arch/x86/kvm/x86.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Kernel-based Virtual Machine driver for Linux
   4  *
   5  * derived from drivers/kvm/kvm_main.c
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright (C) 2008 Qumranet, Inc.
   9  * Copyright IBM Corporation, 2008
  10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  11  *
  12  * Authors:
  13  *   Avi Kivity   <avi@qumranet.com>
  14  *   Yaniv Kamay  <yaniv@qumranet.com>
  15  *   Amit Shah    <amit.shah@qumranet.com>
  16  *   Ben-Ami Yassour <benami@il.ibm.com>
  17  */
  18
  19 #include <linux/kvm_host.h>
  20 #include "irq.h"
  21 #include "mmu.h"
  22 #include "i8254.h"
  23 #include "tss.h"
  24 #include "kvm_cache_regs.h"
  25 #include "x86.h"
  26 #include "cpuid.h"
  27 #include "pmu.h"
  28 #include "hyperv.h"
  29
  30 #include <linux/clocksource.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/kvm.h>
  33 #include <linux/fs.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/export.h>
  36 #include <linux/moduleparam.h>
  37 #include <linux/mman.h>
  38 #include <linux/highmem.h>
  39 #include <linux/iommu.h>
  40 #include <linux/intel-iommu.h>
  41 #include <linux/cpufreq.h>
  42 #include <linux/user-return-notifier.h>
  43 #include <linux/srcu.h>
  44 #include <linux/slab.h>
  45 #include <linux/perf_event.h>
  46 #include <linux/uaccess.h>
  47 #include <linux/hash.h>
  48 #include <linux/pci.h>
  49 #include <linux/timekeeper_internal.h>
  50 #include <linux/pvclock_gtod.h>
  51 #include <linux/kvm_irqfd.h>
  52 #include <linux/irqbypass.h>
  53 #include <linux/sched/stat.h>
  54 #include <linux/sched/isolation.h>
  55 #include <linux/mem_encrypt.h>
  56
  57 #include <trace/events/kvm.h>
  58
  59 #include <asm/debugreg.h>
  60 #include <asm/msr.h>
  61 #include <asm/desc.h>
  62 #include <asm/mce.h>
  63 #include <linux/kernel_stat.h>
  64 #include <asm/fpu/internal.h> /* Ugh! */
  65 #include <asm/pvclock.h>
  66 #include <asm/div64.h>
  67 #include <asm/irq_remapping.h>
  68 #include <asm/mshyperv.h>
  69 #include <asm/hypervisor.h>
  70 #include <asm/intel_pt.h>
  71 #include <clocksource/hyperv_timer.h>
  72
  73 #define CREATE_TRACE_POINTS
  74 #include "trace.h"
  75
  76 #define MAX_IO_MSRS 256
  77 #define KVM_MAX_MCE_BANKS 32
  78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
  79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
  80
  81 #define emul_to_vcpu(ctxt) \
  82         container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
  83
  84 /* EFER defaults:
  85  * - enable syscall per default because its emulated by KVM
  86  * - enable LME and LMA per default on 64 bit KVM
  87  */
  88 #ifdef CONFIG_X86_64
  89 static
  90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  91 #else
  92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  93 #endif
  94
  95 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  97
  98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
  99                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
 100
 101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
 102 static void process_nmi(struct kvm_vcpu *vcpu);
 103 static void enter_smm(struct kvm_vcpu *vcpu);
 104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
 105 static void store_regs(struct kvm_vcpu *vcpu);
 106 static int sync_regs(struct kvm_vcpu *vcpu);
 107
 108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
 109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
 110
 111 static bool __read_mostly ignore_msrs = 0;
 112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
 113
 114 static bool __read_mostly report_ignored_msrs = true;
 115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
 116
 117 unsigned int min_timer_period_us = 200;
 118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
 119
 120 static bool __read_mostly kvmclock_periodic_sync = true;
 121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
 122
 123 bool __read_mostly kvm_has_tsc_control;
 124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
 125 u32  __read_mostly kvm_max_guest_tsc_khz;
 126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
 127 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
 128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
 129 u64  __read_mostly kvm_max_tsc_scaling_ratio;
 130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
 131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
 132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
 133
 134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
 135 static u32 __read_mostly tsc_tolerance_ppm = 250;
 136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
 137
 138 /*
 139  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
 140  * adaptive tuning starting from default advancment of 1000ns.  '0' disables
 141  * advancement entirely.  Any other value is used as-is and disables adaptive
 142  * tuning, i.e. allows priveleged userspace to set an exact advancement time.
 143  */
 144 static int __read_mostly lapic_timer_advance_ns = -1;
 145 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
 146
 147 static bool __read_mostly vector_hashing = true;
 148 module_param(vector_hashing, bool, S_IRUGO);
 149
 150 bool __read_mostly enable_vmware_backdoor = false;
 151 module_param(enable_vmware_backdoor, bool, S_IRUGO);
 152 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
 153
 154 static bool __read_mostly force_emulation_prefix = false;
 155 module_param(force_emulation_prefix, bool, S_IRUGO);
 156
 157 int __read_mostly pi_inject_timer = -1;
 158 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
 159
 160 #define KVM_NR_SHARED_MSRS 16
 161
 162 struct kvm_shared_msrs_global {
 163         int nr;
 164         u32 msrs[KVM_NR_SHARED_MSRS];
 165 };
 166
 167 struct kvm_shared_msrs {
 168         struct user_return_notifier urn;
 169         bool registered;
 170         struct kvm_shared_msr_values {
 171                 u64 host;
 172                 u64 curr;
 173         } values[KVM_NR_SHARED_MSRS];
 174 };
 175
 176 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 177 static struct kvm_shared_msrs __percpu *shared_msrs;
 178
 179 struct kvm_stats_debugfs_item debugfs_entries[] = {
 180         { "pf_fixed", VCPU_STAT(pf_fixed) },
 181         { "pf_guest", VCPU_STAT(pf_guest) },
 182         { "tlb_flush", VCPU_STAT(tlb_flush) },
 183         { "invlpg", VCPU_STAT(invlpg) },
 184         { "exits", VCPU_STAT(exits) },
 185         { "io_exits", VCPU_STAT(io_exits) },
 186         { "mmio_exits", VCPU_STAT(mmio_exits) },
 187         { "signal_exits", VCPU_STAT(signal_exits) },
 188         { "irq_window", VCPU_STAT(irq_window_exits) },
 189         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 190         { "halt_exits", VCPU_STAT(halt_exits) },
 191         { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
 192         { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
 193         { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
 194         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 195         { "hypercalls", VCPU_STAT(hypercalls) },
 196         { "request_irq", VCPU_STAT(request_irq_exits) },
 197         { "irq_exits", VCPU_STAT(irq_exits) },
 198         { "host_state_reload", VCPU_STAT(host_state_reload) },
 199         { "fpu_reload", VCPU_STAT(fpu_reload) },
 200         { "insn_emulation", VCPU_STAT(insn_emulation) },
 201         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 202         { "irq_injections", VCPU_STAT(irq_injections) },
 203         { "nmi_injections", VCPU_STAT(nmi_injections) },
 204         { "req_event", VCPU_STAT(req_event) },
 205         { "l1d_flush", VCPU_STAT(l1d_flush) },
 206         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 207         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 208         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 209         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 210         { "mmu_flooded", VM_STAT(mmu_flooded) },
 211         { "mmu_recycled", VM_STAT(mmu_recycled) },
 212         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 213         { "mmu_unsync", VM_STAT(mmu_unsync) },
 214         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 215         { "largepages", VM_STAT(lpages) },
 216         { "max_mmu_page_hash_collisions",
 217                 VM_STAT(max_mmu_page_hash_collisions) },
 218         { NULL }
 219 };
 220
 221 u64 __read_mostly host_xcr0;
 222
 223 struct kmem_cache *x86_fpu_cache;
 224 EXPORT_SYMBOL_GPL(x86_fpu_cache);
 225
 226 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
 227
 228 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 229 {
 230         int i;
 231         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 232                 vcpu->arch.apf.gfns[i] = ~0;
 233 }
 234
 235 static void kvm_on_user_return(struct user_return_notifier *urn)
 236 {
 237         unsigned slot;
 238         struct kvm_shared_msrs *locals
 239                 = container_of(urn, struct kvm_shared_msrs, urn);
 240         struct kvm_shared_msr_values *values;
 241         unsigned long flags;
 242
 243         /*
 244          * Disabling irqs at this point since the following code could be
 245          * interrupted and executed through kvm_arch_hardware_disable()
 246          */
 247         local_irq_save(flags);
 248         if (locals->registered) {
 249                 locals->registered = false;
 250                 user_return_notifier_unregister(urn);
 251         }
 252         local_irq_restore(flags);
 253         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 254                 values = &locals->values[slot];
 255                 if (values->host != values->curr) {
 256                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 257                         values->curr = values->host;
 258                 }
 259         }
 260 }
 261
 262 static void shared_msr_update(unsigned slot, u32 msr)
 263 {
 264         u64 value;
 265         unsigned int cpu = smp_processor_id();
 266         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 267
 268         /* only read, and nobody should modify it at this time,
 269          * so don't need lock */
 270         if (slot >= shared_msrs_global.nr) {
 271                 printk(KERN_ERR "kvm: invalid MSR slot!");
 272                 return;
 273         }
 274         rdmsrl_safe(msr, &value);
 275         smsr->values[slot].host = value;
 276         smsr->values[slot].curr = value;
 277 }
 278
 279 void kvm_define_shared_msr(unsigned slot, u32 msr)
 280 {
 281         BUG_ON(slot >= KVM_NR_SHARED_MSRS);
 282         shared_msrs_global.msrs[slot] = msr;
 283         if (slot >= shared_msrs_global.nr)
 284                 shared_msrs_global.nr = slot + 1;
 285 }
 286 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 287
 288 static void kvm_shared_msr_cpu_online(void)
 289 {
 290         unsigned i;
 291
 292         for (i = 0; i < shared_msrs_global.nr; ++i)
 293                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 294 }
 295
 296 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 297 {
 298         unsigned int cpu = smp_processor_id();
 299         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 300         int err;
 301
 302         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 303                 return 0;
 304         smsr->values[slot].curr = value;
 305         err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
 306         if (err)
 307                 return 1;
 308
 309         if (!smsr->registered) {
 310                 smsr->urn.on_user_return = kvm_on_user_return;
 311                 user_return_notifier_register(&smsr->urn);
 312                 smsr->registered = true;
 313         }
 314         return 0;
 315 }
 316 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 317
 318 static void drop_user_return_notifiers(void)
 319 {
 320         unsigned int cpu = smp_processor_id();
 321         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 322
 323         if (smsr->registered)
 324                 kvm_on_user_return(&smsr->urn);
 325 }
 326
 327 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 328 {
 329         return vcpu->arch.apic_base;
 330 }
 331 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 332
 333 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
 334 {
 335         return kvm_apic_mode(kvm_get_apic_base(vcpu));
 336 }
 337 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
 338
 339 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 340 {
 341         enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
 342         enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
 343         u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
 344                 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
 345
 346         if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
 347                 return 1;
 348         if (!msr_info->host_initiated) {
 349                 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
 350                         return 1;
 351                 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
 352                         return 1;
 353         }
 354
 355         kvm_lapic_set_base(vcpu, msr_info->data);
 356         return 0;
 357 }
 358 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 359
 360 asmlinkage __visible void kvm_spurious_fault(void)
 361 {
 362         /* Fault while not rebooting.  We want the trace. */
 363         BUG();
 364 }
 365 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
 366
 367 #define EXCPT_BENIGN            0
 368 #define EXCPT_CONTRIBUTORY      1
 369 #define EXCPT_PF                2
 370
 371 static int exception_class(int vector)
 372 {
 373         switch (vector) {
 374         case PF_VECTOR:
 375                 return EXCPT_PF;
 376         case DE_VECTOR:
 377         case TS_VECTOR:
 378         case NP_VECTOR:
 379         case SS_VECTOR:
 380         case GP_VECTOR:
 381                 return EXCPT_CONTRIBUTORY;
 382         default:
 383                 break;
 384         }
 385         return EXCPT_BENIGN;
 386 }
 387
 388 #define EXCPT_FAULT             0
 389 #define EXCPT_TRAP              1
 390 #define EXCPT_ABORT             2
 391 #define EXCPT_INTERRUPT         3
 392
 393 static int exception_type(int vector)
 394 {
 395         unsigned int mask;
 396
 397         if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
 398                 return EXCPT_INTERRUPT;
 399
 400         mask = 1 << vector;
 401
 402         /* #DB is trap, as instruction watchpoints are handled elsewhere */
 403         if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
 404                 return EXCPT_TRAP;
 405
 406         if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
 407                 return EXCPT_ABORT;
 408
 409         /* Reserved exceptions will result in fault */
 410         return EXCPT_FAULT;
 411 }
 412
 413 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
 414 {
 415         unsigned nr = vcpu->arch.exception.nr;
 416         bool has_payload = vcpu->arch.exception.has_payload;
 417         unsigned long payload = vcpu->arch.exception.payload;
 418
 419         if (!has_payload)
 420                 return;
 421
 422         switch (nr) {
 423         case DB_VECTOR:
 424                 /*
 425                  * "Certain debug exceptions may clear bit 0-3.  The
 426                  * remaining contents of the DR6 register are never
 427                  * cleared by the processor".
 428                  */
 429                 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
 430                 /*
 431                  * DR6.RTM is set by all #DB exceptions that don't clear it.
 432                  */
 433                 vcpu->arch.dr6 |= DR6_RTM;
 434                 vcpu->arch.dr6 |= payload;
 435                 /*
 436                  * Bit 16 should be set in the payload whenever the #DB
 437                  * exception should clear DR6.RTM. This makes the payload
 438                  * compatible with the pending debug exceptions under VMX.
 439                  * Though not currently documented in the SDM, this also
 440                  * makes the payload compatible with the exit qualification
 441                  * for #DB exceptions under VMX.
 442                  */
 443                 vcpu->arch.dr6 ^= payload & DR6_RTM;
 444                 break;
 445         case PF_VECTOR:
 446                 vcpu->arch.cr2 = payload;
 447                 break;
 448         }
 449
 450         vcpu->arch.exception.has_payload = false;
 451         vcpu->arch.exception.payload = 0;
 452 }
 453 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
 454
 455 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 456                 unsigned nr, bool has_error, u32 error_code,
 457                 bool has_payload, unsigned long payload, bool reinject)
 458 {
 459         u32 prev_nr;
 460         int class1, class2;
 461
 462         kvm_make_request(KVM_REQ_EVENT, vcpu);
 463
 464         if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
 465         queue:
 466                 if (has_error && !is_protmode(vcpu))
 467                         has_error = false;
 468                 if (reinject) {
 469                         /*
 470                          * On vmentry, vcpu->arch.exception.pending is only
 471                          * true if an event injection was blocked by
 472                          * nested_run_pending.  In that case, however,
 473                          * vcpu_enter_guest requests an immediate exit,
 474                          * and the guest shouldn't proceed far enough to
 475                          * need reinjection.
 476                          */
 477                         WARN_ON_ONCE(vcpu->arch.exception.pending);
 478                         vcpu->arch.exception.injected = true;
 479                         if (WARN_ON_ONCE(has_payload)) {
 480                                 /*
 481                                  * A reinjected event has already
 482                                  * delivered its payload.
 483                                  */
 484                                 has_payload = false;
 485                                 payload = 0;
 486                         }
 487                 } else {
 488                         vcpu->arch.exception.pending = true;
 489                         vcpu->arch.exception.injected = false;
 490                 }
 491                 vcpu->arch.exception.has_error_code = has_error;
 492                 vcpu->arch.exception.nr = nr;
 493                 vcpu->arch.exception.error_code = error_code;
 494                 vcpu->arch.exception.has_payload = has_payload;
 495                 vcpu->arch.exception.payload = payload;
 496                 /*
 497                  * In guest mode, payload delivery should be deferred,
 498                  * so that the L1 hypervisor can intercept #PF before
 499                  * CR2 is modified (or intercept #DB before DR6 is
 500                  * modified under nVMX).  However, for ABI
 501                  * compatibility with KVM_GET_VCPU_EVENTS and
 502                  * KVM_SET_VCPU_EVENTS, we can't delay payload
 503                  * delivery unless userspace has enabled this
 504                  * functionality via the per-VM capability,
 505                  * KVM_CAP_EXCEPTION_PAYLOAD.
 506                  */
 507                 if (!vcpu->kvm->arch.exception_payload_enabled ||
 508                     !is_guest_mode(vcpu))
 509                         kvm_deliver_exception_payload(vcpu);
 510                 return;
 511         }
 512
 513         /* to check exception */
 514         prev_nr = vcpu->arch.exception.nr;
 515         if (prev_nr == DF_VECTOR) {
 516                 /* triple fault -> shutdown */
 517                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 518                 return;
 519         }
 520         class1 = exception_class(prev_nr);
 521         class2 = exception_class(nr);
 522         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 523                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 524                 /*
 525                  * Generate double fault per SDM Table 5-5.  Set
 526                  * exception.pending = true so that the double fault
 527                  * can trigger a nested vmexit.
 528                  */
 529                 vcpu->arch.exception.pending = true;
 530                 vcpu->arch.exception.injected = false;
 531                 vcpu->arch.exception.has_error_code = true;
 532                 vcpu->arch.exception.nr = DF_VECTOR;
 533                 vcpu->arch.exception.error_code = 0;
 534                 vcpu->arch.exception.has_payload = false;
 535                 vcpu->arch.exception.payload = 0;
 536         } else
 537                 /* replace previous exception with a new one in a hope
 538                    that instruction re-execution will regenerate lost
 539                    exception */
 540                 goto queue;
 541 }
 542
 543 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 544 {
 545         kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
 546 }
 547 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 548
 549 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 550 {
 551         kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
 552 }
 553 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 554
 555 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
 556                                   unsigned long payload)
 557 {
 558         kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
 559 }
 560
 561 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
 562                                     u32 error_code, unsigned long payload)
 563 {
 564         kvm_multiple_exception(vcpu, nr, true, error_code,
 565                                true, payload, false);
 566 }
 567
 568 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 569 {
 570         if (err)
 571                 kvm_inject_gp(vcpu, 0);
 572         else
 573                 return kvm_skip_emulated_instruction(vcpu);
 574
 575         return 1;
 576 }
 577 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 578
 579 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 580 {
 581         ++vcpu->stat.pf_guest;
 582         vcpu->arch.exception.nested_apf =
 583                 is_guest_mode(vcpu) && fault->async_page_fault;
 584         if (vcpu->arch.exception.nested_apf) {
 585                 vcpu->arch.apf.nested_apf_token = fault->address;
 586                 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 587         } else {
 588                 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
 589                                         fault->address);
 590         }
 591 }
 592 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
 593
 594 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 595 {
 596         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 597                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 598         else
 599                 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
 600
 601         return fault->nested_page_fault;
 602 }
 603
 604 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 605 {
 606         atomic_inc(&vcpu->arch.nmi_queued);
 607         kvm_make_request(KVM_REQ_NMI, vcpu);
 608 }
 609 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 610
 611 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 612 {
 613         kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
 614 }
 615 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 616
 617 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 618 {
 619         kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
 620 }
 621 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 622
 623 /*
 624  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 625  * a #GP and return false.
 626  */
 627 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 628 {
 629         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 630                 return true;
 631         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 632         return false;
 633 }
 634 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 635
 636 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
 637 {
 638         if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 639                 return true;
 640
 641         kvm_queue_exception(vcpu, UD_VECTOR);
 642         return false;
 643 }
 644 EXPORT_SYMBOL_GPL(kvm_require_dr);
 645
 646 /*
 647  * This function will be used to read from the physical memory of the currently
 648  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
 649  * can read from guest physical or from the guest's guest physical memory.
 650  */
 651 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 652                             gfn_t ngfn, void *data, int offset, int len,
 653                             u32 access)
 654 {
 655         struct x86_exception exception;
 656         gfn_t real_gfn;
 657         gpa_t ngpa;
 658
 659         ngpa     = gfn_to_gpa(ngfn);
 660         real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
 661         if (real_gfn == UNMAPPED_GVA)
 662                 return -EFAULT;
 663
 664         real_gfn = gpa_to_gfn(real_gfn);
 665
 666         return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
 667 }
 668 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 669
 670 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 671                                void *data, int offset, int len, u32 access)
 672 {
 673         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 674                                        data, offset, len, access);
 675 }
 676
 677 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
 678 {
 679         return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
 680                rsvd_bits(1, 2);
 681 }
 682
 683 /*
 684  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
 685  */
 686 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 687 {
 688         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 689         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 690         int i;
 691         int ret;
 692         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 693
 694         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 695                                       offset * sizeof(u64), sizeof(pdpte),
 696                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 697         if (ret < 0) {
 698                 ret = 0;
 699                 goto out;
 700         }
 701         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 702                 if ((pdpte[i] & PT_PRESENT_MASK) &&
 703                     (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
 704                         ret = 0;
 705                         goto out;
 706                 }
 707         }
 708         ret = 1;
 709
 710         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 711         __set_bit(VCPU_EXREG_PDPTR,
 712                   (unsigned long *)&vcpu->arch.regs_avail);
 713         __set_bit(VCPU_EXREG_PDPTR,
 714                   (unsigned long *)&vcpu->arch.regs_dirty);
 715 out:
 716
 717         return ret;
 718 }
 719 EXPORT_SYMBOL_GPL(load_pdptrs);
 720
 721 bool pdptrs_changed(struct kvm_vcpu *vcpu)
 722 {
 723         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 724         bool changed = true;
 725         int offset;
 726         gfn_t gfn;
 727         int r;
 728
 729         if (!is_pae_paging(vcpu))
 730                 return false;
 731
 732         if (!test_bit(VCPU_EXREG_PDPTR,
 733                       (unsigned long *)&vcpu->arch.regs_avail))
 734                 return true;
 735
 736         gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
 737         offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
 738         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 739                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 740         if (r < 0)
 741                 goto out;
 742         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 743 out:
 744
 745         return changed;
 746 }
 747 EXPORT_SYMBOL_GPL(pdptrs_changed);
 748
 749 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 750 {
 751         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 752         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
 753
 754         cr0 |= X86_CR0_ET;
 755
 756 #ifdef CONFIG_X86_64
 757         if (cr0 & 0xffffffff00000000UL)
 758                 return 1;
 759 #endif
 760
 761         cr0 &= ~CR0_RESERVED_BITS;
 762
 763         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 764                 return 1;
 765
 766         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 767                 return 1;
 768
 769         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 770 #ifdef CONFIG_X86_64
 771                 if ((vcpu->arch.efer & EFER_LME)) {
 772                         int cs_db, cs_l;
 773
 774                         if (!is_pae(vcpu))
 775                                 return 1;
 776                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 777                         if (cs_l)
 778                                 return 1;
 779                 } else
 780 #endif
 781                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 782                                                  kvm_read_cr3(vcpu)))
 783                         return 1;
 784         }
 785
 786         if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
 787                 return 1;
 788
 789         kvm_x86_ops->set_cr0(vcpu, cr0);
 790
 791         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 792                 kvm_clear_async_pf_completion_queue(vcpu);
 793                 kvm_async_pf_hash_reset(vcpu);
 794         }
 795
 796         if ((cr0 ^ old_cr0) & update_bits)
 797                 kvm_mmu_reset_context(vcpu);
 798
 799         if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
 800             kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
 801             !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
 802                 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
 803
 804         return 0;
 805 }
 806 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 807
 808 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 809 {
 810         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 811 }
 812 EXPORT_SYMBOL_GPL(kvm_lmsw);
 813
 814 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 815 {
 816         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
 817                         !vcpu->guest_xcr0_loaded) {
 818                 /* kvm_set_xcr() also depends on this */
 819                 if (vcpu->arch.xcr0 != host_xcr0)
 820                         xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
 821                 vcpu->guest_xcr0_loaded = 1;
 822         }
 823 }
 824 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
 825
 826 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 827 {
 828         if (vcpu->guest_xcr0_loaded) {
 829                 if (vcpu->arch.xcr0 != host_xcr0)
 830                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
 831                 vcpu->guest_xcr0_loaded = 0;
 832         }
 833 }
 834 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
 835
 836 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 837 {
 838         u64 xcr0 = xcr;
 839         u64 old_xcr0 = vcpu->arch.xcr0;
 840         u64 valid_bits;
 841
 842         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 843         if (index != XCR_XFEATURE_ENABLED_MASK)
 844                 return 1;
 845         if (!(xcr0 & XFEATURE_MASK_FP))
 846                 return 1;
 847         if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
 848                 return 1;
 849
 850         /*
 851          * Do not allow the guest to set bits that we do not support
 852          * saving.  However, xcr0 bit 0 is always set, even if the
 853          * emulated CPU does not support XSAVE (see fx_init).
 854          */
 855         valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
 856         if (xcr0 & ~valid_bits)
 857                 return 1;
 858
 859         if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
 860             (!(xcr0 & XFEATURE_MASK_BNDCSR)))
 861                 return 1;
 862
 863         if (xcr0 & XFEATURE_MASK_AVX512) {
 864                 if (!(xcr0 & XFEATURE_MASK_YMM))
 865                         return 1;
 866                 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
 867                         return 1;
 868         }
 869         vcpu->arch.xcr0 = xcr0;
 870
 871         if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
 872                 kvm_update_cpuid(vcpu);
 873         return 0;
 874 }
 875
 876 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 877 {
 878         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
 879             __kvm_set_xcr(vcpu, index, xcr)) {
 880                 kvm_inject_gp(vcpu, 0);
 881                 return 1;
 882         }
 883         return 0;
 884 }
 885 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 886
 887 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 888 {
 889         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 890         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
 891                                    X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
 892
 893         if (cr4 & CR4_RESERVED_BITS)
 894                 return 1;
 895
 896         if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
 897                 return 1;
 898
 899         if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
 900                 return 1;
 901
 902         if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
 903                 return 1;
 904
 905         if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
 906                 return 1;
 907
 908         if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
 909                 return 1;
 910
 911         if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
 912                 return 1;
 913
 914         if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
 915                 return 1;
 916
 917         if (is_long_mode(vcpu)) {
 918                 if (!(cr4 & X86_CR4_PAE))
 919                         return 1;
 920         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 921                    && ((cr4 ^ old_cr4) & pdptr_bits)
 922                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 923                                    kvm_read_cr3(vcpu)))
 924                 return 1;
 925
 926         if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
 927                 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
 928                         return 1;
 929
 930                 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
 931                 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
 932                         return 1;
 933         }
 934
 935         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 936                 return 1;
 937
 938         if (((cr4 ^ old_cr4) & pdptr_bits) ||
 939             (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
 940                 kvm_mmu_reset_context(vcpu);
 941
 942         if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
 943                 kvm_update_cpuid(vcpu);
 944
 945         return 0;
 946 }
 947 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 948
 949 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 950 {
 951         bool skip_tlb_flush = false;
 952 #ifdef CONFIG_X86_64
 953         bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
 954
 955         if (pcid_enabled) {
 956                 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
 957                 cr3 &= ~X86_CR3_PCID_NOFLUSH;
 958         }
 959 #endif
 960
 961         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 962                 if (!skip_tlb_flush) {
 963                         kvm_mmu_sync_roots(vcpu);
 964                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 965                 }
 966                 return 0;
 967         }
 968
 969         if (is_long_mode(vcpu) &&
 970             (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
 971                 return 1;
 972         else if (is_pae_paging(vcpu) &&
 973                  !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 974                 return 1;
 975
 976         kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
 977         vcpu->arch.cr3 = cr3;
 978         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 979
 980         return 0;
 981 }
 982 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 983
 984 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 985 {
 986         if (cr8 & CR8_RESERVED_BITS)
 987                 return 1;
 988         if (lapic_in_kernel(vcpu))
 989                 kvm_lapic_set_tpr(vcpu, cr8);
 990         else
 991                 vcpu->arch.cr8 = cr8;
 992         return 0;
 993 }
 994 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 995
 996 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 997 {
 998         if (lapic_in_kernel(vcpu))
 999                 return kvm_lapic_get_cr8(vcpu);
1000         else
1001                 return vcpu->arch.cr8;
1002 }
1003 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1004
1005 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1006 {
1007         int i;
1008
1009         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1010                 for (i = 0; i < KVM_NR_DB_REGS; i++)
1011                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1012                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1013         }
1014 }
1015
1016 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1017 {
1018         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1019                 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1020 }
1021
1022 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1023 {
1024         unsigned long dr7;
1025
1026         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1027                 dr7 = vcpu->arch.guest_debug_dr7;
1028         else
1029                 dr7 = vcpu->arch.dr7;
1030         kvm_x86_ops->set_dr7(vcpu, dr7);
1031         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1032         if (dr7 & DR7_BP_EN_MASK)
1033                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1034 }
1035
1036 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1037 {
1038         u64 fixed = DR6_FIXED_1;
1039
1040         if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1041                 fixed |= DR6_RTM;
1042         return fixed;
1043 }
1044
1045 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1046 {
1047         switch (dr) {
1048         case 0 ... 3:
1049                 vcpu->arch.db[dr] = val;
1050                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1051                         vcpu->arch.eff_db[dr] = val;
1052                 break;
1053         case 4:
1054                 /* fall through */
1055         case 6:
1056                 if (val & 0xffffffff00000000ULL)
1057                         return -1; /* #GP */
1058                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1059                 kvm_update_dr6(vcpu);
1060                 break;
1061         case 5:
1062                 /* fall through */
1063         default: /* 7 */
1064                 if (val & 0xffffffff00000000ULL)
1065                         return -1; /* #GP */
1066                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1067                 kvm_update_dr7(vcpu);
1068                 break;
1069         }
1070
1071         return 0;
1072 }
1073
1074 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1075 {
1076         if (__kvm_set_dr(vcpu, dr, val)) {
1077                 kvm_inject_gp(vcpu, 0);
1078                 return 1;
1079         }
1080         return 0;
1081 }
1082 EXPORT_SYMBOL_GPL(kvm_set_dr);
1083
1084 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1085 {
1086         switch (dr) {
1087         case 0 ... 3:
1088                 *val = vcpu->arch.db[dr];
1089                 break;
1090         case 4:
1091                 /* fall through */
1092         case 6:
1093                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1094                         *val = vcpu->arch.dr6;
1095                 else
1096                         *val = kvm_x86_ops->get_dr6(vcpu);
1097                 break;
1098         case 5:
1099                 /* fall through */
1100         default: /* 7 */
1101                 *val = vcpu->arch.dr7;
1102                 break;
1103         }
1104         return 0;
1105 }
1106 EXPORT_SYMBOL_GPL(kvm_get_dr);
1107
1108 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1109 {
1110         u32 ecx = kvm_rcx_read(vcpu);
1111         u64 data;
1112         int err;
1113
1114         err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1115         if (err)
1116                 return err;
1117         kvm_rax_write(vcpu, (u32)data);
1118         kvm_rdx_write(vcpu, data >> 32);
1119         return err;
1120 }
1121 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1122
1123 /*
1124  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1125  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1126  *
1127  * This list is modified at module load time to reflect the
1128  * capabilities of the host cpu. This capabilities test skips MSRs that are
1129  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1130  * may depend on host virtualization features rather than host cpu features.
1131  */
1132
1133 static u32 msrs_to_save[] = {
1134         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1135         MSR_STAR,
1136 #ifdef CONFIG_X86_64
1137         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1138 #endif
1139         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1140         MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1141         MSR_IA32_SPEC_CTRL,
1142         MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1143         MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1144         MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1145         MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1146         MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1147         MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1148 };
1149
1150 static unsigned num_msrs_to_save;
1151
1152 static u32 emulated_msrs[] = {
1153         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1154         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1155         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1156         HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1157         HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1158         HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1159         HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1160         HV_X64_MSR_RESET,
1161         HV_X64_MSR_VP_INDEX,
1162         HV_X64_MSR_VP_RUNTIME,
1163         HV_X64_MSR_SCONTROL,
1164         HV_X64_MSR_STIMER0_CONFIG,
1165         HV_X64_MSR_VP_ASSIST_PAGE,
1166         HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1167         HV_X64_MSR_TSC_EMULATION_STATUS,
1168
1169         MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1170         MSR_KVM_PV_EOI_EN,
1171
1172         MSR_IA32_TSC_ADJUST,
1173         MSR_IA32_TSCDEADLINE,
1174         MSR_IA32_ARCH_CAPABILITIES,
1175         MSR_IA32_MISC_ENABLE,
1176         MSR_IA32_MCG_STATUS,
1177         MSR_IA32_MCG_CTL,
1178         MSR_IA32_MCG_EXT_CTL,
1179         MSR_IA32_SMBASE,
1180         MSR_SMI_COUNT,
1181         MSR_PLATFORM_INFO,
1182         MSR_MISC_FEATURES_ENABLES,
1183         MSR_AMD64_VIRT_SPEC_CTRL,
1184         MSR_IA32_POWER_CTL,
1185
1186         /*
1187          * The following list leaves out MSRs whose values are determined
1188          * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1189          * We always support the "true" VMX control MSRs, even if the host
1190          * processor does not, so I am putting these registers here rather
1191          * than in msrs_to_save.
1192          */
1193         MSR_IA32_VMX_BASIC,
1194         MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1195         MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1196         MSR_IA32_VMX_TRUE_EXIT_CTLS,
1197         MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1198         MSR_IA32_VMX_MISC,
1199         MSR_IA32_VMX_CR0_FIXED0,
1200         MSR_IA32_VMX_CR4_FIXED0,
1201         MSR_IA32_VMX_VMCS_ENUM,
1202         MSR_IA32_VMX_PROCBASED_CTLS2,
1203         MSR_IA32_VMX_EPT_VPID_CAP,
1204         MSR_IA32_VMX_VMFUNC,
1205
1206         MSR_K7_HWCR,
1207         MSR_KVM_POLL_CONTROL,
1208 };
1209
1210 static unsigned num_emulated_msrs;
1211
1212 /*
1213  * List of msr numbers which are used to expose MSR-based features that
1214  * can be used by a hypervisor to validate requested CPU features.
1215  */
1216 static u32 msr_based_features[] = {
1217         MSR_IA32_VMX_BASIC,
1218         MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1219         MSR_IA32_VMX_PINBASED_CTLS,
1220         MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1221         MSR_IA32_VMX_PROCBASED_CTLS,
1222         MSR_IA32_VMX_TRUE_EXIT_CTLS,
1223         MSR_IA32_VMX_EXIT_CTLS,
1224         MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1225         MSR_IA32_VMX_ENTRY_CTLS,
1226         MSR_IA32_VMX_MISC,
1227         MSR_IA32_VMX_CR0_FIXED0,
1228         MSR_IA32_VMX_CR0_FIXED1,
1229         MSR_IA32_VMX_CR4_FIXED0,
1230         MSR_IA32_VMX_CR4_FIXED1,
1231         MSR_IA32_VMX_VMCS_ENUM,
1232         MSR_IA32_VMX_PROCBASED_CTLS2,
1233         MSR_IA32_VMX_EPT_VPID_CAP,
1234         MSR_IA32_VMX_VMFUNC,
1235
1236         MSR_F10H_DECFG,
1237         MSR_IA32_UCODE_REV,
1238         MSR_IA32_ARCH_CAPABILITIES,
1239 };
1240
1241 static unsigned int num_msr_based_features;
1242
1243 static u64 kvm_get_arch_capabilities(void)
1244 {
1245         u64 data = 0;
1246
1247         if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1248                 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1249
1250         /*
1251          * If we're doing cache flushes (either "always" or "cond")
1252          * we will do one whenever the guest does a vmlaunch/vmresume.
1253          * If an outer hypervisor is doing the cache flush for us
1254          * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1255          * capability to the guest too, and if EPT is disabled we're not
1256          * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1257          * require a nested hypervisor to do a flush of its own.
1258          */
1259         if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1260                 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1261
1262         if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1263                 data |= ARCH_CAP_RDCL_NO;
1264         if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1265                 data |= ARCH_CAP_SSB_NO;
1266         if (!boot_cpu_has_bug(X86_BUG_MDS))
1267                 data |= ARCH_CAP_MDS_NO;
1268
1269         return data;
1270 }
1271
1272 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1273 {
1274         switch (msr->index) {
1275         case MSR_IA32_ARCH_CAPABILITIES:
1276                 msr->data = kvm_get_arch_capabilities();
1277                 break;
1278         case MSR_IA32_UCODE_REV:
1279                 rdmsrl_safe(msr->index, &msr->data);
1280                 break;
1281         default:
1282                 if (kvm_x86_ops->get_msr_feature(msr))
1283                         return 1;
1284         }
1285         return 0;
1286 }
1287
1288 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1289 {
1290         struct kvm_msr_entry msr;
1291         int r;
1292
1293         msr.index = index;
1294         r = kvm_get_msr_feature(&msr);
1295         if (r)
1296                 return r;
1297
1298         *data = msr.data;
1299
1300         return 0;
1301 }
1302
1303 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1304 {
1305         if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1306                 return false;
1307
1308         if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1309                 return false;
1310
1311         if (efer & (EFER_LME | EFER_LMA) &&
1312             !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1313                 return false;
1314
1315         if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1316                 return false;
1317
1318         return true;
1319
1320 }
1321 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1322 {
1323         if (efer & efer_reserved_bits)
1324                 return false;
1325
1326         return __kvm_valid_efer(vcpu, efer);
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1329
1330 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1331 {
1332         u64 old_efer = vcpu->arch.efer;
1333         u64 efer = msr_info->data;
1334
1335         if (efer & efer_reserved_bits)
1336                 return 1;
1337
1338         if (!msr_info->host_initiated) {
1339                 if (!__kvm_valid_efer(vcpu, efer))
1340                         return 1;
1341
1342                 if (is_paging(vcpu) &&
1343                     (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1344                         return 1;
1345         }
1346
1347         efer &= ~EFER_LMA;
1348         efer |= vcpu->arch.efer & EFER_LMA;
1349
1350         kvm_x86_ops->set_efer(vcpu, efer);
1351
1352         /* Update reserved bits */
1353         if ((efer ^ old_efer) & EFER_NX)
1354                 kvm_mmu_reset_context(vcpu);
1355
1356         return 0;
1357 }
1358
1359 void kvm_enable_efer_bits(u64 mask)
1360 {
1361        efer_reserved_bits &= ~mask;
1362 }
1363 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1364
1365 /*
1366  * Write @data into the MSR specified by @index.  Select MSR specific fault
1367  * checks are bypassed if @host_initiated is %true.
1368  * Returns 0 on success, non-0 otherwise.
1369  * Assumes vcpu_load() was already called.
1370  */
1371 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1372                          bool host_initiated)
1373 {
1374         struct msr_data msr;
1375
1376         switch (index) {
1377         case MSR_FS_BASE:
1378         case MSR_GS_BASE:
1379         case MSR_KERNEL_GS_BASE:
1380         case MSR_CSTAR:
1381         case MSR_LSTAR:
1382                 if (is_noncanonical_address(data, vcpu))
1383                         return 1;
1384                 break;
1385         case MSR_IA32_SYSENTER_EIP:
1386         case MSR_IA32_SYSENTER_ESP:
1387                 /*
1388                  * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1389                  * non-canonical address is written on Intel but not on
1390                  * AMD (which ignores the top 32-bits, because it does
1391                  * not implement 64-bit SYSENTER).
1392                  *
1393                  * 64-bit code should hence be able to write a non-canonical
1394                  * value on AMD.  Making the address canonical ensures that
1395                  * vmentry does not fail on Intel after writing a non-canonical
1396                  * value, and that something deterministic happens if the guest
1397                  * invokes 64-bit SYSENTER.
1398                  */
1399                 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1400         }
1401
1402         msr.data = data;
1403         msr.index = index;
1404         msr.host_initiated = host_initiated;
1405
1406         return kvm_x86_ops->set_msr(vcpu, &msr);
1407 }
1408
1409 /*
1410  * Read the MSR specified by @index into @data.  Select MSR specific fault
1411  * checks are bypassed if @host_initiated is %true.
1412  * Returns 0 on success, non-0 otherwise.
1413  * Assumes vcpu_load() was already called.
1414  */
1415 static int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1416                          bool host_initiated)
1417 {
1418         struct msr_data msr;
1419         int ret;
1420
1421         msr.index = index;
1422         msr.host_initiated = host_initiated;
1423
1424         ret = kvm_x86_ops->get_msr(vcpu, &msr);
1425         if (!ret)
1426                 *data = msr.data;
1427         return ret;
1428 }
1429
1430 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1431 {
1432         return __kvm_get_msr(vcpu, index, data, false);
1433 }
1434 EXPORT_SYMBOL_GPL(kvm_get_msr);
1435
1436 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1437 {
1438         return __kvm_set_msr(vcpu, index, data, false);
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_set_msr);
1441
1442 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1443 {
1444         u32 ecx = kvm_rcx_read(vcpu);
1445         u64 data;
1446
1447         if (kvm_get_msr(vcpu, ecx, &data)) {
1448                 trace_kvm_msr_read_ex(ecx);
1449                 kvm_inject_gp(vcpu, 0);
1450                 return 1;
1451         }
1452
1453         trace_kvm_msr_read(ecx, data);
1454
1455         kvm_rax_write(vcpu, data & -1u);
1456         kvm_rdx_write(vcpu, (data >> 32) & -1u);
1457         return kvm_skip_emulated_instruction(vcpu);
1458 }
1459 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1460
1461 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1462 {
1463         u32 ecx = kvm_rcx_read(vcpu);
1464         u64 data = kvm_read_edx_eax(vcpu);
1465
1466         if (kvm_set_msr(vcpu, ecx, data)) {
1467                 trace_kvm_msr_write_ex(ecx, data);
1468                 kvm_inject_gp(vcpu, 0);
1469                 return 1;
1470         }
1471
1472         trace_kvm_msr_write(ecx, data);
1473         return kvm_skip_emulated_instruction(vcpu);
1474 }
1475 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1476
1477 /*
1478  * Adapt set_msr() to msr_io()'s calling convention
1479  */
1480 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1481 {
1482         return __kvm_get_msr(vcpu, index, data, true);
1483 }
1484
1485 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1486 {
1487         return __kvm_set_msr(vcpu, index, *data, true);
1488 }
1489
1490 #ifdef CONFIG_X86_64
1491 struct pvclock_gtod_data {
1492         seqcount_t      seq;
1493
1494         struct { /* extract of a clocksource struct */
1495                 int vclock_mode;
1496                 u64     cycle_last;
1497                 u64     mask;
1498                 u32     mult;
1499                 u32     shift;
1500         } clock;
1501
1502         u64             boot_ns;
1503         u64             nsec_base;
1504         u64             wall_time_sec;
1505 };
1506
1507 static struct pvclock_gtod_data pvclock_gtod_data;
1508
1509 static void update_pvclock_gtod(struct timekeeper *tk)
1510 {
1511         struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1512         u64 boot_ns;
1513
1514         boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1515
1516         write_seqcount_begin(&vdata->seq);
1517
1518         /* copy pvclock gtod data */
1519         vdata->clock.vclock_mode        = tk->tkr_mono.clock->archdata.vclock_mode;
1520         vdata->clock.cycle_last         = tk->tkr_mono.cycle_last;
1521         vdata->clock.mask               = tk->tkr_mono.mask;
1522         vdata->clock.mult               = tk->tkr_mono.mult;
1523         vdata->clock.shift              = tk->tkr_mono.shift;
1524
1525         vdata->boot_ns                  = boot_ns;
1526         vdata->nsec_base                = tk->tkr_mono.xtime_nsec;
1527
1528         vdata->wall_time_sec            = tk->xtime_sec;
1529
1530         write_seqcount_end(&vdata->seq);
1531 }
1532 #endif
1533
1534 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1535 {
1536         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1537         kvm_vcpu_kick(vcpu);
1538 }
1539
1540 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1541 {
1542         int version;
1543         int r;
1544         struct pvclock_wall_clock wc;
1545         struct timespec64 boot;
1546
1547         if (!wall_clock)
1548                 return;
1549
1550         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1551         if (r)
1552                 return;
1553
1554         if (version & 1)
1555                 ++version;  /* first time write, random junk */
1556
1557         ++version;
1558
1559         if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1560                 return;
1561
1562         /*
1563          * The guest calculates current wall clock time by adding
1564          * system time (updated by kvm_guest_time_update below) to the
1565          * wall clock specified here.  guest system time equals host
1566          * system time for us, thus we must fill in host boot time here.
1567          */
1568         getboottime64(&boot);
1569
1570         if (kvm->arch.kvmclock_offset) {
1571                 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1572                 boot = timespec64_sub(boot, ts);
1573         }
1574         wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1575         wc.nsec = boot.tv_nsec;
1576         wc.version = version;
1577
1578         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1579
1580         version++;
1581         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1582 }
1583
1584 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1585 {
1586         do_shl32_div32(dividend, divisor);
1587         return dividend;
1588 }
1589
1590 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1591                                s8 *pshift, u32 *pmultiplier)
1592 {
1593         uint64_t scaled64;
1594         int32_t  shift = 0;
1595         uint64_t tps64;
1596         uint32_t tps32;
1597
1598         tps64 = base_hz;
1599         scaled64 = scaled_hz;
1600         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1601                 tps64 >>= 1;
1602                 shift--;
1603         }
1604
1605         tps32 = (uint32_t)tps64;
1606         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1607                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1608                         scaled64 >>= 1;
1609                 else
1610                         tps32 <<= 1;
1611                 shift++;
1612         }
1613
1614         *pshift = shift;
1615         *pmultiplier = div_frac(scaled64, tps32);
1616 }
1617
1618 #ifdef CONFIG_X86_64
1619 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1620 #endif
1621
1622 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1623 static unsigned long max_tsc_khz;
1624
1625 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1626 {
1627         u64 v = (u64)khz * (1000000 + ppm);
1628         do_div(v, 1000000);
1629         return v;
1630 }
1631
1632 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1633 {
1634         u64 ratio;
1635
1636         /* Guest TSC same frequency as host TSC? */
1637         if (!scale) {
1638                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1639                 return 0;
1640         }
1641
1642         /* TSC scaling supported? */
1643         if (!kvm_has_tsc_control) {
1644                 if (user_tsc_khz > tsc_khz) {
1645                         vcpu->arch.tsc_catchup = 1;
1646                         vcpu->arch.tsc_always_catchup = 1;
1647                         return 0;
1648                 } else {
1649                         pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1650                         return -1;
1651                 }
1652         }
1653
1654         /* TSC scaling required  - calculate ratio */
1655         ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1656                                 user_tsc_khz, tsc_khz);
1657
1658         if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1659                 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1660                                     user_tsc_khz);
1661                 return -1;
1662         }
1663
1664         vcpu->arch.tsc_scaling_ratio = ratio;
1665         return 0;
1666 }
1667
1668 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1669 {
1670         u32 thresh_lo, thresh_hi;
1671         int use_scaling = 0;
1672
1673         /* tsc_khz can be zero if TSC calibration fails */
1674         if (user_tsc_khz == 0) {
1675                 /* set tsc_scaling_ratio to a safe value */
1676                 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1677                 return -1;
1678         }
1679
1680         /* Compute a scale to convert nanoseconds in TSC cycles */
1681         kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1682                            &vcpu->arch.virtual_tsc_shift,
1683                            &vcpu->arch.virtual_tsc_mult);
1684         vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1685
1686         /*
1687          * Compute the variation in TSC rate which is acceptable
1688          * within the range of tolerance and decide if the
1689          * rate being applied is within that bounds of the hardware
1690          * rate.  If so, no scaling or compensation need be done.
1691          */
1692         thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1693         thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1694         if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1695                 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1696                 use_scaling = 1;
1697         }
1698         return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1699 }
1700
1701 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1702 {
1703         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1704                                       vcpu->arch.virtual_tsc_mult,
1705                                       vcpu->arch.virtual_tsc_shift);
1706         tsc += vcpu->arch.this_tsc_write;
1707         return tsc;
1708 }
1709
1710 static inline int gtod_is_based_on_tsc(int mode)
1711 {
1712         return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1713 }
1714
1715 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1716 {
1717 #ifdef CONFIG_X86_64
1718         bool vcpus_matched;
1719         struct kvm_arch *ka = &vcpu->kvm->arch;
1720         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1721
1722         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1723                          atomic_read(&vcpu->kvm->online_vcpus));
1724
1725         /*
1726          * Once the masterclock is enabled, always perform request in
1727          * order to update it.
1728          *
1729          * In order to enable masterclock, the host clocksource must be TSC
1730          * and the vcpus need to have matched TSCs.  When that happens,
1731          * perform request to enable masterclock.
1732          */
1733         if (ka->use_master_clock ||
1734             (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1735                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1736
1737         trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1738                             atomic_read(&vcpu->kvm->online_vcpus),
1739                             ka->use_master_clock, gtod->clock.vclock_mode);
1740 #endif
1741 }
1742
1743 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1744 {
1745         u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1746         vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1747 }
1748
1749 /*
1750  * Multiply tsc by a fixed point number represented by ratio.
1751  *
1752  * The most significant 64-N bits (mult) of ratio represent the
1753  * integral part of the fixed point number; the remaining N bits
1754  * (frac) represent the fractional part, ie. ratio represents a fixed
1755  * point number (mult + frac * 2^(-N)).
1756  *
1757  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1758  */
1759 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1760 {
1761         return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1762 }
1763
1764 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1765 {
1766         u64 _tsc = tsc;
1767         u64 ratio = vcpu->arch.tsc_scaling_ratio;
1768
1769         if (ratio != kvm_default_tsc_scaling_ratio)
1770                 _tsc = __scale_tsc(ratio, tsc);
1771
1772         return _tsc;
1773 }
1774 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1775
1776 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1777 {
1778         u64 tsc;
1779
1780         tsc = kvm_scale_tsc(vcpu, rdtsc());
1781
1782         return target_tsc - tsc;
1783 }
1784
1785 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1786 {
1787         u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1788
1789         return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1790 }
1791 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1792
1793 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1794 {
1795         vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1796 }
1797
1798 static inline bool kvm_check_tsc_unstable(void)
1799 {
1800 #ifdef CONFIG_X86_64
1801         /*
1802          * TSC is marked unstable when we're running on Hyper-V,
1803          * 'TSC page' clocksource is good.
1804          */
1805         if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1806                 return false;
1807 #endif
1808         return check_tsc_unstable();
1809 }
1810
1811 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1812 {
1813         struct kvm *kvm = vcpu->kvm;
1814         u64 offset, ns, elapsed;
1815         unsigned long flags;
1816         bool matched;
1817         bool already_matched;
1818         u64 data = msr->data;
1819         bool synchronizing = false;
1820
1821         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1822         offset = kvm_compute_tsc_offset(vcpu, data);
1823         ns = ktime_get_boottime_ns();
1824         elapsed = ns - kvm->arch.last_tsc_nsec;
1825
1826         if (vcpu->arch.virtual_tsc_khz) {
1827                 if (data == 0 && msr->host_initiated) {
1828                         /*
1829                          * detection of vcpu initialization -- need to sync
1830                          * with other vCPUs. This particularly helps to keep
1831                          * kvm_clock stable after CPU hotplug
1832                          */
1833                         synchronizing = true;
1834                 } else {
1835                         u64 tsc_exp = kvm->arch.last_tsc_write +
1836                                                 nsec_to_cycles(vcpu, elapsed);
1837                         u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1838                         /*
1839                          * Special case: TSC write with a small delta (1 second)
1840                          * of virtual cycle time against real time is
1841                          * interpreted as an attempt to synchronize the CPU.
1842                          */
1843                         synchronizing = data < tsc_exp + tsc_hz &&
1844                                         data + tsc_hz > tsc_exp;
1845                 }
1846         }
1847
1848         /*
1849          * For a reliable TSC, we can match TSC offsets, and for an unstable
1850          * TSC, we add elapsed time in this computation.  We could let the
1851          * compensation code attempt to catch up if we fall behind, but
1852          * it's better to try to match offsets from the beginning.
1853          */
1854         if (synchronizing &&
1855             vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1856                 if (!kvm_check_tsc_unstable()) {
1857                         offset = kvm->arch.cur_tsc_offset;
1858                 } else {
1859                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1860                         data += delta;
1861                         offset = kvm_compute_tsc_offset(vcpu, data);
1862                 }
1863                 matched = true;
1864                 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1865         } else {
1866                 /*
1867                  * We split periods of matched TSC writes into generations.
1868                  * For each generation, we track the original measured
1869                  * nanosecond time, offset, and write, so if TSCs are in
1870                  * sync, we can match exact offset, and if not, we can match
1871                  * exact software computation in compute_guest_tsc()
1872                  *
1873                  * These values are tracked in kvm->arch.cur_xxx variables.
1874                  */
1875                 kvm->arch.cur_tsc_generation++;
1876                 kvm->arch.cur_tsc_nsec = ns;
1877                 kvm->arch.cur_tsc_write = data;
1878                 kvm->arch.cur_tsc_offset = offset;
1879                 matched = false;
1880         }
1881
1882         /*
1883          * We also track th most recent recorded KHZ, write and time to
1884          * allow the matching interval to be extended at each write.
1885          */
1886         kvm->arch.last_tsc_nsec = ns;
1887         kvm->arch.last_tsc_write = data;
1888         kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1889
1890         vcpu->arch.last_guest_tsc = data;
1891
1892         /* Keep track of which generation this VCPU has synchronized to */
1893         vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1894         vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1895         vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1896
1897         if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1898                 update_ia32_tsc_adjust_msr(vcpu, offset);
1899
1900         kvm_vcpu_write_tsc_offset(vcpu, offset);
1901         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1902
1903         spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1904         if (!matched) {
1905                 kvm->arch.nr_vcpus_matched_tsc = 0;
1906         } else if (!already_matched) {
1907                 kvm->arch.nr_vcpus_matched_tsc++;
1908         }
1909
1910         kvm_track_tsc_matching(vcpu);
1911         spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1912 }
1913
1914 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1915
1916 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1917                                            s64 adjustment)
1918 {
1919         u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1920         kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1921 }
1922
1923 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1924 {
1925         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1926                 WARN_ON(adjustment < 0);
1927         adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1928         adjust_tsc_offset_guest(vcpu, adjustment);
1929 }
1930
1931 #ifdef CONFIG_X86_64
1932
1933 static u64 read_tsc(void)
1934 {
1935         u64 ret = (u64)rdtsc_ordered();
1936         u64 last = pvclock_gtod_data.clock.cycle_last;
1937
1938         if (likely(ret >= last))
1939                 return ret;
1940
1941         /*
1942          * GCC likes to generate cmov here, but this branch is extremely
1943          * predictable (it's just a function of time and the likely is
1944          * very likely) and there's a data dependence, so force GCC
1945          * to generate a branch instead.  I don't barrier() because
1946          * we don't actually need a barrier, and if this function
1947          * ever gets inlined it will generate worse code.
1948          */
1949         asm volatile ("");
1950         return last;
1951 }
1952
1953 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1954 {
1955         long v;
1956         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1957         u64 tsc_pg_val;
1958
1959         switch (gtod->clock.vclock_mode) {
1960         case VCLOCK_HVCLOCK:
1961                 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1962                                                   tsc_timestamp);
1963                 if (tsc_pg_val != U64_MAX) {
1964                         /* TSC page valid */
1965                         *mode = VCLOCK_HVCLOCK;
1966                         v = (tsc_pg_val - gtod->clock.cycle_last) &
1967                                 gtod->clock.mask;
1968                 } else {
1969                         /* TSC page invalid */
1970                         *mode = VCLOCK_NONE;
1971                 }
1972                 break;
1973         case VCLOCK_TSC:
1974                 *mode = VCLOCK_TSC;
1975                 *tsc_timestamp = read_tsc();
1976                 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1977                         gtod->clock.mask;
1978                 break;
1979         default:
1980                 *mode = VCLOCK_NONE;
1981         }
1982
1983         if (*mode == VCLOCK_NONE)
1984                 *tsc_timestamp = v = 0;
1985
1986         return v * gtod->clock.mult;
1987 }
1988
1989 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1990 {
1991         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1992         unsigned long seq;
1993         int mode;
1994         u64 ns;
1995
1996         do {
1997                 seq = read_seqcount_begin(&gtod->seq);
1998                 ns = gtod->nsec_base;
1999                 ns += vgettsc(tsc_timestamp, &mode);
2000                 ns >>= gtod->clock.shift;
2001                 ns += gtod->boot_ns;
2002         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2003         *t = ns;
2004
2005         return mode;
2006 }
2007
2008 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2009 {
2010         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2011         unsigned long seq;
2012         int mode;
2013         u64 ns;
2014
2015         do {
2016                 seq = read_seqcount_begin(&gtod->seq);
2017                 ts->tv_sec = gtod->wall_time_sec;
2018                 ns = gtod->nsec_base;
2019                 ns += vgettsc(tsc_timestamp, &mode);
2020                 ns >>= gtod->clock.shift;
2021         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2022
2023         ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2024         ts->tv_nsec = ns;
2025
2026         return mode;
2027 }
2028
2029 /* returns true if host is using TSC based clocksource */
2030 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2031 {
2032         /* checked again under seqlock below */
2033         if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2034                 return false;
2035
2036         return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
2037                                                       tsc_timestamp));
2038 }
2039
2040 /* returns true if host is using TSC based clocksource */
2041 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2042                                            u64 *tsc_timestamp)
2043 {
2044         /* checked again under seqlock below */
2045         if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2046                 return false;
2047
2048         return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2049 }
2050 #endif
2051
2052 /*
2053  *
2054  * Assuming a stable TSC across physical CPUS, and a stable TSC
2055  * across virtual CPUs, the following condition is possible.
2056  * Each numbered line represents an event visible to both
2057  * CPUs at the next numbered event.
2058  *
2059  * "timespecX" represents host monotonic time. "tscX" represents
2060  * RDTSC value.
2061  *
2062  *              VCPU0 on CPU0           |       VCPU1 on CPU1
2063  *
2064  * 1.  read timespec0,tsc0
2065  * 2.                                   | timespec1 = timespec0 + N
2066  *                                      | tsc1 = tsc0 + M
2067  * 3. transition to guest               | transition to guest
2068  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2069  * 5.                                   | ret1 = timespec1 + (rdtsc - tsc1)
2070  *                                      | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2071  *
2072  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2073  *
2074  *      - ret0 < ret1
2075  *      - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2076  *              ...
2077  *      - 0 < N - M => M < N
2078  *
2079  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2080  * always the case (the difference between two distinct xtime instances
2081  * might be smaller then the difference between corresponding TSC reads,
2082  * when updating guest vcpus pvclock areas).
2083  *
2084  * To avoid that problem, do not allow visibility of distinct
2085  * system_timestamp/tsc_timestamp values simultaneously: use a master
2086  * copy of host monotonic time values. Update that master copy
2087  * in lockstep.
2088  *
2089  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2090  *
2091  */
2092
2093 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2094 {
2095 #ifdef CONFIG_X86_64
2096         struct kvm_arch *ka = &kvm->arch;
2097         int vclock_mode;
2098         bool host_tsc_clocksource, vcpus_matched;
2099
2100         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2101                         atomic_read(&kvm->online_vcpus));
2102
2103         /*
2104          * If the host uses TSC clock, then passthrough TSC as stable
2105          * to the guest.
2106          */
2107         host_tsc_clocksource = kvm_get_time_and_clockread(
2108                                         &ka->master_kernel_ns,
2109                                         &ka->master_cycle_now);
2110
2111         ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2112                                 && !ka->backwards_tsc_observed
2113                                 && !ka->boot_vcpu_runs_old_kvmclock;
2114
2115         if (ka->use_master_clock)
2116                 atomic_set(&kvm_guest_has_master_clock, 1);
2117
2118         vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2119         trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2120                                         vcpus_matched);
2121 #endif
2122 }
2123
2124 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2125 {
2126         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2127 }
2128
2129 static void kvm_gen_update_masterclock(struct kvm *kvm)
2130 {
2131 #ifdef CONFIG_X86_64
2132         int i;
2133         struct kvm_vcpu *vcpu;
2134         struct kvm_arch *ka = &kvm->arch;
2135
2136         spin_lock(&ka->pvclock_gtod_sync_lock);
2137         kvm_make_mclock_inprogress_request(kvm);
2138         /* no guest entries from this point */
2139         pvclock_update_vm_gtod_copy(kvm);
2140
2141         kvm_for_each_vcpu(i, vcpu, kvm)
2142                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2143
2144         /* guest entries allowed */
2145         kvm_for_each_vcpu(i, vcpu, kvm)
2146                 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2147
2148         spin_unlock(&ka->pvclock_gtod_sync_lock);
2149 #endif
2150 }
2151
2152 u64 get_kvmclock_ns(struct kvm *kvm)
2153 {
2154         struct kvm_arch *ka = &kvm->arch;
2155         struct pvclock_vcpu_time_info hv_clock;
2156         u64 ret;
2157
2158         spin_lock(&ka->pvclock_gtod_sync_lock);
2159         if (!ka->use_master_clock) {
2160                 spin_unlock(&ka->pvclock_gtod_sync_lock);
2161                 return ktime_get_boottime_ns() + ka->kvmclock_offset;
2162         }
2163
2164         hv_clock.tsc_timestamp = ka->master_cycle_now;
2165         hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2166         spin_unlock(&ka->pvclock_gtod_sync_lock);
2167
2168         /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2169         get_cpu();
2170
2171         if (__this_cpu_read(cpu_tsc_khz)) {
2172                 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2173                                    &hv_clock.tsc_shift,
2174                                    &hv_clock.tsc_to_system_mul);
2175                 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2176         } else
2177                 ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2178
2179         put_cpu();
2180
2181         return ret;
2182 }
2183
2184 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2185 {
2186         struct kvm_vcpu_arch *vcpu = &v->arch;
2187         struct pvclock_vcpu_time_info guest_hv_clock;
2188
2189         if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2190                 &guest_hv_clock, sizeof(guest_hv_clock))))
2191                 return;
2192
2193         /* This VCPU is paused, but it's legal for a guest to read another
2194          * VCPU's kvmclock, so we really have to follow the specification where
2195          * it says that version is odd if data is being modified, and even after
2196          * it is consistent.
2197          *
2198          * Version field updates must be kept separate.  This is because
2199          * kvm_write_guest_cached might use a "rep movs" instruction, and
2200          * writes within a string instruction are weakly ordered.  So there
2201          * are three writes overall.
2202          *
2203          * As a small optimization, only write the version field in the first
2204          * and third write.  The vcpu->pv_time cache is still valid, because the
2205          * version field is the first in the struct.
2206          */
2207         BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2208
2209         if (guest_hv_clock.version & 1)
2210                 ++guest_hv_clock.version;  /* first time write, random junk */
2211
2212         vcpu->hv_clock.version = guest_hv_clock.version + 1;
2213         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2214                                 &vcpu->hv_clock,
2215                                 sizeof(vcpu->hv_clock.version));
2216
2217         smp_wmb();
2218
2219         /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2220         vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2221
2222         if (vcpu->pvclock_set_guest_stopped_request) {
2223                 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2224                 vcpu->pvclock_set_guest_stopped_request = false;
2225         }
2226
2227         trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2228
2229         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2230                                 &vcpu->hv_clock,
2231                                 sizeof(vcpu->hv_clock));
2232
2233         smp_wmb();
2234
2235         vcpu->hv_clock.version++;
2236         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2237                                 &vcpu->hv_clock,
2238                                 sizeof(vcpu->hv_clock.version));
2239 }
2240
2241 static int kvm_guest_time_update(struct kvm_vcpu *v)
2242 {
2243         unsigned long flags, tgt_tsc_khz;
2244         struct kvm_vcpu_arch *vcpu = &v->arch;
2245         struct kvm_arch *ka = &v->kvm->arch;
2246         s64 kernel_ns;
2247         u64 tsc_timestamp, host_tsc;
2248         u8 pvclock_flags;
2249         bool use_master_clock;
2250
2251         kernel_ns = 0;
2252         host_tsc = 0;
2253
2254         /*
2255          * If the host uses TSC clock, then passthrough TSC as stable
2256          * to the guest.
2257          */
2258         spin_lock(&ka->pvclock_gtod_sync_lock);
2259         use_master_clock = ka->use_master_clock;
2260         if (use_master_clock) {
2261                 host_tsc = ka->master_cycle_now;
2262                 kernel_ns = ka->master_kernel_ns;
2263         }
2264         spin_unlock(&ka->pvclock_gtod_sync_lock);
2265
2266         /* Keep irq disabled to prevent changes to the clock */
2267         local_irq_save(flags);
2268         tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2269         if (unlikely(tgt_tsc_khz == 0)) {
2270                 local_irq_restore(flags);
2271                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2272                 return 1;
2273         }
2274         if (!use_master_clock) {
2275                 host_tsc = rdtsc();
2276                 kernel_ns = ktime_get_boottime_ns();
2277         }
2278
2279         tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2280
2281         /*
2282          * We may have to catch up the TSC to match elapsed wall clock
2283          * time for two reasons, even if kvmclock is used.
2284          *   1) CPU could have been running below the maximum TSC rate
2285          *   2) Broken TSC compensation resets the base at each VCPU
2286          *      entry to avoid unknown leaps of TSC even when running
2287          *      again on the same CPU.  This may cause apparent elapsed
2288          *      time to disappear, and the guest to stand still or run
2289          *      very slowly.
2290          */
2291         if (vcpu->tsc_catchup) {
2292                 u64 tsc = compute_guest_tsc(v, kernel_ns);
2293                 if (tsc > tsc_timestamp) {
2294                         adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2295                         tsc_timestamp = tsc;
2296                 }
2297         }
2298
2299         local_irq_restore(flags);
2300
2301         /* With all the info we got, fill in the values */
2302
2303         if (kvm_has_tsc_control)
2304                 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2305
2306         if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2307                 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2308                                    &vcpu->hv_clock.tsc_shift,
2309                                    &vcpu->hv_clock.tsc_to_system_mul);
2310                 vcpu->hw_tsc_khz = tgt_tsc_khz;
2311         }
2312
2313         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2314         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2315         vcpu->last_guest_tsc = tsc_timestamp;
2316
2317         /* If the host uses TSC clocksource, then it is stable */
2318         pvclock_flags = 0;
2319         if (use_master_clock)
2320                 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2321
2322         vcpu->hv_clock.flags = pvclock_flags;
2323
2324         if (vcpu->pv_time_enabled)
2325                 kvm_setup_pvclock_page(v);
2326         if (v == kvm_get_vcpu(v->kvm, 0))
2327                 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2328         return 0;
2329 }
2330
2331 /*
2332  * kvmclock updates which are isolated to a given vcpu, such as
2333  * vcpu->cpu migration, should not allow system_timestamp from
2334  * the rest of the vcpus to remain static. Otherwise ntp frequency
2335  * correction applies to one vcpu's system_timestamp but not
2336  * the others.
2337  *
2338  * So in those cases, request a kvmclock update for all vcpus.
2339  * We need to rate-limit these requests though, as they can
2340  * considerably slow guests that have a large number of vcpus.
2341  * The time for a remote vcpu to update its kvmclock is bound
2342  * by the delay we use to rate-limit the updates.
2343  */
2344
2345 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2346
2347 static void kvmclock_update_fn(struct work_struct *work)
2348 {
2349         int i;
2350         struct delayed_work *dwork = to_delayed_work(work);
2351         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2352                                            kvmclock_update_work);
2353         struct kvm *kvm = container_of(ka, struct kvm, arch);
2354         struct kvm_vcpu *vcpu;
2355
2356         kvm_for_each_vcpu(i, vcpu, kvm) {
2357                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2358                 kvm_vcpu_kick(vcpu);
2359         }
2360 }
2361
2362 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2363 {
2364         struct kvm *kvm = v->kvm;
2365
2366         kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2367         schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2368                                         KVMCLOCK_UPDATE_DELAY);
2369 }
2370
2371 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2372
2373 static void kvmclock_sync_fn(struct work_struct *work)
2374 {
2375         struct delayed_work *dwork = to_delayed_work(work);
2376         struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2377                                            kvmclock_sync_work);
2378         struct kvm *kvm = container_of(ka, struct kvm, arch);
2379
2380         if (!kvmclock_periodic_sync)
2381                 return;
2382
2383         schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2384         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2385                                         KVMCLOCK_SYNC_PERIOD);
2386 }
2387
2388 /*
2389  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2390  */
2391 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2392 {
2393         /* McStatusWrEn enabled? */
2394         if (guest_cpuid_is_amd(vcpu))
2395                 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2396
2397         return false;
2398 }
2399
2400 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2401 {
2402         u64 mcg_cap = vcpu->arch.mcg_cap;
2403         unsigned bank_num = mcg_cap & 0xff;
2404         u32 msr = msr_info->index;
2405         u64 data = msr_info->data;
2406
2407         switch (msr) {
2408         case MSR_IA32_MCG_STATUS:
2409                 vcpu->arch.mcg_status = data;
2410                 break;
2411         case MSR_IA32_MCG_CTL:
2412                 if (!(mcg_cap & MCG_CTL_P) &&
2413                     (data || !msr_info->host_initiated))
2414                         return 1;
2415                 if (data != 0 && data != ~(u64)0)
2416                         return 1;
2417                 vcpu->arch.mcg_ctl = data;
2418                 break;
2419         default:
2420                 if (msr >= MSR_IA32_MC0_CTL &&
2421                     msr < MSR_IA32_MCx_CTL(bank_num)) {
2422                         u32 offset = msr - MSR_IA32_MC0_CTL;
2423                         /* only 0 or all 1s can be written to IA32_MCi_CTL
2424                          * some Linux kernels though clear bit 10 in bank 4 to
2425                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2426                          * this to avoid an uncatched #GP in the guest
2427                          */
2428                         if ((offset & 0x3) == 0 &&
2429                             data != 0 && (data | (1 << 10)) != ~(u64)0)
2430                                 return -1;
2431
2432                         /* MCi_STATUS */
2433                         if (!msr_info->host_initiated &&
2434                             (offset & 0x3) == 1 && data != 0) {
2435                                 if (!can_set_mci_status(vcpu))
2436                                         return -1;
2437                         }
2438
2439                         vcpu->arch.mce_banks[offset] = data;
2440                         break;
2441                 }
2442                 return 1;
2443         }
2444         return 0;
2445 }
2446
2447 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2448 {
2449         struct kvm *kvm = vcpu->kvm;
2450         int lm = is_long_mode(vcpu);
2451         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2452                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2453         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2454                 : kvm->arch.xen_hvm_config.blob_size_32;
2455         u32 page_num = data & ~PAGE_MASK;
2456         u64 page_addr = data & PAGE_MASK;
2457         u8 *page;
2458         int r;
2459
2460         r = -E2BIG;
2461         if (page_num >= blob_size)
2462                 goto out;
2463         r = -ENOMEM;
2464         page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2465         if (IS_ERR(page)) {
2466                 r = PTR_ERR(page);
2467                 goto out;
2468         }
2469         if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2470                 goto out_free;
2471         r = 0;
2472 out_free:
2473         kfree(page);
2474 out:
2475         return r;
2476 }
2477
2478 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2479 {
2480         gpa_t gpa = data & ~0x3f;
2481
2482         /* Bits 3:5 are reserved, Should be zero */
2483         if (data & 0x38)
2484                 return 1;
2485
2486         vcpu->arch.apf.msr_val = data;
2487
2488         if (!(data & KVM_ASYNC_PF_ENABLED)) {
2489                 kvm_clear_async_pf_completion_queue(vcpu);
2490                 kvm_async_pf_hash_reset(vcpu);
2491                 return 0;
2492         }
2493
2494         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2495                                         sizeof(u32)))
2496                 return 1;
2497
2498         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2499         vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2500         kvm_async_pf_wakeup_all(vcpu);
2501         return 0;
2502 }
2503
2504 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2505 {
2506         vcpu->arch.pv_time_enabled = false;
2507 }
2508
2509 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2510 {
2511         ++vcpu->stat.tlb_flush;
2512         kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2513 }
2514
2515 static void record_steal_time(struct kvm_vcpu *vcpu)
2516 {
2517         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2518                 return;
2519
2520         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2521                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2522                 return;
2523
2524         /*
2525          * Doing a TLB flush here, on the guest's behalf, can avoid
2526          * expensive IPIs.
2527          */
2528         trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2529                 vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2530         if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2531                 kvm_vcpu_flush_tlb(vcpu, false);
2532
2533         if (vcpu->arch.st.steal.version & 1)
2534                 vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2535
2536         vcpu->arch.st.steal.version += 1;
2537
2538         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2539                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2540
2541         smp_wmb();
2542
2543         vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2544                 vcpu->arch.st.last_steal;
2545         vcpu->arch.st.last_steal = current->sched_info.run_delay;
2546
2547         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2548                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2549
2550         smp_wmb();
2551
2552         vcpu->arch.st.steal.version += 1;
2553
2554         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2555                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2556 }
2557
2558 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2559 {
2560         bool pr = false;
2561         u32 msr = msr_info->index;
2562         u64 data = msr_info->data;
2563
2564         switch (msr) {
2565         case MSR_AMD64_NB_CFG:
2566         case MSR_IA32_UCODE_WRITE:
2567         case MSR_VM_HSAVE_PA:
2568         case MSR_AMD64_PATCH_LOADER:
2569         case MSR_AMD64_BU_CFG2:
2570         case MSR_AMD64_DC_CFG:
2571         case MSR_F15H_EX_CFG:
2572                 break;
2573
2574         case MSR_IA32_UCODE_REV:
2575                 if (msr_info->host_initiated)
2576                         vcpu->arch.microcode_version = data;
2577                 break;
2578         case MSR_IA32_ARCH_CAPABILITIES:
2579                 if (!msr_info->host_initiated)
2580                         return 1;
2581                 vcpu->arch.arch_capabilities = data;
2582                 break;
2583         case MSR_EFER:
2584                 return set_efer(vcpu, msr_info);
2585         case MSR_K7_HWCR:
2586                 data &= ~(u64)0x40;     /* ignore flush filter disable */
2587                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
2588                 data &= ~(u64)0x8;      /* ignore TLB cache disable */
2589
2590                 /* Handle McStatusWrEn */
2591                 if (data == BIT_ULL(18)) {
2592                         vcpu->arch.msr_hwcr = data;
2593                 } else if (data != 0) {
2594                         vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2595                                     data);
2596                         return 1;
2597                 }
2598                 break;
2599         case MSR_FAM10H_MMIO_CONF_BASE:
2600                 if (data != 0) {
2601                         vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2602                                     "0x%llx\n", data);
2603                         return 1;
2604                 }
2605                 break;
2606         case MSR_IA32_DEBUGCTLMSR:
2607                 if (!data) {
2608                         /* We support the non-activated case already */
2609                         break;
2610                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2611                         /* Values other than LBR and BTF are vendor-specific,
2612                            thus reserved and should throw a #GP */
2613                         return 1;
2614                 }
2615                 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2616                             __func__, data);
2617                 break;
2618         case 0x200 ... 0x2ff:
2619                 return kvm_mtrr_set_msr(vcpu, msr, data);
2620         case MSR_IA32_APICBASE:
2621                 return kvm_set_apic_base(vcpu, msr_info);
2622         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2623                 return kvm_x2apic_msr_write(vcpu, msr, data);
2624         case MSR_IA32_TSCDEADLINE:
2625                 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2626                 break;
2627         case MSR_IA32_TSC_ADJUST:
2628                 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2629                         if (!msr_info->host_initiated) {
2630                                 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2631                                 adjust_tsc_offset_guest(vcpu, adj);
2632                         }
2633                         vcpu->arch.ia32_tsc_adjust_msr = data;
2634                 }
2635                 break;
2636         case MSR_IA32_MISC_ENABLE:
2637                 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2638                     ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2639                         if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2640                                 return 1;
2641                         vcpu->arch.ia32_misc_enable_msr = data;
2642                         kvm_update_cpuid(vcpu);
2643                 } else {
2644                         vcpu->arch.ia32_misc_enable_msr = data;
2645                 }
2646                 break;
2647         case MSR_IA32_SMBASE:
2648                 if (!msr_info->host_initiated)
2649                         return 1;
2650                 vcpu->arch.smbase = data;
2651                 break;
2652         case MSR_IA32_POWER_CTL:
2653                 vcpu->arch.msr_ia32_power_ctl = data;
2654                 break;
2655         case MSR_IA32_TSC:
2656                 kvm_write_tsc(vcpu, msr_info);
2657                 break;
2658         case MSR_SMI_COUNT:
2659                 if (!msr_info->host_initiated)
2660                         return 1;
2661                 vcpu->arch.smi_count = data;
2662                 break;
2663         case MSR_KVM_WALL_CLOCK_NEW:
2664         case MSR_KVM_WALL_CLOCK:
2665                 vcpu->kvm->arch.wall_clock = data;
2666                 kvm_write_wall_clock(vcpu->kvm, data);
2667                 break;
2668         case MSR_KVM_SYSTEM_TIME_NEW:
2669         case MSR_KVM_SYSTEM_TIME: {
2670                 struct kvm_arch *ka = &vcpu->kvm->arch;
2671
2672                 kvmclock_reset(vcpu);
2673
2674                 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2675                         bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2676
2677                         if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2678                                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2679
2680                         ka->boot_vcpu_runs_old_kvmclock = tmp;
2681                 }
2682
2683                 vcpu->arch.time = data;
2684                 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2685
2686                 /* we verify if the enable bit is set... */
2687                 if (!(data & 1))
2688                         break;
2689
2690                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2691                      &vcpu->arch.pv_time, data & ~1ULL,
2692                      sizeof(struct pvclock_vcpu_time_info)))
2693                         vcpu->arch.pv_time_enabled = false;
2694                 else
2695                         vcpu->arch.pv_time_enabled = true;
2696
2697                 break;
2698         }
2699         case MSR_KVM_ASYNC_PF_EN:
2700                 if (kvm_pv_enable_async_pf(vcpu, data))
2701                         return 1;
2702                 break;
2703         case MSR_KVM_STEAL_TIME:
2704
2705                 if (unlikely(!sched_info_on()))
2706                         return 1;
2707
2708                 if (data & KVM_STEAL_RESERVED_MASK)
2709                         return 1;
2710
2711                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2712                                                 data & KVM_STEAL_VALID_BITS,
2713                                                 sizeof(struct kvm_steal_time)))
2714                         return 1;
2715
2716                 vcpu->arch.st.msr_val = data;
2717
2718                 if (!(data & KVM_MSR_ENABLED))
2719                         break;
2720
2721                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2722
2723                 break;
2724         case MSR_KVM_PV_EOI_EN:
2725                 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2726                         return 1;
2727                 break;
2728
2729         case MSR_KVM_POLL_CONTROL:
2730                 /* only enable bit supported */
2731                 if (data & (-1ULL << 1))
2732                         return 1;
2733
2734                 vcpu->arch.msr_kvm_poll_control = data;
2735                 break;
2736
2737         case MSR_IA32_MCG_CTL:
2738         case MSR_IA32_MCG_STATUS:
2739         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2740                 return set_msr_mce(vcpu, msr_info);
2741
2742         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2743         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2744                 pr = true; /* fall through */
2745         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2746         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2747                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2748                         return kvm_pmu_set_msr(vcpu, msr_info);
2749
2750                 if (pr || data != 0)
2751                         vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2752                                     "0x%x data 0x%llx\n", msr, data);
2753                 break;
2754         case MSR_K7_CLK_CTL:
2755                 /*
2756                  * Ignore all writes to this no longer documented MSR.
2757                  * Writes are only relevant for old K7 processors,
2758                  * all pre-dating SVM, but a recommended workaround from
2759                  * AMD for these chips. It is possible to specify the
2760                  * affected processor models on the command line, hence
2761                  * the need to ignore the workaround.
2762                  */
2763                 break;
2764         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2765         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2766         case HV_X64_MSR_CRASH_CTL:
2767         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2768         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2769         case HV_X64_MSR_TSC_EMULATION_CONTROL:
2770         case HV_X64_MSR_TSC_EMULATION_STATUS:
2771                 return kvm_hv_set_msr_common(vcpu, msr, data,
2772                                              msr_info->host_initiated);
2773         case MSR_IA32_BBL_CR_CTL3:
2774                 /* Drop writes to this legacy MSR -- see rdmsr
2775                  * counterpart for further detail.
2776                  */
2777                 if (report_ignored_msrs)
2778                         vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2779                                 msr, data);
2780                 break;
2781         case MSR_AMD64_OSVW_ID_LENGTH:
2782                 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2783                         return 1;
2784                 vcpu->arch.osvw.length = data;
2785                 break;
2786         case MSR_AMD64_OSVW_STATUS:
2787                 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2788                         return 1;
2789                 vcpu->arch.osvw.status = data;
2790                 break;
2791         case MSR_PLATFORM_INFO:
2792                 if (!msr_info->host_initiated ||
2793                     (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2794                      cpuid_fault_enabled(vcpu)))
2795                         return 1;
2796                 vcpu->arch.msr_platform_info = data;
2797                 break;
2798         case MSR_MISC_FEATURES_ENABLES:
2799                 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2800                     (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2801                      !supports_cpuid_fault(vcpu)))
2802                         return 1;
2803                 vcpu->arch.msr_misc_features_enables = data;
2804                 break;
2805         default:
2806                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2807                         return xen_hvm_config(vcpu, data);
2808                 if (kvm_pmu_is_valid_msr(vcpu, msr))
2809                         return kvm_pmu_set_msr(vcpu, msr_info);
2810                 if (!ignore_msrs) {
2811                         vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2812                                     msr, data);
2813                         return 1;
2814                 } else {
2815                         if (report_ignored_msrs)
2816                                 vcpu_unimpl(vcpu,
2817                                         "ignored wrmsr: 0x%x data 0x%llx\n",
2818                                         msr, data);
2819                         break;
2820                 }
2821         }
2822         return 0;
2823 }
2824 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2825
2826 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2827 {
2828         u64 data;
2829         u64 mcg_cap = vcpu->arch.mcg_cap;
2830         unsigned bank_num = mcg_cap & 0xff;
2831
2832         switch (msr) {
2833         case MSR_IA32_P5_MC_ADDR:
2834         case MSR_IA32_P5_MC_TYPE:
2835                 data = 0;
2836                 break;
2837         case MSR_IA32_MCG_CAP:
2838                 data = vcpu->arch.mcg_cap;
2839                 break;
2840         case MSR_IA32_MCG_CTL:
2841                 if (!(mcg_cap & MCG_CTL_P) && !host)
2842                         return 1;
2843                 data = vcpu->arch.mcg_ctl;
2844                 break;
2845         case MSR_IA32_MCG_STATUS:
2846                 data = vcpu->arch.mcg_status;
2847                 break;
2848         default:
2849                 if (msr >= MSR_IA32_MC0_CTL &&
2850                     msr < MSR_IA32_MCx_CTL(bank_num)) {
2851                         u32 offset = msr - MSR_IA32_MC0_CTL;
2852                         data = vcpu->arch.mce_banks[offset];
2853                         break;
2854                 }
2855                 return 1;
2856         }
2857         *pdata = data;
2858         return 0;
2859 }
2860
2861 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2862 {
2863         switch (msr_info->index) {
2864         case MSR_IA32_PLATFORM_ID:
2865         case MSR_IA32_EBL_CR_POWERON:
2866         case MSR_IA32_DEBUGCTLMSR:
2867         case MSR_IA32_LASTBRANCHFROMIP:
2868         case MSR_IA32_LASTBRANCHTOIP:
2869         case MSR_IA32_LASTINTFROMIP:
2870         case MSR_IA32_LASTINTTOIP:
2871         case MSR_K8_SYSCFG:
2872         case MSR_K8_TSEG_ADDR:
2873         case MSR_K8_TSEG_MASK:
2874         case MSR_VM_HSAVE_PA:
2875         case MSR_K8_INT_PENDING_MSG:
2876         case MSR_AMD64_NB_CFG:
2877         case MSR_FAM10H_MMIO_CONF_BASE:
2878         case MSR_AMD64_BU_CFG2:
2879         case MSR_IA32_PERF_CTL:
2880         case MSR_AMD64_DC_CFG:
2881         case MSR_F15H_EX_CFG:
2882                 msr_info->data = 0;
2883                 break;
2884         case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2885         case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2886         case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2887         case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2888         case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2889                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2890                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2891                 msr_info->data = 0;
2892                 break;
2893         case MSR_IA32_UCODE_REV:
2894                 msr_info->data = vcpu->arch.microcode_version;
2895                 break;
2896         case MSR_IA32_ARCH_CAPABILITIES:
2897                 if (!msr_info->host_initiated &&
2898                     !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2899                         return 1;
2900                 msr_info->data = vcpu->arch.arch_capabilities;
2901                 break;
2902         case MSR_IA32_POWER_CTL:
2903                 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2904                 break;
2905         case MSR_IA32_TSC:
2906                 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2907                 break;
2908         case MSR_MTRRcap:
2909         case 0x200 ... 0x2ff:
2910                 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2911         case 0xcd: /* fsb frequency */
2912                 msr_info->data = 3;
2913                 break;
2914                 /*
2915                  * MSR_EBC_FREQUENCY_ID
2916                  * Conservative value valid for even the basic CPU models.
2917                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2918                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2919                  * and 266MHz for model 3, or 4. Set Core Clock
2920                  * Frequency to System Bus Frequency Ratio to 1 (bits
2921                  * 31:24) even though these are only valid for CPU
2922                  * models > 2, however guests may end up dividing or
2923                  * multiplying by zero otherwise.
2924                  */
2925         case MSR_EBC_FREQUENCY_ID:
2926                 msr_info->data = 1 << 24;
2927                 break;
2928         case MSR_IA32_APICBASE:
2929                 msr_info->data = kvm_get_apic_base(vcpu);
2930                 break;
2931         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2932                 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2933                 break;
2934         case MSR_IA32_TSCDEADLINE:
2935                 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2936                 break;
2937         case MSR_IA32_TSC_ADJUST:
2938                 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2939                 break;
2940         case MSR_IA32_MISC_ENABLE:
2941                 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2942                 break;
2943         case MSR_IA32_SMBASE:
2944                 if (!msr_info->host_initiated)
2945                         return 1;
2946                 msr_info->data = vcpu->arch.smbase;
2947                 break;
2948         case MSR_SMI_COUNT:
2949                 msr_info->data = vcpu->arch.smi_count;
2950                 break;
2951         case MSR_IA32_PERF_STATUS:
2952                 /* TSC increment by tick */
2953                 msr_info->data = 1000ULL;
2954                 /* CPU multiplier */
2955                 msr_info->data |= (((uint64_t)4ULL) << 40);
2956                 break;
2957         case MSR_EFER:
2958                 msr_info->data = vcpu->arch.efer;
2959                 break;
2960         case MSR_KVM_WALL_CLOCK:
2961         case MSR_KVM_WALL_CLOCK_NEW:
2962                 msr_info->data = vcpu->kvm->arch.wall_clock;
2963                 break;
2964         case MSR_KVM_SYSTEM_TIME:
2965         case MSR_KVM_SYSTEM_TIME_NEW:
2966                 msr_info->data = vcpu->arch.time;
2967                 break;
2968         case MSR_KVM_ASYNC_PF_EN:
2969                 msr_info->data = vcpu->arch.apf.msr_val;
2970                 break;
2971         case MSR_KVM_STEAL_TIME:
2972                 msr_info->data = vcpu->arch.st.msr_val;
2973                 break;
2974         case MSR_KVM_PV_EOI_EN:
2975                 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2976                 break;
2977         case MSR_KVM_POLL_CONTROL:
2978                 msr_info->data = vcpu->arch.msr_kvm_poll_control;
2979                 break;
2980         case MSR_IA32_P5_MC_ADDR:
2981         case MSR_IA32_P5_MC_TYPE:
2982         case MSR_IA32_MCG_CAP:
2983         case MSR_IA32_MCG_CTL:
2984         case MSR_IA32_MCG_STATUS:
2985         case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2986                 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2987                                    msr_info->host_initiated);
2988         case MSR_K7_CLK_CTL:
2989                 /*
2990                  * Provide expected ramp-up count for K7. All other
2991                  * are set to zero, indicating minimum divisors for
2992                  * every field.
2993                  *
2994                  * This prevents guest kernels on AMD host with CPU
2995                  * type 6, model 8 and higher from exploding due to
2996                  * the rdmsr failing.
2997                  */
2998                 msr_info->data = 0x20000000;
2999                 break;
3000         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3001         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3002         case HV_X64_MSR_CRASH_CTL:
3003         case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3004         case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3005         case HV_X64_MSR_TSC_EMULATION_CONTROL:
3006         case HV_X64_MSR_TSC_EMULATION_STATUS:
3007                 return kvm_hv_get_msr_common(vcpu,
3008                                              msr_info->index, &msr_info->data,
3009                                              msr_info->host_initiated);
3010                 break;
3011         case MSR_IA32_BBL_CR_CTL3:
3012                 /* This legacy MSR exists but isn't fully documented in current
3013                  * silicon.  It is however accessed by winxp in very narrow
3014                  * scenarios where it sets bit #19, itself documented as
3015                  * a "reserved" bit.  Best effort attempt to source coherent
3016                  * read data here should the balance of the register be
3017                  * interpreted by the guest:
3018                  *
3019                  * L2 cache control register 3: 64GB range, 256KB size,
3020                  * enabled, latency 0x1, configured
3021                  */
3022                 msr_info->data = 0xbe702111;
3023                 break;
3024         case MSR_AMD64_OSVW_ID_LENGTH:
3025                 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3026                         return 1;
3027                 msr_info->data = vcpu->arch.osvw.length;
3028                 break;
3029         case MSR_AMD64_OSVW_STATUS:
3030                 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3031                         return 1;
3032                 msr_info->data = vcpu->arch.osvw.status;
3033                 break;
3034         case MSR_PLATFORM_INFO:
3035                 if (!msr_info->host_initiated &&
3036                     !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3037                         return 1;
3038                 msr_info->data = vcpu->arch.msr_platform_info;
3039                 break;
3040         case MSR_MISC_FEATURES_ENABLES:
3041                 msr_info->data = vcpu->arch.msr_misc_features_enables;
3042                 break;
3043         case MSR_K7_HWCR:
3044                 msr_info->data = vcpu->arch.msr_hwcr;
3045                 break;
3046         default:
3047                 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3048                         return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3049                 if (!ignore_msrs) {
3050                         vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3051                                                msr_info->index);
3052                         return 1;
3053                 } else {
3054                         if (report_ignored_msrs)
3055                                 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3056                                         msr_info->index);
3057                         msr_info->data = 0;
3058                 }
3059                 break;
3060         }
3061         return 0;
3062 }
3063 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3064
3065 /*
3066  * Read or write a bunch of msrs. All parameters are kernel addresses.
3067  *
3068  * @return number of msrs set successfully.
3069  */
3070 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3071                     struct kvm_msr_entry *entries,
3072                     int (*do_msr)(struct kvm_vcpu *vcpu,
3073                                   unsigned index, u64 *data))
3074 {
3075         int i;
3076
3077         for (i = 0; i < msrs->nmsrs; ++i)
3078                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3079                         break;
3080
3081         return i;
3082 }
3083
3084 /*
3085  * Read or write a bunch of msrs. Parameters are user addresses.
3086  *
3087  * @return number of msrs set successfully.
3088  */
3089 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3090                   int (*do_msr)(struct kvm_vcpu *vcpu,
3091                                 unsigned index, u64 *data),
3092                   int writeback)
3093 {
3094         struct kvm_msrs msrs;
3095         struct kvm_msr_entry *entries;
3096         int r, n;
3097         unsigned size;
3098
3099         r = -EFAULT;
3100         if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3101                 goto out;
3102
3103         r = -E2BIG;
3104         if (msrs.nmsrs >= MAX_IO_MSRS)
3105                 goto out;
3106
3107         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3108         entries = memdup_user(user_msrs->entries, size);
3109         if (IS_ERR(entries)) {
3110                 r = PTR_ERR(entries);
3111                 goto out;
3112         }
3113
3114         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3115         if (r < 0)
3116                 goto out_free;
3117
3118         r = -EFAULT;
3119         if (writeback && copy_to_user(user_msrs->entries, entries, size))
3120                 goto out_free;
3121
3122         r = n;
3123
3124 out_free:
3125         kfree(entries);
3126 out:
3127         return r;
3128 }
3129
3130 static inline bool kvm_can_mwait_in_guest(void)
3131 {
3132         return boot_cpu_has(X86_FEATURE_MWAIT) &&
3133                 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3134                 boot_cpu_has(X86_FEATURE_ARAT);
3135 }
3136
3137 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3138 {
3139         int r = 0;
3140
3141         switch (ext) {
3142         case KVM_CAP_IRQCHIP:
3143         case KVM_CAP_HLT:
3144         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3145         case KVM_CAP_SET_TSS_ADDR:
3146         case KVM_CAP_EXT_CPUID:
3147         case KVM_CAP_EXT_EMUL_CPUID:
3148         case KVM_CAP_CLOCKSOURCE:
3149         case KVM_CAP_PIT:
3150         case KVM_CAP_NOP_IO_DELAY:
3151         case KVM_CAP_MP_STATE:
3152         case KVM_CAP_SYNC_MMU:
3153         case KVM_CAP_USER_NMI:
3154         case KVM_CAP_REINJECT_CONTROL:
3155         case KVM_CAP_IRQ_INJECT_STATUS:
3156         case KVM_CAP_IOEVENTFD:
3157         case KVM_CAP_IOEVENTFD_NO_LENGTH:
3158         case KVM_CAP_PIT2:
3159         case KVM_CAP_PIT_STATE2:
3160         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3161         case KVM_CAP_XEN_HVM:
3162         case KVM_CAP_VCPU_EVENTS:
3163         case KVM_CAP_HYPERV:
3164         case KVM_CAP_HYPERV_VAPIC:
3165         case KVM_CAP_HYPERV_SPIN:
3166         case KVM_CAP_HYPERV_SYNIC:
3167         case KVM_CAP_HYPERV_SYNIC2:
3168         case KVM_CAP_HYPERV_VP_INDEX:
3169         case KVM_CAP_HYPERV_EVENTFD:
3170         case KVM_CAP_HYPERV_TLBFLUSH:
3171         case KVM_CAP_HYPERV_SEND_IPI:
3172         case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3173         case KVM_CAP_HYPERV_CPUID:
3174         case KVM_CAP_PCI_SEGMENT:
3175         case KVM_CAP_DEBUGREGS:
3176         case KVM_CAP_X86_ROBUST_SINGLESTEP:
3177         case KVM_CAP_XSAVE:
3178         case KVM_CAP_ASYNC_PF:
3179         case KVM_CAP_GET_TSC_KHZ:
3180         case KVM_CAP_KVMCLOCK_CTRL:
3181         case KVM_CAP_READONLY_MEM:
3182         case KVM_CAP_HYPERV_TIME:
3183         case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3184         case KVM_CAP_TSC_DEADLINE_TIMER:
3185         case KVM_CAP_DISABLE_QUIRKS:
3186         case KVM_CAP_SET_BOOT_CPU_ID:
3187         case KVM_CAP_SPLIT_IRQCHIP:
3188         case KVM_CAP_IMMEDIATE_EXIT:
3189         case KVM_CAP_PMU_EVENT_FILTER:
3190         case KVM_CAP_GET_MSR_FEATURES:
3191         case KVM_CAP_MSR_PLATFORM_INFO:
3192         case KVM_CAP_EXCEPTION_PAYLOAD:
3193                 r = 1;
3194                 break;
3195         case KVM_CAP_SYNC_REGS:
3196                 r = KVM_SYNC_X86_VALID_FIELDS;
3197                 break;
3198         case KVM_CAP_ADJUST_CLOCK:
3199                 r = KVM_CLOCK_TSC_STABLE;
3200                 break;
3201         case KVM_CAP_X86_DISABLE_EXITS:
3202                 r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3203                       KVM_X86_DISABLE_EXITS_CSTATE;
3204                 if(kvm_can_mwait_in_guest())
3205                         r |= KVM_X86_DISABLE_EXITS_MWAIT;
3206                 break;
3207         case KVM_CAP_X86_SMM:
3208                 /* SMBASE is usually relocated above 1M on modern chipsets,
3209                  * and SMM handlers might indeed rely on 4G segment limits,
3210                  * so do not report SMM to be available if real mode is
3211                  * emulated via vm86 mode.  Still, do not go to great lengths
3212                  * to avoid userspace's usage of the feature, because it is a
3213                  * fringe case that is not enabled except via specific settings
3214                  * of the module parameters.
3215                  */
3216                 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3217                 break;
3218         case KVM_CAP_VAPIC:
3219                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3220                 break;
3221         case KVM_CAP_NR_VCPUS:
3222                 r = KVM_SOFT_MAX_VCPUS;
3223                 break;
3224         case KVM_CAP_MAX_VCPUS:
3225                 r = KVM_MAX_VCPUS;
3226                 break;
3227         case KVM_CAP_MAX_VCPU_ID:
3228                 r = KVM_MAX_VCPU_ID;
3229                 break;
3230         case KVM_CAP_PV_MMU:    /* obsolete */
3231                 r = 0;
3232                 break;
3233         case KVM_CAP_MCE:
3234                 r = KVM_MAX_MCE_BANKS;
3235                 break;
3236         case KVM_CAP_XCRS:
3237                 r = boot_cpu_has(X86_FEATURE_XSAVE);
3238                 break;
3239         case KVM_CAP_TSC_CONTROL:
3240                 r = kvm_has_tsc_control;
3241                 break;
3242         case KVM_CAP_X2APIC_API:
3243                 r = KVM_X2APIC_API_VALID_FLAGS;
3244                 break;
3245         case KVM_CAP_NESTED_STATE:
3246                 r = kvm_x86_ops->get_nested_state ?
3247                         kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3248                 break;
3249         default:
3250                 break;
3251         }
3252         return r;
3253
3254 }
3255
3256 long kvm_arch_dev_ioctl(struct file *filp,
3257                         unsigned int ioctl, unsigned long arg)
3258 {
3259         void __user *argp = (void __user *)arg;
3260         long r;
3261
3262         switch (ioctl) {
3263         case KVM_GET_MSR_INDEX_LIST: {
3264                 struct kvm_msr_list __user *user_msr_list = argp;
3265                 struct kvm_msr_list msr_list;
3266                 unsigned n;
3267
3268                 r = -EFAULT;
3269                 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3270                         goto out;
3271                 n = msr_list.nmsrs;
3272                 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3273                 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3274                         goto out;
3275                 r = -E2BIG;
3276                 if (n < msr_list.nmsrs)
3277                         goto out;
3278                 r = -EFAULT;
3279                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3280                                  num_msrs_to_save * sizeof(u32)))
3281                         goto out;
3282                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3283                                  &emulated_msrs,
3284                                  num_emulated_msrs * sizeof(u32)))
3285                         goto out;
3286                 r = 0;
3287                 break;
3288         }
3289         case KVM_GET_SUPPORTED_CPUID:
3290         case KVM_GET_EMULATED_CPUID: {
3291                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3292                 struct kvm_cpuid2 cpuid;
3293
3294                 r = -EFAULT;
3295                 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3296                         goto out;
3297
3298                 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3299                                             ioctl);
3300                 if (r)
3301                         goto out;
3302
3303                 r = -EFAULT;
3304                 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3305                         goto out;
3306                 r = 0;
3307                 break;
3308         }
3309         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3310                 r = -EFAULT;
3311                 if (copy_to_user(argp, &kvm_mce_cap_supported,
3312                                  sizeof(kvm_mce_cap_supported)))
3313                         goto out;
3314                 r = 0;
3315                 break;
3316         case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3317                 struct kvm_msr_list __user *user_msr_list = argp;
3318                 struct kvm_msr_list msr_list;
3319                 unsigned int n;
3320
3321                 r = -EFAULT;
3322                 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3323                         goto out;
3324                 n = msr_list.nmsrs;
3325                 msr_list.nmsrs = num_msr_based_features;
3326                 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3327                         goto out;
3328                 r = -E2BIG;
3329                 if (n < msr_list.nmsrs)
3330                         goto out;
3331                 r = -EFAULT;
3332                 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3333                                  num_msr_based_features * sizeof(u32)))
3334                         goto out;
3335                 r = 0;
3336                 break;
3337         }
3338         case KVM_GET_MSRS:
3339                 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3340                 break;
3341         }
3342         default:
3343                 r = -EINVAL;
3344         }
3345 out:
3346         return r;
3347 }
3348
3349 static void wbinvd_ipi(void *garbage)
3350 {
3351         wbinvd();
3352 }
3353
3354 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3355 {
3356         return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3357 }
3358
3359 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3360 {
3361         /* Address WBINVD may be executed by guest */
3362         if (need_emulate_wbinvd(vcpu)) {
3363                 if (kvm_x86_ops->has_wbinvd_exit())
3364                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3365                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3366                         smp_call_function_single(vcpu->cpu,
3367                                         wbinvd_ipi, NULL, 1);
3368         }
3369
3370         kvm_x86_ops->vcpu_load(vcpu, cpu);
3371
3372         fpregs_assert_state_consistent();
3373         if (test_thread_flag(TIF_NEED_FPU_LOAD))
3374                 switch_fpu_return();
3375
3376         /* Apply any externally detected TSC adjustments (due to suspend) */
3377         if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3378                 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3379                 vcpu->arch.tsc_offset_adjustment = 0;
3380                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3381         }
3382
3383         if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3384                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3385                                 rdtsc() - vcpu->arch.last_host_tsc;
3386                 if (tsc_delta < 0)
3387                         mark_tsc_unstable("KVM discovered backwards TSC");
3388
3389                 if (kvm_check_tsc_unstable()) {
3390                         u64 offset = kvm_compute_tsc_offset(vcpu,
3391                                                 vcpu->arch.last_guest_tsc);
3392                         kvm_vcpu_write_tsc_offset(vcpu, offset);
3393                         vcpu->arch.tsc_catchup = 1;
3394                 }
3395
3396                 if (kvm_lapic_hv_timer_in_use(vcpu))
3397                         kvm_lapic_restart_hv_timer(vcpu);
3398
3399                 /*
3400                  * On a host with synchronized TSC, there is no need to update
3401                  * kvmclock on vcpu->cpu migration
3402                  */
3403                 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3404                         kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3405                 if (vcpu->cpu != cpu)
3406                         kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3407                 vcpu->cpu = cpu;
3408         }
3409
3410         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3411 }
3412
3413 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3414 {
3415         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3416                 return;
3417
3418         vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3419
3420         kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3421                         &vcpu->arch.st.steal.preempted,
3422                         offsetof(struct kvm_steal_time, preempted),
3423                         sizeof(vcpu->arch.st.steal.preempted));
3424 }
3425
3426 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3427 {
3428         int idx;
3429
3430         if (vcpu->preempted)
3431                 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3432
3433         /*
3434          * Disable page faults because we're in atomic context here.
3435          * kvm_write_guest_offset_cached() would call might_fault()
3436          * that relies on pagefault_disable() to tell if there's a
3437          * bug. NOTE: the write to guest memory may not go through if
3438          * during postcopy live migration or if there's heavy guest
3439          * paging.
3440          */
3441         pagefault_disable();
3442         /*
3443          * kvm_memslots() will be called by
3444          * kvm_write_guest_offset_cached() so take the srcu lock.
3445          */
3446         idx = srcu_read_lock(&vcpu->kvm->srcu);
3447         kvm_steal_time_set_preempted(vcpu);
3448         srcu_read_unlock(&vcpu->kvm->srcu, idx);
3449         pagefault_enable();
3450         kvm_x86_ops->vcpu_put(vcpu);
3451         vcpu->arch.last_host_tsc = rdtsc();
3452         /*
3453          * If userspace has set any breakpoints or watchpoints, dr6 is restored
3454          * on every vmexit, but if not, we might have a stale dr6 from the
3455          * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3456          */
3457         set_debugreg(0, 6);
3458 }
3459
3460 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3461                                     struct kvm_lapic_state *s)
3462 {
3463         if (vcpu->arch.apicv_active)
3464                 kvm_x86_ops->sync_pir_to_irr(vcpu);
3465
3466         return kvm_apic_get_state(vcpu, s);
3467 }
3468
3469 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3470                                     struct kvm_lapic_state *s)
3471 {
3472         int r;
3473
3474         r = kvm_apic_set_state(vcpu, s);
3475         if (r)
3476                 return r;
3477         update_cr8_intercept(vcpu);
3478
3479         return 0;
3480 }
3481
3482 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3483 {
3484         return (!lapic_in_kernel(vcpu) ||
3485                 kvm_apic_accept_pic_intr(vcpu));
3486 }
3487
3488 /*
3489  * if userspace requested an interrupt window, check that the
3490  * interrupt window is open.
3491  *
3492  * No need to exit to userspace if we already have an interrupt queued.
3493  */
3494 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3495 {
3496         return kvm_arch_interrupt_allowed(vcpu) &&
3497                 !kvm_cpu_has_interrupt(vcpu) &&
3498                 !kvm_event_needs_reinjection(vcpu) &&
3499                 kvm_cpu_accept_dm_intr(vcpu);
3500 }
3501
3502 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3503                                     struct kvm_interrupt *irq)
3504 {
3505         if (irq->irq >= KVM_NR_INTERRUPTS)
3506                 return -EINVAL;
3507
3508         if (!irqchip_in_kernel(vcpu->kvm)) {
3509                 kvm_queue_interrupt(vcpu, irq->irq, false);
3510                 kvm_make_request(KVM_REQ_EVENT, vcpu);
3511                 return 0;
3512         }
3513
3514         /*
3515          * With in-kernel LAPIC, we only use this to inject EXTINT, so
3516          * fail for in-kernel 8259.
3517          */
3518         if (pic_in_kernel(vcpu->kvm))
3519                 return -ENXIO;
3520
3521         if (vcpu->arch.pending_external_vector != -1)
3522                 return -EEXIST;
3523
3524         vcpu->arch.pending_external_vector = irq->irq;
3525         kvm_make_request(KVM_REQ_EVENT, vcpu);
3526         return 0;
3527 }
3528
3529 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3530 {
3531         kvm_inject_nmi(vcpu);
3532
3533         return 0;
3534 }
3535
3536 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3537 {
3538         kvm_make_request(KVM_REQ_SMI, vcpu);
3539
3540         return 0;
3541 }
3542
3543 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3544                                            struct kvm_tpr_access_ctl *tac)
3545 {
3546         if (tac->flags)
3547                 return -EINVAL;
3548         vcpu->arch.tpr_access_reporting = !!tac->enabled;
3549         return 0;
3550 }
3551
3552 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3553                                         u64 mcg_cap)
3554 {
3555         int r;
3556         unsigned bank_num = mcg_cap & 0xff, bank;
3557
3558         r = -EINVAL;
3559         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3560                 goto out;
3561         if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3562                 goto out;
3563         r = 0;
3564         vcpu->arch.mcg_cap = mcg_cap;
3565         /* Init IA32_MCG_CTL to all 1s */
3566         if (mcg_cap & MCG_CTL_P)
3567                 vcpu->arch.mcg_ctl = ~(u64)0;
3568         /* Init IA32_MCi_CTL to all 1s */
3569         for (bank = 0; bank < bank_num; bank++)
3570                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3571
3572         kvm_x86_ops->setup_mce(vcpu);
3573 out:
3574         return r;
3575 }
3576
3577 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3578                                       struct kvm_x86_mce *mce)
3579 {
3580         u64 mcg_cap = vcpu->arch.mcg_cap;
3581         unsigned bank_num = mcg_cap & 0xff;
3582         u64 *banks = vcpu->arch.mce_banks;
3583
3584         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3585                 return -EINVAL;
3586         /*
3587          * if IA32_MCG_CTL is not all 1s, the uncorrected error
3588          * reporting is disabled
3589          */
3590         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3591             vcpu->arch.mcg_ctl != ~(u64)0)
3592                 return 0;
3593         banks += 4 * mce->bank;
3594         /*
3595          * if IA32_MCi_CTL is not all 1s, the uncorrected error
3596          * reporting is disabled for the bank
3597          */
3598         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3599                 return 0;
3600         if (mce->status & MCI_STATUS_UC) {
3601                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3602                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3603                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3604                         return 0;
3605                 }
3606                 if (banks[1] & MCI_STATUS_VAL)
3607                         mce->status |= MCI_STATUS_OVER;
3608                 banks[2] = mce->addr;
3609                 banks[3] = mce->misc;
3610                 vcpu->arch.mcg_status = mce->mcg_status;
3611                 banks[1] = mce->status;
3612                 kvm_queue_exception(vcpu, MC_VECTOR);
3613         } else if (!(banks[1] & MCI_STATUS_VAL)
3614                    || !(banks[1] & MCI_STATUS_UC)) {
3615                 if (banks[1] & MCI_STATUS_VAL)
3616                         mce->status |= MCI_STATUS_OVER;
3617                 banks[2] = mce->addr;
3618                 banks[3] = mce->misc;
3619                 banks[1] = mce->status;
3620         } else
3621                 banks[1] |= MCI_STATUS_OVER;
3622         return 0;
3623 }
3624
3625 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3626                                                struct kvm_vcpu_events *events)
3627 {
3628         process_nmi(vcpu);
3629
3630         /*
3631          * The API doesn't provide the instruction length for software
3632          * exceptions, so don't report them. As long as the guest RIP
3633          * isn't advanced, we should expect to encounter the exception
3634          * again.
3635          */
3636         if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3637                 events->exception.injected = 0;
3638                 events->exception.pending = 0;
3639         } else {
3640                 events->exception.injected = vcpu->arch.exception.injected;
3641                 events->exception.pending = vcpu->arch.exception.pending;
3642                 /*
3643                  * For ABI compatibility, deliberately conflate
3644                  * pending and injected exceptions when
3645                  * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3646                  */
3647                 if (!vcpu->kvm->arch.exception_payload_enabled)
3648                         events->exception.injected |=
3649                                 vcpu->arch.exception.pending;
3650         }
3651         events->exception.nr = vcpu->arch.exception.nr;
3652         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3653         events->exception.error_code = vcpu->arch.exception.error_code;
3654         events->exception_has_payload = vcpu->arch.exception.has_payload;
3655         events->exception_payload = vcpu->arch.exception.payload;
3656
3657         events->interrupt.injected =
3658                 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3659         events->interrupt.nr = vcpu->arch.interrupt.nr;
3660         events->interrupt.soft = 0;
3661         events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3662
3663         events->nmi.injected = vcpu->arch.nmi_injected;
3664         events->nmi.pending = vcpu->arch.nmi_pending != 0;
3665         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3666         events->nmi.pad = 0;
3667
3668         events->sipi_vector = 0; /* never valid when reporting to user space */
3669
3670         events->smi.smm = is_smm(vcpu);
3671         events->smi.pending = vcpu->arch.smi_pending;
3672         events->smi.smm_inside_nmi =
3673                 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3674         events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3675
3676         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3677                          | KVM_VCPUEVENT_VALID_SHADOW
3678                          | KVM_VCPUEVENT_VALID_SMM);
3679         if (vcpu->kvm->arch.exception_payload_enabled)
3680                 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3681
3682         memset(&events->reserved, 0, sizeof(events->reserved));
3683 }
3684
3685 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3686
3687 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3688                                               struct kvm_vcpu_events *events)
3689 {
3690         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3691                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3692                               | KVM_VCPUEVENT_VALID_SHADOW
3693                               | KVM_VCPUEVENT_VALID_SMM
3694                               | KVM_VCPUEVENT_VALID_PAYLOAD))
3695                 return -EINVAL;
3696
3697         if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3698                 if (!vcpu->kvm->arch.exception_payload_enabled)
3699                         return -EINVAL;
3700                 if (events->exception.pending)
3701                         events->exception.injected = 0;
3702                 else
3703                         events->exception_has_payload = 0;
3704         } else {
3705                 events->exception.pending = 0;
3706                 events->exception_has_payload = 0;
3707         }
3708
3709         if ((events->exception.injected || events->exception.pending) &&
3710             (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3711                 return -EINVAL;
3712
3713         /* INITs are latched while in SMM */
3714         if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3715             (events->smi.smm || events->smi.pending) &&
3716             vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3717                 return -EINVAL;
3718
3719         process_nmi(vcpu);
3720         vcpu->arch.exception.injected = events->exception.injected;
3721         vcpu->arch.exception.pending = events->exception.pending;
3722         vcpu->arch.exception.nr = events->exception.nr;
3723         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3724         vcpu->arch.exception.error_code = events->exception.error_code;
3725         vcpu->arch.exception.has_payload = events->exception_has_payload;
3726         vcpu->arch.exception.payload = events->exception_payload;
3727
3728         vcpu->arch.interrupt.injected = events->interrupt.injected;
3729         vcpu->arch.interrupt.nr = events->interrupt.nr;
3730         vcpu->arch.interrupt.soft = events->interrupt.soft;
3731         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3732                 kvm_x86_ops->set_interrupt_shadow(vcpu,
3733                                                   events->interrupt.shadow);
3734
3735         vcpu->arch.nmi_injected = events->nmi.injected;
3736         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3737                 vcpu->arch.nmi_pending = events->nmi.pending;
3738         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3739
3740         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3741             lapic_in_kernel(vcpu))
3742                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3743
3744         if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3745                 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3746                         if (events->smi.smm)
3747                                 vcpu->arch.hflags |= HF_SMM_MASK;
3748                         else
3749                                 vcpu->arch.hflags &= ~HF_SMM_MASK;
3750                         kvm_smm_changed(vcpu);
3751                 }
3752
3753                 vcpu->arch.smi_pending = events->smi.pending;
3754
3755                 if (events->smi.smm) {
3756                         if (events->smi.smm_inside_nmi)
3757                                 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3758                         else
3759                                 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3760                         if (lapic_in_kernel(vcpu)) {
3761                                 if (events->smi.latched_init)
3762                                         set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3763                                 else
3764                                         clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3765                         }
3766                 }
3767         }
3768
3769         kvm_make_request(KVM_REQ_EVENT, vcpu);
3770
3771         return 0;
3772 }
3773
3774 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3775                                              struct kvm_debugregs *dbgregs)
3776 {
3777         unsigned long val;
3778
3779         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3780         kvm_get_dr(vcpu, 6, &val);
3781         dbgregs->dr6 = val;
3782         dbgregs->dr7 = vcpu->arch.dr7;
3783         dbgregs->flags = 0;
3784         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3785 }
3786
3787 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3788                                             struct kvm_debugregs *dbgregs)
3789 {
3790         if (dbgregs->flags)
3791                 return -EINVAL;
3792
3793         if (dbgregs->dr6 & ~0xffffffffull)
3794                 return -EINVAL;
3795         if (dbgregs->dr7 & ~0xffffffffull)
3796                 return -EINVAL;
3797
3798         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3799         kvm_update_dr0123(vcpu);
3800         vcpu->arch.dr6 = dbgregs->dr6;
3801         kvm_update_dr6(vcpu);
3802         vcpu->arch.dr7 = dbgregs->dr7;
3803         kvm_update_dr7(vcpu);
3804
3805         return 0;
3806 }
3807
3808 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3809
3810 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3811 {
3812         struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3813         u64 xstate_bv = xsave->header.xfeatures;
3814         u64 valid;
3815
3816         /*
3817          * Copy legacy XSAVE area, to avoid complications with CPUID
3818          * leaves 0 and 1 in the loop below.
3819          */
3820         memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3821
3822         /* Set XSTATE_BV */
3823         xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3824         *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3825
3826         /*
3827          * Copy each region from the possibly compacted offset to the
3828          * non-compacted offset.
3829          */
3830         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3831         while (valid) {
3832                 u64 xfeature_mask = valid & -valid;
3833                 int xfeature_nr = fls64(xfeature_mask) - 1;
3834                 void *src = get_xsave_addr(xsave, xfeature_nr);
3835
3836                 if (src) {
3837                         u32 size, offset, ecx, edx;
3838                         cpuid_count(XSTATE_CPUID, xfeature_nr,
3839                                     &size, &offset, &ecx, &edx);
3840                         if (xfeature_nr == XFEATURE_PKRU)
3841                                 memcpy(dest + offset, &vcpu->arch.pkru,
3842                                        sizeof(vcpu->arch.pkru));
3843                         else
3844                                 memcpy(dest + offset, src, size);
3845
3846                 }
3847
3848                 valid -= xfeature_mask;
3849         }
3850 }
3851
3852 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3853 {
3854         struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3855         u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3856         u64 valid;
3857
3858         /*
3859          * Copy legacy XSAVE area, to avoid complications with CPUID
3860          * leaves 0 and 1 in the loop below.
3861          */
3862         memcpy(xsave, src, XSAVE_HDR_OFFSET);
3863
3864         /* Set XSTATE_BV and possibly XCOMP_BV.  */
3865         xsave->header.xfeatures = xstate_bv;
3866         if (boot_cpu_has(X86_FEATURE_XSAVES))
3867                 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3868
3869         /*
3870          * Copy each region from the non-compacted offset to the
3871          * possibly compacted offset.
3872          */
3873         valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3874         while (valid) {
3875                 u64 xfeature_mask = valid & -valid;
3876                 int xfeature_nr = fls64(xfeature_mask) - 1;
3877                 void *dest = get_xsave_addr(xsave, xfeature_nr);
3878
3879                 if (dest) {
3880                         u32 size, offset, ecx, edx;
3881                         cpuid_count(XSTATE_CPUID, xfeature_nr,
3882                                     &size, &offset, &ecx, &edx);
3883                         if (xfeature_nr == XFEATURE_PKRU)
3884                                 memcpy(&vcpu->arch.pkru, src + offset,
3885                                        sizeof(vcpu->arch.pkru));
3886                         else
3887                                 memcpy(dest, src + offset, size);
3888                 }
3889
3890                 valid -= xfeature_mask;
3891         }
3892 }
3893
3894 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3895                                          struct kvm_xsave *guest_xsave)
3896 {
3897         if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3898                 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3899                 fill_xsave((u8 *) guest_xsave->region, vcpu);
3900         } else {
3901                 memcpy(guest_xsave->region,
3902                         &vcpu->arch.guest_fpu->state.fxsave,
3903                         sizeof(struct fxregs_state));
3904                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3905                         XFEATURE_MASK_FPSSE;
3906         }
3907 }
3908
3909 #define XSAVE_MXCSR_OFFSET 24
3910
3911 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3912                                         struct kvm_xsave *guest_xsave)
3913 {
3914         u64 xstate_bv =
3915                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3916         u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3917
3918         if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3919                 /*
3920                  * Here we allow setting states that are not present in
3921                  * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3922                  * with old userspace.
3923                  */
3924                 if (xstate_bv & ~kvm_supported_xcr0() ||
3925                         mxcsr & ~mxcsr_feature_mask)
3926                         return -EINVAL;
3927                 load_xsave(vcpu, (u8 *)guest_xsave->region);
3928         } else {
3929                 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3930                         mxcsr & ~mxcsr_feature_mask)
3931                         return -EINVAL;
3932                 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3933                         guest_xsave->region, sizeof(struct fxregs_state));
3934         }
3935         return 0;
3936 }
3937
3938 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3939                                         struct kvm_xcrs *guest_xcrs)
3940 {
3941         if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3942                 guest_xcrs->nr_xcrs = 0;
3943                 return;
3944         }
3945
3946         guest_xcrs->nr_xcrs = 1;
3947         guest_xcrs->flags = 0;
3948         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3949         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3950 }
3951
3952 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3953                                        struct kvm_xcrs *guest_xcrs)
3954 {
3955         int i, r = 0;
3956
3957         if (!boot_cpu_has(X86_FEATURE_XSAVE))
3958                 return -EINVAL;
3959
3960         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3961                 return -EINVAL;
3962
3963         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3964                 /* Only support XCR0 currently */
3965                 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3966                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3967                                 guest_xcrs->xcrs[i].value);
3968                         break;
3969                 }
3970         if (r)
3971                 r = -EINVAL;
3972         return r;
3973 }
3974
3975 /*
3976  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3977  * stopped by the hypervisor.  This function will be called from the host only.
3978  * EINVAL is returned when the host attempts to set the flag for a guest that
3979  * does not support pv clocks.
3980  */
3981 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3982 {
3983         if (!vcpu->arch.pv_time_enabled)
3984                 return -EINVAL;
3985         vcpu->arch.pvclock_set_guest_stopped_request = true;
3986         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3987         return 0;
3988 }
3989
3990 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3991                                      struct kvm_enable_cap *cap)
3992 {
3993         int r;
3994         uint16_t vmcs_version;
3995         void __user *user_ptr;
3996
3997         if (cap->flags)
3998                 return -EINVAL;
3999
4000         switch (cap->cap) {
4001         case KVM_CAP_HYPERV_SYNIC2:
4002                 if (cap->args[0])
4003                         return -EINVAL;
4004                 /* fall through */
4005
4006         case KVM_CAP_HYPERV_SYNIC:
4007                 if (!irqchip_in_kernel(vcpu->kvm))
4008                         return -EINVAL;
4009                 return kvm_hv_activate_synic(vcpu, cap->cap ==
4010                                              KVM_CAP_HYPERV_SYNIC2);
4011         case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4012                 if (!kvm_x86_ops->nested_enable_evmcs)
4013                         return -ENOTTY;
4014                 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4015                 if (!r) {
4016                         user_ptr = (void __user *)(uintptr_t)cap->args[0];
4017                         if (copy_to_user(user_ptr, &vmcs_version,
4018                                          sizeof(vmcs_version)))
4019                                 r = -EFAULT;
4020                 }
4021                 return r;
4022
4023         default:
4024                 return -EINVAL;
4025         }
4026 }
4027
4028 long kvm_arch_vcpu_ioctl(struct file *filp,
4029                          unsigned int ioctl, unsigned long arg)
4030 {
4031         struct kvm_vcpu *vcpu = filp->private_data;
4032         void __user *argp = (void __user *)arg;
4033         int r;
4034         union {
4035                 struct kvm_lapic_state *lapic;
4036                 struct kvm_xsave *xsave;
4037                 struct kvm_xcrs *xcrs;
4038                 void *buffer;
4039         } u;
4040
4041         vcpu_load(vcpu);
4042
4043         u.buffer = NULL;
4044         switch (ioctl) {
4045         case KVM_GET_LAPIC: {
4046                 r = -EINVAL;
4047                 if (!lapic_in_kernel(vcpu))
4048                         goto out;
4049                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4050                                 GFP_KERNEL_ACCOUNT);
4051
4052                 r = -ENOMEM;
4053                 if (!u.lapic)
4054                         goto out;
4055                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4056                 if (r)
4057                         goto out;
4058                 r = -EFAULT;
4059                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4060                         goto out;
4061                 r = 0;
4062                 break;
4063         }
4064         case KVM_SET_LAPIC: {
4065                 r = -EINVAL;
4066                 if (!lapic_in_kernel(vcpu))
4067                         goto out;
4068                 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4069                 if (IS_ERR(u.lapic)) {
4070                         r = PTR_ERR(u.lapic);
4071                         goto out_nofree;
4072                 }
4073
4074                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4075                 break;
4076         }
4077         case KVM_INTERRUPT: {
4078                 struct kvm_interrupt irq;
4079
4080                 r = -EFAULT;
4081                 if (copy_from_user(&irq, argp, sizeof(irq)))
4082                         goto out;
4083                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4084                 break;
4085         }
4086         case KVM_NMI: {
4087                 r = kvm_vcpu_ioctl_nmi(vcpu);
4088                 break;
4089         }
4090         case KVM_SMI: {
4091                 r = kvm_vcpu_ioctl_smi(vcpu);
4092                 break;
4093         }
4094         case KVM_SET_CPUID: {
4095                 struct kvm_cpuid __user *cpuid_arg = argp;
4096                 struct kvm_cpuid cpuid;
4097
4098                 r = -EFAULT;
4099                 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4100                         goto out;
4101                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4102                 break;
4103         }
4104         case KVM_SET_CPUID2: {
4105                 struct kvm_cpuid2 __user *cpuid_arg = argp;
4106                 struct kvm_cpuid2 cpuid;
4107
4108                 r = -EFAULT;
4109                 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4110                         goto out;
4111                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4112                                               cpuid_arg->entries);
4113                 break;
4114         }
4115         case KVM_GET_CPUID2: {
4116                 struct kvm_cpuid2 __user *cpuid_arg = argp;
4117                 struct kvm_cpuid2 cpuid;
4118
4119                 r = -EFAULT;
4120                 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4121                         goto out;
4122                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4123                                               cpuid_arg->entries);
4124                 if (r)
4125                         goto out;
4126                 r = -EFAULT;
4127                 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4128                         goto out;
4129                 r = 0;
4130                 break;
4131         }
4132         case KVM_GET_MSRS: {
4133                 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4134                 r = msr_io(vcpu, argp, do_get_msr, 1);
4135                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4136                 break;
4137         }
4138         case KVM_SET_MSRS: {
4139                 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4140                 r = msr_io(vcpu, argp, do_set_msr, 0);
4141                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4142                 break;
4143         }
4144         case KVM_TPR_ACCESS_REPORTING: {
4145                 struct kvm_tpr_access_ctl tac;
4146
4147                 r = -EFAULT;
4148                 if (copy_from_user(&tac, argp, sizeof(tac)))
4149                         goto out;
4150                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4151                 if (r)
4152                         goto out;
4153                 r = -EFAULT;
4154                 if (copy_to_user(argp, &tac, sizeof(tac)))
4155                         goto out;
4156                 r = 0;
4157                 break;
4158         };
4159         case KVM_SET_VAPIC_ADDR: {
4160                 struct kvm_vapic_addr va;
4161                 int idx;
4162
4163                 r = -EINVAL;
4164                 if (!lapic_in_kernel(vcpu))
4165                         goto out;
4166                 r = -EFAULT;
4167                 if (copy_from_user(&va, argp, sizeof(va)))
4168                         goto out;
4169                 idx = srcu_read_lock(&vcpu->kvm->srcu);
4170                 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4171                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4172                 break;
4173         }
4174         case KVM_X86_SETUP_MCE: {
4175                 u64 mcg_cap;
4176
4177                 r = -EFAULT;
4178                 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4179                         goto out;
4180                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4181                 break;
4182         }
4183         case KVM_X86_SET_MCE: {
4184                 struct kvm_x86_mce mce;
4185
4186                 r = -EFAULT;
4187                 if (copy_from_user(&mce, argp, sizeof(mce)))
4188                         goto out;
4189                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4190                 break;
4191         }
4192         case KVM_GET_VCPU_EVENTS: {
4193                 struct kvm_vcpu_events events;
4194
4195                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4196
4197                 r = -EFAULT;
4198                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4199                         break;
4200                 r = 0;
4201                 break;
4202         }
4203         case KVM_SET_VCPU_EVENTS: {
4204                 struct kvm_vcpu_events events;
4205
4206                 r = -EFAULT;
4207                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4208                         break;
4209
4210                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4211                 break;
4212         }
4213         case KVM_GET_DEBUGREGS: {
4214                 struct kvm_debugregs dbgregs;
4215
4216                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4217
4218                 r = -EFAULT;
4219                 if (copy_to_user(argp, &dbgregs,
4220                                  sizeof(struct kvm_debugregs)))
4221                         break;
4222                 r = 0;
4223                 break;
4224         }
4225         case KVM_SET_DEBUGREGS: {
4226                 struct kvm_debugregs dbgregs;
4227
4228                 r = -EFAULT;
4229                 if (copy_from_user(&dbgregs, argp,
4230                                    sizeof(struct kvm_debugregs)))
4231                         break;
4232
4233                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4234                 break;
4235         }
4236         case KVM_GET_XSAVE: {
4237                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4238                 r = -ENOMEM;
4239                 if (!u.xsave)
4240                         break;
4241
4242                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4243
4244                 r = -EFAULT;
4245                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4246                         break;
4247                 r = 0;
4248                 break;
4249         }
4250         case KVM_SET_XSAVE: {
4251                 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4252                 if (IS_ERR(u.xsave)) {
4253                         r = PTR_ERR(u.xsave);
4254                         goto out_nofree;
4255                 }
4256
4257                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4258                 break;
4259         }
4260         case KVM_GET_XCRS: {
4261                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4262                 r = -ENOMEM;
4263                 if (!u.xcrs)
4264                         break;
4265
4266                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4267
4268                 r = -EFAULT;
4269                 if (copy_to_user(argp, u.xcrs,
4270                                  sizeof(struct kvm_xcrs)))
4271                         break;
4272                 r = 0;
4273                 break;
4274         }
4275         case KVM_SET_XCRS: {
4276                 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4277                 if (IS_ERR(u.xcrs)) {
4278                         r = PTR_ERR(u.xcrs);
4279                         goto out_nofree;
4280                 }
4281
4282                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4283                 break;
4284         }
4285         case KVM_SET_TSC_KHZ: {
4286                 u32 user_tsc_khz;
4287
4288                 r = -EINVAL;
4289                 user_tsc_khz = (u32)arg;
4290
4291                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4292                         goto out;
4293
4294                 if (user_tsc_khz == 0)
4295                         user_tsc_khz = tsc_khz;
4296
4297                 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4298                         r = 0;
4299
4300                 goto out;
4301         }
4302         case KVM_GET_TSC_KHZ: {
4303                 r = vcpu->arch.virtual_tsc_khz;
4304                 goto out;
4305         }
4306         case KVM_KVMCLOCK_CTRL: {
4307                 r = kvm_set_guest_paused(vcpu);
4308                 goto out;
4309         }
4310         case KVM_ENABLE_CAP: {
4311                 struct kvm_enable_cap cap;
4312
4313                 r = -EFAULT;
4314                 if (copy_from_user(&cap, argp, sizeof(cap)))
4315                         goto out;
4316                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4317                 break;
4318         }
4319         case KVM_GET_NESTED_STATE: {
4320                 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4321                 u32 user_data_size;
4322
4323                 r = -EINVAL;
4324                 if (!kvm_x86_ops->get_nested_state)
4325                         break;
4326
4327                 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4328                 r = -EFAULT;
4329                 if (get_user(user_data_size, &user_kvm_nested_state->size))
4330                         break;
4331
4332                 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4333                                                   user_data_size);
4334                 if (r < 0)
4335                         break;
4336
4337                 if (r > user_data_size) {
4338                         if (put_user(r, &user_kvm_nested_state->size))
4339                                 r = -EFAULT;
4340                         else
4341                                 r = -E2BIG;
4342                         break;
4343                 }
4344
4345                 r = 0;
4346                 break;
4347         }
4348         case KVM_SET_NESTED_STATE: {
4349                 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4350                 struct kvm_nested_state kvm_state;
4351
4352                 r = -EINVAL;
4353                 if (!kvm_x86_ops->set_nested_state)
4354                         break;
4355
4356                 r = -EFAULT;
4357                 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4358                         break;
4359
4360                 r = -EINVAL;
4361                 if (kvm_state.size < sizeof(kvm_state))
4362                         break;
4363
4364                 if (kvm_state.flags &
4365                     ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4366                       | KVM_STATE_NESTED_EVMCS))
4367                         break;
4368
4369                 /* nested_run_pending implies guest_mode.  */
4370                 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4371                     && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4372                         break;
4373
4374                 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4375                 break;
4376         }
4377         case KVM_GET_SUPPORTED_HV_CPUID: {
4378                 struct kvm_cpuid2 __user *cpuid_arg = argp;
4379                 struct kvm_cpuid2 cpuid;
4380
4381                 r = -EFAULT;
4382                 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4383                         goto out;
4384
4385                 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4386                                                 cpuid_arg->entries);
4387                 if (r)
4388                         goto out;
4389
4390                 r = -EFAULT;
4391                 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4392                         goto out;
4393                 r = 0;
4394                 break;
4395         }
4396         default:
4397                 r = -EINVAL;
4398         }
4399 out:
4400         kfree(u.buffer);
4401 out_nofree:
4402         vcpu_put(vcpu);
4403         return r;
4404 }
4405
4406 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4407 {
4408         return VM_FAULT_SIGBUS;
4409 }
4410
4411 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4412 {
4413         int ret;
4414
4415         if (addr > (unsigned int)(-3 * PAGE_SIZE))
4416                 return -EINVAL;
4417         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4418         return ret;
4419 }
4420
4421 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4422                                               u64 ident_addr)
4423 {
4424         return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4425 }
4426
4427 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4428                                          unsigned long kvm_nr_mmu_pages)
4429 {
4430         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4431                 return -EINVAL;
4432
4433         mutex_lock(&kvm->slots_lock);
4434
4435         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4436         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4437
4438         mutex_unlock(&kvm->slots_lock);
4439         return 0;
4440 }
4441
4442 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4443 {
4444         return kvm->arch.n_max_mmu_pages;
4445 }
4446
4447 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4448 {
4449         struct kvm_pic *pic = kvm->arch.vpic;
4450         int r;
4451
4452         r = 0;
4453         switch (chip->chip_id) {
4454         case KVM_IRQCHIP_PIC_MASTER:
4455                 memcpy(&chip->chip.pic, &pic->pics[0],
4456                         sizeof(struct kvm_pic_state));
4457                 break;
4458         case KVM_IRQCHIP_PIC_SLAVE:
4459                 memcpy(&chip->chip.pic, &pic->pics[1],
4460                         sizeof(struct kvm_pic_state));
4461                 break;
4462         case KVM_IRQCHIP_IOAPIC:
4463                 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4464                 break;
4465         default:
4466                 r = -EINVAL;
4467                 break;
4468         }
4469         return r;
4470 }
4471
4472 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4473 {
4474         struct kvm_pic *pic = kvm->arch.vpic;
4475         int r;
4476
4477         r = 0;
4478         switch (chip->chip_id) {
4479         case KVM_IRQCHIP_PIC_MASTER:
4480                 spin_lock(&pic->lock);
4481                 memcpy(&pic->pics[0], &chip->chip.pic,
4482                         sizeof(struct kvm_pic_state));
4483                 spin_unlock(&pic->lock);
4484                 break;
4485         case KVM_IRQCHIP_PIC_SLAVE:
4486                 spin_lock(&pic->lock);
4487                 memcpy(&pic->pics[1], &chip->chip.pic,
4488                         sizeof(struct kvm_pic_state));
4489                 spin_unlock(&pic->lock);
4490                 break;
4491         case KVM_IRQCHIP_IOAPIC:
4492                 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4493                 break;
4494         default:
4495                 r = -EINVAL;
4496                 break;
4497         }
4498         kvm_pic_update_irq(pic);
4499         return r;
4500 }
4501
4502 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4503 {
4504         struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4505
4506         BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4507
4508         mutex_lock(&kps->lock);
4509         memcpy(ps, &kps->channels, sizeof(*ps));
4510         mutex_unlock(&kps->lock);
4511         return 0;
4512 }
4513
4514 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4515 {
4516         int i;
4517         struct kvm_pit *pit = kvm->arch.vpit;
4518
4519         mutex_lock(&pit->pit_state.lock);
4520         memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4521         for (i = 0; i < 3; i++)
4522                 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4523         mutex_unlock(&pit->pit_state.lock);
4524         return 0;
4525 }
4526
4527 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4528 {
4529         mutex_lock(&kvm->arch.vpit->pit_state.lock);
4530         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4531                 sizeof(ps->channels));
4532         ps->flags = kvm->arch.vpit->pit_state.flags;
4533         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4534         memset(&ps->reserved, 0, sizeof(ps->reserved));
4535         return 0;
4536 }
4537
4538 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4539 {
4540         int start = 0;
4541         int i;
4542         u32 prev_legacy, cur_legacy;
4543         struct kvm_pit *pit = kvm->arch.vpit;
4544
4545         mutex_lock(&pit->pit_state.lock);
4546         prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4547         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4548         if (!prev_legacy && cur_legacy)
4549                 start = 1;
4550         memcpy(&pit->pit_state.channels, &ps->channels,
4551                sizeof(pit->pit_state.channels));
4552         pit->pit_state.flags = ps->flags;
4553         for (i = 0; i < 3; i++)
4554                 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4555                                    start && i == 0);
4556         mutex_unlock(&pit->pit_state.lock);
4557         return 0;
4558 }
4559
4560 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4561                                  struct kvm_reinject_control *control)
4562 {
4563         struct kvm_pit *pit = kvm->arch.vpit;
4564
4565         if (!pit)
4566                 return -ENXIO;
4567
4568         /* pit->pit_state.lock was overloaded to prevent userspace from getting
4569          * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4570          * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4571          */
4572         mutex_lock(&pit->pit_state.lock);
4573         kvm_pit_set_reinject(pit, control->pit_reinject);
4574         mutex_unlock(&pit->pit_state.lock);
4575
4576         return 0;
4577 }
4578
4579 /**
4580  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4581  * @kvm: kvm instance
4582  * @log: slot id and address to which we copy the log
4583  *
4584  * Steps 1-4 below provide general overview of dirty page logging. See
4585  * kvm_get_dirty_log_protect() function description for additional details.
4586  *
4587  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4588  * always flush the TLB (step 4) even if previous step failed  and the dirty
4589  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4590  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4591  * writes will be marked dirty for next log read.
4592  *
4593  *   1. Take a snapshot of the bit and clear it if needed.
4594  *   2. Write protect the corresponding page.
4595  *   3. Copy the snapshot to the userspace.
4596  *   4. Flush TLB's if needed.
4597  */
4598 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4599 {
4600         bool flush = false;
4601         int r;
4602
4603         mutex_lock(&kvm->slots_lock);
4604
4605         /*
4606          * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4607          */
4608         if (kvm_x86_ops->flush_log_dirty)
4609                 kvm_x86_ops->flush_log_dirty(kvm);
4610
4611         r = kvm_get_dirty_log_protect(kvm, log, &flush);
4612
4613         /*
4614          * All the TLBs can be flushed out of mmu lock, see the comments in
4615          * kvm_mmu_slot_remove_write_access().
4616          */
4617         lockdep_assert_held(&kvm->slots_lock);
4618         if (flush)
4619                 kvm_flush_remote_tlbs(kvm);
4620
4621         mutex_unlock(&kvm->slots_lock);
4622         return r;
4623 }
4624
4625 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4626 {
4627         bool flush = false;
4628         int r;
4629
4630         mutex_lock(&kvm->slots_lock);
4631
4632         /*
4633          * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4634          */
4635         if (kvm_x86_ops->flush_log_dirty)
4636                 kvm_x86_ops->flush_log_dirty(kvm);
4637
4638         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4639
4640         /*
4641          * All the TLBs can be flushed out of mmu lock, see the comments in
4642          * kvm_mmu_slot_remove_write_access().
4643          */
4644         lockdep_assert_held(&kvm->slots_lock);
4645         if (flush)
4646                 kvm_flush_remote_tlbs(kvm);
4647
4648         mutex_unlock(&kvm->slots_lock);
4649         return r;
4650 }
4651
4652 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4653                         bool line_status)
4654 {
4655         if (!irqchip_in_kernel(kvm))
4656                 return -ENXIO;
4657
4658         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4659                                         irq_event->irq, irq_event->level,
4660                                         line_status);
4661         return 0;
4662 }
4663
4664 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4665                             struct kvm_enable_cap *cap)
4666 {
4667         int r;
4668
4669         if (cap->flags)
4670                 return -EINVAL;
4671
4672         switch (cap->cap) {
4673         case KVM_CAP_DISABLE_QUIRKS:
4674                 kvm->arch.disabled_quirks = cap->args[0];
4675                 r = 0;
4676                 break;
4677         case KVM_CAP_SPLIT_IRQCHIP: {
4678                 mutex_lock(&kvm->lock);
4679                 r = -EINVAL;
4680                 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4681                         goto split_irqchip_unlock;
4682                 r = -EEXIST;
4683                 if (irqchip_in_kernel(kvm))
4684                         goto split_irqchip_unlock;
4685                 if (kvm->created_vcpus)
4686                         goto split_irqchip_unlock;
4687                 r = kvm_setup_empty_irq_routing(kvm);
4688                 if (r)
4689                         goto split_irqchip_unlock;
4690                 /* Pairs with irqchip_in_kernel. */
4691                 smp_wmb();
4692                 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4693                 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4694                 r = 0;
4695 split_irqchip_unlock:
4696                 mutex_unlock(&kvm->lock);
4697                 break;
4698         }
4699         case KVM_CAP_X2APIC_API:
4700                 r = -EINVAL;
4701                 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4702                         break;
4703
4704                 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4705                         kvm->arch.x2apic_format = true;
4706                 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4707                         kvm->arch.x2apic_broadcast_quirk_disabled = true;
4708
4709                 r = 0;
4710                 break;
4711         case KVM_CAP_X86_DISABLE_EXITS:
4712                 r = -EINVAL;
4713                 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4714                         break;
4715
4716                 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4717                         kvm_can_mwait_in_guest())
4718                         kvm->arch.mwait_in_guest = true;
4719                 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4720                         kvm->arch.hlt_in_guest = true;
4721                 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4722                         kvm->arch.pause_in_guest = true;
4723                 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4724                         kvm->arch.cstate_in_guest = true;
4725                 r = 0;
4726                 break;
4727         case KVM_CAP_MSR_PLATFORM_INFO:
4728                 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4729                 r = 0;
4730                 break;
4731         case KVM_CAP_EXCEPTION_PAYLOAD:
4732                 kvm->arch.exception_payload_enabled = cap->args[0];
4733                 r = 0;
4734                 break;
4735         default:
4736                 r = -EINVAL;
4737                 break;
4738         }
4739         return r;
4740 }
4741
4742 long kvm_arch_vm_ioctl(struct file *filp,
4743                        unsigned int ioctl, unsigned long arg)
4744 {
4745         struct kvm *kvm = filp->private_data;
4746         void __user *argp = (void __user *)arg;
4747         int r = -ENOTTY;
4748         /*
4749          * This union makes it completely explicit to gcc-3.x
4750          * that these two variables' stack usage should be
4751          * combined, not added together.
4752          */
4753         union {
4754                 struct kvm_pit_state ps;
4755                 struct kvm_pit_state2 ps2;
4756                 struct kvm_pit_config pit_config;
4757         } u;
4758
4759         switch (ioctl) {
4760         case KVM_SET_TSS_ADDR:
4761                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4762                 break;
4763         case KVM_SET_IDENTITY_MAP_ADDR: {
4764                 u64 ident_addr;
4765
4766                 mutex_lock(&kvm->lock);
4767                 r = -EINVAL;
4768                 if (kvm->created_vcpus)
4769                         goto set_identity_unlock;
4770                 r = -EFAULT;
4771                 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4772                         goto set_identity_unlock;
4773                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4774 set_identity_unlock:
4775                 mutex_unlock(&kvm->lock);
4776                 break;
4777         }
4778         case KVM_SET_NR_MMU_PAGES:
4779                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4780                 break;
4781         case KVM_GET_NR_MMU_PAGES:
4782                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4783                 break;
4784         case KVM_CREATE_IRQCHIP: {
4785                 mutex_lock(&kvm->lock);
4786
4787                 r = -EEXIST;
4788                 if (irqchip_in_kernel(kvm))
4789                         goto create_irqchip_unlock;
4790
4791                 r = -EINVAL;
4792                 if (kvm->created_vcpus)
4793                         goto create_irqchip_unlock;
4794
4795                 r = kvm_pic_init(kvm);
4796                 if (r)
4797                         goto create_irqchip_unlock;
4798
4799                 r = kvm_ioapic_init(kvm);
4800                 if (r) {
4801                         kvm_pic_destroy(kvm);
4802                         goto create_irqchip_unlock;
4803                 }
4804
4805                 r = kvm_setup_default_irq_routing(kvm);
4806                 if (r) {
4807                         kvm_ioapic_destroy(kvm);
4808                         kvm_pic_destroy(kvm);
4809                         goto create_irqchip_unlock;
4810                 }
4811                 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4812                 smp_wmb();
4813                 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4814         create_irqchip_unlock:
4815                 mutex_unlock(&kvm->lock);
4816                 break;
4817         }
4818         case KVM_CREATE_PIT:
4819                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4820                 goto create_pit;
4821         case KVM_CREATE_PIT2:
4822                 r = -EFAULT;
4823                 if (copy_from_user(&u.pit_config, argp,
4824                                    sizeof(struct kvm_pit_config)))
4825                         goto out;
4826         create_pit:
4827                 mutex_lock(&kvm->lock);
4828                 r = -EEXIST;
4829                 if (kvm->arch.vpit)
4830                         goto create_pit_unlock;
4831                 r = -ENOMEM;
4832                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4833                 if (kvm->arch.vpit)
4834                         r = 0;
4835         create_pit_unlock:
4836                 mutex_unlock(&kvm->lock);
4837                 break;
4838         case KVM_GET_IRQCHIP: {
4839                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4840                 struct kvm_irqchip *chip;
4841
4842                 chip = memdup_user(argp, sizeof(*chip));
4843                 if (IS_ERR(chip)) {
4844                         r = PTR_ERR(chip);
4845                         goto out;
4846                 }
4847
4848                 r = -ENXIO;
4849                 if (!irqchip_kernel(kvm))
4850                         goto get_irqchip_out;
4851                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4852                 if (r)
4853                         goto get_irqchip_out;
4854                 r = -EFAULT;
4855                 if (copy_to_user(argp, chip, sizeof(*chip)))
4856                         goto get_irqchip_out;
4857                 r = 0;
4858         get_irqchip_out:
4859                 kfree(chip);
4860                 break;
4861         }
4862         case KVM_SET_IRQCHIP: {
4863                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4864                 struct kvm_irqchip *chip;
4865
4866                 chip = memdup_user(argp, sizeof(*chip));
4867                 if (IS_ERR(chip)) {
4868                         r = PTR_ERR(chip);
4869                         goto out;
4870                 }
4871
4872                 r = -ENXIO;
4873                 if (!irqchip_kernel(kvm))
4874                         goto set_irqchip_out;
4875                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4876                 if (r)
4877                         goto set_irqchip_out;
4878                 r = 0;
4879         set_irqchip_out:
4880                 kfree(chip);
4881                 break;
4882         }
4883         case KVM_GET_PIT: {
4884                 r = -EFAULT;
4885                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4886                         goto out;
4887                 r = -ENXIO;
4888                 if (!kvm->arch.vpit)
4889                         goto out;
4890                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4891                 if (r)
4892                         goto out;
4893                 r = -EFAULT;
4894                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4895                         goto out;
4896                 r = 0;
4897                 break;
4898         }
4899         case KVM_SET_PIT: {
4900                 r = -EFAULT;
4901                 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4902                         goto out;
4903                 r = -ENXIO;
4904                 if (!kvm->arch.vpit)
4905                         goto out;
4906                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4907                 break;
4908         }
4909         case KVM_GET_PIT2: {
4910                 r = -ENXIO;
4911                 if (!kvm->arch.vpit)
4912                         goto out;
4913                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4914                 if (r)
4915                         goto out;
4916                 r = -EFAULT;
4917                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4918                         goto out;
4919                 r = 0;
4920                 break;
4921         }
4922         case KVM_SET_PIT2: {
4923                 r = -EFAULT;
4924                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4925                         goto out;
4926                 r = -ENXIO;
4927                 if (!kvm->arch.vpit)
4928                         goto out;
4929                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4930                 break;
4931         }
4932         case KVM_REINJECT_CONTROL: {
4933                 struct kvm_reinject_control control;
4934                 r =  -EFAULT;
4935                 if (copy_from_user(&control, argp, sizeof(control)))
4936                         goto out;
4937                 r = kvm_vm_ioctl_reinject(kvm, &control);
4938                 break;
4939         }
4940         case KVM_SET_BOOT_CPU_ID:
4941                 r = 0;
4942                 mutex_lock(&kvm->lock);
4943                 if (kvm->created_vcpus)
4944                         r = -EBUSY;
4945                 else
4946                         kvm->arch.bsp_vcpu_id = arg;
4947                 mutex_unlock(&kvm->lock);
4948                 break;
4949         case KVM_XEN_HVM_CONFIG: {
4950                 struct kvm_xen_hvm_config xhc;
4951                 r = -EFAULT;
4952                 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4953                         goto out;
4954                 r = -EINVAL;
4955                 if (xhc.flags)
4956                         goto out;
4957                 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4958                 r = 0;
4959                 break;
4960         }
4961         case KVM_SET_CLOCK: {
4962                 struct kvm_clock_data user_ns;
4963                 u64 now_ns;
4964
4965                 r = -EFAULT;
4966                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4967                         goto out;
4968
4969                 r = -EINVAL;
4970                 if (user_ns.flags)
4971                         goto out;
4972
4973                 r = 0;
4974                 /*
4975                  * TODO: userspace has to take care of races with VCPU_RUN, so
4976                  * kvm_gen_update_masterclock() can be cut down to locked
4977                  * pvclock_update_vm_gtod_copy().
4978                  */
4979                 kvm_gen_update_masterclock(kvm);
4980                 now_ns = get_kvmclock_ns(kvm);
4981                 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4982                 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4983                 break;
4984         }
4985         case KVM_GET_CLOCK: {
4986                 struct kvm_clock_data user_ns;
4987                 u64 now_ns;
4988
4989                 now_ns = get_kvmclock_ns(kvm);
4990                 user_ns.clock = now_ns;
4991                 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4992                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4993
4994                 r = -EFAULT;
4995                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4996                         goto out;
4997                 r = 0;
4998                 break;
4999         }
5000         case KVM_MEMORY_ENCRYPT_OP: {
5001                 r = -ENOTTY;
5002                 if (kvm_x86_ops->mem_enc_op)
5003                         r = kvm_x86_ops->mem_enc_op(kvm, argp);
5004                 break;
5005         }
5006         case KVM_MEMORY_ENCRYPT_REG_REGION: {
5007                 struct kvm_enc_region region;
5008
5009                 r = -EFAULT;
5010                 if (copy_from_user(&region, argp, sizeof(region)))
5011                         goto out;
5012
5013                 r = -ENOTTY;
5014                 if (kvm_x86_ops->mem_enc_reg_region)
5015                         r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
5016                 break;
5017         }
5018         case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5019                 struct kvm_enc_region region;
5020
5021                 r = -EFAULT;
5022                 if (copy_from_user(&region, argp, sizeof(region)))
5023                         goto out;
5024
5025                 r = -ENOTTY;
5026                 if (kvm_x86_ops->mem_enc_unreg_region)
5027                         r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
5028                 break;
5029         }
5030         case KVM_HYPERV_EVENTFD: {
5031                 struct kvm_hyperv_eventfd hvevfd;
5032
5033                 r = -EFAULT;
5034                 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5035                         goto out;
5036                 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5037                 break;
5038         }
5039         case KVM_SET_PMU_EVENT_FILTER:
5040                 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5041                 break;
5042         default:
5043                 r = -ENOTTY;
5044         }
5045 out:
5046         return r;
5047 }
5048
5049 static void kvm_init_msr_list(void)
5050 {
5051         u32 dummy[2];
5052         unsigned i, j;
5053
5054         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
5055                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
5056                         continue;
5057
5058                 /*
5059                  * Even MSRs that are valid in the host may not be exposed
5060                  * to the guests in some cases.
5061                  */
5062                 switch (msrs_to_save[i]) {
5063                 case MSR_IA32_BNDCFGS:
5064                         if (!kvm_mpx_supported())
5065                                 continue;
5066                         break;
5067                 case MSR_TSC_AUX:
5068                         if (!kvm_x86_ops->rdtscp_supported())
5069                                 continue;
5070                         break;
5071                 case MSR_IA32_RTIT_CTL:
5072                 case MSR_IA32_RTIT_STATUS:
5073                         if (!kvm_x86_ops->pt_supported())
5074                                 continue;
5075                         break;
5076                 case MSR_IA32_RTIT_CR3_MATCH:
5077                         if (!kvm_x86_ops->pt_supported() ||
5078                             !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5079                                 continue;
5080                         break;
5081                 case MSR_IA32_RTIT_OUTPUT_BASE:
5082                 case MSR_IA32_RTIT_OUTPUT_MASK:
5083                         if (!kvm_x86_ops->pt_supported() ||
5084                                 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5085                                  !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5086                                 continue;
5087                         break;
5088                 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5089                         if (!kvm_x86_ops->pt_supported() ||
5090                                 msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5091                                 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5092                                 continue;
5093                         break;
5094                 }
5095                 default:
5096                         break;
5097                 }
5098
5099                 if (j < i)
5100                         msrs_to_save[j] = msrs_to_save[i];
5101                 j++;
5102         }
5103         num_msrs_to_save = j;
5104
5105         for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5106                 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5107                         continue;
5108
5109                 if (j < i)
5110                         emulated_msrs[j] = emulated_msrs[i];
5111                 j++;
5112         }
5113         num_emulated_msrs = j;
5114
5115         for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5116                 struct kvm_msr_entry msr;
5117
5118                 msr.index = msr_based_features[i];
5119                 if (kvm_get_msr_feature(&msr))
5120                         continue;
5121
5122                 if (j < i)
5123                         msr_based_features[j] = msr_based_features[i];
5124                 j++;
5125         }
5126         num_msr_based_features = j;
5127 }
5128
5129 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5130                            const void *v)
5131 {
5132         int handled = 0;
5133         int n;
5134
5135         do {
5136                 n = min(len, 8);
5137                 if (!(lapic_in_kernel(vcpu) &&
5138                       !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5139                     && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5140                         break;
5141                 handled += n;
5142                 addr += n;
5143                 len -= n;
5144                 v += n;
5145         } while (len);
5146
5147         return handled;
5148 }
5149
5150 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5151 {
5152         int handled = 0;
5153         int n;
5154
5155         do {
5156                 n = min(len, 8);
5157                 if (!(lapic_in_kernel(vcpu) &&
5158                       !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5159                                          addr, n, v))
5160                     && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5161                         break;
5162                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5163                 handled += n;
5164                 addr += n;
5165                 len -= n;
5166                 v += n;
5167         } while (len);
5168
5169         return handled;
5170 }
5171
5172 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5173                         struct kvm_segment *var, int seg)
5174 {
5175         kvm_x86_ops->set_segment(vcpu, var, seg);
5176 }
5177
5178 void kvm_get_segment(struct kvm_vcpu *vcpu,
5179                      struct kvm_segment *var, int seg)
5180 {
5181         kvm_x86_ops->get_segment(vcpu, var, seg);
5182 }
5183
5184 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5185                            struct x86_exception *exception)
5186 {
5187         gpa_t t_gpa;
5188
5189         BUG_ON(!mmu_is_nested(vcpu));
5190
5191         /* NPT walks are always user-walks */
5192         access |= PFERR_USER_MASK;
5193         t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5194
5195         return t_gpa;
5196 }
5197
5198 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5199                               struct x86_exception *exception)
5200 {
5201         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5202         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5203 }
5204
5205  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5206                                 struct x86_exception *exception)
5207 {
5208         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5209         access |= PFERR_FETCH_MASK;
5210         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5211 }
5212
5213 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5214                                struct x86_exception *exception)
5215 {
5216         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5217         access |= PFERR_WRITE_MASK;
5218         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5219 }
5220
5221 /* uses this to access any guest's mapped memory without checking CPL */
5222 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5223                                 struct x86_exception *exception)
5224 {
5225         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5226 }
5227
5228 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5229                                       struct kvm_vcpu *vcpu, u32 access,
5230                                       struct x86_exception *exception)
5231 {
5232         void *data = val;
5233         int r = X86EMUL_CONTINUE;
5234
5235         while (bytes) {
5236                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5237                                                             exception);
5238                 unsigned offset = addr & (PAGE_SIZE-1);
5239                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5240                 int ret;
5241
5242                 if (gpa == UNMAPPED_GVA)
5243                         return X86EMUL_PROPAGATE_FAULT;
5244                 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5245                                                offset, toread);
5246                 if (ret < 0) {
5247                         r = X86EMUL_IO_NEEDED;
5248                         goto out;
5249                 }
5250
5251                 bytes -= toread;
5252                 data += toread;
5253                 addr += toread;
5254         }
5255 out:
5256         return r;
5257 }
5258
5259 /* used for instruction fetching */
5260 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5261                                 gva_t addr, void *val, unsigned int bytes,
5262                                 struct x86_exception *exception)
5263 {
5264         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5265         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5266         unsigned offset;
5267         int ret;
5268
5269         /* Inline kvm_read_guest_virt_helper for speed.  */
5270         gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5271                                                     exception);
5272         if (unlikely(gpa == UNMAPPED_GVA))
5273                 return X86EMUL_PROPAGATE_FAULT;
5274
5275         offset = addr & (PAGE_SIZE-1);
5276         if (WARN_ON(offset + bytes > PAGE_SIZE))
5277                 bytes = (unsigned)PAGE_SIZE - offset;
5278         ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5279                                        offset, bytes);
5280         if (unlikely(ret < 0))
5281                 return X86EMUL_IO_NEEDED;
5282
5283         return X86EMUL_CONTINUE;
5284 }
5285
5286 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5287                                gva_t addr, void *val, unsigned int bytes,
5288                                struct x86_exception *exception)
5289 {
5290         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5291
5292         /*
5293          * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5294          * is returned, but our callers are not ready for that and they blindly
5295          * call kvm_inject_page_fault.  Ensure that they at least do not leak
5296          * uninitialized kernel stack memory into cr2 and error code.
5297          */
5298         memset(exception, 0, sizeof(*exception));
5299         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5300                                           exception);
5301 }
5302 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5303
5304 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5305                              gva_t addr, void *val, unsigned int bytes,
5306                              struct x86_exception *exception, bool system)
5307 {
5308         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5309         u32 access = 0;
5310
5311         if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5312                 access |= PFERR_USER_MASK;
5313
5314         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5315 }
5316
5317 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5318                 unsigned long addr, void *val, unsigned int bytes)
5319 {
5320         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5321         int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5322
5323         return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5324 }
5325
5326 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5327                                       struct kvm_vcpu *vcpu, u32 access,
5328                                       struct x86_exception *exception)
5329 {
5330         void *data = val;
5331         int r = X86EMUL_CONTINUE;
5332
5333         while (bytes) {
5334                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5335                                                              access,
5336                                                              exception);
5337                 unsigned offset = addr & (PAGE_SIZE-1);
5338                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5339                 int ret;
5340
5341                 if (gpa == UNMAPPED_GVA)
5342                         return X86EMUL_PROPAGATE_FAULT;
5343                 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5344                 if (ret < 0) {
5345                         r = X86EMUL_IO_NEEDED;
5346                         goto out;
5347                 }
5348
5349                 bytes -= towrite;
5350                 data += towrite;
5351                 addr += towrite;
5352         }
5353 out:
5354         return r;
5355 }
5356
5357 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5358                               unsigned int bytes, struct x86_exception *exception,
5359                               bool system)
5360 {
5361         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5362         u32 access = PFERR_WRITE_MASK;
5363
5364         if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5365                 access |= PFERR_USER_MASK;
5366
5367         return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5368                                            access, exception);
5369 }
5370
5371 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5372                                 unsigned int bytes, struct x86_exception *exception)
5373 {
5374         /* kvm_write_guest_virt_system can pull in tons of pages. */
5375         vcpu->arch.l1tf_flush_l1d = true;
5376
5377         /*
5378          * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5379          * is returned, but our callers are not ready for that and they blindly
5380          * call kvm_inject_page_fault.  Ensure that they at least do not leak
5381          * uninitialized kernel stack memory into cr2 and error code.
5382          */
5383         memset(exception, 0, sizeof(*exception));
5384         return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5385                                            PFERR_WRITE_MASK, exception);
5386 }
5387 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5388
5389 int handle_ud(struct kvm_vcpu *vcpu)
5390 {
5391         int emul_type = EMULTYPE_TRAP_UD;
5392         enum emulation_result er;
5393         char sig[5]; /* ud2; .ascii "kvm" */
5394         struct x86_exception e;
5395
5396         if (force_emulation_prefix &&
5397             kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5398                                 sig, sizeof(sig), &e) == 0 &&
5399             memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5400                 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5401                 emul_type = 0;
5402         }
5403
5404         er = kvm_emulate_instruction(vcpu, emul_type);
5405         if (er == EMULATE_USER_EXIT)
5406                 return 0;
5407         if (er != EMULATE_DONE)
5408                 kvm_queue_exception(vcpu, UD_VECTOR);
5409         return 1;
5410 }
5411 EXPORT_SYMBOL_GPL(handle_ud);
5412
5413 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5414                             gpa_t gpa, bool write)
5415 {
5416         /* For APIC access vmexit */
5417         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5418                 return 1;
5419
5420         if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5421                 trace_vcpu_match_mmio(gva, gpa, write, true);
5422                 return 1;
5423         }
5424
5425         return 0;
5426 }
5427
5428 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5429                                 gpa_t *gpa, struct x86_exception *exception,
5430                                 bool write)
5431 {
5432         u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5433                 | (write ? PFERR_WRITE_MASK : 0);
5434
5435         /*
5436          * currently PKRU is only applied to ept enabled guest so
5437          * there is no pkey in EPT page table for L1 guest or EPT
5438          * shadow page table for L2 guest.
5439          */
5440         if (vcpu_match_mmio_gva(vcpu, gva)
5441             && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5442                                  vcpu->arch.mmio_access, 0, access)) {
5443                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5444                                         (gva & (PAGE_SIZE - 1));
5445                 trace_vcpu_match_mmio(gva, *gpa, write, false);
5446                 return 1;
5447         }
5448
5449         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5450
5451         if (*gpa == UNMAPPED_GVA)
5452                 return -1;
5453
5454         return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5455 }
5456
5457 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5458                         const void *val, int bytes)
5459 {
5460         int ret;
5461
5462         ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5463         if (ret < 0)
5464                 return 0;
5465         kvm_page_track_write(vcpu, gpa, val, bytes);
5466         return 1;
5467 }
5468
5469 struct read_write_emulator_ops {
5470         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5471                                   int bytes);
5472         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5473                                   void *val, int bytes);
5474         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5475                                int bytes, void *val);
5476         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5477                                     void *val, int bytes);
5478         bool write;
5479 };
5480
5481 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5482 {
5483         if (vcpu->mmio_read_completed) {
5484                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5485                                vcpu->mmio_fragments[0].gpa, val);
5486                 vcpu->mmio_read_completed = 0;
5487                 return 1;
5488         }
5489
5490         return 0;
5491 }
5492
5493 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5494                         void *val, int bytes)
5495 {
5496         return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5497 }
5498
5499 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5500                          void *val, int bytes)
5501 {
5502         return emulator_write_phys(vcpu, gpa, val, bytes);
5503 }
5504
5505 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5506 {
5507         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5508         return vcpu_mmio_write(vcpu, gpa, bytes, val);
5509 }
5510
5511 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5512                           void *val, int bytes)
5513 {
5514         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5515         return X86EMUL_IO_NEEDED;
5516 }
5517
5518 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5519                            void *val, int bytes)
5520 {
5521         struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5522
5523         memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5524         return X86EMUL_CONTINUE;
5525 }
5526
5527 static const struct read_write_emulator_ops read_emultor = {
5528         .read_write_prepare = read_prepare,
5529         .read_write_emulate = read_emulate,
5530         .read_write_mmio = vcpu_mmio_read,
5531         .read_write_exit_mmio = read_exit_mmio,
5532 };
5533
5534 static const struct read_write_emulator_ops write_emultor = {
5535         .read_write_emulate = write_emulate,
5536         .read_write_mmio = write_mmio,
5537         .read_write_exit_mmio = write_exit_mmio,
5538         .write = true,
5539 };
5540
5541 static int emulator_read_write_onepage(unsigned long addr, void *val,
5542                                        unsigned int bytes,
5543                                        struct x86_exception *exception,
5544                                        struct kvm_vcpu *vcpu,
5545                                        const struct read_write_emulator_ops *ops)
5546 {
5547         gpa_t gpa;
5548         int handled, ret;
5549         bool write = ops->write;
5550         struct kvm_mmio_fragment *frag;
5551         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5552
5553         /*
5554          * If the exit was due to a NPF we may already have a GPA.
5555          * If the GPA is present, use it to avoid the GVA to GPA table walk.
5556          * Note, this cannot be used on string operations since string
5557          * operation using rep will only have the initial GPA from the NPF
5558          * occurred.
5559          */
5560         if (vcpu->arch.gpa_available &&
5561             emulator_can_use_gpa(ctxt) &&
5562             (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5563                 gpa = vcpu->arch.gpa_val;
5564                 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5565         } else {
5566                 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5567                 if (ret < 0)
5568                         return X86EMUL_PROPAGATE_FAULT;
5569         }
5570
5571         if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5572                 return X86EMUL_CONTINUE;
5573
5574         /*
5575          * Is this MMIO handled locally?
5576          */
5577         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5578         if (handled == bytes)
5579                 return X86EMUL_CONTINUE;
5580
5581         gpa += handled;
5582         bytes -= handled;
5583         val += handled;
5584
5585         WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5586         frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5587         frag->gpa = gpa;
5588         frag->data = val;
5589         frag->len = bytes;
5590         return X86EMUL_CONTINUE;
5591 }
5592
5593 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5594                         unsigned long addr,
5595                         void *val, unsigned int bytes,
5596                         struct x86_exception *exception,
5597                         const struct read_write_emulator_ops *ops)
5598 {
5599         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5600         gpa_t gpa;
5601         int rc;
5602
5603         if (ops->read_write_prepare &&
5604                   ops->read_write_prepare(vcpu, val, bytes))
5605                 return X86EMUL_CONTINUE;
5606
5607         vcpu->mmio_nr_fragments = 0;
5608
5609         /* Crossing a page boundary? */
5610         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5611                 int now;
5612
5613                 now = -addr & ~PAGE_MASK;
5614                 rc = emulator_read_write_onepage(addr, val, now, exception,
5615                                                  vcpu, ops);
5616
5617                 if (rc != X86EMUL_CONTINUE)
5618                         return rc;
5619                 addr += now;
5620                 if (ctxt->mode != X86EMUL_MODE_PROT64)
5621                         addr = (u32)addr;
5622                 val += now;
5623                 bytes -= now;
5624         }
5625
5626         rc = emulator_read_write_onepage(addr, val, bytes, exception,
5627                                          vcpu, ops);
5628         if (rc != X86EMUL_CONTINUE)
5629                 return rc;
5630
5631         if (!vcpu->mmio_nr_fragments)
5632                 return rc;
5633
5634         gpa = vcpu->mmio_fragments[0].gpa;
5635
5636         vcpu->mmio_needed = 1;
5637         vcpu->mmio_cur_fragment = 0;
5638
5639         vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5640         vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5641         vcpu->run->exit_reason = KVM_EXIT_MMIO;
5642         vcpu->run->mmio.phys_addr = gpa;
5643
5644         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5645 }
5646
5647 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5648                                   unsigned long addr,
5649                                   void *val,
5650                                   unsigned int bytes,
5651                                   struct x86_exception *exception)
5652 {
5653         return emulator_read_write(ctxt, addr, val, bytes,
5654                                    exception, &read_emultor);
5655 }
5656
5657 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5658                             unsigned long addr,
5659                             const void *val,
5660                             unsigned int bytes,
5661                             struct x86_exception *exception)
5662 {
5663         return emulator_read_write(ctxt, addr, (void *)val, bytes,
5664                                    exception, &write_emultor);
5665 }
5666
5667 #define CMPXCHG_TYPE(t, ptr, old, new) \
5668         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5669
5670 #ifdef CONFIG_X86_64
5671 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5672 #else
5673 #  define CMPXCHG64(ptr, old, new) \
5674         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5675 #endif
5676
5677 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5678                                      unsigned long addr,
5679                                      const void *old,
5680                                      const void *new,
5681                                      unsigned int bytes,
5682                                      struct x86_exception *exception)
5683 {
5684         struct kvm_host_map map;
5685         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5686         gpa_t gpa;
5687         char *kaddr;
5688         bool exchanged;
5689
5690         /* guests cmpxchg8b have to be emulated atomically */
5691         if (bytes > 8 || (bytes & (bytes - 1)))
5692                 goto emul_write;
5693
5694         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5695
5696         if (gpa == UNMAPPED_GVA ||
5697             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5698                 goto emul_write;
5699
5700         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5701                 goto emul_write;
5702
5703         if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5704                 goto emul_write;
5705
5706         kaddr = map.hva + offset_in_page(gpa);
5707
5708         switch (bytes) {
5709         case 1:
5710                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5711                 break;
5712         case 2:
5713                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5714                 break;
5715         case 4:
5716                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5717                 break;
5718         case 8:
5719                 exchanged = CMPXCHG64(kaddr, old, new);
5720                 break;
5721         default:
5722                 BUG();
5723         }
5724
5725         kvm_vcpu_unmap(vcpu, &map, true);
5726
5727         if (!exchanged)
5728                 return X86EMUL_CMPXCHG_FAILED;
5729
5730         kvm_page_track_write(vcpu, gpa, new, bytes);
5731
5732         return X86EMUL_CONTINUE;
5733
5734 emul_write:
5735         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5736
5737         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5738 }
5739
5740 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5741 {
5742         int r = 0, i;
5743
5744         for (i = 0; i < vcpu->arch.pio.count; i++) {
5745                 if (vcpu->arch.pio.in)
5746                         r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5747                                             vcpu->arch.pio.size, pd);
5748                 else
5749                         r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5750                                              vcpu->arch.pio.port, vcpu->arch.pio.size,
5751                                              pd);
5752                 if (r)
5753                         break;
5754                 pd += vcpu->arch.pio.size;
5755         }
5756         return r;
5757 }
5758
5759 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5760                                unsigned short port, void *val,
5761                                unsigned int count, bool in)
5762 {
5763         vcpu->arch.pio.port = port;
5764         vcpu->arch.pio.in = in;
5765         vcpu->arch.pio.count  = count;
5766         vcpu->arch.pio.size = size;
5767
5768         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5769                 vcpu->arch.pio.count = 0;
5770                 return 1;
5771         }
5772
5773         vcpu->run->exit_reason = KVM_EXIT_IO;
5774         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5775         vcpu->run->io.size = size;
5776         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5777         vcpu->run->io.count = count;
5778         vcpu->run->io.port = port;
5779
5780         return 0;
5781 }
5782
5783 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5784                                     int size, unsigned short port, void *val,
5785                                     unsigned int count)
5786 {
5787         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5788         int ret;
5789
5790         if (vcpu->arch.pio.count)
5791                 goto data_avail;
5792
5793         memset(vcpu->arch.pio_data, 0, size * count);
5794
5795         ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5796         if (ret) {
5797 data_avail:
5798                 memcpy(val, vcpu->arch.pio_data, size * count);
5799                 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5800                 vcpu->arch.pio.count = 0;
5801                 return 1;
5802         }
5803
5804         return 0;
5805 }
5806
5807 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5808                                      int size, unsigned short port,
5809                                      const void *val, unsigned int count)
5810 {
5811         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5812
5813         memcpy(vcpu->arch.pio_data, val, size * count);
5814         trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5815         return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5816 }
5817
5818 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5819 {
5820         return kvm_x86_ops->get_segment_base(vcpu, seg);
5821 }
5822
5823 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5824 {
5825         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5826 }
5827
5828 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5829 {
5830         if (!need_emulate_wbinvd(vcpu))
5831                 return X86EMUL_CONTINUE;
5832
5833         if (kvm_x86_ops->has_wbinvd_exit()) {
5834                 int cpu = get_cpu();
5835
5836                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5837                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5838                                 wbinvd_ipi, NULL, 1);
5839                 put_cpu();
5840                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5841         } else
5842                 wbinvd();
5843         return X86EMUL_CONTINUE;
5844 }
5845
5846 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5847 {
5848         kvm_emulate_wbinvd_noskip(vcpu);
5849         return kvm_skip_emulated_instruction(vcpu);
5850 }
5851 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5852
5853
5854
5855 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5856 {
5857         kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5858 }
5859
5860 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5861                            unsigned long *dest)
5862 {
5863         return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5864 }
5865
5866 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5867                            unsigned long value)
5868 {
5869
5870         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5871 }
5872
5873 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5874 {
5875         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5876 }
5877
5878 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5879 {
5880         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5881         unsigned long value;
5882
5883         switch (cr) {
5884         case 0:
5885                 value = kvm_read_cr0(vcpu);
5886                 break;
5887         case 2:
5888                 value = vcpu->arch.cr2;
5889                 break;
5890         case 3:
5891                 value = kvm_read_cr3(vcpu);
5892                 break;
5893         case 4:
5894                 value = kvm_read_cr4(vcpu);
5895                 break;
5896         case 8:
5897                 value = kvm_get_cr8(vcpu);
5898                 break;
5899         default:
5900                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5901                 return 0;
5902         }
5903
5904         return value;
5905 }
5906
5907 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5908 {
5909         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5910         int res = 0;
5911
5912         switch (cr) {
5913         case 0:
5914                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5915                 break;
5916         case 2:
5917                 vcpu->arch.cr2 = val;
5918                 break;
5919         case 3:
5920                 res = kvm_set_cr3(vcpu, val);
5921                 break;
5922         case 4:
5923                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5924                 break;
5925         case 8:
5926                 res = kvm_set_cr8(vcpu, val);
5927                 break;
5928         default:
5929                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5930                 res = -1;
5931         }
5932
5933         return res;
5934 }
5935
5936 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5937 {
5938         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5939 }
5940
5941 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5942 {
5943         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5944 }
5945
5946 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5947 {
5948         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5949 }
5950
5951 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5952 {
5953         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5954 }
5955
5956 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5957 {
5958         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5959 }
5960
5961 static unsigned long emulator_get_cached_segment_base(
5962         struct x86_emulate_ctxt *ctxt, int seg)
5963 {
5964         return get_segment_base(emul_to_vcpu(ctxt), seg);
5965 }
5966
5967 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5968                                  struct desc_struct *desc, u32 *base3,
5969                                  int seg)
5970 {
5971         struct kvm_segment var;
5972
5973         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5974         *selector = var.selector;
5975
5976         if (var.unusable) {
5977                 memset(desc, 0, sizeof(*desc));
5978                 if (base3)
5979                         *base3 = 0;
5980                 return false;
5981         }
5982
5983         if (var.g)
5984                 var.limit >>= 12;
5985         set_desc_limit(desc, var.limit);
5986         set_desc_base(desc, (unsigned long)var.base);
5987 #ifdef CONFIG_X86_64
5988         if (base3)
5989                 *base3 = var.base >> 32;
5990 #endif
5991         desc->type = var.type;
5992         desc->s = var.s;
5993         desc->dpl = var.dpl;
5994         desc->p = var.present;
5995         desc->avl = var.avl;
5996         desc->l = var.l;
5997         desc->d = var.db;
5998         desc->g = var.g;
5999
6000         return true;
6001 }
6002
6003 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6004                                  struct desc_struct *desc, u32 base3,
6005                                  int seg)
6006 {
6007         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6008         struct kvm_segment var;
6009
6010         var.selector = selector;
6011         var.base = get_desc_base(desc);
6012 #ifdef CONFIG_X86_64
6013         var.base |= ((u64)base3) << 32;
6014 #endif
6015         var.limit = get_desc_limit(desc);
6016         if (desc->g)
6017                 var.limit = (var.limit << 12) | 0xfff;
6018         var.type = desc->type;
6019         var.dpl = desc->dpl;
6020         var.db = desc->d;
6021         var.s = desc->s;
6022         var.l = desc->l;
6023         var.g = desc->g;
6024         var.avl = desc->avl;
6025         var.present = desc->p;
6026         var.unusable = !var.present;
6027         var.padding = 0;
6028
6029         kvm_set_segment(vcpu, &var, seg);
6030         return;
6031 }
6032
6033 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6034                             u32 msr_index, u64 *pdata)
6035 {
6036         return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6037 }
6038
6039 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6040                             u32 msr_index, u64 data)
6041 {
6042         return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6043 }
6044
6045 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6046 {
6047         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6048
6049         return vcpu->arch.smbase;
6050 }
6051
6052 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6053 {
6054         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6055
6056         vcpu->arch.smbase = smbase;
6057 }
6058
6059 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6060                               u32 pmc)
6061 {
6062         return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6063 }
6064
6065 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6066                              u32 pmc, u64 *pdata)
6067 {
6068         return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6069 }
6070
6071 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6072 {
6073         emul_to_vcpu(ctxt)->arch.halt_request = 1;
6074 }
6075
6076 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6077                               struct x86_instruction_info *info,
6078                               enum x86_intercept_stage stage)
6079 {
6080         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6081 }
6082
6083 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6084                         u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6085 {
6086         return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6087 }
6088
6089 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6090 {
6091         return kvm_register_read(emul_to_vcpu(ctxt), reg);
6092 }
6093
6094 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6095 {
6096         kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6097 }
6098
6099 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6100 {
6101         kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6102 }
6103
6104 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6105 {
6106         return emul_to_vcpu(ctxt)->arch.hflags;
6107 }
6108
6109 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6110 {
6111         emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6112 }
6113
6114 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6115                                   const char *smstate)
6116 {
6117         return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6118 }
6119
6120 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6121 {
6122         kvm_smm_changed(emul_to_vcpu(ctxt));
6123 }
6124
6125 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6126 {
6127         return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6128 }
6129
6130 static const struct x86_emulate_ops emulate_ops = {
6131         .read_gpr            = emulator_read_gpr,
6132         .write_gpr           = emulator_write_gpr,
6133         .read_std            = emulator_read_std,
6134         .write_std           = emulator_write_std,
6135         .read_phys           = kvm_read_guest_phys_system,
6136         .fetch               = kvm_fetch_guest_virt,
6137         .read_emulated       = emulator_read_emulated,
6138         .write_emulated      = emulator_write_emulated,
6139         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
6140         .invlpg              = emulator_invlpg,
6141         .pio_in_emulated     = emulator_pio_in_emulated,
6142         .pio_out_emulated    = emulator_pio_out_emulated,
6143         .get_segment         = emulator_get_segment,
6144         .set_segment         = emulator_set_segment,
6145         .get_cached_segment_base = emulator_get_cached_segment_base,
6146         .get_gdt             = emulator_get_gdt,
6147         .get_idt             = emulator_get_idt,
6148         .set_gdt             = emulator_set_gdt,
6149         .set_idt             = emulator_set_idt,
6150         .get_cr              = emulator_get_cr,
6151         .set_cr              = emulator_set_cr,
6152         .cpl                 = emulator_get_cpl,
6153         .get_dr              = emulator_get_dr,
6154         .set_dr              = emulator_set_dr,
6155         .get_smbase          = emulator_get_smbase,
6156         .set_smbase          = emulator_set_smbase,
6157         .set_msr             = emulator_set_msr,
6158         .get_msr             = emulator_get_msr,
6159         .check_pmc           = emulator_check_pmc,
6160         .read_pmc            = emulator_read_pmc,
6161         .halt                = emulator_halt,
6162         .wbinvd              = emulator_wbinvd,
6163         .fix_hypercall       = emulator_fix_hypercall,
6164         .intercept           = emulator_intercept,
6165         .get_cpuid           = emulator_get_cpuid,
6166         .set_nmi_mask        = emulator_set_nmi_mask,
6167         .get_hflags          = emulator_get_hflags,
6168         .set_hflags          = emulator_set_hflags,
6169         .pre_leave_smm       = emulator_pre_leave_smm,
6170         .post_leave_smm      = emulator_post_leave_smm,
6171         .set_xcr             = emulator_set_xcr,
6172 };
6173
6174 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6175 {
6176         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6177         /*
6178          * an sti; sti; sequence only disable interrupts for the first
6179          * instruction. So, if the last instruction, be it emulated or
6180          * not, left the system with the INT_STI flag enabled, it
6181          * means that the last instruction is an sti. We should not
6182          * leave the flag on in this case. The same goes for mov ss
6183          */
6184         if (int_shadow & mask)
6185                 mask = 0;
6186         if (unlikely(int_shadow || mask)) {
6187                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6188                 if (!mask)
6189                         kvm_make_request(KVM_REQ_EVENT, vcpu);
6190         }
6191 }
6192
6193 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6194 {
6195         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6196         if (ctxt->exception.vector == PF_VECTOR)
6197                 return kvm_propagate_fault(vcpu, &ctxt->exception);
6198
6199         if (ctxt->exception.error_code_valid)
6200                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6201                                       ctxt->exception.error_code);
6202         else
6203                 kvm_queue_exception(vcpu, ctxt->exception.vector);
6204         return false;
6205 }
6206
6207 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6208 {
6209         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6210         int cs_db, cs_l;
6211
6212         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6213
6214         ctxt->eflags = kvm_get_rflags(vcpu);
6215         ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6216
6217         ctxt->eip = kvm_rip_read(vcpu);
6218         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
6219                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
6220                      (cs_l && is_long_mode(vcpu))       ? X86EMUL_MODE_PROT64 :
6221                      cs_db                              ? X86EMUL_MODE_PROT32 :
6222                                                           X86EMUL_MODE_PROT16;
6223         BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6224         BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6225         BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6226
6227         init_decode_cache(ctxt);
6228         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6229 }
6230
6231 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6232 {
6233         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6234         int ret;
6235
6236         init_emulate_ctxt(vcpu);
6237
6238         ctxt->op_bytes = 2;
6239         ctxt->ad_bytes = 2;
6240         ctxt->_eip = ctxt->eip + inc_eip;
6241         ret = emulate_int_real(ctxt, irq);
6242
6243         if (ret != X86EMUL_CONTINUE)
6244                 return EMULATE_FAIL;
6245
6246         ctxt->eip = ctxt->_eip;
6247         kvm_rip_write(vcpu, ctxt->eip);
6248         kvm_set_rflags(vcpu, ctxt->eflags);
6249
6250         return EMULATE_DONE;
6251 }
6252 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6253
6254 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6255 {
6256         int r = EMULATE_DONE;
6257
6258         ++vcpu->stat.insn_emulation_fail;
6259         trace_kvm_emulate_insn_failed(vcpu);
6260
6261         if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6262                 return EMULATE_FAIL;
6263
6264         if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6265                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6266                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6267                 vcpu->run->internal.ndata = 0;
6268                 r = EMULATE_USER_EXIT;
6269         }
6270
6271         kvm_queue_exception(vcpu, UD_VECTOR);
6272
6273         return r;
6274 }
6275
6276 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6277                                   bool write_fault_to_shadow_pgtable,
6278                                   int emulation_type)
6279 {
6280         gpa_t gpa = cr2;
6281         kvm_pfn_t pfn;
6282
6283         if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6284                 return false;
6285
6286         if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6287                 return false;
6288
6289         if (!vcpu->arch.mmu->direct_map) {
6290                 /*
6291                  * Write permission should be allowed since only
6292                  * write access need to be emulated.
6293                  */
6294                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6295
6296                 /*
6297                  * If the mapping is invalid in guest, let cpu retry
6298                  * it to generate fault.
6299                  */
6300                 if (gpa == UNMAPPED_GVA)
6301                         return true;
6302         }
6303
6304         /*
6305          * Do not retry the unhandleable instruction if it faults on the
6306          * readonly host memory, otherwise it will goto a infinite loop:
6307          * retry instruction -> write #PF -> emulation fail -> retry
6308          * instruction -> ...
6309          */
6310         pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6311
6312         /*
6313          * If the instruction failed on the error pfn, it can not be fixed,
6314          * report the error to userspace.
6315          */
6316         if (is_error_noslot_pfn(pfn))
6317                 return false;
6318
6319         kvm_release_pfn_clean(pfn);
6320
6321         /* The instructions are well-emulated on direct mmu. */
6322         if (vcpu->arch.mmu->direct_map) {
6323                 unsigned int indirect_shadow_pages;
6324
6325                 spin_lock(&vcpu->kvm->mmu_lock);
6326                 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6327                 spin_unlock(&vcpu->kvm->mmu_lock);
6328
6329                 if (indirect_shadow_pages)
6330                         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6331
6332                 return true;
6333         }
6334
6335         /*
6336          * if emulation was due to access to shadowed page table
6337          * and it failed try to unshadow page and re-enter the
6338          * guest to let CPU execute the instruction.
6339          */
6340         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6341
6342         /*
6343          * If the access faults on its page table, it can not
6344          * be fixed by unprotecting shadow page and it should
6345          * be reported to userspace.
6346          */
6347         return !write_fault_to_shadow_pgtable;
6348 }
6349
6350 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6351                               unsigned long cr2,  int emulation_type)
6352 {
6353         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6354         unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6355
6356         last_retry_eip = vcpu->arch.last_retry_eip;
6357         last_retry_addr = vcpu->arch.last_retry_addr;
6358
6359         /*
6360          * If the emulation is caused by #PF and it is non-page_table
6361          * writing instruction, it means the VM-EXIT is caused by shadow
6362          * page protected, we can zap the shadow page and retry this
6363          * instruction directly.
6364          *
6365          * Note: if the guest uses a non-page-table modifying instruction
6366          * on the PDE that points to the instruction, then we will unmap
6367          * the instruction and go to an infinite loop. So, we cache the
6368          * last retried eip and the last fault address, if we meet the eip
6369          * and the address again, we can break out of the potential infinite
6370          * loop.
6371          */
6372         vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6373
6374         if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6375                 return false;
6376
6377         if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6378                 return false;
6379
6380         if (x86_page_table_writing_insn(ctxt))
6381                 return false;
6382
6383         if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6384                 return false;
6385
6386         vcpu->arch.last_retry_eip = ctxt->eip;
6387         vcpu->arch.last_retry_addr = cr2;
6388
6389         if (!vcpu->arch.mmu->direct_map)
6390                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6391
6392         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6393
6394         return true;
6395 }
6396
6397 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6398 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6399
6400 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6401 {
6402         if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6403                 /* This is a good place to trace that we are exiting SMM.  */
6404                 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6405
6406                 /* Process a latched INIT or SMI, if any.  */
6407                 kvm_make_request(KVM_REQ_EVENT, vcpu);
6408         }
6409
6410         kvm_mmu_reset_context(vcpu);
6411 }
6412
6413 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6414                                 unsigned long *db)
6415 {
6416         u32 dr6 = 0;
6417         int i;
6418         u32 enable, rwlen;
6419
6420         enable = dr7;
6421         rwlen = dr7 >> 16;
6422         for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6423                 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6424                         dr6 |= (1 << i);
6425         return dr6;
6426 }
6427
6428 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6429 {
6430         struct kvm_run *kvm_run = vcpu->run;
6431
6432         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6433                 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6434                 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6435                 kvm_run->debug.arch.exception = DB_VECTOR;
6436                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6437                 *r = EMULATE_USER_EXIT;
6438         } else {
6439                 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6440         }
6441 }
6442
6443 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6444 {
6445         unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6446         int r;
6447
6448         r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6449         if (unlikely(r != EMULATE_DONE))
6450                 return 0;
6451
6452         /*
6453          * rflags is the old, "raw" value of the flags.  The new value has
6454          * not been saved yet.
6455          *
6456          * This is correct even for TF set by the guest, because "the
6457          * processor will not generate this exception after the instruction
6458          * that sets the TF flag".
6459          */
6460         if (unlikely(rflags & X86_EFLAGS_TF))
6461                 kvm_vcpu_do_singlestep(vcpu, &r);
6462         return r == EMULATE_DONE;
6463 }
6464 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6465
6466 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6467 {
6468         if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6469             (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6470                 struct kvm_run *kvm_run = vcpu->run;
6471                 unsigned long eip = kvm_get_linear_rip(vcpu);
6472                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6473                                            vcpu->arch.guest_debug_dr7,
6474                                            vcpu->arch.eff_db);
6475
6476                 if (dr6 != 0) {
6477                         kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6478                         kvm_run->debug.arch.pc = eip;
6479                         kvm_run->debug.arch.exception = DB_VECTOR;
6480                         kvm_run->exit_reason = KVM_EXIT_DEBUG;
6481                         *r = EMULATE_USER_EXIT;
6482                         return true;
6483                 }
6484         }
6485
6486         if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6487             !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6488                 unsigned long eip = kvm_get_linear_rip(vcpu);
6489                 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6490                                            vcpu->arch.dr7,
6491                                            vcpu->arch.db);
6492
6493                 if (dr6 != 0) {
6494                         vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6495                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
6496                         kvm_queue_exception(vcpu, DB_VECTOR);
6497                         *r = EMULATE_DONE;
6498                         return true;
6499                 }
6500         }
6501
6502         return false;
6503 }
6504
6505 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6506 {
6507         switch (ctxt->opcode_len) {
6508         case 1:
6509                 switch (ctxt->b) {
6510                 case 0xe4:      /* IN */
6511                 case 0xe5:
6512                 case 0xec:
6513                 case 0xed:
6514                 case 0xe6:      /* OUT */
6515                 case 0xe7:
6516                 case 0xee:
6517                 case 0xef:
6518                 case 0x6c:      /* INS */
6519                 case 0x6d:
6520                 case 0x6e:      /* OUTS */
6521                 case 0x6f:
6522                         return true;
6523                 }
6524                 break;
6525         case 2:
6526                 switch (ctxt->b) {
6527                 case 0x33:      /* RDPMC */
6528                         return true;
6529                 }
6530                 break;
6531         }
6532
6533         return false;
6534 }
6535
6536 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6537                             unsigned long cr2,
6538                             int emulation_type,
6539                             void *insn,
6540                             int insn_len)
6541 {
6542         int r;
6543         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6544         bool writeback = true;
6545         bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6546
6547         vcpu->arch.l1tf_flush_l1d = true;
6548
6549         /*
6550          * Clear write_fault_to_shadow_pgtable here to ensure it is
6551          * never reused.
6552          */
6553         vcpu->arch.write_fault_to_shadow_pgtable = false;
6554         kvm_clear_exception_queue(vcpu);
6555
6556         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6557                 init_emulate_ctxt(vcpu);
6558
6559                 /*
6560                  * We will reenter on the same instruction since
6561                  * we do not set complete_userspace_io.  This does not
6562                  * handle watchpoints yet, those would be handled in
6563                  * the emulate_ops.
6564                  */
6565                 if (!(emulation_type & EMULTYPE_SKIP) &&
6566                     kvm_vcpu_check_breakpoint(vcpu, &r))
6567                         return r;
6568
6569                 ctxt->interruptibility = 0;
6570                 ctxt->have_exception = false;
6571                 ctxt->exception.vector = -1;
6572                 ctxt->perm_ok = false;
6573
6574                 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6575
6576                 r = x86_decode_insn(ctxt, insn, insn_len);
6577
6578                 trace_kvm_emulate_insn_start(vcpu);
6579                 ++vcpu->stat.insn_emulation;
6580                 if (r != EMULATION_OK)  {
6581                         if (emulation_type & EMULTYPE_TRAP_UD)
6582                                 return EMULATE_FAIL;
6583                         if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6584                                                 emulation_type))
6585                                 return EMULATE_DONE;
6586                         if (ctxt->have_exception) {
6587                                 /*
6588                                  * #UD should result in just EMULATION_FAILED, and trap-like
6589                                  * exception should not be encountered during decode.
6590                                  */
6591                                 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6592                                              exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6593                                 inject_emulated_exception(vcpu);
6594                                 return EMULATE_DONE;
6595                         }
6596                         if (emulation_type & EMULTYPE_SKIP)
6597                                 return EMULATE_FAIL;
6598                         return handle_emulation_failure(vcpu, emulation_type);
6599                 }
6600         }
6601
6602         if ((emulation_type & EMULTYPE_VMWARE) &&
6603             !is_vmware_backdoor_opcode(ctxt))
6604                 return EMULATE_FAIL;
6605
6606         if (emulation_type & EMULTYPE_SKIP) {
6607                 kvm_rip_write(vcpu, ctxt->_eip);
6608                 if (ctxt->eflags & X86_EFLAGS_RF)
6609                         kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6610                 kvm_x86_ops->set_interrupt_shadow(vcpu, 0);
6611                 return EMULATE_DONE;
6612         }
6613
6614         if (retry_instruction(ctxt, cr2, emulation_type))
6615                 return EMULATE_DONE;
6616
6617         /* this is needed for vmware backdoor interface to work since it
6618            changes registers values  during IO operation */
6619         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6620                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6621                 emulator_invalidate_register_cache(ctxt);
6622         }
6623
6624 restart:
6625         /* Save the faulting GPA (cr2) in the address field */
6626         ctxt->exception.address = cr2;
6627
6628         r = x86_emulate_insn(ctxt);
6629
6630         if (r == EMULATION_INTERCEPTED)
6631                 return EMULATE_DONE;
6632
6633         if (r == EMULATION_FAILED) {
6634                 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6635                                         emulation_type))
6636                         return EMULATE_DONE;
6637
6638                 return handle_emulation_failure(vcpu, emulation_type);
6639         }
6640
6641         if (ctxt->have_exception) {
6642                 r = EMULATE_DONE;
6643                 if (inject_emulated_exception(vcpu))
6644                         return r;
6645         } else if (vcpu->arch.pio.count) {
6646                 if (!vcpu->arch.pio.in) {
6647                         /* FIXME: return into emulator if single-stepping.  */
6648                         vcpu->arch.pio.count = 0;
6649                 } else {
6650                         writeback = false;
6651                         vcpu->arch.complete_userspace_io = complete_emulated_pio;
6652                 }
6653                 r = EMULATE_USER_EXIT;
6654         } else if (vcpu->mmio_needed) {
6655                 if (!vcpu->mmio_is_write)
6656                         writeback = false;
6657                 r = EMULATE_USER_EXIT;
6658                 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6659         } else if (r == EMULATION_RESTART)
6660                 goto restart;
6661         else
6662                 r = EMULATE_DONE;
6663
6664         if (writeback) {
6665                 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6666                 toggle_interruptibility(vcpu, ctxt->interruptibility);
6667                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6668                 if (!ctxt->have_exception ||
6669                     exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6670                         kvm_rip_write(vcpu, ctxt->eip);
6671                         if (r == EMULATE_DONE && ctxt->tf)
6672                                 kvm_vcpu_do_singlestep(vcpu, &r);
6673                         __kvm_set_rflags(vcpu, ctxt->eflags);
6674                 }
6675
6676                 /*
6677                  * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6678                  * do nothing, and it will be requested again as soon as
6679                  * the shadow expires.  But we still need to check here,
6680                  * because POPF has no interrupt shadow.
6681                  */
6682                 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6683                         kvm_make_request(KVM_REQ_EVENT, vcpu);
6684         } else
6685                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6686
6687         return r;
6688 }
6689
6690 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6691 {
6692         return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6693 }
6694 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6695
6696 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6697                                         void *insn, int insn_len)
6698 {
6699         return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6700 }
6701 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6702
6703 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6704 {
6705         vcpu->arch.pio.count = 0;
6706         return 1;
6707 }
6708
6709 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6710 {
6711         vcpu->arch.pio.count = 0;
6712
6713         if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6714                 return 1;
6715
6716         return kvm_skip_emulated_instruction(vcpu);
6717 }
6718
6719 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6720                             unsigned short port)
6721 {
6722         unsigned long val = kvm_rax_read(vcpu);
6723         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6724                                             size, port, &val, 1);
6725         if (ret)
6726                 return ret;
6727
6728         /*
6729          * Workaround userspace that relies on old KVM behavior of %rip being
6730          * incremented prior to exiting to userspace to handle "OUT 0x7e".
6731          */
6732         if (port == 0x7e &&
6733             kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6734                 vcpu->arch.complete_userspace_io =
6735                         complete_fast_pio_out_port_0x7e;
6736                 kvm_skip_emulated_instruction(vcpu);
6737         } else {
6738                 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6739                 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6740         }
6741         return 0;
6742 }
6743
6744 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6745 {
6746         unsigned long val;
6747
6748         /* We should only ever be called with arch.pio.count equal to 1 */
6749         BUG_ON(vcpu->arch.pio.count != 1);
6750
6751         if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6752                 vcpu->arch.pio.count = 0;
6753                 return 1;
6754         }
6755
6756         /* For size less than 4 we merge, else we zero extend */
6757         val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6758
6759         /*
6760          * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6761          * the copy and tracing
6762          */
6763         emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6764                                  vcpu->arch.pio.port, &val, 1);
6765         kvm_rax_write(vcpu, val);
6766
6767         return kvm_skip_emulated_instruction(vcpu);
6768 }
6769
6770 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6771                            unsigned short port)
6772 {
6773         unsigned long val;
6774         int ret;
6775
6776         /* For size less than 4 we merge, else we zero extend */
6777         val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6778
6779         ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6780                                        &val, 1);
6781         if (ret) {
6782                 kvm_rax_write(vcpu, val);
6783                 return ret;
6784         }
6785
6786         vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6787         vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6788
6789         return 0;
6790 }
6791
6792 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6793 {
6794         int ret;
6795
6796         if (in)
6797                 ret = kvm_fast_pio_in(vcpu, size, port);
6798         else
6799                 ret = kvm_fast_pio_out(vcpu, size, port);
6800         return ret && kvm_skip_emulated_instruction(vcpu);
6801 }
6802 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6803
6804 static int kvmclock_cpu_down_prep(unsigned int cpu)
6805 {
6806         __this_cpu_write(cpu_tsc_khz, 0);
6807         return 0;
6808 }
6809
6810 static void tsc_khz_changed(void *data)
6811 {
6812         struct cpufreq_freqs *freq = data;
6813         unsigned long khz = 0;
6814
6815         if (data)
6816                 khz = freq->new;
6817         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6818                 khz = cpufreq_quick_get(raw_smp_processor_id());
6819         if (!khz)
6820                 khz = tsc_khz;
6821         __this_cpu_write(cpu_tsc_khz, khz);
6822 }
6823
6824 #ifdef CONFIG_X86_64
6825 static void kvm_hyperv_tsc_notifier(void)
6826 {
6827         struct kvm *kvm;
6828         struct kvm_vcpu *vcpu;
6829         int cpu;
6830
6831         mutex_lock(&kvm_lock);
6832         list_for_each_entry(kvm, &vm_list, vm_list)
6833                 kvm_make_mclock_inprogress_request(kvm);
6834
6835         hyperv_stop_tsc_emulation();
6836
6837         /* TSC frequency always matches when on Hyper-V */
6838         for_each_present_cpu(cpu)
6839                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6840         kvm_max_guest_tsc_khz = tsc_khz;
6841
6842         list_for_each_entry(kvm, &vm_list, vm_list) {
6843                 struct kvm_arch *ka = &kvm->arch;
6844
6845                 spin_lock(&ka->pvclock_gtod_sync_lock);
6846
6847                 pvclock_update_vm_gtod_copy(kvm);
6848
6849                 kvm_for_each_vcpu(cpu, vcpu, kvm)
6850                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6851
6852                 kvm_for_each_vcpu(cpu, vcpu, kvm)
6853                         kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6854
6855                 spin_unlock(&ka->pvclock_gtod_sync_lock);
6856         }
6857         mutex_unlock(&kvm_lock);
6858 }
6859 #endif
6860
6861 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6862 {
6863         struct kvm *kvm;
6864         struct kvm_vcpu *vcpu;
6865         int i, send_ipi = 0;
6866
6867         /*
6868          * We allow guests to temporarily run on slowing clocks,
6869          * provided we notify them after, or to run on accelerating
6870          * clocks, provided we notify them before.  Thus time never
6871          * goes backwards.
6872          *
6873          * However, we have a problem.  We can't atomically update
6874          * the frequency of a given CPU from this function; it is
6875          * merely a notifier, which can be called from any CPU.
6876          * Changing the TSC frequency at arbitrary points in time
6877          * requires a recomputation of local variables related to
6878          * the TSC for each VCPU.  We must flag these local variables
6879          * to be updated and be sure the update takes place with the
6880          * new frequency before any guests proceed.
6881          *
6882          * Unfortunately, the combination of hotplug CPU and frequency
6883          * change creates an intractable locking scenario; the order
6884          * of when these callouts happen is undefined with respect to
6885          * CPU hotplug, and they can race with each other.  As such,
6886          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6887          * undefined; you can actually have a CPU frequency change take
6888          * place in between the computation of X and the setting of the
6889          * variable.  To protect against this problem, all updates of
6890          * the per_cpu tsc_khz variable are done in an interrupt
6891          * protected IPI, and all callers wishing to update the value
6892          * must wait for a synchronous IPI to complete (which is trivial
6893          * if the caller is on the CPU already).  This establishes the
6894          * necessary total order on variable updates.
6895          *
6896          * Note that because a guest time update may take place
6897          * anytime after the setting of the VCPU's request bit, the
6898          * correct TSC value must be set before the request.  However,
6899          * to ensure the update actually makes it to any guest which
6900          * starts running in hardware virtualization between the set
6901          * and the acquisition of the spinlock, we must also ping the
6902          * CPU after setting the request bit.
6903          *
6904          */
6905
6906         smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6907
6908         mutex_lock(&kvm_lock);
6909         list_for_each_entry(kvm, &vm_list, vm_list) {
6910                 kvm_for_each_vcpu(i, vcpu, kvm) {
6911                         if (vcpu->cpu != cpu)
6912                                 continue;
6913                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6914                         if (vcpu->cpu != raw_smp_processor_id())
6915                                 send_ipi = 1;
6916                 }
6917         }
6918         mutex_unlock(&kvm_lock);
6919
6920         if (freq->old < freq->new && send_ipi) {
6921                 /*
6922                  * We upscale the frequency.  Must make the guest
6923                  * doesn't see old kvmclock values while running with
6924                  * the new frequency, otherwise we risk the guest sees
6925                  * time go backwards.
6926                  *
6927                  * In case we update the frequency for another cpu
6928                  * (which might be in guest context) send an interrupt
6929                  * to kick the cpu out of guest context.  Next time
6930                  * guest context is entered kvmclock will be updated,
6931                  * so the guest will not see stale values.
6932                  */
6933                 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6934         }
6935 }
6936
6937 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6938                                      void *data)
6939 {
6940         struct cpufreq_freqs *freq = data;
6941         int cpu;
6942
6943         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6944                 return 0;
6945         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6946                 return 0;
6947
6948         for_each_cpu(cpu, freq->policy->cpus)
6949                 __kvmclock_cpufreq_notifier(freq, cpu);
6950
6951         return 0;
6952 }
6953
6954 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6955         .notifier_call  = kvmclock_cpufreq_notifier
6956 };
6957
6958 static int kvmclock_cpu_online(unsigned int cpu)
6959 {
6960         tsc_khz_changed(NULL);
6961         return 0;
6962 }
6963
6964 static void kvm_timer_init(void)
6965 {
6966         max_tsc_khz = tsc_khz;
6967
6968         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6969 #ifdef CONFIG_CPU_FREQ
6970                 struct cpufreq_policy policy;
6971                 int cpu;
6972
6973                 memset(&policy, 0, sizeof(policy));
6974                 cpu = get_cpu();
6975                 cpufreq_get_policy(&policy, cpu);
6976                 if (policy.cpuinfo.max_freq)
6977                         max_tsc_khz = policy.cpuinfo.max_freq;
6978                 put_cpu();
6979 #endif
6980                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6981                                           CPUFREQ_TRANSITION_NOTIFIER);
6982         }
6983
6984         cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6985                           kvmclock_cpu_online, kvmclock_cpu_down_prep);
6986 }
6987
6988 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6989 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6990
6991 int kvm_is_in_guest(void)
6992 {
6993         return __this_cpu_read(current_vcpu) != NULL;
6994 }
6995
6996 static int kvm_is_user_mode(void)
6997 {
6998         int user_mode = 3;
6999
7000         if (__this_cpu_read(current_vcpu))
7001                 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7002
7003         return user_mode != 0;
7004 }
7005
7006 static unsigned long kvm_get_guest_ip(void)
7007 {
7008         unsigned long ip = 0;
7009
7010         if (__this_cpu_read(current_vcpu))
7011                 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7012
7013         return ip;
7014 }
7015
7016 static void kvm_handle_intel_pt_intr(void)
7017 {
7018         struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7019
7020         kvm_make_request(KVM_REQ_PMI, vcpu);
7021         __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7022                         (unsigned long *)&vcpu->arch.pmu.global_status);
7023 }
7024
7025 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7026         .is_in_guest            = kvm_is_in_guest,
7027         .is_user_mode           = kvm_is_user_mode,
7028         .get_guest_ip           = kvm_get_guest_ip,
7029         .handle_intel_pt_intr   = kvm_handle_intel_pt_intr,
7030 };
7031
7032 #ifdef CONFIG_X86_64
7033 static void pvclock_gtod_update_fn(struct work_struct *work)
7034 {
7035         struct kvm *kvm;
7036
7037         struct kvm_vcpu *vcpu;
7038         int i;
7039
7040         mutex_lock(&kvm_lock);
7041         list_for_each_entry(kvm, &vm_list, vm_list)
7042                 kvm_for_each_vcpu(i, vcpu, kvm)
7043                         kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7044         atomic_set(&kvm_guest_has_master_clock, 0);
7045         mutex_unlock(&kvm_lock);
7046 }
7047
7048 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7049
7050 /*
7051  * Notification about pvclock gtod data update.
7052  */
7053 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7054                                void *priv)
7055 {
7056         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7057         struct timekeeper *tk = priv;
7058
7059         update_pvclock_gtod(tk);
7060
7061         /* disable master clock if host does not trust, or does not
7062          * use, TSC based clocksource.
7063          */
7064         if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7065             atomic_read(&kvm_guest_has_master_clock) != 0)
7066                 queue_work(system_long_wq, &pvclock_gtod_work);
7067
7068         return 0;
7069 }
7070
7071 static struct notifier_block pvclock_gtod_notifier = {
7072         .notifier_call = pvclock_gtod_notify,
7073 };
7074 #endif
7075
7076 int kvm_arch_init(void *opaque)
7077 {
7078         int r;
7079         struct kvm_x86_ops *ops = opaque;
7080
7081         if (kvm_x86_ops) {
7082                 printk(KERN_ERR "kvm: already loaded the other module\n");
7083                 r = -EEXIST;
7084                 goto out;
7085         }
7086
7087         if (!ops->cpu_has_kvm_support()) {
7088                 printk(KERN_ERR "kvm: no hardware support\n");
7089                 r = -EOPNOTSUPP;
7090                 goto out;
7091         }
7092         if (ops->disabled_by_bios()) {
7093                 printk(KERN_ERR "kvm: disabled by bios\n");
7094                 r = -EOPNOTSUPP;
7095                 goto out;
7096         }
7097
7098         /*
7099          * KVM explicitly assumes that the guest has an FPU and
7100          * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7101          * vCPU's FPU state as a fxregs_state struct.
7102          */
7103         if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7104                 printk(KERN_ERR "kvm: inadequate fpu\n");
7105                 r = -EOPNOTSUPP;
7106                 goto out;
7107         }
7108
7109         r = -ENOMEM;
7110         x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7111                                           __alignof__(struct fpu), SLAB_ACCOUNT,
7112                                           NULL);
7113         if (!x86_fpu_cache) {
7114                 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7115                 goto out;
7116         }
7117
7118         shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7119         if (!shared_msrs) {
7120                 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7121                 goto out_free_x86_fpu_cache;
7122         }
7123
7124         r = kvm_mmu_module_init();
7125         if (r)
7126                 goto out_free_percpu;
7127
7128         kvm_x86_ops = ops;
7129
7130         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7131                         PT_DIRTY_MASK, PT64_NX_MASK, 0,
7132                         PT_PRESENT_MASK, 0, sme_me_mask);
7133         kvm_timer_init();
7134
7135         perf_register_guest_info_callbacks(&kvm_guest_cbs);
7136
7137         if (boot_cpu_has(X86_FEATURE_XSAVE))
7138                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7139
7140         kvm_lapic_init();
7141         if (pi_inject_timer == -1)
7142                 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7143 #ifdef CONFIG_X86_64
7144         pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7145
7146         if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7147                 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7148 #endif
7149
7150         return 0;
7151
7152 out_free_percpu:
7153         free_percpu(shared_msrs);
7154 out_free_x86_fpu_cache:
7155         kmem_cache_destroy(x86_fpu_cache);
7156 out:
7157         return r;
7158 }
7159
7160 void kvm_arch_exit(void)
7161 {
7162 #ifdef CONFIG_X86_64
7163         if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7164                 clear_hv_tscchange_cb();
7165 #endif
7166         kvm_lapic_exit();
7167         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7168
7169         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7170                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7171                                             CPUFREQ_TRANSITION_NOTIFIER);
7172         cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7173 #ifdef CONFIG_X86_64
7174         pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7175 #endif
7176         kvm_x86_ops = NULL;
7177         kvm_mmu_module_exit();
7178         free_percpu(shared_msrs);
7179         kmem_cache_destroy(x86_fpu_cache);
7180 }
7181
7182 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7183 {
7184         ++vcpu->stat.halt_exits;
7185         if (lapic_in_kernel(vcpu)) {
7186                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7187                 return 1;
7188         } else {
7189                 vcpu->run->exit_reason = KVM_EXIT_HLT;
7190                 return 0;
7191         }
7192 }
7193 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7194
7195 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7196 {
7197         int ret = kvm_skip_emulated_instruction(vcpu);
7198         /*
7199          * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7200          * KVM_EXIT_DEBUG here.
7201          */
7202         return kvm_vcpu_halt(vcpu) && ret;
7203 }
7204 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7205
7206 #ifdef CONFIG_X86_64
7207 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7208                                 unsigned long clock_type)
7209 {
7210         struct kvm_clock_pairing clock_pairing;
7211         struct timespec64 ts;
7212         u64 cycle;
7213         int ret;
7214
7215         if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7216                 return -KVM_EOPNOTSUPP;
7217
7218         if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7219                 return -KVM_EOPNOTSUPP;
7220
7221         clock_pairing.sec = ts.tv_sec;
7222         clock_pairing.nsec = ts.tv_nsec;
7223         clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7224         clock_pairing.flags = 0;
7225         memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7226
7227         ret = 0;
7228         if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7229                             sizeof(struct kvm_clock_pairing)))
7230                 ret = -KVM_EFAULT;
7231
7232         return ret;
7233 }
7234 #endif
7235
7236 /*
7237  * kvm_pv_kick_cpu_op:  Kick a vcpu.
7238  *
7239  * @apicid - apicid of vcpu to be kicked.
7240  */
7241 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7242 {
7243         struct kvm_lapic_irq lapic_irq;
7244
7245         lapic_irq.shorthand = 0;
7246         lapic_irq.dest_mode = 0;
7247         lapic_irq.level = 0;
7248         lapic_irq.dest_id = apicid;
7249         lapic_irq.msi_redir_hint = false;
7250
7251         lapic_irq.delivery_mode = APIC_DM_REMRD;
7252         kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7253 }
7254
7255 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7256 {
7257         if (!lapic_in_kernel(vcpu)) {
7258                 WARN_ON_ONCE(vcpu->arch.apicv_active);
7259                 return;
7260         }
7261         if (!vcpu->arch.apicv_active)
7262                 return;
7263
7264         vcpu->arch.apicv_active = false;
7265         kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7266 }
7267
7268 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7269 {
7270         struct kvm_vcpu *target = NULL;
7271         struct kvm_apic_map *map;
7272
7273         rcu_read_lock();
7274         map = rcu_dereference(kvm->arch.apic_map);
7275
7276         if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7277                 target = map->phys_map[dest_id]->vcpu;
7278
7279         rcu_read_unlock();
7280
7281         if (target && READ_ONCE(target->ready))
7282                 kvm_vcpu_yield_to(target);
7283 }
7284
7285 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7286 {
7287         unsigned long nr, a0, a1, a2, a3, ret;
7288         int op_64_bit;
7289
7290         if (kvm_hv_hypercall_enabled(vcpu->kvm))
7291                 return kvm_hv_hypercall(vcpu);
7292
7293         nr = kvm_rax_read(vcpu);
7294         a0 = kvm_rbx_read(vcpu);
7295         a1 = kvm_rcx_read(vcpu);
7296         a2 = kvm_rdx_read(vcpu);
7297         a3 = kvm_rsi_read(vcpu);
7298
7299         trace_kvm_hypercall(nr, a0, a1, a2, a3);
7300
7301         op_64_bit = is_64_bit_mode(vcpu);
7302         if (!op_64_bit) {
7303                 nr &= 0xFFFFFFFF;
7304                 a0 &= 0xFFFFFFFF;
7305                 a1 &= 0xFFFFFFFF;
7306                 a2 &= 0xFFFFFFFF;
7307                 a3 &= 0xFFFFFFFF;
7308         }
7309
7310         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7311                 ret = -KVM_EPERM;
7312                 goto out;
7313         }
7314
7315         switch (nr) {
7316         case KVM_HC_VAPIC_POLL_IRQ:
7317                 ret = 0;
7318                 break;
7319         case KVM_HC_KICK_CPU:
7320                 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7321                 kvm_sched_yield(vcpu->kvm, a1);
7322                 ret = 0;
7323                 break;
7324 #ifdef CONFIG_X86_64
7325         case KVM_HC_CLOCK_PAIRING:
7326                 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7327                 break;
7328 #endif
7329         case KVM_HC_SEND_IPI:
7330                 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7331                 break;
7332         case KVM_HC_SCHED_YIELD:
7333                 kvm_sched_yield(vcpu->kvm, a0);
7334                 ret = 0;
7335                 break;
7336         default:
7337                 ret = -KVM_ENOSYS;
7338                 break;
7339         }
7340 out:
7341         if (!op_64_bit)
7342                 ret = (u32)ret;
7343         kvm_rax_write(vcpu, ret);
7344
7345         ++vcpu->stat.hypercalls;
7346         return kvm_skip_emulated_instruction(vcpu);
7347 }
7348 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7349
7350 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7351 {
7352         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7353         char instruction[3];
7354         unsigned long rip = kvm_rip_read(vcpu);
7355
7356         kvm_x86_ops->patch_hypercall(vcpu, instruction);
7357
7358         return emulator_write_emulated(ctxt, rip, instruction, 3,
7359                 &ctxt->exception);
7360 }
7361
7362 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7363 {
7364         return vcpu->run->request_interrupt_window &&
7365                 likely(!pic_in_kernel(vcpu->kvm));
7366 }
7367
7368 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7369 {
7370         struct kvm_run *kvm_run = vcpu->run;
7371
7372         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7373         kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7374         kvm_run->cr8 = kvm_get_cr8(vcpu);
7375         kvm_run->apic_base = kvm_get_apic_base(vcpu);
7376         kvm_run->ready_for_interrupt_injection =
7377                 pic_in_kernel(vcpu->kvm) ||
7378                 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7379 }
7380
7381 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7382 {
7383         int max_irr, tpr;
7384
7385         if (!kvm_x86_ops->update_cr8_intercept)
7386                 return;
7387
7388         if (!lapic_in_kernel(vcpu))
7389                 return;
7390
7391         if (vcpu->arch.apicv_active)
7392                 return;
7393
7394         if (!vcpu->arch.apic->vapic_addr)
7395                 max_irr = kvm_lapic_find_highest_irr(vcpu);
7396         else
7397                 max_irr = -1;
7398
7399         if (max_irr != -1)
7400                 max_irr >>= 4;
7401
7402         tpr = kvm_lapic_get_cr8(vcpu);
7403
7404         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7405 }
7406
7407 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7408 {
7409         int r;
7410
7411         /* try to reinject previous events if any */
7412
7413         if (vcpu->arch.exception.injected)
7414                 kvm_x86_ops->queue_exception(vcpu);
7415         /*
7416          * Do not inject an NMI or interrupt if there is a pending
7417          * exception.  Exceptions and interrupts are recognized at
7418          * instruction boundaries, i.e. the start of an instruction.
7419          * Trap-like exceptions, e.g. #DB, have higher priority than
7420          * NMIs and interrupts, i.e. traps are recognized before an
7421          * NMI/interrupt that's pending on the same instruction.
7422          * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7423          * priority, but are only generated (pended) during instruction
7424          * execution, i.e. a pending fault-like exception means the
7425          * fault occurred on the *previous* instruction and must be
7426          * serviced prior to recognizing any new events in order to
7427          * fully complete the previous instruction.
7428          */
7429         else if (!vcpu->arch.exception.pending) {
7430                 if (vcpu->arch.nmi_injected)
7431                         kvm_x86_ops->set_nmi(vcpu);
7432                 else if (vcpu->arch.interrupt.injected)
7433                         kvm_x86_ops->set_irq(vcpu);
7434         }
7435
7436         /*
7437          * Call check_nested_events() even if we reinjected a previous event
7438          * in order for caller to determine if it should require immediate-exit
7439          * from L2 to L1 due to pending L1 events which require exit
7440          * from L2 to L1.
7441          */
7442         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7443                 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7444                 if (r != 0)
7445                         return r;
7446         }
7447
7448         /* try to inject new event if pending */
7449         if (vcpu->arch.exception.pending) {
7450                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7451                                         vcpu->arch.exception.has_error_code,
7452                                         vcpu->arch.exception.error_code);
7453
7454                 WARN_ON_ONCE(vcpu->arch.exception.injected);
7455                 vcpu->arch.exception.pending = false;
7456                 vcpu->arch.exception.injected = true;
7457
7458                 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7459                         __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7460                                              X86_EFLAGS_RF);
7461
7462                 if (vcpu->arch.exception.nr == DB_VECTOR) {
7463                         /*
7464                          * This code assumes that nSVM doesn't use
7465                          * check_nested_events(). If it does, the
7466                          * DR6/DR7 changes should happen before L1
7467                          * gets a #VMEXIT for an intercepted #DB in
7468                          * L2.  (Under VMX, on the other hand, the
7469                          * DR6/DR7 changes should not happen in the
7470                          * event of a VM-exit to L1 for an intercepted
7471                          * #DB in L2.)
7472                          */
7473                         kvm_deliver_exception_payload(vcpu);
7474                         if (vcpu->arch.dr7 & DR7_GD) {
7475                                 vcpu->arch.dr7 &= ~DR7_GD;
7476                                 kvm_update_dr7(vcpu);
7477                         }
7478                 }
7479
7480                 kvm_x86_ops->queue_exception(vcpu);
7481         }
7482
7483         /* Don't consider new event if we re-injected an event */
7484         if (kvm_event_needs_reinjection(vcpu))
7485                 return 0;
7486
7487         if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7488             kvm_x86_ops->smi_allowed(vcpu)) {
7489                 vcpu->arch.smi_pending = false;
7490                 ++vcpu->arch.smi_count;
7491                 enter_smm(vcpu);
7492         } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7493                 --vcpu->arch.nmi_pending;
7494                 vcpu->arch.nmi_injected = true;
7495                 kvm_x86_ops->set_nmi(vcpu);
7496         } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7497                 /*
7498                  * Because interrupts can be injected asynchronously, we are
7499                  * calling check_nested_events again here to avoid a race condition.
7500                  * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7501                  * proposal and current concerns.  Perhaps we should be setting
7502                  * KVM_REQ_EVENT only on certain events and not unconditionally?
7503                  */
7504                 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7505                         r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7506                         if (r != 0)
7507                                 return r;
7508                 }
7509                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7510                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7511                                             false);
7512                         kvm_x86_ops->set_irq(vcpu);
7513                 }
7514         }
7515
7516         return 0;
7517 }
7518
7519 static void process_nmi(struct kvm_vcpu *vcpu)
7520 {
7521         unsigned limit = 2;
7522
7523         /*
7524          * x86 is limited to one NMI running, and one NMI pending after it.
7525          * If an NMI is already in progress, limit further NMIs to just one.
7526          * Otherwise, allow two (and we'll inject the first one immediately).
7527          */
7528         if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7529                 limit = 1;
7530
7531         vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7532         vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7533         kvm_make_request(KVM_REQ_EVENT, vcpu);
7534 }
7535
7536 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7537 {
7538         u32 flags = 0;
7539         flags |= seg->g       << 23;
7540         flags |= seg->db      << 22;
7541         flags |= seg->l       << 21;
7542         flags |= seg->avl     << 20;
7543         flags |= seg->present << 15;
7544         flags |= seg->dpl     << 13;
7545         flags |= seg->s       << 12;
7546         flags |= seg->type    << 8;
7547         return flags;
7548 }
7549
7550 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7551 {
7552         struct kvm_segment seg;
7553         int offset;
7554
7555         kvm_get_segment(vcpu, &seg, n);
7556         put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7557
7558         if (n < 3)
7559                 offset = 0x7f84 + n * 12;
7560         else
7561                 offset = 0x7f2c + (n - 3) * 12;
7562
7563         put_smstate(u32, buf, offset + 8, seg.base);
7564         put_smstate(u32, buf, offset + 4, seg.limit);
7565         put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7566 }
7567
7568 #ifdef CONFIG_X86_64
7569 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7570 {
7571         struct kvm_segment seg;
7572         int offset;
7573         u16 flags;
7574
7575         kvm_get_segment(vcpu, &seg, n);
7576         offset = 0x7e00 + n * 16;
7577
7578         flags = enter_smm_get_segment_flags(&seg) >> 8;
7579         put_smstate(u16, buf, offset, seg.selector);
7580         put_smstate(u16, buf, offset + 2, flags);
7581         put_smstate(u32, buf, offset + 4, seg.limit);
7582         put_smstate(u64, buf, offset + 8, seg.base);
7583 }
7584 #endif
7585
7586 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7587 {
7588         struct desc_ptr dt;
7589         struct kvm_segment seg;
7590         unsigned long val;
7591         int i;
7592
7593         put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7594         put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7595         put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7596         put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7597
7598         for (i = 0; i < 8; i++)
7599                 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7600
7601         kvm_get_dr(vcpu, 6, &val);
7602         put_smstate(u32, buf, 0x7fcc, (u32)val);
7603         kvm_get_dr(vcpu, 7, &val);
7604         put_smstate(u32, buf, 0x7fc8, (u32)val);
7605
7606         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7607         put_smstate(u32, buf, 0x7fc4, seg.selector);
7608         put_smstate(u32, buf, 0x7f64, seg.base);
7609         put_smstate(u32, buf, 0x7f60, seg.limit);
7610         put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7611
7612         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7613         put_smstate(u32, buf, 0x7fc0, seg.selector);
7614         put_smstate(u32, buf, 0x7f80, seg.base);
7615         put_smstate(u32, buf, 0x7f7c, seg.limit);
7616         put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7617
7618         kvm_x86_ops->get_gdt(vcpu, &dt);
7619         put_smstate(u32, buf, 0x7f74, dt.address);
7620         put_smstate(u32, buf, 0x7f70, dt.size);
7621
7622         kvm_x86_ops->get_idt(vcpu, &dt);
7623         put_smstate(u32, buf, 0x7f58, dt.address);
7624         put_smstate(u32, buf, 0x7f54, dt.size);
7625
7626         for (i = 0; i < 6; i++)
7627                 enter_smm_save_seg_32(vcpu, buf, i);
7628
7629         put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7630
7631         /* revision id */
7632         put_smstate(u32, buf, 0x7efc, 0x00020000);
7633         put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7634 }
7635
7636 #ifdef CONFIG_X86_64
7637 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7638 {
7639         struct desc_ptr dt;
7640         struct kvm_segment seg;
7641         unsigned long val;
7642         int i;
7643
7644         for (i = 0; i < 16; i++)
7645                 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7646
7647         put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7648         put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7649
7650         kvm_get_dr(vcpu, 6, &val);
7651         put_smstate(u64, buf, 0x7f68, val);
7652         kvm_get_dr(vcpu, 7, &val);
7653         put_smstate(u64, buf, 0x7f60, val);
7654
7655         put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7656         put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7657         put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7658
7659         put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7660
7661         /* revision id */
7662         put_smstate(u32, buf, 0x7efc, 0x00020064);
7663
7664         put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7665
7666         kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7667         put_smstate(u16, buf, 0x7e90, seg.selector);
7668         put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7669         put_smstate(u32, buf, 0x7e94, seg.limit);
7670         put_smstate(u64, buf, 0x7e98, seg.base);
7671
7672         kvm_x86_ops->get_idt(vcpu, &dt);
7673         put_smstate(u32, buf, 0x7e84, dt.size);
7674         put_smstate(u64, buf, 0x7e88, dt.address);
7675
7676         kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7677         put_smstate(u16, buf, 0x7e70, seg.selector);
7678         put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7679         put_smstate(u32, buf, 0x7e74, seg.limit);
7680         put_smstate(u64, buf, 0x7e78, seg.base);
7681
7682         kvm_x86_ops->get_gdt(vcpu, &dt);
7683         put_smstate(u32, buf, 0x7e64, dt.size);
7684         put_smstate(u64, buf, 0x7e68, dt.address);
7685
7686         for (i = 0; i < 6; i++)
7687                 enter_smm_save_seg_64(vcpu, buf, i);
7688 }
7689 #endif
7690
7691 static void enter_smm(struct kvm_vcpu *vcpu)
7692 {
7693         struct kvm_segment cs, ds;
7694         struct desc_ptr dt;
7695         char buf[512];
7696         u32 cr0;
7697
7698         trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7699         memset(buf, 0, 512);
7700 #ifdef CONFIG_X86_64
7701         if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7702                 enter_smm_save_state_64(vcpu, buf);
7703         else
7704 #endif
7705                 enter_smm_save_state_32(vcpu, buf);
7706
7707         /*
7708          * Give pre_enter_smm() a chance to make ISA-specific changes to the
7709          * vCPU state (e.g. leave guest mode) after we've saved the state into
7710          * the SMM state-save area.
7711          */
7712         kvm_x86_ops->pre_enter_smm(vcpu, buf);
7713
7714         vcpu->arch.hflags |= HF_SMM_MASK;
7715         kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7716
7717         if (kvm_x86_ops->get_nmi_mask(vcpu))
7718                 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7719         else
7720                 kvm_x86_ops->set_nmi_mask(vcpu, true);
7721
7722         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7723         kvm_rip_write(vcpu, 0x8000);
7724
7725         cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7726         kvm_x86_ops->set_cr0(vcpu, cr0);
7727         vcpu->arch.cr0 = cr0;
7728
7729         kvm_x86_ops->set_cr4(vcpu, 0);
7730
7731         /* Undocumented: IDT limit is set to zero on entry to SMM.  */
7732         dt.address = dt.size = 0;
7733         kvm_x86_ops->set_idt(vcpu, &dt);
7734
7735         __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7736
7737         cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7738         cs.base = vcpu->arch.smbase;
7739
7740         ds.selector = 0;
7741         ds.base = 0;
7742
7743         cs.limit    = ds.limit = 0xffffffff;
7744         cs.type     = ds.type = 0x3;
7745         cs.dpl      = ds.dpl = 0;
7746         cs.db       = ds.db = 0;
7747         cs.s        = ds.s = 1;
7748         cs.l        = ds.l = 0;
7749         cs.g        = ds.g = 1;
7750         cs.avl      = ds.avl = 0;
7751         cs.present  = ds.present = 1;
7752         cs.unusable = ds.unusable = 0;
7753         cs.padding  = ds.padding = 0;
7754
7755         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7756         kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7757         kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7758         kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7759         kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7760         kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7761
7762 #ifdef CONFIG_X86_64
7763         if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7764                 kvm_x86_ops->set_efer(vcpu, 0);
7765 #endif
7766
7767         kvm_update_cpuid(vcpu);
7768         kvm_mmu_reset_context(vcpu);
7769 }
7770
7771 static void process_smi(struct kvm_vcpu *vcpu)
7772 {
7773         vcpu->arch.smi_pending = true;
7774         kvm_make_request(KVM_REQ_EVENT, vcpu);
7775 }
7776
7777 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7778 {
7779         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7780 }
7781
7782 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7783 {
7784         if (!kvm_apic_present(vcpu))
7785                 return;
7786
7787         bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7788
7789         if (irqchip_split(vcpu->kvm))
7790                 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7791         else {
7792                 if (vcpu->arch.apicv_active)
7793                         kvm_x86_ops->sync_pir_to_irr(vcpu);
7794                 if (ioapic_in_kernel(vcpu->kvm))
7795                         kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7796         }
7797
7798         if (is_guest_mode(vcpu))
7799                 vcpu->arch.load_eoi_exitmap_pending = true;
7800         else
7801                 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7802 }
7803
7804 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7805 {
7806         u64 eoi_exit_bitmap[4];
7807
7808         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7809                 return;
7810
7811         bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7812                   vcpu_to_synic(vcpu)->vec_bitmap, 256);
7813         kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7814 }
7815
7816 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7817                 unsigned long start, unsigned long end,
7818                 bool blockable)
7819 {
7820         unsigned long apic_address;
7821
7822         /*
7823          * The physical address of apic access page is stored in the VMCS.
7824          * Update it when it becomes invalid.
7825          */
7826         apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7827         if (start <= apic_address && apic_address < end)
7828                 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7829
7830         return 0;
7831 }
7832
7833 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7834 {
7835         struct page *page = NULL;
7836
7837         if (!lapic_in_kernel(vcpu))
7838                 return;
7839
7840         if (!kvm_x86_ops->set_apic_access_page_addr)
7841                 return;
7842
7843         page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7844         if (is_error_page(page))
7845                 return;
7846         kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7847
7848         /*
7849          * Do not pin apic access page in memory, the MMU notifier
7850          * will call us again if it is migrated or swapped out.
7851          */
7852         put_page(page);
7853 }
7854 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7855
7856 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7857 {
7858         smp_send_reschedule(vcpu->cpu);
7859 }
7860 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7861
7862 /*
7863  * Returns 1 to let vcpu_run() continue the guest execution loop without
7864  * exiting to the userspace.  Otherwise, the value will be returned to the
7865  * userspace.
7866  */
7867 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7868 {
7869         int r;
7870         bool req_int_win =
7871                 dm_request_for_irq_injection(vcpu) &&
7872                 kvm_cpu_accept_dm_intr(vcpu);
7873
7874         bool req_immediate_exit = false;
7875
7876         if (kvm_request_pending(vcpu)) {
7877                 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7878                         kvm_x86_ops->get_vmcs12_pages(vcpu);
7879                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7880                         kvm_mmu_unload(vcpu);
7881                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7882                         __kvm_migrate_timers(vcpu);
7883                 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7884                         kvm_gen_update_masterclock(vcpu->kvm);
7885                 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7886                         kvm_gen_kvmclock_update(vcpu);
7887                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7888                         r = kvm_guest_time_update(vcpu);
7889                         if (unlikely(r))
7890                                 goto out;
7891                 }
7892                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7893                         kvm_mmu_sync_roots(vcpu);
7894                 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7895                         kvm_mmu_load_cr3(vcpu);
7896                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7897                         kvm_vcpu_flush_tlb(vcpu, true);
7898                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7899                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7900                         r = 0;
7901                         goto out;
7902                 }
7903                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7904                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7905                         vcpu->mmio_needed = 0;
7906                         r = 0;
7907                         goto out;
7908                 }
7909                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7910                         /* Page is swapped out. Do synthetic halt */
7911                         vcpu->arch.apf.halted = true;
7912                         r = 1;
7913                         goto out;
7914                 }
7915                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7916                         record_steal_time(vcpu);
7917                 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7918                         process_smi(vcpu);
7919                 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7920                         process_nmi(vcpu);
7921                 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7922                         kvm_pmu_handle_event(vcpu);
7923                 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7924                         kvm_pmu_deliver_pmi(vcpu);
7925                 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7926                         BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7927                         if (test_bit(vcpu->arch.pending_ioapic_eoi,
7928                                      vcpu->arch.ioapic_handled_vectors)) {
7929                                 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7930                                 vcpu->run->eoi.vector =
7931                                                 vcpu->arch.pending_ioapic_eoi;
7932                                 r = 0;
7933                                 goto out;
7934                         }
7935                 }
7936                 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7937                         vcpu_scan_ioapic(vcpu);
7938                 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7939                         vcpu_load_eoi_exitmap(vcpu);
7940                 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7941                         kvm_vcpu_reload_apic_access_page(vcpu);
7942                 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7943                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7944                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7945                         r = 0;
7946                         goto out;
7947                 }
7948                 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7949                         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7950                         vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7951                         r = 0;
7952                         goto out;
7953                 }
7954                 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7955                         vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7956                         vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7957                         r = 0;
7958                         goto out;
7959                 }
7960
7961                 /*
7962                  * KVM_REQ_HV_STIMER has to be processed after
7963                  * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7964                  * depend on the guest clock being up-to-date
7965                  */
7966                 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7967                         kvm_hv_process_stimers(vcpu);
7968         }
7969
7970         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7971                 ++vcpu->stat.req_event;
7972                 kvm_apic_accept_events(vcpu);
7973                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7974                         r = 1;
7975                         goto out;
7976                 }
7977
7978                 if (inject_pending_event(vcpu, req_int_win) != 0)
7979                         req_immediate_exit = true;
7980                 else {
7981                         /* Enable SMI/NMI/IRQ window open exits if needed.
7982                          *
7983                          * SMIs have three cases:
7984                          * 1) They can be nested, and then there is nothing to
7985                          *    do here because RSM will cause a vmexit anyway.
7986                          * 2) There is an ISA-specific reason why SMI cannot be
7987                          *    injected, and the moment when this changes can be
7988                          *    intercepted.
7989                          * 3) Or the SMI can be pending because
7990                          *    inject_pending_event has completed the injection
7991                          *    of an IRQ or NMI from the previous vmexit, and
7992                          *    then we request an immediate exit to inject the
7993                          *    SMI.
7994                          */
7995                         if (vcpu->arch.smi_pending && !is_smm(vcpu))
7996                                 if (!kvm_x86_ops->enable_smi_window(vcpu))
7997                                         req_immediate_exit = true;
7998                         if (vcpu->arch.nmi_pending)
7999                                 kvm_x86_ops->enable_nmi_window(vcpu);
8000                         if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8001                                 kvm_x86_ops->enable_irq_window(vcpu);
8002                         WARN_ON(vcpu->arch.exception.pending);
8003                 }
8004
8005                 if (kvm_lapic_enabled(vcpu)) {
8006                         update_cr8_intercept(vcpu);
8007                         kvm_lapic_sync_to_vapic(vcpu);
8008                 }
8009         }
8010
8011         r = kvm_mmu_reload(vcpu);
8012         if (unlikely(r)) {
8013                 goto cancel_injection;
8014         }
8015
8016         preempt_disable();
8017
8018         kvm_x86_ops->prepare_guest_switch(vcpu);
8019
8020         /*
8021          * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
8022          * IPI are then delayed after guest entry, which ensures that they
8023          * result in virtual interrupt delivery.
8024          */
8025         local_irq_disable();
8026         vcpu->mode = IN_GUEST_MODE;
8027
8028         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8029
8030         /*
8031          * 1) We should set ->mode before checking ->requests.  Please see
8032          * the comment in kvm_vcpu_exiting_guest_mode().
8033          *
8034          * 2) For APICv, we should set ->mode before checking PID.ON. This
8035          * pairs with the memory barrier implicit in pi_test_and_set_on
8036          * (see vmx_deliver_posted_interrupt).
8037          *
8038          * 3) This also orders the write to mode from any reads to the page
8039          * tables done while the VCPU is running.  Please see the comment
8040          * in kvm_flush_remote_tlbs.
8041          */
8042         smp_mb__after_srcu_read_unlock();
8043
8044         /*
8045          * This handles the case where a posted interrupt was
8046          * notified with kvm_vcpu_kick.
8047          */
8048         if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8049                 kvm_x86_ops->sync_pir_to_irr(vcpu);
8050
8051         if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8052             || need_resched() || signal_pending(current)) {
8053                 vcpu->mode = OUTSIDE_GUEST_MODE;
8054                 smp_wmb();
8055                 local_irq_enable();
8056                 preempt_enable();
8057                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8058                 r = 1;
8059                 goto cancel_injection;
8060         }
8061
8062         if (req_immediate_exit) {
8063                 kvm_make_request(KVM_REQ_EVENT, vcpu);
8064                 kvm_x86_ops->request_immediate_exit(vcpu);
8065         }
8066
8067         trace_kvm_entry(vcpu->vcpu_id);
8068         guest_enter_irqoff();
8069
8070         /* The preempt notifier should have taken care of the FPU already.  */
8071         WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8072
8073         if (unlikely(vcpu->arch.switch_db_regs)) {
8074                 set_debugreg(0, 7);
8075                 set_debugreg(vcpu->arch.eff_db[0], 0);
8076                 set_debugreg(vcpu->arch.eff_db[1], 1);
8077                 set_debugreg(vcpu->arch.eff_db[2], 2);
8078                 set_debugreg(vcpu->arch.eff_db[3], 3);
8079                 set_debugreg(vcpu->arch.dr6, 6);
8080                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8081         }
8082
8083         kvm_x86_ops->run(vcpu);
8084
8085         /*
8086          * Do this here before restoring debug registers on the host.  And
8087          * since we do this before handling the vmexit, a DR access vmexit
8088          * can (a) read the correct value of the debug registers, (b) set
8089          * KVM_DEBUGREG_WONT_EXIT again.
8090          */
8091         if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8092                 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8093                 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8094                 kvm_update_dr0123(vcpu);
8095                 kvm_update_dr6(vcpu);
8096                 kvm_update_dr7(vcpu);
8097                 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8098         }
8099
8100         /*
8101          * If the guest has used debug registers, at least dr7
8102          * will be disabled while returning to the host.
8103          * If we don't have active breakpoints in the host, we don't
8104          * care about the messed up debug address registers. But if
8105          * we have some of them active, restore the old state.
8106          */
8107         if (hw_breakpoint_active())
8108                 hw_breakpoint_restore();
8109
8110         vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8111
8112         vcpu->mode = OUTSIDE_GUEST_MODE;
8113         smp_wmb();
8114
8115         kvm_x86_ops->handle_exit_irqoff(vcpu);
8116
8117         /*
8118          * Consume any pending interrupts, including the possible source of
8119          * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8120          * An instruction is required after local_irq_enable() to fully unblock
8121          * interrupts on processors that implement an interrupt shadow, the
8122          * stat.exits increment will do nicely.
8123          */
8124         kvm_before_interrupt(vcpu);
8125         local_irq_enable();
8126         ++vcpu->stat.exits;
8127         local_irq_disable();
8128         kvm_after_interrupt(vcpu);
8129
8130         guest_exit_irqoff();
8131         if (lapic_in_kernel(vcpu)) {
8132                 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8133                 if (delta != S64_MIN) {
8134                         trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8135                         vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8136                 }
8137         }
8138
8139         local_irq_enable();
8140         preempt_enable();
8141
8142         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8143
8144         /*
8145          * Profile KVM exit RIPs:
8146          */
8147         if (unlikely(prof_on == KVM_PROFILING)) {
8148                 unsigned long rip = kvm_rip_read(vcpu);
8149                 profile_hit(KVM_PROFILING, (void *)rip);
8150         }
8151
8152         if (unlikely(vcpu->arch.tsc_always_catchup))
8153                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8154
8155         if (vcpu->arch.apic_attention)
8156                 kvm_lapic_sync_from_vapic(vcpu);
8157
8158         vcpu->arch.gpa_available = false;
8159         r = kvm_x86_ops->handle_exit(vcpu);
8160         return r;
8161
8162 cancel_injection:
8163         kvm_x86_ops->cancel_injection(vcpu);
8164         if (unlikely(vcpu->arch.apic_attention))
8165                 kvm_lapic_sync_from_vapic(vcpu);
8166 out:
8167         return r;
8168 }
8169
8170 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8171 {
8172         if (!kvm_arch_vcpu_runnable(vcpu) &&
8173             (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8174                 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8175                 kvm_vcpu_block(vcpu);
8176                 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8177
8178                 if (kvm_x86_ops->post_block)
8179                         kvm_x86_ops->post_block(vcpu);
8180
8181                 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8182                         return 1;
8183         }
8184
8185         kvm_apic_accept_events(vcpu);
8186         switch(vcpu->arch.mp_state) {
8187         case KVM_MP_STATE_HALTED:
8188                 vcpu->arch.pv.pv_unhalted = false;
8189                 vcpu->arch.mp_state =
8190                         KVM_MP_STATE_RUNNABLE;
8191                 /* fall through */
8192         case KVM_MP_STATE_RUNNABLE:
8193                 vcpu->arch.apf.halted = false;
8194                 break;
8195         case KVM_MP_STATE_INIT_RECEIVED:
8196                 break;
8197         default:
8198                 return -EINTR;
8199                 break;
8200         }
8201         return 1;
8202 }
8203
8204 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8205 {
8206         if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8207                 kvm_x86_ops->check_nested_events(vcpu, false);
8208
8209         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8210                 !vcpu->arch.apf.halted);
8211 }
8212
8213 static int vcpu_run(struct kvm_vcpu *vcpu)
8214 {
8215         int r;
8216         struct kvm *kvm = vcpu->kvm;
8217
8218         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8219         vcpu->arch.l1tf_flush_l1d = true;
8220
8221         for (;;) {
8222                 if (kvm_vcpu_running(vcpu)) {
8223                         r = vcpu_enter_guest(vcpu);
8224                 } else {
8225                         r = vcpu_block(kvm, vcpu);
8226                 }
8227
8228                 if (r <= 0)
8229                         break;
8230
8231                 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8232                 if (kvm_cpu_has_pending_timer(vcpu))
8233                         kvm_inject_pending_timer_irqs(vcpu);
8234
8235                 if (dm_request_for_irq_injection(vcpu) &&
8236                         kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8237                         r = 0;
8238                         vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8239                         ++vcpu->stat.request_irq_exits;
8240                         break;
8241                 }
8242
8243                 kvm_check_async_pf_completion(vcpu);
8244
8245                 if (signal_pending(current)) {
8246                         r = -EINTR;
8247                         vcpu->run->exit_reason = KVM_EXIT_INTR;
8248                         ++vcpu->stat.signal_exits;
8249                         break;
8250                 }
8251                 if (need_resched()) {
8252                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8253                         cond_resched();
8254                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8255                 }
8256         }
8257
8258         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8259
8260         return r;
8261 }
8262
8263 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8264 {
8265         int r;
8266         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8267         r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8268         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8269         if (r != EMULATE_DONE)
8270                 return 0;
8271         return 1;
8272 }
8273
8274 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8275 {
8276         BUG_ON(!vcpu->arch.pio.count);
8277
8278         return complete_emulated_io(vcpu);
8279 }
8280
8281 /*
8282  * Implements the following, as a state machine:
8283  *
8284  * read:
8285  *   for each fragment
8286  *     for each mmio piece in the fragment
8287  *       write gpa, len
8288  *       exit
8289  *       copy data
8290  *   execute insn
8291  *
8292  * write:
8293  *   for each fragment
8294  *     for each mmio piece in the fragment
8295  *       write gpa, len
8296  *       copy data
8297  *       exit
8298  */
8299 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8300 {
8301         struct kvm_run *run = vcpu->run;
8302         struct kvm_mmio_fragment *frag;
8303         unsigned len;
8304
8305         BUG_ON(!vcpu->mmio_needed);
8306
8307         /* Complete previous fragment */
8308         frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8309         len = min(8u, frag->len);
8310         if (!vcpu->mmio_is_write)
8311                 memcpy(frag->data, run->mmio.data, len);
8312
8313         if (frag->len <= 8) {
8314                 /* Switch to the next fragment. */
8315                 frag++;
8316                 vcpu->mmio_cur_fragment++;
8317         } else {
8318                 /* Go forward to the next mmio piece. */
8319                 frag->data += len;
8320                 frag->gpa += len;
8321                 frag->len -= len;
8322         }
8323
8324         if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8325                 vcpu->mmio_needed = 0;
8326
8327                 /* FIXME: return into emulator if single-stepping.  */
8328                 if (vcpu->mmio_is_write)
8329                         return 1;
8330                 vcpu->mmio_read_completed = 1;
8331                 return complete_emulated_io(vcpu);
8332         }
8333
8334         run->exit_reason = KVM_EXIT_MMIO;
8335         run->mmio.phys_addr = frag->gpa;
8336         if (vcpu->mmio_is_write)
8337                 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8338         run->mmio.len = min(8u, frag->len);
8339         run->mmio.is_write = vcpu->mmio_is_write;
8340         vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8341         return 0;
8342 }
8343
8344 /* Swap (qemu) user FPU context for the guest FPU context. */
8345 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8346 {
8347         fpregs_lock();
8348
8349         copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8350         /* PKRU is separately restored in kvm_x86_ops->run.  */
8351         __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8352                                 ~XFEATURE_MASK_PKRU);
8353
8354         fpregs_mark_activate();
8355         fpregs_unlock();
8356
8357         trace_kvm_fpu(1);
8358 }
8359
8360 /* When vcpu_run ends, restore user space FPU context. */
8361 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8362 {
8363         fpregs_lock();
8364
8365         copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8366         copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8367
8368         fpregs_mark_activate();
8369         fpregs_unlock();
8370
8371         ++vcpu->stat.fpu_reload;
8372         trace_kvm_fpu(0);
8373 }
8374
8375 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8376 {
8377         int r;
8378
8379         vcpu_load(vcpu);
8380         kvm_sigset_activate(vcpu);
8381         kvm_load_guest_fpu(vcpu);
8382
8383         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8384                 if (kvm_run->immediate_exit) {
8385                         r = -EINTR;
8386                         goto out;
8387                 }
8388                 kvm_vcpu_block(vcpu);
8389                 kvm_apic_accept_events(vcpu);
8390                 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8391                 r = -EAGAIN;
8392                 if (signal_pending(current)) {
8393                         r = -EINTR;
8394                         vcpu->run->exit_reason = KVM_EXIT_INTR;
8395                         ++vcpu->stat.signal_exits;
8396                 }
8397                 goto out;
8398         }
8399
8400         if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8401                 r = -EINVAL;
8402                 goto out;
8403         }
8404
8405         if (vcpu->run->kvm_dirty_regs) {
8406                 r = sync_regs(vcpu);
8407                 if (r != 0)
8408                         goto out;
8409         }
8410
8411         /* re-sync apic's tpr */
8412         if (!lapic_in_kernel(vcpu)) {
8413                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8414                         r = -EINVAL;
8415                         goto out;
8416                 }
8417         }
8418
8419         if (unlikely(vcpu->arch.complete_userspace_io)) {
8420                 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8421                 vcpu->arch.complete_userspace_io = NULL;
8422                 r = cui(vcpu);
8423                 if (r <= 0)
8424                         goto out;
8425         } else
8426                 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8427
8428         if (kvm_run->immediate_exit)
8429                 r = -EINTR;
8430         else
8431                 r = vcpu_run(vcpu);
8432
8433 out:
8434         kvm_put_guest_fpu(vcpu);
8435         if (vcpu->run->kvm_valid_regs)
8436                 store_regs(vcpu);
8437         post_kvm_run_save(vcpu);
8438         kvm_sigset_deactivate(vcpu);
8439
8440         vcpu_put(vcpu);
8441         return r;
8442 }
8443
8444 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8445 {
8446         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8447                 /*
8448                  * We are here if userspace calls get_regs() in the middle of
8449                  * instruction emulation. Registers state needs to be copied
8450                  * back from emulation context to vcpu. Userspace shouldn't do
8451                  * that usually, but some bad designed PV devices (vmware
8452                  * backdoor interface) need this to work
8453                  */
8454                 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8455                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8456         }
8457         regs->rax = kvm_rax_read(vcpu);
8458         regs->rbx = kvm_rbx_read(vcpu);
8459         regs->rcx = kvm_rcx_read(vcpu);
8460         regs->rdx = kvm_rdx_read(vcpu);
8461         regs->rsi = kvm_rsi_read(vcpu);
8462         regs->rdi = kvm_rdi_read(vcpu);
8463         regs->rsp = kvm_rsp_read(vcpu);
8464         regs->rbp = kvm_rbp_read(vcpu);
8465 #ifdef CONFIG_X86_64
8466         regs->r8 = kvm_r8_read(vcpu);
8467         regs->r9 = kvm_r9_read(vcpu);
8468         regs->r10 = kvm_r10_read(vcpu);
8469         regs->r11 = kvm_r11_read(vcpu);
8470         regs->r12 = kvm_r12_read(vcpu);
8471         regs->r13 = kvm_r13_read(vcpu);
8472         regs->r14 = kvm_r14_read(vcpu);
8473         regs->r15 = kvm_r15_read(vcpu);
8474 #endif
8475
8476         regs->rip = kvm_rip_read(vcpu);
8477         regs->rflags = kvm_get_rflags(vcpu);
8478 }
8479
8480 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8481 {
8482         vcpu_load(vcpu);
8483         __get_regs(vcpu, regs);
8484         vcpu_put(vcpu);
8485         return 0;
8486 }
8487
8488 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8489 {
8490         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8491         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8492
8493         kvm_rax_write(vcpu, regs->rax);
8494         kvm_rbx_write(vcpu, regs->rbx);
8495         kvm_rcx_write(vcpu, regs->rcx);
8496         kvm_rdx_write(vcpu, regs->rdx);
8497         kvm_rsi_write(vcpu, regs->rsi);
8498         kvm_rdi_write(vcpu, regs->rdi);
8499         kvm_rsp_write(vcpu, regs->rsp);
8500         kvm_rbp_write(vcpu, regs->rbp);
8501 #ifdef CONFIG_X86_64
8502         kvm_r8_write(vcpu, regs->r8);
8503         kvm_r9_write(vcpu, regs->r9);
8504         kvm_r10_write(vcpu, regs->r10);
8505         kvm_r11_write(vcpu, regs->r11);
8506         kvm_r12_write(vcpu, regs->r12);
8507         kvm_r13_write(vcpu, regs->r13);
8508         kvm_r14_write(vcpu, regs->r14);
8509         kvm_r15_write(vcpu, regs->r15);
8510 #endif
8511
8512         kvm_rip_write(vcpu, regs->rip);
8513         kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8514
8515         vcpu->arch.exception.pending = false;
8516
8517         kvm_make_request(KVM_REQ_EVENT, vcpu);
8518 }
8519
8520 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8521 {
8522         vcpu_load(vcpu);
8523         __set_regs(vcpu, regs);
8524         vcpu_put(vcpu);
8525         return 0;
8526 }
8527
8528 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8529 {
8530         struct kvm_segment cs;
8531
8532         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8533         *db = cs.db;
8534         *l = cs.l;
8535 }
8536 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8537
8538 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8539 {
8540         struct desc_ptr dt;
8541
8542         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8543         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8544         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8545         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8546         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8547         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8548
8549         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8550         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8551
8552         kvm_x86_ops->get_idt(vcpu, &dt);
8553         sregs->idt.limit = dt.size;
8554         sregs->idt.base = dt.address;
8555         kvm_x86_ops->get_gdt(vcpu, &dt);
8556         sregs->gdt.limit = dt.size;
8557         sregs->gdt.base = dt.address;
8558
8559         sregs->cr0 = kvm_read_cr0(vcpu);
8560         sregs->cr2 = vcpu->arch.cr2;
8561         sregs->cr3 = kvm_read_cr3(vcpu);
8562         sregs->cr4 = kvm_read_cr4(vcpu);
8563         sregs->cr8 = kvm_get_cr8(vcpu);
8564         sregs->efer = vcpu->arch.efer;
8565         sregs->apic_base = kvm_get_apic_base(vcpu);
8566
8567         memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8568
8569         if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8570                 set_bit(vcpu->arch.interrupt.nr,
8571                         (unsigned long *)sregs->interrupt_bitmap);
8572 }
8573
8574 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8575                                   struct kvm_sregs *sregs)
8576 {
8577         vcpu_load(vcpu);
8578         __get_sregs(vcpu, sregs);
8579         vcpu_put(vcpu);
8580         return 0;
8581 }
8582
8583 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8584                                     struct kvm_mp_state *mp_state)
8585 {
8586         vcpu_load(vcpu);
8587
8588         kvm_apic_accept_events(vcpu);
8589         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8590                                         vcpu->arch.pv.pv_unhalted)
8591                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8592         else
8593                 mp_state->mp_state = vcpu->arch.mp_state;
8594
8595         vcpu_put(vcpu);
8596         return 0;
8597 }
8598
8599 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8600                                     struct kvm_mp_state *mp_state)
8601 {
8602         int ret = -EINVAL;
8603
8604         vcpu_load(vcpu);
8605
8606         if (!lapic_in_kernel(vcpu) &&
8607             mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8608                 goto out;
8609
8610         /* INITs are latched while in SMM */
8611         if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8612             (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8613              mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8614                 goto out;
8615
8616         if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8617                 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8618                 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8619         } else
8620                 vcpu->arch.mp_state = mp_state->mp_state;
8621         kvm_make_request(KVM_REQ_EVENT, vcpu);
8622
8623         ret = 0;
8624 out:
8625         vcpu_put(vcpu);
8626         return ret;
8627 }
8628
8629 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8630                     int reason, bool has_error_code, u32 error_code)
8631 {
8632         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8633         int ret;
8634
8635         init_emulate_ctxt(vcpu);
8636
8637         ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8638                                    has_error_code, error_code);
8639
8640         if (ret)
8641                 return EMULATE_FAIL;
8642
8643         kvm_rip_write(vcpu, ctxt->eip);
8644         kvm_set_rflags(vcpu, ctxt->eflags);
8645         kvm_make_request(KVM_REQ_EVENT, vcpu);
8646         return EMULATE_DONE;
8647 }
8648 EXPORT_SYMBOL_GPL(kvm_task_switch);
8649
8650 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8651 {
8652         if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8653                         (sregs->cr4 & X86_CR4_OSXSAVE))
8654                 return  -EINVAL;
8655
8656         if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8657                 /*
8658                  * When EFER.LME and CR0.PG are set, the processor is in
8659                  * 64-bit mode (though maybe in a 32-bit code segment).
8660                  * CR4.PAE and EFER.LMA must be set.
8661                  */
8662                 if (!(sregs->cr4 & X86_CR4_PAE)
8663                     || !(sregs->efer & EFER_LMA))
8664                         return -EINVAL;
8665         } else {
8666                 /*
8667                  * Not in 64-bit mode: EFER.LMA is clear and the code
8668                  * segment cannot be 64-bit.
8669                  */
8670                 if (sregs->efer & EFER_LMA || sregs->cs.l)
8671                         return -EINVAL;
8672         }
8673
8674         return 0;
8675 }
8676
8677 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8678 {
8679         struct msr_data apic_base_msr;
8680         int mmu_reset_needed = 0;
8681         int cpuid_update_needed = 0;
8682         int pending_vec, max_bits, idx;
8683         struct desc_ptr dt;
8684         int ret = -EINVAL;
8685
8686         if (kvm_valid_sregs(vcpu, sregs))
8687                 goto out;
8688
8689         apic_base_msr.data = sregs->apic_base;
8690         apic_base_msr.host_initiated = true;
8691         if (kvm_set_apic_base(vcpu, &apic_base_msr))
8692                 goto out;
8693
8694         dt.size = sregs->idt.limit;
8695         dt.address = sregs->idt.base;
8696         kvm_x86_ops->set_idt(vcpu, &dt);
8697         dt.size = sregs->gdt.limit;
8698         dt.address = sregs->gdt.base;
8699         kvm_x86_ops->set_gdt(vcpu, &dt);
8700
8701         vcpu->arch.cr2 = sregs->cr2;
8702         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8703         vcpu->arch.cr3 = sregs->cr3;
8704         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8705
8706         kvm_set_cr8(vcpu, sregs->cr8);
8707
8708         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8709         kvm_x86_ops->set_efer(vcpu, sregs->efer);
8710
8711         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8712         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8713         vcpu->arch.cr0 = sregs->cr0;
8714
8715         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8716         cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8717                                 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8718         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8719         if (cpuid_update_needed)
8720                 kvm_update_cpuid(vcpu);
8721
8722         idx = srcu_read_lock(&vcpu->kvm->srcu);
8723         if (is_pae_paging(vcpu)) {
8724                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8725                 mmu_reset_needed = 1;
8726         }
8727         srcu_read_unlock(&vcpu->kvm->srcu, idx);
8728
8729         if (mmu_reset_needed)
8730                 kvm_mmu_reset_context(vcpu);
8731
8732         max_bits = KVM_NR_INTERRUPTS;
8733         pending_vec = find_first_bit(
8734                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8735         if (pending_vec < max_bits) {
8736                 kvm_queue_interrupt(vcpu, pending_vec, false);
8737                 pr_debug("Set back pending irq %d\n", pending_vec);
8738         }
8739
8740         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8741         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8742         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8743         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8744         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8745         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8746
8747         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8748         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8749
8750         update_cr8_intercept(vcpu);
8751
8752         /* Older userspace won't unhalt the vcpu on reset. */
8753         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8754             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8755             !is_protmode(vcpu))
8756                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8757
8758         kvm_make_request(KVM_REQ_EVENT, vcpu);
8759
8760         ret = 0;
8761 out:
8762         return ret;
8763 }
8764
8765 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8766                                   struct kvm_sregs *sregs)
8767 {
8768         int ret;
8769
8770         vcpu_load(vcpu);
8771         ret = __set_sregs(vcpu, sregs);
8772         vcpu_put(vcpu);
8773         return ret;
8774 }
8775
8776 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8777                                         struct kvm_guest_debug *dbg)
8778 {
8779         unsigned long rflags;
8780         int i, r;
8781
8782         vcpu_load(vcpu);
8783
8784         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8785                 r = -EBUSY;
8786                 if (vcpu->arch.exception.pending)
8787                         goto out;
8788                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8789                         kvm_queue_exception(vcpu, DB_VECTOR);
8790                 else
8791                         kvm_queue_exception(vcpu, BP_VECTOR);
8792         }
8793
8794         /*
8795          * Read rflags as long as potentially injected trace flags are still
8796          * filtered out.
8797          */
8798         rflags = kvm_get_rflags(vcpu);
8799
8800         vcpu->guest_debug = dbg->control;
8801         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8802                 vcpu->guest_debug = 0;
8803
8804         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8805                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8806                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8807                 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8808         } else {
8809                 for (i = 0; i < KVM_NR_DB_REGS; i++)
8810                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8811         }
8812         kvm_update_dr7(vcpu);
8813
8814         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8815                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8816                         get_segment_base(vcpu, VCPU_SREG_CS);
8817
8818         /*
8819          * Trigger an rflags update that will inject or remove the trace
8820          * flags.
8821          */
8822         kvm_set_rflags(vcpu, rflags);
8823
8824         kvm_x86_ops->update_bp_intercept(vcpu);
8825
8826         r = 0;
8827
8828 out:
8829         vcpu_put(vcpu);
8830         return r;
8831 }
8832
8833 /*
8834  * Translate a guest virtual address to a guest physical address.
8835  */
8836 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8837                                     struct kvm_translation *tr)
8838 {
8839         unsigned long vaddr = tr->linear_address;
8840         gpa_t gpa;
8841         int idx;
8842
8843         vcpu_load(vcpu);
8844
8845         idx = srcu_read_lock(&vcpu->kvm->srcu);
8846         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8847         srcu_read_unlock(&vcpu->kvm->srcu, idx);
8848         tr->physical_address = gpa;
8849         tr->valid = gpa != UNMAPPED_GVA;
8850         tr->writeable = 1;
8851         tr->usermode = 0;
8852
8853         vcpu_put(vcpu);
8854         return 0;
8855 }
8856
8857 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8858 {
8859         struct fxregs_state *fxsave;
8860
8861         vcpu_load(vcpu);
8862
8863         fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8864         memcpy(fpu->fpr, fxsave->st_space, 128);
8865         fpu->fcw = fxsave->cwd;
8866         fpu->fsw = fxsave->swd;
8867         fpu->ftwx = fxsave->twd;
8868         fpu->last_opcode = fxsave->fop;
8869         fpu->last_ip = fxsave->rip;
8870         fpu->last_dp = fxsave->rdp;
8871         memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8872
8873         vcpu_put(vcpu);
8874         return 0;
8875 }
8876
8877 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8878 {
8879         struct fxregs_state *fxsave;
8880
8881         vcpu_load(vcpu);
8882
8883         fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8884
8885         memcpy(fxsave->st_space, fpu->fpr, 128);
8886         fxsave->cwd = fpu->fcw;
8887         fxsave->swd = fpu->fsw;
8888         fxsave->twd = fpu->ftwx;
8889         fxsave->fop = fpu->last_opcode;
8890         fxsave->rip = fpu->last_ip;
8891         fxsave->rdp = fpu->last_dp;
8892         memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8893
8894         vcpu_put(vcpu);
8895         return 0;
8896 }
8897
8898 static void store_regs(struct kvm_vcpu *vcpu)
8899 {
8900         BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8901
8902         if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8903                 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8904
8905         if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8906                 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8907
8908         if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8909                 kvm_vcpu_ioctl_x86_get_vcpu_events(
8910                                 vcpu, &vcpu->run->s.regs.events);
8911 }
8912
8913 static int sync_regs(struct kvm_vcpu *vcpu)
8914 {
8915         if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8916                 return -EINVAL;
8917
8918         if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8919                 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8920                 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8921         }
8922         if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8923                 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8924                         return -EINVAL;
8925                 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8926         }
8927         if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8928                 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8929                                 vcpu, &vcpu->run->s.regs.events))
8930                         return -EINVAL;
8931                 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8932         }
8933
8934         return 0;
8935 }
8936
8937 static void fx_init(struct kvm_vcpu *vcpu)
8938 {
8939         fpstate_init(&vcpu->arch.guest_fpu->state);
8940         if (boot_cpu_has(X86_FEATURE_XSAVES))
8941                 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8942                         host_xcr0 | XSTATE_COMPACTION_ENABLED;
8943
8944         /*
8945          * Ensure guest xcr0 is valid for loading
8946          */
8947         vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8948
8949         vcpu->arch.cr0 |= X86_CR0_ET;
8950 }
8951
8952 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8953 {
8954         void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8955
8956         kvmclock_reset(vcpu);
8957
8958         kvm_x86_ops->vcpu_free(vcpu);
8959         free_cpumask_var(wbinvd_dirty_mask);
8960 }
8961
8962 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8963                                                 unsigned int id)
8964 {
8965         struct kvm_vcpu *vcpu;
8966
8967         if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8968                 printk_once(KERN_WARNING
8969                 "kvm: SMP vm created on host with unstable TSC; "
8970                 "guest TSC will not be reliable\n");
8971
8972         vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8973
8974         return vcpu;
8975 }
8976
8977 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8978 {
8979         vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
8980         vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8981         kvm_vcpu_mtrr_init(vcpu);
8982         vcpu_load(vcpu);
8983         kvm_vcpu_reset(vcpu, false);
8984         kvm_init_mmu(vcpu, false);
8985         vcpu_put(vcpu);
8986         return 0;
8987 }
8988
8989 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8990 {
8991         struct msr_data msr;
8992         struct kvm *kvm = vcpu->kvm;
8993
8994         kvm_hv_vcpu_postcreate(vcpu);
8995
8996         if (mutex_lock_killable(&vcpu->mutex))
8997                 return;
8998         vcpu_load(vcpu);
8999         msr.data = 0x0;
9000         msr.index = MSR_IA32_TSC;
9001         msr.host_initiated = true;
9002         kvm_write_tsc(vcpu, &msr);
9003         vcpu_put(vcpu);
9004
9005         /* poll control enabled by default */
9006         vcpu->arch.msr_kvm_poll_control = 1;
9007
9008         mutex_unlock(&vcpu->mutex);
9009
9010         if (!kvmclock_periodic_sync)
9011                 return;
9012
9013         schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9014                                         KVMCLOCK_SYNC_PERIOD);
9015 }
9016
9017 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9018 {
9019         vcpu->arch.apf.msr_val = 0;
9020
9021         vcpu_load(vcpu);
9022         kvm_mmu_unload(vcpu);
9023         vcpu_put(vcpu);
9024
9025         kvm_x86_ops->vcpu_free(vcpu);
9026 }
9027
9028 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9029 {
9030         kvm_lapic_reset(vcpu, init_event);
9031
9032         vcpu->arch.hflags = 0;
9033
9034         vcpu->arch.smi_pending = 0;
9035         vcpu->arch.smi_count = 0;
9036         atomic_set(&vcpu->arch.nmi_queued, 0);
9037         vcpu->arch.nmi_pending = 0;
9038         vcpu->arch.nmi_injected = false;
9039         kvm_clear_interrupt_queue(vcpu);
9040         kvm_clear_exception_queue(vcpu);
9041         vcpu->arch.exception.pending = false;
9042
9043         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9044         kvm_update_dr0123(vcpu);
9045         vcpu->arch.dr6 = DR6_INIT;
9046         kvm_update_dr6(vcpu);
9047         vcpu->arch.dr7 = DR7_FIXED_1;
9048         kvm_update_dr7(vcpu);
9049
9050         vcpu->arch.cr2 = 0;
9051
9052         kvm_make_request(KVM_REQ_EVENT, vcpu);
9053         vcpu->arch.apf.msr_val = 0;
9054         vcpu->arch.st.msr_val = 0;
9055
9056         kvmclock_reset(vcpu);
9057
9058         kvm_clear_async_pf_completion_queue(vcpu);
9059         kvm_async_pf_hash_reset(vcpu);
9060         vcpu->arch.apf.halted = false;
9061
9062         if (kvm_mpx_supported()) {
9063                 void *mpx_state_buffer;
9064
9065                 /*
9066                  * To avoid have the INIT path from kvm_apic_has_events() that be
9067                  * called with loaded FPU and does not let userspace fix the state.
9068                  */
9069                 if (init_event)
9070                         kvm_put_guest_fpu(vcpu);
9071                 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9072                                         XFEATURE_BNDREGS);
9073                 if (mpx_state_buffer)
9074                         memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9075                 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9076                                         XFEATURE_BNDCSR);
9077                 if (mpx_state_buffer)
9078                         memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9079                 if (init_event)
9080                         kvm_load_guest_fpu(vcpu);
9081         }
9082
9083         if (!init_event) {
9084                 kvm_pmu_reset(vcpu);
9085                 vcpu->arch.smbase = 0x30000;
9086
9087                 vcpu->arch.msr_misc_features_enables = 0;
9088
9089                 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9090         }
9091
9092         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9093         vcpu->arch.regs_avail = ~0;
9094         vcpu->arch.regs_dirty = ~0;
9095
9096         vcpu->arch.ia32_xss = 0;
9097
9098         kvm_x86_ops->vcpu_reset(vcpu, init_event);
9099 }
9100
9101 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9102 {
9103         struct kvm_segment cs;
9104
9105         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9106         cs.selector = vector << 8;
9107         cs.base = vector << 12;
9108         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9109         kvm_rip_write(vcpu, 0);
9110 }
9111
9112 int kvm_arch_hardware_enable(void)
9113 {
9114         struct kvm *kvm;
9115         struct kvm_vcpu *vcpu;
9116         int i;
9117         int ret;
9118         u64 local_tsc;
9119         u64 max_tsc = 0;
9120         bool stable, backwards_tsc = false;
9121
9122         kvm_shared_msr_cpu_online();
9123         ret = kvm_x86_ops->hardware_enable();
9124         if (ret != 0)
9125                 return ret;
9126
9127         local_tsc = rdtsc();
9128         stable = !kvm_check_tsc_unstable();
9129         list_for_each_entry(kvm, &vm_list, vm_list) {
9130                 kvm_for_each_vcpu(i, vcpu, kvm) {
9131                         if (!stable && vcpu->cpu == smp_processor_id())
9132                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9133                         if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9134                                 backwards_tsc = true;
9135                                 if (vcpu->arch.last_host_tsc > max_tsc)
9136                                         max_tsc = vcpu->arch.last_host_tsc;
9137                         }
9138                 }
9139         }
9140
9141         /*
9142          * Sometimes, even reliable TSCs go backwards.  This happens on
9143          * platforms that reset TSC during suspend or hibernate actions, but
9144          * maintain synchronization.  We must compensate.  Fortunately, we can
9145          * detect that condition here, which happens early in CPU bringup,
9146          * before any KVM threads can be running.  Unfortunately, we can't
9147          * bring the TSCs fully up to date with real time, as we aren't yet far
9148          * enough into CPU bringup that we know how much real time has actually
9149          * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9150          * variables that haven't been updated yet.
9151          *
9152          * So we simply find the maximum observed TSC above, then record the
9153          * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
9154          * the adjustment will be applied.  Note that we accumulate
9155          * adjustments, in case multiple suspend cycles happen before some VCPU
9156          * gets a chance to run again.  In the event that no KVM threads get a
9157          * chance to run, we will miss the entire elapsed period, as we'll have
9158          * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9159          * loose cycle time.  This isn't too big a deal, since the loss will be
9160          * uniform across all VCPUs (not to mention the scenario is extremely
9161          * unlikely). It is possible that a second hibernate recovery happens
9162          * much faster than a first, causing the observed TSC here to be
9163          * smaller; this would require additional padding adjustment, which is
9164          * why we set last_host_tsc to the local tsc observed here.
9165          *
9166          * N.B. - this code below runs only on platforms with reliable TSC,
9167          * as that is the only way backwards_tsc is set above.  Also note
9168          * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9169          * have the same delta_cyc adjustment applied if backwards_tsc
9170          * is detected.  Note further, this adjustment is only done once,
9171          * as we reset last_host_tsc on all VCPUs to stop this from being
9172          * called multiple times (one for each physical CPU bringup).
9173          *
9174          * Platforms with unreliable TSCs don't have to deal with this, they
9175          * will be compensated by the logic in vcpu_load, which sets the TSC to
9176          * catchup mode.  This will catchup all VCPUs to real time, but cannot
9177          * guarantee that they stay in perfect synchronization.
9178          */
9179         if (backwards_tsc) {
9180                 u64 delta_cyc = max_tsc - local_tsc;
9181                 list_for_each_entry(kvm, &vm_list, vm_list) {
9182                         kvm->arch.backwards_tsc_observed = true;
9183                         kvm_for_each_vcpu(i, vcpu, kvm) {
9184                                 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9185                                 vcpu->arch.last_host_tsc = local_tsc;
9186                                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9187                         }
9188
9189                         /*
9190                          * We have to disable TSC offset matching.. if you were
9191                          * booting a VM while issuing an S4 host suspend....
9192                          * you may have some problem.  Solving this issue is
9193                          * left as an exercise to the reader.
9194                          */
9195                         kvm->arch.last_tsc_nsec = 0;
9196                         kvm->arch.last_tsc_write = 0;
9197                 }
9198
9199         }
9200         return 0;
9201 }
9202
9203 void kvm_arch_hardware_disable(void)
9204 {
9205         kvm_x86_ops->hardware_disable();
9206         drop_user_return_notifiers();
9207 }
9208
9209 int kvm_arch_hardware_setup(void)
9210 {
9211         int r;
9212
9213         r = kvm_x86_ops->hardware_setup();
9214         if (r != 0)
9215                 return r;
9216
9217         if (kvm_has_tsc_control) {
9218                 /*
9219                  * Make sure the user can only configure tsc_khz values that
9220                  * fit into a signed integer.
9221                  * A min value is not calculated because it will always
9222                  * be 1 on all machines.
9223                  */
9224                 u64 max = min(0x7fffffffULL,
9225                               __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9226                 kvm_max_guest_tsc_khz = max;
9227
9228                 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9229         }
9230
9231         kvm_init_msr_list();
9232         return 0;
9233 }
9234
9235 void kvm_arch_hardware_unsetup(void)
9236 {
9237         kvm_x86_ops->hardware_unsetup();
9238 }
9239
9240 int kvm_arch_check_processor_compat(void)
9241 {
9242         return kvm_x86_ops->check_processor_compatibility();
9243 }
9244
9245 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9246 {
9247         return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9248 }
9249 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9250
9251 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9252 {
9253         return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9254 }
9255
9256 struct static_key kvm_no_apic_vcpu __read_mostly;
9257 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9258
9259 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9260 {
9261         struct page *page;
9262         int r;
9263
9264         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9265         if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9266                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9267         else
9268                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9269
9270         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9271         if (!page) {
9272                 r = -ENOMEM;
9273                 goto fail;
9274         }
9275         vcpu->arch.pio_data = page_address(page);
9276
9277         kvm_set_tsc_khz(vcpu, max_tsc_khz);
9278
9279         r = kvm_mmu_create(vcpu);
9280         if (r < 0)
9281                 goto fail_free_pio_data;
9282
9283         if (irqchip_in_kernel(vcpu->kvm)) {
9284                 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9285                 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9286                 if (r < 0)
9287                         goto fail_mmu_destroy;
9288         } else
9289                 static_key_slow_inc(&kvm_no_apic_vcpu);
9290
9291         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9292                                        GFP_KERNEL_ACCOUNT);
9293         if (!vcpu->arch.mce_banks) {
9294                 r = -ENOMEM;
9295                 goto fail_free_lapic;
9296         }
9297         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9298
9299         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9300                                 GFP_KERNEL_ACCOUNT)) {
9301                 r = -ENOMEM;
9302                 goto fail_free_mce_banks;
9303         }
9304
9305         fx_init(vcpu);
9306
9307         vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9308
9309         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9310
9311         vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9312
9313         kvm_async_pf_hash_reset(vcpu);
9314         kvm_pmu_init(vcpu);
9315
9316         vcpu->arch.pending_external_vector = -1;
9317         vcpu->arch.preempted_in_kernel = false;
9318
9319         kvm_hv_vcpu_init(vcpu);
9320
9321         return 0;
9322
9323 fail_free_mce_banks:
9324         kfree(vcpu->arch.mce_banks);
9325 fail_free_lapic:
9326         kvm_free_lapic(vcpu);
9327 fail_mmu_destroy:
9328         kvm_mmu_destroy(vcpu);
9329 fail_free_pio_data:
9330         free_page((unsigned long)vcpu->arch.pio_data);
9331 fail:
9332         return r;
9333 }
9334
9335 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9336 {
9337         int idx;
9338
9339         kvm_hv_vcpu_uninit(vcpu);
9340         kvm_pmu_destroy(vcpu);
9341         kfree(vcpu->arch.mce_banks);
9342         kvm_free_lapic(vcpu);
9343         idx = srcu_read_lock(&vcpu->kvm->srcu);
9344         kvm_mmu_destroy(vcpu);
9345         srcu_read_unlock(&vcpu->kvm->srcu, idx);
9346         free_page((unsigned long)vcpu->arch.pio_data);
9347         if (!lapic_in_kernel(vcpu))
9348                 static_key_slow_dec(&kvm_no_apic_vcpu);
9349 }
9350
9351 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9352 {
9353         vcpu->arch.l1tf_flush_l1d = true;
9354         kvm_x86_ops->sched_in(vcpu, cpu);
9355 }
9356
9357 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9358 {
9359         if (type)
9360                 return -EINVAL;
9361
9362         INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9363         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9364         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9365         atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9366
9367         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9368         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9369         /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9370         set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9371                 &kvm->arch.irq_sources_bitmap);
9372
9373         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9374         mutex_init(&kvm->arch.apic_map_lock);
9375         spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9376
9377         kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9378         pvclock_update_vm_gtod_copy(kvm);
9379
9380         kvm->arch.guest_can_read_msr_platform_info = true;
9381
9382         INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9383         INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9384
9385         kvm_hv_init_vm(kvm);
9386         kvm_page_track_init(kvm);
9387         kvm_mmu_init_vm(kvm);
9388
9389         return kvm_x86_ops->vm_init(kvm);
9390 }
9391
9392 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9393 {
9394         vcpu_load(vcpu);
9395         kvm_mmu_unload(vcpu);
9396         vcpu_put(vcpu);
9397 }
9398
9399 static void kvm_free_vcpus(struct kvm *kvm)
9400 {
9401         unsigned int i;
9402         struct kvm_vcpu *vcpu;
9403
9404         /*
9405          * Unpin any mmu pages first.
9406          */
9407         kvm_for_each_vcpu(i, vcpu, kvm) {
9408                 kvm_clear_async_pf_completion_queue(vcpu);
9409                 kvm_unload_vcpu_mmu(vcpu);
9410         }
9411         kvm_for_each_vcpu(i, vcpu, kvm)
9412                 kvm_arch_vcpu_free(vcpu);
9413
9414         mutex_lock(&kvm->lock);
9415         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9416                 kvm->vcpus[i] = NULL;
9417
9418         atomic_set(&kvm->online_vcpus, 0);
9419         mutex_unlock(&kvm->lock);
9420 }
9421
9422 void kvm_arch_sync_events(struct kvm *kvm)
9423 {
9424         cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9425         cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9426         kvm_free_pit(kvm);
9427 }
9428
9429 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9430 {
9431         int i, r;
9432         unsigned long hva;
9433         struct kvm_memslots *slots = kvm_memslots(kvm);
9434         struct kvm_memory_slot *slot, old;
9435
9436         /* Called with kvm->slots_lock held.  */
9437         if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9438                 return -EINVAL;
9439
9440         slot = id_to_memslot(slots, id);
9441         if (size) {
9442                 if (slot->npages)
9443                         return -EEXIST;
9444
9445                 /*
9446                  * MAP_SHARED to prevent internal slot pages from being moved
9447                  * by fork()/COW.
9448                  */
9449                 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9450                               MAP_SHARED | MAP_ANONYMOUS, 0);
9451                 if (IS_ERR((void *)hva))
9452                         return PTR_ERR((void *)hva);
9453         } else {
9454                 if (!slot->npages)
9455                         return 0;
9456
9457                 hva = 0;
9458         }
9459
9460         old = *slot;
9461         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9462                 struct kvm_userspace_memory_region m;
9463
9464                 m.slot = id | (i << 16);
9465                 m.flags = 0;
9466                 m.guest_phys_addr = gpa;
9467                 m.userspace_addr = hva;
9468                 m.memory_size = size;
9469                 r = __kvm_set_memory_region(kvm, &m);
9470                 if (r < 0)
9471                         return r;
9472         }
9473
9474         if (!size)
9475                 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9476
9477         return 0;
9478 }
9479 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9480
9481 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9482 {
9483         int r;
9484
9485         mutex_lock(&kvm->slots_lock);
9486         r = __x86_set_memory_region(kvm, id, gpa, size);
9487         mutex_unlock(&kvm->slots_lock);
9488
9489         return r;
9490 }
9491 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9492
9493 void kvm_arch_destroy_vm(struct kvm *kvm)
9494 {
9495         if (current->mm == kvm->mm) {
9496                 /*
9497                  * Free memory regions allocated on behalf of userspace,
9498                  * unless the the memory map has changed due to process exit
9499                  * or fd copying.
9500                  */
9501                 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9502                 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9503                 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9504         }
9505         if (kvm_x86_ops->vm_destroy)
9506                 kvm_x86_ops->vm_destroy(kvm);
9507         kvm_pic_destroy(kvm);
9508         kvm_ioapic_destroy(kvm);
9509         kvm_free_vcpus(kvm);
9510         kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9511         kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9512         kvm_mmu_uninit_vm(kvm);
9513         kvm_page_track_cleanup(kvm);
9514         kvm_hv_destroy_vm(kvm);
9515 }
9516
9517 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9518                            struct kvm_memory_slot *dont)
9519 {
9520         int i;
9521
9522         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9523                 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9524                         kvfree(free->arch.rmap[i]);
9525                         free->arch.rmap[i] = NULL;
9526                 }
9527                 if (i == 0)
9528                         continue;
9529
9530                 if (!dont || free->arch.lpage_info[i - 1] !=
9531                              dont->arch.lpage_info[i - 1]) {
9532                         kvfree(free->arch.lpage_info[i - 1]);
9533                         free->arch.lpage_info[i - 1] = NULL;
9534                 }
9535         }
9536
9537         kvm_page_track_free_memslot(free, dont);
9538 }
9539
9540 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9541                             unsigned long npages)
9542 {
9543         int i;
9544
9545         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9546                 struct kvm_lpage_info *linfo;
9547                 unsigned long ugfn;
9548                 int lpages;
9549                 int level = i + 1;
9550
9551                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9552                                       slot->base_gfn, level) + 1;
9553
9554                 slot->arch.rmap[i] =
9555                         kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9556                                  GFP_KERNEL_ACCOUNT);
9557                 if (!slot->arch.rmap[i])
9558                         goto out_free;
9559                 if (i == 0)
9560                         continue;
9561
9562                 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9563                 if (!linfo)
9564                         goto out_free;
9565
9566                 slot->arch.lpage_info[i - 1] = linfo;
9567
9568                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9569                         linfo[0].disallow_lpage = 1;
9570                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9571                         linfo[lpages - 1].disallow_lpage = 1;
9572                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9573                 /*
9574                  * If the gfn and userspace address are not aligned wrt each
9575                  * other, or if explicitly asked to, disable large page
9576                  * support for this slot
9577                  */
9578                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9579                     !kvm_largepages_enabled()) {
9580                         unsigned long j;
9581
9582                         for (j = 0; j < lpages; ++j)
9583                                 linfo[j].disallow_lpage = 1;
9584                 }
9585         }
9586
9587         if (kvm_page_track_create_memslot(slot, npages))
9588                 goto out_free;
9589
9590         return 0;
9591
9592 out_free:
9593         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9594                 kvfree(slot->arch.rmap[i]);
9595                 slot->arch.rmap[i] = NULL;
9596                 if (i == 0)
9597                         continue;
9598
9599                 kvfree(slot->arch.lpage_info[i - 1]);
9600                 slot->arch.lpage_info[i - 1] = NULL;
9601         }
9602         return -ENOMEM;
9603 }
9604
9605 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9606 {
9607         /*
9608          * memslots->generation has been incremented.
9609          * mmio generation may have reached its maximum value.
9610          */
9611         kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9612 }
9613
9614 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9615                                 struct kvm_memory_slot *memslot,
9616                                 const struct kvm_userspace_memory_region *mem,
9617                                 enum kvm_mr_change change)
9618 {
9619         return 0;
9620 }
9621
9622 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9623                                      struct kvm_memory_slot *new)
9624 {
9625         /* Still write protect RO slot */
9626         if (new->flags & KVM_MEM_READONLY) {
9627                 kvm_mmu_slot_remove_write_access(kvm, new);
9628                 return;
9629         }
9630
9631         /*
9632          * Call kvm_x86_ops dirty logging hooks when they are valid.
9633          *
9634          * kvm_x86_ops->slot_disable_log_dirty is called when:
9635          *
9636          *  - KVM_MR_CREATE with dirty logging is disabled
9637          *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9638          *
9639          * The reason is, in case of PML, we need to set D-bit for any slots
9640          * with dirty logging disabled in order to eliminate unnecessary GPA
9641          * logging in PML buffer (and potential PML buffer full VMEXT). This
9642          * guarantees leaving PML enabled during guest's lifetime won't have
9643          * any additional overhead from PML when guest is running with dirty
9644          * logging disabled for memory slots.
9645          *
9646          * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9647          * to dirty logging mode.
9648          *
9649          * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9650          *
9651          * In case of write protect:
9652          *
9653          * Write protect all pages for dirty logging.
9654          *
9655          * All the sptes including the large sptes which point to this
9656          * slot are set to readonly. We can not create any new large
9657          * spte on this slot until the end of the logging.
9658          *
9659          * See the comments in fast_page_fault().
9660          */
9661         if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9662                 if (kvm_x86_ops->slot_enable_log_dirty)
9663                         kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9664                 else
9665                         kvm_mmu_slot_remove_write_access(kvm, new);
9666         } else {
9667                 if (kvm_x86_ops->slot_disable_log_dirty)
9668                         kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9669         }
9670 }
9671
9672 void kvm_arch_commit_memory_region(struct kvm *kvm,
9673                                 const struct kvm_userspace_memory_region *mem,
9674                                 const struct kvm_memory_slot *old,
9675                                 const struct kvm_memory_slot *new,
9676                                 enum kvm_mr_change change)
9677 {
9678         if (!kvm->arch.n_requested_mmu_pages)
9679                 kvm_mmu_change_mmu_pages(kvm,
9680                                 kvm_mmu_calculate_default_mmu_pages(kvm));
9681
9682         /*
9683          * Dirty logging tracks sptes in 4k granularity, meaning that large
9684          * sptes have to be split.  If live migration is successful, the guest
9685          * in the source machine will be destroyed and large sptes will be
9686          * created in the destination. However, if the guest continues to run
9687          * in the source machine (for example if live migration fails), small
9688          * sptes will remain around and cause bad performance.
9689          *
9690          * Scan sptes if dirty logging has been stopped, dropping those
9691          * which can be collapsed into a single large-page spte.  Later
9692          * page faults will create the large-page sptes.
9693          */
9694         if ((change != KVM_MR_DELETE) &&
9695                 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9696                 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9697                 kvm_mmu_zap_collapsible_sptes(kvm, new);
9698
9699         /*
9700          * Set up write protection and/or dirty logging for the new slot.
9701          *
9702          * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9703          * been zapped so no dirty logging staff is needed for old slot. For
9704          * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9705          * new and it's also covered when dealing with the new slot.
9706          *
9707          * FIXME: const-ify all uses of struct kvm_memory_slot.
9708          */
9709         if (change != KVM_MR_DELETE)
9710                 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9711 }
9712
9713 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9714 {
9715         kvm_mmu_zap_all(kvm);
9716 }
9717
9718 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9719                                    struct kvm_memory_slot *slot)
9720 {
9721         kvm_page_track_flush_slot(kvm, slot);
9722 }
9723
9724 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9725 {
9726         return (is_guest_mode(vcpu) &&
9727                         kvm_x86_ops->guest_apic_has_interrupt &&
9728                         kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9729 }
9730
9731 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9732 {
9733         if (!list_empty_careful(&vcpu->async_pf.done))
9734                 return true;
9735
9736         if (kvm_apic_has_events(vcpu))
9737                 return true;
9738
9739         if (vcpu->arch.pv.pv_unhalted)
9740                 return true;
9741
9742         if (vcpu->arch.exception.pending)
9743                 return true;
9744
9745         if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9746             (vcpu->arch.nmi_pending &&
9747              kvm_x86_ops->nmi_allowed(vcpu)))
9748                 return true;
9749
9750         if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9751             (vcpu->arch.smi_pending && !is_smm(vcpu)))
9752                 return true;
9753
9754         if (kvm_arch_interrupt_allowed(vcpu) &&
9755             (kvm_cpu_has_interrupt(vcpu) ||
9756             kvm_guest_apic_has_interrupt(vcpu)))
9757                 return true;
9758
9759         if (kvm_hv_has_stimer_pending(vcpu))
9760                 return true;
9761
9762         return false;
9763 }
9764
9765 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9766 {
9767         return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9768 }
9769
9770 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9771 {
9772         if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9773                 return true;
9774
9775         if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9776                 kvm_test_request(KVM_REQ_SMI, vcpu) ||
9777                  kvm_test_request(KVM_REQ_EVENT, vcpu))
9778                 return true;
9779
9780         if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9781                 return true;
9782
9783         return false;
9784 }
9785
9786 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9787 {
9788         return vcpu->arch.preempted_in_kernel;
9789 }
9790
9791 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9792 {
9793         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9794 }
9795
9796 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9797 {
9798         return kvm_x86_ops->interrupt_allowed(vcpu);
9799 }
9800
9801 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9802 {
9803         if (is_64_bit_mode(vcpu))
9804                 return kvm_rip_read(vcpu);
9805         return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9806                      kvm_rip_read(vcpu));
9807 }
9808 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9809
9810 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9811 {
9812         return kvm_get_linear_rip(vcpu) == linear_rip;
9813 }
9814 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9815
9816 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9817 {
9818         unsigned long rflags;
9819
9820         rflags = kvm_x86_ops->get_rflags(vcpu);
9821         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9822                 rflags &= ~X86_EFLAGS_TF;
9823         return rflags;
9824 }
9825 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9826
9827 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9828 {
9829         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9830             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9831                 rflags |= X86_EFLAGS_TF;
9832         kvm_x86_ops->set_rflags(vcpu, rflags);
9833 }
9834
9835 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9836 {
9837         __kvm_set_rflags(vcpu, rflags);
9838         kvm_make_request(KVM_REQ_EVENT, vcpu);
9839 }
9840 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9841
9842 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9843 {
9844         int r;
9845
9846         if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9847               work->wakeup_all)
9848                 return;
9849
9850         r = kvm_mmu_reload(vcpu);
9851         if (unlikely(r))
9852                 return;
9853
9854         if (!vcpu->arch.mmu->direct_map &&
9855               work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9856                 return;
9857
9858         vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9859 }
9860
9861 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9862 {
9863         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9864 }
9865
9866 static inline u32 kvm_async_pf_next_probe(u32 key)
9867 {
9868         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9869 }
9870
9871 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9872 {
9873         u32 key = kvm_async_pf_hash_fn(gfn);
9874
9875         while (vcpu->arch.apf.gfns[key] != ~0)
9876                 key = kvm_async_pf_next_probe(key);
9877
9878         vcpu->arch.apf.gfns[key] = gfn;
9879 }
9880
9881 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9882 {
9883         int i;
9884         u32 key = kvm_async_pf_hash_fn(gfn);
9885
9886         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9887                      (vcpu->arch.apf.gfns[key] != gfn &&
9888                       vcpu->arch.apf.gfns[key] != ~0); i++)
9889                 key = kvm_async_pf_next_probe(key);
9890
9891         return key;
9892 }
9893
9894 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9895 {
9896         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9897 }
9898
9899 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9900 {
9901         u32 i, j, k;
9902
9903         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9904         while (true) {
9905                 vcpu->arch.apf.gfns[i] = ~0;
9906                 do {
9907                         j = kvm_async_pf_next_probe(j);
9908                         if (vcpu->arch.apf.gfns[j] == ~0)
9909                                 return;
9910                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9911                         /*
9912                          * k lies cyclically in ]i,j]
9913                          * |    i.k.j |
9914                          * |....j i.k.| or  |.k..j i...|
9915                          */
9916                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9917                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9918                 i = j;
9919         }
9920 }
9921
9922 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9923 {
9924
9925         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9926                                       sizeof(val));
9927 }
9928
9929 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9930 {
9931
9932         return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9933                                       sizeof(u32));
9934 }
9935
9936 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
9937 {
9938         if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
9939                 return false;
9940
9941         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9942             (vcpu->arch.apf.send_user_only &&
9943              kvm_x86_ops->get_cpl(vcpu) == 0))
9944                 return false;
9945
9946         return true;
9947 }
9948
9949 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
9950 {
9951         if (unlikely(!lapic_in_kernel(vcpu) ||
9952                      kvm_event_needs_reinjection(vcpu) ||
9953                      vcpu->arch.exception.pending))
9954                 return false;
9955
9956         if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
9957                 return false;
9958
9959         /*
9960          * If interrupts are off we cannot even use an artificial
9961          * halt state.
9962          */
9963         return kvm_x86_ops->interrupt_allowed(vcpu);
9964 }
9965
9966 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9967                                      struct kvm_async_pf *work)
9968 {
9969         struct x86_exception fault;
9970
9971         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9972         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9973
9974         if (kvm_can_deliver_async_pf(vcpu) &&
9975             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9976                 fault.vector = PF_VECTOR;
9977                 fault.error_code_valid = true;
9978                 fault.error_code = 0;
9979                 fault.nested_page_fault = false;
9980                 fault.address = work->arch.token;
9981                 fault.async_page_fault = true;
9982                 kvm_inject_page_fault(vcpu, &fault);
9983         } else {
9984                 /*
9985                  * It is not possible to deliver a paravirtualized asynchronous
9986                  * page fault, but putting the guest in an artificial halt state
9987                  * can be beneficial nevertheless: if an interrupt arrives, we
9988                  * can deliver it timely and perhaps the guest will schedule
9989                  * another process.  When the instruction that triggered a page
9990                  * fault is retried, hopefully the page will be ready in the host.
9991                  */
9992                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9993         }
9994 }
9995
9996 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9997                                  struct kvm_async_pf *work)
9998 {
9999         struct x86_exception fault;
10000         u32 val;
10001
10002         if (work->wakeup_all)
10003                 work->arch.token = ~0; /* broadcast wakeup */
10004         else
10005                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10006         trace_kvm_async_pf_ready(work->arch.token, work->gva);
10007
10008         if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10009             !apf_get_user(vcpu, &val)) {
10010                 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10011                     vcpu->arch.exception.pending &&
10012                     vcpu->arch.exception.nr == PF_VECTOR &&
10013                     !apf_put_user(vcpu, 0)) {
10014                         vcpu->arch.exception.injected = false;
10015                         vcpu->arch.exception.pending = false;
10016                         vcpu->arch.exception.nr = 0;
10017                         vcpu->arch.exception.has_error_code = false;
10018                         vcpu->arch.exception.error_code = 0;
10019                         vcpu->arch.exception.has_payload = false;
10020                         vcpu->arch.exception.payload = 0;
10021                 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10022                         fault.vector = PF_VECTOR;
10023                         fault.error_code_valid = true;
10024                         fault.error_code = 0;
10025                         fault.nested_page_fault = false;
10026                         fault.address = work->arch.token;
10027                         fault.async_page_fault = true;
10028                         kvm_inject_page_fault(vcpu, &fault);
10029                 }
10030         }
10031         vcpu->arch.apf.halted = false;
10032         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10033 }
10034
10035 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10036 {
10037         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10038                 return true;
10039         else
10040                 return kvm_can_do_async_pf(vcpu);
10041 }
10042
10043 void kvm_arch_start_assignment(struct kvm *kvm)
10044 {
10045         atomic_inc(&kvm->arch.assigned_device_count);
10046 }
10047 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10048
10049 void kvm_arch_end_assignment(struct kvm *kvm)
10050 {
10051         atomic_dec(&kvm->arch.assigned_device_count);
10052 }
10053 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10054
10055 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10056 {
10057         return atomic_read(&kvm->arch.assigned_device_count);
10058 }
10059 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10060
10061 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10062 {
10063         atomic_inc(&kvm->arch.noncoherent_dma_count);
10064 }
10065 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10066
10067 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10068 {
10069         atomic_dec(&kvm->arch.noncoherent_dma_count);
10070 }
10071 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10072
10073 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10074 {
10075         return atomic_read(&kvm->arch.noncoherent_dma_count);
10076 }
10077 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10078
10079 bool kvm_arch_has_irq_bypass(void)
10080 {
10081         return true;
10082 }
10083
10084 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10085                                       struct irq_bypass_producer *prod)
10086 {
10087         struct kvm_kernel_irqfd *irqfd =
10088                 container_of(cons, struct kvm_kernel_irqfd, consumer);
10089
10090         irqfd->producer = prod;
10091
10092         return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10093                                            prod->irq, irqfd->gsi, 1);
10094 }
10095
10096 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10097                                       struct irq_bypass_producer *prod)
10098 {
10099         int ret;
10100         struct kvm_kernel_irqfd *irqfd =
10101                 container_of(cons, struct kvm_kernel_irqfd, consumer);
10102
10103         WARN_ON(irqfd->producer != prod);
10104         irqfd->producer = NULL;
10105
10106         /*
10107          * When producer of consumer is unregistered, we change back to
10108          * remapped mode, so we can re-use the current implementation
10109          * when the irq is masked/disabled or the consumer side (KVM
10110          * int this case doesn't want to receive the interrupts.
10111         */
10112         ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10113         if (ret)
10114                 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10115                        " fails: %d\n", irqfd->consumer.token, ret);
10116 }
10117
10118 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10119                                    uint32_t guest_irq, bool set)
10120 {
10121         return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10122 }
10123
10124 bool kvm_vector_hashing_enabled(void)
10125 {
10126         return vector_hashing;
10127 }
10128 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10129
10130 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10131 {
10132         return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10133 }
10134 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10135
10136
10137 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10138 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10139 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10140 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10141 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10142 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10143 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10144 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10145 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10146 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10147 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10148 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10149 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10150 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10151 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10152 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10153 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10154 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10155 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10156 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);