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
  56 #define MAX_IO_MSRS 256
  57 #define CR0_RESERVED_BITS                                               \
  58         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  59                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  60                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  61 #define CR4_RESERVED_BITS                                               \
  62         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  63                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  64                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  65                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  66
  67 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  68
  69 #define KVM_MAX_MCE_BANKS 32
  70 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  71
  72 /* EFER defaults:
  73  * - enable syscall per default because its emulated by KVM
  74  * - enable LME and LMA per default on 64 bit KVM
  75  */
  76 #ifdef CONFIG_X86_64
  77 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  78 #else
  79 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  80 #endif
  81
  82 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  83 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  84
  85 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  86 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  87                                     struct kvm_cpuid_entry2 __user *entries);
  88
  89 struct kvm_x86_ops *kvm_x86_ops;
  90 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  91
  92 int ignore_msrs = 0;
  93 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  94
  95 #define KVM_NR_SHARED_MSRS 16
  96
  97 struct kvm_shared_msrs_global {
  98         int nr;
  99         u32 msrs[KVM_NR_SHARED_MSRS];
 100 };
 101
 102 struct kvm_shared_msrs {
 103         struct user_return_notifier urn;
 104         bool registered;
 105         struct kvm_shared_msr_values {
 106                 u64 host;
 107                 u64 curr;
 108         } values[KVM_NR_SHARED_MSRS];
 109 };
 110
 111 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 112 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 113
 114 struct kvm_stats_debugfs_item debugfs_entries[] = {
 115         { "pf_fixed", VCPU_STAT(pf_fixed) },
 116         { "pf_guest", VCPU_STAT(pf_guest) },
 117         { "tlb_flush", VCPU_STAT(tlb_flush) },
 118         { "invlpg", VCPU_STAT(invlpg) },
 119         { "exits", VCPU_STAT(exits) },
 120         { "io_exits", VCPU_STAT(io_exits) },
 121         { "mmio_exits", VCPU_STAT(mmio_exits) },
 122         { "signal_exits", VCPU_STAT(signal_exits) },
 123         { "irq_window", VCPU_STAT(irq_window_exits) },
 124         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 125         { "halt_exits", VCPU_STAT(halt_exits) },
 126         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 127         { "hypercalls", VCPU_STAT(hypercalls) },
 128         { "request_irq", VCPU_STAT(request_irq_exits) },
 129         { "irq_exits", VCPU_STAT(irq_exits) },
 130         { "host_state_reload", VCPU_STAT(host_state_reload) },
 131         { "efer_reload", VCPU_STAT(efer_reload) },
 132         { "fpu_reload", VCPU_STAT(fpu_reload) },
 133         { "insn_emulation", VCPU_STAT(insn_emulation) },
 134         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 135         { "irq_injections", VCPU_STAT(irq_injections) },
 136         { "nmi_injections", VCPU_STAT(nmi_injections) },
 137         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 138         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 139         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 140         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 141         { "mmu_flooded", VM_STAT(mmu_flooded) },
 142         { "mmu_recycled", VM_STAT(mmu_recycled) },
 143         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 144         { "mmu_unsync", VM_STAT(mmu_unsync) },
 145         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 146         { "largepages", VM_STAT(lpages) },
 147         { NULL }
 148 };
 149
 150 static void kvm_on_user_return(struct user_return_notifier *urn)
 151 {
 152         unsigned slot;
 153         struct kvm_shared_msrs *locals
 154                 = container_of(urn, struct kvm_shared_msrs, urn);
 155         struct kvm_shared_msr_values *values;
 156
 157         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 158                 values = &locals->values[slot];
 159                 if (values->host != values->curr) {
 160                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 161                         values->curr = values->host;
 162                 }
 163         }
 164         locals->registered = false;
 165         user_return_notifier_unregister(urn);
 166 }
 167
 168 static void shared_msr_update(unsigned slot, u32 msr)
 169 {
 170         struct kvm_shared_msrs *smsr;
 171         u64 value;
 172
 173         smsr = &__get_cpu_var(shared_msrs);
 174         /* only read, and nobody should modify it at this time,
 175          * so don't need lock */
 176         if (slot >= shared_msrs_global.nr) {
 177                 printk(KERN_ERR "kvm: invalid MSR slot!");
 178                 return;
 179         }
 180         rdmsrl_safe(msr, &value);
 181         smsr->values[slot].host = value;
 182         smsr->values[slot].curr = value;
 183 }
 184
 185 void kvm_define_shared_msr(unsigned slot, u32 msr)
 186 {
 187         if (slot >= shared_msrs_global.nr)
 188                 shared_msrs_global.nr = slot + 1;
 189         shared_msrs_global.msrs[slot] = msr;
 190         /* we need ensured the shared_msr_global have been updated */
 191         smp_wmb();
 192 }
 193 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 194
 195 static void kvm_shared_msr_cpu_online(void)
 196 {
 197         unsigned i;
 198
 199         for (i = 0; i < shared_msrs_global.nr; ++i)
 200                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 201 }
 202
 203 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 204 {
 205         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 206
 207         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 208                 return;
 209         smsr->values[slot].curr = value;
 210         wrmsrl(shared_msrs_global.msrs[slot], value);
 211         if (!smsr->registered) {
 212                 smsr->urn.on_user_return = kvm_on_user_return;
 213                 user_return_notifier_register(&smsr->urn);
 214                 smsr->registered = true;
 215         }
 216 }
 217 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 218
 219 static void drop_user_return_notifiers(void *ignore)
 220 {
 221         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 222
 223         if (smsr->registered)
 224                 kvm_on_user_return(&smsr->urn);
 225 }
 226
 227 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 228 {
 229         if (irqchip_in_kernel(vcpu->kvm))
 230                 return vcpu->arch.apic_base;
 231         else
 232                 return vcpu->arch.apic_base;
 233 }
 234 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 235
 236 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 237 {
 238         /* TODO: reserve bits check */
 239         if (irqchip_in_kernel(vcpu->kvm))
 240                 kvm_lapic_set_base(vcpu, data);
 241         else
 242                 vcpu->arch.apic_base = data;
 243 }
 244 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 245
 246 #define EXCPT_BENIGN            0
 247 #define EXCPT_CONTRIBUTORY      1
 248 #define EXCPT_PF                2
 249
 250 static int exception_class(int vector)
 251 {
 252         switch (vector) {
 253         case PF_VECTOR:
 254                 return EXCPT_PF;
 255         case DE_VECTOR:
 256         case TS_VECTOR:
 257         case NP_VECTOR:
 258         case SS_VECTOR:
 259         case GP_VECTOR:
 260                 return EXCPT_CONTRIBUTORY;
 261         default:
 262                 break;
 263         }
 264         return EXCPT_BENIGN;
 265 }
 266
 267 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 268                 unsigned nr, bool has_error, u32 error_code,
 269                 bool reinject)
 270 {
 271         u32 prev_nr;
 272         int class1, class2;
 273
 274         if (!vcpu->arch.exception.pending) {
 275         queue:
 276                 vcpu->arch.exception.pending = true;
 277                 vcpu->arch.exception.has_error_code = has_error;
 278                 vcpu->arch.exception.nr = nr;
 279                 vcpu->arch.exception.error_code = error_code;
 280                 vcpu->arch.exception.reinject = reinject;
 281                 return;
 282         }
 283
 284         /* to check exception */
 285         prev_nr = vcpu->arch.exception.nr;
 286         if (prev_nr == DF_VECTOR) {
 287                 /* triple fault -> shutdown */
 288                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 289                 return;
 290         }
 291         class1 = exception_class(prev_nr);
 292         class2 = exception_class(nr);
 293         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 294                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 295                 /* generate double fault per SDM Table 5-5 */
 296                 vcpu->arch.exception.pending = true;
 297                 vcpu->arch.exception.has_error_code = true;
 298                 vcpu->arch.exception.nr = DF_VECTOR;
 299                 vcpu->arch.exception.error_code = 0;
 300         } else
 301                 /* replace previous exception with a new one in a hope
 302                    that instruction re-execution will regenerate lost
 303                    exception */
 304                 goto queue;
 305 }
 306
 307 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 308 {
 309         kvm_multiple_exception(vcpu, nr, false, 0, false);
 310 }
 311 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 312
 313 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 314 {
 315         kvm_multiple_exception(vcpu, nr, false, 0, true);
 316 }
 317 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
 318
 319 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 320                            u32 error_code)
 321 {
 322         ++vcpu->stat.pf_guest;
 323         vcpu->arch.cr2 = addr;
 324         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 325 }
 326
 327 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 328 {
 329         vcpu->arch.nmi_pending = 1;
 330 }
 331 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 332
 333 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 334 {
 335         kvm_multiple_exception(vcpu, nr, true, error_code, false);
 336 }
 337 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 338
 339 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 340 {
 341         kvm_multiple_exception(vcpu, nr, true, error_code, true);
 342 }
 343 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
 344
 345 /*
 346  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 347  * a #GP and return false.
 348  */
 349 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 350 {
 351         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 352                 return true;
 353         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 354         return false;
 355 }
 356 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 357
 358 /*
 359  * Load the pae pdptrs.  Return true is they are all valid.
 360  */
 361 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 362 {
 363         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 364         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 365         int i;
 366         int ret;
 367         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 368
 369         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 370                                   offset * sizeof(u64), sizeof(pdpte));
 371         if (ret < 0) {
 372                 ret = 0;
 373                 goto out;
 374         }
 375         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 376                 if (is_present_gpte(pdpte[i]) &&
 377                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 378                         ret = 0;
 379                         goto out;
 380                 }
 381         }
 382         ret = 1;
 383
 384         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 385         __set_bit(VCPU_EXREG_PDPTR,
 386                   (unsigned long *)&vcpu->arch.regs_avail);
 387         __set_bit(VCPU_EXREG_PDPTR,
 388                   (unsigned long *)&vcpu->arch.regs_dirty);
 389 out:
 390
 391         return ret;
 392 }
 393 EXPORT_SYMBOL_GPL(load_pdptrs);
 394
 395 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 396 {
 397         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 398         bool changed = true;
 399         int r;
 400
 401         if (is_long_mode(vcpu) || !is_pae(vcpu))
 402                 return false;
 403
 404         if (!test_bit(VCPU_EXREG_PDPTR,
 405                       (unsigned long *)&vcpu->arch.regs_avail))
 406                 return true;
 407
 408         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 409         if (r < 0)
 410                 goto out;
 411         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 412 out:
 413
 414         return changed;
 415 }
 416
 417 static int __kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 418 {
 419         unsigned long old_cr0 = kvm_read_cr0(vcpu);
 420         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
 421                                     X86_CR0_CD | X86_CR0_NW;
 422
 423         cr0 |= X86_CR0_ET;
 424
 425 #ifdef CONFIG_X86_64
 426         if (cr0 & 0xffffffff00000000UL)
 427                 return 1;
 428 #endif
 429
 430         cr0 &= ~CR0_RESERVED_BITS;
 431
 432         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
 433                 return 1;
 434
 435         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
 436                 return 1;
 437
 438         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 439 #ifdef CONFIG_X86_64
 440                 if ((vcpu->arch.efer & EFER_LME)) {
 441                         int cs_db, cs_l;
 442
 443                         if (!is_pae(vcpu))
 444                                 return 1;
 445                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 446                         if (cs_l)
 447                                 return 1;
 448                 } else
 449 #endif
 450                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3))
 451                         return 1;
 452         }
 453
 454         kvm_x86_ops->set_cr0(vcpu, cr0);
 455
 456         if ((cr0 ^ old_cr0) & update_bits)
 457                 kvm_mmu_reset_context(vcpu);
 458         return 0;
 459 }
 460
 461 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 462 {
 463         if (__kvm_set_cr0(vcpu, cr0))
 464                 kvm_inject_gp(vcpu, 0);
 465 }
 466 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 467
 468 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 469 {
 470         kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
 471 }
 472 EXPORT_SYMBOL_GPL(kvm_lmsw);
 473
 474 int __kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 475 {
 476         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 477         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 478
 479         if (cr4 & CR4_RESERVED_BITS)
 480                 return 1;
 481
 482         if (is_long_mode(vcpu)) {
 483                 if (!(cr4 & X86_CR4_PAE))
 484                         return 1;
 485         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 486                    && ((cr4 ^ old_cr4) & pdptr_bits)
 487                    && !load_pdptrs(vcpu, vcpu->arch.cr3))
 488                 return 1;
 489
 490         if (cr4 & X86_CR4_VMXE)
 491                 return 1;
 492
 493         kvm_x86_ops->set_cr4(vcpu, cr4);
 494
 495         if ((cr4 ^ old_cr4) & pdptr_bits)
 496                 kvm_mmu_reset_context(vcpu);
 497
 498         return 0;
 499 }
 500
 501 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 502 {
 503         if (__kvm_set_cr4(vcpu, cr4))
 504                 kvm_inject_gp(vcpu, 0);
 505 }
 506 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 507
 508 static int __kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 509 {
 510         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 511                 kvm_mmu_sync_roots(vcpu);
 512                 kvm_mmu_flush_tlb(vcpu);
 513                 return 0;
 514         }
 515
 516         if (is_long_mode(vcpu)) {
 517                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
 518                         return 1;
 519         } else {
 520                 if (is_pae(vcpu)) {
 521                         if (cr3 & CR3_PAE_RESERVED_BITS)
 522                                 return 1;
 523                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3))
 524                                 return 1;
 525                 }
 526                 /*
 527                  * We don't check reserved bits in nonpae mode, because
 528                  * this isn't enforced, and VMware depends on this.
 529                  */
 530         }
 531
 532         /*
 533          * Does the new cr3 value map to physical memory? (Note, we
 534          * catch an invalid cr3 even in real-mode, because it would
 535          * cause trouble later on when we turn on paging anyway.)
 536          *
 537          * A real CPU would silently accept an invalid cr3 and would
 538          * attempt to use it - with largely undefined (and often hard
 539          * to debug) behavior on the guest side.
 540          */
 541         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 542                 return 1;
 543         vcpu->arch.cr3 = cr3;
 544         vcpu->arch.mmu.new_cr3(vcpu);
 545         return 0;
 546 }
 547
 548 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 549 {
 550         if (__kvm_set_cr3(vcpu, cr3))
 551                 kvm_inject_gp(vcpu, 0);
 552 }
 553 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 554
 555 int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 556 {
 557         if (cr8 & CR8_RESERVED_BITS)
 558                 return 1;
 559         if (irqchip_in_kernel(vcpu->kvm))
 560                 kvm_lapic_set_tpr(vcpu, cr8);
 561         else
 562                 vcpu->arch.cr8 = cr8;
 563         return 0;
 564 }
 565
 566 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 567 {
 568         if (__kvm_set_cr8(vcpu, cr8))
 569                 kvm_inject_gp(vcpu, 0);
 570 }
 571 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 572
 573 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 574 {
 575         if (irqchip_in_kernel(vcpu->kvm))
 576                 return kvm_lapic_get_cr8(vcpu);
 577         else
 578                 return vcpu->arch.cr8;
 579 }
 580 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 581
 582 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 583 {
 584         switch (dr) {
 585         case 0 ... 3:
 586                 vcpu->arch.db[dr] = val;
 587                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
 588                         vcpu->arch.eff_db[dr] = val;
 589                 break;
 590         case 4:
 591                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 592                         return 1; /* #UD */
 593                 /* fall through */
 594         case 6:
 595                 if (val & 0xffffffff00000000ULL)
 596                         return -1; /* #GP */
 597                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
 598                 break;
 599         case 5:
 600                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 601                         return 1; /* #UD */
 602                 /* fall through */
 603         default: /* 7 */
 604                 if (val & 0xffffffff00000000ULL)
 605                         return -1; /* #GP */
 606                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
 607                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
 608                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
 609                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
 610                 }
 611                 break;
 612         }
 613
 614         return 0;
 615 }
 616
 617 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
 618 {
 619         int res;
 620
 621         res = __kvm_set_dr(vcpu, dr, val);
 622         if (res > 0)
 623                 kvm_queue_exception(vcpu, UD_VECTOR);
 624         else if (res < 0)
 625                 kvm_inject_gp(vcpu, 0);
 626
 627         return res;
 628 }
 629 EXPORT_SYMBOL_GPL(kvm_set_dr);
 630
 631 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 632 {
 633         switch (dr) {
 634         case 0 ... 3:
 635                 *val = vcpu->arch.db[dr];
 636                 break;
 637         case 4:
 638                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 639                         return 1;
 640                 /* fall through */
 641         case 6:
 642                 *val = vcpu->arch.dr6;
 643                 break;
 644         case 5:
 645                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
 646                         return 1;
 647                 /* fall through */
 648         default: /* 7 */
 649                 *val = vcpu->arch.dr7;
 650                 break;
 651         }
 652
 653         return 0;
 654 }
 655
 656 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
 657 {
 658         if (_kvm_get_dr(vcpu, dr, val)) {
 659                 kvm_queue_exception(vcpu, UD_VECTOR);
 660                 return 1;
 661         }
 662         return 0;
 663 }
 664 EXPORT_SYMBOL_GPL(kvm_get_dr);
 665
 666 static inline u32 bit(int bitno)
 667 {
 668         return 1 << (bitno & 31);
 669 }
 670
 671 /*
 672  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 673  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 674  *
 675  * This list is modified at module load time to reflect the
 676  * capabilities of the host cpu. This capabilities test skips MSRs that are
 677  * kvm-specific. Those are put in the beginning of the list.
 678  */
 679
 680 #define KVM_SAVE_MSRS_BEGIN     7
 681 static u32 msrs_to_save[] = {
 682         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 683         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
 684         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 685         HV_X64_MSR_APIC_ASSIST_PAGE,
 686         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 687         MSR_K6_STAR,
 688 #ifdef CONFIG_X86_64
 689         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 690 #endif
 691         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 692 };
 693
 694 static unsigned num_msrs_to_save;
 695
 696 static u32 emulated_msrs[] = {
 697         MSR_IA32_MISC_ENABLE,
 698 };
 699
 700 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
 701 {
 702         u64 old_efer = vcpu->arch.efer;
 703
 704         if (efer & efer_reserved_bits)
 705                 return 1;
 706
 707         if (is_paging(vcpu)
 708             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
 709                 return 1;
 710
 711         if (efer & EFER_FFXSR) {
 712                 struct kvm_cpuid_entry2 *feat;
 713
 714                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 715                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
 716                         return 1;
 717         }
 718
 719         if (efer & EFER_SVME) {
 720                 struct kvm_cpuid_entry2 *feat;
 721
 722                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 723                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
 724                         return 1;
 725         }
 726
 727         efer &= ~EFER_LMA;
 728         efer |= vcpu->arch.efer & EFER_LMA;
 729
 730         kvm_x86_ops->set_efer(vcpu, efer);
 731
 732         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 733         kvm_mmu_reset_context(vcpu);
 734
 735         /* Update reserved bits */
 736         if ((efer ^ old_efer) & EFER_NX)
 737                 kvm_mmu_reset_context(vcpu);
 738
 739         return 0;
 740 }
 741
 742 void kvm_enable_efer_bits(u64 mask)
 743 {
 744        efer_reserved_bits &= ~mask;
 745 }
 746 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 747
 748
 749 /*
 750  * Writes msr value into into the appropriate "register".
 751  * Returns 0 on success, non-0 otherwise.
 752  * Assumes vcpu_load() was already called.
 753  */
 754 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 755 {
 756         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 757 }
 758
 759 /*
 760  * Adapt set_msr() to msr_io()'s calling convention
 761  */
 762 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 763 {
 764         return kvm_set_msr(vcpu, index, *data);
 765 }
 766
 767 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 768 {
 769         int version;
 770         int r;
 771         struct pvclock_wall_clock wc;
 772         struct timespec boot;
 773
 774         if (!wall_clock)
 775                 return;
 776
 777         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
 778         if (r)
 779                 return;
 780
 781         if (version & 1)
 782                 ++version;  /* first time write, random junk */
 783
 784         ++version;
 785
 786         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 787
 788         /*
 789          * The guest calculates current wall clock time by adding
 790          * system time (updated by kvm_write_guest_time below) to the
 791          * wall clock specified here.  guest system time equals host
 792          * system time for us, thus we must fill in host boot time here.
 793          */
 794         getboottime(&boot);
 795
 796         wc.sec = boot.tv_sec;
 797         wc.nsec = boot.tv_nsec;
 798         wc.version = version;
 799
 800         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 801
 802         version++;
 803         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 804 }
 805
 806 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 807 {
 808         uint32_t quotient, remainder;
 809
 810         /* Don't try to replace with do_div(), this one calculates
 811          * "(dividend << 32) / divisor" */
 812         __asm__ ( "divl %4"
 813                   : "=a" (quotient), "=d" (remainder)
 814                   : "0" (0), "1" (dividend), "r" (divisor) );
 815         return quotient;
 816 }
 817
 818 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 819 {
 820         uint64_t nsecs = 1000000000LL;
 821         int32_t  shift = 0;
 822         uint64_t tps64;
 823         uint32_t tps32;
 824
 825         tps64 = tsc_khz * 1000LL;
 826         while (tps64 > nsecs*2) {
 827                 tps64 >>= 1;
 828                 shift--;
 829         }
 830
 831         tps32 = (uint32_t)tps64;
 832         while (tps32 <= (uint32_t)nsecs) {
 833                 tps32 <<= 1;
 834                 shift++;
 835         }
 836
 837         hv_clock->tsc_shift = shift;
 838         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 839
 840         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 841                  __func__, tsc_khz, hv_clock->tsc_shift,
 842                  hv_clock->tsc_to_system_mul);
 843 }
 844
 845 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 846
 847 static void kvm_write_guest_time(struct kvm_vcpu *v)
 848 {
 849         struct timespec ts;
 850         unsigned long flags;
 851         struct kvm_vcpu_arch *vcpu = &v->arch;
 852         void *shared_kaddr;
 853         unsigned long this_tsc_khz;
 854
 855         if ((!vcpu->time_page))
 856                 return;
 857
 858         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 859         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 860                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 861                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 862         }
 863         put_cpu_var(cpu_tsc_khz);
 864
 865         /* Keep irq disabled to prevent changes to the clock */
 866         local_irq_save(flags);
 867         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 868         ktime_get_ts(&ts);
 869         monotonic_to_bootbased(&ts);
 870         local_irq_restore(flags);
 871
 872         /* With all the info we got, fill in the values */
 873
 874         vcpu->hv_clock.system_time = ts.tv_nsec +
 875                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 876
 877         vcpu->hv_clock.flags = 0;
 878
 879         /*
 880          * The interface expects us to write an even number signaling that the
 881          * update is finished. Since the guest won't see the intermediate
 882          * state, we just increase by 2 at the end.
 883          */
 884         vcpu->hv_clock.version += 2;
 885
 886         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 887
 888         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 889                sizeof(vcpu->hv_clock));
 890
 891         kunmap_atomic(shared_kaddr, KM_USER0);
 892
 893         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 894 }
 895
 896 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 897 {
 898         struct kvm_vcpu_arch *vcpu = &v->arch;
 899
 900         if (!vcpu->time_page)
 901                 return 0;
 902         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 903         return 1;
 904 }
 905
 906 static bool msr_mtrr_valid(unsigned msr)
 907 {
 908         switch (msr) {
 909         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 910         case MSR_MTRRfix64K_00000:
 911         case MSR_MTRRfix16K_80000:
 912         case MSR_MTRRfix16K_A0000:
 913         case MSR_MTRRfix4K_C0000:
 914         case MSR_MTRRfix4K_C8000:
 915         case MSR_MTRRfix4K_D0000:
 916         case MSR_MTRRfix4K_D8000:
 917         case MSR_MTRRfix4K_E0000:
 918         case MSR_MTRRfix4K_E8000:
 919         case MSR_MTRRfix4K_F0000:
 920         case MSR_MTRRfix4K_F8000:
 921         case MSR_MTRRdefType:
 922         case MSR_IA32_CR_PAT:
 923                 return true;
 924         case 0x2f8:
 925                 return true;
 926         }
 927         return false;
 928 }
 929
 930 static bool valid_pat_type(unsigned t)
 931 {
 932         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 933 }
 934
 935 static bool valid_mtrr_type(unsigned t)
 936 {
 937         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 938 }
 939
 940 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 941 {
 942         int i;
 943
 944         if (!msr_mtrr_valid(msr))
 945                 return false;
 946
 947         if (msr == MSR_IA32_CR_PAT) {
 948                 for (i = 0; i < 8; i++)
 949                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 950                                 return false;
 951                 return true;
 952         } else if (msr == MSR_MTRRdefType) {
 953                 if (data & ~0xcff)
 954                         return false;
 955                 return valid_mtrr_type(data & 0xff);
 956         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 957                 for (i = 0; i < 8 ; i++)
 958                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 959                                 return false;
 960                 return true;
 961         }
 962
 963         /* variable MTRRs */
 964         return valid_mtrr_type(data & 0xff);
 965 }
 966
 967 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 968 {
 969         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 970
 971         if (!mtrr_valid(vcpu, msr, data))
 972                 return 1;
 973
 974         if (msr == MSR_MTRRdefType) {
 975                 vcpu->arch.mtrr_state.def_type = data;
 976                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 977         } else if (msr == MSR_MTRRfix64K_00000)
 978                 p[0] = data;
 979         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 980                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 981         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 982                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 983         else if (msr == MSR_IA32_CR_PAT)
 984                 vcpu->arch.pat = data;
 985         else {  /* Variable MTRRs */
 986                 int idx, is_mtrr_mask;
 987                 u64 *pt;
 988
 989                 idx = (msr - 0x200) / 2;
 990                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 991                 if (!is_mtrr_mask)
 992                         pt =
 993                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 994                 else
 995                         pt =
 996                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 997                 *pt = data;
 998         }
 999
1000         kvm_mmu_reset_context(vcpu);
1001         return 0;
1002 }
1003
1004 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1005 {
1006         u64 mcg_cap = vcpu->arch.mcg_cap;
1007         unsigned bank_num = mcg_cap & 0xff;
1008
1009         switch (msr) {
1010         case MSR_IA32_MCG_STATUS:
1011                 vcpu->arch.mcg_status = data;
1012                 break;
1013         case MSR_IA32_MCG_CTL:
1014                 if (!(mcg_cap & MCG_CTL_P))
1015                         return 1;
1016                 if (data != 0 && data != ~(u64)0)
1017                         return -1;
1018                 vcpu->arch.mcg_ctl = data;
1019                 break;
1020         default:
1021                 if (msr >= MSR_IA32_MC0_CTL &&
1022                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1023                         u32 offset = msr - MSR_IA32_MC0_CTL;
1024                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1025                          * some Linux kernels though clear bit 10 in bank 4 to
1026                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1027                          * this to avoid an uncatched #GP in the guest
1028                          */
1029                         if ((offset & 0x3) == 0 &&
1030                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1031                                 return -1;
1032                         vcpu->arch.mce_banks[offset] = data;
1033                         break;
1034                 }
1035                 return 1;
1036         }
1037         return 0;
1038 }
1039
1040 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1041 {
1042         struct kvm *kvm = vcpu->kvm;
1043         int lm = is_long_mode(vcpu);
1044         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1045                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1046         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1047                 : kvm->arch.xen_hvm_config.blob_size_32;
1048         u32 page_num = data & ~PAGE_MASK;
1049         u64 page_addr = data & PAGE_MASK;
1050         u8 *page;
1051         int r;
1052
1053         r = -E2BIG;
1054         if (page_num >= blob_size)
1055                 goto out;
1056         r = -ENOMEM;
1057         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1058         if (!page)
1059                 goto out;
1060         r = -EFAULT;
1061         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1062                 goto out_free;
1063         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1064                 goto out_free;
1065         r = 0;
1066 out_free:
1067         kfree(page);
1068 out:
1069         return r;
1070 }
1071
1072 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1073 {
1074         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1075 }
1076
1077 static bool kvm_hv_msr_partition_wide(u32 msr)
1078 {
1079         bool r = false;
1080         switch (msr) {
1081         case HV_X64_MSR_GUEST_OS_ID:
1082         case HV_X64_MSR_HYPERCALL:
1083                 r = true;
1084                 break;
1085         }
1086
1087         return r;
1088 }
1089
1090 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1091 {
1092         struct kvm *kvm = vcpu->kvm;
1093
1094         switch (msr) {
1095         case HV_X64_MSR_GUEST_OS_ID:
1096                 kvm->arch.hv_guest_os_id = data;
1097                 /* setting guest os id to zero disables hypercall page */
1098                 if (!kvm->arch.hv_guest_os_id)
1099                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1100                 break;
1101         case HV_X64_MSR_HYPERCALL: {
1102                 u64 gfn;
1103                 unsigned long addr;
1104                 u8 instructions[4];
1105
1106                 /* if guest os id is not set hypercall should remain disabled */
1107                 if (!kvm->arch.hv_guest_os_id)
1108                         break;
1109                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1110                         kvm->arch.hv_hypercall = data;
1111                         break;
1112                 }
1113                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1114                 addr = gfn_to_hva(kvm, gfn);
1115                 if (kvm_is_error_hva(addr))
1116                         return 1;
1117                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1118                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1119                 if (copy_to_user((void __user *)addr, instructions, 4))
1120                         return 1;
1121                 kvm->arch.hv_hypercall = data;
1122                 break;
1123         }
1124         default:
1125                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1126                           "data 0x%llx\n", msr, data);
1127                 return 1;
1128         }
1129         return 0;
1130 }
1131
1132 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1133 {
1134         switch (msr) {
1135         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1136                 unsigned long addr;
1137
1138                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1139                         vcpu->arch.hv_vapic = data;
1140                         break;
1141                 }
1142                 addr = gfn_to_hva(vcpu->kvm, data >>
1143                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1144                 if (kvm_is_error_hva(addr))
1145                         return 1;
1146                 if (clear_user((void __user *)addr, PAGE_SIZE))
1147                         return 1;
1148                 vcpu->arch.hv_vapic = data;
1149                 break;
1150         }
1151         case HV_X64_MSR_EOI:
1152                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1153         case HV_X64_MSR_ICR:
1154                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1155         case HV_X64_MSR_TPR:
1156                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1157         default:
1158                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1159                           "data 0x%llx\n", msr, data);
1160                 return 1;
1161         }
1162
1163         return 0;
1164 }
1165
1166 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1167 {
1168         switch (msr) {
1169         case MSR_EFER:
1170                 return set_efer(vcpu, data);
1171         case MSR_K7_HWCR:
1172                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1173                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1174                 if (data != 0) {
1175                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1176                                 data);
1177                         return 1;
1178                 }
1179                 break;
1180         case MSR_FAM10H_MMIO_CONF_BASE:
1181                 if (data != 0) {
1182                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1183                                 "0x%llx\n", data);
1184                         return 1;
1185                 }
1186                 break;
1187         case MSR_AMD64_NB_CFG:
1188                 break;
1189         case MSR_IA32_DEBUGCTLMSR:
1190                 if (!data) {
1191                         /* We support the non-activated case already */
1192                         break;
1193                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1194                         /* Values other than LBR and BTF are vendor-specific,
1195                            thus reserved and should throw a #GP */
1196                         return 1;
1197                 }
1198                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1199                         __func__, data);
1200                 break;
1201         case MSR_IA32_UCODE_REV:
1202         case MSR_IA32_UCODE_WRITE:
1203         case MSR_VM_HSAVE_PA:
1204         case MSR_AMD64_PATCH_LOADER:
1205                 break;
1206         case 0x200 ... 0x2ff:
1207                 return set_msr_mtrr(vcpu, msr, data);
1208         case MSR_IA32_APICBASE:
1209                 kvm_set_apic_base(vcpu, data);
1210                 break;
1211         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1212                 return kvm_x2apic_msr_write(vcpu, msr, data);
1213         case MSR_IA32_MISC_ENABLE:
1214                 vcpu->arch.ia32_misc_enable_msr = data;
1215                 break;
1216         case MSR_KVM_WALL_CLOCK_NEW:
1217         case MSR_KVM_WALL_CLOCK:
1218                 vcpu->kvm->arch.wall_clock = data;
1219                 kvm_write_wall_clock(vcpu->kvm, data);
1220                 break;
1221         case MSR_KVM_SYSTEM_TIME_NEW:
1222         case MSR_KVM_SYSTEM_TIME: {
1223                 if (vcpu->arch.time_page) {
1224                         kvm_release_page_dirty(vcpu->arch.time_page);
1225                         vcpu->arch.time_page = NULL;
1226                 }
1227
1228                 vcpu->arch.time = data;
1229
1230                 /* we verify if the enable bit is set... */
1231                 if (!(data & 1))
1232                         break;
1233
1234                 /* ...but clean it before doing the actual write */
1235                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1236
1237                 vcpu->arch.time_page =
1238                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1239
1240                 if (is_error_page(vcpu->arch.time_page)) {
1241                         kvm_release_page_clean(vcpu->arch.time_page);
1242                         vcpu->arch.time_page = NULL;
1243                 }
1244
1245                 kvm_request_guest_time_update(vcpu);
1246                 break;
1247         }
1248         case MSR_IA32_MCG_CTL:
1249         case MSR_IA32_MCG_STATUS:
1250         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1251                 return set_msr_mce(vcpu, msr, data);
1252
1253         /* Performance counters are not protected by a CPUID bit,
1254          * so we should check all of them in the generic path for the sake of
1255          * cross vendor migration.
1256          * Writing a zero into the event select MSRs disables them,
1257          * which we perfectly emulate ;-). Any other value should be at least
1258          * reported, some guests depend on them.
1259          */
1260         case MSR_P6_EVNTSEL0:
1261         case MSR_P6_EVNTSEL1:
1262         case MSR_K7_EVNTSEL0:
1263         case MSR_K7_EVNTSEL1:
1264         case MSR_K7_EVNTSEL2:
1265         case MSR_K7_EVNTSEL3:
1266                 if (data != 0)
1267                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1268                                 "0x%x data 0x%llx\n", msr, data);
1269                 break;
1270         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1271          * so we ignore writes to make it happy.
1272          */
1273         case MSR_P6_PERFCTR0:
1274         case MSR_P6_PERFCTR1:
1275         case MSR_K7_PERFCTR0:
1276         case MSR_K7_PERFCTR1:
1277         case MSR_K7_PERFCTR2:
1278         case MSR_K7_PERFCTR3:
1279                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1280                         "0x%x data 0x%llx\n", msr, data);
1281                 break;
1282         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1283                 if (kvm_hv_msr_partition_wide(msr)) {
1284                         int r;
1285                         mutex_lock(&vcpu->kvm->lock);
1286                         r = set_msr_hyperv_pw(vcpu, msr, data);
1287                         mutex_unlock(&vcpu->kvm->lock);
1288                         return r;
1289                 } else
1290                         return set_msr_hyperv(vcpu, msr, data);
1291                 break;
1292         default:
1293                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1294                         return xen_hvm_config(vcpu, data);
1295                 if (!ignore_msrs) {
1296                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1297                                 msr, data);
1298                         return 1;
1299                 } else {
1300                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1301                                 msr, data);
1302                         break;
1303                 }
1304         }
1305         return 0;
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1308
1309
1310 /*
1311  * Reads an msr value (of 'msr_index') into 'pdata'.
1312  * Returns 0 on success, non-0 otherwise.
1313  * Assumes vcpu_load() was already called.
1314  */
1315 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1316 {
1317         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1318 }
1319
1320 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1321 {
1322         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1323
1324         if (!msr_mtrr_valid(msr))
1325                 return 1;
1326
1327         if (msr == MSR_MTRRdefType)
1328                 *pdata = vcpu->arch.mtrr_state.def_type +
1329                          (vcpu->arch.mtrr_state.enabled << 10);
1330         else if (msr == MSR_MTRRfix64K_00000)
1331                 *pdata = p[0];
1332         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1333                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1334         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1335                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1336         else if (msr == MSR_IA32_CR_PAT)
1337                 *pdata = vcpu->arch.pat;
1338         else {  /* Variable MTRRs */
1339                 int idx, is_mtrr_mask;
1340                 u64 *pt;
1341
1342                 idx = (msr - 0x200) / 2;
1343                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1344                 if (!is_mtrr_mask)
1345                         pt =
1346                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1347                 else
1348                         pt =
1349                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1350                 *pdata = *pt;
1351         }
1352
1353         return 0;
1354 }
1355
1356 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1357 {
1358         u64 data;
1359         u64 mcg_cap = vcpu->arch.mcg_cap;
1360         unsigned bank_num = mcg_cap & 0xff;
1361
1362         switch (msr) {
1363         case MSR_IA32_P5_MC_ADDR:
1364         case MSR_IA32_P5_MC_TYPE:
1365                 data = 0;
1366                 break;
1367         case MSR_IA32_MCG_CAP:
1368                 data = vcpu->arch.mcg_cap;
1369                 break;
1370         case MSR_IA32_MCG_CTL:
1371                 if (!(mcg_cap & MCG_CTL_P))
1372                         return 1;
1373                 data = vcpu->arch.mcg_ctl;
1374                 break;
1375         case MSR_IA32_MCG_STATUS:
1376                 data = vcpu->arch.mcg_status;
1377                 break;
1378         default:
1379                 if (msr >= MSR_IA32_MC0_CTL &&
1380                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1381                         u32 offset = msr - MSR_IA32_MC0_CTL;
1382                         data = vcpu->arch.mce_banks[offset];
1383                         break;
1384                 }
1385                 return 1;
1386         }
1387         *pdata = data;
1388         return 0;
1389 }
1390
1391 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1392 {
1393         u64 data = 0;
1394         struct kvm *kvm = vcpu->kvm;
1395
1396         switch (msr) {
1397         case HV_X64_MSR_GUEST_OS_ID:
1398                 data = kvm->arch.hv_guest_os_id;
1399                 break;
1400         case HV_X64_MSR_HYPERCALL:
1401                 data = kvm->arch.hv_hypercall;
1402                 break;
1403         default:
1404                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1405                 return 1;
1406         }
1407
1408         *pdata = data;
1409         return 0;
1410 }
1411
1412 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1413 {
1414         u64 data = 0;
1415
1416         switch (msr) {
1417         case HV_X64_MSR_VP_INDEX: {
1418                 int r;
1419                 struct kvm_vcpu *v;
1420                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1421                         if (v == vcpu)
1422                                 data = r;
1423                 break;
1424         }
1425         case HV_X64_MSR_EOI:
1426                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1427         case HV_X64_MSR_ICR:
1428                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1429         case HV_X64_MSR_TPR:
1430                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1431         default:
1432                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1433                 return 1;
1434         }
1435         *pdata = data;
1436         return 0;
1437 }
1438
1439 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1440 {
1441         u64 data;
1442
1443         switch (msr) {
1444         case MSR_IA32_PLATFORM_ID:
1445         case MSR_IA32_UCODE_REV:
1446         case MSR_IA32_EBL_CR_POWERON:
1447         case MSR_IA32_DEBUGCTLMSR:
1448         case MSR_IA32_LASTBRANCHFROMIP:
1449         case MSR_IA32_LASTBRANCHTOIP:
1450         case MSR_IA32_LASTINTFROMIP:
1451         case MSR_IA32_LASTINTTOIP:
1452         case MSR_K8_SYSCFG:
1453         case MSR_K7_HWCR:
1454         case MSR_VM_HSAVE_PA:
1455         case MSR_P6_PERFCTR0:
1456         case MSR_P6_PERFCTR1:
1457         case MSR_P6_EVNTSEL0:
1458         case MSR_P6_EVNTSEL1:
1459         case MSR_K7_EVNTSEL0:
1460         case MSR_K7_PERFCTR0:
1461         case MSR_K8_INT_PENDING_MSG:
1462         case MSR_AMD64_NB_CFG:
1463         case MSR_FAM10H_MMIO_CONF_BASE:
1464                 data = 0;
1465                 break;
1466         case MSR_MTRRcap:
1467                 data = 0x500 | KVM_NR_VAR_MTRR;
1468                 break;
1469         case 0x200 ... 0x2ff:
1470                 return get_msr_mtrr(vcpu, msr, pdata);
1471         case 0xcd: /* fsb frequency */
1472                 data = 3;
1473                 break;
1474         case MSR_IA32_APICBASE:
1475                 data = kvm_get_apic_base(vcpu);
1476                 break;
1477         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1478                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1479                 break;
1480         case MSR_IA32_MISC_ENABLE:
1481                 data = vcpu->arch.ia32_misc_enable_msr;
1482                 break;
1483         case MSR_IA32_PERF_STATUS:
1484                 /* TSC increment by tick */
1485                 data = 1000ULL;
1486                 /* CPU multiplier */
1487                 data |= (((uint64_t)4ULL) << 40);
1488                 break;
1489         case MSR_EFER:
1490                 data = vcpu->arch.efer;
1491                 break;
1492         case MSR_KVM_WALL_CLOCK:
1493         case MSR_KVM_WALL_CLOCK_NEW:
1494                 data = vcpu->kvm->arch.wall_clock;
1495                 break;
1496         case MSR_KVM_SYSTEM_TIME:
1497         case MSR_KVM_SYSTEM_TIME_NEW:
1498                 data = vcpu->arch.time;
1499                 break;
1500         case MSR_IA32_P5_MC_ADDR:
1501         case MSR_IA32_P5_MC_TYPE:
1502         case MSR_IA32_MCG_CAP:
1503         case MSR_IA32_MCG_CTL:
1504         case MSR_IA32_MCG_STATUS:
1505         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1506                 return get_msr_mce(vcpu, msr, pdata);
1507         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1508                 if (kvm_hv_msr_partition_wide(msr)) {
1509                         int r;
1510                         mutex_lock(&vcpu->kvm->lock);
1511                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1512                         mutex_unlock(&vcpu->kvm->lock);
1513                         return r;
1514                 } else
1515                         return get_msr_hyperv(vcpu, msr, pdata);
1516                 break;
1517         default:
1518                 if (!ignore_msrs) {
1519                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1520                         return 1;
1521                 } else {
1522                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1523                         data = 0;
1524                 }
1525                 break;
1526         }
1527         *pdata = data;
1528         return 0;
1529 }
1530 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1531
1532 /*
1533  * Read or write a bunch of msrs. All parameters are kernel addresses.
1534  *
1535  * @return number of msrs set successfully.
1536  */
1537 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1538                     struct kvm_msr_entry *entries,
1539                     int (*do_msr)(struct kvm_vcpu *vcpu,
1540                                   unsigned index, u64 *data))
1541 {
1542         int i, idx;
1543
1544         idx = srcu_read_lock(&vcpu->kvm->srcu);
1545         for (i = 0; i < msrs->nmsrs; ++i)
1546                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1547                         break;
1548         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1549
1550         return i;
1551 }
1552
1553 /*
1554  * Read or write a bunch of msrs. Parameters are user addresses.
1555  *
1556  * @return number of msrs set successfully.
1557  */
1558 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1559                   int (*do_msr)(struct kvm_vcpu *vcpu,
1560                                 unsigned index, u64 *data),
1561                   int writeback)
1562 {
1563         struct kvm_msrs msrs;
1564         struct kvm_msr_entry *entries;
1565         int r, n;
1566         unsigned size;
1567
1568         r = -EFAULT;
1569         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1570                 goto out;
1571
1572         r = -E2BIG;
1573         if (msrs.nmsrs >= MAX_IO_MSRS)
1574                 goto out;
1575
1576         r = -ENOMEM;
1577         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1578         entries = kmalloc(size, GFP_KERNEL);
1579         if (!entries)
1580                 goto out;
1581
1582         r = -EFAULT;
1583         if (copy_from_user(entries, user_msrs->entries, size))
1584                 goto out_free;
1585
1586         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1587         if (r < 0)
1588                 goto out_free;
1589
1590         r = -EFAULT;
1591         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1592                 goto out_free;
1593
1594         r = n;
1595
1596 out_free:
1597         kfree(entries);
1598 out:
1599         return r;
1600 }
1601
1602 int kvm_dev_ioctl_check_extension(long ext)
1603 {
1604         int r;
1605
1606         switch (ext) {
1607         case KVM_CAP_IRQCHIP:
1608         case KVM_CAP_HLT:
1609         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1610         case KVM_CAP_SET_TSS_ADDR:
1611         case KVM_CAP_EXT_CPUID:
1612         case KVM_CAP_CLOCKSOURCE:
1613         case KVM_CAP_PIT:
1614         case KVM_CAP_NOP_IO_DELAY:
1615         case KVM_CAP_MP_STATE:
1616         case KVM_CAP_SYNC_MMU:
1617         case KVM_CAP_REINJECT_CONTROL:
1618         case KVM_CAP_IRQ_INJECT_STATUS:
1619         case KVM_CAP_ASSIGN_DEV_IRQ:
1620         case KVM_CAP_IRQFD:
1621         case KVM_CAP_IOEVENTFD:
1622         case KVM_CAP_PIT2:
1623         case KVM_CAP_PIT_STATE2:
1624         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1625         case KVM_CAP_XEN_HVM:
1626         case KVM_CAP_ADJUST_CLOCK:
1627         case KVM_CAP_VCPU_EVENTS:
1628         case KVM_CAP_HYPERV:
1629         case KVM_CAP_HYPERV_VAPIC:
1630         case KVM_CAP_HYPERV_SPIN:
1631         case KVM_CAP_PCI_SEGMENT:
1632         case KVM_CAP_DEBUGREGS:
1633         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1634                 r = 1;
1635                 break;
1636         case KVM_CAP_COALESCED_MMIO:
1637                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1638                 break;
1639         case KVM_CAP_VAPIC:
1640                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1641                 break;
1642         case KVM_CAP_NR_VCPUS:
1643                 r = KVM_MAX_VCPUS;
1644                 break;
1645         case KVM_CAP_NR_MEMSLOTS:
1646                 r = KVM_MEMORY_SLOTS;
1647                 break;
1648         case KVM_CAP_PV_MMU:    /* obsolete */
1649                 r = 0;
1650                 break;
1651         case KVM_CAP_IOMMU:
1652                 r = iommu_found();
1653                 break;
1654         case KVM_CAP_MCE:
1655                 r = KVM_MAX_MCE_BANKS;
1656                 break;
1657         default:
1658                 r = 0;
1659                 break;
1660         }
1661         return r;
1662
1663 }
1664
1665 long kvm_arch_dev_ioctl(struct file *filp,
1666                         unsigned int ioctl, unsigned long arg)
1667 {
1668         void __user *argp = (void __user *)arg;
1669         long r;
1670
1671         switch (ioctl) {
1672         case KVM_GET_MSR_INDEX_LIST: {
1673                 struct kvm_msr_list __user *user_msr_list = argp;
1674                 struct kvm_msr_list msr_list;
1675                 unsigned n;
1676
1677                 r = -EFAULT;
1678                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1679                         goto out;
1680                 n = msr_list.nmsrs;
1681                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1682                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1683                         goto out;
1684                 r = -E2BIG;
1685                 if (n < msr_list.nmsrs)
1686                         goto out;
1687                 r = -EFAULT;
1688                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1689                                  num_msrs_to_save * sizeof(u32)))
1690                         goto out;
1691                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1692                                  &emulated_msrs,
1693                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1694                         goto out;
1695                 r = 0;
1696                 break;
1697         }
1698         case KVM_GET_SUPPORTED_CPUID: {
1699                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1700                 struct kvm_cpuid2 cpuid;
1701
1702                 r = -EFAULT;
1703                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1704                         goto out;
1705                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1706                                                       cpuid_arg->entries);
1707                 if (r)
1708                         goto out;
1709
1710                 r = -EFAULT;
1711                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1712                         goto out;
1713                 r = 0;
1714                 break;
1715         }
1716         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1717                 u64 mce_cap;
1718
1719                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1720                 r = -EFAULT;
1721                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1722                         goto out;
1723                 r = 0;
1724                 break;
1725         }
1726         default:
1727                 r = -EINVAL;
1728         }
1729 out:
1730         return r;
1731 }
1732
1733 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1734 {
1735         kvm_x86_ops->vcpu_load(vcpu, cpu);
1736         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1737                 unsigned long khz = cpufreq_quick_get(cpu);
1738                 if (!khz)
1739                         khz = tsc_khz;
1740                 per_cpu(cpu_tsc_khz, cpu) = khz;
1741         }
1742         kvm_request_guest_time_update(vcpu);
1743 }
1744
1745 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1746 {
1747         kvm_x86_ops->vcpu_put(vcpu);
1748         kvm_put_guest_fpu(vcpu);
1749 }
1750
1751 static int is_efer_nx(void)
1752 {
1753         unsigned long long efer = 0;
1754
1755         rdmsrl_safe(MSR_EFER, &efer);
1756         return efer & EFER_NX;
1757 }
1758
1759 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1760 {
1761         int i;
1762         struct kvm_cpuid_entry2 *e, *entry;
1763
1764         entry = NULL;
1765         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1766                 e = &vcpu->arch.cpuid_entries[i];
1767                 if (e->function == 0x80000001) {
1768                         entry = e;
1769                         break;
1770                 }
1771         }
1772         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1773                 entry->edx &= ~(1 << 20);
1774                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1775         }
1776 }
1777
1778 /* when an old userspace process fills a new kernel module */
1779 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1780                                     struct kvm_cpuid *cpuid,
1781                                     struct kvm_cpuid_entry __user *entries)
1782 {
1783         int r, i;
1784         struct kvm_cpuid_entry *cpuid_entries;
1785
1786         r = -E2BIG;
1787         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1788                 goto out;
1789         r = -ENOMEM;
1790         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1791         if (!cpuid_entries)
1792                 goto out;
1793         r = -EFAULT;
1794         if (copy_from_user(cpuid_entries, entries,
1795                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1796                 goto out_free;
1797         for (i = 0; i < cpuid->nent; i++) {
1798                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1799                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1800                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1801                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1802                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1803                 vcpu->arch.cpuid_entries[i].index = 0;
1804                 vcpu->arch.cpuid_entries[i].flags = 0;
1805                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1806                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1807                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1808         }
1809         vcpu->arch.cpuid_nent = cpuid->nent;
1810         cpuid_fix_nx_cap(vcpu);
1811         r = 0;
1812         kvm_apic_set_version(vcpu);
1813         kvm_x86_ops->cpuid_update(vcpu);
1814
1815 out_free:
1816         vfree(cpuid_entries);
1817 out:
1818         return r;
1819 }
1820
1821 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1822                                      struct kvm_cpuid2 *cpuid,
1823                                      struct kvm_cpuid_entry2 __user *entries)
1824 {
1825         int r;
1826
1827         r = -E2BIG;
1828         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1829                 goto out;
1830         r = -EFAULT;
1831         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1832                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1833                 goto out;
1834         vcpu->arch.cpuid_nent = cpuid->nent;
1835         kvm_apic_set_version(vcpu);
1836         kvm_x86_ops->cpuid_update(vcpu);
1837         return 0;
1838
1839 out:
1840         return r;
1841 }
1842
1843 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1844                                      struct kvm_cpuid2 *cpuid,
1845                                      struct kvm_cpuid_entry2 __user *entries)
1846 {
1847         int r;
1848
1849         r = -E2BIG;
1850         if (cpuid->nent < vcpu->arch.cpuid_nent)
1851                 goto out;
1852         r = -EFAULT;
1853         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1854                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1855                 goto out;
1856         return 0;
1857
1858 out:
1859         cpuid->nent = vcpu->arch.cpuid_nent;
1860         return r;
1861 }
1862
1863 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1864                            u32 index)
1865 {
1866         entry->function = function;
1867         entry->index = index;
1868         cpuid_count(entry->function, entry->index,
1869                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1870         entry->flags = 0;
1871 }
1872
1873 #define F(x) bit(X86_FEATURE_##x)
1874
1875 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1876                          u32 index, int *nent, int maxnent)
1877 {
1878         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1879 #ifdef CONFIG_X86_64
1880         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1881                                 ? F(GBPAGES) : 0;
1882         unsigned f_lm = F(LM);
1883 #else
1884         unsigned f_gbpages = 0;
1885         unsigned f_lm = 0;
1886 #endif
1887         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1888
1889         /* cpuid 1.edx */
1890         const u32 kvm_supported_word0_x86_features =
1891                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1892                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1893                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1894                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1895                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1896                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1897                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1898                 0 /* HTT, TM, Reserved, PBE */;
1899         /* cpuid 0x80000001.edx */
1900         const u32 kvm_supported_word1_x86_features =
1901                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1902                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1903                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1904                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1905                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1906                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1907                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1908                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1909         /* cpuid 1.ecx */
1910         const u32 kvm_supported_word4_x86_features =
1911                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1912                 0 /* DS-CPL, VMX, SMX, EST */ |
1913                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1914                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1915                 0 /* Reserved, DCA */ | F(XMM4_1) |
1916                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1917                 0 /* Reserved, XSAVE, OSXSAVE */;
1918         /* cpuid 0x80000001.ecx */
1919         const u32 kvm_supported_word6_x86_features =
1920                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1921                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1922                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1923                 0 /* SKINIT */ | 0 /* WDT */;
1924
1925         /* all calls to cpuid_count() should be made on the same cpu */
1926         get_cpu();
1927         do_cpuid_1_ent(entry, function, index);
1928         ++*nent;
1929
1930         switch (function) {
1931         case 0:
1932                 entry->eax = min(entry->eax, (u32)0xb);
1933                 break;
1934         case 1:
1935                 entry->edx &= kvm_supported_word0_x86_features;
1936                 entry->ecx &= kvm_supported_word4_x86_features;
1937                 /* we support x2apic emulation even if host does not support
1938                  * it since we emulate x2apic in software */
1939                 entry->ecx |= F(X2APIC);
1940                 break;
1941         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1942          * may return different values. This forces us to get_cpu() before
1943          * issuing the first command, and also to emulate this annoying behavior
1944          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1945         case 2: {
1946                 int t, times = entry->eax & 0xff;
1947
1948                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1949                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1950                 for (t = 1; t < times && *nent < maxnent; ++t) {
1951                         do_cpuid_1_ent(&entry[t], function, 0);
1952                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1953                         ++*nent;
1954                 }
1955                 break;
1956         }
1957         /* function 4 and 0xb have additional index. */
1958         case 4: {
1959                 int i, cache_type;
1960
1961                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1962                 /* read more entries until cache_type is zero */
1963                 for (i = 1; *nent < maxnent; ++i) {
1964                         cache_type = entry[i - 1].eax & 0x1f;
1965                         if (!cache_type)
1966                                 break;
1967                         do_cpuid_1_ent(&entry[i], function, i);
1968                         entry[i].flags |=
1969                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1970                         ++*nent;
1971                 }
1972                 break;
1973         }
1974         case 0xb: {
1975                 int i, level_type;
1976
1977                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1978                 /* read more entries until level_type is zero */
1979                 for (i = 1; *nent < maxnent; ++i) {
1980                         level_type = entry[i - 1].ecx & 0xff00;
1981                         if (!level_type)
1982                                 break;
1983                         do_cpuid_1_ent(&entry[i], function, i);
1984                         entry[i].flags |=
1985                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1986                         ++*nent;
1987                 }
1988                 break;
1989         }
1990         case KVM_CPUID_SIGNATURE: {
1991                 char signature[12] = "KVMKVMKVM\0\0";
1992                 u32 *sigptr = (u32 *)signature;
1993                 entry->eax = 0;
1994                 entry->ebx = sigptr[0];
1995                 entry->ecx = sigptr[1];
1996                 entry->edx = sigptr[2];
1997                 break;
1998         }
1999         case KVM_CPUID_FEATURES:
2000                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2001                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2002                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2003                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2004                 entry->ebx = 0;
2005                 entry->ecx = 0;
2006                 entry->edx = 0;
2007                 break;
2008         case 0x80000000:
2009                 entry->eax = min(entry->eax, 0x8000001a);
2010                 break;
2011         case 0x80000001:
2012                 entry->edx &= kvm_supported_word1_x86_features;
2013                 entry->ecx &= kvm_supported_word6_x86_features;
2014                 break;
2015         }
2016
2017         kvm_x86_ops->set_supported_cpuid(function, entry);
2018
2019         put_cpu();
2020 }
2021
2022 #undef F
2023
2024 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2025                                      struct kvm_cpuid_entry2 __user *entries)
2026 {
2027         struct kvm_cpuid_entry2 *cpuid_entries;
2028         int limit, nent = 0, r = -E2BIG;
2029         u32 func;
2030
2031         if (cpuid->nent < 1)
2032                 goto out;
2033         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2034                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2035         r = -ENOMEM;
2036         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2037         if (!cpuid_entries)
2038                 goto out;
2039
2040         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2041         limit = cpuid_entries[0].eax;
2042         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2043                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2044                              &nent, cpuid->nent);
2045         r = -E2BIG;
2046         if (nent >= cpuid->nent)
2047                 goto out_free;
2048
2049         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2050         limit = cpuid_entries[nent - 1].eax;
2051         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2052                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2053                              &nent, cpuid->nent);
2054
2055
2056
2057         r = -E2BIG;
2058         if (nent >= cpuid->nent)
2059                 goto out_free;
2060
2061         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2062                      cpuid->nent);
2063
2064         r = -E2BIG;
2065         if (nent >= cpuid->nent)
2066                 goto out_free;
2067
2068         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2069                      cpuid->nent);
2070
2071         r = -E2BIG;
2072         if (nent >= cpuid->nent)
2073                 goto out_free;
2074
2075         r = -EFAULT;
2076         if (copy_to_user(entries, cpuid_entries,
2077                          nent * sizeof(struct kvm_cpuid_entry2)))
2078                 goto out_free;
2079         cpuid->nent = nent;
2080         r = 0;
2081
2082 out_free:
2083         vfree(cpuid_entries);
2084 out:
2085         return r;
2086 }
2087
2088 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2089                                     struct kvm_lapic_state *s)
2090 {
2091         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2092
2093         return 0;
2094 }
2095
2096 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2097                                     struct kvm_lapic_state *s)
2098 {
2099         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2100         kvm_apic_post_state_restore(vcpu);
2101         update_cr8_intercept(vcpu);
2102
2103         return 0;
2104 }
2105
2106 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2107                                     struct kvm_interrupt *irq)
2108 {
2109         if (irq->irq < 0 || irq->irq >= 256)
2110                 return -EINVAL;
2111         if (irqchip_in_kernel(vcpu->kvm))
2112                 return -ENXIO;
2113
2114         kvm_queue_interrupt(vcpu, irq->irq, false);
2115
2116         return 0;
2117 }
2118
2119 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2120 {
2121         kvm_inject_nmi(vcpu);
2122
2123         return 0;
2124 }
2125
2126 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2127                                            struct kvm_tpr_access_ctl *tac)
2128 {
2129         if (tac->flags)
2130                 return -EINVAL;
2131         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2132         return 0;
2133 }
2134
2135 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2136                                         u64 mcg_cap)
2137 {
2138         int r;
2139         unsigned bank_num = mcg_cap & 0xff, bank;
2140
2141         r = -EINVAL;
2142         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2143                 goto out;
2144         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2145                 goto out;
2146         r = 0;
2147         vcpu->arch.mcg_cap = mcg_cap;
2148         /* Init IA32_MCG_CTL to all 1s */
2149         if (mcg_cap & MCG_CTL_P)
2150                 vcpu->arch.mcg_ctl = ~(u64)0;
2151         /* Init IA32_MCi_CTL to all 1s */
2152         for (bank = 0; bank < bank_num; bank++)
2153                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2154 out:
2155         return r;
2156 }
2157
2158 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2159                                       struct kvm_x86_mce *mce)
2160 {
2161         u64 mcg_cap = vcpu->arch.mcg_cap;
2162         unsigned bank_num = mcg_cap & 0xff;
2163         u64 *banks = vcpu->arch.mce_banks;
2164
2165         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2166                 return -EINVAL;
2167         /*
2168          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2169          * reporting is disabled
2170          */
2171         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2172             vcpu->arch.mcg_ctl != ~(u64)0)
2173                 return 0;
2174         banks += 4 * mce->bank;
2175         /*
2176          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2177          * reporting is disabled for the bank
2178          */
2179         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2180                 return 0;
2181         if (mce->status & MCI_STATUS_UC) {
2182                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2183                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2184                         printk(KERN_DEBUG "kvm: set_mce: "
2185                                "injects mce exception while "
2186                                "previous one is in progress!\n");
2187                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2188                         return 0;
2189                 }
2190                 if (banks[1] & MCI_STATUS_VAL)
2191                         mce->status |= MCI_STATUS_OVER;
2192                 banks[2] = mce->addr;
2193                 banks[3] = mce->misc;
2194                 vcpu->arch.mcg_status = mce->mcg_status;
2195                 banks[1] = mce->status;
2196                 kvm_queue_exception(vcpu, MC_VECTOR);
2197         } else if (!(banks[1] & MCI_STATUS_VAL)
2198                    || !(banks[1] & MCI_STATUS_UC)) {
2199                 if (banks[1] & MCI_STATUS_VAL)
2200                         mce->status |= MCI_STATUS_OVER;
2201                 banks[2] = mce->addr;
2202                 banks[3] = mce->misc;
2203                 banks[1] = mce->status;
2204         } else
2205                 banks[1] |= MCI_STATUS_OVER;
2206         return 0;
2207 }
2208
2209 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2210                                                struct kvm_vcpu_events *events)
2211 {
2212         events->exception.injected =
2213                 vcpu->arch.exception.pending &&
2214                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2215         events->exception.nr = vcpu->arch.exception.nr;
2216         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2217         events->exception.error_code = vcpu->arch.exception.error_code;
2218
2219         events->interrupt.injected =
2220                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2221         events->interrupt.nr = vcpu->arch.interrupt.nr;
2222         events->interrupt.soft = 0;
2223         events->interrupt.shadow =
2224                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2225                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2226
2227         events->nmi.injected = vcpu->arch.nmi_injected;
2228         events->nmi.pending = vcpu->arch.nmi_pending;
2229         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2230
2231         events->sipi_vector = vcpu->arch.sipi_vector;
2232
2233         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2234                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2235                          | KVM_VCPUEVENT_VALID_SHADOW);
2236 }
2237
2238 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2239                                               struct kvm_vcpu_events *events)
2240 {
2241         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2242                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2243                               | KVM_VCPUEVENT_VALID_SHADOW))
2244                 return -EINVAL;
2245
2246         vcpu->arch.exception.pending = events->exception.injected;
2247         vcpu->arch.exception.nr = events->exception.nr;
2248         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2249         vcpu->arch.exception.error_code = events->exception.error_code;
2250
2251         vcpu->arch.interrupt.pending = events->interrupt.injected;
2252         vcpu->arch.interrupt.nr = events->interrupt.nr;
2253         vcpu->arch.interrupt.soft = events->interrupt.soft;
2254         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2255                 kvm_pic_clear_isr_ack(vcpu->kvm);
2256         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2257                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2258                                                   events->interrupt.shadow);
2259
2260         vcpu->arch.nmi_injected = events->nmi.injected;
2261         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2262                 vcpu->arch.nmi_pending = events->nmi.pending;
2263         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2264
2265         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2266                 vcpu->arch.sipi_vector = events->sipi_vector;
2267
2268         return 0;
2269 }
2270
2271 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2272                                              struct kvm_debugregs *dbgregs)
2273 {
2274         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2275         dbgregs->dr6 = vcpu->arch.dr6;
2276         dbgregs->dr7 = vcpu->arch.dr7;
2277         dbgregs->flags = 0;
2278 }
2279
2280 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2281                                             struct kvm_debugregs *dbgregs)
2282 {
2283         if (dbgregs->flags)
2284                 return -EINVAL;
2285
2286         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2287         vcpu->arch.dr6 = dbgregs->dr6;
2288         vcpu->arch.dr7 = dbgregs->dr7;
2289
2290         return 0;
2291 }
2292
2293 long kvm_arch_vcpu_ioctl(struct file *filp,
2294                          unsigned int ioctl, unsigned long arg)
2295 {
2296         struct kvm_vcpu *vcpu = filp->private_data;
2297         void __user *argp = (void __user *)arg;
2298         int r;
2299         struct kvm_lapic_state *lapic = NULL;
2300
2301         vcpu_load(vcpu);
2302         switch (ioctl) {
2303         case KVM_GET_LAPIC: {
2304                 r = -EINVAL;
2305                 if (!vcpu->arch.apic)
2306                         goto out;
2307                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2308
2309                 r = -ENOMEM;
2310                 if (!lapic)
2311                         goto out;
2312                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2313                 if (r)
2314                         goto out;
2315                 r = -EFAULT;
2316                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2317                         goto out;
2318                 r = 0;
2319                 break;
2320         }
2321         case KVM_SET_LAPIC: {
2322                 r = -EINVAL;
2323                 if (!vcpu->arch.apic)
2324                         goto out;
2325                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2326                 r = -ENOMEM;
2327                 if (!lapic)
2328                         goto out;
2329                 r = -EFAULT;
2330                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2331                         goto out;
2332                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2333                 if (r)
2334                         goto out;
2335                 r = 0;
2336                 break;
2337         }
2338         case KVM_INTERRUPT: {
2339                 struct kvm_interrupt irq;
2340
2341                 r = -EFAULT;
2342                 if (copy_from_user(&irq, argp, sizeof irq))
2343                         goto out;
2344                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2345                 if (r)
2346                         goto out;
2347                 r = 0;
2348                 break;
2349         }
2350         case KVM_NMI: {
2351                 r = kvm_vcpu_ioctl_nmi(vcpu);
2352                 if (r)
2353                         goto out;
2354                 r = 0;
2355                 break;
2356         }
2357         case KVM_SET_CPUID: {
2358                 struct kvm_cpuid __user *cpuid_arg = argp;
2359                 struct kvm_cpuid cpuid;
2360
2361                 r = -EFAULT;
2362                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2363                         goto out;
2364                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2365                 if (r)
2366                         goto out;
2367                 break;
2368         }
2369         case KVM_SET_CPUID2: {
2370                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2371                 struct kvm_cpuid2 cpuid;
2372
2373                 r = -EFAULT;
2374                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2375                         goto out;
2376                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2377                                               cpuid_arg->entries);
2378                 if (r)
2379                         goto out;
2380                 break;
2381         }
2382         case KVM_GET_CPUID2: {
2383                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2384                 struct kvm_cpuid2 cpuid;
2385
2386                 r = -EFAULT;
2387                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2388                         goto out;
2389                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2390                                               cpuid_arg->entries);
2391                 if (r)
2392                         goto out;
2393                 r = -EFAULT;
2394                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2395                         goto out;
2396                 r = 0;
2397                 break;
2398         }
2399         case KVM_GET_MSRS:
2400                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2401                 break;
2402         case KVM_SET_MSRS:
2403                 r = msr_io(vcpu, argp, do_set_msr, 0);
2404                 break;
2405         case KVM_TPR_ACCESS_REPORTING: {
2406                 struct kvm_tpr_access_ctl tac;
2407
2408                 r = -EFAULT;
2409                 if (copy_from_user(&tac, argp, sizeof tac))
2410                         goto out;
2411                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2412                 if (r)
2413                         goto out;
2414                 r = -EFAULT;
2415                 if (copy_to_user(argp, &tac, sizeof tac))
2416                         goto out;
2417                 r = 0;
2418                 break;
2419         };
2420         case KVM_SET_VAPIC_ADDR: {
2421                 struct kvm_vapic_addr va;
2422
2423                 r = -EINVAL;
2424                 if (!irqchip_in_kernel(vcpu->kvm))
2425                         goto out;
2426                 r = -EFAULT;
2427                 if (copy_from_user(&va, argp, sizeof va))
2428                         goto out;
2429                 r = 0;
2430                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2431                 break;
2432         }
2433         case KVM_X86_SETUP_MCE: {
2434                 u64 mcg_cap;
2435
2436                 r = -EFAULT;
2437                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2438                         goto out;
2439                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2440                 break;
2441         }
2442         case KVM_X86_SET_MCE: {
2443                 struct kvm_x86_mce mce;
2444
2445                 r = -EFAULT;
2446                 if (copy_from_user(&mce, argp, sizeof mce))
2447                         goto out;
2448                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2449                 break;
2450         }
2451         case KVM_GET_VCPU_EVENTS: {
2452                 struct kvm_vcpu_events events;
2453
2454                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2455
2456                 r = -EFAULT;
2457                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2458                         break;
2459                 r = 0;
2460                 break;
2461         }
2462         case KVM_SET_VCPU_EVENTS: {
2463                 struct kvm_vcpu_events events;
2464
2465                 r = -EFAULT;
2466                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2467                         break;
2468
2469                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2470                 break;
2471         }
2472         case KVM_GET_DEBUGREGS: {
2473                 struct kvm_debugregs dbgregs;
2474
2475                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2476
2477                 r = -EFAULT;
2478                 if (copy_to_user(argp, &dbgregs,
2479                                  sizeof(struct kvm_debugregs)))
2480                         break;
2481                 r = 0;
2482                 break;
2483         }
2484         case KVM_SET_DEBUGREGS: {
2485                 struct kvm_debugregs dbgregs;
2486
2487                 r = -EFAULT;
2488                 if (copy_from_user(&dbgregs, argp,
2489                                    sizeof(struct kvm_debugregs)))
2490                         break;
2491
2492                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2493                 break;
2494         }
2495         default:
2496                 r = -EINVAL;
2497         }
2498 out:
2499         vcpu_put(vcpu);
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  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
5063  * we have asm/x86/processor.h
5064  */
5065 struct fxsave {
5066         u16     cwd;
5067         u16     swd;
5068         u16     twd;
5069         u16     fop;
5070         u64     rip;
5071         u64     rdp;
5072         u32     mxcsr;
5073         u32     mxcsr_mask;
5074         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
5075 #ifdef CONFIG_X86_64
5076         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
5077 #else
5078         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
5079 #endif
5080 };
5081
5082 /*
5083  * Translate a guest virtual address to a guest physical address.
5084  */
5085 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5086                                     struct kvm_translation *tr)
5087 {
5088         unsigned long vaddr = tr->linear_address;
5089         gpa_t gpa;
5090         int idx;
5091
5092         idx = srcu_read_lock(&vcpu->kvm->srcu);
5093         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5094         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5095         tr->physical_address = gpa;
5096         tr->valid = gpa != UNMAPPED_GVA;
5097         tr->writeable = 1;
5098         tr->usermode = 0;
5099
5100         return 0;
5101 }
5102
5103 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5104 {
5105         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5106
5107         memcpy(fpu->fpr, fxsave->st_space, 128);
5108         fpu->fcw = fxsave->cwd;
5109         fpu->fsw = fxsave->swd;
5110         fpu->ftwx = fxsave->twd;
5111         fpu->last_opcode = fxsave->fop;
5112         fpu->last_ip = fxsave->rip;
5113         fpu->last_dp = fxsave->rdp;
5114         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5115
5116         return 0;
5117 }
5118
5119 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5120 {
5121         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5122
5123         memcpy(fxsave->st_space, fpu->fpr, 128);
5124         fxsave->cwd = fpu->fcw;
5125         fxsave->swd = fpu->fsw;
5126         fxsave->twd = fpu->ftwx;
5127         fxsave->fop = fpu->last_opcode;
5128         fxsave->rip = fpu->last_ip;
5129         fxsave->rdp = fpu->last_dp;
5130         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5131
5132         return 0;
5133 }
5134
5135 void fx_init(struct kvm_vcpu *vcpu)
5136 {
5137         unsigned after_mxcsr_mask;
5138
5139         /*
5140          * Touch the fpu the first time in non atomic context as if
5141          * this is the first fpu instruction the exception handler
5142          * will fire before the instruction returns and it'll have to
5143          * allocate ram with GFP_KERNEL.
5144          */
5145         if (!used_math())
5146                 kvm_fx_save(&vcpu->arch.host_fx_image);
5147
5148         /* Initialize guest FPU by resetting ours and saving into guest's */
5149         preempt_disable();
5150         kvm_fx_save(&vcpu->arch.host_fx_image);
5151         kvm_fx_finit();
5152         kvm_fx_save(&vcpu->arch.guest_fx_image);
5153         kvm_fx_restore(&vcpu->arch.host_fx_image);
5154         preempt_enable();
5155
5156         vcpu->arch.cr0 |= X86_CR0_ET;
5157         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5158         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5159         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5160                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5161 }
5162 EXPORT_SYMBOL_GPL(fx_init);
5163
5164 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5165 {
5166         if (vcpu->guest_fpu_loaded)
5167                 return;
5168
5169         vcpu->guest_fpu_loaded = 1;
5170         kvm_fx_save(&vcpu->arch.host_fx_image);
5171         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5172         trace_kvm_fpu(1);
5173 }
5174
5175 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5176 {
5177         if (!vcpu->guest_fpu_loaded)
5178                 return;
5179
5180         vcpu->guest_fpu_loaded = 0;
5181         kvm_fx_save(&vcpu->arch.guest_fx_image);
5182         kvm_fx_restore(&vcpu->arch.host_fx_image);
5183         ++vcpu->stat.fpu_reload;
5184         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5185         trace_kvm_fpu(0);
5186 }
5187
5188 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5189 {
5190         if (vcpu->arch.time_page) {
5191                 kvm_release_page_dirty(vcpu->arch.time_page);
5192                 vcpu->arch.time_page = NULL;
5193         }
5194
5195         kvm_x86_ops->vcpu_free(vcpu);
5196 }
5197
5198 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5199                                                 unsigned int id)
5200 {
5201         return kvm_x86_ops->vcpu_create(kvm, id);
5202 }
5203
5204 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5205 {
5206         int r;
5207
5208         /* We do fxsave: this must be aligned. */
5209         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5210
5211         vcpu->arch.mtrr_state.have_fixed = 1;
5212         vcpu_load(vcpu);
5213         r = kvm_arch_vcpu_reset(vcpu);
5214         if (r == 0)
5215                 r = kvm_mmu_setup(vcpu);
5216         vcpu_put(vcpu);
5217         if (r < 0)
5218                 goto free_vcpu;
5219
5220         return 0;
5221 free_vcpu:
5222         kvm_x86_ops->vcpu_free(vcpu);
5223         return r;
5224 }
5225
5226 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5227 {
5228         vcpu_load(vcpu);
5229         kvm_mmu_unload(vcpu);
5230         vcpu_put(vcpu);
5231
5232         kvm_x86_ops->vcpu_free(vcpu);
5233 }
5234
5235 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5236 {
5237         vcpu->arch.nmi_pending = false;
5238         vcpu->arch.nmi_injected = false;
5239
5240         vcpu->arch.switch_db_regs = 0;
5241         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5242         vcpu->arch.dr6 = DR6_FIXED_1;
5243         vcpu->arch.dr7 = DR7_FIXED_1;
5244
5245         return kvm_x86_ops->vcpu_reset(vcpu);
5246 }
5247
5248 int kvm_arch_hardware_enable(void *garbage)
5249 {
5250         /*
5251          * Since this may be called from a hotplug notifcation,
5252          * we can't get the CPU frequency directly.
5253          */
5254         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5255                 int cpu = raw_smp_processor_id();
5256                 per_cpu(cpu_tsc_khz, cpu) = 0;
5257         }
5258
5259         kvm_shared_msr_cpu_online();
5260
5261         return kvm_x86_ops->hardware_enable(garbage);
5262 }
5263
5264 void kvm_arch_hardware_disable(void *garbage)
5265 {
5266         kvm_x86_ops->hardware_disable(garbage);
5267         drop_user_return_notifiers(garbage);
5268 }
5269
5270 int kvm_arch_hardware_setup(void)
5271 {
5272         return kvm_x86_ops->hardware_setup();
5273 }
5274
5275 void kvm_arch_hardware_unsetup(void)
5276 {
5277         kvm_x86_ops->hardware_unsetup();
5278 }
5279
5280 void kvm_arch_check_processor_compat(void *rtn)
5281 {
5282         kvm_x86_ops->check_processor_compatibility(rtn);
5283 }
5284
5285 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5286 {
5287         struct page *page;
5288         struct kvm *kvm;
5289         int r;
5290
5291         BUG_ON(vcpu->kvm == NULL);
5292         kvm = vcpu->kvm;
5293
5294         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5295         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5296                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5297         else
5298                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5299
5300         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5301         if (!page) {
5302                 r = -ENOMEM;
5303                 goto fail;
5304         }
5305         vcpu->arch.pio_data = page_address(page);
5306
5307         r = kvm_mmu_create(vcpu);
5308         if (r < 0)
5309                 goto fail_free_pio_data;
5310
5311         if (irqchip_in_kernel(kvm)) {
5312                 r = kvm_create_lapic(vcpu);
5313                 if (r < 0)
5314                         goto fail_mmu_destroy;
5315         }
5316
5317         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5318                                        GFP_KERNEL);
5319         if (!vcpu->arch.mce_banks) {
5320                 r = -ENOMEM;
5321                 goto fail_free_lapic;
5322         }
5323         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5324
5325         return 0;
5326 fail_free_lapic:
5327         kvm_free_lapic(vcpu);
5328 fail_mmu_destroy:
5329         kvm_mmu_destroy(vcpu);
5330 fail_free_pio_data:
5331         free_page((unsigned long)vcpu->arch.pio_data);
5332 fail:
5333         return r;
5334 }
5335
5336 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5337 {
5338         int idx;
5339
5340         kfree(vcpu->arch.mce_banks);
5341         kvm_free_lapic(vcpu);
5342         idx = srcu_read_lock(&vcpu->kvm->srcu);
5343         kvm_mmu_destroy(vcpu);
5344         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5345         free_page((unsigned long)vcpu->arch.pio_data);
5346 }
5347
5348 struct  kvm *kvm_arch_create_vm(void)
5349 {
5350         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5351
5352         if (!kvm)
5353                 return ERR_PTR(-ENOMEM);
5354
5355         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5356         if (!kvm->arch.aliases) {
5357                 kfree(kvm);
5358                 return ERR_PTR(-ENOMEM);
5359         }
5360
5361         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5362         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5363
5364         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5365         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5366
5367         rdtscll(kvm->arch.vm_init_tsc);
5368
5369         return kvm;
5370 }
5371
5372 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5373 {
5374         vcpu_load(vcpu);
5375         kvm_mmu_unload(vcpu);
5376         vcpu_put(vcpu);
5377 }
5378
5379 static void kvm_free_vcpus(struct kvm *kvm)
5380 {
5381         unsigned int i;
5382         struct kvm_vcpu *vcpu;
5383
5384         /*
5385          * Unpin any mmu pages first.
5386          */
5387         kvm_for_each_vcpu(i, vcpu, kvm)
5388                 kvm_unload_vcpu_mmu(vcpu);
5389         kvm_for_each_vcpu(i, vcpu, kvm)
5390                 kvm_arch_vcpu_free(vcpu);
5391
5392         mutex_lock(&kvm->lock);
5393         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5394                 kvm->vcpus[i] = NULL;
5395
5396         atomic_set(&kvm->online_vcpus, 0);
5397         mutex_unlock(&kvm->lock);
5398 }
5399
5400 void kvm_arch_sync_events(struct kvm *kvm)
5401 {
5402         kvm_free_all_assigned_devices(kvm);
5403 }
5404
5405 void kvm_arch_destroy_vm(struct kvm *kvm)
5406 {
5407         kvm_iommu_unmap_guest(kvm);
5408         kvm_free_pit(kvm);
5409         kfree(kvm->arch.vpic);
5410         kfree(kvm->arch.vioapic);
5411         kvm_free_vcpus(kvm);
5412         kvm_free_physmem(kvm);
5413         if (kvm->arch.apic_access_page)
5414                 put_page(kvm->arch.apic_access_page);
5415         if (kvm->arch.ept_identity_pagetable)
5416                 put_page(kvm->arch.ept_identity_pagetable);
5417         cleanup_srcu_struct(&kvm->srcu);
5418         kfree(kvm->arch.aliases);
5419         kfree(kvm);
5420 }
5421
5422 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5423                                 struct kvm_memory_slot *memslot,
5424                                 struct kvm_memory_slot old,
5425                                 struct kvm_userspace_memory_region *mem,
5426                                 int user_alloc)
5427 {
5428         int npages = memslot->npages;
5429
5430         /*To keep backward compatibility with older userspace,
5431          *x86 needs to hanlde !user_alloc case.
5432          */
5433         if (!user_alloc) {
5434                 if (npages && !old.rmap) {
5435                         unsigned long userspace_addr;
5436
5437                         down_write(&current->mm->mmap_sem);
5438                         userspace_addr = do_mmap(NULL, 0,
5439                                                  npages * PAGE_SIZE,
5440                                                  PROT_READ | PROT_WRITE,
5441                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5442                                                  0);
5443                         up_write(&current->mm->mmap_sem);
5444
5445                         if (IS_ERR((void *)userspace_addr))
5446                                 return PTR_ERR((void *)userspace_addr);
5447
5448                         memslot->userspace_addr = userspace_addr;
5449                 }
5450         }
5451
5452
5453         return 0;
5454 }
5455
5456 void kvm_arch_commit_memory_region(struct kvm *kvm,
5457                                 struct kvm_userspace_memory_region *mem,
5458                                 struct kvm_memory_slot old,
5459                                 int user_alloc)
5460 {
5461
5462         int npages = mem->memory_size >> PAGE_SHIFT;
5463
5464         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5465                 int ret;
5466
5467                 down_write(&current->mm->mmap_sem);
5468                 ret = do_munmap(current->mm, old.userspace_addr,
5469                                 old.npages * PAGE_SIZE);
5470                 up_write(&current->mm->mmap_sem);
5471                 if (ret < 0)
5472                         printk(KERN_WARNING
5473                                "kvm_vm_ioctl_set_memory_region: "
5474                                "failed to munmap memory\n");
5475         }
5476
5477         spin_lock(&kvm->mmu_lock);
5478         if (!kvm->arch.n_requested_mmu_pages) {
5479                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5480                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5481         }
5482
5483         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5484         spin_unlock(&kvm->mmu_lock);
5485 }
5486
5487 void kvm_arch_flush_shadow(struct kvm *kvm)
5488 {
5489         kvm_mmu_zap_all(kvm);
5490         kvm_reload_remote_mmus(kvm);
5491 }
5492
5493 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5494 {
5495         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5496                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5497                 || vcpu->arch.nmi_pending ||
5498                 (kvm_arch_interrupt_allowed(vcpu) &&
5499                  kvm_cpu_has_interrupt(vcpu));
5500 }
5501
5502 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5503 {
5504         int me;
5505         int cpu = vcpu->cpu;
5506
5507         if (waitqueue_active(&vcpu->wq)) {
5508                 wake_up_interruptible(&vcpu->wq);
5509                 ++vcpu->stat.halt_wakeup;
5510         }
5511
5512         me = get_cpu();
5513         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5514                 if (atomic_xchg(&vcpu->guest_mode, 0))
5515                         smp_send_reschedule(cpu);
5516         put_cpu();
5517 }
5518
5519 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5520 {
5521         return kvm_x86_ops->interrupt_allowed(vcpu);
5522 }
5523
5524 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5525 {
5526         unsigned long current_rip = kvm_rip_read(vcpu) +
5527                 get_segment_base(vcpu, VCPU_SREG_CS);
5528
5529         return current_rip == linear_rip;
5530 }
5531 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5532
5533 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5534 {
5535         unsigned long rflags;
5536
5537         rflags = kvm_x86_ops->get_rflags(vcpu);
5538         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5539                 rflags &= ~X86_EFLAGS_TF;
5540         return rflags;
5541 }
5542 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5543
5544 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5545 {
5546         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5547             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5548                 rflags |= X86_EFLAGS_TF;
5549         kvm_x86_ops->set_rflags(vcpu, rflags);
5550 }
5551 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5552
5553 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5554 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5555 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5556 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5557 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5558 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5559 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5560 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5561 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5562 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5563 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5564 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);