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         kvm_put_guest_xcr0(vcpu);
 701         vcpu->arch.xcr0 = xcr0;
 702
 703         if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
 704                 kvm_update_cpuid(vcpu);
 705         return 0;
 706 }
 707
 708 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 709 {
 710         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
 711             __kvm_set_xcr(vcpu, index, xcr)) {
 712                 kvm_inject_gp(vcpu, 0);
 713                 return 1;
 714         }
 715         return 0;
 716 }
 717 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 718
 719 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 720 {
 721         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 722         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
 723                                    X86_CR4_SMEP | X86_CR4_SMAP;
 724
 725         if (cr4 & CR4_RESERVED_BITS)
 726                 return 1;
 727
 728         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 729                 return 1;
 730
 731         if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
 732                 return 1;
 733
 734         if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
 735                 return 1;
 736
 737         if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
 738                 return 1;
 739
 740         if (is_long_mode(vcpu)) {
 741                 if (!(cr4 & X86_CR4_PAE))
 742                         return 1;
 743         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 744                    && ((cr4 ^ old_cr4) & pdptr_bits)
 745                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 746                                    kvm_read_cr3(vcpu)))
 747                 return 1;
 748
 749         if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
 750                 if (!guest_cpuid_has_pcid(vcpu))
 751                         return 1;
 752
 753                 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
 754                 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
 755                         return 1;
 756         }
 757
 758         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 759                 return 1;
 760
 761         if (((cr4 ^ old_cr4) & pdptr_bits) ||
 762             (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
 763                 kvm_mmu_reset_context(vcpu);
 764
 765         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 766                 kvm_update_cpuid(vcpu);
 767
 768         return 0;
 769 }
 770 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 771
 772 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 773 {
 774 #ifdef CONFIG_X86_64
 775         cr3 &= ~CR3_PCID_INVD;
 776 #endif
 777
 778         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 779                 kvm_mmu_sync_roots(vcpu);
 780                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 781                 return 0;
 782         }
 783
 784         if (is_long_mode(vcpu)) {
 785                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 786                         return 1;
 787         } else if (is_pae(vcpu) && is_paging(vcpu) &&
 788                    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 789                 return 1;
 790
 791         vcpu->arch.cr3 = cr3;
 792         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 793         kvm_mmu_new_cr3(vcpu);
 794         return 0;
 795 }
 796 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 797
 798 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 799 {
 800         if (cr8 & CR8_RESERVED_BITS)
 801                 return 1;
 802         if (lapic_in_kernel(vcpu))
 803                 kvm_lapic_set_tpr(vcpu, cr8);
 804         else
 805                 vcpu->arch.cr8 = cr8;
 806         return 0;
 807 }
 808 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 809
 810 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 811 {
 812         if (lapic_in_kernel(vcpu))
 813                 return kvm_lapic_get_cr8(vcpu);
 814         else
 815                 return vcpu->arch.cr8;
 816 }
 817 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 818
 819 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
 820 {
 821         int i;
 822
 823         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 824                 for (i = 0; i < KVM_NR_DB_REGS; i++)
 825                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
 826                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
 827         }
 828 }
 829
 830 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
 831 {
 832         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 833                 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
 834 }
 835
 836 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
 837 {
 838         unsigned long dr7;
 839
 840         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 841                 dr7 = vcpu->arch.guest_debug_dr7;
 842         else
 843                 dr7 = vcpu->arch.dr7;
 844         kvm_x86_ops->set_dr7(vcpu, dr7);
 845         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
 846         if (dr7 & DR7_BP_EN_MASK)
 847                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
 848 }
 849
 850 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
 851 {
 852         u64 fixed = DR6_FIXED_1;
 853
 854         if (!guest_cpuid_has_rtm(vcpu))
 855                 fixed |= DR6_RTM;
 856         return fixed;
 857 }
 858
 859 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 860 {
 861         switch (dr) {
 862         case 0 ... 3:
 863                 vcpu->arch.db[dr] = val;
 864                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 865                         vcpu->arch.eff_db[dr] = val;
 866                 break;
 867         case 4:
 868                 /* fall through */
 869         case 6:
 870                 if (val & 0xffffffff00000000ULL)
 871                         return -1; /* #GP */
 872                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
 873                 kvm_update_dr6(vcpu);
 874                 break;
 875         case 5:
 876                 /* fall through */
 877         default: /* 7 */
 878                 if (val & 0xffffffff00000000ULL)
 879                         return -1; /* #GP */
 880                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 881                 kvm_update_dr7(vcpu);
 882                 break;
 883         }
 884
 885         return 0;
 886 }
 887
 888 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 889 {
 890         if (__kvm_set_dr(vcpu, dr, val)) {
 891                 kvm_inject_gp(vcpu, 0);
 892                 return 1;
 893         }
 894         return 0;
 895 }
 896 EXPORT_SYMBOL_GPL(kvm_set_dr);
 897
 898 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 899 {
 900         switch (dr) {
 901         case 0 ... 3:
 902                 *val = vcpu->arch.db[dr];
 903                 break;
 904         case 4:
 905                 /* fall through */
 906         case 6:
 907                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 908                         *val = vcpu->arch.dr6;
 909                 else
 910                         *val = kvm_x86_ops->get_dr6(vcpu);
 911                 break;
 912         case 5:
 913                 /* fall through */
 914         default: /* 7 */
 915                 *val = vcpu->arch.dr7;
 916                 break;
 917         }
 918         return 0;
 919 }
 920 EXPORT_SYMBOL_GPL(kvm_get_dr);
 921
 922 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
 923 {
 924         u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 925         u64 data;
 926         int err;
 927
 928         err = kvm_pmu_rdpmc(vcpu, ecx, &data);
 929         if (err)
 930                 return err;
 931         kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
 932         kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
 933         return err;
 934 }
 935 EXPORT_SYMBOL_GPL(kvm_rdpmc);
 936
 937 /*
 938  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 939  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 940  *
 941  * This list is modified at module load time to reflect the
 942  * capabilities of the host cpu. This capabilities test skips MSRs that are
 943  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
 944  * may depend on host virtualization features rather than host cpu features.
 945  */
 946
 947 static u32 msrs_to_save[] = {
 948         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 949         MSR_STAR,
 950 #ifdef CONFIG_X86_64
 951         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 952 #endif
 953         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
 954         MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
 955 };
 956
 957 static unsigned num_msrs_to_save;
 958
 959 static u32 emulated_msrs[] = {
 960         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 961         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 962         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 963         HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
 964         HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
 965         HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
 966         HV_X64_MSR_RESET,
 967         HV_X64_MSR_VP_INDEX,
 968         HV_X64_MSR_VP_RUNTIME,
 969         HV_X64_MSR_SCONTROL,
 970         HV_X64_MSR_STIMER0_CONFIG,
 971         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
 972         MSR_KVM_PV_EOI_EN,
 973
 974         MSR_IA32_TSC_ADJUST,
 975         MSR_IA32_TSCDEADLINE,
 976         MSR_IA32_MISC_ENABLE,
 977         MSR_IA32_MCG_STATUS,
 978         MSR_IA32_MCG_CTL,
 979         MSR_IA32_SMBASE,
 980 };
 981
 982 static unsigned num_emulated_msrs;
 983
 984 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
 985 {
 986         if (efer & efer_reserved_bits)
 987                 return false;
 988
 989         if (efer & EFER_FFXSR) {
 990                 struct kvm_cpuid_entry2 *feat;
 991
 992                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 993                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 994                         return false;
 995         }
 996
 997         if (efer & EFER_SVME) {
 998                 struct kvm_cpuid_entry2 *feat;
 999
1000                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1001                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1002                         return false;
1003         }
1004
1005         return true;
1006 }
1007 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1008
1009 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1010 {
1011         u64 old_efer = vcpu->arch.efer;
1012
1013         if (!kvm_valid_efer(vcpu, efer))
1014                 return 1;
1015
1016         if (is_paging(vcpu)
1017             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1018                 return 1;
1019
1020         efer &= ~EFER_LMA;
1021         efer |= vcpu->arch.efer & EFER_LMA;
1022
1023         kvm_x86_ops->set_efer(vcpu, efer);
1024
1025         /* Update reserved bits */
1026         if ((efer ^ old_efer) & EFER_NX)
1027                 kvm_mmu_reset_context(vcpu);
1028
1029         return 0;
1030 }
1031
1032 void kvm_enable_efer_bits(u64 mask)
1033 {
1034        efer_reserved_bits &= ~mask;
1035 }
1036 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1037
1038 /*
1039  * Writes msr value into into the appropriate "register".
1040  * Returns 0 on success, non-0 otherwise.
1041  * Assumes vcpu_load() was already called.
1042  */
1043 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1044 {
1045         switch (msr->index) {
1046         case MSR_FS_BASE:
1047         case MSR_GS_BASE:
1048         case MSR_KERNEL_GS_BASE:
1049         case MSR_CSTAR:
1050         case MSR_LSTAR:
1051                 if (is_noncanonical_address(msr->data))
1052                         return 1;
1053                 break;
1054         case MSR_IA32_SYSENTER_EIP:
1055         case MSR_IA32_SYSENTER_ESP:
1056                 /*
1057                  * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1058                  * non-canonical address is written on Intel but not on
1059                  * AMD (which ignores the top 32-bits, because it does
1060                  * not implement 64-bit SYSENTER).
1061                  *
1062                  * 64-bit code should hence be able to write a non-canonical
1063                  * value on AMD.  Making the address canonical ensures that
1064                  * vmentry does not fail on Intel after writing a non-canonical
1065                  * value, and that something deterministic happens if the guest
1066                  * invokes 64-bit SYSENTER.
1067                  */
1068                 msr->data = get_canonical(msr->data);
1069         }
1070         return kvm_x86_ops->set_msr(vcpu, msr);
1071 }
1072 EXPORT_SYMBOL_GPL(kvm_set_msr);
1073
1074 /*
1075  * Adapt set_msr() to msr_io()'s calling convention
1076  */
1077 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1078 {
1079         struct msr_data msr;
1080         int r;
1081
1082         msr.index = index;
1083         msr.host_initiated = true;
1084         r = kvm_get_msr(vcpu, &msr);
1085         if (r)
1086                 return r;
1087
1088         *data = msr.data;
1089         return 0;
1090 }
1091
1092 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1093 {
1094         struct msr_data msr;
1095
1096         msr.data = *data;
1097         msr.index = index;
1098         msr.host_initiated = true;
1099         return kvm_set_msr(vcpu, &msr);
1100 }
1101
1102 #ifdef CONFIG_X86_64
1103 struct pvclock_gtod_data {
1104         seqcount_t      seq;
1105
1106         struct { /* extract of a clocksource struct */
1107                 int vclock_mode;
1108                 cycle_t cycle_last;
1109                 cycle_t mask;
1110                 u32     mult;
1111                 u32     shift;
1112         } clock;
1113
1114         u64             boot_ns;
1115         u64             nsec_base;
1116 };
1117
1118 static struct pvclock_gtod_data pvclock_gtod_data;
1119
1120 static void update_pvclock_gtod(struct timekeeper *tk)
1121 {
1122         struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1123         u64 boot_ns;
1124
1125         boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1126
1127         write_seqcount_begin(&vdata->seq);
1128
1129         /* copy pvclock gtod data */
1130         vdata->clock.vclock_mode        = tk->tkr_mono.clock->archdata.vclock_mode;
1131         vdata->clock.cycle_last         = tk->tkr_mono.cycle_last;
1132         vdata->clock.mask               = tk->tkr_mono.mask;
1133         vdata->clock.mult               = tk->tkr_mono.mult;
1134         vdata->clock.shift              = tk->tkr_mono.shift;
1135
1136         vdata->boot_ns                  = boot_ns;
1137         vdata->nsec_base                = tk->tkr_mono.xtime_nsec;
1138
1139         write_seqcount_end(&vdata->seq);
1140 }
1141 #endif
1142
1143 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1144 {
1145         /*
1146          * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1147          * vcpu_enter_guest.  This function is only called from
1148          * the physical CPU that is running vcpu.
1149          */
1150         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1151 }
1152
1153 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1154 {
1155         int version;
1156         int r;
1157         struct pvclock_wall_clock wc;
1158         struct timespec boot;
1159
1160         if (!wall_clock)
1161                 return;
1162
1163         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1164         if (r)
1165                 return;
1166
1167         if (version & 1)
1168                 ++version;  /* first time write, random junk */
1169
1170         ++version;
1171
1172         if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1173                 return;
1174
1175         /*
1176          * The guest calculates current wall clock time by adding
1177          * system time (updated by kvm_guest_time_update below) to the
1178          * wall clock specified here.  guest system time equals host
1179          * system time for us, thus we must fill in host boot time here.
1180          */
1181         getboottime(&boot);
1182
1183         if (kvm->arch.kvmclock_offset) {
1184                 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1185                 boot = timespec_sub(boot, ts);
1186         }
1187         wc.sec = boot.tv_sec;
1188         wc.nsec = boot.tv_nsec;
1189         wc.version = version;
1190
1191         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1192
1193         version++;
1194         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1195 }
1196
1197 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1198 {
1199         uint32_t quotient, remainder;
1200
1201         /* Don't try to replace with do_div(), this one calculates
1202          * "(dividend << 32) / divisor" */
1203         __asm__ ( "divl %4"
1204                   : "=a" (quotient), "=d" (remainder)
1205                   : "0" (0), "1" (dividend), "r" (divisor) );
1206         return quotient;
1207 }
1208
1209 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1210                                s8 *pshift, u32 *pmultiplier)
1211 {
1212         uint64_t scaled64;
1213         int32_t  shift = 0;
1214         uint64_t tps64;
1215         uint32_t tps32;
1216
1217         tps64 = base_khz * 1000LL;
1218         scaled64 = scaled_khz * 1000LL;
1219         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1220                 tps64 >>= 1;
1221                 shift--;
1222         }
1223
1224         tps32 = (uint32_t)tps64;
1225         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1226                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1227                         scaled64 >>= 1;
1228                 else
1229                         tps32 <<= 1;
1230                 shift++;
1231         }
1232
1233         *pshift = shift;
1234         *pmultiplier = div_frac(scaled64, tps32);
1235
1236         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1237                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
1238 }
1239
1240 #ifdef CONFIG_X86_64
1241 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1242 #endif
1243
1244 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1245 static unsigned long max_tsc_khz;
1246
1247 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1248 {
1249         return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1250                                    vcpu->arch.virtual_tsc_shift);
1251 }
1252
1253 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1254 {
1255         u64 v = (u64)khz * (1000000 + ppm);
1256         do_div(v, 1000000);
1257         return v;
1258 }
1259
1260 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1261 {
1262         u64 ratio;
1263
1264         /* Guest TSC same frequency as host TSC? */
1265         if (!scale) {
1266                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1267                 return 0;
1268         }
1269
1270         /* TSC scaling supported? */
1271         if (!kvm_has_tsc_control) {
1272                 if (user_tsc_khz > tsc_khz) {
1273                         vcpu->arch.tsc_catchup = 1;
1274                         vcpu->arch.tsc_always_catchup = 1;
1275                         return 0;
1276                 } else {
1277                         WARN(1, "user requested TSC rate below hardware speed\n");
1278                         return -1;
1279                 }
1280         }
1281
1282         /* TSC scaling required  - calculate ratio */
1283         ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1284                                 user_tsc_khz, tsc_khz);
1285
1286         if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1287                 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1288                           user_tsc_khz);
1289                 return -1;
1290         }
1291
1292         vcpu->arch.tsc_scaling_ratio = ratio;
1293         return 0;
1294 }
1295
1296 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1297 {
1298         u32 thresh_lo, thresh_hi;
1299         int use_scaling = 0;
1300
1301         /* tsc_khz can be zero if TSC calibration fails */
1302         if (this_tsc_khz == 0) {
1303                 /* set tsc_scaling_ratio to a safe value */
1304                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1305                 return -1;
1306         }
1307
1308         /* Compute a scale to convert nanoseconds in TSC cycles */
1309         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1310                            &vcpu->arch.virtual_tsc_shift,
1311                            &vcpu->arch.virtual_tsc_mult);
1312         vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1313
1314         /*
1315          * Compute the variation in TSC rate which is acceptable
1316          * within the range of tolerance and decide if the
1317          * rate being applied is within that bounds of the hardware
1318          * rate.  If so, no scaling or compensation need be done.
1319          */
1320         thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1321         thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1322         if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1323                 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1324                 use_scaling = 1;
1325         }
1326         return set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1327 }
1328
1329 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1330 {
1331         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1332                                       vcpu->arch.virtual_tsc_mult,
1333                                       vcpu->arch.virtual_tsc_shift);
1334         tsc += vcpu->arch.this_tsc_write;
1335         return tsc;
1336 }
1337
1338 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1339 {
1340 #ifdef CONFIG_X86_64
1341         bool vcpus_matched;
1342         struct kvm_arch *ka = &vcpu->kvm->arch;
1343         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1344
1345         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1346                          atomic_read(&vcpu->kvm->online_vcpus));
1347
1348         /*
1349          * Once the masterclock is enabled, always perform request in
1350          * order to update it.
1351          *
1352          * In order to enable masterclock, the host clocksource must be TSC
1353          * and the vcpus need to have matched TSCs.  When that happens,
1354          * perform request to enable masterclock.
1355          */
1356         if (ka->use_master_clock ||
1357             (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1358                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1359
1360         trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1361                             atomic_read(&vcpu->kvm->online_vcpus),
1362                             ka->use_master_clock, gtod->clock.vclock_mode);
1363 #endif
1364 }
1365
1366 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1367 {
1368         u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1369         vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1370 }
1371
1372 /*
1373  * Multiply tsc by a fixed point number represented by ratio.
1374  *
1375  * The most significant 64-N bits (mult) of ratio represent the
1376  * integral part of the fixed point number; the remaining N bits
1377  * (frac) represent the fractional part, ie. ratio represents a fixed
1378  * point number (mult + frac * 2^(-N)).
1379  *
1380  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1381  */
1382 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1383 {
1384         return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1385 }
1386
1387 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1388 {
1389         u64 _tsc = tsc;
1390         u64 ratio = vcpu->arch.tsc_scaling_ratio;
1391
1392         if (ratio != kvm_default_tsc_scaling_ratio)
1393                 _tsc = __scale_tsc(ratio, tsc);
1394
1395         return _tsc;
1396 }
1397 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1398
1399 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1400 {
1401         u64 tsc;
1402
1403         tsc = kvm_scale_tsc(vcpu, rdtsc());
1404
1405         return target_tsc - tsc;
1406 }
1407
1408 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1409 {
1410         return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1411 }
1412 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1413
1414 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1415 {
1416         struct kvm *kvm = vcpu->kvm;
1417         u64 offset, ns, elapsed;
1418         unsigned long flags;
1419         s64 usdiff;
1420         bool matched;
1421         bool already_matched;
1422         u64 data = msr->data;
1423
1424         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1425         offset = kvm_compute_tsc_offset(vcpu, data);
1426         ns = get_kernel_ns();
1427         elapsed = ns - kvm->arch.last_tsc_nsec;
1428
1429         if (vcpu->arch.virtual_tsc_khz) {
1430                 int faulted = 0;
1431
1432                 /* n.b - signed multiplication and division required */
1433                 usdiff = data - kvm->arch.last_tsc_write;
1434 #ifdef CONFIG_X86_64
1435                 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1436 #else
1437                 /* do_div() only does unsigned */
1438                 asm("1: idivl %[divisor]\n"
1439                     "2: xor %%edx, %%edx\n"
1440                     "   movl $0, %[faulted]\n"
1441                     "3:\n"
1442                     ".section .fixup,\"ax\"\n"
1443                     "4: movl $1, %[faulted]\n"
1444                     "   jmp  3b\n"
1445                     ".previous\n"
1446
1447                 _ASM_EXTABLE(1b, 4b)
1448
1449                 : "=A"(usdiff), [faulted] "=r" (faulted)
1450                 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1451
1452 #endif
1453                 do_div(elapsed, 1000);
1454                 usdiff -= elapsed;
1455                 if (usdiff < 0)
1456                         usdiff = -usdiff;
1457
1458                 /* idivl overflow => difference is larger than USEC_PER_SEC */
1459                 if (faulted)
1460                         usdiff = USEC_PER_SEC;
1461         } else
1462                 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1463
1464         /*
1465          * Special case: TSC write with a small delta (1 second) of virtual
1466          * cycle time against real time is interpreted as an attempt to
1467          * synchronize the CPU.
1468          *
1469          * For a reliable TSC, we can match TSC offsets, and for an unstable
1470          * TSC, we add elapsed time in this computation.  We could let the
1471          * compensation code attempt to catch up if we fall behind, but
1472          * it's better to try to match offsets from the beginning.
1473          */
1474         if (usdiff < USEC_PER_SEC &&
1475             vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1476                 if (!check_tsc_unstable()) {
1477                         offset = kvm->arch.cur_tsc_offset;
1478                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1479                 } else {
1480                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1481                         data += delta;
1482                         offset = kvm_compute_tsc_offset(vcpu, data);
1483                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1484                 }
1485                 matched = true;
1486                 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1487         } else {
1488                 /*
1489                  * We split periods of matched TSC writes into generations.
1490                  * For each generation, we track the original measured
1491                  * nanosecond time, offset, and write, so if TSCs are in
1492                  * sync, we can match exact offset, and if not, we can match
1493                  * exact software computation in compute_guest_tsc()
1494                  *
1495                  * These values are tracked in kvm->arch.cur_xxx variables.
1496                  */
1497                 kvm->arch.cur_tsc_generation++;
1498                 kvm->arch.cur_tsc_nsec = ns;
1499                 kvm->arch.cur_tsc_write = data;
1500                 kvm->arch.cur_tsc_offset = offset;
1501                 matched = false;
1502                 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1503                          kvm->arch.cur_tsc_generation, data);
1504         }
1505
1506         /*
1507          * We also track th most recent recorded KHZ, write and time to
1508          * allow the matching interval to be extended at each write.
1509          */
1510         kvm->arch.last_tsc_nsec = ns;
1511         kvm->arch.last_tsc_write = data;
1512         kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1513
1514         vcpu->arch.last_guest_tsc = data;
1515
1516         /* Keep track of which generation this VCPU has synchronized to */
1517         vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1518         vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1519         vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1520
1521         if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1522                 update_ia32_tsc_adjust_msr(vcpu, offset);
1523         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1524         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1525
1526         spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1527         if (!matched) {
1528                 kvm->arch.nr_vcpus_matched_tsc = 0;
1529         } else if (!already_matched) {
1530                 kvm->arch.nr_vcpus_matched_tsc++;
1531         }
1532
1533         kvm_track_tsc_matching(vcpu);
1534         spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1535 }
1536
1537 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1538
1539 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1540                                            s64 adjustment)
1541 {
1542         kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1543 }
1544
1545 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1546 {
1547         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1548                 WARN_ON(adjustment < 0);
1549         adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1550         kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1551 }
1552
1553 #ifdef CONFIG_X86_64
1554
1555 static cycle_t read_tsc(void)
1556 {
1557         cycle_t ret = (cycle_t)rdtsc_ordered();
1558         u64 last = pvclock_gtod_data.clock.cycle_last;
1559
1560         if (likely(ret >= last))
1561                 return ret;
1562
1563         /*
1564          * GCC likes to generate cmov here, but this branch is extremely
1565          * predictable (it's just a funciton of time and the likely is
1566          * very likely) and there's a data dependence, so force GCC
1567          * to generate a branch instead.  I don't barrier() because
1568          * we don't actually need a barrier, and if this function
1569          * ever gets inlined it will generate worse code.
1570          */
1571         asm volatile ("");
1572         return last;
1573 }
1574
1575 static inline u64 vgettsc(cycle_t *cycle_now)
1576 {
1577         long v;
1578         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1579
1580         *cycle_now = read_tsc();
1581
1582         v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1583         return v * gtod->clock.mult;
1584 }
1585
1586 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1587 {
1588         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1589         unsigned long seq;
1590         int mode;
1591         u64 ns;
1592
1593         do {
1594                 seq = read_seqcount_begin(&gtod->seq);
1595                 mode = gtod->clock.vclock_mode;
1596                 ns = gtod->nsec_base;
1597                 ns += vgettsc(cycle_now);
1598                 ns >>= gtod->clock.shift;
1599                 ns += gtod->boot_ns;
1600         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1601         *t = ns;
1602
1603         return mode;
1604 }
1605
1606 /* returns true if host is using tsc clocksource */
1607 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1608 {
1609         /* checked again under seqlock below */
1610         if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1611                 return false;
1612
1613         return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1614 }
1615 #endif
1616
1617 /*
1618  *
1619  * Assuming a stable TSC across physical CPUS, and a stable TSC
1620  * across virtual CPUs, the following condition is possible.
1621  * Each numbered line represents an event visible to both
1622  * CPUs at the next numbered event.
1623  *
1624  * "timespecX" represents host monotonic time. "tscX" represents
1625  * RDTSC value.
1626  *
1627  *              VCPU0 on CPU0           |       VCPU1 on CPU1
1628  *
1629  * 1.  read timespec0,tsc0
1630  * 2.                                   | timespec1 = timespec0 + N
1631  *                                      | tsc1 = tsc0 + M
1632  * 3. transition to guest               | transition to guest
1633  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1634  * 5.                                   | ret1 = timespec1 + (rdtsc - tsc1)
1635  *                                      | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1636  *
1637  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1638  *
1639  *      - ret0 < ret1
1640  *      - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1641  *              ...
1642  *      - 0 < N - M => M < N
1643  *
1644  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1645  * always the case (the difference between two distinct xtime instances
1646  * might be smaller then the difference between corresponding TSC reads,
1647  * when updating guest vcpus pvclock areas).
1648  *
1649  * To avoid that problem, do not allow visibility of distinct
1650  * system_timestamp/tsc_timestamp values simultaneously: use a master
1651  * copy of host monotonic time values. Update that master copy
1652  * in lockstep.
1653  *
1654  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1655  *
1656  */
1657
1658 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1659 {
1660 #ifdef CONFIG_X86_64
1661         struct kvm_arch *ka = &kvm->arch;
1662         int vclock_mode;
1663         bool host_tsc_clocksource, vcpus_matched;
1664
1665         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1666                         atomic_read(&kvm->online_vcpus));
1667
1668         /*
1669          * If the host uses TSC clock, then passthrough TSC as stable
1670          * to the guest.
1671          */
1672         host_tsc_clocksource = kvm_get_time_and_clockread(
1673                                         &ka->master_kernel_ns,
1674                                         &ka->master_cycle_now);
1675
1676         ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1677                                 && !backwards_tsc_observed
1678                                 && !ka->boot_vcpu_runs_old_kvmclock;
1679
1680         if (ka->use_master_clock)
1681                 atomic_set(&kvm_guest_has_master_clock, 1);
1682
1683         vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1684         trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1685                                         vcpus_matched);
1686 #endif
1687 }
1688
1689 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1690 {
1691         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1692 }
1693
1694 static void kvm_gen_update_masterclock(struct kvm *kvm)
1695 {
1696 #ifdef CONFIG_X86_64
1697         int i;
1698         struct kvm_vcpu *vcpu;
1699         struct kvm_arch *ka = &kvm->arch;
1700
1701         spin_lock(&ka->pvclock_gtod_sync_lock);
1702         kvm_make_mclock_inprogress_request(kvm);
1703         /* no guest entries from this point */
1704         pvclock_update_vm_gtod_copy(kvm);
1705
1706         kvm_for_each_vcpu(i, vcpu, kvm)
1707                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1708
1709         /* guest entries allowed */
1710         kvm_for_each_vcpu(i, vcpu, kvm)
1711                 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1712
1713         spin_unlock(&ka->pvclock_gtod_sync_lock);
1714 #endif
1715 }
1716
1717 static int kvm_guest_time_update(struct kvm_vcpu *v)
1718 {
1719         unsigned long flags, this_tsc_khz, tgt_tsc_khz;
1720         struct kvm_vcpu_arch *vcpu = &v->arch;
1721         struct kvm_arch *ka = &v->kvm->arch;
1722         s64 kernel_ns;
1723         u64 tsc_timestamp, host_tsc;
1724         struct pvclock_vcpu_time_info guest_hv_clock;
1725         u8 pvclock_flags;
1726         bool use_master_clock;
1727
1728         kernel_ns = 0;
1729         host_tsc = 0;
1730
1731         /*
1732          * If the host uses TSC clock, then passthrough TSC as stable
1733          * to the guest.
1734          */
1735         spin_lock(&ka->pvclock_gtod_sync_lock);
1736         use_master_clock = ka->use_master_clock;
1737         if (use_master_clock) {
1738                 host_tsc = ka->master_cycle_now;
1739                 kernel_ns = ka->master_kernel_ns;
1740         }
1741         spin_unlock(&ka->pvclock_gtod_sync_lock);
1742
1743         /* Keep irq disabled to prevent changes to the clock */
1744         local_irq_save(flags);
1745         this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1746         if (unlikely(this_tsc_khz == 0)) {
1747                 local_irq_restore(flags);
1748                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1749                 return 1;
1750         }
1751         if (!use_master_clock) {
1752                 host_tsc = rdtsc();
1753                 kernel_ns = get_kernel_ns();
1754         }
1755
1756         tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1757
1758         /*
1759          * We may have to catch up the TSC to match elapsed wall clock
1760          * time for two reasons, even if kvmclock is used.
1761          *   1) CPU could have been running below the maximum TSC rate
1762          *   2) Broken TSC compensation resets the base at each VCPU
1763          *      entry to avoid unknown leaps of TSC even when running
1764          *      again on the same CPU.  This may cause apparent elapsed
1765          *      time to disappear, and the guest to stand still or run
1766          *      very slowly.
1767          */
1768         if (vcpu->tsc_catchup) {
1769                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1770                 if (tsc > tsc_timestamp) {
1771                         adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1772                         tsc_timestamp = tsc;
1773                 }
1774         }
1775
1776         local_irq_restore(flags);
1777
1778         if (!vcpu->pv_time_enabled)
1779                 return 0;
1780
1781         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1782                 tgt_tsc_khz = kvm_has_tsc_control ?
1783                         vcpu->virtual_tsc_khz : this_tsc_khz;
1784                 kvm_get_time_scale(NSEC_PER_SEC / 1000, tgt_tsc_khz,
1785                                    &vcpu->hv_clock.tsc_shift,
1786                                    &vcpu->hv_clock.tsc_to_system_mul);
1787                 vcpu->hw_tsc_khz = this_tsc_khz;
1788         }
1789
1790         /* With all the info we got, fill in the values */
1791         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1792         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1793         vcpu->last_guest_tsc = tsc_timestamp;
1794
1795         if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1796                 &guest_hv_clock, sizeof(guest_hv_clock))))
1797                 return 0;
1798
1799         /* This VCPU is paused, but it's legal for a guest to read another
1800          * VCPU's kvmclock, so we really have to follow the specification where
1801          * it says that version is odd if data is being modified, and even after
1802          * it is consistent.
1803          *
1804          * Version field updates must be kept separate.  This is because
1805          * kvm_write_guest_cached might use a "rep movs" instruction, and
1806          * writes within a string instruction are weakly ordered.  So there
1807          * are three writes overall.
1808          *
1809          * As a small optimization, only write the version field in the first
1810          * and third write.  The vcpu->pv_time cache is still valid, because the
1811          * version field is the first in the struct.
1812          */
1813         BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1814
1815         vcpu->hv_clock.version = guest_hv_clock.version + 1;
1816         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1817                                 &vcpu->hv_clock,
1818                                 sizeof(vcpu->hv_clock.version));
1819
1820         smp_wmb();
1821
1822         /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1823         pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1824
1825         if (vcpu->pvclock_set_guest_stopped_request) {
1826                 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1827                 vcpu->pvclock_set_guest_stopped_request = false;
1828         }
1829
1830         /* If the host uses TSC clocksource, then it is stable */
1831         if (use_master_clock)
1832                 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1833
1834         vcpu->hv_clock.flags = pvclock_flags;
1835
1836         trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1837
1838         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1839                                 &vcpu->hv_clock,
1840                                 sizeof(vcpu->hv_clock));
1841
1842         smp_wmb();
1843
1844         vcpu->hv_clock.version++;
1845         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1846                                 &vcpu->hv_clock,
1847                                 sizeof(vcpu->hv_clock.version));
1848         return 0;
1849 }
1850
1851 /*
1852  * kvmclock updates which are isolated to a given vcpu, such as
1853  * vcpu->cpu migration, should not allow system_timestamp from
1854  * the rest of the vcpus to remain static. Otherwise ntp frequency
1855  * correction applies to one vcpu's system_timestamp but not
1856  * the others.
1857  *
1858  * So in those cases, request a kvmclock update for all vcpus.
1859  * We need to rate-limit these requests though, as they can
1860  * considerably slow guests that have a large number of vcpus.
1861  * The time for a remote vcpu to update its kvmclock is bound
1862  * by the delay we use to rate-limit the updates.
1863  */
1864
1865 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1866
1867 static void kvmclock_update_fn(struct work_struct *work)
1868 {
1869         int i;
1870         struct delayed_work *dwork = to_delayed_work(work);
1871         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1872                                            kvmclock_update_work);
1873         struct kvm *kvm = container_of(ka, struct kvm, arch);
1874         struct kvm_vcpu *vcpu;
1875
1876         kvm_for_each_vcpu(i, vcpu, kvm) {
1877                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1878                 kvm_vcpu_kick(vcpu);
1879         }
1880 }
1881
1882 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1883 {
1884         struct kvm *kvm = v->kvm;
1885
1886         kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1887         schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1888                                         KVMCLOCK_UPDATE_DELAY);
1889 }
1890
1891 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1892
1893 static void kvmclock_sync_fn(struct work_struct *work)
1894 {
1895         struct delayed_work *dwork = to_delayed_work(work);
1896         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1897                                            kvmclock_sync_work);
1898         struct kvm *kvm = container_of(ka, struct kvm, arch);
1899
1900         if (!kvmclock_periodic_sync)
1901                 return;
1902
1903         schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1904         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1905                                         KVMCLOCK_SYNC_PERIOD);
1906 }
1907
1908 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1909 {
1910         u64 mcg_cap = vcpu->arch.mcg_cap;
1911         unsigned bank_num = mcg_cap & 0xff;
1912
1913         switch (msr) {
1914         case MSR_IA32_MCG_STATUS:
1915                 vcpu->arch.mcg_status = data;
1916                 break;
1917         case MSR_IA32_MCG_CTL:
1918                 if (!(mcg_cap & MCG_CTL_P))
1919                         return 1;
1920                 if (data != 0 && data != ~(u64)0)
1921                         return -1;
1922                 vcpu->arch.mcg_ctl = data;
1923                 break;
1924         default:
1925                 if (msr >= MSR_IA32_MC0_CTL &&
1926                     msr < MSR_IA32_MCx_CTL(bank_num)) {
1927                         u32 offset = msr - MSR_IA32_MC0_CTL;
1928                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1929                          * some Linux kernels though clear bit 10 in bank 4 to
1930                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1931                          * this to avoid an uncatched #GP in the guest
1932                          */
1933                         if ((offset & 0x3) == 0 &&
1934                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1935                                 return -1;
1936                         vcpu->arch.mce_banks[offset] = data;
1937                         break;
1938                 }
1939                 return 1;
1940         }
1941         return 0;
1942 }
1943
1944 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1945 {
1946         struct kvm *kvm = vcpu->kvm;
1947         int lm = is_long_mode(vcpu);
1948         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1949                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1950         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1951                 : kvm->arch.xen_hvm_config.blob_size_32;
1952         u32 page_num = data & ~PAGE_MASK;
1953         u64 page_addr = data & PAGE_MASK;
1954         u8 *page;
1955         int r;
1956
1957         r = -E2BIG;
1958         if (page_num >= blob_size)
1959                 goto out;
1960         r = -ENOMEM;
1961         page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1962         if (IS_ERR(page)) {
1963                 r = PTR_ERR(page);
1964                 goto out;
1965         }
1966         if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1967                 goto out_free;
1968         r = 0;
1969 out_free:
1970         kfree(page);
1971 out:
1972         return r;
1973 }
1974
1975 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1976 {
1977         gpa_t gpa = data & ~0x3f;
1978
1979         /* Bits 2:5 are reserved, Should be zero */
1980         if (data & 0x3c)
1981                 return 1;
1982
1983         vcpu->arch.apf.msr_val = data;
1984
1985         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1986                 kvm_clear_async_pf_completion_queue(vcpu);
1987                 kvm_async_pf_hash_reset(vcpu);
1988                 return 0;
1989         }
1990
1991         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1992                                         sizeof(u32)))
1993                 return 1;
1994
1995         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1996         kvm_async_pf_wakeup_all(vcpu);
1997         return 0;
1998 }
1999
2000 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2001 {
2002         vcpu->arch.pv_time_enabled = false;
2003 }
2004
2005 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2006 {
2007         u64 delta;
2008
2009         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2010                 return;
2011
2012         delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2013         vcpu->arch.st.last_steal = current->sched_info.run_delay;
2014         vcpu->arch.st.accum_steal = delta;
2015 }
2016
2017 static void record_steal_time(struct kvm_vcpu *vcpu)
2018 {
2019         accumulate_steal_time(vcpu);
2020
2021         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2022                 return;
2023
2024         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2025                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2026                 return;
2027
2028         vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2029         vcpu->arch.st.steal.version += 2;
2030         vcpu->arch.st.accum_steal = 0;
2031
2032         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2033                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2034 }
2035
2036 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2037 {
2038         bool pr = false;
2039         u32 msr = msr_info->index;
2040         u64 data = msr_info->data;
2041
2042         switch (msr) {
2043         case MSR_AMD64_NB_CFG:
2044         case MSR_IA32_UCODE_REV:
2045         case MSR_IA32_UCODE_WRITE:
2046         case MSR_VM_HSAVE_PA:
2047         case MSR_AMD64_PATCH_LOADER:
2048         case MSR_AMD64_BU_CFG2:
2049                 break;
2050
2051         case MSR_EFER:
2052                 return set_efer(vcpu, data);
2053         case MSR_K7_HWCR:
2054                 data &= ~(u64)0x40;     /* ignore flush filter disable */
2055                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
2056                 data &= ~(u64)0x8;      /* ignore TLB cache disable */
2057                 data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2058                 if (data != 0) {
2059                         vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2060                                     data);
2061                         return 1;
2062                 }
2063                 break;
2064         case MSR_FAM10H_MMIO_CONF_BASE:
2065                 if (data != 0) {
2066                         vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2067                                     "0x%llx\n", data);
2068                         return 1;
2069                 }
2070                 break;
2071         case MSR_IA32_DEBUGCTLMSR:
2072                 if (!data) {
2073                         /* We support the non-activated case already */
2074                         break;
2075                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2076                         /* Values other than LBR and BTF are vendor-specific,
2077                            thus reserved and should throw a #GP */
2078                         return 1;
2079                 }
2080                 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2081                             __func__, data);
2082                 break;
2083         case 0x200 ... 0x2ff:
2084                 return kvm_mtrr_set_msr(vcpu, msr, data);
2085         case MSR_IA32_APICBASE:
2086                 return kvm_set_apic_base(vcpu, msr_info);
2087         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2088                 return kvm_x2apic_msr_write(vcpu, msr, data);
2089         case MSR_IA32_TSCDEADLINE:
2090                 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2091                 break;
2092         case MSR_IA32_TSC_ADJUST:
2093                 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2094                         if (!msr_info->host_initiated) {
2095                                 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2096                                 adjust_tsc_offset_guest(vcpu, adj);
2097                         }
2098                         vcpu->arch.ia32_tsc_adjust_msr = data;
2099                 }
2100                 break;
2101         case MSR_IA32_MISC_ENABLE:
2102                 vcpu->arch.ia32_misc_enable_msr = data;
2103                 break;
2104         case MSR_IA32_SMBASE:
2105                 if (!msr_info->host_initiated)
2106                         return 1;
2107                 vcpu->arch.smbase = data;
2108                 break;
2109         case MSR_KVM_WALL_CLOCK_NEW:
2110         case MSR_KVM_WALL_CLOCK:
2111                 vcpu->kvm->arch.wall_clock = data;
2112                 kvm_write_wall_clock(vcpu->kvm, data);
2113                 break;
2114         case MSR_KVM_SYSTEM_TIME_NEW:
2115         case MSR_KVM_SYSTEM_TIME: {
2116                 u64 gpa_offset;
2117                 struct kvm_arch *ka = &vcpu->kvm->arch;
2118
2119                 kvmclock_reset(vcpu);
2120
2121                 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2122                         bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2123
2124                         if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2125                                 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2126                                         &vcpu->requests);
2127
2128                         ka->boot_vcpu_runs_old_kvmclock = tmp;
2129                 }
2130
2131                 vcpu->arch.time = data;
2132                 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2133
2134                 /* we verify if the enable bit is set... */
2135                 if (!(data & 1))
2136                         break;
2137
2138                 gpa_offset = data & ~(PAGE_MASK | 1);
2139
2140                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2141                      &vcpu->arch.pv_time, data & ~1ULL,
2142                      sizeof(struct pvclock_vcpu_time_info)))
2143                         vcpu->arch.pv_time_enabled = false;
2144                 else
2145                         vcpu->arch.pv_time_enabled = true;
2146
2147                 break;
2148         }
2149         case MSR_KVM_ASYNC_PF_EN:
2150                 if (kvm_pv_enable_async_pf(vcpu, data))
2151                         return 1;
2152                 break;
2153         case MSR_KVM_STEAL_TIME:
2154
2155                 if (unlikely(!sched_info_on()))
2156                         return 1;
2157
2158                 if (data & KVM_STEAL_RESERVED_MASK)
2159                         return 1;
2160
2161                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2162                                                 data & KVM_STEAL_VALID_BITS,
2163                                                 sizeof(struct kvm_steal_time)))
2164                         return 1;
2165
2166                 vcpu->arch.st.msr_val = data;
2167
2168                 if (!(data & KVM_MSR_ENABLED))
2169                         break;
2170
2171                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2172
2173                 break;
2174         case MSR_KVM_PV_EOI_EN:
2175                 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2176                         return 1;
2177                 break;
2178
2179         case MSR_IA32_MCG_CTL:
2180         case MSR_IA32_MCG_STATUS:
2181         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2182                 return set_msr_mce(vcpu, msr, data);
2183
2184         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2185         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2186                 pr = true; /* fall through */
2187         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2188         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2189                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2190                         return kvm_pmu_set_msr(vcpu, msr_info);
2191
2192                 if (pr || data != 0)
2193                         vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2194                                     "0x%x data 0x%llx\n", msr, data);
2195                 break;
2196         case MSR_K7_CLK_CTL:
2197                 /*
2198                  * Ignore all writes to this no longer documented MSR.
2199                  * Writes are only relevant for old K7 processors,
2200                  * all pre-dating SVM, but a recommended workaround from
2201                  * AMD for these chips. It is possible to specify the
2202                  * affected processor models on the command line, hence
2203                  * the need to ignore the workaround.
2204                  */
2205                 break;
2206         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2207         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2208         case HV_X64_MSR_CRASH_CTL:
2209         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2210                 return kvm_hv_set_msr_common(vcpu, msr, data,
2211                                              msr_info->host_initiated);
2212         case MSR_IA32_BBL_CR_CTL3:
2213                 /* Drop writes to this legacy MSR -- see rdmsr
2214                  * counterpart for further detail.
2215                  */
2216                 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2217                 break;
2218         case MSR_AMD64_OSVW_ID_LENGTH:
2219                 if (!guest_cpuid_has_osvw(vcpu))
2220                         return 1;
2221                 vcpu->arch.osvw.length = data;
2222                 break;
2223         case MSR_AMD64_OSVW_STATUS:
2224                 if (!guest_cpuid_has_osvw(vcpu))
2225                         return 1;
2226                 vcpu->arch.osvw.status = data;
2227                 break;
2228         default:
2229                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2230                         return xen_hvm_config(vcpu, data);
2231                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2232                         return kvm_pmu_set_msr(vcpu, msr_info);
2233                 if (!ignore_msrs) {
2234                         vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2235                                     msr, data);
2236                         return 1;
2237                 } else {
2238                         vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2239                                     msr, data);
2240                         break;
2241                 }
2242         }
2243         return 0;
2244 }
2245 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2246
2247
2248 /*
2249  * Reads an msr value (of 'msr_index') into 'pdata'.
2250  * Returns 0 on success, non-0 otherwise.
2251  * Assumes vcpu_load() was already called.
2252  */
2253 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2254 {
2255         return kvm_x86_ops->get_msr(vcpu, msr);
2256 }
2257 EXPORT_SYMBOL_GPL(kvm_get_msr);
2258
2259 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2260 {
2261         u64 data;
2262         u64 mcg_cap = vcpu->arch.mcg_cap;
2263         unsigned bank_num = mcg_cap & 0xff;
2264
2265         switch (msr) {
2266         case MSR_IA32_P5_MC_ADDR:
2267         case MSR_IA32_P5_MC_TYPE:
2268                 data = 0;
2269                 break;
2270         case MSR_IA32_MCG_CAP:
2271                 data = vcpu->arch.mcg_cap;
2272                 break;
2273         case MSR_IA32_MCG_CTL:
2274                 if (!(mcg_cap & MCG_CTL_P))
2275                         return 1;
2276                 data = vcpu->arch.mcg_ctl;
2277                 break;
2278         case MSR_IA32_MCG_STATUS:
2279                 data = vcpu->arch.mcg_status;
2280                 break;
2281         default:
2282                 if (msr >= MSR_IA32_MC0_CTL &&
2283                     msr < MSR_IA32_MCx_CTL(bank_num)) {
2284                         u32 offset = msr - MSR_IA32_MC0_CTL;
2285                         data = vcpu->arch.mce_banks[offset];
2286                         break;
2287                 }
2288                 return 1;
2289         }
2290         *pdata = data;
2291         return 0;
2292 }
2293
2294 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2295 {
2296         switch (msr_info->index) {
2297         case MSR_IA32_PLATFORM_ID:
2298         case MSR_IA32_EBL_CR_POWERON:
2299         case MSR_IA32_DEBUGCTLMSR:
2300         case MSR_IA32_LASTBRANCHFROMIP:
2301         case MSR_IA32_LASTBRANCHTOIP:
2302         case MSR_IA32_LASTINTFROMIP:
2303         case MSR_IA32_LASTINTTOIP:
2304         case MSR_K8_SYSCFG:
2305         case MSR_K8_TSEG_ADDR:
2306         case MSR_K8_TSEG_MASK:
2307         case MSR_K7_HWCR:
2308         case MSR_VM_HSAVE_PA:
2309         case MSR_K8_INT_PENDING_MSG:
2310         case MSR_AMD64_NB_CFG:
2311         case MSR_FAM10H_MMIO_CONF_BASE:
2312         case MSR_AMD64_BU_CFG2:
2313                 msr_info->data = 0;
2314                 break;
2315         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2316         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2317         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2318         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2319                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2320                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2321                 msr_info->data = 0;
2322                 break;
2323         case MSR_IA32_UCODE_REV:
2324                 msr_info->data = 0x100000000ULL;
2325                 break;
2326         case MSR_MTRRcap:
2327         case 0x200 ... 0x2ff:
2328                 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2329         case 0xcd: /* fsb frequency */
2330                 msr_info->data = 3;
2331                 break;
2332                 /*
2333                  * MSR_EBC_FREQUENCY_ID
2334                  * Conservative value valid for even the basic CPU models.
2335                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2336                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2337                  * and 266MHz for model 3, or 4. Set Core Clock
2338                  * Frequency to System Bus Frequency Ratio to 1 (bits
2339                  * 31:24) even though these are only valid for CPU
2340                  * models > 2, however guests may end up dividing or
2341                  * multiplying by zero otherwise.
2342                  */
2343         case MSR_EBC_FREQUENCY_ID:
2344                 msr_info->data = 1 << 24;
2345                 break;
2346         case MSR_IA32_APICBASE:
2347                 msr_info->data = kvm_get_apic_base(vcpu);
2348                 break;
2349         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2350                 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2351                 break;
2352         case MSR_IA32_TSCDEADLINE:
2353                 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2354                 break;
2355         case MSR_IA32_TSC_ADJUST:
2356                 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2357                 break;
2358         case MSR_IA32_MISC_ENABLE:
2359                 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2360                 break;
2361         case MSR_IA32_SMBASE:
2362                 if (!msr_info->host_initiated)
2363                         return 1;
2364                 msr_info->data = vcpu->arch.smbase;
2365                 break;
2366         case MSR_IA32_PERF_STATUS:
2367                 /* TSC increment by tick */
2368                 msr_info->data = 1000ULL;
2369                 /* CPU multiplier */
2370                 msr_info->data |= (((uint64_t)4ULL) << 40);
2371                 break;
2372         case MSR_EFER:
2373                 msr_info->data = vcpu->arch.efer;
2374                 break;
2375         case MSR_KVM_WALL_CLOCK:
2376         case MSR_KVM_WALL_CLOCK_NEW:
2377                 msr_info->data = vcpu->kvm->arch.wall_clock;
2378                 break;
2379         case MSR_KVM_SYSTEM_TIME:
2380         case MSR_KVM_SYSTEM_TIME_NEW:
2381                 msr_info->data = vcpu->arch.time;
2382                 break;
2383         case MSR_KVM_ASYNC_PF_EN:
2384                 msr_info->data = vcpu->arch.apf.msr_val;
2385                 break;
2386         case MSR_KVM_STEAL_TIME:
2387                 msr_info->data = vcpu->arch.st.msr_val;
2388                 break;
2389         case MSR_KVM_PV_EOI_EN:
2390                 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2391                 break;
2392         case MSR_IA32_P5_MC_ADDR:
2393         case MSR_IA32_P5_MC_TYPE:
2394         case MSR_IA32_MCG_CAP:
2395         case MSR_IA32_MCG_CTL:
2396         case MSR_IA32_MCG_STATUS:
2397         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2398                 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2399         case MSR_K7_CLK_CTL:
2400                 /*
2401                  * Provide expected ramp-up count for K7. All other
2402                  * are set to zero, indicating minimum divisors for
2403                  * every field.
2404                  *
2405                  * This prevents guest kernels on AMD host with CPU
2406                  * type 6, model 8 and higher from exploding due to
2407                  * the rdmsr failing.
2408                  */
2409                 msr_info->data = 0x20000000;
2410                 break;
2411         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2412         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2413         case HV_X64_MSR_CRASH_CTL:
2414         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2415                 return kvm_hv_get_msr_common(vcpu,
2416                                              msr_info->index, &msr_info->data);
2417                 break;
2418         case MSR_IA32_BBL_CR_CTL3:
2419                 /* This legacy MSR exists but isn't fully documented in current
2420                  * silicon.  It is however accessed by winxp in very narrow
2421                  * scenarios where it sets bit #19, itself documented as
2422                  * a "reserved" bit.  Best effort attempt to source coherent
2423                  * read data here should the balance of the register be
2424                  * interpreted by the guest:
2425                  *
2426                  * L2 cache control register 3: 64GB range, 256KB size,
2427                  * enabled, latency 0x1, configured
2428                  */
2429                 msr_info->data = 0xbe702111;
2430                 break;
2431         case MSR_AMD64_OSVW_ID_LENGTH:
2432                 if (!guest_cpuid_has_osvw(vcpu))
2433                         return 1;
2434                 msr_info->data = vcpu->arch.osvw.length;
2435                 break;
2436         case MSR_AMD64_OSVW_STATUS:
2437                 if (!guest_cpuid_has_osvw(vcpu))
2438                         return 1;
2439                 msr_info->data = vcpu->arch.osvw.status;
2440                 break;
2441         default:
2442                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2443                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2444                 if (!ignore_msrs) {
2445                         vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2446                         return 1;
2447                 } else {
2448                         vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2449                         msr_info->data = 0;
2450                 }
2451                 break;
2452         }
2453         return 0;
2454 }
2455 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2456
2457 /*
2458  * Read or write a bunch of msrs. All parameters are kernel addresses.
2459  *
2460  * @return number of msrs set successfully.
2461  */
2462 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2463                     struct kvm_msr_entry *entries,
2464                     int (*do_msr)(struct kvm_vcpu *vcpu,
2465                                   unsigned index, u64 *data))
2466 {
2467         int i, idx;
2468
2469         idx = srcu_read_lock(&vcpu->kvm->srcu);
2470         for (i = 0; i < msrs->nmsrs; ++i)
2471                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2472                         break;
2473         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2474
2475         return i;
2476 }
2477
2478 /*
2479  * Read or write a bunch of msrs. Parameters are user addresses.
2480  *
2481  * @return number of msrs set successfully.
2482  */
2483 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2484                   int (*do_msr)(struct kvm_vcpu *vcpu,
2485                                 unsigned index, u64 *data),
2486                   int writeback)
2487 {
2488         struct kvm_msrs msrs;
2489         struct kvm_msr_entry *entries;
2490         int r, n;
2491         unsigned size;
2492
2493         r = -EFAULT;
2494         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2495                 goto out;
2496
2497         r = -E2BIG;
2498         if (msrs.nmsrs >= MAX_IO_MSRS)
2499                 goto out;
2500
2501         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2502         entries = memdup_user(user_msrs->entries, size);
2503         if (IS_ERR(entries)) {
2504                 r = PTR_ERR(entries);
2505                 goto out;
2506         }
2507
2508         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2509         if (r < 0)
2510                 goto out_free;
2511
2512         r = -EFAULT;
2513         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2514                 goto out_free;
2515
2516         r = n;
2517
2518 out_free:
2519         kfree(entries);
2520 out:
2521         return r;
2522 }
2523
2524 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2525 {
2526         int r;
2527
2528         switch (ext) {
2529         case KVM_CAP_IRQCHIP:
2530         case KVM_CAP_HLT:
2531         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2532         case KVM_CAP_SET_TSS_ADDR:
2533         case KVM_CAP_EXT_CPUID:
2534         case KVM_CAP_EXT_EMUL_CPUID:
2535         case KVM_CAP_CLOCKSOURCE:
2536         case KVM_CAP_PIT:
2537         case KVM_CAP_NOP_IO_DELAY:
2538         case KVM_CAP_MP_STATE:
2539         case KVM_CAP_SYNC_MMU:
2540         case KVM_CAP_USER_NMI:
2541         case KVM_CAP_REINJECT_CONTROL:
2542         case KVM_CAP_IRQ_INJECT_STATUS:
2543         case KVM_CAP_IOEVENTFD:
2544         case KVM_CAP_IOEVENTFD_NO_LENGTH:
2545         case KVM_CAP_PIT2:
2546         case KVM_CAP_PIT_STATE2:
2547         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2548         case KVM_CAP_XEN_HVM:
2549         case KVM_CAP_ADJUST_CLOCK:
2550         case KVM_CAP_VCPU_EVENTS:
2551         case KVM_CAP_HYPERV:
2552         case KVM_CAP_HYPERV_VAPIC:
2553         case KVM_CAP_HYPERV_SPIN:
2554         case KVM_CAP_HYPERV_SYNIC:
2555         case KVM_CAP_PCI_SEGMENT:
2556         case KVM_CAP_DEBUGREGS:
2557         case KVM_CAP_X86_ROBUST_SINGLESTEP:
2558         case KVM_CAP_XSAVE:
2559         case KVM_CAP_ASYNC_PF:
2560         case KVM_CAP_GET_TSC_KHZ:
2561         case KVM_CAP_KVMCLOCK_CTRL:
2562         case KVM_CAP_READONLY_MEM:
2563         case KVM_CAP_HYPERV_TIME:
2564         case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2565         case KVM_CAP_TSC_DEADLINE_TIMER:
2566         case KVM_CAP_ENABLE_CAP_VM:
2567         case KVM_CAP_DISABLE_QUIRKS:
2568         case KVM_CAP_SET_BOOT_CPU_ID:
2569         case KVM_CAP_SPLIT_IRQCHIP:
2570 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2571         case KVM_CAP_ASSIGN_DEV_IRQ:
2572         case KVM_CAP_PCI_2_3:
2573 #endif
2574                 r = 1;
2575                 break;
2576         case KVM_CAP_X86_SMM:
2577                 /* SMBASE is usually relocated above 1M on modern chipsets,
2578                  * and SMM handlers might indeed rely on 4G segment limits,
2579                  * so do not report SMM to be available if real mode is
2580                  * emulated via vm86 mode.  Still, do not go to great lengths
2581                  * to avoid userspace's usage of the feature, because it is a
2582                  * fringe case that is not enabled except via specific settings
2583                  * of the module parameters.
2584                  */
2585                 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2586                 break;
2587         case KVM_CAP_COALESCED_MMIO:
2588                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2589                 break;
2590         case KVM_CAP_VAPIC:
2591                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2592                 break;
2593         case KVM_CAP_NR_VCPUS:
2594                 r = KVM_SOFT_MAX_VCPUS;
2595                 break;
2596         case KVM_CAP_MAX_VCPUS:
2597                 r = KVM_MAX_VCPUS;
2598                 break;
2599         case KVM_CAP_NR_MEMSLOTS:
2600                 r = KVM_USER_MEM_SLOTS;
2601                 break;
2602         case KVM_CAP_PV_MMU:    /* obsolete */
2603                 r = 0;
2604                 break;
2605 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2606         case KVM_CAP_IOMMU:
2607                 r = iommu_present(&pci_bus_type);
2608                 break;
2609 #endif
2610         case KVM_CAP_MCE:
2611                 r = KVM_MAX_MCE_BANKS;
2612                 break;
2613         case KVM_CAP_XCRS:
2614                 r = cpu_has_xsave;
2615                 break;
2616         case KVM_CAP_TSC_CONTROL:
2617                 r = kvm_has_tsc_control;
2618                 break;
2619         default:
2620                 r = 0;
2621                 break;
2622         }
2623         return r;
2624
2625 }
2626
2627 long kvm_arch_dev_ioctl(struct file *filp,
2628                         unsigned int ioctl, unsigned long arg)
2629 {
2630         void __user *argp = (void __user *)arg;
2631         long r;
2632
2633         switch (ioctl) {
2634         case KVM_GET_MSR_INDEX_LIST: {
2635                 struct kvm_msr_list __user *user_msr_list = argp;
2636                 struct kvm_msr_list msr_list;
2637                 unsigned n;
2638
2639                 r = -EFAULT;
2640                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2641                         goto out;
2642                 n = msr_list.nmsrs;
2643                 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2644                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2645                         goto out;
2646                 r = -E2BIG;
2647                 if (n < msr_list.nmsrs)
2648                         goto out;
2649                 r = -EFAULT;
2650                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2651                                  num_msrs_to_save * sizeof(u32)))
2652                         goto out;
2653                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2654                                  &emulated_msrs,
2655                                  num_emulated_msrs * sizeof(u32)))
2656                         goto out;
2657                 r = 0;
2658                 break;
2659         }
2660         case KVM_GET_SUPPORTED_CPUID:
2661         case KVM_GET_EMULATED_CPUID: {
2662                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2663                 struct kvm_cpuid2 cpuid;
2664
2665                 r = -EFAULT;
2666                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2667                         goto out;
2668
2669                 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2670                                             ioctl);
2671                 if (r)
2672                         goto out;
2673
2674                 r = -EFAULT;
2675                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2676                         goto out;
2677                 r = 0;
2678                 break;
2679         }
2680         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2681                 u64 mce_cap;
2682
2683                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2684                 r = -EFAULT;
2685                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2686                         goto out;
2687                 r = 0;
2688                 break;
2689         }
2690         default:
2691                 r = -EINVAL;
2692         }
2693 out:
2694         return r;
2695 }
2696
2697 static void wbinvd_ipi(void *garbage)
2698 {
2699         wbinvd();
2700 }
2701
2702 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2703 {
2704         return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2705 }
2706
2707 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2708 {
2709         set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2710 }
2711
2712 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2713 {
2714         /* Address WBINVD may be executed by guest */
2715         if (need_emulate_wbinvd(vcpu)) {
2716                 if (kvm_x86_ops->has_wbinvd_exit())
2717                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2718                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2719                         smp_call_function_single(vcpu->cpu,
2720                                         wbinvd_ipi, NULL, 1);
2721         }
2722
2723         kvm_x86_ops->vcpu_load(vcpu, cpu);
2724
2725         /* Apply any externally detected TSC adjustments (due to suspend) */
2726         if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2727                 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2728                 vcpu->arch.tsc_offset_adjustment = 0;
2729                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2730         }
2731
2732         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2733                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2734                                 rdtsc() - vcpu->arch.last_host_tsc;
2735                 if (tsc_delta < 0)
2736                         mark_tsc_unstable("KVM discovered backwards TSC");
2737                 if (check_tsc_unstable()) {
2738                         u64 offset = kvm_compute_tsc_offset(vcpu,
2739                                                 vcpu->arch.last_guest_tsc);
2740                         kvm_x86_ops->write_tsc_offset(vcpu, offset);
2741                         vcpu->arch.tsc_catchup = 1;
2742                 }
2743                 /*
2744                  * On a host with synchronized TSC, there is no need to update
2745                  * kvmclock on vcpu->cpu migration
2746                  */
2747                 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2748                         kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2749                 if (vcpu->cpu != cpu)
2750                         kvm_migrate_timers(vcpu);
2751                 vcpu->cpu = cpu;
2752         }
2753
2754         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2755 }
2756
2757 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2758 {
2759         kvm_x86_ops->vcpu_put(vcpu);
2760         kvm_put_guest_fpu(vcpu);
2761         vcpu->arch.last_host_tsc = rdtsc();
2762 }
2763
2764 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2765                                     struct kvm_lapic_state *s)
2766 {
2767         if (vcpu->arch.apicv_active)
2768                 kvm_x86_ops->sync_pir_to_irr(vcpu);
2769
2770         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2771
2772         return 0;
2773 }
2774
2775 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2776                                     struct kvm_lapic_state *s)
2777 {
2778         kvm_apic_post_state_restore(vcpu, s);
2779         update_cr8_intercept(vcpu);
2780
2781         return 0;
2782 }
2783
2784 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2785 {
2786         return (!lapic_in_kernel(vcpu) ||
2787                 kvm_apic_accept_pic_intr(vcpu));
2788 }
2789
2790 /*
2791  * if userspace requested an interrupt window, check that the
2792  * interrupt window is open.
2793  *
2794  * No need to exit to userspace if we already have an interrupt queued.
2795  */
2796 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2797 {
2798         return kvm_arch_interrupt_allowed(vcpu) &&
2799                 !kvm_cpu_has_interrupt(vcpu) &&
2800                 !kvm_event_needs_reinjection(vcpu) &&
2801                 kvm_cpu_accept_dm_intr(vcpu);
2802 }
2803
2804 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2805                                     struct kvm_interrupt *irq)
2806 {
2807         if (irq->irq >= KVM_NR_INTERRUPTS)
2808                 return -EINVAL;
2809
2810         if (!irqchip_in_kernel(vcpu->kvm)) {
2811                 kvm_queue_interrupt(vcpu, irq->irq, false);
2812                 kvm_make_request(KVM_REQ_EVENT, vcpu);
2813                 return 0;
2814         }
2815
2816         /*
2817          * With in-kernel LAPIC, we only use this to inject EXTINT, so
2818          * fail for in-kernel 8259.
2819          */
2820         if (pic_in_kernel(vcpu->kvm))
2821                 return -ENXIO;
2822
2823         if (vcpu->arch.pending_external_vector != -1)
2824                 return -EEXIST;
2825
2826         vcpu->arch.pending_external_vector = irq->irq;
2827         kvm_make_request(KVM_REQ_EVENT, vcpu);
2828         return 0;
2829 }
2830
2831 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2832 {
2833         kvm_inject_nmi(vcpu);
2834
2835         return 0;
2836 }
2837
2838 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2839 {
2840         kvm_make_request(KVM_REQ_SMI, vcpu);
2841
2842         return 0;
2843 }
2844
2845 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2846                                            struct kvm_tpr_access_ctl *tac)
2847 {
2848         if (tac->flags)
2849                 return -EINVAL;
2850         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2851         return 0;
2852 }
2853
2854 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2855                                         u64 mcg_cap)
2856 {
2857         int r;
2858         unsigned bank_num = mcg_cap & 0xff, bank;
2859
2860         r = -EINVAL;
2861         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2862                 goto out;
2863         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2864                 goto out;
2865         r = 0;
2866         vcpu->arch.mcg_cap = mcg_cap;
2867         /* Init IA32_MCG_CTL to all 1s */
2868         if (mcg_cap & MCG_CTL_P)
2869                 vcpu->arch.mcg_ctl = ~(u64)0;
2870         /* Init IA32_MCi_CTL to all 1s */
2871         for (bank = 0; bank < bank_num; bank++)
2872                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2873 out:
2874         return r;
2875 }
2876
2877 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2878                                       struct kvm_x86_mce *mce)
2879 {
2880         u64 mcg_cap = vcpu->arch.mcg_cap;
2881         unsigned bank_num = mcg_cap & 0xff;
2882         u64 *banks = vcpu->arch.mce_banks;
2883
2884         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2885                 return -EINVAL;
2886         /*
2887          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2888          * reporting is disabled
2889          */
2890         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2891             vcpu->arch.mcg_ctl != ~(u64)0)
2892                 return 0;
2893         banks += 4 * mce->bank;
2894         /*
2895          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2896          * reporting is disabled for the bank
2897          */
2898         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2899                 return 0;
2900         if (mce->status & MCI_STATUS_UC) {
2901                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2902                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2903                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2904                         return 0;
2905                 }
2906                 if (banks[1] & MCI_STATUS_VAL)
2907                         mce->status |= MCI_STATUS_OVER;
2908                 banks[2] = mce->addr;
2909                 banks[3] = mce->misc;
2910                 vcpu->arch.mcg_status = mce->mcg_status;
2911                 banks[1] = mce->status;
2912                 kvm_queue_exception(vcpu, MC_VECTOR);
2913         } else if (!(banks[1] & MCI_STATUS_VAL)
2914                    || !(banks[1] & MCI_STATUS_UC)) {
2915                 if (banks[1] & MCI_STATUS_VAL)
2916                         mce->status |= MCI_STATUS_OVER;
2917                 banks[2] = mce->addr;
2918                 banks[3] = mce->misc;
2919                 banks[1] = mce->status;
2920         } else
2921                 banks[1] |= MCI_STATUS_OVER;
2922         return 0;
2923 }
2924
2925 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2926                                                struct kvm_vcpu_events *events)
2927 {
2928         process_nmi(vcpu);
2929         events->exception.injected =
2930                 vcpu->arch.exception.pending &&
2931                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2932         events->exception.nr = vcpu->arch.exception.nr;
2933         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2934         events->exception.pad = 0;
2935         events->exception.error_code = vcpu->arch.exception.error_code;
2936
2937         events->interrupt.injected =
2938                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2939         events->interrupt.nr = vcpu->arch.interrupt.nr;
2940         events->interrupt.soft = 0;
2941         events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2942
2943         events->nmi.injected = vcpu->arch.nmi_injected;
2944         events->nmi.pending = vcpu->arch.nmi_pending != 0;
2945         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2946         events->nmi.pad = 0;
2947
2948         events->sipi_vector = 0; /* never valid when reporting to user space */
2949
2950         events->smi.smm = is_smm(vcpu);
2951         events->smi.pending = vcpu->arch.smi_pending;
2952         events->smi.smm_inside_nmi =
2953                 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2954         events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2955
2956         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2957                          | KVM_VCPUEVENT_VALID_SHADOW
2958                          | KVM_VCPUEVENT_VALID_SMM);
2959         memset(&events->reserved, 0, sizeof(events->reserved));
2960 }
2961
2962 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2963                                               struct kvm_vcpu_events *events)
2964 {
2965         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2966                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2967                               | KVM_VCPUEVENT_VALID_SHADOW
2968                               | KVM_VCPUEVENT_VALID_SMM))
2969                 return -EINVAL;
2970
2971         process_nmi(vcpu);
2972         vcpu->arch.exception.pending = events->exception.injected;
2973         vcpu->arch.exception.nr = events->exception.nr;
2974         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2975         vcpu->arch.exception.error_code = events->exception.error_code;
2976
2977         vcpu->arch.interrupt.pending = events->interrupt.injected;
2978         vcpu->arch.interrupt.nr = events->interrupt.nr;
2979         vcpu->arch.interrupt.soft = events->interrupt.soft;
2980         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2981                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2982                                                   events->interrupt.shadow);
2983
2984         vcpu->arch.nmi_injected = events->nmi.injected;
2985         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2986                 vcpu->arch.nmi_pending = events->nmi.pending;
2987         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2988
2989         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2990             kvm_vcpu_has_lapic(vcpu))
2991                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2992
2993         if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2994                 if (events->smi.smm)
2995                         vcpu->arch.hflags |= HF_SMM_MASK;
2996                 else
2997                         vcpu->arch.hflags &= ~HF_SMM_MASK;
2998                 vcpu->arch.smi_pending = events->smi.pending;
2999                 if (events->smi.smm_inside_nmi)
3000                         vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3001                 else
3002                         vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3003                 if (kvm_vcpu_has_lapic(vcpu)) {
3004                         if (events->smi.latched_init)
3005                                 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3006                         else
3007                                 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3008                 }
3009         }
3010
3011         kvm_make_request(KVM_REQ_EVENT, vcpu);
3012
3013         return 0;
3014 }
3015
3016 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3017                                              struct kvm_debugregs *dbgregs)
3018 {
3019         unsigned long val;
3020
3021         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3022         kvm_get_dr(vcpu, 6, &val);
3023         dbgregs->dr6 = val;
3024         dbgregs->dr7 = vcpu->arch.dr7;
3025         dbgregs->flags = 0;
3026         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3027 }
3028
3029 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3030                                             struct kvm_debugregs *dbgregs)
3031 {
3032         if (dbgregs->flags)
3033                 return -EINVAL;
3034
3035         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3036         kvm_update_dr0123(vcpu);
3037         vcpu->arch.dr6 = dbgregs->dr6;
3038         kvm_update_dr6(vcpu);
3039         vcpu->arch.dr7 = dbgregs->dr7;
3040         kvm_update_dr7(vcpu);
3041
3042         return 0;
3043 }
3044
3045 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3046
3047 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3048 {
3049         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3050         u64 xstate_bv = xsave->header.xfeatures;
3051         u64 valid;
3052
3053         /*
3054          * Copy legacy XSAVE area, to avoid complications with CPUID
3055          * leaves 0 and 1 in the loop below.
3056          */
3057         memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3058
3059         /* Set XSTATE_BV */
3060         *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3061
3062         /*
3063          * Copy each region from the possibly compacted offset to the
3064          * non-compacted offset.
3065          */
3066         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3067         while (valid) {
3068                 u64 feature = valid & -valid;
3069                 int index = fls64(feature) - 1;
3070                 void *src = get_xsave_addr(xsave, feature);
3071
3072                 if (src) {
3073                         u32 size, offset, ecx, edx;
3074                         cpuid_count(XSTATE_CPUID, index,
3075                                     &size, &offset, &ecx, &edx);
3076                         memcpy(dest + offset, src, size);
3077                 }
3078
3079                 valid -= feature;
3080         }
3081 }
3082
3083 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3084 {
3085         struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3086         u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3087         u64 valid;
3088
3089         /*
3090          * Copy legacy XSAVE area, to avoid complications with CPUID
3091          * leaves 0 and 1 in the loop below.
3092          */
3093         memcpy(xsave, src, XSAVE_HDR_OFFSET);
3094
3095         /* Set XSTATE_BV and possibly XCOMP_BV.  */
3096         xsave->header.xfeatures = xstate_bv;
3097         if (cpu_has_xsaves)
3098                 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3099
3100         /*
3101          * Copy each region from the non-compacted offset to the
3102          * possibly compacted offset.
3103          */
3104         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3105         while (valid) {
3106                 u64 feature = valid & -valid;
3107                 int index = fls64(feature) - 1;
3108                 void *dest = get_xsave_addr(xsave, feature);
3109
3110                 if (dest) {
3111                         u32 size, offset, ecx, edx;
3112                         cpuid_count(XSTATE_CPUID, index,
3113                                     &size, &offset, &ecx, &edx);
3114                         memcpy(dest, src + offset, size);
3115                 }
3116
3117                 valid -= feature;
3118         }
3119 }
3120
3121 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3122                                          struct kvm_xsave *guest_xsave)
3123 {
3124         if (cpu_has_xsave) {
3125                 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3126                 fill_xsave((u8 *) guest_xsave->region, vcpu);
3127         } else {
3128                 memcpy(guest_xsave->region,
3129                         &vcpu->arch.guest_fpu.state.fxsave,
3130                         sizeof(struct fxregs_state));
3131                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3132                         XFEATURE_MASK_FPSSE;
3133         }
3134 }
3135
3136 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3137                                         struct kvm_xsave *guest_xsave)
3138 {
3139         u64 xstate_bv =
3140                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3141
3142         if (cpu_has_xsave) {
3143                 /*
3144                  * Here we allow setting states that are not present in
3145                  * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3146                  * with old userspace.
3147                  */
3148                 if (xstate_bv & ~kvm_supported_xcr0())
3149                         return -EINVAL;
3150                 load_xsave(vcpu, (u8 *)guest_xsave->region);
3151         } else {
3152                 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3153                         return -EINVAL;
3154                 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3155                         guest_xsave->region, sizeof(struct fxregs_state));
3156         }
3157         return 0;
3158 }
3159
3160 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3161                                         struct kvm_xcrs *guest_xcrs)
3162 {
3163         if (!cpu_has_xsave) {
3164                 guest_xcrs->nr_xcrs = 0;
3165                 return;
3166         }
3167
3168         guest_xcrs->nr_xcrs = 1;
3169         guest_xcrs->flags = 0;
3170         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3171         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3172 }
3173
3174 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3175                                        struct kvm_xcrs *guest_xcrs)
3176 {
3177         int i, r = 0;
3178
3179         if (!cpu_has_xsave)
3180                 return -EINVAL;
3181
3182         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3183                 return -EINVAL;
3184
3185         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3186                 /* Only support XCR0 currently */
3187                 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3188                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3189                                 guest_xcrs->xcrs[i].value);
3190                         break;
3191                 }
3192         if (r)
3193                 r = -EINVAL;
3194         return r;
3195 }
3196
3197 /*
3198  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3199  * stopped by the hypervisor.  This function will be called from the host only.
3200  * EINVAL is returned when the host attempts to set the flag for a guest that
3201  * does not support pv clocks.
3202  */
3203 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3204 {
3205         if (!vcpu->arch.pv_time_enabled)
3206                 return -EINVAL;
3207         vcpu->arch.pvclock_set_guest_stopped_request = true;
3208         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3209         return 0;
3210 }
3211
3212 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3213                                      struct kvm_enable_cap *cap)
3214 {
3215         if (cap->flags)
3216                 return -EINVAL;
3217
3218         switch (cap->cap) {
3219         case KVM_CAP_HYPERV_SYNIC:
3220                 return kvm_hv_activate_synic(vcpu);
3221         default:
3222                 return -EINVAL;
3223         }
3224 }
3225
3226 long kvm_arch_vcpu_ioctl(struct file *filp,
3227                          unsigned int ioctl, unsigned long arg)
3228 {
3229         struct kvm_vcpu *vcpu = filp->private_data;
3230         void __user *argp = (void __user *)arg;
3231         int r;
3232         union {
3233                 struct kvm_lapic_state *lapic;
3234                 struct kvm_xsave *xsave;
3235                 struct kvm_xcrs *xcrs;
3236                 void *buffer;
3237         } u;
3238
3239         u.buffer = NULL;
3240         switch (ioctl) {
3241         case KVM_GET_LAPIC: {
3242                 r = -EINVAL;
3243                 if (!vcpu->arch.apic)
3244                         goto out;
3245                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3246
3247                 r = -ENOMEM;
3248                 if (!u.lapic)
3249                         goto out;
3250                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3251                 if (r)
3252                         goto out;
3253                 r = -EFAULT;
3254                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3255                         goto out;
3256                 r = 0;
3257                 break;
3258         }
3259         case KVM_SET_LAPIC: {
3260                 r = -EINVAL;
3261                 if (!vcpu->arch.apic)
3262                         goto out;
3263                 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3264                 if (IS_ERR(u.lapic))
3265                         return PTR_ERR(u.lapic);
3266
3267                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3268                 break;
3269         }
3270         case KVM_INTERRUPT: {
3271                 struct kvm_interrupt irq;
3272
3273                 r = -EFAULT;
3274                 if (copy_from_user(&irq, argp, sizeof irq))
3275                         goto out;
3276                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3277                 break;
3278         }
3279         case KVM_NMI: {
3280                 r = kvm_vcpu_ioctl_nmi(vcpu);
3281                 break;
3282         }
3283         case KVM_SMI: {
3284                 r = kvm_vcpu_ioctl_smi(vcpu);
3285                 break;
3286         }
3287         case KVM_SET_CPUID: {
3288                 struct kvm_cpuid __user *cpuid_arg = argp;
3289                 struct kvm_cpuid cpuid;
3290
3291                 r = -EFAULT;
3292                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3293                         goto out;
3294                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3295                 break;
3296         }
3297         case KVM_SET_CPUID2: {
3298                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3299                 struct kvm_cpuid2 cpuid;
3300
3301                 r = -EFAULT;
3302                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3303                         goto out;
3304                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3305                                               cpuid_arg->entries);
3306                 break;
3307         }
3308         case KVM_GET_CPUID2: {
3309                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3310                 struct kvm_cpuid2 cpuid;
3311
3312                 r = -EFAULT;
3313                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3314                         goto out;
3315                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3316                                               cpuid_arg->entries);
3317                 if (r)
3318                         goto out;
3319                 r = -EFAULT;
3320                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3321                         goto out;
3322                 r = 0;
3323                 break;
3324         }
3325         case KVM_GET_MSRS:
3326                 r = msr_io(vcpu, argp, do_get_msr, 1);
3327                 break;
3328         case KVM_SET_MSRS:
3329                 r = msr_io(vcpu, argp, do_set_msr, 0);
3330                 break;
3331         case KVM_TPR_ACCESS_REPORTING: {
3332                 struct kvm_tpr_access_ctl tac;
3333
3334                 r = -EFAULT;
3335                 if (copy_from_user(&tac, argp, sizeof tac))
3336                         goto out;
3337                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3338                 if (r)
3339                         goto out;
3340                 r = -EFAULT;
3341                 if (copy_to_user(argp, &tac, sizeof tac))
3342                         goto out;
3343                 r = 0;
3344                 break;
3345         };
3346         case KVM_SET_VAPIC_ADDR: {
3347                 struct kvm_vapic_addr va;
3348
3349                 r = -EINVAL;
3350                 if (!lapic_in_kernel(vcpu))
3351                         goto out;
3352                 r = -EFAULT;
3353                 if (copy_from_user(&va, argp, sizeof va))
3354                         goto out;
3355                 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3356                 break;
3357         }
3358         case KVM_X86_SETUP_MCE: {
3359                 u64 mcg_cap;
3360
3361                 r = -EFAULT;
3362                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3363                         goto out;
3364                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3365                 break;
3366         }
3367         case KVM_X86_SET_MCE: {
3368                 struct kvm_x86_mce mce;
3369
3370                 r = -EFAULT;
3371                 if (copy_from_user(&mce, argp, sizeof mce))
3372                         goto out;
3373                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3374                 break;
3375         }
3376         case KVM_GET_VCPU_EVENTS: {
3377                 struct kvm_vcpu_events events;
3378
3379                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3380
3381                 r = -EFAULT;
3382                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3383                         break;
3384                 r = 0;
3385                 break;
3386         }
3387         case KVM_SET_VCPU_EVENTS: {
3388                 struct kvm_vcpu_events events;
3389
3390                 r = -EFAULT;
3391                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3392                         break;
3393
3394                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3395                 break;
3396         }
3397         case KVM_GET_DEBUGREGS: {
3398                 struct kvm_debugregs dbgregs;
3399
3400                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3401
3402                 r = -EFAULT;
3403                 if (copy_to_user(argp, &dbgregs,
3404                                  sizeof(struct kvm_debugregs)))
3405                         break;
3406                 r = 0;
3407                 break;
3408         }
3409         case KVM_SET_DEBUGREGS: {
3410                 struct kvm_debugregs dbgregs;
3411
3412                 r = -EFAULT;
3413                 if (copy_from_user(&dbgregs, argp,
3414                                    sizeof(struct kvm_debugregs)))
3415                         break;
3416
3417                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3418                 break;
3419         }
3420         case KVM_GET_XSAVE: {
3421                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3422                 r = -ENOMEM;
3423                 if (!u.xsave)
3424                         break;
3425
3426                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3427
3428                 r = -EFAULT;
3429                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3430                         break;
3431                 r = 0;
3432                 break;
3433         }
3434         case KVM_SET_XSAVE: {
3435                 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3436                 if (IS_ERR(u.xsave))
3437                         return PTR_ERR(u.xsave);
3438
3439                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3440                 break;
3441         }
3442         case KVM_GET_XCRS: {
3443                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3444                 r = -ENOMEM;
3445                 if (!u.xcrs)
3446                         break;
3447
3448                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3449
3450                 r = -EFAULT;
3451                 if (copy_to_user(argp, u.xcrs,
3452                                  sizeof(struct kvm_xcrs)))
3453                         break;
3454                 r = 0;
3455                 break;
3456         }
3457         case KVM_SET_XCRS: {
3458                 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3459                 if (IS_ERR(u.xcrs))
3460                         return PTR_ERR(u.xcrs);
3461
3462                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3463                 break;
3464         }
3465         case KVM_SET_TSC_KHZ: {
3466                 u32 user_tsc_khz;
3467
3468                 r = -EINVAL;
3469                 user_tsc_khz = (u32)arg;
3470
3471                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3472                         goto out;
3473
3474                 if (user_tsc_khz == 0)
3475                         user_tsc_khz = tsc_khz;
3476
3477                 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3478                         r = 0;
3479
3480                 goto out;
3481         }
3482         case KVM_GET_TSC_KHZ: {
3483                 r = vcpu->arch.virtual_tsc_khz;
3484                 goto out;
3485         }
3486         case KVM_KVMCLOCK_CTRL: {
3487                 r = kvm_set_guest_paused(vcpu);
3488                 goto out;
3489         }
3490         case KVM_ENABLE_CAP: {
3491                 struct kvm_enable_cap cap;
3492
3493                 r = -EFAULT;
3494                 if (copy_from_user(&cap, argp, sizeof(cap)))
3495                         goto out;
3496                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3497                 break;
3498         }
3499         default:
3500                 r = -EINVAL;
3501         }
3502 out:
3503         kfree(u.buffer);
3504         return r;
3505 }
3506
3507 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3508 {
3509         return VM_FAULT_SIGBUS;
3510 }
3511
3512 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3513 {
3514         int ret;
3515
3516         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3517                 return -EINVAL;
3518         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3519         return ret;
3520 }
3521
3522 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3523                                               u64 ident_addr)
3524 {
3525         kvm->arch.ept_identity_map_addr = ident_addr;
3526         return 0;
3527 }
3528
3529 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3530                                           u32 kvm_nr_mmu_pages)
3531 {
3532         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3533                 return -EINVAL;
3534
3535         mutex_lock(&kvm->slots_lock);
3536
3537         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3538         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3539
3540         mutex_unlock(&kvm->slots_lock);
3541         return 0;
3542 }
3543
3544 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3545 {
3546         return kvm->arch.n_max_mmu_pages;
3547 }
3548
3549 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3550 {
3551         int r;
3552
3553         r = 0;
3554         switch (chip->chip_id) {
3555         case KVM_IRQCHIP_PIC_MASTER:
3556                 memcpy(&chip->chip.pic,
3557                         &pic_irqchip(kvm)->pics[0],
3558                         sizeof(struct kvm_pic_state));
3559                 break;
3560         case KVM_IRQCHIP_PIC_SLAVE:
3561                 memcpy(&chip->chip.pic,
3562                         &pic_irqchip(kvm)->pics[1],
3563                         sizeof(struct kvm_pic_state));
3564                 break;
3565         case KVM_IRQCHIP_IOAPIC:
3566                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3567                 break;
3568         default:
3569                 r = -EINVAL;
3570                 break;
3571         }
3572         return r;
3573 }
3574
3575 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3576 {
3577         int r;
3578
3579         r = 0;
3580         switch (chip->chip_id) {
3581         case KVM_IRQCHIP_PIC_MASTER:
3582                 spin_lock(&pic_irqchip(kvm)->lock);
3583                 memcpy(&pic_irqchip(kvm)->pics[0],
3584                         &chip->chip.pic,
3585                         sizeof(struct kvm_pic_state));
3586                 spin_unlock(&pic_irqchip(kvm)->lock);
3587                 break;
3588         case KVM_IRQCHIP_PIC_SLAVE:
3589                 spin_lock(&pic_irqchip(kvm)->lock);
3590                 memcpy(&pic_irqchip(kvm)->pics[1],
3591                         &chip->chip.pic,
3592                         sizeof(struct kvm_pic_state));
3593                 spin_unlock(&pic_irqchip(kvm)->lock);
3594                 break;
3595         case KVM_IRQCHIP_IOAPIC:
3596                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3597                 break;
3598         default:
3599                 r = -EINVAL;
3600                 break;
3601         }
3602         kvm_pic_update_irq(pic_irqchip(kvm));
3603         return r;
3604 }
3605
3606 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3607 {
3608         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3609         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3610         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3611         return 0;
3612 }
3613
3614 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3615 {
3616         int i;
3617         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3618         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3619         for (i = 0; i < 3; i++)
3620                 kvm_pit_load_count(kvm, i, ps->channels[i].count, 0);
3621         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3622         return 0;
3623 }
3624
3625 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3626 {
3627         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3628         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3629                 sizeof(ps->channels));
3630         ps->flags = kvm->arch.vpit->pit_state.flags;
3631         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3632         memset(&ps->reserved, 0, sizeof(ps->reserved));
3633         return 0;
3634 }
3635
3636 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3637 {
3638         int start = 0;
3639         int i;
3640         u32 prev_legacy, cur_legacy;
3641         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3642         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3643         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3644         if (!prev_legacy && cur_legacy)
3645                 start = 1;
3646         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3647                sizeof(kvm->arch.vpit->pit_state.channels));
3648         kvm->arch.vpit->pit_state.flags = ps->flags;
3649         for (i = 0; i < 3; i++)
3650                 kvm_pit_load_count(kvm, i, kvm->arch.vpit->pit_state.channels[i].count,
3651                                    start && i == 0);
3652         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3653         return 0;
3654 }
3655
3656 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3657                                  struct kvm_reinject_control *control)
3658 {
3659         if (!kvm->arch.vpit)
3660                 return -ENXIO;
3661         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3662         kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3663         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3664         return 0;
3665 }
3666
3667 /**
3668  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3669  * @kvm: kvm instance
3670  * @log: slot id and address to which we copy the log
3671  *
3672  * Steps 1-4 below provide general overview of dirty page logging. See
3673  * kvm_get_dirty_log_protect() function description for additional details.
3674  *
3675  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3676  * always flush the TLB (step 4) even if previous step failed  and the dirty
3677  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3678  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3679  * writes will be marked dirty for next log read.
3680  *
3681  *   1. Take a snapshot of the bit and clear it if needed.
3682  *   2. Write protect the corresponding page.
3683  *   3. Copy the snapshot to the userspace.
3684  *   4. Flush TLB's if needed.
3685  */
3686 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3687 {
3688         bool is_dirty = false;
3689         int r;
3690
3691         mutex_lock(&kvm->slots_lock);
3692
3693         /*
3694          * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3695          */
3696         if (kvm_x86_ops->flush_log_dirty)
3697                 kvm_x86_ops->flush_log_dirty(kvm);
3698
3699         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3700
3701         /*
3702          * All the TLBs can be flushed out of mmu lock, see the comments in
3703          * kvm_mmu_slot_remove_write_access().
3704          */
3705         lockdep_assert_held(&kvm->slots_lock);
3706         if (is_dirty)
3707                 kvm_flush_remote_tlbs(kvm);
3708
3709         mutex_unlock(&kvm->slots_lock);
3710         return r;
3711 }
3712
3713 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3714                         bool line_status)
3715 {
3716         if (!irqchip_in_kernel(kvm))
3717                 return -ENXIO;
3718
3719         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3720                                         irq_event->irq, irq_event->level,
3721                                         line_status);
3722         return 0;
3723 }
3724
3725 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3726                                    struct kvm_enable_cap *cap)
3727 {
3728         int r;
3729
3730         if (cap->flags)
3731                 return -EINVAL;
3732
3733         switch (cap->cap) {
3734         case KVM_CAP_DISABLE_QUIRKS:
3735                 kvm->arch.disabled_quirks = cap->args[0];
3736                 r = 0;
3737                 break;
3738         case KVM_CAP_SPLIT_IRQCHIP: {
3739                 mutex_lock(&kvm->lock);
3740                 r = -EINVAL;
3741                 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3742                         goto split_irqchip_unlock;
3743                 r = -EEXIST;
3744                 if (irqchip_in_kernel(kvm))
3745                         goto split_irqchip_unlock;
3746                 if (atomic_read(&kvm->online_vcpus))
3747                         goto split_irqchip_unlock;
3748                 r = kvm_setup_empty_irq_routing(kvm);
3749                 if (r)
3750                         goto split_irqchip_unlock;
3751                 /* Pairs with irqchip_in_kernel. */
3752                 smp_wmb();
3753                 kvm->arch.irqchip_split = true;
3754                 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3755                 r = 0;
3756 split_irqchip_unlock:
3757                 mutex_unlock(&kvm->lock);
3758                 break;
3759         }
3760         default:
3761                 r = -EINVAL;
3762                 break;
3763         }
3764         return r;
3765 }
3766
3767 long kvm_arch_vm_ioctl(struct file *filp,
3768                        unsigned int ioctl, unsigned long arg)
3769 {
3770         struct kvm *kvm = filp->private_data;
3771         void __user *argp = (void __user *)arg;
3772         int r = -ENOTTY;
3773         /*
3774          * This union makes it completely explicit to gcc-3.x
3775          * that these two variables' stack usage should be
3776          * combined, not added together.
3777          */
3778         union {
3779                 struct kvm_pit_state ps;
3780                 struct kvm_pit_state2 ps2;
3781                 struct kvm_pit_config pit_config;
3782         } u;
3783
3784         switch (ioctl) {
3785         case KVM_SET_TSS_ADDR:
3786                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3787                 break;
3788         case KVM_SET_IDENTITY_MAP_ADDR: {
3789                 u64 ident_addr;
3790
3791                 r = -EFAULT;
3792                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3793                         goto out;
3794                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3795                 break;
3796         }
3797         case KVM_SET_NR_MMU_PAGES:
3798                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3799                 break;
3800         case KVM_GET_NR_MMU_PAGES:
3801                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3802                 break;
3803         case KVM_CREATE_IRQCHIP: {
3804                 struct kvm_pic *vpic;
3805
3806                 mutex_lock(&kvm->lock);
3807                 r = -EEXIST;
3808                 if (kvm->arch.vpic)
3809                         goto create_irqchip_unlock;
3810                 r = -EINVAL;
3811                 if (atomic_read(&kvm->online_vcpus))
3812                         goto create_irqchip_unlock;
3813                 r = -ENOMEM;
3814                 vpic = kvm_create_pic(kvm);
3815                 if (vpic) {
3816                         r = kvm_ioapic_init(kvm);
3817                         if (r) {
3818                                 mutex_lock(&kvm->slots_lock);
3819                                 kvm_destroy_pic(vpic);
3820                                 mutex_unlock(&kvm->slots_lock);
3821                                 goto create_irqchip_unlock;
3822                         }
3823                 } else
3824                         goto create_irqchip_unlock;
3825                 r = kvm_setup_default_irq_routing(kvm);
3826                 if (r) {
3827                         mutex_lock(&kvm->slots_lock);
3828                         mutex_lock(&kvm->irq_lock);
3829                         kvm_ioapic_destroy(kvm);
3830                         kvm_destroy_pic(vpic);
3831                         mutex_unlock(&kvm->irq_lock);
3832                         mutex_unlock(&kvm->slots_lock);
3833                         goto create_irqchip_unlock;
3834                 }
3835                 /* Write kvm->irq_routing before kvm->arch.vpic.  */
3836                 smp_wmb();
3837                 kvm->arch.vpic = vpic;
3838         create_irqchip_unlock:
3839                 mutex_unlock(&kvm->lock);
3840                 break;
3841         }
3842         case KVM_CREATE_PIT:
3843                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3844                 goto create_pit;
3845         case KVM_CREATE_PIT2:
3846                 r = -EFAULT;
3847                 if (copy_from_user(&u.pit_config, argp,
3848                                    sizeof(struct kvm_pit_config)))
3849                         goto out;
3850         create_pit:
3851                 mutex_lock(&kvm->slots_lock);
3852                 r = -EEXIST;
3853                 if (kvm->arch.vpit)
3854                         goto create_pit_unlock;
3855                 r = -ENOMEM;
3856                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3857                 if (kvm->arch.vpit)
3858                         r = 0;
3859         create_pit_unlock:
3860                 mutex_unlock(&kvm->slots_lock);
3861                 break;
3862         case KVM_GET_IRQCHIP: {
3863                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3864                 struct kvm_irqchip *chip;
3865
3866                 chip = memdup_user(argp, sizeof(*chip));
3867                 if (IS_ERR(chip)) {
3868                         r = PTR_ERR(chip);
3869                         goto out;
3870                 }
3871
3872                 r = -ENXIO;
3873                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3874                         goto get_irqchip_out;
3875                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3876                 if (r)
3877                         goto get_irqchip_out;
3878                 r = -EFAULT;
3879                 if (copy_to_user(argp, chip, sizeof *chip))
3880                         goto get_irqchip_out;
3881                 r = 0;
3882         get_irqchip_out:
3883                 kfree(chip);
3884                 break;
3885         }
3886         case KVM_SET_IRQCHIP: {
3887                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3888                 struct kvm_irqchip *chip;
3889
3890                 chip = memdup_user(argp, sizeof(*chip));
3891                 if (IS_ERR(chip)) {
3892                         r = PTR_ERR(chip);
3893                         goto out;
3894                 }
3895
3896                 r = -ENXIO;
3897                 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3898                         goto set_irqchip_out;
3899                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3900                 if (r)
3901                         goto set_irqchip_out;
3902                 r = 0;
3903         set_irqchip_out:
3904                 kfree(chip);
3905                 break;
3906         }
3907         case KVM_GET_PIT: {
3908                 r = -EFAULT;
3909                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3910                         goto out;
3911                 r = -ENXIO;
3912                 if (!kvm->arch.vpit)
3913                         goto out;
3914                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3915                 if (r)
3916                         goto out;
3917                 r = -EFAULT;
3918                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3919                         goto out;
3920                 r = 0;
3921                 break;
3922         }
3923         case KVM_SET_PIT: {
3924                 r = -EFAULT;
3925                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3926                         goto out;
3927                 r = -ENXIO;
3928                 if (!kvm->arch.vpit)
3929                         goto out;
3930                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3931                 break;
3932         }
3933         case KVM_GET_PIT2: {
3934                 r = -ENXIO;
3935                 if (!kvm->arch.vpit)
3936                         goto out;
3937                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3938                 if (r)
3939                         goto out;
3940                 r = -EFAULT;
3941                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3942                         goto out;
3943                 r = 0;
3944                 break;
3945         }
3946         case KVM_SET_PIT2: {
3947                 r = -EFAULT;
3948                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3949                         goto out;
3950                 r = -ENXIO;
3951                 if (!kvm->arch.vpit)
3952                         goto out;
3953                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3954                 break;
3955         }
3956         case KVM_REINJECT_CONTROL: {
3957                 struct kvm_reinject_control control;
3958                 r =  -EFAULT;
3959                 if (copy_from_user(&control, argp, sizeof(control)))
3960                         goto out;
3961                 r = kvm_vm_ioctl_reinject(kvm, &control);
3962                 break;
3963         }
3964         case KVM_SET_BOOT_CPU_ID:
3965                 r = 0;
3966                 mutex_lock(&kvm->lock);
3967                 if (atomic_read(&kvm->online_vcpus) != 0)
3968                         r = -EBUSY;
3969                 else
3970                         kvm->arch.bsp_vcpu_id = arg;
3971                 mutex_unlock(&kvm->lock);
3972                 break;
3973         case KVM_XEN_HVM_CONFIG: {
3974                 r = -EFAULT;
3975                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3976                                    sizeof(struct kvm_xen_hvm_config)))
3977                         goto out;
3978                 r = -EINVAL;
3979                 if (kvm->arch.xen_hvm_config.flags)
3980                         goto out;
3981                 r = 0;
3982                 break;
3983         }
3984         case KVM_SET_CLOCK: {
3985                 struct kvm_clock_data user_ns;
3986                 u64 now_ns;
3987                 s64 delta;
3988
3989                 r = -EFAULT;
3990                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3991                         goto out;
3992
3993                 r = -EINVAL;
3994                 if (user_ns.flags)
3995                         goto out;
3996
3997                 r = 0;
3998                 local_irq_disable();
3999                 now_ns = get_kernel_ns();
4000                 delta = user_ns.clock - now_ns;
4001                 local_irq_enable();
4002                 kvm->arch.kvmclock_offset = delta;
4003                 kvm_gen_update_masterclock(kvm);
4004                 break;
4005         }
4006         case KVM_GET_CLOCK: {
4007                 struct kvm_clock_data user_ns;
4008                 u64 now_ns;
4009
4010                 local_irq_disable();
4011                 now_ns = get_kernel_ns();
4012                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4013                 local_irq_enable();
4014                 user_ns.flags = 0;
4015                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4016
4017                 r = -EFAULT;
4018                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4019                         goto out;
4020                 r = 0;
4021                 break;
4022         }
4023         case KVM_ENABLE_CAP: {
4024                 struct kvm_enable_cap cap;
4025
4026                 r = -EFAULT;
4027                 if (copy_from_user(&cap, argp, sizeof(cap)))
4028                         goto out;
4029                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4030                 break;
4031         }
4032         default:
4033                 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4034         }
4035 out:
4036         return r;
4037 }
4038
4039 static void kvm_init_msr_list(void)
4040 {
4041         u32 dummy[2];
4042         unsigned i, j;
4043
4044         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4045                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4046                         continue;
4047
4048                 /*
4049                  * Even MSRs that are valid in the host may not be exposed
4050                  * to the guests in some cases.
4051                  */
4052                 switch (msrs_to_save[i]) {
4053                 case MSR_IA32_BNDCFGS:
4054                         if (!kvm_x86_ops->mpx_supported())
4055                                 continue;
4056                         break;
4057                 case MSR_TSC_AUX:
4058                         if (!kvm_x86_ops->rdtscp_supported())
4059                                 continue;
4060                         break;
4061                 default:
4062                         break;
4063                 }
4064
4065                 if (j < i)
4066                         msrs_to_save[j] = msrs_to_save[i];
4067                 j++;
4068         }
4069         num_msrs_to_save = j;
4070
4071         for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4072                 switch (emulated_msrs[i]) {
4073                 case MSR_IA32_SMBASE:
4074                         if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4075                                 continue;
4076                         break;
4077                 default:
4078                         break;
4079                 }
4080
4081                 if (j < i)
4082                         emulated_msrs[j] = emulated_msrs[i];
4083                 j++;
4084         }
4085         num_emulated_msrs = j;
4086 }
4087
4088 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4089                            const void *v)
4090 {
4091         int handled = 0;
4092         int n;
4093
4094         do {
4095                 n = min(len, 8);
4096                 if (!(vcpu->arch.apic &&
4097                       !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4098                     && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4099                         break;
4100                 handled += n;
4101                 addr += n;
4102                 len -= n;
4103                 v += n;
4104         } while (len);
4105
4106         return handled;
4107 }
4108
4109 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4110 {
4111         int handled = 0;
4112         int n;
4113
4114         do {
4115                 n = min(len, 8);
4116                 if (!(vcpu->arch.apic &&
4117                       !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4118                                          addr, n, v))
4119                     && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4120                         break;
4121                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4122                 handled += n;
4123                 addr += n;
4124                 len -= n;
4125                 v += n;
4126         } while (len);
4127
4128         return handled;
4129 }
4130
4131 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4132                         struct kvm_segment *var, int seg)
4133 {
4134         kvm_x86_ops->set_segment(vcpu, var, seg);
4135 }
4136
4137 void kvm_get_segment(struct kvm_vcpu *vcpu,
4138                      struct kvm_segment *var, int seg)
4139 {
4140         kvm_x86_ops->get_segment(vcpu, var, seg);
4141 }
4142
4143 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4144                            struct x86_exception *exception)
4145 {
4146         gpa_t t_gpa;
4147
4148         BUG_ON(!mmu_is_nested(vcpu));
4149
4150         /* NPT walks are always user-walks */
4151         access |= PFERR_USER_MASK;
4152         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4153
4154         return t_gpa;
4155 }
4156
4157 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4158                               struct x86_exception *exception)
4159 {
4160         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4161         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4162 }
4163
4164  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4165                                 struct x86_exception *exception)
4166 {
4167         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4168         access |= PFERR_FETCH_MASK;
4169         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4170 }
4171
4172 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4173                                struct x86_exception *exception)
4174 {
4175         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4176         access |= PFERR_WRITE_MASK;
4177         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4178 }
4179
4180 /* uses this to access any guest's mapped memory without checking CPL */
4181 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4182                                 struct x86_exception *exception)
4183 {
4184         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4185 }
4186
4187 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4188                                       struct kvm_vcpu *vcpu, u32 access,
4189                                       struct x86_exception *exception)
4190 {
4191         void *data = val;
4192         int r = X86EMUL_CONTINUE;
4193
4194         while (bytes) {
4195                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4196                                                             exception);
4197                 unsigned offset = addr & (PAGE_SIZE-1);
4198                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4199                 int ret;
4200
4201                 if (gpa == UNMAPPED_GVA)
4202                         return X86EMUL_PROPAGATE_FAULT;
4203                 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4204                                                offset, toread);
4205                 if (ret < 0) {
4206                         r = X86EMUL_IO_NEEDED;
4207                         goto out;
4208                 }
4209
4210                 bytes -= toread;
4211                 data += toread;
4212                 addr += toread;
4213         }
4214 out:
4215         return r;
4216 }
4217
4218 /* used for instruction fetching */
4219 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4220                                 gva_t addr, void *val, unsigned int bytes,
4221                                 struct x86_exception *exception)
4222 {
4223         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4224         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4225         unsigned offset;
4226         int ret;
4227
4228         /* Inline kvm_read_guest_virt_helper for speed.  */
4229         gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4230                                                     exception);
4231         if (unlikely(gpa == UNMAPPED_GVA))
4232                 return X86EMUL_PROPAGATE_FAULT;
4233
4234         offset = addr & (PAGE_SIZE-1);
4235         if (WARN_ON(offset + bytes > PAGE_SIZE))
4236                 bytes = (unsigned)PAGE_SIZE - offset;
4237         ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4238                                        offset, bytes);
4239         if (unlikely(ret < 0))
4240                 return X86EMUL_IO_NEEDED;
4241
4242         return X86EMUL_CONTINUE;
4243 }
4244
4245 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4246                                gva_t addr, void *val, unsigned int bytes,
4247                                struct x86_exception *exception)
4248 {
4249         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4250         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4251
4252         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4253                                           exception);
4254 }
4255 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4256
4257 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4258                                       gva_t addr, void *val, unsigned int bytes,
4259                                       struct x86_exception *exception)
4260 {
4261         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4262         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4263 }
4264
4265 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4266                 unsigned long addr, void *val, unsigned int bytes)
4267 {
4268         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4269         int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4270
4271         return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4272 }
4273
4274 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4275                                        gva_t addr, void *val,
4276                                        unsigned int bytes,
4277                                        struct x86_exception *exception)
4278 {
4279         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4280         void *data = val;
4281         int r = X86EMUL_CONTINUE;
4282
4283         while (bytes) {
4284                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4285                                                              PFERR_WRITE_MASK,
4286                                                              exception);
4287                 unsigned offset = addr & (PAGE_SIZE-1);
4288                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4289                 int ret;
4290
4291                 if (gpa == UNMAPPED_GVA)
4292                         return X86EMUL_PROPAGATE_FAULT;
4293                 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4294                 if (ret < 0) {
4295                         r = X86EMUL_IO_NEEDED;
4296                         goto out;
4297                 }
4298
4299                 bytes -= towrite;
4300                 data += towrite;
4301                 addr += towrite;
4302         }
4303 out:
4304         return r;
4305 }
4306 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4307
4308 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4309                                 gpa_t *gpa, struct x86_exception *exception,
4310                                 bool write)
4311 {
4312         u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4313                 | (write ? PFERR_WRITE_MASK : 0);
4314
4315         if (vcpu_match_mmio_gva(vcpu, gva)
4316             && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4317                                  vcpu->arch.access, access)) {
4318                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4319                                         (gva & (PAGE_SIZE - 1));
4320                 trace_vcpu_match_mmio(gva, *gpa, write, false);
4321                 return 1;
4322         }
4323
4324         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4325
4326         if (*gpa == UNMAPPED_GVA)
4327                 return -1;
4328
4329         /* For APIC access vmexit */
4330         if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4331                 return 1;
4332
4333         if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4334                 trace_vcpu_match_mmio(gva, *gpa, write, true);
4335                 return 1;
4336         }
4337
4338         return 0;
4339 }
4340
4341 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4342                         const void *val, int bytes)
4343 {
4344         int ret;
4345
4346         ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4347         if (ret < 0)
4348                 return 0;
4349         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4350         return 1;
4351 }
4352
4353 struct read_write_emulator_ops {
4354         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4355                                   int bytes);
4356         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4357                                   void *val, int bytes);
4358         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4359                                int bytes, void *val);
4360         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4361                                     void *val, int bytes);
4362         bool write;
4363 };
4364
4365 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4366 {
4367         if (vcpu->mmio_read_completed) {
4368                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4369                                vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4370                 vcpu->mmio_read_completed = 0;
4371                 return 1;
4372         }
4373
4374         return 0;
4375 }
4376
4377 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4378                         void *val, int bytes)
4379 {
4380         return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4381 }
4382
4383 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4384                          void *val, int bytes)
4385 {
4386         return emulator_write_phys(vcpu, gpa, val, bytes);
4387 }
4388
4389 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4390 {
4391         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4392         return vcpu_mmio_write(vcpu, gpa, bytes, val);
4393 }
4394
4395 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4396                           void *val, int bytes)
4397 {
4398         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4399         return X86EMUL_IO_NEEDED;
4400 }
4401
4402 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4403                            void *val, int bytes)
4404 {
4405         struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4406
4407         memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4408         return X86EMUL_CONTINUE;
4409 }
4410
4411 static const struct read_write_emulator_ops read_emultor = {
4412         .read_write_prepare = read_prepare,
4413         .read_write_emulate = read_emulate,
4414         .read_write_mmio = vcpu_mmio_read,
4415         .read_write_exit_mmio = read_exit_mmio,
4416 };
4417
4418 static const struct read_write_emulator_ops write_emultor = {
4419         .read_write_emulate = write_emulate,
4420         .read_write_mmio = write_mmio,
4421         .read_write_exit_mmio = write_exit_mmio,
4422         .write = true,
4423 };
4424
4425 static int emulator_read_write_onepage(unsigned long addr, void *val,
4426                                        unsigned int bytes,
4427                                        struct x86_exception *exception,
4428                                        struct kvm_vcpu *vcpu,
4429                                        const struct read_write_emulator_ops *ops)
4430 {
4431         gpa_t gpa;
4432         int handled, ret;
4433         bool write = ops->write;
4434         struct kvm_mmio_fragment *frag;
4435
4436         ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4437
4438         if (ret < 0)
4439                 return X86EMUL_PROPAGATE_FAULT;
4440
4441         /* For APIC access vmexit */
4442         if (ret)
4443                 goto mmio;
4444
4445         if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4446                 return X86EMUL_CONTINUE;
4447
4448 mmio:
4449         /*
4450          * Is this MMIO handled locally?
4451          */
4452         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4453         if (handled == bytes)
4454                 return X86EMUL_CONTINUE;
4455
4456         gpa += handled;
4457         bytes -= handled;
4458         val += handled;
4459
4460         WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4461         frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4462         frag->gpa = gpa;
4463         frag->data = val;
4464         frag->len = bytes;
4465         return X86EMUL_CONTINUE;
4466 }
4467
4468 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4469                         unsigned long addr,
4470                         void *val, unsigned int bytes,
4471                         struct x86_exception *exception,
4472                         const struct read_write_emulator_ops *ops)
4473 {
4474         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4475         gpa_t gpa;
4476         int rc;
4477
4478         if (ops->read_write_prepare &&
4479                   ops->read_write_prepare(vcpu, val, bytes))
4480                 return X86EMUL_CONTINUE;
4481
4482         vcpu->mmio_nr_fragments = 0;
4483
4484         /* Crossing a page boundary? */
4485         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4486                 int now;
4487
4488                 now = -addr & ~PAGE_MASK;
4489                 rc = emulator_read_write_onepage(addr, val, now, exception,
4490                                                  vcpu, ops);
4491
4492                 if (rc != X86EMUL_CONTINUE)
4493                         return rc;
4494                 addr += now;
4495                 if (ctxt->mode != X86EMUL_MODE_PROT64)
4496                         addr = (u32)addr;
4497                 val += now;
4498                 bytes -= now;
4499         }
4500
4501         rc = emulator_read_write_onepage(addr, val, bytes, exception,
4502                                          vcpu, ops);
4503         if (rc != X86EMUL_CONTINUE)
4504                 return rc;
4505
4506         if (!vcpu->mmio_nr_fragments)
4507                 return rc;
4508
4509         gpa = vcpu->mmio_fragments[0].gpa;
4510
4511         vcpu->mmio_needed = 1;
4512         vcpu->mmio_cur_fragment = 0;
4513
4514         vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4515         vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4516         vcpu->run->exit_reason = KVM_EXIT_MMIO;
4517         vcpu->run->mmio.phys_addr = gpa;
4518
4519         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4520 }
4521
4522 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4523                                   unsigned long addr,
4524                                   void *val,
4525                                   unsigned int bytes,
4526                                   struct x86_exception *exception)
4527 {
4528         return emulator_read_write(ctxt, addr, val, bytes,
4529                                    exception, &read_emultor);
4530 }
4531
4532 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4533                             unsigned long addr,
4534                             const void *val,
4535                             unsigned int bytes,
4536                             struct x86_exception *exception)
4537 {
4538         return emulator_read_write(ctxt, addr, (void *)val, bytes,
4539                                    exception, &write_emultor);
4540 }
4541
4542 #define CMPXCHG_TYPE(t, ptr, old, new) \
4543         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4544
4545 #ifdef CONFIG_X86_64
4546 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4547 #else
4548 #  define CMPXCHG64(ptr, old, new) \
4549         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4550 #endif
4551
4552 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4553                                      unsigned long addr,
4554                                      const void *old,
4555                                      const void *new,
4556                                      unsigned int bytes,
4557                                      struct x86_exception *exception)
4558 {
4559         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4560         gpa_t gpa;
4561         struct page *page;
4562         char *kaddr;
4563         bool exchanged;
4564
4565         /* guests cmpxchg8b have to be emulated atomically */
4566         if (bytes > 8 || (bytes & (bytes - 1)))
4567                 goto emul_write;
4568
4569         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4570
4571         if (gpa == UNMAPPED_GVA ||
4572             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4573                 goto emul_write;
4574
4575         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4576                 goto emul_write;
4577
4578         page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4579         if (is_error_page(page))
4580                 goto emul_write;
4581
4582         kaddr = kmap_atomic(page);
4583         kaddr += offset_in_page(gpa);
4584         switch (bytes) {
4585         case 1:
4586                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4587                 break;
4588         case 2:
4589                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4590                 break;
4591         case 4:
4592                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4593                 break;
4594         case 8:
4595                 exchanged = CMPXCHG64(kaddr, old, new);
4596                 break;
4597         default:
4598                 BUG();
4599         }
4600         kunmap_atomic(kaddr);
4601         kvm_release_page_dirty(page);
4602
4603         if (!exchanged)
4604                 return X86EMUL_CMPXCHG_FAILED;
4605
4606         kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4607         kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4608
4609         return X86EMUL_CONTINUE;
4610
4611 emul_write:
4612         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4613
4614         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4615 }
4616
4617 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4618 {
4619         /* TODO: String I/O for in kernel device */
4620         int r;
4621
4622         if (vcpu->arch.pio.in)
4623                 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4624                                     vcpu->arch.pio.size, pd);
4625         else
4626                 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4627                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4628                                      pd);
4629         return r;
4630 }
4631
4632 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4633                                unsigned short port, void *val,
4634                                unsigned int count, bool in)
4635 {
4636         vcpu->arch.pio.port = port;
4637         vcpu->arch.pio.in = in;
4638         vcpu->arch.pio.count  = count;
4639         vcpu->arch.pio.size = size;
4640
4641         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4642                 vcpu->arch.pio.count = 0;
4643                 return 1;
4644         }
4645
4646         vcpu->run->exit_reason = KVM_EXIT_IO;
4647         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4648         vcpu->run->io.size = size;
4649         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4650         vcpu->run->io.count = count;
4651         vcpu->run->io.port = port;
4652
4653         return 0;
4654 }
4655
4656 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4657                                     int size, unsigned short port, void *val,
4658                                     unsigned int count)
4659 {
4660         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4661         int ret;
4662
4663         if (vcpu->arch.pio.count)
4664                 goto data_avail;
4665
4666         ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4667         if (ret) {
4668 data_avail:
4669                 memcpy(val, vcpu->arch.pio_data, size * count);
4670                 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4671                 vcpu->arch.pio.count = 0;
4672                 return 1;
4673         }
4674
4675         return 0;
4676 }
4677
4678 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4679                                      int size, unsigned short port,
4680                                      const void *val, unsigned int count)
4681 {
4682         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4683
4684         memcpy(vcpu->arch.pio_data, val, size * count);
4685         trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4686         return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4687 }
4688
4689 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4690 {
4691         return kvm_x86_ops->get_segment_base(vcpu, seg);
4692 }
4693
4694 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4695 {
4696         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4697 }
4698
4699 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4700 {
4701         if (!need_emulate_wbinvd(vcpu))
4702                 return X86EMUL_CONTINUE;
4703
4704         if (kvm_x86_ops->has_wbinvd_exit()) {
4705                 int cpu = get_cpu();
4706
4707                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4708                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4709                                 wbinvd_ipi, NULL, 1);
4710                 put_cpu();
4711                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4712         } else
4713                 wbinvd();
4714         return X86EMUL_CONTINUE;
4715 }
4716
4717 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4718 {
4719         kvm_x86_ops->skip_emulated_instruction(vcpu);
4720         return kvm_emulate_wbinvd_noskip(vcpu);
4721 }
4722 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4723
4724
4725
4726 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4727 {
4728         kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4729 }
4730
4731 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4732                            unsigned long *dest)
4733 {
4734         return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4735 }
4736
4737 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4738                            unsigned long value)
4739 {
4740
4741         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4742 }
4743
4744 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4745 {
4746         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4747 }
4748
4749 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4750 {
4751         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4752         unsigned long value;
4753
4754         switch (cr) {
4755         case 0:
4756                 value = kvm_read_cr0(vcpu);
4757                 break;
4758         case 2:
4759                 value = vcpu->arch.cr2;
4760                 break;
4761         case 3:
4762                 value = kvm_read_cr3(vcpu);
4763                 break;
4764         case 4:
4765                 value = kvm_read_cr4(vcpu);
4766                 break;
4767         case 8:
4768                 value = kvm_get_cr8(vcpu);
4769                 break;
4770         default:
4771                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4772                 return 0;
4773         }
4774
4775         return value;
4776 }
4777
4778 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4779 {
4780         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4781         int res = 0;
4782
4783         switch (cr) {
4784         case 0:
4785                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4786                 break;
4787         case 2:
4788                 vcpu->arch.cr2 = val;
4789                 break;
4790         case 3:
4791                 res = kvm_set_cr3(vcpu, val);
4792                 break;
4793         case 4:
4794                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4795                 break;
4796         case 8:
4797                 res = kvm_set_cr8(vcpu, val);
4798                 break;
4799         default:
4800                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4801                 res = -1;
4802         }
4803
4804         return res;
4805 }
4806
4807 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4808 {
4809         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4810 }
4811
4812 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4813 {
4814         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4815 }
4816
4817 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4818 {
4819         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4820 }
4821
4822 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4823 {
4824         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4825 }
4826
4827 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4828 {
4829         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4830 }
4831
4832 static unsigned long emulator_get_cached_segment_base(
4833         struct x86_emulate_ctxt *ctxt, int seg)
4834 {
4835         return get_segment_base(emul_to_vcpu(ctxt), seg);
4836 }
4837
4838 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4839                                  struct desc_struct *desc, u32 *base3,
4840                                  int seg)
4841 {
4842         struct kvm_segment var;
4843
4844         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4845         *selector = var.selector;
4846
4847         if (var.unusable) {
4848                 memset(desc, 0, sizeof(*desc));
4849                 return false;
4850         }
4851
4852         if (var.g)
4853                 var.limit >>= 12;
4854         set_desc_limit(desc, var.limit);
4855         set_desc_base(desc, (unsigned long)var.base);
4856 #ifdef CONFIG_X86_64
4857         if (base3)
4858                 *base3 = var.base >> 32;
4859 #endif
4860         desc->type = var.type;
4861         desc->s = var.s;
4862         desc->dpl = var.dpl;
4863         desc->p = var.present;
4864         desc->avl = var.avl;
4865         desc->l = var.l;
4866         desc->d = var.db;
4867         desc->g = var.g;
4868
4869         return true;
4870 }
4871
4872 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4873                                  struct desc_struct *desc, u32 base3,
4874                                  int seg)
4875 {
4876         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4877         struct kvm_segment var;
4878
4879         var.selector = selector;
4880         var.base = get_desc_base(desc);
4881 #ifdef CONFIG_X86_64
4882         var.base |= ((u64)base3) << 32;
4883 #endif
4884         var.limit = get_desc_limit(desc);
4885         if (desc->g)
4886                 var.limit = (var.limit << 12) | 0xfff;
4887         var.type = desc->type;
4888         var.dpl = desc->dpl;
4889         var.db = desc->d;
4890         var.s = desc->s;
4891         var.l = desc->l;
4892         var.g = desc->g;
4893         var.avl = desc->avl;
4894         var.present = desc->p;
4895         var.unusable = !var.present;
4896         var.padding = 0;
4897
4898         kvm_set_segment(vcpu, &var, seg);
4899         return;
4900 }
4901
4902 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4903                             u32 msr_index, u64 *pdata)
4904 {
4905         struct msr_data msr;
4906         int r;
4907
4908         msr.index = msr_index;
4909         msr.host_initiated = false;
4910         r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4911         if (r)
4912                 return r;
4913
4914         *pdata = msr.data;
4915         return 0;
4916 }
4917
4918 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4919                             u32 msr_index, u64 data)
4920 {
4921         struct msr_data msr;
4922
4923         msr.data = data;
4924         msr.index = msr_index;
4925         msr.host_initiated = false;
4926         return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4927 }
4928
4929 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4930 {
4931         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4932
4933         return vcpu->arch.smbase;
4934 }
4935
4936 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4937 {
4938         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4939
4940         vcpu->arch.smbase = smbase;
4941 }
4942
4943 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4944                               u32 pmc)
4945 {
4946         return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4947 }
4948
4949 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4950                              u32 pmc, u64 *pdata)
4951 {
4952         return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4953 }
4954
4955 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4956 {
4957         emul_to_vcpu(ctxt)->arch.halt_request = 1;
4958 }
4959
4960 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4961 {
4962         preempt_disable();
4963         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4964         /*
4965          * CR0.TS may reference the host fpu state, not the guest fpu state,
4966          * so it may be clear at this point.
4967          */
4968         clts();
4969 }
4970
4971 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4972 {
4973         preempt_enable();
4974 }
4975
4976 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4977                               struct x86_instruction_info *info,
4978                               enum x86_intercept_stage stage)
4979 {
4980         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4981 }
4982
4983 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4984                                u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4985 {
4986         kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4987 }
4988
4989 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4990 {
4991         return kvm_register_read(emul_to_vcpu(ctxt), reg);
4992 }
4993
4994 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4995 {
4996         kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4997 }
4998
4999 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5000 {
5001         kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5002 }
5003
5004 static const struct x86_emulate_ops emulate_ops = {
5005         .read_gpr            = emulator_read_gpr,
5006         .write_gpr           = emulator_write_gpr,
5007         .read_std            = kvm_read_guest_virt_system,
5008         .write_std           = kvm_write_guest_virt_system,
5009         .read_phys           = kvm_read_guest_phys_system,
5010         .fetch               = kvm_fetch_guest_virt,
5011         .read_emulated       = emulator_read_emulated,
5012         .write_emulated      = emulator_write_emulated,
5013         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
5014         .invlpg              = emulator_invlpg,
5015         .pio_in_emulated     = emulator_pio_in_emulated,
5016         .pio_out_emulated    = emulator_pio_out_emulated,
5017         .get_segment         = emulator_get_segment,
5018         .set_segment         = emulator_set_segment,
5019         .get_cached_segment_base = emulator_get_cached_segment_base,
5020         .get_gdt             = emulator_get_gdt,
5021         .get_idt             = emulator_get_idt,
5022         .set_gdt             = emulator_set_gdt,
5023         .set_idt             = emulator_set_idt,
5024         .get_cr              = emulator_get_cr,
5025         .set_cr              = emulator_set_cr,
5026         .cpl                 = emulator_get_cpl,
5027         .get_dr              = emulator_get_dr,
5028         .set_dr              = emulator_set_dr,
5029         .get_smbase          = emulator_get_smbase,
5030         .set_smbase          = emulator_set_smbase,
5031         .set_msr             = emulator_set_msr,
5032         .get_msr             = emulator_get_msr,
5033         .check_pmc           = emulator_check_pmc,
5034         .read_pmc            = emulator_read_pmc,
5035         .halt                = emulator_halt,
5036         .wbinvd              = emulator_wbinvd,
5037         .fix_hypercall       = emulator_fix_hypercall,
5038         .get_fpu             = emulator_get_fpu,
5039         .put_fpu             = emulator_put_fpu,
5040         .intercept           = emulator_intercept,
5041         .get_cpuid           = emulator_get_cpuid,
5042         .set_nmi_mask        = emulator_set_nmi_mask,
5043 };
5044
5045 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5046 {
5047         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5048         /*
5049          * an sti; sti; sequence only disable interrupts for the first
5050          * instruction. So, if the last instruction, be it emulated or
5051          * not, left the system with the INT_STI flag enabled, it
5052          * means that the last instruction is an sti. We should not
5053          * leave the flag on in this case. The same goes for mov ss
5054          */
5055         if (int_shadow & mask)
5056                 mask = 0;
5057         if (unlikely(int_shadow || mask)) {
5058                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5059                 if (!mask)
5060                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5061         }
5062 }
5063
5064 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5065 {
5066         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5067         if (ctxt->exception.vector == PF_VECTOR)
5068                 return kvm_propagate_fault(vcpu, &ctxt->exception);
5069
5070         if (ctxt->exception.error_code_valid)
5071                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5072                                       ctxt->exception.error_code);
5073         else
5074                 kvm_queue_exception(vcpu, ctxt->exception.vector);
5075         return false;
5076 }
5077
5078 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5079 {
5080         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5081         int cs_db, cs_l;
5082
5083         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5084
5085         ctxt->eflags = kvm_get_rflags(vcpu);
5086         ctxt->eip = kvm_rip_read(vcpu);
5087         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
5088                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
5089                      (cs_l && is_long_mode(vcpu))       ? X86EMUL_MODE_PROT64 :
5090                      cs_db                              ? X86EMUL_MODE_PROT32 :
5091                                                           X86EMUL_MODE_PROT16;
5092         BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5093         BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5094         BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5095         ctxt->emul_flags = vcpu->arch.hflags;
5096
5097         init_decode_cache(ctxt);
5098         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5099 }
5100
5101 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5102 {
5103         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5104         int ret;
5105
5106         init_emulate_ctxt(vcpu);
5107
5108         ctxt->op_bytes = 2;
5109         ctxt->ad_bytes = 2;
5110         ctxt->_eip = ctxt->eip + inc_eip;
5111         ret = emulate_int_real(ctxt, irq);
5112
5113         if (ret != X86EMUL_CONTINUE)
5114                 return EMULATE_FAIL;
5115
5116         ctxt->eip = ctxt->_eip;
5117         kvm_rip_write(vcpu, ctxt->eip);
5118         kvm_set_rflags(vcpu, ctxt->eflags);
5119
5120         if (irq == NMI_VECTOR)
5121                 vcpu->arch.nmi_pending = 0;
5122         else
5123                 vcpu->arch.interrupt.pending = false;
5124
5125         return EMULATE_DONE;
5126 }
5127 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5128
5129 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5130 {
5131         int r = EMULATE_DONE;
5132
5133         ++vcpu->stat.insn_emulation_fail;
5134         trace_kvm_emulate_insn_failed(vcpu);
5135         if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5136                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5137                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5138                 vcpu->run->internal.ndata = 0;
5139                 r = EMULATE_FAIL;
5140         }
5141         kvm_queue_exception(vcpu, UD_VECTOR);
5142
5143         return r;
5144 }
5145
5146 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5147                                   bool write_fault_to_shadow_pgtable,
5148                                   int emulation_type)
5149 {
5150         gpa_t gpa = cr2;
5151         pfn_t pfn;
5152
5153         if (emulation_type & EMULTYPE_NO_REEXECUTE)
5154                 return false;
5155
5156         if (!vcpu->arch.mmu.direct_map) {
5157                 /*
5158                  * Write permission should be allowed since only
5159                  * write access need to be emulated.
5160                  */
5161                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5162
5163                 /*
5164                  * If the mapping is invalid in guest, let cpu retry
5165                  * it to generate fault.
5166                  */
5167                 if (gpa == UNMAPPED_GVA)
5168                         return true;
5169         }
5170
5171         /*
5172          * Do not retry the unhandleable instruction if it faults on the
5173          * readonly host memory, otherwise it will goto a infinite loop:
5174          * retry instruction -> write #PF -> emulation fail -> retry
5175          * instruction -> ...
5176          */
5177         pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5178
5179         /*
5180          * If the instruction failed on the error pfn, it can not be fixed,
5181          * report the error to userspace.
5182          */
5183         if (is_error_noslot_pfn(pfn))
5184                 return false;
5185
5186         kvm_release_pfn_clean(pfn);
5187
5188         /* The instructions are well-emulated on direct mmu. */
5189         if (vcpu->arch.mmu.direct_map) {
5190                 unsigned int indirect_shadow_pages;
5191
5192                 spin_lock(&vcpu->kvm->mmu_lock);
5193                 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5194                 spin_unlock(&vcpu->kvm->mmu_lock);
5195
5196                 if (indirect_shadow_pages)
5197                         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5198
5199                 return true;
5200         }
5201
5202         /*
5203          * if emulation was due to access to shadowed page table
5204          * and it failed try to unshadow page and re-enter the
5205          * guest to let CPU execute the instruction.
5206          */
5207         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5208
5209         /*
5210          * If the access faults on its page table, it can not
5211          * be fixed by unprotecting shadow page and it should
5212          * be reported to userspace.
5213          */
5214         return !write_fault_to_shadow_pgtable;
5215 }
5216
5217 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5218                               unsigned long cr2,  int emulation_type)
5219 {
5220         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5221         unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5222
5223         last_retry_eip = vcpu->arch.last_retry_eip;
5224         last_retry_addr = vcpu->arch.last_retry_addr;
5225
5226         /*
5227          * If the emulation is caused by #PF and it is non-page_table
5228          * writing instruction, it means the VM-EXIT is caused by shadow
5229          * page protected, we can zap the shadow page and retry this
5230          * instruction directly.
5231          *
5232          * Note: if the guest uses a non-page-table modifying instruction
5233          * on the PDE that points to the instruction, then we will unmap
5234          * the instruction and go to an infinite loop. So, we cache the
5235          * last retried eip and the last fault address, if we meet the eip
5236          * and the address again, we can break out of the potential infinite
5237          * loop.
5238          */
5239         vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5240
5241         if (!(emulation_type & EMULTYPE_RETRY))
5242                 return false;
5243
5244         if (x86_page_table_writing_insn(ctxt))
5245                 return false;
5246
5247         if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5248                 return false;
5249
5250         vcpu->arch.last_retry_eip = ctxt->eip;
5251         vcpu->arch.last_retry_addr = cr2;
5252
5253         if (!vcpu->arch.mmu.direct_map)
5254                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5255
5256         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5257
5258         return true;
5259 }
5260
5261 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5262 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5263
5264 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5265 {
5266         if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5267                 /* This is a good place to trace that we are exiting SMM.  */
5268                 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5269
5270                 if (unlikely(vcpu->arch.smi_pending)) {
5271                         kvm_make_request(KVM_REQ_SMI, vcpu);
5272                         vcpu->arch.smi_pending = 0;
5273                 } else {
5274                         /* Process a latched INIT, if any.  */
5275                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5276                 }
5277         }
5278
5279         kvm_mmu_reset_context(vcpu);
5280 }
5281
5282 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5283 {
5284         unsigned changed = vcpu->arch.hflags ^ emul_flags;
5285
5286         vcpu->arch.hflags = emul_flags;
5287
5288         if (changed & HF_SMM_MASK)
5289                 kvm_smm_changed(vcpu);
5290 }
5291
5292 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5293                                 unsigned long *db)
5294 {
5295         u32 dr6 = 0;
5296         int i;
5297         u32 enable, rwlen;
5298
5299         enable = dr7;
5300         rwlen = dr7 >> 16;
5301         for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5302                 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5303                         dr6 |= (1 << i);
5304         return dr6;
5305 }
5306
5307 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5308 {
5309         struct kvm_run *kvm_run = vcpu->run;
5310
5311         /*
5312          * rflags is the old, "raw" value of the flags.  The new value has
5313          * not been saved yet.
5314          *
5315          * This is correct even for TF set by the guest, because "the
5316          * processor will not generate this exception after the instruction
5317          * that sets the TF flag".
5318          */
5319         if (unlikely(rflags & X86_EFLAGS_TF)) {
5320                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5321                         kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5322                                                   DR6_RTM;
5323                         kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5324                         kvm_run->debug.arch.exception = DB_VECTOR;
5325                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5326                         *r = EMULATE_USER_EXIT;
5327                 } else {
5328                         vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5329                         /*
5330                          * "Certain debug exceptions may clear bit 0-3.  The
5331                          * remaining contents of the DR6 register are never
5332                          * cleared by the processor".
5333                          */
5334                         vcpu->arch.dr6 &= ~15;
5335                         vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5336                         kvm_queue_exception(vcpu, DB_VECTOR);
5337                 }
5338         }
5339 }
5340
5341 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5342 {
5343         if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5344             (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5345                 struct kvm_run *kvm_run = vcpu->run;
5346                 unsigned long eip = kvm_get_linear_rip(vcpu);
5347                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5348                                            vcpu->arch.guest_debug_dr7,
5349                                            vcpu->arch.eff_db);
5350
5351                 if (dr6 != 0) {
5352                         kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5353                         kvm_run->debug.arch.pc = eip;
5354                         kvm_run->debug.arch.exception = DB_VECTOR;
5355                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
5356                         *r = EMULATE_USER_EXIT;
5357                         return true;
5358                 }
5359         }
5360
5361         if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5362             !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5363                 unsigned long eip = kvm_get_linear_rip(vcpu);
5364                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5365                                            vcpu->arch.dr7,
5366                                            vcpu->arch.db);
5367
5368                 if (dr6 != 0) {
5369                         vcpu->arch.dr6 &= ~15;
5370                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5371                         kvm_queue_exception(vcpu, DB_VECTOR);
5372                         *r = EMULATE_DONE;
5373                         return true;
5374                 }
5375         }
5376
5377         return false;
5378 }
5379
5380 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5381                             unsigned long cr2,
5382                             int emulation_type,
5383                             void *insn,
5384                             int insn_len)
5385 {
5386         int r;
5387         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5388         bool writeback = true;
5389         bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5390
5391         /*
5392          * Clear write_fault_to_shadow_pgtable here to ensure it is
5393          * never reused.
5394          */
5395         vcpu->arch.write_fault_to_shadow_pgtable = false;
5396         kvm_clear_exception_queue(vcpu);
5397
5398         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5399                 init_emulate_ctxt(vcpu);
5400
5401                 /*
5402                  * We will reenter on the same instruction since
5403                  * we do not set complete_userspace_io.  This does not
5404                  * handle watchpoints yet, those would be handled in
5405                  * the emulate_ops.
5406                  */
5407                 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5408                         return r;
5409
5410                 ctxt->interruptibility = 0;
5411                 ctxt->have_exception = false;
5412                 ctxt->exception.vector = -1;
5413                 ctxt->perm_ok = false;
5414
5415                 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5416
5417                 r = x86_decode_insn(ctxt, insn, insn_len);
5418
5419                 trace_kvm_emulate_insn_start(vcpu);
5420                 ++vcpu->stat.insn_emulation;
5421                 if (r != EMULATION_OK)  {
5422                         if (emulation_type & EMULTYPE_TRAP_UD)
5423                                 return EMULATE_FAIL;
5424                         if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5425                                                 emulation_type))
5426                                 return EMULATE_DONE;
5427                         if (emulation_type & EMULTYPE_SKIP)
5428                                 return EMULATE_FAIL;
5429                         return handle_emulation_failure(vcpu);
5430                 }
5431         }
5432
5433         if (emulation_type & EMULTYPE_SKIP) {
5434                 kvm_rip_write(vcpu, ctxt->_eip);
5435                 if (ctxt->eflags & X86_EFLAGS_RF)
5436                         kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5437                 return EMULATE_DONE;
5438         }
5439
5440         if (retry_instruction(ctxt, cr2, emulation_type))
5441                 return EMULATE_DONE;
5442
5443         /* this is needed for vmware backdoor interface to work since it
5444            changes registers values  during IO operation */
5445         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5446                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5447                 emulator_invalidate_register_cache(ctxt);
5448         }
5449
5450 restart:
5451         r = x86_emulate_insn(ctxt);
5452
5453         if (r == EMULATION_INTERCEPTED)
5454                 return EMULATE_DONE;
5455
5456         if (r == EMULATION_FAILED) {
5457                 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5458                                         emulation_type))
5459                         return EMULATE_DONE;
5460
5461                 return handle_emulation_failure(vcpu);
5462         }
5463
5464         if (ctxt->have_exception) {
5465                 r = EMULATE_DONE;
5466                 if (inject_emulated_exception(vcpu))
5467                         return r;
5468         } else if (vcpu->arch.pio.count) {
5469                 if (!vcpu->arch.pio.in) {
5470                         /* FIXME: return into emulator if single-stepping.  */
5471                         vcpu->arch.pio.count = 0;
5472                 } else {
5473                         writeback = false;
5474                         vcpu->arch.complete_userspace_io = complete_emulated_pio;
5475                 }
5476                 r = EMULATE_USER_EXIT;
5477         } else if (vcpu->mmio_needed) {
5478                 if (!vcpu->mmio_is_write)
5479                         writeback = false;
5480                 r = EMULATE_USER_EXIT;
5481                 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5482         } else if (r == EMULATION_RESTART)
5483                 goto restart;
5484         else
5485                 r = EMULATE_DONE;
5486
5487         if (writeback) {
5488                 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5489                 toggle_interruptibility(vcpu, ctxt->interruptibility);
5490                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5491                 if (vcpu->arch.hflags != ctxt->emul_flags)
5492                         kvm_set_hflags(vcpu, ctxt->emul_flags);
5493                 kvm_rip_write(vcpu, ctxt->eip);
5494                 if (r == EMULATE_DONE)
5495                         kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5496                 if (!ctxt->have_exception ||
5497                     exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5498                         __kvm_set_rflags(vcpu, ctxt->eflags);
5499
5500                 /*
5501                  * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5502                  * do nothing, and it will be requested again as soon as
5503                  * the shadow expires.  But we still need to check here,
5504                  * because POPF has no interrupt shadow.
5505                  */
5506                 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5507                         kvm_make_request(KVM_REQ_EVENT, vcpu);
5508         } else
5509                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5510
5511         return r;
5512 }
5513 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5514
5515 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5516 {
5517         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5518         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5519                                             size, port, &val, 1);
5520         /* do not return to emulator after return from userspace */
5521         vcpu->arch.pio.count = 0;
5522         return ret;
5523 }
5524 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5525
5526 static void tsc_bad(void *info)
5527 {
5528         __this_cpu_write(cpu_tsc_khz, 0);
5529 }
5530
5531 static void tsc_khz_changed(void *data)
5532 {
5533         struct cpufreq_freqs *freq = data;
5534         unsigned long khz = 0;
5535
5536         if (data)
5537                 khz = freq->new;
5538         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5539                 khz = cpufreq_quick_get(raw_smp_processor_id());
5540         if (!khz)
5541                 khz = tsc_khz;
5542         __this_cpu_write(cpu_tsc_khz, khz);
5543 }
5544
5545 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5546                                      void *data)
5547 {
5548         struct cpufreq_freqs *freq = data;
5549         struct kvm *kvm;
5550         struct kvm_vcpu *vcpu;
5551         int i, send_ipi = 0;
5552
5553         /*
5554          * We allow guests to temporarily run on slowing clocks,
5555          * provided we notify them after, or to run on accelerating
5556          * clocks, provided we notify them before.  Thus time never
5557          * goes backwards.
5558          *
5559          * However, we have a problem.  We can't atomically update
5560          * the frequency of a given CPU from this function; it is
5561          * merely a notifier, which can be called from any CPU.
5562          * Changing the TSC frequency at arbitrary points in time
5563          * requires a recomputation of local variables related to
5564          * the TSC for each VCPU.  We must flag these local variables
5565          * to be updated and be sure the update takes place with the
5566          * new frequency before any guests proceed.
5567          *
5568          * Unfortunately, the combination of hotplug CPU and frequency
5569          * change creates an intractable locking scenario; the order
5570          * of when these callouts happen is undefined with respect to
5571          * CPU hotplug, and they can race with each other.  As such,
5572          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5573          * undefined; you can actually have a CPU frequency change take
5574          * place in between the computation of X and the setting of the
5575          * variable.  To protect against this problem, all updates of
5576          * the per_cpu tsc_khz variable are done in an interrupt
5577          * protected IPI, and all callers wishing to update the value
5578          * must wait for a synchronous IPI to complete (which is trivial
5579          * if the caller is on the CPU already).  This establishes the
5580          * necessary total order on variable updates.
5581          *
5582          * Note that because a guest time update may take place
5583          * anytime after the setting of the VCPU's request bit, the
5584          * correct TSC value must be set before the request.  However,
5585          * to ensure the update actually makes it to any guest which
5586          * starts running in hardware virtualization between the set
5587          * and the acquisition of the spinlock, we must also ping the
5588          * CPU after setting the request bit.
5589          *
5590          */
5591
5592         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5593                 return 0;
5594         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5595                 return 0;
5596
5597         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5598
5599         spin_lock(&kvm_lock);
5600         list_for_each_entry(kvm, &vm_list, vm_list) {
5601                 kvm_for_each_vcpu(i, vcpu, kvm) {
5602                         if (vcpu->cpu != freq->cpu)
5603                                 continue;
5604                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5605                         if (vcpu->cpu != smp_processor_id())
5606                                 send_ipi = 1;
5607                 }
5608         }
5609         spin_unlock(&kvm_lock);
5610
5611         if (freq->old < freq->new && send_ipi) {
5612                 /*
5613                  * We upscale the frequency.  Must make the guest
5614                  * doesn't see old kvmclock values while running with
5615                  * the new frequency, otherwise we risk the guest sees
5616                  * time go backwards.
5617                  *
5618                  * In case we update the frequency for another cpu
5619                  * (which might be in guest context) send an interrupt
5620                  * to kick the cpu out of guest context.  Next time
5621                  * guest context is entered kvmclock will be updated,
5622                  * so the guest will not see stale values.
5623                  */
5624                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5625         }
5626         return 0;
5627 }
5628
5629 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5630         .notifier_call  = kvmclock_cpufreq_notifier
5631 };
5632
5633 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5634                                         unsigned long action, void *hcpu)
5635 {
5636         unsigned int cpu = (unsigned long)hcpu;
5637
5638         switch (action) {
5639                 case CPU_ONLINE:
5640                 case CPU_DOWN_FAILED:
5641                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5642                         break;
5643                 case CPU_DOWN_PREPARE:
5644                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
5645                         break;
5646         }
5647         return NOTIFY_OK;
5648 }
5649
5650 static struct notifier_block kvmclock_cpu_notifier_block = {
5651         .notifier_call  = kvmclock_cpu_notifier,
5652         .priority = -INT_MAX
5653 };
5654
5655 static void kvm_timer_init(void)
5656 {
5657         int cpu;
5658
5659         max_tsc_khz = tsc_khz;
5660
5661         cpu_notifier_register_begin();
5662         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5663 #ifdef CONFIG_CPU_FREQ
5664                 struct cpufreq_policy policy;
5665                 memset(&policy, 0, sizeof(policy));
5666                 cpu = get_cpu();
5667                 cpufreq_get_policy(&policy, cpu);
5668                 if (policy.cpuinfo.max_freq)
5669                         max_tsc_khz = policy.cpuinfo.max_freq;
5670                 put_cpu();
5671 #endif
5672                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5673                                           CPUFREQ_TRANSITION_NOTIFIER);
5674         }
5675         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5676         for_each_online_cpu(cpu)
5677                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5678
5679         __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5680         cpu_notifier_register_done();
5681
5682 }
5683
5684 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5685
5686 int kvm_is_in_guest(void)
5687 {
5688         return __this_cpu_read(current_vcpu) != NULL;
5689 }
5690
5691 static int kvm_is_user_mode(void)
5692 {
5693         int user_mode = 3;
5694
5695         if (__this_cpu_read(current_vcpu))
5696                 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5697
5698         return user_mode != 0;
5699 }
5700
5701 static unsigned long kvm_get_guest_ip(void)
5702 {
5703         unsigned long ip = 0;
5704
5705         if (__this_cpu_read(current_vcpu))
5706                 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5707
5708         return ip;
5709 }
5710
5711 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5712         .is_in_guest            = kvm_is_in_guest,
5713         .is_user_mode           = kvm_is_user_mode,
5714         .get_guest_ip           = kvm_get_guest_ip,
5715 };
5716
5717 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5718 {
5719         __this_cpu_write(current_vcpu, vcpu);
5720 }
5721 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5722
5723 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5724 {
5725         __this_cpu_write(current_vcpu, NULL);
5726 }
5727 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5728
5729 static void kvm_set_mmio_spte_mask(void)
5730 {
5731         u64 mask;
5732         int maxphyaddr = boot_cpu_data.x86_phys_bits;
5733
5734         /*
5735          * Set the reserved bits and the present bit of an paging-structure
5736          * entry to generate page fault with PFER.RSV = 1.
5737          */
5738          /* Mask the reserved physical address bits. */
5739         mask = rsvd_bits(maxphyaddr, 51);
5740
5741         /* Bit 62 is always reserved for 32bit host. */
5742         mask |= 0x3ull << 62;
5743
5744         /* Set the present bit. */
5745         mask |= 1ull;
5746
5747 #ifdef CONFIG_X86_64
5748         /*
5749          * If reserved bit is not supported, clear the present bit to disable
5750          * mmio page fault.
5751          */
5752         if (maxphyaddr == 52)
5753                 mask &= ~1ull;
5754 #endif
5755
5756         kvm_mmu_set_mmio_spte_mask(mask);
5757 }
5758
5759 #ifdef CONFIG_X86_64
5760 static void pvclock_gtod_update_fn(struct work_struct *work)
5761 {
5762         struct kvm *kvm;
5763
5764         struct kvm_vcpu *vcpu;
5765         int i;
5766
5767         spin_lock(&kvm_lock);
5768         list_for_each_entry(kvm, &vm_list, vm_list)
5769                 kvm_for_each_vcpu(i, vcpu, kvm)
5770                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5771         atomic_set(&kvm_guest_has_master_clock, 0);
5772         spin_unlock(&kvm_lock);
5773 }
5774
5775 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5776
5777 /*
5778  * Notification about pvclock gtod data update.
5779  */
5780 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5781                                void *priv)
5782 {
5783         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5784         struct timekeeper *tk = priv;
5785
5786         update_pvclock_gtod(tk);
5787
5788         /* disable master clock if host does not trust, or does not
5789          * use, TSC clocksource
5790          */
5791         if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5792             atomic_read(&kvm_guest_has_master_clock) != 0)
5793                 queue_work(system_long_wq, &pvclock_gtod_work);
5794
5795         return 0;
5796 }
5797
5798 static struct notifier_block pvclock_gtod_notifier = {
5799         .notifier_call = pvclock_gtod_notify,
5800 };
5801 #endif
5802
5803 int kvm_arch_init(void *opaque)
5804 {
5805         int r;
5806         struct kvm_x86_ops *ops = opaque;
5807
5808         if (kvm_x86_ops) {
5809                 printk(KERN_ERR "kvm: already loaded the other module\n");
5810                 r = -EEXIST;
5811                 goto out;
5812         }
5813
5814         if (!ops->cpu_has_kvm_support()) {
5815                 printk(KERN_ERR "kvm: no hardware support\n");
5816                 r = -EOPNOTSUPP;
5817                 goto out;
5818         }
5819         if (ops->disabled_by_bios()) {
5820                 printk(KERN_ERR "kvm: disabled by bios\n");
5821                 r = -EOPNOTSUPP;
5822                 goto out;
5823         }
5824
5825         r = -ENOMEM;
5826         shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5827         if (!shared_msrs) {
5828                 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5829                 goto out;
5830         }
5831
5832         r = kvm_mmu_module_init();
5833         if (r)
5834                 goto out_free_percpu;
5835
5836         kvm_set_mmio_spte_mask();
5837
5838         kvm_x86_ops = ops;
5839
5840         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5841                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
5842
5843         kvm_timer_init();
5844
5845         perf_register_guest_info_callbacks(&kvm_guest_cbs);
5846
5847         if (cpu_has_xsave)
5848                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5849
5850         kvm_lapic_init();
5851 #ifdef CONFIG_X86_64
5852         pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5853 #endif
5854
5855         return 0;
5856
5857 out_free_percpu:
5858         free_percpu(shared_msrs);
5859 out:
5860         return r;
5861 }
5862
5863 void kvm_arch_exit(void)
5864 {
5865         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5866
5867         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5868                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5869                                             CPUFREQ_TRANSITION_NOTIFIER);
5870         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5871 #ifdef CONFIG_X86_64
5872         pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5873 #endif
5874         kvm_x86_ops = NULL;
5875         kvm_mmu_module_exit();
5876         free_percpu(shared_msrs);
5877 }
5878
5879 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5880 {
5881         ++vcpu->stat.halt_exits;
5882         if (lapic_in_kernel(vcpu)) {
5883                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5884                 return 1;
5885         } else {
5886                 vcpu->run->exit_reason = KVM_EXIT_HLT;
5887                 return 0;
5888         }
5889 }
5890 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5891
5892 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5893 {
5894         kvm_x86_ops->skip_emulated_instruction(vcpu);
5895         return kvm_vcpu_halt(vcpu);
5896 }
5897 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5898
5899 /*
5900  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5901  *
5902  * @apicid - apicid of vcpu to be kicked.
5903  */
5904 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5905 {
5906         struct kvm_lapic_irq lapic_irq;
5907
5908         lapic_irq.shorthand = 0;
5909         lapic_irq.dest_mode = 0;
5910         lapic_irq.dest_id = apicid;
5911         lapic_irq.msi_redir_hint = false;
5912
5913         lapic_irq.delivery_mode = APIC_DM_REMRD;
5914         kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5915 }
5916
5917 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5918 {
5919         vcpu->arch.apicv_active = false;
5920         kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5921 }
5922
5923 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5924 {
5925         unsigned long nr, a0, a1, a2, a3, ret;
5926         int op_64_bit, r = 1;
5927
5928         kvm_x86_ops->skip_emulated_instruction(vcpu);
5929
5930         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5931                 return kvm_hv_hypercall(vcpu);
5932
5933         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5934         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5935         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5936         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5937         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5938
5939         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5940
5941         op_64_bit = is_64_bit_mode(vcpu);
5942         if (!op_64_bit) {
5943                 nr &= 0xFFFFFFFF;
5944                 a0 &= 0xFFFFFFFF;
5945                 a1 &= 0xFFFFFFFF;
5946                 a2 &= 0xFFFFFFFF;
5947                 a3 &= 0xFFFFFFFF;
5948         }
5949
5950         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5951                 ret = -KVM_EPERM;
5952                 goto out;
5953         }
5954
5955         switch (nr) {
5956         case KVM_HC_VAPIC_POLL_IRQ:
5957                 ret = 0;
5958                 break;
5959         case KVM_HC_KICK_CPU:
5960                 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5961                 ret = 0;
5962                 break;
5963         default:
5964                 ret = -KVM_ENOSYS;
5965                 break;
5966         }
5967 out:
5968         if (!op_64_bit)
5969                 ret = (u32)ret;
5970         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5971         ++vcpu->stat.hypercalls;
5972         return r;
5973 }
5974 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5975
5976 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5977 {
5978         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5979         char instruction[3];
5980         unsigned long rip = kvm_rip_read(vcpu);
5981
5982         kvm_x86_ops->patch_hypercall(vcpu, instruction);
5983
5984         return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5985 }
5986
5987 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5988 {
5989         return vcpu->run->request_interrupt_window &&
5990                 likely(!pic_in_kernel(vcpu->kvm));
5991 }
5992
5993 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5994 {
5995         struct kvm_run *kvm_run = vcpu->run;
5996
5997         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5998         kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
5999         kvm_run->cr8 = kvm_get_cr8(vcpu);
6000         kvm_run->apic_base = kvm_get_apic_base(vcpu);
6001         kvm_run->ready_for_interrupt_injection =
6002                 pic_in_kernel(vcpu->kvm) ||
6003                 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6004 }
6005
6006 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6007 {
6008         int max_irr, tpr;
6009
6010         if (!kvm_x86_ops->update_cr8_intercept)
6011                 return;
6012
6013         if (!vcpu->arch.apic)
6014                 return;
6015
6016         if (vcpu->arch.apicv_active)
6017                 return;
6018
6019         if (!vcpu->arch.apic->vapic_addr)
6020                 max_irr = kvm_lapic_find_highest_irr(vcpu);
6021         else
6022                 max_irr = -1;
6023
6024         if (max_irr != -1)
6025                 max_irr >>= 4;
6026
6027         tpr = kvm_lapic_get_cr8(vcpu);
6028
6029         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6030 }
6031
6032 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6033 {
6034         int r;
6035
6036         /* try to reinject previous events if any */
6037         if (vcpu->arch.exception.pending) {
6038                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6039                                         vcpu->arch.exception.has_error_code,
6040                                         vcpu->arch.exception.error_code);
6041
6042                 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6043                         __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6044                                              X86_EFLAGS_RF);
6045
6046                 if (vcpu->arch.exception.nr == DB_VECTOR &&
6047                     (vcpu->arch.dr7 & DR7_GD)) {
6048                         vcpu->arch.dr7 &= ~DR7_GD;
6049                         kvm_update_dr7(vcpu);
6050                 }
6051
6052                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6053                                           vcpu->arch.exception.has_error_code,
6054                                           vcpu->arch.exception.error_code,
6055                                           vcpu->arch.exception.reinject);
6056                 return 0;
6057         }
6058
6059         if (vcpu->arch.nmi_injected) {
6060                 kvm_x86_ops->set_nmi(vcpu);
6061                 return 0;
6062         }
6063
6064         if (vcpu->arch.interrupt.pending) {
6065                 kvm_x86_ops->set_irq(vcpu);
6066                 return 0;
6067         }
6068
6069         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6070                 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6071                 if (r != 0)
6072                         return r;
6073         }
6074
6075         /* try to inject new event if pending */
6076         if (vcpu->arch.nmi_pending) {
6077                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
6078                         --vcpu->arch.nmi_pending;
6079                         vcpu->arch.nmi_injected = true;
6080                         kvm_x86_ops->set_nmi(vcpu);
6081                 }
6082         } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6083                 /*
6084                  * Because interrupts can be injected asynchronously, we are
6085                  * calling check_nested_events again here to avoid a race condition.
6086                  * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6087                  * proposal and current concerns.  Perhaps we should be setting
6088                  * KVM_REQ_EVENT only on certain events and not unconditionally?
6089                  */
6090                 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6091                         r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6092                         if (r != 0)
6093                                 return r;
6094                 }
6095                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6096                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6097                                             false);
6098                         kvm_x86_ops->set_irq(vcpu);
6099                 }
6100         }
6101         return 0;
6102 }
6103
6104 static void process_nmi(struct kvm_vcpu *vcpu)
6105 {
6106         unsigned limit = 2;
6107
6108         /*
6109          * x86 is limited to one NMI running, and one NMI pending after it.
6110          * If an NMI is already in progress, limit further NMIs to just one.
6111          * Otherwise, allow two (and we'll inject the first one immediately).
6112          */
6113         if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6114                 limit = 1;
6115
6116         vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6117         vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6118         kvm_make_request(KVM_REQ_EVENT, vcpu);
6119 }
6120
6121 #define put_smstate(type, buf, offset, val)                       \
6122         *(type *)((buf) + (offset) - 0x7e00) = val
6123
6124 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6125 {
6126         u32 flags = 0;
6127         flags |= seg->g       << 23;
6128         flags |= seg->db      << 22;
6129         flags |= seg->l       << 21;
6130         flags |= seg->avl     << 20;
6131         flags |= seg->present << 15;
6132         flags |= seg->dpl     << 13;
6133         flags |= seg->s       << 12;
6134         flags |= seg->type    << 8;
6135         return flags;
6136 }
6137
6138 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6139 {
6140         struct kvm_segment seg;
6141         int offset;
6142
6143         kvm_get_segment(vcpu, &seg, n);
6144         put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6145
6146         if (n < 3)
6147                 offset = 0x7f84 + n * 12;
6148         else
6149                 offset = 0x7f2c + (n - 3) * 12;
6150
6151         put_smstate(u32, buf, offset + 8, seg.base);
6152         put_smstate(u32, buf, offset + 4, seg.limit);
6153         put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6154 }
6155
6156 #ifdef CONFIG_X86_64
6157 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6158 {
6159         struct kvm_segment seg;
6160         int offset;
6161         u16 flags;
6162
6163         kvm_get_segment(vcpu, &seg, n);
6164         offset = 0x7e00 + n * 16;
6165
6166         flags = process_smi_get_segment_flags(&seg) >> 8;
6167         put_smstate(u16, buf, offset, seg.selector);
6168         put_smstate(u16, buf, offset + 2, flags);
6169         put_smstate(u32, buf, offset + 4, seg.limit);
6170         put_smstate(u64, buf, offset + 8, seg.base);
6171 }
6172 #endif
6173
6174 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6175 {
6176         struct desc_ptr dt;
6177         struct kvm_segment seg;
6178         unsigned long val;
6179         int i;
6180
6181         put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6182         put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6183         put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6184         put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6185
6186         for (i = 0; i < 8; i++)
6187                 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6188
6189         kvm_get_dr(vcpu, 6, &val);
6190         put_smstate(u32, buf, 0x7fcc, (u32)val);
6191         kvm_get_dr(vcpu, 7, &val);
6192         put_smstate(u32, buf, 0x7fc8, (u32)val);
6193
6194         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6195         put_smstate(u32, buf, 0x7fc4, seg.selector);
6196         put_smstate(u32, buf, 0x7f64, seg.base);
6197         put_smstate(u32, buf, 0x7f60, seg.limit);
6198         put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6199
6200         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6201         put_smstate(u32, buf, 0x7fc0, seg.selector);
6202         put_smstate(u32, buf, 0x7f80, seg.base);
6203         put_smstate(u32, buf, 0x7f7c, seg.limit);
6204         put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6205
6206         kvm_x86_ops->get_gdt(vcpu, &dt);
6207         put_smstate(u32, buf, 0x7f74, dt.address);
6208         put_smstate(u32, buf, 0x7f70, dt.size);
6209
6210         kvm_x86_ops->get_idt(vcpu, &dt);
6211         put_smstate(u32, buf, 0x7f58, dt.address);
6212         put_smstate(u32, buf, 0x7f54, dt.size);
6213
6214         for (i = 0; i < 6; i++)
6215                 process_smi_save_seg_32(vcpu, buf, i);
6216
6217         put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6218
6219         /* revision id */
6220         put_smstate(u32, buf, 0x7efc, 0x00020000);
6221         put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6222 }
6223
6224 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6225 {
6226 #ifdef CONFIG_X86_64
6227         struct desc_ptr dt;
6228         struct kvm_segment seg;
6229         unsigned long val;
6230         int i;
6231
6232         for (i = 0; i < 16; i++)
6233                 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6234
6235         put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6236         put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6237
6238         kvm_get_dr(vcpu, 6, &val);
6239         put_smstate(u64, buf, 0x7f68, val);
6240         kvm_get_dr(vcpu, 7, &val);
6241         put_smstate(u64, buf, 0x7f60, val);
6242
6243         put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6244         put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6245         put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6246
6247         put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6248
6249         /* revision id */
6250         put_smstate(u32, buf, 0x7efc, 0x00020064);
6251
6252         put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6253
6254         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6255         put_smstate(u16, buf, 0x7e90, seg.selector);
6256         put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6257         put_smstate(u32, buf, 0x7e94, seg.limit);
6258         put_smstate(u64, buf, 0x7e98, seg.base);
6259
6260         kvm_x86_ops->get_idt(vcpu, &dt);
6261         put_smstate(u32, buf, 0x7e84, dt.size);
6262         put_smstate(u64, buf, 0x7e88, dt.address);
6263
6264         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6265         put_smstate(u16, buf, 0x7e70, seg.selector);
6266         put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6267         put_smstate(u32, buf, 0x7e74, seg.limit);
6268         put_smstate(u64, buf, 0x7e78, seg.base);
6269
6270         kvm_x86_ops->get_gdt(vcpu, &dt);
6271         put_smstate(u32, buf, 0x7e64, dt.size);
6272         put_smstate(u64, buf, 0x7e68, dt.address);
6273
6274         for (i = 0; i < 6; i++)
6275                 process_smi_save_seg_64(vcpu, buf, i);
6276 #else
6277         WARN_ON_ONCE(1);
6278 #endif
6279 }
6280
6281 static void process_smi(struct kvm_vcpu *vcpu)
6282 {
6283         struct kvm_segment cs, ds;
6284         struct desc_ptr dt;
6285         char buf[512];
6286         u32 cr0;
6287
6288         if (is_smm(vcpu)) {
6289                 vcpu->arch.smi_pending = true;
6290                 return;
6291         }
6292
6293         trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6294         vcpu->arch.hflags |= HF_SMM_MASK;
6295         memset(buf, 0, 512);
6296         if (guest_cpuid_has_longmode(vcpu))
6297                 process_smi_save_state_64(vcpu, buf);
6298         else
6299                 process_smi_save_state_32(vcpu, buf);
6300
6301         kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6302
6303         if (kvm_x86_ops->get_nmi_mask(vcpu))
6304                 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6305         else
6306                 kvm_x86_ops->set_nmi_mask(vcpu, true);
6307
6308         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6309         kvm_rip_write(vcpu, 0x8000);
6310
6311         cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6312         kvm_x86_ops->set_cr0(vcpu, cr0);
6313         vcpu->arch.cr0 = cr0;
6314
6315         kvm_x86_ops->set_cr4(vcpu, 0);
6316
6317         /* Undocumented: IDT limit is set to zero on entry to SMM.  */
6318         dt.address = dt.size = 0;
6319         kvm_x86_ops->set_idt(vcpu, &dt);
6320
6321         __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6322
6323         cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6324         cs.base = vcpu->arch.smbase;
6325
6326         ds.selector = 0;
6327         ds.base = 0;
6328
6329         cs.limit    = ds.limit = 0xffffffff;
6330         cs.type     = ds.type = 0x3;
6331         cs.dpl      = ds.dpl = 0;
6332         cs.db       = ds.db = 0;
6333         cs.s        = ds.s = 1;
6334         cs.l        = ds.l = 0;
6335         cs.g        = ds.g = 1;
6336         cs.avl      = ds.avl = 0;
6337         cs.present  = ds.present = 1;
6338         cs.unusable = ds.unusable = 0;
6339         cs.padding  = ds.padding = 0;
6340
6341         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6342         kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6343         kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6344         kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6345         kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6346         kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6347
6348         if (guest_cpuid_has_longmode(vcpu))
6349                 kvm_x86_ops->set_efer(vcpu, 0);
6350
6351         kvm_update_cpuid(vcpu);
6352         kvm_mmu_reset_context(vcpu);
6353 }
6354
6355 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6356 {
6357         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6358 }
6359
6360 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6361 {
6362         u64 eoi_exit_bitmap[4];
6363
6364         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6365                 return;
6366
6367         bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6368
6369         if (irqchip_split(vcpu->kvm))
6370                 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6371         else {
6372                 if (vcpu->arch.apicv_active)
6373                         kvm_x86_ops->sync_pir_to_irr(vcpu);
6374                 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6375         }
6376         bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6377                   vcpu_to_synic(vcpu)->vec_bitmap, 256);
6378         kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6379 }
6380
6381 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6382 {
6383         ++vcpu->stat.tlb_flush;
6384         kvm_x86_ops->tlb_flush(vcpu);
6385 }
6386
6387 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6388 {
6389         struct page *page = NULL;
6390
6391         if (!lapic_in_kernel(vcpu))
6392                 return;
6393
6394         if (!kvm_x86_ops->set_apic_access_page_addr)
6395                 return;
6396
6397         page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6398         if (is_error_page(page))
6399                 return;
6400         kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6401
6402         /*
6403          * Do not pin apic access page in memory, the MMU notifier
6404          * will call us again if it is migrated or swapped out.
6405          */
6406         put_page(page);
6407 }
6408 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6409
6410 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6411                                            unsigned long address)
6412 {
6413         /*
6414          * The physical address of apic access page is stored in the VMCS.
6415          * Update it when it becomes invalid.
6416          */
6417         if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6418                 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6419 }
6420
6421 /*
6422  * Returns 1 to let vcpu_run() continue the guest execution loop without
6423  * exiting to the userspace.  Otherwise, the value will be returned to the
6424  * userspace.
6425  */
6426 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6427 {
6428         int r;
6429         bool req_int_win =
6430                 dm_request_for_irq_injection(vcpu) &&
6431                 kvm_cpu_accept_dm_intr(vcpu);
6432
6433         bool req_immediate_exit = false;
6434
6435         if (vcpu->requests) {
6436                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6437                         kvm_mmu_unload(vcpu);
6438                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6439                         __kvm_migrate_timers(vcpu);
6440                 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6441                         kvm_gen_update_masterclock(vcpu->kvm);
6442                 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6443                         kvm_gen_kvmclock_update(vcpu);
6444                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6445                         r = kvm_guest_time_update(vcpu);
6446                         if (unlikely(r))
6447                                 goto out;
6448                 }
6449                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6450                         kvm_mmu_sync_roots(vcpu);
6451                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6452                         kvm_vcpu_flush_tlb(vcpu);
6453                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6454                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6455                         r = 0;
6456                         goto out;
6457                 }
6458                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6459                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6460                         r = 0;
6461                         goto out;
6462                 }
6463                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6464                         vcpu->fpu_active = 0;
6465                         kvm_x86_ops->fpu_deactivate(vcpu);
6466                 }
6467                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6468                         /* Page is swapped out. Do synthetic halt */
6469                         vcpu->arch.apf.halted = true;
6470                         r = 1;
6471                         goto out;
6472                 }
6473                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6474                         record_steal_time(vcpu);
6475                 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6476                         process_smi(vcpu);
6477                 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6478                         process_nmi(vcpu);
6479                 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6480                         kvm_pmu_handle_event(vcpu);
6481                 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6482                         kvm_pmu_deliver_pmi(vcpu);
6483                 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6484                         BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6485                         if (test_bit(vcpu->arch.pending_ioapic_eoi,
6486                                      vcpu->arch.ioapic_handled_vectors)) {
6487                                 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6488                                 vcpu->run->eoi.vector =
6489                                                 vcpu->arch.pending_ioapic_eoi;
6490                                 r = 0;
6491                                 goto out;
6492                         }
6493                 }
6494                 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6495                         vcpu_scan_ioapic(vcpu);
6496                 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6497                         kvm_vcpu_reload_apic_access_page(vcpu);
6498                 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6499                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6500                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6501                         r = 0;
6502                         goto out;
6503                 }
6504                 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6505                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6506                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6507                         r = 0;
6508                         goto out;
6509                 }
6510                 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6511                         vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6512                         vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6513                         r = 0;
6514                         goto out;
6515                 }
6516
6517                 /*
6518                  * KVM_REQ_HV_STIMER has to be processed after
6519                  * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6520                  * depend on the guest clock being up-to-date
6521                  */
6522                 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6523                         kvm_hv_process_stimers(vcpu);
6524         }
6525
6526         /*
6527          * KVM_REQ_EVENT is not set when posted interrupts are set by
6528          * VT-d hardware, so we have to update RVI unconditionally.
6529          */
6530         if (kvm_lapic_enabled(vcpu)) {
6531                 /*
6532                  * Update architecture specific hints for APIC
6533                  * virtual interrupt delivery.
6534                  */
6535                 if (vcpu->arch.apicv_active)
6536                         kvm_x86_ops->hwapic_irr_update(vcpu,
6537                                 kvm_lapic_find_highest_irr(vcpu));
6538         }
6539
6540         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6541                 kvm_apic_accept_events(vcpu);
6542                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6543                         r = 1;
6544                         goto out;
6545                 }
6546
6547                 if (inject_pending_event(vcpu, req_int_win) != 0)
6548                         req_immediate_exit = true;
6549                 /* enable NMI/IRQ window open exits if needed */
6550                 else if (vcpu->arch.nmi_pending)
6551                         kvm_x86_ops->enable_nmi_window(vcpu);
6552                 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6553                         kvm_x86_ops->enable_irq_window(vcpu);
6554
6555                 if (kvm_lapic_enabled(vcpu)) {
6556                         update_cr8_intercept(vcpu);
6557                         kvm_lapic_sync_to_vapic(vcpu);
6558                 }
6559         }
6560
6561         r = kvm_mmu_reload(vcpu);
6562         if (unlikely(r)) {
6563                 goto cancel_injection;
6564         }
6565
6566         preempt_disable();
6567
6568         kvm_x86_ops->prepare_guest_switch(vcpu);
6569         if (vcpu->fpu_active)
6570                 kvm_load_guest_fpu(vcpu);
6571         kvm_load_guest_xcr0(vcpu);
6572
6573         vcpu->mode = IN_GUEST_MODE;
6574
6575         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6576
6577         /* We should set ->mode before check ->requests,
6578          * see the comment in make_all_cpus_request.
6579          */
6580         smp_mb__after_srcu_read_unlock();
6581
6582         local_irq_disable();
6583
6584         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6585             || need_resched() || signal_pending(current)) {
6586                 vcpu->mode = OUTSIDE_GUEST_MODE;
6587                 smp_wmb();
6588                 local_irq_enable();
6589                 preempt_enable();
6590                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6591                 r = 1;
6592                 goto cancel_injection;
6593         }
6594
6595         if (req_immediate_exit)
6596                 smp_send_reschedule(vcpu->cpu);
6597
6598         trace_kvm_entry(vcpu->vcpu_id);
6599         wait_lapic_expire(vcpu);
6600         __kvm_guest_enter();
6601
6602         if (unlikely(vcpu->arch.switch_db_regs)) {
6603                 set_debugreg(0, 7);
6604                 set_debugreg(vcpu->arch.eff_db[0], 0);
6605                 set_debugreg(vcpu->arch.eff_db[1], 1);
6606                 set_debugreg(vcpu->arch.eff_db[2], 2);
6607                 set_debugreg(vcpu->arch.eff_db[3], 3);
6608                 set_debugreg(vcpu->arch.dr6, 6);
6609                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6610         }
6611
6612         kvm_x86_ops->run(vcpu);
6613
6614         /*
6615          * Do this here before restoring debug registers on the host.  And
6616          * since we do this before handling the vmexit, a DR access vmexit
6617          * can (a) read the correct value of the debug registers, (b) set
6618          * KVM_DEBUGREG_WONT_EXIT again.
6619          */
6620         if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6621                 int i;
6622
6623                 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6624                 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6625                 for (i = 0; i < KVM_NR_DB_REGS; i++)
6626                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6627         }
6628
6629         /*
6630          * If the guest has used debug registers, at least dr7
6631          * will be disabled while returning to the host.
6632          * If we don't have active breakpoints in the host, we don't
6633          * care about the messed up debug address registers. But if
6634          * we have some of them active, restore the old state.
6635          */
6636         if (hw_breakpoint_active())
6637                 hw_breakpoint_restore();
6638
6639         vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6640
6641         vcpu->mode = OUTSIDE_GUEST_MODE;
6642         smp_wmb();
6643
6644         /* Interrupt is enabled by handle_external_intr() */
6645         kvm_x86_ops->handle_external_intr(vcpu);
6646
6647         ++vcpu->stat.exits;
6648
6649         /*
6650          * We must have an instruction between local_irq_enable() and
6651          * kvm_guest_exit(), so the timer interrupt isn't delayed by
6652          * the interrupt shadow.  The stat.exits increment will do nicely.
6653          * But we need to prevent reordering, hence this barrier():
6654          */
6655         barrier();
6656
6657         kvm_guest_exit();
6658
6659         preempt_enable();
6660
6661         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6662
6663         /*
6664          * Profile KVM exit RIPs:
6665          */
6666         if (unlikely(prof_on == KVM_PROFILING)) {
6667                 unsigned long rip = kvm_rip_read(vcpu);
6668                 profile_hit(KVM_PROFILING, (void *)rip);
6669         }
6670
6671         if (unlikely(vcpu->arch.tsc_always_catchup))
6672                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6673
6674         if (vcpu->arch.apic_attention)
6675                 kvm_lapic_sync_from_vapic(vcpu);
6676
6677         r = kvm_x86_ops->handle_exit(vcpu);
6678         return r;
6679
6680 cancel_injection:
6681         kvm_x86_ops->cancel_injection(vcpu);
6682         if (unlikely(vcpu->arch.apic_attention))
6683                 kvm_lapic_sync_from_vapic(vcpu);
6684 out:
6685         return r;
6686 }
6687
6688 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6689 {
6690         if (!kvm_arch_vcpu_runnable(vcpu) &&
6691             (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6692                 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6693                 kvm_vcpu_block(vcpu);
6694                 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6695
6696                 if (kvm_x86_ops->post_block)
6697                         kvm_x86_ops->post_block(vcpu);
6698
6699                 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6700                         return 1;
6701         }
6702
6703         kvm_apic_accept_events(vcpu);
6704         switch(vcpu->arch.mp_state) {
6705         case KVM_MP_STATE_HALTED:
6706                 vcpu->arch.pv.pv_unhalted = false;
6707                 vcpu->arch.mp_state =
6708                         KVM_MP_STATE_RUNNABLE;
6709         case KVM_MP_STATE_RUNNABLE:
6710                 vcpu->arch.apf.halted = false;
6711                 break;
6712         case KVM_MP_STATE_INIT_RECEIVED:
6713                 break;
6714         default:
6715                 return -EINTR;
6716                 break;
6717         }
6718         return 1;
6719 }
6720
6721 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6722 {
6723         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6724                 !vcpu->arch.apf.halted);
6725 }
6726
6727 static int vcpu_run(struct kvm_vcpu *vcpu)
6728 {
6729         int r;
6730         struct kvm *kvm = vcpu->kvm;
6731
6732         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6733
6734         for (;;) {
6735                 if (kvm_vcpu_running(vcpu)) {
6736                         r = vcpu_enter_guest(vcpu);
6737                 } else {
6738                         r = vcpu_block(kvm, vcpu);
6739                 }
6740
6741                 if (r <= 0)
6742                         break;
6743
6744                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6745                 if (kvm_cpu_has_pending_timer(vcpu))
6746                         kvm_inject_pending_timer_irqs(vcpu);
6747
6748                 if (dm_request_for_irq_injection(vcpu) &&
6749                         kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6750                         r = 0;
6751                         vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6752                         ++vcpu->stat.request_irq_exits;
6753                         break;
6754                 }
6755
6756                 kvm_check_async_pf_completion(vcpu);
6757
6758                 if (signal_pending(current)) {
6759                         r = -EINTR;
6760                         vcpu->run->exit_reason = KVM_EXIT_INTR;
6761                         ++vcpu->stat.signal_exits;
6762                         break;
6763                 }
6764                 if (need_resched()) {
6765                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6766                         cond_resched();
6767                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6768                 }
6769         }
6770
6771         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6772
6773         return r;
6774 }
6775
6776 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6777 {
6778         int r;
6779         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6780         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6781         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6782         if (r != EMULATE_DONE)
6783                 return 0;
6784         return 1;
6785 }
6786
6787 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6788 {
6789         BUG_ON(!vcpu->arch.pio.count);
6790
6791         return complete_emulated_io(vcpu);
6792 }
6793
6794 /*
6795  * Implements the following, as a state machine:
6796  *
6797  * read:
6798  *   for each fragment
6799  *     for each mmio piece in the fragment
6800  *       write gpa, len
6801  *       exit
6802  *       copy data
6803  *   execute insn
6804  *
6805  * write:
6806  *   for each fragment
6807  *     for each mmio piece in the fragment
6808  *       write gpa, len
6809  *       copy data
6810  *       exit
6811  */
6812 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6813 {
6814         struct kvm_run *run = vcpu->run;
6815         struct kvm_mmio_fragment *frag;
6816         unsigned len;
6817
6818         BUG_ON(!vcpu->mmio_needed);
6819
6820         /* Complete previous fragment */
6821         frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6822         len = min(8u, frag->len);
6823         if (!vcpu->mmio_is_write)
6824                 memcpy(frag->data, run->mmio.data, len);
6825
6826         if (frag->len <= 8) {
6827                 /* Switch to the next fragment. */
6828                 frag++;
6829                 vcpu->mmio_cur_fragment++;
6830         } else {
6831                 /* Go forward to the next mmio piece. */
6832                 frag->data += len;
6833                 frag->gpa += len;
6834                 frag->len -= len;
6835         }
6836
6837         if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6838                 vcpu->mmio_needed = 0;
6839
6840                 /* FIXME: return into emulator if single-stepping.  */
6841                 if (vcpu->mmio_is_write)
6842                         return 1;
6843                 vcpu->mmio_read_completed = 1;
6844                 return complete_emulated_io(vcpu);
6845         }
6846
6847         run->exit_reason = KVM_EXIT_MMIO;
6848         run->mmio.phys_addr = frag->gpa;
6849         if (vcpu->mmio_is_write)
6850                 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6851         run->mmio.len = min(8u, frag->len);
6852         run->mmio.is_write = vcpu->mmio_is_write;
6853         vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6854         return 0;
6855 }
6856
6857
6858 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6859 {
6860         struct fpu *fpu = &current->thread.fpu;
6861         int r;
6862         sigset_t sigsaved;
6863
6864         fpu__activate_curr(fpu);
6865
6866         if (vcpu->sigset_active)
6867                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6868
6869         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6870                 kvm_vcpu_block(vcpu);
6871                 kvm_apic_accept_events(vcpu);
6872                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6873                 r = -EAGAIN;
6874                 goto out;
6875         }
6876
6877         /* re-sync apic's tpr */
6878         if (!lapic_in_kernel(vcpu)) {
6879                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6880                         r = -EINVAL;
6881                         goto out;
6882                 }
6883         }
6884
6885         if (unlikely(vcpu->arch.complete_userspace_io)) {
6886                 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6887                 vcpu->arch.complete_userspace_io = NULL;
6888                 r = cui(vcpu);
6889                 if (r <= 0)
6890                         goto out;
6891         } else
6892                 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6893
6894         r = vcpu_run(vcpu);
6895
6896 out:
6897         post_kvm_run_save(vcpu);
6898         if (vcpu->sigset_active)
6899                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6900
6901         return r;
6902 }
6903
6904 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6905 {
6906         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6907                 /*
6908                  * We are here if userspace calls get_regs() in the middle of
6909                  * instruction emulation. Registers state needs to be copied
6910                  * back from emulation context to vcpu. Userspace shouldn't do
6911                  * that usually, but some bad designed PV devices (vmware
6912                  * backdoor interface) need this to work
6913                  */
6914                 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6915                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6916         }
6917         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6918         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6919         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6920         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6921         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6922         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6923         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6924         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6925 #ifdef CONFIG_X86_64
6926         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6927         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6928         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6929         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6930         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6931         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6932         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6933         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6934 #endif
6935
6936         regs->rip = kvm_rip_read(vcpu);
6937         regs->rflags = kvm_get_rflags(vcpu);
6938
6939         return 0;
6940 }
6941
6942 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6943 {
6944         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6945         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6946
6947         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6948         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6949         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6950         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6951         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6952         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6953         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6954         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6955 #ifdef CONFIG_X86_64
6956         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6957         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6958         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6959         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6960         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6961         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6962         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6963         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6964 #endif
6965
6966         kvm_rip_write(vcpu, regs->rip);
6967         kvm_set_rflags(vcpu, regs->rflags);
6968
6969         vcpu->arch.exception.pending = false;
6970
6971         kvm_make_request(KVM_REQ_EVENT, vcpu);
6972
6973         return 0;
6974 }
6975
6976 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6977 {
6978         struct kvm_segment cs;
6979
6980         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6981         *db = cs.db;
6982         *l = cs.l;
6983 }
6984 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6985
6986 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6987                                   struct kvm_sregs *sregs)
6988 {
6989         struct desc_ptr dt;
6990
6991         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6992         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6993         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6994         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6995         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6996         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6997
6998         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6999         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7000
7001         kvm_x86_ops->get_idt(vcpu, &dt);
7002         sregs->idt.limit = dt.size;
7003         sregs->idt.base = dt.address;
7004         kvm_x86_ops->get_gdt(vcpu, &dt);
7005         sregs->gdt.limit = dt.size;
7006         sregs->gdt.base = dt.address;
7007
7008         sregs->cr0 = kvm_read_cr0(vcpu);
7009         sregs->cr2 = vcpu->arch.cr2;
7010         sregs->cr3 = kvm_read_cr3(vcpu);
7011         sregs->cr4 = kvm_read_cr4(vcpu);
7012         sregs->cr8 = kvm_get_cr8(vcpu);
7013         sregs->efer = vcpu->arch.efer;
7014         sregs->apic_base = kvm_get_apic_base(vcpu);
7015
7016         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7017
7018         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7019                 set_bit(vcpu->arch.interrupt.nr,
7020                         (unsigned long *)sregs->interrupt_bitmap);
7021
7022         return 0;
7023 }
7024
7025 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7026                                     struct kvm_mp_state *mp_state)
7027 {
7028         kvm_apic_accept_events(vcpu);
7029         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7030                                         vcpu->arch.pv.pv_unhalted)
7031                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7032         else
7033                 mp_state->mp_state = vcpu->arch.mp_state;
7034
7035         return 0;
7036 }
7037
7038 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7039                                     struct kvm_mp_state *mp_state)
7040 {
7041         if (!kvm_vcpu_has_lapic(vcpu) &&
7042             mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7043                 return -EINVAL;
7044
7045         if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7046                 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7047                 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7048         } else
7049                 vcpu->arch.mp_state = mp_state->mp_state;
7050         kvm_make_request(KVM_REQ_EVENT, vcpu);
7051         return 0;
7052 }
7053
7054 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7055                     int reason, bool has_error_code, u32 error_code)
7056 {
7057         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7058         int ret;
7059
7060         init_emulate_ctxt(vcpu);
7061
7062         ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7063                                    has_error_code, error_code);
7064
7065         if (ret)
7066                 return EMULATE_FAIL;
7067
7068         kvm_rip_write(vcpu, ctxt->eip);
7069         kvm_set_rflags(vcpu, ctxt->eflags);
7070         kvm_make_request(KVM_REQ_EVENT, vcpu);
7071         return EMULATE_DONE;
7072 }
7073 EXPORT_SYMBOL_GPL(kvm_task_switch);
7074
7075 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7076                                   struct kvm_sregs *sregs)
7077 {
7078         struct msr_data apic_base_msr;
7079         int mmu_reset_needed = 0;
7080         int pending_vec, max_bits, idx;
7081         struct desc_ptr dt;
7082
7083         if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7084                 return -EINVAL;
7085
7086         dt.size = sregs->idt.limit;
7087         dt.address = sregs->idt.base;
7088         kvm_x86_ops->set_idt(vcpu, &dt);
7089         dt.size = sregs->gdt.limit;
7090         dt.address = sregs->gdt.base;
7091         kvm_x86_ops->set_gdt(vcpu, &dt);
7092
7093         vcpu->arch.cr2 = sregs->cr2;
7094         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7095         vcpu->arch.cr3 = sregs->cr3;
7096         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7097
7098         kvm_set_cr8(vcpu, sregs->cr8);
7099
7100         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7101         kvm_x86_ops->set_efer(vcpu, sregs->efer);
7102         apic_base_msr.data = sregs->apic_base;
7103         apic_base_msr.host_initiated = true;
7104         kvm_set_apic_base(vcpu, &apic_base_msr);
7105
7106         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7107         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7108         vcpu->arch.cr0 = sregs->cr0;
7109
7110         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7111         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7112         if (sregs->cr4 & X86_CR4_OSXSAVE)
7113                 kvm_update_cpuid(vcpu);
7114
7115         idx = srcu_read_lock(&vcpu->kvm->srcu);
7116         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7117                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7118                 mmu_reset_needed = 1;
7119         }
7120         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7121
7122         if (mmu_reset_needed)
7123                 kvm_mmu_reset_context(vcpu);
7124
7125         max_bits = KVM_NR_INTERRUPTS;
7126         pending_vec = find_first_bit(
7127                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7128         if (pending_vec < max_bits) {
7129                 kvm_queue_interrupt(vcpu, pending_vec, false);
7130                 pr_debug("Set back pending irq %d\n", pending_vec);
7131         }
7132
7133         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7134         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7135         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7136         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7137         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7138         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7139
7140         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7141         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7142
7143         update_cr8_intercept(vcpu);
7144
7145         /* Older userspace won't unhalt the vcpu on reset. */
7146         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7147             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7148             !is_protmode(vcpu))
7149                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7150
7151         kvm_make_request(KVM_REQ_EVENT, vcpu);
7152
7153         return 0;
7154 }
7155
7156 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7157                                         struct kvm_guest_debug *dbg)
7158 {
7159         unsigned long rflags;
7160         int i, r;
7161
7162         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7163                 r = -EBUSY;
7164                 if (vcpu->arch.exception.pending)
7165                         goto out;
7166                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7167                         kvm_queue_exception(vcpu, DB_VECTOR);
7168                 else
7169                         kvm_queue_exception(vcpu, BP_VECTOR);
7170         }
7171
7172         /*
7173          * Read rflags as long as potentially injected trace flags are still
7174          * filtered out.
7175          */
7176         rflags = kvm_get_rflags(vcpu);
7177
7178         vcpu->guest_debug = dbg->control;
7179         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7180                 vcpu->guest_debug = 0;
7181
7182         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7183                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7184                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7185                 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7186         } else {
7187                 for (i = 0; i < KVM_NR_DB_REGS; i++)
7188                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7189         }
7190         kvm_update_dr7(vcpu);
7191
7192         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7193                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7194                         get_segment_base(vcpu, VCPU_SREG_CS);
7195
7196         /*
7197          * Trigger an rflags update that will inject or remove the trace
7198          * flags.
7199          */
7200         kvm_set_rflags(vcpu, rflags);
7201
7202         kvm_x86_ops->update_bp_intercept(vcpu);
7203
7204         r = 0;
7205
7206 out:
7207
7208         return r;
7209 }
7210
7211 /*
7212  * Translate a guest virtual address to a guest physical address.
7213  */
7214 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7215                                     struct kvm_translation *tr)
7216 {
7217         unsigned long vaddr = tr->linear_address;
7218         gpa_t gpa;
7219         int idx;
7220
7221         idx = srcu_read_lock(&vcpu->kvm->srcu);
7222         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7223         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7224         tr->physical_address = gpa;
7225         tr->valid = gpa != UNMAPPED_GVA;
7226         tr->writeable = 1;
7227         tr->usermode = 0;
7228
7229         return 0;
7230 }
7231
7232 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7233 {
7234         struct fxregs_state *fxsave =
7235                         &vcpu->arch.guest_fpu.state.fxsave;
7236
7237         memcpy(fpu->fpr, fxsave->st_space, 128);
7238         fpu->fcw = fxsave->cwd;
7239         fpu->fsw = fxsave->swd;
7240         fpu->ftwx = fxsave->twd;
7241         fpu->last_opcode = fxsave->fop;
7242         fpu->last_ip = fxsave->rip;
7243         fpu->last_dp = fxsave->rdp;
7244         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7245
7246         return 0;
7247 }
7248
7249 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7250 {
7251         struct fxregs_state *fxsave =
7252                         &vcpu->arch.guest_fpu.state.fxsave;
7253
7254         memcpy(fxsave->st_space, fpu->fpr, 128);
7255         fxsave->cwd = fpu->fcw;
7256         fxsave->swd = fpu->fsw;
7257         fxsave->twd = fpu->ftwx;
7258         fxsave->fop = fpu->last_opcode;
7259         fxsave->rip = fpu->last_ip;
7260         fxsave->rdp = fpu->last_dp;
7261         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7262
7263         return 0;
7264 }
7265
7266 static void fx_init(struct kvm_vcpu *vcpu)
7267 {
7268         fpstate_init(&vcpu->arch.guest_fpu.state);
7269         if (cpu_has_xsaves)
7270                 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7271                         host_xcr0 | XSTATE_COMPACTION_ENABLED;
7272
7273         /*
7274          * Ensure guest xcr0 is valid for loading
7275          */
7276         vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7277
7278         vcpu->arch.cr0 |= X86_CR0_ET;
7279 }
7280
7281 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7282 {
7283         if (vcpu->guest_fpu_loaded)
7284                 return;
7285
7286         /*
7287          * Restore all possible states in the guest,
7288          * and assume host would use all available bits.
7289          * Guest xcr0 would be loaded later.
7290          */
7291         kvm_put_guest_xcr0(vcpu);
7292         vcpu->guest_fpu_loaded = 1;
7293         __kernel_fpu_begin();
7294         __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7295         trace_kvm_fpu(1);
7296 }
7297
7298 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7299 {
7300         kvm_put_guest_xcr0(vcpu);
7301
7302         if (!vcpu->guest_fpu_loaded) {
7303                 vcpu->fpu_counter = 0;
7304                 return;
7305         }
7306
7307         vcpu->guest_fpu_loaded = 0;
7308         copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7309         __kernel_fpu_end();
7310         ++vcpu->stat.fpu_reload;
7311         /*
7312          * If using eager FPU mode, or if the guest is a frequent user
7313          * of the FPU, just leave the FPU active for next time.
7314          * Every 255 times fpu_counter rolls over to 0; a guest that uses
7315          * the FPU in bursts will revert to loading it on demand.
7316          */
7317         if (!vcpu->arch.eager_fpu) {
7318                 if (++vcpu->fpu_counter < 5)
7319                         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7320         }
7321         trace_kvm_fpu(0);
7322 }
7323
7324 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7325 {
7326         kvmclock_reset(vcpu);
7327
7328         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7329         kvm_x86_ops->vcpu_free(vcpu);
7330 }
7331
7332 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7333                                                 unsigned int id)
7334 {
7335         struct kvm_vcpu *vcpu;
7336
7337         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7338                 printk_once(KERN_WARNING
7339                 "kvm: SMP vm created on host with unstable TSC; "
7340                 "guest TSC will not be reliable\n");
7341
7342         vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7343
7344         return vcpu;
7345 }
7346
7347 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7348 {
7349         int r;
7350
7351         kvm_vcpu_mtrr_init(vcpu);
7352         r = vcpu_load(vcpu);
7353         if (r)
7354                 return r;
7355         kvm_vcpu_reset(vcpu, false);
7356         kvm_mmu_setup(vcpu);
7357         vcpu_put(vcpu);
7358         return r;
7359 }
7360
7361 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7362 {
7363         struct msr_data msr;
7364         struct kvm *kvm = vcpu->kvm;
7365
7366         if (vcpu_load(vcpu))
7367                 return;
7368         msr.data = 0x0;
7369         msr.index = MSR_IA32_TSC;
7370         msr.host_initiated = true;
7371         kvm_write_tsc(vcpu, &msr);
7372         vcpu_put(vcpu);
7373
7374         if (!kvmclock_periodic_sync)
7375                 return;
7376
7377         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7378                                         KVMCLOCK_SYNC_PERIOD);
7379 }
7380
7381 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7382 {
7383         int r;
7384         vcpu->arch.apf.msr_val = 0;
7385
7386         r = vcpu_load(vcpu);
7387         BUG_ON(r);
7388         kvm_mmu_unload(vcpu);
7389         vcpu_put(vcpu);
7390
7391         kvm_x86_ops->vcpu_free(vcpu);
7392 }
7393
7394 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7395 {
7396         vcpu->arch.hflags = 0;
7397
7398         atomic_set(&vcpu->arch.nmi_queued, 0);
7399         vcpu->arch.nmi_pending = 0;
7400         vcpu->arch.nmi_injected = false;
7401         kvm_clear_interrupt_queue(vcpu);
7402         kvm_clear_exception_queue(vcpu);
7403
7404         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7405         kvm_update_dr0123(vcpu);
7406         vcpu->arch.dr6 = DR6_INIT;
7407         kvm_update_dr6(vcpu);
7408         vcpu->arch.dr7 = DR7_FIXED_1;
7409         kvm_update_dr7(vcpu);
7410
7411         vcpu->arch.cr2 = 0;
7412
7413         kvm_make_request(KVM_REQ_EVENT, vcpu);
7414         vcpu->arch.apf.msr_val = 0;
7415         vcpu->arch.st.msr_val = 0;
7416
7417         kvmclock_reset(vcpu);
7418
7419         kvm_clear_async_pf_completion_queue(vcpu);
7420         kvm_async_pf_hash_reset(vcpu);
7421         vcpu->arch.apf.halted = false;
7422
7423         if (!init_event) {
7424                 kvm_pmu_reset(vcpu);
7425                 vcpu->arch.smbase = 0x30000;
7426         }
7427
7428         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7429         vcpu->arch.regs_avail = ~0;
7430         vcpu->arch.regs_dirty = ~0;
7431
7432         kvm_x86_ops->vcpu_reset(vcpu, init_event);
7433 }
7434
7435 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7436 {
7437         struct kvm_segment cs;
7438
7439         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7440         cs.selector = vector << 8;
7441         cs.base = vector << 12;
7442         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7443         kvm_rip_write(vcpu, 0);
7444 }
7445
7446 int kvm_arch_hardware_enable(void)
7447 {
7448         struct kvm *kvm;
7449         struct kvm_vcpu *vcpu;
7450         int i;
7451         int ret;
7452         u64 local_tsc;
7453         u64 max_tsc = 0;
7454         bool stable, backwards_tsc = false;
7455
7456         kvm_shared_msr_cpu_online();
7457         ret = kvm_x86_ops->hardware_enable();
7458         if (ret != 0)
7459                 return ret;
7460
7461         local_tsc = rdtsc();
7462         stable = !check_tsc_unstable();
7463         list_for_each_entry(kvm, &vm_list, vm_list) {
7464                 kvm_for_each_vcpu(i, vcpu, kvm) {
7465                         if (!stable && vcpu->cpu == smp_processor_id())
7466                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7467                         if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7468                                 backwards_tsc = true;
7469                                 if (vcpu->arch.last_host_tsc > max_tsc)
7470                                         max_tsc = vcpu->arch.last_host_tsc;
7471                         }
7472                 }
7473         }
7474
7475         /*
7476          * Sometimes, even reliable TSCs go backwards.  This happens on
7477          * platforms that reset TSC during suspend or hibernate actions, but
7478          * maintain synchronization.  We must compensate.  Fortunately, we can
7479          * detect that condition here, which happens early in CPU bringup,
7480          * before any KVM threads can be running.  Unfortunately, we can't
7481          * bring the TSCs fully up to date with real time, as we aren't yet far
7482          * enough into CPU bringup that we know how much real time has actually
7483          * elapsed; our helper function, get_kernel_ns() will be using boot
7484          * variables that haven't been updated yet.
7485          *
7486          * So we simply find the maximum observed TSC above, then record the
7487          * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7488          * the adjustment will be applied.  Note that we accumulate
7489          * adjustments, in case multiple suspend cycles happen before some VCPU
7490          * gets a chance to run again.  In the event that no KVM threads get a
7491          * chance to run, we will miss the entire elapsed period, as we'll have
7492          * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7493          * loose cycle time.  This isn't too big a deal, since the loss will be
7494          * uniform across all VCPUs (not to mention the scenario is extremely
7495          * unlikely). It is possible that a second hibernate recovery happens
7496          * much faster than a first, causing the observed TSC here to be
7497          * smaller; this would require additional padding adjustment, which is
7498          * why we set last_host_tsc to the local tsc observed here.
7499          *
7500          * N.B. - this code below runs only on platforms with reliable TSC,
7501          * as that is the only way backwards_tsc is set above.  Also note
7502          * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7503          * have the same delta_cyc adjustment applied if backwards_tsc
7504          * is detected.  Note further, this adjustment is only done once,
7505          * as we reset last_host_tsc on all VCPUs to stop this from being
7506          * called multiple times (one for each physical CPU bringup).
7507          *
7508          * Platforms with unreliable TSCs don't have to deal with this, they
7509          * will be compensated by the logic in vcpu_load, which sets the TSC to
7510          * catchup mode.  This will catchup all VCPUs to real time, but cannot
7511          * guarantee that they stay in perfect synchronization.
7512          */
7513         if (backwards_tsc) {
7514                 u64 delta_cyc = max_tsc - local_tsc;
7515                 backwards_tsc_observed = true;
7516                 list_for_each_entry(kvm, &vm_list, vm_list) {
7517                         kvm_for_each_vcpu(i, vcpu, kvm) {
7518                                 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7519                                 vcpu->arch.last_host_tsc = local_tsc;
7520                                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7521                         }
7522
7523                         /*
7524                          * We have to disable TSC offset matching.. if you were
7525                          * booting a VM while issuing an S4 host suspend....
7526                          * you may have some problem.  Solving this issue is
7527                          * left as an exercise to the reader.
7528                          */
7529                         kvm->arch.last_tsc_nsec = 0;
7530                         kvm->arch.last_tsc_write = 0;
7531                 }
7532
7533         }
7534         return 0;
7535 }
7536
7537 void kvm_arch_hardware_disable(void)
7538 {
7539         kvm_x86_ops->hardware_disable();
7540         drop_user_return_notifiers();
7541 }
7542
7543 int kvm_arch_hardware_setup(void)
7544 {
7545         int r;
7546
7547         r = kvm_x86_ops->hardware_setup();
7548         if (r != 0)
7549                 return r;
7550
7551         if (kvm_has_tsc_control) {
7552                 /*
7553                  * Make sure the user can only configure tsc_khz values that
7554                  * fit into a signed integer.
7555                  * A min value is not calculated needed because it will always
7556                  * be 1 on all machines.
7557                  */
7558                 u64 max = min(0x7fffffffULL,
7559                               __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7560                 kvm_max_guest_tsc_khz = max;
7561
7562                 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7563         }
7564
7565         kvm_init_msr_list();
7566         return 0;
7567 }
7568
7569 void kvm_arch_hardware_unsetup(void)
7570 {
7571         kvm_x86_ops->hardware_unsetup();
7572 }
7573
7574 void kvm_arch_check_processor_compat(void *rtn)
7575 {
7576         kvm_x86_ops->check_processor_compatibility(rtn);
7577 }
7578
7579 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7580 {
7581         return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7582 }
7583 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7584
7585 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7586 {
7587         return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7588 }
7589
7590 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7591 {
7592         return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7593 }
7594
7595 struct static_key kvm_no_apic_vcpu __read_mostly;
7596
7597 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7598 {
7599         struct page *page;
7600         struct kvm *kvm;
7601         int r;
7602
7603         BUG_ON(vcpu->kvm == NULL);
7604         kvm = vcpu->kvm;
7605
7606         vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7607         vcpu->arch.pv.pv_unhalted = false;
7608         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7609         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7610                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7611         else
7612                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7613
7614         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7615         if (!page) {
7616                 r = -ENOMEM;
7617                 goto fail;
7618         }
7619         vcpu->arch.pio_data = page_address(page);
7620
7621         kvm_set_tsc_khz(vcpu, max_tsc_khz);
7622
7623         r = kvm_mmu_create(vcpu);
7624         if (r < 0)
7625                 goto fail_free_pio_data;
7626
7627         if (irqchip_in_kernel(kvm)) {
7628                 r = kvm_create_lapic(vcpu);
7629                 if (r < 0)
7630                         goto fail_mmu_destroy;
7631         } else
7632                 static_key_slow_inc(&kvm_no_apic_vcpu);
7633
7634         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7635                                        GFP_KERNEL);
7636         if (!vcpu->arch.mce_banks) {
7637                 r = -ENOMEM;
7638                 goto fail_free_lapic;
7639         }
7640         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7641
7642         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7643                 r = -ENOMEM;
7644                 goto fail_free_mce_banks;
7645         }
7646
7647         fx_init(vcpu);
7648
7649         vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7650         vcpu->arch.pv_time_enabled = false;
7651
7652         vcpu->arch.guest_supported_xcr0 = 0;
7653         vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7654
7655         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7656
7657         vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7658
7659         kvm_async_pf_hash_reset(vcpu);
7660         kvm_pmu_init(vcpu);
7661
7662         vcpu->arch.pending_external_vector = -1;
7663
7664         kvm_hv_vcpu_init(vcpu);
7665
7666         return 0;
7667
7668 fail_free_mce_banks:
7669         kfree(vcpu->arch.mce_banks);
7670 fail_free_lapic:
7671         kvm_free_lapic(vcpu);
7672 fail_mmu_destroy:
7673         kvm_mmu_destroy(vcpu);
7674 fail_free_pio_data:
7675         free_page((unsigned long)vcpu->arch.pio_data);
7676 fail:
7677         return r;
7678 }
7679
7680 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7681 {
7682         int idx;
7683
7684         kvm_hv_vcpu_uninit(vcpu);
7685         kvm_pmu_destroy(vcpu);
7686         kfree(vcpu->arch.mce_banks);
7687         kvm_free_lapic(vcpu);
7688         idx = srcu_read_lock(&vcpu->kvm->srcu);
7689         kvm_mmu_destroy(vcpu);
7690         srcu_read_unlock(&vcpu->kvm->srcu, idx);
7691         free_page((unsigned long)vcpu->arch.pio_data);
7692         if (!lapic_in_kernel(vcpu))
7693                 static_key_slow_dec(&kvm_no_apic_vcpu);
7694 }
7695
7696 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7697 {
7698         kvm_x86_ops->sched_in(vcpu, cpu);
7699 }
7700
7701 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7702 {
7703         if (type)
7704                 return -EINVAL;
7705
7706         INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7707         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7708         INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7709         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7710         atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7711
7712         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7713         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7714         /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7715         set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7716                 &kvm->arch.irq_sources_bitmap);
7717
7718         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7719         mutex_init(&kvm->arch.apic_map_lock);
7720         spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7721
7722         pvclock_update_vm_gtod_copy(kvm);
7723
7724         INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7725         INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7726
7727         return 0;
7728 }
7729
7730 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7731 {
7732         int r;
7733         r = vcpu_load(vcpu);
7734         BUG_ON(r);
7735         kvm_mmu_unload(vcpu);
7736         vcpu_put(vcpu);
7737 }
7738
7739 static void kvm_free_vcpus(struct kvm *kvm)
7740 {
7741         unsigned int i;
7742         struct kvm_vcpu *vcpu;
7743
7744         /*
7745          * Unpin any mmu pages first.
7746          */
7747         kvm_for_each_vcpu(i, vcpu, kvm) {
7748                 kvm_clear_async_pf_completion_queue(vcpu);
7749                 kvm_unload_vcpu_mmu(vcpu);
7750         }
7751         kvm_for_each_vcpu(i, vcpu, kvm)
7752                 kvm_arch_vcpu_free(vcpu);
7753
7754         mutex_lock(&kvm->lock);
7755         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7756                 kvm->vcpus[i] = NULL;
7757
7758         atomic_set(&kvm->online_vcpus, 0);
7759         mutex_unlock(&kvm->lock);
7760 }
7761
7762 void kvm_arch_sync_events(struct kvm *kvm)
7763 {
7764         cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7765         cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7766         kvm_free_all_assigned_devices(kvm);
7767         kvm_free_pit(kvm);
7768 }
7769
7770 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7771 {
7772         int i, r;
7773         unsigned long hva;
7774         struct kvm_memslots *slots = kvm_memslots(kvm);
7775         struct kvm_memory_slot *slot, old;
7776
7777         /* Called with kvm->slots_lock held.  */
7778         if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7779                 return -EINVAL;
7780
7781         slot = id_to_memslot(slots, id);
7782         if (size) {
7783                 if (WARN_ON(slot->npages))
7784                         return -EEXIST;
7785
7786                 /*
7787                  * MAP_SHARED to prevent internal slot pages from being moved
7788                  * by fork()/COW.
7789                  */
7790                 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7791                               MAP_SHARED | MAP_ANONYMOUS, 0);
7792                 if (IS_ERR((void *)hva))
7793                         return PTR_ERR((void *)hva);
7794         } else {
7795                 if (!slot->npages)
7796                         return 0;
7797
7798                 hva = 0;
7799         }
7800
7801         old = *slot;
7802         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7803                 struct kvm_userspace_memory_region m;
7804
7805                 m.slot = id | (i << 16);
7806                 m.flags = 0;
7807                 m.guest_phys_addr = gpa;
7808                 m.userspace_addr = hva;
7809                 m.memory_size = size;
7810                 r = __kvm_set_memory_region(kvm, &m);
7811                 if (r < 0)
7812                         return r;
7813         }
7814
7815         if (!size) {
7816                 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7817                 WARN_ON(r < 0);
7818         }
7819
7820         return 0;
7821 }
7822 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7823
7824 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7825 {
7826         int r;
7827
7828         mutex_lock(&kvm->slots_lock);
7829         r = __x86_set_memory_region(kvm, id, gpa, size);
7830         mutex_unlock(&kvm->slots_lock);
7831
7832         return r;
7833 }
7834 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7835
7836 void kvm_arch_destroy_vm(struct kvm *kvm)
7837 {
7838         if (current->mm == kvm->mm) {
7839                 /*
7840                  * Free memory regions allocated on behalf of userspace,
7841                  * unless the the memory map has changed due to process exit
7842                  * or fd copying.
7843                  */
7844                 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7845                 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7846                 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7847         }
7848         kvm_iommu_unmap_guest(kvm);
7849         kfree(kvm->arch.vpic);
7850         kfree(kvm->arch.vioapic);
7851         kvm_free_vcpus(kvm);
7852         kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7853 }
7854
7855 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7856                            struct kvm_memory_slot *dont)
7857 {
7858         int i;
7859
7860         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7861                 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7862                         kvfree(free->arch.rmap[i]);
7863                         free->arch.rmap[i] = NULL;
7864                 }
7865                 if (i == 0)
7866                         continue;
7867
7868                 if (!dont || free->arch.lpage_info[i - 1] !=
7869                              dont->arch.lpage_info[i - 1]) {
7870                         kvfree(free->arch.lpage_info[i - 1]);
7871                         free->arch.lpage_info[i - 1] = NULL;
7872                 }
7873         }
7874 }
7875
7876 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7877                             unsigned long npages)
7878 {
7879         int i;
7880
7881         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7882                 unsigned long ugfn;
7883                 int lpages;
7884                 int level = i + 1;
7885
7886                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7887                                       slot->base_gfn, level) + 1;
7888
7889                 slot->arch.rmap[i] =
7890                         kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7891                 if (!slot->arch.rmap[i])
7892                         goto out_free;
7893                 if (i == 0)
7894                         continue;
7895
7896                 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7897                                         sizeof(*slot->arch.lpage_info[i - 1]));
7898                 if (!slot->arch.lpage_info[i - 1])
7899                         goto out_free;
7900
7901                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7902                         slot->arch.lpage_info[i - 1][0].write_count = 1;
7903                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7904                         slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7905                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7906                 /*
7907                  * If the gfn and userspace address are not aligned wrt each
7908                  * other, or if explicitly asked to, disable large page
7909                  * support for this slot
7910                  */
7911                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7912                     !kvm_largepages_enabled()) {
7913                         unsigned long j;
7914
7915                         for (j = 0; j < lpages; ++j)
7916                                 slot->arch.lpage_info[i - 1][j].write_count = 1;
7917                 }
7918         }
7919
7920         return 0;
7921
7922 out_free:
7923         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7924                 kvfree(slot->arch.rmap[i]);
7925                 slot->arch.rmap[i] = NULL;
7926                 if (i == 0)
7927                         continue;
7928
7929                 kvfree(slot->arch.lpage_info[i - 1]);
7930                 slot->arch.lpage_info[i - 1] = NULL;
7931         }
7932         return -ENOMEM;
7933 }
7934
7935 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7936 {
7937         /*
7938          * memslots->generation has been incremented.
7939          * mmio generation may have reached its maximum value.
7940          */
7941         kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7942 }
7943
7944 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7945                                 struct kvm_memory_slot *memslot,
7946                                 const struct kvm_userspace_memory_region *mem,
7947                                 enum kvm_mr_change change)
7948 {
7949         return 0;
7950 }
7951
7952 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7953                                      struct kvm_memory_slot *new)
7954 {
7955         /* Still write protect RO slot */
7956         if (new->flags & KVM_MEM_READONLY) {
7957                 kvm_mmu_slot_remove_write_access(kvm, new);
7958                 return;
7959         }
7960
7961         /*
7962          * Call kvm_x86_ops dirty logging hooks when they are valid.
7963          *
7964          * kvm_x86_ops->slot_disable_log_dirty is called when:
7965          *
7966          *  - KVM_MR_CREATE with dirty logging is disabled
7967          *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7968          *
7969          * The reason is, in case of PML, we need to set D-bit for any slots
7970          * with dirty logging disabled in order to eliminate unnecessary GPA
7971          * logging in PML buffer (and potential PML buffer full VMEXT). This
7972          * guarantees leaving PML enabled during guest's lifetime won't have
7973          * any additonal overhead from PML when guest is running with dirty
7974          * logging disabled for memory slots.
7975          *
7976          * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7977          * to dirty logging mode.
7978          *
7979          * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7980          *
7981          * In case of write protect:
7982          *
7983          * Write protect all pages for dirty logging.
7984          *
7985          * All the sptes including the large sptes which point to this
7986          * slot are set to readonly. We can not create any new large
7987          * spte on this slot until the end of the logging.
7988          *
7989          * See the comments in fast_page_fault().
7990          */
7991         if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7992                 if (kvm_x86_ops->slot_enable_log_dirty)
7993                         kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7994                 else
7995                         kvm_mmu_slot_remove_write_access(kvm, new);
7996         } else {
7997                 if (kvm_x86_ops->slot_disable_log_dirty)
7998                         kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7999         }
8000 }
8001
8002 void kvm_arch_commit_memory_region(struct kvm *kvm,
8003                                 const struct kvm_userspace_memory_region *mem,
8004                                 const struct kvm_memory_slot *old,
8005                                 const struct kvm_memory_slot *new,
8006                                 enum kvm_mr_change change)
8007 {
8008         int nr_mmu_pages = 0;
8009
8010         if (!kvm->arch.n_requested_mmu_pages)
8011                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8012
8013         if (nr_mmu_pages)
8014                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8015
8016         /*
8017          * Dirty logging tracks sptes in 4k granularity, meaning that large
8018          * sptes have to be split.  If live migration is successful, the guest
8019          * in the source machine will be destroyed and large sptes will be
8020          * created in the destination. However, if the guest continues to run
8021          * in the source machine (for example if live migration fails), small
8022          * sptes will remain around and cause bad performance.
8023          *
8024          * Scan sptes if dirty logging has been stopped, dropping those
8025          * which can be collapsed into a single large-page spte.  Later
8026          * page faults will create the large-page sptes.
8027          */
8028         if ((change != KVM_MR_DELETE) &&
8029                 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8030                 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8031                 kvm_mmu_zap_collapsible_sptes(kvm, new);
8032
8033         /*
8034          * Set up write protection and/or dirty logging for the new slot.
8035          *
8036          * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8037          * been zapped so no dirty logging staff is needed for old slot. For
8038          * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8039          * new and it's also covered when dealing with the new slot.
8040          *
8041          * FIXME: const-ify all uses of struct kvm_memory_slot.
8042          */
8043         if (change != KVM_MR_DELETE)
8044                 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8045 }
8046
8047 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8048 {
8049         kvm_mmu_invalidate_zap_all_pages(kvm);
8050 }
8051
8052 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8053                                    struct kvm_memory_slot *slot)
8054 {
8055         kvm_mmu_invalidate_zap_all_pages(kvm);
8056 }
8057
8058 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8059 {
8060         if (!list_empty_careful(&vcpu->async_pf.done))
8061                 return true;
8062
8063         if (kvm_apic_has_events(vcpu))
8064                 return true;
8065
8066         if (vcpu->arch.pv.pv_unhalted)
8067                 return true;
8068
8069         if (atomic_read(&vcpu->arch.nmi_queued))
8070                 return true;
8071
8072         if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8073                 return true;
8074
8075         if (kvm_arch_interrupt_allowed(vcpu) &&
8076             kvm_cpu_has_interrupt(vcpu))
8077                 return true;
8078
8079         if (kvm_hv_has_stimer_pending(vcpu))
8080                 return true;
8081
8082         return false;
8083 }
8084
8085 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8086 {
8087         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8088                 kvm_x86_ops->check_nested_events(vcpu, false);
8089
8090         return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8091 }
8092
8093 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8094 {
8095         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8096 }
8097
8098 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8099 {
8100         return kvm_x86_ops->interrupt_allowed(vcpu);
8101 }
8102
8103 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8104 {
8105         if (is_64_bit_mode(vcpu))
8106                 return kvm_rip_read(vcpu);
8107         return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8108                      kvm_rip_read(vcpu));
8109 }
8110 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8111
8112 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8113 {
8114         return kvm_get_linear_rip(vcpu) == linear_rip;
8115 }
8116 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8117
8118 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8119 {
8120         unsigned long rflags;
8121
8122         rflags = kvm_x86_ops->get_rflags(vcpu);
8123         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8124                 rflags &= ~X86_EFLAGS_TF;
8125         return rflags;
8126 }
8127 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8128
8129 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8130 {
8131         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8132             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8133                 rflags |= X86_EFLAGS_TF;
8134         kvm_x86_ops->set_rflags(vcpu, rflags);
8135 }
8136
8137 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8138 {
8139         __kvm_set_rflags(vcpu, rflags);
8140         kvm_make_request(KVM_REQ_EVENT, vcpu);
8141 }
8142 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8143
8144 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8145 {
8146         int r;
8147
8148         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8149               work->wakeup_all)
8150                 return;
8151
8152         r = kvm_mmu_reload(vcpu);
8153         if (unlikely(r))
8154                 return;
8155
8156         if (!vcpu->arch.mmu.direct_map &&
8157               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8158                 return;
8159
8160         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8161 }
8162
8163 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8164 {
8165         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8166 }
8167
8168 static inline u32 kvm_async_pf_next_probe(u32 key)
8169 {
8170         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8171 }
8172
8173 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8174 {
8175         u32 key = kvm_async_pf_hash_fn(gfn);
8176
8177         while (vcpu->arch.apf.gfns[key] != ~0)
8178                 key = kvm_async_pf_next_probe(key);
8179
8180         vcpu->arch.apf.gfns[key] = gfn;
8181 }
8182
8183 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8184 {
8185         int i;
8186         u32 key = kvm_async_pf_hash_fn(gfn);
8187
8188         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8189                      (vcpu->arch.apf.gfns[key] != gfn &&
8190                       vcpu->arch.apf.gfns[key] != ~0); i++)
8191                 key = kvm_async_pf_next_probe(key);
8192
8193         return key;
8194 }
8195
8196 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8197 {
8198         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8199 }
8200
8201 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8202 {
8203         u32 i, j, k;
8204
8205         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8206         while (true) {
8207                 vcpu->arch.apf.gfns[i] = ~0;
8208                 do {
8209                         j = kvm_async_pf_next_probe(j);
8210                         if (vcpu->arch.apf.gfns[j] == ~0)
8211                                 return;
8212                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8213                         /*
8214                          * k lies cyclically in ]i,j]
8215                          * |    i.k.j |
8216                          * |....j i.k.| or  |.k..j i...|
8217                          */
8218                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8219                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8220                 i = j;
8221         }
8222 }
8223
8224 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8225 {
8226
8227         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8228                                       sizeof(val));
8229 }
8230
8231 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8232                                      struct kvm_async_pf *work)
8233 {
8234         struct x86_exception fault;
8235
8236         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8237         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8238
8239         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8240             (vcpu->arch.apf.send_user_only &&
8241              kvm_x86_ops->get_cpl(vcpu) == 0))
8242                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8243         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8244                 fault.vector = PF_VECTOR;
8245                 fault.error_code_valid = true;
8246                 fault.error_code = 0;
8247                 fault.nested_page_fault = false;
8248                 fault.address = work->arch.token;
8249                 kvm_inject_page_fault(vcpu, &fault);
8250         }
8251 }
8252
8253 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8254                                  struct kvm_async_pf *work)
8255 {
8256         struct x86_exception fault;
8257
8258         trace_kvm_async_pf_ready(work->arch.token, work->gva);
8259         if (work->wakeup_all)
8260                 work->arch.token = ~0; /* broadcast wakeup */
8261         else
8262                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8263
8264         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8265             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8266                 fault.vector = PF_VECTOR;
8267                 fault.error_code_valid = true;
8268                 fault.error_code = 0;
8269                 fault.nested_page_fault = false;
8270                 fault.address = work->arch.token;
8271                 kvm_inject_page_fault(vcpu, &fault);
8272         }
8273         vcpu->arch.apf.halted = false;
8274         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8275 }
8276
8277 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8278 {
8279         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8280                 return true;
8281         else
8282                 return !kvm_event_needs_reinjection(vcpu) &&
8283                         kvm_x86_ops->interrupt_allowed(vcpu);
8284 }
8285
8286 void kvm_arch_start_assignment(struct kvm *kvm)
8287 {
8288         atomic_inc(&kvm->arch.assigned_device_count);
8289 }
8290 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8291
8292 void kvm_arch_end_assignment(struct kvm *kvm)
8293 {
8294         atomic_dec(&kvm->arch.assigned_device_count);
8295 }
8296 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8297
8298 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8299 {
8300         return atomic_read(&kvm->arch.assigned_device_count);
8301 }
8302 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8303
8304 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8305 {
8306         atomic_inc(&kvm->arch.noncoherent_dma_count);
8307 }
8308 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8309
8310 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8311 {
8312         atomic_dec(&kvm->arch.noncoherent_dma_count);
8313 }
8314 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8315
8316 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8317 {
8318         return atomic_read(&kvm->arch.noncoherent_dma_count);
8319 }
8320 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8321
8322 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8323                                       struct irq_bypass_producer *prod)
8324 {
8325         struct kvm_kernel_irqfd *irqfd =
8326                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8327
8328         if (kvm_x86_ops->update_pi_irte) {
8329                 irqfd->producer = prod;
8330                 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8331                                 prod->irq, irqfd->gsi, 1);
8332         }
8333
8334         return -EINVAL;
8335 }
8336
8337 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8338                                       struct irq_bypass_producer *prod)
8339 {
8340         int ret;
8341         struct kvm_kernel_irqfd *irqfd =
8342                 container_of(cons, struct kvm_kernel_irqfd, consumer);
8343
8344         if (!kvm_x86_ops->update_pi_irte) {
8345                 WARN_ON(irqfd->producer != NULL);
8346                 return;
8347         }
8348
8349         WARN_ON(irqfd->producer != prod);
8350         irqfd->producer = NULL;
8351
8352         /*
8353          * When producer of consumer is unregistered, we change back to
8354          * remapped mode, so we can re-use the current implementation
8355          * when the irq is masked/disabed or the consumer side (KVM
8356          * int this case doesn't want to receive the interrupts.
8357         */
8358         ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8359         if (ret)
8360                 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8361                        " fails: %d\n", irqfd->consumer.token, ret);
8362 }
8363
8364 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8365                                    uint32_t guest_irq, bool set)
8366 {
8367         if (!kvm_x86_ops->update_pi_irte)
8368                 return -EINVAL;
8369
8370         return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8371 }
8372
8373 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8374 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8375 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8376 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8377 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8378 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8379 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8380 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8381 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8382 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8383 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8384 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8385 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8386 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8387 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8388 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8389 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);