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  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *   Amit Shah    <amit.shah@qumranet.com>
  14  *   Ben-Ami Yassour <benami@il.ibm.com>
  15  *
  16  * This work is licensed under the terms of the GNU GPL, version 2.  See
  17  * the COPYING file in the top-level directory.
  18  *
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "i8254.h"
  25 #include "tss.h"
  26 #include "kvm_cache_regs.h"
  27 #include "x86.h"
  28
  29 #include <linux/clocksource.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/kvm.h>
  32 #include <linux/fs.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/module.h>
  35 #include <linux/mman.h>
  36 #include <linux/highmem.h>
  37 #include <linux/iommu.h>
  38 #include <linux/intel-iommu.h>
  39 #include <linux/cpufreq.h>
  40 #include <linux/user-return-notifier.h>
  41 #include <linux/srcu.h>
  42 #include <linux/slab.h>
  43 #include <linux/perf_event.h>
  44 #include <trace/events/kvm.h>
  45
  46 #define CREATE_TRACE_POINTS
  47 #include "trace.h"
  48
  49 #include <asm/debugreg.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/msr.h>
  52 #include <asm/desc.h>
  53 #include <asm/mtrr.h>
  54 #include <asm/mce.h>
  55 #include <asm/i387.h>
  56 #include <asm/xcr.h>
  57
  58 #define MAX_IO_MSRS 256
  59 #define CR0_RESERVED_BITS                                               \
  60         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  61                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  62                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  63 #define CR4_RESERVED_BITS                                               \
  64         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  65                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  66                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  67                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  68
  69 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  70
  71 #define KVM_MAX_MCE_BANKS 32
  72 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  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 u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  80 #else
  81 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  82 #endif
  83
  84 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  85 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  86
  87 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  88 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  89                                     struct kvm_cpuid_entry2 __user *entries);
  90
  91 struct kvm_x86_ops *kvm_x86_ops;
  92 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  93
  94 int ignore_msrs = 0;
  95 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  96
  97 #define KVM_NR_SHARED_MSRS 16
  98
  99 struct kvm_shared_msrs_global {
 100         int nr;
 101         u32 msrs[KVM_NR_SHARED_MSRS];
 102 };
 103
 104 struct kvm_shared_msrs {
 105         struct user_return_notifier urn;
 106         bool registered;
 107         struct kvm_shared_msr_values {
 108                 u64 host;
 109                 u64 curr;
 110         } values[KVM_NR_SHARED_MSRS];
 111 };
 112
 113 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 114 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 115
 116 struct kvm_stats_debugfs_item debugfs_entries[] = {
 117         { "pf_fixed", VCPU_STAT(pf_fixed) },
 118         { "pf_guest", VCPU_STAT(pf_guest) },
 119         { "tlb_flush", VCPU_STAT(tlb_flush) },
 120         { "invlpg", VCPU_STAT(invlpg) },
 121         { "exits", VCPU_STAT(exits) },
 122         { "io_exits", VCPU_STAT(io_exits) },
 123         { "mmio_exits", VCPU_STAT(mmio_exits) },
 124         { "signal_exits", VCPU_STAT(signal_exits) },
 125         { "irq_window", VCPU_STAT(irq_window_exits) },
 126         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 127         { "halt_exits", VCPU_STAT(halt_exits) },
 128         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 129         { "hypercalls", VCPU_STAT(hypercalls) },
 130         { "request_irq", VCPU_STAT(request_irq_exits) },
 131         { "irq_exits", VCPU_STAT(irq_exits) },
 132         { "host_state_reload", VCPU_STAT(host_state_reload) },
 133         { "efer_reload", VCPU_STAT(efer_reload) },
 134         { "fpu_reload", VCPU_STAT(fpu_reload) },
 135         { "insn_emulation", VCPU_STAT(insn_emulation) },
 136         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 137         { "irq_injections", VCPU_STAT(irq_injections) },
 138         { "nmi_injections", VCPU_STAT(nmi_injections) },
 139         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 140         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 141         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 142         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 143         { "mmu_flooded", VM_STAT(mmu_flooded) },
 144         { "mmu_recycled", VM_STAT(mmu_recycled) },
 145         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 146         { "mmu_unsync", VM_STAT(mmu_unsync) },
 147         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 148         { "largepages", VM_STAT(lpages) },
 149         { NULL }
 150 };
 151
 152 static void kvm_on_user_return(struct user_return_notifier *urn)
 153 {
 154         unsigned slot;
 155         struct kvm_shared_msrs *locals
 156                 = container_of(urn, struct kvm_shared_msrs, urn);
 157         struct kvm_shared_msr_values *values;
 158
 159         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 160                 values = &locals->values[slot];
 161                 if (values->host != values->curr) {
 162                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 163                         values->curr = values->host;
 164                 }
 165         }
 166         locals->registered = false;
 167         user_return_notifier_unregister(urn);
 168 }
 169
 170 static void shared_msr_update(unsigned slot, u32 msr)
 171 {
 172         struct kvm_shared_msrs *smsr;
 173         u64 value;
 174
 175         smsr = &__get_cpu_var(shared_msrs);
 176         /* only read, and nobody should modify it at this time,
 177          * so don't need lock */
 178         if (slot >= shared_msrs_global.nr) {
 179                 printk(KERN_ERR "kvm: invalid MSR slot!");
 180                 return;
 181         }
 182         rdmsrl_safe(msr, &value);
 183         smsr->values[slot].host = value;
 184         smsr->values[slot].curr = value;
 185 }
 186
 187 void kvm_define_shared_msr(unsigned slot, u32 msr)
 188 {
 189         if (slot >= shared_msrs_global.nr)
 190                 shared_msrs_global.nr = slot + 1;
 191         shared_msrs_global.msrs[slot] = msr;
 192         /* we need ensured the shared_msr_global have been updated */
 193         smp_wmb();
 194 }
 195 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 196
 197 static void kvm_shared_msr_cpu_online(void)
 198 {
 199         unsigned i;
 200
 201         for (i = 0; i < shared_msrs_global.nr; ++i)
 202                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 203 }
 204
 205 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 206 {
 207         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 208
 209         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 210                 return;
 211         smsr->values[slot].curr = value;
 212         wrmsrl(shared_msrs_global.msrs[slot], value);
 213         if (!smsr->registered) {
 214                 smsr->urn.on_user_return = kvm_on_user_return;
 215                 user_return_notifier_register(&smsr->urn);
 216                 smsr->registered = true;
 217         }
 218 }
 219 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 220
 221 static void drop_user_return_notifiers(void *ignore)
 222 {
 223         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 224
 225         if (smsr->registered)
 226                 kvm_on_user_return(&smsr->urn);
 227 }
 228
 229 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 230 {
 231         if (irqchip_in_kernel(vcpu->kvm))
 232                 return vcpu->arch.apic_base;
 233         else
 234                 return vcpu->arch.apic_base;
 235 }
 236 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 237
 238 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 239 {
 240         /* TODO: reserve bits check */
 241         if (irqchip_in_kernel(vcpu->kvm))
 242                 kvm_lapic_set_base(vcpu, data);
 243         else
 244                 vcpu->arch.apic_base = data;
 245 }
 246 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 247
 248 #define EXCPT_BENIGN            0
 249 #define EXCPT_CONTRIBUTORY      1
 250 #define EXCPT_PF                2
 251
 252 static int exception_class(int vector)
 253 {
 254         switch (vector) {
 255         case PF_VECTOR:
 256                 return EXCPT_PF;
 257         case DE_VECTOR:
 258         case TS_VECTOR:
 259         case NP_VECTOR:
 260         case SS_VECTOR:
 261         case GP_VECTOR:
 262                 return EXCPT_CONTRIBUTORY;
 263         default:
 264                 break;
 265         }
 266         return EXCPT_BENIGN;
 267 }
 268
 269 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 270                 unsigned nr, bool has_error, u32 error_code,
 271                 bool reinject)
 272 {
 273         u32 prev_nr;
 274         int class1, class2;
 275
 276         if (!vcpu->arch.exception.pending) {
 277         queue:
 278                 vcpu->arch.exception.pending = true;
 279                 vcpu->arch.exception.has_error_code = has_error;
 280                 vcpu->arch.exception.nr = nr;
 281                 vcpu->arch.exception.error_code = error_code;
 282                 vcpu->arch.exception.reinject = reinject;
 283                 return;
 284         }
 285
 286         /* to check exception */
 287         prev_nr = vcpu->arch.exception.nr;
 288         if (prev_nr == DF_VECTOR) {
 289                 /* triple fault -> shutdown */
 290                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 291                 return;
 292         }
 293         class1 = exception_class(prev_nr);
 294         class2 = exception_class(nr);
 295         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 296                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 297                 /* generate double fault per SDM Table 5-5 */
 298                 vcpu->arch.exception.pending = true;
 299                 vcpu->arch.exception.has_error_code = true;
 300                 vcpu->arch.exception.nr = DF_VECTOR;
 301                 vcpu->arch.exception.error_code = 0;
 302         } else
 303                 /* replace previous exception with a new one in a hope
 304                    that instruction re-execution will regenerate lost
 305                    exception */
 306                 goto queue;
 307 }
 308
 309 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 310 {
 311         kvm_multiple_exception(vcpu, nr, false, 0, false);
 312 }
 313 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 314
 315 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 316 {
 317         kvm_multiple_exception(vcpu, nr, false, 0, true);
 318 }
 319 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 320
 321 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 322                            u32 error_code)
 323 {
 324         ++vcpu->stat.pf_guest;
 325         vcpu->arch.cr2 = addr;
 326         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 327 }
 328
 329 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 330 {
 331         vcpu->arch.nmi_pending = 1;
 332 }
 333 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 334
 335 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 336 {
 337         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 338 }
 339 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 340
 341 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 342 {
 343         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 344 }
 345 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 346
 347 /*
 348  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 349  * a #GP and return false.
 350  */
 351 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 352 {
 353         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 354                 return true;
 355         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 356         return false;
 357 }
 358 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 359
 360 /*
 361  * Load the pae pdptrs.  Return true is they are all valid.
 362  */
 363 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 364 {
 365         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 366         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 367         int i;
 368         int ret;
 369         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 370
 371         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 372                                   offset * sizeof(u64), sizeof(pdpte));
 373         if (ret < 0) {
 374                 ret = 0;
 375                 goto out;
 376         }
 377         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 378                 if (is_present_gpte(pdpte[i]) &&
 379                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 380                         ret = 0;
 381                         goto out;
 382                 }
 383         }
 384         ret = 1;
 385
 386         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 387         __set_bit(VCPU_EXREG_PDPTR,
 388                   (unsigned long *)&vcpu->arch.regs_avail);
 389         __set_bit(VCPU_EXREG_PDPTR,
 390                   (unsigned long *)&vcpu->arch.regs_dirty);
 391 out:
 392
 393         return ret;
 394 }
 395 EXPORT_SYMBOL_GPL(load_pdptrs);
 396
 397 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 398 {
 399         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 400         bool changed = true;
 401         int r;
 402
 403         if (is_long_mode(vcpu) || !is_pae(vcpu))
 404                 return false;
 405
 406         if (!test_bit(VCPU_EXREG_PDPTR,
 407                       (unsigned long *)&vcpu->arch.regs_avail))
 408                 return true;
 409
 410         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 411         if (r < 0)
 412                 goto out;
 413         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 414 out:
 415
 416         return changed;
 417 }
 418
 419 static int __kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 420 {
 421         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 422         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 423                                     X86_CR0_CD | X86_CR0_NW;
 424
 425         cr0 |= X86_CR0_ET;
 426
 427 #ifdef CONFIG_X86_64
 428         if (cr0 & 0xffffffff00000000UL)
 429                 return 1;
 430 #endif
 431
 432         cr0 &= ~CR0_RESERVED_BITS;
 433
 434         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 435                 return 1;
 436
 437         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 438                 return 1;
 439
 440         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 441 #ifdef CONFIG_X86_64
 442                 if ((vcpu->arch.efer & EFER_LME)) {
 443                         int cs_db, cs_l;
 444
 445                         if (!is_pae(vcpu))
 446                                 return 1;
 447                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 448                         if (cs_l)
 449                                 return 1;
 450                 } else
 451 #endif
 452                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3))
 453                         return 1;
 454         }
 455
 456         kvm_x86_ops->set_cr0(vcpu, cr0);
 457
 458         if ((cr0 ^ old_cr0) & update_bits)
 459                 kvm_mmu_reset_context(vcpu);
 460         return 0;
 461 }
 462
 463 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 464 {
 465         if (__kvm_set_cr0(vcpu, cr0))
 466                 kvm_inject_gp(vcpu, 0);
 467 }
 468 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 469
 470 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 471 {
 472         kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 473 }
 474 EXPORT_SYMBOL_GPL(kvm_lmsw);
 475
 476 int __kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 477 {
 478         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 479         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 480
 481         if (cr4 & CR4_RESERVED_BITS)
 482                 return 1;
 483
 484         if (is_long_mode(vcpu)) {
 485                 if (!(cr4 & X86_CR4_PAE))
 486                         return 1;
 487         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 488                    && ((cr4 ^ old_cr4) & pdptr_bits)
 489                    && !load_pdptrs(vcpu, vcpu->arch.cr3))
 490                 return 1;
 491
 492         if (cr4 & X86_CR4_VMXE)
 493                 return 1;
 494
 495         kvm_x86_ops->set_cr4(vcpu, cr4);
 496
 497         if ((cr4 ^ old_cr4) & pdptr_bits)
 498                 kvm_mmu_reset_context(vcpu);
 499
 500         return 0;
 501 }
 502
 503 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 504 {
 505         if (__kvm_set_cr4(vcpu, cr4))
 506                 kvm_inject_gp(vcpu, 0);
 507 }
 508 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 509
 510 static int __kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 511 {
 512         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 513                 kvm_mmu_sync_roots(vcpu);
 514                 kvm_mmu_flush_tlb(vcpu);
 515                 return 0;
 516         }
 517
 518         if (is_long_mode(vcpu)) {
 519                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 520                         return 1;
 521         } else {
 522                 if (is_pae(vcpu)) {
 523                         if (cr3 & CR3_PAE_RESERVED_BITS)
 524                                 return 1;
 525                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3))
 526                                 return 1;
 527                 }
 528                 /*
 529                  * We don't check reserved bits in nonpae mode, because
 530                  * this isn't enforced, and VMware depends on this.
 531                  */
 532         }
 533
 534         /*
 535          * Does the new cr3 value map to physical memory? (Note, we
 536          * catch an invalid cr3 even in real-mode, because it would
 537          * cause trouble later on when we turn on paging anyway.)
 538          *
 539          * A real CPU would silently accept an invalid cr3 and would
 540          * attempt to use it - with largely undefined (and often hard
 541          * to debug) behavior on the guest side.
 542          */
 543         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 544                 return 1;
 545         vcpu->arch.cr3 = cr3;
 546         vcpu->arch.mmu.new_cr3(vcpu);
 547         return 0;
 548 }
 549
 550 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 551 {
 552         if (__kvm_set_cr3(vcpu, cr3))
 553                 kvm_inject_gp(vcpu, 0);
 554 }
 555 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 556
 557 int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 558 {
 559         if (cr8 & CR8_RESERVED_BITS)
 560                 return 1;
 561         if (irqchip_in_kernel(vcpu->kvm))
 562                 kvm_lapic_set_tpr(vcpu, cr8);
 563         else
 564                 vcpu->arch.cr8 = cr8;
 565         return 0;
 566 }
 567
 568 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 569 {
 570         if (__kvm_set_cr8(vcpu, cr8))
 571                 kvm_inject_gp(vcpu, 0);
 572 }
 573 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 574
 575 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 576 {
 577         if (irqchip_in_kernel(vcpu->kvm))
 578                 return kvm_lapic_get_cr8(vcpu);
 579         else
 580                 return vcpu->arch.cr8;
 581 }
 582 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 583
 584 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 585 {
 586         switch (dr) {
 587         case 0 ... 3:
 588                 vcpu->arch.db[dr] = val;
 589                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 590                         vcpu->arch.eff_db[dr] = val;
 591                 break;
 592         case 4:
 593                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 594                         return 1; /* #UD */
 595                 /* fall through */
 596         case 6:
 597                 if (val & 0xffffffff00000000ULL)
 598                         return -1; /* #GP */
 599                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 600                 break;
 601         case 5:
 602                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 603                         return 1; /* #UD */
 604                 /* fall through */
 605         default: /* 7 */
 606                 if (val & 0xffffffff00000000ULL)
 607                         return -1; /* #GP */
 608                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 609                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 610                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
 611                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
 612                 }
 613                 break;
 614         }
 615
 616         return 0;
 617 }
 618
 619 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 620 {
 621         int res;
 622
 623         res = __kvm_set_dr(vcpu, dr, val);
 624         if (res > 0)
 625                 kvm_queue_exception(vcpu, UD_VECTOR);
 626         else if (res < 0)
 627                 kvm_inject_gp(vcpu, 0);
 628
 629         return res;
 630 }
 631 EXPORT_SYMBOL_GPL(kvm_set_dr);
 632
 633 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 634 {
 635         switch (dr) {
 636         case 0 ... 3:
 637                 *val = vcpu->arch.db[dr];
 638                 break;
 639         case 4:
 640                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 641                         return 1;
 642                 /* fall through */
 643         case 6:
 644                 *val = vcpu->arch.dr6;
 645                 break;
 646         case 5:
 647                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 648                         return 1;
 649                 /* fall through */
 650         default: /* 7 */
 651                 *val = vcpu->arch.dr7;
 652                 break;
 653         }
 654
 655         return 0;
 656 }
 657
 658 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 659 {
 660         if (_kvm_get_dr(vcpu, dr, val)) {
 661                 kvm_queue_exception(vcpu, UD_VECTOR);
 662                 return 1;
 663         }
 664         return 0;
 665 }
 666 EXPORT_SYMBOL_GPL(kvm_get_dr);
 667
 668 static inline u32 bit(int bitno)
 669 {
 670         return 1 << (bitno & 31);
 671 }
 672
 673 /*
 674  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 675  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 676  *
 677  * This list is modified at module load time to reflect the
 678  * capabilities of the host cpu. This capabilities test skips MSRs that are
 679  * kvm-specific. Those are put in the beginning of the list.
 680  */
 681
 682 #define KVM_SAVE_MSRS_BEGIN     7
 683 static u32 msrs_to_save[] = {
 684         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 685         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 686         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 687         HV_X64_MSR_APIC_ASSIST_PAGE,
 688         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 689         MSR_K6_STAR,
 690 #ifdef CONFIG_X86_64
 691         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 692 #endif
 693         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 694 };
 695
 696 static unsigned num_msrs_to_save;
 697
 698 static u32 emulated_msrs[] = {
 699         MSR_IA32_MISC_ENABLE,
 700 };
 701
 702 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 703 {
 704         u64 old_efer = vcpu->arch.efer;
 705
 706         if (efer & efer_reserved_bits)
 707                 return 1;
 708
 709         if (is_paging(vcpu)
 710             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 711                 return 1;
 712
 713         if (efer & EFER_FFXSR) {
 714                 struct kvm_cpuid_entry2 *feat;
 715
 716                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 717                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 718                         return 1;
 719         }
 720
 721         if (efer & EFER_SVME) {
 722                 struct kvm_cpuid_entry2 *feat;
 723
 724                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 725                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 726                         return 1;
 727         }
 728
 729         efer &= ~EFER_LMA;
 730         efer |= vcpu->arch.efer & EFER_LMA;
 731
 732         kvm_x86_ops->set_efer(vcpu, efer);
 733
 734         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 735         kvm_mmu_reset_context(vcpu);
 736
 737         /* Update reserved bits */
 738         if ((efer ^ old_efer) & EFER_NX)
 739                 kvm_mmu_reset_context(vcpu);
 740
 741         return 0;
 742 }
 743
 744 void kvm_enable_efer_bits(u64 mask)
 745 {
 746        efer_reserved_bits &= ~mask;
 747 }
 748 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 749
 750
 751 /*
 752  * Writes msr value into into the appropriate "register".
 753  * Returns 0 on success, non-0 otherwise.
 754  * Assumes vcpu_load() was already called.
 755  */
 756 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 757 {
 758         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 759 }
 760
 761 /*
 762  * Adapt set_msr() to msr_io()'s calling convention
 763  */
 764 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 765 {
 766         return kvm_set_msr(vcpu, index, *data);
 767 }
 768
 769 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 770 {
 771         int version;
 772         int r;
 773         struct pvclock_wall_clock wc;
 774         struct timespec boot;
 775
 776         if (!wall_clock)
 777                 return;
 778
 779         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
 780         if (r)
 781                 return;
 782
 783         if (version & 1)
 784                 ++version;  /* first time write, random junk */
 785
 786         ++version;
 787
 788         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 789
 790         /*
 791          * The guest calculates current wall clock time by adding
 792          * system time (updated by kvm_write_guest_time below) to the
 793          * wall clock specified here.  guest system time equals host
 794          * system time for us, thus we must fill in host boot time here.
 795          */
 796         getboottime(&boot);
 797
 798         wc.sec = boot.tv_sec;
 799         wc.nsec = boot.tv_nsec;
 800         wc.version = version;
 801
 802         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 803
 804         version++;
 805         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 806 }
 807
 808 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 809 {
 810         uint32_t quotient, remainder;
 811
 812         /* Don't try to replace with do_div(), this one calculates
 813          * "(dividend << 32) / divisor" */
 814         __asm__ ( "divl %4"
 815                   : "=a" (quotient), "=d" (remainder)
 816                   : "0" (0), "1" (dividend), "r" (divisor) );
 817         return quotient;
 818 }
 819
 820 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 821 {
 822         uint64_t nsecs = 1000000000LL;
 823         int32_t  shift = 0;
 824         uint64_t tps64;
 825         uint32_t tps32;
 826
 827         tps64 = tsc_khz * 1000LL;
 828         while (tps64 > nsecs*2) {
 829                 tps64 >>= 1;
 830                 shift--;
 831         }
 832
 833         tps32 = (uint32_t)tps64;
 834         while (tps32 <= (uint32_t)nsecs) {
 835                 tps32 <<= 1;
 836                 shift++;
 837         }
 838
 839         hv_clock->tsc_shift = shift;
 840         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 841
 842         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 843                  __func__, tsc_khz, hv_clock->tsc_shift,
 844                  hv_clock->tsc_to_system_mul);
 845 }
 846
 847 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 848
 849 static void kvm_write_guest_time(struct kvm_vcpu *v)
 850 {
 851         struct timespec ts;
 852         unsigned long flags;
 853         struct kvm_vcpu_arch *vcpu = &v->arch;
 854         void *shared_kaddr;
 855         unsigned long this_tsc_khz;
 856
 857         if ((!vcpu->time_page))
 858                 return;
 859
 860         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 861         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 862                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 863                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 864         }
 865         put_cpu_var(cpu_tsc_khz);
 866
 867         /* Keep irq disabled to prevent changes to the clock */
 868         local_irq_save(flags);
 869         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 870         ktime_get_ts(&ts);
 871         monotonic_to_bootbased(&ts);
 872         local_irq_restore(flags);
 873
 874         /* With all the info we got, fill in the values */
 875
 876         vcpu->hv_clock.system_time = ts.tv_nsec +
 877                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 878
 879         vcpu->hv_clock.flags = 0;
 880
 881         /*
 882          * The interface expects us to write an even number signaling that the
 883          * update is finished. Since the guest won't see the intermediate
 884          * state, we just increase by 2 at the end.
 885          */
 886         vcpu->hv_clock.version += 2;
 887
 888         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 889
 890         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 891                sizeof(vcpu->hv_clock));
 892
 893         kunmap_atomic(shared_kaddr, KM_USER0);
 894
 895         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 896 }
 897
 898 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 899 {
 900         struct kvm_vcpu_arch *vcpu = &v->arch;
 901
 902         if (!vcpu->time_page)
 903                 return 0;
 904         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 905         return 1;
 906 }
 907
 908 static bool msr_mtrr_valid(unsigned msr)
 909 {
 910         switch (msr) {
 911         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 912         case MSR_MTRRfix64K_00000:
 913         case MSR_MTRRfix16K_80000:
 914         case MSR_MTRRfix16K_A0000:
 915         case MSR_MTRRfix4K_C0000:
 916         case MSR_MTRRfix4K_C8000:
 917         case MSR_MTRRfix4K_D0000:
 918         case MSR_MTRRfix4K_D8000:
 919         case MSR_MTRRfix4K_E0000:
 920         case MSR_MTRRfix4K_E8000:
 921         case MSR_MTRRfix4K_F0000:
 922         case MSR_MTRRfix4K_F8000:
 923         case MSR_MTRRdefType:
 924         case MSR_IA32_CR_PAT:
 925                 return true;
 926         case 0x2f8:
 927                 return true;
 928         }
 929         return false;
 930 }
 931
 932 static bool valid_pat_type(unsigned t)
 933 {
 934         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 935 }
 936
 937 static bool valid_mtrr_type(unsigned t)
 938 {
 939         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 940 }
 941
 942 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 943 {
 944         int i;
 945
 946         if (!msr_mtrr_valid(msr))
 947                 return false;
 948
 949         if (msr == MSR_IA32_CR_PAT) {
 950                 for (i = 0; i < 8; i++)
 951                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 952                                 return false;
 953                 return true;
 954         } else if (msr == MSR_MTRRdefType) {
 955                 if (data & ~0xcff)
 956                         return false;
 957                 return valid_mtrr_type(data & 0xff);
 958         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 959                 for (i = 0; i < 8 ; i++)
 960                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 961                                 return false;
 962                 return true;
 963         }
 964
 965         /* variable MTRRs */
 966         return valid_mtrr_type(data & 0xff);
 967 }
 968
 969 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 970 {
 971         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 972
 973         if (!mtrr_valid(vcpu, msr, data))
 974                 return 1;
 975
 976         if (msr == MSR_MTRRdefType) {
 977                 vcpu->arch.mtrr_state.def_type = data;
 978                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 979         } else if (msr == MSR_MTRRfix64K_00000)
 980                 p[0] = data;
 981         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 982                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 983         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 984                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 985         else if (msr == MSR_IA32_CR_PAT)
 986                 vcpu->arch.pat = data;
 987         else {  /* Variable MTRRs */
 988                 int idx, is_mtrr_mask;
 989                 u64 *pt;
 990
 991                 idx = (msr - 0x200) / 2;
 992                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 993                 if (!is_mtrr_mask)
 994                         pt =
 995                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 996                 else
 997                         pt =
 998                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 999                 *pt = data;
1000         }
1001
1002         kvm_mmu_reset_context(vcpu);
1003         return 0;
1004 }
1005
1006 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1007 {
1008         u64 mcg_cap = vcpu->arch.mcg_cap;
1009         unsigned bank_num = mcg_cap & 0xff;
1010
1011         switch (msr) {
1012         case MSR_IA32_MCG_STATUS:
1013                 vcpu->arch.mcg_status = data;
1014                 break;
1015         case MSR_IA32_MCG_CTL:
1016                 if (!(mcg_cap & MCG_CTL_P))
1017                         return 1;
1018                 if (data != 0 && data != ~(u64)0)
1019                         return -1;
1020                 vcpu->arch.mcg_ctl = data;
1021                 break;
1022         default:
1023                 if (msr >= MSR_IA32_MC0_CTL &&
1024                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1025                         u32 offset = msr - MSR_IA32_MC0_CTL;
1026                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1027                          * some Linux kernels though clear bit 10 in bank 4 to
1028                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1029                          * this to avoid an uncatched #GP in the guest
1030                          */
1031                         if ((offset & 0x3) == 0 &&
1032                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1033                                 return -1;
1034                         vcpu->arch.mce_banks[offset] = data;
1035                         break;
1036                 }
1037                 return 1;
1038         }
1039         return 0;
1040 }
1041
1042 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1043 {
1044         struct kvm *kvm = vcpu->kvm;
1045         int lm = is_long_mode(vcpu);
1046         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1047                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1048         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1049                 : kvm->arch.xen_hvm_config.blob_size_32;
1050         u32 page_num = data & ~PAGE_MASK;
1051         u64 page_addr = data & PAGE_MASK;
1052         u8 *page;
1053         int r;
1054
1055         r = -E2BIG;
1056         if (page_num >= blob_size)
1057                 goto out;
1058         r = -ENOMEM;
1059         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1060         if (!page)
1061                 goto out;
1062         r = -EFAULT;
1063         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1064                 goto out_free;
1065         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1066                 goto out_free;
1067         r = 0;
1068 out_free:
1069         kfree(page);
1070 out:
1071         return r;
1072 }
1073
1074 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1075 {
1076         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1077 }
1078
1079 static bool kvm_hv_msr_partition_wide(u32 msr)
1080 {
1081         bool r = false;
1082         switch (msr) {
1083         case HV_X64_MSR_GUEST_OS_ID:
1084         case HV_X64_MSR_HYPERCALL:
1085                 r = true;
1086                 break;
1087         }
1088
1089         return r;
1090 }
1091
1092 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1093 {
1094         struct kvm *kvm = vcpu->kvm;
1095
1096         switch (msr) {
1097         case HV_X64_MSR_GUEST_OS_ID:
1098                 kvm->arch.hv_guest_os_id = data;
1099                 /* setting guest os id to zero disables hypercall page */
1100                 if (!kvm->arch.hv_guest_os_id)
1101                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1102                 break;
1103         case HV_X64_MSR_HYPERCALL: {
1104                 u64 gfn;
1105                 unsigned long addr;
1106                 u8 instructions[4];
1107
1108                 /* if guest os id is not set hypercall should remain disabled */
1109                 if (!kvm->arch.hv_guest_os_id)
1110                         break;
1111                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1112                         kvm->arch.hv_hypercall = data;
1113                         break;
1114                 }
1115                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1116                 addr = gfn_to_hva(kvm, gfn);
1117                 if (kvm_is_error_hva(addr))
1118                         return 1;
1119                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1120                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1121                 if (copy_to_user((void __user *)addr, instructions, 4))
1122                         return 1;
1123                 kvm->arch.hv_hypercall = data;
1124                 break;
1125         }
1126         default:
1127                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1128                           "data 0x%llx\n", msr, data);
1129                 return 1;
1130         }
1131         return 0;
1132 }
1133
1134 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1135 {
1136         switch (msr) {
1137         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1138                 unsigned long addr;
1139
1140                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1141                         vcpu->arch.hv_vapic = data;
1142                         break;
1143                 }
1144                 addr = gfn_to_hva(vcpu->kvm, data >>
1145                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1146                 if (kvm_is_error_hva(addr))
1147                         return 1;
1148                 if (clear_user((void __user *)addr, PAGE_SIZE))
1149                         return 1;
1150                 vcpu->arch.hv_vapic = data;
1151                 break;
1152         }
1153         case HV_X64_MSR_EOI:
1154                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1155         case HV_X64_MSR_ICR:
1156                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1157         case HV_X64_MSR_TPR:
1158                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1159         default:
1160                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1161                           "data 0x%llx\n", msr, data);
1162                 return 1;
1163         }
1164
1165         return 0;
1166 }
1167
1168 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1169 {
1170         switch (msr) {
1171         case MSR_EFER:
1172                 return set_efer(vcpu, data);
1173         case MSR_K7_HWCR:
1174                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1175                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1176                 if (data != 0) {
1177                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1178                                 data);
1179                         return 1;
1180                 }
1181                 break;
1182         case MSR_FAM10H_MMIO_CONF_BASE:
1183                 if (data != 0) {
1184                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1185                                 "0x%llx\n", data);
1186                         return 1;
1187                 }
1188                 break;
1189         case MSR_AMD64_NB_CFG:
1190                 break;
1191         case MSR_IA32_DEBUGCTLMSR:
1192                 if (!data) {
1193                         /* We support the non-activated case already */
1194                         break;
1195                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1196                         /* Values other than LBR and BTF are vendor-specific,
1197                            thus reserved and should throw a #GP */
1198                         return 1;
1199                 }
1200                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1201                         __func__, data);
1202                 break;
1203         case MSR_IA32_UCODE_REV:
1204         case MSR_IA32_UCODE_WRITE:
1205         case MSR_VM_HSAVE_PA:
1206         case MSR_AMD64_PATCH_LOADER:
1207                 break;
1208         case 0x200 ... 0x2ff:
1209                 return set_msr_mtrr(vcpu, msr, data);
1210         case MSR_IA32_APICBASE:
1211                 kvm_set_apic_base(vcpu, data);
1212                 break;
1213         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1214                 return kvm_x2apic_msr_write(vcpu, msr, data);
1215         case MSR_IA32_MISC_ENABLE:
1216                 vcpu->arch.ia32_misc_enable_msr = data;
1217                 break;
1218         case MSR_KVM_WALL_CLOCK_NEW:
1219         case MSR_KVM_WALL_CLOCK:
1220                 vcpu->kvm->arch.wall_clock = data;
1221                 kvm_write_wall_clock(vcpu->kvm, data);
1222                 break;
1223         case MSR_KVM_SYSTEM_TIME_NEW:
1224         case MSR_KVM_SYSTEM_TIME: {
1225                 if (vcpu->arch.time_page) {
1226                         kvm_release_page_dirty(vcpu->arch.time_page);
1227                         vcpu->arch.time_page = NULL;
1228                 }
1229
1230                 vcpu->arch.time = data;
1231
1232                 /* we verify if the enable bit is set... */
1233                 if (!(data & 1))
1234                         break;
1235
1236                 /* ...but clean it before doing the actual write */
1237                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1238
1239                 vcpu->arch.time_page =
1240                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1241
1242                 if (is_error_page(vcpu->arch.time_page)) {
1243                         kvm_release_page_clean(vcpu->arch.time_page);
1244                         vcpu->arch.time_page = NULL;
1245                 }
1246
1247                 kvm_request_guest_time_update(vcpu);
1248                 break;
1249         }
1250         case MSR_IA32_MCG_CTL:
1251         case MSR_IA32_MCG_STATUS:
1252         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1253                 return set_msr_mce(vcpu, msr, data);
1254
1255         /* Performance counters are not protected by a CPUID bit,
1256          * so we should check all of them in the generic path for the sake of
1257          * cross vendor migration.
1258          * Writing a zero into the event select MSRs disables them,
1259          * which we perfectly emulate ;-). Any other value should be at least
1260          * reported, some guests depend on them.
1261          */
1262         case MSR_P6_EVNTSEL0:
1263         case MSR_P6_EVNTSEL1:
1264         case MSR_K7_EVNTSEL0:
1265         case MSR_K7_EVNTSEL1:
1266         case MSR_K7_EVNTSEL2:
1267         case MSR_K7_EVNTSEL3:
1268                 if (data != 0)
1269                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1270                                 "0x%x data 0x%llx\n", msr, data);
1271                 break;
1272         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1273          * so we ignore writes to make it happy.
1274          */
1275         case MSR_P6_PERFCTR0:
1276         case MSR_P6_PERFCTR1:
1277         case MSR_K7_PERFCTR0:
1278         case MSR_K7_PERFCTR1:
1279         case MSR_K7_PERFCTR2:
1280         case MSR_K7_PERFCTR3:
1281                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1282                         "0x%x data 0x%llx\n", msr, data);
1283                 break;
1284         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1285                 if (kvm_hv_msr_partition_wide(msr)) {
1286                         int r;
1287                         mutex_lock(&vcpu->kvm->lock);
1288                         r = set_msr_hyperv_pw(vcpu, msr, data);
1289                         mutex_unlock(&vcpu->kvm->lock);
1290                         return r;
1291                 } else
1292                         return set_msr_hyperv(vcpu, msr, data);
1293                 break;
1294         default:
1295                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1296                         return xen_hvm_config(vcpu, data);
1297                 if (!ignore_msrs) {
1298                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1299                                 msr, data);
1300                         return 1;
1301                 } else {
1302                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1303                                 msr, data);
1304                         break;
1305                 }
1306         }
1307         return 0;
1308 }
1309 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1310
1311
1312 /*
1313  * Reads an msr value (of 'msr_index') into 'pdata'.
1314  * Returns 0 on success, non-0 otherwise.
1315  * Assumes vcpu_load() was already called.
1316  */
1317 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1318 {
1319         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1320 }
1321
1322 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1323 {
1324         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1325
1326         if (!msr_mtrr_valid(msr))
1327                 return 1;
1328
1329         if (msr == MSR_MTRRdefType)
1330                 *pdata = vcpu->arch.mtrr_state.def_type +
1331                          (vcpu->arch.mtrr_state.enabled << 10);
1332         else if (msr == MSR_MTRRfix64K_00000)
1333                 *pdata = p[0];
1334         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1335                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1336         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1337                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1338         else if (msr == MSR_IA32_CR_PAT)
1339                 *pdata = vcpu->arch.pat;
1340         else {  /* Variable MTRRs */
1341                 int idx, is_mtrr_mask;
1342                 u64 *pt;
1343
1344                 idx = (msr - 0x200) / 2;
1345                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1346                 if (!is_mtrr_mask)
1347                         pt =
1348                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1349                 else
1350                         pt =
1351                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1352                 *pdata = *pt;
1353         }
1354
1355         return 0;
1356 }
1357
1358 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1359 {
1360         u64 data;
1361         u64 mcg_cap = vcpu->arch.mcg_cap;
1362         unsigned bank_num = mcg_cap & 0xff;
1363
1364         switch (msr) {
1365         case MSR_IA32_P5_MC_ADDR:
1366         case MSR_IA32_P5_MC_TYPE:
1367                 data = 0;
1368                 break;
1369         case MSR_IA32_MCG_CAP:
1370                 data = vcpu->arch.mcg_cap;
1371                 break;
1372         case MSR_IA32_MCG_CTL:
1373                 if (!(mcg_cap & MCG_CTL_P))
1374                         return 1;
1375                 data = vcpu->arch.mcg_ctl;
1376                 break;
1377         case MSR_IA32_MCG_STATUS:
1378                 data = vcpu->arch.mcg_status;
1379                 break;
1380         default:
1381                 if (msr >= MSR_IA32_MC0_CTL &&
1382                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1383                         u32 offset = msr - MSR_IA32_MC0_CTL;
1384                         data = vcpu->arch.mce_banks[offset];
1385                         break;
1386                 }
1387                 return 1;
1388         }
1389         *pdata = data;
1390         return 0;
1391 }
1392
1393 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1394 {
1395         u64 data = 0;
1396         struct kvm *kvm = vcpu->kvm;
1397
1398         switch (msr) {
1399         case HV_X64_MSR_GUEST_OS_ID:
1400                 data = kvm->arch.hv_guest_os_id;
1401                 break;
1402         case HV_X64_MSR_HYPERCALL:
1403                 data = kvm->arch.hv_hypercall;
1404                 break;
1405         default:
1406                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1407                 return 1;
1408         }
1409
1410         *pdata = data;
1411         return 0;
1412 }
1413
1414 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1415 {
1416         u64 data = 0;
1417
1418         switch (msr) {
1419         case HV_X64_MSR_VP_INDEX: {
1420                 int r;
1421                 struct kvm_vcpu *v;
1422                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1423                         if (v == vcpu)
1424                                 data = r;
1425                 break;
1426         }
1427         case HV_X64_MSR_EOI:
1428                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1429         case HV_X64_MSR_ICR:
1430                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1431         case HV_X64_MSR_TPR:
1432                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1433         default:
1434                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1435                 return 1;
1436         }
1437         *pdata = data;
1438         return 0;
1439 }
1440
1441 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1442 {
1443         u64 data;
1444
1445         switch (msr) {
1446         case MSR_IA32_PLATFORM_ID:
1447         case MSR_IA32_UCODE_REV:
1448         case MSR_IA32_EBL_CR_POWERON:
1449         case MSR_IA32_DEBUGCTLMSR:
1450         case MSR_IA32_LASTBRANCHFROMIP:
1451         case MSR_IA32_LASTBRANCHTOIP:
1452         case MSR_IA32_LASTINTFROMIP:
1453         case MSR_IA32_LASTINTTOIP:
1454         case MSR_K8_SYSCFG:
1455         case MSR_K7_HWCR:
1456         case MSR_VM_HSAVE_PA:
1457         case MSR_P6_PERFCTR0:
1458         case MSR_P6_PERFCTR1:
1459         case MSR_P6_EVNTSEL0:
1460         case MSR_P6_EVNTSEL1:
1461         case MSR_K7_EVNTSEL0:
1462         case MSR_K7_PERFCTR0:
1463         case MSR_K8_INT_PENDING_MSG:
1464         case MSR_AMD64_NB_CFG:
1465         case MSR_FAM10H_MMIO_CONF_BASE:
1466                 data = 0;
1467                 break;
1468         case MSR_MTRRcap:
1469                 data = 0x500 | KVM_NR_VAR_MTRR;
1470                 break;
1471         case 0x200 ... 0x2ff:
1472                 return get_msr_mtrr(vcpu, msr, pdata);
1473         case 0xcd: /* fsb frequency */
1474                 data = 3;
1475                 break;
1476         case MSR_IA32_APICBASE:
1477                 data = kvm_get_apic_base(vcpu);
1478                 break;
1479         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1480                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1481                 break;
1482         case MSR_IA32_MISC_ENABLE:
1483                 data = vcpu->arch.ia32_misc_enable_msr;
1484                 break;
1485         case MSR_IA32_PERF_STATUS:
1486                 /* TSC increment by tick */
1487                 data = 1000ULL;
1488                 /* CPU multiplier */
1489                 data |= (((uint64_t)4ULL) << 40);
1490                 break;
1491         case MSR_EFER:
1492                 data = vcpu->arch.efer;
1493                 break;
1494         case MSR_KVM_WALL_CLOCK:
1495         case MSR_KVM_WALL_CLOCK_NEW:
1496                 data = vcpu->kvm->arch.wall_clock;
1497                 break;
1498         case MSR_KVM_SYSTEM_TIME:
1499         case MSR_KVM_SYSTEM_TIME_NEW:
1500                 data = vcpu->arch.time;
1501                 break;
1502         case MSR_IA32_P5_MC_ADDR:
1503         case MSR_IA32_P5_MC_TYPE:
1504         case MSR_IA32_MCG_CAP:
1505         case MSR_IA32_MCG_CTL:
1506         case MSR_IA32_MCG_STATUS:
1507         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1508                 return get_msr_mce(vcpu, msr, pdata);
1509         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1510                 if (kvm_hv_msr_partition_wide(msr)) {
1511                         int r;
1512                         mutex_lock(&vcpu->kvm->lock);
1513                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1514                         mutex_unlock(&vcpu->kvm->lock);
1515                         return r;
1516                 } else
1517                         return get_msr_hyperv(vcpu, msr, pdata);
1518                 break;
1519         default:
1520                 if (!ignore_msrs) {
1521                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1522                         return 1;
1523                 } else {
1524                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1525                         data = 0;
1526                 }
1527                 break;
1528         }
1529         *pdata = data;
1530         return 0;
1531 }
1532 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1533
1534 /*
1535  * Read or write a bunch of msrs. All parameters are kernel addresses.
1536  *
1537  * @return number of msrs set successfully.
1538  */
1539 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1540                     struct kvm_msr_entry *entries,
1541                     int (*do_msr)(struct kvm_vcpu *vcpu,
1542                                   unsigned index, u64 *data))
1543 {
1544         int i, idx;
1545
1546         idx = srcu_read_lock(&vcpu->kvm->srcu);
1547         for (i = 0; i < msrs->nmsrs; ++i)
1548                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1549                         break;
1550         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1551
1552         return i;
1553 }
1554
1555 /*
1556  * Read or write a bunch of msrs. Parameters are user addresses.
1557  *
1558  * @return number of msrs set successfully.
1559  */
1560 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1561                   int (*do_msr)(struct kvm_vcpu *vcpu,
1562                                 unsigned index, u64 *data),
1563                   int writeback)
1564 {
1565         struct kvm_msrs msrs;
1566         struct kvm_msr_entry *entries;
1567         int r, n;
1568         unsigned size;
1569
1570         r = -EFAULT;
1571         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1572                 goto out;
1573
1574         r = -E2BIG;
1575         if (msrs.nmsrs >= MAX_IO_MSRS)
1576                 goto out;
1577
1578         r = -ENOMEM;
1579         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1580         entries = kmalloc(size, GFP_KERNEL);
1581         if (!entries)
1582                 goto out;
1583
1584         r = -EFAULT;
1585         if (copy_from_user(entries, user_msrs->entries, size))
1586                 goto out_free;
1587
1588         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1589         if (r < 0)
1590                 goto out_free;
1591
1592         r = -EFAULT;
1593         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1594                 goto out_free;
1595
1596         r = n;
1597
1598 out_free:
1599         kfree(entries);
1600 out:
1601         return r;
1602 }
1603
1604 int kvm_dev_ioctl_check_extension(long ext)
1605 {
1606         int r;
1607
1608         switch (ext) {
1609         case KVM_CAP_IRQCHIP:
1610         case KVM_CAP_HLT:
1611         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1612         case KVM_CAP_SET_TSS_ADDR:
1613         case KVM_CAP_EXT_CPUID:
1614         case KVM_CAP_CLOCKSOURCE:
1615         case KVM_CAP_PIT:
1616         case KVM_CAP_NOP_IO_DELAY:
1617         case KVM_CAP_MP_STATE:
1618         case KVM_CAP_SYNC_MMU:
1619         case KVM_CAP_REINJECT_CONTROL:
1620         case KVM_CAP_IRQ_INJECT_STATUS:
1621         case KVM_CAP_ASSIGN_DEV_IRQ:
1622         case KVM_CAP_IRQFD:
1623         case KVM_CAP_IOEVENTFD:
1624         case KVM_CAP_PIT2:
1625         case KVM_CAP_PIT_STATE2:
1626         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1627         case KVM_CAP_XEN_HVM:
1628         case KVM_CAP_ADJUST_CLOCK:
1629         case KVM_CAP_VCPU_EVENTS:
1630         case KVM_CAP_HYPERV:
1631         case KVM_CAP_HYPERV_VAPIC:
1632         case KVM_CAP_HYPERV_SPIN:
1633         case KVM_CAP_PCI_SEGMENT:
1634         case KVM_CAP_DEBUGREGS:
1635         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1636                 r = 1;
1637                 break;
1638         case KVM_CAP_COALESCED_MMIO:
1639                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1640                 break;
1641         case KVM_CAP_VAPIC:
1642                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1643                 break;
1644         case KVM_CAP_NR_VCPUS:
1645                 r = KVM_MAX_VCPUS;
1646                 break;
1647         case KVM_CAP_NR_MEMSLOTS:
1648                 r = KVM_MEMORY_SLOTS;
1649                 break;
1650         case KVM_CAP_PV_MMU:    /* obsolete */
1651                 r = 0;
1652                 break;
1653         case KVM_CAP_IOMMU:
1654                 r = iommu_found();
1655                 break;
1656         case KVM_CAP_MCE:
1657                 r = KVM_MAX_MCE_BANKS;
1658                 break;
1659         default:
1660                 r = 0;
1661                 break;
1662         }
1663         return r;
1664
1665 }
1666
1667 long kvm_arch_dev_ioctl(struct file *filp,
1668                         unsigned int ioctl, unsigned long arg)
1669 {
1670         void __user *argp = (void __user *)arg;
1671         long r;
1672
1673         switch (ioctl) {
1674         case KVM_GET_MSR_INDEX_LIST: {
1675                 struct kvm_msr_list __user *user_msr_list = argp;
1676                 struct kvm_msr_list msr_list;
1677                 unsigned n;
1678
1679                 r = -EFAULT;
1680                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1681                         goto out;
1682                 n = msr_list.nmsrs;
1683                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1684                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1685                         goto out;
1686                 r = -E2BIG;
1687                 if (n < msr_list.nmsrs)
1688                         goto out;
1689                 r = -EFAULT;
1690                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1691                                  num_msrs_to_save * sizeof(u32)))
1692                         goto out;
1693                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1694                                  &emulated_msrs,
1695                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1696                         goto out;
1697                 r = 0;
1698                 break;
1699         }
1700         case KVM_GET_SUPPORTED_CPUID: {
1701                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1702                 struct kvm_cpuid2 cpuid;
1703
1704                 r = -EFAULT;
1705                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1706                         goto out;
1707                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1708                                                       cpuid_arg->entries);
1709                 if (r)
1710                         goto out;
1711
1712                 r = -EFAULT;
1713                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1714                         goto out;
1715                 r = 0;
1716                 break;
1717         }
1718         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1719                 u64 mce_cap;
1720
1721                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1722                 r = -EFAULT;
1723                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1724                         goto out;
1725                 r = 0;
1726                 break;
1727         }
1728         default:
1729                 r = -EINVAL;
1730         }
1731 out:
1732         return r;
1733 }
1734
1735 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1736 {
1737         kvm_x86_ops->vcpu_load(vcpu, cpu);
1738         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1739                 unsigned long khz = cpufreq_quick_get(cpu);
1740                 if (!khz)
1741                         khz = tsc_khz;
1742                 per_cpu(cpu_tsc_khz, cpu) = khz;
1743         }
1744         kvm_request_guest_time_update(vcpu);
1745 }
1746
1747 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1748 {
1749         kvm_x86_ops->vcpu_put(vcpu);
1750         kvm_put_guest_fpu(vcpu);
1751 }
1752
1753 static int is_efer_nx(void)
1754 {
1755         unsigned long long efer = 0;
1756
1757         rdmsrl_safe(MSR_EFER, &efer);
1758         return efer & EFER_NX;
1759 }
1760
1761 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1762 {
1763         int i;
1764         struct kvm_cpuid_entry2 *e, *entry;
1765
1766         entry = NULL;
1767         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1768                 e = &vcpu->arch.cpuid_entries[i];
1769                 if (e->function == 0x80000001) {
1770                         entry = e;
1771                         break;
1772                 }
1773         }
1774         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1775                 entry->edx &= ~(1 << 20);
1776                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1777         }
1778 }
1779
1780 /* when an old userspace process fills a new kernel module */
1781 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1782                                     struct kvm_cpuid *cpuid,
1783                                     struct kvm_cpuid_entry __user *entries)
1784 {
1785         int r, i;
1786         struct kvm_cpuid_entry *cpuid_entries;
1787
1788         r = -E2BIG;
1789         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1790                 goto out;
1791         r = -ENOMEM;
1792         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1793         if (!cpuid_entries)
1794                 goto out;
1795         r = -EFAULT;
1796         if (copy_from_user(cpuid_entries, entries,
1797                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1798                 goto out_free;
1799         for (i = 0; i < cpuid->nent; i++) {
1800                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1801                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1802                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1803                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1804                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1805                 vcpu->arch.cpuid_entries[i].index = 0;
1806                 vcpu->arch.cpuid_entries[i].flags = 0;
1807                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1808                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1809                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1810         }
1811         vcpu->arch.cpuid_nent = cpuid->nent;
1812         cpuid_fix_nx_cap(vcpu);
1813         r = 0;
1814         kvm_apic_set_version(vcpu);
1815         kvm_x86_ops->cpuid_update(vcpu);
1816
1817 out_free:
1818         vfree(cpuid_entries);
1819 out:
1820         return r;
1821 }
1822
1823 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1824                                      struct kvm_cpuid2 *cpuid,
1825                                      struct kvm_cpuid_entry2 __user *entries)
1826 {
1827         int r;
1828
1829         r = -E2BIG;
1830         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1831                 goto out;
1832         r = -EFAULT;
1833         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1834                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1835                 goto out;
1836         vcpu->arch.cpuid_nent = cpuid->nent;
1837         kvm_apic_set_version(vcpu);
1838         kvm_x86_ops->cpuid_update(vcpu);
1839         return 0;
1840
1841 out:
1842         return r;
1843 }
1844
1845 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1846                                      struct kvm_cpuid2 *cpuid,
1847                                      struct kvm_cpuid_entry2 __user *entries)
1848 {
1849         int r;
1850
1851         r = -E2BIG;
1852         if (cpuid->nent < vcpu->arch.cpuid_nent)
1853                 goto out;
1854         r = -EFAULT;
1855         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1856                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1857                 goto out;
1858         return 0;
1859
1860 out:
1861         cpuid->nent = vcpu->arch.cpuid_nent;
1862         return r;
1863 }
1864
1865 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1866                            u32 index)
1867 {
1868         entry->function = function;
1869         entry->index = index;
1870         cpuid_count(entry->function, entry->index,
1871                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1872         entry->flags = 0;
1873 }
1874
1875 #define F(x) bit(X86_FEATURE_##x)
1876
1877 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1878                          u32 index, int *nent, int maxnent)
1879 {
1880         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1881 #ifdef CONFIG_X86_64
1882         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1883                                 ? F(GBPAGES) : 0;
1884         unsigned f_lm = F(LM);
1885 #else
1886         unsigned f_gbpages = 0;
1887         unsigned f_lm = 0;
1888 #endif
1889         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1890
1891         /* cpuid 1.edx */
1892         const u32 kvm_supported_word0_x86_features =
1893                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1894                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1895                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1896                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1897                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1898                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1899                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1900                 0 /* HTT, TM, Reserved, PBE */;
1901         /* cpuid 0x80000001.edx */
1902         const u32 kvm_supported_word1_x86_features =
1903                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1904                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1905                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1906                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1907                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1908                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1909                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1910                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1911         /* cpuid 1.ecx */
1912         const u32 kvm_supported_word4_x86_features =
1913                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1914                 0 /* DS-CPL, VMX, SMX, EST */ |
1915                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1916                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1917                 0 /* Reserved, DCA */ | F(XMM4_1) |
1918                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1919                 0 /* Reserved, XSAVE, OSXSAVE */;
1920         /* cpuid 0x80000001.ecx */
1921         const u32 kvm_supported_word6_x86_features =
1922                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1923                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1924                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1925                 0 /* SKINIT */ | 0 /* WDT */;
1926
1927         /* all calls to cpuid_count() should be made on the same cpu */
1928         get_cpu();
1929         do_cpuid_1_ent(entry, function, index);
1930         ++*nent;
1931
1932         switch (function) {
1933         case 0:
1934                 entry->eax = min(entry->eax, (u32)0xb);
1935                 break;
1936         case 1:
1937                 entry->edx &= kvm_supported_word0_x86_features;
1938                 entry->ecx &= kvm_supported_word4_x86_features;
1939                 /* we support x2apic emulation even if host does not support
1940                  * it since we emulate x2apic in software */
1941                 entry->ecx |= F(X2APIC);
1942                 break;
1943         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1944          * may return different values. This forces us to get_cpu() before
1945          * issuing the first command, and also to emulate this annoying behavior
1946          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1947         case 2: {
1948                 int t, times = entry->eax & 0xff;
1949
1950                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1951                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1952                 for (t = 1; t < times && *nent < maxnent; ++t) {
1953                         do_cpuid_1_ent(&entry[t], function, 0);
1954                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1955                         ++*nent;
1956                 }
1957                 break;
1958         }
1959         /* function 4 and 0xb have additional index. */
1960         case 4: {
1961                 int i, cache_type;
1962
1963                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1964                 /* read more entries until cache_type is zero */
1965                 for (i = 1; *nent < maxnent; ++i) {
1966                         cache_type = entry[i - 1].eax & 0x1f;
1967                         if (!cache_type)
1968                                 break;
1969                         do_cpuid_1_ent(&entry[i], function, i);
1970                         entry[i].flags |=
1971                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1972                         ++*nent;
1973                 }
1974                 break;
1975         }
1976         case 0xb: {
1977                 int i, level_type;
1978
1979                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1980                 /* read more entries until level_type is zero */
1981                 for (i = 1; *nent < maxnent; ++i) {
1982                         level_type = entry[i - 1].ecx & 0xff00;
1983                         if (!level_type)
1984                                 break;
1985                         do_cpuid_1_ent(&entry[i], function, i);
1986                         entry[i].flags |=
1987                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1988                         ++*nent;
1989                 }
1990                 break;
1991         }
1992         case KVM_CPUID_SIGNATURE: {
1993                 char signature[12] = "KVMKVMKVM\0\0";
1994                 u32 *sigptr = (u32 *)signature;
1995                 entry->eax = 0;
1996                 entry->ebx = sigptr[0];
1997                 entry->ecx = sigptr[1];
1998                 entry->edx = sigptr[2];
1999                 break;
2000         }
2001         case KVM_CPUID_FEATURES:
2002                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2003                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2004                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2005                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2006                 entry->ebx = 0;
2007                 entry->ecx = 0;
2008                 entry->edx = 0;
2009                 break;
2010         case 0x80000000:
2011                 entry->eax = min(entry->eax, 0x8000001a);
2012                 break;
2013         case 0x80000001:
2014                 entry->edx &= kvm_supported_word1_x86_features;
2015                 entry->ecx &= kvm_supported_word6_x86_features;
2016                 break;
2017         }
2018
2019         kvm_x86_ops->set_supported_cpuid(function, entry);
2020
2021         put_cpu();
2022 }
2023
2024 #undef F
2025
2026 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2027                                      struct kvm_cpuid_entry2 __user *entries)
2028 {
2029         struct kvm_cpuid_entry2 *cpuid_entries;
2030         int limit, nent = 0, r = -E2BIG;
2031         u32 func;
2032
2033         if (cpuid->nent < 1)
2034                 goto out;
2035         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2036                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2037         r = -ENOMEM;
2038         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2039         if (!cpuid_entries)
2040                 goto out;
2041
2042         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2043         limit = cpuid_entries[0].eax;
2044         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2045                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2046                              &nent, cpuid->nent);
2047         r = -E2BIG;
2048         if (nent >= cpuid->nent)
2049                 goto out_free;
2050
2051         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2052         limit = cpuid_entries[nent - 1].eax;
2053         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2054                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2055                              &nent, cpuid->nent);
2056
2057
2058
2059         r = -E2BIG;
2060         if (nent >= cpuid->nent)
2061                 goto out_free;
2062
2063         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2064                      cpuid->nent);
2065
2066         r = -E2BIG;
2067         if (nent >= cpuid->nent)
2068                 goto out_free;
2069
2070         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2071                      cpuid->nent);
2072
2073         r = -E2BIG;
2074         if (nent >= cpuid->nent)
2075                 goto out_free;
2076
2077         r = -EFAULT;
2078         if (copy_to_user(entries, cpuid_entries,
2079                          nent * sizeof(struct kvm_cpuid_entry2)))
2080                 goto out_free;
2081         cpuid->nent = nent;
2082         r = 0;
2083
2084 out_free:
2085         vfree(cpuid_entries);
2086 out:
2087         return r;
2088 }
2089
2090 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2091                                     struct kvm_lapic_state *s)
2092 {
2093         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2094
2095         return 0;
2096 }
2097
2098 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2099                                     struct kvm_lapic_state *s)
2100 {
2101         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2102         kvm_apic_post_state_restore(vcpu);
2103         update_cr8_intercept(vcpu);
2104
2105         return 0;
2106 }
2107
2108 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2109                                     struct kvm_interrupt *irq)
2110 {
2111         if (irq->irq < 0 || irq->irq >= 256)
2112                 return -EINVAL;
2113         if (irqchip_in_kernel(vcpu->kvm))
2114                 return -ENXIO;
2115
2116         kvm_queue_interrupt(vcpu, irq->irq, false);
2117
2118         return 0;
2119 }
2120
2121 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2122 {
2123         kvm_inject_nmi(vcpu);
2124
2125         return 0;
2126 }
2127
2128 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2129                                            struct kvm_tpr_access_ctl *tac)
2130 {
2131         if (tac->flags)
2132                 return -EINVAL;
2133         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2134         return 0;
2135 }
2136
2137 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2138                                         u64 mcg_cap)
2139 {
2140         int r;
2141         unsigned bank_num = mcg_cap & 0xff, bank;
2142
2143         r = -EINVAL;
2144         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2145                 goto out;
2146         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2147                 goto out;
2148         r = 0;
2149         vcpu->arch.mcg_cap = mcg_cap;
2150         /* Init IA32_MCG_CTL to all 1s */
2151         if (mcg_cap & MCG_CTL_P)
2152                 vcpu->arch.mcg_ctl = ~(u64)0;
2153         /* Init IA32_MCi_CTL to all 1s */
2154         for (bank = 0; bank < bank_num; bank++)
2155                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2156 out:
2157         return r;
2158 }
2159
2160 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2161                                       struct kvm_x86_mce *mce)
2162 {
2163         u64 mcg_cap = vcpu->arch.mcg_cap;
2164         unsigned bank_num = mcg_cap & 0xff;
2165         u64 *banks = vcpu->arch.mce_banks;
2166
2167         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2168                 return -EINVAL;
2169         /*
2170          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2171          * reporting is disabled
2172          */
2173         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2174             vcpu->arch.mcg_ctl != ~(u64)0)
2175                 return 0;
2176         banks += 4 * mce->bank;
2177         /*
2178          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2179          * reporting is disabled for the bank
2180          */
2181         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2182                 return 0;
2183         if (mce->status & MCI_STATUS_UC) {
2184                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2185                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2186                         printk(KERN_DEBUG "kvm: set_mce: "
2187                                "injects mce exception while "
2188                                "previous one is in progress!\n");
2189                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2190                         return 0;
2191                 }
2192                 if (banks[1] & MCI_STATUS_VAL)
2193                         mce->status |= MCI_STATUS_OVER;
2194                 banks[2] = mce->addr;
2195                 banks[3] = mce->misc;
2196                 vcpu->arch.mcg_status = mce->mcg_status;
2197                 banks[1] = mce->status;
2198                 kvm_queue_exception(vcpu, MC_VECTOR);
2199         } else if (!(banks[1] & MCI_STATUS_VAL)
2200                    || !(banks[1] & MCI_STATUS_UC)) {
2201                 if (banks[1] & MCI_STATUS_VAL)
2202                         mce->status |= MCI_STATUS_OVER;
2203                 banks[2] = mce->addr;
2204                 banks[3] = mce->misc;
2205                 banks[1] = mce->status;
2206         } else
2207                 banks[1] |= MCI_STATUS_OVER;
2208         return 0;
2209 }
2210
2211 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2212                                                struct kvm_vcpu_events *events)
2213 {
2214         events->exception.injected =
2215                 vcpu->arch.exception.pending &&
2216                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2217         events->exception.nr = vcpu->arch.exception.nr;
2218         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2219         events->exception.error_code = vcpu->arch.exception.error_code;
2220
2221         events->interrupt.injected =
2222                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2223         events->interrupt.nr = vcpu->arch.interrupt.nr;
2224         events->interrupt.soft = 0;
2225         events->interrupt.shadow =
2226                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2227                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2228
2229         events->nmi.injected = vcpu->arch.nmi_injected;
2230         events->nmi.pending = vcpu->arch.nmi_pending;
2231         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2232
2233         events->sipi_vector = vcpu->arch.sipi_vector;
2234
2235         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2236                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2237                          | KVM_VCPUEVENT_VALID_SHADOW);
2238 }
2239
2240 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2241                                               struct kvm_vcpu_events *events)
2242 {
2243         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2244                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2245                               | KVM_VCPUEVENT_VALID_SHADOW))
2246                 return -EINVAL;
2247
2248         vcpu->arch.exception.pending = events->exception.injected;
2249         vcpu->arch.exception.nr = events->exception.nr;
2250         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2251         vcpu->arch.exception.error_code = events->exception.error_code;
2252
2253         vcpu->arch.interrupt.pending = events->interrupt.injected;
2254         vcpu->arch.interrupt.nr = events->interrupt.nr;
2255         vcpu->arch.interrupt.soft = events->interrupt.soft;
2256         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2257                 kvm_pic_clear_isr_ack(vcpu->kvm);
2258         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2259                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2260                                                   events->interrupt.shadow);
2261
2262         vcpu->arch.nmi_injected = events->nmi.injected;
2263         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2264                 vcpu->arch.nmi_pending = events->nmi.pending;
2265         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2266
2267         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2268                 vcpu->arch.sipi_vector = events->sipi_vector;
2269
2270         return 0;
2271 }
2272
2273 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2274                                              struct kvm_debugregs *dbgregs)
2275 {
2276         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2277         dbgregs->dr6 = vcpu->arch.dr6;
2278         dbgregs->dr7 = vcpu->arch.dr7;
2279         dbgregs->flags = 0;
2280 }
2281
2282 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2283                                             struct kvm_debugregs *dbgregs)
2284 {
2285         if (dbgregs->flags)
2286                 return -EINVAL;
2287
2288         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2289         vcpu->arch.dr6 = dbgregs->dr6;
2290         vcpu->arch.dr7 = dbgregs->dr7;
2291
2292         return 0;
2293 }
2294
2295 long kvm_arch_vcpu_ioctl(struct file *filp,
2296                          unsigned int ioctl, unsigned long arg)
2297 {
2298         struct kvm_vcpu *vcpu = filp->private_data;
2299         void __user *argp = (void __user *)arg;
2300         int r;
2301         struct kvm_lapic_state *lapic = NULL;
2302
2303         switch (ioctl) {
2304         case KVM_GET_LAPIC: {
2305                 r = -EINVAL;
2306                 if (!vcpu->arch.apic)
2307                         goto out;
2308                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2309
2310                 r = -ENOMEM;
2311                 if (!lapic)
2312                         goto out;
2313                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2314                 if (r)
2315                         goto out;
2316                 r = -EFAULT;
2317                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2318                         goto out;
2319                 r = 0;
2320                 break;
2321         }
2322         case KVM_SET_LAPIC: {
2323                 r = -EINVAL;
2324                 if (!vcpu->arch.apic)
2325                         goto out;
2326                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2327                 r = -ENOMEM;
2328                 if (!lapic)
2329                         goto out;
2330                 r = -EFAULT;
2331                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2332                         goto out;
2333                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2334                 if (r)
2335                         goto out;
2336                 r = 0;
2337                 break;
2338         }
2339         case KVM_INTERRUPT: {
2340                 struct kvm_interrupt irq;
2341
2342                 r = -EFAULT;
2343                 if (copy_from_user(&irq, argp, sizeof irq))
2344                         goto out;
2345                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2346                 if (r)
2347                         goto out;
2348                 r = 0;
2349                 break;
2350         }
2351         case KVM_NMI: {
2352                 r = kvm_vcpu_ioctl_nmi(vcpu);
2353                 if (r)
2354                         goto out;
2355                 r = 0;
2356                 break;
2357         }
2358         case KVM_SET_CPUID: {
2359                 struct kvm_cpuid __user *cpuid_arg = argp;
2360                 struct kvm_cpuid cpuid;
2361
2362                 r = -EFAULT;
2363                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2364                         goto out;
2365                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2366                 if (r)
2367                         goto out;
2368                 break;
2369         }
2370         case KVM_SET_CPUID2: {
2371                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2372                 struct kvm_cpuid2 cpuid;
2373
2374                 r = -EFAULT;
2375                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2376                         goto out;
2377                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2378                                               cpuid_arg->entries);
2379                 if (r)
2380                         goto out;
2381                 break;
2382         }
2383         case KVM_GET_CPUID2: {
2384                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2385                 struct kvm_cpuid2 cpuid;
2386
2387                 r = -EFAULT;
2388                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2389                         goto out;
2390                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2391                                               cpuid_arg->entries);
2392                 if (r)
2393                         goto out;
2394                 r = -EFAULT;
2395                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2396                         goto out;
2397                 r = 0;
2398                 break;
2399         }
2400         case KVM_GET_MSRS:
2401                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2402                 break;
2403         case KVM_SET_MSRS:
2404                 r = msr_io(vcpu, argp, do_set_msr, 0);
2405                 break;
2406         case KVM_TPR_ACCESS_REPORTING: {
2407                 struct kvm_tpr_access_ctl tac;
2408
2409                 r = -EFAULT;
2410                 if (copy_from_user(&tac, argp, sizeof tac))
2411                         goto out;
2412                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2413                 if (r)
2414                         goto out;
2415                 r = -EFAULT;
2416                 if (copy_to_user(argp, &tac, sizeof tac))
2417                         goto out;
2418                 r = 0;
2419                 break;
2420         };
2421         case KVM_SET_VAPIC_ADDR: {
2422                 struct kvm_vapic_addr va;
2423
2424                 r = -EINVAL;
2425                 if (!irqchip_in_kernel(vcpu->kvm))
2426                         goto out;
2427                 r = -EFAULT;
2428                 if (copy_from_user(&va, argp, sizeof va))
2429                         goto out;
2430                 r = 0;
2431                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2432                 break;
2433         }
2434         case KVM_X86_SETUP_MCE: {
2435                 u64 mcg_cap;
2436
2437                 r = -EFAULT;
2438                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2439                         goto out;
2440                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2441                 break;
2442         }
2443         case KVM_X86_SET_MCE: {
2444                 struct kvm_x86_mce mce;
2445
2446                 r = -EFAULT;
2447                 if (copy_from_user(&mce, argp, sizeof mce))
2448                         goto out;
2449                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2450                 break;
2451         }
2452         case KVM_GET_VCPU_EVENTS: {
2453                 struct kvm_vcpu_events events;
2454
2455                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2456
2457                 r = -EFAULT;
2458                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2459                         break;
2460                 r = 0;
2461                 break;
2462         }
2463         case KVM_SET_VCPU_EVENTS: {
2464                 struct kvm_vcpu_events events;
2465
2466                 r = -EFAULT;
2467                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2468                         break;
2469
2470                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2471                 break;
2472         }
2473         case KVM_GET_DEBUGREGS: {
2474                 struct kvm_debugregs dbgregs;
2475
2476                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2477
2478                 r = -EFAULT;
2479                 if (copy_to_user(argp, &dbgregs,
2480                                  sizeof(struct kvm_debugregs)))
2481                         break;
2482                 r = 0;
2483                 break;
2484         }
2485         case KVM_SET_DEBUGREGS: {
2486                 struct kvm_debugregs dbgregs;
2487
2488                 r = -EFAULT;
2489                 if (copy_from_user(&dbgregs, argp,
2490                                    sizeof(struct kvm_debugregs)))
2491                         break;
2492
2493                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2494                 break;
2495         }
2496         default:
2497                 r = -EINVAL;
2498         }
2499 out:
2500         kfree(lapic);
2501         return r;
2502 }
2503
2504 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2505 {
2506         int ret;
2507
2508         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2509                 return -1;
2510         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2511         return ret;
2512 }
2513
2514 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2515                                               u64 ident_addr)
2516 {
2517         kvm->arch.ept_identity_map_addr = ident_addr;
2518         return 0;
2519 }
2520
2521 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2522                                           u32 kvm_nr_mmu_pages)
2523 {
2524         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2525                 return -EINVAL;
2526
2527         mutex_lock(&kvm->slots_lock);
2528         spin_lock(&kvm->mmu_lock);
2529
2530         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2531         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2532
2533         spin_unlock(&kvm->mmu_lock);
2534         mutex_unlock(&kvm->slots_lock);
2535         return 0;
2536 }
2537
2538 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2539 {
2540         return kvm->arch.n_alloc_mmu_pages;
2541 }
2542
2543 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2544 {
2545         int i;
2546         struct kvm_mem_alias *alias;
2547         struct kvm_mem_aliases *aliases;
2548
2549         aliases = kvm_aliases(kvm);
2550
2551         for (i = 0; i < aliases->naliases; ++i) {
2552                 alias = &aliases->aliases[i];
2553                 if (alias->flags & KVM_ALIAS_INVALID)
2554                         continue;
2555                 if (gfn >= alias->base_gfn
2556                     && gfn < alias->base_gfn + alias->npages)
2557                         return alias->target_gfn + gfn - alias->base_gfn;
2558         }
2559         return gfn;
2560 }
2561
2562 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2563 {
2564         int i;
2565         struct kvm_mem_alias *alias;
2566         struct kvm_mem_aliases *aliases;
2567
2568         aliases = kvm_aliases(kvm);
2569
2570         for (i = 0; i < aliases->naliases; ++i) {
2571                 alias = &aliases->aliases[i];
2572                 if (gfn >= alias->base_gfn
2573                     && gfn < alias->base_gfn + alias->npages)
2574                         return alias->target_gfn + gfn - alias->base_gfn;
2575         }
2576         return gfn;
2577 }
2578
2579 /*
2580  * Set a new alias region.  Aliases map a portion of physical memory into
2581  * another portion.  This is useful for memory windows, for example the PC
2582  * VGA region.
2583  */
2584 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2585                                          struct kvm_memory_alias *alias)
2586 {
2587         int r, n;
2588         struct kvm_mem_alias *p;
2589         struct kvm_mem_aliases *aliases, *old_aliases;
2590
2591         r = -EINVAL;
2592         /* General sanity checks */
2593         if (alias->memory_size & (PAGE_SIZE - 1))
2594                 goto out;
2595         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2596                 goto out;
2597         if (alias->slot >= KVM_ALIAS_SLOTS)
2598                 goto out;
2599         if (alias->guest_phys_addr + alias->memory_size
2600             < alias->guest_phys_addr)
2601                 goto out;
2602         if (alias->target_phys_addr + alias->memory_size
2603             < alias->target_phys_addr)
2604                 goto out;
2605
2606         r = -ENOMEM;
2607         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2608         if (!aliases)
2609                 goto out;
2610
2611         mutex_lock(&kvm->slots_lock);
2612
2613         /* invalidate any gfn reference in case of deletion/shrinking */
2614         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2615         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2616         old_aliases = kvm->arch.aliases;
2617         rcu_assign_pointer(kvm->arch.aliases, aliases);
2618         synchronize_srcu_expedited(&kvm->srcu);
2619         kvm_mmu_zap_all(kvm);
2620         kfree(old_aliases);
2621
2622         r = -ENOMEM;
2623         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2624         if (!aliases)
2625                 goto out_unlock;
2626
2627         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2628
2629         p = &aliases->aliases[alias->slot];
2630         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2631         p->npages = alias->memory_size >> PAGE_SHIFT;
2632         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2633         p->flags &= ~(KVM_ALIAS_INVALID);
2634
2635         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2636                 if (aliases->aliases[n - 1].npages)
2637                         break;
2638         aliases->naliases = n;
2639
2640         old_aliases = kvm->arch.aliases;
2641         rcu_assign_pointer(kvm->arch.aliases, aliases);
2642         synchronize_srcu_expedited(&kvm->srcu);
2643         kfree(old_aliases);
2644         r = 0;
2645
2646 out_unlock:
2647         mutex_unlock(&kvm->slots_lock);
2648 out:
2649         return r;
2650 }
2651
2652 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2653 {
2654         int r;
2655
2656         r = 0;
2657         switch (chip->chip_id) {
2658         case KVM_IRQCHIP_PIC_MASTER:
2659                 memcpy(&chip->chip.pic,
2660                         &pic_irqchip(kvm)->pics[0],
2661                         sizeof(struct kvm_pic_state));
2662                 break;
2663         case KVM_IRQCHIP_PIC_SLAVE:
2664                 memcpy(&chip->chip.pic,
2665                         &pic_irqchip(kvm)->pics[1],
2666                         sizeof(struct kvm_pic_state));
2667                 break;
2668         case KVM_IRQCHIP_IOAPIC:
2669                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2670                 break;
2671         default:
2672                 r = -EINVAL;
2673                 break;
2674         }
2675         return r;
2676 }
2677
2678 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2679 {
2680         int r;
2681
2682         r = 0;
2683         switch (chip->chip_id) {
2684         case KVM_IRQCHIP_PIC_MASTER:
2685                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2686                 memcpy(&pic_irqchip(kvm)->pics[0],
2687                         &chip->chip.pic,
2688                         sizeof(struct kvm_pic_state));
2689                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2690                 break;
2691         case KVM_IRQCHIP_PIC_SLAVE:
2692                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2693                 memcpy(&pic_irqchip(kvm)->pics[1],
2694                         &chip->chip.pic,
2695                         sizeof(struct kvm_pic_state));
2696                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2697                 break;
2698         case KVM_IRQCHIP_IOAPIC:
2699                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2700                 break;
2701         default:
2702                 r = -EINVAL;
2703                 break;
2704         }
2705         kvm_pic_update_irq(pic_irqchip(kvm));
2706         return r;
2707 }
2708
2709 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2710 {
2711         int r = 0;
2712
2713         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2714         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2715         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2716         return r;
2717 }
2718
2719 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2720 {
2721         int r = 0;
2722
2723         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2724         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2725         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2726         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2727         return r;
2728 }
2729
2730 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2731 {
2732         int r = 0;
2733
2734         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2735         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2736                 sizeof(ps->channels));
2737         ps->flags = kvm->arch.vpit->pit_state.flags;
2738         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2739         return r;
2740 }
2741
2742 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2743 {
2744         int r = 0, start = 0;
2745         u32 prev_legacy, cur_legacy;
2746         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2747         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2748         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2749         if (!prev_legacy && cur_legacy)
2750                 start = 1;
2751         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2752                sizeof(kvm->arch.vpit->pit_state.channels));
2753         kvm->arch.vpit->pit_state.flags = ps->flags;
2754         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2755         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2756         return r;
2757 }
2758
2759 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2760                                  struct kvm_reinject_control *control)
2761 {
2762         if (!kvm->arch.vpit)
2763                 return -ENXIO;
2764         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2765         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2766         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2767         return 0;
2768 }
2769
2770 /*
2771  * Get (and clear) the dirty memory log for a memory slot.
2772  */
2773 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2774                                       struct kvm_dirty_log *log)
2775 {
2776         int r, i;
2777         struct kvm_memory_slot *memslot;
2778         unsigned long n;
2779         unsigned long is_dirty = 0;
2780
2781         mutex_lock(&kvm->slots_lock);
2782
2783         r = -EINVAL;
2784         if (log->slot >= KVM_MEMORY_SLOTS)
2785                 goto out;
2786
2787         memslot = &kvm->memslots->memslots[log->slot];
2788         r = -ENOENT;
2789         if (!memslot->dirty_bitmap)
2790                 goto out;
2791
2792         n = kvm_dirty_bitmap_bytes(memslot);
2793
2794         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2795                 is_dirty = memslot->dirty_bitmap[i];
2796
2797         /* If nothing is dirty, don't bother messing with page tables. */
2798         if (is_dirty) {
2799                 struct kvm_memslots *slots, *old_slots;
2800                 unsigned long *dirty_bitmap;
2801
2802                 spin_lock(&kvm->mmu_lock);
2803                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2804                 spin_unlock(&kvm->mmu_lock);
2805
2806                 r = -ENOMEM;
2807                 dirty_bitmap = vmalloc(n);
2808                 if (!dirty_bitmap)
2809                         goto out;
2810                 memset(dirty_bitmap, 0, n);
2811
2812                 r = -ENOMEM;
2813                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2814                 if (!slots) {
2815                         vfree(dirty_bitmap);
2816                         goto out;
2817                 }
2818                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2819                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2820
2821                 old_slots = kvm->memslots;
2822                 rcu_assign_pointer(kvm->memslots, slots);
2823                 synchronize_srcu_expedited(&kvm->srcu);
2824                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2825                 kfree(old_slots);
2826
2827                 r = -EFAULT;
2828                 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) {
2829                         vfree(dirty_bitmap);
2830                         goto out;
2831                 }
2832                 vfree(dirty_bitmap);
2833         } else {
2834                 r = -EFAULT;
2835                 if (clear_user(log->dirty_bitmap, n))
2836                         goto out;
2837         }
2838
2839         r = 0;
2840 out:
2841         mutex_unlock(&kvm->slots_lock);
2842         return r;
2843 }
2844
2845 long kvm_arch_vm_ioctl(struct file *filp,
2846                        unsigned int ioctl, unsigned long arg)
2847 {
2848         struct kvm *kvm = filp->private_data;
2849         void __user *argp = (void __user *)arg;
2850         int r = -ENOTTY;
2851         /*
2852          * This union makes it completely explicit to gcc-3.x
2853          * that these two variables' stack usage should be
2854          * combined, not added together.
2855          */
2856         union {
2857                 struct kvm_pit_state ps;
2858                 struct kvm_pit_state2 ps2;
2859                 struct kvm_memory_alias alias;
2860                 struct kvm_pit_config pit_config;
2861         } u;
2862
2863         switch (ioctl) {
2864         case KVM_SET_TSS_ADDR:
2865                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2866                 if (r < 0)
2867                         goto out;
2868                 break;
2869         case KVM_SET_IDENTITY_MAP_ADDR: {
2870                 u64 ident_addr;
2871
2872                 r = -EFAULT;
2873                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2874                         goto out;
2875                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2876                 if (r < 0)
2877                         goto out;
2878                 break;
2879         }
2880         case KVM_SET_MEMORY_REGION: {
2881                 struct kvm_memory_region kvm_mem;
2882                 struct kvm_userspace_memory_region kvm_userspace_mem;
2883
2884                 r = -EFAULT;
2885                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2886                         goto out;
2887                 kvm_userspace_mem.slot = kvm_mem.slot;
2888                 kvm_userspace_mem.flags = kvm_mem.flags;
2889                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2890                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2891                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2892                 if (r)
2893                         goto out;
2894                 break;
2895         }
2896         case KVM_SET_NR_MMU_PAGES:
2897                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2898                 if (r)
2899                         goto out;
2900                 break;
2901         case KVM_GET_NR_MMU_PAGES:
2902                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2903                 break;
2904         case KVM_SET_MEMORY_ALIAS:
2905                 r = -EFAULT;
2906                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2907                         goto out;
2908                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2909                 if (r)
2910                         goto out;
2911                 break;
2912         case KVM_CREATE_IRQCHIP: {
2913                 struct kvm_pic *vpic;
2914
2915                 mutex_lock(&kvm->lock);
2916                 r = -EEXIST;
2917                 if (kvm->arch.vpic)
2918                         goto create_irqchip_unlock;
2919                 r = -ENOMEM;
2920                 vpic = kvm_create_pic(kvm);
2921                 if (vpic) {
2922                         r = kvm_ioapic_init(kvm);
2923                         if (r) {
2924                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2925                                                           &vpic->dev);
2926                                 kfree(vpic);
2927                                 goto create_irqchip_unlock;
2928                         }
2929                 } else
2930                         goto create_irqchip_unlock;
2931                 smp_wmb();
2932                 kvm->arch.vpic = vpic;
2933                 smp_wmb();
2934                 r = kvm_setup_default_irq_routing(kvm);
2935                 if (r) {
2936                         mutex_lock(&kvm->irq_lock);
2937                         kvm_ioapic_destroy(kvm);
2938                         kvm_destroy_pic(kvm);
2939                         mutex_unlock(&kvm->irq_lock);
2940                 }
2941         create_irqchip_unlock:
2942                 mutex_unlock(&kvm->lock);
2943                 break;
2944         }
2945         case KVM_CREATE_PIT:
2946                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2947                 goto create_pit;
2948         case KVM_CREATE_PIT2:
2949                 r = -EFAULT;
2950                 if (copy_from_user(&u.pit_config, argp,
2951                                    sizeof(struct kvm_pit_config)))
2952                         goto out;
2953         create_pit:
2954                 mutex_lock(&kvm->slots_lock);
2955                 r = -EEXIST;
2956                 if (kvm->arch.vpit)
2957                         goto create_pit_unlock;
2958                 r = -ENOMEM;
2959                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2960                 if (kvm->arch.vpit)
2961                         r = 0;
2962         create_pit_unlock:
2963                 mutex_unlock(&kvm->slots_lock);
2964                 break;
2965         case KVM_IRQ_LINE_STATUS:
2966         case KVM_IRQ_LINE: {
2967                 struct kvm_irq_level irq_event;
2968
2969                 r = -EFAULT;
2970                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2971                         goto out;
2972                 r = -ENXIO;
2973                 if (irqchip_in_kernel(kvm)) {
2974                         __s32 status;
2975                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2976                                         irq_event.irq, irq_event.level);
2977                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2978                                 r = -EFAULT;
2979                                 irq_event.status = status;
2980                                 if (copy_to_user(argp, &irq_event,
2981                                                         sizeof irq_event))
2982                                         goto out;
2983                         }
2984                         r = 0;
2985                 }
2986                 break;
2987         }
2988         case KVM_GET_IRQCHIP: {
2989                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2990                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2991
2992                 r = -ENOMEM;
2993                 if (!chip)
2994                         goto out;
2995                 r = -EFAULT;
2996                 if (copy_from_user(chip, argp, sizeof *chip))
2997                         goto get_irqchip_out;
2998                 r = -ENXIO;
2999                 if (!irqchip_in_kernel(kvm))
3000                         goto get_irqchip_out;
3001                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3002                 if (r)
3003                         goto get_irqchip_out;
3004                 r = -EFAULT;
3005                 if (copy_to_user(argp, chip, sizeof *chip))
3006                         goto get_irqchip_out;
3007                 r = 0;
3008         get_irqchip_out:
3009                 kfree(chip);
3010                 if (r)
3011                         goto out;
3012                 break;
3013         }
3014         case KVM_SET_IRQCHIP: {
3015                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3016                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3017
3018                 r = -ENOMEM;
3019                 if (!chip)
3020                         goto out;
3021                 r = -EFAULT;
3022                 if (copy_from_user(chip, argp, sizeof *chip))
3023                         goto set_irqchip_out;
3024                 r = -ENXIO;
3025                 if (!irqchip_in_kernel(kvm))
3026                         goto set_irqchip_out;
3027                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3028                 if (r)
3029                         goto set_irqchip_out;
3030                 r = 0;
3031         set_irqchip_out:
3032                 kfree(chip);
3033                 if (r)
3034                         goto out;
3035                 break;
3036         }
3037         case KVM_GET_PIT: {
3038                 r = -EFAULT;
3039                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3040                         goto out;
3041                 r = -ENXIO;
3042                 if (!kvm->arch.vpit)
3043                         goto out;
3044                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3045                 if (r)
3046                         goto out;
3047                 r = -EFAULT;
3048                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3049                         goto out;
3050                 r = 0;
3051                 break;
3052         }
3053         case KVM_SET_PIT: {
3054                 r = -EFAULT;
3055                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3056                         goto out;
3057                 r = -ENXIO;
3058                 if (!kvm->arch.vpit)
3059                         goto out;
3060                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3061                 if (r)
3062                         goto out;
3063                 r = 0;
3064                 break;
3065         }
3066         case KVM_GET_PIT2: {
3067                 r = -ENXIO;
3068                 if (!kvm->arch.vpit)
3069                         goto out;
3070                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3071                 if (r)
3072                         goto out;
3073                 r = -EFAULT;
3074                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3075                         goto out;
3076                 r = 0;
3077                 break;
3078         }
3079         case KVM_SET_PIT2: {
3080                 r = -EFAULT;
3081                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3082                         goto out;
3083                 r = -ENXIO;
3084                 if (!kvm->arch.vpit)
3085                         goto out;
3086                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3087                 if (r)
3088                         goto out;
3089                 r = 0;
3090                 break;
3091         }
3092         case KVM_REINJECT_CONTROL: {
3093                 struct kvm_reinject_control control;
3094                 r =  -EFAULT;
3095                 if (copy_from_user(&control, argp, sizeof(control)))
3096                         goto out;
3097                 r = kvm_vm_ioctl_reinject(kvm, &control);
3098                 if (r)
3099                         goto out;
3100                 r = 0;
3101                 break;
3102         }
3103         case KVM_XEN_HVM_CONFIG: {
3104                 r = -EFAULT;
3105                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3106                                    sizeof(struct kvm_xen_hvm_config)))
3107                         goto out;
3108                 r = -EINVAL;
3109                 if (kvm->arch.xen_hvm_config.flags)
3110                         goto out;
3111                 r = 0;
3112                 break;
3113         }
3114         case KVM_SET_CLOCK: {
3115                 struct timespec now;
3116                 struct kvm_clock_data user_ns;
3117                 u64 now_ns;
3118                 s64 delta;
3119
3120                 r = -EFAULT;
3121                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3122                         goto out;
3123
3124                 r = -EINVAL;
3125                 if (user_ns.flags)
3126                         goto out;
3127
3128                 r = 0;
3129                 ktime_get_ts(&now);
3130                 now_ns = timespec_to_ns(&now);
3131                 delta = user_ns.clock - now_ns;
3132                 kvm->arch.kvmclock_offset = delta;
3133                 break;
3134         }
3135         case KVM_GET_CLOCK: {
3136                 struct timespec now;
3137                 struct kvm_clock_data user_ns;
3138                 u64 now_ns;
3139
3140                 ktime_get_ts(&now);
3141                 now_ns = timespec_to_ns(&now);
3142                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3143                 user_ns.flags = 0;
3144
3145                 r = -EFAULT;
3146                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3147                         goto out;
3148                 r = 0;
3149                 break;
3150         }
3151
3152         default:
3153                 ;
3154         }
3155 out:
3156         return r;
3157 }
3158
3159 static void kvm_init_msr_list(void)
3160 {
3161         u32 dummy[2];
3162         unsigned i, j;
3163
3164         /* skip the first msrs in the list. KVM-specific */
3165         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3166                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3167                         continue;
3168                 if (j < i)
3169                         msrs_to_save[j] = msrs_to_save[i];
3170                 j++;
3171         }
3172         num_msrs_to_save = j;
3173 }
3174
3175 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3176                            const void *v)
3177 {
3178         if (vcpu->arch.apic &&
3179             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3180                 return 0;
3181
3182         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3183 }
3184
3185 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3186 {
3187         if (vcpu->arch.apic &&
3188             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3189                 return 0;
3190
3191         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3192 }
3193
3194 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3195                         struct kvm_segment *var, int seg)
3196 {
3197         kvm_x86_ops->set_segment(vcpu, var, seg);
3198 }
3199
3200 void kvm_get_segment(struct kvm_vcpu *vcpu,
3201                      struct kvm_segment *var, int seg)
3202 {
3203         kvm_x86_ops->get_segment(vcpu, var, seg);
3204 }
3205
3206 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3207 {
3208         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3209         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3210 }
3211
3212  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3213 {
3214         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3215         access |= PFERR_FETCH_MASK;
3216         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3217 }
3218
3219 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3220 {
3221         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3222         access |= PFERR_WRITE_MASK;
3223         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3224 }
3225
3226 /* uses this to access any guest's mapped memory without checking CPL */
3227 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3228 {
3229         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3230 }
3231
3232 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3233                                       struct kvm_vcpu *vcpu, u32 access,
3234                                       u32 *error)
3235 {
3236         void *data = val;
3237         int r = X86EMUL_CONTINUE;
3238
3239         while (bytes) {
3240                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3241                 unsigned offset = addr & (PAGE_SIZE-1);
3242                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3243                 int ret;
3244
3245                 if (gpa == UNMAPPED_GVA) {
3246                         r = X86EMUL_PROPAGATE_FAULT;
3247                         goto out;
3248                 }
3249                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3250                 if (ret < 0) {
3251                         r = X86EMUL_IO_NEEDED;
3252                         goto out;
3253                 }
3254
3255                 bytes -= toread;
3256                 data += toread;
3257                 addr += toread;
3258         }
3259 out:
3260         return r;
3261 }
3262
3263 /* used for instruction fetching */
3264 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3265                                 struct kvm_vcpu *vcpu, u32 *error)
3266 {
3267         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3268         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3269                                           access | PFERR_FETCH_MASK, error);
3270 }
3271
3272 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3273                                struct kvm_vcpu *vcpu, u32 *error)
3274 {
3275         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3276         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3277                                           error);
3278 }
3279
3280 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3281                                struct kvm_vcpu *vcpu, u32 *error)
3282 {
3283         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3284 }
3285
3286 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3287                                        unsigned int bytes,
3288                                        struct kvm_vcpu *vcpu,
3289                                        u32 *error)
3290 {
3291         void *data = val;
3292         int r = X86EMUL_CONTINUE;
3293
3294         while (bytes) {
3295                 gpa_t gpa =  vcpu->arch.mmu.gva_to_gpa(vcpu, addr,
3296                                                        PFERR_WRITE_MASK, error);
3297                 unsigned offset = addr & (PAGE_SIZE-1);
3298                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3299                 int ret;
3300
3301                 if (gpa == UNMAPPED_GVA) {
3302                         r = X86EMUL_PROPAGATE_FAULT;
3303                         goto out;
3304                 }
3305                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3306                 if (ret < 0) {
3307                         r = X86EMUL_IO_NEEDED;
3308                         goto out;
3309                 }
3310
3311                 bytes -= towrite;
3312                 data += towrite;
3313                 addr += towrite;
3314         }
3315 out:
3316         return r;
3317 }
3318
3319 static int emulator_read_emulated(unsigned long addr,
3320                                   void *val,
3321                                   unsigned int bytes,
3322                                   unsigned int *error_code,
3323                                   struct kvm_vcpu *vcpu)
3324 {
3325         gpa_t                 gpa;
3326
3327         if (vcpu->mmio_read_completed) {
3328                 memcpy(val, vcpu->mmio_data, bytes);
3329                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3330                                vcpu->mmio_phys_addr, *(u64 *)val);
3331                 vcpu->mmio_read_completed = 0;
3332                 return X86EMUL_CONTINUE;
3333         }
3334
3335         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, error_code);
3336
3337         if (gpa == UNMAPPED_GVA)
3338                 return X86EMUL_PROPAGATE_FAULT;
3339
3340         /* For APIC access vmexit */
3341         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3342                 goto mmio;
3343
3344         if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3345                                 == X86EMUL_CONTINUE)
3346                 return X86EMUL_CONTINUE;
3347
3348 mmio:
3349         /*
3350          * Is this MMIO handled locally?
3351          */
3352         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3353                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3354                 return X86EMUL_CONTINUE;
3355         }
3356
3357         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3358
3359         vcpu->mmio_needed = 1;
3360         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3361         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3362         vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3363         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3364
3365         return X86EMUL_IO_NEEDED;
3366 }
3367
3368 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3369                           const void *val, int bytes)
3370 {
3371         int ret;
3372
3373         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3374         if (ret < 0)
3375                 return 0;
3376         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3377         return 1;
3378 }
3379
3380 static int emulator_write_emulated_onepage(unsigned long addr,
3381                                            const void *val,
3382                                            unsigned int bytes,
3383                                            unsigned int *error_code,
3384                                            struct kvm_vcpu *vcpu)
3385 {
3386         gpa_t                 gpa;
3387
3388         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error_code);
3389
3390         if (gpa == UNMAPPED_GVA)
3391                 return X86EMUL_PROPAGATE_FAULT;
3392
3393         /* For APIC access vmexit */
3394         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3395                 goto mmio;
3396
3397         if (emulator_write_phys(vcpu, gpa, val, bytes))
3398                 return X86EMUL_CONTINUE;
3399
3400 mmio:
3401         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3402         /*
3403          * Is this MMIO handled locally?
3404          */
3405         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3406                 return X86EMUL_CONTINUE;
3407
3408         vcpu->mmio_needed = 1;
3409         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3410         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3411         vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3412         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3413         memcpy(vcpu->run->mmio.data, val, bytes);
3414
3415         return X86EMUL_CONTINUE;
3416 }
3417
3418 int emulator_write_emulated(unsigned long addr,
3419                             const void *val,
3420                             unsigned int bytes,
3421                             unsigned int *error_code,
3422                             struct kvm_vcpu *vcpu)
3423 {
3424         /* Crossing a page boundary? */
3425         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3426                 int rc, now;
3427
3428                 now = -addr & ~PAGE_MASK;
3429                 rc = emulator_write_emulated_onepage(addr, val, now, error_code,
3430                                                      vcpu);
3431                 if (rc != X86EMUL_CONTINUE)
3432                         return rc;
3433                 addr += now;
3434                 val += now;
3435                 bytes -= now;
3436         }
3437         return emulator_write_emulated_onepage(addr, val, bytes, error_code,
3438                                                vcpu);
3439 }
3440
3441 #define CMPXCHG_TYPE(t, ptr, old, new) \
3442         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3443
3444 #ifdef CONFIG_X86_64
3445 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3446 #else
3447 #  define CMPXCHG64(ptr, old, new) \
3448         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3449 #endif
3450
3451 static int emulator_cmpxchg_emulated(unsigned long addr,
3452                                      const void *old,
3453                                      const void *new,
3454                                      unsigned int bytes,
3455                                      unsigned int *error_code,
3456                                      struct kvm_vcpu *vcpu)
3457 {
3458         gpa_t gpa;
3459         struct page *page;
3460         char *kaddr;
3461         bool exchanged;
3462
3463         /* guests cmpxchg8b have to be emulated atomically */
3464         if (bytes > 8 || (bytes & (bytes - 1)))
3465                 goto emul_write;
3466
3467         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3468
3469         if (gpa == UNMAPPED_GVA ||
3470             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3471                 goto emul_write;
3472
3473         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3474                 goto emul_write;
3475
3476         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3477
3478         kaddr = kmap_atomic(page, KM_USER0);
3479         kaddr += offset_in_page(gpa);
3480         switch (bytes) {
3481         case 1:
3482                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3483                 break;
3484         case 2:
3485                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3486                 break;
3487         case 4:
3488                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3489                 break;
3490         case 8:
3491                 exchanged = CMPXCHG64(kaddr, old, new);
3492                 break;
3493         default:
3494                 BUG();
3495         }
3496         kunmap_atomic(kaddr, KM_USER0);
3497         kvm_release_page_dirty(page);
3498
3499         if (!exchanged)
3500                 return X86EMUL_CMPXCHG_FAILED;
3501
3502         kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3503
3504         return X86EMUL_CONTINUE;
3505
3506 emul_write:
3507         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3508
3509         return emulator_write_emulated(addr, new, bytes, error_code, vcpu);
3510 }
3511
3512 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3513 {
3514         /* TODO: String I/O for in kernel device */
3515         int r;
3516
3517         if (vcpu->arch.pio.in)
3518                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3519                                     vcpu->arch.pio.size, pd);
3520         else
3521                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3522                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3523                                      pd);
3524         return r;
3525 }
3526
3527
3528 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3529                              unsigned int count, struct kvm_vcpu *vcpu)
3530 {
3531         if (vcpu->arch.pio.count)
3532                 goto data_avail;
3533
3534         trace_kvm_pio(1, port, size, 1);
3535
3536         vcpu->arch.pio.port = port;
3537         vcpu->arch.pio.in = 1;
3538         vcpu->arch.pio.count  = count;
3539         vcpu->arch.pio.size = size;
3540
3541         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3542         data_avail:
3543                 memcpy(val, vcpu->arch.pio_data, size * count);
3544                 vcpu->arch.pio.count = 0;
3545                 return 1;
3546         }
3547
3548         vcpu->run->exit_reason = KVM_EXIT_IO;
3549         vcpu->run->io.direction = KVM_EXIT_IO_IN;
3550         vcpu->run->io.size = size;
3551         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3552         vcpu->run->io.count = count;
3553         vcpu->run->io.port = port;
3554
3555         return 0;
3556 }
3557
3558 static int emulator_pio_out_emulated(int size, unsigned short port,
3559                               const void *val, unsigned int count,
3560                               struct kvm_vcpu *vcpu)
3561 {
3562         trace_kvm_pio(0, port, size, 1);
3563
3564         vcpu->arch.pio.port = port;
3565         vcpu->arch.pio.in = 0;
3566         vcpu->arch.pio.count = count;
3567         vcpu->arch.pio.size = size;
3568
3569         memcpy(vcpu->arch.pio_data, val, size * count);
3570
3571         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3572                 vcpu->arch.pio.count = 0;
3573                 return 1;
3574         }
3575
3576         vcpu->run->exit_reason = KVM_EXIT_IO;
3577         vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3578         vcpu->run->io.size = size;
3579         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3580         vcpu->run->io.count = count;
3581         vcpu->run->io.port = port;
3582
3583         return 0;
3584 }
3585
3586 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3587 {
3588         return kvm_x86_ops->get_segment_base(vcpu, seg);
3589 }
3590
3591 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3592 {
3593         kvm_mmu_invlpg(vcpu, address);
3594         return X86EMUL_CONTINUE;
3595 }
3596
3597 int emulate_clts(struct kvm_vcpu *vcpu)
3598 {
3599         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3600         kvm_x86_ops->fpu_activate(vcpu);
3601         return X86EMUL_CONTINUE;
3602 }
3603
3604 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
3605 {
3606         return _kvm_get_dr(vcpu, dr, dest);
3607 }
3608
3609 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
3610 {
3611
3612         return __kvm_set_dr(vcpu, dr, value);
3613 }
3614
3615 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3616 {
3617         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3618 }
3619
3620 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
3621 {
3622         unsigned long value;
3623
3624         switch (cr) {
3625         case 0:
3626                 value = kvm_read_cr0(vcpu);
3627                 break;
3628         case 2:
3629                 value = vcpu->arch.cr2;
3630                 break;
3631         case 3:
3632                 value = vcpu->arch.cr3;
3633                 break;
3634         case 4:
3635                 value = kvm_read_cr4(vcpu);
3636                 break;
3637         case 8:
3638                 value = kvm_get_cr8(vcpu);
3639                 break;
3640         default:
3641                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3642                 return 0;
3643         }
3644
3645         return value;
3646 }
3647
3648 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
3649 {
3650         int res = 0;
3651
3652         switch (cr) {
3653         case 0:
3654                 res = __kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3655                 break;
3656         case 2:
3657                 vcpu->arch.cr2 = val;
3658                 break;
3659         case 3:
3660                 res = __kvm_set_cr3(vcpu, val);
3661                 break;
3662         case 4:
3663                 res = __kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3664                 break;
3665         case 8:
3666                 res = __kvm_set_cr8(vcpu, val & 0xfUL);
3667                 break;
3668         default:
3669                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3670                 res = -1;
3671         }
3672
3673         return res;
3674 }
3675
3676 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
3677 {
3678         return kvm_x86_ops->get_cpl(vcpu);
3679 }
3680
3681 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
3682 {
3683         kvm_x86_ops->get_gdt(vcpu, dt);
3684 }
3685
3686 static unsigned long emulator_get_cached_segment_base(int seg,
3687                                                       struct kvm_vcpu *vcpu)
3688 {
3689         return get_segment_base(vcpu, seg);
3690 }
3691
3692 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
3693                                            struct kvm_vcpu *vcpu)
3694 {
3695         struct kvm_segment var;
3696
3697         kvm_get_segment(vcpu, &var, seg);
3698
3699         if (var.unusable)
3700                 return false;
3701
3702         if (var.g)
3703                 var.limit >>= 12;
3704         set_desc_limit(desc, var.limit);
3705         set_desc_base(desc, (unsigned long)var.base);
3706         desc->type = var.type;
3707         desc->s = var.s;
3708         desc->dpl = var.dpl;
3709         desc->p = var.present;
3710         desc->avl = var.avl;
3711         desc->l = var.l;
3712         desc->d = var.db;
3713         desc->g = var.g;
3714
3715         return true;
3716 }
3717
3718 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
3719                                            struct kvm_vcpu *vcpu)
3720 {
3721         struct kvm_segment var;
3722
3723         /* needed to preserve selector */
3724         kvm_get_segment(vcpu, &var, seg);
3725
3726         var.base = get_desc_base(desc);
3727         var.limit = get_desc_limit(desc);
3728         if (desc->g)
3729                 var.limit = (var.limit << 12) | 0xfff;
3730         var.type = desc->type;
3731         var.present = desc->p;
3732         var.dpl = desc->dpl;
3733         var.db = desc->d;
3734         var.s = desc->s;
3735         var.l = desc->l;
3736         var.g = desc->g;
3737         var.avl = desc->avl;
3738         var.present = desc->p;
3739         var.unusable = !var.present;
3740         var.padding = 0;
3741
3742         kvm_set_segment(vcpu, &var, seg);
3743         return;
3744 }
3745
3746 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
3747 {
3748         struct kvm_segment kvm_seg;
3749
3750         kvm_get_segment(vcpu, &kvm_seg, seg);
3751         return kvm_seg.selector;
3752 }
3753
3754 static void emulator_set_segment_selector(u16 sel, int seg,
3755                                           struct kvm_vcpu *vcpu)
3756 {
3757         struct kvm_segment kvm_seg;
3758
3759         kvm_get_segment(vcpu, &kvm_seg, seg);
3760         kvm_seg.selector = sel;
3761         kvm_set_segment(vcpu, &kvm_seg, seg);
3762 }
3763
3764 static struct x86_emulate_ops emulate_ops = {
3765         .read_std            = kvm_read_guest_virt_system,
3766         .write_std           = kvm_write_guest_virt_system,
3767         .fetch               = kvm_fetch_guest_virt,
3768         .read_emulated       = emulator_read_emulated,
3769         .write_emulated      = emulator_write_emulated,
3770         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3771         .pio_in_emulated     = emulator_pio_in_emulated,
3772         .pio_out_emulated    = emulator_pio_out_emulated,
3773         .get_cached_descriptor = emulator_get_cached_descriptor,
3774         .set_cached_descriptor = emulator_set_cached_descriptor,
3775         .get_segment_selector = emulator_get_segment_selector,
3776         .set_segment_selector = emulator_set_segment_selector,
3777         .get_cached_segment_base = emulator_get_cached_segment_base,
3778         .get_gdt             = emulator_get_gdt,
3779         .get_cr              = emulator_get_cr,
3780         .set_cr              = emulator_set_cr,
3781         .cpl                 = emulator_get_cpl,
3782         .get_dr              = emulator_get_dr,
3783         .set_dr              = emulator_set_dr,
3784         .set_msr             = kvm_set_msr,
3785         .get_msr             = kvm_get_msr,
3786 };
3787
3788 static void cache_all_regs(struct kvm_vcpu *vcpu)
3789 {
3790         kvm_register_read(vcpu, VCPU_REGS_RAX);
3791         kvm_register_read(vcpu, VCPU_REGS_RSP);
3792         kvm_register_read(vcpu, VCPU_REGS_RIP);
3793         vcpu->arch.regs_dirty = ~0;
3794 }
3795
3796 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
3797 {
3798         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
3799         /*
3800          * an sti; sti; sequence only disable interrupts for the first
3801          * instruction. So, if the last instruction, be it emulated or
3802          * not, left the system with the INT_STI flag enabled, it
3803          * means that the last instruction is an sti. We should not
3804          * leave the flag on in this case. The same goes for mov ss
3805          */
3806         if (!(int_shadow & mask))
3807                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
3808 }
3809
3810 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
3811 {
3812         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
3813         if (ctxt->exception == PF_VECTOR)
3814                 kvm_inject_page_fault(vcpu, ctxt->cr2, ctxt->error_code);
3815         else if (ctxt->error_code_valid)
3816                 kvm_queue_exception_e(vcpu, ctxt->exception, ctxt->error_code);
3817         else
3818                 kvm_queue_exception(vcpu, ctxt->exception);
3819 }
3820
3821 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
3822 {
3823         ++vcpu->stat.insn_emulation_fail;
3824         trace_kvm_emulate_insn_failed(vcpu);
3825         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3826         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3827         vcpu->run->internal.ndata = 0;
3828         kvm_queue_exception(vcpu, UD_VECTOR);
3829         return EMULATE_FAIL;
3830 }
3831
3832 int emulate_instruction(struct kvm_vcpu *vcpu,
3833                         unsigned long cr2,
3834                         u16 error_code,
3835                         int emulation_type)
3836 {
3837         int r;
3838         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
3839
3840         kvm_clear_exception_queue(vcpu);
3841         vcpu->arch.mmio_fault_cr2 = cr2;
3842         /*
3843          * TODO: fix emulate.c to use guest_read/write_register
3844          * instead of direct ->regs accesses, can save hundred cycles
3845          * on Intel for instructions that don't read/change RSP, for
3846          * for example.
3847          */
3848         cache_all_regs(vcpu);
3849
3850         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3851                 int cs_db, cs_l;
3852                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3853
3854                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3855                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
3856                 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
3857                 vcpu->arch.emulate_ctxt.mode =
3858                         (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3859                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3860                         ? X86EMUL_MODE_VM86 : cs_l
3861                         ? X86EMUL_MODE_PROT64 : cs_db
3862                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3863                 memset(c, 0, sizeof(struct decode_cache));
3864                 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
3865                 vcpu->arch.emulate_ctxt.interruptibility = 0;
3866                 vcpu->arch.emulate_ctxt.exception = -1;
3867
3868                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3869                 trace_kvm_emulate_insn_start(vcpu);
3870
3871                 /* Only allow emulation of specific instructions on #UD
3872                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3873                 if (emulation_type & EMULTYPE_TRAP_UD) {
3874                         if (!c->twobyte)
3875                                 return EMULATE_FAIL;
3876                         switch (c->b) {
3877                         case 0x01: /* VMMCALL */
3878                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3879                                         return EMULATE_FAIL;
3880                                 break;
3881                         case 0x34: /* sysenter */
3882                         case 0x35: /* sysexit */
3883                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3884                                         return EMULATE_FAIL;
3885                                 break;
3886                         case 0x05: /* syscall */
3887                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3888                                         return EMULATE_FAIL;
3889                                 break;
3890                         default:
3891                                 return EMULATE_FAIL;
3892                         }
3893
3894                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3895                                 return EMULATE_FAIL;
3896                 }
3897
3898                 ++vcpu->stat.insn_emulation;
3899                 if (r)  {
3900                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3901                                 return EMULATE_DONE;
3902                         if (emulation_type & EMULTYPE_SKIP)
3903                                 return EMULATE_FAIL;
3904                         return handle_emulation_failure(vcpu);
3905                 }
3906         }
3907
3908         if (emulation_type & EMULTYPE_SKIP) {
3909                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3910                 return EMULATE_DONE;
3911         }
3912
3913         /* this is needed for vmware backdor interface to work since it
3914            changes registers values  during IO operation */
3915         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
3916
3917 restart:
3918         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3919
3920         if (r) { /* emulation failed */
3921                 /*
3922                  * if emulation was due to access to shadowed page table
3923                  * and it failed try to unshadow page and re-entetr the
3924                  * guest to let CPU execute the instruction.
3925                  */
3926                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3927                         return EMULATE_DONE;
3928
3929                 return handle_emulation_failure(vcpu);
3930         }
3931
3932         toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
3933         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3934         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
3935         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
3936
3937         if (vcpu->arch.emulate_ctxt.exception >= 0) {
3938                 inject_emulated_exception(vcpu);
3939                 return EMULATE_DONE;
3940         }
3941
3942         if (vcpu->arch.pio.count) {
3943                 if (!vcpu->arch.pio.in)
3944                         vcpu->arch.pio.count = 0;
3945                 return EMULATE_DO_MMIO;
3946         }
3947
3948         if (vcpu->mmio_needed) {
3949                 if (vcpu->mmio_is_write)
3950                         vcpu->mmio_needed = 0;
3951                 return EMULATE_DO_MMIO;
3952         }
3953
3954         if (vcpu->arch.emulate_ctxt.restart)
3955                 goto restart;
3956
3957         return EMULATE_DONE;
3958 }
3959 EXPORT_SYMBOL_GPL(emulate_instruction);
3960
3961 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
3962 {
3963         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3964         int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
3965         /* do not return to emulator after return from userspace */
3966         vcpu->arch.pio.count = 0;
3967         return ret;
3968 }
3969 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
3970
3971 static void bounce_off(void *info)
3972 {
3973         /* nothing */
3974 }
3975
3976 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3977                                      void *data)
3978 {
3979         struct cpufreq_freqs *freq = data;
3980         struct kvm *kvm;
3981         struct kvm_vcpu *vcpu;
3982         int i, send_ipi = 0;
3983
3984         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3985                 return 0;
3986         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3987                 return 0;
3988         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3989
3990         spin_lock(&kvm_lock);
3991         list_for_each_entry(kvm, &vm_list, vm_list) {
3992                 kvm_for_each_vcpu(i, vcpu, kvm) {
3993                         if (vcpu->cpu != freq->cpu)
3994                                 continue;
3995                         if (!kvm_request_guest_time_update(vcpu))
3996                                 continue;
3997                         if (vcpu->cpu != smp_processor_id())
3998                                 send_ipi++;
3999                 }
4000         }
4001         spin_unlock(&kvm_lock);
4002
4003         if (freq->old < freq->new && send_ipi) {
4004                 /*
4005                  * We upscale the frequency.  Must make the guest
4006                  * doesn't see old kvmclock values while running with
4007                  * the new frequency, otherwise we risk the guest sees
4008                  * time go backwards.
4009                  *
4010                  * In case we update the frequency for another cpu
4011                  * (which might be in guest context) send an interrupt
4012                  * to kick the cpu out of guest context.  Next time
4013                  * guest context is entered kvmclock will be updated,
4014                  * so the guest will not see stale values.
4015                  */
4016                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
4017         }
4018         return 0;
4019 }
4020
4021 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4022         .notifier_call  = kvmclock_cpufreq_notifier
4023 };
4024
4025 static void kvm_timer_init(void)
4026 {
4027         int cpu;
4028
4029         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4030                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4031                                           CPUFREQ_TRANSITION_NOTIFIER);
4032                 for_each_online_cpu(cpu) {
4033                         unsigned long khz = cpufreq_get(cpu);
4034                         if (!khz)
4035                                 khz = tsc_khz;
4036                         per_cpu(cpu_tsc_khz, cpu) = khz;
4037                 }
4038         } else {
4039                 for_each_possible_cpu(cpu)
4040                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
4041         }
4042 }
4043
4044 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4045
4046 static int kvm_is_in_guest(void)
4047 {
4048         return percpu_read(current_vcpu) != NULL;
4049 }
4050
4051 static int kvm_is_user_mode(void)
4052 {
4053         int user_mode = 3;
4054
4055         if (percpu_read(current_vcpu))
4056                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4057
4058         return user_mode != 0;
4059 }
4060
4061 static unsigned long kvm_get_guest_ip(void)
4062 {
4063         unsigned long ip = 0;
4064
4065         if (percpu_read(current_vcpu))
4066                 ip = kvm_rip_read(percpu_read(current_vcpu));
4067
4068         return ip;
4069 }
4070
4071 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4072         .is_in_guest            = kvm_is_in_guest,
4073         .is_user_mode           = kvm_is_user_mode,
4074         .get_guest_ip           = kvm_get_guest_ip,
4075 };
4076
4077 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4078 {
4079         percpu_write(current_vcpu, vcpu);
4080 }
4081 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4082
4083 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4084 {
4085         percpu_write(current_vcpu, NULL);
4086 }
4087 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4088
4089 int kvm_arch_init(void *opaque)
4090 {
4091         int r;
4092         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4093
4094         if (kvm_x86_ops) {
4095                 printk(KERN_ERR "kvm: already loaded the other module\n");
4096                 r = -EEXIST;
4097                 goto out;
4098         }
4099
4100         if (!ops->cpu_has_kvm_support()) {
4101                 printk(KERN_ERR "kvm: no hardware support\n");
4102                 r = -EOPNOTSUPP;
4103                 goto out;
4104         }
4105         if (ops->disabled_by_bios()) {
4106                 printk(KERN_ERR "kvm: disabled by bios\n");
4107                 r = -EOPNOTSUPP;
4108                 goto out;
4109         }
4110
4111         r = kvm_mmu_module_init();
4112         if (r)
4113                 goto out;
4114
4115         kvm_init_msr_list();
4116
4117         kvm_x86_ops = ops;
4118         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4119         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4120         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4121                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
4122
4123         kvm_timer_init();
4124
4125         perf_register_guest_info_callbacks(&kvm_guest_cbs);
4126
4127         return 0;
4128
4129 out:
4130         return r;
4131 }
4132
4133 void kvm_arch_exit(void)
4134 {
4135         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4136
4137         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4138                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4139                                             CPUFREQ_TRANSITION_NOTIFIER);
4140         kvm_x86_ops = NULL;
4141         kvm_mmu_module_exit();
4142 }
4143
4144 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4145 {
4146         ++vcpu->stat.halt_exits;
4147         if (irqchip_in_kernel(vcpu->kvm)) {
4148                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4149                 return 1;
4150         } else {
4151                 vcpu->run->exit_reason = KVM_EXIT_HLT;
4152                 return 0;
4153         }
4154 }
4155 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4156
4157 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4158                            unsigned long a1)
4159 {
4160         if (is_long_mode(vcpu))
4161                 return a0;
4162         else
4163                 return a0 | ((gpa_t)a1 << 32);
4164 }
4165
4166 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4167 {
4168         u64 param, ingpa, outgpa, ret;
4169         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4170         bool fast, longmode;
4171         int cs_db, cs_l;
4172
4173         /*
4174          * hypercall generates UD from non zero cpl and real mode
4175          * per HYPER-V spec
4176          */
4177         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4178                 kvm_queue_exception(vcpu, UD_VECTOR);
4179                 return 0;
4180         }
4181
4182         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4183         longmode = is_long_mode(vcpu) && cs_l == 1;
4184
4185         if (!longmode) {
4186                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4187                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4188                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4189                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4190                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4191                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4192         }
4193 #ifdef CONFIG_X86_64
4194         else {
4195                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4196                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4197                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4198         }
4199 #endif
4200
4201         code = param & 0xffff;
4202         fast = (param >> 16) & 0x1;
4203         rep_cnt = (param >> 32) & 0xfff;
4204         rep_idx = (param >> 48) & 0xfff;
4205
4206         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4207
4208         switch (code) {
4209         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4210                 kvm_vcpu_on_spin(vcpu);
4211                 break;
4212         default:
4213                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4214                 break;
4215         }
4216
4217         ret = res | (((u64)rep_done & 0xfff) << 32);
4218         if (longmode) {
4219                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4220         } else {
4221                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4222                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4223         }
4224
4225         return 1;
4226 }
4227
4228 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4229 {
4230         unsigned long nr, a0, a1, a2, a3, ret;
4231         int r = 1;
4232
4233         if (kvm_hv_hypercall_enabled(vcpu->kvm))
4234                 return kvm_hv_hypercall(vcpu);
4235
4236         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4237         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4238         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4239         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4240         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4241
4242         trace_kvm_hypercall(nr, a0, a1, a2, a3);
4243
4244         if (!is_long_mode(vcpu)) {
4245                 nr &= 0xFFFFFFFF;
4246                 a0 &= 0xFFFFFFFF;
4247                 a1 &= 0xFFFFFFFF;
4248                 a2 &= 0xFFFFFFFF;
4249                 a3 &= 0xFFFFFFFF;
4250         }
4251
4252         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4253                 ret = -KVM_EPERM;
4254                 goto out;
4255         }
4256
4257         switch (nr) {
4258         case KVM_HC_VAPIC_POLL_IRQ:
4259                 ret = 0;
4260                 break;
4261         case KVM_HC_MMU_OP:
4262                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4263                 break;
4264         default:
4265                 ret = -KVM_ENOSYS;
4266                 break;
4267         }
4268 out:
4269         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4270         ++vcpu->stat.hypercalls;
4271         return r;
4272 }
4273 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4274
4275 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4276 {
4277         char instruction[3];
4278         unsigned long rip = kvm_rip_read(vcpu);
4279
4280         /*
4281          * Blow out the MMU to ensure that no other VCPU has an active mapping
4282          * to ensure that the updated hypercall appears atomically across all
4283          * VCPUs.
4284          */
4285         kvm_mmu_zap_all(vcpu->kvm);
4286
4287         kvm_x86_ops->patch_hypercall(vcpu, instruction);
4288
4289         return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4290 }
4291
4292 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4293 {
4294         struct desc_ptr dt = { limit, base };
4295
4296         kvm_x86_ops->set_gdt(vcpu, &dt);
4297 }
4298
4299 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4300 {
4301         struct desc_ptr dt = { limit, base };
4302
4303         kvm_x86_ops->set_idt(vcpu, &dt);
4304 }
4305
4306 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4307 {
4308         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4309         int j, nent = vcpu->arch.cpuid_nent;
4310
4311         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4312         /* when no next entry is found, the current entry[i] is reselected */
4313         for (j = i + 1; ; j = (j + 1) % nent) {
4314                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4315                 if (ej->function == e->function) {
4316                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4317                         return j;
4318                 }
4319         }
4320         return 0; /* silence gcc, even though control never reaches here */
4321 }
4322
4323 /* find an entry with matching function, matching index (if needed), and that
4324  * should be read next (if it's stateful) */
4325 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4326         u32 function, u32 index)
4327 {
4328         if (e->function != function)
4329                 return 0;
4330         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4331                 return 0;
4332         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4333             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4334                 return 0;
4335         return 1;
4336 }
4337
4338 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4339                                               u32 function, u32 index)
4340 {
4341         int i;
4342         struct kvm_cpuid_entry2 *best = NULL;
4343
4344         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4345                 struct kvm_cpuid_entry2 *e;
4346
4347                 e = &vcpu->arch.cpuid_entries[i];
4348                 if (is_matching_cpuid_entry(e, function, index)) {
4349                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4350                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4351                         best = e;
4352                         break;
4353                 }
4354                 /*
4355                  * Both basic or both extended?
4356                  */
4357                 if (((e->function ^ function) & 0x80000000) == 0)
4358                         if (!best || e->function > best->function)
4359                                 best = e;
4360         }
4361         return best;
4362 }
4363 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4364
4365 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4366 {
4367         struct kvm_cpuid_entry2 *best;
4368
4369         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4370         if (!best || best->eax < 0x80000008)
4371                 goto not_found;
4372         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4373         if (best)
4374                 return best->eax & 0xff;
4375 not_found:
4376         return 36;
4377 }
4378
4379 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4380 {
4381         u32 function, index;
4382         struct kvm_cpuid_entry2 *best;
4383
4384         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4385         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4386         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4387         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4388         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4389         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4390         best = kvm_find_cpuid_entry(vcpu, function, index);
4391         if (best) {
4392                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4393                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4394                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4395                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4396         }
4397         kvm_x86_ops->skip_emulated_instruction(vcpu);
4398         trace_kvm_cpuid(function,
4399                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4400                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4401                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4402                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4403 }
4404 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4405
4406 /*
4407  * Check if userspace requested an interrupt window, and that the
4408  * interrupt window is open.
4409  *
4410  * No need to exit to userspace if we already have an interrupt queued.
4411  */
4412 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4413 {
4414         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4415                 vcpu->run->request_interrupt_window &&
4416                 kvm_arch_interrupt_allowed(vcpu));
4417 }
4418
4419 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4420 {
4421         struct kvm_run *kvm_run = vcpu->run;
4422
4423         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4424         kvm_run->cr8 = kvm_get_cr8(vcpu);
4425         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4426         if (irqchip_in_kernel(vcpu->kvm))
4427                 kvm_run->ready_for_interrupt_injection = 1;
4428         else
4429                 kvm_run->ready_for_interrupt_injection =
4430                         kvm_arch_interrupt_allowed(vcpu) &&
4431                         !kvm_cpu_has_interrupt(vcpu) &&
4432                         !kvm_event_needs_reinjection(vcpu);
4433 }
4434
4435 static void vapic_enter(struct kvm_vcpu *vcpu)
4436 {
4437         struct kvm_lapic *apic = vcpu->arch.apic;
4438         struct page *page;
4439
4440         if (!apic || !apic->vapic_addr)
4441                 return;
4442
4443         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4444
4445         vcpu->arch.apic->vapic_page = page;
4446 }
4447
4448 static void vapic_exit(struct kvm_vcpu *vcpu)
4449 {
4450         struct kvm_lapic *apic = vcpu->arch.apic;
4451         int idx;
4452
4453         if (!apic || !apic->vapic_addr)
4454                 return;
4455
4456         idx = srcu_read_lock(&vcpu->kvm->srcu);
4457         kvm_release_page_dirty(apic->vapic_page);
4458         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4459         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4460 }
4461
4462 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4463 {
4464         int max_irr, tpr;
4465
4466         if (!kvm_x86_ops->update_cr8_intercept)
4467                 return;
4468
4469         if (!vcpu->arch.apic)
4470                 return;
4471
4472         if (!vcpu->arch.apic->vapic_addr)
4473                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4474         else
4475                 max_irr = -1;
4476
4477         if (max_irr != -1)
4478                 max_irr >>= 4;
4479
4480         tpr = kvm_lapic_get_cr8(vcpu);
4481
4482         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4483 }
4484
4485 static void inject_pending_event(struct kvm_vcpu *vcpu)
4486 {
4487         /* try to reinject previous events if any */
4488         if (vcpu->arch.exception.pending) {
4489                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
4490                                         vcpu->arch.exception.has_error_code,
4491                                         vcpu->arch.exception.error_code);
4492                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4493                                           vcpu->arch.exception.has_error_code,
4494                                           vcpu->arch.exception.error_code,
4495                                           vcpu->arch.exception.reinject);
4496                 return;
4497         }
4498
4499         if (vcpu->arch.nmi_injected) {
4500                 kvm_x86_ops->set_nmi(vcpu);
4501                 return;
4502         }
4503
4504         if (vcpu->arch.interrupt.pending) {
4505                 kvm_x86_ops->set_irq(vcpu);
4506                 return;
4507         }
4508
4509         /* try to inject new event if pending */
4510         if (vcpu->arch.nmi_pending) {
4511                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4512                         vcpu->arch.nmi_pending = false;
4513                         vcpu->arch.nmi_injected = true;
4514                         kvm_x86_ops->set_nmi(vcpu);
4515                 }
4516         } else if (kvm_cpu_has_interrupt(vcpu)) {
4517                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4518                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4519                                             false);
4520                         kvm_x86_ops->set_irq(vcpu);
4521                 }
4522         }
4523 }
4524
4525 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4526 {
4527         int r;
4528         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4529                 vcpu->run->request_interrupt_window;
4530
4531         if (vcpu->requests)
4532                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4533                         kvm_mmu_unload(vcpu);
4534
4535         r = kvm_mmu_reload(vcpu);
4536         if (unlikely(r))
4537                 goto out;
4538
4539         if (vcpu->requests) {
4540                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4541                         __kvm_migrate_timers(vcpu);
4542                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4543                         kvm_write_guest_time(vcpu);
4544                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4545                         kvm_mmu_sync_roots(vcpu);
4546                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4547                         kvm_x86_ops->tlb_flush(vcpu);
4548                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4549                                        &vcpu->requests)) {
4550                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4551                         r = 0;
4552                         goto out;
4553                 }
4554                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4555                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4556                         r = 0;
4557                         goto out;
4558                 }
4559                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4560                         vcpu->fpu_active = 0;
4561                         kvm_x86_ops->fpu_deactivate(vcpu);
4562                 }
4563         }
4564
4565         preempt_disable();
4566
4567         kvm_x86_ops->prepare_guest_switch(vcpu);
4568         if (vcpu->fpu_active)
4569                 kvm_load_guest_fpu(vcpu);
4570
4571         atomic_set(&vcpu->guest_mode, 1);
4572         smp_wmb();
4573
4574         local_irq_disable();
4575
4576         if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
4577             || need_resched() || signal_pending(current)) {
4578                 atomic_set(&vcpu->guest_mode, 0);
4579                 smp_wmb();
4580                 local_irq_enable();
4581                 preempt_enable();
4582                 r = 1;
4583                 goto out;
4584         }
4585
4586         inject_pending_event(vcpu);
4587
4588         /* enable NMI/IRQ window open exits if needed */
4589         if (vcpu->arch.nmi_pending)
4590                 kvm_x86_ops->enable_nmi_window(vcpu);
4591         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4592                 kvm_x86_ops->enable_irq_window(vcpu);
4593
4594         if (kvm_lapic_enabled(vcpu)) {
4595                 update_cr8_intercept(vcpu);
4596                 kvm_lapic_sync_to_vapic(vcpu);
4597         }
4598
4599         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4600
4601         kvm_guest_enter();
4602
4603         if (unlikely(vcpu->arch.switch_db_regs)) {
4604                 set_debugreg(0, 7);
4605                 set_debugreg(vcpu->arch.eff_db[0], 0);
4606                 set_debugreg(vcpu->arch.eff_db[1], 1);
4607                 set_debugreg(vcpu->arch.eff_db[2], 2);
4608                 set_debugreg(vcpu->arch.eff_db[3], 3);
4609         }
4610
4611         trace_kvm_entry(vcpu->vcpu_id);
4612         kvm_x86_ops->run(vcpu);
4613
4614         /*
4615          * If the guest has used debug registers, at least dr7
4616          * will be disabled while returning to the host.
4617          * If we don't have active breakpoints in the host, we don't
4618          * care about the messed up debug address registers. But if
4619          * we have some of them active, restore the old state.
4620          */
4621         if (hw_breakpoint_active())
4622                 hw_breakpoint_restore();
4623
4624         atomic_set(&vcpu->guest_mode, 0);
4625         smp_wmb();
4626         local_irq_enable();
4627
4628         ++vcpu->stat.exits;
4629
4630         /*
4631          * We must have an instruction between local_irq_enable() and
4632          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4633          * the interrupt shadow.  The stat.exits increment will do nicely.
4634          * But we need to prevent reordering, hence this barrier():
4635          */
4636         barrier();
4637
4638         kvm_guest_exit();
4639
4640         preempt_enable();
4641
4642         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4643
4644         /*
4645          * Profile KVM exit RIPs:
4646          */
4647         if (unlikely(prof_on == KVM_PROFILING)) {
4648                 unsigned long rip = kvm_rip_read(vcpu);
4649                 profile_hit(KVM_PROFILING, (void *)rip);
4650         }
4651
4652
4653         kvm_lapic_sync_from_vapic(vcpu);
4654
4655         r = kvm_x86_ops->handle_exit(vcpu);
4656 out:
4657         return r;
4658 }
4659
4660
4661 static int __vcpu_run(struct kvm_vcpu *vcpu)
4662 {
4663         int r;
4664         struct kvm *kvm = vcpu->kvm;
4665
4666         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4667                 pr_debug("vcpu %d received sipi with vector # %x\n",
4668                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4669                 kvm_lapic_reset(vcpu);
4670                 r = kvm_arch_vcpu_reset(vcpu);
4671                 if (r)
4672                         return r;
4673                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4674         }
4675
4676         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4677         vapic_enter(vcpu);
4678
4679         r = 1;
4680         while (r > 0) {
4681                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4682                         r = vcpu_enter_guest(vcpu);
4683                 else {
4684                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4685                         kvm_vcpu_block(vcpu);
4686                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4687                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4688                         {
4689                                 switch(vcpu->arch.mp_state) {
4690                                 case KVM_MP_STATE_HALTED:
4691                                         vcpu->arch.mp_state =
4692                                                 KVM_MP_STATE_RUNNABLE;
4693                                 case KVM_MP_STATE_RUNNABLE:
4694                                         break;
4695                                 case KVM_MP_STATE_SIPI_RECEIVED:
4696                                 default:
4697                                         r = -EINTR;
4698                                         break;
4699                                 }
4700                         }
4701                 }
4702
4703                 if (r <= 0)
4704                         break;
4705
4706                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4707                 if (kvm_cpu_has_pending_timer(vcpu))
4708                         kvm_inject_pending_timer_irqs(vcpu);
4709
4710                 if (dm_request_for_irq_injection(vcpu)) {
4711                         r = -EINTR;
4712                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4713                         ++vcpu->stat.request_irq_exits;
4714                 }
4715                 if (signal_pending(current)) {
4716                         r = -EINTR;
4717                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4718                         ++vcpu->stat.signal_exits;
4719                 }
4720                 if (need_resched()) {
4721                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4722                         kvm_resched(vcpu);
4723                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4724                 }
4725         }
4726
4727         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4728
4729         vapic_exit(vcpu);
4730
4731         return r;
4732 }
4733
4734 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4735 {
4736         int r;
4737         sigset_t sigsaved;
4738
4739         if (vcpu->sigset_active)
4740                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4741
4742         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4743                 kvm_vcpu_block(vcpu);
4744                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4745                 r = -EAGAIN;
4746                 goto out;
4747         }
4748
4749         /* re-sync apic's tpr */
4750         if (!irqchip_in_kernel(vcpu->kvm))
4751                 kvm_set_cr8(vcpu, kvm_run->cr8);
4752
4753         if (vcpu->arch.pio.count || vcpu->mmio_needed ||
4754             vcpu->arch.emulate_ctxt.restart) {
4755                 if (vcpu->mmio_needed) {
4756                         memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4757                         vcpu->mmio_read_completed = 1;
4758                         vcpu->mmio_needed = 0;
4759                 }
4760                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4761                 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
4762                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4763                 if (r != EMULATE_DONE) {
4764                         r = 0;
4765                         goto out;
4766                 }
4767         }
4768         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4769                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4770                                      kvm_run->hypercall.ret);
4771
4772         r = __vcpu_run(vcpu);
4773
4774 out:
4775         post_kvm_run_save(vcpu);
4776         if (vcpu->sigset_active)
4777                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4778
4779         return r;
4780 }
4781
4782 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4783 {
4784         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4785         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4786         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4787         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4788         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4789         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4790         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4791         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4792 #ifdef CONFIG_X86_64
4793         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4794         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4795         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4796         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4797         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4798         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4799         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4800         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4801 #endif
4802
4803         regs->rip = kvm_rip_read(vcpu);
4804         regs->rflags = kvm_get_rflags(vcpu);
4805
4806         return 0;
4807 }
4808
4809 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4810 {
4811         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4812         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4813         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4814         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4815         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4816         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4817         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4818         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4819 #ifdef CONFIG_X86_64
4820         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4821         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4822         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4823         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4824         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4825         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4826         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4827         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4828 #endif
4829
4830         kvm_rip_write(vcpu, regs->rip);
4831         kvm_set_rflags(vcpu, regs->rflags);
4832
4833         vcpu->arch.exception.pending = false;
4834
4835         return 0;
4836 }
4837
4838 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4839 {
4840         struct kvm_segment cs;
4841
4842         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4843         *db = cs.db;
4844         *l = cs.l;
4845 }
4846 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4847
4848 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4849                                   struct kvm_sregs *sregs)
4850 {
4851         struct desc_ptr dt;
4852
4853         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4854         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4855         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4856         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4857         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4858         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4859
4860         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4861         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4862
4863         kvm_x86_ops->get_idt(vcpu, &dt);
4864         sregs->idt.limit = dt.size;
4865         sregs->idt.base = dt.address;
4866         kvm_x86_ops->get_gdt(vcpu, &dt);
4867         sregs->gdt.limit = dt.size;
4868         sregs->gdt.base = dt.address;
4869
4870         sregs->cr0 = kvm_read_cr0(vcpu);
4871         sregs->cr2 = vcpu->arch.cr2;
4872         sregs->cr3 = vcpu->arch.cr3;
4873         sregs->cr4 = kvm_read_cr4(vcpu);
4874         sregs->cr8 = kvm_get_cr8(vcpu);
4875         sregs->efer = vcpu->arch.efer;
4876         sregs->apic_base = kvm_get_apic_base(vcpu);
4877
4878         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4879
4880         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4881                 set_bit(vcpu->arch.interrupt.nr,
4882                         (unsigned long *)sregs->interrupt_bitmap);
4883
4884         return 0;
4885 }
4886
4887 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4888                                     struct kvm_mp_state *mp_state)
4889 {
4890         mp_state->mp_state = vcpu->arch.mp_state;
4891         return 0;
4892 }
4893
4894 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4895                                     struct kvm_mp_state *mp_state)
4896 {
4897         vcpu->arch.mp_state = mp_state->mp_state;
4898         return 0;
4899 }
4900
4901 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
4902                     bool has_error_code, u32 error_code)
4903 {
4904         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4905         int cs_db, cs_l, ret;
4906         cache_all_regs(vcpu);
4907
4908         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4909
4910         vcpu->arch.emulate_ctxt.vcpu = vcpu;
4911         vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4912         vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4913         vcpu->arch.emulate_ctxt.mode =
4914                 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4915                 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4916                 ? X86EMUL_MODE_VM86 : cs_l
4917                 ? X86EMUL_MODE_PROT64 : cs_db
4918                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4919         memset(c, 0, sizeof(struct decode_cache));
4920         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4921
4922         ret = emulator_task_switch(&vcpu->arch.emulate_ctxt, &emulate_ops,
4923                                    tss_selector, reason, has_error_code,
4924                                    error_code);
4925
4926         if (ret)
4927                 return EMULATE_FAIL;
4928
4929         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4930         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4931         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4932         return EMULATE_DONE;
4933 }
4934 EXPORT_SYMBOL_GPL(kvm_task_switch);
4935
4936 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4937                                   struct kvm_sregs *sregs)
4938 {
4939         int mmu_reset_needed = 0;
4940         int pending_vec, max_bits;
4941         struct desc_ptr dt;
4942
4943         dt.size = sregs->idt.limit;
4944         dt.address = sregs->idt.base;
4945         kvm_x86_ops->set_idt(vcpu, &dt);
4946         dt.size = sregs->gdt.limit;
4947         dt.address = sregs->gdt.base;
4948         kvm_x86_ops->set_gdt(vcpu, &dt);
4949
4950         vcpu->arch.cr2 = sregs->cr2;
4951         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4952         vcpu->arch.cr3 = sregs->cr3;
4953
4954         kvm_set_cr8(vcpu, sregs->cr8);
4955
4956         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
4957         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4958         kvm_set_apic_base(vcpu, sregs->apic_base);
4959
4960         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
4961         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4962         vcpu->arch.cr0 = sregs->cr0;
4963
4964         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
4965         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4966         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
4967                 load_pdptrs(vcpu, vcpu->arch.cr3);
4968                 mmu_reset_needed = 1;
4969         }
4970
4971         if (mmu_reset_needed)
4972                 kvm_mmu_reset_context(vcpu);
4973
4974         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4975         pending_vec = find_first_bit(
4976                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4977         if (pending_vec < max_bits) {
4978                 kvm_queue_interrupt(vcpu, pending_vec, false);
4979                 pr_debug("Set back pending irq %d\n", pending_vec);
4980                 if (irqchip_in_kernel(vcpu->kvm))
4981                         kvm_pic_clear_isr_ack(vcpu->kvm);
4982         }
4983
4984         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4985         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4986         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4987         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4988         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4989         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4990
4991         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4992         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4993
4994         update_cr8_intercept(vcpu);
4995
4996         /* Older userspace won't unhalt the vcpu on reset. */
4997         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4998             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4999             !is_protmode(vcpu))
5000                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5001
5002         return 0;
5003 }
5004
5005 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5006                                         struct kvm_guest_debug *dbg)
5007 {
5008         unsigned long rflags;
5009         int i, r;
5010
5011         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5012                 r = -EBUSY;
5013                 if (vcpu->arch.exception.pending)
5014                         goto out;
5015                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5016                         kvm_queue_exception(vcpu, DB_VECTOR);
5017                 else
5018                         kvm_queue_exception(vcpu, BP_VECTOR);
5019         }
5020
5021         /*
5022          * Read rflags as long as potentially injected trace flags are still
5023          * filtered out.
5024          */
5025         rflags = kvm_get_rflags(vcpu);
5026
5027         vcpu->guest_debug = dbg->control;
5028         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5029                 vcpu->guest_debug = 0;
5030
5031         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5032                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5033                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5034                 vcpu->arch.switch_db_regs =
5035                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5036         } else {
5037                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5038                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5039                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5040         }
5041
5042         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5043                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5044                         get_segment_base(vcpu, VCPU_SREG_CS);
5045
5046         /*
5047          * Trigger an rflags update that will inject or remove the trace
5048          * flags.
5049          */
5050         kvm_set_rflags(vcpu, rflags);
5051
5052         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5053
5054         r = 0;
5055
5056 out:
5057
5058         return r;
5059 }
5060
5061 /*
5062  * Translate a guest virtual address to a guest physical address.
5063  */
5064 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5065                                     struct kvm_translation *tr)
5066 {
5067         unsigned long vaddr = tr->linear_address;
5068         gpa_t gpa;
5069         int idx;
5070
5071         idx = srcu_read_lock(&vcpu->kvm->srcu);
5072         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5073         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5074         tr->physical_address = gpa;
5075         tr->valid = gpa != UNMAPPED_GVA;
5076         tr->writeable = 1;
5077         tr->usermode = 0;
5078
5079         return 0;
5080 }
5081
5082 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5083 {
5084         struct i387_fxsave_struct *fxsave =
5085                         &vcpu->arch.guest_fpu.state->fxsave;
5086
5087         memcpy(fpu->fpr, fxsave->st_space, 128);
5088         fpu->fcw = fxsave->cwd;
5089         fpu->fsw = fxsave->swd;
5090         fpu->ftwx = fxsave->twd;
5091         fpu->last_opcode = fxsave->fop;
5092         fpu->last_ip = fxsave->rip;
5093         fpu->last_dp = fxsave->rdp;
5094         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5095
5096         return 0;
5097 }
5098
5099 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5100 {
5101         struct i387_fxsave_struct *fxsave =
5102                         &vcpu->arch.guest_fpu.state->fxsave;
5103
5104         memcpy(fxsave->st_space, fpu->fpr, 128);
5105         fxsave->cwd = fpu->fcw;
5106         fxsave->swd = fpu->fsw;
5107         fxsave->twd = fpu->ftwx;
5108         fxsave->fop = fpu->last_opcode;
5109         fxsave->rip = fpu->last_ip;
5110         fxsave->rdp = fpu->last_dp;
5111         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5112
5113         return 0;
5114 }
5115
5116 void fx_init(struct kvm_vcpu *vcpu)
5117 {
5118         fpu_alloc(&vcpu->arch.guest_fpu);
5119         fpu_finit(&vcpu->arch.guest_fpu);
5120
5121         vcpu->arch.cr0 |= X86_CR0_ET;
5122 }
5123 EXPORT_SYMBOL_GPL(fx_init);
5124
5125 static void fx_free(struct kvm_vcpu *vcpu)
5126 {
5127         fpu_free(&vcpu->arch.guest_fpu);
5128 }
5129
5130 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5131 {
5132         if (vcpu->guest_fpu_loaded)
5133                 return;
5134
5135         vcpu->guest_fpu_loaded = 1;
5136         unlazy_fpu(current);
5137         fpu_restore_checking(&vcpu->arch.guest_fpu);
5138         trace_kvm_fpu(1);
5139 }
5140
5141 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5142 {
5143         if (!vcpu->guest_fpu_loaded)
5144                 return;
5145
5146         vcpu->guest_fpu_loaded = 0;
5147         fpu_save_init(&vcpu->arch.guest_fpu);
5148         ++vcpu->stat.fpu_reload;
5149         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5150         trace_kvm_fpu(0);
5151 }
5152
5153 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5154 {
5155         if (vcpu->arch.time_page) {
5156                 kvm_release_page_dirty(vcpu->arch.time_page);
5157                 vcpu->arch.time_page = NULL;
5158         }
5159
5160         fx_free(vcpu);
5161         kvm_x86_ops->vcpu_free(vcpu);
5162 }
5163
5164 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5165                                                 unsigned int id)
5166 {
5167         return kvm_x86_ops->vcpu_create(kvm, id);
5168 }
5169
5170 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5171 {
5172         int r;
5173
5174         vcpu->arch.mtrr_state.have_fixed = 1;
5175         vcpu_load(vcpu);
5176         r = kvm_arch_vcpu_reset(vcpu);
5177         if (r == 0)
5178                 r = kvm_mmu_setup(vcpu);
5179         vcpu_put(vcpu);
5180         if (r < 0)
5181                 goto free_vcpu;
5182
5183         return 0;
5184 free_vcpu:
5185         kvm_x86_ops->vcpu_free(vcpu);
5186         return r;
5187 }
5188
5189 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5190 {
5191         vcpu_load(vcpu);
5192         kvm_mmu_unload(vcpu);
5193         vcpu_put(vcpu);
5194
5195         fx_free(vcpu);
5196         kvm_x86_ops->vcpu_free(vcpu);
5197 }
5198
5199 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5200 {
5201         vcpu->arch.nmi_pending = false;
5202         vcpu->arch.nmi_injected = false;
5203
5204         vcpu->arch.switch_db_regs = 0;
5205         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5206         vcpu->arch.dr6 = DR6_FIXED_1;
5207         vcpu->arch.dr7 = DR7_FIXED_1;
5208
5209         return kvm_x86_ops->vcpu_reset(vcpu);
5210 }
5211
5212 int kvm_arch_hardware_enable(void *garbage)
5213 {
5214         /*
5215          * Since this may be called from a hotplug notifcation,
5216          * we can't get the CPU frequency directly.
5217          */
5218         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5219                 int cpu = raw_smp_processor_id();
5220                 per_cpu(cpu_tsc_khz, cpu) = 0;
5221         }
5222
5223         kvm_shared_msr_cpu_online();
5224
5225         return kvm_x86_ops->hardware_enable(garbage);
5226 }
5227
5228 void kvm_arch_hardware_disable(void *garbage)
5229 {
5230         kvm_x86_ops->hardware_disable(garbage);
5231         drop_user_return_notifiers(garbage);
5232 }
5233
5234 int kvm_arch_hardware_setup(void)
5235 {
5236         return kvm_x86_ops->hardware_setup();
5237 }
5238
5239 void kvm_arch_hardware_unsetup(void)
5240 {
5241         kvm_x86_ops->hardware_unsetup();
5242 }
5243
5244 void kvm_arch_check_processor_compat(void *rtn)
5245 {
5246         kvm_x86_ops->check_processor_compatibility(rtn);
5247 }
5248
5249 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5250 {
5251         struct page *page;
5252         struct kvm *kvm;
5253         int r;
5254
5255         BUG_ON(vcpu->kvm == NULL);
5256         kvm = vcpu->kvm;
5257
5258         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5259         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5260                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5261         else
5262                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5263
5264         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5265         if (!page) {
5266                 r = -ENOMEM;
5267                 goto fail;
5268         }
5269         vcpu->arch.pio_data = page_address(page);
5270
5271         r = kvm_mmu_create(vcpu);
5272         if (r < 0)
5273                 goto fail_free_pio_data;
5274
5275         if (irqchip_in_kernel(kvm)) {
5276                 r = kvm_create_lapic(vcpu);
5277                 if (r < 0)
5278                         goto fail_mmu_destroy;
5279         }
5280
5281         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5282                                        GFP_KERNEL);
5283         if (!vcpu->arch.mce_banks) {
5284                 r = -ENOMEM;
5285                 goto fail_free_lapic;
5286         }
5287         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5288
5289         return 0;
5290 fail_free_lapic:
5291         kvm_free_lapic(vcpu);
5292 fail_mmu_destroy:
5293         kvm_mmu_destroy(vcpu);
5294 fail_free_pio_data:
5295         free_page((unsigned long)vcpu->arch.pio_data);
5296 fail:
5297         return r;
5298 }
5299
5300 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5301 {
5302         int idx;
5303
5304         kfree(vcpu->arch.mce_banks);
5305         kvm_free_lapic(vcpu);
5306         idx = srcu_read_lock(&vcpu->kvm->srcu);
5307         kvm_mmu_destroy(vcpu);
5308         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5309         free_page((unsigned long)vcpu->arch.pio_data);
5310 }
5311
5312 struct  kvm *kvm_arch_create_vm(void)
5313 {
5314         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5315
5316         if (!kvm)
5317                 return ERR_PTR(-ENOMEM);
5318
5319         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5320         if (!kvm->arch.aliases) {
5321                 kfree(kvm);
5322                 return ERR_PTR(-ENOMEM);
5323         }
5324
5325         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5326         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5327
5328         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5329         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5330
5331         rdtscll(kvm->arch.vm_init_tsc);
5332
5333         return kvm;
5334 }
5335
5336 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5337 {
5338         vcpu_load(vcpu);
5339         kvm_mmu_unload(vcpu);
5340         vcpu_put(vcpu);
5341 }
5342
5343 static void kvm_free_vcpus(struct kvm *kvm)
5344 {
5345         unsigned int i;
5346         struct kvm_vcpu *vcpu;
5347
5348         /*
5349          * Unpin any mmu pages first.
5350          */
5351         kvm_for_each_vcpu(i, vcpu, kvm)
5352                 kvm_unload_vcpu_mmu(vcpu);
5353         kvm_for_each_vcpu(i, vcpu, kvm)
5354                 kvm_arch_vcpu_free(vcpu);
5355
5356         mutex_lock(&kvm->lock);
5357         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5358                 kvm->vcpus[i] = NULL;
5359
5360         atomic_set(&kvm->online_vcpus, 0);
5361         mutex_unlock(&kvm->lock);
5362 }
5363
5364 void kvm_arch_sync_events(struct kvm *kvm)
5365 {
5366         kvm_free_all_assigned_devices(kvm);
5367 }
5368
5369 void kvm_arch_destroy_vm(struct kvm *kvm)
5370 {
5371         kvm_iommu_unmap_guest(kvm);
5372         kvm_free_pit(kvm);
5373         kfree(kvm->arch.vpic);
5374         kfree(kvm->arch.vioapic);
5375         kvm_free_vcpus(kvm);
5376         kvm_free_physmem(kvm);
5377         if (kvm->arch.apic_access_page)
5378                 put_page(kvm->arch.apic_access_page);
5379         if (kvm->arch.ept_identity_pagetable)
5380                 put_page(kvm->arch.ept_identity_pagetable);
5381         cleanup_srcu_struct(&kvm->srcu);
5382         kfree(kvm->arch.aliases);
5383         kfree(kvm);
5384 }
5385
5386 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5387                                 struct kvm_memory_slot *memslot,
5388                                 struct kvm_memory_slot old,
5389                                 struct kvm_userspace_memory_region *mem,
5390                                 int user_alloc)
5391 {
5392         int npages = memslot->npages;
5393
5394         /*To keep backward compatibility with older userspace,
5395          *x86 needs to hanlde !user_alloc case.
5396          */
5397         if (!user_alloc) {
5398                 if (npages && !old.rmap) {
5399                         unsigned long userspace_addr;
5400
5401                         down_write(&current->mm->mmap_sem);
5402                         userspace_addr = do_mmap(NULL, 0,
5403                                                  npages * PAGE_SIZE,
5404                                                  PROT_READ | PROT_WRITE,
5405                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5406                                                  0);
5407                         up_write(&current->mm->mmap_sem);
5408
5409                         if (IS_ERR((void *)userspace_addr))
5410                                 return PTR_ERR((void *)userspace_addr);
5411
5412                         memslot->userspace_addr = userspace_addr;
5413                 }
5414         }
5415
5416
5417         return 0;
5418 }
5419
5420 void kvm_arch_commit_memory_region(struct kvm *kvm,
5421                                 struct kvm_userspace_memory_region *mem,
5422                                 struct kvm_memory_slot old,
5423                                 int user_alloc)
5424 {
5425
5426         int npages = mem->memory_size >> PAGE_SHIFT;
5427
5428         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5429                 int ret;
5430
5431                 down_write(&current->mm->mmap_sem);
5432                 ret = do_munmap(current->mm, old.userspace_addr,
5433                                 old.npages * PAGE_SIZE);
5434                 up_write(&current->mm->mmap_sem);
5435                 if (ret < 0)
5436                         printk(KERN_WARNING
5437                                "kvm_vm_ioctl_set_memory_region: "
5438                                "failed to munmap memory\n");
5439         }
5440
5441         spin_lock(&kvm->mmu_lock);
5442         if (!kvm->arch.n_requested_mmu_pages) {
5443                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5444                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5445         }
5446
5447         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5448         spin_unlock(&kvm->mmu_lock);
5449 }
5450
5451 void kvm_arch_flush_shadow(struct kvm *kvm)
5452 {
5453         kvm_mmu_zap_all(kvm);
5454         kvm_reload_remote_mmus(kvm);
5455 }
5456
5457 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5458 {
5459         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5460                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5461                 || vcpu->arch.nmi_pending ||
5462                 (kvm_arch_interrupt_allowed(vcpu) &&
5463                  kvm_cpu_has_interrupt(vcpu));
5464 }
5465
5466 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5467 {
5468         int me;
5469         int cpu = vcpu->cpu;
5470
5471         if (waitqueue_active(&vcpu->wq)) {
5472                 wake_up_interruptible(&vcpu->wq);
5473                 ++vcpu->stat.halt_wakeup;
5474         }
5475
5476         me = get_cpu();
5477         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5478                 if (atomic_xchg(&vcpu->guest_mode, 0))
5479                         smp_send_reschedule(cpu);
5480         put_cpu();
5481 }
5482
5483 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5484 {
5485         return kvm_x86_ops->interrupt_allowed(vcpu);
5486 }
5487
5488 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5489 {
5490         unsigned long current_rip = kvm_rip_read(vcpu) +
5491                 get_segment_base(vcpu, VCPU_SREG_CS);
5492
5493         return current_rip == linear_rip;
5494 }
5495 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5496
5497 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5498 {
5499         unsigned long rflags;
5500
5501         rflags = kvm_x86_ops->get_rflags(vcpu);
5502         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5503                 rflags &= ~X86_EFLAGS_TF;
5504         return rflags;
5505 }
5506 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5507
5508 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5509 {
5510         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5511             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5512                 rflags |= X86_EFLAGS_TF;
5513         kvm_x86_ops->set_rflags(vcpu, rflags);
5514 }
5515 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5516
5517 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5518 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5519 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5520 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5521 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5522 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5523 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5524 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5525 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5526 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5527 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5528 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);