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 #include "cpuid.h"
  30 #include "assigned-dev.h"
  31 #include "pmu.h"
  32 #include "hyperv.h"
  33
  34 #include <linux/clocksource.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/kvm.h>
  37 #include <linux/fs.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/module.h>
  40 #include <linux/mman.h>
  41 #include <linux/highmem.h>
  42 #include <linux/iommu.h>
  43 #include <linux/intel-iommu.h>
  44 #include <linux/cpufreq.h>
  45 #include <linux/user-return-notifier.h>
  46 #include <linux/srcu.h>
  47 #include <linux/slab.h>
  48 #include <linux/perf_event.h>
  49 #include <linux/uaccess.h>
  50 #include <linux/hash.h>
  51 #include <linux/pci.h>
  52 #include <linux/timekeeper_internal.h>
  53 #include <linux/pvclock_gtod.h>
  54 #include <linux/kvm_irqfd.h>
  55 #include <linux/irqbypass.h>
  56 #include <trace/events/kvm.h>
  57
  58 #define CREATE_TRACE_POINTS
  59 #include "trace.h"
  60
  61 #include <asm/debugreg.h>
  62 #include <asm/msr.h>
  63 #include <asm/desc.h>
  64 #include <asm/mce.h>
  65 #include <linux/kernel_stat.h>
  66 #include <asm/fpu/internal.h> /* Ugh! */
  67 #include <asm/pvclock.h>
  68 #include <asm/div64.h>
  69 #include <asm/irq_remapping.h>
  70
  71 #define MAX_IO_MSRS 256
  72 #define KVM_MAX_MCE_BANKS 32
  73 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
  74
  75 #define emul_to_vcpu(ctxt) \
  76         container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
  77
  78 /* EFER defaults:
  79  * - enable syscall per default because its emulated by KVM
  80  * - enable LME and LMA per default on 64 bit KVM
  81  */
  82 #ifdef CONFIG_X86_64
  83 static
  84 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  85 #else
  86 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  87 #endif
  88
  89 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  90 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  91
  92 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  93 static void process_nmi(struct kvm_vcpu *vcpu);
  94 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
  95
  96 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
  97 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  98
  99 static bool __read_mostly ignore_msrs = 0;
 100 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
 101
 102 unsigned int min_timer_period_us = 500;
 103 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
 104
 105 static bool __read_mostly kvmclock_periodic_sync = true;
 106 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
 107
 108 bool __read_mostly kvm_has_tsc_control;
 109 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
 110 u32  __read_mostly kvm_max_guest_tsc_khz;
 111 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
 112 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
 113 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
 114 u64  __read_mostly kvm_max_tsc_scaling_ratio;
 115 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
 116 static u64 __read_mostly kvm_default_tsc_scaling_ratio;
 117
 118 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
 119 static u32 __read_mostly tsc_tolerance_ppm = 250;
 120 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
 121
 122 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
 123 unsigned int __read_mostly lapic_timer_advance_ns = 0;
 124 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
 125
 126 static bool __read_mostly backwards_tsc_observed = false;
 127
 128 #define KVM_NR_SHARED_MSRS 16
 129
 130 struct kvm_shared_msrs_global {
 131         int nr;
 132         u32 msrs[KVM_NR_SHARED_MSRS];
 133 };
 134
 135 struct kvm_shared_msrs {
 136         struct user_return_notifier urn;
 137         bool registered;
 138         struct kvm_shared_msr_values {
 139                 u64 host;
 140                 u64 curr;
 141         } values[KVM_NR_SHARED_MSRS];
 142 };
 143
 144 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 145 static struct kvm_shared_msrs __percpu *shared_msrs;
 146
 147 struct kvm_stats_debugfs_item debugfs_entries[] = {
 148         { "pf_fixed", VCPU_STAT(pf_fixed) },
 149         { "pf_guest", VCPU_STAT(pf_guest) },
 150         { "tlb_flush", VCPU_STAT(tlb_flush) },
 151         { "invlpg", VCPU_STAT(invlpg) },
 152         { "exits", VCPU_STAT(exits) },
 153         { "io_exits", VCPU_STAT(io_exits) },
 154         { "mmio_exits", VCPU_STAT(mmio_exits) },
 155         { "signal_exits", VCPU_STAT(signal_exits) },
 156         { "irq_window", VCPU_STAT(irq_window_exits) },
 157         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 158         { "halt_exits", VCPU_STAT(halt_exits) },
 159         { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
 160         { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
 161         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 162         { "hypercalls", VCPU_STAT(hypercalls) },
 163         { "request_irq", VCPU_STAT(request_irq_exits) },
 164         { "irq_exits", VCPU_STAT(irq_exits) },
 165         { "host_state_reload", VCPU_STAT(host_state_reload) },
 166         { "efer_reload", VCPU_STAT(efer_reload) },
 167         { "fpu_reload", VCPU_STAT(fpu_reload) },
 168         { "insn_emulation", VCPU_STAT(insn_emulation) },
 169         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 170         { "irq_injections", VCPU_STAT(irq_injections) },
 171         { "nmi_injections", VCPU_STAT(nmi_injections) },
 172         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 173         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 174         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 175         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 176         { "mmu_flooded", VM_STAT(mmu_flooded) },
 177         { "mmu_recycled", VM_STAT(mmu_recycled) },
 178         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 179         { "mmu_unsync", VM_STAT(mmu_unsync) },
 180         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 181         { "largepages", VM_STAT(lpages) },
 182         { NULL }
 183 };
 184
 185 u64 __read_mostly host_xcr0;
 186
 187 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
 188
 189 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 190 {
 191         int i;
 192         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 193                 vcpu->arch.apf.gfns[i] = ~0;
 194 }
 195
 196 static void kvm_on_user_return(struct user_return_notifier *urn)
 197 {
 198         unsigned slot;
 199         struct kvm_shared_msrs *locals
 200                 = container_of(urn, struct kvm_shared_msrs, urn);
 201         struct kvm_shared_msr_values *values;
 202
 203         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 204                 values = &locals->values[slot];
 205                 if (values->host != values->curr) {
 206                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 207                         values->curr = values->host;
 208                 }
 209         }
 210         locals->registered = false;
 211         user_return_notifier_unregister(urn);
 212 }
 213
 214 static void shared_msr_update(unsigned slot, u32 msr)
 215 {
 216         u64 value;
 217         unsigned int cpu = smp_processor_id();
 218         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 219
 220         /* only read, and nobody should modify it at this time,
 221          * so don't need lock */
 222         if (slot >= shared_msrs_global.nr) {
 223                 printk(KERN_ERR "kvm: invalid MSR slot!");
 224                 return;
 225         }
 226         rdmsrl_safe(msr, &value);
 227         smsr->values[slot].host = value;
 228         smsr->values[slot].curr = value;
 229 }
 230
 231 void kvm_define_shared_msr(unsigned slot, u32 msr)
 232 {
 233         BUG_ON(slot >= KVM_NR_SHARED_MSRS);
 234         shared_msrs_global.msrs[slot] = msr;
 235         if (slot >= shared_msrs_global.nr)
 236                 shared_msrs_global.nr = slot + 1;
 237 }
 238 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 239
 240 static void kvm_shared_msr_cpu_online(void)
 241 {
 242         unsigned i;
 243
 244         for (i = 0; i < shared_msrs_global.nr; ++i)
 245                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 246 }
 247
 248 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 249 {
 250         unsigned int cpu = smp_processor_id();
 251         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 252         int err;
 253
 254         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 255                 return 0;
 256         smsr->values[slot].curr = value;
 257         err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
 258         if (err)
 259                 return 1;
 260
 261         if (!smsr->registered) {
 262                 smsr->urn.on_user_return = kvm_on_user_return;
 263                 user_return_notifier_register(&smsr->urn);
 264                 smsr->registered = true;
 265         }
 266         return 0;
 267 }
 268 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 269
 270 static void drop_user_return_notifiers(void)
 271 {
 272         unsigned int cpu = smp_processor_id();
 273         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 274
 275         if (smsr->registered)
 276                 kvm_on_user_return(&smsr->urn);
 277 }
 278
 279 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 280 {
 281         return vcpu->arch.apic_base;
 282 }
 283 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 284
 285 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 286 {
 287         u64 old_state = vcpu->arch.apic_base &
 288                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 289         u64 new_state = msr_info->data &
 290                 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
 291         u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
 292                 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
 293
 294         if (!msr_info->host_initiated &&
 295             ((msr_info->data & reserved_bits) != 0 ||
 296              new_state == X2APIC_ENABLE ||
 297              (new_state == MSR_IA32_APICBASE_ENABLE &&
 298               old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
 299              (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
 300               old_state == 0)))
 301                 return 1;
 302
 303         kvm_lapic_set_base(vcpu, msr_info->data);
 304         return 0;
 305 }
 306 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 307
 308 asmlinkage __visible void kvm_spurious_fault(void)
 309 {
 310         /* Fault while not rebooting.  We want the trace. */
 311         BUG();
 312 }
 313 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
 314
 315 #define EXCPT_BENIGN            0
 316 #define EXCPT_CONTRIBUTORY      1
 317 #define EXCPT_PF                2
 318
 319 static int exception_class(int vector)
 320 {
 321         switch (vector) {
 322         case PF_VECTOR:
 323                 return EXCPT_PF;
 324         case DE_VECTOR:
 325         case TS_VECTOR:
 326         case NP_VECTOR:
 327         case SS_VECTOR:
 328         case GP_VECTOR:
 329                 return EXCPT_CONTRIBUTORY;
 330         default:
 331                 break;
 332         }
 333         return EXCPT_BENIGN;
 334 }
 335
 336 #define EXCPT_FAULT             0
 337 #define EXCPT_TRAP              1
 338 #define EXCPT_ABORT             2
 339 #define EXCPT_INTERRUPT         3
 340
 341 static int exception_type(int vector)
 342 {
 343         unsigned int mask;
 344
 345         if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
 346                 return EXCPT_INTERRUPT;
 347
 348         mask = 1 << vector;
 349
 350         /* #DB is trap, as instruction watchpoints are handled elsewhere */
 351         if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
 352                 return EXCPT_TRAP;
 353
 354         if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
 355                 return EXCPT_ABORT;
 356
 357         /* Reserved exceptions will result in fault */
 358         return EXCPT_FAULT;
 359 }
 360
 361 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 362                 unsigned nr, bool has_error, u32 error_code,
 363                 bool reinject)
 364 {
 365         u32 prev_nr;
 366         int class1, class2;
 367
 368         kvm_make_request(KVM_REQ_EVENT, vcpu);
 369
 370         if (!vcpu->arch.exception.pending) {
 371         queue:
 372                 if (has_error && !is_protmode(vcpu))
 373                         has_error = false;
 374                 vcpu->arch.exception.pending = true;
 375                 vcpu->arch.exception.has_error_code = has_error;
 376                 vcpu->arch.exception.nr = nr;
 377                 vcpu->arch.exception.error_code = error_code;
 378                 vcpu->arch.exception.reinject = reinject;
 379                 return;
 380         }
 381
 382         /* to check exception */
 383         prev_nr = vcpu->arch.exception.nr;
 384         if (prev_nr == DF_VECTOR) {
 385                 /* triple fault -> shutdown */
 386                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 387                 return;
 388         }
 389         class1 = exception_class(prev_nr);
 390         class2 = exception_class(nr);
 391         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 392                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 393                 /* generate double fault per SDM Table 5-5 */
 394                 vcpu->arch.exception.pending = true;
 395                 vcpu->arch.exception.has_error_code = true;
 396                 vcpu->arch.exception.nr = DF_VECTOR;
 397                 vcpu->arch.exception.error_code = 0;
 398         } else
 399                 /* replace previous exception with a new one in a hope
 400                    that instruction re-execution will regenerate lost
 401                    exception */
 402                 goto queue;
 403 }
 404
 405 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 406 {
 407         kvm_multiple_exception(vcpu, nr, false, 0, false);
 408 }
 409 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 410
 411 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 412 {
 413         kvm_multiple_exception(vcpu, nr, false, 0, true);
 414 }
 415 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 416
 417 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 418 {
 419         if (err)
 420                 kvm_inject_gp(vcpu, 0);
 421         else
 422                 kvm_x86_ops->skip_emulated_instruction(vcpu);
 423 }
 424 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 425
 426 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 427 {
 428         ++vcpu->stat.pf_guest;
 429         vcpu->arch.cr2 = fault->address;
 430         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 431 }
 432 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
 433
 434 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 435 {
 436         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 437                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 438         else
 439                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 440
 441         return fault->nested_page_fault;
 442 }
 443
 444 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 445 {
 446         atomic_inc(&vcpu->arch.nmi_queued);
 447         kvm_make_request(KVM_REQ_NMI, vcpu);
 448 }
 449 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 450
 451 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 452 {
 453         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 454 }
 455 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 456
 457 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 458 {
 459         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 460 }
 461 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 462
 463 /*
 464  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 465  * a #GP and return false.
 466  */
 467 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 468 {
 469         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 470                 return true;
 471         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 472         return false;
 473 }
 474 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 475
 476 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
 477 {
 478         if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 479                 return true;
 480
 481         kvm_queue_exception(vcpu, UD_VECTOR);
 482         return false;
 483 }
 484 EXPORT_SYMBOL_GPL(kvm_require_dr);
 485
 486 /*
 487  * This function will be used to read from the physical memory of the currently
 488  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
 489  * can read from guest physical or from the guest's guest physical memory.
 490  */
 491 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 492                             gfn_t ngfn, void *data, int offset, int len,
 493                             u32 access)
 494 {
 495         struct x86_exception exception;
 496         gfn_t real_gfn;
 497         gpa_t ngpa;
 498
 499         ngpa     = gfn_to_gpa(ngfn);
 500         real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
 501         if (real_gfn == UNMAPPED_GVA)
 502                 return -EFAULT;
 503
 504         real_gfn = gpa_to_gfn(real_gfn);
 505
 506         return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
 507 }
 508 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 509
 510 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 511                                void *data, int offset, int len, u32 access)
 512 {
 513         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 514                                        data, offset, len, access);
 515 }
 516
 517 /*
 518  * Load the pae pdptrs.  Return true is they are all valid.
 519  */
 520 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 521 {
 522         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 523         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 524         int i;
 525         int ret;
 526         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 527
 528         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 529                                       offset * sizeof(u64), sizeof(pdpte),
 530                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 531         if (ret < 0) {
 532                 ret = 0;
 533                 goto out;
 534         }
 535         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 536                 if (is_present_gpte(pdpte[i]) &&
 537                     (pdpte[i] &
 538                      vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
 539                         ret = 0;
 540                         goto out;
 541                 }
 542         }
 543         ret = 1;
 544
 545         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 546         __set_bit(VCPU_EXREG_PDPTR,
 547                   (unsigned long *)&vcpu->arch.regs_avail);
 548         __set_bit(VCPU_EXREG_PDPTR,
 549                   (unsigned long *)&vcpu->arch.regs_dirty);
 550 out:
 551
 552         return ret;
 553 }
 554 EXPORT_SYMBOL_GPL(load_pdptrs);
 555
 556 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 557 {
 558         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 559         bool changed = true;
 560         int offset;
 561         gfn_t gfn;
 562         int r;
 563
 564         if (is_long_mode(vcpu) || !is_pae(vcpu))
 565                 return false;
 566
 567         if (!test_bit(VCPU_EXREG_PDPTR,
 568                       (unsigned long *)&vcpu->arch.regs_avail))
 569                 return true;
 570
 571         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 572         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 573         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 574                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 575         if (r < 0)
 576                 goto out;
 577         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 578 out:
 579
 580         return changed;
 581 }
 582
 583 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 584 {
 585         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 586         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
 587
 588         cr0 |= X86_CR0_ET;
 589
 590 #ifdef CONFIG_X86_64
 591         if (cr0 & 0xffffffff00000000UL)
 592                 return 1;
 593 #endif
 594
 595         cr0 &= ~CR0_RESERVED_BITS;
 596
 597         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 598                 return 1;
 599
 600         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 601                 return 1;
 602
 603         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 604 #ifdef CONFIG_X86_64
 605                 if ((vcpu->arch.efer & EFER_LME)) {
 606                         int cs_db, cs_l;
 607
 608                         if (!is_pae(vcpu))
 609                                 return 1;
 610                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 611                         if (cs_l)
 612                                 return 1;
 613                 } else
 614 #endif
 615                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 616                                                  kvm_read_cr3(vcpu)))
 617                         return 1;
 618         }
 619
 620         if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
 621                 return 1;
 622
 623         kvm_x86_ops->set_cr0(vcpu, cr0);
 624
 625         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 626                 kvm_clear_async_pf_completion_queue(vcpu);
 627                 kvm_async_pf_hash_reset(vcpu);
 628         }
 629
 630         if ((cr0 ^ old_cr0) & update_bits)
 631                 kvm_mmu_reset_context(vcpu);
 632
 633         if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
 634             kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
 635             !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
 636                 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
 637
 638         return 0;
 639 }
 640 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 641
 642 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 643 {
 644         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 645 }
 646 EXPORT_SYMBOL_GPL(kvm_lmsw);
 647
 648 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 649 {
 650         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
 651                         !vcpu->guest_xcr0_loaded) {
 652                 /* kvm_set_xcr() also depends on this */
 653                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
 654                 vcpu->guest_xcr0_loaded = 1;
 655         }
 656 }
 657
 658 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 659 {
 660         if (vcpu->guest_xcr0_loaded) {
 661                 if (vcpu->arch.xcr0 != host_xcr0)
 662                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
 663                 vcpu->guest_xcr0_loaded = 0;
 664         }
 665 }
 666
 667 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 668 {
 669         u64 xcr0 = xcr;
 670         u64 old_xcr0 = vcpu->arch.xcr0;
 671         u64 valid_bits;
 672
 673         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 674         if (index != XCR_XFEATURE_ENABLED_MASK)
 675                 return 1;
 676         if (!(xcr0 & XFEATURE_MASK_FP))
 677                 return 1;
 678         if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
 679                 return 1;
 680
 681         /*
 682          * Do not allow the guest to set bits that we do not support
 683          * saving.  However, xcr0 bit 0 is always set, even if the
 684          * emulated CPU does not support XSAVE (see fx_init).
 685          */
 686         valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
 687         if (xcr0 & ~valid_bits)
 688                 return 1;
 689
 690         if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
 691             (!(xcr0 & XFEATURE_MASK_BNDCSR)))
 692                 return 1;
 693
 694         if (xcr0 & XFEATURE_MASK_AVX512) {
 695                 if (!(xcr0 & XFEATURE_MASK_YMM))
 696                         return 1;
 697                 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
 698                         return 1;
 699         }
 700         vcpu->arch.xcr0 = xcr0;
 701
 702         if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
 703                 kvm_update_cpuid(vcpu);
 704         return 0;
 705 }
 706
 707 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 708 {
 709         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
 710             __kvm_set_xcr(vcpu, index, xcr)) {
 711                 kvm_inject_gp(vcpu, 0);
 712                 return 1;
 713         }
 714         return 0;
 715 }
 716 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 717
 718 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 719 {
 720         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 721         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
 722                                    X86_CR4_SMEP | X86_CR4_SMAP;
 723
 724         if (cr4 & CR4_RESERVED_BITS)
 725                 return 1;
 726
 727         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 728                 return 1;
 729
 730         if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
 731                 return 1;
 732
 733         if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
 734                 return 1;
 735
 736         if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
 737                 return 1;
 738
 739         if (is_long_mode(vcpu)) {
 740                 if (!(cr4 & X86_CR4_PAE))
 741                         return 1;
 742         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 743                    && ((cr4 ^ old_cr4) & pdptr_bits)
 744                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 745                                    kvm_read_cr3(vcpu)))
 746                 return 1;
 747
 748         if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
 749                 if (!guest_cpuid_has_pcid(vcpu))
 750                         return 1;
 751
 752                 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
 753                 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
 754                         return 1;
 755         }
 756
 757         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 758                 return 1;
 759
 760         if (((cr4 ^ old_cr4) & pdptr_bits) ||
 761             (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
 762                 kvm_mmu_reset_context(vcpu);
 763
 764         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 765                 kvm_update_cpuid(vcpu);
 766
 767         return 0;
 768 }
 769 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 770
 771 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 772 {
 773 #ifdef CONFIG_X86_64
 774         cr3 &= ~CR3_PCID_INVD;
 775 #endif
 776
 777         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 778                 kvm_mmu_sync_roots(vcpu);
 779                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 780                 return 0;
 781         }
 782
 783         if (is_long_mode(vcpu)) {
 784                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 785                         return 1;
 786         } else if (is_pae(vcpu) && is_paging(vcpu) &&
 787                    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 788                 return 1;
 789
 790         vcpu->arch.cr3 = cr3;
 791         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 792         kvm_mmu_new_cr3(vcpu);
 793         return 0;
 794 }
 795 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 796
 797 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 798 {
 799         if (cr8 & CR8_RESERVED_BITS)
 800                 return 1;
 801         if (lapic_in_kernel(vcpu))
 802                 kvm_lapic_set_tpr(vcpu, cr8);
 803         else
 804                 vcpu->arch.cr8 = cr8;
 805         return 0;
 806 }
 807 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 808
 809 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 810 {
 811         if (lapic_in_kernel(vcpu))
 812                 return kvm_lapic_get_cr8(vcpu);
 813         else
 814                 return vcpu->arch.cr8;
 815 }
 816 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 817
 818 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
 819 {
 820         int i;
 821
 822         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 823                 for (i = 0; i < KVM_NR_DB_REGS; i++)
 824                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
 825                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
 826         }
 827 }
 828
 829 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
 830 {
 831         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 832                 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
 833 }
 834
 835 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
 836 {
 837         unsigned long dr7;
 838
 839         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 840                 dr7 = vcpu->arch.guest_debug_dr7;
 841         else
 842                 dr7 = vcpu->arch.dr7;
 843         kvm_x86_ops->set_dr7(vcpu, dr7);
 844         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
 845         if (dr7 & DR7_BP_EN_MASK)
 846                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
 847 }
 848
 849 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
 850 {
 851         u64 fixed = DR6_FIXED_1;
 852
 853         if (!guest_cpuid_has_rtm(vcpu))
 854                 fixed |= DR6_RTM;
 855         return fixed;
 856 }
 857
 858 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 859 {
 860         switch (dr) {
 861         case 0 ... 3:
 862                 vcpu->arch.db[dr] = val;
 863                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 864                         vcpu->arch.eff_db[dr] = val;
 865                 break;
 866         case 4:
 867                 /* fall through */
 868         case 6:
 869                 if (val & 0xffffffff00000000ULL)
 870                         return -1; /* #GP */
 871                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
 872                 kvm_update_dr6(vcpu);
 873                 break;
 874         case 5:
 875                 /* fall through */
 876         default: /* 7 */
 877                 if (val & 0xffffffff00000000ULL)
 878                         return -1; /* #GP */
 879                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 880                 kvm_update_dr7(vcpu);
 881                 break;
 882         }
 883
 884         return 0;
 885 }
 886
 887 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 888 {
 889         if (__kvm_set_dr(vcpu, dr, val)) {
 890                 kvm_inject_gp(vcpu, 0);
 891                 return 1;
 892         }
 893         return 0;
 894 }
 895 EXPORT_SYMBOL_GPL(kvm_set_dr);
 896
 897 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 898 {
 899         switch (dr) {
 900         case 0 ... 3:
 901                 *val = vcpu->arch.db[dr];
 902                 break;
 903         case 4:
 904                 /* fall through */
 905         case 6:
 906                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 907                         *val = vcpu->arch.dr6;
 908                 else
 909                         *val = kvm_x86_ops->get_dr6(vcpu);
 910                 break;
 911         case 5:
 912                 /* fall through */
 913         default: /* 7 */
 914                 *val = vcpu->arch.dr7;
 915                 break;
 916         }
 917         return 0;
 918 }
 919 EXPORT_SYMBOL_GPL(kvm_get_dr);
 920
 921 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
 922 {
 923         u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 924         u64 data;
 925         int err;
 926
 927         err = kvm_pmu_rdpmc(vcpu, ecx, &data);
 928         if (err)
 929                 return err;
 930         kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
 931         kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
 932         return err;
 933 }
 934 EXPORT_SYMBOL_GPL(kvm_rdpmc);
 935
 936 /*
 937  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 938  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 939  *
 940  * This list is modified at module load time to reflect the
 941  * capabilities of the host cpu. This capabilities test skips MSRs that are
 942  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
 943  * may depend on host virtualization features rather than host cpu features.
 944  */
 945
 946 static u32 msrs_to_save[] = {
 947         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 948         MSR_STAR,
 949 #ifdef CONFIG_X86_64
 950         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 951 #endif
 952         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
 953         MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
 954 };
 955
 956 static unsigned num_msrs_to_save;
 957
 958 static u32 emulated_msrs[] = {
 959         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 960         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 961         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 962         HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
 963         HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
 964         HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
 965         HV_X64_MSR_RESET,
 966         HV_X64_MSR_VP_INDEX,
 967         HV_X64_MSR_VP_RUNTIME,
 968         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
 969         MSR_KVM_PV_EOI_EN,
 970
 971         MSR_IA32_TSC_ADJUST,
 972         MSR_IA32_TSCDEADLINE,
 973         MSR_IA32_MISC_ENABLE,
 974         MSR_IA32_MCG_STATUS,
 975         MSR_IA32_MCG_CTL,
 976         MSR_IA32_SMBASE,
 977 };
 978
 979 static unsigned num_emulated_msrs;
 980
 981 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
 982 {
 983         if (efer & efer_reserved_bits)
 984                 return false;
 985
 986         if (efer & EFER_FFXSR) {
 987                 struct kvm_cpuid_entry2 *feat;
 988
 989                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 990                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 991                         return false;
 992         }
 993
 994         if (efer & EFER_SVME) {
 995                 struct kvm_cpuid_entry2 *feat;
 996
 997                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 998                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 999                         return false;
1000         }
1001
1002         return true;
1003 }
1004 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1005
1006 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1007 {
1008         u64 old_efer = vcpu->arch.efer;
1009
1010         if (!kvm_valid_efer(vcpu, efer))
1011                 return 1;
1012
1013         if (is_paging(vcpu)
1014             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1015                 return 1;
1016
1017         efer &= ~EFER_LMA;
1018         efer |= vcpu->arch.efer & EFER_LMA;
1019
1020         kvm_x86_ops->set_efer(vcpu, efer);
1021
1022         /* Update reserved bits */
1023         if ((efer ^ old_efer) & EFER_NX)
1024                 kvm_mmu_reset_context(vcpu);
1025
1026         return 0;
1027 }
1028
1029 void kvm_enable_efer_bits(u64 mask)
1030 {
1031        efer_reserved_bits &= ~mask;
1032 }
1033 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1034
1035 /*
1036  * Writes msr value into into the appropriate "register".
1037  * Returns 0 on success, non-0 otherwise.
1038  * Assumes vcpu_load() was already called.
1039  */
1040 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1041 {
1042         switch (msr->index) {
1043         case MSR_FS_BASE:
1044         case MSR_GS_BASE:
1045         case MSR_KERNEL_GS_BASE:
1046         case MSR_CSTAR:
1047         case MSR_LSTAR:
1048                 if (is_noncanonical_address(msr->data))
1049                         return 1;
1050                 break;
1051         case MSR_IA32_SYSENTER_EIP:
1052         case MSR_IA32_SYSENTER_ESP:
1053                 /*
1054                  * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1055                  * non-canonical address is written on Intel but not on
1056                  * AMD (which ignores the top 32-bits, because it does
1057                  * not implement 64-bit SYSENTER).
1058                  *
1059                  * 64-bit code should hence be able to write a non-canonical
1060                  * value on AMD.  Making the address canonical ensures that
1061                  * vmentry does not fail on Intel after writing a non-canonical
1062                  * value, and that something deterministic happens if the guest
1063                  * invokes 64-bit SYSENTER.
1064                  */
1065                 msr->data = get_canonical(msr->data);
1066         }
1067         return kvm_x86_ops->set_msr(vcpu, msr);
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_set_msr);
1070
1071 /*
1072  * Adapt set_msr() to msr_io()'s calling convention
1073  */
1074 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1075 {
1076         struct msr_data msr;
1077         int r;
1078
1079         msr.index = index;
1080         msr.host_initiated = true;
1081         r = kvm_get_msr(vcpu, &msr);
1082         if (r)
1083                 return r;
1084
1085         *data = msr.data;
1086         return 0;
1087 }
1088
1089 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1090 {
1091         struct msr_data msr;
1092
1093         msr.data = *data;
1094         msr.index = index;
1095         msr.host_initiated = true;
1096         return kvm_set_msr(vcpu, &msr);
1097 }
1098
1099 #ifdef CONFIG_X86_64
1100 struct pvclock_gtod_data {
1101         seqcount_t      seq;
1102
1103         struct { /* extract of a clocksource struct */
1104                 int vclock_mode;
1105                 cycle_t cycle_last;
1106                 cycle_t mask;
1107                 u32     mult;
1108                 u32     shift;
1109         } clock;
1110
1111         u64             boot_ns;
1112         u64             nsec_base;
1113 };
1114
1115 static struct pvclock_gtod_data pvclock_gtod_data;
1116
1117 static void update_pvclock_gtod(struct timekeeper *tk)
1118 {
1119         struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1120         u64 boot_ns;
1121
1122         boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1123
1124         write_seqcount_begin(&vdata->seq);
1125
1126         /* copy pvclock gtod data */
1127         vdata->clock.vclock_mode        = tk->tkr_mono.clock->archdata.vclock_mode;
1128         vdata->clock.cycle_last         = tk->tkr_mono.cycle_last;
1129         vdata->clock.mask               = tk->tkr_mono.mask;
1130         vdata->clock.mult               = tk->tkr_mono.mult;
1131         vdata->clock.shift              = tk->tkr_mono.shift;
1132
1133         vdata->boot_ns                  = boot_ns;
1134         vdata->nsec_base                = tk->tkr_mono.xtime_nsec;
1135
1136         write_seqcount_end(&vdata->seq);
1137 }
1138 #endif
1139
1140 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1141 {
1142         /*
1143          * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1144          * vcpu_enter_guest.  This function is only called from
1145          * the physical CPU that is running vcpu.
1146          */
1147         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1148 }
1149
1150 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1151 {
1152         int version;
1153         int r;
1154         struct pvclock_wall_clock wc;
1155         struct timespec boot;
1156
1157         if (!wall_clock)
1158                 return;
1159
1160         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1161         if (r)
1162                 return;
1163
1164         if (version & 1)
1165                 ++version;  /* first time write, random junk */
1166
1167         ++version;
1168
1169         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1170
1171         /*
1172          * The guest calculates current wall clock time by adding
1173          * system time (updated by kvm_guest_time_update below) to the
1174          * wall clock specified here.  guest system time equals host
1175          * system time for us, thus we must fill in host boot time here.
1176          */
1177         getboottime(&boot);
1178
1179         if (kvm->arch.kvmclock_offset) {
1180                 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1181                 boot = timespec_sub(boot, ts);
1182         }
1183         wc.sec = boot.tv_sec;
1184         wc.nsec = boot.tv_nsec;
1185         wc.version = version;
1186
1187         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1188
1189         version++;
1190         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1191 }
1192
1193 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1194 {
1195         uint32_t quotient, remainder;
1196
1197         /* Don't try to replace with do_div(), this one calculates
1198          * "(dividend << 32) / divisor" */
1199         __asm__ ( "divl %4"
1200                   : "=a" (quotient), "=d" (remainder)
1201                   : "0" (0), "1" (dividend), "r" (divisor) );
1202         return quotient;
1203 }
1204
1205 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1206                                s8 *pshift, u32 *pmultiplier)
1207 {
1208         uint64_t scaled64;
1209         int32_t  shift = 0;
1210         uint64_t tps64;
1211         uint32_t tps32;
1212
1213         tps64 = base_khz * 1000LL;
1214         scaled64 = scaled_khz * 1000LL;
1215         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1216                 tps64 >>= 1;
1217                 shift--;
1218         }
1219
1220         tps32 = (uint32_t)tps64;
1221         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1222                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1223                         scaled64 >>= 1;
1224                 else
1225                         tps32 <<= 1;
1226                 shift++;
1227         }
1228
1229         *pshift = shift;
1230         *pmultiplier = div_frac(scaled64, tps32);
1231
1232         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1233                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
1234 }
1235
1236 #ifdef CONFIG_X86_64
1237 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1238 #endif
1239
1240 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1241 static unsigned long max_tsc_khz;
1242
1243 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1244 {
1245         return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1246                                    vcpu->arch.virtual_tsc_shift);
1247 }
1248
1249 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1250 {
1251         u64 v = (u64)khz * (1000000 + ppm);
1252         do_div(v, 1000000);
1253         return v;
1254 }
1255
1256 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1257 {
1258         u64 ratio;
1259
1260         /* Guest TSC same frequency as host TSC? */
1261         if (!scale) {
1262                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1263                 return 0;
1264         }
1265
1266         /* TSC scaling supported? */
1267         if (!kvm_has_tsc_control) {
1268                 if (user_tsc_khz > tsc_khz) {
1269                         vcpu->arch.tsc_catchup = 1;
1270                         vcpu->arch.tsc_always_catchup = 1;
1271                         return 0;
1272                 } else {
1273                         WARN(1, "user requested TSC rate below hardware speed\n");
1274                         return -1;
1275                 }
1276         }
1277
1278         /* TSC scaling required  - calculate ratio */
1279         ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1280                                 user_tsc_khz, tsc_khz);
1281
1282         if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1283                 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1284                           user_tsc_khz);
1285                 return -1;
1286         }
1287
1288         vcpu->arch.tsc_scaling_ratio = ratio;
1289         return 0;
1290 }
1291
1292 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1293 {
1294         u32 thresh_lo, thresh_hi;
1295         int use_scaling = 0;
1296
1297         /* tsc_khz can be zero if TSC calibration fails */
1298         if (this_tsc_khz == 0) {
1299                 /* set tsc_scaling_ratio to a safe value */
1300                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1301                 return -1;
1302         }
1303
1304         /* Compute a scale to convert nanoseconds in TSC cycles */
1305         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1306                            &vcpu->arch.virtual_tsc_shift,
1307                            &vcpu->arch.virtual_tsc_mult);
1308         vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1309
1310         /*
1311          * Compute the variation in TSC rate which is acceptable
1312          * within the range of tolerance and decide if the
1313          * rate being applied is within that bounds of the hardware
1314          * rate.  If so, no scaling or compensation need be done.
1315          */
1316         thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1317         thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1318         if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1319                 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1320                 use_scaling = 1;
1321         }
1322         return set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1323 }
1324
1325 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1326 {
1327         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1328                                       vcpu->arch.virtual_tsc_mult,
1329                                       vcpu->arch.virtual_tsc_shift);
1330         tsc += vcpu->arch.this_tsc_write;
1331         return tsc;
1332 }
1333
1334 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1335 {
1336 #ifdef CONFIG_X86_64
1337         bool vcpus_matched;
1338         struct kvm_arch *ka = &vcpu->kvm->arch;
1339         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1340
1341         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1342                          atomic_read(&vcpu->kvm->online_vcpus));
1343
1344         /*
1345          * Once the masterclock is enabled, always perform request in
1346          * order to update it.
1347          *
1348          * In order to enable masterclock, the host clocksource must be TSC
1349          * and the vcpus need to have matched TSCs.  When that happens,
1350          * perform request to enable masterclock.
1351          */
1352         if (ka->use_master_clock ||
1353             (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1354                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1355
1356         trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1357                             atomic_read(&vcpu->kvm->online_vcpus),
1358                             ka->use_master_clock, gtod->clock.vclock_mode);
1359 #endif
1360 }
1361
1362 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1363 {
1364         u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1365         vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1366 }
1367
1368 /*
1369  * Multiply tsc by a fixed point number represented by ratio.
1370  *
1371  * The most significant 64-N bits (mult) of ratio represent the
1372  * integral part of the fixed point number; the remaining N bits
1373  * (frac) represent the fractional part, ie. ratio represents a fixed
1374  * point number (mult + frac * 2^(-N)).
1375  *
1376  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1377  */
1378 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1379 {
1380         return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1381 }
1382
1383 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1384 {
1385         u64 _tsc = tsc;
1386         u64 ratio = vcpu->arch.tsc_scaling_ratio;
1387
1388         if (ratio != kvm_default_tsc_scaling_ratio)
1389                 _tsc = __scale_tsc(ratio, tsc);
1390
1391         return _tsc;
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1394
1395 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1396 {
1397         u64 tsc;
1398
1399         tsc = kvm_scale_tsc(vcpu, rdtsc());
1400
1401         return target_tsc - tsc;
1402 }
1403
1404 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1405 {
1406         return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1407 }
1408 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1409
1410 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1411 {
1412         struct kvm *kvm = vcpu->kvm;
1413         u64 offset, ns, elapsed;
1414         unsigned long flags;
1415         s64 usdiff;
1416         bool matched;
1417         bool already_matched;
1418         u64 data = msr->data;
1419
1420         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1421         offset = kvm_compute_tsc_offset(vcpu, data);
1422         ns = get_kernel_ns();
1423         elapsed = ns - kvm->arch.last_tsc_nsec;
1424
1425         if (vcpu->arch.virtual_tsc_khz) {
1426                 int faulted = 0;
1427
1428                 /* n.b - signed multiplication and division required */
1429                 usdiff = data - kvm->arch.last_tsc_write;
1430 #ifdef CONFIG_X86_64
1431                 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1432 #else
1433                 /* do_div() only does unsigned */
1434                 asm("1: idivl %[divisor]\n"
1435                     "2: xor %%edx, %%edx\n"
1436                     "   movl $0, %[faulted]\n"
1437                     "3:\n"
1438                     ".section .fixup,\"ax\"\n"
1439                     "4: movl $1, %[faulted]\n"
1440                     "   jmp  3b\n"
1441                     ".previous\n"
1442
1443                 _ASM_EXTABLE(1b, 4b)
1444
1445                 : "=A"(usdiff), [faulted] "=r" (faulted)
1446                 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1447
1448 #endif
1449                 do_div(elapsed, 1000);
1450                 usdiff -= elapsed;
1451                 if (usdiff < 0)
1452                         usdiff = -usdiff;
1453
1454                 /* idivl overflow => difference is larger than USEC_PER_SEC */
1455                 if (faulted)
1456                         usdiff = USEC_PER_SEC;
1457         } else
1458                 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1459
1460         /*
1461          * Special case: TSC write with a small delta (1 second) of virtual
1462          * cycle time against real time is interpreted as an attempt to
1463          * synchronize the CPU.
1464          *
1465          * For a reliable TSC, we can match TSC offsets, and for an unstable
1466          * TSC, we add elapsed time in this computation.  We could let the
1467          * compensation code attempt to catch up if we fall behind, but
1468          * it's better to try to match offsets from the beginning.
1469          */
1470         if (usdiff < USEC_PER_SEC &&
1471             vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1472                 if (!check_tsc_unstable()) {
1473                         offset = kvm->arch.cur_tsc_offset;
1474                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1475                 } else {
1476                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1477                         data += delta;
1478                         offset = kvm_compute_tsc_offset(vcpu, data);
1479                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1480                 }
1481                 matched = true;
1482                 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1483         } else {
1484                 /*
1485                  * We split periods of matched TSC writes into generations.
1486                  * For each generation, we track the original measured
1487                  * nanosecond time, offset, and write, so if TSCs are in
1488                  * sync, we can match exact offset, and if not, we can match
1489                  * exact software computation in compute_guest_tsc()
1490                  *
1491                  * These values are tracked in kvm->arch.cur_xxx variables.
1492                  */
1493                 kvm->arch.cur_tsc_generation++;
1494                 kvm->arch.cur_tsc_nsec = ns;
1495                 kvm->arch.cur_tsc_write = data;
1496                 kvm->arch.cur_tsc_offset = offset;
1497                 matched = false;
1498                 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1499                          kvm->arch.cur_tsc_generation, data);
1500         }
1501
1502         /*
1503          * We also track th most recent recorded KHZ, write and time to
1504          * allow the matching interval to be extended at each write.
1505          */
1506         kvm->arch.last_tsc_nsec = ns;
1507         kvm->arch.last_tsc_write = data;
1508         kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1509
1510         vcpu->arch.last_guest_tsc = data;
1511
1512         /* Keep track of which generation this VCPU has synchronized to */
1513         vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1514         vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1515         vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1516
1517         if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1518                 update_ia32_tsc_adjust_msr(vcpu, offset);
1519         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1520         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1521
1522         spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1523         if (!matched) {
1524                 kvm->arch.nr_vcpus_matched_tsc = 0;
1525         } else if (!already_matched) {
1526                 kvm->arch.nr_vcpus_matched_tsc++;
1527         }
1528
1529         kvm_track_tsc_matching(vcpu);
1530         spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1531 }
1532
1533 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1534
1535 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1536                                            s64 adjustment)
1537 {
1538         kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1539 }
1540
1541 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1542 {
1543         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1544                 WARN_ON(adjustment < 0);
1545         adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1546         kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1547 }
1548
1549 #ifdef CONFIG_X86_64
1550
1551 static cycle_t read_tsc(void)
1552 {
1553         cycle_t ret = (cycle_t)rdtsc_ordered();
1554         u64 last = pvclock_gtod_data.clock.cycle_last;
1555
1556         if (likely(ret >= last))
1557                 return ret;
1558
1559         /*
1560          * GCC likes to generate cmov here, but this branch is extremely
1561          * predictable (it's just a funciton of time and the likely is
1562          * very likely) and there's a data dependence, so force GCC
1563          * to generate a branch instead.  I don't barrier() because
1564          * we don't actually need a barrier, and if this function
1565          * ever gets inlined it will generate worse code.
1566          */
1567         asm volatile ("");
1568         return last;
1569 }
1570
1571 static inline u64 vgettsc(cycle_t *cycle_now)
1572 {
1573         long v;
1574         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1575
1576         *cycle_now = read_tsc();
1577
1578         v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1579         return v * gtod->clock.mult;
1580 }
1581
1582 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1583 {
1584         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1585         unsigned long seq;
1586         int mode;
1587         u64 ns;
1588
1589         do {
1590                 seq = read_seqcount_begin(&gtod->seq);
1591                 mode = gtod->clock.vclock_mode;
1592                 ns = gtod->nsec_base;
1593                 ns += vgettsc(cycle_now);
1594                 ns >>= gtod->clock.shift;
1595                 ns += gtod->boot_ns;
1596         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1597         *t = ns;
1598
1599         return mode;
1600 }
1601
1602 /* returns true if host is using tsc clocksource */
1603 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1604 {
1605         /* checked again under seqlock below */
1606         if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1607                 return false;
1608
1609         return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1610 }
1611 #endif
1612
1613 /*
1614  *
1615  * Assuming a stable TSC across physical CPUS, and a stable TSC
1616  * across virtual CPUs, the following condition is possible.
1617  * Each numbered line represents an event visible to both
1618  * CPUs at the next numbered event.
1619  *
1620  * "timespecX" represents host monotonic time. "tscX" represents
1621  * RDTSC value.
1622  *
1623  *              VCPU0 on CPU0           |       VCPU1 on CPU1
1624  *
1625  * 1.  read timespec0,tsc0
1626  * 2.                                   | timespec1 = timespec0 + N
1627  *                                      | tsc1 = tsc0 + M
1628  * 3. transition to guest               | transition to guest
1629  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1630  * 5.                                   | ret1 = timespec1 + (rdtsc - tsc1)
1631  *                                      | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1632  *
1633  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1634  *
1635  *      - ret0 < ret1
1636  *      - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1637  *              ...
1638  *      - 0 < N - M => M < N
1639  *
1640  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1641  * always the case (the difference between two distinct xtime instances
1642  * might be smaller then the difference between corresponding TSC reads,
1643  * when updating guest vcpus pvclock areas).
1644  *
1645  * To avoid that problem, do not allow visibility of distinct
1646  * system_timestamp/tsc_timestamp values simultaneously: use a master
1647  * copy of host monotonic time values. Update that master copy
1648  * in lockstep.
1649  *
1650  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1651  *
1652  */
1653
1654 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1655 {
1656 #ifdef CONFIG_X86_64
1657         struct kvm_arch *ka = &kvm->arch;
1658         int vclock_mode;
1659         bool host_tsc_clocksource, vcpus_matched;
1660
1661         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1662                         atomic_read(&kvm->online_vcpus));
1663
1664         /*
1665          * If the host uses TSC clock, then passthrough TSC as stable
1666          * to the guest.
1667          */
1668         host_tsc_clocksource = kvm_get_time_and_clockread(
1669                                         &ka->master_kernel_ns,
1670                                         &ka->master_cycle_now);
1671
1672         ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1673                                 && !backwards_tsc_observed
1674                                 && !ka->boot_vcpu_runs_old_kvmclock;
1675
1676         if (ka->use_master_clock)
1677                 atomic_set(&kvm_guest_has_master_clock, 1);
1678
1679         vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1680         trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1681                                         vcpus_matched);
1682 #endif
1683 }
1684
1685 static void kvm_gen_update_masterclock(struct kvm *kvm)
1686 {
1687 #ifdef CONFIG_X86_64
1688         int i;
1689         struct kvm_vcpu *vcpu;
1690         struct kvm_arch *ka = &kvm->arch;
1691
1692         spin_lock(&ka->pvclock_gtod_sync_lock);
1693         kvm_make_mclock_inprogress_request(kvm);
1694         /* no guest entries from this point */
1695         pvclock_update_vm_gtod_copy(kvm);
1696
1697         kvm_for_each_vcpu(i, vcpu, kvm)
1698                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1699
1700         /* guest entries allowed */
1701         kvm_for_each_vcpu(i, vcpu, kvm)
1702                 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1703
1704         spin_unlock(&ka->pvclock_gtod_sync_lock);
1705 #endif
1706 }
1707
1708 static int kvm_guest_time_update(struct kvm_vcpu *v)
1709 {
1710         unsigned long flags, this_tsc_khz, tgt_tsc_khz;
1711         struct kvm_vcpu_arch *vcpu = &v->arch;
1712         struct kvm_arch *ka = &v->kvm->arch;
1713         s64 kernel_ns;
1714         u64 tsc_timestamp, host_tsc;
1715         struct pvclock_vcpu_time_info guest_hv_clock;
1716         u8 pvclock_flags;
1717         bool use_master_clock;
1718
1719         kernel_ns = 0;
1720         host_tsc = 0;
1721
1722         /*
1723          * If the host uses TSC clock, then passthrough TSC as stable
1724          * to the guest.
1725          */
1726         spin_lock(&ka->pvclock_gtod_sync_lock);
1727         use_master_clock = ka->use_master_clock;
1728         if (use_master_clock) {
1729                 host_tsc = ka->master_cycle_now;
1730                 kernel_ns = ka->master_kernel_ns;
1731         }
1732         spin_unlock(&ka->pvclock_gtod_sync_lock);
1733
1734         /* Keep irq disabled to prevent changes to the clock */
1735         local_irq_save(flags);
1736         this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1737         if (unlikely(this_tsc_khz == 0)) {
1738                 local_irq_restore(flags);
1739                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1740                 return 1;
1741         }
1742         if (!use_master_clock) {
1743                 host_tsc = rdtsc();
1744                 kernel_ns = get_kernel_ns();
1745         }
1746
1747         tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1748
1749         /*
1750          * We may have to catch up the TSC to match elapsed wall clock
1751          * time for two reasons, even if kvmclock is used.
1752          *   1) CPU could have been running below the maximum TSC rate
1753          *   2) Broken TSC compensation resets the base at each VCPU
1754          *      entry to avoid unknown leaps of TSC even when running
1755          *      again on the same CPU.  This may cause apparent elapsed
1756          *      time to disappear, and the guest to stand still or run
1757          *      very slowly.
1758          */
1759         if (vcpu->tsc_catchup) {
1760                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1761                 if (tsc > tsc_timestamp) {
1762                         adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1763                         tsc_timestamp = tsc;
1764                 }
1765         }
1766
1767         local_irq_restore(flags);
1768
1769         if (!vcpu->pv_time_enabled)
1770                 return 0;
1771
1772         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1773                 tgt_tsc_khz = kvm_has_tsc_control ?
1774                         vcpu->virtual_tsc_khz : this_tsc_khz;
1775                 kvm_get_time_scale(NSEC_PER_SEC / 1000, tgt_tsc_khz,
1776                                    &vcpu->hv_clock.tsc_shift,
1777                                    &vcpu->hv_clock.tsc_to_system_mul);
1778                 vcpu->hw_tsc_khz = this_tsc_khz;
1779         }
1780
1781         /* With all the info we got, fill in the values */
1782         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1783         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1784         vcpu->last_guest_tsc = tsc_timestamp;
1785
1786         if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1787                 &guest_hv_clock, sizeof(guest_hv_clock))))
1788                 return 0;
1789
1790         /* This VCPU is paused, but it's legal for a guest to read another
1791          * VCPU's kvmclock, so we really have to follow the specification where
1792          * it says that version is odd if data is being modified, and even after
1793          * it is consistent.
1794          *
1795          * Version field updates must be kept separate.  This is because
1796          * kvm_write_guest_cached might use a "rep movs" instruction, and
1797          * writes within a string instruction are weakly ordered.  So there
1798          * are three writes overall.
1799          *
1800          * As a small optimization, only write the version field in the first
1801          * and third write.  The vcpu->pv_time cache is still valid, because the
1802          * version field is the first in the struct.
1803          */
1804         BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1805
1806         vcpu->hv_clock.version = guest_hv_clock.version + 1;
1807         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1808                                 &vcpu->hv_clock,
1809                                 sizeof(vcpu->hv_clock.version));
1810
1811         smp_wmb();
1812
1813         /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1814         pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1815
1816         if (vcpu->pvclock_set_guest_stopped_request) {
1817                 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1818                 vcpu->pvclock_set_guest_stopped_request = false;
1819         }
1820
1821         /* If the host uses TSC clocksource, then it is stable */
1822         if (use_master_clock)
1823                 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1824
1825         vcpu->hv_clock.flags = pvclock_flags;
1826
1827         trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1828
1829         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1830                                 &vcpu->hv_clock,
1831                                 sizeof(vcpu->hv_clock));
1832
1833         smp_wmb();
1834
1835         vcpu->hv_clock.version++;
1836         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1837                                 &vcpu->hv_clock,
1838                                 sizeof(vcpu->hv_clock.version));
1839         return 0;
1840 }
1841
1842 /*
1843  * kvmclock updates which are isolated to a given vcpu, such as
1844  * vcpu->cpu migration, should not allow system_timestamp from
1845  * the rest of the vcpus to remain static. Otherwise ntp frequency
1846  * correction applies to one vcpu's system_timestamp but not
1847  * the others.
1848  *
1849  * So in those cases, request a kvmclock update for all vcpus.
1850  * We need to rate-limit these requests though, as they can
1851  * considerably slow guests that have a large number of vcpus.
1852  * The time for a remote vcpu to update its kvmclock is bound
1853  * by the delay we use to rate-limit the updates.
1854  */
1855
1856 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1857
1858 static void kvmclock_update_fn(struct work_struct *work)
1859 {
1860         int i;
1861         struct delayed_work *dwork = to_delayed_work(work);
1862         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1863                                            kvmclock_update_work);
1864         struct kvm *kvm = container_of(ka, struct kvm, arch);
1865         struct kvm_vcpu *vcpu;
1866
1867         kvm_for_each_vcpu(i, vcpu, kvm) {
1868                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1869                 kvm_vcpu_kick(vcpu);
1870         }
1871 }
1872
1873 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1874 {
1875         struct kvm *kvm = v->kvm;
1876
1877         kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1878         schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1879                                         KVMCLOCK_UPDATE_DELAY);
1880 }
1881
1882 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1883
1884 static void kvmclock_sync_fn(struct work_struct *work)
1885 {
1886         struct delayed_work *dwork = to_delayed_work(work);
1887         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1888                                            kvmclock_sync_work);
1889         struct kvm *kvm = container_of(ka, struct kvm, arch);
1890
1891         if (!kvmclock_periodic_sync)
1892                 return;
1893
1894         schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1895         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1896                                         KVMCLOCK_SYNC_PERIOD);
1897 }
1898
1899 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1900 {
1901         u64 mcg_cap = vcpu->arch.mcg_cap;
1902         unsigned bank_num = mcg_cap & 0xff;
1903
1904         switch (msr) {
1905         case MSR_IA32_MCG_STATUS:
1906                 vcpu->arch.mcg_status = data;
1907                 break;
1908         case MSR_IA32_MCG_CTL:
1909                 if (!(mcg_cap & MCG_CTL_P))
1910                         return 1;
1911                 if (data != 0 && data != ~(u64)0)
1912                         return -1;
1913                 vcpu->arch.mcg_ctl = data;
1914                 break;
1915         default:
1916                 if (msr >= MSR_IA32_MC0_CTL &&
1917                     msr < MSR_IA32_MCx_CTL(bank_num)) {
1918                         u32 offset = msr - MSR_IA32_MC0_CTL;
1919                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1920                          * some Linux kernels though clear bit 10 in bank 4 to
1921                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1922                          * this to avoid an uncatched #GP in the guest
1923                          */
1924                         if ((offset & 0x3) == 0 &&
1925                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1926                                 return -1;
1927                         vcpu->arch.mce_banks[offset] = data;
1928                         break;
1929                 }
1930                 return 1;
1931         }
1932         return 0;
1933 }
1934
1935 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1936 {
1937         struct kvm *kvm = vcpu->kvm;
1938         int lm = is_long_mode(vcpu);
1939         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1940                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1941         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1942                 : kvm->arch.xen_hvm_config.blob_size_32;
1943         u32 page_num = data & ~PAGE_MASK;
1944         u64 page_addr = data & PAGE_MASK;
1945         u8 *page;
1946         int r;
1947
1948         r = -E2BIG;
1949         if (page_num >= blob_size)
1950                 goto out;
1951         r = -ENOMEM;
1952         page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1953         if (IS_ERR(page)) {
1954                 r = PTR_ERR(page);
1955                 goto out;
1956         }
1957         if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1958                 goto out_free;
1959         r = 0;
1960 out_free:
1961         kfree(page);
1962 out:
1963         return r;
1964 }
1965
1966 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1967 {
1968         gpa_t gpa = data & ~0x3f;
1969
1970         /* Bits 2:5 are reserved, Should be zero */
1971         if (data & 0x3c)
1972                 return 1;
1973
1974         vcpu->arch.apf.msr_val = data;
1975
1976         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1977                 kvm_clear_async_pf_completion_queue(vcpu);
1978                 kvm_async_pf_hash_reset(vcpu);
1979                 return 0;
1980         }
1981
1982         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1983                                         sizeof(u32)))
1984                 return 1;
1985
1986         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1987         kvm_async_pf_wakeup_all(vcpu);
1988         return 0;
1989 }
1990
1991 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1992 {
1993         vcpu->arch.pv_time_enabled = false;
1994 }
1995
1996 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1997 {
1998         u64 delta;
1999
2000         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2001                 return;
2002
2003         delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2004         vcpu->arch.st.last_steal = current->sched_info.run_delay;
2005         vcpu->arch.st.accum_steal = delta;
2006 }
2007
2008 static void record_steal_time(struct kvm_vcpu *vcpu)
2009 {
2010         accumulate_steal_time(vcpu);
2011
2012         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2013                 return;
2014
2015         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2016                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2017                 return;
2018
2019         vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2020         vcpu->arch.st.steal.version += 2;
2021         vcpu->arch.st.accum_steal = 0;
2022
2023         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2024                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2025 }
2026
2027 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2028 {
2029         bool pr = false;
2030         u32 msr = msr_info->index;
2031         u64 data = msr_info->data;
2032
2033         switch (msr) {
2034         case MSR_AMD64_NB_CFG:
2035         case MSR_IA32_UCODE_REV:
2036         case MSR_IA32_UCODE_WRITE:
2037         case MSR_VM_HSAVE_PA:
2038         case MSR_AMD64_PATCH_LOADER:
2039         case MSR_AMD64_BU_CFG2:
2040                 break;
2041
2042         case MSR_EFER:
2043                 return set_efer(vcpu, data);
2044         case MSR_K7_HWCR:
2045                 data &= ~(u64)0x40;     /* ignore flush filter disable */
2046                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
2047                 data &= ~(u64)0x8;      /* ignore TLB cache disable */
2048                 data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2049                 if (data != 0) {
2050                         vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2051                                     data);
2052                         return 1;
2053                 }
2054                 break;
2055         case MSR_FAM10H_MMIO_CONF_BASE:
2056                 if (data != 0) {
2057                         vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2058                                     "0x%llx\n", data);
2059                         return 1;
2060                 }
2061                 break;
2062         case MSR_IA32_DEBUGCTLMSR:
2063                 if (!data) {
2064                         /* We support the non-activated case already */
2065                         break;
2066                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2067                         /* Values other than LBR and BTF are vendor-specific,
2068                            thus reserved and should throw a #GP */
2069                         return 1;
2070                 }
2071                 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2072                             __func__, data);
2073                 break;
2074         case 0x200 ... 0x2ff:
2075                 return kvm_mtrr_set_msr(vcpu, msr, data);
2076         case MSR_IA32_APICBASE:
2077                 return kvm_set_apic_base(vcpu, msr_info);
2078         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2079                 return kvm_x2apic_msr_write(vcpu, msr, data);
2080         case MSR_IA32_TSCDEADLINE:
2081                 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2082                 break;
2083         case MSR_IA32_TSC_ADJUST:
2084                 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2085                         if (!msr_info->host_initiated) {
2086                                 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2087                                 adjust_tsc_offset_guest(vcpu, adj);
2088                         }
2089                         vcpu->arch.ia32_tsc_adjust_msr = data;
2090                 }
2091                 break;
2092         case MSR_IA32_MISC_ENABLE:
2093                 vcpu->arch.ia32_misc_enable_msr = data;
2094                 break;
2095         case MSR_IA32_SMBASE:
2096                 if (!msr_info->host_initiated)
2097                         return 1;
2098                 vcpu->arch.smbase = data;
2099                 break;
2100         case MSR_KVM_WALL_CLOCK_NEW:
2101         case MSR_KVM_WALL_CLOCK:
2102                 vcpu->kvm->arch.wall_clock = data;
2103                 kvm_write_wall_clock(vcpu->kvm, data);
2104                 break;
2105         case MSR_KVM_SYSTEM_TIME_NEW:
2106         case MSR_KVM_SYSTEM_TIME: {
2107                 u64 gpa_offset;
2108                 struct kvm_arch *ka = &vcpu->kvm->arch;
2109
2110                 kvmclock_reset(vcpu);
2111
2112                 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2113                         bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2114
2115                         if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2116                                 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2117                                         &vcpu->requests);
2118
2119                         ka->boot_vcpu_runs_old_kvmclock = tmp;
2120                 }
2121
2122                 vcpu->arch.time = data;
2123                 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2124
2125                 /* we verify if the enable bit is set... */
2126                 if (!(data & 1))
2127                         break;
2128
2129                 gpa_offset = data & ~(PAGE_MASK | 1);
2130
2131                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2132                      &vcpu->arch.pv_time, data & ~1ULL,
2133                      sizeof(struct pvclock_vcpu_time_info)))
2134                         vcpu->arch.pv_time_enabled = false;
2135                 else
2136                         vcpu->arch.pv_time_enabled = true;
2137
2138                 break;
2139         }
2140         case MSR_KVM_ASYNC_PF_EN:
2141                 if (kvm_pv_enable_async_pf(vcpu, data))
2142                         return 1;
2143                 break;
2144         case MSR_KVM_STEAL_TIME:
2145
2146                 if (unlikely(!sched_info_on()))
2147                         return 1;
2148
2149                 if (data & KVM_STEAL_RESERVED_MASK)
2150                         return 1;
2151
2152                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2153                                                 data & KVM_STEAL_VALID_BITS,
2154                                                 sizeof(struct kvm_steal_time)))
2155                         return 1;
2156
2157                 vcpu->arch.st.msr_val = data;
2158
2159                 if (!(data & KVM_MSR_ENABLED))
2160                         break;
2161
2162                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2163
2164                 break;
2165         case MSR_KVM_PV_EOI_EN:
2166                 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2167                         return 1;
2168                 break;
2169
2170         case MSR_IA32_MCG_CTL:
2171         case MSR_IA32_MCG_STATUS:
2172         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2173                 return set_msr_mce(vcpu, msr, data);
2174
2175         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2176         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2177                 pr = true; /* fall through */
2178         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2179         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2180                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2181                         return kvm_pmu_set_msr(vcpu, msr_info);
2182
2183                 if (pr || data != 0)
2184                         vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2185                                     "0x%x data 0x%llx\n", msr, data);
2186                 break;
2187         case MSR_K7_CLK_CTL:
2188                 /*
2189                  * Ignore all writes to this no longer documented MSR.
2190                  * Writes are only relevant for old K7 processors,
2191                  * all pre-dating SVM, but a recommended workaround from
2192                  * AMD for these chips. It is possible to specify the
2193                  * affected processor models on the command line, hence
2194                  * the need to ignore the workaround.
2195                  */
2196                 break;
2197         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2198         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2199         case HV_X64_MSR_CRASH_CTL:
2200                 return kvm_hv_set_msr_common(vcpu, msr, data,
2201                                              msr_info->host_initiated);
2202         case MSR_IA32_BBL_CR_CTL3:
2203                 /* Drop writes to this legacy MSR -- see rdmsr
2204                  * counterpart for further detail.
2205                  */
2206                 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2207                 break;
2208         case MSR_AMD64_OSVW_ID_LENGTH:
2209                 if (!guest_cpuid_has_osvw(vcpu))
2210                         return 1;
2211                 vcpu->arch.osvw.length = data;
2212                 break;
2213         case MSR_AMD64_OSVW_STATUS:
2214                 if (!guest_cpuid_has_osvw(vcpu))
2215                         return 1;
2216                 vcpu->arch.osvw.status = data;
2217                 break;
2218         default:
2219                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2220                         return xen_hvm_config(vcpu, data);
2221                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2222                         return kvm_pmu_set_msr(vcpu, msr_info);
2223                 if (!ignore_msrs) {
2224                         vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2225                                     msr, data);
2226                         return 1;
2227                 } else {
2228                         vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2229                                     msr, data);
2230                         break;
2231                 }
2232         }
2233         return 0;
2234 }
2235 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2236
2237
2238 /*
2239  * Reads an msr value (of 'msr_index') into 'pdata'.
2240  * Returns 0 on success, non-0 otherwise.
2241  * Assumes vcpu_load() was already called.
2242  */
2243 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2244 {
2245         return kvm_x86_ops->get_msr(vcpu, msr);
2246 }
2247 EXPORT_SYMBOL_GPL(kvm_get_msr);
2248
2249 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2250 {
2251         u64 data;
2252         u64 mcg_cap = vcpu->arch.mcg_cap;
2253         unsigned bank_num = mcg_cap & 0xff;
2254
2255         switch (msr) {
2256         case MSR_IA32_P5_MC_ADDR:
2257         case MSR_IA32_P5_MC_TYPE:
2258                 data = 0;
2259                 break;
2260         case MSR_IA32_MCG_CAP:
2261                 data = vcpu->arch.mcg_cap;
2262                 break;
2263         case MSR_IA32_MCG_CTL:
2264                 if (!(mcg_cap & MCG_CTL_P))
2265                         return 1;
2266                 data = vcpu->arch.mcg_ctl;
2267                 break;
2268         case MSR_IA32_MCG_STATUS:
2269                 data = vcpu->arch.mcg_status;
2270                 break;
2271         default:
2272                 if (msr >= MSR_IA32_MC0_CTL &&
2273                     msr < MSR_IA32_MCx_CTL(bank_num)) {
2274                         u32 offset = msr - MSR_IA32_MC0_CTL;
2275                         data = vcpu->arch.mce_banks[offset];
2276                         break;
2277                 }
2278                 return 1;
2279         }
2280         *pdata = data;
2281         return 0;
2282 }
2283
2284 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2285 {
2286         switch (msr_info->index) {
2287         case MSR_IA32_PLATFORM_ID:
2288         case MSR_IA32_EBL_CR_POWERON:
2289         case MSR_IA32_DEBUGCTLMSR:
2290         case MSR_IA32_LASTBRANCHFROMIP:
2291         case MSR_IA32_LASTBRANCHTOIP:
2292         case MSR_IA32_LASTINTFROMIP:
2293         case MSR_IA32_LASTINTTOIP:
2294         case MSR_K8_SYSCFG:
2295         case MSR_K8_TSEG_ADDR:
2296         case MSR_K8_TSEG_MASK:
2297         case MSR_K7_HWCR:
2298         case MSR_VM_HSAVE_PA:
2299         case MSR_K8_INT_PENDING_MSG:
2300         case MSR_AMD64_NB_CFG:
2301         case MSR_FAM10H_MMIO_CONF_BASE:
2302         case MSR_AMD64_BU_CFG2:
2303                 msr_info->data = 0;
2304                 break;
2305         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2306         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2307         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2308         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2309                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2310                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2311                 msr_info->data = 0;
2312                 break;
2313         case MSR_IA32_UCODE_REV:
2314                 msr_info->data = 0x100000000ULL;
2315                 break;
2316         case MSR_MTRRcap:
2317         case 0x200 ... 0x2ff:
2318                 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2319         case 0xcd: /* fsb frequency */
2320                 msr_info->data = 3;
2321                 break;
2322                 /*
2323                  * MSR_EBC_FREQUENCY_ID
2324                  * Conservative value valid for even the basic CPU models.
2325                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2326                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2327                  * and 266MHz for model 3, or 4. Set Core Clock
2328                  * Frequency to System Bus Frequency Ratio to 1 (bits
2329                  * 31:24) even though these are only valid for CPU
2330                  * models > 2, however guests may end up dividing or
2331                  * multiplying by zero otherwise.
2332                  */
2333         case MSR_EBC_FREQUENCY_ID:
2334                 msr_info->data = 1 << 24;
2335                 break;
2336         case MSR_IA32_APICBASE:
2337                 msr_info->data = kvm_get_apic_base(vcpu);
2338                 break;
2339         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2340                 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2341                 break;
2342         case MSR_IA32_TSCDEADLINE:
2343                 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2344                 break;
2345         case MSR_IA32_TSC_ADJUST:
2346                 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2347                 break;
2348         case MSR_IA32_MISC_ENABLE:
2349                 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2350                 break;
2351         case MSR_IA32_SMBASE:
2352                 if (!msr_info->host_initiated)
2353                         return 1;
2354                 msr_info->data = vcpu->arch.smbase;
2355                 break;
2356         case MSR_IA32_PERF_STATUS:
2357                 /* TSC increment by tick */
2358                 msr_info->data = 1000ULL;
2359                 /* CPU multiplier */
2360                 msr_info->data |= (((uint64_t)4ULL) << 40);
2361                 break;
2362         case MSR_EFER:
2363                 msr_info->data = vcpu->arch.efer;
2364                 break;
2365         case MSR_KVM_WALL_CLOCK:
2366         case MSR_KVM_WALL_CLOCK_NEW:
2367                 msr_info->data = vcpu->kvm->arch.wall_clock;
2368                 break;
2369         case MSR_KVM_SYSTEM_TIME:
2370         case MSR_KVM_SYSTEM_TIME_NEW:
2371                 msr_info->data = vcpu->arch.time;
2372                 break;
2373         case MSR_KVM_ASYNC_PF_EN:
2374                 msr_info->data = vcpu->arch.apf.msr_val;
2375                 break;
2376         case MSR_KVM_STEAL_TIME:
2377                 msr_info->data = vcpu->arch.st.msr_val;
2378                 break;
2379         case MSR_KVM_PV_EOI_EN:
2380                 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2381                 break;
2382         case MSR_IA32_P5_MC_ADDR:
2383         case MSR_IA32_P5_MC_TYPE:
2384         case MSR_IA32_MCG_CAP:
2385         case MSR_IA32_MCG_CTL:
2386         case MSR_IA32_MCG_STATUS:
2387         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2388                 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2389         case MSR_K7_CLK_CTL:
2390                 /*
2391                  * Provide expected ramp-up count for K7. All other
2392                  * are set to zero, indicating minimum divisors for
2393                  * every field.
2394                  *
2395                  * This prevents guest kernels on AMD host with CPU
2396                  * type 6, model 8 and higher from exploding due to
2397                  * the rdmsr failing.
2398                  */
2399                 msr_info->data = 0x20000000;
2400                 break;
2401         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2402         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2403         case HV_X64_MSR_CRASH_CTL:
2404                 return kvm_hv_get_msr_common(vcpu,
2405                                              msr_info->index, &msr_info->data);
2406                 break;
2407         case MSR_IA32_BBL_CR_CTL3:
2408                 /* This legacy MSR exists but isn't fully documented in current
2409                  * silicon.  It is however accessed by winxp in very narrow
2410                  * scenarios where it sets bit #19, itself documented as
2411                  * a "reserved" bit.  Best effort attempt to source coherent
2412                  * read data here should the balance of the register be
2413                  * interpreted by the guest:
2414                  *
2415                  * L2 cache control register 3: 64GB range, 256KB size,
2416                  * enabled, latency 0x1, configured
2417                  */
2418                 msr_info->data = 0xbe702111;
2419                 break;
2420         case MSR_AMD64_OSVW_ID_LENGTH:
2421                 if (!guest_cpuid_has_osvw(vcpu))
2422                         return 1;
2423                 msr_info->data = vcpu->arch.osvw.length;
2424                 break;
2425         case MSR_AMD64_OSVW_STATUS:
2426                 if (!guest_cpuid_has_osvw(vcpu))
2427                         return 1;
2428                 msr_info->data = vcpu->arch.osvw.status;
2429                 break;
2430         default:
2431                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2432                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2433                 if (!ignore_msrs) {
2434                         vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2435                         return 1;
2436                 } else {
2437                         vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2438                         msr_info->data = 0;
2439                 }
2440                 break;
2441         }
2442         return 0;
2443 }
2444 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2445
2446 /*
2447  * Read or write a bunch of msrs. All parameters are kernel addresses.
2448  *
2449  * @return number of msrs set successfully.
2450  */
2451 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2452                     struct kvm_msr_entry *entries,
2453                     int (*do_msr)(struct kvm_vcpu *vcpu,
2454                                   unsigned index, u64 *data))
2455 {
2456         int i, idx;
2457
2458         idx = srcu_read_lock(&vcpu->kvm->srcu);
2459         for (i = 0; i < msrs->nmsrs; ++i)
2460                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2461                         break;
2462         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2463
2464         return i;
2465 }
2466
2467 /*
2468  * Read or write a bunch of msrs. Parameters are user addresses.
2469  *
2470  * @return number of msrs set successfully.
2471  */
2472 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2473                   int (*do_msr)(struct kvm_vcpu *vcpu,
2474                                 unsigned index, u64 *data),
2475                   int writeback)
2476 {
2477         struct kvm_msrs msrs;
2478         struct kvm_msr_entry *entries;
2479         int r, n;
2480         unsigned size;
2481
2482         r = -EFAULT;
2483         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2484                 goto out;
2485
2486         r = -E2BIG;
2487         if (msrs.nmsrs >= MAX_IO_MSRS)
2488                 goto out;
2489
2490         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2491         entries = memdup_user(user_msrs->entries, size);
2492         if (IS_ERR(entries)) {
2493                 r = PTR_ERR(entries);
2494                 goto out;
2495         }
2496
2497         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2498         if (r < 0)
2499                 goto out_free;
2500
2501         r = -EFAULT;
2502         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2503                 goto out_free;
2504
2505         r = n;
2506
2507 out_free:
2508         kfree(entries);
2509 out:
2510         return r;
2511 }
2512
2513 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2514 {
2515         int r;
2516
2517         switch (ext) {
2518         case KVM_CAP_IRQCHIP:
2519         case KVM_CAP_HLT:
2520         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2521         case KVM_CAP_SET_TSS_ADDR:
2522         case KVM_CAP_EXT_CPUID:
2523         case KVM_CAP_EXT_EMUL_CPUID:
2524         case KVM_CAP_CLOCKSOURCE:
2525         case KVM_CAP_PIT:
2526         case KVM_CAP_NOP_IO_DELAY:
2527         case KVM_CAP_MP_STATE:
2528         case KVM_CAP_SYNC_MMU:
2529         case KVM_CAP_USER_NMI:
2530         case KVM_CAP_REINJECT_CONTROL:
2531         case KVM_CAP_IRQ_INJECT_STATUS:
2532         case KVM_CAP_IOEVENTFD:
2533         case KVM_CAP_IOEVENTFD_NO_LENGTH:
2534         case KVM_CAP_PIT2:
2535         case KVM_CAP_PIT_STATE2:
2536         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2537         case KVM_CAP_XEN_HVM:
2538         case KVM_CAP_ADJUST_CLOCK:
2539         case KVM_CAP_VCPU_EVENTS:
2540         case KVM_CAP_HYPERV:
2541         case KVM_CAP_HYPERV_VAPIC:
2542         case KVM_CAP_HYPERV_SPIN:
2543         case KVM_CAP_PCI_SEGMENT:
2544         case KVM_CAP_DEBUGREGS:
2545         case KVM_CAP_X86_ROBUST_SINGLESTEP:
2546         case KVM_CAP_XSAVE:
2547         case KVM_CAP_ASYNC_PF:
2548         case KVM_CAP_GET_TSC_KHZ:
2549         case KVM_CAP_KVMCLOCK_CTRL:
2550         case KVM_CAP_READONLY_MEM:
2551         case KVM_CAP_HYPERV_TIME:
2552         case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2553         case KVM_CAP_TSC_DEADLINE_TIMER:
2554         case KVM_CAP_ENABLE_CAP_VM:
2555         case KVM_CAP_DISABLE_QUIRKS:
2556         case KVM_CAP_SET_BOOT_CPU_ID:
2557         case KVM_CAP_SPLIT_IRQCHIP:
2558 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2559         case KVM_CAP_ASSIGN_DEV_IRQ:
2560         case KVM_CAP_PCI_2_3:
2561 #endif
2562                 r = 1;
2563                 break;
2564         case KVM_CAP_X86_SMM:
2565                 /* SMBASE is usually relocated above 1M on modern chipsets,
2566                  * and SMM handlers might indeed rely on 4G segment limits,
2567                  * so do not report SMM to be available if real mode is
2568                  * emulated via vm86 mode.  Still, do not go to great lengths
2569                  * to avoid userspace's usage of the feature, because it is a
2570                  * fringe case that is not enabled except via specific settings
2571                  * of the module parameters.
2572                  */
2573                 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2574                 break;
2575         case KVM_CAP_COALESCED_MMIO:
2576                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2577                 break;
2578         case KVM_CAP_VAPIC:
2579                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2580                 break;
2581         case KVM_CAP_NR_VCPUS:
2582                 r = KVM_SOFT_MAX_VCPUS;
2583                 break;
2584         case KVM_CAP_MAX_VCPUS:
2585                 r = KVM_MAX_VCPUS;
2586                 break;
2587         case KVM_CAP_NR_MEMSLOTS:
2588                 r = KVM_USER_MEM_SLOTS;
2589                 break;
2590         case KVM_CAP_PV_MMU:    /* obsolete */
2591                 r = 0;
2592                 break;
2593 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2594         case KVM_CAP_IOMMU:
2595                 r = iommu_present(&pci_bus_type);
2596                 break;
2597 #endif
2598         case KVM_CAP_MCE:
2599                 r = KVM_MAX_MCE_BANKS;
2600                 break;
2601         case KVM_CAP_XCRS:
2602                 r = cpu_has_xsave;
2603                 break;
2604         case KVM_CAP_TSC_CONTROL:
2605                 r = kvm_has_tsc_control;
2606                 break;
2607         default:
2608                 r = 0;
2609                 break;
2610         }
2611         return r;
2612
2613 }
2614
2615 long kvm_arch_dev_ioctl(struct file *filp,
2616                         unsigned int ioctl, unsigned long arg)
2617 {
2618         void __user *argp = (void __user *)arg;
2619         long r;
2620
2621         switch (ioctl) {
2622         case KVM_GET_MSR_INDEX_LIST: {
2623                 struct kvm_msr_list __user *user_msr_list = argp;
2624                 struct kvm_msr_list msr_list;
2625                 unsigned n;
2626
2627                 r = -EFAULT;
2628                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2629                         goto out;
2630                 n = msr_list.nmsrs;
2631                 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2632                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2633                         goto out;
2634                 r = -E2BIG;
2635                 if (n < msr_list.nmsrs)
2636                         goto out;
2637                 r = -EFAULT;
2638                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2639                                  num_msrs_to_save * sizeof(u32)))
2640                         goto out;
2641                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2642                                  &emulated_msrs,
2643                                  num_emulated_msrs * sizeof(u32)))
2644                         goto out;
2645                 r = 0;
2646                 break;
2647         }
2648         case KVM_GET_SUPPORTED_CPUID:
2649         case KVM_GET_EMULATED_CPUID: {
2650                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2651                 struct kvm_cpuid2 cpuid;
2652
2653                 r = -EFAULT;
2654                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2655                         goto out;
2656
2657                 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2658                                             ioctl);
2659                 if (r)
2660                         goto out;
2661
2662                 r = -EFAULT;
2663                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2664                         goto out;
2665                 r = 0;
2666                 break;
2667         }
2668         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2669                 u64 mce_cap;
2670
2671                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2672                 r = -EFAULT;
2673                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2674                         goto out;
2675                 r = 0;
2676                 break;
2677         }
2678         default:
2679                 r = -EINVAL;
2680         }
2681 out:
2682         return r;
2683 }
2684
2685 static void wbinvd_ipi(void *garbage)
2686 {
2687         wbinvd();
2688 }
2689
2690 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2691 {
2692         return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2693 }
2694
2695 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2696 {
2697         /* Address WBINVD may be executed by guest */
2698         if (need_emulate_wbinvd(vcpu)) {
2699                 if (kvm_x86_ops->has_wbinvd_exit())
2700                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2701                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2702                         smp_call_function_single(vcpu->cpu,
2703                                         wbinvd_ipi, NULL, 1);
2704         }
2705
2706         kvm_x86_ops->vcpu_load(vcpu, cpu);
2707
2708         /* Apply any externally detected TSC adjustments (due to suspend) */
2709         if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2710                 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2711                 vcpu->arch.tsc_offset_adjustment = 0;
2712                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2713         }
2714
2715         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2716                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2717                                 rdtsc() - vcpu->arch.last_host_tsc;
2718                 if (tsc_delta < 0)
2719                         mark_tsc_unstable("KVM discovered backwards TSC");
2720                 if (check_tsc_unstable()) {
2721                         u64 offset = kvm_compute_tsc_offset(vcpu,
2722                                                 vcpu->arch.last_guest_tsc);
2723                         kvm_x86_ops->write_tsc_offset(vcpu, offset);
2724                         vcpu->arch.tsc_catchup = 1;
2725                 }
2726                 /*
2727                  * On a host with synchronized TSC, there is no need to update
2728                  * kvmclock on vcpu->cpu migration
2729                  */
2730                 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2731                         kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2732                 if (vcpu->cpu != cpu)
2733                         kvm_migrate_timers(vcpu);
2734                 vcpu->cpu = cpu;
2735         }
2736
2737         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2738         vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
2739 }
2740
2741 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2742 {
2743         kvm_x86_ops->vcpu_put(vcpu);
2744         kvm_put_guest_fpu(vcpu);
2745         vcpu->arch.last_host_tsc = rdtsc();
2746 }
2747
2748 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2749                                     struct kvm_lapic_state *s)
2750 {
2751         kvm_x86_ops->sync_pir_to_irr(vcpu);
2752         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2753
2754         return 0;
2755 }
2756
2757 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2758                                     struct kvm_lapic_state *s)
2759 {
2760         kvm_apic_post_state_restore(vcpu, s);
2761         update_cr8_intercept(vcpu);
2762
2763         return 0;
2764 }
2765
2766 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2767 {
2768         return (!lapic_in_kernel(vcpu) ||
2769                 kvm_apic_accept_pic_intr(vcpu));
2770 }
2771
2772 /*
2773  * if userspace requested an interrupt window, check that the
2774  * interrupt window is open.
2775  *
2776  * No need to exit to userspace if we already have an interrupt queued.
2777  */
2778 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2779 {
2780         return kvm_arch_interrupt_allowed(vcpu) &&
2781                 !kvm_cpu_has_interrupt(vcpu) &&
2782                 !kvm_event_needs_reinjection(vcpu) &&
2783                 kvm_cpu_accept_dm_intr(vcpu);
2784 }
2785
2786 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2787                                     struct kvm_interrupt *irq)
2788 {
2789         if (irq->irq >= KVM_NR_INTERRUPTS)
2790                 return -EINVAL;
2791
2792         if (!irqchip_in_kernel(vcpu->kvm)) {
2793                 kvm_queue_interrupt(vcpu, irq->irq, false);
2794                 kvm_make_request(KVM_REQ_EVENT, vcpu);
2795                 return 0;
2796         }
2797
2798         /*
2799          * With in-kernel LAPIC, we only use this to inject EXTINT, so
2800          * fail for in-kernel 8259.
2801          */
2802         if (pic_in_kernel(vcpu->kvm))
2803                 return -ENXIO;
2804
2805         if (vcpu->arch.pending_external_vector != -1)
2806                 return -EEXIST;
2807
2808         vcpu->arch.pending_external_vector = irq->irq;
2809         kvm_make_request(KVM_REQ_EVENT, vcpu);
2810         return 0;
2811 }
2812
2813 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2814 {
2815         kvm_inject_nmi(vcpu);
2816
2817         return 0;
2818 }
2819
2820 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2821 {
2822         kvm_make_request(KVM_REQ_SMI, vcpu);
2823
2824         return 0;
2825 }
2826
2827 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2828                                            struct kvm_tpr_access_ctl *tac)
2829 {
2830         if (tac->flags)
2831                 return -EINVAL;
2832         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2833         return 0;
2834 }
2835
2836 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2837                                         u64 mcg_cap)
2838 {
2839         int r;
2840         unsigned bank_num = mcg_cap & 0xff, bank;
2841
2842         r = -EINVAL;
2843         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2844                 goto out;
2845         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2846                 goto out;
2847         r = 0;
2848         vcpu->arch.mcg_cap = mcg_cap;
2849         /* Init IA32_MCG_CTL to all 1s */
2850         if (mcg_cap & MCG_CTL_P)
2851                 vcpu->arch.mcg_ctl = ~(u64)0;
2852         /* Init IA32_MCi_CTL to all 1s */
2853         for (bank = 0; bank < bank_num; bank++)
2854                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2855 out:
2856         return r;
2857 }
2858
2859 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2860                                       struct kvm_x86_mce *mce)
2861 {
2862         u64 mcg_cap = vcpu->arch.mcg_cap;
2863         unsigned bank_num = mcg_cap & 0xff;
2864         u64 *banks = vcpu->arch.mce_banks;
2865
2866         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2867                 return -EINVAL;
2868         /*
2869          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2870          * reporting is disabled
2871          */
2872         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2873             vcpu->arch.mcg_ctl != ~(u64)0)
2874                 return 0;
2875         banks += 4 * mce->bank;
2876         /*
2877          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2878          * reporting is disabled for the bank
2879          */
2880         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2881                 return 0;
2882         if (mce->status & MCI_STATUS_UC) {
2883                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2884                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2885                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2886                         return 0;
2887                 }
2888                 if (banks[1] & MCI_STATUS_VAL)
2889                         mce->status |= MCI_STATUS_OVER;
2890                 banks[2] = mce->addr;
2891                 banks[3] = mce->misc;
2892                 vcpu->arch.mcg_status = mce->mcg_status;
2893                 banks[1] = mce->status;
2894                 kvm_queue_exception(vcpu, MC_VECTOR);
2895         } else if (!(banks[1] & MCI_STATUS_VAL)
2896                    || !(banks[1] & MCI_STATUS_UC)) {
2897                 if (banks[1] & MCI_STATUS_VAL)
2898                         mce->status |= MCI_STATUS_OVER;
2899                 banks[2] = mce->addr;
2900                 banks[3] = mce->misc;
2901                 banks[1] = mce->status;
2902         } else
2903                 banks[1] |= MCI_STATUS_OVER;
2904         return 0;
2905 }
2906
2907 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2908                                                struct kvm_vcpu_events *events)
2909 {
2910         process_nmi(vcpu);
2911         events->exception.injected =
2912                 vcpu->arch.exception.pending &&
2913                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2914         events->exception.nr = vcpu->arch.exception.nr;
2915         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2916         events->exception.pad = 0;
2917         events->exception.error_code = vcpu->arch.exception.error_code;
2918
2919         events->interrupt.injected =
2920                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2921         events->interrupt.nr = vcpu->arch.interrupt.nr;
2922         events->interrupt.soft = 0;
2923         events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2924
2925         events->nmi.injected = vcpu->arch.nmi_injected;
2926         events->nmi.pending = vcpu->arch.nmi_pending != 0;
2927         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2928         events->nmi.pad = 0;
2929
2930         events->sipi_vector = 0; /* never valid when reporting to user space */
2931
2932         events->smi.smm = is_smm(vcpu);
2933         events->smi.pending = vcpu->arch.smi_pending;
2934         events->smi.smm_inside_nmi =
2935                 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2936         events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2937
2938         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2939                          | KVM_VCPUEVENT_VALID_SHADOW
2940                          | KVM_VCPUEVENT_VALID_SMM);
2941         memset(&events->reserved, 0, sizeof(events->reserved));
2942 }
2943
2944 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2945                                               struct kvm_vcpu_events *events)
2946 {
2947         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2948                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2949                               | KVM_VCPUEVENT_VALID_SHADOW
2950                               | KVM_VCPUEVENT_VALID_SMM))
2951                 return -EINVAL;
2952
2953         process_nmi(vcpu);
2954         vcpu->arch.exception.pending = events->exception.injected;
2955         vcpu->arch.exception.nr = events->exception.nr;
2956         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2957         vcpu->arch.exception.error_code = events->exception.error_code;
2958
2959         vcpu->arch.interrupt.pending = events->interrupt.injected;
2960         vcpu->arch.interrupt.nr = events->interrupt.nr;
2961         vcpu->arch.interrupt.soft = events->interrupt.soft;
2962         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2963                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2964                                                   events->interrupt.shadow);
2965
2966         vcpu->arch.nmi_injected = events->nmi.injected;
2967         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2968                 vcpu->arch.nmi_pending = events->nmi.pending;
2969         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2970
2971         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2972             kvm_vcpu_has_lapic(vcpu))
2973                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2974
2975         if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2976                 if (events->smi.smm)
2977                         vcpu->arch.hflags |= HF_SMM_MASK;
2978                 else
2979                         vcpu->arch.hflags &= ~HF_SMM_MASK;
2980                 vcpu->arch.smi_pending = events->smi.pending;
2981                 if (events->smi.smm_inside_nmi)
2982                         vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2983                 else
2984                         vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2985                 if (kvm_vcpu_has_lapic(vcpu)) {
2986                         if (events->smi.latched_init)
2987                                 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2988                         else
2989                                 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2990                 }
2991         }
2992
2993         kvm_make_request(KVM_REQ_EVENT, vcpu);
2994
2995         return 0;
2996 }
2997
2998 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2999                                              struct kvm_debugregs *dbgregs)
3000 {
3001         unsigned long val;
3002
3003         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3004         kvm_get_dr(vcpu, 6, &val);
3005         dbgregs->dr6 = val;
3006         dbgregs->dr7 = vcpu->arch.dr7;
3007         dbgregs->flags = 0;
3008         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3009 }
3010
3011 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3012                                             struct kvm_debugregs *dbgregs)
3013 {
3014         if (dbgregs->flags)
3015                 return -EINVAL;
3016
3017         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3018         kvm_update_dr0123(vcpu);
3019         vcpu->arch.dr6 = dbgregs->dr6;
3020         kvm_update_dr6(vcpu);
3021         vcpu->arch.dr7 = dbgregs->dr7;
3022         kvm_update_dr7(vcpu);
3023
3024         return 0;
3025 }
3026
3027 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3028
3029 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3030 {
3031         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3032         u64 xstate_bv = xsave->header.xfeatures;
3033         u64 valid;
3034
3035         /*
3036          * Copy legacy XSAVE area, to avoid complications with CPUID
3037          * leaves 0 and 1 in the loop below.
3038          */
3039         memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3040
3041         /* Set XSTATE_BV */
3042         *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3043
3044         /*
3045          * Copy each region from the possibly compacted offset to the
3046          * non-compacted offset.
3047          */
3048         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3049         while (valid) {
3050                 u64 feature = valid & -valid;
3051                 int index = fls64(feature) - 1;
3052                 void *src = get_xsave_addr(xsave, feature);
3053
3054                 if (src) {
3055                         u32 size, offset, ecx, edx;
3056                         cpuid_count(XSTATE_CPUID, index,
3057                                     &size, &offset, &ecx, &edx);
3058                         memcpy(dest + offset, src, size);
3059                 }
3060
3061                 valid -= feature;
3062         }
3063 }
3064
3065 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3066 {
3067         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3068         u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3069         u64 valid;
3070
3071         /*
3072          * Copy legacy XSAVE area, to avoid complications with CPUID
3073          * leaves 0 and 1 in the loop below.
3074          */
3075         memcpy(xsave, src, XSAVE_HDR_OFFSET);
3076
3077         /* Set XSTATE_BV and possibly XCOMP_BV.  */
3078         xsave->header.xfeatures = xstate_bv;
3079         if (cpu_has_xsaves)
3080                 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3081
3082         /*
3083          * Copy each region from the non-compacted offset to the
3084          * possibly compacted offset.
3085          */
3086         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3087         while (valid) {
3088                 u64 feature = valid & -valid;
3089                 int index = fls64(feature) - 1;
3090                 void *dest = get_xsave_addr(xsave, feature);
3091
3092                 if (dest) {
3093                         u32 size, offset, ecx, edx;
3094                         cpuid_count(XSTATE_CPUID, index,
3095                                     &size, &offset, &ecx, &edx);
3096                         memcpy(dest, src + offset, size);
3097                 }
3098
3099                 valid -= feature;
3100         }
3101 }
3102
3103 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3104                                          struct kvm_xsave *guest_xsave)
3105 {
3106         if (cpu_has_xsave) {
3107                 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3108                 fill_xsave((u8 *) guest_xsave->region, vcpu);
3109         } else {
3110                 memcpy(guest_xsave->region,
3111                         &vcpu->arch.guest_fpu.state.fxsave,
3112                         sizeof(struct fxregs_state));
3113                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3114                         XFEATURE_MASK_FPSSE;
3115         }
3116 }
3117
3118 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3119                                         struct kvm_xsave *guest_xsave)
3120 {
3121         u64 xstate_bv =
3122                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3123
3124         if (cpu_has_xsave) {
3125                 /*
3126                  * Here we allow setting states that are not present in
3127                  * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3128                  * with old userspace.
3129                  */
3130                 if (xstate_bv & ~kvm_supported_xcr0())
3131                         return -EINVAL;
3132                 load_xsave(vcpu, (u8 *)guest_xsave->region);
3133         } else {
3134                 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3135                         return -EINVAL;
3136                 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3137                         guest_xsave->region, sizeof(struct fxregs_state));
3138         }
3139         return 0;
3140 }
3141
3142 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3143                                         struct kvm_xcrs *guest_xcrs)
3144 {
3145         if (!cpu_has_xsave) {
3146                 guest_xcrs->nr_xcrs = 0;
3147                 return;
3148         }
3149
3150         guest_xcrs->nr_xcrs = 1;
3151         guest_xcrs->flags = 0;
3152         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3153         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3154 }
3155
3156 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3157                                        struct kvm_xcrs *guest_xcrs)
3158 {
3159         int i, r = 0;
3160
3161         if (!cpu_has_xsave)
3162                 return -EINVAL;
3163
3164         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3165                 return -EINVAL;
3166
3167         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3168                 /* Only support XCR0 currently */
3169                 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3170                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3171                                 guest_xcrs->xcrs[i].value);
3172                         break;
3173                 }
3174         if (r)
3175                 r = -EINVAL;
3176         return r;
3177 }
3178
3179 /*
3180  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3181  * stopped by the hypervisor.  This function will be called from the host only.
3182  * EINVAL is returned when the host attempts to set the flag for a guest that
3183  * does not support pv clocks.
3184  */
3185 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3186 {
3187         if (!vcpu->arch.pv_time_enabled)
3188                 return -EINVAL;
3189         vcpu->arch.pvclock_set_guest_stopped_request = true;
3190         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3191         return 0;
3192 }
3193
3194 long kvm_arch_vcpu_ioctl(struct file *filp,
3195                          unsigned int ioctl, unsigned long arg)
3196 {
3197         struct kvm_vcpu *vcpu = filp->private_data;
3198         void __user *argp = (void __user *)arg;
3199         int r;
3200         union {
3201                 struct kvm_lapic_state *lapic;
3202                 struct kvm_xsave *xsave;
3203                 struct kvm_xcrs *xcrs;
3204                 void *buffer;
3205         } u;
3206
3207         u.buffer = NULL;
3208         switch (ioctl) {
3209         case KVM_GET_LAPIC: {
3210                 r = -EINVAL;
3211                 if (!vcpu->arch.apic)
3212                         goto out;
3213                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3214
3215                 r = -ENOMEM;
3216                 if (!u.lapic)
3217                         goto out;
3218                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3219                 if (r)
3220                         goto out;
3221                 r = -EFAULT;
3222                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3223                         goto out;
3224                 r = 0;
3225                 break;
3226         }
3227         case KVM_SET_LAPIC: {
3228                 r = -EINVAL;
3229                 if (!vcpu->arch.apic)
3230                         goto out;
3231                 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3232                 if (IS_ERR(u.lapic))
3233                         return PTR_ERR(u.lapic);
3234
3235                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3236                 break;
3237         }
3238         case KVM_INTERRUPT: {
3239                 struct kvm_interrupt irq;
3240
3241                 r = -EFAULT;
3242                 if (copy_from_user(&irq, argp, sizeof irq))
3243                         goto out;
3244                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3245                 break;
3246         }
3247         case KVM_NMI: {
3248                 r = kvm_vcpu_ioctl_nmi(vcpu);
3249                 break;
3250         }
3251         case KVM_SMI: {
3252                 r = kvm_vcpu_ioctl_smi(vcpu);
3253                 break;
3254         }
3255         case KVM_SET_CPUID: {
3256                 struct kvm_cpuid __user *cpuid_arg = argp;
3257                 struct kvm_cpuid cpuid;
3258
3259                 r = -EFAULT;
3260                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3261                         goto out;
3262                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3263                 break;
3264         }
3265         case KVM_SET_CPUID2: {
3266                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3267                 struct kvm_cpuid2 cpuid;
3268
3269                 r = -EFAULT;
3270                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3271                         goto out;
3272                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3273                                               cpuid_arg->entries);
3274                 break;
3275         }
3276         case KVM_GET_CPUID2: {
3277                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3278                 struct kvm_cpuid2 cpuid;
3279
3280                 r = -EFAULT;
3281                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3282                         goto out;
3283                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3284                                               cpuid_arg->entries);
3285                 if (r)
3286                         goto out;
3287                 r = -EFAULT;
3288                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3289                         goto out;
3290                 r = 0;
3291                 break;
3292         }
3293         case KVM_GET_MSRS:
3294                 r = msr_io(vcpu, argp, do_get_msr, 1);
3295                 break;
3296         case KVM_SET_MSRS:
3297                 r = msr_io(vcpu, argp, do_set_msr, 0);
3298                 break;
3299         case KVM_TPR_ACCESS_REPORTING: {
3300                 struct kvm_tpr_access_ctl tac;
3301
3302                 r = -EFAULT;
3303                 if (copy_from_user(&tac, argp, sizeof tac))
3304                         goto out;
3305                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3306                 if (r)
3307                         goto out;
3308                 r = -EFAULT;
3309                 if (copy_to_user(argp, &tac, sizeof tac))
3310                         goto out;
3311                 r = 0;
3312                 break;
3313         };
3314         case KVM_SET_VAPIC_ADDR: {
3315                 struct kvm_vapic_addr va;
3316
3317                 r = -EINVAL;
3318                 if (!lapic_in_kernel(vcpu))
3319                         goto out;
3320                 r = -EFAULT;
3321                 if (copy_from_user(&va, argp, sizeof va))
3322                         goto out;
3323                 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3324                 break;
3325         }
3326         case KVM_X86_SETUP_MCE: {
3327                 u64 mcg_cap;
3328
3329                 r = -EFAULT;
3330                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3331                         goto out;
3332                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3333                 break;
3334         }
3335         case KVM_X86_SET_MCE: {
3336                 struct kvm_x86_mce mce;
3337
3338                 r = -EFAULT;
3339                 if (copy_from_user(&mce, argp, sizeof mce))
3340                         goto out;
3341                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3342                 break;
3343         }
3344         case KVM_GET_VCPU_EVENTS: {
3345                 struct kvm_vcpu_events events;
3346
3347                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3348
3349                 r = -EFAULT;
3350                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3351                         break;
3352                 r = 0;
3353                 break;
3354         }
3355         case KVM_SET_VCPU_EVENTS: {
3356                 struct kvm_vcpu_events events;
3357
3358                 r = -EFAULT;
3359                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3360                         break;
3361
3362                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3363                 break;
3364         }
3365         case KVM_GET_DEBUGREGS: {
3366                 struct kvm_debugregs dbgregs;
3367
3368                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3369
3370                 r = -EFAULT;
3371                 if (copy_to_user(argp, &dbgregs,
3372                                  sizeof(struct kvm_debugregs)))
3373                         break;
3374                 r = 0;
3375                 break;
3376         }
3377         case KVM_SET_DEBUGREGS: {
3378                 struct kvm_debugregs dbgregs;
3379
3380                 r = -EFAULT;
3381                 if (copy_from_user(&dbgregs, argp,
3382                                    sizeof(struct kvm_debugregs)))
3383                         break;
3384
3385                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3386                 break;
3387         }
3388         case KVM_GET_XSAVE: {
3389                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3390                 r = -ENOMEM;
3391                 if (!u.xsave)
3392                         break;
3393
3394                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3395
3396                 r = -EFAULT;
3397                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3398                         break;
3399                 r = 0;
3400                 break;
3401         }
3402         case KVM_SET_XSAVE: {
3403                 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3404                 if (IS_ERR(u.xsave))
3405                         return PTR_ERR(u.xsave);
3406
3407                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3408                 break;
3409         }
3410         case KVM_GET_XCRS: {
3411                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3412                 r = -ENOMEM;
3413                 if (!u.xcrs)
3414                         break;
3415
3416                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3417
3418                 r = -EFAULT;
3419                 if (copy_to_user(argp, u.xcrs,
3420                                  sizeof(struct kvm_xcrs)))
3421                         break;
3422                 r = 0;
3423                 break;
3424         }
3425         case KVM_SET_XCRS: {
3426                 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3427                 if (IS_ERR(u.xcrs))
3428                         return PTR_ERR(u.xcrs);
3429
3430                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3431                 break;
3432         }
3433         case KVM_SET_TSC_KHZ: {
3434                 u32 user_tsc_khz;
3435
3436                 r = -EINVAL;
3437                 user_tsc_khz = (u32)arg;
3438
3439                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3440                         goto out;
3441
3442                 if (user_tsc_khz == 0)
3443                         user_tsc_khz = tsc_khz;
3444
3445                 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3446                         r = 0;
3447
3448                 goto out;
3449         }
3450         case KVM_GET_TSC_KHZ: {
3451                 r = vcpu->arch.virtual_tsc_khz;
3452                 goto out;
3453         }
3454         case KVM_KVMCLOCK_CTRL: {
3455                 r = kvm_set_guest_paused(vcpu);
3456                 goto out;
3457         }
3458         default:
3459                 r = -EINVAL;
3460         }
3461 out:
3462         kfree(u.buffer);
3463         return r;
3464 }
3465
3466 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3467 {
3468         return VM_FAULT_SIGBUS;
3469 }
3470
3471 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3472 {
3473         int ret;
3474
3475         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3476                 return -EINVAL;
3477         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3478         return ret;
3479 }
3480
3481 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3482                                               u64 ident_addr)
3483 {
3484         kvm->arch.ept_identity_map_addr = ident_addr;
3485         return 0;
3486 }
3487
3488 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3489                                           u32 kvm_nr_mmu_pages)
3490 {
3491         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3492                 return -EINVAL;
3493
3494         mutex_lock(&kvm->slots_lock);
3495
3496         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3497         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3498
3499         mutex_unlock(&kvm->slots_lock);
3500         return 0;
3501 }
3502
3503 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3504 {
3505         return kvm->arch.n_max_mmu_pages;
3506 }
3507
3508 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3509 {
3510         int r;
3511
3512         r = 0;
3513         switch (chip->chip_id) {
3514         case KVM_IRQCHIP_PIC_MASTER:
3515                 memcpy(&chip->chip.pic,
3516                         &pic_irqchip(kvm)->pics[0],
3517                         sizeof(struct kvm_pic_state));
3518                 break;
3519         case KVM_IRQCHIP_PIC_SLAVE:
3520                 memcpy(&chip->chip.pic,
3521                         &pic_irqchip(kvm)->pics[1],
3522                         sizeof(struct kvm_pic_state));
3523                 break;
3524         case KVM_IRQCHIP_IOAPIC:
3525                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3526                 break;
3527         default:
3528                 r = -EINVAL;
3529                 break;
3530         }
3531         return r;
3532 }
3533
3534 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3535 {
3536         int r;
3537
3538         r = 0;
3539         switch (chip->chip_id) {
3540         case KVM_IRQCHIP_PIC_MASTER:
3541                 spin_lock(&pic_irqchip(kvm)->lock);
3542                 memcpy(&pic_irqchip(kvm)->pics[0],
3543                         &chip->chip.pic,
3544                         sizeof(struct kvm_pic_state));
3545                 spin_unlock(&pic_irqchip(kvm)->lock);
3546                 break;
3547         case KVM_IRQCHIP_PIC_SLAVE:
3548                 spin_lock(&pic_irqchip(kvm)->lock);
3549                 memcpy(&pic_irqchip(kvm)->pics[1],
3550                         &chip->chip.pic,
3551                         sizeof(struct kvm_pic_state));
3552                 spin_unlock(&pic_irqchip(kvm)->lock);
3553                 break;
3554         case KVM_IRQCHIP_IOAPIC:
3555                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3556                 break;
3557         default:
3558                 r = -EINVAL;
3559                 break;
3560         }
3561         kvm_pic_update_irq(pic_irqchip(kvm));
3562         return r;
3563 }
3564
3565 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3566 {
3567         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3568         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3569         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3570         return 0;
3571 }
3572
3573 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3574 {
3575         int i;
3576         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3577         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3578         for (i = 0; i < 3; i++)
3579                 kvm_pit_load_count(kvm, i, ps->channels[i].count, 0);
3580         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3581         return 0;
3582 }
3583
3584 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3585 {
3586         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3587         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3588                 sizeof(ps->channels));
3589         ps->flags = kvm->arch.vpit->pit_state.flags;
3590         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3591         memset(&ps->reserved, 0, sizeof(ps->reserved));
3592         return 0;
3593 }
3594
3595 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3596 {
3597         int start = 0;
3598         int i;
3599         u32 prev_legacy, cur_legacy;
3600         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3601         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3602         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3603         if (!prev_legacy && cur_legacy)
3604                 start = 1;
3605         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3606                sizeof(kvm->arch.vpit->pit_state.channels));
3607         kvm->arch.vpit->pit_state.flags = ps->flags;
3608         for (i = 0; i < 3; i++)
3609                 kvm_pit_load_count(kvm, i, kvm->arch.vpit->pit_state.channels[i].count,
3610                                    start && i == 0);
3611         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3612         return 0;
3613 }
3614
3615 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3616                                  struct kvm_reinject_control *control)
3617 {
3618         if (!kvm->arch.vpit)
3619                 return -ENXIO;
3620         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3621         kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3622         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3623         return 0;
3624 }
3625
3626 /**
3627  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3628  * @kvm: kvm instance
3629  * @log: slot id and address to which we copy the log
3630  *
3631  * Steps 1-4 below provide general overview of dirty page logging. See
3632  * kvm_get_dirty_log_protect() function description for additional details.
3633  *
3634  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3635  * always flush the TLB (step 4) even if previous step failed  and the dirty
3636  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3637  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3638  * writes will be marked dirty for next log read.
3639  *
3640  *   1. Take a snapshot of the bit and clear it if needed.
3641  *   2. Write protect the corresponding page.
3642  *   3. Copy the snapshot to the userspace.
3643  *   4. Flush TLB's if needed.
3644  */
3645 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3646 {
3647         bool is_dirty = false;
3648         int r;
3649
3650         mutex_lock(&kvm->slots_lock);
3651
3652         /*
3653          * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3654          */
3655         if (kvm_x86_ops->flush_log_dirty)
3656                 kvm_x86_ops->flush_log_dirty(kvm);
3657
3658         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3659
3660         /*
3661          * All the TLBs can be flushed out of mmu lock, see the comments in
3662          * kvm_mmu_slot_remove_write_access().
3663          */
3664         lockdep_assert_held(&kvm->slots_lock);
3665         if (is_dirty)
3666                 kvm_flush_remote_tlbs(kvm);
3667
3668         mutex_unlock(&kvm->slots_lock);
3669         return r;
3670 }
3671
3672 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3673                         bool line_status)
3674 {
3675         if (!irqchip_in_kernel(kvm))
3676                 return -ENXIO;
3677
3678         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3679                                         irq_event->irq, irq_event->level,
3680                                         line_status);
3681         return 0;
3682 }
3683
3684 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3685                                    struct kvm_enable_cap *cap)
3686 {
3687         int r;
3688
3689         if (cap->flags)
3690                 return -EINVAL;
3691
3692         switch (cap->cap) {
3693         case KVM_CAP_DISABLE_QUIRKS:
3694                 kvm->arch.disabled_quirks = cap->args[0];
3695                 r = 0;
3696                 break;
3697         case KVM_CAP_SPLIT_IRQCHIP: {
3698                 mutex_lock(&kvm->lock);
3699                 r = -EINVAL;
3700                 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3701                         goto split_irqchip_unlock;
3702                 r = -EEXIST;
3703                 if (irqchip_in_kernel(kvm))
3704                         goto split_irqchip_unlock;
3705                 if (atomic_read(&kvm->online_vcpus))
3706                         goto split_irqchip_unlock;
3707                 r = kvm_setup_empty_irq_routing(kvm);
3708                 if (r)
3709                         goto split_irqchip_unlock;
3710                 /* Pairs with irqchip_in_kernel. */
3711                 smp_wmb();
3712                 kvm->arch.irqchip_split = true;
3713                 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3714                 r = 0;
3715 split_irqchip_unlock:
3716                 mutex_unlock(&kvm->lock);
3717                 break;
3718         }
3719         default:
3720                 r = -EINVAL;
3721                 break;
3722         }
3723         return r;
3724 }
3725
3726 long kvm_arch_vm_ioctl(struct file *filp,
3727                        unsigned int ioctl, unsigned long arg)
3728 {
3729         struct kvm *kvm = filp->private_data;
3730         void __user *argp = (void __user *)arg;
3731         int r = -ENOTTY;
3732         /*
3733          * This union makes it completely explicit to gcc-3.x
3734          * that these two variables' stack usage should be
3735          * combined, not added together.
3736          */
3737         union {
3738                 struct kvm_pit_state ps;
3739                 struct kvm_pit_state2 ps2;
3740                 struct kvm_pit_config pit_config;
3741         } u;
3742
3743         switch (ioctl) {
3744         case KVM_SET_TSS_ADDR:
3745                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3746                 break;
3747         case KVM_SET_IDENTITY_MAP_ADDR: {
3748                 u64 ident_addr;
3749
3750                 r = -EFAULT;
3751                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3752                         goto out;
3753                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3754                 break;
3755         }
3756         case KVM_SET_NR_MMU_PAGES:
3757                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3758                 break;
3759         case KVM_GET_NR_MMU_PAGES:
3760                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3761                 break;
3762         case KVM_CREATE_IRQCHIP: {
3763                 struct kvm_pic *vpic;
3764
3765                 mutex_lock(&kvm->lock);
3766                 r = -EEXIST;
3767                 if (kvm->arch.vpic)
3768                         goto create_irqchip_unlock;
3769                 r = -EINVAL;
3770                 if (atomic_read(&kvm->online_vcpus))
3771                         goto create_irqchip_unlock;
3772                 r = -ENOMEM;
3773                 vpic = kvm_create_pic(kvm);
3774                 if (vpic) {
3775                         r = kvm_ioapic_init(kvm);
3776                         if (r) {
3777                                 mutex_lock(&kvm->slots_lock);
3778                                 kvm_destroy_pic(vpic);
3779                                 mutex_unlock(&kvm->slots_lock);
3780                                 goto create_irqchip_unlock;
3781                         }
3782                 } else
3783                         goto create_irqchip_unlock;
3784                 r = kvm_setup_default_irq_routing(kvm);
3785                 if (r) {
3786                         mutex_lock(&kvm->slots_lock);
3787                         mutex_lock(&kvm->irq_lock);
3788                         kvm_ioapic_destroy(kvm);
3789                         kvm_destroy_pic(vpic);
3790                         mutex_unlock(&kvm->irq_lock);
3791                         mutex_unlock(&kvm->slots_lock);
3792                         goto create_irqchip_unlock;
3793                 }
3794                 /* Write kvm->irq_routing before kvm->arch.vpic.  */
3795                 smp_wmb();
3796                 kvm->arch.vpic = vpic;
3797         create_irqchip_unlock:
3798                 mutex_unlock(&kvm->lock);
3799                 break;
3800         }
3801         case KVM_CREATE_PIT:
3802                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3803                 goto create_pit;
3804         case KVM_CREATE_PIT2:
3805                 r = -EFAULT;
3806                 if (copy_from_user(&u.pit_config, argp,
3807                                    sizeof(struct kvm_pit_config)))
3808                         goto out;
3809         create_pit:
3810                 mutex_lock(&kvm->slots_lock);
3811                 r = -EEXIST;
3812                 if (kvm->arch.vpit)
3813                         goto create_pit_unlock;
3814                 r = -ENOMEM;
3815                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3816                 if (kvm->arch.vpit)
3817                         r = 0;
3818         create_pit_unlock:
3819                 mutex_unlock(&kvm->slots_lock);
3820                 break;
3821         case KVM_GET_IRQCHIP: {
3822                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3823                 struct kvm_irqchip *chip;
3824
3825                 chip = memdup_user(argp, sizeof(*chip));
3826                 if (IS_ERR(chip)) {
3827                         r = PTR_ERR(chip);
3828                         goto out;
3829                 }
3830
3831                 r = -ENXIO;
3832                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3833                         goto get_irqchip_out;
3834                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3835                 if (r)
3836                         goto get_irqchip_out;
3837                 r = -EFAULT;
3838                 if (copy_to_user(argp, chip, sizeof *chip))
3839                         goto get_irqchip_out;
3840                 r = 0;
3841         get_irqchip_out:
3842                 kfree(chip);
3843                 break;
3844         }
3845         case KVM_SET_IRQCHIP: {
3846                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3847                 struct kvm_irqchip *chip;
3848
3849                 chip = memdup_user(argp, sizeof(*chip));
3850                 if (IS_ERR(chip)) {
3851                         r = PTR_ERR(chip);
3852                         goto out;
3853                 }
3854
3855                 r = -ENXIO;
3856                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3857                         goto set_irqchip_out;
3858                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3859                 if (r)
3860                         goto set_irqchip_out;
3861                 r = 0;
3862         set_irqchip_out:
3863                 kfree(chip);
3864                 break;
3865         }
3866         case KVM_GET_PIT: {
3867                 r = -EFAULT;
3868                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3869                         goto out;
3870                 r = -ENXIO;
3871                 if (!kvm->arch.vpit)
3872                         goto out;
3873                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3874                 if (r)
3875                         goto out;
3876                 r = -EFAULT;
3877                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3878                         goto out;
3879                 r = 0;
3880                 break;
3881         }
3882         case KVM_SET_PIT: {
3883                 r = -EFAULT;
3884                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3885                         goto out;
3886                 r = -ENXIO;
3887                 if (!kvm->arch.vpit)
3888                         goto out;
3889                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3890                 break;
3891         }
3892         case KVM_GET_PIT2: {
3893                 r = -ENXIO;
3894                 if (!kvm->arch.vpit)
3895                         goto out;
3896                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3897                 if (r)
3898                         goto out;
3899                 r = -EFAULT;
3900                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3901                         goto out;
3902                 r = 0;
3903                 break;
3904         }
3905         case KVM_SET_PIT2: {
3906                 r = -EFAULT;
3907                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3908                         goto out;
3909                 r = -ENXIO;
3910                 if (!kvm->arch.vpit)
3911                         goto out;
3912                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3913                 break;
3914         }
3915         case KVM_REINJECT_CONTROL: {
3916                 struct kvm_reinject_control control;
3917                 r =  -EFAULT;
3918                 if (copy_from_user(&control, argp, sizeof(control)))
3919                         goto out;
3920                 r = kvm_vm_ioctl_reinject(kvm, &control);
3921                 break;
3922         }
3923         case KVM_SET_BOOT_CPU_ID:
3924                 r = 0;
3925                 mutex_lock(&kvm->lock);
3926                 if (atomic_read(&kvm->online_vcpus) != 0)
3927                         r = -EBUSY;
3928                 else
3929                         kvm->arch.bsp_vcpu_id = arg;
3930                 mutex_unlock(&kvm->lock);
3931                 break;
3932         case KVM_XEN_HVM_CONFIG: {
3933                 r = -EFAULT;
3934                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3935                                    sizeof(struct kvm_xen_hvm_config)))
3936                         goto out;
3937                 r = -EINVAL;
3938                 if (kvm->arch.xen_hvm_config.flags)
3939                         goto out;
3940                 r = 0;
3941                 break;
3942         }
3943         case KVM_SET_CLOCK: {
3944                 struct kvm_clock_data user_ns;
3945                 u64 now_ns;
3946                 s64 delta;
3947
3948                 r = -EFAULT;
3949                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3950                         goto out;
3951
3952                 r = -EINVAL;
3953                 if (user_ns.flags)
3954                         goto out;
3955
3956                 r = 0;
3957                 local_irq_disable();
3958                 now_ns = get_kernel_ns();
3959                 delta = user_ns.clock - now_ns;
3960                 local_irq_enable();
3961                 kvm->arch.kvmclock_offset = delta;
3962                 kvm_gen_update_masterclock(kvm);
3963                 break;
3964         }
3965         case KVM_GET_CLOCK: {
3966                 struct kvm_clock_data user_ns;
3967                 u64 now_ns;
3968
3969                 local_irq_disable();
3970                 now_ns = get_kernel_ns();
3971                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3972                 local_irq_enable();
3973                 user_ns.flags = 0;
3974                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3975
3976                 r = -EFAULT;
3977                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3978                         goto out;
3979                 r = 0;
3980                 break;
3981         }
3982         case KVM_ENABLE_CAP: {
3983                 struct kvm_enable_cap cap;
3984
3985                 r = -EFAULT;
3986                 if (copy_from_user(&cap, argp, sizeof(cap)))
3987                         goto out;
3988                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
3989                 break;
3990         }
3991         default:
3992                 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
3993         }
3994 out:
3995         return r;
3996 }
3997
3998 static void kvm_init_msr_list(void)
3999 {
4000         u32 dummy[2];
4001         unsigned i, j;
4002
4003         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4004                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4005                         continue;
4006
4007                 /*
4008                  * Even MSRs that are valid in the host may not be exposed
4009                  * to the guests in some cases.
4010                  */
4011                 switch (msrs_to_save[i]) {
4012                 case MSR_IA32_BNDCFGS:
4013                         if (!kvm_x86_ops->mpx_supported())
4014                                 continue;
4015                         break;
4016                 case MSR_TSC_AUX:
4017                         if (!kvm_x86_ops->rdtscp_supported())
4018                                 continue;
4019                         break;
4020                 default:
4021                         break;
4022                 }
4023
4024                 if (j < i)
4025                         msrs_to_save[j] = msrs_to_save[i];
4026                 j++;
4027         }
4028         num_msrs_to_save = j;
4029
4030         for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4031                 switch (emulated_msrs[i]) {
4032                 case MSR_IA32_SMBASE:
4033                         if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4034                                 continue;
4035                         break;
4036                 default:
4037                         break;
4038                 }
4039
4040                 if (j < i)
4041                         emulated_msrs[j] = emulated_msrs[i];
4042                 j++;
4043         }
4044         num_emulated_msrs = j;
4045 }
4046
4047 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4048                            const void *v)
4049 {
4050         int handled = 0;
4051         int n;
4052
4053         do {
4054                 n = min(len, 8);
4055                 if (!(vcpu->arch.apic &&
4056                       !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4057                     && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4058                         break;
4059                 handled += n;
4060                 addr += n;
4061                 len -= n;
4062                 v += n;
4063         } while (len);
4064
4065         return handled;
4066 }
4067
4068 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4069 {
4070         int handled = 0;
4071         int n;
4072
4073         do {
4074                 n = min(len, 8);
4075                 if (!(vcpu->arch.apic &&
4076                       !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4077                                          addr, n, v))
4078                     && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4079                         break;
4080                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4081                 handled += n;
4082                 addr += n;
4083                 len -= n;
4084                 v += n;
4085         } while (len);
4086
4087         return handled;
4088 }
4089
4090 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4091                         struct kvm_segment *var, int seg)
4092 {
4093         kvm_x86_ops->set_segment(vcpu, var, seg);
4094 }
4095
4096 void kvm_get_segment(struct kvm_vcpu *vcpu,
4097                      struct kvm_segment *var, int seg)
4098 {
4099         kvm_x86_ops->get_segment(vcpu, var, seg);
4100 }
4101
4102 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4103                            struct x86_exception *exception)
4104 {
4105         gpa_t t_gpa;
4106
4107         BUG_ON(!mmu_is_nested(vcpu));
4108
4109         /* NPT walks are always user-walks */
4110         access |= PFERR_USER_MASK;
4111         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4112
4113         return t_gpa;
4114 }
4115
4116 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4117                               struct x86_exception *exception)
4118 {
4119         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4120         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4121 }
4122
4123  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4124                                 struct x86_exception *exception)
4125 {
4126         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4127         access |= PFERR_FETCH_MASK;
4128         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4129 }
4130
4131 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4132                                struct x86_exception *exception)
4133 {
4134         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4135         access |= PFERR_WRITE_MASK;
4136         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4137 }
4138
4139 /* uses this to access any guest's mapped memory without checking CPL */
4140 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4141                                 struct x86_exception *exception)
4142 {
4143         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4144 }
4145
4146 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4147                                       struct kvm_vcpu *vcpu, u32 access,
4148                                       struct x86_exception *exception)
4149 {
4150         void *data = val;
4151         int r = X86EMUL_CONTINUE;
4152
4153         while (bytes) {
4154                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4155                                                             exception);
4156                 unsigned offset = addr & (PAGE_SIZE-1);
4157                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4158                 int ret;
4159
4160                 if (gpa == UNMAPPED_GVA)
4161                         return X86EMUL_PROPAGATE_FAULT;
4162                 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4163                                                offset, toread);
4164                 if (ret < 0) {
4165                         r = X86EMUL_IO_NEEDED;
4166                         goto out;
4167                 }
4168
4169                 bytes -= toread;
4170                 data += toread;
4171                 addr += toread;
4172         }
4173 out:
4174         return r;
4175 }
4176
4177 /* used for instruction fetching */
4178 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4179                                 gva_t addr, void *val, unsigned int bytes,
4180                                 struct x86_exception *exception)
4181 {
4182         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4183         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4184         unsigned offset;
4185         int ret;
4186
4187         /* Inline kvm_read_guest_virt_helper for speed.  */
4188         gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4189                                                     exception);
4190         if (unlikely(gpa == UNMAPPED_GVA))
4191                 return X86EMUL_PROPAGATE_FAULT;
4192
4193         offset = addr & (PAGE_SIZE-1);
4194         if (WARN_ON(offset + bytes > PAGE_SIZE))
4195                 bytes = (unsigned)PAGE_SIZE - offset;
4196         ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4197                                        offset, bytes);
4198         if (unlikely(ret < 0))
4199                 return X86EMUL_IO_NEEDED;
4200
4201         return X86EMUL_CONTINUE;
4202 }
4203
4204 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4205                                gva_t addr, void *val, unsigned int bytes,
4206                                struct x86_exception *exception)
4207 {
4208         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4209         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4210
4211         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4212                                           exception);
4213 }
4214 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4215
4216 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4217                                       gva_t addr, void *val, unsigned int bytes,
4218                                       struct x86_exception *exception)
4219 {
4220         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4221         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4222 }
4223
4224 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4225                 unsigned long addr, void *val, unsigned int bytes)
4226 {
4227         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4228         int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4229
4230         return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4231 }
4232
4233 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4234                                        gva_t addr, void *val,
4235                                        unsigned int bytes,
4236                                        struct x86_exception *exception)
4237 {
4238         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4239         void *data = val;
4240         int r = X86EMUL_CONTINUE;
4241
4242         while (bytes) {
4243                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4244                                                              PFERR_WRITE_MASK,
4245                                                              exception);
4246                 unsigned offset = addr & (PAGE_SIZE-1);
4247                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4248                 int ret;
4249
4250                 if (gpa == UNMAPPED_GVA)
4251                         return X86EMUL_PROPAGATE_FAULT;
4252                 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4253                 if (ret < 0) {
4254                         r = X86EMUL_IO_NEEDED;
4255                         goto out;
4256                 }
4257
4258                 bytes -= towrite;
4259                 data += towrite;
4260                 addr += towrite;
4261         }
4262 out:
4263         return r;
4264 }
4265 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4266
4267 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4268                                 gpa_t *gpa, struct x86_exception *exception,
4269                                 bool write)
4270 {
4271         u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4272                 | (write ? PFERR_WRITE_MASK : 0);
4273
4274         if (vcpu_match_mmio_gva(vcpu, gva)
4275             && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4276                                  vcpu->arch.access, access)) {
4277                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4278                                         (gva & (PAGE_SIZE - 1));
4279                 trace_vcpu_match_mmio(gva, *gpa, write, false);
4280                 return 1;
4281         }
4282
4283         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4284
4285         if (*gpa == UNMAPPED_GVA)
4286                 return -1;
4287
4288         /* For APIC access vmexit */
4289         if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4290                 return 1;
4291
4292         if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4293                 trace_vcpu_match_mmio(gva, *gpa, write, true);
4294                 return 1;
4295         }
4296
4297         return 0;
4298 }
4299
4300 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4301                         const void *val, int bytes)
4302 {
4303         int ret;
4304
4305         ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4306         if (ret < 0)
4307                 return 0;
4308         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4309         return 1;
4310 }
4311
4312 struct read_write_emulator_ops {
4313         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4314                                   int bytes);
4315         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4316                                   void *val, int bytes);
4317         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4318                                int bytes, void *val);
4319         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4320                                     void *val, int bytes);
4321         bool write;
4322 };
4323
4324 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4325 {
4326         if (vcpu->mmio_read_completed) {
4327                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4328                                vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4329                 vcpu->mmio_read_completed = 0;
4330                 return 1;
4331         }
4332
4333         return 0;
4334 }
4335
4336 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4337                         void *val, int bytes)
4338 {
4339         return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4340 }
4341
4342 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4343                          void *val, int bytes)
4344 {
4345         return emulator_write_phys(vcpu, gpa, val, bytes);
4346 }
4347
4348 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4349 {
4350         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4351         return vcpu_mmio_write(vcpu, gpa, bytes, val);
4352 }
4353
4354 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4355                           void *val, int bytes)
4356 {
4357         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4358         return X86EMUL_IO_NEEDED;
4359 }
4360
4361 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4362                            void *val, int bytes)
4363 {
4364         struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4365
4366         memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4367         return X86EMUL_CONTINUE;
4368 }
4369
4370 static const struct read_write_emulator_ops read_emultor = {
4371         .read_write_prepare = read_prepare,
4372         .read_write_emulate = read_emulate,
4373         .read_write_mmio = vcpu_mmio_read,
4374         .read_write_exit_mmio = read_exit_mmio,
4375 };
4376
4377 static const struct read_write_emulator_ops write_emultor = {
4378         .read_write_emulate = write_emulate,
4379         .read_write_mmio = write_mmio,
4380         .read_write_exit_mmio = write_exit_mmio,
4381         .write = true,
4382 };
4383
4384 static int emulator_read_write_onepage(unsigned long addr, void *val,
4385                                        unsigned int bytes,
4386                                        struct x86_exception *exception,
4387                                        struct kvm_vcpu *vcpu,
4388                                        const struct read_write_emulator_ops *ops)
4389 {
4390         gpa_t gpa;
4391         int handled, ret;
4392         bool write = ops->write;
4393         struct kvm_mmio_fragment *frag;
4394
4395         ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4396
4397         if (ret < 0)
4398                 return X86EMUL_PROPAGATE_FAULT;
4399
4400         /* For APIC access vmexit */
4401         if (ret)
4402                 goto mmio;
4403
4404         if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4405                 return X86EMUL_CONTINUE;
4406
4407 mmio:
4408         /*
4409          * Is this MMIO handled locally?
4410          */
4411         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4412         if (handled == bytes)
4413                 return X86EMUL_CONTINUE;
4414
4415         gpa += handled;
4416         bytes -= handled;
4417         val += handled;
4418
4419         WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4420         frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4421         frag->gpa = gpa;
4422         frag->data = val;
4423         frag->len = bytes;
4424         return X86EMUL_CONTINUE;
4425 }
4426
4427 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4428                         unsigned long addr,
4429                         void *val, unsigned int bytes,
4430                         struct x86_exception *exception,
4431                         const struct read_write_emulator_ops *ops)
4432 {
4433         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4434         gpa_t gpa;
4435         int rc;
4436
4437         if (ops->read_write_prepare &&
4438                   ops->read_write_prepare(vcpu, val, bytes))
4439                 return X86EMUL_CONTINUE;
4440
4441         vcpu->mmio_nr_fragments = 0;
4442
4443         /* Crossing a page boundary? */
4444         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4445                 int now;
4446
4447                 now = -addr & ~PAGE_MASK;
4448                 rc = emulator_read_write_onepage(addr, val, now, exception,
4449                                                  vcpu, ops);
4450
4451                 if (rc != X86EMUL_CONTINUE)
4452                         return rc;
4453                 addr += now;
4454                 if (ctxt->mode != X86EMUL_MODE_PROT64)
4455                         addr = (u32)addr;
4456                 val += now;
4457                 bytes -= now;
4458         }
4459
4460         rc = emulator_read_write_onepage(addr, val, bytes, exception,
4461                                          vcpu, ops);
4462         if (rc != X86EMUL_CONTINUE)
4463                 return rc;
4464
4465         if (!vcpu->mmio_nr_fragments)
4466                 return rc;
4467
4468         gpa = vcpu->mmio_fragments[0].gpa;
4469
4470         vcpu->mmio_needed = 1;
4471         vcpu->mmio_cur_fragment = 0;
4472
4473         vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4474         vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4475         vcpu->run->exit_reason = KVM_EXIT_MMIO;
4476         vcpu->run->mmio.phys_addr = gpa;
4477
4478         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4479 }
4480
4481 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4482                                   unsigned long addr,
4483                                   void *val,
4484                                   unsigned int bytes,
4485                                   struct x86_exception *exception)
4486 {
4487         return emulator_read_write(ctxt, addr, val, bytes,
4488                                    exception, &read_emultor);
4489 }
4490
4491 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4492                             unsigned long addr,
4493                             const void *val,
4494                             unsigned int bytes,
4495                             struct x86_exception *exception)
4496 {
4497         return emulator_read_write(ctxt, addr, (void *)val, bytes,
4498                                    exception, &write_emultor);
4499 }
4500
4501 #define CMPXCHG_TYPE(t, ptr, old, new) \
4502         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4503
4504 #ifdef CONFIG_X86_64
4505 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4506 #else
4507 #  define CMPXCHG64(ptr, old, new) \
4508         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4509 #endif
4510
4511 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4512                                      unsigned long addr,
4513                                      const void *old,
4514                                      const void *new,
4515                                      unsigned int bytes,
4516                                      struct x86_exception *exception)
4517 {
4518         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4519         gpa_t gpa;
4520         struct page *page;
4521         char *kaddr;
4522         bool exchanged;
4523
4524         /* guests cmpxchg8b have to be emulated atomically */
4525         if (bytes > 8 || (bytes & (bytes - 1)))
4526                 goto emul_write;
4527
4528         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4529
4530         if (gpa == UNMAPPED_GVA ||
4531             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4532                 goto emul_write;
4533
4534         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4535                 goto emul_write;
4536
4537         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4538         if (is_error_page(page))
4539                 goto emul_write;
4540
4541         kaddr = kmap_atomic(page);
4542         kaddr += offset_in_page(gpa);
4543         switch (bytes) {
4544         case 1:
4545                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4546                 break;
4547         case 2:
4548                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4549                 break;
4550         case 4:
4551                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4552                 break;
4553         case 8:
4554                 exchanged = CMPXCHG64(kaddr, old, new);
4555                 break;
4556         default:
4557                 BUG();
4558         }
4559         kunmap_atomic(kaddr);
4560         kvm_release_page_dirty(page);
4561
4562         if (!exchanged)
4563                 return X86EMUL_CMPXCHG_FAILED;
4564
4565         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4566         kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4567
4568         return X86EMUL_CONTINUE;
4569
4570 emul_write:
4571         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4572
4573         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4574 }
4575
4576 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4577 {
4578         /* TODO: String I/O for in kernel device */
4579         int r;
4580
4581         if (vcpu->arch.pio.in)
4582                 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4583                                     vcpu->arch.pio.size, pd);
4584         else
4585                 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4586                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4587                                      pd);
4588         return r;
4589 }
4590
4591 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4592                                unsigned short port, void *val,
4593                                unsigned int count, bool in)
4594 {
4595         vcpu->arch.pio.port = port;
4596         vcpu->arch.pio.in = in;
4597         vcpu->arch.pio.count  = count;
4598         vcpu->arch.pio.size = size;
4599
4600         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4601                 vcpu->arch.pio.count = 0;
4602                 return 1;
4603         }
4604
4605         vcpu->run->exit_reason = KVM_EXIT_IO;
4606         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4607         vcpu->run->io.size = size;
4608         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4609         vcpu->run->io.count = count;
4610         vcpu->run->io.port = port;
4611
4612         return 0;
4613 }
4614
4615 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4616                                     int size, unsigned short port, void *val,
4617                                     unsigned int count)
4618 {
4619         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4620         int ret;
4621
4622         if (vcpu->arch.pio.count)
4623                 goto data_avail;
4624
4625         ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4626         if (ret) {
4627 data_avail:
4628                 memcpy(val, vcpu->arch.pio_data, size * count);
4629                 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4630                 vcpu->arch.pio.count = 0;
4631                 return 1;
4632         }
4633
4634         return 0;
4635 }
4636
4637 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4638                                      int size, unsigned short port,
4639                                      const void *val, unsigned int count)
4640 {
4641         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4642
4643         memcpy(vcpu->arch.pio_data, val, size * count);
4644         trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4645         return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4646 }
4647
4648 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4649 {
4650         return kvm_x86_ops->get_segment_base(vcpu, seg);
4651 }
4652
4653 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4654 {
4655         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4656 }
4657
4658 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4659 {
4660         if (!need_emulate_wbinvd(vcpu))
4661                 return X86EMUL_CONTINUE;
4662
4663         if (kvm_x86_ops->has_wbinvd_exit()) {
4664                 int cpu = get_cpu();
4665
4666                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4667                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4668                                 wbinvd_ipi, NULL, 1);
4669                 put_cpu();
4670                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4671         } else
4672                 wbinvd();
4673         return X86EMUL_CONTINUE;
4674 }
4675
4676 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4677 {
4678         kvm_x86_ops->skip_emulated_instruction(vcpu);
4679         return kvm_emulate_wbinvd_noskip(vcpu);
4680 }
4681 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4682
4683
4684
4685 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4686 {
4687         kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4688 }
4689
4690 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4691                            unsigned long *dest)
4692 {
4693         return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4694 }
4695
4696 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4697                            unsigned long value)
4698 {
4699
4700         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4701 }
4702
4703 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4704 {
4705         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4706 }
4707
4708 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4709 {
4710         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4711         unsigned long value;
4712
4713         switch (cr) {
4714         case 0:
4715                 value = kvm_read_cr0(vcpu);
4716                 break;
4717         case 2:
4718                 value = vcpu->arch.cr2;
4719                 break;
4720         case 3:
4721                 value = kvm_read_cr3(vcpu);
4722                 break;
4723         case 4:
4724                 value = kvm_read_cr4(vcpu);
4725                 break;
4726         case 8:
4727                 value = kvm_get_cr8(vcpu);
4728                 break;
4729         default:
4730                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4731                 return 0;
4732         }
4733
4734         return value;
4735 }
4736
4737 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4738 {
4739         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4740         int res = 0;
4741
4742         switch (cr) {
4743         case 0:
4744                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4745                 break;
4746         case 2:
4747                 vcpu->arch.cr2 = val;
4748                 break;
4749         case 3:
4750                 res = kvm_set_cr3(vcpu, val);
4751                 break;
4752         case 4:
4753                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4754                 break;
4755         case 8:
4756                 res = kvm_set_cr8(vcpu, val);
4757                 break;
4758         default:
4759                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4760                 res = -1;
4761         }
4762
4763         return res;
4764 }
4765
4766 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4767 {
4768         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4769 }
4770
4771 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4772 {
4773         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4774 }
4775
4776 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4777 {
4778         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4779 }
4780
4781 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4782 {
4783         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4784 }
4785
4786 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4787 {
4788         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4789 }
4790
4791 static unsigned long emulator_get_cached_segment_base(
4792         struct x86_emulate_ctxt *ctxt, int seg)
4793 {
4794         return get_segment_base(emul_to_vcpu(ctxt), seg);
4795 }
4796
4797 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4798                                  struct desc_struct *desc, u32 *base3,
4799                                  int seg)
4800 {
4801         struct kvm_segment var;
4802
4803         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4804         *selector = var.selector;
4805
4806         if (var.unusable) {
4807                 memset(desc, 0, sizeof(*desc));
4808                 return false;
4809         }
4810
4811         if (var.g)
4812                 var.limit >>= 12;
4813         set_desc_limit(desc, var.limit);
4814         set_desc_base(desc, (unsigned long)var.base);
4815 #ifdef CONFIG_X86_64
4816         if (base3)
4817                 *base3 = var.base >> 32;
4818 #endif
4819         desc->type = var.type;
4820         desc->s = var.s;
4821         desc->dpl = var.dpl;
4822         desc->p = var.present;
4823         desc->avl = var.avl;
4824         desc->l = var.l;
4825         desc->d = var.db;
4826         desc->g = var.g;
4827
4828         return true;
4829 }
4830
4831 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4832                                  struct desc_struct *desc, u32 base3,
4833                                  int seg)
4834 {
4835         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4836         struct kvm_segment var;
4837
4838         var.selector = selector;
4839         var.base = get_desc_base(desc);
4840 #ifdef CONFIG_X86_64
4841         var.base |= ((u64)base3) << 32;
4842 #endif
4843         var.limit = get_desc_limit(desc);
4844         if (desc->g)
4845                 var.limit = (var.limit << 12) | 0xfff;
4846         var.type = desc->type;
4847         var.dpl = desc->dpl;
4848         var.db = desc->d;
4849         var.s = desc->s;
4850         var.l = desc->l;
4851         var.g = desc->g;
4852         var.avl = desc->avl;
4853         var.present = desc->p;
4854         var.unusable = !var.present;
4855         var.padding = 0;
4856
4857         kvm_set_segment(vcpu, &var, seg);
4858         return;
4859 }
4860
4861 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4862                             u32 msr_index, u64 *pdata)
4863 {
4864         struct msr_data msr;
4865         int r;
4866
4867         msr.index = msr_index;
4868         msr.host_initiated = false;
4869         r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4870         if (r)
4871                 return r;
4872
4873         *pdata = msr.data;
4874         return 0;
4875 }
4876
4877 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4878                             u32 msr_index, u64 data)
4879 {
4880         struct msr_data msr;
4881
4882         msr.data = data;
4883         msr.index = msr_index;
4884         msr.host_initiated = false;
4885         return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4886 }
4887
4888 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4889 {
4890         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4891
4892         return vcpu->arch.smbase;
4893 }
4894
4895 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4896 {
4897         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4898
4899         vcpu->arch.smbase = smbase;
4900 }
4901
4902 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4903                               u32 pmc)
4904 {
4905         return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4906 }
4907
4908 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4909                              u32 pmc, u64 *pdata)
4910 {
4911         return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4912 }
4913
4914 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4915 {
4916         emul_to_vcpu(ctxt)->arch.halt_request = 1;
4917 }
4918
4919 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4920 {
4921         preempt_disable();
4922         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4923         /*
4924          * CR0.TS may reference the host fpu state, not the guest fpu state,
4925          * so it may be clear at this point.
4926          */
4927         clts();
4928 }
4929
4930 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4931 {
4932         preempt_enable();
4933 }
4934
4935 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4936                               struct x86_instruction_info *info,
4937                               enum x86_intercept_stage stage)
4938 {
4939         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4940 }
4941
4942 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4943                                u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4944 {
4945         kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4946 }
4947
4948 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4949 {
4950         return kvm_register_read(emul_to_vcpu(ctxt), reg);
4951 }
4952
4953 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4954 {
4955         kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4956 }
4957
4958 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4959 {
4960         kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4961 }
4962
4963 static const struct x86_emulate_ops emulate_ops = {
4964         .read_gpr            = emulator_read_gpr,
4965         .write_gpr           = emulator_write_gpr,
4966         .read_std            = kvm_read_guest_virt_system,
4967         .write_std           = kvm_write_guest_virt_system,
4968         .read_phys           = kvm_read_guest_phys_system,
4969         .fetch               = kvm_fetch_guest_virt,
4970         .read_emulated       = emulator_read_emulated,
4971         .write_emulated      = emulator_write_emulated,
4972         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
4973         .invlpg              = emulator_invlpg,
4974         .pio_in_emulated     = emulator_pio_in_emulated,
4975         .pio_out_emulated    = emulator_pio_out_emulated,
4976         .get_segment         = emulator_get_segment,
4977         .set_segment         = emulator_set_segment,
4978         .get_cached_segment_base = emulator_get_cached_segment_base,
4979         .get_gdt             = emulator_get_gdt,
4980         .get_idt             = emulator_get_idt,
4981         .set_gdt             = emulator_set_gdt,
4982         .set_idt             = emulator_set_idt,
4983         .get_cr              = emulator_get_cr,
4984         .set_cr              = emulator_set_cr,
4985         .cpl                 = emulator_get_cpl,
4986         .get_dr              = emulator_get_dr,
4987         .set_dr              = emulator_set_dr,
4988         .get_smbase          = emulator_get_smbase,
4989         .set_smbase          = emulator_set_smbase,
4990         .set_msr             = emulator_set_msr,
4991         .get_msr             = emulator_get_msr,
4992         .check_pmc           = emulator_check_pmc,
4993         .read_pmc            = emulator_read_pmc,
4994         .halt                = emulator_halt,
4995         .wbinvd              = emulator_wbinvd,
4996         .fix_hypercall       = emulator_fix_hypercall,
4997         .get_fpu             = emulator_get_fpu,
4998         .put_fpu             = emulator_put_fpu,
4999         .intercept           = emulator_intercept,
5000         .get_cpuid           = emulator_get_cpuid,
5001         .set_nmi_mask        = emulator_set_nmi_mask,
5002 };
5003
5004 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5005 {
5006         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5007         /*
5008          * an sti; sti; sequence only disable interrupts for the first
5009          * instruction. So, if the last instruction, be it emulated or
5010          * not, left the system with the INT_STI flag enabled, it
5011          * means that the last instruction is an sti. We should not
5012          * leave the flag on in this case. The same goes for mov ss
5013          */
5014         if (int_shadow & mask)
5015                 mask = 0;
5016         if (unlikely(int_shadow || mask)) {
5017                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5018                 if (!mask)
5019                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5020         }
5021 }
5022
5023 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5024 {
5025         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5026         if (ctxt->exception.vector == PF_VECTOR)
5027                 return kvm_propagate_fault(vcpu, &ctxt->exception);
5028
5029         if (ctxt->exception.error_code_valid)
5030                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5031                                       ctxt->exception.error_code);
5032         else
5033                 kvm_queue_exception(vcpu, ctxt->exception.vector);
5034         return false;
5035 }
5036
5037 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5038 {
5039         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5040         int cs_db, cs_l;
5041
5042         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5043
5044         ctxt->eflags = kvm_get_rflags(vcpu);
5045         ctxt->eip = kvm_rip_read(vcpu);
5046         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
5047                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
5048                      (cs_l && is_long_mode(vcpu))       ? X86EMUL_MODE_PROT64 :
5049                      cs_db                              ? X86EMUL_MODE_PROT32 :
5050                                                           X86EMUL_MODE_PROT16;
5051         BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5052         BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5053         BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5054         ctxt->emul_flags = vcpu->arch.hflags;
5055
5056         init_decode_cache(ctxt);
5057         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5058 }
5059
5060 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5061 {
5062         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5063         int ret;
5064
5065         init_emulate_ctxt(vcpu);
5066
5067         ctxt->op_bytes = 2;
5068         ctxt->ad_bytes = 2;
5069         ctxt->_eip = ctxt->eip + inc_eip;
5070         ret = emulate_int_real(ctxt, irq);
5071
5072         if (ret != X86EMUL_CONTINUE)
5073                 return EMULATE_FAIL;
5074
5075         ctxt->eip = ctxt->_eip;
5076         kvm_rip_write(vcpu, ctxt->eip);
5077         kvm_set_rflags(vcpu, ctxt->eflags);
5078
5079         if (irq == NMI_VECTOR)
5080                 vcpu->arch.nmi_pending = 0;
5081         else
5082                 vcpu->arch.interrupt.pending = false;
5083
5084         return EMULATE_DONE;
5085 }
5086 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5087
5088 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5089 {
5090         int r = EMULATE_DONE;
5091
5092         ++vcpu->stat.insn_emulation_fail;
5093         trace_kvm_emulate_insn_failed(vcpu);
5094         if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5095                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5096                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5097                 vcpu->run->internal.ndata = 0;
5098                 r = EMULATE_FAIL;
5099         }
5100         kvm_queue_exception(vcpu, UD_VECTOR);
5101
5102         return r;
5103 }
5104
5105 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5106                                   bool write_fault_to_shadow_pgtable,
5107                                   int emulation_type)
5108 {
5109         gpa_t gpa = cr2;
5110         pfn_t pfn;
5111
5112         if (emulation_type & EMULTYPE_NO_REEXECUTE)
5113                 return false;
5114
5115         if (!vcpu->arch.mmu.direct_map) {
5116                 /*
5117                  * Write permission should be allowed since only
5118                  * write access need to be emulated.
5119                  */
5120                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5121
5122                 /*
5123                  * If the mapping is invalid in guest, let cpu retry
5124                  * it to generate fault.
5125                  */
5126                 if (gpa == UNMAPPED_GVA)
5127                         return true;
5128         }
5129
5130         /*
5131          * Do not retry the unhandleable instruction if it faults on the
5132          * readonly host memory, otherwise it will goto a infinite loop:
5133          * retry instruction -> write #PF -> emulation fail -> retry
5134          * instruction -> ...
5135          */
5136         pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5137
5138         /*
5139          * If the instruction failed on the error pfn, it can not be fixed,
5140          * report the error to userspace.
5141          */
5142         if (is_error_noslot_pfn(pfn))
5143                 return false;
5144
5145         kvm_release_pfn_clean(pfn);
5146
5147         /* The instructions are well-emulated on direct mmu. */
5148         if (vcpu->arch.mmu.direct_map) {
5149                 unsigned int indirect_shadow_pages;
5150
5151                 spin_lock(&vcpu->kvm->mmu_lock);
5152                 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5153                 spin_unlock(&vcpu->kvm->mmu_lock);
5154
5155                 if (indirect_shadow_pages)
5156                         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5157
5158                 return true;
5159         }
5160
5161         /*
5162          * if emulation was due to access to shadowed page table
5163          * and it failed try to unshadow page and re-enter the
5164          * guest to let CPU execute the instruction.
5165          */
5166         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5167
5168         /*
5169          * If the access faults on its page table, it can not
5170          * be fixed by unprotecting shadow page and it should
5171          * be reported to userspace.
5172          */
5173         return !write_fault_to_shadow_pgtable;
5174 }
5175
5176 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5177                               unsigned long cr2,  int emulation_type)
5178 {
5179         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5180         unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5181
5182         last_retry_eip = vcpu->arch.last_retry_eip;
5183         last_retry_addr = vcpu->arch.last_retry_addr;
5184
5185         /*
5186          * If the emulation is caused by #PF and it is non-page_table
5187          * writing instruction, it means the VM-EXIT is caused by shadow
5188          * page protected, we can zap the shadow page and retry this
5189          * instruction directly.
5190          *
5191          * Note: if the guest uses a non-page-table modifying instruction
5192          * on the PDE that points to the instruction, then we will unmap
5193          * the instruction and go to an infinite loop. So, we cache the
5194          * last retried eip and the last fault address, if we meet the eip
5195          * and the address again, we can break out of the potential infinite
5196          * loop.
5197          */
5198         vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5199
5200         if (!(emulation_type & EMULTYPE_RETRY))
5201                 return false;
5202
5203         if (x86_page_table_writing_insn(ctxt))
5204                 return false;
5205
5206         if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5207                 return false;
5208
5209         vcpu->arch.last_retry_eip = ctxt->eip;
5210         vcpu->arch.last_retry_addr = cr2;
5211
5212         if (!vcpu->arch.mmu.direct_map)
5213                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5214
5215         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5216
5217         return true;
5218 }
5219
5220 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5221 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5222
5223 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5224 {
5225         if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5226                 /* This is a good place to trace that we are exiting SMM.  */
5227                 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5228
5229                 if (unlikely(vcpu->arch.smi_pending)) {
5230                         kvm_make_request(KVM_REQ_SMI, vcpu);
5231                         vcpu->arch.smi_pending = 0;
5232                 } else {
5233                         /* Process a latched INIT, if any.  */
5234                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5235                 }
5236         }
5237
5238         kvm_mmu_reset_context(vcpu);
5239 }
5240
5241 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5242 {
5243         unsigned changed = vcpu->arch.hflags ^ emul_flags;
5244
5245         vcpu->arch.hflags = emul_flags;
5246
5247         if (changed & HF_SMM_MASK)
5248                 kvm_smm_changed(vcpu);
5249 }
5250
5251 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5252                                 unsigned long *db)
5253 {
5254         u32 dr6 = 0;
5255         int i;
5256         u32 enable, rwlen;
5257
5258         enable = dr7;
5259         rwlen = dr7 >> 16;
5260         for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5261                 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5262                         dr6 |= (1 << i);
5263         return dr6;
5264 }
5265
5266 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5267 {
5268         struct kvm_run *kvm_run = vcpu->run;
5269
5270         /*
5271          * rflags is the old, "raw" value of the flags.  The new value has
5272          * not been saved yet.
5273          *
5274          * This is correct even for TF set by the guest, because "the
5275          * processor will not generate this exception after the instruction
5276          * that sets the TF flag".
5277          */
5278         if (unlikely(rflags & X86_EFLAGS_TF)) {
5279                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5280                         kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5281                                                   DR6_RTM;
5282                         kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5283                         kvm_run->debug.arch.exception = DB_VECTOR;
5284                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5285                         *r = EMULATE_USER_EXIT;
5286                 } else {
5287                         vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5288                         /*
5289                          * "Certain debug exceptions may clear bit 0-3.  The
5290                          * remaining contents of the DR6 register are never
5291                          * cleared by the processor".
5292                          */
5293                         vcpu->arch.dr6 &= ~15;
5294                         vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5295                         kvm_queue_exception(vcpu, DB_VECTOR);
5296                 }
5297         }
5298 }
5299
5300 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5301 {
5302         if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5303             (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5304                 struct kvm_run *kvm_run = vcpu->run;
5305                 unsigned long eip = kvm_get_linear_rip(vcpu);
5306                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5307                                            vcpu->arch.guest_debug_dr7,
5308                                            vcpu->arch.eff_db);
5309
5310                 if (dr6 != 0) {
5311                         kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5312                         kvm_run->debug.arch.pc = eip;
5313                         kvm_run->debug.arch.exception = DB_VECTOR;
5314                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5315                         *r = EMULATE_USER_EXIT;
5316                         return true;
5317                 }
5318         }
5319
5320         if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5321             !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5322                 unsigned long eip = kvm_get_linear_rip(vcpu);
5323                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5324                                            vcpu->arch.dr7,
5325                                            vcpu->arch.db);
5326
5327                 if (dr6 != 0) {
5328                         vcpu->arch.dr6 &= ~15;
5329                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5330                         kvm_queue_exception(vcpu, DB_VECTOR);
5331                         *r = EMULATE_DONE;
5332                         return true;
5333                 }
5334         }
5335
5336         return false;
5337 }
5338
5339 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5340                             unsigned long cr2,
5341                             int emulation_type,
5342                             void *insn,
5343                             int insn_len)
5344 {
5345         int r;
5346         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5347         bool writeback = true;
5348         bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5349
5350         /*
5351          * Clear write_fault_to_shadow_pgtable here to ensure it is
5352          * never reused.
5353          */
5354         vcpu->arch.write_fault_to_shadow_pgtable = false;
5355         kvm_clear_exception_queue(vcpu);
5356
5357         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5358                 init_emulate_ctxt(vcpu);
5359
5360                 /*
5361                  * We will reenter on the same instruction since
5362                  * we do not set complete_userspace_io.  This does not
5363                  * handle watchpoints yet, those would be handled in
5364                  * the emulate_ops.
5365                  */
5366                 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5367                         return r;
5368
5369                 ctxt->interruptibility = 0;
5370                 ctxt->have_exception = false;
5371                 ctxt->exception.vector = -1;
5372                 ctxt->perm_ok = false;
5373
5374                 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5375
5376                 r = x86_decode_insn(ctxt, insn, insn_len);
5377
5378                 trace_kvm_emulate_insn_start(vcpu);
5379                 ++vcpu->stat.insn_emulation;
5380                 if (r != EMULATION_OK)  {
5381                         if (emulation_type & EMULTYPE_TRAP_UD)
5382                                 return EMULATE_FAIL;
5383                         if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5384                                                 emulation_type))
5385                                 return EMULATE_DONE;
5386                         if (emulation_type & EMULTYPE_SKIP)
5387                                 return EMULATE_FAIL;
5388                         return handle_emulation_failure(vcpu);
5389                 }
5390         }
5391
5392         if (emulation_type & EMULTYPE_SKIP) {
5393                 kvm_rip_write(vcpu, ctxt->_eip);
5394                 if (ctxt->eflags & X86_EFLAGS_RF)
5395                         kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5396                 return EMULATE_DONE;
5397         }
5398
5399         if (retry_instruction(ctxt, cr2, emulation_type))
5400                 return EMULATE_DONE;
5401
5402         /* this is needed for vmware backdoor interface to work since it
5403            changes registers values  during IO operation */
5404         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5405                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5406                 emulator_invalidate_register_cache(ctxt);
5407         }
5408
5409 restart:
5410         r = x86_emulate_insn(ctxt);
5411
5412         if (r == EMULATION_INTERCEPTED)
5413                 return EMULATE_DONE;
5414
5415         if (r == EMULATION_FAILED) {
5416                 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5417                                         emulation_type))
5418                         return EMULATE_DONE;
5419
5420                 return handle_emulation_failure(vcpu);
5421         }
5422
5423         if (ctxt->have_exception) {
5424                 r = EMULATE_DONE;
5425                 if (inject_emulated_exception(vcpu))
5426                         return r;
5427         } else if (vcpu->arch.pio.count) {
5428                 if (!vcpu->arch.pio.in) {
5429                         /* FIXME: return into emulator if single-stepping.  */
5430                         vcpu->arch.pio.count = 0;
5431                 } else {
5432                         writeback = false;
5433                         vcpu->arch.complete_userspace_io = complete_emulated_pio;
5434                 }
5435                 r = EMULATE_USER_EXIT;
5436         } else if (vcpu->mmio_needed) {
5437                 if (!vcpu->mmio_is_write)
5438                         writeback = false;
5439                 r = EMULATE_USER_EXIT;
5440                 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5441         } else if (r == EMULATION_RESTART)
5442                 goto restart;
5443         else
5444                 r = EMULATE_DONE;
5445
5446         if (writeback) {
5447                 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5448                 toggle_interruptibility(vcpu, ctxt->interruptibility);
5449                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5450                 if (vcpu->arch.hflags != ctxt->emul_flags)
5451                         kvm_set_hflags(vcpu, ctxt->emul_flags);
5452                 kvm_rip_write(vcpu, ctxt->eip);
5453                 if (r == EMULATE_DONE)
5454                         kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5455                 if (!ctxt->have_exception ||
5456                     exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5457                         __kvm_set_rflags(vcpu, ctxt->eflags);
5458
5459                 /*
5460                  * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5461                  * do nothing, and it will be requested again as soon as
5462                  * the shadow expires.  But we still need to check here,
5463                  * because POPF has no interrupt shadow.
5464                  */
5465                 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5466                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5467         } else
5468                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5469
5470         return r;
5471 }
5472 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5473
5474 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5475 {
5476         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5477         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5478                                             size, port, &val, 1);
5479         /* do not return to emulator after return from userspace */
5480         vcpu->arch.pio.count = 0;
5481         return ret;
5482 }
5483 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5484
5485 static void tsc_bad(void *info)
5486 {
5487         __this_cpu_write(cpu_tsc_khz, 0);
5488 }
5489
5490 static void tsc_khz_changed(void *data)
5491 {
5492         struct cpufreq_freqs *freq = data;
5493         unsigned long khz = 0;
5494
5495         if (data)
5496                 khz = freq->new;
5497         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5498                 khz = cpufreq_quick_get(raw_smp_processor_id());
5499         if (!khz)
5500                 khz = tsc_khz;
5501         __this_cpu_write(cpu_tsc_khz, khz);
5502 }
5503
5504 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5505                                      void *data)
5506 {
5507         struct cpufreq_freqs *freq = data;
5508         struct kvm *kvm;
5509         struct kvm_vcpu *vcpu;
5510         int i, send_ipi = 0;
5511
5512         /*
5513          * We allow guests to temporarily run on slowing clocks,
5514          * provided we notify them after, or to run on accelerating
5515          * clocks, provided we notify them before.  Thus time never
5516          * goes backwards.
5517          *
5518          * However, we have a problem.  We can't atomically update
5519          * the frequency of a given CPU from this function; it is
5520          * merely a notifier, which can be called from any CPU.
5521          * Changing the TSC frequency at arbitrary points in time
5522          * requires a recomputation of local variables related to
5523          * the TSC for each VCPU.  We must flag these local variables
5524          * to be updated and be sure the update takes place with the
5525          * new frequency before any guests proceed.
5526          *
5527          * Unfortunately, the combination of hotplug CPU and frequency
5528          * change creates an intractable locking scenario; the order
5529          * of when these callouts happen is undefined with respect to
5530          * CPU hotplug, and they can race with each other.  As such,
5531          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5532          * undefined; you can actually have a CPU frequency change take
5533          * place in between the computation of X and the setting of the
5534          * variable.  To protect against this problem, all updates of
5535          * the per_cpu tsc_khz variable are done in an interrupt
5536          * protected IPI, and all callers wishing to update the value
5537          * must wait for a synchronous IPI to complete (which is trivial
5538          * if the caller is on the CPU already).  This establishes the
5539          * necessary total order on variable updates.
5540          *
5541          * Note that because a guest time update may take place
5542          * anytime after the setting of the VCPU's request bit, the
5543          * correct TSC value must be set before the request.  However,
5544          * to ensure the update actually makes it to any guest which
5545          * starts running in hardware virtualization between the set
5546          * and the acquisition of the spinlock, we must also ping the
5547          * CPU after setting the request bit.
5548          *
5549          */
5550
5551         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5552                 return 0;
5553         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5554                 return 0;
5555
5556         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5557
5558         spin_lock(&kvm_lock);
5559         list_for_each_entry(kvm, &vm_list, vm_list) {
5560                 kvm_for_each_vcpu(i, vcpu, kvm) {
5561                         if (vcpu->cpu != freq->cpu)
5562                                 continue;
5563                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5564                         if (vcpu->cpu != smp_processor_id())
5565                                 send_ipi = 1;
5566                 }
5567         }
5568         spin_unlock(&kvm_lock);
5569
5570         if (freq->old < freq->new && send_ipi) {
5571                 /*
5572                  * We upscale the frequency.  Must make the guest
5573                  * doesn't see old kvmclock values while running with
5574                  * the new frequency, otherwise we risk the guest sees
5575                  * time go backwards.
5576                  *
5577                  * In case we update the frequency for another cpu
5578                  * (which might be in guest context) send an interrupt
5579                  * to kick the cpu out of guest context.  Next time
5580                  * guest context is entered kvmclock will be updated,
5581                  * so the guest will not see stale values.
5582                  */
5583                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5584         }
5585         return 0;
5586 }
5587
5588 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5589         .notifier_call  = kvmclock_cpufreq_notifier
5590 };
5591
5592 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5593                                         unsigned long action, void *hcpu)
5594 {
5595         unsigned int cpu = (unsigned long)hcpu;
5596
5597         switch (action) {
5598                 case CPU_ONLINE:
5599                 case CPU_DOWN_FAILED:
5600                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5601                         break;
5602                 case CPU_DOWN_PREPARE:
5603                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
5604                         break;
5605         }
5606         return NOTIFY_OK;
5607 }
5608
5609 static struct notifier_block kvmclock_cpu_notifier_block = {
5610         .notifier_call  = kvmclock_cpu_notifier,
5611         .priority = -INT_MAX
5612 };
5613
5614 static void kvm_timer_init(void)
5615 {
5616         int cpu;
5617
5618         max_tsc_khz = tsc_khz;
5619
5620         cpu_notifier_register_begin();
5621         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5622 #ifdef CONFIG_CPU_FREQ
5623                 struct cpufreq_policy policy;
5624                 memset(&policy, 0, sizeof(policy));
5625                 cpu = get_cpu();
5626                 cpufreq_get_policy(&policy, cpu);
5627                 if (policy.cpuinfo.max_freq)
5628                         max_tsc_khz = policy.cpuinfo.max_freq;
5629                 put_cpu();
5630 #endif
5631                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5632                                           CPUFREQ_TRANSITION_NOTIFIER);
5633         }
5634         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5635         for_each_online_cpu(cpu)
5636                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5637
5638         __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5639         cpu_notifier_register_done();
5640
5641 }
5642
5643 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5644
5645 int kvm_is_in_guest(void)
5646 {
5647         return __this_cpu_read(current_vcpu) != NULL;
5648 }
5649
5650 static int kvm_is_user_mode(void)
5651 {
5652         int user_mode = 3;
5653
5654         if (__this_cpu_read(current_vcpu))
5655                 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5656
5657         return user_mode != 0;
5658 }
5659
5660 static unsigned long kvm_get_guest_ip(void)
5661 {
5662         unsigned long ip = 0;
5663
5664         if (__this_cpu_read(current_vcpu))
5665                 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5666
5667         return ip;
5668 }
5669
5670 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5671         .is_in_guest            = kvm_is_in_guest,
5672         .is_user_mode           = kvm_is_user_mode,
5673         .get_guest_ip           = kvm_get_guest_ip,
5674 };
5675
5676 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5677 {
5678         __this_cpu_write(current_vcpu, vcpu);
5679 }
5680 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5681
5682 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5683 {
5684         __this_cpu_write(current_vcpu, NULL);
5685 }
5686 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5687
5688 static void kvm_set_mmio_spte_mask(void)
5689 {
5690         u64 mask;
5691         int maxphyaddr = boot_cpu_data.x86_phys_bits;
5692
5693         /*
5694          * Set the reserved bits and the present bit of an paging-structure
5695          * entry to generate page fault with PFER.RSV = 1.
5696          */
5697          /* Mask the reserved physical address bits. */
5698         mask = rsvd_bits(maxphyaddr, 51);
5699
5700         /* Bit 62 is always reserved for 32bit host. */
5701         mask |= 0x3ull << 62;
5702
5703         /* Set the present bit. */
5704         mask |= 1ull;
5705
5706 #ifdef CONFIG_X86_64
5707         /*
5708          * If reserved bit is not supported, clear the present bit to disable
5709          * mmio page fault.
5710          */
5711         if (maxphyaddr == 52)
5712                 mask &= ~1ull;
5713 #endif
5714
5715         kvm_mmu_set_mmio_spte_mask(mask);
5716 }
5717
5718 #ifdef CONFIG_X86_64
5719 static void pvclock_gtod_update_fn(struct work_struct *work)
5720 {
5721         struct kvm *kvm;
5722
5723         struct kvm_vcpu *vcpu;
5724         int i;
5725
5726         spin_lock(&kvm_lock);
5727         list_for_each_entry(kvm, &vm_list, vm_list)
5728                 kvm_for_each_vcpu(i, vcpu, kvm)
5729                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5730         atomic_set(&kvm_guest_has_master_clock, 0);
5731         spin_unlock(&kvm_lock);
5732 }
5733
5734 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5735
5736 /*
5737  * Notification about pvclock gtod data update.
5738  */
5739 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5740                                void *priv)
5741 {
5742         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5743         struct timekeeper *tk = priv;
5744
5745         update_pvclock_gtod(tk);
5746
5747         /* disable master clock if host does not trust, or does not
5748          * use, TSC clocksource
5749          */
5750         if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5751             atomic_read(&kvm_guest_has_master_clock) != 0)
5752                 queue_work(system_long_wq, &pvclock_gtod_work);
5753
5754         return 0;
5755 }
5756
5757 static struct notifier_block pvclock_gtod_notifier = {
5758         .notifier_call = pvclock_gtod_notify,
5759 };
5760 #endif
5761
5762 int kvm_arch_init(void *opaque)
5763 {
5764         int r;
5765         struct kvm_x86_ops *ops = opaque;
5766
5767         if (kvm_x86_ops) {
5768                 printk(KERN_ERR "kvm: already loaded the other module\n");
5769                 r = -EEXIST;
5770                 goto out;
5771         }
5772
5773         if (!ops->cpu_has_kvm_support()) {
5774                 printk(KERN_ERR "kvm: no hardware support\n");
5775                 r = -EOPNOTSUPP;
5776                 goto out;
5777         }
5778         if (ops->disabled_by_bios()) {
5779                 printk(KERN_ERR "kvm: disabled by bios\n");
5780                 r = -EOPNOTSUPP;
5781                 goto out;
5782         }
5783
5784         r = -ENOMEM;
5785         shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5786         if (!shared_msrs) {
5787                 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5788                 goto out;
5789         }
5790
5791         r = kvm_mmu_module_init();
5792         if (r)
5793                 goto out_free_percpu;
5794
5795         kvm_set_mmio_spte_mask();
5796
5797         kvm_x86_ops = ops;
5798
5799         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5800                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
5801
5802         kvm_timer_init();
5803
5804         perf_register_guest_info_callbacks(&kvm_guest_cbs);
5805
5806         if (cpu_has_xsave)
5807                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5808
5809         kvm_lapic_init();
5810 #ifdef CONFIG_X86_64
5811         pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5812 #endif
5813
5814         return 0;
5815
5816 out_free_percpu:
5817         free_percpu(shared_msrs);
5818 out:
5819         return r;
5820 }
5821
5822 void kvm_arch_exit(void)
5823 {
5824         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5825
5826         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5827                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5828                                             CPUFREQ_TRANSITION_NOTIFIER);
5829         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5830 #ifdef CONFIG_X86_64
5831         pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5832 #endif
5833         kvm_x86_ops = NULL;
5834         kvm_mmu_module_exit();
5835         free_percpu(shared_msrs);
5836 }
5837
5838 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5839 {
5840         ++vcpu->stat.halt_exits;
5841         if (lapic_in_kernel(vcpu)) {
5842                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5843                 return 1;
5844         } else {
5845                 vcpu->run->exit_reason = KVM_EXIT_HLT;
5846                 return 0;
5847         }
5848 }
5849 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5850
5851 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5852 {
5853         kvm_x86_ops->skip_emulated_instruction(vcpu);
5854         return kvm_vcpu_halt(vcpu);
5855 }
5856 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5857
5858 /*
5859  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5860  *
5861  * @apicid - apicid of vcpu to be kicked.
5862  */
5863 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5864 {
5865         struct kvm_lapic_irq lapic_irq;
5866
5867         lapic_irq.shorthand = 0;
5868         lapic_irq.dest_mode = 0;
5869         lapic_irq.dest_id = apicid;
5870         lapic_irq.msi_redir_hint = false;
5871
5872         lapic_irq.delivery_mode = APIC_DM_REMRD;
5873         kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5874 }
5875
5876 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5877 {
5878         unsigned long nr, a0, a1, a2, a3, ret;
5879         int op_64_bit, r = 1;
5880
5881         kvm_x86_ops->skip_emulated_instruction(vcpu);
5882
5883         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5884                 return kvm_hv_hypercall(vcpu);
5885
5886         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5887         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5888         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5889         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5890         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5891
5892         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5893
5894         op_64_bit = is_64_bit_mode(vcpu);
5895         if (!op_64_bit) {
5896                 nr &= 0xFFFFFFFF;
5897                 a0 &= 0xFFFFFFFF;
5898                 a1 &= 0xFFFFFFFF;
5899                 a2 &= 0xFFFFFFFF;
5900                 a3 &= 0xFFFFFFFF;
5901         }
5902
5903         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5904                 ret = -KVM_EPERM;
5905                 goto out;
5906         }
5907
5908         switch (nr) {
5909         case KVM_HC_VAPIC_POLL_IRQ:
5910                 ret = 0;
5911                 break;
5912         case KVM_HC_KICK_CPU:
5913                 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5914                 ret = 0;
5915                 break;
5916         default:
5917                 ret = -KVM_ENOSYS;
5918                 break;
5919         }
5920 out:
5921         if (!op_64_bit)
5922                 ret = (u32)ret;
5923         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5924         ++vcpu->stat.hypercalls;
5925         return r;
5926 }
5927 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5928
5929 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5930 {
5931         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5932         char instruction[3];
5933         unsigned long rip = kvm_rip_read(vcpu);
5934
5935         kvm_x86_ops->patch_hypercall(vcpu, instruction);
5936
5937         return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5938 }
5939
5940 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5941 {
5942         return vcpu->run->request_interrupt_window &&
5943                 likely(!pic_in_kernel(vcpu->kvm));
5944 }
5945
5946 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5947 {
5948         struct kvm_run *kvm_run = vcpu->run;
5949
5950         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5951         kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
5952         kvm_run->cr8 = kvm_get_cr8(vcpu);
5953         kvm_run->apic_base = kvm_get_apic_base(vcpu);
5954         kvm_run->ready_for_interrupt_injection =
5955                 pic_in_kernel(vcpu->kvm) ||
5956                 kvm_vcpu_ready_for_interrupt_injection(vcpu);
5957 }
5958
5959 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5960 {
5961         int max_irr, tpr;
5962
5963         if (!kvm_x86_ops->update_cr8_intercept)
5964                 return;
5965
5966         if (!vcpu->arch.apic)
5967                 return;
5968
5969         if (!vcpu->arch.apic->vapic_addr)
5970                 max_irr = kvm_lapic_find_highest_irr(vcpu);
5971         else
5972                 max_irr = -1;
5973
5974         if (max_irr != -1)
5975                 max_irr >>= 4;
5976
5977         tpr = kvm_lapic_get_cr8(vcpu);
5978
5979         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5980 }
5981
5982 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5983 {
5984         int r;
5985
5986         /* try to reinject previous events if any */
5987         if (vcpu->arch.exception.pending) {
5988                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5989                                         vcpu->arch.exception.has_error_code,
5990                                         vcpu->arch.exception.error_code);
5991
5992                 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5993                         __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5994                                              X86_EFLAGS_RF);
5995
5996                 if (vcpu->arch.exception.nr == DB_VECTOR &&
5997                     (vcpu->arch.dr7 & DR7_GD)) {
5998                         vcpu->arch.dr7 &= ~DR7_GD;
5999                         kvm_update_dr7(vcpu);
6000                 }
6001
6002                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6003                                           vcpu->arch.exception.has_error_code,
6004                                           vcpu->arch.exception.error_code,
6005                                           vcpu->arch.exception.reinject);
6006                 return 0;
6007         }
6008
6009         if (vcpu->arch.nmi_injected) {
6010                 kvm_x86_ops->set_nmi(vcpu);
6011                 return 0;
6012         }
6013
6014         if (vcpu->arch.interrupt.pending) {
6015                 kvm_x86_ops->set_irq(vcpu);
6016                 return 0;
6017         }
6018
6019         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6020                 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6021                 if (r != 0)
6022                         return r;
6023         }
6024
6025         /* try to inject new event if pending */
6026         if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6027                 --vcpu->arch.nmi_pending;
6028                 vcpu->arch.nmi_injected = true;
6029                 kvm_x86_ops->set_nmi(vcpu);
6030         } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6031                 /*
6032                  * Because interrupts can be injected asynchronously, we are
6033                  * calling check_nested_events again here to avoid a race condition.
6034                  * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6035                  * proposal and current concerns.  Perhaps we should be setting
6036                  * KVM_REQ_EVENT only on certain events and not unconditionally?
6037                  */
6038                 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6039                         r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6040                         if (r != 0)
6041                                 return r;
6042                 }
6043                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6044                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6045                                             false);
6046                         kvm_x86_ops->set_irq(vcpu);
6047                 }
6048         }
6049         return 0;
6050 }
6051
6052 static void process_nmi(struct kvm_vcpu *vcpu)
6053 {
6054         unsigned limit = 2;
6055
6056         /*
6057          * x86 is limited to one NMI running, and one NMI pending after it.
6058          * If an NMI is already in progress, limit further NMIs to just one.
6059          * Otherwise, allow two (and we'll inject the first one immediately).
6060          */
6061         if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6062                 limit = 1;
6063
6064         vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6065         vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6066         kvm_make_request(KVM_REQ_EVENT, vcpu);
6067 }
6068
6069 #define put_smstate(type, buf, offset, val)                       \
6070         *(type *)((buf) + (offset) - 0x7e00) = val
6071
6072 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6073 {
6074         u32 flags = 0;
6075         flags |= seg->g       << 23;
6076         flags |= seg->db      << 22;
6077         flags |= seg->l       << 21;
6078         flags |= seg->avl     << 20;
6079         flags |= seg->present << 15;
6080         flags |= seg->dpl     << 13;
6081         flags |= seg->s       << 12;
6082         flags |= seg->type    << 8;
6083         return flags;
6084 }
6085
6086 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6087 {
6088         struct kvm_segment seg;
6089         int offset;
6090
6091         kvm_get_segment(vcpu, &seg, n);
6092         put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6093
6094         if (n < 3)
6095                 offset = 0x7f84 + n * 12;
6096         else
6097                 offset = 0x7f2c + (n - 3) * 12;
6098
6099         put_smstate(u32, buf, offset + 8, seg.base);
6100         put_smstate(u32, buf, offset + 4, seg.limit);
6101         put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6102 }
6103
6104 #ifdef CONFIG_X86_64
6105 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6106 {
6107         struct kvm_segment seg;
6108         int offset;
6109         u16 flags;
6110
6111         kvm_get_segment(vcpu, &seg, n);
6112         offset = 0x7e00 + n * 16;
6113
6114         flags = process_smi_get_segment_flags(&seg) >> 8;
6115         put_smstate(u16, buf, offset, seg.selector);
6116         put_smstate(u16, buf, offset + 2, flags);
6117         put_smstate(u32, buf, offset + 4, seg.limit);
6118         put_smstate(u64, buf, offset + 8, seg.base);
6119 }
6120 #endif
6121
6122 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6123 {
6124         struct desc_ptr dt;
6125         struct kvm_segment seg;
6126         unsigned long val;
6127         int i;
6128
6129         put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6130         put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6131         put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6132         put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6133
6134         for (i = 0; i < 8; i++)
6135                 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6136
6137         kvm_get_dr(vcpu, 6, &val);
6138         put_smstate(u32, buf, 0x7fcc, (u32)val);
6139         kvm_get_dr(vcpu, 7, &val);
6140         put_smstate(u32, buf, 0x7fc8, (u32)val);
6141
6142         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6143         put_smstate(u32, buf, 0x7fc4, seg.selector);
6144         put_smstate(u32, buf, 0x7f64, seg.base);
6145         put_smstate(u32, buf, 0x7f60, seg.limit);
6146         put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6147
6148         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6149         put_smstate(u32, buf, 0x7fc0, seg.selector);
6150         put_smstate(u32, buf, 0x7f80, seg.base);
6151         put_smstate(u32, buf, 0x7f7c, seg.limit);
6152         put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6153
6154         kvm_x86_ops->get_gdt(vcpu, &dt);
6155         put_smstate(u32, buf, 0x7f74, dt.address);
6156         put_smstate(u32, buf, 0x7f70, dt.size);
6157
6158         kvm_x86_ops->get_idt(vcpu, &dt);
6159         put_smstate(u32, buf, 0x7f58, dt.address);
6160         put_smstate(u32, buf, 0x7f54, dt.size);
6161
6162         for (i = 0; i < 6; i++)
6163                 process_smi_save_seg_32(vcpu, buf, i);
6164
6165         put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6166
6167         /* revision id */
6168         put_smstate(u32, buf, 0x7efc, 0x00020000);
6169         put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6170 }
6171
6172 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6173 {
6174 #ifdef CONFIG_X86_64
6175         struct desc_ptr dt;
6176         struct kvm_segment seg;
6177         unsigned long val;
6178         int i;
6179
6180         for (i = 0; i < 16; i++)
6181                 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6182
6183         put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6184         put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6185
6186         kvm_get_dr(vcpu, 6, &val);
6187         put_smstate(u64, buf, 0x7f68, val);
6188         kvm_get_dr(vcpu, 7, &val);
6189         put_smstate(u64, buf, 0x7f60, val);
6190
6191         put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6192         put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6193         put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6194
6195         put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6196
6197         /* revision id */
6198         put_smstate(u32, buf, 0x7efc, 0x00020064);
6199
6200         put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6201
6202         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6203         put_smstate(u16, buf, 0x7e90, seg.selector);
6204         put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6205         put_smstate(u32, buf, 0x7e94, seg.limit);
6206         put_smstate(u64, buf, 0x7e98, seg.base);
6207
6208         kvm_x86_ops->get_idt(vcpu, &dt);
6209         put_smstate(u32, buf, 0x7e84, dt.size);
6210         put_smstate(u64, buf, 0x7e88, dt.address);
6211
6212         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6213         put_smstate(u16, buf, 0x7e70, seg.selector);
6214         put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6215         put_smstate(u32, buf, 0x7e74, seg.limit);
6216         put_smstate(u64, buf, 0x7e78, seg.base);
6217
6218         kvm_x86_ops->get_gdt(vcpu, &dt);
6219         put_smstate(u32, buf, 0x7e64, dt.size);
6220         put_smstate(u64, buf, 0x7e68, dt.address);
6221
6222         for (i = 0; i < 6; i++)
6223                 process_smi_save_seg_64(vcpu, buf, i);
6224 #else
6225         WARN_ON_ONCE(1);
6226 #endif
6227 }
6228
6229 static void process_smi(struct kvm_vcpu *vcpu)
6230 {
6231         struct kvm_segment cs, ds;
6232         struct desc_ptr dt;
6233         char buf[512];
6234         u32 cr0;
6235
6236         if (is_smm(vcpu)) {
6237                 vcpu->arch.smi_pending = true;
6238                 return;
6239         }
6240
6241         trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6242         vcpu->arch.hflags |= HF_SMM_MASK;
6243         memset(buf, 0, 512);
6244         if (guest_cpuid_has_longmode(vcpu))
6245                 process_smi_save_state_64(vcpu, buf);
6246         else
6247                 process_smi_save_state_32(vcpu, buf);
6248
6249         kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6250
6251         if (kvm_x86_ops->get_nmi_mask(vcpu))
6252                 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6253         else
6254                 kvm_x86_ops->set_nmi_mask(vcpu, true);
6255
6256         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6257         kvm_rip_write(vcpu, 0x8000);
6258
6259         cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6260         kvm_x86_ops->set_cr0(vcpu, cr0);
6261         vcpu->arch.cr0 = cr0;
6262
6263         kvm_x86_ops->set_cr4(vcpu, 0);
6264
6265         /* Undocumented: IDT limit is set to zero on entry to SMM.  */
6266         dt.address = dt.size = 0;
6267         kvm_x86_ops->set_idt(vcpu, &dt);
6268
6269         __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6270
6271         cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6272         cs.base = vcpu->arch.smbase;
6273
6274         ds.selector = 0;
6275         ds.base = 0;
6276
6277         cs.limit    = ds.limit = 0xffffffff;
6278         cs.type     = ds.type = 0x3;
6279         cs.dpl      = ds.dpl = 0;
6280         cs.db       = ds.db = 0;
6281         cs.s        = ds.s = 1;
6282         cs.l        = ds.l = 0;
6283         cs.g        = ds.g = 1;
6284         cs.avl      = ds.avl = 0;
6285         cs.present  = ds.present = 1;
6286         cs.unusable = ds.unusable = 0;
6287         cs.padding  = ds.padding = 0;
6288
6289         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6290         kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6291         kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6292         kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6293         kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6294         kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6295
6296         if (guest_cpuid_has_longmode(vcpu))
6297                 kvm_x86_ops->set_efer(vcpu, 0);
6298
6299         kvm_update_cpuid(vcpu);
6300         kvm_mmu_reset_context(vcpu);
6301 }
6302
6303 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6304 {
6305         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6306                 return;
6307
6308         memset(vcpu->arch.eoi_exit_bitmap, 0, 256 / 8);
6309
6310         if (irqchip_split(vcpu->kvm))
6311                 kvm_scan_ioapic_routes(vcpu, vcpu->arch.eoi_exit_bitmap);
6312         else {
6313                 kvm_x86_ops->sync_pir_to_irr(vcpu);
6314                 kvm_ioapic_scan_entry(vcpu, vcpu->arch.eoi_exit_bitmap);
6315         }
6316         kvm_x86_ops->load_eoi_exitmap(vcpu);
6317 }
6318
6319 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6320 {
6321         ++vcpu->stat.tlb_flush;
6322         kvm_x86_ops->tlb_flush(vcpu);
6323 }
6324
6325 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6326 {
6327         struct page *page = NULL;
6328
6329         if (!lapic_in_kernel(vcpu))
6330                 return;
6331
6332         if (!kvm_x86_ops->set_apic_access_page_addr)
6333                 return;
6334
6335         page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6336         if (is_error_page(page))
6337                 return;
6338         kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6339
6340         /*
6341          * Do not pin apic access page in memory, the MMU notifier
6342          * will call us again if it is migrated or swapped out.
6343          */
6344         put_page(page);
6345 }
6346 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6347
6348 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6349                                            unsigned long address)
6350 {
6351         /*
6352          * The physical address of apic access page is stored in the VMCS.
6353          * Update it when it becomes invalid.
6354          */
6355         if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6356                 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6357 }
6358
6359 /*
6360  * Returns 1 to let vcpu_run() continue the guest execution loop without
6361  * exiting to the userspace.  Otherwise, the value will be returned to the
6362  * userspace.
6363  */
6364 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6365 {
6366         int r;
6367         bool req_int_win =
6368                 dm_request_for_irq_injection(vcpu) &&
6369                 kvm_cpu_accept_dm_intr(vcpu);
6370
6371         bool req_immediate_exit = false;
6372
6373         if (vcpu->requests) {
6374                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6375                         kvm_mmu_unload(vcpu);
6376                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6377                         __kvm_migrate_timers(vcpu);
6378                 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6379                         kvm_gen_update_masterclock(vcpu->kvm);
6380                 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6381                         kvm_gen_kvmclock_update(vcpu);
6382                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6383                         r = kvm_guest_time_update(vcpu);
6384                         if (unlikely(r))
6385                                 goto out;
6386                 }
6387                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6388                         kvm_mmu_sync_roots(vcpu);
6389                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6390                         kvm_vcpu_flush_tlb(vcpu);
6391                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6392                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6393                         r = 0;
6394                         goto out;
6395                 }
6396                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6397                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6398                         r = 0;
6399                         goto out;
6400                 }
6401                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6402                         vcpu->fpu_active = 0;
6403                         kvm_x86_ops->fpu_deactivate(vcpu);
6404                 }
6405                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6406                         /* Page is swapped out. Do synthetic halt */
6407                         vcpu->arch.apf.halted = true;
6408                         r = 1;
6409                         goto out;
6410                 }
6411                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6412                         record_steal_time(vcpu);
6413                 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6414                         process_smi(vcpu);
6415                 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6416                         process_nmi(vcpu);
6417                 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6418                         kvm_pmu_handle_event(vcpu);
6419                 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6420                         kvm_pmu_deliver_pmi(vcpu);
6421                 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6422                         BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6423                         if (test_bit(vcpu->arch.pending_ioapic_eoi,
6424                                      (void *) vcpu->arch.eoi_exit_bitmap)) {
6425                                 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6426                                 vcpu->run->eoi.vector =
6427                                                 vcpu->arch.pending_ioapic_eoi;
6428                                 r = 0;
6429                                 goto out;
6430                         }
6431                 }
6432                 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6433                         vcpu_scan_ioapic(vcpu);
6434                 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6435                         kvm_vcpu_reload_apic_access_page(vcpu);
6436                 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6437                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6438                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6439                         r = 0;
6440                         goto out;
6441                 }
6442                 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6443                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6444                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6445                         r = 0;
6446                         goto out;
6447                 }
6448         }
6449
6450         /*
6451          * KVM_REQ_EVENT is not set when posted interrupts are set by
6452          * VT-d hardware, so we have to update RVI unconditionally.
6453          */
6454         if (kvm_lapic_enabled(vcpu)) {
6455                 /*
6456                  * Update architecture specific hints for APIC
6457                  * virtual interrupt delivery.
6458                  */
6459                 if (kvm_x86_ops->hwapic_irr_update)
6460                         kvm_x86_ops->hwapic_irr_update(vcpu,
6461                                 kvm_lapic_find_highest_irr(vcpu));
6462         }
6463
6464         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6465                 kvm_apic_accept_events(vcpu);
6466                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6467                         r = 1;
6468                         goto out;
6469                 }
6470
6471                 if (inject_pending_event(vcpu, req_int_win) != 0)
6472                         req_immediate_exit = true;
6473                 /* enable NMI/IRQ window open exits if needed */
6474                 else {
6475                         if (vcpu->arch.nmi_pending)
6476                                 kvm_x86_ops->enable_nmi_window(vcpu);
6477                         if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6478                                 kvm_x86_ops->enable_irq_window(vcpu);
6479                 }
6480
6481                 if (kvm_lapic_enabled(vcpu)) {
6482                         update_cr8_intercept(vcpu);
6483                         kvm_lapic_sync_to_vapic(vcpu);
6484                 }
6485         }
6486
6487         r = kvm_mmu_reload(vcpu);
6488         if (unlikely(r)) {
6489                 goto cancel_injection;
6490         }
6491
6492         preempt_disable();
6493
6494         kvm_x86_ops->prepare_guest_switch(vcpu);
6495         if (vcpu->fpu_active)
6496                 kvm_load_guest_fpu(vcpu);
6497         vcpu->mode = IN_GUEST_MODE;
6498
6499         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6500
6501         /* We should set ->mode before check ->requests,
6502          * see the comment in make_all_cpus_request.
6503          */
6504         smp_mb__after_srcu_read_unlock();
6505
6506         local_irq_disable();
6507
6508         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6509             || need_resched() || signal_pending(current)) {
6510                 vcpu->mode = OUTSIDE_GUEST_MODE;
6511                 smp_wmb();
6512                 local_irq_enable();
6513                 preempt_enable();
6514                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6515                 r = 1;
6516                 goto cancel_injection;
6517         }
6518
6519         kvm_load_guest_xcr0(vcpu);
6520
6521         if (req_immediate_exit)
6522                 smp_send_reschedule(vcpu->cpu);
6523
6524         trace_kvm_entry(vcpu->vcpu_id);
6525         wait_lapic_expire(vcpu);
6526         __kvm_guest_enter();
6527
6528         if (unlikely(vcpu->arch.switch_db_regs)) {
6529                 set_debugreg(0, 7);
6530                 set_debugreg(vcpu->arch.eff_db[0], 0);
6531                 set_debugreg(vcpu->arch.eff_db[1], 1);
6532                 set_debugreg(vcpu->arch.eff_db[2], 2);
6533                 set_debugreg(vcpu->arch.eff_db[3], 3);
6534                 set_debugreg(vcpu->arch.dr6, 6);
6535                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6536         }
6537
6538         kvm_x86_ops->run(vcpu);
6539
6540         /*
6541          * Do this here before restoring debug registers on the host.  And
6542          * since we do this before handling the vmexit, a DR access vmexit
6543          * can (a) read the correct value of the debug registers, (b) set
6544          * KVM_DEBUGREG_WONT_EXIT again.
6545          */
6546         if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6547                 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6548                 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6549                 kvm_update_dr0123(vcpu);
6550                 kvm_update_dr6(vcpu);
6551                 kvm_update_dr7(vcpu);
6552                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6553         }
6554
6555         /*
6556          * If the guest has used debug registers, at least dr7
6557          * will be disabled while returning to the host.
6558          * If we don't have active breakpoints in the host, we don't
6559          * care about the messed up debug address registers. But if
6560          * we have some of them active, restore the old state.
6561          */
6562         if (hw_breakpoint_active())
6563                 hw_breakpoint_restore();
6564
6565         vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6566
6567         vcpu->mode = OUTSIDE_GUEST_MODE;
6568         smp_wmb();
6569
6570         kvm_put_guest_xcr0(vcpu);
6571
6572         /* Interrupt is enabled by handle_external_intr() */
6573         kvm_x86_ops->handle_external_intr(vcpu);
6574
6575         ++vcpu->stat.exits;
6576
6577         /*
6578          * We must have an instruction between local_irq_enable() and
6579          * kvm_guest_exit(), so the timer interrupt isn't delayed by
6580          * the interrupt shadow.  The stat.exits increment will do nicely.
6581          * But we need to prevent reordering, hence this barrier():
6582          */
6583         barrier();
6584
6585         kvm_guest_exit();
6586
6587         preempt_enable();
6588
6589         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6590
6591         /*
6592          * Profile KVM exit RIPs:
6593          */
6594         if (unlikely(prof_on == KVM_PROFILING)) {
6595                 unsigned long rip = kvm_rip_read(vcpu);
6596                 profile_hit(KVM_PROFILING, (void *)rip);
6597         }
6598
6599         if (unlikely(vcpu->arch.tsc_always_catchup))
6600                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6601
6602         if (vcpu->arch.apic_attention)
6603                 kvm_lapic_sync_from_vapic(vcpu);
6604
6605         r = kvm_x86_ops->handle_exit(vcpu);
6606         return r;
6607
6608 cancel_injection:
6609         kvm_x86_ops->cancel_injection(vcpu);
6610         if (unlikely(vcpu->arch.apic_attention))
6611                 kvm_lapic_sync_from_vapic(vcpu);
6612 out:
6613         return r;
6614 }
6615
6616 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6617 {
6618         if (!kvm_arch_vcpu_runnable(vcpu) &&
6619             (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6620                 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6621                 kvm_vcpu_block(vcpu);
6622                 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6623
6624                 if (kvm_x86_ops->post_block)
6625                         kvm_x86_ops->post_block(vcpu);
6626
6627                 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6628                         return 1;
6629         }
6630
6631         kvm_apic_accept_events(vcpu);
6632         switch(vcpu->arch.mp_state) {
6633         case KVM_MP_STATE_HALTED:
6634                 vcpu->arch.pv.pv_unhalted = false;
6635                 vcpu->arch.mp_state =
6636                         KVM_MP_STATE_RUNNABLE;
6637         case KVM_MP_STATE_RUNNABLE:
6638                 vcpu->arch.apf.halted = false;
6639                 break;
6640         case KVM_MP_STATE_INIT_RECEIVED:
6641                 break;
6642         default:
6643                 return -EINTR;
6644                 break;
6645         }
6646         return 1;
6647 }
6648
6649 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6650 {
6651         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6652                 !vcpu->arch.apf.halted);
6653 }
6654
6655 static int vcpu_run(struct kvm_vcpu *vcpu)
6656 {
6657         int r;
6658         struct kvm *kvm = vcpu->kvm;
6659
6660         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6661
6662         for (;;) {
6663                 if (kvm_vcpu_running(vcpu)) {
6664                         r = vcpu_enter_guest(vcpu);
6665                 } else {
6666                         r = vcpu_block(kvm, vcpu);
6667                 }
6668
6669                 if (r <= 0)
6670                         break;
6671
6672                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6673                 if (kvm_cpu_has_pending_timer(vcpu))
6674                         kvm_inject_pending_timer_irqs(vcpu);
6675
6676                 if (dm_request_for_irq_injection(vcpu) &&
6677                         kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6678                         r = 0;
6679                         vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6680                         ++vcpu->stat.request_irq_exits;
6681                         break;
6682                 }
6683
6684                 kvm_check_async_pf_completion(vcpu);
6685
6686                 if (signal_pending(current)) {
6687                         r = -EINTR;
6688                         vcpu->run->exit_reason = KVM_EXIT_INTR;
6689                         ++vcpu->stat.signal_exits;
6690                         break;
6691                 }
6692                 if (need_resched()) {
6693                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6694                         cond_resched();
6695                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6696                 }
6697         }
6698
6699         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6700
6701         return r;
6702 }
6703
6704 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6705 {
6706         int r;
6707         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6708         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6709         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6710         if (r != EMULATE_DONE)
6711                 return 0;
6712         return 1;
6713 }
6714
6715 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6716 {
6717         BUG_ON(!vcpu->arch.pio.count);
6718
6719         return complete_emulated_io(vcpu);
6720 }
6721
6722 /*
6723  * Implements the following, as a state machine:
6724  *
6725  * read:
6726  *   for each fragment
6727  *     for each mmio piece in the fragment
6728  *       write gpa, len
6729  *       exit
6730  *       copy data
6731  *   execute insn
6732  *
6733  * write:
6734  *   for each fragment
6735  *     for each mmio piece in the fragment
6736  *       write gpa, len
6737  *       copy data
6738  *       exit
6739  */
6740 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6741 {
6742         struct kvm_run *run = vcpu->run;
6743         struct kvm_mmio_fragment *frag;
6744         unsigned len;
6745
6746         BUG_ON(!vcpu->mmio_needed);
6747
6748         /* Complete previous fragment */
6749         frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6750         len = min(8u, frag->len);
6751         if (!vcpu->mmio_is_write)
6752                 memcpy(frag->data, run->mmio.data, len);
6753
6754         if (frag->len <= 8) {
6755                 /* Switch to the next fragment. */
6756                 frag++;
6757                 vcpu->mmio_cur_fragment++;
6758         } else {
6759                 /* Go forward to the next mmio piece. */
6760                 frag->data += len;
6761                 frag->gpa += len;
6762                 frag->len -= len;
6763         }
6764
6765         if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6766                 vcpu->mmio_needed = 0;
6767
6768                 /* FIXME: return into emulator if single-stepping.  */
6769                 if (vcpu->mmio_is_write)
6770                         return 1;
6771                 vcpu->mmio_read_completed = 1;
6772                 return complete_emulated_io(vcpu);
6773         }
6774
6775         run->exit_reason = KVM_EXIT_MMIO;
6776         run->mmio.phys_addr = frag->gpa;
6777         if (vcpu->mmio_is_write)
6778                 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6779         run->mmio.len = min(8u, frag->len);
6780         run->mmio.is_write = vcpu->mmio_is_write;
6781         vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6782         return 0;
6783 }
6784
6785
6786 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6787 {
6788         struct fpu *fpu = &current->thread.fpu;
6789         int r;
6790         sigset_t sigsaved;
6791
6792         fpu__activate_curr(fpu);
6793
6794         if (vcpu->sigset_active)
6795                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6796
6797         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6798                 kvm_vcpu_block(vcpu);
6799                 kvm_apic_accept_events(vcpu);
6800                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6801                 r = -EAGAIN;
6802                 goto out;
6803         }
6804
6805         /* re-sync apic's tpr */
6806         if (!lapic_in_kernel(vcpu)) {
6807                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6808                         r = -EINVAL;
6809                         goto out;
6810                 }
6811         }
6812
6813         if (unlikely(vcpu->arch.complete_userspace_io)) {
6814                 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6815                 vcpu->arch.complete_userspace_io = NULL;
6816                 r = cui(vcpu);
6817                 if (r <= 0)
6818                         goto out;
6819         } else
6820                 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6821
6822         r = vcpu_run(vcpu);
6823
6824 out:
6825         post_kvm_run_save(vcpu);
6826         if (vcpu->sigset_active)
6827                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6828
6829         return r;
6830 }
6831
6832 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6833 {
6834         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6835                 /*
6836                  * We are here if userspace calls get_regs() in the middle of
6837                  * instruction emulation. Registers state needs to be copied
6838                  * back from emulation context to vcpu. Userspace shouldn't do
6839                  * that usually, but some bad designed PV devices (vmware
6840                  * backdoor interface) need this to work
6841                  */
6842                 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6843                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6844         }
6845         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6846         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6847         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6848         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6849         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6850         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6851         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6852         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6853 #ifdef CONFIG_X86_64
6854         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6855         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6856         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6857         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6858         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6859         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6860         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6861         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6862 #endif
6863
6864         regs->rip = kvm_rip_read(vcpu);
6865         regs->rflags = kvm_get_rflags(vcpu);
6866
6867         return 0;
6868 }
6869
6870 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6871 {
6872         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6873         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6874
6875         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6876         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6877         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6878         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6879         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6880         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6881         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6882         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6883 #ifdef CONFIG_X86_64
6884         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6885         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6886         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6887         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6888         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6889         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6890         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6891         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6892 #endif
6893
6894         kvm_rip_write(vcpu, regs->rip);
6895         kvm_set_rflags(vcpu, regs->rflags);
6896
6897         vcpu->arch.exception.pending = false;
6898
6899         kvm_make_request(KVM_REQ_EVENT, vcpu);
6900
6901         return 0;
6902 }
6903
6904 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6905 {
6906         struct kvm_segment cs;
6907
6908         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6909         *db = cs.db;
6910         *l = cs.l;
6911 }
6912 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6913
6914 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6915                                   struct kvm_sregs *sregs)
6916 {
6917         struct desc_ptr dt;
6918
6919         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6920         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6921         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6922         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6923         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6924         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6925
6926         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6927         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6928
6929         kvm_x86_ops->get_idt(vcpu, &dt);
6930         sregs->idt.limit = dt.size;
6931         sregs->idt.base = dt.address;
6932         kvm_x86_ops->get_gdt(vcpu, &dt);
6933         sregs->gdt.limit = dt.size;
6934         sregs->gdt.base = dt.address;
6935
6936         sregs->cr0 = kvm_read_cr0(vcpu);
6937         sregs->cr2 = vcpu->arch.cr2;
6938         sregs->cr3 = kvm_read_cr3(vcpu);
6939         sregs->cr4 = kvm_read_cr4(vcpu);
6940         sregs->cr8 = kvm_get_cr8(vcpu);
6941         sregs->efer = vcpu->arch.efer;
6942         sregs->apic_base = kvm_get_apic_base(vcpu);
6943
6944         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6945
6946         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6947                 set_bit(vcpu->arch.interrupt.nr,
6948                         (unsigned long *)sregs->interrupt_bitmap);
6949
6950         return 0;
6951 }
6952
6953 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6954                                     struct kvm_mp_state *mp_state)
6955 {
6956         kvm_apic_accept_events(vcpu);
6957         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6958                                         vcpu->arch.pv.pv_unhalted)
6959                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6960         else
6961                 mp_state->mp_state = vcpu->arch.mp_state;
6962
6963         return 0;
6964 }
6965
6966 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6967                                     struct kvm_mp_state *mp_state)
6968 {
6969         if (!kvm_vcpu_has_lapic(vcpu) &&
6970             mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6971                 return -EINVAL;
6972
6973         if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6974                 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6975                 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6976         } else
6977                 vcpu->arch.mp_state = mp_state->mp_state;
6978         kvm_make_request(KVM_REQ_EVENT, vcpu);
6979         return 0;
6980 }
6981
6982 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6983                     int reason, bool has_error_code, u32 error_code)
6984 {
6985         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6986         int ret;
6987
6988         init_emulate_ctxt(vcpu);
6989
6990         ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6991                                    has_error_code, error_code);
6992
6993         if (ret)
6994                 return EMULATE_FAIL;
6995
6996         kvm_rip_write(vcpu, ctxt->eip);
6997         kvm_set_rflags(vcpu, ctxt->eflags);
6998         kvm_make_request(KVM_REQ_EVENT, vcpu);
6999         return EMULATE_DONE;
7000 }
7001 EXPORT_SYMBOL_GPL(kvm_task_switch);
7002
7003 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7004                                   struct kvm_sregs *sregs)
7005 {
7006         struct msr_data apic_base_msr;
7007         int mmu_reset_needed = 0;
7008         int pending_vec, max_bits, idx;
7009         struct desc_ptr dt;
7010
7011         if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7012                 return -EINVAL;
7013
7014         dt.size = sregs->idt.limit;
7015         dt.address = sregs->idt.base;
7016         kvm_x86_ops->set_idt(vcpu, &dt);
7017         dt.size = sregs->gdt.limit;
7018         dt.address = sregs->gdt.base;
7019         kvm_x86_ops->set_gdt(vcpu, &dt);
7020
7021         vcpu->arch.cr2 = sregs->cr2;
7022         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7023         vcpu->arch.cr3 = sregs->cr3;
7024         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7025
7026         kvm_set_cr8(vcpu, sregs->cr8);
7027
7028         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7029         kvm_x86_ops->set_efer(vcpu, sregs->efer);
7030         apic_base_msr.data = sregs->apic_base;
7031         apic_base_msr.host_initiated = true;
7032         kvm_set_apic_base(vcpu, &apic_base_msr);
7033
7034         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7035         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7036         vcpu->arch.cr0 = sregs->cr0;
7037
7038         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7039         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7040         if (sregs->cr4 & X86_CR4_OSXSAVE)
7041                 kvm_update_cpuid(vcpu);
7042
7043         idx = srcu_read_lock(&vcpu->kvm->srcu);
7044         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7045                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7046                 mmu_reset_needed = 1;
7047         }
7048         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7049
7050         if (mmu_reset_needed)
7051                 kvm_mmu_reset_context(vcpu);
7052
7053         max_bits = KVM_NR_INTERRUPTS;
7054         pending_vec = find_first_bit(
7055                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7056         if (pending_vec < max_bits) {
7057                 kvm_queue_interrupt(vcpu, pending_vec, false);
7058                 pr_debug("Set back pending irq %d\n", pending_vec);
7059         }
7060
7061         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7062         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7063         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7064         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7065         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7066         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7067
7068         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7069         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7070
7071         update_cr8_intercept(vcpu);
7072
7073         /* Older userspace won't unhalt the vcpu on reset. */
7074         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7075             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7076             !is_protmode(vcpu))
7077                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7078
7079         kvm_make_request(KVM_REQ_EVENT, vcpu);
7080
7081         return 0;
7082 }
7083
7084 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7085                                         struct kvm_guest_debug *dbg)
7086 {
7087         unsigned long rflags;
7088         int i, r;
7089
7090         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7091                 r = -EBUSY;
7092                 if (vcpu->arch.exception.pending)
7093                         goto out;
7094                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7095                         kvm_queue_exception(vcpu, DB_VECTOR);
7096                 else
7097                         kvm_queue_exception(vcpu, BP_VECTOR);
7098         }
7099
7100         /*
7101          * Read rflags as long as potentially injected trace flags are still
7102          * filtered out.
7103          */
7104         rflags = kvm_get_rflags(vcpu);
7105
7106         vcpu->guest_debug = dbg->control;
7107         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7108                 vcpu->guest_debug = 0;
7109
7110         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7111                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7112                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7113                 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7114         } else {
7115                 for (i = 0; i < KVM_NR_DB_REGS; i++)
7116                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7117         }
7118         kvm_update_dr7(vcpu);
7119
7120         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7121                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7122                         get_segment_base(vcpu, VCPU_SREG_CS);
7123
7124         /*
7125          * Trigger an rflags update that will inject or remove the trace
7126          * flags.
7127          */
7128         kvm_set_rflags(vcpu, rflags);
7129
7130         kvm_x86_ops->update_bp_intercept(vcpu);
7131
7132         r = 0;
7133
7134 out:
7135
7136         return r;
7137 }
7138
7139 /*
7140  * Translate a guest virtual address to a guest physical address.
7141  */
7142 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7143                                     struct kvm_translation *tr)
7144 {
7145         unsigned long vaddr = tr->linear_address;
7146         gpa_t gpa;
7147         int idx;
7148
7149         idx = srcu_read_lock(&vcpu->kvm->srcu);
7150         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7151         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7152         tr->physical_address = gpa;
7153         tr->valid = gpa != UNMAPPED_GVA;
7154         tr->writeable = 1;
7155         tr->usermode = 0;
7156
7157         return 0;
7158 }
7159
7160 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7161 {
7162         struct fxregs_state *fxsave =
7163                         &vcpu->arch.guest_fpu.state.fxsave;
7164
7165         memcpy(fpu->fpr, fxsave->st_space, 128);
7166         fpu->fcw = fxsave->cwd;
7167         fpu->fsw = fxsave->swd;
7168         fpu->ftwx = fxsave->twd;
7169         fpu->last_opcode = fxsave->fop;
7170         fpu->last_ip = fxsave->rip;
7171         fpu->last_dp = fxsave->rdp;
7172         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7173
7174         return 0;
7175 }
7176
7177 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7178 {
7179         struct fxregs_state *fxsave =
7180                         &vcpu->arch.guest_fpu.state.fxsave;
7181
7182         memcpy(fxsave->st_space, fpu->fpr, 128);
7183         fxsave->cwd = fpu->fcw;
7184         fxsave->swd = fpu->fsw;
7185         fxsave->twd = fpu->ftwx;
7186         fxsave->fop = fpu->last_opcode;
7187         fxsave->rip = fpu->last_ip;
7188         fxsave->rdp = fpu->last_dp;
7189         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7190
7191         return 0;
7192 }
7193
7194 static void fx_init(struct kvm_vcpu *vcpu)
7195 {
7196         fpstate_init(&vcpu->arch.guest_fpu.state);
7197         if (cpu_has_xsaves)
7198                 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7199                         host_xcr0 | XSTATE_COMPACTION_ENABLED;
7200
7201         /*
7202          * Ensure guest xcr0 is valid for loading
7203          */
7204         vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7205
7206         vcpu->arch.cr0 |= X86_CR0_ET;
7207 }
7208
7209 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7210 {
7211         if (vcpu->guest_fpu_loaded)
7212                 return;
7213
7214         /*
7215          * Restore all possible states in the guest,
7216          * and assume host would use all available bits.
7217          * Guest xcr0 would be loaded later.
7218          */
7219         vcpu->guest_fpu_loaded = 1;
7220         __kernel_fpu_begin();
7221         __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7222         trace_kvm_fpu(1);
7223 }
7224
7225 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7226 {
7227         if (!vcpu->guest_fpu_loaded) {
7228                 vcpu->fpu_counter = 0;
7229                 return;
7230         }
7231
7232         vcpu->guest_fpu_loaded = 0;
7233         copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7234         __kernel_fpu_end();
7235         ++vcpu->stat.fpu_reload;
7236         /*
7237          * If using eager FPU mode, or if the guest is a frequent user
7238          * of the FPU, just leave the FPU active for next time.
7239          * Every 255 times fpu_counter rolls over to 0; a guest that uses
7240          * the FPU in bursts will revert to loading it on demand.
7241          */
7242         if (!vcpu->arch.eager_fpu) {
7243                 if (++vcpu->fpu_counter < 5)
7244                         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7245         }
7246         trace_kvm_fpu(0);
7247 }
7248
7249 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7250 {
7251         kvmclock_reset(vcpu);
7252
7253         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7254         kvm_x86_ops->vcpu_free(vcpu);
7255 }
7256
7257 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7258                                                 unsigned int id)
7259 {
7260         struct kvm_vcpu *vcpu;
7261
7262         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7263                 printk_once(KERN_WARNING
7264                 "kvm: SMP vm created on host with unstable TSC; "
7265                 "guest TSC will not be reliable\n");
7266
7267         vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7268
7269         return vcpu;
7270 }
7271
7272 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7273 {
7274         int r;
7275
7276         kvm_vcpu_mtrr_init(vcpu);
7277         r = vcpu_load(vcpu);
7278         if (r)
7279                 return r;
7280         kvm_vcpu_reset(vcpu, false);
7281         kvm_mmu_setup(vcpu);
7282         vcpu_put(vcpu);
7283         return r;
7284 }
7285
7286 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7287 {
7288         struct msr_data msr;
7289         struct kvm *kvm = vcpu->kvm;
7290
7291         if (vcpu_load(vcpu))
7292                 return;
7293         msr.data = 0x0;
7294         msr.index = MSR_IA32_TSC;
7295         msr.host_initiated = true;
7296         kvm_write_tsc(vcpu, &msr);
7297         vcpu_put(vcpu);
7298
7299         if (!kvmclock_periodic_sync)
7300                 return;
7301
7302         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7303                                         KVMCLOCK_SYNC_PERIOD);
7304 }
7305
7306 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7307 {
7308         int r;
7309         vcpu->arch.apf.msr_val = 0;
7310
7311         r = vcpu_load(vcpu);
7312         BUG_ON(r);
7313         kvm_mmu_unload(vcpu);
7314         vcpu_put(vcpu);
7315
7316         kvm_x86_ops->vcpu_free(vcpu);
7317 }
7318
7319 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7320 {
7321         vcpu->arch.hflags = 0;
7322
7323         atomic_set(&vcpu->arch.nmi_queued, 0);
7324         vcpu->arch.nmi_pending = 0;
7325         vcpu->arch.nmi_injected = false;
7326         kvm_clear_interrupt_queue(vcpu);
7327         kvm_clear_exception_queue(vcpu);
7328
7329         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7330         kvm_update_dr0123(vcpu);
7331         vcpu->arch.dr6 = DR6_INIT;
7332         kvm_update_dr6(vcpu);
7333         vcpu->arch.dr7 = DR7_FIXED_1;
7334         kvm_update_dr7(vcpu);
7335
7336         vcpu->arch.cr2 = 0;
7337
7338         kvm_make_request(KVM_REQ_EVENT, vcpu);
7339         vcpu->arch.apf.msr_val = 0;
7340         vcpu->arch.st.msr_val = 0;
7341
7342         kvmclock_reset(vcpu);
7343
7344         kvm_clear_async_pf_completion_queue(vcpu);
7345         kvm_async_pf_hash_reset(vcpu);
7346         vcpu->arch.apf.halted = false;
7347
7348         if (!init_event) {
7349                 kvm_pmu_reset(vcpu);
7350                 vcpu->arch.smbase = 0x30000;
7351         }
7352
7353         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7354         vcpu->arch.regs_avail = ~0;
7355         vcpu->arch.regs_dirty = ~0;
7356
7357         kvm_x86_ops->vcpu_reset(vcpu, init_event);
7358 }
7359
7360 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7361 {
7362         struct kvm_segment cs;
7363
7364         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7365         cs.selector = vector << 8;
7366         cs.base = vector << 12;
7367         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7368         kvm_rip_write(vcpu, 0);
7369 }
7370
7371 int kvm_arch_hardware_enable(void)
7372 {
7373         struct kvm *kvm;
7374         struct kvm_vcpu *vcpu;
7375         int i;
7376         int ret;
7377         u64 local_tsc;
7378         u64 max_tsc = 0;
7379         bool stable, backwards_tsc = false;
7380
7381         kvm_shared_msr_cpu_online();
7382         ret = kvm_x86_ops->hardware_enable();
7383         if (ret != 0)
7384                 return ret;
7385
7386         local_tsc = rdtsc();
7387         stable = !check_tsc_unstable();
7388         list_for_each_entry(kvm, &vm_list, vm_list) {
7389                 kvm_for_each_vcpu(i, vcpu, kvm) {
7390                         if (!stable && vcpu->cpu == smp_processor_id())
7391                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7392                         if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7393                                 backwards_tsc = true;
7394                                 if (vcpu->arch.last_host_tsc > max_tsc)
7395                                         max_tsc = vcpu->arch.last_host_tsc;
7396                         }
7397                 }
7398         }
7399
7400         /*
7401          * Sometimes, even reliable TSCs go backwards.  This happens on
7402          * platforms that reset TSC during suspend or hibernate actions, but
7403          * maintain synchronization.  We must compensate.  Fortunately, we can
7404          * detect that condition here, which happens early in CPU bringup,
7405          * before any KVM threads can be running.  Unfortunately, we can't
7406          * bring the TSCs fully up to date with real time, as we aren't yet far
7407          * enough into CPU bringup that we know how much real time has actually
7408          * elapsed; our helper function, get_kernel_ns() will be using boot
7409          * variables that haven't been updated yet.
7410          *
7411          * So we simply find the maximum observed TSC above, then record the
7412          * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7413          * the adjustment will be applied.  Note that we accumulate
7414          * adjustments, in case multiple suspend cycles happen before some VCPU
7415          * gets a chance to run again.  In the event that no KVM threads get a
7416          * chance to run, we will miss the entire elapsed period, as we'll have
7417          * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7418          * loose cycle time.  This isn't too big a deal, since the loss will be
7419          * uniform across all VCPUs (not to mention the scenario is extremely
7420          * unlikely). It is possible that a second hibernate recovery happens
7421          * much faster than a first, causing the observed TSC here to be
7422          * smaller; this would require additional padding adjustment, which is
7423          * why we set last_host_tsc to the local tsc observed here.
7424          *
7425          * N.B. - this code below runs only on platforms with reliable TSC,
7426          * as that is the only way backwards_tsc is set above.  Also note
7427          * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7428          * have the same delta_cyc adjustment applied if backwards_tsc
7429          * is detected.  Note further, this adjustment is only done once,
7430          * as we reset last_host_tsc on all VCPUs to stop this from being
7431          * called multiple times (one for each physical CPU bringup).
7432          *
7433          * Platforms with unreliable TSCs don't have to deal with this, they
7434          * will be compensated by the logic in vcpu_load, which sets the TSC to
7435          * catchup mode.  This will catchup all VCPUs to real time, but cannot
7436          * guarantee that they stay in perfect synchronization.
7437          */
7438         if (backwards_tsc) {
7439                 u64 delta_cyc = max_tsc - local_tsc;
7440                 backwards_tsc_observed = true;
7441                 list_for_each_entry(kvm, &vm_list, vm_list) {
7442                         kvm_for_each_vcpu(i, vcpu, kvm) {
7443                                 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7444                                 vcpu->arch.last_host_tsc = local_tsc;
7445                                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7446                         }
7447
7448                         /*
7449                          * We have to disable TSC offset matching.. if you were
7450                          * booting a VM while issuing an S4 host suspend....
7451                          * you may have some problem.  Solving this issue is
7452                          * left as an exercise to the reader.
7453                          */
7454                         kvm->arch.last_tsc_nsec = 0;
7455                         kvm->arch.last_tsc_write = 0;
7456                 }
7457
7458         }
7459         return 0;
7460 }
7461
7462 void kvm_arch_hardware_disable(void)
7463 {
7464         kvm_x86_ops->hardware_disable();
7465         drop_user_return_notifiers();
7466 }
7467
7468 int kvm_arch_hardware_setup(void)
7469 {
7470         int r;
7471
7472         r = kvm_x86_ops->hardware_setup();
7473         if (r != 0)
7474                 return r;
7475
7476         if (kvm_has_tsc_control) {
7477                 /*
7478                  * Make sure the user can only configure tsc_khz values that
7479                  * fit into a signed integer.
7480                  * A min value is not calculated needed because it will always
7481                  * be 1 on all machines.
7482                  */
7483                 u64 max = min(0x7fffffffULL,
7484                               __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7485                 kvm_max_guest_tsc_khz = max;
7486
7487                 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7488         }
7489
7490         kvm_init_msr_list();
7491         return 0;
7492 }
7493
7494 void kvm_arch_hardware_unsetup(void)
7495 {
7496         kvm_x86_ops->hardware_unsetup();
7497 }
7498
7499 void kvm_arch_check_processor_compat(void *rtn)
7500 {
7501         kvm_x86_ops->check_processor_compatibility(rtn);
7502 }
7503
7504 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7505 {
7506         return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7507 }
7508 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7509
7510 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7511 {
7512         return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7513 }
7514
7515 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7516 {
7517         return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7518 }
7519
7520 struct static_key kvm_no_apic_vcpu __read_mostly;
7521
7522 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7523 {
7524         struct page *page;
7525         struct kvm *kvm;
7526         int r;
7527
7528         BUG_ON(vcpu->kvm == NULL);
7529         kvm = vcpu->kvm;
7530
7531         vcpu->arch.pv.pv_unhalted = false;
7532         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7533         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7534                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7535         else
7536                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7537
7538         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7539         if (!page) {
7540                 r = -ENOMEM;
7541                 goto fail;
7542         }
7543         vcpu->arch.pio_data = page_address(page);
7544
7545         kvm_set_tsc_khz(vcpu, max_tsc_khz);
7546
7547         r = kvm_mmu_create(vcpu);
7548         if (r < 0)
7549                 goto fail_free_pio_data;
7550
7551         if (irqchip_in_kernel(kvm)) {
7552                 r = kvm_create_lapic(vcpu);
7553                 if (r < 0)
7554                         goto fail_mmu_destroy;
7555         } else
7556                 static_key_slow_inc(&kvm_no_apic_vcpu);
7557
7558         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7559                                        GFP_KERNEL);
7560         if (!vcpu->arch.mce_banks) {
7561                 r = -ENOMEM;
7562                 goto fail_free_lapic;
7563         }
7564         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7565
7566         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7567                 r = -ENOMEM;
7568                 goto fail_free_mce_banks;
7569         }
7570
7571         fx_init(vcpu);
7572
7573         vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7574         vcpu->arch.pv_time_enabled = false;
7575
7576         vcpu->arch.guest_supported_xcr0 = 0;
7577         vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7578
7579         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7580
7581         vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7582
7583         kvm_async_pf_hash_reset(vcpu);
7584         kvm_pmu_init(vcpu);
7585
7586         vcpu->arch.pending_external_vector = -1;
7587
7588         return 0;
7589
7590 fail_free_mce_banks:
7591         kfree(vcpu->arch.mce_banks);
7592 fail_free_lapic:
7593         kvm_free_lapic(vcpu);
7594 fail_mmu_destroy:
7595         kvm_mmu_destroy(vcpu);
7596 fail_free_pio_data:
7597         free_page((unsigned long)vcpu->arch.pio_data);
7598 fail:
7599         return r;
7600 }
7601
7602 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7603 {
7604         int idx;
7605
7606         kvm_pmu_destroy(vcpu);
7607         kfree(vcpu->arch.mce_banks);
7608         kvm_free_lapic(vcpu);
7609         idx = srcu_read_lock(&vcpu->kvm->srcu);
7610         kvm_mmu_destroy(vcpu);
7611         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7612         free_page((unsigned long)vcpu->arch.pio_data);
7613         if (!lapic_in_kernel(vcpu))
7614                 static_key_slow_dec(&kvm_no_apic_vcpu);
7615 }
7616
7617 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7618 {
7619         kvm_x86_ops->sched_in(vcpu, cpu);
7620 }
7621
7622 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7623 {
7624         if (type)
7625                 return -EINVAL;
7626
7627         INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7628         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7629         INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7630         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7631         atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7632
7633         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7634         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7635         /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7636         set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7637                 &kvm->arch.irq_sources_bitmap);
7638
7639         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7640         mutex_init(&kvm->arch.apic_map_lock);
7641         spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7642
7643         pvclock_update_vm_gtod_copy(kvm);
7644
7645         INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7646         INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7647
7648         return 0;
7649 }
7650
7651 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7652 {
7653         int r;
7654         r = vcpu_load(vcpu);
7655         BUG_ON(r);
7656         kvm_mmu_unload(vcpu);
7657         vcpu_put(vcpu);
7658 }
7659
7660 static void kvm_free_vcpus(struct kvm *kvm)
7661 {
7662         unsigned int i;
7663         struct kvm_vcpu *vcpu;
7664
7665         /*
7666          * Unpin any mmu pages first.
7667          */
7668         kvm_for_each_vcpu(i, vcpu, kvm) {
7669                 kvm_clear_async_pf_completion_queue(vcpu);
7670                 kvm_unload_vcpu_mmu(vcpu);
7671         }
7672         kvm_for_each_vcpu(i, vcpu, kvm)
7673                 kvm_arch_vcpu_free(vcpu);
7674
7675         mutex_lock(&kvm->lock);
7676         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7677                 kvm->vcpus[i] = NULL;
7678
7679         atomic_set(&kvm->online_vcpus, 0);
7680         mutex_unlock(&kvm->lock);
7681 }
7682
7683 void kvm_arch_sync_events(struct kvm *kvm)
7684 {
7685         cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7686         cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7687         kvm_free_all_assigned_devices(kvm);
7688         kvm_free_pit(kvm);
7689 }
7690
7691 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7692 {
7693         int i, r;
7694         unsigned long hva;
7695         struct kvm_memslots *slots = kvm_memslots(kvm);
7696         struct kvm_memory_slot *slot, old;
7697
7698         /* Called with kvm->slots_lock held.  */
7699         if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7700                 return -EINVAL;
7701
7702         slot = id_to_memslot(slots, id);
7703         if (size) {
7704                 if (WARN_ON(slot->npages))
7705                         return -EEXIST;
7706
7707                 /*
7708                  * MAP_SHARED to prevent internal slot pages from being moved
7709                  * by fork()/COW.
7710                  */
7711                 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7712                               MAP_SHARED | MAP_ANONYMOUS, 0);
7713                 if (IS_ERR((void *)hva))
7714                         return PTR_ERR((void *)hva);
7715         } else {
7716                 if (!slot->npages)
7717                         return 0;
7718
7719                 hva = 0;
7720         }
7721
7722         old = *slot;
7723         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7724                 struct kvm_userspace_memory_region m;
7725
7726                 m.slot = id | (i << 16);
7727                 m.flags = 0;
7728                 m.guest_phys_addr = gpa;
7729                 m.userspace_addr = hva;
7730                 m.memory_size = size;
7731                 r = __kvm_set_memory_region(kvm, &m);
7732                 if (r < 0)
7733                         return r;
7734         }
7735
7736         if (!size) {
7737                 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7738                 WARN_ON(r < 0);
7739         }
7740
7741         return 0;
7742 }
7743 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7744
7745 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7746 {
7747         int r;
7748
7749         mutex_lock(&kvm->slots_lock);
7750         r = __x86_set_memory_region(kvm, id, gpa, size);
7751         mutex_unlock(&kvm->slots_lock);
7752
7753         return r;
7754 }
7755 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7756
7757 void kvm_arch_destroy_vm(struct kvm *kvm)
7758 {
7759         if (current->mm == kvm->mm) {
7760                 /*
7761                  * Free memory regions allocated on behalf of userspace,
7762                  * unless the the memory map has changed due to process exit
7763                  * or fd copying.
7764                  */
7765                 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7766                 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7767                 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7768         }
7769         kvm_iommu_unmap_guest(kvm);
7770         kfree(kvm->arch.vpic);
7771         kfree(kvm->arch.vioapic);
7772         kvm_free_vcpus(kvm);
7773         kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7774 }
7775
7776 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7777                            struct kvm_memory_slot *dont)
7778 {
7779         int i;
7780
7781         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7782                 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7783                         kvfree(free->arch.rmap[i]);
7784                         free->arch.rmap[i] = NULL;
7785                 }
7786                 if (i == 0)
7787                         continue;
7788
7789                 if (!dont || free->arch.lpage_info[i - 1] !=
7790                              dont->arch.lpage_info[i - 1]) {
7791                         kvfree(free->arch.lpage_info[i - 1]);
7792                         free->arch.lpage_info[i - 1] = NULL;
7793                 }
7794         }
7795 }
7796
7797 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7798                             unsigned long npages)
7799 {
7800         int i;
7801
7802         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7803                 unsigned long ugfn;
7804                 int lpages;
7805                 int level = i + 1;
7806
7807                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7808                                       slot->base_gfn, level) + 1;
7809
7810                 slot->arch.rmap[i] =
7811                         kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7812                 if (!slot->arch.rmap[i])
7813                         goto out_free;
7814                 if (i == 0)
7815                         continue;
7816
7817                 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7818                                         sizeof(*slot->arch.lpage_info[i - 1]));
7819                 if (!slot->arch.lpage_info[i - 1])
7820                         goto out_free;
7821
7822                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7823                         slot->arch.lpage_info[i - 1][0].write_count = 1;
7824                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7825                         slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7826                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7827                 /*
7828                  * If the gfn and userspace address are not aligned wrt each
7829                  * other, or if explicitly asked to, disable large page
7830                  * support for this slot
7831                  */
7832                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7833                     !kvm_largepages_enabled()) {
7834                         unsigned long j;
7835
7836                         for (j = 0; j < lpages; ++j)
7837                                 slot->arch.lpage_info[i - 1][j].write_count = 1;
7838                 }
7839         }
7840
7841         return 0;
7842
7843 out_free:
7844         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7845                 kvfree(slot->arch.rmap[i]);
7846                 slot->arch.rmap[i] = NULL;
7847                 if (i == 0)
7848                         continue;
7849
7850                 kvfree(slot->arch.lpage_info[i - 1]);
7851                 slot->arch.lpage_info[i - 1] = NULL;
7852         }
7853         return -ENOMEM;
7854 }
7855
7856 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7857 {
7858         /*
7859          * memslots->generation has been incremented.
7860          * mmio generation may have reached its maximum value.
7861          */
7862         kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7863 }
7864
7865 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7866                                 struct kvm_memory_slot *memslot,
7867                                 const struct kvm_userspace_memory_region *mem,
7868                                 enum kvm_mr_change change)
7869 {
7870         return 0;
7871 }
7872
7873 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7874                                      struct kvm_memory_slot *new)
7875 {
7876         /* Still write protect RO slot */
7877         if (new->flags & KVM_MEM_READONLY) {
7878                 kvm_mmu_slot_remove_write_access(kvm, new);
7879                 return;
7880         }
7881
7882         /*
7883          * Call kvm_x86_ops dirty logging hooks when they are valid.
7884          *
7885          * kvm_x86_ops->slot_disable_log_dirty is called when:
7886          *
7887          *  - KVM_MR_CREATE with dirty logging is disabled
7888          *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7889          *
7890          * The reason is, in case of PML, we need to set D-bit for any slots
7891          * with dirty logging disabled in order to eliminate unnecessary GPA
7892          * logging in PML buffer (and potential PML buffer full VMEXT). This
7893          * guarantees leaving PML enabled during guest's lifetime won't have
7894          * any additonal overhead from PML when guest is running with dirty
7895          * logging disabled for memory slots.
7896          *
7897          * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7898          * to dirty logging mode.
7899          *
7900          * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7901          *
7902          * In case of write protect:
7903          *
7904          * Write protect all pages for dirty logging.
7905          *
7906          * All the sptes including the large sptes which point to this
7907          * slot are set to readonly. We can not create any new large
7908          * spte on this slot until the end of the logging.
7909          *
7910          * See the comments in fast_page_fault().
7911          */
7912         if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7913                 if (kvm_x86_ops->slot_enable_log_dirty)
7914                         kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7915                 else
7916                         kvm_mmu_slot_remove_write_access(kvm, new);
7917         } else {
7918                 if (kvm_x86_ops->slot_disable_log_dirty)
7919                         kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7920         }
7921 }
7922
7923 void kvm_arch_commit_memory_region(struct kvm *kvm,
7924                                 const struct kvm_userspace_memory_region *mem,
7925                                 const struct kvm_memory_slot *old,
7926                                 const struct kvm_memory_slot *new,
7927                                 enum kvm_mr_change change)
7928 {
7929         int nr_mmu_pages = 0;
7930
7931         if (!kvm->arch.n_requested_mmu_pages)
7932                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7933
7934         if (nr_mmu_pages)
7935                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7936
7937         /*
7938          * Dirty logging tracks sptes in 4k granularity, meaning that large
7939          * sptes have to be split.  If live migration is successful, the guest
7940          * in the source machine will be destroyed and large sptes will be
7941          * created in the destination. However, if the guest continues to run
7942          * in the source machine (for example if live migration fails), small
7943          * sptes will remain around and cause bad performance.
7944          *
7945          * Scan sptes if dirty logging has been stopped, dropping those
7946          * which can be collapsed into a single large-page spte.  Later
7947          * page faults will create the large-page sptes.
7948          */
7949         if ((change != KVM_MR_DELETE) &&
7950                 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7951                 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7952                 kvm_mmu_zap_collapsible_sptes(kvm, new);
7953
7954         /*
7955          * Set up write protection and/or dirty logging for the new slot.
7956          *
7957          * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7958          * been zapped so no dirty logging staff is needed for old slot. For
7959          * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7960          * new and it's also covered when dealing with the new slot.
7961          *
7962          * FIXME: const-ify all uses of struct kvm_memory_slot.
7963          */
7964         if (change != KVM_MR_DELETE)
7965                 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
7966 }
7967
7968 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7969 {
7970         kvm_mmu_invalidate_zap_all_pages(kvm);
7971 }
7972
7973 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7974                                    struct kvm_memory_slot *slot)
7975 {
7976         kvm_mmu_invalidate_zap_all_pages(kvm);
7977 }
7978
7979 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
7980 {
7981         if (!list_empty_careful(&vcpu->async_pf.done))
7982                 return true;
7983
7984         if (kvm_apic_has_events(vcpu))
7985                 return true;
7986
7987         if (vcpu->arch.pv.pv_unhalted)
7988                 return true;
7989
7990         if (atomic_read(&vcpu->arch.nmi_queued))
7991                 return true;
7992
7993         if (test_bit(KVM_REQ_SMI, &vcpu->requests))
7994                 return true;
7995
7996         if (kvm_arch_interrupt_allowed(vcpu) &&
7997             kvm_cpu_has_interrupt(vcpu))
7998                 return true;
7999
8000         return false;
8001 }
8002
8003 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8004 {
8005         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8006                 kvm_x86_ops->check_nested_events(vcpu, false);
8007
8008         return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8009 }
8010
8011 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8012 {
8013         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8014 }
8015
8016 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8017 {
8018         return kvm_x86_ops->interrupt_allowed(vcpu);
8019 }
8020
8021 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8022 {
8023         if (is_64_bit_mode(vcpu))
8024                 return kvm_rip_read(vcpu);
8025         return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8026                      kvm_rip_read(vcpu));
8027 }
8028 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8029
8030 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8031 {
8032         return kvm_get_linear_rip(vcpu) == linear_rip;
8033 }
8034 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8035
8036 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8037 {
8038         unsigned long rflags;
8039
8040         rflags = kvm_x86_ops->get_rflags(vcpu);
8041         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8042                 rflags &= ~X86_EFLAGS_TF;
8043         return rflags;
8044 }
8045 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8046
8047 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8048 {
8049         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8050             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8051                 rflags |= X86_EFLAGS_TF;
8052         kvm_x86_ops->set_rflags(vcpu, rflags);
8053 }
8054
8055 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8056 {
8057         __kvm_set_rflags(vcpu, rflags);
8058         kvm_make_request(KVM_REQ_EVENT, vcpu);
8059 }
8060 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8061
8062 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8063 {
8064         int r;
8065
8066         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8067               work->wakeup_all)
8068                 return;
8069
8070         r = kvm_mmu_reload(vcpu);
8071         if (unlikely(r))
8072                 return;
8073
8074         if (!vcpu->arch.mmu.direct_map &&
8075               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8076                 return;
8077
8078         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8079 }
8080
8081 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8082 {
8083         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8084 }
8085
8086 static inline u32 kvm_async_pf_next_probe(u32 key)
8087 {
8088         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8089 }
8090
8091 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8092 {
8093         u32 key = kvm_async_pf_hash_fn(gfn);
8094
8095         while (vcpu->arch.apf.gfns[key] != ~0)
8096                 key = kvm_async_pf_next_probe(key);
8097
8098         vcpu->arch.apf.gfns[key] = gfn;
8099 }
8100
8101 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8102 {
8103         int i;
8104         u32 key = kvm_async_pf_hash_fn(gfn);
8105
8106         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8107                      (vcpu->arch.apf.gfns[key] != gfn &&
8108                       vcpu->arch.apf.gfns[key] != ~0); i++)
8109                 key = kvm_async_pf_next_probe(key);
8110
8111         return key;
8112 }
8113
8114 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8115 {
8116         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8117 }
8118
8119 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8120 {
8121         u32 i, j, k;
8122
8123         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8124         while (true) {
8125                 vcpu->arch.apf.gfns[i] = ~0;
8126                 do {
8127                         j = kvm_async_pf_next_probe(j);
8128                         if (vcpu->arch.apf.gfns[j] == ~0)
8129                                 return;
8130                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8131                         /*
8132                          * k lies cyclically in ]i,j]
8133                          * |    i.k.j |
8134                          * |....j i.k.| or  |.k..j i...|
8135                          */
8136                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8137                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8138                 i = j;
8139         }
8140 }
8141
8142 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8143 {
8144
8145         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8146                                       sizeof(val));
8147 }
8148
8149 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8150                                      struct kvm_async_pf *work)
8151 {
8152         struct x86_exception fault;
8153
8154         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8155         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8156
8157         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8158             (vcpu->arch.apf.send_user_only &&
8159              kvm_x86_ops->get_cpl(vcpu) == 0))
8160                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8161         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8162                 fault.vector = PF_VECTOR;
8163                 fault.error_code_valid = true;
8164                 fault.error_code = 0;
8165                 fault.nested_page_fault = false;
8166                 fault.address = work->arch.token;
8167                 kvm_inject_page_fault(vcpu, &fault);
8168         }
8169 }
8170
8171 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8172                                  struct kvm_async_pf *work)
8173 {
8174         struct x86_exception fault;
8175
8176         trace_kvm_async_pf_ready(work->arch.token, work->gva);
8177         if (work->wakeup_all)
8178                 work->arch.token = ~0; /* broadcast wakeup */
8179         else
8180                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8181
8182         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8183             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8184                 fault.vector = PF_VECTOR;
8185                 fault.error_code_valid = true;
8186                 fault.error_code = 0;
8187                 fault.nested_page_fault = false;
8188                 fault.address = work->arch.token;
8189                 kvm_inject_page_fault(vcpu, &fault);
8190         }
8191         vcpu->arch.apf.halted = false;
8192         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8193 }
8194
8195 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8196 {
8197         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8198                 return true;
8199         else
8200                 return !kvm_event_needs_reinjection(vcpu) &&
8201                         kvm_x86_ops->interrupt_allowed(vcpu);
8202 }
8203
8204 void kvm_arch_start_assignment(struct kvm *kvm)
8205 {
8206         atomic_inc(&kvm->arch.assigned_device_count);
8207 }
8208 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8209
8210 void kvm_arch_end_assignment(struct kvm *kvm)
8211 {
8212         atomic_dec(&kvm->arch.assigned_device_count);
8213 }
8214 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8215
8216 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8217 {
8218         return atomic_read(&kvm->arch.assigned_device_count);
8219 }
8220 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8221
8222 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8223 {
8224         atomic_inc(&kvm->arch.noncoherent_dma_count);
8225 }
8226 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8227
8228 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8229 {
8230         atomic_dec(&kvm->arch.noncoherent_dma_count);
8231 }
8232 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8233
8234 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8235 {
8236         return atomic_read(&kvm->arch.noncoherent_dma_count);
8237 }
8238 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8239
8240 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8241                                       struct irq_bypass_producer *prod)
8242 {
8243         struct kvm_kernel_irqfd *irqfd =
8244                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8245
8246         if (kvm_x86_ops->update_pi_irte) {
8247                 irqfd->producer = prod;
8248                 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8249                                 prod->irq, irqfd->gsi, 1);
8250         }
8251
8252         return -EINVAL;
8253 }
8254
8255 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8256                                       struct irq_bypass_producer *prod)
8257 {
8258         int ret;
8259         struct kvm_kernel_irqfd *irqfd =
8260                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8261
8262         if (!kvm_x86_ops->update_pi_irte) {
8263                 WARN_ON(irqfd->producer != NULL);
8264                 return;
8265         }
8266
8267         WARN_ON(irqfd->producer != prod);
8268         irqfd->producer = NULL;
8269
8270         /*
8271          * When producer of consumer is unregistered, we change back to
8272          * remapped mode, so we can re-use the current implementation
8273          * when the irq is masked/disabed or the consumer side (KVM
8274          * int this case doesn't want to receive the interrupts.
8275         */
8276         ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8277         if (ret)
8278                 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8279                        " fails: %d\n", irqfd->consumer.token, ret);
8280 }
8281
8282 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8283                                    uint32_t guest_irq, bool set)
8284 {
8285         if (!kvm_x86_ops->update_pi_irte)
8286                 return -EINVAL;
8287
8288         return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8289 }
8290
8291 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8292 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8293 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8294 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8295 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8296 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8297 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8298 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8299 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8300 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8301 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8302 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8303 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8304 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8305 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8306 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8307 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);