload(vcpu);
5373         if (unlikely(r))
5374                 goto out;
5375
5376         /*
5377          * An NMI can be injected between local nmi_pending read and
5378          * vcpu->arch.nmi_pending read inside inject_pending_event().
5379          * But in that case, KVM_REQ_EVENT will be set, which makes
5380          * the race described above benign.
5381          */
5382         nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5383
5384         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5385                 inject_pending_event(vcpu);
5386
5387                 /* enable NMI/IRQ window open exits if needed */
5388                 if (nmi_pending)
5389                         kvm_x86_ops->enable_nmi_window(vcpu);
5390                 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5391                         kvm_x86_ops->enable_irq_window(vcpu);
5392
5393                 if (kvm_lapic_enabled(vcpu)) {
5394                         update_cr8_intercept(vcpu);
5395                         kvm_lapic_sync_to_vapic(vcpu);
5396                 }
5397         }
5398
5399         preempt_disable();
5400
5401         kvm_x86_ops->prepare_guest_switch(vcpu);
5402         if (vcpu->fpu_active)
5403                 kvm_load_guest_fpu(vcpu);
5404         kvm_load_guest_xcr0(vcpu);
5405
5406         vcpu->mode = IN_GUEST_MODE;
5407
5408         /* We should set ->mode before check ->requests,
5409          * see the comment in make_all_cpus_request.
5410          */
5411         smp_mb();
5412
5413         local_irq_disable();
5414
5415         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5416             || need_resched() || signal_pending(current)) {
5417                 vcpu->mode = OUTSIDE_GUEST_MODE;
5418                 smp_wmb();
5419                 local_irq_enable();
5420                 preempt_enable();
5421                 kvm_x86_ops->cancel_injection(vcpu);
5422                 r = 1;
5423                 goto out;
5424         }
5425
5426         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5427
5428         kvm_guest_enter();
5429
5430         if (unlikely(vcpu->arch.switch_db_regs)) {
5431                 set_debugreg(0, 7);
5432                 set_debugreg(vcpu->arch.eff_db[0], 0);
5433                 set_debugreg(vcpu->arch.eff_db[1], 1);
5434                 set_debugreg(vcpu->arch.eff_db[2], 2);
5435                 set_debugreg(vcpu->arch.eff_db[3], 3);
5436         }
5437
5438         trace_kvm_entry(vcpu->vcpu_id);
5439         kvm_x86_ops->run(vcpu);
5440
5441         /*
5442          * If the guest has used debug registers, at least dr7
5443          * will be disabled while returning to the host.
5444          * If we don't have active breakpoints in the host, we don't
5445          * care about the messed up debug address registers. But if
5446          * we have some of them active, restore the old state.
5447          */
5448         if (hw_breakpoint_active())
5449                 hw_breakpoint_restore();
5450
5451         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5452
5453         vcpu->mode = OUTSIDE_GUEST_MODE;
5454         smp_wmb();
5455         local_irq_enable();
5456
5457         ++vcpu->stat.exits;
5458
5459         /*
5460          * We must have an instruction between local_irq_enable() and
5461          * kvm_guest_exit(), so the timer interrupt isn't delayed by
5462          * the interrupt shadow.  The stat.exits increment will do nicely.
5463          * But we need to prevent reordering, hence this barrier():
5464          */
5465         barrier();
5466
5467         kvm_guest_exit();
5468
5469         preempt_enable();
5470
5471         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5472
5473         /*
5474          * Profile KVM exit RIPs:
5475          */
5476         if (unlikely(prof_on == KVM_PROFILING)) {
5477                 unsigned long rip = kvm_rip_read(vcpu);
5478                 profile_hit(KVM_PROFILING, (void *)rip);
5479         }
5480
5481
5482         kvm_lapic_sync_from_vapic(vcpu);
5483
5484         r = kvm_x86_ops->handle_exit(vcpu);
5485 out:
5486         return r;
5487 }
5488
5489
5490 static int __vcpu_run(struct kvm_vcpu *vcpu)
5491 {
5492         int r;
5493         struct kvm *kvm = vcpu->kvm;
5494
5495         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5496                 pr_debug("vcpu %d received sipi with vector # %x\n",
5497                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
5498                 kvm_lapic_reset(vcpu);
5499                 r = kvm_arch_vcpu_reset(vcpu);
5500                 if (r)
5501                         return r;
5502                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5503         }
5504
5505         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5506         vapic_enter(vcpu);
5507
5508         r = 1;
5509         while (r > 0) {
5510                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5511                     !vcpu->arch.apf.halted)
5512                         r = vcpu_enter_guest(vcpu);
5513                 else {
5514                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5515                         kvm_vcpu_block(vcpu);
5516                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5517                         if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5518                         {
5519                                 switch(vcpu->arch.mp_state) {
5520                                 case KVM_MP_STATE_HALTED:
5521                                         vcpu->arch.mp_state =
5522                                                 KVM_MP_STATE_RUNNABLE;
5523                                 case KVM_MP_STATE_RUNNABLE:
5524                                         vcpu->arch.apf.halted = false;
5525                                         break;
5526                                 case KVM_MP_STATE_SIPI_RECEIVED:
5527                                 default:
5528                                         r = -EINTR;
5529                                         break;
5530                                 }
5531                         }
5532                 }
5533
5534                 if (r <= 0)
5535                         break;
5536
5537                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5538                 if (kvm_cpu_has_pending_timer(vcpu))
5539                         kvm_inject_pending_timer_irqs(vcpu);
5540
5541                 if (dm_request_for_irq_injection(vcpu)) {
5542                         r = -EINTR;
5543                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5544                         ++vcpu->stat.request_irq_exits;
5545                 }
5546
5547                 kvm_check_async_pf_completion(vcpu);
5548
5549                 if (signal_pending(current)) {
5550                         r = -EINTR;
5551                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5552                         ++vcpu->stat.signal_exits;
5553                 }
5554                 if (need_resched()) {
5555                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5556                         kvm_resched(vcpu);
5557                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5558                 }
5559         }
5560
5561         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5562
5563         vapic_exit(vcpu);
5564
5565         return r;
5566 }
5567
5568 static int complete_mmio(struct kvm_vcpu *vcpu)
5569 {
5570         struct kvm_run *run = vcpu->run;
5571         int r;
5572
5573         if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5574                 return 1;
5575
5576         if (vcpu->mmio_needed) {
5577                 vcpu->mmio_needed = 0;
5578                 if (!vcpu->mmio_is_write)
5579                         memcpy(vcpu->mmio_data + vcpu->mmio_index,
5580                                run->mmio.data, 8);
5581                 vcpu->mmio_index += 8;
5582                 if (vcpu->mmio_index < vcpu->mmio_size) {
5583                         run->exit_reason = KVM_EXIT_MMIO;
5584                         run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5585                         memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5586                         run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5587                         run->mmio.is_write = vcpu->mmio_is_write;
5588                         vcpu->mmio_needed = 1;
5589                         return 0;
5590                 }
5591                 if (vcpu->mmio_is_write)
5592                         return 1;
5593                 vcpu->mmio_read_completed = 1;
5594         }
5595         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5596         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5597         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5598         if (r != EMULATE_DONE)
5599                 return 0;
5600         return 1;
5601 }
5602
5603 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5604 {
5605         int r;
5606         sigset_t sigsaved;
5607
5608         if (!tsk_used_math(current) && init_fpu(current))
5609                 return -ENOMEM;
5610
5611         if (vcpu->sigset_active)
5612                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5613
5614         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5615                 kvm_vcpu_block(vcpu);
5616                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5617                 r = -EAGAIN;
5618                 goto out;
5619         }
5620
5621         /* re-sync apic's tpr */
5622         if (!irqchip_in_kernel(vcpu->kvm)) {
5623                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5624                         r = -EINVAL;
5625                         goto out;
5626                 }
5627         }
5628
5629         r = complete_mmio(vcpu);
5630         if (r <= 0)
5631                 goto out;
5632
5633         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5634                 kvm_register_write(vcpu, VCPU_REGS_RAX,
5635                                      kvm_run->hypercall.ret);
5636
5637         r = __vcpu_run(vcpu);
5638
5639 out:
5640         post_kvm_run_save(vcpu);
5641         if (vcpu->sigset_active)
5642                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5643
5644         return r;
5645 }
5646
5647 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5648 {
5649         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5650                 /*
5651                  * We are here if userspace calls get_regs() in the middle of
5652                  * instruction emulation. Registers state needs to be copied
5653                  * back from emulation context to vcpu. Usrapace shouldn't do
5654                  * that usually, but some bad designed PV devices (vmware
5655                  * backdoor interface) need this to work
5656                  */
5657                 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5658                 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5659                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5660         }
5661         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5662         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5663         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5664         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5665         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5666         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5667         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5668         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5669 #ifdef CONFIG_X86_64
5670         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5671         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5672         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5673         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5674         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5675         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5676         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5677         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5678 #endif
5679
5680         regs->rip = kvm_rip_read(vcpu);
5681         regs->rflags = kvm_get_rflags(vcpu);
5682
5683         return 0;
5684 }
5685
5686 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5687 {
5688         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5689         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5690
5691         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5692         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5693         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5694         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5695         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5696         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5697         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5698         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5699 #ifdef CONFIG_X86_64
5700         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5701         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5702         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5703         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5704         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5705         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5706         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5707         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5708 #endif
5709
5710         kvm_rip_write(vcpu, regs->rip);
5711         kvm_set_rflags(vcpu, regs->rflags);
5712
5713         vcpu->arch.exception.pending = false;
5714
5715         kvm_make_request(KVM_REQ_EVENT, vcpu);
5716
5717         return 0;
5718 }
5719
5720 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5721 {
5722         struct kvm_segment cs;
5723
5724         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5725         *db = cs.db;
5726         *l = cs.l;
5727 }
5728 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5729
5730 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5731                                   struct kvm_sregs *sregs)
5732 {
5733         struct desc_ptr dt;
5734
5735         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5736         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5737         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5738         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5739         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5740         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5741
5742         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5743         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5744
5745         kvm_x86_ops->get_idt(vcpu, &dt);
5746         sregs->idt.limit = dt.size;
5747         sregs->idt.base = dt.address;
5748         kvm_x86_ops->get_gdt(vcpu, &dt);
5749         sregs->gdt.limit = dt.size;
5750         sregs->gdt.base = dt.address;
5751
5752         sregs->cr0 = kvm_read_cr0(vcpu);
5753         sregs->cr2 = vcpu->arch.cr2;
5754         sregs->cr3 = kvm_read_cr3(vcpu);
5755         sregs->cr4 = kvm_read_cr4(vcpu);
5756         sregs->cr8 = kvm_get_cr8(vcpu);
5757         sregs->efer = vcpu->arch.efer;
5758         sregs->apic_base = kvm_get_apic_base(vcpu);
5759
5760         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5761
5762         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5763                 set_bit(vcpu->arch.interrupt.nr,
5764                         (unsigned long *)sregs->interrupt_bitmap);
5765
5766         return 0;
5767 }
5768
5769 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5770                                     struct kvm_mp_state *mp_state)
5771 {
5772         mp_state->mp_state = vcpu->arch.mp_state;
5773         return 0;
5774 }
5775
5776 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5777                                     struct kvm_mp_state *mp_state)
5778 {
5779         vcpu->arch.mp_state = mp_state->mp_state;
5780         kvm_make_request(KVM_REQ_EVENT, vcpu);
5781         return 0;
5782 }
5783
5784 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5785                     bool has_error_code, u32 error_code)
5786 {
5787         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5788         int ret;
5789
5790         init_emulate_ctxt(vcpu);
5791
5792         ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5793                                    tss_selector, reason, has_error_code,
5794                                    error_code);
5795
5796         if (ret)
5797                 return EMULATE_FAIL;
5798
5799         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5800         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5801         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5802         kvm_make_request(KVM_REQ_EVENT, vcpu);
5803         return EMULATE_DONE;
5804 }
5805 EXPORT_SYMBOL_GPL(kvm_task_switch);
5806
5807 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5808                                   struct kvm_sregs *sregs)
5809 {
5810         int mmu_reset_needed = 0;
5811         int pending_vec, max_bits, idx;
5812         struct desc_ptr dt;
5813
5814         dt.size = sregs->idt.limit;
5815         dt.address = sregs->idt.base;
5816         kvm_x86_ops->set_idt(vcpu, &dt);
5817         dt.size = sregs->gdt.limit;
5818         dt.address = sregs->gdt.base;
5819         kvm_x86_ops->set_gdt(vcpu, &dt);
5820
5821         vcpu->arch.cr2 = sregs->cr2;
5822         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5823         vcpu->arch.cr3 = sregs->cr3;
5824         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5825
5826         kvm_set_cr8(vcpu, sregs->cr8);
5827
5828         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5829         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5830         kvm_set_apic_base(vcpu, sregs->apic_base);
5831
5832         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5833         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5834         vcpu->arch.cr0 = sregs->cr0;
5835
5836         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5837         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5838         if (sregs->cr4 & X86_CR4_OSXSAVE)
5839                 update_cpuid(vcpu);
5840
5841         idx = srcu_read_lock(&vcpu->kvm->srcu);
5842         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5843                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5844                 mmu_reset_needed = 1;
5845         }
5846         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5847
5848         if (mmu_reset_needed)
5849                 kvm_mmu_reset_context(vcpu);
5850
5851         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5852         pending_vec = find_first_bit(
5853                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5854         if (pending_vec < max_bits) {
5855                 kvm_queue_interrupt(vcpu, pending_vec, false);
5856                 pr_debug("Set back pending irq %d\n", pending_vec);
5857         }
5858
5859         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5860         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5861         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5862         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5863         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5864         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5865
5866         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5867         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5868
5869         update_cr8_intercept(vcpu);
5870
5871         /* Older userspace won't unhalt the vcpu on reset. */
5872         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5873             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5874             !is_protmode(vcpu))
5875                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5876
5877         kvm_make_request(KVM_REQ_EVENT, vcpu);
5878
5879         return 0;
5880 }
5881
5882 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5883                                         struct kvm_guest_debug *dbg)
5884 {
5885         unsigned long rflags;
5886         int i, r;
5887
5888         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5889                 r = -EBUSY;
5890                 if (vcpu->arch.exception.pending)
5891                         goto out;
5892                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5893                         kvm_queue_exception(vcpu, DB_VECTOR);
5894                 else
5895                         kvm_queue_exception(vcpu, BP_VECTOR);
5896         }
5897
5898         /*
5899          * Read rflags as long as potentially injected trace flags are still
5900          * filtered out.
5901          */
5902         rflags = kvm_get_rflags(vcpu);
5903
5904         vcpu->guest_debug = dbg->control;
5905         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5906                 vcpu->guest_debug = 0;
5907
5908         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5909                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5910                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5911                 vcpu->arch.switch_db_regs =
5912                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5913         } else {
5914                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5915                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5916                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5917         }
5918
5919         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5920                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5921                         get_segment_base(vcpu, VCPU_SREG_CS);
5922
5923         /*
5924          * Trigger an rflags update that will inject or remove the trace
5925          * flags.
5926          */
5927         kvm_set_rflags(vcpu, rflags);
5928
5929         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5930
5931         r = 0;
5932
5933 out:
5934
5935         return r;
5936 }
5937
5938 /*
5939  * Translate a guest virtual address to a guest physical address.
5940  */
5941 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5942                                     struct kvm_translation *tr)
5943 {
5944         unsigned long vaddr = tr->linear_address;
5945         gpa_t gpa;
5946         int idx;
5947
5948         idx = srcu_read_lock(&vcpu->kvm->srcu);
5949         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5950         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5951         tr->physical_address = gpa;
5952         tr->valid = gpa != UNMAPPED_GVA;
5953         tr->writeable = 1;
5954         tr->usermode = 0;
5955
5956         return 0;
5957 }
5958
5959 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5960 {
5961         struct i387_fxsave_struct *fxsave =
5962                         &vcpu->arch.guest_fpu.state->fxsave;
5963
5964         memcpy(fpu->fpr, fxsave->st_space, 128);
5965         fpu->fcw = fxsave->cwd;
5966         fpu->fsw = fxsave->swd;
5967         fpu->ftwx = fxsave->twd;
5968         fpu->last_opcode = fxsave->fop;
5969         fpu->last_ip = fxsave->rip;
5970         fpu->last_dp = fxsave->rdp;
5971         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5972
5973         return 0;
5974 }
5975
5976 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5977 {
5978         struct i387_fxsave_struct *fxsave =
5979                         &vcpu->arch.guest_fpu.state->fxsave;
5980
5981         memcpy(fxsave->st_space, fpu->fpr, 128);
5982         fxsave->cwd = fpu->fcw;
5983         fxsave->swd = fpu->fsw;
5984         fxsave->twd = fpu->ftwx;
5985         fxsave->fop = fpu->last_opcode;
5986         fxsave->rip = fpu->last_ip;
5987         fxsave->rdp = fpu->last_dp;
5988         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5989
5990         return 0;
5991 }
5992
5993 int fx_init(struct kvm_vcpu *vcpu)
5994 {
5995         int err;
5996
5997         err = fpu_alloc(&vcpu->arch.guest_fpu);
5998         if (err)
5999                 return err;
6000
6001         fpu_finit(&vcpu->arch.guest_fpu);
6002
6003         /*
6004          * Ensure guest xcr0 is valid for loading
6005          */
6006         vcpu->arch.xcr0 = XSTATE_FP;
6007
6008         vcpu->arch.cr0 |= X86_CR0_ET;
6009
6010         return 0;
6011 }
6012 EXPORT_SYMBOL_GPL(fx_init);
6013
6014 static void fx_free(struct kvm_vcpu *vcpu)
6015 {
6016         fpu_free(&vcpu->arch.guest_fpu);
6017 }
6018
6019 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6020 {
6021         if (vcpu->guest_fpu_loaded)
6022                 return;
6023
6024         /*
6025          * Restore all possible states in the guest,
6026          * and assume host would use all available bits.
6027          * Guest xcr0 would be loaded later.
6028          */
6029         kvm_put_guest_xcr0(vcpu);
6030         vcpu->guest_fpu_loaded = 1;
6031         unlazy_fpu(current);
6032         fpu_restore_checking(&vcpu->arch.guest_fpu);
6033         trace_kvm_fpu(1);
6034 }
6035
6036 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6037 {
6038         kvm_put_guest_xcr0(vcpu);
6039
6040         if (!vcpu->guest_fpu_loaded)
6041                 return;
6042
6043         vcpu->guest_fpu_loaded = 0;
6044         fpu_save_init(&vcpu->arch.guest_fpu);
6045         ++vcpu->stat.fpu_reload;
6046         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6047         trace_kvm_fpu(0);
6048 }
6049
6050 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6051 {
6052         kvmclock_reset(vcpu);
6053
6054         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6055         fx_free(vcpu);
6056         kvm_x86_ops->vcpu_free(vcpu);
6057 }
6058
6059 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6060                                                 unsigned int id)
6061 {
6062         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6063                 printk_once(KERN_WARNING
6064                 "kvm: SMP vm created on host with unstable TSC; "
6065                 "guest TSC will not be reliable\n");
6066         return kvm_x86_ops->vcpu_create(kvm, id);
6067 }
6068
6069 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6070 {
6071         int r;
6072
6073         vcpu->arch.mtrr_state.have_fixed = 1;
6074         vcpu_load(vcpu);
6075         r = kvm_arch_vcpu_reset(vcpu);
6076         if (r == 0)
6077                 r = kvm_mmu_setup(vcpu);
6078         vcpu_put(vcpu);
6079         if (r < 0)
6080                 goto free_vcpu;
6081
6082         return 0;
6083 free_vcpu:
6084         kvm_x86_ops->vcpu_free(vcpu);
6085         return r;
6086 }
6087
6088 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6089 {
6090         vcpu->arch.apf.msr_val = 0;
6091
6092         vcpu_load(vcpu);
6093         kvm_mmu_unload(vcpu);
6094         vcpu_put(vcpu);
6095
6096         fx_free(vcpu);
6097         kvm_x86_ops->vcpu_free(vcpu);
6098 }
6099
6100 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6101 {
6102         vcpu->arch.nmi_pending = false;
6103         vcpu->arch.nmi_injected = false;
6104
6105         vcpu->arch.switch_db_regs = 0;
6106         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6107         vcpu->arch.dr6 = DR6_FIXED_1;
6108         vcpu->arch.dr7 = DR7_FIXED_1;
6109
6110         kvm_make_request(KVM_REQ_EVENT, vcpu);
6111         vcpu->arch.apf.msr_val = 0;
6112
6113         kvmclock_reset(vcpu);
6114
6115         kvm_clear_async_pf_completion_queue(vcpu);
6116         kvm_async_pf_hash_reset(vcpu);
6117         vcpu->arch.apf.halted = false;
6118
6119         return kvm_x86_ops->vcpu_reset(vcpu);
6120 }
6121
6122 int kvm_arch_hardware_enable(void *garbage)
6123 {
6124         struct kvm *kvm;
6125         struct kvm_vcpu *vcpu;
6126         int i;
6127
6128         kvm_shared_msr_cpu_online();
6129         list_for_each_entry(kvm, &vm_list, vm_list)
6130                 kvm_for_each_vcpu(i, vcpu, kvm)
6131                         if (vcpu->cpu == smp_processor_id())
6132                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6133         return kvm_x86_ops->hardware_enable(garbage);
6134 }
6135
6136 void kvm_arch_hardware_disable(void *garbage)
6137 {
6138         kvm_x86_ops->hardware_disable(garbage);
6139         drop_user_return_notifiers(garbage);
6140 }
6141
6142 int kvm_arch_hardware_setup(void)
6143 {
6144         return kvm_x86_ops->hardware_setup();
6145 }
6146
6147 void kvm_arch_hardware_unsetup(void)
6148 {
6149         kvm_x86_ops->hardware_unsetup();
6150 }
6151
6152 void kvm_arch_check_processor_compat(void *rtn)
6153 {
6154         kvm_x86_ops->check_processor_compatibility(rtn);
6155 }
6156
6157 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6158 {
6159         struct page *page;
6160         struct kvm *kvm;
6161         int r;
6162
6163         BUG_ON(vcpu->kvm == NULL);
6164         kvm = vcpu->kvm;
6165
6166         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6167         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6168         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6169         vcpu->arch.mmu.translate_gpa = translate_gpa;
6170         vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6171         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6172                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6173         else
6174                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6175
6176         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6177         if (!page) {
6178                 r = -ENOMEM;
6179                 goto fail;
6180         }
6181         vcpu->arch.pio_data = page_address(page);
6182
6183         kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6184
6185         r = kvm_mmu_create(vcpu);
6186         if (r < 0)
6187                 goto fail_free_pio_data;
6188
6189         if (irqchip_in_kernel(kvm)) {
6190                 r = kvm_create_lapic(vcpu);
6191                 if (r < 0)
6192                         goto fail_mmu_destroy;
6193         }
6194
6195         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6196                                        GFP_KERNEL);
6197         if (!vcpu->arch.mce_banks) {
6198                 r = -ENOMEM;
6199                 goto fail_free_lapic;
6200         }
6201         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6202
6203         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6204                 goto fail_free_mce_banks;
6205
6206         kvm_async_pf_hash_reset(vcpu);
6207
6208         return 0;
6209 fail_free_mce_banks:
6210         kfree(vcpu->arch.mce_banks);
6211 fail_free_lapic:
6212         kvm_free_lapic(vcpu);
6213 fail_mmu_destroy:
6214         kvm_mmu_destroy(vcpu);
6215 fail_free_pio_data:
6216         free_page((unsigned long)vcpu->arch.pio_data);
6217 fail:
6218         return r;
6219 }
6220
6221 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6222 {
6223         int idx;
6224
6225         kfree(vcpu->arch.mce_banks);
6226         kvm_free_lapic(vcpu);
6227         idx = srcu_read_lock(&vcpu->kvm->srcu);
6228         kvm_mmu_destroy(vcpu);
6229         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6230         free_page((unsigned long)vcpu->arch.pio_data);
6231 }
6232
6233 int kvm_arch_init_vm(struct kvm *kvm)
6234 {
6235         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6236         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6237
6238         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6239         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6240
6241         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6242
6243         return 0;
6244 }
6245
6246 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6247 {
6248         vcpu_load(vcpu);
6249         kvm_mmu_unload(vcpu);
6250         vcpu_put(vcpu);
6251 }
6252
6253 static void kvm_free_vcpus(struct kvm *kvm)
6254 {
6255         unsigned int i;
6256         struct kvm_vcpu *vcpu;
6257
6258         /*
6259          * Unpin any mmu pages first.
6260          */
6261         kvm_for_each_vcpu(i, vcpu, kvm) {
6262                 kvm_clear_async_pf_completion_queue(vcpu);
6263                 kvm_unload_vcpu_mmu(vcpu);
6264         }
6265         kvm_for_each_vcpu(i, vcpu, kvm)
6266                 kvm_arch_vcpu_free(vcpu);
6267
6268         mutex_lock(&kvm->lock);
6269         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6270                 kvm->vcpus[i] = NULL;
6271
6272         atomic_set(&kvm->online_vcpus, 0);
6273         mutex_unlock(&kvm->lock);
6274 }
6275
6276 void kvm_arch_sync_events(struct kvm *kvm)
6277 {
6278         kvm_free_all_assigned_devices(kvm);
6279         kvm_free_pit(kvm);
6280 }
6281
6282 void kvm_arch_destroy_vm(struct kvm *kvm)
6283 {
6284         kvm_iommu_unmap_guest(kvm);
6285         kfree(kvm->arch.vpic);
6286         kfree(kvm->arch.vioapic);
6287         kvm_free_vcpus(kvm);
6288         if (kvm->arch.apic_access_page)
6289                 put_page(kvm->arch.apic_access_page);
6290         if (kvm->arch.ept_identity_pagetable)
6291                 put_page(kvm->arch.ept_identity_pagetable);
6292 }
6293
6294 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6295                                 struct kvm_memory_slot *memslot,
6296                                 struct kvm_memory_slot old,
6297                                 struct kvm_userspace_memory_region *mem,
6298                                 int user_alloc)
6299 {
6300         int npages = memslot->npages;
6301         int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6302
6303         /* Prevent internal slot pages from being moved by fork()/COW. */
6304         if (memslot->id >= KVM_MEMORY_SLOTS)
6305                 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6306
6307         /*To keep backward compatibility with older userspace,
6308          *x86 needs to hanlde !user_alloc case.
6309          */
6310         if (!user_alloc) {
6311                 if (npages && !old.rmap) {
6312                         unsigned long userspace_addr;
6313
6314                         down_write(&current->mm->mmap_sem);
6315                         userspace_addr = do_mmap(NULL, 0,
6316                                                  npages * PAGE_SIZE,
6317                                                  PROT_READ | PROT_WRITE,
6318                                                  map_flags,
6319                                                  0);
6320                         up_write(&current->mm->mmap_sem);
6321
6322                         if (IS_ERR((void *)userspace_addr))
6323                                 return PTR_ERR((void *)userspace_addr);
6324
6325                         memslot->userspace_addr = userspace_addr;
6326                 }
6327         }
6328
6329
6330         return 0;
6331 }
6332
6333 void kvm_arch_commit_memory_region(struct kvm *kvm,
6334                                 struct kvm_userspace_memory_region *mem,
6335                                 struct kvm_memory_slot old,
6336                                 int user_alloc)
6337 {
6338
6339         int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6340
6341         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6342                 int ret;
6343
6344                 down_write(&current->mm->mmap_sem);
6345                 ret = do_munmap(current->mm, old.userspace_addr,
6346                                 old.npages * PAGE_SIZE);
6347                 up_write(&current->mm->mmap_sem);
6348                 if (ret < 0)
6349                         printk(KERN_WARNING
6350                                "kvm_vm_ioctl_set_memory_region: "
6351                                "failed to munmap memory\n");
6352         }
6353
6354         if (!kvm->arch.n_requested_mmu_pages)
6355                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6356
6357         spin_lock(&kvm->mmu_lock);
6358         if (nr_mmu_pages)
6359                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6360         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6361         spin_unlock(&kvm->mmu_lock);
6362 }
6363
6364 void kvm_arch_flush_shadow(struct kvm *kvm)
6365 {
6366         kvm_mmu_zap_all(kvm);
6367         kvm_reload_remote_mmus(kvm);
6368 }
6369
6370 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6371 {
6372         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6373                 !vcpu->arch.apf.halted)
6374                 || !list_empty_careful(&vcpu->async_pf.done)
6375                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6376                 || vcpu->arch.nmi_pending ||
6377                 (kvm_arch_interrupt_allowed(vcpu) &&
6378                  kvm_cpu_has_interrupt(vcpu));
6379 }
6380
6381 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6382 {
6383         int me;
6384         int cpu = vcpu->cpu;
6385
6386         if (waitqueue_active(&vcpu->wq)) {
6387                 wake_up_interruptible(&vcpu->wq);
6388                 ++vcpu->stat.halt_wakeup;
6389         }
6390
6391         me = get_cpu();
6392         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6393                 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6394                         smp_send_reschedule(cpu);
6395         put_cpu();
6396 }
6397
6398 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6399 {
6400         return kvm_x86_ops->interrupt_allowed(vcpu);
6401 }
6402
6403 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6404 {
6405         unsigned long current_rip = kvm_rip_read(vcpu) +
6406                 get_segment_base(vcpu, VCPU_SREG_CS);
6407
6408         return current_rip == linear_rip;
6409 }
6410 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6411
6412 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6413 {
6414         unsigned long rflags;
6415
6416         rflags = kvm_x86_ops->get_rflags(vcpu);
6417         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6418                 rflags &= ~X86_EFLAGS_TF;
6419         return rflags;
6420 }
6421 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6422
6423 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6424 {
6425         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6426             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6427                 rflags |= X86_EFLAGS_TF;
6428         kvm_x86_ops->set_rflags(vcpu, rflags);
6429         kvm_make_request(KVM_REQ_EVENT, vcpu);
6430 }
6431 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6432
6433 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6434 {
6435         int r;
6436
6437         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6438               is_error_page(work->page))
6439                 return;
6440
6441         r = kvm_mmu_reload(vcpu);
6442         if (unlikely(r))
6443                 return;
6444
6445         if (!vcpu->arch.mmu.direct_map &&
6446               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6447                 return;
6448
6449         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6450 }
6451
6452 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6453 {
6454         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6455 }
6456
6457 static inline u32 kvm_async_pf_next_probe(u32 key)
6458 {
6459         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6460 }
6461
6462 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6463 {
6464         u32 key = kvm_async_pf_hash_fn(gfn);
6465
6466         while (vcpu->arch.apf.gfns[key] != ~0)
6467                 key = kvm_async_pf_next_probe(key);
6468
6469         vcpu->arch.apf.gfns[key] = gfn;
6470 }
6471
6472 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6473 {
6474         int i;
6475         u32 key = kvm_async_pf_hash_fn(gfn);
6476
6477         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6478                      (vcpu->arch.apf.gfns[key] != gfn &&
6479                       vcpu->arch.apf.gfns[key] != ~0); i++)
6480                 key = kvm_async_pf_next_probe(key);
6481
6482         return key;
6483 }
6484
6485 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6486 {
6487         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6488 }
6489
6490 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6491 {
6492         u32 i, j, k;
6493
6494         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6495         while (true) {
6496                 vcpu->arch.apf.gfns[i] = ~0;
6497                 do {
6498                         j = kvm_async_pf_next_probe(j);
6499                         if (vcpu->arch.apf.gfns[j] == ~0)
6500                                 return;
6501                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6502                         /*
6503                          * k lies cyclically in ]i,j]
6504                          * |    i.k.j |
6505                          * |....j i.k.| or  |.k..j i...|
6506                          */
6507                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6508                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6509                 i = j;
6510         }
6511 }
6512
6513 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6514 {
6515
6516         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6517                                       sizeof(val));
6518 }
6519
6520 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6521                                      struct kvm_async_pf *work)
6522 {
6523         struct x86_exception fault;
6524
6525         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6526         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6527
6528         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6529             (vcpu->arch.apf.send_user_only &&
6530              kvm_x86_ops->get_cpl(vcpu) == 0))
6531                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6532         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6533                 fault.vector = PF_VECTOR;
6534                 fault.error_code_valid = true;
6535                 fault.error_code = 0;
6536                 fault.nested_page_fault = false;
6537                 fault.address = work->arch.token;
6538                 kvm_inject_page_fault(vcpu, &fault);
6539         }
6540 }
6541
6542 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6543                                  struct kvm_async_pf *work)
6544 {
6545         struct x86_exception fault;
6546
6547         trace_kvm_async_pf_ready(work->arch.token, work->gva);
6548         if (is_error_page(work->page))
6549                 work->arch.token = ~0; /* broadcast wakeup */
6550         else
6551                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6552
6553         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6554             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6555                 fault.vector = PF_VECTOR;
6556                 fault.error_code_valid = true;
6557                 fault.error_code = 0;
6558                 fault.nested_page_fault = false;
6559                 fault.address = work->arch.token;
6560                 kvm_inject_page_fault(vcpu, &fault);
6561         }
6562         vcpu->arch.apf.halted = false;
6563 }
6564
6565 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6566 {
6567         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6568                 return true;
6569         else
6570                 return !kvm_event_needs_reinjection(vcpu) &&
6571                         kvm_x86_ops->interrupt_allowed(vcpu);
6572 }
6573
6574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6579 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6580 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6581 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);