arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10  *
  11  * Authors:
  12  *   Avi Kivity   <avi@qumranet.com>
  13  *   Yaniv Kamay  <yaniv@qumranet.com>
  14  *   Amit Shah    <amit.shah@qumranet.com>
  15  *   Ben-Ami Yassour <benami@il.ibm.com>
  16  *
  17  * This work is licensed under the terms of the GNU GPL, version 2.  See
  18  * the COPYING file in the top-level directory.
  19  *
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include "irq.h"
  24 #include "mmu.h"
  25 #include "i8254.h"
  26 #include "tss.h"
  27 #include "kvm_cache_regs.h"
  28 #include "x86.h"
  29 #include "cpuid.h"
  30
  31 #include <linux/clocksource.h>
  32 #include <linux/interrupt.h>
  33 #include <linux/kvm.h>
  34 #include <linux/fs.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/module.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 <trace/events/kvm.h>
  52
  53 #define CREATE_TRACE_POINTS
  54 #include "trace.h"
  55
  56 #include <asm/debugreg.h>
  57 #include <asm/msr.h>
  58 #include <asm/desc.h>
  59 #include <asm/mtrr.h>
  60 #include <asm/mce.h>
  61 #include <asm/i387.h>
  62 #include <asm/fpu-internal.h> /* Ugh! */
  63 #include <asm/xcr.h>
  64 #include <asm/pvclock.h>
  65 #include <asm/div64.h>
  66
  67 #define MAX_IO_MSRS 256
  68 #define KVM_MAX_MCE_BANKS 32
  69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
  70
  71 #define emul_to_vcpu(ctxt) \
  72         container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
  73
  74 /* EFER defaults:
  75  * - enable syscall per default because its emulated by KVM
  76  * - enable LME and LMA per default on 64 bit KVM
  77  */
  78 #ifdef CONFIG_X86_64
  79 static
  80 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
  81 #else
  82 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
  83 #endif
  84
  85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  87
  88 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  89 static void process_nmi(struct kvm_vcpu *vcpu);
  90
  91 struct kvm_x86_ops *kvm_x86_ops;
  92 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  93
  94 static bool ignore_msrs = 0;
  95 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
  96
  97 unsigned int min_timer_period_us = 500;
  98 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
  99
 100 bool kvm_has_tsc_control;
 101 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
 102 u32  kvm_max_guest_tsc_khz;
 103 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
 104
 105 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
 106 static u32 tsc_tolerance_ppm = 250;
 107 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
 108
 109 #define KVM_NR_SHARED_MSRS 16
 110
 111 struct kvm_shared_msrs_global {
 112         int nr;
 113         u32 msrs[KVM_NR_SHARED_MSRS];
 114 };
 115
 116 struct kvm_shared_msrs {
 117         struct user_return_notifier urn;
 118         bool registered;
 119         struct kvm_shared_msr_values {
 120                 u64 host;
 121                 u64 curr;
 122         } values[KVM_NR_SHARED_MSRS];
 123 };
 124
 125 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 126 static struct kvm_shared_msrs __percpu *shared_msrs;
 127
 128 struct kvm_stats_debugfs_item debugfs_entries[] = {
 129         { "pf_fixed", VCPU_STAT(pf_fixed) },
 130         { "pf_guest", VCPU_STAT(pf_guest) },
 131         { "tlb_flush", VCPU_STAT(tlb_flush) },
 132         { "invlpg", VCPU_STAT(invlpg) },
 133         { "exits", VCPU_STAT(exits) },
 134         { "io_exits", VCPU_STAT(io_exits) },
 135         { "mmio_exits", VCPU_STAT(mmio_exits) },
 136         { "signal_exits", VCPU_STAT(signal_exits) },
 137         { "irq_window", VCPU_STAT(irq_window_exits) },
 138         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 139         { "halt_exits", VCPU_STAT(halt_exits) },
 140         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 141         { "hypercalls", VCPU_STAT(hypercalls) },
 142         { "request_irq", VCPU_STAT(request_irq_exits) },
 143         { "irq_exits", VCPU_STAT(irq_exits) },
 144         { "host_state_reload", VCPU_STAT(host_state_reload) },
 145         { "efer_reload", VCPU_STAT(efer_reload) },
 146         { "fpu_reload", VCPU_STAT(fpu_reload) },
 147         { "insn_emulation", VCPU_STAT(insn_emulation) },
 148         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 149         { "irq_injections", VCPU_STAT(irq_injections) },
 150         { "nmi_injections", VCPU_STAT(nmi_injections) },
 151         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 152         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 153         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 154         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 155         { "mmu_flooded", VM_STAT(mmu_flooded) },
 156         { "mmu_recycled", VM_STAT(mmu_recycled) },
 157         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 158         { "mmu_unsync", VM_STAT(mmu_unsync) },
 159         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 160         { "largepages", VM_STAT(lpages) },
 161         { NULL }
 162 };
 163
 164 u64 __read_mostly host_xcr0;
 165
 166 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
 167
 168 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
 169 {
 170         int i;
 171         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
 172                 vcpu->arch.apf.gfns[i] = ~0;
 173 }
 174
 175 static void kvm_on_user_return(struct user_return_notifier *urn)
 176 {
 177         unsigned slot;
 178         struct kvm_shared_msrs *locals
 179                 = container_of(urn, struct kvm_shared_msrs, urn);
 180         struct kvm_shared_msr_values *values;
 181
 182         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 183                 values = &locals->values[slot];
 184                 if (values->host != values->curr) {
 185                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 186                         values->curr = values->host;
 187                 }
 188         }
 189         locals->registered = false;
 190         user_return_notifier_unregister(urn);
 191 }
 192
 193 static void shared_msr_update(unsigned slot, u32 msr)
 194 {
 195         u64 value;
 196         unsigned int cpu = smp_processor_id();
 197         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 198
 199         /* only read, and nobody should modify it at this time,
 200          * so don't need lock */
 201         if (slot >= shared_msrs_global.nr) {
 202                 printk(KERN_ERR "kvm: invalid MSR slot!");
 203                 return;
 204         }
 205         rdmsrl_safe(msr, &value);
 206         smsr->values[slot].host = value;
 207         smsr->values[slot].curr = value;
 208 }
 209
 210 void kvm_define_shared_msr(unsigned slot, u32 msr)
 211 {
 212         if (slot >= shared_msrs_global.nr)
 213                 shared_msrs_global.nr = slot + 1;
 214         shared_msrs_global.msrs[slot] = msr;
 215         /* we need ensured the shared_msr_global have been updated */
 216         smp_wmb();
 217 }
 218 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 219
 220 static void kvm_shared_msr_cpu_online(void)
 221 {
 222         unsigned i;
 223
 224         for (i = 0; i < shared_msrs_global.nr; ++i)
 225                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 226 }
 227
 228 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 229 {
 230         unsigned int cpu = smp_processor_id();
 231         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 232         int err;
 233
 234         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 235                 return 0;
 236         smsr->values[slot].curr = value;
 237         err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
 238         if (err)
 239                 return 1;
 240
 241         if (!smsr->registered) {
 242                 smsr->urn.on_user_return = kvm_on_user_return;
 243                 user_return_notifier_register(&smsr->urn);
 244                 smsr->registered = true;
 245         }
 246         return 0;
 247 }
 248 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 249
 250 static void drop_user_return_notifiers(void)
 251 {
 252         unsigned int cpu = smp_processor_id();
 253         struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
 254
 255         if (smsr->registered)
 256                 kvm_on_user_return(&smsr->urn);
 257 }
 258
 259 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 260 {
 261         return vcpu->arch.apic_base;
 262 }
 263 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 264
 265 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 266 {
 267         /* TODO: reserve bits check */
 268         kvm_lapic_set_base(vcpu, data);
 269 }
 270 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 271
 272 asmlinkage void kvm_spurious_fault(void)
 273 {
 274         /* Fault while not rebooting.  We want the trace. */
 275         BUG();
 276 }
 277 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
 278
 279 #define EXCPT_BENIGN            0
 280 #define EXCPT_CONTRIBUTORY      1
 281 #define EXCPT_PF                2
 282
 283 static int exception_class(int vector)
 284 {
 285         switch (vector) {
 286         case PF_VECTOR:
 287                 return EXCPT_PF;
 288         case DE_VECTOR:
 289         case TS_VECTOR:
 290         case NP_VECTOR:
 291         case SS_VECTOR:
 292         case GP_VECTOR:
 293                 return EXCPT_CONTRIBUTORY;
 294         default:
 295                 break;
 296         }
 297         return EXCPT_BENIGN;
 298 }
 299
 300 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 301                 unsigned nr, bool has_error, u32 error_code,
 302                 bool reinject)
 303 {
 304         u32 prev_nr;
 305         int class1, class2;
 306
 307         kvm_make_request(KVM_REQ_EVENT, vcpu);
 308
 309         if (!vcpu->arch.exception.pending) {
 310         queue:
 311                 vcpu->arch.exception.pending = true;
 312                 vcpu->arch.exception.has_error_code = has_error;
 313                 vcpu->arch.exception.nr = nr;
 314                 vcpu->arch.exception.error_code = error_code;
 315                 vcpu->arch.exception.reinject = reinject;
 316                 return;
 317         }
 318
 319         /* to check exception */
 320         prev_nr = vcpu->arch.exception.nr;
 321         if (prev_nr == DF_VECTOR) {
 322                 /* triple fault -> shutdown */
 323                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
 324                 return;
 325         }
 326         class1 = exception_class(prev_nr);
 327         class2 = exception_class(nr);
 328         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 329                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 330                 /* generate double fault per SDM Table 5-5 */
 331                 vcpu->arch.exception.pending = true;
 332                 vcpu->arch.exception.has_error_code = true;
 333                 vcpu->arch.exception.nr = DF_VECTOR;
 334                 vcpu->arch.exception.error_code = 0;
 335         } else
 336                 /* replace previous exception with a new one in a hope
 337                    that instruction re-execution will regenerate lost
 338                    exception */
 339                 goto queue;
 340 }
 341
 342 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 343 {
 344         kvm_multiple_exception(vcpu, nr, false, 0, false);
 345 }
 346 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 347
 348 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 349 {
 350         kvm_multiple_exception(vcpu, nr, false, 0, true);
 351 }
 352 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 353
 354 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
 355 {
 356         if (err)
 357                 kvm_inject_gp(vcpu, 0);
 358         else
 359                 kvm_x86_ops->skip_emulated_instruction(vcpu);
 360 }
 361 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
 362
 363 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 364 {
 365         ++vcpu->stat.pf_guest;
 366         vcpu->arch.cr2 = fault->address;
 367         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
 368 }
 369 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
 370
 371 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
 372 {
 373         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
 374                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
 375         else
 376                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
 377 }
 378
 379 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 380 {
 381         atomic_inc(&vcpu->arch.nmi_queued);
 382         kvm_make_request(KVM_REQ_NMI, vcpu);
 383 }
 384 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 385
 386 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 387 {
 388         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 389 }
 390 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 391
 392 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 393 {
 394         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 395 }
 396 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 397
 398 /*
 399  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 400  * a #GP and return false.
 401  */
 402 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 403 {
 404         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 405                 return true;
 406         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 407         return false;
 408 }
 409 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 410
 411 /*
 412  * This function will be used to read from the physical memory of the currently
 413  * running guest. The difference to kvm_read_guest_page is that this function
 414  * can read from guest physical or from the guest's guest physical memory.
 415  */
 416 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 417                             gfn_t ngfn, void *data, int offset, int len,
 418                             u32 access)
 419 {
 420         gfn_t real_gfn;
 421         gpa_t ngpa;
 422
 423         ngpa     = gfn_to_gpa(ngfn);
 424         real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
 425         if (real_gfn == UNMAPPED_GVA)
 426                 return -EFAULT;
 427
 428         real_gfn = gpa_to_gfn(real_gfn);
 429
 430         return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
 431 }
 432 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
 433
 434 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
 435                                void *data, int offset, int len, u32 access)
 436 {
 437         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
 438                                        data, offset, len, access);
 439 }
 440
 441 /*
 442  * Load the pae pdptrs.  Return true is they are all valid.
 443  */
 444 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
 445 {
 446         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 447         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 448         int i;
 449         int ret;
 450         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
 451
 452         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
 453                                       offset * sizeof(u64), sizeof(pdpte),
 454                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
 455         if (ret < 0) {
 456                 ret = 0;
 457                 goto out;
 458         }
 459         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 460                 if (is_present_gpte(pdpte[i]) &&
 461                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 462                         ret = 0;
 463                         goto out;
 464                 }
 465         }
 466         ret = 1;
 467
 468         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
 469         __set_bit(VCPU_EXREG_PDPTR,
 470                   (unsigned long *)&vcpu->arch.regs_avail);
 471         __set_bit(VCPU_EXREG_PDPTR,
 472                   (unsigned long *)&vcpu->arch.regs_dirty);
 473 out:
 474
 475         return ret;
 476 }
 477 EXPORT_SYMBOL_GPL(load_pdptrs);
 478
 479 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 480 {
 481         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
 482         bool changed = true;
 483         int offset;
 484         gfn_t gfn;
 485         int r;
 486
 487         if (is_long_mode(vcpu) || !is_pae(vcpu))
 488                 return false;
 489
 490         if (!test_bit(VCPU_EXREG_PDPTR,
 491                       (unsigned long *)&vcpu->arch.regs_avail))
 492                 return true;
 493
 494         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
 495         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
 496         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
 497                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
 498         if (r < 0)
 499                 goto out;
 500         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
 501 out:
 502
 503         return changed;
 504 }
 505
 506 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 507 {
 508         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 509         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 510                                     X86_CR0_CD | X86_CR0_NW;
 511
 512         cr0 |= X86_CR0_ET;
 513
 514 #ifdef CONFIG_X86_64
 515         if (cr0 & 0xffffffff00000000UL)
 516                 return 1;
 517 #endif
 518
 519         cr0 &= ~CR0_RESERVED_BITS;
 520
 521         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 522                 return 1;
 523
 524         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 525                 return 1;
 526
 527         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 528 #ifdef CONFIG_X86_64
 529                 if ((vcpu->arch.efer & EFER_LME)) {
 530                         int cs_db, cs_l;
 531
 532                         if (!is_pae(vcpu))
 533                                 return 1;
 534                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 535                         if (cs_l)
 536                                 return 1;
 537                 } else
 538 #endif
 539                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 540                                                  kvm_read_cr3(vcpu)))
 541                         return 1;
 542         }
 543
 544         if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
 545                 return 1;
 546
 547         kvm_x86_ops->set_cr0(vcpu, cr0);
 548
 549         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
 550                 kvm_clear_async_pf_completion_queue(vcpu);
 551                 kvm_async_pf_hash_reset(vcpu);
 552         }
 553
 554         if ((cr0 ^ old_cr0) & update_bits)
 555                 kvm_mmu_reset_context(vcpu);
 556         return 0;
 557 }
 558 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 559
 560 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 561 {
 562         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 563 }
 564 EXPORT_SYMBOL_GPL(kvm_lmsw);
 565
 566 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 567 {
 568         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
 569                         !vcpu->guest_xcr0_loaded) {
 570                 /* kvm_set_xcr() also depends on this */
 571                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
 572                 vcpu->guest_xcr0_loaded = 1;
 573         }
 574 }
 575
 576 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 577 {
 578         if (vcpu->guest_xcr0_loaded) {
 579                 if (vcpu->arch.xcr0 != host_xcr0)
 580                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
 581                 vcpu->guest_xcr0_loaded = 0;
 582         }
 583 }
 584
 585 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 586 {
 587         u64 xcr0;
 588
 589         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
 590         if (index != XCR_XFEATURE_ENABLED_MASK)
 591                 return 1;
 592         xcr0 = xcr;
 593         if (!(xcr0 & XSTATE_FP))
 594                 return 1;
 595         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
 596                 return 1;
 597         if (xcr0 & ~host_xcr0)
 598                 return 1;
 599         kvm_put_guest_xcr0(vcpu);
 600         vcpu->arch.xcr0 = xcr0;
 601         return 0;
 602 }
 603
 604 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 605 {
 606         if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
 607             __kvm_set_xcr(vcpu, index, xcr)) {
 608                 kvm_inject_gp(vcpu, 0);
 609                 return 1;
 610         }
 611         return 0;
 612 }
 613 EXPORT_SYMBOL_GPL(kvm_set_xcr);
 614
 615 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 616 {
 617         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 618         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
 619                                    X86_CR4_PAE | X86_CR4_SMEP;
 620         if (cr4 & CR4_RESERVED_BITS)
 621                 return 1;
 622
 623         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
 624                 return 1;
 625
 626         if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
 627                 return 1;
 628
 629         if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
 630                 return 1;
 631
 632         if (is_long_mode(vcpu)) {
 633                 if (!(cr4 & X86_CR4_PAE))
 634                         return 1;
 635         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 636                    && ((cr4 ^ old_cr4) & pdptr_bits)
 637                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
 638                                    kvm_read_cr3(vcpu)))
 639                 return 1;
 640
 641         if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
 642                 if (!guest_cpuid_has_pcid(vcpu))
 643                         return 1;
 644
 645                 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
 646                 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
 647                         return 1;
 648         }
 649
 650         if (kvm_x86_ops->set_cr4(vcpu, cr4))
 651                 return 1;
 652
 653         if (((cr4 ^ old_cr4) & pdptr_bits) ||
 654             (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
 655                 kvm_mmu_reset_context(vcpu);
 656
 657         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
 658                 kvm_update_cpuid(vcpu);
 659
 660         return 0;
 661 }
 662 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 663
 664 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 665 {
 666         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
 667                 kvm_mmu_sync_roots(vcpu);
 668                 kvm_mmu_flush_tlb(vcpu);
 669                 return 0;
 670         }
 671
 672         if (is_long_mode(vcpu)) {
 673                 if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
 674                         if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
 675                                 return 1;
 676                 } else
 677                         if (cr3 & CR3_L_MODE_RESERVED_BITS)
 678                                 return 1;
 679         } else {
 680                 if (is_pae(vcpu)) {
 681                         if (cr3 & CR3_PAE_RESERVED_BITS)
 682                                 return 1;
 683                         if (is_paging(vcpu) &&
 684                             !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 685                                 return 1;
 686                 }
 687                 /*
 688                  * We don't check reserved bits in nonpae mode, because
 689                  * this isn't enforced, and VMware depends on this.
 690                  */
 691         }
 692
 693         /*
 694          * Does the new cr3 value map to physical memory? (Note, we
 695          * catch an invalid cr3 even in real-mode, because it would
 696          * cause trouble later on when we turn on paging anyway.)
 697          *
 698          * A real CPU would silently accept an invalid cr3 and would
 699          * attempt to use it - with largely undefined (and often hard
 700          * to debug) behavior on the guest side.
 701          */
 702         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 703                 return 1;
 704         vcpu->arch.cr3 = cr3;
 705         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
 706         vcpu->arch.mmu.new_cr3(vcpu);
 707         return 0;
 708 }
 709 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 710
 711 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 712 {
 713         if (cr8 & CR8_RESERVED_BITS)
 714                 return 1;
 715         if (irqchip_in_kernel(vcpu->kvm))
 716                 kvm_lapic_set_tpr(vcpu, cr8);
 717         else
 718                 vcpu->arch.cr8 = cr8;
 719         return 0;
 720 }
 721 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 722
 723 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 724 {
 725         if (irqchip_in_kernel(vcpu->kvm))
 726                 return kvm_lapic_get_cr8(vcpu);
 727         else
 728                 return vcpu->arch.cr8;
 729 }
 730 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 731
 732 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
 733 {
 734         unsigned long dr7;
 735
 736         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 737                 dr7 = vcpu->arch.guest_debug_dr7;
 738         else
 739                 dr7 = vcpu->arch.dr7;
 740         kvm_x86_ops->set_dr7(vcpu, dr7);
 741         vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK);
 742 }
 743
 744 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 745 {
 746         switch (dr) {
 747         case 0 ... 3:
 748                 vcpu->arch.db[dr] = val;
 749                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 750                         vcpu->arch.eff_db[dr] = val;
 751                 break;
 752         case 4:
 753                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 754                         return 1; /* #UD */
 755                 /* fall through */
 756         case 6:
 757                 if (val & 0xffffffff00000000ULL)
 758                         return -1; /* #GP */
 759                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 760                 break;
 761         case 5:
 762                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 763                         return 1; /* #UD */
 764                 /* fall through */
 765         default: /* 7 */
 766                 if (val & 0xffffffff00000000ULL)
 767                         return -1; /* #GP */
 768                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 769                 kvm_update_dr7(vcpu);
 770                 break;
 771         }
 772
 773         return 0;
 774 }
 775
 776 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 777 {
 778         int res;
 779
 780         res = __kvm_set_dr(vcpu, dr, val);
 781         if (res > 0)
 782                 kvm_queue_exception(vcpu, UD_VECTOR);
 783         else if (res < 0)
 784                 kvm_inject_gp(vcpu, 0);
 785
 786         return res;
 787 }
 788 EXPORT_SYMBOL_GPL(kvm_set_dr);
 789
 790 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 791 {
 792         switch (dr) {
 793         case 0 ... 3:
 794                 *val = vcpu->arch.db[dr];
 795                 break;
 796         case 4:
 797                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 798                         return 1;
 799                 /* fall through */
 800         case 6:
 801                 *val = vcpu->arch.dr6;
 802                 break;
 803         case 5:
 804                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 805                         return 1;
 806                 /* fall through */
 807         default: /* 7 */
 808                 *val = vcpu->arch.dr7;
 809                 break;
 810         }
 811
 812         return 0;
 813 }
 814
 815 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 816 {
 817         if (_kvm_get_dr(vcpu, dr, val)) {
 818                 kvm_queue_exception(vcpu, UD_VECTOR);
 819                 return 1;
 820         }
 821         return 0;
 822 }
 823 EXPORT_SYMBOL_GPL(kvm_get_dr);
 824
 825 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
 826 {
 827         u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 828         u64 data;
 829         int err;
 830
 831         err = kvm_pmu_read_pmc(vcpu, ecx, &data);
 832         if (err)
 833                 return err;
 834         kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
 835         kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
 836         return err;
 837 }
 838 EXPORT_SYMBOL_GPL(kvm_rdpmc);
 839
 840 /*
 841  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 842  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 843  *
 844  * This list is modified at module load time to reflect the
 845  * capabilities of the host cpu. This capabilities test skips MSRs that are
 846  * kvm-specific. Those are put in the beginning of the list.
 847  */
 848
 849 #define KVM_SAVE_MSRS_BEGIN     10
 850 static u32 msrs_to_save[] = {
 851         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 852         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 853         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 854         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
 855         MSR_KVM_PV_EOI_EN,
 856         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 857         MSR_STAR,
 858 #ifdef CONFIG_X86_64
 859         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 860 #endif
 861         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 862 };
 863
 864 static unsigned num_msrs_to_save;
 865
 866 static const u32 emulated_msrs[] = {
 867         MSR_IA32_TSC_ADJUST,
 868         MSR_IA32_TSCDEADLINE,
 869         MSR_IA32_MISC_ENABLE,
 870         MSR_IA32_MCG_STATUS,
 871         MSR_IA32_MCG_CTL,
 872 };
 873
 874 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
 875 {
 876         if (efer & efer_reserved_bits)
 877                 return false;
 878
 879         if (efer & EFER_FFXSR) {
 880                 struct kvm_cpuid_entry2 *feat;
 881
 882                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 883                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 884                         return false;
 885         }
 886
 887         if (efer & EFER_SVME) {
 888                 struct kvm_cpuid_entry2 *feat;
 889
 890                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 891                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 892                         return false;
 893         }
 894
 895         return true;
 896 }
 897 EXPORT_SYMBOL_GPL(kvm_valid_efer);
 898
 899 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 900 {
 901         u64 old_efer = vcpu->arch.efer;
 902
 903         if (!kvm_valid_efer(vcpu, efer))
 904                 return 1;
 905
 906         if (is_paging(vcpu)
 907             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 908                 return 1;
 909
 910         efer &= ~EFER_LMA;
 911         efer |= vcpu->arch.efer & EFER_LMA;
 912
 913         kvm_x86_ops->set_efer(vcpu, efer);
 914
 915         /* Update reserved bits */
 916         if ((efer ^ old_efer) & EFER_NX)
 917                 kvm_mmu_reset_context(vcpu);
 918
 919         return 0;
 920 }
 921
 922 void kvm_enable_efer_bits(u64 mask)
 923 {
 924        efer_reserved_bits &= ~mask;
 925 }
 926 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 927
 928 /*
 929  * Writes msr value into into the appropriate "register".
 930  * Returns 0 on success, non-0 otherwise.
 931  * Assumes vcpu_load() was already called.
 932  */
 933 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
 934 {
 935         switch (msr->index) {
 936         case MSR_FS_BASE:
 937         case MSR_GS_BASE:
 938         case MSR_KERNEL_GS_BASE:
 939         case MSR_CSTAR:
 940         case MSR_LSTAR:
 941                 if (is_noncanonical_address(msr->data))
 942                         return 1;
 943                 break;
 944         case MSR_IA32_SYSENTER_EIP:
 945         case MSR_IA32_SYSENTER_ESP:
 946                 /*
 947                  * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
 948                  * non-canonical address is written on Intel but not on
 949                  * AMD (which ignores the top 32-bits, because it does
 950                  * not implement 64-bit SYSENTER).
 951                  *
 952                  * 64-bit code should hence be able to write a non-canonical
 953                  * value on AMD.  Making the address canonical ensures that
 954                  * vmentry does not fail on Intel after writing a non-canonical
 955                  * value, and that something deterministic happens if the guest
 956                  * invokes 64-bit SYSENTER.
 957                  */
 958                 msr->data = get_canonical(msr->data);
 959         }
 960         return kvm_x86_ops->set_msr(vcpu, msr);
 961 }
 962 EXPORT_SYMBOL_GPL(kvm_set_msr);
 963
 964 /*
 965  * Adapt set_msr() to msr_io()'s calling convention
 966  */
 967 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 968 {
 969         struct msr_data msr;
 970
 971         msr.data = *data;
 972         msr.index = index;
 973         msr.host_initiated = true;
 974         return kvm_set_msr(vcpu, &msr);
 975 }
 976
 977 #ifdef CONFIG_X86_64
 978 struct pvclock_gtod_data {
 979         seqcount_t      seq;
 980
 981         struct { /* extract of a clocksource struct */
 982                 int vclock_mode;
 983                 cycle_t cycle_last;
 984                 cycle_t mask;
 985                 u32     mult;
 986                 u32     shift;
 987         } clock;
 988
 989         /* open coded 'struct timespec' */
 990         u64             monotonic_time_snsec;
 991         time_t          monotonic_time_sec;
 992 };
 993
 994 static struct pvclock_gtod_data pvclock_gtod_data;
 995
 996 static void update_pvclock_gtod(struct timekeeper *tk)
 997 {
 998         struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
 999
1000         write_seqcount_begin(&vdata->seq);
1001
1002         /* copy pvclock gtod data */
1003         vdata->clock.vclock_mode        = tk->clock->archdata.vclock_mode;
1004         vdata->clock.cycle_last         = tk->clock->cycle_last;
1005         vdata->clock.mask               = tk->clock->mask;
1006         vdata->clock.mult               = tk->mult;
1007         vdata->clock.shift              = tk->shift;
1008
1009         vdata->monotonic_time_sec       = tk->xtime_sec
1010                                         + tk->wall_to_monotonic.tv_sec;
1011         vdata->monotonic_time_snsec     = tk->xtime_nsec
1012                                         + (tk->wall_to_monotonic.tv_nsec
1013                                                 << tk->shift);
1014         while (vdata->monotonic_time_snsec >=
1015                                         (((u64)NSEC_PER_SEC) << tk->shift)) {
1016                 vdata->monotonic_time_snsec -=
1017                                         ((u64)NSEC_PER_SEC) << tk->shift;
1018                 vdata->monotonic_time_sec++;
1019         }
1020
1021         write_seqcount_end(&vdata->seq);
1022 }
1023 #endif
1024
1025
1026 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1027 {
1028         int version;
1029         int r;
1030         struct pvclock_wall_clock wc;
1031         struct timespec boot;
1032
1033         if (!wall_clock)
1034                 return;
1035
1036         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1037         if (r)
1038                 return;
1039
1040         if (version & 1)
1041                 ++version;  /* first time write, random junk */
1042
1043         ++version;
1044
1045         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1046
1047         /*
1048          * The guest calculates current wall clock time by adding
1049          * system time (updated by kvm_guest_time_update below) to the
1050          * wall clock specified here.  guest system time equals host
1051          * system time for us, thus we must fill in host boot time here.
1052          */
1053         getboottime(&boot);
1054
1055         if (kvm->arch.kvmclock_offset) {
1056                 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1057                 boot = timespec_sub(boot, ts);
1058         }
1059         wc.sec = boot.tv_sec;
1060         wc.nsec = boot.tv_nsec;
1061         wc.version = version;
1062
1063         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1064
1065         version++;
1066         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1067 }
1068
1069 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1070 {
1071         uint32_t quotient, remainder;
1072
1073         /* Don't try to replace with do_div(), this one calculates
1074          * "(dividend << 32) / divisor" */
1075         __asm__ ( "divl %4"
1076                   : "=a" (quotient), "=d" (remainder)
1077                   : "0" (0), "1" (dividend), "r" (divisor) );
1078         return quotient;
1079 }
1080
1081 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1082                                s8 *pshift, u32 *pmultiplier)
1083 {
1084         uint64_t scaled64;
1085         int32_t  shift = 0;
1086         uint64_t tps64;
1087         uint32_t tps32;
1088
1089         tps64 = base_khz * 1000LL;
1090         scaled64 = scaled_khz * 1000LL;
1091         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1092                 tps64 >>= 1;
1093                 shift--;
1094         }
1095
1096         tps32 = (uint32_t)tps64;
1097         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1098                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1099                         scaled64 >>= 1;
1100                 else
1101                         tps32 <<= 1;
1102                 shift++;
1103         }
1104
1105         *pshift = shift;
1106         *pmultiplier = div_frac(scaled64, tps32);
1107
1108         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1109                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
1110 }
1111
1112 static inline u64 get_kernel_ns(void)
1113 {
1114         struct timespec ts;
1115
1116         WARN_ON(preemptible());
1117         ktime_get_ts(&ts);
1118         monotonic_to_bootbased(&ts);
1119         return timespec_to_ns(&ts);
1120 }
1121
1122 #ifdef CONFIG_X86_64
1123 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1124 #endif
1125
1126 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1127 unsigned long max_tsc_khz;
1128
1129 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1130 {
1131         return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1132                                    vcpu->arch.virtual_tsc_shift);
1133 }
1134
1135 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1136 {
1137         u64 v = (u64)khz * (1000000 + ppm);
1138         do_div(v, 1000000);
1139         return v;
1140 }
1141
1142 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1143 {
1144         u32 thresh_lo, thresh_hi;
1145         int use_scaling = 0;
1146
1147         /* tsc_khz can be zero if TSC calibration fails */
1148         if (this_tsc_khz == 0)
1149                 return;
1150
1151         /* Compute a scale to convert nanoseconds in TSC cycles */
1152         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1153                            &vcpu->arch.virtual_tsc_shift,
1154                            &vcpu->arch.virtual_tsc_mult);
1155         vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1156
1157         /*
1158          * Compute the variation in TSC rate which is acceptable
1159          * within the range of tolerance and decide if the
1160          * rate being applied is within that bounds of the hardware
1161          * rate.  If so, no scaling or compensation need be done.
1162          */
1163         thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1164         thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1165         if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1166                 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1167                 use_scaling = 1;
1168         }
1169         kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1170 }
1171
1172 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1173 {
1174         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1175                                       vcpu->arch.virtual_tsc_mult,
1176                                       vcpu->arch.virtual_tsc_shift);
1177         tsc += vcpu->arch.this_tsc_write;
1178         return tsc;
1179 }
1180
1181 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1182 {
1183 #ifdef CONFIG_X86_64
1184         bool vcpus_matched;
1185         bool do_request = false;
1186         struct kvm_arch *ka = &vcpu->kvm->arch;
1187         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1188
1189         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1190                          atomic_read(&vcpu->kvm->online_vcpus));
1191
1192         if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1193                 if (!ka->use_master_clock)
1194                         do_request = 1;
1195
1196         if (!vcpus_matched && ka->use_master_clock)
1197                         do_request = 1;
1198
1199         if (do_request)
1200                 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1201
1202         trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1203                             atomic_read(&vcpu->kvm->online_vcpus),
1204                             ka->use_master_clock, gtod->clock.vclock_mode);
1205 #endif
1206 }
1207
1208 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1209 {
1210         u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1211         vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1212 }
1213
1214 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1215 {
1216         struct kvm *kvm = vcpu->kvm;
1217         u64 offset, ns, elapsed;
1218         unsigned long flags;
1219         s64 usdiff;
1220         bool matched;
1221         u64 data = msr->data;
1222
1223         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1224         offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1225         ns = get_kernel_ns();
1226         elapsed = ns - kvm->arch.last_tsc_nsec;
1227
1228         if (vcpu->arch.virtual_tsc_khz) {
1229                 int faulted = 0;
1230
1231                 /* n.b - signed multiplication and division required */
1232                 usdiff = data - kvm->arch.last_tsc_write;
1233 #ifdef CONFIG_X86_64
1234                 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1235 #else
1236                 /* do_div() only does unsigned */
1237                 asm("1: idivl %[divisor]\n"
1238                     "2: xor %%edx, %%edx\n"
1239                     "   movl $0, %[faulted]\n"
1240                     "3:\n"
1241                     ".section .fixup,\"ax\"\n"
1242                     "4: movl $1, %[faulted]\n"
1243                     "   jmp  3b\n"
1244                     ".previous\n"
1245
1246                 _ASM_EXTABLE(1b, 4b)
1247
1248                 : "=A"(usdiff), [faulted] "=r" (faulted)
1249                 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1250
1251 #endif
1252                 do_div(elapsed, 1000);
1253                 usdiff -= elapsed;
1254                 if (usdiff < 0)
1255                         usdiff = -usdiff;
1256
1257                 /* idivl overflow => difference is larger than USEC_PER_SEC */
1258                 if (faulted)
1259                         usdiff = USEC_PER_SEC;
1260         } else
1261                 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1262
1263         /*
1264          * Special case: TSC write with a small delta (1 second) of virtual
1265          * cycle time against real time is interpreted as an attempt to
1266          * synchronize the CPU.
1267          *
1268          * For a reliable TSC, we can match TSC offsets, and for an unstable
1269          * TSC, we add elapsed time in this computation.  We could let the
1270          * compensation code attempt to catch up if we fall behind, but
1271          * it's better to try to match offsets from the beginning.
1272          */
1273         if (usdiff < USEC_PER_SEC &&
1274             vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1275                 if (!check_tsc_unstable()) {
1276                         offset = kvm->arch.cur_tsc_offset;
1277                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1278                 } else {
1279                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1280                         data += delta;
1281                         offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1282                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1283                 }
1284                 matched = true;
1285         } else {
1286                 /*
1287                  * We split periods of matched TSC writes into generations.
1288                  * For each generation, we track the original measured
1289                  * nanosecond time, offset, and write, so if TSCs are in
1290                  * sync, we can match exact offset, and if not, we can match
1291                  * exact software computation in compute_guest_tsc()
1292                  *
1293                  * These values are tracked in kvm->arch.cur_xxx variables.
1294                  */
1295                 kvm->arch.cur_tsc_generation++;
1296                 kvm->arch.cur_tsc_nsec = ns;
1297                 kvm->arch.cur_tsc_write = data;
1298                 kvm->arch.cur_tsc_offset = offset;
1299                 matched = false;
1300                 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1301                          kvm->arch.cur_tsc_generation, data);
1302         }
1303
1304         /*
1305          * We also track th most recent recorded KHZ, write and time to
1306          * allow the matching interval to be extended at each write.
1307          */
1308         kvm->arch.last_tsc_nsec = ns;
1309         kvm->arch.last_tsc_write = data;
1310         kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1311
1312         /* Reset of TSC must disable overshoot protection below */
1313         vcpu->arch.hv_clock.tsc_timestamp = 0;
1314         vcpu->arch.last_guest_tsc = data;
1315
1316         /* Keep track of which generation this VCPU has synchronized to */
1317         vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1318         vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1319         vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1320
1321         if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1322                 update_ia32_tsc_adjust_msr(vcpu, offset);
1323         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1324         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1325
1326         spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1327         if (matched)
1328                 kvm->arch.nr_vcpus_matched_tsc++;
1329         else
1330                 kvm->arch.nr_vcpus_matched_tsc = 0;
1331
1332         kvm_track_tsc_matching(vcpu);
1333         spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1334 }
1335
1336 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1337
1338 #ifdef CONFIG_X86_64
1339
1340 static cycle_t read_tsc(void)
1341 {
1342         cycle_t ret;
1343         u64 last;
1344
1345         /*
1346          * Empirically, a fence (of type that depends on the CPU)
1347          * before rdtsc is enough to ensure that rdtsc is ordered
1348          * with respect to loads.  The various CPU manuals are unclear
1349          * as to whether rdtsc can be reordered with later loads,
1350          * but no one has ever seen it happen.
1351          */
1352         rdtsc_barrier();
1353         ret = (cycle_t)vget_cycles();
1354
1355         last = pvclock_gtod_data.clock.cycle_last;
1356
1357         if (likely(ret >= last))
1358                 return ret;
1359
1360         /*
1361          * GCC likes to generate cmov here, but this branch is extremely
1362          * predictable (it's just a funciton of time and the likely is
1363          * very likely) and there's a data dependence, so force GCC
1364          * to generate a branch instead.  I don't barrier() because
1365          * we don't actually need a barrier, and if this function
1366          * ever gets inlined it will generate worse code.
1367          */
1368         asm volatile ("");
1369         return last;
1370 }
1371
1372 static inline u64 vgettsc(cycle_t *cycle_now)
1373 {
1374         long v;
1375         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1376
1377         *cycle_now = read_tsc();
1378
1379         v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1380         return v * gtod->clock.mult;
1381 }
1382
1383 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1384 {
1385         unsigned long seq;
1386         u64 ns;
1387         int mode;
1388         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1389
1390         ts->tv_nsec = 0;
1391         do {
1392                 seq = read_seqcount_begin(&gtod->seq);
1393                 mode = gtod->clock.vclock_mode;
1394                 ts->tv_sec = gtod->monotonic_time_sec;
1395                 ns = gtod->monotonic_time_snsec;
1396                 ns += vgettsc(cycle_now);
1397                 ns >>= gtod->clock.shift;
1398         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1399         timespec_add_ns(ts, ns);
1400
1401         return mode;
1402 }
1403
1404 /* returns true if host is using tsc clocksource */
1405 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1406 {
1407         struct timespec ts;
1408
1409         /* checked again under seqlock below */
1410         if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1411                 return false;
1412
1413         if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1414                 return false;
1415
1416         monotonic_to_bootbased(&ts);
1417         *kernel_ns = timespec_to_ns(&ts);
1418
1419         return true;
1420 }
1421 #endif
1422
1423 /*
1424  *
1425  * Assuming a stable TSC across physical CPUS, and a stable TSC
1426  * across virtual CPUs, the following condition is possible.
1427  * Each numbered line represents an event visible to both
1428  * CPUs at the next numbered event.
1429  *
1430  * "timespecX" represents host monotonic time. "tscX" represents
1431  * RDTSC value.
1432  *
1433  *              VCPU0 on CPU0           |       VCPU1 on CPU1
1434  *
1435  * 1.  read timespec0,tsc0
1436  * 2.                                   | timespec1 = timespec0 + N
1437  *                                      | tsc1 = tsc0 + M
1438  * 3. transition to guest               | transition to guest
1439  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1440  * 5.                                   | ret1 = timespec1 + (rdtsc - tsc1)
1441  *                                      | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1442  *
1443  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1444  *
1445  *      - ret0 < ret1
1446  *      - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1447  *              ...
1448  *      - 0 < N - M => M < N
1449  *
1450  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1451  * always the case (the difference between two distinct xtime instances
1452  * might be smaller then the difference between corresponding TSC reads,
1453  * when updating guest vcpus pvclock areas).
1454  *
1455  * To avoid that problem, do not allow visibility of distinct
1456  * system_timestamp/tsc_timestamp values simultaneously: use a master
1457  * copy of host monotonic time values. Update that master copy
1458  * in lockstep.
1459  *
1460  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1461  *
1462  */
1463
1464 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1465 {
1466 #ifdef CONFIG_X86_64
1467         struct kvm_arch *ka = &kvm->arch;
1468         int vclock_mode;
1469         bool host_tsc_clocksource, vcpus_matched;
1470
1471         vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1472                         atomic_read(&kvm->online_vcpus));
1473
1474         /*
1475          * If the host uses TSC clock, then passthrough TSC as stable
1476          * to the guest.
1477          */
1478         host_tsc_clocksource = kvm_get_time_and_clockread(
1479                                         &ka->master_kernel_ns,
1480                                         &ka->master_cycle_now);
1481
1482         ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1483
1484         if (ka->use_master_clock)
1485                 atomic_set(&kvm_guest_has_master_clock, 1);
1486
1487         vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1488         trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1489                                         vcpus_matched);
1490 #endif
1491 }
1492
1493 static int kvm_guest_time_update(struct kvm_vcpu *v)
1494 {
1495         unsigned long flags, this_tsc_khz;
1496         struct kvm_vcpu_arch *vcpu = &v->arch;
1497         struct kvm_arch *ka = &v->kvm->arch;
1498         s64 kernel_ns, max_kernel_ns;
1499         u64 tsc_timestamp, host_tsc;
1500         struct pvclock_vcpu_time_info guest_hv_clock;
1501         u8 pvclock_flags;
1502         bool use_master_clock;
1503
1504         kernel_ns = 0;
1505         host_tsc = 0;
1506
1507         /*
1508          * If the host uses TSC clock, then passthrough TSC as stable
1509          * to the guest.
1510          */
1511         spin_lock(&ka->pvclock_gtod_sync_lock);
1512         use_master_clock = ka->use_master_clock;
1513         if (use_master_clock) {
1514                 host_tsc = ka->master_cycle_now;
1515                 kernel_ns = ka->master_kernel_ns;
1516         }
1517         spin_unlock(&ka->pvclock_gtod_sync_lock);
1518
1519         /* Keep irq disabled to prevent changes to the clock */
1520         local_irq_save(flags);
1521         this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1522         if (unlikely(this_tsc_khz == 0)) {
1523                 local_irq_restore(flags);
1524                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1525                 return 1;
1526         }
1527         if (!use_master_clock) {
1528                 host_tsc = native_read_tsc();
1529                 kernel_ns = get_kernel_ns();
1530         }
1531
1532         tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1533
1534         /*
1535          * We may have to catch up the TSC to match elapsed wall clock
1536          * time for two reasons, even if kvmclock is used.
1537          *   1) CPU could have been running below the maximum TSC rate
1538          *   2) Broken TSC compensation resets the base at each VCPU
1539          *      entry to avoid unknown leaps of TSC even when running
1540          *      again on the same CPU.  This may cause apparent elapsed
1541          *      time to disappear, and the guest to stand still or run
1542          *      very slowly.
1543          */
1544         if (vcpu->tsc_catchup) {
1545                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1546                 if (tsc > tsc_timestamp) {
1547                         adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1548                         tsc_timestamp = tsc;
1549                 }
1550         }
1551
1552         local_irq_restore(flags);
1553
1554         if (!vcpu->pv_time_enabled)
1555                 return 0;
1556
1557         /*
1558          * Time as measured by the TSC may go backwards when resetting the base
1559          * tsc_timestamp.  The reason for this is that the TSC resolution is
1560          * higher than the resolution of the other clock scales.  Thus, many
1561          * possible measurments of the TSC correspond to one measurement of any
1562          * other clock, and so a spread of values is possible.  This is not a
1563          * problem for the computation of the nanosecond clock; with TSC rates
1564          * around 1GHZ, there can only be a few cycles which correspond to one
1565          * nanosecond value, and any path through this code will inevitably
1566          * take longer than that.  However, with the kernel_ns value itself,
1567          * the precision may be much lower, down to HZ granularity.  If the
1568          * first sampling of TSC against kernel_ns ends in the low part of the
1569          * range, and the second in the high end of the range, we can get:
1570          *
1571          * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1572          *
1573          * As the sampling errors potentially range in the thousands of cycles,
1574          * it is possible such a time value has already been observed by the
1575          * guest.  To protect against this, we must compute the system time as
1576          * observed by the guest and ensure the new system time is greater.
1577          */
1578         max_kernel_ns = 0;
1579         if (vcpu->hv_clock.tsc_timestamp) {
1580                 max_kernel_ns = vcpu->last_guest_tsc -
1581                                 vcpu->hv_clock.tsc_timestamp;
1582                 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1583                                     vcpu->hv_clock.tsc_to_system_mul,
1584                                     vcpu->hv_clock.tsc_shift);
1585                 max_kernel_ns += vcpu->last_kernel_ns;
1586         }
1587
1588         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1589                 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1590                                    &vcpu->hv_clock.tsc_shift,
1591                                    &vcpu->hv_clock.tsc_to_system_mul);
1592                 vcpu->hw_tsc_khz = this_tsc_khz;
1593         }
1594
1595         /* with a master <monotonic time, tsc value> tuple,
1596          * pvclock clock reads always increase at the (scaled) rate
1597          * of guest TSC - no need to deal with sampling errors.
1598          */
1599         if (!use_master_clock) {
1600                 if (max_kernel_ns > kernel_ns)
1601                         kernel_ns = max_kernel_ns;
1602         }
1603         /* With all the info we got, fill in the values */
1604         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1605         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1606         vcpu->last_kernel_ns = kernel_ns;
1607         vcpu->last_guest_tsc = tsc_timestamp;
1608
1609         /*
1610          * The interface expects us to write an even number signaling that the
1611          * update is finished. Since the guest won't see the intermediate
1612          * state, we just increase by 2 at the end.
1613          */
1614         vcpu->hv_clock.version += 2;
1615
1616         if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1617                 &guest_hv_clock, sizeof(guest_hv_clock))))
1618                 return 0;
1619
1620         /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1621         pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1622
1623         if (vcpu->pvclock_set_guest_stopped_request) {
1624                 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1625                 vcpu->pvclock_set_guest_stopped_request = false;
1626         }
1627
1628         /* If the host uses TSC clocksource, then it is stable */
1629         if (use_master_clock)
1630                 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1631
1632         vcpu->hv_clock.flags = pvclock_flags;
1633
1634         kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1635                                 &vcpu->hv_clock,
1636                                 sizeof(vcpu->hv_clock));
1637         return 0;
1638 }
1639
1640 static bool msr_mtrr_valid(unsigned msr)
1641 {
1642         switch (msr) {
1643         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1644         case MSR_MTRRfix64K_00000:
1645         case MSR_MTRRfix16K_80000:
1646         case MSR_MTRRfix16K_A0000:
1647         case MSR_MTRRfix4K_C0000:
1648         case MSR_MTRRfix4K_C8000:
1649         case MSR_MTRRfix4K_D0000:
1650         case MSR_MTRRfix4K_D8000:
1651         case MSR_MTRRfix4K_E0000:
1652         case MSR_MTRRfix4K_E8000:
1653         case MSR_MTRRfix4K_F0000:
1654         case MSR_MTRRfix4K_F8000:
1655         case MSR_MTRRdefType:
1656         case MSR_IA32_CR_PAT:
1657                 return true;
1658         case 0x2f8:
1659                 return true;
1660         }
1661         return false;
1662 }
1663
1664 static bool valid_pat_type(unsigned t)
1665 {
1666         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1667 }
1668
1669 static bool valid_mtrr_type(unsigned t)
1670 {
1671         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1672 }
1673
1674 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1675 {
1676         int i;
1677
1678         if (!msr_mtrr_valid(msr))
1679                 return false;
1680
1681         if (msr == MSR_IA32_CR_PAT) {
1682                 for (i = 0; i < 8; i++)
1683                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
1684                                 return false;
1685                 return true;
1686         } else if (msr == MSR_MTRRdefType) {
1687                 if (data & ~0xcff)
1688                         return false;
1689                 return valid_mtrr_type(data & 0xff);
1690         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1691                 for (i = 0; i < 8 ; i++)
1692                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1693                                 return false;
1694                 return true;
1695         }
1696
1697         /* variable MTRRs */
1698         return valid_mtrr_type(data & 0xff);
1699 }
1700
1701 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1702 {
1703         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1704
1705         if (!mtrr_valid(vcpu, msr, data))
1706                 return 1;
1707
1708         if (msr == MSR_MTRRdefType) {
1709                 vcpu->arch.mtrr_state.def_type = data;
1710                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1711         } else if (msr == MSR_MTRRfix64K_00000)
1712                 p[0] = data;
1713         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1714                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1715         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1716                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1717         else if (msr == MSR_IA32_CR_PAT)
1718                 vcpu->arch.pat = data;
1719         else {  /* Variable MTRRs */
1720                 int idx, is_mtrr_mask;
1721                 u64 *pt;
1722
1723                 idx = (msr - 0x200) / 2;
1724                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1725                 if (!is_mtrr_mask)
1726                         pt =
1727                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1728                 else
1729                         pt =
1730                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1731                 *pt = data;
1732         }
1733
1734         kvm_mmu_reset_context(vcpu);
1735         return 0;
1736 }
1737
1738 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1739 {
1740         u64 mcg_cap = vcpu->arch.mcg_cap;
1741         unsigned bank_num = mcg_cap & 0xff;
1742
1743         switch (msr) {
1744         case MSR_IA32_MCG_STATUS:
1745                 vcpu->arch.mcg_status = data;
1746                 break;
1747         case MSR_IA32_MCG_CTL:
1748                 if (!(mcg_cap & MCG_CTL_P))
1749                         return 1;
1750                 if (data != 0 && data != ~(u64)0)
1751                         return -1;
1752                 vcpu->arch.mcg_ctl = data;
1753                 break;
1754         default:
1755                 if (msr >= MSR_IA32_MC0_CTL &&
1756                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1757                         u32 offset = msr - MSR_IA32_MC0_CTL;
1758                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1759                          * some Linux kernels though clear bit 10 in bank 4 to
1760                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1761                          * this to avoid an uncatched #GP in the guest
1762                          */
1763                         if ((offset & 0x3) == 0 &&
1764                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1765                                 return -1;
1766                         vcpu->arch.mce_banks[offset] = data;
1767                         break;
1768                 }
1769                 return 1;
1770         }
1771         return 0;
1772 }
1773
1774 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1775 {
1776         struct kvm *kvm = vcpu->kvm;
1777         int lm = is_long_mode(vcpu);
1778         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1779                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1780         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1781                 : kvm->arch.xen_hvm_config.blob_size_32;
1782         u32 page_num = data & ~PAGE_MASK;
1783         u64 page_addr = data & PAGE_MASK;
1784         u8 *page;
1785         int r;
1786
1787         r = -E2BIG;
1788         if (page_num >= blob_size)
1789                 goto out;
1790         r = -ENOMEM;
1791         page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1792         if (IS_ERR(page)) {
1793                 r = PTR_ERR(page);
1794                 goto out;
1795         }
1796         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1797                 goto out_free;
1798         r = 0;
1799 out_free:
1800         kfree(page);
1801 out:
1802         return r;
1803 }
1804
1805 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1806 {
1807         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1808 }
1809
1810 static bool kvm_hv_msr_partition_wide(u32 msr)
1811 {
1812         bool r = false;
1813         switch (msr) {
1814         case HV_X64_MSR_GUEST_OS_ID:
1815         case HV_X64_MSR_HYPERCALL:
1816                 r = true;
1817                 break;
1818         }
1819
1820         return r;
1821 }
1822
1823 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1824 {
1825         struct kvm *kvm = vcpu->kvm;
1826
1827         switch (msr) {
1828         case HV_X64_MSR_GUEST_OS_ID:
1829                 kvm->arch.hv_guest_os_id = data;
1830                 /* setting guest os id to zero disables hypercall page */
1831                 if (!kvm->arch.hv_guest_os_id)
1832                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1833                 break;
1834         case HV_X64_MSR_HYPERCALL: {
1835                 u64 gfn;
1836                 unsigned long addr;
1837                 u8 instructions[4];
1838
1839                 /* if guest os id is not set hypercall should remain disabled */
1840                 if (!kvm->arch.hv_guest_os_id)
1841                         break;
1842                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1843                         kvm->arch.hv_hypercall = data;
1844                         break;
1845                 }
1846                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1847                 addr = gfn_to_hva(kvm, gfn);
1848                 if (kvm_is_error_hva(addr))
1849                         return 1;
1850                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1851                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1852                 if (__copy_to_user((void __user *)addr, instructions, 4))
1853                         return 1;
1854                 kvm->arch.hv_hypercall = data;
1855                 break;
1856         }
1857         default:
1858                 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1859                             "data 0x%llx\n", msr, data);
1860                 return 1;
1861         }
1862         return 0;
1863 }
1864
1865 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1866 {
1867         switch (msr) {
1868         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1869                 unsigned long addr;
1870
1871                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1872                         vcpu->arch.hv_vapic = data;
1873                         break;
1874                 }
1875                 addr = gfn_to_hva(vcpu->kvm, data >>
1876                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1877                 if (kvm_is_error_hva(addr))
1878                         return 1;
1879                 if (__clear_user((void __user *)addr, PAGE_SIZE))
1880                         return 1;
1881                 vcpu->arch.hv_vapic = data;
1882                 break;
1883         }
1884         case HV_X64_MSR_EOI:
1885                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1886         case HV_X64_MSR_ICR:
1887                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1888         case HV_X64_MSR_TPR:
1889                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1890         default:
1891                 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1892                             "data 0x%llx\n", msr, data);
1893                 return 1;
1894         }
1895
1896         return 0;
1897 }
1898
1899 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1900 {
1901         gpa_t gpa = data & ~0x3f;
1902
1903         /* Bits 2:5 are reserved, Should be zero */
1904         if (data & 0x3c)
1905                 return 1;
1906
1907         vcpu->arch.apf.msr_val = data;
1908
1909         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1910                 kvm_clear_async_pf_completion_queue(vcpu);
1911                 kvm_async_pf_hash_reset(vcpu);
1912                 return 0;
1913         }
1914
1915         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1916                                         sizeof(u32)))
1917                 return 1;
1918
1919         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1920         kvm_async_pf_wakeup_all(vcpu);
1921         return 0;
1922 }
1923
1924 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1925 {
1926         vcpu->arch.pv_time_enabled = false;
1927 }
1928
1929 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1930 {
1931         u64 delta;
1932
1933         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1934                 return;
1935
1936         delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1937         vcpu->arch.st.last_steal = current->sched_info.run_delay;
1938         vcpu->arch.st.accum_steal = delta;
1939 }
1940
1941 static void record_steal_time(struct kvm_vcpu *vcpu)
1942 {
1943         if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1944                 return;
1945
1946         if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1947                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1948                 return;
1949
1950         vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1951         vcpu->arch.st.steal.version += 2;
1952         vcpu->arch.st.accum_steal = 0;
1953
1954         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1955                 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1956 }
1957
1958 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1959 {
1960         bool pr = false;
1961         u32 msr = msr_info->index;
1962         u64 data = msr_info->data;
1963
1964         switch (msr) {
1965         case MSR_AMD64_NB_CFG:
1966         case MSR_IA32_UCODE_REV:
1967         case MSR_IA32_UCODE_WRITE:
1968         case MSR_VM_HSAVE_PA:
1969         case MSR_AMD64_PATCH_LOADER:
1970         case MSR_AMD64_BU_CFG2:
1971                 break;
1972
1973         case MSR_EFER:
1974                 return set_efer(vcpu, data);
1975         case MSR_K7_HWCR:
1976                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1977                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1978                 data &= ~(u64)0x8;      /* ignore TLB cache disable */
1979                 if (data != 0) {
1980                         vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1981                                     data);
1982                         return 1;
1983                 }
1984                 break;
1985         case MSR_FAM10H_MMIO_CONF_BASE:
1986                 if (data != 0) {
1987                         vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1988                                     "0x%llx\n", data);
1989                         return 1;
1990                 }
1991                 break;
1992         case MSR_IA32_DEBUGCTLMSR:
1993                 if (!data) {
1994                         /* We support the non-activated case already */
1995                         break;
1996                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1997                         /* Values other than LBR and BTF are vendor-specific,
1998                            thus reserved and should throw a #GP */
1999                         return 1;
2000                 }
2001                 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2002                             __func__, data);
2003                 break;
2004         case 0x200 ... 0x2ff:
2005                 return set_msr_mtrr(vcpu, msr, data);
2006         case MSR_IA32_APICBASE:
2007                 kvm_set_apic_base(vcpu, data);
2008                 break;
2009         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2010                 return kvm_x2apic_msr_write(vcpu, msr, data);
2011         case MSR_IA32_TSCDEADLINE:
2012                 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2013                 break;
2014         case MSR_IA32_TSC_ADJUST:
2015                 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2016                         if (!msr_info->host_initiated) {
2017                                 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2018                                 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2019                         }
2020                         vcpu->arch.ia32_tsc_adjust_msr = data;
2021                 }
2022                 break;
2023         case MSR_IA32_MISC_ENABLE:
2024                 vcpu->arch.ia32_misc_enable_msr = data;
2025                 break;
2026         case MSR_KVM_WALL_CLOCK_NEW:
2027         case MSR_KVM_WALL_CLOCK:
2028                 vcpu->kvm->arch.wall_clock = data;
2029                 kvm_write_wall_clock(vcpu->kvm, data);
2030                 break;
2031         case MSR_KVM_SYSTEM_TIME_NEW:
2032         case MSR_KVM_SYSTEM_TIME: {
2033                 u64 gpa_offset;
2034                 kvmclock_reset(vcpu);
2035
2036                 vcpu->arch.time = data;
2037                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2038
2039                 /* we verify if the enable bit is set... */
2040                 if (!(data & 1))
2041                         break;
2042
2043                 gpa_offset = data & ~(PAGE_MASK | 1);
2044
2045                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2046                      &vcpu->arch.pv_time, data & ~1ULL,
2047                      sizeof(struct pvclock_vcpu_time_info)))
2048                         vcpu->arch.pv_time_enabled = false;
2049                 else
2050                         vcpu->arch.pv_time_enabled = true;
2051
2052                 break;
2053         }
2054         case MSR_KVM_ASYNC_PF_EN:
2055                 if (kvm_pv_enable_async_pf(vcpu, data))
2056                         return 1;
2057                 break;
2058         case MSR_KVM_STEAL_TIME:
2059
2060                 if (unlikely(!sched_info_on()))
2061                         return 1;
2062
2063                 if (data & KVM_STEAL_RESERVED_MASK)
2064                         return 1;
2065
2066                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2067                                                 data & KVM_STEAL_VALID_BITS,
2068                                                 sizeof(struct kvm_steal_time)))
2069                         return 1;
2070
2071                 vcpu->arch.st.msr_val = data;
2072
2073                 if (!(data & KVM_MSR_ENABLED))
2074                         break;
2075
2076                 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2077
2078                 preempt_disable();
2079                 accumulate_steal_time(vcpu);
2080                 preempt_enable();
2081
2082                 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2083
2084                 break;
2085         case MSR_KVM_PV_EOI_EN:
2086                 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2087                         return 1;
2088                 break;
2089
2090         case MSR_IA32_MCG_CTL:
2091         case MSR_IA32_MCG_STATUS:
2092         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2093                 return set_msr_mce(vcpu, msr, data);
2094
2095         /* Performance counters are not protected by a CPUID bit,
2096          * so we should check all of them in the generic path for the sake of
2097          * cross vendor migration.
2098          * Writing a zero into the event select MSRs disables them,
2099          * which we perfectly emulate ;-). Any other value should be at least
2100          * reported, some guests depend on them.
2101          */
2102         case MSR_K7_EVNTSEL0:
2103         case MSR_K7_EVNTSEL1:
2104         case MSR_K7_EVNTSEL2:
2105         case MSR_K7_EVNTSEL3:
2106                 if (data != 0)
2107                         vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2108                                     "0x%x data 0x%llx\n", msr, data);
2109                 break;
2110         /* at least RHEL 4 unconditionally writes to the perfctr registers,
2111          * so we ignore writes to make it happy.
2112          */
2113         case MSR_K7_PERFCTR0:
2114         case MSR_K7_PERFCTR1:
2115         case MSR_K7_PERFCTR2:
2116         case MSR_K7_PERFCTR3:
2117                 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2118                             "0x%x data 0x%llx\n", msr, data);
2119                 break;
2120         case MSR_P6_PERFCTR0:
2121         case MSR_P6_PERFCTR1:
2122                 pr = true;
2123         case MSR_P6_EVNTSEL0:
2124         case MSR_P6_EVNTSEL1:
2125                 if (kvm_pmu_msr(vcpu, msr))
2126                         return kvm_pmu_set_msr(vcpu, msr_info);
2127
2128                 if (pr || data != 0)
2129                         vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2130                                     "0x%x data 0x%llx\n", msr, data);
2131                 break;
2132         case MSR_K7_CLK_CTL:
2133                 /*
2134                  * Ignore all writes to this no longer documented MSR.
2135                  * Writes are only relevant for old K7 processors,
2136                  * all pre-dating SVM, but a recommended workaround from
2137                  * AMD for these chips. It is possible to specify the
2138                  * affected processor models on the command line, hence
2139                  * the need to ignore the workaround.
2140                  */
2141                 break;
2142         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2143                 if (kvm_hv_msr_partition_wide(msr)) {
2144                         int r;
2145                         mutex_lock(&vcpu->kvm->lock);
2146                         r = set_msr_hyperv_pw(vcpu, msr, data);
2147                         mutex_unlock(&vcpu->kvm->lock);
2148                         return r;
2149                 } else
2150                         return set_msr_hyperv(vcpu, msr, data);
2151                 break;
2152         case MSR_IA32_BBL_CR_CTL3:
2153                 /* Drop writes to this legacy MSR -- see rdmsr
2154                  * counterpart for further detail.
2155                  */
2156                 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2157                 break;
2158         case MSR_AMD64_OSVW_ID_LENGTH:
2159                 if (!guest_cpuid_has_osvw(vcpu))
2160                         return 1;
2161                 vcpu->arch.osvw.length = data;
2162                 break;
2163         case MSR_AMD64_OSVW_STATUS:
2164                 if (!guest_cpuid_has_osvw(vcpu))
2165                         return 1;
2166                 vcpu->arch.osvw.status = data;
2167                 break;
2168         default:
2169                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2170                         return xen_hvm_config(vcpu, data);
2171                 if (kvm_pmu_msr(vcpu, msr))
2172                         return kvm_pmu_set_msr(vcpu, msr_info);
2173                 if (!ignore_msrs) {
2174                         vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2175                                     msr, data);
2176                         return 1;
2177                 } else {
2178                         vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2179                                     msr, data);
2180                         break;
2181                 }
2182         }
2183         return 0;
2184 }
2185 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2186
2187
2188 /*
2189  * Reads an msr value (of 'msr_index') into 'pdata'.
2190  * Returns 0 on success, non-0 otherwise.
2191  * Assumes vcpu_load() was already called.
2192  */
2193 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2194 {
2195         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2196 }
2197
2198 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2199 {
2200         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2201
2202         if (!msr_mtrr_valid(msr))
2203                 return 1;
2204
2205         if (msr == MSR_MTRRdefType)
2206                 *pdata = vcpu->arch.mtrr_state.def_type +
2207                          (vcpu->arch.mtrr_state.enabled << 10);
2208         else if (msr == MSR_MTRRfix64K_00000)
2209                 *pdata = p[0];
2210         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2211                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2212         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2213                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2214         else if (msr == MSR_IA32_CR_PAT)
2215                 *pdata = vcpu->arch.pat;
2216         else {  /* Variable MTRRs */
2217                 int idx, is_mtrr_mask;
2218                 u64 *pt;
2219
2220                 idx = (msr - 0x200) / 2;
2221                 is_mtrr_mask = msr - 0x200 - 2 * idx;
2222                 if (!is_mtrr_mask)
2223                         pt =
2224                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2225                 else
2226                         pt =
2227                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2228                 *pdata = *pt;
2229         }
2230
2231         return 0;
2232 }
2233
2234 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2235 {
2236         u64 data;
2237         u64 mcg_cap = vcpu->arch.mcg_cap;
2238         unsigned bank_num = mcg_cap & 0xff;
2239
2240         switch (msr) {
2241         case MSR_IA32_P5_MC_ADDR:
2242         case MSR_IA32_P5_MC_TYPE:
2243                 data = 0;
2244                 break;
2245         case MSR_IA32_MCG_CAP:
2246                 data = vcpu->arch.mcg_cap;
2247                 break;
2248         case MSR_IA32_MCG_CTL:
2249                 if (!(mcg_cap & MCG_CTL_P))
2250                         return 1;
2251                 data = vcpu->arch.mcg_ctl;
2252                 break;
2253         case MSR_IA32_MCG_STATUS:
2254                 data = vcpu->arch.mcg_status;
2255                 break;
2256         default:
2257                 if (msr >= MSR_IA32_MC0_CTL &&
2258                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2259                         u32 offset = msr - MSR_IA32_MC0_CTL;
2260                         data = vcpu->arch.mce_banks[offset];
2261                         break;
2262                 }
2263                 return 1;
2264         }
2265         *pdata = data;
2266         return 0;
2267 }
2268
2269 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2270 {
2271         u64 data = 0;
2272         struct kvm *kvm = vcpu->kvm;
2273
2274         switch (msr) {
2275         case HV_X64_MSR_GUEST_OS_ID:
2276                 data = kvm->arch.hv_guest_os_id;
2277                 break;
2278         case HV_X64_MSR_HYPERCALL:
2279                 data = kvm->arch.hv_hypercall;
2280                 break;
2281         default:
2282                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2283                 return 1;
2284         }
2285
2286         *pdata = data;
2287         return 0;
2288 }
2289
2290 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2291 {
2292         u64 data = 0;
2293
2294         switch (msr) {
2295         case HV_X64_MSR_VP_INDEX: {
2296                 int r;
2297                 struct kvm_vcpu *v;
2298                 kvm_for_each_vcpu(r, v, vcpu->kvm)
2299                         if (v == vcpu)
2300                                 data = r;
2301                 break;
2302         }
2303         case HV_X64_MSR_EOI:
2304                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2305         case HV_X64_MSR_ICR:
2306                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2307         case HV_X64_MSR_TPR:
2308                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2309         case HV_X64_MSR_APIC_ASSIST_PAGE:
2310                 data = vcpu->arch.hv_vapic;
2311                 break;
2312         default:
2313                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2314                 return 1;
2315         }
2316         *pdata = data;
2317         return 0;
2318 }
2319
2320 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2321 {
2322         u64 data;
2323
2324         switch (msr) {
2325         case MSR_IA32_PLATFORM_ID:
2326         case MSR_IA32_EBL_CR_POWERON:
2327         case MSR_IA32_DEBUGCTLMSR:
2328         case MSR_IA32_LASTBRANCHFROMIP:
2329         case MSR_IA32_LASTBRANCHTOIP:
2330         case MSR_IA32_LASTINTFROMIP:
2331         case MSR_IA32_LASTINTTOIP:
2332         case MSR_K8_SYSCFG:
2333         case MSR_K7_HWCR:
2334         case MSR_VM_HSAVE_PA:
2335         case MSR_K7_EVNTSEL0:
2336         case MSR_K7_PERFCTR0:
2337         case MSR_K8_INT_PENDING_MSG:
2338         case MSR_AMD64_NB_CFG:
2339         case MSR_FAM10H_MMIO_CONF_BASE:
2340         case MSR_AMD64_BU_CFG2:
2341                 data = 0;
2342                 break;
2343         case MSR_P6_PERFCTR0:
2344         case MSR_P6_PERFCTR1:
2345         case MSR_P6_EVNTSEL0:
2346         case MSR_P6_EVNTSEL1:
2347                 if (kvm_pmu_msr(vcpu, msr))
2348                         return kvm_pmu_get_msr(vcpu, msr, pdata);
2349                 data = 0;
2350                 break;
2351         case MSR_IA32_UCODE_REV:
2352                 data = 0x100000000ULL;
2353                 break;
2354         case MSR_MTRRcap:
2355                 data = 0x500 | KVM_NR_VAR_MTRR;
2356                 break;
2357         case 0x200 ... 0x2ff:
2358                 return get_msr_mtrr(vcpu, msr, pdata);
2359         case 0xcd: /* fsb frequency */
2360                 data = 3;
2361                 break;
2362                 /*
2363                  * MSR_EBC_FREQUENCY_ID
2364                  * Conservative value valid for even the basic CPU models.
2365                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2366                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2367                  * and 266MHz for model 3, or 4. Set Core Clock
2368                  * Frequency to System Bus Frequency Ratio to 1 (bits
2369                  * 31:24) even though these are only valid for CPU
2370                  * models > 2, however guests may end up dividing or
2371                  * multiplying by zero otherwise.
2372                  */
2373         case MSR_EBC_FREQUENCY_ID:
2374                 data = 1 << 24;
2375                 break;
2376         case MSR_IA32_APICBASE:
2377                 data = kvm_get_apic_base(vcpu);
2378                 break;
2379         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2380                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2381                 break;
2382         case MSR_IA32_TSCDEADLINE:
2383                 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2384                 break;
2385         case MSR_IA32_TSC_ADJUST:
2386                 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2387                 break;
2388         case MSR_IA32_MISC_ENABLE:
2389                 data = vcpu->arch.ia32_misc_enable_msr;
2390                 break;
2391         case MSR_IA32_PERF_STATUS:
2392                 /* TSC increment by tick */
2393                 data = 1000ULL;
2394                 /* CPU multiplier */
2395                 data |= (((uint64_t)4ULL) << 40);
2396                 break;
2397         case MSR_EFER:
2398                 data = vcpu->arch.efer;
2399                 break;
2400         case MSR_KVM_WALL_CLOCK:
2401         case MSR_KVM_WALL_CLOCK_NEW:
2402                 data = vcpu->kvm->arch.wall_clock;
2403                 break;
2404         case MSR_KVM_SYSTEM_TIME:
2405         case MSR_KVM_SYSTEM_TIME_NEW:
2406                 data = vcpu->arch.time;
2407                 break;
2408         case MSR_KVM_ASYNC_PF_EN:
2409                 data = vcpu->arch.apf.msr_val;
2410                 break;
2411         case MSR_KVM_STEAL_TIME:
2412                 data = vcpu->arch.st.msr_val;
2413                 break;
2414         case MSR_KVM_PV_EOI_EN:
2415                 data = vcpu->arch.pv_eoi.msr_val;
2416                 break;
2417         case MSR_IA32_P5_MC_ADDR:
2418         case MSR_IA32_P5_MC_TYPE:
2419         case MSR_IA32_MCG_CAP:
2420         case MSR_IA32_MCG_CTL:
2421         case MSR_IA32_MCG_STATUS:
2422         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2423                 return get_msr_mce(vcpu, msr, pdata);
2424         case MSR_K7_CLK_CTL:
2425                 /*
2426                  * Provide expected ramp-up count for K7. All other
2427                  * are set to zero, indicating minimum divisors for
2428                  * every field.
2429                  *
2430                  * This prevents guest kernels on AMD host with CPU
2431                  * type 6, model 8 and higher from exploding due to
2432                  * the rdmsr failing.
2433                  */
2434                 data = 0x20000000;
2435                 break;
2436         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2437                 if (kvm_hv_msr_partition_wide(msr)) {
2438                         int r;
2439                         mutex_lock(&vcpu->kvm->lock);
2440                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
2441                         mutex_unlock(&vcpu->kvm->lock);
2442                         return r;
2443                 } else
2444                         return get_msr_hyperv(vcpu, msr, pdata);
2445                 break;
2446         case MSR_IA32_BBL_CR_CTL3:
2447                 /* This legacy MSR exists but isn't fully documented in current
2448                  * silicon.  It is however accessed by winxp in very narrow
2449                  * scenarios where it sets bit #19, itself documented as
2450                  * a "reserved" bit.  Best effort attempt to source coherent
2451                  * read data here should the balance of the register be
2452                  * interpreted by the guest:
2453                  *
2454                  * L2 cache control register 3: 64GB range, 256KB size,
2455                  * enabled, latency 0x1, configured
2456                  */
2457                 data = 0xbe702111;
2458                 break;
2459         case MSR_AMD64_OSVW_ID_LENGTH:
2460                 if (!guest_cpuid_has_osvw(vcpu))
2461                         return 1;
2462                 data = vcpu->arch.osvw.length;
2463                 break;
2464         case MSR_AMD64_OSVW_STATUS:
2465                 if (!guest_cpuid_has_osvw(vcpu))
2466                         return 1;
2467                 data = vcpu->arch.osvw.status;
2468                 break;
2469         default:
2470                 if (kvm_pmu_msr(vcpu, msr))
2471                         return kvm_pmu_get_msr(vcpu, msr, pdata);
2472                 if (!ignore_msrs) {
2473                         vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2474                         return 1;
2475                 } else {
2476                         vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2477                         data = 0;
2478                 }
2479                 break;
2480         }
2481         *pdata = data;
2482         return 0;
2483 }
2484 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2485
2486 /*
2487  * Read or write a bunch of msrs. All parameters are kernel addresses.
2488  *
2489  * @return number of msrs set successfully.
2490  */
2491 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2492                     struct kvm_msr_entry *entries,
2493                     int (*do_msr)(struct kvm_vcpu *vcpu,
2494                                   unsigned index, u64 *data))
2495 {
2496         int i, idx;
2497
2498         idx = srcu_read_lock(&vcpu->kvm->srcu);
2499         for (i = 0; i < msrs->nmsrs; ++i)
2500                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2501                         break;
2502         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2503
2504         return i;
2505 }
2506
2507 /*
2508  * Read or write a bunch of msrs. Parameters are user addresses.
2509  *
2510  * @return number of msrs set successfully.
2511  */
2512 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2513                   int (*do_msr)(struct kvm_vcpu *vcpu,
2514                                 unsigned index, u64 *data),
2515                   int writeback)
2516 {
2517         struct kvm_msrs msrs;
2518         struct kvm_msr_entry *entries;
2519         int r, n;
2520         unsigned size;
2521
2522         r = -EFAULT;
2523         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2524                 goto out;
2525
2526         r = -E2BIG;
2527         if (msrs.nmsrs >= MAX_IO_MSRS)
2528                 goto out;
2529
2530         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2531         entries = memdup_user(user_msrs->entries, size);
2532         if (IS_ERR(entries)) {
2533                 r = PTR_ERR(entries);
2534                 goto out;
2535         }
2536
2537         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2538         if (r < 0)
2539                 goto out_free;
2540
2541         r = -EFAULT;
2542         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2543                 goto out_free;
2544
2545         r = n;
2546
2547 out_free:
2548         kfree(entries);
2549 out:
2550         return r;
2551 }
2552
2553 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2554 {
2555         int r;
2556
2557         switch (ext) {
2558         case KVM_CAP_IRQCHIP:
2559         case KVM_CAP_HLT:
2560         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2561         case KVM_CAP_SET_TSS_ADDR:
2562         case KVM_CAP_EXT_CPUID:
2563         case KVM_CAP_EXT_EMUL_CPUID:
2564         case KVM_CAP_CLOCKSOURCE:
2565         case KVM_CAP_PIT:
2566         case KVM_CAP_NOP_IO_DELAY:
2567         case KVM_CAP_MP_STATE:
2568         case KVM_CAP_SYNC_MMU:
2569         case KVM_CAP_USER_NMI:
2570         case KVM_CAP_REINJECT_CONTROL:
2571         case KVM_CAP_IRQ_INJECT_STATUS:
2572         case KVM_CAP_IRQFD:
2573         case KVM_CAP_IOEVENTFD:
2574         case KVM_CAP_PIT2:
2575         case KVM_CAP_PIT_STATE2:
2576         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2577         case KVM_CAP_XEN_HVM:
2578         case KVM_CAP_ADJUST_CLOCK:
2579         case KVM_CAP_VCPU_EVENTS:
2580         case KVM_CAP_HYPERV:
2581         case KVM_CAP_HYPERV_VAPIC:
2582         case KVM_CAP_HYPERV_SPIN:
2583         case KVM_CAP_PCI_SEGMENT:
2584         case KVM_CAP_DEBUGREGS:
2585         case KVM_CAP_X86_ROBUST_SINGLESTEP:
2586         case KVM_CAP_XSAVE:
2587         case KVM_CAP_ASYNC_PF:
2588         case KVM_CAP_GET_TSC_KHZ:
2589         case KVM_CAP_KVMCLOCK_CTRL:
2590         case KVM_CAP_READONLY_MEM:
2591 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2592         case KVM_CAP_ASSIGN_DEV_IRQ:
2593         case KVM_CAP_PCI_2_3:
2594 #endif
2595                 r = 1;
2596                 break;
2597         case KVM_CAP_COALESCED_MMIO:
2598                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2599                 break;
2600         case KVM_CAP_VAPIC:
2601                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2602                 break;
2603         case KVM_CAP_NR_VCPUS:
2604                 r = KVM_SOFT_MAX_VCPUS;
2605                 break;
2606         case KVM_CAP_MAX_VCPUS:
2607                 r = KVM_MAX_VCPUS;
2608                 break;
2609         case KVM_CAP_NR_MEMSLOTS:
2610                 r = KVM_USER_MEM_SLOTS;
2611                 break;
2612         case KVM_CAP_PV_MMU:    /* obsolete */
2613                 r = 0;
2614                 break;
2615 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2616         case KVM_CAP_IOMMU:
2617                 r = iommu_present(&pci_bus_type);
2618                 break;
2619 #endif
2620         case KVM_CAP_MCE:
2621                 r = KVM_MAX_MCE_BANKS;
2622                 break;
2623         case KVM_CAP_XCRS:
2624                 r = cpu_has_xsave;
2625                 break;
2626         case KVM_CAP_TSC_CONTROL:
2627                 r = kvm_has_tsc_control;
2628                 break;
2629         case KVM_CAP_TSC_DEADLINE_TIMER:
2630                 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2631                 break;
2632         default:
2633                 r = 0;
2634                 break;
2635         }
2636         return r;
2637
2638 }
2639
2640 long kvm_arch_dev_ioctl(struct file *filp,
2641                         unsigned int ioctl, unsigned long arg)
2642 {
2643         void __user *argp = (void __user *)arg;
2644         long r;
2645
2646         switch (ioctl) {
2647         case KVM_GET_MSR_INDEX_LIST: {
2648                 struct kvm_msr_list __user *user_msr_list = argp;
2649                 struct kvm_msr_list msr_list;
2650                 unsigned n;
2651
2652                 r = -EFAULT;
2653                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2654                         goto out;
2655                 n = msr_list.nmsrs;
2656                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2657                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2658                         goto out;
2659                 r = -E2BIG;
2660                 if (n < msr_list.nmsrs)
2661                         goto out;
2662                 r = -EFAULT;
2663                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2664                                  num_msrs_to_save * sizeof(u32)))
2665                         goto out;
2666                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2667                                  &emulated_msrs,
2668                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2669                         goto out;
2670                 r = 0;
2671                 break;
2672         }
2673         case KVM_GET_SUPPORTED_CPUID:
2674         case KVM_GET_EMULATED_CPUID: {
2675                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2676                 struct kvm_cpuid2 cpuid;
2677
2678                 r = -EFAULT;
2679                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2680                         goto out;
2681
2682                 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2683                                             ioctl);
2684                 if (r)
2685                         goto out;
2686
2687                 r = -EFAULT;
2688                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2689                         goto out;
2690                 r = 0;
2691                 break;
2692         }
2693         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2694                 u64 mce_cap;
2695
2696                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2697                 r = -EFAULT;
2698                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2699                         goto out;
2700                 r = 0;
2701                 break;
2702         }
2703         default:
2704                 r = -EINVAL;
2705         }
2706 out:
2707         return r;
2708 }
2709
2710 static void wbinvd_ipi(void *garbage)
2711 {
2712         wbinvd();
2713 }
2714
2715 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2716 {
2717         return vcpu->kvm->arch.iommu_domain &&
2718                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2719 }
2720
2721 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2722 {
2723         /* Address WBINVD may be executed by guest */
2724         if (need_emulate_wbinvd(vcpu)) {
2725                 if (kvm_x86_ops->has_wbinvd_exit())
2726                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2727                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2728                         smp_call_function_single(vcpu->cpu,
2729                                         wbinvd_ipi, NULL, 1);
2730         }
2731
2732         kvm_x86_ops->vcpu_load(vcpu, cpu);
2733
2734         /* Apply any externally detected TSC adjustments (due to suspend) */
2735         if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2736                 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2737                 vcpu->arch.tsc_offset_adjustment = 0;
2738                 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2739         }
2740
2741         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2742                 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2743                                 native_read_tsc() - vcpu->arch.last_host_tsc;
2744                 if (tsc_delta < 0)
2745                         mark_tsc_unstable("KVM discovered backwards TSC");
2746                 if (check_tsc_unstable()) {
2747                         u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2748                                                 vcpu->arch.last_guest_tsc);
2749                         kvm_x86_ops->write_tsc_offset(vcpu, offset);
2750                         vcpu->arch.tsc_catchup = 1;
2751                 }
2752                 /*
2753                  * On a host with synchronized TSC, there is no need to update
2754                  * kvmclock on vcpu->cpu migration
2755                  */
2756                 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2757                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2758                 if (vcpu->cpu != cpu)
2759                         kvm_migrate_timers(vcpu);
2760                 vcpu->cpu = cpu;
2761         }
2762
2763         accumulate_steal_time(vcpu);
2764         kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2765 }
2766
2767 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2768 {
2769         kvm_x86_ops->vcpu_put(vcpu);
2770         kvm_put_guest_fpu(vcpu);
2771         vcpu->arch.last_host_tsc = native_read_tsc();
2772 }
2773
2774 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2775                                     struct kvm_lapic_state *s)
2776 {
2777         kvm_x86_ops->sync_pir_to_irr(vcpu);
2778         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2779
2780         return 0;
2781 }
2782
2783 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2784                                     struct kvm_lapic_state *s)
2785 {
2786         kvm_apic_post_state_restore(vcpu, s);
2787         update_cr8_intercept(vcpu);
2788
2789         return 0;
2790 }
2791
2792 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2793                                     struct kvm_interrupt *irq)
2794 {
2795         if (irq->irq >= KVM_NR_INTERRUPTS)
2796                 return -EINVAL;
2797         if (irqchip_in_kernel(vcpu->kvm))
2798                 return -ENXIO;
2799
2800         kvm_queue_interrupt(vcpu, irq->irq, false);
2801         kvm_make_request(KVM_REQ_EVENT, vcpu);
2802
2803         return 0;
2804 }
2805
2806 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2807 {
2808         kvm_inject_nmi(vcpu);
2809
2810         return 0;
2811 }
2812
2813 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2814                                            struct kvm_tpr_access_ctl *tac)
2815 {
2816         if (tac->flags)
2817                 return -EINVAL;
2818         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2819         return 0;
2820 }
2821
2822 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2823                                         u64 mcg_cap)
2824 {
2825         int r;
2826         unsigned bank_num = mcg_cap & 0xff, bank;
2827
2828         r = -EINVAL;
2829         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2830                 goto out;
2831         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2832                 goto out;
2833         r = 0;
2834         vcpu->arch.mcg_cap = mcg_cap;
2835         /* Init IA32_MCG_CTL to all 1s */
2836         if (mcg_cap & MCG_CTL_P)
2837                 vcpu->arch.mcg_ctl = ~(u64)0;
2838         /* Init IA32_MCi_CTL to all 1s */
2839         for (bank = 0; bank < bank_num; bank++)
2840                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2841 out:
2842         return r;
2843 }
2844
2845 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2846                                       struct kvm_x86_mce *mce)
2847 {
2848         u64 mcg_cap = vcpu->arch.mcg_cap;
2849         unsigned bank_num = mcg_cap & 0xff;
2850         u64 *banks = vcpu->arch.mce_banks;
2851
2852         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2853                 return -EINVAL;
2854         /*
2855          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2856          * reporting is disabled
2857          */
2858         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2859             vcpu->arch.mcg_ctl != ~(u64)0)
2860                 return 0;
2861         banks += 4 * mce->bank;
2862         /*
2863          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2864          * reporting is disabled for the bank
2865          */
2866         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2867                 return 0;
2868         if (mce->status & MCI_STATUS_UC) {
2869                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2870                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2871                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2872                         return 0;
2873                 }
2874                 if (banks[1] & MCI_STATUS_VAL)
2875                         mce->status |= MCI_STATUS_OVER;
2876                 banks[2] = mce->addr;
2877                 banks[3] = mce->misc;
2878                 vcpu->arch.mcg_status = mce->mcg_status;
2879                 banks[1] = mce->status;
2880                 kvm_queue_exception(vcpu, MC_VECTOR);
2881         } else if (!(banks[1] & MCI_STATUS_VAL)
2882                    || !(banks[1] & MCI_STATUS_UC)) {
2883                 if (banks[1] & MCI_STATUS_VAL)
2884                         mce->status |= MCI_STATUS_OVER;
2885                 banks[2] = mce->addr;
2886                 banks[3] = mce->misc;
2887                 banks[1] = mce->status;
2888         } else
2889                 banks[1] |= MCI_STATUS_OVER;
2890         return 0;
2891 }
2892
2893 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2894                                                struct kvm_vcpu_events *events)
2895 {
2896         process_nmi(vcpu);
2897         events->exception.injected =
2898                 vcpu->arch.exception.pending &&
2899                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2900         events->exception.nr = vcpu->arch.exception.nr;
2901         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2902         events->exception.pad = 0;
2903         events->exception.error_code = vcpu->arch.exception.error_code;
2904
2905         events->interrupt.injected =
2906                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2907         events->interrupt.nr = vcpu->arch.interrupt.nr;
2908         events->interrupt.soft = 0;
2909         events->interrupt.shadow =
2910                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2911                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2912
2913         events->nmi.injected = vcpu->arch.nmi_injected;
2914         events->nmi.pending = vcpu->arch.nmi_pending != 0;
2915         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2916         events->nmi.pad = 0;
2917
2918         events->sipi_vector = 0; /* never valid when reporting to user space */
2919
2920         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2921                          | KVM_VCPUEVENT_VALID_SHADOW);
2922         memset(&events->reserved, 0, sizeof(events->reserved));
2923 }
2924
2925 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2926                                               struct kvm_vcpu_events *events)
2927 {
2928         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2929                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2930                               | KVM_VCPUEVENT_VALID_SHADOW))
2931                 return -EINVAL;
2932
2933         process_nmi(vcpu);
2934         vcpu->arch.exception.pending = events->exception.injected;
2935         vcpu->arch.exception.nr = events->exception.nr;
2936         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2937         vcpu->arch.exception.error_code = events->exception.error_code;
2938
2939         vcpu->arch.interrupt.pending = events->interrupt.injected;
2940         vcpu->arch.interrupt.nr = events->interrupt.nr;
2941         vcpu->arch.interrupt.soft = events->interrupt.soft;
2942         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2943                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2944                                                   events->interrupt.shadow);
2945
2946         vcpu->arch.nmi_injected = events->nmi.injected;
2947         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2948                 vcpu->arch.nmi_pending = events->nmi.pending;
2949         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2950
2951         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2952             kvm_vcpu_has_lapic(vcpu))
2953                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2954
2955         kvm_make_request(KVM_REQ_EVENT, vcpu);
2956
2957         return 0;
2958 }
2959
2960 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2961                                              struct kvm_debugregs *dbgregs)
2962 {
2963         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2964         dbgregs->dr6 = vcpu->arch.dr6;
2965         dbgregs->dr7 = vcpu->arch.dr7;
2966         dbgregs->flags = 0;
2967         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2968 }
2969
2970 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2971                                             struct kvm_debugregs *dbgregs)
2972 {
2973         if (dbgregs->flags)
2974                 return -EINVAL;
2975
2976         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2977         vcpu->arch.dr6 = dbgregs->dr6;
2978         vcpu->arch.dr7 = dbgregs->dr7;
2979
2980         return 0;
2981 }
2982
2983 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2984                                          struct kvm_xsave *guest_xsave)
2985 {
2986         if (cpu_has_xsave)
2987                 memcpy(guest_xsave->region,
2988                         &vcpu->arch.guest_fpu.state->xsave,
2989                         xstate_size);
2990         else {
2991                 memcpy(guest_xsave->region,
2992                         &vcpu->arch.guest_fpu.state->fxsave,
2993                         sizeof(struct i387_fxsave_struct));
2994                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2995                         XSTATE_FPSSE;
2996         }
2997 }
2998
2999 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3000                                         struct kvm_xsave *guest_xsave)
3001 {
3002         u64 xstate_bv =
3003                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3004
3005         if (cpu_has_xsave)
3006                 memcpy(&vcpu->arch.guest_fpu.state->xsave,
3007                         guest_xsave->region, xstate_size);
3008         else {
3009                 if (xstate_bv & ~XSTATE_FPSSE)
3010                         return -EINVAL;
3011                 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3012                         guest_xsave->region, sizeof(struct i387_fxsave_struct));
3013         }
3014         return 0;
3015 }
3016
3017 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3018                                         struct kvm_xcrs *guest_xcrs)
3019 {
3020         if (!cpu_has_xsave) {
3021                 guest_xcrs->nr_xcrs = 0;
3022                 return;
3023         }
3024
3025         guest_xcrs->nr_xcrs = 1;
3026         guest_xcrs->flags = 0;
3027         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3028         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3029 }
3030
3031 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3032                                        struct kvm_xcrs *guest_xcrs)
3033 {
3034         int i, r = 0;
3035
3036         if (!cpu_has_xsave)
3037                 return -EINVAL;
3038
3039         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3040                 return -EINVAL;
3041
3042         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3043                 /* Only support XCR0 currently */
3044                 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
3045                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3046                                 guest_xcrs->xcrs[0].value);
3047                         break;
3048                 }
3049         if (r)
3050                 r = -EINVAL;
3051         return r;
3052 }
3053
3054 /*
3055  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3056  * stopped by the hypervisor.  This function will be called from the host only.
3057  * EINVAL is returned when the host attempts to set the flag for a guest that
3058  * does not support pv clocks.
3059  */
3060 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3061 {
3062         if (!vcpu->arch.pv_time_enabled)
3063                 return -EINVAL;
3064         vcpu->arch.pvclock_set_guest_stopped_request = true;
3065         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3066         return 0;
3067 }
3068
3069 long kvm_arch_vcpu_ioctl(struct file *filp,
3070                          unsigned int ioctl, unsigned long arg)
3071 {
3072         struct kvm_vcpu *vcpu = filp->private_data;
3073         void __user *argp = (void __user *)arg;
3074         int r;
3075         union {
3076                 struct kvm_lapic_state *lapic;
3077                 struct kvm_xsave *xsave;
3078                 struct kvm_xcrs *xcrs;
3079                 void *buffer;
3080         } u;
3081
3082         u.buffer = NULL;
3083         switch (ioctl) {
3084         case KVM_GET_LAPIC: {
3085                 r = -EINVAL;
3086                 if (!vcpu->arch.apic)
3087                         goto out;
3088                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3089
3090                 r = -ENOMEM;
3091                 if (!u.lapic)
3092                         goto out;
3093                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3094                 if (r)
3095                         goto out;
3096                 r = -EFAULT;
3097                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3098                         goto out;
3099                 r = 0;
3100                 break;
3101         }
3102         case KVM_SET_LAPIC: {
3103                 r = -EINVAL;
3104                 if (!vcpu->arch.apic)
3105                         goto out;
3106                 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3107                 if (IS_ERR(u.lapic))
3108                         return PTR_ERR(u.lapic);
3109
3110                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3111                 break;
3112         }
3113         case KVM_INTERRUPT: {
3114                 struct kvm_interrupt irq;
3115
3116                 r = -EFAULT;
3117                 if (copy_from_user(&irq, argp, sizeof irq))
3118                         goto out;
3119                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3120                 break;
3121         }
3122         case KVM_NMI: {
3123                 r = kvm_vcpu_ioctl_nmi(vcpu);
3124                 break;
3125         }
3126         case KVM_SET_CPUID: {
3127                 struct kvm_cpuid __user *cpuid_arg = argp;
3128                 struct kvm_cpuid cpuid;
3129
3130                 r = -EFAULT;
3131                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3132                         goto out;
3133                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3134                 break;
3135         }
3136         case KVM_SET_CPUID2: {
3137                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3138                 struct kvm_cpuid2 cpuid;
3139
3140                 r = -EFAULT;
3141                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3142                         goto out;
3143                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3144                                               cpuid_arg->entries);
3145                 break;
3146         }
3147         case KVM_GET_CPUID2: {
3148                 struct kvm_cpuid2 __user *cpuid_arg = argp;
3149                 struct kvm_cpuid2 cpuid;
3150
3151                 r = -EFAULT;
3152                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3153                         goto out;
3154                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3155                                               cpuid_arg->entries);
3156                 if (r)
3157                         goto out;
3158                 r = -EFAULT;
3159                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3160                         goto out;
3161                 r = 0;
3162                 break;
3163         }
3164         case KVM_GET_MSRS:
3165                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3166                 break;
3167         case KVM_SET_MSRS:
3168                 r = msr_io(vcpu, argp, do_set_msr, 0);
3169                 break;
3170         case KVM_TPR_ACCESS_REPORTING: {
3171                 struct kvm_tpr_access_ctl tac;
3172
3173                 r = -EFAULT;
3174                 if (copy_from_user(&tac, argp, sizeof tac))
3175                         goto out;
3176                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3177                 if (r)
3178                         goto out;
3179                 r = -EFAULT;
3180                 if (copy_to_user(argp, &tac, sizeof tac))
3181                         goto out;
3182                 r = 0;
3183                 break;
3184         };
3185         case KVM_SET_VAPIC_ADDR: {
3186                 struct kvm_vapic_addr va;
3187
3188                 r = -EINVAL;
3189                 if (!irqchip_in_kernel(vcpu->kvm))
3190                         goto out;
3191                 r = -EFAULT;
3192                 if (copy_from_user(&va, argp, sizeof va))
3193                         goto out;
3194                 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3195                 break;
3196         }
3197         case KVM_X86_SETUP_MCE: {
3198                 u64 mcg_cap;
3199
3200                 r = -EFAULT;
3201                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3202                         goto out;
3203                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3204                 break;
3205         }
3206         case KVM_X86_SET_MCE: {
3207                 struct kvm_x86_mce mce;
3208
3209                 r = -EFAULT;
3210                 if (copy_from_user(&mce, argp, sizeof mce))
3211                         goto out;
3212                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3213                 break;
3214         }
3215         case KVM_GET_VCPU_EVENTS: {
3216                 struct kvm_vcpu_events events;
3217
3218                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3219
3220                 r = -EFAULT;
3221                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3222                         break;
3223                 r = 0;
3224                 break;
3225         }
3226         case KVM_SET_VCPU_EVENTS: {
3227                 struct kvm_vcpu_events events;
3228
3229                 r = -EFAULT;
3230                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3231                         break;
3232
3233                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3234                 break;
3235         }
3236         case KVM_GET_DEBUGREGS: {
3237                 struct kvm_debugregs dbgregs;
3238
3239                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3240
3241                 r = -EFAULT;
3242                 if (copy_to_user(argp, &dbgregs,
3243                                  sizeof(struct kvm_debugregs)))
3244                         break;
3245                 r = 0;
3246                 break;
3247         }
3248         case KVM_SET_DEBUGREGS: {
3249                 struct kvm_debugregs dbgregs;
3250
3251                 r = -EFAULT;
3252                 if (copy_from_user(&dbgregs, argp,
3253                                    sizeof(struct kvm_debugregs)))
3254                         break;
3255
3256                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3257                 break;
3258         }
3259         case KVM_GET_XSAVE: {
3260                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3261                 r = -ENOMEM;
3262                 if (!u.xsave)
3263                         break;
3264
3265                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3266
3267                 r = -EFAULT;
3268                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3269                         break;
3270                 r = 0;
3271                 break;
3272         }
3273         case KVM_SET_XSAVE: {
3274                 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3275                 if (IS_ERR(u.xsave))
3276                         return PTR_ERR(u.xsave);
3277
3278                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3279                 break;
3280         }
3281         case KVM_GET_XCRS: {
3282                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3283                 r = -ENOMEM;
3284                 if (!u.xcrs)
3285                         break;
3286
3287                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3288
3289                 r = -EFAULT;
3290                 if (copy_to_user(argp, u.xcrs,
3291                                  sizeof(struct kvm_xcrs)))
3292                         break;
3293                 r = 0;
3294                 break;
3295         }
3296         case KVM_SET_XCRS: {
3297                 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3298                 if (IS_ERR(u.xcrs))
3299                         return PTR_ERR(u.xcrs);
3300
3301                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3302                 break;
3303         }
3304         case KVM_SET_TSC_KHZ: {
3305                 u32 user_tsc_khz;
3306
3307                 r = -EINVAL;
3308                 user_tsc_khz = (u32)arg;
3309
3310                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3311                         goto out;
3312
3313                 if (user_tsc_khz == 0)
3314                         user_tsc_khz = tsc_khz;
3315
3316                 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3317
3318                 r = 0;
3319                 goto out;
3320         }
3321         case KVM_GET_TSC_KHZ: {
3322                 r = vcpu->arch.virtual_tsc_khz;
3323                 goto out;
3324         }
3325         case KVM_KVMCLOCK_CTRL: {
3326                 r = kvm_set_guest_paused(vcpu);
3327                 goto out;
3328         }
3329         default:
3330                 r = -EINVAL;
3331         }
3332 out:
3333         kfree(u.buffer);
3334         return r;
3335 }
3336
3337 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3338 {
3339         return VM_FAULT_SIGBUS;
3340 }
3341
3342 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3343 {
3344         int ret;
3345
3346         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3347                 return -EINVAL;
3348         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3349         return ret;
3350 }
3351
3352 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3353                                               u64 ident_addr)
3354 {
3355         kvm->arch.ept_identity_map_addr = ident_addr;
3356         return 0;
3357 }
3358
3359 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3360                                           u32 kvm_nr_mmu_pages)
3361 {
3362         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3363                 return -EINVAL;
3364
3365         mutex_lock(&kvm->slots_lock);
3366
3367         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3368         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3369
3370         mutex_unlock(&kvm->slots_lock);
3371         return 0;
3372 }
3373
3374 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3375 {
3376         return kvm->arch.n_max_mmu_pages;
3377 }
3378
3379 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3380 {
3381         int r;
3382
3383         r = 0;
3384         switch (chip->chip_id) {
3385         case KVM_IRQCHIP_PIC_MASTER:
3386                 memcpy(&chip->chip.pic,
3387                         &pic_irqchip(kvm)->pics[0],
3388                         sizeof(struct kvm_pic_state));
3389                 break;
3390         case KVM_IRQCHIP_PIC_SLAVE:
3391                 memcpy(&chip->chip.pic,
3392                         &pic_irqchip(kvm)->pics[1],
3393                         sizeof(struct kvm_pic_state));
3394                 break;
3395         case KVM_IRQCHIP_IOAPIC:
3396                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3397                 break;
3398         default:
3399                 r = -EINVAL;
3400                 break;
3401         }
3402         return r;
3403 }
3404
3405 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3406 {
3407         int r;
3408
3409         r = 0;
3410         switch (chip->chip_id) {
3411         case KVM_IRQCHIP_PIC_MASTER:
3412                 spin_lock(&pic_irqchip(kvm)->lock);
3413                 memcpy(&pic_irqchip(kvm)->pics[0],
3414                         &chip->chip.pic,
3415                         sizeof(struct kvm_pic_state));
3416                 spin_unlock(&pic_irqchip(kvm)->lock);
3417                 break;
3418         case KVM_IRQCHIP_PIC_SLAVE:
3419                 spin_lock(&pic_irqchip(kvm)->lock);
3420                 memcpy(&pic_irqchip(kvm)->pics[1],
3421                         &chip->chip.pic,
3422                         sizeof(struct kvm_pic_state));
3423                 spin_unlock(&pic_irqchip(kvm)->lock);
3424                 break;
3425         case KVM_IRQCHIP_IOAPIC:
3426                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3427                 break;
3428         default:
3429                 r = -EINVAL;
3430                 break;
3431         }
3432         kvm_pic_update_irq(pic_irqchip(kvm));
3433         return r;
3434 }
3435
3436 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3437 {
3438         int r = 0;
3439
3440         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3441         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3442         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3443         return r;
3444 }
3445
3446 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3447 {
3448         int r = 0;
3449
3450         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3451         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3452         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3453         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3454         return r;
3455 }
3456
3457 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3458 {
3459         int r = 0;
3460
3461         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3462         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3463                 sizeof(ps->channels));
3464         ps->flags = kvm->arch.vpit->pit_state.flags;
3465         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3466         memset(&ps->reserved, 0, sizeof(ps->reserved));
3467         return r;
3468 }
3469
3470 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3471 {
3472         int r = 0, start = 0;
3473         u32 prev_legacy, cur_legacy;
3474         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3475         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3476         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3477         if (!prev_legacy && cur_legacy)
3478                 start = 1;
3479         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3480                sizeof(kvm->arch.vpit->pit_state.channels));
3481         kvm->arch.vpit->pit_state.flags = ps->flags;
3482         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3483         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3484         return r;
3485 }
3486
3487 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3488                                  struct kvm_reinject_control *control)
3489 {
3490         if (!kvm->arch.vpit)
3491                 return -ENXIO;
3492         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3493         kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3494         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3495         return 0;
3496 }
3497
3498 /**
3499  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3500  * @kvm: kvm instance
3501  * @log: slot id and address to which we copy the log
3502  *
3503  * We need to keep it in mind that VCPU threads can write to the bitmap
3504  * concurrently.  So, to avoid losing data, we keep the following order for
3505  * each bit:
3506  *
3507  *   1. Take a snapshot of the bit and clear it if needed.
3508  *   2. Write protect the corresponding page.
3509  *   3. Flush TLB's if needed.
3510  *   4. Copy the snapshot to the userspace.
3511  *
3512  * Between 2 and 3, the guest may write to the page using the remaining TLB
3513  * entry.  This is not a problem because the page will be reported dirty at
3514  * step 4 using the snapshot taken before and step 3 ensures that successive
3515  * writes will be logged for the next call.
3516  */
3517 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3518 {
3519         int r;
3520         struct kvm_memory_slot *memslot;
3521         unsigned long n, i;
3522         unsigned long *dirty_bitmap;
3523         unsigned long *dirty_bitmap_buffer;
3524         bool is_dirty = false;
3525
3526         mutex_lock(&kvm->slots_lock);
3527
3528         r = -EINVAL;
3529         if (log->slot >= KVM_USER_MEM_SLOTS)
3530                 goto out;
3531
3532         memslot = id_to_memslot(kvm->memslots, log->slot);
3533
3534         dirty_bitmap = memslot->dirty_bitmap;
3535         r = -ENOENT;
3536         if (!dirty_bitmap)
3537                 goto out;
3538
3539         n = kvm_dirty_bitmap_bytes(memslot);
3540
3541         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3542         memset(dirty_bitmap_buffer, 0, n);
3543
3544         spin_lock(&kvm->mmu_lock);
3545
3546         for (i = 0; i < n / sizeof(long); i++) {
3547                 unsigned long mask;
3548                 gfn_t offset;
3549
3550                 if (!dirty_bitmap[i])
3551                         continue;
3552
3553                 is_dirty = true;
3554
3555                 mask = xchg(&dirty_bitmap[i], 0);
3556                 dirty_bitmap_buffer[i] = mask;
3557
3558                 offset = i * BITS_PER_LONG;
3559                 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3560         }
3561         if (is_dirty)
3562                 kvm_flush_remote_tlbs(kvm);
3563
3564         spin_unlock(&kvm->mmu_lock);
3565
3566         r = -EFAULT;
3567         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3568                 goto out;
3569
3570         r = 0;
3571 out:
3572         mutex_unlock(&kvm->slots_lock);
3573         return r;
3574 }
3575
3576 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3577                         bool line_status)
3578 {
3579         if (!irqchip_in_kernel(kvm))
3580                 return -ENXIO;
3581
3582         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3583                                         irq_event->irq, irq_event->level,
3584                                         line_status);
3585         return 0;
3586 }
3587
3588 long kvm_arch_vm_ioctl(struct file *filp,
3589                        unsigned int ioctl, unsigned long arg)
3590 {
3591         struct kvm *kvm = filp->private_data;
3592         void __user *argp = (void __user *)arg;
3593         int r = -ENOTTY;
3594         /*
3595          * This union makes it completely explicit to gcc-3.x
3596          * that these two variables' stack usage should be
3597          * combined, not added together.
3598          */
3599         union {
3600                 struct kvm_pit_state ps;
3601                 struct kvm_pit_state2 ps2;
3602                 struct kvm_pit_config pit_config;
3603         } u;
3604
3605         switch (ioctl) {
3606         case KVM_SET_TSS_ADDR:
3607                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3608                 break;
3609         case KVM_SET_IDENTITY_MAP_ADDR: {
3610                 u64 ident_addr;
3611
3612                 r = -EFAULT;
3613                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3614                         goto out;
3615                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3616                 break;
3617         }
3618         case KVM_SET_NR_MMU_PAGES:
3619                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3620                 break;
3621         case KVM_GET_NR_MMU_PAGES:
3622                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3623                 break;
3624         case KVM_CREATE_IRQCHIP: {
3625                 struct kvm_pic *vpic;
3626
3627                 mutex_lock(&kvm->lock);
3628                 r = -EEXIST;
3629                 if (kvm->arch.vpic)
3630                         goto create_irqchip_unlock;
3631                 r = -EINVAL;
3632                 if (atomic_read(&kvm->online_vcpus))
3633                         goto create_irqchip_unlock;
3634                 r = -ENOMEM;
3635                 vpic = kvm_create_pic(kvm);
3636                 if (vpic) {
3637                         r = kvm_ioapic_init(kvm);
3638                         if (r) {
3639                                 mutex_lock(&kvm->slots_lock);
3640                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3641                                                           &vpic->dev_master);
3642                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3643                                                           &vpic->dev_slave);
3644                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3645                                                           &vpic->dev_eclr);
3646                                 mutex_unlock(&kvm->slots_lock);
3647                                 kfree(vpic);
3648                                 goto create_irqchip_unlock;
3649                         }
3650                 } else
3651                         goto create_irqchip_unlock;
3652                 smp_wmb();
3653                 kvm->arch.vpic = vpic;
3654                 smp_wmb();
3655                 r = kvm_setup_default_irq_routing(kvm);
3656                 if (r) {
3657                         mutex_lock(&kvm->slots_lock);
3658                         mutex_lock(&kvm->irq_lock);
3659                         kvm_ioapic_destroy(kvm);
3660                         kvm_destroy_pic(kvm);
3661                         mutex_unlock(&kvm->irq_lock);
3662                         mutex_unlock(&kvm->slots_lock);
3663                 }
3664         create_irqchip_unlock:
3665                 mutex_unlock(&kvm->lock);
3666                 break;
3667         }
3668         case KVM_CREATE_PIT:
3669                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3670                 goto create_pit;
3671         case KVM_CREATE_PIT2:
3672                 r = -EFAULT;
3673                 if (copy_from_user(&u.pit_config, argp,
3674                                    sizeof(struct kvm_pit_config)))
3675                         goto out;
3676         create_pit:
3677                 mutex_lock(&kvm->slots_lock);
3678                 r = -EEXIST;
3679                 if (kvm->arch.vpit)
3680                         goto create_pit_unlock;
3681                 r = -ENOMEM;
3682                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3683                 if (kvm->arch.vpit)
3684                         r = 0;
3685         create_pit_unlock:
3686                 mutex_unlock(&kvm->slots_lock);
3687                 break;
3688         case KVM_GET_IRQCHIP: {
3689                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3690                 struct kvm_irqchip *chip;
3691
3692                 chip = memdup_user(argp, sizeof(*chip));
3693                 if (IS_ERR(chip)) {
3694                         r = PTR_ERR(chip);
3695                         goto out;
3696                 }
3697
3698                 r = -ENXIO;
3699                 if (!irqchip_in_kernel(kvm))
3700                         goto get_irqchip_out;
3701                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3702                 if (r)
3703                         goto get_irqchip_out;
3704                 r = -EFAULT;
3705                 if (copy_to_user(argp, chip, sizeof *chip))
3706                         goto get_irqchip_out;
3707                 r = 0;
3708         get_irqchip_out:
3709                 kfree(chip);
3710                 break;
3711         }
3712         case KVM_SET_IRQCHIP: {
3713                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3714                 struct kvm_irqchip *chip;
3715
3716                 chip = memdup_user(argp, sizeof(*chip));
3717                 if (IS_ERR(chip)) {
3718                         r = PTR_ERR(chip);
3719                         goto out;
3720                 }
3721
3722                 r = -ENXIO;
3723                 if (!irqchip_in_kernel(kvm))
3724                         goto set_irqchip_out;
3725                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3726                 if (r)
3727                         goto set_irqchip_out;
3728                 r = 0;
3729         set_irqchip_out:
3730                 kfree(chip);
3731                 break;
3732         }
3733         case KVM_GET_PIT: {
3734                 r = -EFAULT;
3735                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3736                         goto out;
3737                 r = -ENXIO;
3738                 if (!kvm->arch.vpit)
3739                         goto out;
3740                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3741                 if (r)
3742                         goto out;
3743                 r = -EFAULT;
3744                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3745                         goto out;
3746                 r = 0;
3747                 break;
3748         }
3749         case KVM_SET_PIT: {
3750                 r = -EFAULT;
3751                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3752                         goto out;
3753                 r = -ENXIO;
3754                 if (!kvm->arch.vpit)
3755                         goto out;
3756                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3757                 break;
3758         }
3759         case KVM_GET_PIT2: {
3760                 r = -ENXIO;
3761                 if (!kvm->arch.vpit)
3762                         goto out;
3763                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3764                 if (r)
3765                         goto out;
3766                 r = -EFAULT;
3767                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3768                         goto out;
3769                 r = 0;
3770                 break;
3771         }
3772         case KVM_SET_PIT2: {
3773                 r = -EFAULT;
3774                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3775                         goto out;
3776                 r = -ENXIO;
3777                 if (!kvm->arch.vpit)
3778                         goto out;
3779                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3780                 break;
3781         }
3782         case KVM_REINJECT_CONTROL: {
3783                 struct kvm_reinject_control control;
3784                 r =  -EFAULT;
3785                 if (copy_from_user(&control, argp, sizeof(control)))
3786                         goto out;
3787                 r = kvm_vm_ioctl_reinject(kvm, &control);
3788                 break;
3789         }
3790         case KVM_XEN_HVM_CONFIG: {
3791                 r = -EFAULT;
3792                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3793                                    sizeof(struct kvm_xen_hvm_config)))
3794                         goto out;
3795                 r = -EINVAL;
3796                 if (kvm->arch.xen_hvm_config.flags)
3797                         goto out;
3798                 r = 0;
3799                 break;
3800         }
3801         case KVM_SET_CLOCK: {
3802                 struct kvm_clock_data user_ns;
3803                 u64 now_ns;
3804                 s64 delta;
3805
3806                 r = -EFAULT;
3807                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3808                         goto out;
3809
3810                 r = -EINVAL;
3811                 if (user_ns.flags)
3812                         goto out;
3813
3814                 r = 0;
3815                 local_irq_disable();
3816                 now_ns = get_kernel_ns();
3817                 delta = user_ns.clock - now_ns;
3818                 local_irq_enable();
3819                 kvm->arch.kvmclock_offset = delta;
3820                 break;
3821         }
3822         case KVM_GET_CLOCK: {
3823                 struct kvm_clock_data user_ns;
3824                 u64 now_ns;
3825
3826                 local_irq_disable();
3827                 now_ns = get_kernel_ns();
3828                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3829                 local_irq_enable();
3830                 user_ns.flags = 0;
3831                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3832
3833                 r = -EFAULT;
3834                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3835                         goto out;
3836                 r = 0;
3837                 break;
3838         }
3839
3840         default:
3841                 ;
3842         }
3843 out:
3844         return r;
3845 }
3846
3847 static void kvm_init_msr_list(void)
3848 {
3849         u32 dummy[2];
3850         unsigned i, j;
3851
3852         /* skip the first msrs in the list. KVM-specific */
3853         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3854                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3855                         continue;
3856                 if (j < i)
3857                         msrs_to_save[j] = msrs_to_save[i];
3858                 j++;
3859         }
3860         num_msrs_to_save = j;
3861 }
3862
3863 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3864                            const void *v)
3865 {
3866         int handled = 0;
3867         int n;
3868
3869         do {
3870                 n = min(len, 8);
3871                 if (!(vcpu->arch.apic &&
3872                       !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3873                     && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3874                         break;
3875                 handled += n;
3876                 addr += n;
3877                 len -= n;
3878                 v += n;
3879         } while (len);
3880
3881         return handled;
3882 }
3883
3884 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3885 {
3886         int handled = 0;
3887         int n;
3888
3889         do {
3890                 n = min(len, 8);
3891                 if (!(vcpu->arch.apic &&
3892                       !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3893                     && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3894                         break;
3895                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3896                 handled += n;
3897                 addr += n;
3898                 len -= n;
3899                 v += n;
3900         } while (len);
3901
3902         return handled;
3903 }
3904
3905 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3906                         struct kvm_segment *var, int seg)
3907 {
3908         kvm_x86_ops->set_segment(vcpu, var, seg);
3909 }
3910
3911 void kvm_get_segment(struct kvm_vcpu *vcpu,
3912                      struct kvm_segment *var, int seg)
3913 {
3914         kvm_x86_ops->get_segment(vcpu, var, seg);
3915 }
3916
3917 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3918 {
3919         gpa_t t_gpa;
3920         struct x86_exception exception;
3921
3922         BUG_ON(!mmu_is_nested(vcpu));
3923
3924         /* NPT walks are always user-walks */
3925         access |= PFERR_USER_MASK;
3926         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3927
3928         return t_gpa;
3929 }
3930
3931 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3932                               struct x86_exception *exception)
3933 {
3934         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3935         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3936 }
3937
3938  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3939                                 struct x86_exception *exception)
3940 {
3941         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3942         access |= PFERR_FETCH_MASK;
3943         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3944 }
3945
3946 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3947                                struct x86_exception *exception)
3948 {
3949         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3950         access |= PFERR_WRITE_MASK;
3951         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3952 }
3953
3954 /* uses this to access any guest's mapped memory without checking CPL */
3955 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3956                                 struct x86_exception *exception)
3957 {
3958         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3959 }
3960
3961 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3962                                       struct kvm_vcpu *vcpu, u32 access,
3963                                       struct x86_exception *exception)
3964 {
3965         void *data = val;
3966         int r = X86EMUL_CONTINUE;
3967
3968         while (bytes) {
3969                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3970                                                             exception);
3971                 unsigned offset = addr & (PAGE_SIZE-1);
3972                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3973                 int ret;
3974
3975                 if (gpa == UNMAPPED_GVA)
3976                         return X86EMUL_PROPAGATE_FAULT;
3977                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3978                 if (ret < 0) {
3979                         r = X86EMUL_IO_NEEDED;
3980                         goto out;
3981                 }
3982
3983                 bytes -= toread;
3984                 data += toread;
3985                 addr += toread;
3986         }
3987 out:
3988         return r;
3989 }
3990
3991 /* used for instruction fetching */
3992 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3993                                 gva_t addr, void *val, unsigned int bytes,
3994                                 struct x86_exception *exception)
3995 {
3996         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3997         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3998
3999         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
4000                                           access | PFERR_FETCH_MASK,
4001                                           exception);
4002 }
4003
4004 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4005                                gva_t addr, void *val, unsigned int bytes,
4006                                struct x86_exception *exception)
4007 {
4008         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4009         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4010
4011         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4012                                           exception);
4013 }
4014 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4015
4016 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4017                                       gva_t addr, void *val, unsigned int bytes,
4018                                       struct x86_exception *exception)
4019 {
4020         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4021         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4022 }
4023
4024 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4025                                        gva_t addr, void *val,
4026                                        unsigned int bytes,
4027                                        struct x86_exception *exception)
4028 {
4029         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4030         void *data = val;
4031         int r = X86EMUL_CONTINUE;
4032
4033         while (bytes) {
4034                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4035                                                              PFERR_WRITE_MASK,
4036                                                              exception);
4037                 unsigned offset = addr & (PAGE_SIZE-1);
4038                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4039                 int ret;
4040
4041                 if (gpa == UNMAPPED_GVA)
4042                         return X86EMUL_PROPAGATE_FAULT;
4043                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4044                 if (ret < 0) {
4045                         r = X86EMUL_IO_NEEDED;
4046                         goto out;
4047                 }
4048
4049                 bytes -= towrite;
4050                 data += towrite;
4051                 addr += towrite;
4052         }
4053 out:
4054         return r;
4055 }
4056 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4057
4058 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4059                                 gpa_t *gpa, struct x86_exception *exception,
4060                                 bool write)
4061 {
4062         u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4063                 | (write ? PFERR_WRITE_MASK : 0);
4064
4065         if (vcpu_match_mmio_gva(vcpu, gva)
4066             && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
4067                 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4068                                         (gva & (PAGE_SIZE - 1));
4069                 trace_vcpu_match_mmio(gva, *gpa, write, false);
4070                 return 1;
4071         }
4072
4073         *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4074
4075         if (*gpa == UNMAPPED_GVA)
4076                 return -1;
4077
4078         /* For APIC access vmexit */
4079         if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4080                 return 1;
4081
4082         if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4083                 trace_vcpu_match_mmio(gva, *gpa, write, true);
4084                 return 1;
4085         }
4086
4087         return 0;
4088 }
4089
4090 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4091                         const void *val, int bytes)
4092 {
4093         int ret;
4094
4095         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4096         if (ret < 0)
4097                 return 0;
4098         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4099         return 1;
4100 }
4101
4102 struct read_write_emulator_ops {
4103         int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4104                                   int bytes);
4105         int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4106                                   void *val, int bytes);
4107         int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4108                                int bytes, void *val);
4109         int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4110                                     void *val, int bytes);
4111         bool write;
4112 };
4113
4114 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4115 {
4116         if (vcpu->mmio_read_completed) {
4117                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4118                                vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4119                 vcpu->mmio_read_completed = 0;
4120                 return 1;
4121         }
4122
4123         return 0;
4124 }
4125
4126 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4127                         void *val, int bytes)
4128 {
4129         return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4130 }
4131
4132 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4133                          void *val, int bytes)
4134 {
4135         return emulator_write_phys(vcpu, gpa, val, bytes);
4136 }
4137
4138 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4139 {
4140         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4141         return vcpu_mmio_write(vcpu, gpa, bytes, val);
4142 }
4143
4144 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4145                           void *val, int bytes)
4146 {
4147         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4148         return X86EMUL_IO_NEEDED;
4149 }
4150
4151 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4152                            void *val, int bytes)
4153 {
4154         struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4155
4156         memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4157         return X86EMUL_CONTINUE;
4158 }
4159
4160 static const struct read_write_emulator_ops read_emultor = {
4161         .read_write_prepare = read_prepare,
4162         .read_write_emulate = read_emulate,
4163         .read_write_mmio = vcpu_mmio_read,
4164         .read_write_exit_mmio = read_exit_mmio,
4165 };
4166
4167 static const struct read_write_emulator_ops write_emultor = {
4168         .read_write_emulate = write_emulate,
4169         .read_write_mmio = write_mmio,
4170         .read_write_exit_mmio = write_exit_mmio,
4171         .write = true,
4172 };
4173
4174 static int emulator_read_write_onepage(unsigned long addr, void *val,
4175                                        unsigned int bytes,
4176                                        struct x86_exception *exception,
4177                                        struct kvm_vcpu *vcpu,
4178                                        const struct read_write_emulator_ops *ops)
4179 {
4180         gpa_t gpa;
4181         int handled, ret;
4182         bool write = ops->write;
4183         struct kvm_mmio_fragment *frag;
4184
4185         ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4186
4187         if (ret < 0)
4188                 return X86EMUL_PROPAGATE_FAULT;
4189
4190         /* For APIC access vmexit */
4191         if (ret)
4192                 goto mmio;
4193
4194         if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4195                 return X86EMUL_CONTINUE;
4196
4197 mmio:
4198         /*
4199          * Is this MMIO handled locally?
4200          */
4201         handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4202         if (handled == bytes)
4203                 return X86EMUL_CONTINUE;
4204
4205         gpa += handled;
4206         bytes -= handled;
4207         val += handled;
4208
4209         WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4210         frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4211         frag->gpa = gpa;
4212         frag->data = val;
4213         frag->len = bytes;
4214         return X86EMUL_CONTINUE;
4215 }
4216
4217 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4218                         void *val, unsigned int bytes,
4219                         struct x86_exception *exception,
4220                         const struct read_write_emulator_ops *ops)
4221 {
4222         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4223         gpa_t gpa;
4224         int rc;
4225
4226         if (ops->read_write_prepare &&
4227                   ops->read_write_prepare(vcpu, val, bytes))
4228                 return X86EMUL_CONTINUE;
4229
4230         vcpu->mmio_nr_fragments = 0;
4231
4232         /* Crossing a page boundary? */
4233         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4234                 int now;
4235
4236                 now = -addr & ~PAGE_MASK;
4237                 rc = emulator_read_write_onepage(addr, val, now, exception,
4238                                                  vcpu, ops);
4239
4240                 if (rc != X86EMUL_CONTINUE)
4241                         return rc;
4242                 addr += now;
4243                 val += now;
4244                 bytes -= now;
4245         }
4246
4247         rc = emulator_read_write_onepage(addr, val, bytes, exception,
4248                                          vcpu, ops);
4249         if (rc != X86EMUL_CONTINUE)
4250                 return rc;
4251
4252         if (!vcpu->mmio_nr_fragments)
4253                 return rc;
4254
4255         gpa = vcpu->mmio_fragments[0].gpa;
4256
4257         vcpu->mmio_needed = 1;
4258         vcpu->mmio_cur_fragment = 0;
4259
4260         vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4261         vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4262         vcpu->run->exit_reason = KVM_EXIT_MMIO;
4263         vcpu->run->mmio.phys_addr = gpa;
4264
4265         return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4266 }
4267
4268 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4269                                   unsigned long addr,
4270                                   void *val,
4271                                   unsigned int bytes,
4272                                   struct x86_exception *exception)
4273 {
4274         return emulator_read_write(ctxt, addr, val, bytes,
4275                                    exception, &read_emultor);
4276 }
4277
4278 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4279                             unsigned long addr,
4280                             const void *val,
4281                             unsigned int bytes,
4282                             struct x86_exception *exception)
4283 {
4284         return emulator_read_write(ctxt, addr, (void *)val, bytes,
4285                                    exception, &write_emultor);
4286 }
4287
4288 #define CMPXCHG_TYPE(t, ptr, old, new) \
4289         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4290
4291 #ifdef CONFIG_X86_64
4292 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4293 #else
4294 #  define CMPXCHG64(ptr, old, new) \
4295         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4296 #endif
4297
4298 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4299                                      unsigned long addr,
4300                                      const void *old,
4301                                      const void *new,
4302                                      unsigned int bytes,
4303                                      struct x86_exception *exception)
4304 {
4305         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4306         gpa_t gpa;
4307         struct page *page;
4308         char *kaddr;
4309         bool exchanged;
4310
4311         /* guests cmpxchg8b have to be emulated atomically */
4312         if (bytes > 8 || (bytes & (bytes - 1)))
4313                 goto emul_write;
4314
4315         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4316
4317         if (gpa == UNMAPPED_GVA ||
4318             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4319                 goto emul_write;
4320
4321         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4322                 goto emul_write;
4323
4324         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4325         if (is_error_page(page))
4326                 goto emul_write;
4327
4328         kaddr = kmap_atomic(page);
4329         kaddr += offset_in_page(gpa);
4330         switch (bytes) {
4331         case 1:
4332                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4333                 break;
4334         case 2:
4335                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4336                 break;
4337         case 4:
4338                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4339                 break;
4340         case 8:
4341                 exchanged = CMPXCHG64(kaddr, old, new);
4342                 break;
4343         default:
4344                 BUG();
4345         }
4346         kunmap_atomic(kaddr);
4347         kvm_release_page_dirty(page);
4348
4349         if (!exchanged)
4350                 return X86EMUL_CMPXCHG_FAILED;
4351
4352         kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4353
4354         return X86EMUL_CONTINUE;
4355
4356 emul_write:
4357         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4358
4359         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4360 }
4361
4362 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4363 {
4364         /* TODO: String I/O for in kernel device */
4365         int r;
4366
4367         if (vcpu->arch.pio.in)
4368                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4369                                     vcpu->arch.pio.size, pd);
4370         else
4371                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4372                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4373                                      pd);
4374         return r;
4375 }
4376
4377 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4378                                unsigned short port, void *val,
4379                                unsigned int count, bool in)
4380 {
4381         trace_kvm_pio(!in, port, size, count);
4382
4383         vcpu->arch.pio.port = port;
4384         vcpu->arch.pio.in = in;
4385         vcpu->arch.pio.count  = count;
4386         vcpu->arch.pio.size = size;
4387
4388         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4389                 vcpu->arch.pio.count = 0;
4390                 return 1;
4391         }
4392
4393         vcpu->run->exit_reason = KVM_EXIT_IO;
4394         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4395         vcpu->run->io.size = size;
4396         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4397         vcpu->run->io.count = count;
4398         vcpu->run->io.port = port;
4399
4400         return 0;
4401 }
4402
4403 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4404                                     int size, unsigned short port, void *val,
4405                                     unsigned int count)
4406 {
4407         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4408         int ret;
4409
4410         if (vcpu->arch.pio.count)
4411                 goto data_avail;
4412
4413         ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4414         if (ret) {
4415 data_avail:
4416                 memcpy(val, vcpu->arch.pio_data, size * count);
4417                 vcpu->arch.pio.count = 0;
4418                 return 1;
4419         }
4420
4421         return 0;
4422 }
4423
4424 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4425                                      int size, unsigned short port,
4426                                      const void *val, unsigned int count)
4427 {
4428         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4429
4430         memcpy(vcpu->arch.pio_data, val, size * count);
4431         return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4432 }
4433
4434 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4435 {
4436         return kvm_x86_ops->get_segment_base(vcpu, seg);
4437 }
4438
4439 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4440 {
4441         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4442 }
4443
4444 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4445 {
4446         if (!need_emulate_wbinvd(vcpu))
4447                 return X86EMUL_CONTINUE;
4448
4449         if (kvm_x86_ops->has_wbinvd_exit()) {
4450                 int cpu = get_cpu();
4451
4452                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4453                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4454                                 wbinvd_ipi, NULL, 1);
4455                 put_cpu();
4456                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4457         } else
4458                 wbinvd();
4459         return X86EMUL_CONTINUE;
4460 }
4461 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4462
4463 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4464 {
4465         kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4466 }
4467
4468 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4469 {
4470         return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4471 }
4472
4473 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4474 {
4475
4476         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4477 }
4478
4479 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4480 {
4481         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4482 }
4483
4484 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4485 {
4486         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4487         unsigned long value;
4488
4489         switch (cr) {
4490         case 0:
4491                 value = kvm_read_cr0(vcpu);
4492                 break;
4493         case 2:
4494                 value = vcpu->arch.cr2;
4495                 break;
4496         case 3:
4497                 value = kvm_read_cr3(vcpu);
4498                 break;
4499         case 4:
4500                 value = kvm_read_cr4(vcpu);
4501                 break;
4502         case 8:
4503                 value = kvm_get_cr8(vcpu);
4504                 break;
4505         default:
4506                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4507                 return 0;
4508         }
4509
4510         return value;
4511 }
4512
4513 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4514 {
4515         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4516         int res = 0;
4517
4518         switch (cr) {
4519         case 0:
4520                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4521                 break;
4522         case 2:
4523                 vcpu->arch.cr2 = val;
4524                 break;
4525         case 3:
4526                 res = kvm_set_cr3(vcpu, val);
4527                 break;
4528         case 4:
4529                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4530                 break;
4531         case 8:
4532                 res = kvm_set_cr8(vcpu, val);
4533                 break;
4534         default:
4535                 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4536                 res = -1;
4537         }
4538
4539         return res;
4540 }
4541
4542 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4543 {
4544         kvm_set_rflags(emul_to_vcpu(ctxt), val);
4545 }
4546
4547 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4548 {
4549         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4550 }
4551
4552 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4553 {
4554         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4555 }
4556
4557 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4558 {
4559         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4560 }
4561
4562 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4563 {
4564         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4565 }
4566
4567 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4568 {
4569         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4570 }
4571
4572 static unsigned long emulator_get_cached_segment_base(
4573         struct x86_emulate_ctxt *ctxt, int seg)
4574 {
4575         return get_segment_base(emul_to_vcpu(ctxt), seg);
4576 }
4577
4578 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4579                                  struct desc_struct *desc, u32 *base3,
4580                                  int seg)
4581 {
4582         struct kvm_segment var;
4583
4584         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4585         *selector = var.selector;
4586
4587         if (var.unusable) {
4588                 memset(desc, 0, sizeof(*desc));
4589                 return false;
4590         }
4591
4592         if (var.g)
4593                 var.limit >>= 12;
4594         set_desc_limit(desc, var.limit);
4595         set_desc_base(desc, (unsigned long)var.base);
4596 #ifdef CONFIG_X86_64
4597         if (base3)
4598                 *base3 = var.base >> 32;
4599 #endif
4600         desc->type = var.type;
4601         desc->s = var.s;
4602         desc->dpl = var.dpl;
4603         desc->p = var.present;
4604         desc->avl = var.avl;
4605         desc->l = var.l;
4606         desc->d = var.db;
4607         desc->g = var.g;
4608
4609         return true;
4610 }
4611
4612 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4613                                  struct desc_struct *desc, u32 base3,
4614                                  int seg)
4615 {
4616         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4617         struct kvm_segment var;
4618
4619         var.selector = selector;
4620         var.base = get_desc_base(desc);
4621 #ifdef CONFIG_X86_64
4622         var.base |= ((u64)base3) << 32;
4623 #endif
4624         var.limit = get_desc_limit(desc);
4625         if (desc->g)
4626                 var.limit = (var.limit << 12) | 0xfff;
4627         var.type = desc->type;
4628         var.present = desc->p;
4629         var.dpl = desc->dpl;
4630         var.db = desc->d;
4631         var.s = desc->s;
4632         var.l = desc->l;
4633         var.g = desc->g;
4634         var.avl = desc->avl;
4635         var.present = desc->p;
4636         var.unusable = !var.present;
4637         var.padding = 0;
4638
4639         kvm_set_segment(vcpu, &var, seg);
4640         return;
4641 }
4642
4643 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4644                             u32 msr_index, u64 *pdata)
4645 {
4646         return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4647 }
4648
4649 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4650                             u32 msr_index, u64 data)
4651 {
4652         struct msr_data msr;
4653
4654         msr.data = data;
4655         msr.index = msr_index;
4656         msr.host_initiated = false;
4657         return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4658 }
4659
4660 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4661                              u32 pmc, u64 *pdata)
4662 {
4663         return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4664 }
4665
4666 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4667 {
4668         emul_to_vcpu(ctxt)->arch.halt_request = 1;
4669 }
4670
4671 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4672 {
4673         preempt_disable();
4674         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4675         /*
4676          * CR0.TS may reference the host fpu state, not the guest fpu state,
4677          * so it may be clear at this point.
4678          */
4679         clts();
4680 }
4681
4682 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4683 {
4684         preempt_enable();
4685 }
4686
4687 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4688                               struct x86_instruction_info *info,
4689                               enum x86_intercept_stage stage)
4690 {
4691         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4692 }
4693
4694 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4695                                u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4696 {
4697         kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4698 }
4699
4700 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4701 {
4702         return kvm_register_read(emul_to_vcpu(ctxt), reg);
4703 }
4704
4705 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4706 {
4707         kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4708 }
4709
4710 static const struct x86_emulate_ops emulate_ops = {
4711         .read_gpr            = emulator_read_gpr,
4712         .write_gpr           = emulator_write_gpr,
4713         .read_std            = kvm_read_guest_virt_system,
4714         .write_std           = kvm_write_guest_virt_system,
4715         .fetch               = kvm_fetch_guest_virt,
4716         .read_emulated       = emulator_read_emulated,
4717         .write_emulated      = emulator_write_emulated,
4718         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
4719         .invlpg              = emulator_invlpg,
4720         .pio_in_emulated     = emulator_pio_in_emulated,
4721         .pio_out_emulated    = emulator_pio_out_emulated,
4722         .get_segment         = emulator_get_segment,
4723         .set_segment         = emulator_set_segment,
4724         .get_cached_segment_base = emulator_get_cached_segment_base,
4725         .get_gdt             = emulator_get_gdt,
4726         .get_idt             = emulator_get_idt,
4727         .set_gdt             = emulator_set_gdt,
4728         .set_idt             = emulator_set_idt,
4729         .get_cr              = emulator_get_cr,
4730         .set_cr              = emulator_set_cr,
4731         .set_rflags          = emulator_set_rflags,
4732         .cpl                 = emulator_get_cpl,
4733         .get_dr              = emulator_get_dr,
4734         .set_dr              = emulator_set_dr,
4735         .set_msr             = emulator_set_msr,
4736         .get_msr             = emulator_get_msr,
4737         .read_pmc            = emulator_read_pmc,
4738         .halt                = emulator_halt,
4739         .wbinvd              = emulator_wbinvd,
4740         .fix_hypercall       = emulator_fix_hypercall,
4741         .get_fpu             = emulator_get_fpu,
4742         .put_fpu             = emulator_put_fpu,
4743         .intercept           = emulator_intercept,
4744         .get_cpuid           = emulator_get_cpuid,
4745 };
4746
4747 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4748 {
4749         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4750         /*
4751          * an sti; sti; sequence only disable interrupts for the first
4752          * instruction. So, if the last instruction, be it emulated or
4753          * not, left the system with the INT_STI flag enabled, it
4754          * means that the last instruction is an sti. We should not
4755          * leave the flag on in this case. The same goes for mov ss
4756          */
4757         if (!(int_shadow & mask))
4758                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4759 }
4760
4761 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4762 {
4763         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4764         if (ctxt->exception.vector == PF_VECTOR)
4765                 kvm_propagate_fault(vcpu, &ctxt->exception);
4766         else if (ctxt->exception.error_code_valid)
4767                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4768                                       ctxt->exception.error_code);
4769         else
4770                 kvm_queue_exception(vcpu, ctxt->exception.vector);
4771 }
4772
4773 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4774 {
4775         memset(&ctxt->twobyte, 0,
4776                (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
4777
4778         ctxt->fetch.start = 0;
4779         ctxt->fetch.end = 0;
4780         ctxt->io_read.pos = 0;
4781         ctxt->io_read.end = 0;
4782         ctxt->mem_read.pos = 0;
4783         ctxt->mem_read.end = 0;
4784 }
4785
4786 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4787 {
4788         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4789         int cs_db, cs_l;
4790
4791         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4792
4793         ctxt->eflags = kvm_get_rflags(vcpu);
4794         ctxt->eip = kvm_rip_read(vcpu);
4795         ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
4796                      (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
4797                      cs_l                               ? X86EMUL_MODE_PROT64 :
4798                      cs_db                              ? X86EMUL_MODE_PROT32 :
4799                                                           X86EMUL_MODE_PROT16;
4800         ctxt->guest_mode = is_guest_mode(vcpu);
4801
4802         init_decode_cache(ctxt);
4803         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4804 }
4805
4806 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4807 {
4808         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4809         int ret;
4810
4811         init_emulate_ctxt(vcpu);
4812
4813         ctxt->op_bytes = 2;
4814         ctxt->ad_bytes = 2;
4815         ctxt->_eip = ctxt->eip + inc_eip;
4816         ret = emulate_int_real(ctxt, irq);
4817
4818         if (ret != X86EMUL_CONTINUE)
4819                 return EMULATE_FAIL;
4820
4821         ctxt->eip = ctxt->_eip;
4822         kvm_rip_write(vcpu, ctxt->eip);
4823         kvm_set_rflags(vcpu, ctxt->eflags);
4824
4825         if (irq == NMI_VECTOR)
4826                 vcpu->arch.nmi_pending = 0;
4827         else
4828                 vcpu->arch.interrupt.pending = false;
4829
4830         return EMULATE_DONE;
4831 }
4832 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4833
4834 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4835 {
4836         int r = EMULATE_DONE;
4837
4838         ++vcpu->stat.insn_emulation_fail;
4839         trace_kvm_emulate_insn_failed(vcpu);
4840         if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
4841                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4842                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4843                 vcpu->run->internal.ndata = 0;
4844                 r = EMULATE_FAIL;
4845         }
4846         kvm_queue_exception(vcpu, UD_VECTOR);
4847
4848         return r;
4849 }
4850
4851 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4852                                   bool write_fault_to_shadow_pgtable,
4853                                   int emulation_type)
4854 {
4855         gpa_t gpa = cr2;
4856         pfn_t pfn;
4857
4858         if (emulation_type & EMULTYPE_NO_REEXECUTE)
4859                 return false;
4860
4861         if (!vcpu->arch.mmu.direct_map) {
4862                 /*
4863                  * Write permission should be allowed since only
4864                  * write access need to be emulated.
4865                  */
4866                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4867
4868                 /*
4869                  * If the mapping is invalid in guest, let cpu retry
4870                  * it to generate fault.
4871                  */
4872                 if (gpa == UNMAPPED_GVA)
4873                         return true;
4874         }
4875
4876         /*
4877          * Do not retry the unhandleable instruction if it faults on the
4878          * readonly host memory, otherwise it will goto a infinite loop:
4879          * retry instruction -> write #PF -> emulation fail -> retry
4880          * instruction -> ...
4881          */
4882         pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4883
4884         /*
4885          * If the instruction failed on the error pfn, it can not be fixed,
4886          * report the error to userspace.
4887          */
4888         if (is_error_noslot_pfn(pfn))
4889                 return false;
4890
4891         kvm_release_pfn_clean(pfn);
4892
4893         /* The instructions are well-emulated on direct mmu. */
4894         if (vcpu->arch.mmu.direct_map) {
4895                 unsigned int indirect_shadow_pages;
4896
4897                 spin_lock(&vcpu->kvm->mmu_lock);
4898                 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4899                 spin_unlock(&vcpu->kvm->mmu_lock);
4900
4901                 if (indirect_shadow_pages)
4902                         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4903
4904                 return true;
4905         }
4906
4907         /*
4908          * if emulation was due to access to shadowed page table
4909          * and it failed try to unshadow page and re-enter the
4910          * guest to let CPU execute the instruction.
4911          */
4912         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4913
4914         /*
4915          * If the access faults on its page table, it can not
4916          * be fixed by unprotecting shadow page and it should
4917          * be reported to userspace.
4918          */
4919         return !write_fault_to_shadow_pgtable;
4920 }
4921
4922 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4923                               unsigned long cr2,  int emulation_type)
4924 {
4925         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4926         unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4927
4928         last_retry_eip = vcpu->arch.last_retry_eip;
4929         last_retry_addr = vcpu->arch.last_retry_addr;
4930
4931         /*
4932          * If the emulation is caused by #PF and it is non-page_table
4933          * writing instruction, it means the VM-EXIT is caused by shadow
4934          * page protected, we can zap the shadow page and retry this
4935          * instruction directly.
4936          *
4937          * Note: if the guest uses a non-page-table modifying instruction
4938          * on the PDE that points to the instruction, then we will unmap
4939          * the instruction and go to an infinite loop. So, we cache the
4940          * last retried eip and the last fault address, if we meet the eip
4941          * and the address again, we can break out of the potential infinite
4942          * loop.
4943          */
4944         vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4945
4946         if (!(emulation_type & EMULTYPE_RETRY))
4947                 return false;
4948
4949         if (x86_page_table_writing_insn(ctxt))
4950                 return false;
4951
4952         if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4953                 return false;
4954
4955         vcpu->arch.last_retry_eip = ctxt->eip;
4956         vcpu->arch.last_retry_addr = cr2;
4957
4958         if (!vcpu->arch.mmu.direct_map)
4959                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4960
4961         kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4962
4963         return true;
4964 }
4965
4966 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4967 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4968
4969 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4970                             unsigned long cr2,
4971                             int emulation_type,
4972                             void *insn,
4973                             int insn_len)
4974 {
4975         int r;
4976         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4977         bool writeback = true;
4978         bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
4979
4980         /*
4981          * Clear write_fault_to_shadow_pgtable here to ensure it is
4982          * never reused.
4983          */
4984         vcpu->arch.write_fault_to_shadow_pgtable = false;
4985         kvm_clear_exception_queue(vcpu);
4986
4987         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4988                 init_emulate_ctxt(vcpu);
4989                 ctxt->interruptibility = 0;
4990                 ctxt->have_exception = false;
4991                 ctxt->perm_ok = false;
4992
4993                 ctxt->only_vendor_specific_insn
4994                         = emulation_type & EMULTYPE_TRAP_UD;
4995
4996                 r = x86_decode_insn(ctxt, insn, insn_len);
4997
4998                 trace_kvm_emulate_insn_start(vcpu);
4999                 ++vcpu->stat.insn_emulation;
5000                 if (r != EMULATION_OK)  {
5001                         if (emulation_type & EMULTYPE_TRAP_UD)
5002                                 return EMULATE_FAIL;
5003                         if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5004                                                 emulation_type))
5005                                 return EMULATE_DONE;
5006                         if (emulation_type & EMULTYPE_SKIP)
5007                                 return EMULATE_FAIL;
5008                         return handle_emulation_failure(vcpu);
5009                 }
5010         }
5011
5012         if (emulation_type & EMULTYPE_SKIP) {
5013                 kvm_rip_write(vcpu, ctxt->_eip);
5014                 return EMULATE_DONE;
5015         }
5016
5017         if (retry_instruction(ctxt, cr2, emulation_type))
5018                 return EMULATE_DONE;
5019
5020         /* this is needed for vmware backdoor interface to work since it
5021            changes registers values  during IO operation */
5022         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5023                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5024                 emulator_invalidate_register_cache(ctxt);
5025         }
5026
5027 restart:
5028         r = x86_emulate_insn(ctxt);
5029
5030         if (r == EMULATION_INTERCEPTED)
5031                 return EMULATE_DONE;
5032
5033         if (r == EMULATION_FAILED) {
5034                 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5035                                         emulation_type))
5036                         return EMULATE_DONE;
5037
5038                 return handle_emulation_failure(vcpu);
5039         }
5040
5041         if (ctxt->have_exception) {
5042                 inject_emulated_exception(vcpu);
5043                 r = EMULATE_DONE;
5044         } else if (vcpu->arch.pio.count) {
5045                 if (!vcpu->arch.pio.in)
5046                         vcpu->arch.pio.count = 0;
5047                 else {
5048                         writeback = false;
5049                         vcpu->arch.complete_userspace_io = complete_emulated_pio;
5050                 }
5051                 r = EMULATE_DO_MMIO;
5052         } else if (vcpu->mmio_needed) {
5053                 if (!vcpu->mmio_is_write)
5054                         writeback = false;
5055                 r = EMULATE_DO_MMIO;
5056                 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5057         } else if (r == EMULATION_RESTART)
5058                 goto restart;
5059         else
5060                 r = EMULATE_DONE;
5061
5062         if (writeback) {
5063                 toggle_interruptibility(vcpu, ctxt->interruptibility);
5064                 kvm_set_rflags(vcpu, ctxt->eflags);
5065                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5066                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5067                 kvm_rip_write(vcpu, ctxt->eip);
5068         } else
5069                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5070
5071         return r;
5072 }
5073 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5074
5075 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5076 {
5077         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5078         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5079                                             size, port, &val, 1);
5080         /* do not return to emulator after return from userspace */
5081         vcpu->arch.pio.count = 0;
5082         return ret;
5083 }
5084 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5085
5086 static void tsc_bad(void *info)
5087 {
5088         __this_cpu_write(cpu_tsc_khz, 0);
5089 }
5090
5091 static void tsc_khz_changed(void *data)
5092 {
5093         struct cpufreq_freqs *freq = data;
5094         unsigned long khz = 0;
5095
5096         if (data)
5097                 khz = freq->new;
5098         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5099                 khz = cpufreq_quick_get(raw_smp_processor_id());
5100         if (!khz)
5101                 khz = tsc_khz;
5102         __this_cpu_write(cpu_tsc_khz, khz);
5103 }
5104
5105 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5106                                      void *data)
5107 {
5108         struct cpufreq_freqs *freq = data;
5109         struct kvm *kvm;
5110         struct kvm_vcpu *vcpu;
5111         int i, send_ipi = 0;
5112
5113         /*
5114          * We allow guests to temporarily run on slowing clocks,
5115          * provided we notify them after, or to run on accelerating
5116          * clocks, provided we notify them before.  Thus time never
5117          * goes backwards.
5118          *
5119          * However, we have a problem.  We can't atomically update
5120          * the frequency of a given CPU from this function; it is
5121          * merely a notifier, which can be called from any CPU.
5122          * Changing the TSC frequency at arbitrary points in time
5123          * requires a recomputation of local variables related to
5124          * the TSC for each VCPU.  We must flag these local variables
5125          * to be updated and be sure the update takes place with the
5126          * new frequency before any guests proceed.
5127          *
5128          * Unfortunately, the combination of hotplug CPU and frequency
5129          * change creates an intractable locking scenario; the order
5130          * of when these callouts happen is undefined with respect to
5131          * CPU hotplug, and they can race with each other.  As such,
5132          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5133          * undefined; you can actually have a CPU frequency change take
5134          * place in between the computation of X and the setting of the
5135          * variable.  To protect against this problem, all updates of
5136          * the per_cpu tsc_khz variable are done in an interrupt
5137          * protected IPI, and all callers wishing to update the value
5138          * must wait for a synchronous IPI to complete (which is trivial
5139          * if the caller is on the CPU already).  This establishes the
5140          * necessary total order on variable updates.
5141          *
5142          * Note that because a guest time update may take place
5143          * anytime after the setting of the VCPU's request bit, the
5144          * correct TSC value must be set before the request.  However,
5145          * to ensure the update actually makes it to any guest which
5146          * starts running in hardware virtualization between the set
5147          * and the acquisition of the spinlock, we must also ping the
5148          * CPU after setting the request bit.
5149          *
5150          */
5151
5152         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5153                 return 0;
5154         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5155                 return 0;
5156
5157         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5158
5159         spin_lock(&kvm_lock);
5160         list_for_each_entry(kvm, &vm_list, vm_list) {
5161                 kvm_for_each_vcpu(i, vcpu, kvm) {
5162                         if (vcpu->cpu != freq->cpu)
5163                                 continue;
5164                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5165                         if (vcpu->cpu != smp_processor_id())
5166                                 send_ipi = 1;
5167                 }
5168         }
5169         spin_unlock(&kvm_lock);
5170
5171         if (freq->old < freq->new && send_ipi) {
5172                 /*
5173                  * We upscale the frequency.  Must make the guest
5174                  * doesn't see old kvmclock values while running with
5175                  * the new frequency, otherwise we risk the guest sees
5176                  * time go backwards.
5177                  *
5178                  * In case we update the frequency for another cpu
5179                  * (which might be in guest context) send an interrupt
5180                  * to kick the cpu out of guest context.  Next time
5181                  * guest context is entered kvmclock will be updated,
5182                  * so the guest will not see stale values.
5183                  */
5184                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5185         }
5186         return 0;
5187 }
5188
5189 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5190         .notifier_call  = kvmclock_cpufreq_notifier
5191 };
5192
5193 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5194                                         unsigned long action, void *hcpu)
5195 {
5196         unsigned int cpu = (unsigned long)hcpu;
5197
5198         switch (action) {
5199                 case CPU_ONLINE:
5200                 case CPU_DOWN_FAILED:
5201                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5202                         break;
5203                 case CPU_DOWN_PREPARE:
5204                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
5205                         break;
5206         }
5207         return NOTIFY_OK;
5208 }
5209
5210 static struct notifier_block kvmclock_cpu_notifier_block = {
5211         .notifier_call  = kvmclock_cpu_notifier,
5212         .priority = -INT_MAX
5213 };
5214
5215 static void kvm_timer_init(void)
5216 {
5217         int cpu;
5218
5219         max_tsc_khz = tsc_khz;
5220         register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5221         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5222 #ifdef CONFIG_CPU_FREQ
5223                 struct cpufreq_policy policy;
5224                 memset(&policy, 0, sizeof(policy));
5225                 cpu = get_cpu();
5226                 cpufreq_get_policy(&policy, cpu);
5227                 if (policy.cpuinfo.max_freq)
5228                         max_tsc_khz = policy.cpuinfo.max_freq;
5229                 put_cpu();
5230 #endif
5231                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5232                                           CPUFREQ_TRANSITION_NOTIFIER);
5233         }
5234         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5235         for_each_online_cpu(cpu)
5236                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5237 }
5238
5239 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5240
5241 int kvm_is_in_guest(void)
5242 {
5243         return __this_cpu_read(current_vcpu) != NULL;
5244 }
5245
5246 static int kvm_is_user_mode(void)
5247 {
5248         int user_mode = 3;
5249
5250         if (__this_cpu_read(current_vcpu))
5251                 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5252
5253         return user_mode != 0;
5254 }
5255
5256 static unsigned long kvm_get_guest_ip(void)
5257 {
5258         unsigned long ip = 0;
5259
5260         if (__this_cpu_read(current_vcpu))
5261                 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5262
5263         return ip;
5264 }
5265
5266 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5267         .is_in_guest            = kvm_is_in_guest,
5268         .is_user_mode           = kvm_is_user_mode,
5269         .get_guest_ip           = kvm_get_guest_ip,
5270 };
5271
5272 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5273 {
5274         __this_cpu_write(current_vcpu, vcpu);
5275 }
5276 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5277
5278 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5279 {
5280         __this_cpu_write(current_vcpu, NULL);
5281 }
5282 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5283
5284 static void kvm_set_mmio_spte_mask(void)
5285 {
5286         u64 mask;
5287         int maxphyaddr = boot_cpu_data.x86_phys_bits;
5288
5289         /*
5290          * Set the reserved bits and the present bit of an paging-structure
5291          * entry to generate page fault with PFER.RSV = 1.
5292          */
5293         mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5294         mask |= 1ull;
5295
5296 #ifdef CONFIG_X86_64
5297         /*
5298          * If reserved bit is not supported, clear the present bit to disable
5299          * mmio page fault.
5300          */
5301         if (maxphyaddr == 52)
5302                 mask &= ~1ull;
5303 #endif
5304
5305         kvm_mmu_set_mmio_spte_mask(mask);
5306 }
5307
5308 #ifdef CONFIG_X86_64
5309 static void pvclock_gtod_update_fn(struct work_struct *work)
5310 {
5311         struct kvm *kvm;
5312
5313         struct kvm_vcpu *vcpu;
5314         int i;
5315
5316         spin_lock(&kvm_lock);
5317         list_for_each_entry(kvm, &vm_list, vm_list)
5318                 kvm_for_each_vcpu(i, vcpu, kvm)
5319                         set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5320         atomic_set(&kvm_guest_has_master_clock, 0);
5321         spin_unlock(&kvm_lock);
5322 }
5323
5324 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5325
5326 /*
5327  * Notification about pvclock gtod data update.
5328  */
5329 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5330                                void *priv)
5331 {
5332         struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5333         struct timekeeper *tk = priv;
5334
5335         update_pvclock_gtod(tk);
5336
5337         /* disable master clock if host does not trust, or does not
5338          * use, TSC clocksource
5339          */
5340         if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5341             atomic_read(&kvm_guest_has_master_clock) != 0)
5342                 queue_work(system_long_wq, &pvclock_gtod_work);
5343
5344         return 0;
5345 }
5346
5347 static struct notifier_block pvclock_gtod_notifier = {
5348         .notifier_call = pvclock_gtod_notify,
5349 };
5350 #endif
5351
5352 int kvm_arch_init(void *opaque)
5353 {
5354         int r;
5355         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5356
5357         if (kvm_x86_ops) {
5358                 printk(KERN_ERR "kvm: already loaded the other module\n");
5359                 r = -EEXIST;
5360                 goto out;
5361         }
5362
5363         if (!ops->cpu_has_kvm_support()) {
5364                 printk(KERN_ERR "kvm: no hardware support\n");
5365                 r = -EOPNOTSUPP;
5366                 goto out;
5367         }
5368         if (ops->disabled_by_bios()) {
5369                 printk(KERN_ERR "kvm: disabled by bios\n");
5370                 r = -EOPNOTSUPP;
5371                 goto out;
5372         }
5373
5374         r = -ENOMEM;
5375         shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5376         if (!shared_msrs) {
5377                 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5378                 goto out;
5379         }
5380
5381         r = kvm_mmu_module_init();
5382         if (r)
5383                 goto out_free_percpu;
5384
5385         kvm_set_mmio_spte_mask();
5386         kvm_init_msr_list();
5387
5388         kvm_x86_ops = ops;
5389         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5390                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
5391
5392         kvm_timer_init();
5393
5394         perf_register_guest_info_callbacks(&kvm_guest_cbs);
5395
5396         if (cpu_has_xsave)
5397                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5398
5399         kvm_lapic_init();
5400 #ifdef CONFIG_X86_64
5401         pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5402 #endif
5403
5404         return 0;
5405
5406 out_free_percpu:
5407         free_percpu(shared_msrs);
5408 out:
5409         return r;
5410 }
5411
5412 void kvm_arch_exit(void)
5413 {
5414         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5415
5416         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5417                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5418                                             CPUFREQ_TRANSITION_NOTIFIER);
5419         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5420 #ifdef CONFIG_X86_64
5421         pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5422 #endif
5423         kvm_x86_ops = NULL;
5424         kvm_mmu_module_exit();
5425         free_percpu(shared_msrs);
5426 }
5427
5428 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5429 {
5430         ++vcpu->stat.halt_exits;
5431         if (irqchip_in_kernel(vcpu->kvm)) {
5432                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5433                 return 1;
5434         } else {
5435                 vcpu->run->exit_reason = KVM_EXIT_HLT;
5436                 return 0;
5437         }
5438 }
5439 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5440
5441 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5442 {
5443         u64 param, ingpa, outgpa, ret;
5444         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5445         bool fast, longmode;
5446         int cs_db, cs_l;
5447
5448         /*
5449          * hypercall generates UD from non zero cpl and real mode
5450          * per HYPER-V spec
5451          */
5452         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5453                 kvm_queue_exception(vcpu, UD_VECTOR);
5454                 return 0;
5455         }
5456
5457         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5458         longmode = is_long_mode(vcpu) && cs_l == 1;
5459
5460         if (!longmode) {
5461                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5462                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5463                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5464                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5465                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5466                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5467         }
5468 #ifdef CONFIG_X86_64
5469         else {
5470                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5471                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5472                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5473         }
5474 #endif
5475
5476         code = param & 0xffff;
5477         fast = (param >> 16) & 0x1;
5478         rep_cnt = (param >> 32) & 0xfff;
5479         rep_idx = (param >> 48) & 0xfff;
5480
5481         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5482
5483         switch (code) {
5484         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5485                 kvm_vcpu_on_spin(vcpu);
5486                 break;
5487         default:
5488                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5489                 break;
5490         }
5491
5492         ret = res | (((u64)rep_done & 0xfff) << 32);
5493         if (longmode) {
5494                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5495         } else {
5496                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5497                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5498         }
5499
5500         return 1;
5501 }
5502
5503 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5504 {
5505         unsigned long nr, a0, a1, a2, a3, ret;
5506         int r = 1;
5507
5508         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5509                 return kvm_hv_hypercall(vcpu);
5510
5511         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5512         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5513         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5514         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5515         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5516
5517         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5518
5519         if (!is_long_mode(vcpu)) {
5520                 nr &= 0xFFFFFFFF;
5521                 a0 &= 0xFFFFFFFF;
5522                 a1 &= 0xFFFFFFFF;
5523                 a2 &= 0xFFFFFFFF;
5524                 a3 &= 0xFFFFFFFF;
5525         }
5526
5527         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5528                 ret = -KVM_EPERM;
5529                 goto out;
5530         }
5531
5532         switch (nr) {
5533         case KVM_HC_VAPIC_POLL_IRQ:
5534                 ret = 0;
5535                 break;
5536         default:
5537                 ret = -KVM_ENOSYS;
5538                 break;
5539         }
5540 out:
5541         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5542         ++vcpu->stat.hypercalls;
5543         return r;
5544 }
5545 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5546
5547 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5548 {
5549         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5550         char instruction[3];
5551         unsigned long rip = kvm_rip_read(vcpu);
5552
5553         /*
5554          * Blow out the MMU to ensure that no other VCPU has an active mapping
5555          * to ensure that the updated hypercall appears atomically across all
5556          * VCPUs.
5557          */
5558         kvm_mmu_zap_all(vcpu->kvm);
5559
5560         kvm_x86_ops->patch_hypercall(vcpu, instruction);
5561
5562         return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5563 }
5564
5565 /*
5566  * Check if userspace requested an interrupt window, and that the
5567  * interrupt window is open.
5568  *
5569  * No need to exit to userspace if we already have an interrupt queued.
5570  */
5571 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5572 {
5573         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5574                 vcpu->run->request_interrupt_window &&
5575                 kvm_arch_interrupt_allowed(vcpu));
5576 }
5577
5578 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5579 {
5580         struct kvm_run *kvm_run = vcpu->run;
5581
5582         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5583         kvm_run->cr8 = kvm_get_cr8(vcpu);
5584         kvm_run->apic_base = kvm_get_apic_base(vcpu);
5585         if (irqchip_in_kernel(vcpu->kvm))
5586                 kvm_run->ready_for_interrupt_injection = 1;
5587         else
5588                 kvm_run->ready_for_interrupt_injection =
5589                         kvm_arch_interrupt_allowed(vcpu) &&
5590                         !kvm_cpu_has_interrupt(vcpu) &&
5591                         !kvm_event_needs_reinjection(vcpu);
5592 }
5593
5594 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5595 {
5596         int max_irr, tpr;
5597
5598         if (!kvm_x86_ops->update_cr8_intercept)
5599                 return;
5600
5601         if (!vcpu->arch.apic)
5602                 return;
5603
5604         if (!vcpu->arch.apic->vapic_addr)
5605                 max_irr = kvm_lapic_find_highest_irr(vcpu);
5606         else
5607                 max_irr = -1;
5608
5609         if (max_irr != -1)
5610                 max_irr >>= 4;
5611
5612         tpr = kvm_lapic_get_cr8(vcpu);
5613
5614         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5615 }
5616
5617 static void inject_pending_event(struct kvm_vcpu *vcpu)
5618 {
5619         /* try to reinject previous events if any */
5620         if (vcpu->arch.exception.pending) {
5621                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5622                                         vcpu->arch.exception.has_error_code,
5623                                         vcpu->arch.exception.error_code);
5624                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5625                                           vcpu->arch.exception.has_error_code,
5626                                           vcpu->arch.exception.error_code,
5627                                           vcpu->arch.exception.reinject);
5628                 return;
5629         }
5630
5631         if (vcpu->arch.nmi_injected) {
5632                 kvm_x86_ops->set_nmi(vcpu);
5633                 return;
5634         }
5635
5636         if (vcpu->arch.interrupt.pending) {
5637                 kvm_x86_ops->set_irq(vcpu);
5638                 return;
5639         }
5640
5641         /* try to inject new event if pending */
5642         if (vcpu->arch.nmi_pending) {
5643                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5644                         --vcpu->arch.nmi_pending;
5645                         vcpu->arch.nmi_injected = true;
5646                         kvm_x86_ops->set_nmi(vcpu);
5647                 }
5648         } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5649                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5650                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5651                                             false);
5652                         kvm_x86_ops->set_irq(vcpu);
5653                 }
5654         }
5655 }
5656
5657 static void process_nmi(struct kvm_vcpu *vcpu)
5658 {
5659         unsigned limit = 2;
5660
5661         /*
5662          * x86 is limited to one NMI running, and one NMI pending after it.
5663          * If an NMI is already in progress, limit further NMIs to just one.
5664          * Otherwise, allow two (and we'll inject the first one immediately).
5665          */
5666         if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5667                 limit = 1;
5668
5669         vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5670         vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5671         kvm_make_request(KVM_REQ_EVENT, vcpu);
5672 }
5673
5674 static void kvm_gen_update_masterclock(struct kvm *kvm)
5675 {
5676 #ifdef CONFIG_X86_64
5677         int i;
5678         struct kvm_vcpu *vcpu;
5679         struct kvm_arch *ka = &kvm->arch;
5680
5681         spin_lock(&ka->pvclock_gtod_sync_lock);
5682         kvm_make_mclock_inprogress_request(kvm);
5683         /* no guest entries from this point */
5684         pvclock_update_vm_gtod_copy(kvm);
5685
5686         kvm_for_each_vcpu(i, vcpu, kvm)
5687                 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5688
5689         /* guest entries allowed */
5690         kvm_for_each_vcpu(i, vcpu, kvm)
5691                 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5692
5693         spin_unlock(&ka->pvclock_gtod_sync_lock);
5694 #endif
5695 }
5696
5697 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5698 {
5699         u64 eoi_exit_bitmap[4];
5700         u32 tmr[8];
5701
5702         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
5703                 return;
5704
5705         memset(eoi_exit_bitmap, 0, 32);
5706         memset(tmr, 0, 32);
5707
5708         kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
5709         kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5710         kvm_apic_update_tmr(vcpu, tmr);
5711 }
5712
5713 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5714 {
5715         int r;
5716         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5717                 vcpu->run->request_interrupt_window;
5718         bool req_immediate_exit = false;
5719
5720         if (vcpu->requests) {
5721                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5722                         kvm_mmu_unload(vcpu);
5723                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5724                         __kvm_migrate_timers(vcpu);
5725                 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5726                         kvm_gen_update_masterclock(vcpu->kvm);
5727                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5728                         r = kvm_guest_time_update(vcpu);
5729                         if (unlikely(r))
5730                                 goto out;
5731                 }
5732                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5733                         kvm_mmu_sync_roots(vcpu);
5734                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5735                         kvm_x86_ops->tlb_flush(vcpu);
5736                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5737                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5738                         r = 0;
5739                         goto out;
5740                 }
5741                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5742                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5743                         r = 0;
5744                         goto out;
5745                 }
5746                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5747                         vcpu->fpu_active = 0;
5748                         kvm_x86_ops->fpu_deactivate(vcpu);
5749                 }
5750                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5751                         /* Page is swapped out. Do synthetic halt */
5752                         vcpu->arch.apf.halted = true;
5753                         r = 1;
5754                         goto out;
5755                 }
5756                 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5757                         record_steal_time(vcpu);
5758                 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5759                         process_nmi(vcpu);
5760                 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5761                         kvm_handle_pmu_event(vcpu);
5762                 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5763                         kvm_deliver_pmi(vcpu);
5764                 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
5765                         vcpu_scan_ioapic(vcpu);
5766         }
5767
5768         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5769                 kvm_apic_accept_events(vcpu);
5770                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
5771                         r = 1;
5772                         goto out;
5773                 }
5774
5775                 inject_pending_event(vcpu);
5776
5777                 /* enable NMI/IRQ window open exits if needed */
5778                 if (vcpu->arch.nmi_pending)
5779                         req_immediate_exit =
5780                                 kvm_x86_ops->enable_nmi_window(vcpu) != 0;
5781                 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
5782                         req_immediate_exit =
5783                                 kvm_x86_ops->enable_irq_window(vcpu) != 0;
5784
5785                 if (kvm_lapic_enabled(vcpu)) {
5786                         /*
5787                          * Update architecture specific hints for APIC
5788                          * virtual interrupt delivery.
5789                          */
5790                         if (kvm_x86_ops->hwapic_irr_update)
5791                                 kvm_x86_ops->hwapic_irr_update(vcpu,
5792                                         kvm_lapic_find_highest_irr(vcpu));
5793                         update_cr8_intercept(vcpu);
5794                         kvm_lapic_sync_to_vapic(vcpu);
5795                 }
5796         }
5797
5798         r = kvm_mmu_reload(vcpu);
5799         if (unlikely(r)) {
5800                 goto cancel_injection;
5801         }
5802
5803         preempt_disable();
5804
5805         kvm_x86_ops->prepare_guest_switch(vcpu);
5806         if (vcpu->fpu_active)
5807                 kvm_load_guest_fpu(vcpu);
5808         kvm_load_guest_xcr0(vcpu);
5809
5810         vcpu->mode = IN_GUEST_MODE;
5811
5812         /* We should set ->mode before check ->requests,
5813          * see the comment in make_all_cpus_request.
5814          */
5815         smp_mb();
5816
5817         local_irq_disable();
5818
5819         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5820             || need_resched() || signal_pending(current)) {
5821                 vcpu->mode = OUTSIDE_GUEST_MODE;
5822                 smp_wmb();
5823                 local_irq_enable();
5824                 preempt_enable();
5825                 r = 1;
5826                 goto cancel_injection;
5827         }
5828
5829         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5830
5831         if (req_immediate_exit)
5832                 smp_send_reschedule(vcpu->cpu);
5833
5834         kvm_guest_enter();
5835
5836         if (unlikely(vcpu->arch.switch_db_regs)) {
5837                 set_debugreg(0, 7);
5838                 set_debugreg(vcpu->arch.eff_db[0], 0);
5839                 set_debugreg(vcpu->arch.eff_db[1], 1);
5840                 set_debugreg(vcpu->arch.eff_db[2], 2);
5841                 set_debugreg(vcpu->arch.eff_db[3], 3);
5842         }
5843
5844         trace_kvm_entry(vcpu->vcpu_id);
5845         kvm_x86_ops->run(vcpu);
5846
5847         /*
5848          * If the guest has used debug registers, at least dr7
5849          * will be disabled while returning to the host.
5850          * If we don't have active breakpoints in the host, we don't
5851          * care about the messed up debug address registers. But if
5852          * we have some of them active, restore the old state.
5853          */
5854         if (hw_breakpoint_active())
5855                 hw_breakpoint_restore();
5856
5857         vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5858                                                            native_read_tsc());
5859
5860         vcpu->mode = OUTSIDE_GUEST_MODE;
5861         smp_wmb();
5862
5863         /* Interrupt is enabled by handle_external_intr() */
5864         kvm_x86_ops->handle_external_intr(vcpu);
5865
5866         ++vcpu->stat.exits;
5867
5868         /*
5869          * We must have an instruction between local_irq_enable() and
5870          * kvm_guest_exit(), so the timer interrupt isn't delayed by
5871          * the interrupt shadow.  The stat.exits increment will do nicely.
5872          * But we need to prevent reordering, hence this barrier():
5873          */
5874         barrier();
5875
5876         kvm_guest_exit();
5877
5878         preempt_enable();
5879
5880         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5881
5882         /*
5883          * Profile KVM exit RIPs:
5884          */
5885         if (unlikely(prof_on == KVM_PROFILING)) {
5886                 unsigned long rip = kvm_rip_read(vcpu);
5887                 profile_hit(KVM_PROFILING, (void *)rip);
5888         }
5889
5890         if (unlikely(vcpu->arch.tsc_always_catchup))
5891                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5892
5893         if (vcpu->arch.apic_attention)
5894                 kvm_lapic_sync_from_vapic(vcpu);
5895
5896         r = kvm_x86_ops->handle_exit(vcpu);
5897         return r;
5898
5899 cancel_injection:
5900         kvm_x86_ops->cancel_injection(vcpu);
5901         if (unlikely(vcpu->arch.apic_attention))
5902                 kvm_lapic_sync_from_vapic(vcpu);
5903 out:
5904         return r;
5905 }
5906
5907
5908 static int __vcpu_run(struct kvm_vcpu *vcpu)
5909 {
5910         int r;
5911         struct kvm *kvm = vcpu->kvm;
5912
5913         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5914
5915         r = 1;
5916         while (r > 0) {
5917                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5918                     !vcpu->arch.apf.halted)
5919                         r = vcpu_enter_guest(vcpu);
5920                 else {
5921                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5922                         kvm_vcpu_block(vcpu);
5923                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5924                         if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
5925                                 kvm_apic_accept_events(vcpu);
5926                                 switch(vcpu->arch.mp_state) {
5927                                 case KVM_MP_STATE_HALTED:
5928                                         vcpu->arch.mp_state =
5929                                                 KVM_MP_STATE_RUNNABLE;
5930                                 case KVM_MP_STATE_RUNNABLE:
5931                                         vcpu->arch.apf.halted = false;
5932                                         break;
5933                                 case KVM_MP_STATE_INIT_RECEIVED:
5934                                         break;
5935                                 default:
5936                                         r = -EINTR;
5937                                         break;
5938                                 }
5939                         }
5940                 }
5941
5942                 if (r <= 0)
5943                         break;
5944
5945                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5946                 if (kvm_cpu_has_pending_timer(vcpu))
5947                         kvm_inject_pending_timer_irqs(vcpu);
5948
5949                 if (dm_request_for_irq_injection(vcpu)) {
5950                         r = -EINTR;
5951                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5952                         ++vcpu->stat.request_irq_exits;
5953                 }
5954
5955                 kvm_check_async_pf_completion(vcpu);
5956
5957                 if (signal_pending(current)) {
5958                         r = -EINTR;
5959                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5960                         ++vcpu->stat.signal_exits;
5961                 }
5962                 if (need_resched()) {
5963                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5964                         cond_resched();
5965                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5966                 }
5967         }
5968
5969         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5970
5971         return r;
5972 }
5973
5974 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5975 {
5976         int r;
5977         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5978         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5979         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5980         if (r != EMULATE_DONE)
5981                 return 0;
5982         return 1;
5983 }
5984
5985 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5986 {
5987         BUG_ON(!vcpu->arch.pio.count);
5988
5989         return complete_emulated_io(vcpu);
5990 }
5991
5992 /*
5993  * Implements the following, as a state machine:
5994  *
5995  * read:
5996  *   for each fragment
5997  *     for each mmio piece in the fragment
5998  *       write gpa, len
5999  *       exit
6000  *       copy data
6001  *   execute insn
6002  *
6003  * write:
6004  *   for each fragment
6005  *     for each mmio piece in the fragment
6006  *       write gpa, len
6007  *       copy data
6008  *       exit
6009  */
6010 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6011 {
6012         struct kvm_run *run = vcpu->run;
6013         struct kvm_mmio_fragment *frag;
6014         unsigned len;
6015
6016         BUG_ON(!vcpu->mmio_needed);
6017
6018         /* Complete previous fragment */
6019         frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6020         len = min(8u, frag->len);
6021         if (!vcpu->mmio_is_write)
6022                 memcpy(frag->data, run->mmio.data, len);
6023
6024         if (frag->len <= 8) {
6025                 /* Switch to the next fragment. */
6026                 frag++;
6027                 vcpu->mmio_cur_fragment++;
6028         } else {
6029                 /* Go forward to the next mmio piece. */
6030                 frag->data += len;
6031                 frag->gpa += len;
6032                 frag->len -= len;
6033         }
6034
6035         if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6036                 vcpu->mmio_needed = 0;
6037                 if (vcpu->mmio_is_write)
6038                         return 1;
6039                 vcpu->mmio_read_completed = 1;
6040                 return complete_emulated_io(vcpu);
6041         }
6042
6043         run->exit_reason = KVM_EXIT_MMIO;
6044         run->mmio.phys_addr = frag->gpa;
6045         if (vcpu->mmio_is_write)
6046                 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6047         run->mmio.len = min(8u, frag->len);
6048         run->mmio.is_write = vcpu->mmio_is_write;
6049         vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6050         return 0;
6051 }
6052
6053
6054 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6055 {
6056         int r;
6057         sigset_t sigsaved;
6058
6059         if (!tsk_used_math(current) && init_fpu(current))
6060                 return -ENOMEM;
6061
6062         if (vcpu->sigset_active)
6063                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6064
6065         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6066                 kvm_vcpu_block(vcpu);
6067                 kvm_apic_accept_events(vcpu);
6068                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6069                 r = -EAGAIN;
6070                 goto out;
6071         }
6072
6073         /* re-sync apic's tpr */
6074         if (!irqchip_in_kernel(vcpu->kvm)) {
6075                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6076                         r = -EINVAL;
6077                         goto out;
6078                 }
6079         }
6080
6081         if (unlikely(vcpu->arch.complete_userspace_io)) {
6082                 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6083                 vcpu->arch.complete_userspace_io = NULL;
6084                 r = cui(vcpu);
6085                 if (r <= 0)
6086                         goto out;
6087         } else
6088                 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6089
6090         r = __vcpu_run(vcpu);
6091
6092 out:
6093         post_kvm_run_save(vcpu);
6094         if (vcpu->sigset_active)
6095                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6096
6097         return r;
6098 }
6099
6100 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6101 {
6102         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6103                 /*
6104                  * We are here if userspace calls get_regs() in the middle of
6105                  * instruction emulation. Registers state needs to be copied
6106                  * back from emulation context to vcpu. Userspace shouldn't do
6107                  * that usually, but some bad designed PV devices (vmware
6108                  * backdoor interface) need this to work
6109                  */
6110                 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6111                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6112         }
6113         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6114         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6115         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6116         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6117         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6118         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6119         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6120         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6121 #ifdef CONFIG_X86_64
6122         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6123         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6124         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6125         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6126         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6127         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6128         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6129         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6130 #endif
6131
6132         regs->rip = kvm_rip_read(vcpu);
6133         regs->rflags = kvm_get_rflags(vcpu);
6134
6135         return 0;
6136 }
6137
6138 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6139 {
6140         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6141         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6142
6143         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6144         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6145         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6146         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6147         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6148         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6149         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6150         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6151 #ifdef CONFIG_X86_64
6152         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6153         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6154         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6155         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6156         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6157         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6158         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6159         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6160 #endif
6161
6162         kvm_rip_write(vcpu, regs->rip);
6163         kvm_set_rflags(vcpu, regs->rflags);
6164
6165         vcpu->arch.exception.pending = false;
6166
6167         kvm_make_request(KVM_REQ_EVENT, vcpu);
6168
6169         return 0;
6170 }
6171
6172 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6173 {
6174         struct kvm_segment cs;
6175
6176         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6177         *db = cs.db;
6178         *l = cs.l;
6179 }
6180 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6181
6182 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6183                                   struct kvm_sregs *sregs)
6184 {
6185         struct desc_ptr dt;
6186
6187         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6188         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6189         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6190         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6191         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6192         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6193
6194         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6195         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6196
6197         kvm_x86_ops->get_idt(vcpu, &dt);
6198         sregs->idt.limit = dt.size;
6199         sregs->idt.base = dt.address;
6200         kvm_x86_ops->get_gdt(vcpu, &dt);
6201         sregs->gdt.limit = dt.size;
6202         sregs->gdt.base = dt.address;
6203
6204         sregs->cr0 = kvm_read_cr0(vcpu);
6205         sregs->cr2 = vcpu->arch.cr2;
6206         sregs->cr3 = kvm_read_cr3(vcpu);
6207         sregs->cr4 = kvm_read_cr4(vcpu);
6208         sregs->cr8 = kvm_get_cr8(vcpu);
6209         sregs->efer = vcpu->arch.efer;
6210         sregs->apic_base = kvm_get_apic_base(vcpu);
6211
6212         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6213
6214         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6215                 set_bit(vcpu->arch.interrupt.nr,
6216                         (unsigned long *)sregs->interrupt_bitmap);
6217
6218         return 0;
6219 }
6220
6221 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6222                                     struct kvm_mp_state *mp_state)
6223 {
6224         kvm_apic_accept_events(vcpu);
6225         mp_state->mp_state = vcpu->arch.mp_state;
6226         return 0;
6227 }
6228
6229 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6230                                     struct kvm_mp_state *mp_state)
6231 {
6232         if (!kvm_vcpu_has_lapic(vcpu) &&
6233             mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6234                 return -EINVAL;
6235
6236         if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6237                 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6238                 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6239         } else
6240                 vcpu->arch.mp_state = mp_state->mp_state;
6241         kvm_make_request(KVM_REQ_EVENT, vcpu);
6242         return 0;
6243 }
6244
6245 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6246                     int reason, bool has_error_code, u32 error_code)
6247 {
6248         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6249         int ret;
6250
6251         init_emulate_ctxt(vcpu);
6252
6253         ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6254                                    has_error_code, error_code);
6255
6256         if (ret)
6257                 return EMULATE_FAIL;
6258
6259         kvm_rip_write(vcpu, ctxt->eip);
6260         kvm_set_rflags(vcpu, ctxt->eflags);
6261         kvm_make_request(KVM_REQ_EVENT, vcpu);
6262         return EMULATE_DONE;
6263 }
6264 EXPORT_SYMBOL_GPL(kvm_task_switch);
6265
6266 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6267                                   struct kvm_sregs *sregs)
6268 {
6269         int mmu_reset_needed = 0;
6270         int pending_vec, max_bits, idx;
6271         struct desc_ptr dt;
6272
6273         if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6274                 return -EINVAL;
6275
6276         dt.size = sregs->idt.limit;
6277         dt.address = sregs->idt.base;
6278         kvm_x86_ops->set_idt(vcpu, &dt);
6279         dt.size = sregs->gdt.limit;
6280         dt.address = sregs->gdt.base;
6281         kvm_x86_ops->set_gdt(vcpu, &dt);
6282
6283         vcpu->arch.cr2 = sregs->cr2;
6284         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6285         vcpu->arch.cr3 = sregs->cr3;
6286         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6287
6288         kvm_set_cr8(vcpu, sregs->cr8);
6289
6290         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6291         kvm_x86_ops->set_efer(vcpu, sregs->efer);
6292         kvm_set_apic_base(vcpu, sregs->apic_base);
6293
6294         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6295         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6296         vcpu->arch.cr0 = sregs->cr0;
6297
6298         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6299         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6300         if (sregs->cr4 & X86_CR4_OSXSAVE)
6301                 kvm_update_cpuid(vcpu);
6302
6303         idx = srcu_read_lock(&vcpu->kvm->srcu);
6304         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6305                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6306                 mmu_reset_needed = 1;
6307         }
6308         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6309
6310         if (mmu_reset_needed)
6311                 kvm_mmu_reset_context(vcpu);
6312
6313         max_bits = KVM_NR_INTERRUPTS;
6314         pending_vec = find_first_bit(
6315                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6316         if (pending_vec < max_bits) {
6317                 kvm_queue_interrupt(vcpu, pending_vec, false);
6318                 pr_debug("Set back pending irq %d\n", pending_vec);
6319         }
6320
6321         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6322         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6323         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6324         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6325         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6326         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6327
6328         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6329         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6330
6331         update_cr8_intercept(vcpu);
6332
6333         /* Older userspace won't unhalt the vcpu on reset. */
6334         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6335             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6336             !is_protmode(vcpu))
6337                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6338
6339         kvm_make_request(KVM_REQ_EVENT, vcpu);
6340
6341         return 0;
6342 }
6343
6344 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6345                                         struct kvm_guest_debug *dbg)
6346 {
6347         unsigned long rflags;
6348         int i, r;
6349
6350         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6351                 r = -EBUSY;
6352                 if (vcpu->arch.exception.pending)
6353                         goto out;
6354                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6355                         kvm_queue_exception(vcpu, DB_VECTOR);
6356                 else
6357                         kvm_queue_exception(vcpu, BP_VECTOR);
6358         }
6359
6360         /*
6361          * Read rflags as long as potentially injected trace flags are still
6362          * filtered out.
6363          */
6364         rflags = kvm_get_rflags(vcpu);
6365
6366         vcpu->guest_debug = dbg->control;
6367         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6368                 vcpu->guest_debug = 0;
6369
6370         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6371                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6372                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6373                 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6374         } else {
6375                 for (i = 0; i < KVM_NR_DB_REGS; i++)
6376                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6377         }
6378         kvm_update_dr7(vcpu);
6379
6380         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6381                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6382                         get_segment_base(vcpu, VCPU_SREG_CS);
6383
6384         /*
6385          * Trigger an rflags update that will inject or remove the trace
6386          * flags.
6387          */
6388         kvm_set_rflags(vcpu, rflags);
6389
6390         kvm_x86_ops->update_db_bp_intercept(vcpu);
6391
6392         r = 0;
6393
6394 out:
6395
6396         return r;
6397 }
6398
6399 /*
6400  * Translate a guest virtual address to a guest physical address.
6401  */
6402 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6403                                     struct kvm_translation *tr)
6404 {
6405         unsigned long vaddr = tr->linear_address;
6406         gpa_t gpa;
6407         int idx;
6408
6409         idx = srcu_read_lock(&vcpu->kvm->srcu);
6410         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6411         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6412         tr->physical_address = gpa;
6413         tr->valid = gpa != UNMAPPED_GVA;
6414         tr->writeable = 1;
6415         tr->usermode = 0;
6416
6417         return 0;
6418 }
6419
6420 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6421 {
6422         struct i387_fxsave_struct *fxsave =
6423                         &vcpu->arch.guest_fpu.state->fxsave;
6424
6425         memcpy(fpu->fpr, fxsave->st_space, 128);
6426         fpu->fcw = fxsave->cwd;
6427         fpu->fsw = fxsave->swd;
6428         fpu->ftwx = fxsave->twd;
6429         fpu->last_opcode = fxsave->fop;
6430         fpu->last_ip = fxsave->rip;
6431         fpu->last_dp = fxsave->rdp;
6432         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6433
6434         return 0;
6435 }
6436
6437 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6438 {
6439         struct i387_fxsave_struct *fxsave =
6440                         &vcpu->arch.guest_fpu.state->fxsave;
6441
6442         memcpy(fxsave->st_space, fpu->fpr, 128);
6443         fxsave->cwd = fpu->fcw;
6444         fxsave->swd = fpu->fsw;
6445         fxsave->twd = fpu->ftwx;
6446         fxsave->fop = fpu->last_opcode;
6447         fxsave->rip = fpu->last_ip;
6448         fxsave->rdp = fpu->last_dp;
6449         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6450
6451         return 0;
6452 }
6453
6454 int fx_init(struct kvm_vcpu *vcpu)
6455 {
6456         int err;
6457
6458         err = fpu_alloc(&vcpu->arch.guest_fpu);
6459         if (err)
6460                 return err;
6461
6462         fpu_finit(&vcpu->arch.guest_fpu);
6463
6464         /*
6465          * Ensure guest xcr0 is valid for loading
6466          */
6467         vcpu->arch.xcr0 = XSTATE_FP;
6468
6469         vcpu->arch.cr0 |= X86_CR0_ET;
6470
6471         return 0;
6472 }
6473 EXPORT_SYMBOL_GPL(fx_init);
6474
6475 static void fx_free(struct kvm_vcpu *vcpu)
6476 {
6477         fpu_free(&vcpu->arch.guest_fpu);
6478 }
6479
6480 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6481 {
6482         if (vcpu->guest_fpu_loaded)
6483                 return;
6484
6485         /*
6486          * Restore all possible states in the guest,
6487          * and assume host would use all available bits.
6488          * Guest xcr0 would be loaded later.
6489          */
6490         kvm_put_guest_xcr0(vcpu);
6491         vcpu->guest_fpu_loaded = 1;
6492         __kernel_fpu_begin();
6493         fpu_restore_checking(&vcpu->arch.guest_fpu);
6494         trace_kvm_fpu(1);
6495 }
6496
6497 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6498 {
6499         kvm_put_guest_xcr0(vcpu);
6500
6501         if (!vcpu->guest_fpu_loaded)
6502                 return;
6503
6504         vcpu->guest_fpu_loaded = 0;
6505         fpu_save_init(&vcpu->arch.guest_fpu);
6506         __kernel_fpu_end();
6507         ++vcpu->stat.fpu_reload;
6508         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6509         trace_kvm_fpu(0);
6510 }
6511
6512 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6513 {
6514         kvmclock_reset(vcpu);
6515
6516         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6517         fx_free(vcpu);
6518         kvm_x86_ops->vcpu_free(vcpu);
6519 }
6520
6521 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6522                                                 unsigned int id)
6523 {
6524         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6525                 printk_once(KERN_WARNING
6526                 "kvm: SMP vm created on host with unstable TSC; "
6527                 "guest TSC will not be reliable\n");
6528         return kvm_x86_ops->vcpu_create(kvm, id);
6529 }
6530
6531 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6532 {
6533         int r;
6534
6535         vcpu->arch.mtrr_state.have_fixed = 1;
6536         r = vcpu_load(vcpu);
6537         if (r)
6538                 return r;
6539         kvm_vcpu_reset(vcpu);
6540         r = kvm_mmu_setup(vcpu);
6541         vcpu_put(vcpu);
6542
6543         return r;
6544 }
6545
6546 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6547 {
6548         int r;
6549         struct msr_data msr;
6550
6551         r = vcpu_load(vcpu);
6552         if (r)
6553                 return r;
6554         msr.data = 0x0;
6555         msr.index = MSR_IA32_TSC;
6556         msr.host_initiated = true;
6557         kvm_write_tsc(vcpu, &msr);
6558         vcpu_put(vcpu);
6559
6560         return r;
6561 }
6562
6563 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6564 {
6565         int r;
6566         vcpu->arch.apf.msr_val = 0;
6567
6568         r = vcpu_load(vcpu);
6569         BUG_ON(r);
6570         kvm_mmu_unload(vcpu);
6571         vcpu_put(vcpu);
6572
6573         fx_free(vcpu);
6574         kvm_x86_ops->vcpu_free(vcpu);
6575 }
6576
6577 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6578 {
6579         atomic_set(&vcpu->arch.nmi_queued, 0);
6580         vcpu->arch.nmi_pending = 0;
6581         vcpu->arch.nmi_injected = false;
6582
6583         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6584         vcpu->arch.dr6 = DR6_FIXED_1;
6585         vcpu->arch.dr7 = DR7_FIXED_1;
6586         kvm_update_dr7(vcpu);
6587
6588         kvm_make_request(KVM_REQ_EVENT, vcpu);
6589         vcpu->arch.apf.msr_val = 0;
6590         vcpu->arch.st.msr_val = 0;
6591
6592         kvmclock_reset(vcpu);
6593
6594         kvm_clear_async_pf_completion_queue(vcpu);
6595         kvm_async_pf_hash_reset(vcpu);
6596         vcpu->arch.apf.halted = false;
6597
6598         kvm_pmu_reset(vcpu);
6599
6600         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6601         vcpu->arch.regs_avail = ~0;
6602         vcpu->arch.regs_dirty = ~0;
6603
6604         kvm_x86_ops->vcpu_reset(vcpu);
6605 }
6606
6607 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6608 {
6609         struct kvm_segment cs;
6610
6611         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6612         cs.selector = vector << 8;
6613         cs.base = vector << 12;
6614         kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6615         kvm_rip_write(vcpu, 0);
6616 }
6617
6618 int kvm_arch_hardware_enable(void)
6619 {
6620         struct kvm *kvm;
6621         struct kvm_vcpu *vcpu;
6622         int i;
6623         int ret;
6624         u64 local_tsc;
6625         u64 max_tsc = 0;
6626         bool stable, backwards_tsc = false;
6627
6628         kvm_shared_msr_cpu_online();
6629         ret = kvm_x86_ops->hardware_enable();
6630         if (ret != 0)
6631                 return ret;
6632
6633         local_tsc = native_read_tsc();
6634         stable = !check_tsc_unstable();
6635         list_for_each_entry(kvm, &vm_list, vm_list) {
6636                 kvm_for_each_vcpu(i, vcpu, kvm) {
6637                         if (!stable && vcpu->cpu == smp_processor_id())
6638                                 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6639                         if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6640                                 backwards_tsc = true;
6641                                 if (vcpu->arch.last_host_tsc > max_tsc)
6642                                         max_tsc = vcpu->arch.last_host_tsc;
6643                         }
6644                 }
6645         }
6646
6647         /*
6648          * Sometimes, even reliable TSCs go backwards.  This happens on
6649          * platforms that reset TSC during suspend or hibernate actions, but
6650          * maintain synchronization.  We must compensate.  Fortunately, we can
6651          * detect that condition here, which happens early in CPU bringup,
6652          * before any KVM threads can be running.  Unfortunately, we can't
6653          * bring the TSCs fully up to date with real time, as we aren't yet far
6654          * enough into CPU bringup that we know how much real time has actually
6655          * elapsed; our helper function, get_kernel_ns() will be using boot
6656          * variables that haven't been updated yet.
6657          *
6658          * So we simply find the maximum observed TSC above, then record the
6659          * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
6660          * the adjustment will be applied.  Note that we accumulate
6661          * adjustments, in case multiple suspend cycles happen before some VCPU
6662          * gets a chance to run again.  In the event that no KVM threads get a
6663          * chance to run, we will miss the entire elapsed period, as we'll have
6664          * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6665          * loose cycle time.  This isn't too big a deal, since the loss will be
6666          * uniform across all VCPUs (not to mention the scenario is extremely
6667          * unlikely). It is possible that a second hibernate recovery happens
6668          * much faster than a first, causing the observed TSC here to be
6669          * smaller; this would require additional padding adjustment, which is
6670          * why we set last_host_tsc to the local tsc observed here.
6671          *
6672          * N.B. - this code below runs only on platforms with reliable TSC,
6673          * as that is the only way backwards_tsc is set above.  Also note
6674          * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6675          * have the same delta_cyc adjustment applied if backwards_tsc
6676          * is detected.  Note further, this adjustment is only done once,
6677          * as we reset last_host_tsc on all VCPUs to stop this from being
6678          * called multiple times (one for each physical CPU bringup).
6679          *
6680          * Platforms with unreliable TSCs don't have to deal with this, they
6681          * will be compensated by the logic in vcpu_load, which sets the TSC to
6682          * catchup mode.  This will catchup all VCPUs to real time, but cannot
6683          * guarantee that they stay in perfect synchronization.
6684          */
6685         if (backwards_tsc) {
6686                 u64 delta_cyc = max_tsc - local_tsc;
6687                 list_for_each_entry(kvm, &vm_list, vm_list) {
6688                         kvm_for_each_vcpu(i, vcpu, kvm) {
6689                                 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6690                                 vcpu->arch.last_host_tsc = local_tsc;
6691                                 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6692                                         &vcpu->requests);
6693                         }
6694
6695                         /*
6696                          * We have to disable TSC offset matching.. if you were
6697                          * booting a VM while issuing an S4 host suspend....
6698                          * you may have some problem.  Solving this issue is
6699                          * left as an exercise to the reader.
6700                          */
6701                         kvm->arch.last_tsc_nsec = 0;
6702                         kvm->arch.last_tsc_write = 0;
6703                 }
6704
6705         }
6706         return 0;
6707 }
6708
6709 void kvm_arch_hardware_disable(void)
6710 {
6711         kvm_x86_ops->hardware_disable();
6712         drop_user_return_notifiers();
6713 }
6714
6715 int kvm_arch_hardware_setup(void)
6716 {
6717         return kvm_x86_ops->hardware_setup();
6718 }
6719
6720 void kvm_arch_hardware_unsetup(void)
6721 {
6722         kvm_x86_ops->hardware_unsetup();
6723 }
6724
6725 void kvm_arch_check_processor_compat(void *rtn)
6726 {
6727         kvm_x86_ops->check_processor_compatibility(rtn);
6728 }
6729
6730 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6731 {
6732         return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6733 }
6734
6735 struct static_key kvm_no_apic_vcpu __read_mostly;
6736
6737 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6738 {
6739         struct page *page;
6740         struct kvm *kvm;
6741         int r;
6742
6743         BUG_ON(vcpu->kvm == NULL);
6744         kvm = vcpu->kvm;
6745
6746         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6747         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6748                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6749         else
6750                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6751
6752         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6753         if (!page) {
6754                 r = -ENOMEM;
6755                 goto fail;
6756         }
6757         vcpu->arch.pio_data = page_address(page);
6758
6759         kvm_set_tsc_khz(vcpu, max_tsc_khz);
6760
6761         r = kvm_mmu_create(vcpu);
6762         if (r < 0)
6763                 goto fail_free_pio_data;
6764
6765         if (irqchip_in_kernel(kvm)) {
6766                 r = kvm_create_lapic(vcpu);
6767                 if (r < 0)
6768                         goto fail_mmu_destroy;
6769         } else
6770                 static_key_slow_inc(&kvm_no_apic_vcpu);
6771
6772         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6773                                        GFP_KERNEL);
6774         if (!vcpu->arch.mce_banks) {
6775                 r = -ENOMEM;
6776                 goto fail_free_lapic;
6777         }
6778         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6779
6780         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
6781                 r = -ENOMEM;
6782                 goto fail_free_mce_banks;
6783         }
6784
6785         r = fx_init(vcpu);
6786         if (r)
6787                 goto fail_free_wbinvd_dirty_mask;
6788
6789         vcpu->arch.ia32_tsc_adjust_msr = 0x0;
6790         vcpu->arch.pv_time_enabled = false;
6791         kvm_async_pf_hash_reset(vcpu);
6792         kvm_pmu_init(vcpu);
6793
6794         return 0;
6795 fail_free_wbinvd_dirty_mask:
6796         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6797 fail_free_mce_banks:
6798         kfree(vcpu->arch.mce_banks);
6799 fail_free_lapic:
6800         kvm_free_lapic(vcpu);
6801 fail_mmu_destroy:
6802         kvm_mmu_destroy(vcpu);
6803 fail_free_pio_data:
6804         free_page((unsigned long)vcpu->arch.pio_data);
6805 fail:
6806         return r;
6807 }
6808
6809 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6810 {
6811         int idx;
6812
6813         kvm_pmu_destroy(vcpu);
6814         kfree(vcpu->arch.mce_banks);
6815         kvm_free_lapic(vcpu);
6816         idx = srcu_read_lock(&vcpu->kvm->srcu);
6817         kvm_mmu_destroy(vcpu);
6818         srcu_read_unlock(&vcpu->kvm->srcu, idx);
6819         free_page((unsigned long)vcpu->arch.pio_data);
6820         if (!irqchip_in_kernel(vcpu->kvm))
6821                 static_key_slow_dec(&kvm_no_apic_vcpu);
6822 }
6823
6824 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
6825 {
6826 }
6827
6828 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6829 {
6830         if (type)
6831                 return -EINVAL;
6832
6833         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6834         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6835
6836         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6837         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6838         /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6839         set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6840                 &kvm->arch.irq_sources_bitmap);
6841
6842         raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6843         mutex_init(&kvm->arch.apic_map_lock);
6844         spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6845
6846         pvclock_update_vm_gtod_copy(kvm);
6847
6848         return 0;
6849 }
6850
6851 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6852 {
6853         int r;
6854         r = vcpu_load(vcpu);
6855         BUG_ON(r);
6856         kvm_mmu_unload(vcpu);
6857         vcpu_put(vcpu);
6858 }
6859
6860 static void kvm_free_vcpus(struct kvm *kvm)
6861 {
6862         unsigned int i;
6863         struct kvm_vcpu *vcpu;
6864
6865         /*
6866          * Unpin any mmu pages first.
6867          */
6868         kvm_for_each_vcpu(i, vcpu, kvm) {
6869                 kvm_clear_async_pf_completion_queue(vcpu);
6870                 kvm_unload_vcpu_mmu(vcpu);
6871         }
6872         kvm_for_each_vcpu(i, vcpu, kvm)
6873                 kvm_arch_vcpu_free(vcpu);
6874
6875         mutex_lock(&kvm->lock);
6876         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6877                 kvm->vcpus[i] = NULL;
6878
6879         atomic_set(&kvm->online_vcpus, 0);
6880         mutex_unlock(&kvm->lock);
6881 }
6882
6883 void kvm_arch_sync_events(struct kvm *kvm)
6884 {
6885         kvm_free_all_assigned_devices(kvm);
6886         kvm_free_pit(kvm);
6887 }
6888
6889 void kvm_arch_destroy_vm(struct kvm *kvm)
6890 {
6891         if (current->mm == kvm->mm) {
6892                 /*
6893                  * Free memory regions allocated on behalf of userspace,
6894                  * unless the the memory map has changed due to process exit
6895                  * or fd copying.
6896                  */
6897                 struct kvm_userspace_memory_region mem;
6898                 memset(&mem, 0, sizeof(mem));
6899                 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
6900                 kvm_set_memory_region(kvm, &mem);
6901
6902                 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
6903                 kvm_set_memory_region(kvm, &mem);
6904
6905                 mem.slot = TSS_PRIVATE_MEMSLOT;
6906                 kvm_set_memory_region(kvm, &mem);
6907         }
6908         kvm_iommu_unmap_guest(kvm);
6909         kfree(kvm->arch.vpic);
6910         kfree(kvm->arch.vioapic);
6911         kvm_free_vcpus(kvm);
6912         if (kvm->arch.apic_access_page)
6913                 put_page(kvm->arch.apic_access_page);
6914         if (kvm->arch.ept_identity_pagetable)
6915                 put_page(kvm->arch.ept_identity_pagetable);
6916         kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
6917 }
6918
6919 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
6920                            struct kvm_memory_slot *dont)
6921 {
6922         int i;
6923
6924         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6925                 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6926                         kvm_kvfree(free->arch.rmap[i]);
6927                         free->arch.rmap[i] = NULL;
6928                 }
6929                 if (i == 0)
6930                         continue;
6931
6932                 if (!dont || free->arch.lpage_info[i - 1] !=
6933                              dont->arch.lpage_info[i - 1]) {
6934                         kvm_kvfree(free->arch.lpage_info[i - 1]);
6935                         free->arch.lpage_info[i - 1] = NULL;
6936                 }
6937         }
6938 }
6939
6940 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
6941                             unsigned long npages)
6942 {
6943         int i;
6944
6945         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6946                 unsigned long ugfn;
6947                 int lpages;
6948                 int level = i + 1;
6949
6950                 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6951                                       slot->base_gfn, level) + 1;
6952
6953                 slot->arch.rmap[i] =
6954                         kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6955                 if (!slot->arch.rmap[i])
6956                         goto out_free;
6957                 if (i == 0)
6958                         continue;
6959
6960                 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6961                                         sizeof(*slot->arch.lpage_info[i - 1]));
6962                 if (!slot->arch.lpage_info[i - 1])
6963                         goto out_free;
6964
6965                 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6966                         slot->arch.lpage_info[i - 1][0].write_count = 1;
6967                 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6968                         slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
6969                 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6970                 /*
6971                  * If the gfn and userspace address are not aligned wrt each
6972                  * other, or if explicitly asked to, disable large page
6973                  * support for this slot
6974                  */
6975                 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6976                     !kvm_largepages_enabled()) {
6977                         unsigned long j;
6978
6979                         for (j = 0; j < lpages; ++j)
6980                                 slot->arch.lpage_info[i - 1][j].write_count = 1;
6981                 }
6982         }
6983
6984         return 0;
6985
6986 out_free:
6987         for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6988                 kvm_kvfree(slot->arch.rmap[i]);
6989                 slot->arch.rmap[i] = NULL;
6990                 if (i == 0)
6991                         continue;
6992
6993                 kvm_kvfree(slot->arch.lpage_info[i - 1]);
6994                 slot->arch.lpage_info[i - 1] = NULL;
6995         }
6996         return -ENOMEM;
6997 }
6998
6999 void kvm_arch_memslots_updated(struct kvm *kvm)
7000 {
7001 }
7002
7003 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7004                                 struct kvm_memory_slot *memslot,
7005                                 struct kvm_userspace_memory_region *mem,
7006                                 enum kvm_mr_change change)
7007 {
7008         /*
7009          * Only private memory slots need to be mapped here since
7010          * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7011          */
7012         if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7013                 unsigned long userspace_addr;
7014
7015                 /*
7016                  * MAP_SHARED to prevent internal slot pages from being moved
7017                  * by fork()/COW.
7018                  */
7019                 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7020                                          PROT_READ | PROT_WRITE,
7021                                          MAP_SHARED | MAP_ANONYMOUS, 0);
7022
7023                 if (IS_ERR((void *)userspace_addr))
7024                         return PTR_ERR((void *)userspace_addr);
7025
7026                 memslot->userspace_addr = userspace_addr;
7027         }
7028
7029         return 0;
7030 }
7031
7032 void kvm_arch_commit_memory_region(struct kvm *kvm,
7033                                 struct kvm_userspace_memory_region *mem,
7034                                 const struct kvm_memory_slot *old,
7035                                 enum kvm_mr_change change)
7036 {
7037
7038         int nr_mmu_pages = 0;
7039
7040         if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7041                 int ret;
7042
7043                 ret = vm_munmap(old->userspace_addr,
7044                                 old->npages * PAGE_SIZE);
7045                 if (ret < 0)
7046                         printk(KERN_WARNING
7047                                "kvm_vm_ioctl_set_memory_region: "
7048                                "failed to munmap memory\n");
7049         }
7050
7051         if (!kvm->arch.n_requested_mmu_pages)
7052                 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7053
7054         if (nr_mmu_pages)
7055                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7056         /*
7057          * Write protect all pages for dirty logging.
7058          * Existing largepage mappings are destroyed here and new ones will
7059          * not be created until the end of the logging.
7060          */
7061         if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7062                 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7063         /*
7064          * If memory slot is created, or moved, we need to clear all
7065          * mmio sptes.
7066          */
7067         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
7068                 kvm_mmu_zap_mmio_sptes(kvm);
7069                 kvm_reload_remote_mmus(kvm);
7070         }
7071 }
7072
7073 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7074 {
7075         kvm_mmu_zap_all(kvm);
7076         kvm_reload_remote_mmus(kvm);
7077 }
7078
7079 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7080                                    struct kvm_memory_slot *slot)
7081 {
7082         kvm_arch_flush_shadow_all(kvm);
7083 }
7084
7085 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7086 {
7087         return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7088                 !vcpu->arch.apf.halted)
7089                 || !list_empty_careful(&vcpu->async_pf.done)
7090                 || kvm_apic_has_events(vcpu)
7091                 || atomic_read(&vcpu->arch.nmi_queued) ||
7092                 (kvm_arch_interrupt_allowed(vcpu) &&
7093                  kvm_cpu_has_interrupt(vcpu));
7094 }
7095
7096 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7097 {
7098         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7099 }
7100
7101 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7102 {
7103         return kvm_x86_ops->interrupt_allowed(vcpu);
7104 }
7105
7106 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7107 {
7108         unsigned long current_rip = kvm_rip_read(vcpu) +
7109                 get_segment_base(vcpu, VCPU_SREG_CS);
7110
7111         return current_rip == linear_rip;
7112 }
7113 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7114
7115 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7116 {
7117         unsigned long rflags;
7118
7119         rflags = kvm_x86_ops->get_rflags(vcpu);
7120         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7121                 rflags &= ~X86_EFLAGS_TF;
7122         return rflags;
7123 }
7124 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7125
7126 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7127 {
7128         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7129             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7130                 rflags |= X86_EFLAGS_TF;
7131         kvm_x86_ops->set_rflags(vcpu, rflags);
7132         kvm_make_request(KVM_REQ_EVENT, vcpu);
7133 }
7134 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7135
7136 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7137 {
7138         int r;
7139
7140         if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7141               work->wakeup_all)
7142                 return;
7143
7144         r = kvm_mmu_reload(vcpu);
7145         if (unlikely(r))
7146                 return;
7147
7148         if (!vcpu->arch.mmu.direct_map &&
7149               work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7150                 return;
7151
7152         vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7153 }
7154
7155 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7156 {
7157         return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7158 }
7159
7160 static inline u32 kvm_async_pf_next_probe(u32 key)
7161 {
7162         return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7163 }
7164
7165 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7166 {
7167         u32 key = kvm_async_pf_hash_fn(gfn);
7168
7169         while (vcpu->arch.apf.gfns[key] != ~0)
7170                 key = kvm_async_pf_next_probe(key);
7171
7172         vcpu->arch.apf.gfns[key] = gfn;
7173 }
7174
7175 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7176 {
7177         int i;
7178         u32 key = kvm_async_pf_hash_fn(gfn);
7179
7180         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7181                      (vcpu->arch.apf.gfns[key] != gfn &&
7182                       vcpu->arch.apf.gfns[key] != ~0); i++)
7183                 key = kvm_async_pf_next_probe(key);
7184
7185         return key;
7186 }
7187
7188 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7189 {
7190         return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7191 }
7192
7193 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7194 {
7195         u32 i, j, k;
7196
7197         i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7198         while (true) {
7199                 vcpu->arch.apf.gfns[i] = ~0;
7200                 do {
7201                         j = kvm_async_pf_next_probe(j);
7202                         if (vcpu->arch.apf.gfns[j] == ~0)
7203                                 return;
7204                         k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7205                         /*
7206                          * k lies cyclically in ]i,j]
7207                          * |    i.k.j |
7208                          * |....j i.k.| or  |.k..j i...|
7209                          */
7210                 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7211                 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7212                 i = j;
7213         }
7214 }
7215
7216 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7217 {
7218
7219         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7220                                       sizeof(val));
7221 }
7222
7223 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7224                                      struct kvm_async_pf *work)
7225 {
7226         struct x86_exception fault;
7227
7228         trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7229         kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7230
7231         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7232             (vcpu->arch.apf.send_user_only &&
7233              kvm_x86_ops->get_cpl(vcpu) == 0))
7234                 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7235         else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7236                 fault.vector = PF_VECTOR;
7237                 fault.error_code_valid = true;
7238                 fault.error_code = 0;
7239                 fault.nested_page_fault = false;
7240                 fault.address = work->arch.token;
7241                 kvm_inject_page_fault(vcpu, &fault);
7242         }
7243 }
7244
7245 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7246                                  struct kvm_async_pf *work)
7247 {
7248         struct x86_exception fault;
7249
7250         trace_kvm_async_pf_ready(work->arch.token, work->gva);
7251         if (work->wakeup_all)
7252                 work->arch.token = ~0; /* broadcast wakeup */
7253         else
7254                 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7255
7256         if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7257             !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7258                 fault.vector = PF_VECTOR;
7259                 fault.error_code_valid = true;
7260                 fault.error_code = 0;
7261                 fault.nested_page_fault = false;
7262                 fault.address = work->arch.token;
7263                 kvm_inject_page_fault(vcpu, &fault);
7264         }
7265         vcpu->arch.apf.halted = false;
7266         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7267 }
7268
7269 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7270 {
7271         if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7272                 return true;
7273         else
7274                 return !kvm_event_needs_reinjection(vcpu) &&
7275                         kvm_x86_ops->interrupt_allowed(vcpu);
7276 }
7277
7278 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7279 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7280 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7281 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7282 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7283 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7284 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7285 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7286 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7287 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7288 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7289 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);