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 /* EFER defaults:
  67  * - enable syscall per default because its emulated by KVM
  68  * - enable LME and LMA per default on 64 bit KVM
  69  */
  70 #ifdef CONFIG_X86_64
  71 static
  72 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  73 #else
  74 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  75 #endif
  76
  77 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  78 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  79
  80 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  81 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  82                                     struct kvm_cpuid_entry2 __user *entries);
  83
  84 struct kvm_x86_ops *kvm_x86_ops;
  85 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  86
  87 int ignore_msrs = 0;
  88 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  89
  90 bool kvm_has_tsc_control;
  91 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
  92 u32  kvm_max_guest_tsc_khz;
  93 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
  94
  95 #define KVM_NR_SHARED_MSRS 16
  96
  97 struct kvm_shared_msrs_global {
  98         int nr;
  99         u32 msrs[KVM_NR_SHARED_MSRS];
 100 };
 101
 102 struct kvm_shared_msrs {
 103         struct user_return_notifier urn;
 104         bool registered;
 105         struct kvm_shared_msr_values {
 106                 u64 host;
 107                 u64 curr;
 108         } values[KVM_NR_SHARED_MSRS];
 109 };
 110
 111 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 112 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 113
 114 struct kvm_stats_debugfs_item debugfs_entries[] = {
 115         { "pf_fixed", VCPU_STAT(pf_fixed) },
 116         { "pf_guest", VCPU_STAT(pf_guest) },
 117         { "tlb_flush", VCPU_STAT(tlb_flush) },
 118         { "invlpg", VCPU_STAT(invlpg) },
 119         { "exits", VCPU_STAT(exits) },
 120         { "io_exits", VCPU_STAT(io_exits) },
 121         { "mmio_exits", VCPU_STAT(mmio_exits) },
 122         { "signal_exits", VCPU_STAT(signal_exits) },
 123         { "irq_window", VCPU_STAT(irq_window_exits) },
 124         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 125         { "halt_exits", VCPU_STAT(halt_exits) },
 126         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 127         { "hypercalls", VCPU_STAT(hypercalls) },
 128         { "request_irq", VCPU_STAT(request_irq_exits) },
 129         { "irq_exits", VCPU_STAT(irq_exits) },
 130         { "host_state_reload", VCPU_STAT(host_state_reload) },
 131         { "efer_reload", VCPU_STAT(efer_reload) },
 132         { "fpu_reload", VCPU_STAT(fpu_reload) },
 133         { "insn_emulation", VCPU_STAT(insn_emulation) },
 134         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 135         { "irq_injections", VCPU_STAT(irq_injections) },
 136         { "nmi_injections", VCPU_STAT(nmi_injections) },
 137         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 138         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 139         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 140         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 141         { "mmu_flooded", VM_STAT(mmu_flooded) },
 142         { "mmu_recycled", VM_STAT(mmu_recycled) },
 143         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 144         { "mmu_unsync", VM_STAT(mmu_unsync) },
 145         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 146         { "largepages", VM_STAT(lpages) },
 147         { NULL }
 148 };
 149
 150 u64 __read_mostly host_xcr0;
 151
 152 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 153 {
 154         int i;
 155         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 156                 vcpu->arch.apf.gfns[i] = ~0;
 157 }
 158
 159 static void kvm_on_user_return(struct user_return_notifier *urn)
 160 {
 161         unsigned slot;
 162         struct kvm_shared_msrs *locals
 163                 = container_of(urn, struct kvm_shared_msrs, urn);
 164         struct kvm_shared_msr_values *values;
 165
 166         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 167                 values = &locals->values[slot];
 168                 if (values->host != values->curr) {
 169                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 170                         values->curr = values->host;
 171                 }
 172         }
 173         locals->registered = false;
 174         user_return_notifier_unregister(urn);
 175 }
 176
 177 static void shared_msr_update(unsigned slot, u32 msr)
 178 {
 179         struct kvm_shared_msrs *smsr;
 180         u64 value;
 181
 182         smsr = &__get_cpu_var(shared_msrs);
 183         /* only read, and nobody should modify it at this time,
 184          * so don't need lock */
 185         if (slot >= shared_msrs_global.nr) {
 186                 printk(KERN_ERR "kvm: invalid MSR slot!");
 187                 return;
 188         }
 189         rdmsrl_safe(msr, &value);
 190         smsr->values[slot].host = value;
 191         smsr->values[slot].curr = value;
 192 }
 193
 194 void kvm_define_shared_msr(unsigned slot, u32 msr)
 195 {
 196         if (slot >= shared_msrs_global.nr)
 197                 shared_msrs_global.nr = slot + 1;
 198         shared_msrs_global.msrs[slot] = msr;
 199         /* we need ensured the shared_msr_global have been updated */
 200         smp_wmb();
 201 }
 202 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 203
 204 static void kvm_shared_msr_cpu_online(void)
 205 {
 206         unsigned i;
 207
 208         for (i = 0; i < shared_msrs_global.nr; ++i)
 209                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 210 }
 211
 212 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 213 {
 214         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 215
 216         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 217                 return;
 218         smsr->values[slot].curr = value;
 219         wrmsrl(shared_msrs_global.msrs[slot], value);
 220         if (!smsr->registered) {
 221                 smsr->urn.on_user_return = kvm_on_user_return;
 222                 user_return_notifier_register(&smsr->urn);
 223                 smsr->registered = true;
 224         }
 225 }
 226 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 227
 228 static void drop_user_return_notifiers(void *ignore)
 229 {
 230         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 231
 232         if (smsr->registered)
 233                 kvm_on_user_return(&smsr->urn);
 234 }
 235
 236 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 237 {
 238         if (irqchip_in_kernel(vcpu->kvm))
 239                 return vcpu->arch.apic_base;
 240         else
 241                 return vcpu->arch.apic_base;
 242 }
 243 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 244
 245 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 246 {
 247         /* TODO: reserve bits check */
 248         if (irqchip_in_kernel(vcpu->kvm))
 249                 kvm_lapic_set_base(vcpu, data);
 250         else
 251                 vcpu->arch.apic_base = data;
 252 }
 253 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 254
 255 #define EXCPT_BENIGN            0
 256 #define EXCPT_CONTRIBUTORY      1
 257 #define EXCPT_PF                2
 258
 259 static int exception_class(int vector)
 260 {
 261         switch (vector) {
 262         case PF_VECTOR:
 263                 return EXCPT_PF;
 264         case DE_VECTOR:
 265         case TS_VECTOR:
 266         case NP_VECTOR:
 267         case SS_VECTOR:
 268         case GP_VECTOR:
 269                 return EXCPT_CONTRIBUTORY;
 270         default:
 271                 break;
 272         }
 273         return EXCPT_BENIGN;
 274 }
 275
 276 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 277                 unsigned nr, bool has_error, u32 error_code,
 278                 bool reinject)
 279 {
 280         u32 prev_nr;
 281         int class1, class2;
 282
 283         kvm_make_request(KVM_REQ_EVENT, vcpu);
 284
 285         if (!vcpu->arch.exception.pending) {
 286         queue:
 287                 vcpu->arch.exception.pending = true;
 288                 vcpu->arch.exception.has_error_code = has_error;
 289                 vcpu->arch.exception.nr = nr;
 290                 vcpu->arch.exception.error_code = error_code;
 291                 vcpu->arch.exception.reinject = reinject;
 292                 return;
 293         }
 294
 295         /* to check exception */
 296         prev_nr = vcpu->arch.exception.nr;
 297         if (prev_nr == DF_VECTOR) {
 298                 /* triple fault -> shutdown */
 299                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 300                 return;
 301         }
 302         class1 = exception_class(prev_nr);
 303         class2 = exception_class(nr);
 304         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 305                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 306                 /* generate double fault per SDM Table 5-5 */
 307                 vcpu->arch.exception.pending = true;
 308                 vcpu->arch.exception.has_error_code = true;
 309                 vcpu->arch.exception.nr = DF_VECTOR;
 310                 vcpu->arch.exception.error_code = 0;
 311         } else
 312                 /* replace previous exception with a new one in a hope
 313                    that instruction re-execution will regenerate lost
 314                    exception */
 315                 goto queue;
 316 }
 317
 318 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 319 {
 320         kvm_multiple_exception(vcpu, nr, false, 0, false);
 321 }
 322 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 323
 324 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 325 {
 326         kvm_multiple_exception(vcpu, nr, false, 0, true);
 327 }
 328 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 329
 330 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 331 {
 332         if (err)
 333                 kvm_inject_gp(vcpu, 0);
 334         else
 335                 kvm_x86_ops->skip_emulated_instruction(vcpu);
 336 }
 337 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 338
 339 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 340 {
 341         ++vcpu->stat.pf_guest;
 342         vcpu->arch.cr2 = fault->address;
 343         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 344 }
 345
 346 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 347 {
 348         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 349                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 350         else
 351                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 352 }
 353
 354 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 355 {
 356         kvm_make_request(KVM_REQ_EVENT, vcpu);
 357         vcpu->arch.nmi_pending = 1;
 358 }
 359 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 360
 361 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 362 {
 363         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 364 }
 365 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 366
 367 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 368 {
 369         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 370 }
 371 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 372
 373 /*
 374  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 375  * a #GP and return false.
 376  */
 377 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 378 {
 379         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 380                 return true;
 381         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 382         return false;
 383 }
 384 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 385
 386 /*
 387  * This function will be used to read from the physical memory of the currently
 388  * running guest. The difference to kvm_read_guest_page is that this function
 389  * can read from guest physical or from the guest's guest physical memory.
 390  */
 391 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 392                             gfn_t ngfn, void *data, int offset, int len,
 393                             u32 access)
 394 {
 395         gfn_t real_gfn;
 396         gpa_t ngpa;
 397
 398         ngpa     = gfn_to_gpa(ngfn);
 399         real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
 400         if (real_gfn == UNMAPPED_GVA)
 401                 return -EFAULT;
 402
 403         real_gfn = gpa_to_gfn(real_gfn);
 404
 405         return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
 406 }
 407 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 408
 409 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 410                                void *data, int offset, int len, u32 access)
 411 {
 412         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 413                                        data, offset, len, access);
 414 }
 415
 416 /*
 417  * Load the pae pdptrs.  Return true is they are all valid.
 418  */
 419 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 420 {
 421         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 422         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 423         int i;
 424         int ret;
 425         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 426
 427         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 428                                       offset * sizeof(u64), sizeof(pdpte),
 429                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 430         if (ret < 0) {
 431                 ret = 0;
 432                 goto out;
 433         }
 434         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 435                 if (is_present_gpte(pdpte[i]) &&
 436                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 437                         ret = 0;
 438                         goto out;
 439                 }
 440         }
 441         ret = 1;
 442
 443         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 444         __set_bit(VCPU_EXREG_PDPTR,
 445                   (unsigned long *)&vcpu->arch.regs_avail);
 446         __set_bit(VCPU_EXREG_PDPTR,
 447                   (unsigned long *)&vcpu->arch.regs_dirty);
 448 out:
 449
 450         return ret;
 451 }
 452 EXPORT_SYMBOL_GPL(load_pdptrs);
 453
 454 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 455 {
 456         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 457         bool changed = true;
 458         int offset;
 459         gfn_t gfn;
 460         int r;
 461
 462         if (is_long_mode(vcpu) || !is_pae(vcpu))
 463                 return false;
 464
 465         if (!test_bit(VCPU_EXREG_PDPTR,
 466                       (unsigned long *)&vcpu->arch.regs_avail))
 467                 return true;
 468
 469         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 470         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 471         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 472                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 473         if (r < 0)
 474                 goto out;
 475         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 476 out:
 477
 478         return changed;
 479 }
 480
 481 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 482 {
 483         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 484         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 485                                     X86_CR0_CD | X86_CR0_NW;
 486
 487         cr0 |= X86_CR0_ET;
 488
 489 #ifdef CONFIG_X86_64
 490         if (cr0 & 0xffffffff00000000UL)
 491                 return 1;
 492 #endif
 493
 494         cr0 &= ~CR0_RESERVED_BITS;
 495
 496         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 497                 return 1;
 498
 499         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 500                 return 1;
 501
 502         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 503 #ifdef CONFIG_X86_64
 504                 if ((vcpu->arch.efer & EFER_LME)) {
 505                         int cs_db, cs_l;
 506
 507                         if (!is_pae(vcpu))
 508                                 return 1;
 509                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 510                         if (cs_l)
 511                                 return 1;
 512                 } else
 513 #endif
 514                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 515                                                  kvm_read_cr3(vcpu)))
 516                         return 1;
 517         }
 518
 519         kvm_x86_ops->set_cr0(vcpu, cr0);
 520
 521         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 522                 kvm_clear_async_pf_completion_queue(vcpu);
 523                 kvm_async_pf_hash_reset(vcpu);
 524         }
 525
 526         if ((cr0 ^ old_cr0) & update_bits)
 527                 kvm_mmu_reset_context(vcpu);
 528         return 0;
 529 }
 530 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 531
 532 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 533 {
 534         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 535 }
 536 EXPORT_SYMBOL_GPL(kvm_lmsw);
 537
 538 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 539 {
 540         u64 xcr0;
 541
 542         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 543         if (index != XCR_XFEATURE_ENABLED_MASK)
 544                 return 1;
 545         xcr0 = xcr;
 546         if (kvm_x86_ops->get_cpl(vcpu) != 0)
 547                 return 1;
 548         if (!(xcr0 & XSTATE_FP))
 549                 return 1;
 550         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
 551                 return 1;
 552         if (xcr0 & ~host_xcr0)
 553                 return 1;
 554         vcpu->arch.xcr0 = xcr0;
 555         vcpu->guest_xcr0_loaded = 0;
 556         return 0;
 557 }
 558
 559 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 560 {
 561         if (__kvm_set_xcr(vcpu, index, xcr)) {
 562                 kvm_inject_gp(vcpu, 0);
 563                 return 1;
 564         }
 565         return 0;
 566 }
 567 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 568
 569 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
 570 {
 571         struct kvm_cpuid_entry2 *best;
 572
 573         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 574         return best && (best->ecx & bit(X86_FEATURE_XSAVE));
 575 }
 576
 577 static void update_cpuid(struct kvm_vcpu *vcpu)
 578 {
 579         struct kvm_cpuid_entry2 *best;
 580
 581         best = kvm_find_cpuid_entry(vcpu, 1, 0);
 582         if (!best)
 583                 return;
 584
 585         /* Update OSXSAVE bit */
 586         if (cpu_has_xsave && best->function == 0x1) {
 587                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
 588                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
 589                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
 590         }
 591 }
 592
 593 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 594 {
 595         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 596         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 597
 598         if (cr4 & CR4_RESERVED_BITS)
 599                 return 1;
 600
 601         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 602                 return 1;
 603
 604         if (is_long_mode(vcpu)) {
 605                 if (!(cr4 & X86_CR4_PAE))
 606                         return 1;
 607         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 608                    && ((cr4 ^ old_cr4) & pdptr_bits)
 609                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 610                                    kvm_read_cr3(vcpu)))
 611                 return 1;
 612
 613         if (cr4 & X86_CR4_VMXE)
 614                 return 1;
 615
 616         kvm_x86_ops->set_cr4(vcpu, cr4);
 617
 618         if ((cr4 ^ old_cr4) & pdptr_bits)
 619                 kvm_mmu_reset_context(vcpu);
 620
 621         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 622                 update_cpuid(vcpu);
 623
 624         return 0;
 625 }
 626 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 627
 628 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 629 {
 630         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 631                 kvm_mmu_sync_roots(vcpu);
 632                 kvm_mmu_flush_tlb(vcpu);
 633                 return 0;
 634         }
 635
 636         if (is_long_mode(vcpu)) {
 637                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 638                         return 1;
 639         } else {
 640                 if (is_pae(vcpu)) {
 641                         if (cr3 & CR3_PAE_RESERVED_BITS)
 642                                 return 1;
 643                         if (is_paging(vcpu) &&
 644                             !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 645                                 return 1;
 646                 }
 647                 /*
 648                  * We don't check reserved bits in nonpae mode, because
 649                  * this isn't enforced, and VMware depends on this.
 650                  */
 651         }
 652
 653         /*
 654          * Does the new cr3 value map to physical memory? (Note, we
 655          * catch an invalid cr3 even in real-mode, because it would
 656          * cause trouble later on when we turn on paging anyway.)
 657          *
 658          * A real CPU would silently accept an invalid cr3 and would
 659          * attempt to use it - with largely undefined (and often hard
 660          * to debug) behavior on the guest side.
 661          */
 662         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 663                 return 1;
 664         vcpu->arch.cr3 = cr3;
 665         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 666         vcpu->arch.mmu.new_cr3(vcpu);
 667         return 0;
 668 }
 669 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 670
 671 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 672 {
 673         if (cr8 & CR8_RESERVED_BITS)
 674                 return 1;
 675         if (irqchip_in_kernel(vcpu->kvm))
 676                 kvm_lapic_set_tpr(vcpu, cr8);
 677         else
 678                 vcpu->arch.cr8 = cr8;
 679         return 0;
 680 }
 681 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 682
 683 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 684 {
 685         if (irqchip_in_kernel(vcpu->kvm))
 686                 return kvm_lapic_get_cr8(vcpu);
 687         else
 688                 return vcpu->arch.cr8;
 689 }
 690 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 691
 692 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 693 {
 694         switch (dr) {
 695         case 0 ... 3:
 696                 vcpu->arch.db[dr] = val;
 697                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 698                         vcpu->arch.eff_db[dr] = val;
 699                 break;
 700         case 4:
 701                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 702                         return 1; /* #UD */
 703                 /* fall through */
 704         case 6:
 705                 if (val & 0xffffffff00000000ULL)
 706                         return -1; /* #GP */
 707                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 708                 break;
 709         case 5:
 710                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 711                         return 1; /* #UD */
 712                 /* fall through */
 713         default: /* 7 */
 714                 if (val & 0xffffffff00000000ULL)
 715                         return -1; /* #GP */
 716                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 717                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 718                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
 719                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
 720                 }
 721                 break;
 722         }
 723
 724         return 0;
 725 }
 726
 727 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 728 {
 729         int res;
 730
 731         res = __kvm_set_dr(vcpu, dr, val);
 732         if (res > 0)
 733                 kvm_queue_exception(vcpu, UD_VECTOR);
 734         else if (res < 0)
 735                 kvm_inject_gp(vcpu, 0);
 736
 737         return res;
 738 }
 739 EXPORT_SYMBOL_GPL(kvm_set_dr);
 740
 741 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 742 {
 743         switch (dr) {
 744         case 0 ... 3:
 745                 *val = vcpu->arch.db[dr];
 746                 break;
 747         case 4:
 748                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 749                         return 1;
 750                 /* fall through */
 751         case 6:
 752                 *val = vcpu->arch.dr6;
 753                 break;
 754         case 5:
 755                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 756                         return 1;
 757                 /* fall through */
 758         default: /* 7 */
 759                 *val = vcpu->arch.dr7;
 760                 break;
 761         }
 762
 763         return 0;
 764 }
 765
 766 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 767 {
 768         if (_kvm_get_dr(vcpu, dr, val)) {
 769                 kvm_queue_exception(vcpu, UD_VECTOR);
 770                 return 1;
 771         }
 772         return 0;
 773 }
 774 EXPORT_SYMBOL_GPL(kvm_get_dr);
 775
 776 /*
 777  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 778  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 779  *
 780  * This list is modified at module load time to reflect the
 781  * capabilities of the host cpu. This capabilities test skips MSRs that are
 782  * kvm-specific. Those are put in the beginning of the list.
 783  */
 784
 785 #define KVM_SAVE_MSRS_BEGIN     8
 786 static u32 msrs_to_save[] = {
 787         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 788         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 789         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 790         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
 791         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 792         MSR_STAR,
 793 #ifdef CONFIG_X86_64
 794         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 795 #endif
 796         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 797 };
 798
 799 static unsigned num_msrs_to_save;
 800
 801 static u32 emulated_msrs[] = {
 802         MSR_IA32_MISC_ENABLE,
 803         MSR_IA32_MCG_STATUS,
 804         MSR_IA32_MCG_CTL,
 805 };
 806
 807 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 808 {
 809         u64 old_efer = vcpu->arch.efer;
 810
 811         if (efer & efer_reserved_bits)
 812                 return 1;
 813
 814         if (is_paging(vcpu)
 815             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 816                 return 1;
 817
 818         if (efer & EFER_FFXSR) {
 819                 struct kvm_cpuid_entry2 *feat;
 820
 821                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 822                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 823                         return 1;
 824         }
 825
 826         if (efer & EFER_SVME) {
 827                 struct kvm_cpuid_entry2 *feat;
 828
 829                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 830                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 831                         return 1;
 832         }
 833
 834         efer &= ~EFER_LMA;
 835         efer |= vcpu->arch.efer & EFER_LMA;
 836
 837         kvm_x86_ops->set_efer(vcpu, efer);
 838
 839         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 840
 841         /* Update reserved bits */
 842         if ((efer ^ old_efer) & EFER_NX)
 843                 kvm_mmu_reset_context(vcpu);
 844
 845         return 0;
 846 }
 847
 848 void kvm_enable_efer_bits(u64 mask)
 849 {
 850        efer_reserved_bits &= ~mask;
 851 }
 852 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 853
 854
 855 /*
 856  * Writes msr value into into the appropriate "register".
 857  * Returns 0 on success, non-0 otherwise.
 858  * Assumes vcpu_load() was already called.
 859  */
 860 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 861 {
 862         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 863 }
 864
 865 /*
 866  * Adapt set_msr() to msr_io()'s calling convention
 867  */
 868 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 869 {
 870         return kvm_set_msr(vcpu, index, *data);
 871 }
 872
 873 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 874 {
 875         int version;
 876         int r;
 877         struct pvclock_wall_clock wc;
 878         struct timespec boot;
 879
 880         if (!wall_clock)
 881                 return;
 882
 883         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
 884         if (r)
 885                 return;
 886
 887         if (version & 1)
 888                 ++version;  /* first time write, random junk */
 889
 890         ++version;
 891
 892         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 893
 894         /*
 895          * The guest calculates current wall clock time by adding
 896          * system time (updated by kvm_guest_time_update below) to the
 897          * wall clock specified here.  guest system time equals host
 898          * system time for us, thus we must fill in host boot time here.
 899          */
 900         getboottime(&boot);
 901
 902         wc.sec = boot.tv_sec;
 903         wc.nsec = boot.tv_nsec;
 904         wc.version = version;
 905
 906         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 907
 908         version++;
 909         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 910 }
 911
 912 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 913 {
 914         uint32_t quotient, remainder;
 915
 916         /* Don't try to replace with do_div(), this one calculates
 917          * "(dividend << 32) / divisor" */
 918         __asm__ ( "divl %4"
 919                   : "=a" (quotient), "=d" (remainder)
 920                   : "0" (0), "1" (dividend), "r" (divisor) );
 921         return quotient;
 922 }
 923
 924 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
 925                                s8 *pshift, u32 *pmultiplier)
 926 {
 927         uint64_t scaled64;
 928         int32_t  shift = 0;
 929         uint64_t tps64;
 930         uint32_t tps32;
 931
 932         tps64 = base_khz * 1000LL;
 933         scaled64 = scaled_khz * 1000LL;
 934         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
 935                 tps64 >>= 1;
 936                 shift--;
 937         }
 938
 939         tps32 = (uint32_t)tps64;
 940         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
 941                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
 942                         scaled64 >>= 1;
 943                 else
 944                         tps32 <<= 1;
 945                 shift++;
 946         }
 947
 948         *pshift = shift;
 949         *pmultiplier = div_frac(scaled64, tps32);
 950
 951         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
 952                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
 953 }
 954
 955 static inline u64 get_kernel_ns(void)
 956 {
 957         struct timespec ts;
 958
 959         WARN_ON(preemptible());
 960         ktime_get_ts(&ts);
 961         monotonic_to_bootbased(&ts);
 962         return timespec_to_ns(&ts);
 963 }
 964
 965 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 966 unsigned long max_tsc_khz;
 967
 968 static inline int kvm_tsc_changes_freq(void)
 969 {
 970         int cpu = get_cpu();
 971         int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
 972                   cpufreq_quick_get(cpu) != 0;
 973         put_cpu();
 974         return ret;
 975 }
 976
 977 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
 978 {
 979         if (vcpu->arch.virtual_tsc_khz)
 980                 return vcpu->arch.virtual_tsc_khz;
 981         else
 982                 return __this_cpu_read(cpu_tsc_khz);
 983 }
 984
 985 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
 986 {
 987         u64 ret;
 988
 989         WARN_ON(preemptible());
 990         if (kvm_tsc_changes_freq())
 991                 printk_once(KERN_WARNING
 992                  "kvm: unreliable cycle conversion on adjustable rate TSC\n");
 993         ret = nsec * vcpu_tsc_khz(vcpu);
 994         do_div(ret, USEC_PER_SEC);
 995         return ret;
 996 }
 997
 998 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
 999 {
1000         /* Compute a scale to convert nanoseconds in TSC cycles */
1001         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1002                            &vcpu->arch.tsc_catchup_shift,
1003                            &vcpu->arch.tsc_catchup_mult);
1004 }
1005
1006 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1007 {
1008         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1009                                       vcpu->arch.tsc_catchup_mult,
1010                                       vcpu->arch.tsc_catchup_shift);
1011         tsc += vcpu->arch.last_tsc_write;
1012         return tsc;
1013 }
1014
1015 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1016 {
1017         struct kvm *kvm = vcpu->kvm;
1018         u64 offset, ns, elapsed;
1019         unsigned long flags;
1020         s64 sdiff;
1021
1022         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1023         offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1024         ns = get_kernel_ns();
1025         elapsed = ns - kvm->arch.last_tsc_nsec;
1026         sdiff = data - kvm->arch.last_tsc_write;
1027         if (sdiff < 0)
1028                 sdiff = -sdiff;
1029
1030         /*
1031          * Special case: close write to TSC within 5 seconds of
1032          * another CPU is interpreted as an attempt to synchronize
1033          * The 5 seconds is to accommodate host load / swapping as
1034          * well as any reset of TSC during the boot process.
1035          *
1036          * In that case, for a reliable TSC, we can match TSC offsets,
1037          * or make a best guest using elapsed value.
1038          */
1039         if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1040             elapsed < 5ULL * NSEC_PER_SEC) {
1041                 if (!check_tsc_unstable()) {
1042                         offset = kvm->arch.last_tsc_offset;
1043                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1044                 } else {
1045                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1046                         offset += delta;
1047                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1048                 }
1049                 ns = kvm->arch.last_tsc_nsec;
1050         }
1051         kvm->arch.last_tsc_nsec = ns;
1052         kvm->arch.last_tsc_write = data;
1053         kvm->arch.last_tsc_offset = offset;
1054         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1055         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1056
1057         /* Reset of TSC must disable overshoot protection below */
1058         vcpu->arch.hv_clock.tsc_timestamp = 0;
1059         vcpu->arch.last_tsc_write = data;
1060         vcpu->arch.last_tsc_nsec = ns;
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1063
1064 static int kvm_guest_time_update(struct kvm_vcpu *v)
1065 {
1066         unsigned long flags;
1067         struct kvm_vcpu_arch *vcpu = &v->arch;
1068         void *shared_kaddr;
1069         unsigned long this_tsc_khz;
1070         s64 kernel_ns, max_kernel_ns;
1071         u64 tsc_timestamp;
1072
1073         /* Keep irq disabled to prevent changes to the clock */
1074         local_irq_save(flags);
1075         kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1076         kernel_ns = get_kernel_ns();
1077         this_tsc_khz = vcpu_tsc_khz(v);
1078         if (unlikely(this_tsc_khz == 0)) {
1079                 local_irq_restore(flags);
1080                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1081                 return 1;
1082         }
1083
1084         /*
1085          * We may have to catch up the TSC to match elapsed wall clock
1086          * time for two reasons, even if kvmclock is used.
1087          *   1) CPU could have been running below the maximum TSC rate
1088          *   2) Broken TSC compensation resets the base at each VCPU
1089          *      entry to avoid unknown leaps of TSC even when running
1090          *      again on the same CPU.  This may cause apparent elapsed
1091          *      time to disappear, and the guest to stand still or run
1092          *      very slowly.
1093          */
1094         if (vcpu->tsc_catchup) {
1095                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1096                 if (tsc > tsc_timestamp) {
1097                         kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1098                         tsc_timestamp = tsc;
1099                 }
1100         }
1101
1102         local_irq_restore(flags);
1103
1104         if (!vcpu->time_page)
1105                 return 0;
1106
1107         /*
1108          * Time as measured by the TSC may go backwards when resetting the base
1109          * tsc_timestamp.  The reason for this is that the TSC resolution is
1110          * higher than the resolution of the other clock scales.  Thus, many
1111          * possible measurments of the TSC correspond to one measurement of any
1112          * other clock, and so a spread of values is possible.  This is not a
1113          * problem for the computation of the nanosecond clock; with TSC rates
1114          * around 1GHZ, there can only be a few cycles which correspond to one
1115          * nanosecond value, and any path through this code will inevitably
1116          * take longer than that.  However, with the kernel_ns value itself,
1117          * the precision may be much lower, down to HZ granularity.  If the
1118          * first sampling of TSC against kernel_ns ends in the low part of the
1119          * range, and the second in the high end of the range, we can get:
1120          *
1121          * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1122          *
1123          * As the sampling errors potentially range in the thousands of cycles,
1124          * it is possible such a time value has already been observed by the
1125          * guest.  To protect against this, we must compute the system time as
1126          * observed by the guest and ensure the new system time is greater.
1127          */
1128         max_kernel_ns = 0;
1129         if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1130                 max_kernel_ns = vcpu->last_guest_tsc -
1131                                 vcpu->hv_clock.tsc_timestamp;
1132                 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1133                                     vcpu->hv_clock.tsc_to_system_mul,
1134                                     vcpu->hv_clock.tsc_shift);
1135                 max_kernel_ns += vcpu->last_kernel_ns;
1136         }
1137
1138         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1139                 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1140                                    &vcpu->hv_clock.tsc_shift,
1141                                    &vcpu->hv_clock.tsc_to_system_mul);
1142                 vcpu->hw_tsc_khz = this_tsc_khz;
1143         }
1144
1145         if (max_kernel_ns > kernel_ns)
1146                 kernel_ns = max_kernel_ns;
1147
1148         /* With all the info we got, fill in the values */
1149         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1150         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1151         vcpu->last_kernel_ns = kernel_ns;
1152         vcpu->last_guest_tsc = tsc_timestamp;
1153         vcpu->hv_clock.flags = 0;
1154
1155         /*
1156          * The interface expects us to write an even number signaling that the
1157          * update is finished. Since the guest won't see the intermediate
1158          * state, we just increase by 2 at the end.
1159          */
1160         vcpu->hv_clock.version += 2;
1161
1162         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1163
1164         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1165                sizeof(vcpu->hv_clock));
1166
1167         kunmap_atomic(shared_kaddr, KM_USER0);
1168
1169         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1170         return 0;
1171 }
1172
1173 static bool msr_mtrr_valid(unsigned msr)
1174 {
1175         switch (msr) {
1176         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1177         case MSR_MTRRfix64K_00000:
1178         case MSR_MTRRfix16K_80000:
1179         case MSR_MTRRfix16K_A0000:
1180         case MSR_MTRRfix4K_C0000:
1181         case MSR_MTRRfix4K_C8000:
1182         case MSR_MTRRfix4K_D0000:
1183         case MSR_MTRRfix4K_D8000:
1184         case MSR_MTRRfix4K_E0000:
1185         case MSR_MTRRfix4K_E8000:
1186         case MSR_MTRRfix4K_F0000:
1187         case MSR_MTRRfix4K_F8000:
1188         case MSR_MTRRdefType:
1189         case MSR_IA32_CR_PAT:
1190                 return true;
1191         case 0x2f8:
1192                 return true;
1193         }
1194         return false;
1195 }
1196
1197 static bool valid_pat_type(unsigned t)
1198 {
1199         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1200 }
1201
1202 static bool valid_mtrr_type(unsigned t)
1203 {
1204         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1205 }
1206
1207 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1208 {
1209         int i;
1210
1211         if (!msr_mtrr_valid(msr))
1212                 return false;
1213
1214         if (msr == MSR_IA32_CR_PAT) {
1215                 for (i = 0; i < 8; i++)
1216                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
1217                                 return false;
1218                 return true;
1219         } else if (msr == MSR_MTRRdefType) {
1220                 if (data & ~0xcff)
1221                         return false;
1222                 return valid_mtrr_type(data & 0xff);
1223         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1224                 for (i = 0; i < 8 ; i++)
1225                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1226                                 return false;
1227                 return true;
1228         }
1229
1230         /* variable MTRRs */
1231         return valid_mtrr_type(data & 0xff);
1232 }
1233
1234 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1235 {
1236         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1237
1238         if (!mtrr_valid(vcpu, msr, data))
1239                 return 1;
1240
1241         if (msr == MSR_MTRRdefType) {
1242                 vcpu->arch.mtrr_state.def_type = data;
1243                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1244         } else if (msr == MSR_MTRRfix64K_00000)
1245                 p[0] = data;
1246         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1247                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1248         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1249                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1250         else if (msr == MSR_IA32_CR_PAT)
1251                 vcpu->arch.pat = data;
1252         else {  /* Variable MTRRs */
1253                 int idx, is_mtrr_mask;
1254                 u64 *pt;
1255
1256                 idx = (msr - 0x200) / 2;
1257                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1258                 if (!is_mtrr_mask)
1259                         pt =
1260                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1261                 else
1262                         pt =
1263                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1264                 *pt = data;
1265         }
1266
1267         kvm_mmu_reset_context(vcpu);
1268         return 0;
1269 }
1270
1271 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1272 {
1273         u64 mcg_cap = vcpu->arch.mcg_cap;
1274         unsigned bank_num = mcg_cap & 0xff;
1275
1276         switch (msr) {
1277         case MSR_IA32_MCG_STATUS:
1278                 vcpu->arch.mcg_status = data;
1279                 break;
1280         case MSR_IA32_MCG_CTL:
1281                 if (!(mcg_cap & MCG_CTL_P))
1282                         return 1;
1283                 if (data != 0 && data != ~(u64)0)
1284                         return -1;
1285                 vcpu->arch.mcg_ctl = data;
1286                 break;
1287         default:
1288                 if (msr >= MSR_IA32_MC0_CTL &&
1289                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1290                         u32 offset = msr - MSR_IA32_MC0_CTL;
1291                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1292                          * some Linux kernels though clear bit 10 in bank 4 to
1293                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1294                          * this to avoid an uncatched #GP in the guest
1295                          */
1296                         if ((offset & 0x3) == 0 &&
1297                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1298                                 return -1;
1299                         vcpu->arch.mce_banks[offset] = data;
1300                         break;
1301                 }
1302                 return 1;
1303         }
1304         return 0;
1305 }
1306
1307 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1308 {
1309         struct kvm *kvm = vcpu->kvm;
1310         int lm = is_long_mode(vcpu);
1311         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1312                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1313         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1314                 : kvm->arch.xen_hvm_config.blob_size_32;
1315         u32 page_num = data & ~PAGE_MASK;
1316         u64 page_addr = data & PAGE_MASK;
1317         u8 *page;
1318         int r;
1319
1320         r = -E2BIG;
1321         if (page_num >= blob_size)
1322                 goto out;
1323         r = -ENOMEM;
1324         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1325         if (!page)
1326                 goto out;
1327         r = -EFAULT;
1328         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1329                 goto out_free;
1330         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1331                 goto out_free;
1332         r = 0;
1333 out_free:
1334         kfree(page);
1335 out:
1336         return r;
1337 }
1338
1339 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1340 {
1341         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1342 }
1343
1344 static bool kvm_hv_msr_partition_wide(u32 msr)
1345 {
1346         bool r = false;
1347         switch (msr) {
1348         case HV_X64_MSR_GUEST_OS_ID:
1349         case HV_X64_MSR_HYPERCALL:
1350                 r = true;
1351                 break;
1352         }
1353
1354         return r;
1355 }
1356
1357 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1358 {
1359         struct kvm *kvm = vcpu->kvm;
1360
1361         switch (msr) {
1362         case HV_X64_MSR_GUEST_OS_ID:
1363                 kvm->arch.hv_guest_os_id = data;
1364                 /* setting guest os id to zero disables hypercall page */
1365                 if (!kvm->arch.hv_guest_os_id)
1366                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1367                 break;
1368         case HV_X64_MSR_HYPERCALL: {
1369                 u64 gfn;
1370                 unsigned long addr;
1371                 u8 instructions[4];
1372
1373                 /* if guest os id is not set hypercall should remain disabled */
1374                 if (!kvm->arch.hv_guest_os_id)
1375                         break;
1376                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1377                         kvm->arch.hv_hypercall = data;
1378                         break;
1379                 }
1380                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1381                 addr = gfn_to_hva(kvm, gfn);
1382                 if (kvm_is_error_hva(addr))
1383                         return 1;
1384                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1385                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1386                 if (copy_to_user((void __user *)addr, instructions, 4))
1387                         return 1;
1388                 kvm->arch.hv_hypercall = data;
1389                 break;
1390         }
1391         default:
1392                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1393                           "data 0x%llx\n", msr, data);
1394                 return 1;
1395         }
1396         return 0;
1397 }
1398
1399 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1400 {
1401         switch (msr) {
1402         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1403                 unsigned long addr;
1404
1405                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1406                         vcpu->arch.hv_vapic = data;
1407                         break;
1408                 }
1409                 addr = gfn_to_hva(vcpu->kvm, data >>
1410                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1411                 if (kvm_is_error_hva(addr))
1412                         return 1;
1413                 if (clear_user((void __user *)addr, PAGE_SIZE))
1414                         return 1;
1415                 vcpu->arch.hv_vapic = data;
1416                 break;
1417         }
1418         case HV_X64_MSR_EOI:
1419                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1420         case HV_X64_MSR_ICR:
1421                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1422         case HV_X64_MSR_TPR:
1423                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1424         default:
1425                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1426                           "data 0x%llx\n", msr, data);
1427                 return 1;
1428         }
1429
1430         return 0;
1431 }
1432
1433 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1434 {
1435         gpa_t gpa = data & ~0x3f;
1436
1437         /* Bits 2:5 are resrved, Should be zero */
1438         if (data & 0x3c)
1439                 return 1;
1440
1441         vcpu->arch.apf.msr_val = data;
1442
1443         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1444                 kvm_clear_async_pf_completion_queue(vcpu);
1445                 kvm_async_pf_hash_reset(vcpu);
1446                 return 0;
1447         }
1448
1449         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1450                 return 1;
1451
1452         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1453         kvm_async_pf_wakeup_all(vcpu);
1454         return 0;
1455 }
1456
1457 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1458 {
1459         if (vcpu->arch.time_page) {
1460                 kvm_release_page_dirty(vcpu->arch.time_page);
1461                 vcpu->arch.time_page = NULL;
1462         }
1463 }
1464
1465 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1466 {
1467         switch (msr) {
1468         case MSR_EFER:
1469                 return set_efer(vcpu, data);
1470         case MSR_K7_HWCR:
1471                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1472                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1473                 if (data != 0) {
1474                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1475                                 data);
1476                         return 1;
1477                 }
1478                 break;
1479         case MSR_FAM10H_MMIO_CONF_BASE:
1480                 if (data != 0) {
1481                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1482                                 "0x%llx\n", data);
1483                         return 1;
1484                 }
1485                 break;
1486         case MSR_AMD64_NB_CFG:
1487                 break;
1488         case MSR_IA32_DEBUGCTLMSR:
1489                 if (!data) {
1490                         /* We support the non-activated case already */
1491                         break;
1492                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1493                         /* Values other than LBR and BTF are vendor-specific,
1494                            thus reserved and should throw a #GP */
1495                         return 1;
1496                 }
1497                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1498                         __func__, data);
1499                 break;
1500         case MSR_IA32_UCODE_REV:
1501         case MSR_IA32_UCODE_WRITE:
1502         case MSR_VM_HSAVE_PA:
1503         case MSR_AMD64_PATCH_LOADER:
1504                 break;
1505         case 0x200 ... 0x2ff:
1506                 return set_msr_mtrr(vcpu, msr, data);
1507         case MSR_IA32_APICBASE:
1508                 kvm_set_apic_base(vcpu, data);
1509                 break;
1510         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1511                 return kvm_x2apic_msr_write(vcpu, msr, data);
1512         case MSR_IA32_MISC_ENABLE:
1513                 vcpu->arch.ia32_misc_enable_msr = data;
1514                 break;
1515         case MSR_KVM_WALL_CLOCK_NEW:
1516         case MSR_KVM_WALL_CLOCK:
1517                 vcpu->kvm->arch.wall_clock = data;
1518                 kvm_write_wall_clock(vcpu->kvm, data);
1519                 break;
1520         case MSR_KVM_SYSTEM_TIME_NEW:
1521         case MSR_KVM_SYSTEM_TIME: {
1522                 kvmclock_reset(vcpu);
1523
1524                 vcpu->arch.time = data;
1525                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1526
1527                 /* we verify if the enable bit is set... */
1528                 if (!(data & 1))
1529                         break;
1530
1531                 /* ...but clean it before doing the actual write */
1532                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1533
1534                 vcpu->arch.time_page =
1535                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1536
1537                 if (is_error_page(vcpu->arch.time_page)) {
1538                         kvm_release_page_clean(vcpu->arch.time_page);
1539                         vcpu->arch.time_page = NULL;
1540                 }
1541                 break;
1542         }
1543         case MSR_KVM_ASYNC_PF_EN:
1544                 if (kvm_pv_enable_async_pf(vcpu, data))
1545                         return 1;
1546                 break;
1547         case MSR_IA32_MCG_CTL:
1548         case MSR_IA32_MCG_STATUS:
1549         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1550                 return set_msr_mce(vcpu, msr, data);
1551
1552         /* Performance counters are not protected by a CPUID bit,
1553          * so we should check all of them in the generic path for the sake of
1554          * cross vendor migration.
1555          * Writing a zero into the event select MSRs disables them,
1556          * which we perfectly emulate ;-). Any other value should be at least
1557          * reported, some guests depend on them.
1558          */
1559         case MSR_P6_EVNTSEL0:
1560         case MSR_P6_EVNTSEL1:
1561         case MSR_K7_EVNTSEL0:
1562         case MSR_K7_EVNTSEL1:
1563         case MSR_K7_EVNTSEL2:
1564         case MSR_K7_EVNTSEL3:
1565                 if (data != 0)
1566                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1567                                 "0x%x data 0x%llx\n", msr, data);
1568                 break;
1569         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1570          * so we ignore writes to make it happy.
1571          */
1572         case MSR_P6_PERFCTR0:
1573         case MSR_P6_PERFCTR1:
1574         case MSR_K7_PERFCTR0:
1575         case MSR_K7_PERFCTR1:
1576         case MSR_K7_PERFCTR2:
1577         case MSR_K7_PERFCTR3:
1578                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1579                         "0x%x data 0x%llx\n", msr, data);
1580                 break;
1581         case MSR_K7_CLK_CTL:
1582                 /*
1583                  * Ignore all writes to this no longer documented MSR.
1584                  * Writes are only relevant for old K7 processors,
1585                  * all pre-dating SVM, but a recommended workaround from
1586                  * AMD for these chips. It is possible to speicify the
1587                  * affected processor models on the command line, hence
1588                  * the need to ignore the workaround.
1589                  */
1590                 break;
1591         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1592                 if (kvm_hv_msr_partition_wide(msr)) {
1593                         int r;
1594                         mutex_lock(&vcpu->kvm->lock);
1595                         r = set_msr_hyperv_pw(vcpu, msr, data);
1596                         mutex_unlock(&vcpu->kvm->lock);
1597                         return r;
1598                 } else
1599                         return set_msr_hyperv(vcpu, msr, data);
1600                 break;
1601         case MSR_IA32_BBL_CR_CTL3:
1602                 /* Drop writes to this legacy MSR -- see rdmsr
1603                  * counterpart for further detail.
1604                  */
1605                 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1606                 break;
1607         default:
1608                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1609                         return xen_hvm_config(vcpu, data);
1610                 if (!ignore_msrs) {
1611                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1612                                 msr, data);
1613                         return 1;
1614                 } else {
1615                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1616                                 msr, data);
1617                         break;
1618                 }
1619         }
1620         return 0;
1621 }
1622 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1623
1624
1625 /*
1626  * Reads an msr value (of 'msr_index') into 'pdata'.
1627  * Returns 0 on success, non-0 otherwise.
1628  * Assumes vcpu_load() was already called.
1629  */
1630 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1631 {
1632         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1633 }
1634
1635 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1636 {
1637         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1638
1639         if (!msr_mtrr_valid(msr))
1640                 return 1;
1641
1642         if (msr == MSR_MTRRdefType)
1643                 *pdata = vcpu->arch.mtrr_state.def_type +
1644                          (vcpu->arch.mtrr_state.enabled << 10);
1645         else if (msr == MSR_MTRRfix64K_00000)
1646                 *pdata = p[0];
1647         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1648                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1649         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1650                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1651         else if (msr == MSR_IA32_CR_PAT)
1652                 *pdata = vcpu->arch.pat;
1653         else {  /* Variable MTRRs */
1654                 int idx, is_mtrr_mask;
1655                 u64 *pt;
1656
1657                 idx = (msr - 0x200) / 2;
1658                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1659                 if (!is_mtrr_mask)
1660                         pt =
1661                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1662                 else
1663                         pt =
1664                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1665                 *pdata = *pt;
1666         }
1667
1668         return 0;
1669 }
1670
1671 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1672 {
1673         u64 data;
1674         u64 mcg_cap = vcpu->arch.mcg_cap;
1675         unsigned bank_num = mcg_cap & 0xff;
1676
1677         switch (msr) {
1678         case MSR_IA32_P5_MC_ADDR:
1679         case MSR_IA32_P5_MC_TYPE:
1680                 data = 0;
1681                 break;
1682         case MSR_IA32_MCG_CAP:
1683                 data = vcpu->arch.mcg_cap;
1684                 break;
1685         case MSR_IA32_MCG_CTL:
1686                 if (!(mcg_cap & MCG_CTL_P))
1687                         return 1;
1688                 data = vcpu->arch.mcg_ctl;
1689                 break;
1690         case MSR_IA32_MCG_STATUS:
1691                 data = vcpu->arch.mcg_status;
1692                 break;
1693         default:
1694                 if (msr >= MSR_IA32_MC0_CTL &&
1695                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1696                         u32 offset = msr - MSR_IA32_MC0_CTL;
1697                         data = vcpu->arch.mce_banks[offset];
1698                         break;
1699                 }
1700                 return 1;
1701         }
1702         *pdata = data;
1703         return 0;
1704 }
1705
1706 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1707 {
1708         u64 data = 0;
1709         struct kvm *kvm = vcpu->kvm;
1710
1711         switch (msr) {
1712         case HV_X64_MSR_GUEST_OS_ID:
1713                 data = kvm->arch.hv_guest_os_id;
1714                 break;
1715         case HV_X64_MSR_HYPERCALL:
1716                 data = kvm->arch.hv_hypercall;
1717                 break;
1718         default:
1719                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1720                 return 1;
1721         }
1722
1723         *pdata = data;
1724         return 0;
1725 }
1726
1727 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1728 {
1729         u64 data = 0;
1730
1731         switch (msr) {
1732         case HV_X64_MSR_VP_INDEX: {
1733                 int r;
1734                 struct kvm_vcpu *v;
1735                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1736                         if (v == vcpu)
1737                                 data = r;
1738                 break;
1739         }
1740         case HV_X64_MSR_EOI:
1741                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1742         case HV_X64_MSR_ICR:
1743                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1744         case HV_X64_MSR_TPR:
1745                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1746         default:
1747                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1748                 return 1;
1749         }
1750         *pdata = data;
1751         return 0;
1752 }
1753
1754 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1755 {
1756         u64 data;
1757
1758         switch (msr) {
1759         case MSR_IA32_PLATFORM_ID:
1760         case MSR_IA32_UCODE_REV:
1761         case MSR_IA32_EBL_CR_POWERON:
1762         case MSR_IA32_DEBUGCTLMSR:
1763         case MSR_IA32_LASTBRANCHFROMIP:
1764         case MSR_IA32_LASTBRANCHTOIP:
1765         case MSR_IA32_LASTINTFROMIP:
1766         case MSR_IA32_LASTINTTOIP:
1767         case MSR_K8_SYSCFG:
1768         case MSR_K7_HWCR:
1769         case MSR_VM_HSAVE_PA:
1770         case MSR_P6_PERFCTR0:
1771         case MSR_P6_PERFCTR1:
1772         case MSR_P6_EVNTSEL0:
1773         case MSR_P6_EVNTSEL1:
1774         case MSR_K7_EVNTSEL0:
1775         case MSR_K7_PERFCTR0:
1776         case MSR_K8_INT_PENDING_MSG:
1777         case MSR_AMD64_NB_CFG:
1778         case MSR_FAM10H_MMIO_CONF_BASE:
1779                 data = 0;
1780                 break;
1781         case MSR_MTRRcap:
1782                 data = 0x500 | KVM_NR_VAR_MTRR;
1783                 break;
1784         case 0x200 ... 0x2ff:
1785                 return get_msr_mtrr(vcpu, msr, pdata);
1786         case 0xcd: /* fsb frequency */
1787                 data = 3;
1788                 break;
1789                 /*
1790                  * MSR_EBC_FREQUENCY_ID
1791                  * Conservative value valid for even the basic CPU models.
1792                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1793                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1794                  * and 266MHz for model 3, or 4. Set Core Clock
1795                  * Frequency to System Bus Frequency Ratio to 1 (bits
1796                  * 31:24) even though these are only valid for CPU
1797                  * models > 2, however guests may end up dividing or
1798                  * multiplying by zero otherwise.
1799                  */
1800         case MSR_EBC_FREQUENCY_ID:
1801                 data = 1 << 24;
1802                 break;
1803         case MSR_IA32_APICBASE:
1804                 data = kvm_get_apic_base(vcpu);
1805                 break;
1806         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1807                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1808                 break;
1809         case MSR_IA32_MISC_ENABLE:
1810                 data = vcpu->arch.ia32_misc_enable_msr;
1811                 break;
1812         case MSR_IA32_PERF_STATUS:
1813                 /* TSC increment by tick */
1814                 data = 1000ULL;
1815                 /* CPU multiplier */
1816                 data |= (((uint64_t)4ULL) << 40);
1817                 break;
1818         case MSR_EFER:
1819                 data = vcpu->arch.efer;
1820                 break;
1821         case MSR_KVM_WALL_CLOCK:
1822         case MSR_KVM_WALL_CLOCK_NEW:
1823                 data = vcpu->kvm->arch.wall_clock;
1824                 break;
1825         case MSR_KVM_SYSTEM_TIME:
1826         case MSR_KVM_SYSTEM_TIME_NEW:
1827                 data = vcpu->arch.time;
1828                 break;
1829         case MSR_KVM_ASYNC_PF_EN:
1830                 data = vcpu->arch.apf.msr_val;
1831                 break;
1832         case MSR_IA32_P5_MC_ADDR:
1833         case MSR_IA32_P5_MC_TYPE:
1834         case MSR_IA32_MCG_CAP:
1835         case MSR_IA32_MCG_CTL:
1836         case MSR_IA32_MCG_STATUS:
1837         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1838                 return get_msr_mce(vcpu, msr, pdata);
1839         case MSR_K7_CLK_CTL:
1840                 /*
1841                  * Provide expected ramp-up count for K7. All other
1842                  * are set to zero, indicating minimum divisors for
1843                  * every field.
1844                  *
1845                  * This prevents guest kernels on AMD host with CPU
1846                  * type 6, model 8 and higher from exploding due to
1847                  * the rdmsr failing.
1848                  */
1849                 data = 0x20000000;
1850                 break;
1851         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1852                 if (kvm_hv_msr_partition_wide(msr)) {
1853                         int r;
1854                         mutex_lock(&vcpu->kvm->lock);
1855                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1856                         mutex_unlock(&vcpu->kvm->lock);
1857                         return r;
1858                 } else
1859                         return get_msr_hyperv(vcpu, msr, pdata);
1860                 break;
1861         case MSR_IA32_BBL_CR_CTL3:
1862                 /* This legacy MSR exists but isn't fully documented in current
1863                  * silicon.  It is however accessed by winxp in very narrow
1864                  * scenarios where it sets bit #19, itself documented as
1865                  * a "reserved" bit.  Best effort attempt to source coherent
1866                  * read data here should the balance of the register be
1867                  * interpreted by the guest:
1868                  *
1869                  * L2 cache control register 3: 64GB range, 256KB size,
1870                  * enabled, latency 0x1, configured
1871                  */
1872                 data = 0xbe702111;
1873                 break;
1874         default:
1875                 if (!ignore_msrs) {
1876                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1877                         return 1;
1878                 } else {
1879                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1880                         data = 0;
1881                 }
1882                 break;
1883         }
1884         *pdata = data;
1885         return 0;
1886 }
1887 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1888
1889 /*
1890  * Read or write a bunch of msrs. All parameters are kernel addresses.
1891  *
1892  * @return number of msrs set successfully.
1893  */
1894 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1895                     struct kvm_msr_entry *entries,
1896                     int (*do_msr)(struct kvm_vcpu *vcpu,
1897                                   unsigned index, u64 *data))
1898 {
1899         int i, idx;
1900
1901         idx = srcu_read_lock(&vcpu->kvm->srcu);
1902         for (i = 0; i < msrs->nmsrs; ++i)
1903                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1904                         break;
1905         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1906
1907         return i;
1908 }
1909
1910 /*
1911  * Read or write a bunch of msrs. Parameters are user addresses.
1912  *
1913  * @return number of msrs set successfully.
1914  */
1915 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1916                   int (*do_msr)(struct kvm_vcpu *vcpu,
1917                                 unsigned index, u64 *data),
1918                   int writeback)
1919 {
1920         struct kvm_msrs msrs;
1921         struct kvm_msr_entry *entries;
1922         int r, n;
1923         unsigned size;
1924
1925         r = -EFAULT;
1926         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1927                 goto out;
1928
1929         r = -E2BIG;
1930         if (msrs.nmsrs >= MAX_IO_MSRS)
1931                 goto out;
1932
1933         r = -ENOMEM;
1934         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1935         entries = kmalloc(size, GFP_KERNEL);
1936         if (!entries)
1937                 goto out;
1938
1939         r = -EFAULT;
1940         if (copy_from_user(entries, user_msrs->entries, size))
1941                 goto out_free;
1942
1943         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1944         if (r < 0)
1945                 goto out_free;
1946
1947         r = -EFAULT;
1948         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1949                 goto out_free;
1950
1951         r = n;
1952
1953 out_free:
1954         kfree(entries);
1955 out:
1956         return r;
1957 }
1958
1959 int kvm_dev_ioctl_check_extension(long ext)
1960 {
1961         int r;
1962
1963         switch (ext) {
1964         case KVM_CAP_IRQCHIP:
1965         case KVM_CAP_HLT:
1966         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1967         case KVM_CAP_SET_TSS_ADDR:
1968         case KVM_CAP_EXT_CPUID:
1969         case KVM_CAP_CLOCKSOURCE:
1970         case KVM_CAP_PIT:
1971         case KVM_CAP_NOP_IO_DELAY:
1972         case KVM_CAP_MP_STATE:
1973         case KVM_CAP_SYNC_MMU:
1974         case KVM_CAP_USER_NMI:
1975         case KVM_CAP_REINJECT_CONTROL:
1976         case KVM_CAP_IRQ_INJECT_STATUS:
1977         case KVM_CAP_ASSIGN_DEV_IRQ:
1978         case KVM_CAP_IRQFD:
1979         case KVM_CAP_IOEVENTFD:
1980         case KVM_CAP_PIT2:
1981         case KVM_CAP_PIT_STATE2:
1982         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1983         case KVM_CAP_XEN_HVM:
1984         case KVM_CAP_ADJUST_CLOCK:
1985         case KVM_CAP_VCPU_EVENTS:
1986         case KVM_CAP_HYPERV:
1987         case KVM_CAP_HYPERV_VAPIC:
1988         case KVM_CAP_HYPERV_SPIN:
1989         case KVM_CAP_PCI_SEGMENT:
1990         case KVM_CAP_DEBUGREGS:
1991         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1992         case KVM_CAP_XSAVE:
1993         case KVM_CAP_ASYNC_PF:
1994         case KVM_CAP_GET_TSC_KHZ:
1995                 r = 1;
1996                 break;
1997         case KVM_CAP_COALESCED_MMIO:
1998                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1999                 break;
2000         case KVM_CAP_VAPIC:
2001                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2002                 break;
2003         case KVM_CAP_NR_VCPUS:
2004                 r = KVM_MAX_VCPUS;
2005                 break;
2006         case KVM_CAP_NR_MEMSLOTS:
2007                 r = KVM_MEMORY_SLOTS;
2008                 break;
2009         case KVM_CAP_PV_MMU:    /* obsolete */
2010                 r = 0;
2011                 break;
2012         case KVM_CAP_IOMMU:
2013                 r = iommu_found();
2014                 break;
2015         case KVM_CAP_MCE:
2016                 r = KVM_MAX_MCE_BANKS;
2017                 break;
2018         case KVM_CAP_XCRS:
2019                 r = cpu_has_xsave;
2020                 break;
2021         case KVM_CAP_TSC_CONTROL:
2022                 r = kvm_has_tsc_control;
2023                 break;
2024         default:
2025                 r = 0;
2026                 break;
2027         }
2028         return r;
2029
2030 }
2031
2032 long kvm_arch_dev_ioctl(struct file *filp,
2033                         unsigned int ioctl, unsigned long arg)
2034 {
2035         void __user *argp = (void __user *)arg;
2036         long r;
2037
2038         switch (ioctl) {
2039         case KVM_GET_MSR_INDEX_LIST: {
2040                 struct kvm_msr_list __user *user_msr_list = argp;
2041                 struct kvm_msr_list msr_list;
2042                 unsigned n;
2043
2044                 r = -EFAULT;
2045                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2046                         goto out;
2047                 n = msr_list.nmsrs;
2048                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2049                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2050                         goto out;
2051                 r = -E2BIG;
2052                 if (n < msr_list.nmsrs)
2053                         goto out;
2054                 r = -EFAULT;
2055                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2056                                  num_msrs_to_save * sizeof(u32)))
2057                         goto out;
2058                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2059                                  &emulated_msrs,
2060                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2061                         goto out;
2062                 r = 0;
2063                 break;
2064         }
2065         case KVM_GET_SUPPORTED_CPUID: {
2066                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2067                 struct kvm_cpuid2 cpuid;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2071                         goto out;
2072                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2073                                                       cpuid_arg->entries);
2074                 if (r)
2075                         goto out;
2076
2077                 r = -EFAULT;
2078                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2079                         goto out;
2080                 r = 0;
2081                 break;
2082         }
2083         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2084                 u64 mce_cap;
2085
2086                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2087                 r = -EFAULT;
2088                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2089                         goto out;
2090                 r = 0;
2091                 break;
2092         }
2093         default:
2094                 r = -EINVAL;
2095         }
2096 out:
2097         return r;
2098 }
2099
2100 static void wbinvd_ipi(void *garbage)
2101 {
2102         wbinvd();
2103 }
2104
2105 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2106 {
2107         return vcpu->kvm->arch.iommu_domain &&
2108                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2109 }
2110
2111 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2112 {
2113         /* Address WBINVD may be executed by guest */
2114         if (need_emulate_wbinvd(vcpu)) {
2115                 if (kvm_x86_ops->has_wbinvd_exit())
2116                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2117                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2118                         smp_call_function_single(vcpu->cpu,
2119                                         wbinvd_ipi, NULL, 1);
2120         }
2121
2122         kvm_x86_ops->vcpu_load(vcpu, cpu);
2123         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2124                 /* Make sure TSC doesn't go backwards */
2125                 s64 tsc_delta;
2126                 u64 tsc;
2127
2128                 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2129                 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2130                              tsc - vcpu->arch.last_guest_tsc;
2131
2132                 if (tsc_delta < 0)
2133                         mark_tsc_unstable("KVM discovered backwards TSC");
2134                 if (check_tsc_unstable()) {
2135                         kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2136                         vcpu->arch.tsc_catchup = 1;
2137                 }
2138                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2139                 if (vcpu->cpu != cpu)
2140                         kvm_migrate_timers(vcpu);
2141                 vcpu->cpu = cpu;
2142         }
2143 }
2144
2145 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2146 {
2147         kvm_x86_ops->vcpu_put(vcpu);
2148         kvm_put_guest_fpu(vcpu);
2149         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2150 }
2151
2152 static int is_efer_nx(void)
2153 {
2154         unsigned long long efer = 0;
2155
2156         rdmsrl_safe(MSR_EFER, &efer);
2157         return efer & EFER_NX;
2158 }
2159
2160 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2161 {
2162         int i;
2163         struct kvm_cpuid_entry2 *e, *entry;
2164
2165         entry = NULL;
2166         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2167                 e = &vcpu->arch.cpuid_entries[i];
2168                 if (e->function == 0x80000001) {
2169                         entry = e;
2170                         break;
2171                 }
2172         }
2173         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2174                 entry->edx &= ~(1 << 20);
2175                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2176         }
2177 }
2178
2179 /* when an old userspace process fills a new kernel module */
2180 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2181                                     struct kvm_cpuid *cpuid,
2182                                     struct kvm_cpuid_entry __user *entries)
2183 {
2184         int r, i;
2185         struct kvm_cpuid_entry *cpuid_entries;
2186
2187         r = -E2BIG;
2188         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2189                 goto out;
2190         r = -ENOMEM;
2191         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2192         if (!cpuid_entries)
2193                 goto out;
2194         r = -EFAULT;
2195         if (copy_from_user(cpuid_entries, entries,
2196                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2197                 goto out_free;
2198         for (i = 0; i < cpuid->nent; i++) {
2199                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2200                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2201                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2202                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2203                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2204                 vcpu->arch.cpuid_entries[i].index = 0;
2205                 vcpu->arch.cpuid_entries[i].flags = 0;
2206                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2207                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2208                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2209         }
2210         vcpu->arch.cpuid_nent = cpuid->nent;
2211         cpuid_fix_nx_cap(vcpu);
2212         r = 0;
2213         kvm_apic_set_version(vcpu);
2214         kvm_x86_ops->cpuid_update(vcpu);
2215         update_cpuid(vcpu);
2216
2217 out_free:
2218         vfree(cpuid_entries);
2219 out:
2220         return r;
2221 }
2222
2223 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2224                                      struct kvm_cpuid2 *cpuid,
2225                                      struct kvm_cpuid_entry2 __user *entries)
2226 {
2227         int r;
2228
2229         r = -E2BIG;
2230         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2231                 goto out;
2232         r = -EFAULT;
2233         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2234                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2235                 goto out;
2236         vcpu->arch.cpuid_nent = cpuid->nent;
2237         kvm_apic_set_version(vcpu);
2238         kvm_x86_ops->cpuid_update(vcpu);
2239         update_cpuid(vcpu);
2240         return 0;
2241
2242 out:
2243         return r;
2244 }
2245
2246 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2247                                      struct kvm_cpuid2 *cpuid,
2248                                      struct kvm_cpuid_entry2 __user *entries)
2249 {
2250         int r;
2251
2252         r = -E2BIG;
2253         if (cpuid->nent < vcpu->arch.cpuid_nent)
2254                 goto out;
2255         r = -EFAULT;
2256         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2257                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2258                 goto out;
2259         return 0;
2260
2261 out:
2262         cpuid->nent = vcpu->arch.cpuid_nent;
2263         return r;
2264 }
2265
2266 static void cpuid_mask(u32 *word, int wordnum)
2267 {
2268         *word &= boot_cpu_data.x86_capability[wordnum];
2269 }
2270
2271 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2272                            u32 index)
2273 {
2274         entry->function = function;
2275         entry->index = index;
2276         cpuid_count(entry->function, entry->index,
2277                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2278         entry->flags = 0;
2279 }
2280
2281 #define F(x) bit(X86_FEATURE_##x)
2282
2283 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2284                          u32 index, int *nent, int maxnent)
2285 {
2286         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2287 #ifdef CONFIG_X86_64
2288         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2289                                 ? F(GBPAGES) : 0;
2290         unsigned f_lm = F(LM);
2291 #else
2292         unsigned f_gbpages = 0;
2293         unsigned f_lm = 0;
2294 #endif
2295         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2296
2297         /* cpuid 1.edx */
2298         const u32 kvm_supported_word0_x86_features =
2299                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2300                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2301                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2302                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2303                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2304                 0 /* Reserved, DS, ACPI */ | F(MMX) |
2305                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2306                 0 /* HTT, TM, Reserved, PBE */;
2307         /* cpuid 0x80000001.edx */
2308         const u32 kvm_supported_word1_x86_features =
2309                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2310                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2311                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2312                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2313                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2314                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2315                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2316                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2317         /* cpuid 1.ecx */
2318         const u32 kvm_supported_word4_x86_features =
2319                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2320                 0 /* DS-CPL, VMX, SMX, EST */ |
2321                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2322                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2323                 0 /* Reserved, DCA */ | F(XMM4_1) |
2324                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2325                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2326                 F(F16C);
2327         /* cpuid 0x80000001.ecx */
2328         const u32 kvm_supported_word6_x86_features =
2329                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2330                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2331                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2332                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2333
2334         /* all calls to cpuid_count() should be made on the same cpu */
2335         get_cpu();
2336         do_cpuid_1_ent(entry, function, index);
2337         ++*nent;
2338
2339         switch (function) {
2340         case 0:
2341                 entry->eax = min(entry->eax, (u32)0xd);
2342                 break;
2343         case 1:
2344                 entry->edx &= kvm_supported_word0_x86_features;
2345                 cpuid_mask(&entry->edx, 0);
2346                 entry->ecx &= kvm_supported_word4_x86_features;
2347                 cpuid_mask(&entry->ecx, 4);
2348                 /* we support x2apic emulation even if host does not support
2349                  * it since we emulate x2apic in software */
2350                 entry->ecx |= F(X2APIC);
2351                 break;
2352         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2353          * may return different values. This forces us to get_cpu() before
2354          * issuing the first command, and also to emulate this annoying behavior
2355          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2356         case 2: {
2357                 int t, times = entry->eax & 0xff;
2358
2359                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2360                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2361                 for (t = 1; t < times && *nent < maxnent; ++t) {
2362                         do_cpuid_1_ent(&entry[t], function, 0);
2363                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2364                         ++*nent;
2365                 }
2366                 break;
2367         }
2368         /* function 4 and 0xb have additional index. */
2369         case 4: {
2370                 int i, cache_type;
2371
2372                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2373                 /* read more entries until cache_type is zero */
2374                 for (i = 1; *nent < maxnent; ++i) {
2375                         cache_type = entry[i - 1].eax & 0x1f;
2376                         if (!cache_type)
2377                                 break;
2378                         do_cpuid_1_ent(&entry[i], function, i);
2379                         entry[i].flags |=
2380                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2381                         ++*nent;
2382                 }
2383                 break;
2384         }
2385         case 0xb: {
2386                 int i, level_type;
2387
2388                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2389                 /* read more entries until level_type is zero */
2390                 for (i = 1; *nent < maxnent; ++i) {
2391                         level_type = entry[i - 1].ecx & 0xff00;
2392                         if (!level_type)
2393                                 break;
2394                         do_cpuid_1_ent(&entry[i], function, i);
2395                         entry[i].flags |=
2396                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2397                         ++*nent;
2398                 }
2399                 break;
2400         }
2401         case 0xd: {
2402                 int i;
2403
2404                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2405                 for (i = 1; *nent < maxnent && i < 64; ++i) {
2406                         if (entry[i].eax == 0)
2407                                 continue;
2408                         do_cpuid_1_ent(&entry[i], function, i);
2409                         entry[i].flags |=
2410                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2411                         ++*nent;
2412                 }
2413                 break;
2414         }
2415         case KVM_CPUID_SIGNATURE: {
2416                 char signature[12] = "KVMKVMKVM\0\0";
2417                 u32 *sigptr = (u32 *)signature;
2418                 entry->eax = 0;
2419                 entry->ebx = sigptr[0];
2420                 entry->ecx = sigptr[1];
2421                 entry->edx = sigptr[2];
2422                 break;
2423         }
2424         case KVM_CPUID_FEATURES:
2425                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2426                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2427                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2428                              (1 << KVM_FEATURE_ASYNC_PF) |
2429                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2430                 entry->ebx = 0;
2431                 entry->ecx = 0;
2432                 entry->edx = 0;
2433                 break;
2434         case 0x80000000:
2435                 entry->eax = min(entry->eax, 0x8000001a);
2436                 break;
2437         case 0x80000001:
2438                 entry->edx &= kvm_supported_word1_x86_features;
2439                 cpuid_mask(&entry->edx, 1);
2440                 entry->ecx &= kvm_supported_word6_x86_features;
2441                 cpuid_mask(&entry->ecx, 6);
2442                 break;
2443         }
2444
2445         kvm_x86_ops->set_supported_cpuid(function, entry);
2446
2447         put_cpu();
2448 }
2449
2450 #undef F
2451
2452 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2453                                      struct kvm_cpuid_entry2 __user *entries)
2454 {
2455         struct kvm_cpuid_entry2 *cpuid_entries;
2456         int limit, nent = 0, r = -E2BIG;
2457         u32 func;
2458
2459         if (cpuid->nent < 1)
2460                 goto out;
2461         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2462                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2463         r = -ENOMEM;
2464         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2465         if (!cpuid_entries)
2466                 goto out;
2467
2468         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2469         limit = cpuid_entries[0].eax;
2470         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2471                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2472                              &nent, cpuid->nent);
2473         r = -E2BIG;
2474         if (nent >= cpuid->nent)
2475                 goto out_free;
2476
2477         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2478         limit = cpuid_entries[nent - 1].eax;
2479         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2480                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2481                              &nent, cpuid->nent);
2482
2483
2484
2485         r = -E2BIG;
2486         if (nent >= cpuid->nent)
2487                 goto out_free;
2488
2489         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2490                      cpuid->nent);
2491
2492         r = -E2BIG;
2493         if (nent >= cpuid->nent)
2494                 goto out_free;
2495
2496         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2497                      cpuid->nent);
2498
2499         r = -E2BIG;
2500         if (nent >= cpuid->nent)
2501                 goto out_free;
2502
2503         r = -EFAULT;
2504         if (copy_to_user(entries, cpuid_entries,
2505                          nent * sizeof(struct kvm_cpuid_entry2)))
2506                 goto out_free;
2507         cpuid->nent = nent;
2508         r = 0;
2509
2510 out_free:
2511         vfree(cpuid_entries);
2512 out:
2513         return r;
2514 }
2515
2516 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2517                                     struct kvm_lapic_state *s)
2518 {
2519         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2520
2521         return 0;
2522 }
2523
2524 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2525                                     struct kvm_lapic_state *s)
2526 {
2527         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2528         kvm_apic_post_state_restore(vcpu);
2529         update_cr8_intercept(vcpu);
2530
2531         return 0;
2532 }
2533
2534 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2535                                     struct kvm_interrupt *irq)
2536 {
2537         if (irq->irq < 0 || irq->irq >= 256)
2538                 return -EINVAL;
2539         if (irqchip_in_kernel(vcpu->kvm))
2540                 return -ENXIO;
2541
2542         kvm_queue_interrupt(vcpu, irq->irq, false);
2543         kvm_make_request(KVM_REQ_EVENT, vcpu);
2544
2545         return 0;
2546 }
2547
2548 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2549 {
2550         kvm_inject_nmi(vcpu);
2551
2552         return 0;
2553 }
2554
2555 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2556                                            struct kvm_tpr_access_ctl *tac)
2557 {
2558         if (tac->flags)
2559                 return -EINVAL;
2560         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2561         return 0;
2562 }
2563
2564 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2565                                         u64 mcg_cap)
2566 {
2567         int r;
2568         unsigned bank_num = mcg_cap & 0xff, bank;
2569
2570         r = -EINVAL;
2571         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2572                 goto out;
2573         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2574                 goto out;
2575         r = 0;
2576         vcpu->arch.mcg_cap = mcg_cap;
2577         /* Init IA32_MCG_CTL to all 1s */
2578         if (mcg_cap & MCG_CTL_P)
2579                 vcpu->arch.mcg_ctl = ~(u64)0;
2580         /* Init IA32_MCi_CTL to all 1s */
2581         for (bank = 0; bank < bank_num; bank++)
2582                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2583 out:
2584         return r;
2585 }
2586
2587 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2588                                       struct kvm_x86_mce *mce)
2589 {
2590         u64 mcg_cap = vcpu->arch.mcg_cap;
2591         unsigned bank_num = mcg_cap & 0xff;
2592         u64 *banks = vcpu->arch.mce_banks;
2593
2594         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2595                 return -EINVAL;
2596         /*
2597          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2598          * reporting is disabled
2599          */
2600         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2601             vcpu->arch.mcg_ctl != ~(u64)0)
2602                 return 0;
2603         banks += 4 * mce->bank;
2604         /*
2605          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2606          * reporting is disabled for the bank
2607          */
2608         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2609                 return 0;
2610         if (mce->status & MCI_STATUS_UC) {
2611                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2612                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2613                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2614                         return 0;
2615                 }
2616                 if (banks[1] & MCI_STATUS_VAL)
2617                         mce->status |= MCI_STATUS_OVER;
2618                 banks[2] = mce->addr;
2619                 banks[3] = mce->misc;
2620                 vcpu->arch.mcg_status = mce->mcg_status;
2621                 banks[1] = mce->status;
2622                 kvm_queue_exception(vcpu, MC_VECTOR);
2623         } else if (!(banks[1] & MCI_STATUS_VAL)
2624                    || !(banks[1] & MCI_STATUS_UC)) {
2625                 if (banks[1] & MCI_STATUS_VAL)
2626                         mce->status |= MCI_STATUS_OVER;
2627                 banks[2] = mce->addr;
2628                 banks[3] = mce->misc;
2629                 banks[1] = mce->status;
2630         } else
2631                 banks[1] |= MCI_STATUS_OVER;
2632         return 0;
2633 }
2634
2635 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2636                                                struct kvm_vcpu_events *events)
2637 {
2638         events->exception.injected =
2639                 vcpu->arch.exception.pending &&
2640                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2641         events->exception.nr = vcpu->arch.exception.nr;
2642         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2643         events->exception.pad = 0;
2644         events->exception.error_code = vcpu->arch.exception.error_code;
2645
2646         events->interrupt.injected =
2647                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2648         events->interrupt.nr = vcpu->arch.interrupt.nr;
2649         events->interrupt.soft = 0;
2650         events->interrupt.shadow =
2651                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2652                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2653
2654         events->nmi.injected = vcpu->arch.nmi_injected;
2655         events->nmi.pending = vcpu->arch.nmi_pending;
2656         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2657         events->nmi.pad = 0;
2658
2659         events->sipi_vector = vcpu->arch.sipi_vector;
2660
2661         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2662                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2663                          | KVM_VCPUEVENT_VALID_SHADOW);
2664         memset(&events->reserved, 0, sizeof(events->reserved));
2665 }
2666
2667 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2668                                               struct kvm_vcpu_events *events)
2669 {
2670         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2671                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2672                               | KVM_VCPUEVENT_VALID_SHADOW))
2673                 return -EINVAL;
2674
2675         vcpu->arch.exception.pending = events->exception.injected;
2676         vcpu->arch.exception.nr = events->exception.nr;
2677         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2678         vcpu->arch.exception.error_code = events->exception.error_code;
2679
2680         vcpu->arch.interrupt.pending = events->interrupt.injected;
2681         vcpu->arch.interrupt.nr = events->interrupt.nr;
2682         vcpu->arch.interrupt.soft = events->interrupt.soft;
2683         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2684                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2685                                                   events->interrupt.shadow);
2686
2687         vcpu->arch.nmi_injected = events->nmi.injected;
2688         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2689                 vcpu->arch.nmi_pending = events->nmi.pending;
2690         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2691
2692         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2693                 vcpu->arch.sipi_vector = events->sipi_vector;
2694
2695         kvm_make_request(KVM_REQ_EVENT, vcpu);
2696
2697         return 0;
2698 }
2699
2700 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2701                                              struct kvm_debugregs *dbgregs)
2702 {
2703         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2704         dbgregs->dr6 = vcpu->arch.dr6;
2705         dbgregs->dr7 = vcpu->arch.dr7;
2706         dbgregs->flags = 0;
2707         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2708 }
2709
2710 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2711                                             struct kvm_debugregs *dbgregs)
2712 {
2713         if (dbgregs->flags)
2714                 return -EINVAL;
2715
2716         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2717         vcpu->arch.dr6 = dbgregs->dr6;
2718         vcpu->arch.dr7 = dbgregs->dr7;
2719
2720         return 0;
2721 }
2722
2723 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2724                                          struct kvm_xsave *guest_xsave)
2725 {
2726         if (cpu_has_xsave)
2727                 memcpy(guest_xsave->region,
2728                         &vcpu->arch.guest_fpu.state->xsave,
2729                         xstate_size);
2730         else {
2731                 memcpy(guest_xsave->region,
2732                         &vcpu->arch.guest_fpu.state->fxsave,
2733                         sizeof(struct i387_fxsave_struct));
2734                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2735                         XSTATE_FPSSE;
2736         }
2737 }
2738
2739 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2740                                         struct kvm_xsave *guest_xsave)
2741 {
2742         u64 xstate_bv =
2743                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2744
2745         if (cpu_has_xsave)
2746                 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2747                         guest_xsave->region, xstate_size);
2748         else {
2749                 if (xstate_bv & ~XSTATE_FPSSE)
2750                         return -EINVAL;
2751                 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2752                         guest_xsave->region, sizeof(struct i387_fxsave_struct));
2753         }
2754         return 0;
2755 }
2756
2757 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2758                                         struct kvm_xcrs *guest_xcrs)
2759 {
2760         if (!cpu_has_xsave) {
2761                 guest_xcrs->nr_xcrs = 0;
2762                 return;
2763         }
2764
2765         guest_xcrs->nr_xcrs = 1;
2766         guest_xcrs->flags = 0;
2767         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2768         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2769 }
2770
2771 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2772                                        struct kvm_xcrs *guest_xcrs)
2773 {
2774         int i, r = 0;
2775
2776         if (!cpu_has_xsave)
2777                 return -EINVAL;
2778
2779         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2780                 return -EINVAL;
2781
2782         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2783                 /* Only support XCR0 currently */
2784                 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2785                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2786                                 guest_xcrs->xcrs[0].value);
2787                         break;
2788                 }
2789         if (r)
2790                 r = -EINVAL;
2791         return r;
2792 }
2793
2794 long kvm_arch_vcpu_ioctl(struct file *filp,
2795                          unsigned int ioctl, unsigned long arg)
2796 {
2797         struct kvm_vcpu *vcpu = filp->private_data;
2798         void __user *argp = (void __user *)arg;
2799         int r;
2800         union {
2801                 struct kvm_lapic_state *lapic;
2802                 struct kvm_xsave *xsave;
2803                 struct kvm_xcrs *xcrs;
2804                 void *buffer;
2805         } u;
2806
2807         u.buffer = NULL;
2808         switch (ioctl) {
2809         case KVM_GET_LAPIC: {
2810                 r = -EINVAL;
2811                 if (!vcpu->arch.apic)
2812                         goto out;
2813                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2814
2815                 r = -ENOMEM;
2816                 if (!u.lapic)
2817                         goto out;
2818                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2819                 if (r)
2820                         goto out;
2821                 r = -EFAULT;
2822                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2823                         goto out;
2824                 r = 0;
2825                 break;
2826         }
2827         case KVM_SET_LAPIC: {
2828                 r = -EINVAL;
2829                 if (!vcpu->arch.apic)
2830                         goto out;
2831                 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2832                 r = -ENOMEM;
2833                 if (!u.lapic)
2834                         goto out;
2835                 r = -EFAULT;
2836                 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2837                         goto out;
2838                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2839                 if (r)
2840                         goto out;
2841                 r = 0;
2842                 break;
2843         }
2844         case KVM_INTERRUPT: {
2845                 struct kvm_interrupt irq;
2846
2847                 r = -EFAULT;
2848                 if (copy_from_user(&irq, argp, sizeof irq))
2849                         goto out;
2850                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2851                 if (r)
2852                         goto out;
2853                 r = 0;
2854                 break;
2855         }
2856         case KVM_NMI: {
2857                 r = kvm_vcpu_ioctl_nmi(vcpu);
2858                 if (r)
2859                         goto out;
2860                 r = 0;
2861                 break;
2862         }
2863         case KVM_SET_CPUID: {
2864                 struct kvm_cpuid __user *cpuid_arg = argp;
2865                 struct kvm_cpuid cpuid;
2866
2867                 r = -EFAULT;
2868                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2869                         goto out;
2870                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2871                 if (r)
2872                         goto out;
2873                 break;
2874         }
2875         case KVM_SET_CPUID2: {
2876                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2877                 struct kvm_cpuid2 cpuid;
2878
2879                 r = -EFAULT;
2880                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2881                         goto out;
2882                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2883                                               cpuid_arg->entries);
2884                 if (r)
2885                         goto out;
2886                 break;
2887         }
2888         case KVM_GET_CPUID2: {
2889                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2890                 struct kvm_cpuid2 cpuid;
2891
2892                 r = -EFAULT;
2893                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2894                         goto out;
2895                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2896                                               cpuid_arg->entries);
2897                 if (r)
2898                         goto out;
2899                 r = -EFAULT;
2900                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2901                         goto out;
2902                 r = 0;
2903                 break;
2904         }
2905         case KVM_GET_MSRS:
2906                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2907                 break;
2908         case KVM_SET_MSRS:
2909                 r = msr_io(vcpu, argp, do_set_msr, 0);
2910                 break;
2911         case KVM_TPR_ACCESS_REPORTING: {
2912                 struct kvm_tpr_access_ctl tac;
2913
2914                 r = -EFAULT;
2915                 if (copy_from_user(&tac, argp, sizeof tac))
2916                         goto out;
2917                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2918                 if (r)
2919                         goto out;
2920                 r = -EFAULT;
2921                 if (copy_to_user(argp, &tac, sizeof tac))
2922                         goto out;
2923                 r = 0;
2924                 break;
2925         };
2926         case KVM_SET_VAPIC_ADDR: {
2927                 struct kvm_vapic_addr va;
2928
2929                 r = -EINVAL;
2930                 if (!irqchip_in_kernel(vcpu->kvm))
2931                         goto out;
2932                 r = -EFAULT;
2933                 if (copy_from_user(&va, argp, sizeof va))
2934                         goto out;
2935                 r = 0;
2936                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2937                 break;
2938         }
2939         case KVM_X86_SETUP_MCE: {
2940                 u64 mcg_cap;
2941
2942                 r = -EFAULT;
2943                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2944                         goto out;
2945                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2946                 break;
2947         }
2948         case KVM_X86_SET_MCE: {
2949                 struct kvm_x86_mce mce;
2950
2951                 r = -EFAULT;
2952                 if (copy_from_user(&mce, argp, sizeof mce))
2953                         goto out;
2954                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2955                 break;
2956         }
2957         case KVM_GET_VCPU_EVENTS: {
2958                 struct kvm_vcpu_events events;
2959
2960                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2961
2962                 r = -EFAULT;
2963                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2964                         break;
2965                 r = 0;
2966                 break;
2967         }
2968         case KVM_SET_VCPU_EVENTS: {
2969                 struct kvm_vcpu_events events;
2970
2971                 r = -EFAULT;
2972                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2973                         break;
2974
2975                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2976                 break;
2977         }
2978         case KVM_GET_DEBUGREGS: {
2979                 struct kvm_debugregs dbgregs;
2980
2981                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2982
2983                 r = -EFAULT;
2984                 if (copy_to_user(argp, &dbgregs,
2985                                  sizeof(struct kvm_debugregs)))
2986                         break;
2987                 r = 0;
2988                 break;
2989         }
2990         case KVM_SET_DEBUGREGS: {
2991                 struct kvm_debugregs dbgregs;
2992
2993                 r = -EFAULT;
2994                 if (copy_from_user(&dbgregs, argp,
2995                                    sizeof(struct kvm_debugregs)))
2996                         break;
2997
2998                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2999                 break;
3000         }
3001         case KVM_GET_XSAVE: {
3002                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3003                 r = -ENOMEM;
3004                 if (!u.xsave)
3005                         break;
3006
3007                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3008
3009                 r = -EFAULT;
3010                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3011                         break;
3012                 r = 0;
3013                 break;
3014         }
3015         case KVM_SET_XSAVE: {
3016                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3017                 r = -ENOMEM;
3018                 if (!u.xsave)
3019                         break;
3020
3021                 r = -EFAULT;
3022                 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3023                         break;
3024
3025                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3026                 break;
3027         }
3028         case KVM_GET_XCRS: {
3029                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3030                 r = -ENOMEM;
3031                 if (!u.xcrs)
3032                         break;
3033
3034                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3035
3036                 r = -EFAULT;
3037                 if (copy_to_user(argp, u.xcrs,
3038                                  sizeof(struct kvm_xcrs)))
3039                         break;
3040                 r = 0;
3041                 break;
3042         }
3043         case KVM_SET_XCRS: {
3044                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3045                 r = -ENOMEM;
3046                 if (!u.xcrs)
3047                         break;
3048
3049                 r = -EFAULT;
3050                 if (copy_from_user(u.xcrs, argp,
3051                                    sizeof(struct kvm_xcrs)))
3052                         break;
3053
3054                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3055                 break;
3056         }
3057         case KVM_SET_TSC_KHZ: {
3058                 u32 user_tsc_khz;
3059
3060                 r = -EINVAL;
3061                 if (!kvm_has_tsc_control)
3062                         break;
3063
3064                 user_tsc_khz = (u32)arg;
3065
3066                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3067                         goto out;
3068
3069                 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3070
3071                 r = 0;
3072                 goto out;
3073         }
3074         case KVM_GET_TSC_KHZ: {
3075                 r = -EIO;
3076                 if (check_tsc_unstable())
3077                         goto out;
3078
3079                 r = vcpu_tsc_khz(vcpu);
3080
3081                 goto out;
3082         }
3083         default:
3084                 r = -EINVAL;
3085         }
3086 out:
3087         kfree(u.buffer);
3088         return r;
3089 }
3090
3091 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3092 {
3093         int ret;
3094
3095         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3096                 return -1;
3097         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3098         return ret;
3099 }
3100
3101 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3102                                               u64 ident_addr)
3103 {
3104         kvm->arch.ept_identity_map_addr = ident_addr;
3105         return 0;
3106 }
3107
3108 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3109                                           u32 kvm_nr_mmu_pages)
3110 {
3111         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3112                 return -EINVAL;
3113
3114         mutex_lock(&kvm->slots_lock);
3115         spin_lock(&kvm->mmu_lock);
3116
3117         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3118         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3119
3120         spin_unlock(&kvm->mmu_lock);
3121         mutex_unlock(&kvm->slots_lock);
3122         return 0;
3123 }
3124
3125 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3126 {
3127         return kvm->arch.n_max_mmu_pages;
3128 }
3129
3130 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3131 {
3132         int r;
3133
3134         r = 0;
3135         switch (chip->chip_id) {
3136         case KVM_IRQCHIP_PIC_MASTER:
3137                 memcpy(&chip->chip.pic,
3138                         &pic_irqchip(kvm)->pics[0],
3139                         sizeof(struct kvm_pic_state));
3140                 break;
3141         case KVM_IRQCHIP_PIC_SLAVE:
3142                 memcpy(&chip->chip.pic,
3143                         &pic_irqchip(kvm)->pics[1],
3144                         sizeof(struct kvm_pic_state));
3145                 break;
3146         case KVM_IRQCHIP_IOAPIC:
3147                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3148                 break;
3149         default:
3150                 r = -EINVAL;
3151                 break;
3152         }
3153         return r;
3154 }
3155
3156 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3157 {
3158         int r;
3159
3160         r = 0;
3161         switch (chip->chip_id) {
3162         case KVM_IRQCHIP_PIC_MASTER:
3163                 spin_lock(&pic_irqchip(kvm)->lock);
3164                 memcpy(&pic_irqchip(kvm)->pics[0],
3165                         &chip->chip.pic,
3166                         sizeof(struct kvm_pic_state));
3167                 spin_unlock(&pic_irqchip(kvm)->lock);
3168                 break;
3169         case KVM_IRQCHIP_PIC_SLAVE:
3170                 spin_lock(&pic_irqchip(kvm)->lock);
3171                 memcpy(&pic_irqchip(kvm)->pics[1],
3172                         &chip->chip.pic,
3173                         sizeof(struct kvm_pic_state));
3174                 spin_unlock(&pic_irqchip(kvm)->lock);
3175                 break;
3176         case KVM_IRQCHIP_IOAPIC:
3177                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3178                 break;
3179         default:
3180                 r = -EINVAL;
3181                 break;
3182         }
3183         kvm_pic_update_irq(pic_irqchip(kvm));
3184         return r;
3185 }
3186
3187 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3188 {
3189         int r = 0;
3190
3191         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3192         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3193         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3194         return r;
3195 }
3196
3197 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3198 {
3199         int r = 0;
3200
3201         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3202         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3203         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3204         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3205         return r;
3206 }
3207
3208 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3209 {
3210         int r = 0;
3211
3212         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3213         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3214                 sizeof(ps->channels));
3215         ps->flags = kvm->arch.vpit->pit_state.flags;
3216         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3217         memset(&ps->reserved, 0, sizeof(ps->reserved));
3218         return r;
3219 }
3220
3221 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3222 {
3223         int r = 0, start = 0;
3224         u32 prev_legacy, cur_legacy;
3225         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3226         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3227         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3228         if (!prev_legacy && cur_legacy)
3229                 start = 1;
3230         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3231                sizeof(kvm->arch.vpit->pit_state.channels));
3232         kvm->arch.vpit->pit_state.flags = ps->flags;
3233         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3234         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3235         return r;
3236 }
3237
3238 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3239                                  struct kvm_reinject_control *control)
3240 {
3241         if (!kvm->arch.vpit)
3242                 return -ENXIO;
3243         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3244         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3245         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3246         return 0;
3247 }
3248
3249 /*
3250  * Get (and clear) the dirty memory log for a memory slot.
3251  */
3252 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3253                                       struct kvm_dirty_log *log)
3254 {
3255         int r, i;
3256         struct kvm_memory_slot *memslot;
3257         unsigned long n;
3258         unsigned long is_dirty = 0;
3259
3260         mutex_lock(&kvm->slots_lock);
3261
3262         r = -EINVAL;
3263         if (log->slot >= KVM_MEMORY_SLOTS)
3264                 goto out;
3265
3266         memslot = &kvm->memslots->memslots[log->slot];
3267         r = -ENOENT;
3268         if (!memslot->dirty_bitmap)
3269                 goto out;
3270
3271         n = kvm_dirty_bitmap_bytes(memslot);
3272
3273         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3274                 is_dirty = memslot->dirty_bitmap[i];
3275
3276         /* If nothing is dirty, don't bother messing with page tables. */
3277         if (is_dirty) {
3278                 struct kvm_memslots *slots, *old_slots;
3279                 unsigned long *dirty_bitmap;
3280
3281                 dirty_bitmap = memslot->dirty_bitmap_head;
3282                 if (memslot->dirty_bitmap == dirty_bitmap)
3283                         dirty_bitmap += n / sizeof(long);
3284                 memset(dirty_bitmap, 0, n);
3285
3286                 r = -ENOMEM;
3287                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3288                 if (!slots)
3289                         goto out;
3290                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3291                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3292                 slots->generation++;
3293
3294                 old_slots = kvm->memslots;
3295                 rcu_assign_pointer(kvm->memslots, slots);
3296                 synchronize_srcu_expedited(&kvm->srcu);
3297                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3298                 kfree(old_slots);
3299
3300                 spin_lock(&kvm->mmu_lock);
3301                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3302                 spin_unlock(&kvm->mmu_lock);
3303
3304                 r = -EFAULT;
3305                 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3306                         goto out;
3307         } else {
3308                 r = -EFAULT;
3309                 if (clear_user(log->dirty_bitmap, n))
3310                         goto out;
3311         }
3312
3313         r = 0;
3314 out:
3315         mutex_unlock(&kvm->slots_lock);
3316         return r;
3317 }
3318
3319 long kvm_arch_vm_ioctl(struct file *filp,
3320                        unsigned int ioctl, unsigned long arg)
3321 {
3322         struct kvm *kvm = filp->private_data;
3323         void __user *argp = (void __user *)arg;
3324         int r = -ENOTTY;
3325         /*
3326          * This union makes it completely explicit to gcc-3.x
3327          * that these two variables' stack usage should be
3328          * combined, not added together.
3329          */
3330         union {
3331                 struct kvm_pit_state ps;
3332                 struct kvm_pit_state2 ps2;
3333                 struct kvm_pit_config pit_config;
3334         } u;
3335
3336         switch (ioctl) {
3337         case KVM_SET_TSS_ADDR:
3338                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3339                 if (r < 0)
3340                         goto out;
3341                 break;
3342         case KVM_SET_IDENTITY_MAP_ADDR: {
3343                 u64 ident_addr;
3344
3345                 r = -EFAULT;
3346                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3347                         goto out;
3348                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3349                 if (r < 0)
3350                         goto out;
3351                 break;
3352         }
3353         case KVM_SET_NR_MMU_PAGES:
3354                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3355                 if (r)
3356                         goto out;
3357                 break;
3358         case KVM_GET_NR_MMU_PAGES:
3359                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3360                 break;
3361         case KVM_CREATE_IRQCHIP: {
3362                 struct kvm_pic *vpic;
3363
3364                 mutex_lock(&kvm->lock);
3365                 r = -EEXIST;
3366                 if (kvm->arch.vpic)
3367                         goto create_irqchip_unlock;
3368                 r = -ENOMEM;
3369                 vpic = kvm_create_pic(kvm);
3370                 if (vpic) {
3371                         r = kvm_ioapic_init(kvm);
3372                         if (r) {
3373                                 mutex_lock(&kvm->slots_lock);
3374                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3375                                                           &vpic->dev);
3376                                 mutex_unlock(&kvm->slots_lock);
3377                                 kfree(vpic);
3378                                 goto create_irqchip_unlock;
3379                         }
3380                 } else
3381                         goto create_irqchip_unlock;
3382                 smp_wmb();
3383                 kvm->arch.vpic = vpic;
3384                 smp_wmb();
3385                 r = kvm_setup_default_irq_routing(kvm);
3386                 if (r) {
3387                         mutex_lock(&kvm->slots_lock);
3388                         mutex_lock(&kvm->irq_lock);
3389                         kvm_ioapic_destroy(kvm);
3390                         kvm_destroy_pic(kvm);
3391                         mutex_unlock(&kvm->irq_lock);
3392                         mutex_unlock(&kvm->slots_lock);
3393                 }
3394         create_irqchip_unlock:
3395                 mutex_unlock(&kvm->lock);
3396                 break;
3397         }
3398         case KVM_CREATE_PIT:
3399                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3400                 goto create_pit;
3401         case KVM_CREATE_PIT2:
3402                 r = -EFAULT;
3403                 if (copy_from_user(&u.pit_config, argp,
3404                                    sizeof(struct kvm_pit_config)))
3405                         goto out;
3406         create_pit:
3407                 mutex_lock(&kvm->slots_lock);
3408                 r = -EEXIST;
3409                 if (kvm->arch.vpit)
3410                         goto create_pit_unlock;
3411                 r = -ENOMEM;
3412                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3413                 if (kvm->arch.vpit)
3414                         r = 0;
3415         create_pit_unlock:
3416                 mutex_unlock(&kvm->slots_lock);
3417                 break;
3418         case KVM_IRQ_LINE_STATUS:
3419         case KVM_IRQ_LINE: {
3420                 struct kvm_irq_level irq_event;
3421
3422                 r = -EFAULT;
3423                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3424                         goto out;
3425                 r = -ENXIO;
3426                 if (irqchip_in_kernel(kvm)) {
3427                         __s32 status;
3428                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3429                                         irq_event.irq, irq_event.level);
3430                         if (ioctl == KVM_IRQ_LINE_STATUS) {
3431                                 r = -EFAULT;
3432                                 irq_event.status = status;
3433                                 if (copy_to_user(argp, &irq_event,
3434                                                         sizeof irq_event))
3435                                         goto out;
3436                         }
3437                         r = 0;
3438                 }
3439                 break;
3440         }
3441         case KVM_GET_IRQCHIP: {
3442                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3443                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3444
3445                 r = -ENOMEM;
3446                 if (!chip)
3447                         goto out;
3448                 r = -EFAULT;
3449                 if (copy_from_user(chip, argp, sizeof *chip))
3450                         goto get_irqchip_out;
3451                 r = -ENXIO;
3452                 if (!irqchip_in_kernel(kvm))
3453                         goto get_irqchip_out;
3454                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3455                 if (r)
3456                         goto get_irqchip_out;
3457                 r = -EFAULT;
3458                 if (copy_to_user(argp, chip, sizeof *chip))
3459                         goto get_irqchip_out;
3460                 r = 0;
3461         get_irqchip_out:
3462                 kfree(chip);
3463                 if (r)
3464                         goto out;
3465                 break;
3466         }
3467         case KVM_SET_IRQCHIP: {
3468                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3469                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3470
3471                 r = -ENOMEM;
3472                 if (!chip)
3473                         goto out;
3474                 r = -EFAULT;
3475                 if (copy_from_user(chip, argp, sizeof *chip))
3476                         goto set_irqchip_out;
3477                 r = -ENXIO;
3478                 if (!irqchip_in_kernel(kvm))
3479                         goto set_irqchip_out;
3480                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3481                 if (r)
3482                         goto set_irqchip_out;
3483                 r = 0;
3484         set_irqchip_out:
3485                 kfree(chip);
3486                 if (r)
3487                         goto out;
3488                 break;
3489         }
3490         case KVM_GET_PIT: {
3491                 r = -EFAULT;
3492                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3493                         goto out;
3494                 r = -ENXIO;
3495                 if (!kvm->arch.vpit)
3496                         goto out;
3497                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3498                 if (r)
3499                         goto out;
3500                 r = -EFAULT;
3501                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3502                         goto out;
3503                 r = 0;
3504                 break;
3505         }
3506         case KVM_SET_PIT: {
3507                 r = -EFAULT;
3508                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3509                         goto out;
3510                 r = -ENXIO;
3511                 if (!kvm->arch.vpit)
3512                         goto out;
3513                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3514                 if (r)
3515                         goto out;
3516                 r = 0;
3517                 break;
3518         }
3519         case KVM_GET_PIT2: {
3520                 r = -ENXIO;
3521                 if (!kvm->arch.vpit)
3522                         goto out;
3523                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3524                 if (r)
3525                         goto out;
3526                 r = -EFAULT;
3527                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3528                         goto out;
3529                 r = 0;
3530                 break;
3531         }
3532         case KVM_SET_PIT2: {
3533                 r = -EFAULT;
3534                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3535                         goto out;
3536                 r = -ENXIO;
3537                 if (!kvm->arch.vpit)
3538                         goto out;
3539                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3540                 if (r)
3541                         goto out;
3542                 r = 0;
3543                 break;
3544         }
3545         case KVM_REINJECT_CONTROL: {
3546                 struct kvm_reinject_control control;
3547                 r =  -EFAULT;
3548                 if (copy_from_user(&control, argp, sizeof(control)))
3549                         goto out;
3550                 r = kvm_vm_ioctl_reinject(kvm, &control);
3551                 if (r)
3552                         goto out;
3553                 r = 0;
3554                 break;
3555         }
3556         case KVM_XEN_HVM_CONFIG: {
3557                 r = -EFAULT;
3558                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3559                                    sizeof(struct kvm_xen_hvm_config)))
3560                         goto out;
3561                 r = -EINVAL;
3562                 if (kvm->arch.xen_hvm_config.flags)
3563                         goto out;
3564                 r = 0;
3565                 break;
3566         }
3567         case KVM_SET_CLOCK: {
3568                 struct kvm_clock_data user_ns;
3569                 u64 now_ns;
3570                 s64 delta;
3571
3572                 r = -EFAULT;
3573                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3574                         goto out;
3575
3576                 r = -EINVAL;
3577                 if (user_ns.flags)
3578                         goto out;
3579
3580                 r = 0;
3581                 local_irq_disable();
3582                 now_ns = get_kernel_ns();
3583                 delta = user_ns.clock - now_ns;
3584                 local_irq_enable();
3585                 kvm->arch.kvmclock_offset = delta;
3586                 break;
3587         }
3588         case KVM_GET_CLOCK: {
3589                 struct kvm_clock_data user_ns;
3590                 u64 now_ns;
3591
3592                 local_irq_disable();
3593                 now_ns = get_kernel_ns();
3594                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3595                 local_irq_enable();
3596                 user_ns.flags = 0;
3597                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3598
3599                 r = -EFAULT;
3600                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3601                         goto out;
3602                 r = 0;
3603                 break;
3604         }
3605
3606         default:
3607                 ;
3608         }
3609 out:
3610         return r;
3611 }
3612
3613 static void kvm_init_msr_list(void)
3614 {
3615         u32 dummy[2];
3616         unsigned i, j;
3617
3618         /* skip the first msrs in the list. KVM-specific */
3619         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3620                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3621                         continue;
3622                 if (j < i)
3623                         msrs_to_save[j] = msrs_to_save[i];
3624                 j++;
3625         }
3626         num_msrs_to_save = j;
3627 }
3628
3629 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3630                            const void *v)
3631 {
3632         int handled = 0;
3633         int n;
3634
3635         do {
3636                 n = min(len, 8);
3637                 if (!(vcpu->arch.apic &&
3638                       !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3639                     && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3640                         break;
3641                 handled += n;
3642                 addr += n;
3643                 len -= n;
3644                 v += n;
3645         } while (len);
3646
3647         return handled;
3648 }
3649
3650 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3651 {
3652         int handled = 0;
3653         int n;
3654
3655         do {
3656                 n = min(len, 8);
3657                 if (!(vcpu->arch.apic &&
3658                       !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3659                     && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3660                         break;
3661                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3662                 handled += n;
3663                 addr += n;
3664                 len -= n;
3665                 v += n;
3666         } while (len);
3667
3668         return handled;
3669 }
3670
3671 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3672                         struct kvm_segment *var, int seg)
3673 {
3674         kvm_x86_ops->set_segment(vcpu, var, seg);
3675 }
3676
3677 void kvm_get_segment(struct kvm_vcpu *vcpu,
3678                      struct kvm_segment *var, int seg)
3679 {
3680         kvm_x86_ops->get_segment(vcpu, var, seg);
3681 }
3682
3683 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3684 {
3685         return gpa;
3686 }
3687
3688 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3689 {
3690         gpa_t t_gpa;
3691         struct x86_exception exception;
3692
3693         BUG_ON(!mmu_is_nested(vcpu));
3694
3695         /* NPT walks are always user-walks */
3696         access |= PFERR_USER_MASK;
3697         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3698
3699         return t_gpa;
3700 }
3701
3702 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3703                               struct x86_exception *exception)
3704 {
3705         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3706         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3707 }
3708
3709  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3710                                 struct x86_exception *exception)
3711 {
3712         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3713         access |= PFERR_FETCH_MASK;
3714         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3715 }
3716
3717 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3718                                struct x86_exception *exception)
3719 {
3720         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3721         access |= PFERR_WRITE_MASK;
3722         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3723 }
3724
3725 /* uses this to access any guest's mapped memory without checking CPL */
3726 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3727                                 struct x86_exception *exception)
3728 {
3729         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3730 }
3731
3732 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3733                                       struct kvm_vcpu *vcpu, u32 access,
3734                                       struct x86_exception *exception)
3735 {
3736         void *data = val;
3737         int r = X86EMUL_CONTINUE;
3738
3739         while (bytes) {
3740                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3741                                                             exception);
3742                 unsigned offset = addr & (PAGE_SIZE-1);
3743                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3744                 int ret;
3745
3746                 if (gpa == UNMAPPED_GVA)
3747                         return X86EMUL_PROPAGATE_FAULT;
3748                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3749                 if (ret < 0) {
3750                         r = X86EMUL_IO_NEEDED;
3751                         goto out;
3752                 }
3753
3754                 bytes -= toread;
3755                 data += toread;
3756                 addr += toread;
3757         }
3758 out:
3759         return r;
3760 }
3761
3762 /* used for instruction fetching */
3763 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3764                                 struct kvm_vcpu *vcpu,
3765                                 struct x86_exception *exception)
3766 {
3767         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3768         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3769                                           access | PFERR_FETCH_MASK,
3770                                           exception);
3771 }
3772
3773 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3774                                struct kvm_vcpu *vcpu,
3775                                struct x86_exception *exception)
3776 {
3777         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3778         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3779                                           exception);
3780 }
3781
3782 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3783                                       struct kvm_vcpu *vcpu,
3784                                       struct x86_exception *exception)
3785 {
3786         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3787 }
3788
3789 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3790                                        unsigned int bytes,
3791                                        struct kvm_vcpu *vcpu,
3792                                        struct x86_exception *exception)
3793 {
3794         void *data = val;
3795         int r = X86EMUL_CONTINUE;
3796
3797         while (bytes) {
3798                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3799                                                              PFERR_WRITE_MASK,
3800                                                              exception);
3801                 unsigned offset = addr & (PAGE_SIZE-1);
3802                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3803                 int ret;
3804
3805                 if (gpa == UNMAPPED_GVA)
3806                         return X86EMUL_PROPAGATE_FAULT;
3807                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3808                 if (ret < 0) {
3809                         r = X86EMUL_IO_NEEDED;
3810                         goto out;
3811                 }
3812
3813                 bytes -= towrite;
3814                 data += towrite;
3815                 addr += towrite;
3816         }
3817 out:
3818         return r;
3819 }
3820
3821 static int emulator_read_emulated(unsigned long addr,
3822                                   void *val,
3823                                   unsigned int bytes,
3824                                   struct x86_exception *exception,
3825                                   struct kvm_vcpu *vcpu)
3826 {
3827         gpa_t                 gpa;
3828         int handled;
3829
3830         if (vcpu->mmio_read_completed) {
3831                 memcpy(val, vcpu->mmio_data, bytes);
3832                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3833                                vcpu->mmio_phys_addr, *(u64 *)val);
3834                 vcpu->mmio_read_completed = 0;
3835                 return X86EMUL_CONTINUE;
3836         }
3837
3838         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3839
3840         if (gpa == UNMAPPED_GVA)
3841                 return X86EMUL_PROPAGATE_FAULT;
3842
3843         /* For APIC access vmexit */
3844         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3845                 goto mmio;
3846
3847         if (kvm_read_guest_virt(addr, val, bytes, vcpu, exception)
3848             == X86EMUL_CONTINUE)
3849                 return X86EMUL_CONTINUE;
3850
3851 mmio:
3852         /*
3853          * Is this MMIO handled locally?
3854          */
3855         handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3856
3857         if (handled == bytes)
3858                 return X86EMUL_CONTINUE;
3859
3860         gpa += handled;
3861         bytes -= handled;
3862         val += handled;
3863
3864         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3865
3866         vcpu->mmio_needed = 1;
3867         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3868         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3869         vcpu->mmio_size = bytes;
3870         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3871         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3872         vcpu->mmio_index = 0;
3873
3874         return X86EMUL_IO_NEEDED;
3875 }
3876
3877 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3878                         const void *val, int bytes)
3879 {
3880         int ret;
3881
3882         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3883         if (ret < 0)
3884                 return 0;
3885         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3886         return 1;
3887 }
3888
3889 static int emulator_write_emulated_onepage(unsigned long addr,
3890                                            const void *val,
3891                                            unsigned int bytes,
3892                                            struct x86_exception *exception,
3893                                            struct kvm_vcpu *vcpu)
3894 {
3895         gpa_t                 gpa;
3896         int handled;
3897
3898         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3899
3900         if (gpa == UNMAPPED_GVA)
3901                 return X86EMUL_PROPAGATE_FAULT;
3902
3903         /* For APIC access vmexit */
3904         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3905                 goto mmio;
3906
3907         if (emulator_write_phys(vcpu, gpa, val, bytes))
3908                 return X86EMUL_CONTINUE;
3909
3910 mmio:
3911         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3912         /*
3913          * Is this MMIO handled locally?
3914          */
3915         handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3916         if (handled == bytes)
3917                 return X86EMUL_CONTINUE;
3918
3919         gpa += handled;
3920         bytes -= handled;
3921         val += handled;
3922
3923         vcpu->mmio_needed = 1;
3924         memcpy(vcpu->mmio_data, val, bytes);
3925         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3926         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3927         vcpu->mmio_size = bytes;
3928         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3929         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3930         memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3931         vcpu->mmio_index = 0;
3932
3933         return X86EMUL_CONTINUE;
3934 }
3935
3936 int emulator_write_emulated(unsigned long addr,
3937                             const void *val,
3938                             unsigned int bytes,
3939                             struct x86_exception *exception,
3940                             struct kvm_vcpu *vcpu)
3941 {
3942         /* Crossing a page boundary? */
3943         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3944                 int rc, now;
3945
3946                 now = -addr & ~PAGE_MASK;
3947                 rc = emulator_write_emulated_onepage(addr, val, now, exception,
3948                                                      vcpu);
3949                 if (rc != X86EMUL_CONTINUE)
3950                         return rc;
3951                 addr += now;
3952                 val += now;
3953                 bytes -= now;
3954         }
3955         return emulator_write_emulated_onepage(addr, val, bytes, exception,
3956                                                vcpu);
3957 }
3958
3959 #define CMPXCHG_TYPE(t, ptr, old, new) \
3960         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3961
3962 #ifdef CONFIG_X86_64
3963 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3964 #else
3965 #  define CMPXCHG64(ptr, old, new) \
3966         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3967 #endif
3968
3969 static int emulator_cmpxchg_emulated(unsigned long addr,
3970                                      const void *old,
3971                                      const void *new,
3972                                      unsigned int bytes,
3973                                      struct x86_exception *exception,
3974                                      struct kvm_vcpu *vcpu)
3975 {
3976         gpa_t gpa;
3977         struct page *page;
3978         char *kaddr;
3979         bool exchanged;
3980
3981         /* guests cmpxchg8b have to be emulated atomically */
3982         if (bytes > 8 || (bytes & (bytes - 1)))
3983                 goto emul_write;
3984
3985         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3986
3987         if (gpa == UNMAPPED_GVA ||
3988             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3989                 goto emul_write;
3990
3991         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3992                 goto emul_write;
3993
3994         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3995         if (is_error_page(page)) {
3996                 kvm_release_page_clean(page);
3997                 goto emul_write;
3998         }
3999
4000         kaddr = kmap_atomic(page, KM_USER0);
4001         kaddr += offset_in_page(gpa);
4002         switch (bytes) {
4003         case 1:
4004                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4005                 break;
4006         case 2:
4007                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4008                 break;
4009         case 4:
4010                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4011                 break;
4012         case 8:
4013                 exchanged = CMPXCHG64(kaddr, old, new);
4014                 break;
4015         default:
4016                 BUG();
4017         }
4018         kunmap_atomic(kaddr, KM_USER0);
4019         kvm_release_page_dirty(page);
4020
4021         if (!exchanged)
4022                 return X86EMUL_CMPXCHG_FAILED;
4023
4024         kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4025
4026         return X86EMUL_CONTINUE;
4027
4028 emul_write:
4029         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4030
4031         return emulator_write_emulated(addr, new, bytes, exception, vcpu);
4032 }
4033
4034 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4035 {
4036         /* TODO: String I/O for in kernel device */
4037         int r;
4038
4039         if (vcpu->arch.pio.in)
4040                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4041                                     vcpu->arch.pio.size, pd);
4042         else
4043                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4044                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4045                                      pd);
4046         return r;
4047 }
4048
4049
4050 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
4051                              unsigned int count, struct kvm_vcpu *vcpu)
4052 {
4053         if (vcpu->arch.pio.count)
4054                 goto data_avail;
4055
4056         trace_kvm_pio(0, port, size, count);
4057
4058         vcpu->arch.pio.port = port;
4059         vcpu->arch.pio.in = 1;
4060         vcpu->arch.pio.count  = count;
4061         vcpu->arch.pio.size = size;
4062
4063         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4064         data_avail:
4065                 memcpy(val, vcpu->arch.pio_data, size * count);
4066                 vcpu->arch.pio.count = 0;
4067                 return 1;
4068         }
4069
4070         vcpu->run->exit_reason = KVM_EXIT_IO;
4071         vcpu->run->io.direction = KVM_EXIT_IO_IN;
4072         vcpu->run->io.size = size;
4073         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4074         vcpu->run->io.count = count;
4075         vcpu->run->io.port = port;
4076
4077         return 0;
4078 }
4079
4080 static int emulator_pio_out_emulated(int size, unsigned short port,
4081                               const void *val, unsigned int count,
4082                               struct kvm_vcpu *vcpu)
4083 {
4084         trace_kvm_pio(1, port, size, count);
4085
4086         vcpu->arch.pio.port = port;
4087         vcpu->arch.pio.in = 0;
4088         vcpu->arch.pio.count = count;
4089         vcpu->arch.pio.size = size;
4090
4091         memcpy(vcpu->arch.pio_data, val, size * count);
4092
4093         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4094                 vcpu->arch.pio.count = 0;
4095                 return 1;
4096         }
4097
4098         vcpu->run->exit_reason = KVM_EXIT_IO;
4099         vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4100         vcpu->run->io.size = size;
4101         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4102         vcpu->run->io.count = count;
4103         vcpu->run->io.port = port;
4104
4105         return 0;
4106 }
4107
4108 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4109 {
4110         return kvm_x86_ops->get_segment_base(vcpu, seg);
4111 }
4112
4113 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
4114 {
4115         kvm_mmu_invlpg(vcpu, address);
4116         return X86EMUL_CONTINUE;
4117 }
4118
4119 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4120 {
4121         if (!need_emulate_wbinvd(vcpu))
4122                 return X86EMUL_CONTINUE;
4123
4124         if (kvm_x86_ops->has_wbinvd_exit()) {
4125                 int cpu = get_cpu();
4126
4127                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4128                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4129                                 wbinvd_ipi, NULL, 1);
4130                 put_cpu();
4131                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4132         } else
4133                 wbinvd();
4134         return X86EMUL_CONTINUE;
4135 }
4136 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4137
4138 int emulate_clts(struct kvm_vcpu *vcpu)
4139 {
4140         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4141         kvm_x86_ops->fpu_activate(vcpu);
4142         return X86EMUL_CONTINUE;
4143 }
4144
4145 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
4146 {
4147         return _kvm_get_dr(vcpu, dr, dest);
4148 }
4149
4150 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
4151 {
4152
4153         return __kvm_set_dr(vcpu, dr, value);
4154 }
4155
4156 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4157 {
4158         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4159 }
4160
4161 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4162 {
4163         unsigned long value;
4164
4165         switch (cr) {
4166         case 0:
4167                 value = kvm_read_cr0(vcpu);
4168                 break;
4169         case 2:
4170                 value = vcpu->arch.cr2;
4171                 break;
4172         case 3:
4173                 value = kvm_read_cr3(vcpu);
4174                 break;
4175         case 4:
4176                 value = kvm_read_cr4(vcpu);
4177                 break;
4178         case 8:
4179                 value = kvm_get_cr8(vcpu);
4180                 break;
4181         default:
4182                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4183                 return 0;
4184         }
4185
4186         return value;
4187 }
4188
4189 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4190 {
4191         int res = 0;
4192
4193         switch (cr) {
4194         case 0:
4195                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4196                 break;
4197         case 2:
4198                 vcpu->arch.cr2 = val;
4199                 break;
4200         case 3:
4201                 res = kvm_set_cr3(vcpu, val);
4202                 break;
4203         case 4:
4204                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4205                 break;
4206         case 8:
4207                 res = kvm_set_cr8(vcpu, val);
4208                 break;
4209         default:
4210                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4211                 res = -1;
4212         }
4213
4214         return res;
4215 }
4216
4217 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4218 {
4219         return kvm_x86_ops->get_cpl(vcpu);
4220 }
4221
4222 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4223 {
4224         kvm_x86_ops->get_gdt(vcpu, dt);
4225 }
4226
4227 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4228 {
4229         kvm_x86_ops->get_idt(vcpu, dt);
4230 }
4231
4232 static unsigned long emulator_get_cached_segment_base(int seg,
4233                                                       struct kvm_vcpu *vcpu)
4234 {
4235         return get_segment_base(vcpu, seg);
4236 }
4237
4238 static bool emulator_get_cached_descriptor(struct desc_struct *desc, u32 *base3,
4239                                            int seg, struct kvm_vcpu *vcpu)
4240 {
4241         struct kvm_segment var;
4242
4243         kvm_get_segment(vcpu, &var, seg);
4244
4245         if (var.unusable)
4246                 return false;
4247
4248         if (var.g)
4249                 var.limit >>= 12;
4250         set_desc_limit(desc, var.limit);
4251         set_desc_base(desc, (unsigned long)var.base);
4252 #ifdef CONFIG_X86_64
4253         if (base3)
4254                 *base3 = var.base >> 32;
4255 #endif
4256         desc->type = var.type;
4257         desc->s = var.s;
4258         desc->dpl = var.dpl;
4259         desc->p = var.present;
4260         desc->avl = var.avl;
4261         desc->l = var.l;
4262         desc->d = var.db;
4263         desc->g = var.g;
4264
4265         return true;
4266 }
4267
4268 static void emulator_set_cached_descriptor(struct desc_struct *desc, u32 base3,
4269                                            int seg, struct kvm_vcpu *vcpu)
4270 {
4271         struct kvm_segment var;
4272
4273         /* needed to preserve selector */
4274         kvm_get_segment(vcpu, &var, seg);
4275
4276         var.base = get_desc_base(desc);
4277 #ifdef CONFIG_X86_64
4278         var.base |= ((u64)base3) << 32;
4279 #endif
4280         var.limit = get_desc_limit(desc);
4281         if (desc->g)
4282                 var.limit = (var.limit << 12) | 0xfff;
4283         var.type = desc->type;
4284         var.present = desc->p;
4285         var.dpl = desc->dpl;
4286         var.db = desc->d;
4287         var.s = desc->s;
4288         var.l = desc->l;
4289         var.g = desc->g;
4290         var.avl = desc->avl;
4291         var.present = desc->p;
4292         var.unusable = !var.present;
4293         var.padding = 0;
4294
4295         kvm_set_segment(vcpu, &var, seg);
4296         return;
4297 }
4298
4299 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4300 {
4301         struct kvm_segment kvm_seg;
4302
4303         kvm_get_segment(vcpu, &kvm_seg, seg);
4304         return kvm_seg.selector;
4305 }
4306
4307 static void emulator_set_segment_selector(u16 sel, int seg,
4308                                           struct kvm_vcpu *vcpu)
4309 {
4310         struct kvm_segment kvm_seg;
4311
4312         kvm_get_segment(vcpu, &kvm_seg, seg);
4313         kvm_seg.selector = sel;
4314         kvm_set_segment(vcpu, &kvm_seg, seg);
4315 }
4316
4317 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4318 {
4319         preempt_disable();
4320         kvm_load_guest_fpu(ctxt->vcpu);
4321         /*
4322          * CR0.TS may reference the host fpu state, not the guest fpu state,
4323          * so it may be clear at this point.
4324          */
4325         clts();
4326 }
4327
4328 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4329 {
4330         preempt_enable();
4331 }
4332
4333 static int emulator_intercept(struct kvm_vcpu *vcpu,
4334                               struct x86_instruction_info *info,
4335                               enum x86_intercept_stage stage)
4336 {
4337         return kvm_x86_ops->check_intercept(vcpu, info, stage);
4338 }
4339
4340 static struct x86_emulate_ops emulate_ops = {
4341         .read_std            = kvm_read_guest_virt_system,
4342         .write_std           = kvm_write_guest_virt_system,
4343         .fetch               = kvm_fetch_guest_virt,
4344         .read_emulated       = emulator_read_emulated,
4345         .write_emulated      = emulator_write_emulated,
4346         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
4347         .pio_in_emulated     = emulator_pio_in_emulated,
4348         .pio_out_emulated    = emulator_pio_out_emulated,
4349         .get_cached_descriptor = emulator_get_cached_descriptor,
4350         .set_cached_descriptor = emulator_set_cached_descriptor,
4351         .get_segment_selector = emulator_get_segment_selector,
4352         .set_segment_selector = emulator_set_segment_selector,
4353         .get_cached_segment_base = emulator_get_cached_segment_base,
4354         .get_gdt             = emulator_get_gdt,
4355         .get_idt             = emulator_get_idt,
4356         .get_cr              = emulator_get_cr,
4357         .set_cr              = emulator_set_cr,
4358         .cpl                 = emulator_get_cpl,
4359         .get_dr              = emulator_get_dr,
4360         .set_dr              = emulator_set_dr,
4361         .set_msr             = kvm_set_msr,
4362         .get_msr             = kvm_get_msr,
4363         .get_fpu             = emulator_get_fpu,
4364         .put_fpu             = emulator_put_fpu,
4365         .intercept           = emulator_intercept,
4366 };
4367
4368 static void cache_all_regs(struct kvm_vcpu *vcpu)
4369 {
4370         kvm_register_read(vcpu, VCPU_REGS_RAX);
4371         kvm_register_read(vcpu, VCPU_REGS_RSP);
4372         kvm_register_read(vcpu, VCPU_REGS_RIP);
4373         vcpu->arch.regs_dirty = ~0;
4374 }
4375
4376 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4377 {
4378         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4379         /*
4380          * an sti; sti; sequence only disable interrupts for the first
4381          * instruction. So, if the last instruction, be it emulated or
4382          * not, left the system with the INT_STI flag enabled, it
4383          * means that the last instruction is an sti. We should not
4384          * leave the flag on in this case. The same goes for mov ss
4385          */
4386         if (!(int_shadow & mask))
4387                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4388 }
4389
4390 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4391 {
4392         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4393         if (ctxt->exception.vector == PF_VECTOR)
4394                 kvm_propagate_fault(vcpu, &ctxt->exception);
4395         else if (ctxt->exception.error_code_valid)
4396                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4397                                       ctxt->exception.error_code);
4398         else
4399                 kvm_queue_exception(vcpu, ctxt->exception.vector);
4400 }
4401
4402 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4403 {
4404         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4405         int cs_db, cs_l;
4406
4407         cache_all_regs(vcpu);
4408
4409         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4410
4411         vcpu->arch.emulate_ctxt.vcpu = vcpu;
4412         vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
4413         vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4414         vcpu->arch.emulate_ctxt.mode =
4415                 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4416                 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4417                 ? X86EMUL_MODE_VM86 : cs_l
4418                 ? X86EMUL_MODE_PROT64 : cs_db
4419                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4420         vcpu->arch.emulate_ctxt.guest_mode = is_guest_mode(vcpu);
4421         memset(c, 0, sizeof(struct decode_cache));
4422         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4423         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4424 }
4425
4426 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4427 {
4428         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4429         int ret;
4430
4431         init_emulate_ctxt(vcpu);
4432
4433         vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4434         vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4435         vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip +
4436                                                                  inc_eip;
4437         ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4438
4439         if (ret != X86EMUL_CONTINUE)
4440                 return EMULATE_FAIL;
4441
4442         vcpu->arch.emulate_ctxt.eip = c->eip;
4443         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4444         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4445         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4446
4447         if (irq == NMI_VECTOR)
4448                 vcpu->arch.nmi_pending = false;
4449         else
4450                 vcpu->arch.interrupt.pending = false;
4451
4452         return EMULATE_DONE;
4453 }
4454 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4455
4456 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4457 {
4458         int r = EMULATE_DONE;
4459
4460         ++vcpu->stat.insn_emulation_fail;
4461         trace_kvm_emulate_insn_failed(vcpu);
4462         if (!is_guest_mode(vcpu)) {
4463                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4464                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4465                 vcpu->run->internal.ndata = 0;
4466                 r = EMULATE_FAIL;
4467         }
4468         kvm_queue_exception(vcpu, UD_VECTOR);
4469
4470         return r;
4471 }
4472
4473 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4474 {
4475         gpa_t gpa;
4476
4477         if (tdp_enabled)
4478                 return false;
4479
4480         /*
4481          * if emulation was due to access to shadowed page table
4482          * and it failed try to unshadow page and re-entetr the
4483          * guest to let CPU execute the instruction.
4484          */
4485         if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4486                 return true;
4487
4488         gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4489
4490         if (gpa == UNMAPPED_GVA)
4491                 return true; /* let cpu generate fault */
4492
4493         if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4494                 return true;
4495
4496         return false;
4497 }
4498
4499 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4500                             unsigned long cr2,
4501                             int emulation_type,
4502                             void *insn,
4503                             int insn_len)
4504 {
4505         int r;
4506         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4507         bool writeback = true;
4508
4509         kvm_clear_exception_queue(vcpu);
4510         vcpu->arch.mmio_fault_cr2 = cr2;
4511         /*
4512          * TODO: fix emulate.c to use guest_read/write_register
4513          * instead of direct ->regs accesses, can save hundred cycles
4514          * on Intel for instructions that don't read/change RSP, for
4515          * for example.
4516          */
4517         cache_all_regs(vcpu);
4518
4519         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4520                 init_emulate_ctxt(vcpu);
4521                 vcpu->arch.emulate_ctxt.interruptibility = 0;
4522                 vcpu->arch.emulate_ctxt.have_exception = false;
4523                 vcpu->arch.emulate_ctxt.perm_ok = false;
4524
4525                 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4526                         = emulation_type & EMULTYPE_TRAP_UD;
4527
4528                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4529
4530                 trace_kvm_emulate_insn_start(vcpu);
4531                 ++vcpu->stat.insn_emulation;
4532                 if (r)  {
4533                         if (emulation_type & EMULTYPE_TRAP_UD)
4534                                 return EMULATE_FAIL;
4535                         if (reexecute_instruction(vcpu, cr2))
4536                                 return EMULATE_DONE;
4537                         if (emulation_type & EMULTYPE_SKIP)
4538                                 return EMULATE_FAIL;
4539                         return handle_emulation_failure(vcpu);
4540                 }
4541         }
4542
4543         if (emulation_type & EMULTYPE_SKIP) {
4544                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4545                 return EMULATE_DONE;
4546         }
4547
4548         /* this is needed for vmware backdoor interface to work since it
4549            changes registers values  during IO operation */
4550         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4551                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4552                 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4553         }
4554
4555 restart:
4556         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4557
4558         if (r == EMULATION_INTERCEPTED)
4559                 return EMULATE_DONE;
4560
4561         if (r == EMULATION_FAILED) {
4562                 if (reexecute_instruction(vcpu, cr2))
4563                         return EMULATE_DONE;
4564
4565                 return handle_emulation_failure(vcpu);
4566         }
4567
4568         if (vcpu->arch.emulate_ctxt.have_exception) {
4569                 inject_emulated_exception(vcpu);
4570                 r = EMULATE_DONE;
4571         } else if (vcpu->arch.pio.count) {
4572                 if (!vcpu->arch.pio.in)
4573                         vcpu->arch.pio.count = 0;
4574                 else
4575                         writeback = false;
4576                 r = EMULATE_DO_MMIO;
4577         } else if (vcpu->mmio_needed) {
4578                 if (!vcpu->mmio_is_write)
4579                         writeback = false;
4580                 r = EMULATE_DO_MMIO;
4581         } else if (r == EMULATION_RESTART)
4582                 goto restart;
4583         else
4584                 r = EMULATE_DONE;
4585
4586         if (writeback) {
4587                 toggle_interruptibility(vcpu,
4588                                 vcpu->arch.emulate_ctxt.interruptibility);
4589                 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4590                 kvm_make_request(KVM_REQ_EVENT, vcpu);
4591                 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4592                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4593                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4594         } else
4595                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4596
4597         return r;
4598 }
4599 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4600
4601 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4602 {
4603         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4604         int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4605         /* do not return to emulator after return from userspace */
4606         vcpu->arch.pio.count = 0;
4607         return ret;
4608 }
4609 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4610
4611 static void tsc_bad(void *info)
4612 {
4613         __this_cpu_write(cpu_tsc_khz, 0);
4614 }
4615
4616 static void tsc_khz_changed(void *data)
4617 {
4618         struct cpufreq_freqs *freq = data;
4619         unsigned long khz = 0;
4620
4621         if (data)
4622                 khz = freq->new;
4623         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4624                 khz = cpufreq_quick_get(raw_smp_processor_id());
4625         if (!khz)
4626                 khz = tsc_khz;
4627         __this_cpu_write(cpu_tsc_khz, khz);
4628 }
4629
4630 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4631                                      void *data)
4632 {
4633         struct cpufreq_freqs *freq = data;
4634         struct kvm *kvm;
4635         struct kvm_vcpu *vcpu;
4636         int i, send_ipi = 0;
4637
4638         /*
4639          * We allow guests to temporarily run on slowing clocks,
4640          * provided we notify them after, or to run on accelerating
4641          * clocks, provided we notify them before.  Thus time never
4642          * goes backwards.
4643          *
4644          * However, we have a problem.  We can't atomically update
4645          * the frequency of a given CPU from this function; it is
4646          * merely a notifier, which can be called from any CPU.
4647          * Changing the TSC frequency at arbitrary points in time
4648          * requires a recomputation of local variables related to
4649          * the TSC for each VCPU.  We must flag these local variables
4650          * to be updated and be sure the update takes place with the
4651          * new frequency before any guests proceed.
4652          *
4653          * Unfortunately, the combination of hotplug CPU and frequency
4654          * change creates an intractable locking scenario; the order
4655          * of when these callouts happen is undefined with respect to
4656          * CPU hotplug, and they can race with each other.  As such,
4657          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4658          * undefined; you can actually have a CPU frequency change take
4659          * place in between the computation of X and the setting of the
4660          * variable.  To protect against this problem, all updates of
4661          * the per_cpu tsc_khz variable are done in an interrupt
4662          * protected IPI, and all callers wishing to update the value
4663          * must wait for a synchronous IPI to complete (which is trivial
4664          * if the caller is on the CPU already).  This establishes the
4665          * necessary total order on variable updates.
4666          *
4667          * Note that because a guest time update may take place
4668          * anytime after the setting of the VCPU's request bit, the
4669          * correct TSC value must be set before the request.  However,
4670          * to ensure the update actually makes it to any guest which
4671          * starts running in hardware virtualization between the set
4672          * and the acquisition of the spinlock, we must also ping the
4673          * CPU after setting the request bit.
4674          *
4675          */
4676
4677         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4678                 return 0;
4679         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4680                 return 0;
4681
4682         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4683
4684         raw_spin_lock(&kvm_lock);
4685         list_for_each_entry(kvm, &vm_list, vm_list) {
4686                 kvm_for_each_vcpu(i, vcpu, kvm) {
4687                         if (vcpu->cpu != freq->cpu)
4688                                 continue;
4689                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4690                         if (vcpu->cpu != smp_processor_id())
4691                                 send_ipi = 1;
4692                 }
4693         }
4694         raw_spin_unlock(&kvm_lock);
4695
4696         if (freq->old < freq->new && send_ipi) {
4697                 /*
4698                  * We upscale the frequency.  Must make the guest
4699                  * doesn't see old kvmclock values while running with
4700                  * the new frequency, otherwise we risk the guest sees
4701                  * time go backwards.
4702                  *
4703                  * In case we update the frequency for another cpu
4704                  * (which might be in guest context) send an interrupt
4705                  * to kick the cpu out of guest context.  Next time
4706                  * guest context is entered kvmclock will be updated,
4707                  * so the guest will not see stale values.
4708                  */
4709                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4710         }
4711         return 0;
4712 }
4713
4714 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4715         .notifier_call  = kvmclock_cpufreq_notifier
4716 };
4717
4718 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4719                                         unsigned long action, void *hcpu)
4720 {
4721         unsigned int cpu = (unsigned long)hcpu;
4722
4723         switch (action) {
4724                 case CPU_ONLINE:
4725                 case CPU_DOWN_FAILED:
4726                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4727                         break;
4728                 case CPU_DOWN_PREPARE:
4729                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
4730                         break;
4731         }
4732         return NOTIFY_OK;
4733 }
4734
4735 static struct notifier_block kvmclock_cpu_notifier_block = {
4736         .notifier_call  = kvmclock_cpu_notifier,
4737         .priority = -INT_MAX
4738 };
4739
4740 static void kvm_timer_init(void)
4741 {
4742         int cpu;
4743
4744         max_tsc_khz = tsc_khz;
4745         register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4746         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4747 #ifdef CONFIG_CPU_FREQ
4748                 struct cpufreq_policy policy;
4749                 memset(&policy, 0, sizeof(policy));
4750                 cpu = get_cpu();
4751                 cpufreq_get_policy(&policy, cpu);
4752                 if (policy.cpuinfo.max_freq)
4753                         max_tsc_khz = policy.cpuinfo.max_freq;
4754                 put_cpu();
4755 #endif
4756                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4757                                           CPUFREQ_TRANSITION_NOTIFIER);
4758         }
4759         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4760         for_each_online_cpu(cpu)
4761                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4762 }
4763
4764 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4765
4766 static int kvm_is_in_guest(void)
4767 {
4768         return percpu_read(current_vcpu) != NULL;
4769 }
4770
4771 static int kvm_is_user_mode(void)
4772 {
4773         int user_mode = 3;
4774
4775         if (percpu_read(current_vcpu))
4776                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4777
4778         return user_mode != 0;
4779 }
4780
4781 static unsigned long kvm_get_guest_ip(void)
4782 {
4783         unsigned long ip = 0;
4784
4785         if (percpu_read(current_vcpu))
4786                 ip = kvm_rip_read(percpu_read(current_vcpu));
4787
4788         return ip;
4789 }
4790
4791 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4792         .is_in_guest            = kvm_is_in_guest,
4793         .is_user_mode           = kvm_is_user_mode,
4794         .get_guest_ip           = kvm_get_guest_ip,
4795 };
4796
4797 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4798 {
4799         percpu_write(current_vcpu, vcpu);
4800 }
4801 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4802
4803 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4804 {
4805         percpu_write(current_vcpu, NULL);
4806 }
4807 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4808
4809 int kvm_arch_init(void *opaque)
4810 {
4811         int r;
4812         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4813
4814         if (kvm_x86_ops) {
4815                 printk(KERN_ERR "kvm: already loaded the other module\n");
4816                 r = -EEXIST;
4817                 goto out;
4818         }
4819
4820         if (!ops->cpu_has_kvm_support()) {
4821                 printk(KERN_ERR "kvm: no hardware support\n");
4822                 r = -EOPNOTSUPP;
4823                 goto out;
4824         }
4825         if (ops->disabled_by_bios()) {
4826                 printk(KERN_ERR "kvm: disabled by bios\n");
4827                 r = -EOPNOTSUPP;
4828                 goto out;
4829         }
4830
4831         r = kvm_mmu_module_init();
4832         if (r)
4833                 goto out;
4834
4835         kvm_init_msr_list();
4836
4837         kvm_x86_ops = ops;
4838         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4839         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4840                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
4841
4842         kvm_timer_init();
4843
4844         perf_register_guest_info_callbacks(&kvm_guest_cbs);
4845
4846         if (cpu_has_xsave)
4847                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4848
4849         return 0;
4850
4851 out:
4852         return r;
4853 }
4854
4855 void kvm_arch_exit(void)
4856 {
4857         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4858
4859         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4860                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4861                                             CPUFREQ_TRANSITION_NOTIFIER);
4862         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4863         kvm_x86_ops = NULL;
4864         kvm_mmu_module_exit();
4865 }
4866
4867 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4868 {
4869         ++vcpu->stat.halt_exits;
4870         if (irqchip_in_kernel(vcpu->kvm)) {
4871                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4872                 return 1;
4873         } else {
4874                 vcpu->run->exit_reason = KVM_EXIT_HLT;
4875                 return 0;
4876         }
4877 }
4878 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4879
4880 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4881                            unsigned long a1)
4882 {
4883         if (is_long_mode(vcpu))
4884                 return a0;
4885         else
4886                 return a0 | ((gpa_t)a1 << 32);
4887 }
4888
4889 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4890 {
4891         u64 param, ingpa, outgpa, ret;
4892         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4893         bool fast, longmode;
4894         int cs_db, cs_l;
4895
4896         /*
4897          * hypercall generates UD from non zero cpl and real mode
4898          * per HYPER-V spec
4899          */
4900         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4901                 kvm_queue_exception(vcpu, UD_VECTOR);
4902                 return 0;
4903         }
4904
4905         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4906         longmode = is_long_mode(vcpu) && cs_l == 1;
4907
4908         if (!longmode) {
4909                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4910                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4911                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4912                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4913                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4914                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4915         }
4916 #ifdef CONFIG_X86_64
4917         else {
4918                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4919                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4920                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4921         }
4922 #endif
4923
4924         code = param & 0xffff;
4925         fast = (param >> 16) & 0x1;
4926         rep_cnt = (param >> 32) & 0xfff;
4927         rep_idx = (param >> 48) & 0xfff;
4928
4929         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4930
4931         switch (code) {
4932         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4933                 kvm_vcpu_on_spin(vcpu);
4934                 break;
4935         default:
4936                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4937                 break;
4938         }
4939
4940         ret = res | (((u64)rep_done & 0xfff) << 32);
4941         if (longmode) {
4942                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4943         } else {
4944                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4945                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4946         }
4947
4948         return 1;
4949 }
4950
4951 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4952 {
4953         unsigned long nr, a0, a1, a2, a3, ret;
4954         int r = 1;
4955
4956         if (kvm_hv_hypercall_enabled(vcpu->kvm))
4957                 return kvm_hv_hypercall(vcpu);
4958
4959         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4960         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4961         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4962         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4963         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4964
4965         trace_kvm_hypercall(nr, a0, a1, a2, a3);
4966
4967         if (!is_long_mode(vcpu)) {
4968                 nr &= 0xFFFFFFFF;
4969                 a0 &= 0xFFFFFFFF;
4970                 a1 &= 0xFFFFFFFF;
4971                 a2 &= 0xFFFFFFFF;
4972                 a3 &= 0xFFFFFFFF;
4973         }
4974
4975         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4976                 ret = -KVM_EPERM;
4977                 goto out;
4978         }
4979
4980         switch (nr) {
4981         case KVM_HC_VAPIC_POLL_IRQ:
4982                 ret = 0;
4983                 break;
4984         case KVM_HC_MMU_OP:
4985                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4986                 break;
4987         default:
4988                 ret = -KVM_ENOSYS;
4989                 break;
4990         }
4991 out:
4992         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4993         ++vcpu->stat.hypercalls;
4994         return r;
4995 }
4996 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4997
4998 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4999 {
5000         char instruction[3];
5001         unsigned long rip = kvm_rip_read(vcpu);
5002
5003         /*
5004          * Blow out the MMU to ensure that no other VCPU has an active mapping
5005          * to ensure that the updated hypercall appears atomically across all
5006          * VCPUs.
5007          */
5008         kvm_mmu_zap_all(vcpu->kvm);
5009
5010         kvm_x86_ops->patch_hypercall(vcpu, instruction);
5011
5012         return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
5013 }
5014
5015 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
5016 {
5017         struct desc_ptr dt = { limit, base };
5018
5019         kvm_x86_ops->set_gdt(vcpu, &dt);
5020 }
5021
5022 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
5023 {
5024         struct desc_ptr dt = { limit, base };
5025
5026         kvm_x86_ops->set_idt(vcpu, &dt);
5027 }
5028
5029 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5030 {
5031         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5032         int j, nent = vcpu->arch.cpuid_nent;
5033
5034         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5035         /* when no next entry is found, the current entry[i] is reselected */
5036         for (j = i + 1; ; j = (j + 1) % nent) {
5037                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5038                 if (ej->function == e->function) {
5039                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5040                         return j;
5041                 }
5042         }
5043         return 0; /* silence gcc, even though control never reaches here */
5044 }
5045
5046 /* find an entry with matching function, matching index (if needed), and that
5047  * should be read next (if it's stateful) */
5048 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5049         u32 function, u32 index)
5050 {
5051         if (e->function != function)
5052                 return 0;
5053         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5054                 return 0;
5055         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5056             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5057                 return 0;
5058         return 1;
5059 }
5060
5061 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5062                                               u32 function, u32 index)
5063 {
5064         int i;
5065         struct kvm_cpuid_entry2 *best = NULL;
5066
5067         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5068                 struct kvm_cpuid_entry2 *e;
5069
5070                 e = &vcpu->arch.cpuid_entries[i];
5071                 if (is_matching_cpuid_entry(e, function, index)) {
5072                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5073                                 move_to_next_stateful_cpuid_entry(vcpu, i);
5074                         best = e;
5075                         break;
5076                 }
5077         }
5078         return best;
5079 }
5080 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5081
5082 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5083 {
5084         struct kvm_cpuid_entry2 *best;
5085
5086         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5087         if (!best || best->eax < 0x80000008)
5088                 goto not_found;
5089         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5090         if (best)
5091                 return best->eax & 0xff;
5092 not_found:
5093         return 36;
5094 }
5095
5096 /*
5097  * If no match is found, check whether we exceed the vCPU's limit
5098  * and return the content of the highest valid _standard_ leaf instead.
5099  * This is to satisfy the CPUID specification.
5100  */
5101 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5102                                                   u32 function, u32 index)
5103 {
5104         struct kvm_cpuid_entry2 *maxlevel;
5105
5106         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5107         if (!maxlevel || maxlevel->eax >= function)
5108                 return NULL;
5109         if (function & 0x80000000) {
5110                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5111                 if (!maxlevel)
5112                         return NULL;
5113         }
5114         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5115 }
5116
5117 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5118 {
5119         u32 function, index;
5120         struct kvm_cpuid_entry2 *best;
5121
5122         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5123         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5124         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5125         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5126         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5127         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5128         best = kvm_find_cpuid_entry(vcpu, function, index);
5129
5130         if (!best)
5131                 best = check_cpuid_limit(vcpu, function, index);
5132
5133         if (best) {
5134                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5135                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5136                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5137                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5138         }
5139         kvm_x86_ops->skip_emulated_instruction(vcpu);
5140         trace_kvm_cpuid(function,
5141                         kvm_register_read(vcpu, VCPU_REGS_RAX),
5142                         kvm_register_read(vcpu, VCPU_REGS_RBX),
5143                         kvm_register_read(vcpu, VCPU_REGS_RCX),
5144                         kvm_register_read(vcpu, VCPU_REGS_RDX));
5145 }
5146 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5147
5148 /*
5149  * Check if userspace requested an interrupt window, and that the
5150  * interrupt window is open.
5151  *
5152  * No need to exit to userspace if we already have an interrupt queued.
5153  */
5154 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5155 {
5156         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5157                 vcpu->run->request_interrupt_window &&
5158                 kvm_arch_interrupt_allowed(vcpu));
5159 }
5160
5161 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5162 {
5163         struct kvm_run *kvm_run = vcpu->run;
5164
5165         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5166         kvm_run->cr8 = kvm_get_cr8(vcpu);
5167         kvm_run->apic_base = kvm_get_apic_base(vcpu);
5168         if (irqchip_in_kernel(vcpu->kvm))
5169                 kvm_run->ready_for_interrupt_injection = 1;
5170         else
5171                 kvm_run->ready_for_interrupt_injection =
5172                         kvm_arch_interrupt_allowed(vcpu) &&
5173                         !kvm_cpu_has_interrupt(vcpu) &&
5174                         !kvm_event_needs_reinjection(vcpu);
5175 }
5176
5177 static void vapic_enter(struct kvm_vcpu *vcpu)
5178 {
5179         struct kvm_lapic *apic = vcpu->arch.apic;
5180         struct page *page;
5181
5182         if (!apic || !apic->vapic_addr)
5183                 return;
5184
5185         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5186
5187         vcpu->arch.apic->vapic_page = page;
5188 }
5189
5190 static void vapic_exit(struct kvm_vcpu *vcpu)
5191 {
5192         struct kvm_lapic *apic = vcpu->arch.apic;
5193         int idx;
5194
5195         if (!apic || !apic->vapic_addr)
5196                 return;
5197
5198         idx = srcu_read_lock(&vcpu->kvm->srcu);
5199         kvm_release_page_dirty(apic->vapic_page);
5200         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5201         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5202 }
5203
5204 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5205 {
5206         int max_irr, tpr;
5207
5208         if (!kvm_x86_ops->update_cr8_intercept)
5209                 return;
5210
5211         if (!vcpu->arch.apic)
5212                 return;
5213
5214         if (!vcpu->arch.apic->vapic_addr)
5215                 max_irr = kvm_lapic_find_highest_irr(vcpu);
5216         else
5217                 max_irr = -1;
5218
5219         if (max_irr != -1)
5220                 max_irr >>= 4;
5221
5222         tpr = kvm_lapic_get_cr8(vcpu);
5223
5224         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5225 }
5226
5227 static void inject_pending_event(struct kvm_vcpu *vcpu)
5228 {
5229         /* try to reinject previous events if any */
5230         if (vcpu->arch.exception.pending) {
5231                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5232                                         vcpu->arch.exception.has_error_code,
5233                                         vcpu->arch.exception.error_code);
5234                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5235                                           vcpu->arch.exception.has_error_code,
5236                                           vcpu->arch.exception.error_code,
5237                                           vcpu->arch.exception.reinject);
5238                 return;
5239         }
5240
5241         if (vcpu->arch.nmi_injected) {
5242                 kvm_x86_ops->set_nmi(vcpu);
5243                 return;
5244         }
5245
5246         if (vcpu->arch.interrupt.pending) {
5247                 kvm_x86_ops->set_irq(vcpu);
5248                 return;
5249         }
5250
5251         /* try to inject new event if pending */
5252         if (vcpu->arch.nmi_pending) {
5253                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5254                         vcpu->arch.nmi_pending = false;
5255                         vcpu->arch.nmi_injected = true;
5256                         kvm_x86_ops->set_nmi(vcpu);
5257                 }
5258         } else if (kvm_cpu_has_interrupt(vcpu)) {
5259                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5260                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5261                                             false);
5262                         kvm_x86_ops->set_irq(vcpu);
5263                 }
5264         }
5265 }
5266
5267 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5268 {
5269         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5270                         !vcpu->guest_xcr0_loaded) {
5271                 /* kvm_set_xcr() also depends on this */
5272                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5273                 vcpu->guest_xcr0_loaded = 1;
5274         }
5275 }
5276
5277 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5278 {
5279         if (vcpu->guest_xcr0_loaded) {
5280                 if (vcpu->arch.xcr0 != host_xcr0)
5281                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5282                 vcpu->guest_xcr0_loaded = 0;
5283         }
5284 }
5285
5286 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5287 {
5288         int r;
5289         bool nmi_pending;
5290         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5291                 vcpu->run->request_interrupt_window;
5292
5293         if (vcpu->requests) {
5294                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5295                         kvm_mmu_unload(vcpu);
5296                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5297                         __kvm_migrate_timers(vcpu);
5298                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5299                         r = kvm_guest_time_update(vcpu);
5300                         if (unlikely(r))
5301                                 goto out;
5302                 }
5303                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5304                         kvm_mmu_sync_roots(vcpu);
5305                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5306                         kvm_x86_ops->tlb_flush(vcpu);
5307                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5308                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5309                         r = 0;
5310                         goto out;
5311                 }
5312                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5313                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5314                         r = 0;
5315                         goto out;
5316                 }
5317                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5318                         vcpu->fpu_active = 0;
5319                         kvm_x86_ops->fpu_deactivate(vcpu);
5320                 }
5321                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5322                         /* Page is swapped out. Do synthetic halt */
5323                         vcpu->arch.apf.halted = true;
5324                         r = 1;
5325                         goto out;
5326                 }
5327         }
5328
5329         r = kvm_mmu_reload(vcpu);
5330         if (unlikely(r))
5331                 goto out;
5332
5333         /*
5334          * An NMI can be injected between local nmi_pending read and
5335          * vcpu->arch.nmi_pending read inside inject_pending_event().
5336          * But in that case, KVM_REQ_EVENT will be set, which makes
5337          * the race described above benign.
5338          */
5339         nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5340
5341         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5342                 inject_pending_event(vcpu);
5343
5344                 /* enable NMI/IRQ window open exits if needed */
5345                 if (nmi_pending)
5346                         kvm_x86_ops->enable_nmi_window(vcpu);
5347                 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5348                         kvm_x86_ops->enable_irq_window(vcpu);
5349
5350                 if (kvm_lapic_enabled(vcpu)) {
5351                         update_cr8_intercept(vcpu);
5352                         kvm_lapic_sync_to_vapic(vcpu);
5353                 }
5354         }
5355
5356         preempt_disable();
5357
5358         kvm_x86_ops->prepare_guest_switch(vcpu);
5359         if (vcpu->fpu_active)
5360                 kvm_load_guest_fpu(vcpu);
5361         kvm_load_guest_xcr0(vcpu);
5362
5363         vcpu->mode = IN_GUEST_MODE;
5364
5365         /* We should set ->mode before check ->requests,
5366          * see the comment in make_all_cpus_request.
5367          */
5368         smp_mb();
5369
5370         local_irq_disable();
5371
5372         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5373             || need_resched() || signal_pending(current)) {
5374                 vcpu->mode = OUTSIDE_GUEST_MODE;
5375                 smp_wmb();
5376                 local_irq_enable();
5377                 preempt_enable();
5378                 kvm_x86_ops->cancel_injection(vcpu);
5379                 r = 1;
5380                 goto out;
5381         }
5382
5383         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5384
5385         kvm_guest_enter();
5386
5387         if (unlikely(vcpu->arch.switch_db_regs)) {
5388                 set_debugreg(0, 7);
5389                 set_debugreg(vcpu->arch.eff_db[0], 0);
5390                 set_debugreg(vcpu->arch.eff_db[1], 1);
5391                 set_debugreg(vcpu->arch.eff_db[2], 2);
5392                 set_debugreg(vcpu->arch.eff_db[3], 3);
5393         }
5394
5395         trace_kvm_entry(vcpu->vcpu_id);
5396         kvm_x86_ops->run(vcpu);
5397
5398         /*
5399          * If the guest has used debug registers, at least dr7
5400          * will be disabled while returning to the host.
5401          * If we don't have active breakpoints in the host, we don't
5402          * care about the messed up debug address registers. But if
5403          * we have some of them active, restore the old state.
5404          */
5405         if (hw_breakpoint_active())
5406                 hw_breakpoint_restore();
5407
5408         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5409
5410         vcpu->mode = OUTSIDE_GUEST_MODE;
5411         smp_wmb();
5412         local_irq_enable();
5413
5414         ++vcpu->stat.exits;
5415
5416         /*
5417          * We must have an instruction between local_irq_enable() and
5418          * kvm_guest_exit(), so the timer interrupt isn't delayed by
5419          * the interrupt shadow.  The stat.exits increment will do nicely.
5420          * But we need to prevent reordering, hence this barrier():
5421          */
5422         barrier();
5423
5424         kvm_guest_exit();
5425
5426         preempt_enable();
5427
5428         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5429
5430         /*
5431          * Profile KVM exit RIPs:
5432          */
5433         if (unlikely(prof_on == KVM_PROFILING)) {
5434                 unsigned long rip = kvm_rip_read(vcpu);
5435                 profile_hit(KVM_PROFILING, (void *)rip);
5436         }
5437
5438
5439         kvm_lapic_sync_from_vapic(vcpu);
5440
5441         r = kvm_x86_ops->handle_exit(vcpu);
5442 out:
5443         return r;
5444 }
5445
5446
5447 static int __vcpu_run(struct kvm_vcpu *vcpu)
5448 {
5449         int r;
5450         struct kvm *kvm = vcpu->kvm;
5451
5452         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5453                 pr_debug("vcpu %d received sipi with vector # %x\n",
5454                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
5455                 kvm_lapic_reset(vcpu);
5456                 r = kvm_arch_vcpu_reset(vcpu);
5457                 if (r)
5458                         return r;
5459                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5460         }
5461
5462         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5463         vapic_enter(vcpu);
5464
5465         r = 1;
5466         while (r > 0) {
5467                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5468                     !vcpu->arch.apf.halted)
5469                         r = vcpu_enter_guest(vcpu);
5470                 else {
5471                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5472                         kvm_vcpu_block(vcpu);
5473                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5474                         if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5475                         {
5476                                 switch(vcpu->arch.mp_state) {
5477                                 case KVM_MP_STATE_HALTED:
5478                                         vcpu->arch.mp_state =
5479                                                 KVM_MP_STATE_RUNNABLE;
5480                                 case KVM_MP_STATE_RUNNABLE:
5481                                         vcpu->arch.apf.halted = false;
5482                                         break;
5483                                 case KVM_MP_STATE_SIPI_RECEIVED:
5484                                 default:
5485                                         r = -EINTR;
5486                                         break;
5487                                 }
5488                         }
5489                 }
5490
5491                 if (r <= 0)
5492                         break;
5493
5494                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5495                 if (kvm_cpu_has_pending_timer(vcpu))
5496                         kvm_inject_pending_timer_irqs(vcpu);
5497
5498                 if (dm_request_for_irq_injection(vcpu)) {
5499                         r = -EINTR;
5500                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5501                         ++vcpu->stat.request_irq_exits;
5502                 }
5503
5504                 kvm_check_async_pf_completion(vcpu);
5505
5506                 if (signal_pending(current)) {
5507                         r = -EINTR;
5508                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5509                         ++vcpu->stat.signal_exits;
5510                 }
5511                 if (need_resched()) {
5512                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5513                         kvm_resched(vcpu);
5514                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5515                 }
5516         }
5517
5518         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5519
5520         vapic_exit(vcpu);
5521
5522         return r;
5523 }
5524
5525 static int complete_mmio(struct kvm_vcpu *vcpu)
5526 {
5527         struct kvm_run *run = vcpu->run;
5528         int r;
5529
5530         if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5531                 return 1;
5532
5533         if (vcpu->mmio_needed) {
5534                 vcpu->mmio_needed = 0;
5535                 if (!vcpu->mmio_is_write)
5536                         memcpy(vcpu->mmio_data, run->mmio.data, 8);
5537                 vcpu->mmio_index += 8;
5538                 if (vcpu->mmio_index < vcpu->mmio_size) {
5539                         run->exit_reason = KVM_EXIT_MMIO;
5540                         run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5541                         memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5542                         run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5543                         run->mmio.is_write = vcpu->mmio_is_write;
5544                         vcpu->mmio_needed = 1;
5545                         return 0;
5546                 }
5547                 if (vcpu->mmio_is_write)
5548                         return 1;
5549                 vcpu->mmio_read_completed = 1;
5550         }
5551         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5552         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5553         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5554         if (r != EMULATE_DONE)
5555                 return 0;
5556         return 1;
5557 }
5558
5559 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5560 {
5561         int r;
5562         sigset_t sigsaved;
5563
5564         if (!tsk_used_math(current) && init_fpu(current))
5565                 return -ENOMEM;
5566
5567         if (vcpu->sigset_active)
5568                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5569
5570         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5571                 kvm_vcpu_block(vcpu);
5572                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5573                 r = -EAGAIN;
5574                 goto out;
5575         }
5576
5577         /* re-sync apic's tpr */
5578         if (!irqchip_in_kernel(vcpu->kvm)) {
5579                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5580                         r = -EINVAL;
5581                         goto out;
5582                 }
5583         }
5584
5585         r = complete_mmio(vcpu);
5586         if (r <= 0)
5587                 goto out;
5588
5589         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5590                 kvm_register_write(vcpu, VCPU_REGS_RAX,
5591                                      kvm_run->hypercall.ret);
5592
5593         r = __vcpu_run(vcpu);
5594
5595 out:
5596         post_kvm_run_save(vcpu);
5597         if (vcpu->sigset_active)
5598                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5599
5600         return r;
5601 }
5602
5603 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5604 {
5605         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5606                 /*
5607                  * We are here if userspace calls get_regs() in the middle of
5608                  * instruction emulation. Registers state needs to be copied
5609                  * back from emulation context to vcpu. Usrapace shouldn't do
5610                  * that usually, but some bad designed PV devices (vmware
5611                  * backdoor interface) need this to work
5612                  */
5613                 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5614                 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5615                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5616         }
5617         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5618         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5619         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5620         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5621         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5622         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5623         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5624         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5625 #ifdef CONFIG_X86_64
5626         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5627         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5628         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5629         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5630         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5631         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5632         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5633         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5634 #endif
5635
5636         regs->rip = kvm_rip_read(vcpu);
5637         regs->rflags = kvm_get_rflags(vcpu);
5638
5639         return 0;
5640 }
5641
5642 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5643 {
5644         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5645         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5646
5647         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5648         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5649         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5650         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5651         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5652         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5653         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5654         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5655 #ifdef CONFIG_X86_64
5656         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5657         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5658         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5659         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5660         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5661         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5662         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5663         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5664 #endif
5665
5666         kvm_rip_write(vcpu, regs->rip);
5667         kvm_set_rflags(vcpu, regs->rflags);
5668
5669         vcpu->arch.exception.pending = false;
5670
5671         kvm_make_request(KVM_REQ_EVENT, vcpu);
5672
5673         return 0;
5674 }
5675
5676 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5677 {
5678         struct kvm_segment cs;
5679
5680         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5681         *db = cs.db;
5682         *l = cs.l;
5683 }
5684 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5685
5686 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5687                                   struct kvm_sregs *sregs)
5688 {
5689         struct desc_ptr dt;
5690
5691         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5692         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5693         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5694         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5695         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5696         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5697
5698         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5699         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5700
5701         kvm_x86_ops->get_idt(vcpu, &dt);
5702         sregs->idt.limit = dt.size;
5703         sregs->idt.base = dt.address;
5704         kvm_x86_ops->get_gdt(vcpu, &dt);
5705         sregs->gdt.limit = dt.size;
5706         sregs->gdt.base = dt.address;
5707
5708         sregs->cr0 = kvm_read_cr0(vcpu);
5709         sregs->cr2 = vcpu->arch.cr2;
5710         sregs->cr3 = kvm_read_cr3(vcpu);
5711         sregs->cr4 = kvm_read_cr4(vcpu);
5712         sregs->cr8 = kvm_get_cr8(vcpu);
5713         sregs->efer = vcpu->arch.efer;
5714         sregs->apic_base = kvm_get_apic_base(vcpu);
5715
5716         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5717
5718         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5719                 set_bit(vcpu->arch.interrupt.nr,
5720                         (unsigned long *)sregs->interrupt_bitmap);
5721
5722         return 0;
5723 }
5724
5725 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5726                                     struct kvm_mp_state *mp_state)
5727 {
5728         mp_state->mp_state = vcpu->arch.mp_state;
5729         return 0;
5730 }
5731
5732 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5733                                     struct kvm_mp_state *mp_state)
5734 {
5735         vcpu->arch.mp_state = mp_state->mp_state;
5736         kvm_make_request(KVM_REQ_EVENT, vcpu);
5737         return 0;
5738 }
5739
5740 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5741                     bool has_error_code, u32 error_code)
5742 {
5743         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5744         int ret;
5745
5746         init_emulate_ctxt(vcpu);
5747
5748         ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5749                                    tss_selector, reason, has_error_code,
5750                                    error_code);
5751
5752         if (ret)
5753                 return EMULATE_FAIL;
5754
5755         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5756         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5757         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5758         kvm_make_request(KVM_REQ_EVENT, vcpu);
5759         return EMULATE_DONE;
5760 }
5761 EXPORT_SYMBOL_GPL(kvm_task_switch);
5762
5763 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5764                                   struct kvm_sregs *sregs)
5765 {
5766         int mmu_reset_needed = 0;
5767         int pending_vec, max_bits, idx;
5768         struct desc_ptr dt;
5769
5770         dt.size = sregs->idt.limit;
5771         dt.address = sregs->idt.base;
5772         kvm_x86_ops->set_idt(vcpu, &dt);
5773         dt.size = sregs->gdt.limit;
5774         dt.address = sregs->gdt.base;
5775         kvm_x86_ops->set_gdt(vcpu, &dt);
5776
5777         vcpu->arch.cr2 = sregs->cr2;
5778         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5779         vcpu->arch.cr3 = sregs->cr3;
5780         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5781
5782         kvm_set_cr8(vcpu, sregs->cr8);
5783
5784         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5785         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5786         kvm_set_apic_base(vcpu, sregs->apic_base);
5787
5788         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5789         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5790         vcpu->arch.cr0 = sregs->cr0;
5791
5792         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5793         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5794         if (sregs->cr4 & X86_CR4_OSXSAVE)
5795                 update_cpuid(vcpu);
5796
5797         idx = srcu_read_lock(&vcpu->kvm->srcu);
5798         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5799                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5800                 mmu_reset_needed = 1;
5801         }
5802         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5803
5804         if (mmu_reset_needed)
5805                 kvm_mmu_reset_context(vcpu);
5806
5807         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5808         pending_vec = find_first_bit(
5809                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5810         if (pending_vec < max_bits) {
5811                 kvm_queue_interrupt(vcpu, pending_vec, false);
5812                 pr_debug("Set back pending irq %d\n", pending_vec);
5813         }
5814
5815         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5816         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5817         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5818         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5819         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5820         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5821
5822         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5823         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5824
5825         update_cr8_intercept(vcpu);
5826
5827         /* Older userspace won't unhalt the vcpu on reset. */
5828         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5829             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5830             !is_protmode(vcpu))
5831                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5832
5833         kvm_make_request(KVM_REQ_EVENT, vcpu);
5834
5835         return 0;
5836 }
5837
5838 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5839                                         struct kvm_guest_debug *dbg)
5840 {
5841         unsigned long rflags;
5842         int i, r;
5843
5844         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5845                 r = -EBUSY;
5846                 if (vcpu->arch.exception.pending)
5847                         goto out;
5848                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5849                         kvm_queue_exception(vcpu, DB_VECTOR);
5850                 else
5851                         kvm_queue_exception(vcpu, BP_VECTOR);
5852         }
5853
5854         /*
5855          * Read rflags as long as potentially injected trace flags are still
5856          * filtered out.
5857          */
5858         rflags = kvm_get_rflags(vcpu);
5859
5860         vcpu->guest_debug = dbg->control;
5861         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5862                 vcpu->guest_debug = 0;
5863
5864         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5865                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5866                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5867                 vcpu->arch.switch_db_regs =
5868                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5869         } else {
5870                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5871                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5872                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5873         }
5874
5875         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5876                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5877                         get_segment_base(vcpu, VCPU_SREG_CS);
5878
5879         /*
5880          * Trigger an rflags update that will inject or remove the trace
5881          * flags.
5882          */
5883         kvm_set_rflags(vcpu, rflags);
5884
5885         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5886
5887         r = 0;
5888
5889 out:
5890
5891         return r;
5892 }
5893
5894 /*
5895  * Translate a guest virtual address to a guest physical address.
5896  */
5897 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5898                                     struct kvm_translation *tr)
5899 {
5900         unsigned long vaddr = tr->linear_address;
5901         gpa_t gpa;
5902         int idx;
5903
5904         idx = srcu_read_lock(&vcpu->kvm->srcu);
5905         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5906         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5907         tr->physical_address = gpa;
5908         tr->valid = gpa != UNMAPPED_GVA;
5909         tr->writeable = 1;
5910         tr->usermode = 0;
5911
5912         return 0;
5913 }
5914
5915 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5916 {
5917         struct i387_fxsave_struct *fxsave =
5918                         &vcpu->arch.guest_fpu.state->fxsave;
5919
5920         memcpy(fpu->fpr, fxsave->st_space, 128);
5921         fpu->fcw = fxsave->cwd;
5922         fpu->fsw = fxsave->swd;
5923         fpu->ftwx = fxsave->twd;
5924         fpu->last_opcode = fxsave->fop;
5925         fpu->last_ip = fxsave->rip;
5926         fpu->last_dp = fxsave->rdp;
5927         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5928
5929         return 0;
5930 }
5931
5932 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5933 {
5934         struct i387_fxsave_struct *fxsave =
5935                         &vcpu->arch.guest_fpu.state->fxsave;
5936
5937         memcpy(fxsave->st_space, fpu->fpr, 128);
5938         fxsave->cwd = fpu->fcw;
5939         fxsave->swd = fpu->fsw;
5940         fxsave->twd = fpu->ftwx;
5941         fxsave->fop = fpu->last_opcode;
5942         fxsave->rip = fpu->last_ip;
5943         fxsave->rdp = fpu->last_dp;
5944         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5945
5946         return 0;
5947 }
5948
5949 int fx_init(struct kvm_vcpu *vcpu)
5950 {
5951         int err;
5952
5953         err = fpu_alloc(&vcpu->arch.guest_fpu);
5954         if (err)
5955                 return err;
5956
5957         fpu_finit(&vcpu->arch.guest_fpu);
5958
5959         /*
5960          * Ensure guest xcr0 is valid for loading
5961          */
5962         vcpu->arch.xcr0 = XSTATE_FP;
5963
5964         vcpu->arch.cr0 |= X86_CR0_ET;
5965
5966         return 0;
5967 }
5968 EXPORT_SYMBOL_GPL(fx_init);
5969
5970 static void fx_free(struct kvm_vcpu *vcpu)
5971 {
5972         fpu_free(&vcpu->arch.guest_fpu);
5973 }
5974
5975 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5976 {
5977         if (vcpu->guest_fpu_loaded)
5978                 return;
5979
5980         /*
5981          * Restore all possible states in the guest,
5982          * and assume host would use all available bits.
5983          * Guest xcr0 would be loaded later.
5984          */
5985         kvm_put_guest_xcr0(vcpu);
5986         vcpu->guest_fpu_loaded = 1;
5987         unlazy_fpu(current);
5988         fpu_restore_checking(&vcpu->arch.guest_fpu);
5989         trace_kvm_fpu(1);
5990 }
5991
5992 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5993 {
5994         kvm_put_guest_xcr0(vcpu);
5995
5996         if (!vcpu->guest_fpu_loaded)
5997                 return;
5998
5999         vcpu->guest_fpu_loaded = 0;
6000         fpu_save_init(&vcpu->arch.guest_fpu);
6001         ++vcpu->stat.fpu_reload;
6002         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6003         trace_kvm_fpu(0);
6004 }
6005
6006 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6007 {
6008         kvmclock_reset(vcpu);
6009
6010         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6011         fx_free(vcpu);
6012         kvm_x86_ops->vcpu_free(vcpu);
6013 }
6014
6015 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6016                                                 unsigned int id)
6017 {
6018         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6019                 printk_once(KERN_WARNING
6020                 "kvm: SMP vm created on host with unstable TSC; "
6021                 "guest TSC will not be reliable\n");
6022         return kvm_x86_ops->vcpu_create(kvm, id);
6023 }
6024
6025 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6026 {
6027         int r;
6028
6029         vcpu->arch.mtrr_state.have_fixed = 1;
6030         vcpu_load(vcpu);
6031         r = kvm_arch_vcpu_reset(vcpu);
6032         if (r == 0)
6033                 r = kvm_mmu_setup(vcpu);
6034         vcpu_put(vcpu);
6035         if (r < 0)
6036                 goto free_vcpu;
6037
6038         return 0;
6039 free_vcpu:
6040         kvm_x86_ops->vcpu_free(vcpu);
6041         return r;
6042 }
6043
6044 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6045 {
6046         vcpu->arch.apf.msr_val = 0;
6047
6048         vcpu_load(vcpu);
6049         kvm_mmu_unload(vcpu);
6050         vcpu_put(vcpu);
6051
6052         fx_free(vcpu);
6053         kvm_x86_ops->vcpu_free(vcpu);
6054 }
6055
6056 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6057 {
6058         vcpu->arch.nmi_pending = false;
6059         vcpu->arch.nmi_injected = false;
6060
6061         vcpu->arch.switch_db_regs = 0;
6062         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6063         vcpu->arch.dr6 = DR6_FIXED_1;
6064         vcpu->arch.dr7 = DR7_FIXED_1;
6065
6066         kvm_make_request(KVM_REQ_EVENT, vcpu);
6067         vcpu->arch.apf.msr_val = 0;
6068
6069         kvmclock_reset(vcpu);
6070
6071         kvm_clear_async_pf_completion_queue(vcpu);
6072         kvm_async_pf_hash_reset(vcpu);
6073         vcpu->arch.apf.halted = false;
6074
6075         return kvm_x86_ops->vcpu_reset(vcpu);
6076 }
6077
6078 int kvm_arch_hardware_enable(void *garbage)
6079 {
6080         struct kvm *kvm;
6081         struct kvm_vcpu *vcpu;
6082         int i;
6083
6084         kvm_shared_msr_cpu_online();
6085         list_for_each_entry(kvm, &vm_list, vm_list)
6086                 kvm_for_each_vcpu(i, vcpu, kvm)
6087                         if (vcpu->cpu == smp_processor_id())
6088                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6089         return kvm_x86_ops->hardware_enable(garbage);
6090 }
6091
6092 void kvm_arch_hardware_disable(void *garbage)
6093 {
6094         kvm_x86_ops->hardware_disable(garbage);
6095         drop_user_return_notifiers(garbage);
6096 }
6097
6098 int kvm_arch_hardware_setup(void)
6099 {
6100         return kvm_x86_ops->hardware_setup();
6101 }
6102
6103 void kvm_arch_hardware_unsetup(void)
6104 {
6105         kvm_x86_ops->hardware_unsetup();
6106 }
6107
6108 void kvm_arch_check_processor_compat(void *rtn)
6109 {
6110         kvm_x86_ops->check_processor_compatibility(rtn);
6111 }
6112
6113 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6114 {
6115         struct page *page;
6116         struct kvm *kvm;
6117         int r;
6118
6119         BUG_ON(vcpu->kvm == NULL);
6120         kvm = vcpu->kvm;
6121
6122         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6123         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6124         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6125         vcpu->arch.mmu.translate_gpa = translate_gpa;
6126         vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6127         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6128                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6129         else
6130                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6131
6132         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6133         if (!page) {
6134                 r = -ENOMEM;
6135                 goto fail;
6136         }
6137         vcpu->arch.pio_data = page_address(page);
6138
6139         kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6140
6141         r = kvm_mmu_create(vcpu);
6142         if (r < 0)
6143                 goto fail_free_pio_data;
6144
6145         if (irqchip_in_kernel(kvm)) {
6146                 r = kvm_create_lapic(vcpu);
6147                 if (r < 0)
6148                         goto fail_mmu_destroy;
6149         }
6150
6151         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6152                                        GFP_KERNEL);
6153         if (!vcpu->arch.mce_banks) {
6154                 r = -ENOMEM;
6155                 goto fail_free_lapic;
6156         }
6157         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6158
6159         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6160                 goto fail_free_mce_banks;
6161
6162         kvm_async_pf_hash_reset(vcpu);
6163
6164         return 0;
6165 fail_free_mce_banks:
6166         kfree(vcpu->arch.mce_banks);
6167 fail_free_lapic:
6168         kvm_free_lapic(vcpu);
6169 fail_mmu_destroy:
6170         kvm_mmu_destroy(vcpu);
6171 fail_free_pio_data:
6172         free_page((unsigned long)vcpu->arch.pio_data);
6173 fail:
6174         return r;
6175 }
6176
6177 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6178 {
6179         int idx;
6180
6181         kfree(vcpu->arch.mce_banks);
6182         kvm_free_lapic(vcpu);
6183         idx = srcu_read_lock(&vcpu->kvm->srcu);
6184         kvm_mmu_destroy(vcpu);
6185         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6186         free_page((unsigned long)vcpu->arch.pio_data);
6187 }
6188
6189 int kvm_arch_init_vm(struct kvm *kvm)
6190 {
6191         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6192         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6193
6194         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6195         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6196
6197         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6198
6199         return 0;
6200 }
6201
6202 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6203 {
6204         vcpu_load(vcpu);
6205         kvm_mmu_unload(vcpu);
6206         vcpu_put(vcpu);
6207 }
6208
6209 static void kvm_free_vcpus(struct kvm *kvm)
6210 {
6211         unsigned int i;
6212         struct kvm_vcpu *vcpu;
6213
6214         /*
6215          * Unpin any mmu pages first.
6216          */
6217         kvm_for_each_vcpu(i, vcpu, kvm) {
6218                 kvm_clear_async_pf_completion_queue(vcpu);
6219                 kvm_unload_vcpu_mmu(vcpu);
6220         }
6221         kvm_for_each_vcpu(i, vcpu, kvm)
6222                 kvm_arch_vcpu_free(vcpu);
6223
6224         mutex_lock(&kvm->lock);
6225         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6226                 kvm->vcpus[i] = NULL;
6227
6228         atomic_set(&kvm->online_vcpus, 0);
6229         mutex_unlock(&kvm->lock);
6230 }
6231
6232 void kvm_arch_sync_events(struct kvm *kvm)
6233 {
6234         kvm_free_all_assigned_devices(kvm);
6235         kvm_free_pit(kvm);
6236 }
6237
6238 void kvm_arch_destroy_vm(struct kvm *kvm)
6239 {
6240         kvm_iommu_unmap_guest(kvm);
6241         kfree(kvm->arch.vpic);
6242         kfree(kvm->arch.vioapic);
6243         kvm_free_vcpus(kvm);
6244         if (kvm->arch.apic_access_page)
6245                 put_page(kvm->arch.apic_access_page);
6246         if (kvm->arch.ept_identity_pagetable)
6247                 put_page(kvm->arch.ept_identity_pagetable);
6248 }
6249
6250 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6251                                 struct kvm_memory_slot *memslot,
6252                                 struct kvm_memory_slot old,
6253                                 struct kvm_userspace_memory_region *mem,
6254                                 int user_alloc)
6255 {
6256         int npages = memslot->npages;
6257         int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6258
6259         /* Prevent internal slot pages from being moved by fork()/COW. */
6260         if (memslot->id >= KVM_MEMORY_SLOTS)
6261                 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6262
6263         /*To keep backward compatibility with older userspace,
6264          *x86 needs to hanlde !user_alloc case.
6265          */
6266         if (!user_alloc) {
6267                 if (npages && !old.rmap) {
6268                         unsigned long userspace_addr;
6269
6270                         down_write(&current->mm->mmap_sem);
6271                         userspace_addr = do_mmap(NULL, 0,
6272                                                  npages * PAGE_SIZE,
6273                                                  PROT_READ | PROT_WRITE,
6274                                                  map_flags,
6275                                                  0);
6276                         up_write(&current->mm->mmap_sem);
6277
6278                         if (IS_ERR((void *)userspace_addr))
6279                                 return PTR_ERR((void *)userspace_addr);
6280
6281                         memslot->userspace_addr = userspace_addr;
6282                 }
6283         }
6284
6285
6286         return 0;
6287 }
6288
6289 void kvm_arch_commit_memory_region(struct kvm *kvm,
6290                                 struct kvm_userspace_memory_region *mem,
6291                                 struct kvm_memory_slot old,
6292                                 int user_alloc)
6293 {
6294
6295         int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6296
6297         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6298                 int ret;
6299
6300                 down_write(&current->mm->mmap_sem);
6301                 ret = do_munmap(current->mm, old.userspace_addr,
6302                                 old.npages * PAGE_SIZE);
6303                 up_write(&current->mm->mmap_sem);
6304                 if (ret < 0)
6305                         printk(KERN_WARNING
6306                                "kvm_vm_ioctl_set_memory_region: "
6307                                "failed to munmap memory\n");
6308         }
6309
6310         if (!kvm->arch.n_requested_mmu_pages)
6311                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6312
6313         spin_lock(&kvm->mmu_lock);
6314         if (nr_mmu_pages)
6315                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6316         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6317         spin_unlock(&kvm->mmu_lock);
6318 }
6319
6320 void kvm_arch_flush_shadow(struct kvm *kvm)
6321 {
6322         kvm_mmu_zap_all(kvm);
6323         kvm_reload_remote_mmus(kvm);
6324 }
6325
6326 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6327 {
6328         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6329                 !vcpu->arch.apf.halted)
6330                 || !list_empty_careful(&vcpu->async_pf.done)
6331                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6332                 || vcpu->arch.nmi_pending ||
6333                 (kvm_arch_interrupt_allowed(vcpu) &&
6334                  kvm_cpu_has_interrupt(vcpu));
6335 }
6336
6337 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6338 {
6339         int me;
6340         int cpu = vcpu->cpu;
6341
6342         if (waitqueue_active(&vcpu->wq)) {
6343                 wake_up_interruptible(&vcpu->wq);
6344                 ++vcpu->stat.halt_wakeup;
6345         }
6346
6347         me = get_cpu();
6348         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6349                 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6350                         smp_send_reschedule(cpu);
6351         put_cpu();
6352 }
6353
6354 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6355 {
6356         return kvm_x86_ops->interrupt_allowed(vcpu);
6357 }
6358
6359 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6360 {
6361         unsigned long current_rip = kvm_rip_read(vcpu) +
6362                 get_segment_base(vcpu, VCPU_SREG_CS);
6363
6364         return current_rip == linear_rip;
6365 }
6366 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6367
6368 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6369 {
6370         unsigned long rflags;
6371
6372         rflags = kvm_x86_ops->get_rflags(vcpu);
6373         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6374                 rflags &= ~X86_EFLAGS_TF;
6375         return rflags;
6376 }
6377 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6378
6379 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6380 {
6381         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6382             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6383                 rflags |= X86_EFLAGS_TF;
6384         kvm_x86_ops->set_rflags(vcpu, rflags);
6385         kvm_make_request(KVM_REQ_EVENT, vcpu);
6386 }
6387 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6388
6389 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6390 {
6391         int r;
6392
6393         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6394               is_error_page(work->page))
6395                 return;
6396
6397         r = kvm_mmu_reload(vcpu);
6398         if (unlikely(r))
6399                 return;
6400
6401         if (!vcpu->arch.mmu.direct_map &&
6402               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6403                 return;
6404
6405         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6406 }
6407
6408 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6409 {
6410         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6411 }
6412
6413 static inline u32 kvm_async_pf_next_probe(u32 key)
6414 {
6415         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6416 }
6417
6418 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6419 {
6420         u32 key = kvm_async_pf_hash_fn(gfn);
6421
6422         while (vcpu->arch.apf.gfns[key] != ~0)
6423                 key = kvm_async_pf_next_probe(key);
6424
6425         vcpu->arch.apf.gfns[key] = gfn;
6426 }
6427
6428 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6429 {
6430         int i;
6431         u32 key = kvm_async_pf_hash_fn(gfn);
6432
6433         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6434                      (vcpu->arch.apf.gfns[key] != gfn &&
6435                       vcpu->arch.apf.gfns[key] != ~0); i++)
6436                 key = kvm_async_pf_next_probe(key);
6437
6438         return key;
6439 }
6440
6441 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6442 {
6443         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6444 }
6445
6446 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6447 {
6448         u32 i, j, k;
6449
6450         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6451         while (true) {
6452                 vcpu->arch.apf.gfns[i] = ~0;
6453                 do {
6454                         j = kvm_async_pf_next_probe(j);
6455                         if (vcpu->arch.apf.gfns[j] == ~0)
6456                                 return;
6457                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6458                         /*
6459                          * k lies cyclically in ]i,j]
6460                          * |    i.k.j |
6461                          * |....j i.k.| or  |.k..j i...|
6462                          */
6463                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6464                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6465                 i = j;
6466         }
6467 }
6468
6469 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6470 {
6471
6472         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6473                                       sizeof(val));
6474 }
6475
6476 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6477                                      struct kvm_async_pf *work)
6478 {
6479         struct x86_exception fault;
6480
6481         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6482         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6483
6484         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6485             (vcpu->arch.apf.send_user_only &&
6486              kvm_x86_ops->get_cpl(vcpu) == 0))
6487                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6488         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6489                 fault.vector = PF_VECTOR;
6490                 fault.error_code_valid = true;
6491                 fault.error_code = 0;
6492                 fault.nested_page_fault = false;
6493                 fault.address = work->arch.token;
6494                 kvm_inject_page_fault(vcpu, &fault);
6495         }
6496 }
6497
6498 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6499                                  struct kvm_async_pf *work)
6500 {
6501         struct x86_exception fault;
6502
6503         trace_kvm_async_pf_ready(work->arch.token, work->gva);
6504         if (is_error_page(work->page))
6505                 work->arch.token = ~0; /* broadcast wakeup */
6506         else
6507                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6508
6509         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6510             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6511                 fault.vector = PF_VECTOR;
6512                 fault.error_code_valid = true;
6513                 fault.error_code = 0;
6514                 fault.nested_page_fault = false;
6515                 fault.address = work->arch.token;
6516                 kvm_inject_page_fault(vcpu, &fault);
6517         }
6518         vcpu->arch.apf.halted = false;
6519 }
6520
6521 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6522 {
6523         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6524                 return true;
6525         else
6526                 return !kvm_event_needs_reinjection(vcpu) &&
6527                         kvm_x86_ops->interrupt_allowed(vcpu);
6528 }
6529
6530 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6531 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6532 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6533 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6534 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6535 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6536 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6537 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6538 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6539 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6540 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6541 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);