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 <trace/events/kvm.h>
  44 #undef TRACE_INCLUDE_FILE
  45 #define CREATE_TRACE_POINTS
  46 #include "trace.h"
  47
  48 #include <asm/debugreg.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/msr.h>
  51 #include <asm/desc.h>
  52 #include <asm/mtrr.h>
  53 #include <asm/mce.h>
  54
  55 #define MAX_IO_MSRS 256
  56 #define CR0_RESERVED_BITS                                               \
  57         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  58                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  59                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  60 #define CR4_RESERVED_BITS                                               \
  61         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  62                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  63                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  64                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  65
  66 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  67
  68 #define KVM_MAX_MCE_BANKS 32
  69 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  70
  71 /* EFER defaults:
  72  * - enable syscall per default because its emulated by KVM
  73  * - enable LME and LMA per default on 64 bit KVM
  74  */
  75 #ifdef CONFIG_X86_64
  76 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  77 #else
  78 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  79 #endif
  80
  81 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  82 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  83
  84 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  85 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  86                                     struct kvm_cpuid_entry2 __user *entries);
  87
  88 struct kvm_x86_ops *kvm_x86_ops;
  89 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  90
  91 int ignore_msrs = 0;
  92 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  93
  94 #define KVM_NR_SHARED_MSRS 16
  95
  96 struct kvm_shared_msrs_global {
  97         int nr;
  98         u32 msrs[KVM_NR_SHARED_MSRS];
  99 };
 100
 101 struct kvm_shared_msrs {
 102         struct user_return_notifier urn;
 103         bool registered;
 104         struct kvm_shared_msr_values {
 105                 u64 host;
 106                 u64 curr;
 107         } values[KVM_NR_SHARED_MSRS];
 108 };
 109
 110 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 111 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 112
 113 struct kvm_stats_debugfs_item debugfs_entries[] = {
 114         { "pf_fixed", VCPU_STAT(pf_fixed) },
 115         { "pf_guest", VCPU_STAT(pf_guest) },
 116         { "tlb_flush", VCPU_STAT(tlb_flush) },
 117         { "invlpg", VCPU_STAT(invlpg) },
 118         { "exits", VCPU_STAT(exits) },
 119         { "io_exits", VCPU_STAT(io_exits) },
 120         { "mmio_exits", VCPU_STAT(mmio_exits) },
 121         { "signal_exits", VCPU_STAT(signal_exits) },
 122         { "irq_window", VCPU_STAT(irq_window_exits) },
 123         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 124         { "halt_exits", VCPU_STAT(halt_exits) },
 125         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 126         { "hypercalls", VCPU_STAT(hypercalls) },
 127         { "request_irq", VCPU_STAT(request_irq_exits) },
 128         { "irq_exits", VCPU_STAT(irq_exits) },
 129         { "host_state_reload", VCPU_STAT(host_state_reload) },
 130         { "efer_reload", VCPU_STAT(efer_reload) },
 131         { "fpu_reload", VCPU_STAT(fpu_reload) },
 132         { "insn_emulation", VCPU_STAT(insn_emulation) },
 133         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 134         { "irq_injections", VCPU_STAT(irq_injections) },
 135         { "nmi_injections", VCPU_STAT(nmi_injections) },
 136         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 137         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 138         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 139         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 140         { "mmu_flooded", VM_STAT(mmu_flooded) },
 141         { "mmu_recycled", VM_STAT(mmu_recycled) },
 142         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 143         { "mmu_unsync", VM_STAT(mmu_unsync) },
 144         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 145         { "largepages", VM_STAT(lpages) },
 146         { NULL }
 147 };
 148
 149 static void kvm_on_user_return(struct user_return_notifier *urn)
 150 {
 151         unsigned slot;
 152         struct kvm_shared_msrs *locals
 153                 = container_of(urn, struct kvm_shared_msrs, urn);
 154         struct kvm_shared_msr_values *values;
 155
 156         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 157                 values = &locals->values[slot];
 158                 if (values->host != values->curr) {
 159                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 160                         values->curr = values->host;
 161                 }
 162         }
 163         locals->registered = false;
 164         user_return_notifier_unregister(urn);
 165 }
 166
 167 static void shared_msr_update(unsigned slot, u32 msr)
 168 {
 169         struct kvm_shared_msrs *smsr;
 170         u64 value;
 171
 172         smsr = &__get_cpu_var(shared_msrs);
 173         /* only read, and nobody should modify it at this time,
 174          * so don't need lock */
 175         if (slot >= shared_msrs_global.nr) {
 176                 printk(KERN_ERR "kvm: invalid MSR slot!");
 177                 return;
 178         }
 179         rdmsrl_safe(msr, &value);
 180         smsr->values[slot].host = value;
 181         smsr->values[slot].curr = value;
 182 }
 183
 184 void kvm_define_shared_msr(unsigned slot, u32 msr)
 185 {
 186         if (slot >= shared_msrs_global.nr)
 187                 shared_msrs_global.nr = slot + 1;
 188         shared_msrs_global.msrs[slot] = msr;
 189         /* we need ensured the shared_msr_global have been updated */
 190         smp_wmb();
 191 }
 192 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 193
 194 static void kvm_shared_msr_cpu_online(void)
 195 {
 196         unsigned i;
 197
 198         for (i = 0; i < shared_msrs_global.nr; ++i)
 199                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 200 }
 201
 202 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 203 {
 204         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 205
 206         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 207                 return;
 208         smsr->values[slot].curr = value;
 209         wrmsrl(shared_msrs_global.msrs[slot], value);
 210         if (!smsr->registered) {
 211                 smsr->urn.on_user_return = kvm_on_user_return;
 212                 user_return_notifier_register(&smsr->urn);
 213                 smsr->registered = true;
 214         }
 215 }
 216 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 217
 218 static void drop_user_return_notifiers(void *ignore)
 219 {
 220         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 221
 222         if (smsr->registered)
 223                 kvm_on_user_return(&smsr->urn);
 224 }
 225
 226 unsigned long segment_base(u16 selector)
 227 {
 228         struct desc_ptr gdt;
 229         struct desc_struct *d;
 230         unsigned long table_base;
 231         unsigned long v;
 232
 233         if (selector == 0)
 234                 return 0;
 235
 236         kvm_get_gdt(&gdt);
 237         table_base = gdt.address;
 238
 239         if (selector & 4) {           /* from ldt */
 240                 u16 ldt_selector = kvm_read_ldt();
 241
 242                 table_base = segment_base(ldt_selector);
 243         }
 244         d = (struct desc_struct *)(table_base + (selector & ~7));
 245         v = get_desc_base(d);
 246 #ifdef CONFIG_X86_64
 247         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 248                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 249 #endif
 250         return v;
 251 }
 252 EXPORT_SYMBOL_GPL(segment_base);
 253
 254 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 255 {
 256         if (irqchip_in_kernel(vcpu->kvm))
 257                 return vcpu->arch.apic_base;
 258         else
 259                 return vcpu->arch.apic_base;
 260 }
 261 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 262
 263 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 264 {
 265         /* TODO: reserve bits check */
 266         if (irqchip_in_kernel(vcpu->kvm))
 267                 kvm_lapic_set_base(vcpu, data);
 268         else
 269                 vcpu->arch.apic_base = data;
 270 }
 271 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 272
 273 #define EXCPT_BENIGN            0
 274 #define EXCPT_CONTRIBUTORY      1
 275 #define EXCPT_PF                2
 276
 277 static int exception_class(int vector)
 278 {
 279         switch (vector) {
 280         case PF_VECTOR:
 281                 return EXCPT_PF;
 282         case DE_VECTOR:
 283         case TS_VECTOR:
 284         case NP_VECTOR:
 285         case SS_VECTOR:
 286         case GP_VECTOR:
 287                 return EXCPT_CONTRIBUTORY;
 288         default:
 289                 break;
 290         }
 291         return EXCPT_BENIGN;
 292 }
 293
 294 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 295                 unsigned nr, bool has_error, u32 error_code)
 296 {
 297         u32 prev_nr;
 298         int class1, class2;
 299
 300         if (!vcpu->arch.exception.pending) {
 301         queue:
 302                 vcpu->arch.exception.pending = true;
 303                 vcpu->arch.exception.has_error_code = has_error;
 304                 vcpu->arch.exception.nr = nr;
 305                 vcpu->arch.exception.error_code = error_code;
 306                 return;
 307         }
 308
 309         /* to check exception */
 310         prev_nr = vcpu->arch.exception.nr;
 311         if (prev_nr == DF_VECTOR) {
 312                 /* triple fault -> shutdown */
 313                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 314                 return;
 315         }
 316         class1 = exception_class(prev_nr);
 317         class2 = exception_class(nr);
 318         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 319                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 320                 /* generate double fault per SDM Table 5-5 */
 321                 vcpu->arch.exception.pending = true;
 322                 vcpu->arch.exception.has_error_code = true;
 323                 vcpu->arch.exception.nr = DF_VECTOR;
 324                 vcpu->arch.exception.error_code = 0;
 325         } else
 326                 /* replace previous exception with a new one in a hope
 327                    that instruction re-execution will regenerate lost
 328                    exception */
 329                 goto queue;
 330 }
 331
 332 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 333 {
 334         kvm_multiple_exception(vcpu, nr, false, 0);
 335 }
 336 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 337
 338 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 339                            u32 error_code)
 340 {
 341         ++vcpu->stat.pf_guest;
 342         vcpu->arch.cr2 = addr;
 343         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 344 }
 345
 346 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 347 {
 348         vcpu->arch.nmi_pending = 1;
 349 }
 350 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 351
 352 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 353 {
 354         kvm_multiple_exception(vcpu, nr, true, error_code);
 355 }
 356 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 357
 358 /*
 359  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 360  * a #GP and return false.
 361  */
 362 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 363 {
 364         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 365                 return true;
 366         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 367         return false;
 368 }
 369 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 370
 371 /*
 372  * Load the pae pdptrs.  Return true is they are all valid.
 373  */
 374 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 375 {
 376         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 377         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 378         int i;
 379         int ret;
 380         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 381
 382         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 383                                   offset * sizeof(u64), sizeof(pdpte));
 384         if (ret < 0) {
 385                 ret = 0;
 386                 goto out;
 387         }
 388         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 389                 if (is_present_gpte(pdpte[i]) &&
 390                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 391                         ret = 0;
 392                         goto out;
 393                 }
 394         }
 395         ret = 1;
 396
 397         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 398         __set_bit(VCPU_EXREG_PDPTR,
 399                   (unsigned long *)&vcpu->arch.regs_avail);
 400         __set_bit(VCPU_EXREG_PDPTR,
 401                   (unsigned long *)&vcpu->arch.regs_dirty);
 402 out:
 403
 404         return ret;
 405 }
 406 EXPORT_SYMBOL_GPL(load_pdptrs);
 407
 408 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 409 {
 410         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 411         bool changed = true;
 412         int r;
 413
 414         if (is_long_mode(vcpu) || !is_pae(vcpu))
 415                 return false;
 416
 417         if (!test_bit(VCPU_EXREG_PDPTR,
 418                       (unsigned long *)&vcpu->arch.regs_avail))
 419                 return true;
 420
 421         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 422         if (r < 0)
 423                 goto out;
 424         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 425 out:
 426
 427         return changed;
 428 }
 429
 430 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 431 {
 432         cr0 |= X86_CR0_ET;
 433
 434 #ifdef CONFIG_X86_64
 435         if (cr0 & 0xffffffff00000000UL) {
 436                 kvm_inject_gp(vcpu, 0);
 437                 return;
 438         }
 439 #endif
 440
 441         cr0 &= ~CR0_RESERVED_BITS;
 442
 443         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 444                 kvm_inject_gp(vcpu, 0);
 445                 return;
 446         }
 447
 448         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 449                 kvm_inject_gp(vcpu, 0);
 450                 return;
 451         }
 452
 453         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 454 #ifdef CONFIG_X86_64
 455                 if ((vcpu->arch.efer & EFER_LME)) {
 456                         int cs_db, cs_l;
 457
 458                         if (!is_pae(vcpu)) {
 459                                 kvm_inject_gp(vcpu, 0);
 460                                 return;
 461                         }
 462                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 463                         if (cs_l) {
 464                                 kvm_inject_gp(vcpu, 0);
 465                                 return;
 466
 467                         }
 468                 } else
 469 #endif
 470                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 471                         kvm_inject_gp(vcpu, 0);
 472                         return;
 473                 }
 474
 475         }
 476
 477         kvm_x86_ops->set_cr0(vcpu, cr0);
 478         vcpu->arch.cr0 = cr0;
 479
 480         kvm_mmu_reset_context(vcpu);
 481         return;
 482 }
 483 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 484
 485 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 486 {
 487         kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
 488 }
 489 EXPORT_SYMBOL_GPL(kvm_lmsw);
 490
 491 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 492 {
 493         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 494         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 495
 496         if (cr4 & CR4_RESERVED_BITS) {
 497                 kvm_inject_gp(vcpu, 0);
 498                 return;
 499         }
 500
 501         if (is_long_mode(vcpu)) {
 502                 if (!(cr4 & X86_CR4_PAE)) {
 503                         kvm_inject_gp(vcpu, 0);
 504                         return;
 505                 }
 506         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 507                    && ((cr4 ^ old_cr4) & pdptr_bits)
 508                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 509                 kvm_inject_gp(vcpu, 0);
 510                 return;
 511         }
 512
 513         if (cr4 & X86_CR4_VMXE) {
 514                 kvm_inject_gp(vcpu, 0);
 515                 return;
 516         }
 517         kvm_x86_ops->set_cr4(vcpu, cr4);
 518         vcpu->arch.cr4 = cr4;
 519         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 520         kvm_mmu_reset_context(vcpu);
 521 }
 522 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 523
 524 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 525 {
 526         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 527                 kvm_mmu_sync_roots(vcpu);
 528                 kvm_mmu_flush_tlb(vcpu);
 529                 return;
 530         }
 531
 532         if (is_long_mode(vcpu)) {
 533                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 534                         kvm_inject_gp(vcpu, 0);
 535                         return;
 536                 }
 537         } else {
 538                 if (is_pae(vcpu)) {
 539                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 540                                 kvm_inject_gp(vcpu, 0);
 541                                 return;
 542                         }
 543                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 544                                 kvm_inject_gp(vcpu, 0);
 545                                 return;
 546                         }
 547                 }
 548                 /*
 549                  * We don't check reserved bits in nonpae mode, because
 550                  * this isn't enforced, and VMware depends on this.
 551                  */
 552         }
 553
 554         /*
 555          * Does the new cr3 value map to physical memory? (Note, we
 556          * catch an invalid cr3 even in real-mode, because it would
 557          * cause trouble later on when we turn on paging anyway.)
 558          *
 559          * A real CPU would silently accept an invalid cr3 and would
 560          * attempt to use it - with largely undefined (and often hard
 561          * to debug) behavior on the guest side.
 562          */
 563         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 564                 kvm_inject_gp(vcpu, 0);
 565         else {
 566                 vcpu->arch.cr3 = cr3;
 567                 vcpu->arch.mmu.new_cr3(vcpu);
 568         }
 569 }
 570 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 571
 572 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 573 {
 574         if (cr8 & CR8_RESERVED_BITS) {
 575                 kvm_inject_gp(vcpu, 0);
 576                 return;
 577         }
 578         if (irqchip_in_kernel(vcpu->kvm))
 579                 kvm_lapic_set_tpr(vcpu, cr8);
 580         else
 581                 vcpu->arch.cr8 = cr8;
 582 }
 583 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 584
 585 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 586 {
 587         if (irqchip_in_kernel(vcpu->kvm))
 588                 return kvm_lapic_get_cr8(vcpu);
 589         else
 590                 return vcpu->arch.cr8;
 591 }
 592 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 593
 594 static inline u32 bit(int bitno)
 595 {
 596         return 1 << (bitno & 31);
 597 }
 598
 599 /*
 600  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 601  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 602  *
 603  * This list is modified at module load time to reflect the
 604  * capabilities of the host cpu. This capabilities test skips MSRs that are
 605  * kvm-specific. Those are put in the beginning of the list.
 606  */
 607
 608 #define KVM_SAVE_MSRS_BEGIN     5
 609 static u32 msrs_to_save[] = {
 610         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 611         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 612         HV_X64_MSR_APIC_ASSIST_PAGE,
 613         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 614         MSR_K6_STAR,
 615 #ifdef CONFIG_X86_64
 616         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 617 #endif
 618         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 619 };
 620
 621 static unsigned num_msrs_to_save;
 622
 623 static u32 emulated_msrs[] = {
 624         MSR_IA32_MISC_ENABLE,
 625 };
 626
 627 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 628 {
 629         if (efer & efer_reserved_bits) {
 630                 kvm_inject_gp(vcpu, 0);
 631                 return;
 632         }
 633
 634         if (is_paging(vcpu)
 635             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) {
 636                 kvm_inject_gp(vcpu, 0);
 637                 return;
 638         }
 639
 640         if (efer & EFER_FFXSR) {
 641                 struct kvm_cpuid_entry2 *feat;
 642
 643                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 644                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 645                         kvm_inject_gp(vcpu, 0);
 646                         return;
 647                 }
 648         }
 649
 650         if (efer & EFER_SVME) {
 651                 struct kvm_cpuid_entry2 *feat;
 652
 653                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 654                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 655                         kvm_inject_gp(vcpu, 0);
 656                         return;
 657                 }
 658         }
 659
 660         kvm_x86_ops->set_efer(vcpu, efer);
 661
 662         efer &= ~EFER_LMA;
 663         efer |= vcpu->arch.efer & EFER_LMA;
 664
 665         vcpu->arch.efer = efer;
 666
 667         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 668         kvm_mmu_reset_context(vcpu);
 669 }
 670
 671 void kvm_enable_efer_bits(u64 mask)
 672 {
 673        efer_reserved_bits &= ~mask;
 674 }
 675 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 676
 677
 678 /*
 679  * Writes msr value into into the appropriate "register".
 680  * Returns 0 on success, non-0 otherwise.
 681  * Assumes vcpu_load() was already called.
 682  */
 683 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 684 {
 685         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 686 }
 687
 688 /*
 689  * Adapt set_msr() to msr_io()'s calling convention
 690  */
 691 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 692 {
 693         return kvm_set_msr(vcpu, index, *data);
 694 }
 695
 696 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 697 {
 698         static int version;
 699         struct pvclock_wall_clock wc;
 700         struct timespec boot;
 701
 702         if (!wall_clock)
 703                 return;
 704
 705         version++;
 706
 707         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 708
 709         /*
 710          * The guest calculates current wall clock time by adding
 711          * system time (updated by kvm_write_guest_time below) to the
 712          * wall clock specified here.  guest system time equals host
 713          * system time for us, thus we must fill in host boot time here.
 714          */
 715         getboottime(&boot);
 716
 717         wc.sec = boot.tv_sec;
 718         wc.nsec = boot.tv_nsec;
 719         wc.version = version;
 720
 721         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 722
 723         version++;
 724         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 725 }
 726
 727 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 728 {
 729         uint32_t quotient, remainder;
 730
 731         /* Don't try to replace with do_div(), this one calculates
 732          * "(dividend << 32) / divisor" */
 733         __asm__ ( "divl %4"
 734                   : "=a" (quotient), "=d" (remainder)
 735                   : "0" (0), "1" (dividend), "r" (divisor) );
 736         return quotient;
 737 }
 738
 739 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 740 {
 741         uint64_t nsecs = 1000000000LL;
 742         int32_t  shift = 0;
 743         uint64_t tps64;
 744         uint32_t tps32;
 745
 746         tps64 = tsc_khz * 1000LL;
 747         while (tps64 > nsecs*2) {
 748                 tps64 >>= 1;
 749                 shift--;
 750         }
 751
 752         tps32 = (uint32_t)tps64;
 753         while (tps32 <= (uint32_t)nsecs) {
 754                 tps32 <<= 1;
 755                 shift++;
 756         }
 757
 758         hv_clock->tsc_shift = shift;
 759         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 760
 761         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 762                  __func__, tsc_khz, hv_clock->tsc_shift,
 763                  hv_clock->tsc_to_system_mul);
 764 }
 765
 766 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 767
 768 static void kvm_write_guest_time(struct kvm_vcpu *v)
 769 {
 770         struct timespec ts;
 771         unsigned long flags;
 772         struct kvm_vcpu_arch *vcpu = &v->arch;
 773         void *shared_kaddr;
 774         unsigned long this_tsc_khz;
 775
 776         if ((!vcpu->time_page))
 777                 return;
 778
 779         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 780         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 781                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 782                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 783         }
 784         put_cpu_var(cpu_tsc_khz);
 785
 786         /* Keep irq disabled to prevent changes to the clock */
 787         local_irq_save(flags);
 788         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 789         ktime_get_ts(&ts);
 790         monotonic_to_bootbased(&ts);
 791         local_irq_restore(flags);
 792
 793         /* With all the info we got, fill in the values */
 794
 795         vcpu->hv_clock.system_time = ts.tv_nsec +
 796                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 797
 798         /*
 799          * The interface expects us to write an even number signaling that the
 800          * update is finished. Since the guest won't see the intermediate
 801          * state, we just increase by 2 at the end.
 802          */
 803         vcpu->hv_clock.version += 2;
 804
 805         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 806
 807         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 808                sizeof(vcpu->hv_clock));
 809
 810         kunmap_atomic(shared_kaddr, KM_USER0);
 811
 812         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 813 }
 814
 815 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 816 {
 817         struct kvm_vcpu_arch *vcpu = &v->arch;
 818
 819         if (!vcpu->time_page)
 820                 return 0;
 821         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 822         return 1;
 823 }
 824
 825 static bool msr_mtrr_valid(unsigned msr)
 826 {
 827         switch (msr) {
 828         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 829         case MSR_MTRRfix64K_00000:
 830         case MSR_MTRRfix16K_80000:
 831         case MSR_MTRRfix16K_A0000:
 832         case MSR_MTRRfix4K_C0000:
 833         case MSR_MTRRfix4K_C8000:
 834         case MSR_MTRRfix4K_D0000:
 835         case MSR_MTRRfix4K_D8000:
 836         case MSR_MTRRfix4K_E0000:
 837         case MSR_MTRRfix4K_E8000:
 838         case MSR_MTRRfix4K_F0000:
 839         case MSR_MTRRfix4K_F8000:
 840         case MSR_MTRRdefType:
 841         case MSR_IA32_CR_PAT:
 842                 return true;
 843         case 0x2f8:
 844                 return true;
 845         }
 846         return false;
 847 }
 848
 849 static bool valid_pat_type(unsigned t)
 850 {
 851         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 852 }
 853
 854 static bool valid_mtrr_type(unsigned t)
 855 {
 856         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 857 }
 858
 859 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 860 {
 861         int i;
 862
 863         if (!msr_mtrr_valid(msr))
 864                 return false;
 865
 866         if (msr == MSR_IA32_CR_PAT) {
 867                 for (i = 0; i < 8; i++)
 868                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 869                                 return false;
 870                 return true;
 871         } else if (msr == MSR_MTRRdefType) {
 872                 if (data & ~0xcff)
 873                         return false;
 874                 return valid_mtrr_type(data & 0xff);
 875         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 876                 for (i = 0; i < 8 ; i++)
 877                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 878                                 return false;
 879                 return true;
 880         }
 881
 882         /* variable MTRRs */
 883         return valid_mtrr_type(data & 0xff);
 884 }
 885
 886 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 887 {
 888         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 889
 890         if (!mtrr_valid(vcpu, msr, data))
 891                 return 1;
 892
 893         if (msr == MSR_MTRRdefType) {
 894                 vcpu->arch.mtrr_state.def_type = data;
 895                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 896         } else if (msr == MSR_MTRRfix64K_00000)
 897                 p[0] = data;
 898         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 899                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 900         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 901                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 902         else if (msr == MSR_IA32_CR_PAT)
 903                 vcpu->arch.pat = data;
 904         else {  /* Variable MTRRs */
 905                 int idx, is_mtrr_mask;
 906                 u64 *pt;
 907
 908                 idx = (msr - 0x200) / 2;
 909                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 910                 if (!is_mtrr_mask)
 911                         pt =
 912                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 913                 else
 914                         pt =
 915                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 916                 *pt = data;
 917         }
 918
 919         kvm_mmu_reset_context(vcpu);
 920         return 0;
 921 }
 922
 923 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 924 {
 925         u64 mcg_cap = vcpu->arch.mcg_cap;
 926         unsigned bank_num = mcg_cap & 0xff;
 927
 928         switch (msr) {
 929         case MSR_IA32_MCG_STATUS:
 930                 vcpu->arch.mcg_status = data;
 931                 break;
 932         case MSR_IA32_MCG_CTL:
 933                 if (!(mcg_cap & MCG_CTL_P))
 934                         return 1;
 935                 if (data != 0 && data != ~(u64)0)
 936                         return -1;
 937                 vcpu->arch.mcg_ctl = data;
 938                 break;
 939         default:
 940                 if (msr >= MSR_IA32_MC0_CTL &&
 941                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 942                         u32 offset = msr - MSR_IA32_MC0_CTL;
 943                         /* only 0 or all 1s can be written to IA32_MCi_CTL
 944                          * some Linux kernels though clear bit 10 in bank 4 to
 945                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
 946                          * this to avoid an uncatched #GP in the guest
 947                          */
 948                         if ((offset & 0x3) == 0 &&
 949                             data != 0 && (data | (1 << 10)) != ~(u64)0)
 950                                 return -1;
 951                         vcpu->arch.mce_banks[offset] = data;
 952                         break;
 953                 }
 954                 return 1;
 955         }
 956         return 0;
 957 }
 958
 959 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 960 {
 961         struct kvm *kvm = vcpu->kvm;
 962         int lm = is_long_mode(vcpu);
 963         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 964                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 965         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 966                 : kvm->arch.xen_hvm_config.blob_size_32;
 967         u32 page_num = data & ~PAGE_MASK;
 968         u64 page_addr = data & PAGE_MASK;
 969         u8 *page;
 970         int r;
 971
 972         r = -E2BIG;
 973         if (page_num >= blob_size)
 974                 goto out;
 975         r = -ENOMEM;
 976         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 977         if (!page)
 978                 goto out;
 979         r = -EFAULT;
 980         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
 981                 goto out_free;
 982         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
 983                 goto out_free;
 984         r = 0;
 985 out_free:
 986         kfree(page);
 987 out:
 988         return r;
 989 }
 990
 991 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
 992 {
 993         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
 994 }
 995
 996 static bool kvm_hv_msr_partition_wide(u32 msr)
 997 {
 998         bool r = false;
 999         switch (msr) {
1000         case HV_X64_MSR_GUEST_OS_ID:
1001         case HV_X64_MSR_HYPERCALL:
1002                 r = true;
1003                 break;
1004         }
1005
1006         return r;
1007 }
1008
1009 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1010 {
1011         struct kvm *kvm = vcpu->kvm;
1012
1013         switch (msr) {
1014         case HV_X64_MSR_GUEST_OS_ID:
1015                 kvm->arch.hv_guest_os_id = data;
1016                 /* setting guest os id to zero disables hypercall page */
1017                 if (!kvm->arch.hv_guest_os_id)
1018                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1019                 break;
1020         case HV_X64_MSR_HYPERCALL: {
1021                 u64 gfn;
1022                 unsigned long addr;
1023                 u8 instructions[4];
1024
1025                 /* if guest os id is not set hypercall should remain disabled */
1026                 if (!kvm->arch.hv_guest_os_id)
1027                         break;
1028                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1029                         kvm->arch.hv_hypercall = data;
1030                         break;
1031                 }
1032                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1033                 addr = gfn_to_hva(kvm, gfn);
1034                 if (kvm_is_error_hva(addr))
1035                         return 1;
1036                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1037                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1038                 if (copy_to_user((void __user *)addr, instructions, 4))
1039                         return 1;
1040                 kvm->arch.hv_hypercall = data;
1041                 break;
1042         }
1043         default:
1044                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1045                           "data 0x%llx\n", msr, data);
1046                 return 1;
1047         }
1048         return 0;
1049 }
1050
1051 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1052 {
1053         switch (msr) {
1054         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1055                 unsigned long addr;
1056
1057                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1058                         vcpu->arch.hv_vapic = data;
1059                         break;
1060                 }
1061                 addr = gfn_to_hva(vcpu->kvm, data >>
1062                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1063                 if (kvm_is_error_hva(addr))
1064                         return 1;
1065                 if (clear_user((void __user *)addr, PAGE_SIZE))
1066                         return 1;
1067                 vcpu->arch.hv_vapic = data;
1068                 break;
1069         }
1070         case HV_X64_MSR_EOI:
1071                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1072         case HV_X64_MSR_ICR:
1073                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1074         case HV_X64_MSR_TPR:
1075                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1076         default:
1077                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1078                           "data 0x%llx\n", msr, data);
1079                 return 1;
1080         }
1081
1082         return 0;
1083 }
1084
1085 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1086 {
1087         switch (msr) {
1088         case MSR_EFER:
1089                 set_efer(vcpu, data);
1090                 break;
1091         case MSR_K7_HWCR:
1092                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1093                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1094                 if (data != 0) {
1095                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1096                                 data);
1097                         return 1;
1098                 }
1099                 break;
1100         case MSR_FAM10H_MMIO_CONF_BASE:
1101                 if (data != 0) {
1102                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1103                                 "0x%llx\n", data);
1104                         return 1;
1105                 }
1106                 break;
1107         case MSR_AMD64_NB_CFG:
1108                 break;
1109         case MSR_IA32_DEBUGCTLMSR:
1110                 if (!data) {
1111                         /* We support the non-activated case already */
1112                         break;
1113                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1114                         /* Values other than LBR and BTF are vendor-specific,
1115                            thus reserved and should throw a #GP */
1116                         return 1;
1117                 }
1118                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1119                         __func__, data);
1120                 break;
1121         case MSR_IA32_UCODE_REV:
1122         case MSR_IA32_UCODE_WRITE:
1123         case MSR_VM_HSAVE_PA:
1124         case MSR_AMD64_PATCH_LOADER:
1125                 break;
1126         case 0x200 ... 0x2ff:
1127                 return set_msr_mtrr(vcpu, msr, data);
1128         case MSR_IA32_APICBASE:
1129                 kvm_set_apic_base(vcpu, data);
1130                 break;
1131         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1132                 return kvm_x2apic_msr_write(vcpu, msr, data);
1133         case MSR_IA32_MISC_ENABLE:
1134                 vcpu->arch.ia32_misc_enable_msr = data;
1135                 break;
1136         case MSR_KVM_WALL_CLOCK:
1137                 vcpu->kvm->arch.wall_clock = data;
1138                 kvm_write_wall_clock(vcpu->kvm, data);
1139                 break;
1140         case MSR_KVM_SYSTEM_TIME: {
1141                 if (vcpu->arch.time_page) {
1142                         kvm_release_page_dirty(vcpu->arch.time_page);
1143                         vcpu->arch.time_page = NULL;
1144                 }
1145
1146                 vcpu->arch.time = data;
1147
1148                 /* we verify if the enable bit is set... */
1149                 if (!(data & 1))
1150                         break;
1151
1152                 /* ...but clean it before doing the actual write */
1153                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1154
1155                 vcpu->arch.time_page =
1156                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1157
1158                 if (is_error_page(vcpu->arch.time_page)) {
1159                         kvm_release_page_clean(vcpu->arch.time_page);
1160                         vcpu->arch.time_page = NULL;
1161                 }
1162
1163                 kvm_request_guest_time_update(vcpu);
1164                 break;
1165         }
1166         case MSR_IA32_MCG_CTL:
1167         case MSR_IA32_MCG_STATUS:
1168         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1169                 return set_msr_mce(vcpu, msr, data);
1170
1171         /* Performance counters are not protected by a CPUID bit,
1172          * so we should check all of them in the generic path for the sake of
1173          * cross vendor migration.
1174          * Writing a zero into the event select MSRs disables them,
1175          * which we perfectly emulate ;-). Any other value should be at least
1176          * reported, some guests depend on them.
1177          */
1178         case MSR_P6_EVNTSEL0:
1179         case MSR_P6_EVNTSEL1:
1180         case MSR_K7_EVNTSEL0:
1181         case MSR_K7_EVNTSEL1:
1182         case MSR_K7_EVNTSEL2:
1183         case MSR_K7_EVNTSEL3:
1184                 if (data != 0)
1185                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1186                                 "0x%x data 0x%llx\n", msr, data);
1187                 break;
1188         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1189          * so we ignore writes to make it happy.
1190          */
1191         case MSR_P6_PERFCTR0:
1192         case MSR_P6_PERFCTR1:
1193         case MSR_K7_PERFCTR0:
1194         case MSR_K7_PERFCTR1:
1195         case MSR_K7_PERFCTR2:
1196         case MSR_K7_PERFCTR3:
1197                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1198                         "0x%x data 0x%llx\n", msr, data);
1199                 break;
1200         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1201                 if (kvm_hv_msr_partition_wide(msr)) {
1202                         int r;
1203                         mutex_lock(&vcpu->kvm->lock);
1204                         r = set_msr_hyperv_pw(vcpu, msr, data);
1205                         mutex_unlock(&vcpu->kvm->lock);
1206                         return r;
1207                 } else
1208                         return set_msr_hyperv(vcpu, msr, data);
1209                 break;
1210         default:
1211                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1212                         return xen_hvm_config(vcpu, data);
1213                 if (!ignore_msrs) {
1214                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1215                                 msr, data);
1216                         return 1;
1217                 } else {
1218                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1219                                 msr, data);
1220                         break;
1221                 }
1222         }
1223         return 0;
1224 }
1225 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1226
1227
1228 /*
1229  * Reads an msr value (of 'msr_index') into 'pdata'.
1230  * Returns 0 on success, non-0 otherwise.
1231  * Assumes vcpu_load() was already called.
1232  */
1233 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1234 {
1235         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1236 }
1237
1238 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1239 {
1240         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1241
1242         if (!msr_mtrr_valid(msr))
1243                 return 1;
1244
1245         if (msr == MSR_MTRRdefType)
1246                 *pdata = vcpu->arch.mtrr_state.def_type +
1247                          (vcpu->arch.mtrr_state.enabled << 10);
1248         else if (msr == MSR_MTRRfix64K_00000)
1249                 *pdata = p[0];
1250         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1251                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1252         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1253                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1254         else if (msr == MSR_IA32_CR_PAT)
1255                 *pdata = vcpu->arch.pat;
1256         else {  /* Variable MTRRs */
1257                 int idx, is_mtrr_mask;
1258                 u64 *pt;
1259
1260                 idx = (msr - 0x200) / 2;
1261                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1262                 if (!is_mtrr_mask)
1263                         pt =
1264                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1265                 else
1266                         pt =
1267                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1268                 *pdata = *pt;
1269         }
1270
1271         return 0;
1272 }
1273
1274 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1275 {
1276         u64 data;
1277         u64 mcg_cap = vcpu->arch.mcg_cap;
1278         unsigned bank_num = mcg_cap & 0xff;
1279
1280         switch (msr) {
1281         case MSR_IA32_P5_MC_ADDR:
1282         case MSR_IA32_P5_MC_TYPE:
1283                 data = 0;
1284                 break;
1285         case MSR_IA32_MCG_CAP:
1286                 data = vcpu->arch.mcg_cap;
1287                 break;
1288         case MSR_IA32_MCG_CTL:
1289                 if (!(mcg_cap & MCG_CTL_P))
1290                         return 1;
1291                 data = vcpu->arch.mcg_ctl;
1292                 break;
1293         case MSR_IA32_MCG_STATUS:
1294                 data = vcpu->arch.mcg_status;
1295                 break;
1296         default:
1297                 if (msr >= MSR_IA32_MC0_CTL &&
1298                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1299                         u32 offset = msr - MSR_IA32_MC0_CTL;
1300                         data = vcpu->arch.mce_banks[offset];
1301                         break;
1302                 }
1303                 return 1;
1304         }
1305         *pdata = data;
1306         return 0;
1307 }
1308
1309 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1310 {
1311         u64 data = 0;
1312         struct kvm *kvm = vcpu->kvm;
1313
1314         switch (msr) {
1315         case HV_X64_MSR_GUEST_OS_ID:
1316                 data = kvm->arch.hv_guest_os_id;
1317                 break;
1318         case HV_X64_MSR_HYPERCALL:
1319                 data = kvm->arch.hv_hypercall;
1320                 break;
1321         default:
1322                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1323                 return 1;
1324         }
1325
1326         *pdata = data;
1327         return 0;
1328 }
1329
1330 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1331 {
1332         u64 data = 0;
1333
1334         switch (msr) {
1335         case HV_X64_MSR_VP_INDEX: {
1336                 int r;
1337                 struct kvm_vcpu *v;
1338                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1339                         if (v == vcpu)
1340                                 data = r;
1341                 break;
1342         }
1343         case HV_X64_MSR_EOI:
1344                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1345         case HV_X64_MSR_ICR:
1346                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1347         case HV_X64_MSR_TPR:
1348                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1349         default:
1350                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1351                 return 1;
1352         }
1353         *pdata = data;
1354         return 0;
1355 }
1356
1357 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1358 {
1359         u64 data;
1360
1361         switch (msr) {
1362         case MSR_IA32_PLATFORM_ID:
1363         case MSR_IA32_UCODE_REV:
1364         case MSR_IA32_EBL_CR_POWERON:
1365         case MSR_IA32_DEBUGCTLMSR:
1366         case MSR_IA32_LASTBRANCHFROMIP:
1367         case MSR_IA32_LASTBRANCHTOIP:
1368         case MSR_IA32_LASTINTFROMIP:
1369         case MSR_IA32_LASTINTTOIP:
1370         case MSR_K8_SYSCFG:
1371         case MSR_K7_HWCR:
1372         case MSR_VM_HSAVE_PA:
1373         case MSR_P6_PERFCTR0:
1374         case MSR_P6_PERFCTR1:
1375         case MSR_P6_EVNTSEL0:
1376         case MSR_P6_EVNTSEL1:
1377         case MSR_K7_EVNTSEL0:
1378         case MSR_K7_PERFCTR0:
1379         case MSR_K8_INT_PENDING_MSG:
1380         case MSR_AMD64_NB_CFG:
1381         case MSR_FAM10H_MMIO_CONF_BASE:
1382                 data = 0;
1383                 break;
1384         case MSR_MTRRcap:
1385                 data = 0x500 | KVM_NR_VAR_MTRR;
1386                 break;
1387         case 0x200 ... 0x2ff:
1388                 return get_msr_mtrr(vcpu, msr, pdata);
1389         case 0xcd: /* fsb frequency */
1390                 data = 3;
1391                 break;
1392         case MSR_IA32_APICBASE:
1393                 data = kvm_get_apic_base(vcpu);
1394                 break;
1395         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1396                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1397                 break;
1398         case MSR_IA32_MISC_ENABLE:
1399                 data = vcpu->arch.ia32_misc_enable_msr;
1400                 break;
1401         case MSR_IA32_PERF_STATUS:
1402                 /* TSC increment by tick */
1403                 data = 1000ULL;
1404                 /* CPU multiplier */
1405                 data |= (((uint64_t)4ULL) << 40);
1406                 break;
1407         case MSR_EFER:
1408                 data = vcpu->arch.efer;
1409                 break;
1410         case MSR_KVM_WALL_CLOCK:
1411                 data = vcpu->kvm->arch.wall_clock;
1412                 break;
1413         case MSR_KVM_SYSTEM_TIME:
1414                 data = vcpu->arch.time;
1415                 break;
1416         case MSR_IA32_P5_MC_ADDR:
1417         case MSR_IA32_P5_MC_TYPE:
1418         case MSR_IA32_MCG_CAP:
1419         case MSR_IA32_MCG_CTL:
1420         case MSR_IA32_MCG_STATUS:
1421         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1422                 return get_msr_mce(vcpu, msr, pdata);
1423         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1424                 if (kvm_hv_msr_partition_wide(msr)) {
1425                         int r;
1426                         mutex_lock(&vcpu->kvm->lock);
1427                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1428                         mutex_unlock(&vcpu->kvm->lock);
1429                         return r;
1430                 } else
1431                         return get_msr_hyperv(vcpu, msr, pdata);
1432                 break;
1433         default:
1434                 if (!ignore_msrs) {
1435                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1436                         return 1;
1437                 } else {
1438                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1439                         data = 0;
1440                 }
1441                 break;
1442         }
1443         *pdata = data;
1444         return 0;
1445 }
1446 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1447
1448 /*
1449  * Read or write a bunch of msrs. All parameters are kernel addresses.
1450  *
1451  * @return number of msrs set successfully.
1452  */
1453 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1454                     struct kvm_msr_entry *entries,
1455                     int (*do_msr)(struct kvm_vcpu *vcpu,
1456                                   unsigned index, u64 *data))
1457 {
1458         int i, idx;
1459
1460         vcpu_load(vcpu);
1461
1462         idx = srcu_read_lock(&vcpu->kvm->srcu);
1463         for (i = 0; i < msrs->nmsrs; ++i)
1464                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1465                         break;
1466         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1467
1468         vcpu_put(vcpu);
1469
1470         return i;
1471 }
1472
1473 /*
1474  * Read or write a bunch of msrs. Parameters are user addresses.
1475  *
1476  * @return number of msrs set successfully.
1477  */
1478 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1479                   int (*do_msr)(struct kvm_vcpu *vcpu,
1480                                 unsigned index, u64 *data),
1481                   int writeback)
1482 {
1483         struct kvm_msrs msrs;
1484         struct kvm_msr_entry *entries;
1485         int r, n;
1486         unsigned size;
1487
1488         r = -EFAULT;
1489         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1490                 goto out;
1491
1492         r = -E2BIG;
1493         if (msrs.nmsrs >= MAX_IO_MSRS)
1494                 goto out;
1495
1496         r = -ENOMEM;
1497         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1498         entries = vmalloc(size);
1499         if (!entries)
1500                 goto out;
1501
1502         r = -EFAULT;
1503         if (copy_from_user(entries, user_msrs->entries, size))
1504                 goto out_free;
1505
1506         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1507         if (r < 0)
1508                 goto out_free;
1509
1510         r = -EFAULT;
1511         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1512                 goto out_free;
1513
1514         r = n;
1515
1516 out_free:
1517         vfree(entries);
1518 out:
1519         return r;
1520 }
1521
1522 int kvm_dev_ioctl_check_extension(long ext)
1523 {
1524         int r;
1525
1526         switch (ext) {
1527         case KVM_CAP_IRQCHIP:
1528         case KVM_CAP_HLT:
1529         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1530         case KVM_CAP_SET_TSS_ADDR:
1531         case KVM_CAP_EXT_CPUID:
1532         case KVM_CAP_CLOCKSOURCE:
1533         case KVM_CAP_PIT:
1534         case KVM_CAP_NOP_IO_DELAY:
1535         case KVM_CAP_MP_STATE:
1536         case KVM_CAP_SYNC_MMU:
1537         case KVM_CAP_REINJECT_CONTROL:
1538         case KVM_CAP_IRQ_INJECT_STATUS:
1539         case KVM_CAP_ASSIGN_DEV_IRQ:
1540         case KVM_CAP_IRQFD:
1541         case KVM_CAP_IOEVENTFD:
1542         case KVM_CAP_PIT2:
1543         case KVM_CAP_PIT_STATE2:
1544         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1545         case KVM_CAP_XEN_HVM:
1546         case KVM_CAP_ADJUST_CLOCK:
1547         case KVM_CAP_VCPU_EVENTS:
1548         case KVM_CAP_HYPERV:
1549         case KVM_CAP_HYPERV_VAPIC:
1550         case KVM_CAP_HYPERV_SPIN:
1551         case KVM_CAP_PCI_SEGMENT:
1552         case KVM_CAP_DEBUGREGS:
1553         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1554                 r = 1;
1555                 break;
1556         case KVM_CAP_COALESCED_MMIO:
1557                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1558                 break;
1559         case KVM_CAP_VAPIC:
1560                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1561                 break;
1562         case KVM_CAP_NR_VCPUS:
1563                 r = KVM_MAX_VCPUS;
1564                 break;
1565         case KVM_CAP_NR_MEMSLOTS:
1566                 r = KVM_MEMORY_SLOTS;
1567                 break;
1568         case KVM_CAP_PV_MMU:    /* obsolete */
1569                 r = 0;
1570                 break;
1571         case KVM_CAP_IOMMU:
1572                 r = iommu_found();
1573                 break;
1574         case KVM_CAP_MCE:
1575                 r = KVM_MAX_MCE_BANKS;
1576                 break;
1577         default:
1578                 r = 0;
1579                 break;
1580         }
1581         return r;
1582
1583 }
1584
1585 long kvm_arch_dev_ioctl(struct file *filp,
1586                         unsigned int ioctl, unsigned long arg)
1587 {
1588         void __user *argp = (void __user *)arg;
1589         long r;
1590
1591         switch (ioctl) {
1592         case KVM_GET_MSR_INDEX_LIST: {
1593                 struct kvm_msr_list __user *user_msr_list = argp;
1594                 struct kvm_msr_list msr_list;
1595                 unsigned n;
1596
1597                 r = -EFAULT;
1598                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1599                         goto out;
1600                 n = msr_list.nmsrs;
1601                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1602                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1603                         goto out;
1604                 r = -E2BIG;
1605                 if (n < msr_list.nmsrs)
1606                         goto out;
1607                 r = -EFAULT;
1608                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1609                                  num_msrs_to_save * sizeof(u32)))
1610                         goto out;
1611                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1612                                  &emulated_msrs,
1613                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1614                         goto out;
1615                 r = 0;
1616                 break;
1617         }
1618         case KVM_GET_SUPPORTED_CPUID: {
1619                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1620                 struct kvm_cpuid2 cpuid;
1621
1622                 r = -EFAULT;
1623                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1624                         goto out;
1625                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1626                                                       cpuid_arg->entries);
1627                 if (r)
1628                         goto out;
1629
1630                 r = -EFAULT;
1631                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1632                         goto out;
1633                 r = 0;
1634                 break;
1635         }
1636         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1637                 u64 mce_cap;
1638
1639                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1640                 r = -EFAULT;
1641                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1642                         goto out;
1643                 r = 0;
1644                 break;
1645         }
1646         default:
1647                 r = -EINVAL;
1648         }
1649 out:
1650         return r;
1651 }
1652
1653 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1654 {
1655         kvm_x86_ops->vcpu_load(vcpu, cpu);
1656         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1657                 unsigned long khz = cpufreq_quick_get(cpu);
1658                 if (!khz)
1659                         khz = tsc_khz;
1660                 per_cpu(cpu_tsc_khz, cpu) = khz;
1661         }
1662         kvm_request_guest_time_update(vcpu);
1663 }
1664
1665 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1666 {
1667         kvm_put_guest_fpu(vcpu);
1668         kvm_x86_ops->vcpu_put(vcpu);
1669 }
1670
1671 static int is_efer_nx(void)
1672 {
1673         unsigned long long efer = 0;
1674
1675         rdmsrl_safe(MSR_EFER, &efer);
1676         return efer & EFER_NX;
1677 }
1678
1679 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1680 {
1681         int i;
1682         struct kvm_cpuid_entry2 *e, *entry;
1683
1684         entry = NULL;
1685         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1686                 e = &vcpu->arch.cpuid_entries[i];
1687                 if (e->function == 0x80000001) {
1688                         entry = e;
1689                         break;
1690                 }
1691         }
1692         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1693                 entry->edx &= ~(1 << 20);
1694                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1695         }
1696 }
1697
1698 /* when an old userspace process fills a new kernel module */
1699 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1700                                     struct kvm_cpuid *cpuid,
1701                                     struct kvm_cpuid_entry __user *entries)
1702 {
1703         int r, i;
1704         struct kvm_cpuid_entry *cpuid_entries;
1705
1706         r = -E2BIG;
1707         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1708                 goto out;
1709         r = -ENOMEM;
1710         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1711         if (!cpuid_entries)
1712                 goto out;
1713         r = -EFAULT;
1714         if (copy_from_user(cpuid_entries, entries,
1715                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1716                 goto out_free;
1717         for (i = 0; i < cpuid->nent; i++) {
1718                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1719                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1720                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1721                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1722                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1723                 vcpu->arch.cpuid_entries[i].index = 0;
1724                 vcpu->arch.cpuid_entries[i].flags = 0;
1725                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1726                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1727                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1728         }
1729         vcpu->arch.cpuid_nent = cpuid->nent;
1730         cpuid_fix_nx_cap(vcpu);
1731         r = 0;
1732         kvm_apic_set_version(vcpu);
1733         kvm_x86_ops->cpuid_update(vcpu);
1734
1735 out_free:
1736         vfree(cpuid_entries);
1737 out:
1738         return r;
1739 }
1740
1741 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1742                                      struct kvm_cpuid2 *cpuid,
1743                                      struct kvm_cpuid_entry2 __user *entries)
1744 {
1745         int r;
1746
1747         r = -E2BIG;
1748         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1749                 goto out;
1750         r = -EFAULT;
1751         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1752                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1753                 goto out;
1754         vcpu->arch.cpuid_nent = cpuid->nent;
1755         kvm_apic_set_version(vcpu);
1756         kvm_x86_ops->cpuid_update(vcpu);
1757         return 0;
1758
1759 out:
1760         return r;
1761 }
1762
1763 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1764                                      struct kvm_cpuid2 *cpuid,
1765                                      struct kvm_cpuid_entry2 __user *entries)
1766 {
1767         int r;
1768
1769         r = -E2BIG;
1770         if (cpuid->nent < vcpu->arch.cpuid_nent)
1771                 goto out;
1772         r = -EFAULT;
1773         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1774                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1775                 goto out;
1776         return 0;
1777
1778 out:
1779         cpuid->nent = vcpu->arch.cpuid_nent;
1780         return r;
1781 }
1782
1783 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1784                            u32 index)
1785 {
1786         entry->function = function;
1787         entry->index = index;
1788         cpuid_count(entry->function, entry->index,
1789                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1790         entry->flags = 0;
1791 }
1792
1793 #define F(x) bit(X86_FEATURE_##x)
1794
1795 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1796                          u32 index, int *nent, int maxnent)
1797 {
1798         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1799 #ifdef CONFIG_X86_64
1800         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1801                                 ? F(GBPAGES) : 0;
1802         unsigned f_lm = F(LM);
1803 #else
1804         unsigned f_gbpages = 0;
1805         unsigned f_lm = 0;
1806 #endif
1807         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1808
1809         /* cpuid 1.edx */
1810         const u32 kvm_supported_word0_x86_features =
1811                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1812                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1813                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1814                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1815                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1816                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1817                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1818                 0 /* HTT, TM, Reserved, PBE */;
1819         /* cpuid 0x80000001.edx */
1820         const u32 kvm_supported_word1_x86_features =
1821                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1822                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1823                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1824                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1825                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1826                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1827                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1828                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1829         /* cpuid 1.ecx */
1830         const u32 kvm_supported_word4_x86_features =
1831                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1832                 0 /* DS-CPL, VMX, SMX, EST */ |
1833                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1834                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1835                 0 /* Reserved, DCA */ | F(XMM4_1) |
1836                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1837                 0 /* Reserved, XSAVE, OSXSAVE */;
1838         /* cpuid 0x80000001.ecx */
1839         const u32 kvm_supported_word6_x86_features =
1840                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1841                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1842                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1843                 0 /* SKINIT */ | 0 /* WDT */;
1844
1845         /* all calls to cpuid_count() should be made on the same cpu */
1846         get_cpu();
1847         do_cpuid_1_ent(entry, function, index);
1848         ++*nent;
1849
1850         switch (function) {
1851         case 0:
1852                 entry->eax = min(entry->eax, (u32)0xb);
1853                 break;
1854         case 1:
1855                 entry->edx &= kvm_supported_word0_x86_features;
1856                 entry->ecx &= kvm_supported_word4_x86_features;
1857                 /* we support x2apic emulation even if host does not support
1858                  * it since we emulate x2apic in software */
1859                 entry->ecx |= F(X2APIC);
1860                 break;
1861         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1862          * may return different values. This forces us to get_cpu() before
1863          * issuing the first command, and also to emulate this annoying behavior
1864          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1865         case 2: {
1866                 int t, times = entry->eax & 0xff;
1867
1868                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1869                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1870                 for (t = 1; t < times && *nent < maxnent; ++t) {
1871                         do_cpuid_1_ent(&entry[t], function, 0);
1872                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1873                         ++*nent;
1874                 }
1875                 break;
1876         }
1877         /* function 4 and 0xb have additional index. */
1878         case 4: {
1879                 int i, cache_type;
1880
1881                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1882                 /* read more entries until cache_type is zero */
1883                 for (i = 1; *nent < maxnent; ++i) {
1884                         cache_type = entry[i - 1].eax & 0x1f;
1885                         if (!cache_type)
1886                                 break;
1887                         do_cpuid_1_ent(&entry[i], function, i);
1888                         entry[i].flags |=
1889                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1890                         ++*nent;
1891                 }
1892                 break;
1893         }
1894         case 0xb: {
1895                 int i, level_type;
1896
1897                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1898                 /* read more entries until level_type is zero */
1899                 for (i = 1; *nent < maxnent; ++i) {
1900                         level_type = entry[i - 1].ecx & 0xff00;
1901                         if (!level_type)
1902                                 break;
1903                         do_cpuid_1_ent(&entry[i], function, i);
1904                         entry[i].flags |=
1905                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1906                         ++*nent;
1907                 }
1908                 break;
1909         }
1910         case 0x80000000:
1911                 entry->eax = min(entry->eax, 0x8000001a);
1912                 break;
1913         case 0x80000001:
1914                 entry->edx &= kvm_supported_word1_x86_features;
1915                 entry->ecx &= kvm_supported_word6_x86_features;
1916                 break;
1917         }
1918         put_cpu();
1919 }
1920
1921 #undef F
1922
1923 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1924                                      struct kvm_cpuid_entry2 __user *entries)
1925 {
1926         struct kvm_cpuid_entry2 *cpuid_entries;
1927         int limit, nent = 0, r = -E2BIG;
1928         u32 func;
1929
1930         if (cpuid->nent < 1)
1931                 goto out;
1932         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1933                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1934         r = -ENOMEM;
1935         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1936         if (!cpuid_entries)
1937                 goto out;
1938
1939         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1940         limit = cpuid_entries[0].eax;
1941         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1942                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1943                              &nent, cpuid->nent);
1944         r = -E2BIG;
1945         if (nent >= cpuid->nent)
1946                 goto out_free;
1947
1948         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1949         limit = cpuid_entries[nent - 1].eax;
1950         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1951                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1952                              &nent, cpuid->nent);
1953         r = -E2BIG;
1954         if (nent >= cpuid->nent)
1955                 goto out_free;
1956
1957         r = -EFAULT;
1958         if (copy_to_user(entries, cpuid_entries,
1959                          nent * sizeof(struct kvm_cpuid_entry2)))
1960                 goto out_free;
1961         cpuid->nent = nent;
1962         r = 0;
1963
1964 out_free:
1965         vfree(cpuid_entries);
1966 out:
1967         return r;
1968 }
1969
1970 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1971                                     struct kvm_lapic_state *s)
1972 {
1973         vcpu_load(vcpu);
1974         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1975         vcpu_put(vcpu);
1976
1977         return 0;
1978 }
1979
1980 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1981                                     struct kvm_lapic_state *s)
1982 {
1983         vcpu_load(vcpu);
1984         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1985         kvm_apic_post_state_restore(vcpu);
1986         update_cr8_intercept(vcpu);
1987         vcpu_put(vcpu);
1988
1989         return 0;
1990 }
1991
1992 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1993                                     struct kvm_interrupt *irq)
1994 {
1995         if (irq->irq < 0 || irq->irq >= 256)
1996                 return -EINVAL;
1997         if (irqchip_in_kernel(vcpu->kvm))
1998                 return -ENXIO;
1999         vcpu_load(vcpu);
2000
2001         kvm_queue_interrupt(vcpu, irq->irq, false);
2002
2003         vcpu_put(vcpu);
2004
2005         return 0;
2006 }
2007
2008 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2009 {
2010         vcpu_load(vcpu);
2011         kvm_inject_nmi(vcpu);
2012         vcpu_put(vcpu);
2013
2014         return 0;
2015 }
2016
2017 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2018                                            struct kvm_tpr_access_ctl *tac)
2019 {
2020         if (tac->flags)
2021                 return -EINVAL;
2022         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2023         return 0;
2024 }
2025
2026 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2027                                         u64 mcg_cap)
2028 {
2029         int r;
2030         unsigned bank_num = mcg_cap & 0xff, bank;
2031
2032         r = -EINVAL;
2033         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2034                 goto out;
2035         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2036                 goto out;
2037         r = 0;
2038         vcpu->arch.mcg_cap = mcg_cap;
2039         /* Init IA32_MCG_CTL to all 1s */
2040         if (mcg_cap & MCG_CTL_P)
2041                 vcpu->arch.mcg_ctl = ~(u64)0;
2042         /* Init IA32_MCi_CTL to all 1s */
2043         for (bank = 0; bank < bank_num; bank++)
2044                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2045 out:
2046         return r;
2047 }
2048
2049 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2050                                       struct kvm_x86_mce *mce)
2051 {
2052         u64 mcg_cap = vcpu->arch.mcg_cap;
2053         unsigned bank_num = mcg_cap & 0xff;
2054         u64 *banks = vcpu->arch.mce_banks;
2055
2056         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2057                 return -EINVAL;
2058         /*
2059          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2060          * reporting is disabled
2061          */
2062         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2063             vcpu->arch.mcg_ctl != ~(u64)0)
2064                 return 0;
2065         banks += 4 * mce->bank;
2066         /*
2067          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2068          * reporting is disabled for the bank
2069          */
2070         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2071                 return 0;
2072         if (mce->status & MCI_STATUS_UC) {
2073                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2074                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2075                         printk(KERN_DEBUG "kvm: set_mce: "
2076                                "injects mce exception while "
2077                                "previous one is in progress!\n");
2078                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2079                         return 0;
2080                 }
2081                 if (banks[1] & MCI_STATUS_VAL)
2082                         mce->status |= MCI_STATUS_OVER;
2083                 banks[2] = mce->addr;
2084                 banks[3] = mce->misc;
2085                 vcpu->arch.mcg_status = mce->mcg_status;
2086                 banks[1] = mce->status;
2087                 kvm_queue_exception(vcpu, MC_VECTOR);
2088         } else if (!(banks[1] & MCI_STATUS_VAL)
2089                    || !(banks[1] & MCI_STATUS_UC)) {
2090                 if (banks[1] & MCI_STATUS_VAL)
2091                         mce->status |= MCI_STATUS_OVER;
2092                 banks[2] = mce->addr;
2093                 banks[3] = mce->misc;
2094                 banks[1] = mce->status;
2095         } else
2096                 banks[1] |= MCI_STATUS_OVER;
2097         return 0;
2098 }
2099
2100 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2101                                                struct kvm_vcpu_events *events)
2102 {
2103         vcpu_load(vcpu);
2104
2105         events->exception.injected =
2106                 vcpu->arch.exception.pending &&
2107                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2108         events->exception.nr = vcpu->arch.exception.nr;
2109         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2110         events->exception.error_code = vcpu->arch.exception.error_code;
2111
2112         events->interrupt.injected =
2113                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2114         events->interrupt.nr = vcpu->arch.interrupt.nr;
2115         events->interrupt.soft = 0;
2116         events->interrupt.shadow =
2117                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2118                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2119
2120         events->nmi.injected = vcpu->arch.nmi_injected;
2121         events->nmi.pending = vcpu->arch.nmi_pending;
2122         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2123
2124         events->sipi_vector = vcpu->arch.sipi_vector;
2125
2126         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2127                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2128                          | KVM_VCPUEVENT_VALID_SHADOW);
2129
2130         vcpu_put(vcpu);
2131 }
2132
2133 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2134                                               struct kvm_vcpu_events *events)
2135 {
2136         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2137                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2138                               | KVM_VCPUEVENT_VALID_SHADOW))
2139                 return -EINVAL;
2140
2141         vcpu_load(vcpu);
2142
2143         vcpu->arch.exception.pending = events->exception.injected;
2144         vcpu->arch.exception.nr = events->exception.nr;
2145         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2146         vcpu->arch.exception.error_code = events->exception.error_code;
2147
2148         vcpu->arch.interrupt.pending = events->interrupt.injected;
2149         vcpu->arch.interrupt.nr = events->interrupt.nr;
2150         vcpu->arch.interrupt.soft = events->interrupt.soft;
2151         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2152                 kvm_pic_clear_isr_ack(vcpu->kvm);
2153         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2154                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2155                                                   events->interrupt.shadow);
2156
2157         vcpu->arch.nmi_injected = events->nmi.injected;
2158         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2159                 vcpu->arch.nmi_pending = events->nmi.pending;
2160         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2161
2162         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2163                 vcpu->arch.sipi_vector = events->sipi_vector;
2164
2165         vcpu_put(vcpu);
2166
2167         return 0;
2168 }
2169
2170 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2171                                              struct kvm_debugregs *dbgregs)
2172 {
2173         vcpu_load(vcpu);
2174
2175         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2176         dbgregs->dr6 = vcpu->arch.dr6;
2177         dbgregs->dr7 = vcpu->arch.dr7;
2178         dbgregs->flags = 0;
2179
2180         vcpu_put(vcpu);
2181 }
2182
2183 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2184                                             struct kvm_debugregs *dbgregs)
2185 {
2186         if (dbgregs->flags)
2187                 return -EINVAL;
2188
2189         vcpu_load(vcpu);
2190
2191         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2192         vcpu->arch.dr6 = dbgregs->dr6;
2193         vcpu->arch.dr7 = dbgregs->dr7;
2194
2195         vcpu_put(vcpu);
2196
2197         return 0;
2198 }
2199
2200 long kvm_arch_vcpu_ioctl(struct file *filp,
2201                          unsigned int ioctl, unsigned long arg)
2202 {
2203         struct kvm_vcpu *vcpu = filp->private_data;
2204         void __user *argp = (void __user *)arg;
2205         int r;
2206         struct kvm_lapic_state *lapic = NULL;
2207
2208         switch (ioctl) {
2209         case KVM_GET_LAPIC: {
2210                 r = -EINVAL;
2211                 if (!vcpu->arch.apic)
2212                         goto out;
2213                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2214
2215                 r = -ENOMEM;
2216                 if (!lapic)
2217                         goto out;
2218                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2219                 if (r)
2220                         goto out;
2221                 r = -EFAULT;
2222                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2223                         goto out;
2224                 r = 0;
2225                 break;
2226         }
2227         case KVM_SET_LAPIC: {
2228                 r = -EINVAL;
2229                 if (!vcpu->arch.apic)
2230                         goto out;
2231                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2232                 r = -ENOMEM;
2233                 if (!lapic)
2234                         goto out;
2235                 r = -EFAULT;
2236                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2237                         goto out;
2238                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2239                 if (r)
2240                         goto out;
2241                 r = 0;
2242                 break;
2243         }
2244         case KVM_INTERRUPT: {
2245                 struct kvm_interrupt irq;
2246
2247                 r = -EFAULT;
2248                 if (copy_from_user(&irq, argp, sizeof irq))
2249                         goto out;
2250                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2251                 if (r)
2252                         goto out;
2253                 r = 0;
2254                 break;
2255         }
2256         case KVM_NMI: {
2257                 r = kvm_vcpu_ioctl_nmi(vcpu);
2258                 if (r)
2259                         goto out;
2260                 r = 0;
2261                 break;
2262         }
2263         case KVM_SET_CPUID: {
2264                 struct kvm_cpuid __user *cpuid_arg = argp;
2265                 struct kvm_cpuid cpuid;
2266
2267                 r = -EFAULT;
2268                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2269                         goto out;
2270                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2271                 if (r)
2272                         goto out;
2273                 break;
2274         }
2275         case KVM_SET_CPUID2: {
2276                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2277                 struct kvm_cpuid2 cpuid;
2278
2279                 r = -EFAULT;
2280                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2281                         goto out;
2282                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2283                                               cpuid_arg->entries);
2284                 if (r)
2285                         goto out;
2286                 break;
2287         }
2288         case KVM_GET_CPUID2: {
2289                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2290                 struct kvm_cpuid2 cpuid;
2291
2292                 r = -EFAULT;
2293                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2294                         goto out;
2295                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2296                                               cpuid_arg->entries);
2297                 if (r)
2298                         goto out;
2299                 r = -EFAULT;
2300                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2301                         goto out;
2302                 r = 0;
2303                 break;
2304         }
2305         case KVM_GET_MSRS:
2306                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2307                 break;
2308         case KVM_SET_MSRS:
2309                 r = msr_io(vcpu, argp, do_set_msr, 0);
2310                 break;
2311         case KVM_TPR_ACCESS_REPORTING: {
2312                 struct kvm_tpr_access_ctl tac;
2313
2314                 r = -EFAULT;
2315                 if (copy_from_user(&tac, argp, sizeof tac))
2316                         goto out;
2317                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2318                 if (r)
2319                         goto out;
2320                 r = -EFAULT;
2321                 if (copy_to_user(argp, &tac, sizeof tac))
2322                         goto out;
2323                 r = 0;
2324                 break;
2325         };
2326         case KVM_SET_VAPIC_ADDR: {
2327                 struct kvm_vapic_addr va;
2328
2329                 r = -EINVAL;
2330                 if (!irqchip_in_kernel(vcpu->kvm))
2331                         goto out;
2332                 r = -EFAULT;
2333                 if (copy_from_user(&va, argp, sizeof va))
2334                         goto out;
2335                 r = 0;
2336                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2337                 break;
2338         }
2339         case KVM_X86_SETUP_MCE: {
2340                 u64 mcg_cap;
2341
2342                 r = -EFAULT;
2343                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2344                         goto out;
2345                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2346                 break;
2347         }
2348         case KVM_X86_SET_MCE: {
2349                 struct kvm_x86_mce mce;
2350
2351                 r = -EFAULT;
2352                 if (copy_from_user(&mce, argp, sizeof mce))
2353                         goto out;
2354                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2355                 break;
2356         }
2357         case KVM_GET_VCPU_EVENTS: {
2358                 struct kvm_vcpu_events events;
2359
2360                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2361
2362                 r = -EFAULT;
2363                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2364                         break;
2365                 r = 0;
2366                 break;
2367         }
2368         case KVM_SET_VCPU_EVENTS: {
2369                 struct kvm_vcpu_events events;
2370
2371                 r = -EFAULT;
2372                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2373                         break;
2374
2375                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2376                 break;
2377         }
2378         case KVM_GET_DEBUGREGS: {
2379                 struct kvm_debugregs dbgregs;
2380
2381                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2382
2383                 r = -EFAULT;
2384                 if (copy_to_user(argp, &dbgregs,
2385                                  sizeof(struct kvm_debugregs)))
2386                         break;
2387                 r = 0;
2388                 break;
2389         }
2390         case KVM_SET_DEBUGREGS: {
2391                 struct kvm_debugregs dbgregs;
2392
2393                 r = -EFAULT;
2394                 if (copy_from_user(&dbgregs, argp,
2395                                    sizeof(struct kvm_debugregs)))
2396                         break;
2397
2398                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2399                 break;
2400         }
2401         default:
2402                 r = -EINVAL;
2403         }
2404 out:
2405         kfree(lapic);
2406         return r;
2407 }
2408
2409 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2410 {
2411         int ret;
2412
2413         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2414                 return -1;
2415         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2416         return ret;
2417 }
2418
2419 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2420                                               u64 ident_addr)
2421 {
2422         kvm->arch.ept_identity_map_addr = ident_addr;
2423         return 0;
2424 }
2425
2426 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2427                                           u32 kvm_nr_mmu_pages)
2428 {
2429         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2430                 return -EINVAL;
2431
2432         mutex_lock(&kvm->slots_lock);
2433         spin_lock(&kvm->mmu_lock);
2434
2435         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2436         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2437
2438         spin_unlock(&kvm->mmu_lock);
2439         mutex_unlock(&kvm->slots_lock);
2440         return 0;
2441 }
2442
2443 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2444 {
2445         return kvm->arch.n_alloc_mmu_pages;
2446 }
2447
2448 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2449 {
2450         int i;
2451         struct kvm_mem_alias *alias;
2452         struct kvm_mem_aliases *aliases;
2453
2454         aliases = rcu_dereference(kvm->arch.aliases);
2455
2456         for (i = 0; i < aliases->naliases; ++i) {
2457                 alias = &aliases->aliases[i];
2458                 if (alias->flags & KVM_ALIAS_INVALID)
2459                         continue;
2460                 if (gfn >= alias->base_gfn
2461                     && gfn < alias->base_gfn + alias->npages)
2462                         return alias->target_gfn + gfn - alias->base_gfn;
2463         }
2464         return gfn;
2465 }
2466
2467 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2468 {
2469         int i;
2470         struct kvm_mem_alias *alias;
2471         struct kvm_mem_aliases *aliases;
2472
2473         aliases = rcu_dereference(kvm->arch.aliases);
2474
2475         for (i = 0; i < aliases->naliases; ++i) {
2476                 alias = &aliases->aliases[i];
2477                 if (gfn >= alias->base_gfn
2478                     && gfn < alias->base_gfn + alias->npages)
2479                         return alias->target_gfn + gfn - alias->base_gfn;
2480         }
2481         return gfn;
2482 }
2483
2484 /*
2485  * Set a new alias region.  Aliases map a portion of physical memory into
2486  * another portion.  This is useful for memory windows, for example the PC
2487  * VGA region.
2488  */
2489 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2490                                          struct kvm_memory_alias *alias)
2491 {
2492         int r, n;
2493         struct kvm_mem_alias *p;
2494         struct kvm_mem_aliases *aliases, *old_aliases;
2495
2496         r = -EINVAL;
2497         /* General sanity checks */
2498         if (alias->memory_size & (PAGE_SIZE - 1))
2499                 goto out;
2500         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2501                 goto out;
2502         if (alias->slot >= KVM_ALIAS_SLOTS)
2503                 goto out;
2504         if (alias->guest_phys_addr + alias->memory_size
2505             < alias->guest_phys_addr)
2506                 goto out;
2507         if (alias->target_phys_addr + alias->memory_size
2508             < alias->target_phys_addr)
2509                 goto out;
2510
2511         r = -ENOMEM;
2512         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2513         if (!aliases)
2514                 goto out;
2515
2516         mutex_lock(&kvm->slots_lock);
2517
2518         /* invalidate any gfn reference in case of deletion/shrinking */
2519         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2520         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2521         old_aliases = kvm->arch.aliases;
2522         rcu_assign_pointer(kvm->arch.aliases, aliases);
2523         synchronize_srcu_expedited(&kvm->srcu);
2524         kvm_mmu_zap_all(kvm);
2525         kfree(old_aliases);
2526
2527         r = -ENOMEM;
2528         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2529         if (!aliases)
2530                 goto out_unlock;
2531
2532         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2533
2534         p = &aliases->aliases[alias->slot];
2535         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2536         p->npages = alias->memory_size >> PAGE_SHIFT;
2537         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2538         p->flags &= ~(KVM_ALIAS_INVALID);
2539
2540         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2541                 if (aliases->aliases[n - 1].npages)
2542                         break;
2543         aliases->naliases = n;
2544
2545         old_aliases = kvm->arch.aliases;
2546         rcu_assign_pointer(kvm->arch.aliases, aliases);
2547         synchronize_srcu_expedited(&kvm->srcu);
2548         kfree(old_aliases);
2549         r = 0;
2550
2551 out_unlock:
2552         mutex_unlock(&kvm->slots_lock);
2553 out:
2554         return r;
2555 }
2556
2557 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2558 {
2559         int r;
2560
2561         r = 0;
2562         switch (chip->chip_id) {
2563         case KVM_IRQCHIP_PIC_MASTER:
2564                 memcpy(&chip->chip.pic,
2565                         &pic_irqchip(kvm)->pics[0],
2566                         sizeof(struct kvm_pic_state));
2567                 break;
2568         case KVM_IRQCHIP_PIC_SLAVE:
2569                 memcpy(&chip->chip.pic,
2570                         &pic_irqchip(kvm)->pics[1],
2571                         sizeof(struct kvm_pic_state));
2572                 break;
2573         case KVM_IRQCHIP_IOAPIC:
2574                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2575                 break;
2576         default:
2577                 r = -EINVAL;
2578                 break;
2579         }
2580         return r;
2581 }
2582
2583 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2584 {
2585         int r;
2586
2587         r = 0;
2588         switch (chip->chip_id) {
2589         case KVM_IRQCHIP_PIC_MASTER:
2590                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2591                 memcpy(&pic_irqchip(kvm)->pics[0],
2592                         &chip->chip.pic,
2593                         sizeof(struct kvm_pic_state));
2594                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2595                 break;
2596         case KVM_IRQCHIP_PIC_SLAVE:
2597                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2598                 memcpy(&pic_irqchip(kvm)->pics[1],
2599                         &chip->chip.pic,
2600                         sizeof(struct kvm_pic_state));
2601                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2602                 break;
2603         case KVM_IRQCHIP_IOAPIC:
2604                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2605                 break;
2606         default:
2607                 r = -EINVAL;
2608                 break;
2609         }
2610         kvm_pic_update_irq(pic_irqchip(kvm));
2611         return r;
2612 }
2613
2614 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2615 {
2616         int r = 0;
2617
2618         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2619         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2620         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2621         return r;
2622 }
2623
2624 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2625 {
2626         int r = 0;
2627
2628         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2629         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2630         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2631         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2632         return r;
2633 }
2634
2635 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2636 {
2637         int r = 0;
2638
2639         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2640         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2641                 sizeof(ps->channels));
2642         ps->flags = kvm->arch.vpit->pit_state.flags;
2643         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2644         return r;
2645 }
2646
2647 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2648 {
2649         int r = 0, start = 0;
2650         u32 prev_legacy, cur_legacy;
2651         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2652         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2653         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2654         if (!prev_legacy && cur_legacy)
2655                 start = 1;
2656         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2657                sizeof(kvm->arch.vpit->pit_state.channels));
2658         kvm->arch.vpit->pit_state.flags = ps->flags;
2659         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2660         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2661         return r;
2662 }
2663
2664 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2665                                  struct kvm_reinject_control *control)
2666 {
2667         if (!kvm->arch.vpit)
2668                 return -ENXIO;
2669         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2670         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2671         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2672         return 0;
2673 }
2674
2675 /*
2676  * Get (and clear) the dirty memory log for a memory slot.
2677  */
2678 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2679                                       struct kvm_dirty_log *log)
2680 {
2681         int r, i;
2682         struct kvm_memory_slot *memslot;
2683         unsigned long n;
2684         unsigned long is_dirty = 0;
2685         unsigned long *dirty_bitmap = NULL;
2686
2687         mutex_lock(&kvm->slots_lock);
2688
2689         r = -EINVAL;
2690         if (log->slot >= KVM_MEMORY_SLOTS)
2691                 goto out;
2692
2693         memslot = &kvm->memslots->memslots[log->slot];
2694         r = -ENOENT;
2695         if (!memslot->dirty_bitmap)
2696                 goto out;
2697
2698         n = kvm_dirty_bitmap_bytes(memslot);
2699
2700         r = -ENOMEM;
2701         dirty_bitmap = vmalloc(n);
2702         if (!dirty_bitmap)
2703                 goto out;
2704         memset(dirty_bitmap, 0, n);
2705
2706         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2707                 is_dirty = memslot->dirty_bitmap[i];
2708
2709         /* If nothing is dirty, don't bother messing with page tables. */
2710         if (is_dirty) {
2711                 struct kvm_memslots *slots, *old_slots;
2712
2713                 spin_lock(&kvm->mmu_lock);
2714                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2715                 spin_unlock(&kvm->mmu_lock);
2716
2717                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2718                 if (!slots)
2719                         goto out_free;
2720
2721                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2722                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2723
2724                 old_slots = kvm->memslots;
2725                 rcu_assign_pointer(kvm->memslots, slots);
2726                 synchronize_srcu_expedited(&kvm->srcu);
2727                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2728                 kfree(old_slots);
2729         }
2730
2731         r = 0;
2732         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2733                 r = -EFAULT;
2734 out_free:
2735         vfree(dirty_bitmap);
2736 out:
2737         mutex_unlock(&kvm->slots_lock);
2738         return r;
2739 }
2740
2741 long kvm_arch_vm_ioctl(struct file *filp,
2742                        unsigned int ioctl, unsigned long arg)
2743 {
2744         struct kvm *kvm = filp->private_data;
2745         void __user *argp = (void __user *)arg;
2746         int r = -ENOTTY;
2747         /*
2748          * This union makes it completely explicit to gcc-3.x
2749          * that these two variables' stack usage should be
2750          * combined, not added together.
2751          */
2752         union {
2753                 struct kvm_pit_state ps;
2754                 struct kvm_pit_state2 ps2;
2755                 struct kvm_memory_alias alias;
2756                 struct kvm_pit_config pit_config;
2757         } u;
2758
2759         switch (ioctl) {
2760         case KVM_SET_TSS_ADDR:
2761                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2762                 if (r < 0)
2763                         goto out;
2764                 break;
2765         case KVM_SET_IDENTITY_MAP_ADDR: {
2766                 u64 ident_addr;
2767
2768                 r = -EFAULT;
2769                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2770                         goto out;
2771                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2772                 if (r < 0)
2773                         goto out;
2774                 break;
2775         }
2776         case KVM_SET_MEMORY_REGION: {
2777                 struct kvm_memory_region kvm_mem;
2778                 struct kvm_userspace_memory_region kvm_userspace_mem;
2779
2780                 r = -EFAULT;
2781                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2782                         goto out;
2783                 kvm_userspace_mem.slot = kvm_mem.slot;
2784                 kvm_userspace_mem.flags = kvm_mem.flags;
2785                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2786                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2787                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2788                 if (r)
2789                         goto out;
2790                 break;
2791         }
2792         case KVM_SET_NR_MMU_PAGES:
2793                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2794                 if (r)
2795                         goto out;
2796                 break;
2797         case KVM_GET_NR_MMU_PAGES:
2798                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2799                 break;
2800         case KVM_SET_MEMORY_ALIAS:
2801                 r = -EFAULT;
2802                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2803                         goto out;
2804                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2805                 if (r)
2806                         goto out;
2807                 break;
2808         case KVM_CREATE_IRQCHIP: {
2809                 struct kvm_pic *vpic;
2810
2811                 mutex_lock(&kvm->lock);
2812                 r = -EEXIST;
2813                 if (kvm->arch.vpic)
2814                         goto create_irqchip_unlock;
2815                 r = -ENOMEM;
2816                 vpic = kvm_create_pic(kvm);
2817                 if (vpic) {
2818                         r = kvm_ioapic_init(kvm);
2819                         if (r) {
2820                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2821                                                           &vpic->dev);
2822                                 kfree(vpic);
2823                                 goto create_irqchip_unlock;
2824                         }
2825                 } else
2826                         goto create_irqchip_unlock;
2827                 smp_wmb();
2828                 kvm->arch.vpic = vpic;
2829                 smp_wmb();
2830                 r = kvm_setup_default_irq_routing(kvm);
2831                 if (r) {
2832                         mutex_lock(&kvm->irq_lock);
2833                         kvm_ioapic_destroy(kvm);
2834                         kvm_destroy_pic(kvm);
2835                         mutex_unlock(&kvm->irq_lock);
2836                 }
2837         create_irqchip_unlock:
2838                 mutex_unlock(&kvm->lock);
2839                 break;
2840         }
2841         case KVM_CREATE_PIT:
2842                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2843                 goto create_pit;
2844         case KVM_CREATE_PIT2:
2845                 r = -EFAULT;
2846                 if (copy_from_user(&u.pit_config, argp,
2847                                    sizeof(struct kvm_pit_config)))
2848                         goto out;
2849         create_pit:
2850                 mutex_lock(&kvm->slots_lock);
2851                 r = -EEXIST;
2852                 if (kvm->arch.vpit)
2853                         goto create_pit_unlock;
2854                 r = -ENOMEM;
2855                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2856                 if (kvm->arch.vpit)
2857                         r = 0;
2858         create_pit_unlock:
2859                 mutex_unlock(&kvm->slots_lock);
2860                 break;
2861         case KVM_IRQ_LINE_STATUS:
2862         case KVM_IRQ_LINE: {
2863                 struct kvm_irq_level irq_event;
2864
2865                 r = -EFAULT;
2866                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2867                         goto out;
2868                 if (irqchip_in_kernel(kvm)) {
2869                         __s32 status;
2870                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2871                                         irq_event.irq, irq_event.level);
2872                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2873                                 irq_event.status = status;
2874                                 if (copy_to_user(argp, &irq_event,
2875                                                         sizeof irq_event))
2876                                         goto out;
2877                         }
2878                         r = 0;
2879                 }
2880                 break;
2881         }
2882         case KVM_GET_IRQCHIP: {
2883                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2884                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2885
2886                 r = -ENOMEM;
2887                 if (!chip)
2888                         goto out;
2889                 r = -EFAULT;
2890                 if (copy_from_user(chip, argp, sizeof *chip))
2891                         goto get_irqchip_out;
2892                 r = -ENXIO;
2893                 if (!irqchip_in_kernel(kvm))
2894                         goto get_irqchip_out;
2895                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2896                 if (r)
2897                         goto get_irqchip_out;
2898                 r = -EFAULT;
2899                 if (copy_to_user(argp, chip, sizeof *chip))
2900                         goto get_irqchip_out;
2901                 r = 0;
2902         get_irqchip_out:
2903                 kfree(chip);
2904                 if (r)
2905                         goto out;
2906                 break;
2907         }
2908         case KVM_SET_IRQCHIP: {
2909                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2910                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2911
2912                 r = -ENOMEM;
2913                 if (!chip)
2914                         goto out;
2915                 r = -EFAULT;
2916                 if (copy_from_user(chip, argp, sizeof *chip))
2917                         goto set_irqchip_out;
2918                 r = -ENXIO;
2919                 if (!irqchip_in_kernel(kvm))
2920                         goto set_irqchip_out;
2921                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2922                 if (r)
2923                         goto set_irqchip_out;
2924                 r = 0;
2925         set_irqchip_out:
2926                 kfree(chip);
2927                 if (r)
2928                         goto out;
2929                 break;
2930         }
2931         case KVM_GET_PIT: {
2932                 r = -EFAULT;
2933                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2934                         goto out;
2935                 r = -ENXIO;
2936                 if (!kvm->arch.vpit)
2937                         goto out;
2938                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2939                 if (r)
2940                         goto out;
2941                 r = -EFAULT;
2942                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2943                         goto out;
2944                 r = 0;
2945                 break;
2946         }
2947         case KVM_SET_PIT: {
2948                 r = -EFAULT;
2949                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2950                         goto out;
2951                 r = -ENXIO;
2952                 if (!kvm->arch.vpit)
2953                         goto out;
2954                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2955                 if (r)
2956                         goto out;
2957                 r = 0;
2958                 break;
2959         }
2960         case KVM_GET_PIT2: {
2961                 r = -ENXIO;
2962                 if (!kvm->arch.vpit)
2963                         goto out;
2964                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2965                 if (r)
2966                         goto out;
2967                 r = -EFAULT;
2968                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2969                         goto out;
2970                 r = 0;
2971                 break;
2972         }
2973         case KVM_SET_PIT2: {
2974                 r = -EFAULT;
2975                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2976                         goto out;
2977                 r = -ENXIO;
2978                 if (!kvm->arch.vpit)
2979                         goto out;
2980                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2981                 if (r)
2982                         goto out;
2983                 r = 0;
2984                 break;
2985         }
2986         case KVM_REINJECT_CONTROL: {
2987                 struct kvm_reinject_control control;
2988                 r =  -EFAULT;
2989                 if (copy_from_user(&control, argp, sizeof(control)))
2990                         goto out;
2991                 r = kvm_vm_ioctl_reinject(kvm, &control);
2992                 if (r)
2993                         goto out;
2994                 r = 0;
2995                 break;
2996         }
2997         case KVM_XEN_HVM_CONFIG: {
2998                 r = -EFAULT;
2999                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3000                                    sizeof(struct kvm_xen_hvm_config)))
3001                         goto out;
3002                 r = -EINVAL;
3003                 if (kvm->arch.xen_hvm_config.flags)
3004                         goto out;
3005                 r = 0;
3006                 break;
3007         }
3008         case KVM_SET_CLOCK: {
3009                 struct timespec now;
3010                 struct kvm_clock_data user_ns;
3011                 u64 now_ns;
3012                 s64 delta;
3013
3014                 r = -EFAULT;
3015                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3016                         goto out;
3017
3018                 r = -EINVAL;
3019                 if (user_ns.flags)
3020                         goto out;
3021
3022                 r = 0;
3023                 ktime_get_ts(&now);
3024                 now_ns = timespec_to_ns(&now);
3025                 delta = user_ns.clock - now_ns;
3026                 kvm->arch.kvmclock_offset = delta;
3027                 break;
3028         }
3029         case KVM_GET_CLOCK: {
3030                 struct timespec now;
3031                 struct kvm_clock_data user_ns;
3032                 u64 now_ns;
3033
3034                 ktime_get_ts(&now);
3035                 now_ns = timespec_to_ns(&now);
3036                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3037                 user_ns.flags = 0;
3038
3039                 r = -EFAULT;
3040                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3041                         goto out;
3042                 r = 0;
3043                 break;
3044         }
3045
3046         default:
3047                 ;
3048         }
3049 out:
3050         return r;
3051 }
3052
3053 static void kvm_init_msr_list(void)
3054 {
3055         u32 dummy[2];
3056         unsigned i, j;
3057
3058         /* skip the first msrs in the list. KVM-specific */
3059         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3060                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3061                         continue;
3062                 if (j < i)
3063                         msrs_to_save[j] = msrs_to_save[i];
3064                 j++;
3065         }
3066         num_msrs_to_save = j;
3067 }
3068
3069 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3070                            const void *v)
3071 {
3072         if (vcpu->arch.apic &&
3073             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3074                 return 0;
3075
3076         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3077 }
3078
3079 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3080 {
3081         if (vcpu->arch.apic &&
3082             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3083                 return 0;
3084
3085         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3086 }
3087
3088 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3089 {
3090         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3091         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3092 }
3093
3094  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3095 {
3096         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3097         access |= PFERR_FETCH_MASK;
3098         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3099 }
3100
3101 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3102 {
3103         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3104         access |= PFERR_WRITE_MASK;
3105         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3106 }
3107
3108 /* uses this to access any guest's mapped memory without checking CPL */
3109 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3110 {
3111         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3112 }
3113
3114 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3115                                       struct kvm_vcpu *vcpu, u32 access,
3116                                       u32 *error)
3117 {
3118         void *data = val;
3119         int r = X86EMUL_CONTINUE;
3120
3121         while (bytes) {
3122                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3123                 unsigned offset = addr & (PAGE_SIZE-1);
3124                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3125                 int ret;
3126
3127                 if (gpa == UNMAPPED_GVA) {
3128                         r = X86EMUL_PROPAGATE_FAULT;
3129                         goto out;
3130                 }
3131                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3132                 if (ret < 0) {
3133                         r = X86EMUL_UNHANDLEABLE;
3134                         goto out;
3135                 }
3136
3137                 bytes -= toread;
3138                 data += toread;
3139                 addr += toread;
3140         }
3141 out:
3142         return r;
3143 }
3144
3145 /* used for instruction fetching */
3146 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3147                                 struct kvm_vcpu *vcpu, u32 *error)
3148 {
3149         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3150         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3151                                           access | PFERR_FETCH_MASK, error);
3152 }
3153
3154 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3155                                struct kvm_vcpu *vcpu, u32 *error)
3156 {
3157         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3158         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3159                                           error);
3160 }
3161
3162 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3163                                struct kvm_vcpu *vcpu, u32 *error)
3164 {
3165         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3166 }
3167
3168 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3169                                 struct kvm_vcpu *vcpu, u32 *error)
3170 {
3171         void *data = val;
3172         int r = X86EMUL_CONTINUE;
3173
3174         while (bytes) {
3175                 gpa_t gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error);
3176                 unsigned offset = addr & (PAGE_SIZE-1);
3177                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3178                 int ret;
3179
3180                 if (gpa == UNMAPPED_GVA) {
3181                         r = X86EMUL_PROPAGATE_FAULT;
3182                         goto out;
3183                 }
3184                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3185                 if (ret < 0) {
3186                         r = X86EMUL_UNHANDLEABLE;
3187                         goto out;
3188                 }
3189
3190                 bytes -= towrite;
3191                 data += towrite;
3192                 addr += towrite;
3193         }
3194 out:
3195         return r;
3196 }
3197
3198
3199 static int emulator_read_emulated(unsigned long addr,
3200                                   void *val,
3201                                   unsigned int bytes,
3202                                   struct kvm_vcpu *vcpu)
3203 {
3204         gpa_t                 gpa;
3205         u32 error_code;
3206
3207         if (vcpu->mmio_read_completed) {
3208                 memcpy(val, vcpu->mmio_data, bytes);
3209                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3210                                vcpu->mmio_phys_addr, *(u64 *)val);
3211                 vcpu->mmio_read_completed = 0;
3212                 return X86EMUL_CONTINUE;
3213         }
3214
3215         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, &error_code);
3216
3217         if (gpa == UNMAPPED_GVA) {
3218                 kvm_inject_page_fault(vcpu, addr, error_code);
3219                 return X86EMUL_PROPAGATE_FAULT;
3220         }
3221
3222         /* For APIC access vmexit */
3223         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3224                 goto mmio;
3225
3226         if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3227                                 == X86EMUL_CONTINUE)
3228                 return X86EMUL_CONTINUE;
3229
3230 mmio:
3231         /*
3232          * Is this MMIO handled locally?
3233          */
3234         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3235                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3236                 return X86EMUL_CONTINUE;
3237         }
3238
3239         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3240
3241         vcpu->mmio_needed = 1;
3242         vcpu->mmio_phys_addr = gpa;
3243         vcpu->mmio_size = bytes;
3244         vcpu->mmio_is_write = 0;
3245
3246         return X86EMUL_UNHANDLEABLE;
3247 }
3248
3249 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3250                           const void *val, int bytes)
3251 {
3252         int ret;
3253
3254         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3255         if (ret < 0)
3256                 return 0;
3257         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3258         return 1;
3259 }
3260
3261 static int emulator_write_emulated_onepage(unsigned long addr,
3262                                            const void *val,
3263                                            unsigned int bytes,
3264                                            struct kvm_vcpu *vcpu)
3265 {
3266         gpa_t                 gpa;
3267         u32 error_code;
3268
3269         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, &error_code);
3270
3271         if (gpa == UNMAPPED_GVA) {
3272                 kvm_inject_page_fault(vcpu, addr, error_code);
3273                 return X86EMUL_PROPAGATE_FAULT;
3274         }
3275
3276         /* For APIC access vmexit */
3277         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3278                 goto mmio;
3279
3280         if (emulator_write_phys(vcpu, gpa, val, bytes))
3281                 return X86EMUL_CONTINUE;
3282
3283 mmio:
3284         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3285         /*
3286          * Is this MMIO handled locally?
3287          */
3288         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3289                 return X86EMUL_CONTINUE;
3290
3291         vcpu->mmio_needed = 1;
3292         vcpu->mmio_phys_addr = gpa;
3293         vcpu->mmio_size = bytes;
3294         vcpu->mmio_is_write = 1;
3295         memcpy(vcpu->mmio_data, val, bytes);
3296
3297         return X86EMUL_CONTINUE;
3298 }
3299
3300 int emulator_write_emulated(unsigned long addr,
3301                                    const void *val,
3302                                    unsigned int bytes,
3303                                    struct kvm_vcpu *vcpu)
3304 {
3305         /* Crossing a page boundary? */
3306         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3307                 int rc, now;
3308
3309                 now = -addr & ~PAGE_MASK;
3310                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3311                 if (rc != X86EMUL_CONTINUE)
3312                         return rc;
3313                 addr += now;
3314                 val += now;
3315                 bytes -= now;
3316         }
3317         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3318 }
3319 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3320
3321 static int emulator_cmpxchg_emulated(unsigned long addr,
3322                                      const void *old,
3323                                      const void *new,
3324                                      unsigned int bytes,
3325                                      struct kvm_vcpu *vcpu)
3326 {
3327         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3328 #ifndef CONFIG_X86_64
3329         /* guests cmpxchg8b have to be emulated atomically */
3330         if (bytes == 8) {
3331                 gpa_t gpa;
3332                 struct page *page;
3333                 char *kaddr;
3334                 u64 val;
3335
3336                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3337
3338                 if (gpa == UNMAPPED_GVA ||
3339                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3340                         goto emul_write;
3341
3342                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3343                         goto emul_write;
3344
3345                 val = *(u64 *)new;
3346
3347                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3348
3349                 kaddr = kmap_atomic(page, KM_USER0);
3350                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3351                 kunmap_atomic(kaddr, KM_USER0);
3352                 kvm_release_page_dirty(page);
3353         }
3354 emul_write:
3355 #endif
3356
3357         return emulator_write_emulated(addr, new, bytes, vcpu);
3358 }
3359
3360 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3361 {
3362         return kvm_x86_ops->get_segment_base(vcpu, seg);
3363 }
3364
3365 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3366 {
3367         kvm_mmu_invlpg(vcpu, address);
3368         return X86EMUL_CONTINUE;
3369 }
3370
3371 int emulate_clts(struct kvm_vcpu *vcpu)
3372 {
3373         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3374         kvm_x86_ops->fpu_activate(vcpu);
3375         return X86EMUL_CONTINUE;
3376 }
3377
3378 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3379 {
3380         return kvm_x86_ops->get_dr(ctxt->vcpu, dr, dest);
3381 }
3382
3383 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3384 {
3385         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3386
3387         return kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask);
3388 }
3389
3390 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3391 {
3392         u8 opcodes[4];
3393         unsigned long rip = kvm_rip_read(vcpu);
3394         unsigned long rip_linear;
3395
3396         if (!printk_ratelimit())
3397                 return;
3398
3399         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3400
3401         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu, NULL);
3402
3403         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3404                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3405 }
3406 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3407
3408 static struct x86_emulate_ops emulate_ops = {
3409         .read_std            = kvm_read_guest_virt_system,
3410         .fetch               = kvm_fetch_guest_virt,
3411         .read_emulated       = emulator_read_emulated,
3412         .write_emulated      = emulator_write_emulated,
3413         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3414 };
3415
3416 static void cache_all_regs(struct kvm_vcpu *vcpu)
3417 {
3418         kvm_register_read(vcpu, VCPU_REGS_RAX);
3419         kvm_register_read(vcpu, VCPU_REGS_RSP);
3420         kvm_register_read(vcpu, VCPU_REGS_RIP);
3421         vcpu->arch.regs_dirty = ~0;
3422 }
3423
3424 int emulate_instruction(struct kvm_vcpu *vcpu,
3425                         unsigned long cr2,
3426                         u16 error_code,
3427                         int emulation_type)
3428 {
3429         int r, shadow_mask;
3430         struct decode_cache *c;
3431         struct kvm_run *run = vcpu->run;
3432
3433         kvm_clear_exception_queue(vcpu);
3434         vcpu->arch.mmio_fault_cr2 = cr2;
3435         /*
3436          * TODO: fix emulate.c to use guest_read/write_register
3437          * instead of direct ->regs accesses, can save hundred cycles
3438          * on Intel for instructions that don't read/change RSP, for
3439          * for example.
3440          */
3441         cache_all_regs(vcpu);
3442
3443         vcpu->mmio_is_write = 0;
3444         vcpu->arch.pio.string = 0;
3445
3446         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3447                 int cs_db, cs_l;
3448                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3449
3450                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3451                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
3452                 vcpu->arch.emulate_ctxt.mode =
3453                         (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3454                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3455                         ? X86EMUL_MODE_VM86 : cs_l
3456                         ? X86EMUL_MODE_PROT64 : cs_db
3457                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3458
3459                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3460
3461                 /* Only allow emulation of specific instructions on #UD
3462                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3463                 c = &vcpu->arch.emulate_ctxt.decode;
3464                 if (emulation_type & EMULTYPE_TRAP_UD) {
3465                         if (!c->twobyte)
3466                                 return EMULATE_FAIL;
3467                         switch (c->b) {
3468                         case 0x01: /* VMMCALL */
3469                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3470                                         return EMULATE_FAIL;
3471                                 break;
3472                         case 0x34: /* sysenter */
3473                         case 0x35: /* sysexit */
3474                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3475                                         return EMULATE_FAIL;
3476                                 break;
3477                         case 0x05: /* syscall */
3478                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3479                                         return EMULATE_FAIL;
3480                                 break;
3481                         default:
3482                                 return EMULATE_FAIL;
3483                         }
3484
3485                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3486                                 return EMULATE_FAIL;
3487                 }
3488
3489                 ++vcpu->stat.insn_emulation;
3490                 if (r)  {
3491                         ++vcpu->stat.insn_emulation_fail;
3492                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3493                                 return EMULATE_DONE;
3494                         return EMULATE_FAIL;
3495                 }
3496         }
3497
3498         if (emulation_type & EMULTYPE_SKIP) {
3499                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3500                 return EMULATE_DONE;
3501         }
3502
3503         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3504         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3505
3506         if (r == 0)
3507                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3508
3509         if (vcpu->arch.pio.string)
3510                 return EMULATE_DO_MMIO;
3511
3512         if (r || vcpu->mmio_is_write) {
3513                 run->exit_reason = KVM_EXIT_MMIO;
3514                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3515                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3516                 run->mmio.len = vcpu->mmio_size;
3517                 run->mmio.is_write = vcpu->mmio_is_write;
3518         }
3519
3520         if (r) {
3521                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3522                         return EMULATE_DONE;
3523                 if (!vcpu->mmio_needed) {
3524                         kvm_report_emulation_failure(vcpu, "mmio");
3525                         return EMULATE_FAIL;
3526                 }
3527                 return EMULATE_DO_MMIO;
3528         }
3529
3530         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3531
3532         if (vcpu->mmio_is_write) {
3533                 vcpu->mmio_needed = 0;
3534                 return EMULATE_DO_MMIO;
3535         }
3536
3537         return EMULATE_DONE;
3538 }
3539 EXPORT_SYMBOL_GPL(emulate_instruction);
3540
3541 static int pio_copy_data(struct kvm_vcpu *vcpu)
3542 {
3543         void *p = vcpu->arch.pio_data;
3544         gva_t q = vcpu->arch.pio.guest_gva;
3545         unsigned bytes;
3546         int ret;
3547         u32 error_code;
3548
3549         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3550         if (vcpu->arch.pio.in)
3551                 ret = kvm_write_guest_virt(q, p, bytes, vcpu, &error_code);
3552         else
3553                 ret = kvm_read_guest_virt(q, p, bytes, vcpu, &error_code);
3554
3555         if (ret == X86EMUL_PROPAGATE_FAULT)
3556                 kvm_inject_page_fault(vcpu, q, error_code);
3557
3558         return ret;
3559 }
3560
3561 int complete_pio(struct kvm_vcpu *vcpu)
3562 {
3563         struct kvm_pio_request *io = &vcpu->arch.pio;
3564         long delta;
3565         int r;
3566         unsigned long val;
3567
3568         if (!io->string) {
3569                 if (io->in) {
3570                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3571                         memcpy(&val, vcpu->arch.pio_data, io->size);
3572                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3573                 }
3574         } else {
3575                 if (io->in) {
3576                         r = pio_copy_data(vcpu);
3577                         if (r)
3578                                 goto out;
3579                 }
3580
3581                 delta = 1;
3582                 if (io->rep) {
3583                         delta *= io->cur_count;
3584                         /*
3585                          * The size of the register should really depend on
3586                          * current address size.
3587                          */
3588                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3589                         val -= delta;
3590                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3591                 }
3592                 if (io->down)
3593                         delta = -delta;
3594                 delta *= io->size;
3595                 if (io->in) {
3596                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3597                         val += delta;
3598                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3599                 } else {
3600                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3601                         val += delta;
3602                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3603                 }
3604         }
3605 out:
3606         io->count -= io->cur_count;
3607         io->cur_count = 0;
3608
3609         return 0;
3610 }
3611
3612 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3613 {
3614         /* TODO: String I/O for in kernel device */
3615         int r;
3616
3617         if (vcpu->arch.pio.in)
3618                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3619                                     vcpu->arch.pio.size, pd);
3620         else
3621                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3622                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3623                                      pd);
3624         return r;
3625 }
3626
3627 static int pio_string_write(struct kvm_vcpu *vcpu)
3628 {
3629         struct kvm_pio_request *io = &vcpu->arch.pio;
3630         void *pd = vcpu->arch.pio_data;
3631         int i, r = 0;
3632
3633         for (i = 0; i < io->cur_count; i++) {
3634                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3635                                      io->port, io->size, pd)) {
3636                         r = -EOPNOTSUPP;
3637                         break;
3638                 }
3639                 pd += io->size;
3640         }
3641         return r;
3642 }
3643
3644 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3645 {
3646         unsigned long val;
3647
3648         trace_kvm_pio(!in, port, size, 1);
3649
3650         vcpu->run->exit_reason = KVM_EXIT_IO;
3651         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3652         vcpu->run->io.size = vcpu->arch.pio.size = size;
3653         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3654         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3655         vcpu->run->io.port = vcpu->arch.pio.port = port;
3656         vcpu->arch.pio.in = in;
3657         vcpu->arch.pio.string = 0;
3658         vcpu->arch.pio.down = 0;
3659         vcpu->arch.pio.rep = 0;
3660
3661         if (!vcpu->arch.pio.in) {
3662                 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3663                 memcpy(vcpu->arch.pio_data, &val, 4);
3664         }
3665
3666         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3667                 complete_pio(vcpu);
3668                 return 1;
3669         }
3670         return 0;
3671 }
3672 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3673
3674 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3675                   int size, unsigned long count, int down,
3676                   gva_t address, int rep, unsigned port)
3677 {
3678         unsigned now, in_page;
3679         int ret = 0;
3680
3681         trace_kvm_pio(!in, port, size, count);
3682
3683         vcpu->run->exit_reason = KVM_EXIT_IO;
3684         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3685         vcpu->run->io.size = vcpu->arch.pio.size = size;
3686         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3687         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3688         vcpu->run->io.port = vcpu->arch.pio.port = port;
3689         vcpu->arch.pio.in = in;
3690         vcpu->arch.pio.string = 1;
3691         vcpu->arch.pio.down = down;
3692         vcpu->arch.pio.rep = rep;
3693
3694         if (!count) {
3695                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3696                 return 1;
3697         }
3698
3699         if (!down)
3700                 in_page = PAGE_SIZE - offset_in_page(address);
3701         else
3702                 in_page = offset_in_page(address) + size;
3703         now = min(count, (unsigned long)in_page / size);
3704         if (!now)
3705                 now = 1;
3706         if (down) {
3707                 /*
3708                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3709                  */
3710                 pr_unimpl(vcpu, "guest string pio down\n");
3711                 kvm_inject_gp(vcpu, 0);
3712                 return 1;
3713         }
3714         vcpu->run->io.count = now;
3715         vcpu->arch.pio.cur_count = now;
3716
3717         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3718                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3719
3720         vcpu->arch.pio.guest_gva = address;
3721
3722         if (!vcpu->arch.pio.in) {
3723                 /* string PIO write */
3724                 ret = pio_copy_data(vcpu);
3725                 if (ret == X86EMUL_PROPAGATE_FAULT)
3726                         return 1;
3727                 if (ret == 0 && !pio_string_write(vcpu)) {
3728                         complete_pio(vcpu);
3729                         if (vcpu->arch.pio.count == 0)
3730                                 ret = 1;
3731                 }
3732         }
3733         /* no string PIO read support yet */
3734
3735         return ret;
3736 }
3737 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3738
3739 static void bounce_off(void *info)
3740 {
3741         /* nothing */
3742 }
3743
3744 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3745                                      void *data)
3746 {
3747         struct cpufreq_freqs *freq = data;
3748         struct kvm *kvm;
3749         struct kvm_vcpu *vcpu;
3750         int i, send_ipi = 0;
3751
3752         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3753                 return 0;
3754         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3755                 return 0;
3756         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3757
3758         spin_lock(&kvm_lock);
3759         list_for_each_entry(kvm, &vm_list, vm_list) {
3760                 kvm_for_each_vcpu(i, vcpu, kvm) {
3761                         if (vcpu->cpu != freq->cpu)
3762                                 continue;
3763                         if (!kvm_request_guest_time_update(vcpu))
3764                                 continue;
3765                         if (vcpu->cpu != smp_processor_id())
3766                                 send_ipi++;
3767                 }
3768         }
3769         spin_unlock(&kvm_lock);
3770
3771         if (freq->old < freq->new && send_ipi) {
3772                 /*
3773                  * We upscale the frequency.  Must make the guest
3774                  * doesn't see old kvmclock values while running with
3775                  * the new frequency, otherwise we risk the guest sees
3776                  * time go backwards.
3777                  *
3778                  * In case we update the frequency for another cpu
3779                  * (which might be in guest context) send an interrupt
3780                  * to kick the cpu out of guest context.  Next time
3781                  * guest context is entered kvmclock will be updated,
3782                  * so the guest will not see stale values.
3783                  */
3784                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3785         }
3786         return 0;
3787 }
3788
3789 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3790         .notifier_call  = kvmclock_cpufreq_notifier
3791 };
3792
3793 static void kvm_timer_init(void)
3794 {
3795         int cpu;
3796
3797         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3798                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3799                                           CPUFREQ_TRANSITION_NOTIFIER);
3800                 for_each_online_cpu(cpu) {
3801                         unsigned long khz = cpufreq_get(cpu);
3802                         if (!khz)
3803                                 khz = tsc_khz;
3804                         per_cpu(cpu_tsc_khz, cpu) = khz;
3805                 }
3806         } else {
3807                 for_each_possible_cpu(cpu)
3808                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3809         }
3810 }
3811
3812 int kvm_arch_init(void *opaque)
3813 {
3814         int r;
3815         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3816
3817         if (kvm_x86_ops) {
3818                 printk(KERN_ERR "kvm: already loaded the other module\n");
3819                 r = -EEXIST;
3820                 goto out;
3821         }
3822
3823         if (!ops->cpu_has_kvm_support()) {
3824                 printk(KERN_ERR "kvm: no hardware support\n");
3825                 r = -EOPNOTSUPP;
3826                 goto out;
3827         }
3828         if (ops->disabled_by_bios()) {
3829                 printk(KERN_ERR "kvm: disabled by bios\n");
3830                 r = -EOPNOTSUPP;
3831                 goto out;
3832         }
3833
3834         r = kvm_mmu_module_init();
3835         if (r)
3836                 goto out;
3837
3838         kvm_init_msr_list();
3839
3840         kvm_x86_ops = ops;
3841         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3842         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3843         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3844                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3845
3846         kvm_timer_init();
3847
3848         return 0;
3849
3850 out:
3851         return r;
3852 }
3853
3854 void kvm_arch_exit(void)
3855 {
3856         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3857                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3858                                             CPUFREQ_TRANSITION_NOTIFIER);
3859         kvm_x86_ops = NULL;
3860         kvm_mmu_module_exit();
3861 }
3862
3863 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3864 {
3865         ++vcpu->stat.halt_exits;
3866         if (irqchip_in_kernel(vcpu->kvm)) {
3867                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3868                 return 1;
3869         } else {
3870                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3871                 return 0;
3872         }
3873 }
3874 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3875
3876 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3877                            unsigned long a1)
3878 {
3879         if (is_long_mode(vcpu))
3880                 return a0;
3881         else
3882                 return a0 | ((gpa_t)a1 << 32);
3883 }
3884
3885 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3886 {
3887         u64 param, ingpa, outgpa, ret;
3888         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3889         bool fast, longmode;
3890         int cs_db, cs_l;
3891
3892         /*
3893          * hypercall generates UD from non zero cpl and real mode
3894          * per HYPER-V spec
3895          */
3896         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
3897                 kvm_queue_exception(vcpu, UD_VECTOR);
3898                 return 0;
3899         }
3900
3901         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3902         longmode = is_long_mode(vcpu) && cs_l == 1;
3903
3904         if (!longmode) {
3905                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3906                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
3907                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3908                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
3909                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3910                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
3911         }
3912 #ifdef CONFIG_X86_64
3913         else {
3914                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3915                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3916                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3917         }
3918 #endif
3919
3920         code = param & 0xffff;
3921         fast = (param >> 16) & 0x1;
3922         rep_cnt = (param >> 32) & 0xfff;
3923         rep_idx = (param >> 48) & 0xfff;
3924
3925         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3926
3927         switch (code) {
3928         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
3929                 kvm_vcpu_on_spin(vcpu);
3930                 break;
3931         default:
3932                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
3933                 break;
3934         }
3935
3936         ret = res | (((u64)rep_done & 0xfff) << 32);
3937         if (longmode) {
3938                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3939         } else {
3940                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3941                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3942         }
3943
3944         return 1;
3945 }
3946
3947 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3948 {
3949         unsigned long nr, a0, a1, a2, a3, ret;
3950         int r = 1;
3951
3952         if (kvm_hv_hypercall_enabled(vcpu->kvm))
3953                 return kvm_hv_hypercall(vcpu);
3954
3955         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3956         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3957         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3958         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3959         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3960
3961         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3962
3963         if (!is_long_mode(vcpu)) {
3964                 nr &= 0xFFFFFFFF;
3965                 a0 &= 0xFFFFFFFF;
3966                 a1 &= 0xFFFFFFFF;
3967                 a2 &= 0xFFFFFFFF;
3968                 a3 &= 0xFFFFFFFF;
3969         }
3970
3971         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3972                 ret = -KVM_EPERM;
3973                 goto out;
3974         }
3975
3976         switch (nr) {
3977         case KVM_HC_VAPIC_POLL_IRQ:
3978                 ret = 0;
3979                 break;
3980         case KVM_HC_MMU_OP:
3981                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3982                 break;
3983         default:
3984                 ret = -KVM_ENOSYS;
3985                 break;
3986         }
3987 out:
3988         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3989         ++vcpu->stat.hypercalls;
3990         return r;
3991 }
3992 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3993
3994 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3995 {
3996         char instruction[3];
3997         unsigned long rip = kvm_rip_read(vcpu);
3998
3999         /*
4000          * Blow out the MMU to ensure that no other VCPU has an active mapping
4001          * to ensure that the updated hypercall appears atomically across all
4002          * VCPUs.
4003          */
4004         kvm_mmu_zap_all(vcpu->kvm);
4005
4006         kvm_x86_ops->patch_hypercall(vcpu, instruction);
4007
4008         return emulator_write_emulated(rip, instruction, 3, vcpu);
4009 }
4010
4011 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4012 {
4013         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4014 }
4015
4016 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4017 {
4018         struct desc_ptr dt = { limit, base };
4019
4020         kvm_x86_ops->set_gdt(vcpu, &dt);
4021 }
4022
4023 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4024 {
4025         struct desc_ptr dt = { limit, base };
4026
4027         kvm_x86_ops->set_idt(vcpu, &dt);
4028 }
4029
4030 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
4031                    unsigned long *rflags)
4032 {
4033         kvm_lmsw(vcpu, msw);
4034         *rflags = kvm_get_rflags(vcpu);
4035 }
4036
4037 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
4038 {
4039         unsigned long value;
4040
4041         switch (cr) {
4042         case 0:
4043                 value = kvm_read_cr0(vcpu);
4044                 break;
4045         case 2:
4046                 value = vcpu->arch.cr2;
4047                 break;
4048         case 3:
4049                 value = vcpu->arch.cr3;
4050                 break;
4051         case 4:
4052                 value = kvm_read_cr4(vcpu);
4053                 break;
4054         case 8:
4055                 value = kvm_get_cr8(vcpu);
4056                 break;
4057         default:
4058                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4059                 return 0;
4060         }
4061
4062         return value;
4063 }
4064
4065 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
4066                      unsigned long *rflags)
4067 {
4068         switch (cr) {
4069         case 0:
4070                 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4071                 *rflags = kvm_get_rflags(vcpu);
4072                 break;
4073         case 2:
4074                 vcpu->arch.cr2 = val;
4075                 break;
4076         case 3:
4077                 kvm_set_cr3(vcpu, val);
4078                 break;
4079         case 4:
4080                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4081                 break;
4082         case 8:
4083                 kvm_set_cr8(vcpu, val & 0xfUL);
4084                 break;
4085         default:
4086                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4087         }
4088 }
4089
4090 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4091 {
4092         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4093         int j, nent = vcpu->arch.cpuid_nent;
4094
4095         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4096         /* when no next entry is found, the current entry[i] is reselected */
4097         for (j = i + 1; ; j = (j + 1) % nent) {
4098                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4099                 if (ej->function == e->function) {
4100                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4101                         return j;
4102                 }
4103         }
4104         return 0; /* silence gcc, even though control never reaches here */
4105 }
4106
4107 /* find an entry with matching function, matching index (if needed), and that
4108  * should be read next (if it's stateful) */
4109 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4110         u32 function, u32 index)
4111 {
4112         if (e->function != function)
4113                 return 0;
4114         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4115                 return 0;
4116         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4117             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4118                 return 0;
4119         return 1;
4120 }
4121
4122 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4123                                               u32 function, u32 index)
4124 {
4125         int i;
4126         struct kvm_cpuid_entry2 *best = NULL;
4127
4128         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4129                 struct kvm_cpuid_entry2 *e;
4130
4131                 e = &vcpu->arch.cpuid_entries[i];
4132                 if (is_matching_cpuid_entry(e, function, index)) {
4133                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4134                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4135                         best = e;
4136                         break;
4137                 }
4138                 /*
4139                  * Both basic or both extended?
4140                  */
4141                 if (((e->function ^ function) & 0x80000000) == 0)
4142                         if (!best || e->function > best->function)
4143                                 best = e;
4144         }
4145         return best;
4146 }
4147 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4148
4149 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4150 {
4151         struct kvm_cpuid_entry2 *best;
4152
4153         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4154         if (best)
4155                 return best->eax & 0xff;
4156         return 36;
4157 }
4158
4159 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4160 {
4161         u32 function, index;
4162         struct kvm_cpuid_entry2 *best;
4163
4164         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4165         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4166         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4167         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4168         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4169         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4170         best = kvm_find_cpuid_entry(vcpu, function, index);
4171         if (best) {
4172                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4173                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4174                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4175                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4176         }
4177         kvm_x86_ops->skip_emulated_instruction(vcpu);
4178         trace_kvm_cpuid(function,
4179                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4180                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4181                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4182                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4183 }
4184 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4185
4186 /*
4187  * Check if userspace requested an interrupt window, and that the
4188  * interrupt window is open.
4189  *
4190  * No need to exit to userspace if we already have an interrupt queued.
4191  */
4192 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4193 {
4194         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4195                 vcpu->run->request_interrupt_window &&
4196                 kvm_arch_interrupt_allowed(vcpu));
4197 }
4198
4199 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4200 {
4201         struct kvm_run *kvm_run = vcpu->run;
4202
4203         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4204         kvm_run->cr8 = kvm_get_cr8(vcpu);
4205         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4206         if (irqchip_in_kernel(vcpu->kvm))
4207                 kvm_run->ready_for_interrupt_injection = 1;
4208         else
4209                 kvm_run->ready_for_interrupt_injection =
4210                         kvm_arch_interrupt_allowed(vcpu) &&
4211                         !kvm_cpu_has_interrupt(vcpu) &&
4212                         !kvm_event_needs_reinjection(vcpu);
4213 }
4214
4215 static void vapic_enter(struct kvm_vcpu *vcpu)
4216 {
4217         struct kvm_lapic *apic = vcpu->arch.apic;
4218         struct page *page;
4219
4220         if (!apic || !apic->vapic_addr)
4221                 return;
4222
4223         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4224
4225         vcpu->arch.apic->vapic_page = page;
4226 }
4227
4228 static void vapic_exit(struct kvm_vcpu *vcpu)
4229 {
4230         struct kvm_lapic *apic = vcpu->arch.apic;
4231         int idx;
4232
4233         if (!apic || !apic->vapic_addr)
4234                 return;
4235
4236         idx = srcu_read_lock(&vcpu->kvm->srcu);
4237         kvm_release_page_dirty(apic->vapic_page);
4238         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4239         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4240 }
4241
4242 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4243 {
4244         int max_irr, tpr;
4245
4246         if (!kvm_x86_ops->update_cr8_intercept)
4247                 return;
4248
4249         if (!vcpu->arch.apic)
4250                 return;
4251
4252         if (!vcpu->arch.apic->vapic_addr)
4253                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4254         else
4255                 max_irr = -1;
4256
4257         if (max_irr != -1)
4258                 max_irr >>= 4;
4259
4260         tpr = kvm_lapic_get_cr8(vcpu);
4261
4262         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4263 }
4264
4265 static void inject_pending_event(struct kvm_vcpu *vcpu)
4266 {
4267         /* try to reinject previous events if any */
4268         if (vcpu->arch.exception.pending) {
4269                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4270                                           vcpu->arch.exception.has_error_code,
4271                                           vcpu->arch.exception.error_code);
4272                 return;
4273         }
4274
4275         if (vcpu->arch.nmi_injected) {
4276                 kvm_x86_ops->set_nmi(vcpu);
4277                 return;
4278         }
4279
4280         if (vcpu->arch.interrupt.pending) {
4281                 kvm_x86_ops->set_irq(vcpu);
4282                 return;
4283         }
4284
4285         /* try to inject new event if pending */
4286         if (vcpu->arch.nmi_pending) {
4287                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4288                         vcpu->arch.nmi_pending = false;
4289                         vcpu->arch.nmi_injected = true;
4290                         kvm_x86_ops->set_nmi(vcpu);
4291                 }
4292         } else if (kvm_cpu_has_interrupt(vcpu)) {
4293                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4294                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4295                                             false);
4296                         kvm_x86_ops->set_irq(vcpu);
4297                 }
4298         }
4299 }
4300
4301 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4302 {
4303         int r;
4304         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4305                 vcpu->run->request_interrupt_window;
4306
4307         if (vcpu->requests)
4308                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4309                         kvm_mmu_unload(vcpu);
4310
4311         r = kvm_mmu_reload(vcpu);
4312         if (unlikely(r))
4313                 goto out;
4314
4315         if (vcpu->requests) {
4316                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4317                         __kvm_migrate_timers(vcpu);
4318                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4319                         kvm_write_guest_time(vcpu);
4320                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4321                         kvm_mmu_sync_roots(vcpu);
4322                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4323                         kvm_x86_ops->tlb_flush(vcpu);
4324                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4325                                        &vcpu->requests)) {
4326                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4327                         r = 0;
4328                         goto out;
4329                 }
4330                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4331                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4332                         r = 0;
4333                         goto out;
4334                 }
4335                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4336                         vcpu->fpu_active = 0;
4337                         kvm_x86_ops->fpu_deactivate(vcpu);
4338                 }
4339         }
4340
4341         preempt_disable();
4342
4343         kvm_x86_ops->prepare_guest_switch(vcpu);
4344         if (vcpu->fpu_active)
4345                 kvm_load_guest_fpu(vcpu);
4346
4347         local_irq_disable();
4348
4349         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4350         smp_mb__after_clear_bit();
4351
4352         if (vcpu->requests || need_resched() || signal_pending(current)) {
4353                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4354                 local_irq_enable();
4355                 preempt_enable();
4356                 r = 1;
4357                 goto out;
4358         }
4359
4360         inject_pending_event(vcpu);
4361
4362         /* enable NMI/IRQ window open exits if needed */
4363         if (vcpu->arch.nmi_pending)
4364                 kvm_x86_ops->enable_nmi_window(vcpu);
4365         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4366                 kvm_x86_ops->enable_irq_window(vcpu);
4367
4368         if (kvm_lapic_enabled(vcpu)) {
4369                 update_cr8_intercept(vcpu);
4370                 kvm_lapic_sync_to_vapic(vcpu);
4371         }
4372
4373         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4374
4375         kvm_guest_enter();
4376
4377         if (unlikely(vcpu->arch.switch_db_regs)) {
4378                 set_debugreg(0, 7);
4379                 set_debugreg(vcpu->arch.eff_db[0], 0);
4380                 set_debugreg(vcpu->arch.eff_db[1], 1);
4381                 set_debugreg(vcpu->arch.eff_db[2], 2);
4382                 set_debugreg(vcpu->arch.eff_db[3], 3);
4383         }
4384
4385         trace_kvm_entry(vcpu->vcpu_id);
4386         kvm_x86_ops->run(vcpu);
4387
4388         /*
4389          * If the guest has used debug registers, at least dr7
4390          * will be disabled while returning to the host.
4391          * If we don't have active breakpoints in the host, we don't
4392          * care about the messed up debug address registers. But if
4393          * we have some of them active, restore the old state.
4394          */
4395         if (hw_breakpoint_active())
4396                 hw_breakpoint_restore();
4397
4398         set_bit(KVM_REQ_KICK, &vcpu->requests);
4399         local_irq_enable();
4400
4401         ++vcpu->stat.exits;
4402
4403         /*
4404          * We must have an instruction between local_irq_enable() and
4405          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4406          * the interrupt shadow.  The stat.exits increment will do nicely.
4407          * But we need to prevent reordering, hence this barrier():
4408          */
4409         barrier();
4410
4411         kvm_guest_exit();
4412
4413         preempt_enable();
4414
4415         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4416
4417         /*
4418          * Profile KVM exit RIPs:
4419          */
4420         if (unlikely(prof_on == KVM_PROFILING)) {
4421                 unsigned long rip = kvm_rip_read(vcpu);
4422                 profile_hit(KVM_PROFILING, (void *)rip);
4423         }
4424
4425
4426         kvm_lapic_sync_from_vapic(vcpu);
4427
4428         r = kvm_x86_ops->handle_exit(vcpu);
4429 out:
4430         return r;
4431 }
4432
4433
4434 static int __vcpu_run(struct kvm_vcpu *vcpu)
4435 {
4436         int r;
4437         struct kvm *kvm = vcpu->kvm;
4438
4439         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4440                 pr_debug("vcpu %d received sipi with vector # %x\n",
4441                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4442                 kvm_lapic_reset(vcpu);
4443                 r = kvm_arch_vcpu_reset(vcpu);
4444                 if (r)
4445                         return r;
4446                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4447         }
4448
4449         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4450         vapic_enter(vcpu);
4451
4452         r = 1;
4453         while (r > 0) {
4454                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4455                         r = vcpu_enter_guest(vcpu);
4456                 else {
4457                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4458                         kvm_vcpu_block(vcpu);
4459                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4460                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4461                         {
4462                                 switch(vcpu->arch.mp_state) {
4463                                 case KVM_MP_STATE_HALTED:
4464                                         vcpu->arch.mp_state =
4465                                                 KVM_MP_STATE_RUNNABLE;
4466                                 case KVM_MP_STATE_RUNNABLE:
4467                                         break;
4468                                 case KVM_MP_STATE_SIPI_RECEIVED:
4469                                 default:
4470                                         r = -EINTR;
4471                                         break;
4472                                 }
4473                         }
4474                 }
4475
4476                 if (r <= 0)
4477                         break;
4478
4479                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4480                 if (kvm_cpu_has_pending_timer(vcpu))
4481                         kvm_inject_pending_timer_irqs(vcpu);
4482
4483                 if (dm_request_for_irq_injection(vcpu)) {
4484                         r = -EINTR;
4485                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4486                         ++vcpu->stat.request_irq_exits;
4487                 }
4488                 if (signal_pending(current)) {
4489                         r = -EINTR;
4490                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4491                         ++vcpu->stat.signal_exits;
4492                 }
4493                 if (need_resched()) {
4494                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4495                         kvm_resched(vcpu);
4496                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4497                 }
4498         }
4499
4500         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4501         post_kvm_run_save(vcpu);
4502
4503         vapic_exit(vcpu);
4504
4505         return r;
4506 }
4507
4508 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4509 {
4510         int r;
4511         sigset_t sigsaved;
4512
4513         vcpu_load(vcpu);
4514
4515         if (vcpu->sigset_active)
4516                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4517
4518         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4519                 kvm_vcpu_block(vcpu);
4520                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4521                 r = -EAGAIN;
4522                 goto out;
4523         }
4524
4525         /* re-sync apic's tpr */
4526         if (!irqchip_in_kernel(vcpu->kvm))
4527                 kvm_set_cr8(vcpu, kvm_run->cr8);
4528
4529         if (vcpu->arch.pio.cur_count) {
4530                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4531                 r = complete_pio(vcpu);
4532                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4533                 if (r)
4534                         goto out;
4535         }
4536         if (vcpu->mmio_needed) {
4537                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4538                 vcpu->mmio_read_completed = 1;
4539                 vcpu->mmio_needed = 0;
4540
4541                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4542                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4543                                         EMULTYPE_NO_DECODE);
4544                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4545                 if (r == EMULATE_DO_MMIO) {
4546                         /*
4547                          * Read-modify-write.  Back to userspace.
4548                          */
4549                         r = 0;
4550                         goto out;
4551                 }
4552         }
4553         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4554                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4555                                      kvm_run->hypercall.ret);
4556
4557         r = __vcpu_run(vcpu);
4558
4559 out:
4560         if (vcpu->sigset_active)
4561                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4562
4563         vcpu_put(vcpu);
4564         return r;
4565 }
4566
4567 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4568 {
4569         vcpu_load(vcpu);
4570
4571         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4572         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4573         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4574         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4575         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4576         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4577         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4578         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4579 #ifdef CONFIG_X86_64
4580         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4581         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4582         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4583         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4584         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4585         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4586         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4587         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4588 #endif
4589
4590         regs->rip = kvm_rip_read(vcpu);
4591         regs->rflags = kvm_get_rflags(vcpu);
4592
4593         vcpu_put(vcpu);
4594
4595         return 0;
4596 }
4597
4598 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4599 {
4600         vcpu_load(vcpu);
4601
4602         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4603         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4604         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4605         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4606         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4607         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4608         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4609         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4610 #ifdef CONFIG_X86_64
4611         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4612         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4613         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4614         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4615         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4616         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4617         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4618         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4619 #endif
4620
4621         kvm_rip_write(vcpu, regs->rip);
4622         kvm_set_rflags(vcpu, regs->rflags);
4623
4624         vcpu->arch.exception.pending = false;
4625
4626         vcpu_put(vcpu);
4627
4628         return 0;
4629 }
4630
4631 void kvm_get_segment(struct kvm_vcpu *vcpu,
4632                      struct kvm_segment *var, int seg)
4633 {
4634         kvm_x86_ops->get_segment(vcpu, var, seg);
4635 }
4636
4637 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4638 {
4639         struct kvm_segment cs;
4640
4641         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4642         *db = cs.db;
4643         *l = cs.l;
4644 }
4645 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4646
4647 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4648                                   struct kvm_sregs *sregs)
4649 {
4650         struct desc_ptr dt;
4651
4652         vcpu_load(vcpu);
4653
4654         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4655         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4656         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4657         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4658         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4659         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4660
4661         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4662         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4663
4664         kvm_x86_ops->get_idt(vcpu, &dt);
4665         sregs->idt.limit = dt.size;
4666         sregs->idt.base = dt.address;
4667         kvm_x86_ops->get_gdt(vcpu, &dt);
4668         sregs->gdt.limit = dt.size;
4669         sregs->gdt.base = dt.address;
4670
4671         sregs->cr0 = kvm_read_cr0(vcpu);
4672         sregs->cr2 = vcpu->arch.cr2;
4673         sregs->cr3 = vcpu->arch.cr3;
4674         sregs->cr4 = kvm_read_cr4(vcpu);
4675         sregs->cr8 = kvm_get_cr8(vcpu);
4676         sregs->efer = vcpu->arch.efer;
4677         sregs->apic_base = kvm_get_apic_base(vcpu);
4678
4679         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4680
4681         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4682                 set_bit(vcpu->arch.interrupt.nr,
4683                         (unsigned long *)sregs->interrupt_bitmap);
4684
4685         vcpu_put(vcpu);
4686
4687         return 0;
4688 }
4689
4690 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4691                                     struct kvm_mp_state *mp_state)
4692 {
4693         vcpu_load(vcpu);
4694         mp_state->mp_state = vcpu->arch.mp_state;
4695         vcpu_put(vcpu);
4696         return 0;
4697 }
4698
4699 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4700                                     struct kvm_mp_state *mp_state)
4701 {
4702         vcpu_load(vcpu);
4703         vcpu->arch.mp_state = mp_state->mp_state;
4704         vcpu_put(vcpu);
4705         return 0;
4706 }
4707
4708 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4709                         struct kvm_segment *var, int seg)
4710 {
4711         kvm_x86_ops->set_segment(vcpu, var, seg);
4712 }
4713
4714 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4715                                    struct kvm_segment *kvm_desct)
4716 {
4717         kvm_desct->base = get_desc_base(seg_desc);
4718         kvm_desct->limit = get_desc_limit(seg_desc);
4719         if (seg_desc->g) {
4720                 kvm_desct->limit <<= 12;
4721                 kvm_desct->limit |= 0xfff;
4722         }
4723         kvm_desct->selector = selector;
4724         kvm_desct->type = seg_desc->type;
4725         kvm_desct->present = seg_desc->p;
4726         kvm_desct->dpl = seg_desc->dpl;
4727         kvm_desct->db = seg_desc->d;
4728         kvm_desct->s = seg_desc->s;
4729         kvm_desct->l = seg_desc->l;
4730         kvm_desct->g = seg_desc->g;
4731         kvm_desct->avl = seg_desc->avl;
4732         if (!selector)
4733                 kvm_desct->unusable = 1;
4734         else
4735                 kvm_desct->unusable = 0;
4736         kvm_desct->padding = 0;
4737 }
4738
4739 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4740                                           u16 selector,
4741                                           struct desc_ptr *dtable)
4742 {
4743         if (selector & 1 << 2) {
4744                 struct kvm_segment kvm_seg;
4745
4746                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4747
4748                 if (kvm_seg.unusable)
4749                         dtable->size = 0;
4750                 else
4751                         dtable->size = kvm_seg.limit;
4752                 dtable->address = kvm_seg.base;
4753         }
4754         else
4755                 kvm_x86_ops->get_gdt(vcpu, dtable);
4756 }
4757
4758 /* allowed just for 8 bytes segments */
4759 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4760                                          struct desc_struct *seg_desc)
4761 {
4762         struct desc_ptr dtable;
4763         u16 index = selector >> 3;
4764         int ret;
4765         u32 err;
4766         gva_t addr;
4767
4768         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4769
4770         if (dtable.size < index * 8 + 7) {
4771                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4772                 return X86EMUL_PROPAGATE_FAULT;
4773         }
4774         addr = dtable.base + index * 8;
4775         ret = kvm_read_guest_virt_system(addr, seg_desc, sizeof(*seg_desc),
4776                                          vcpu,  &err);
4777         if (ret == X86EMUL_PROPAGATE_FAULT)
4778                 kvm_inject_page_fault(vcpu, addr, err);
4779
4780        return ret;
4781 }
4782
4783 /* allowed just for 8 bytes segments */
4784 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4785                                          struct desc_struct *seg_desc)
4786 {
4787         struct desc_ptr dtable;
4788         u16 index = selector >> 3;
4789
4790         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4791
4792         if (dtable.size < index * 8 + 7)
4793                 return 1;
4794         return kvm_write_guest_virt(dtable.address + index*8, seg_desc, sizeof(*seg_desc), vcpu, NULL);
4795 }
4796
4797 static gpa_t get_tss_base_addr_write(struct kvm_vcpu *vcpu,
4798                                struct desc_struct *seg_desc)
4799 {
4800         u32 base_addr = get_desc_base(seg_desc);
4801
4802         return kvm_mmu_gva_to_gpa_write(vcpu, base_addr, NULL);
4803 }
4804
4805 static gpa_t get_tss_base_addr_read(struct kvm_vcpu *vcpu,
4806                              struct desc_struct *seg_desc)
4807 {
4808         u32 base_addr = get_desc_base(seg_desc);
4809
4810         return kvm_mmu_gva_to_gpa_read(vcpu, base_addr, NULL);
4811 }
4812
4813 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4814 {
4815         struct kvm_segment kvm_seg;
4816
4817         kvm_get_segment(vcpu, &kvm_seg, seg);
4818         return kvm_seg.selector;
4819 }
4820
4821 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4822 {
4823         struct kvm_segment segvar = {
4824                 .base = selector << 4,
4825                 .limit = 0xffff,
4826                 .selector = selector,
4827                 .type = 3,
4828                 .present = 1,
4829                 .dpl = 3,
4830                 .db = 0,
4831                 .s = 1,
4832                 .l = 0,
4833                 .g = 0,
4834                 .avl = 0,
4835                 .unusable = 0,
4836         };
4837         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4838         return X86EMUL_CONTINUE;
4839 }
4840
4841 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4842 {
4843         return (seg != VCPU_SREG_LDTR) &&
4844                 (seg != VCPU_SREG_TR) &&
4845                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4846 }
4847
4848 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector, int seg)
4849 {
4850         struct kvm_segment kvm_seg;
4851         struct desc_struct seg_desc;
4852         u8 dpl, rpl, cpl;
4853         unsigned err_vec = GP_VECTOR;
4854         u32 err_code = 0;
4855         bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
4856         int ret;
4857
4858         if (is_vm86_segment(vcpu, seg) || !is_protmode(vcpu))
4859                 return kvm_load_realmode_segment(vcpu, selector, seg);
4860
4861         /* NULL selector is not valid for TR, CS and SS */
4862         if ((seg == VCPU_SREG_CS || seg == VCPU_SREG_SS || seg == VCPU_SREG_TR)
4863             && null_selector)
4864                 goto exception;
4865
4866         /* TR should be in GDT only */
4867         if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
4868                 goto exception;
4869
4870         ret = load_guest_segment_descriptor(vcpu, selector, &seg_desc);
4871         if (ret)
4872                 return ret;
4873
4874         seg_desct_to_kvm_desct(&seg_desc, selector, &kvm_seg);
4875
4876         if (null_selector) { /* for NULL selector skip all following checks */
4877                 kvm_seg.unusable = 1;
4878                 goto load;
4879         }
4880
4881         err_code = selector & 0xfffc;
4882         err_vec = GP_VECTOR;
4883
4884         /* can't load system descriptor into segment selecor */
4885         if (seg <= VCPU_SREG_GS && !kvm_seg.s)
4886                 goto exception;
4887
4888         if (!kvm_seg.present) {
4889                 err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
4890                 goto exception;
4891         }
4892
4893         rpl = selector & 3;
4894         dpl = kvm_seg.dpl;
4895         cpl = kvm_x86_ops->get_cpl(vcpu);
4896
4897         switch (seg) {
4898         case VCPU_SREG_SS:
4899                 /*
4900                  * segment is not a writable data segment or segment
4901                  * selector's RPL != CPL or segment selector's RPL != CPL
4902                  */
4903                 if (rpl != cpl || (kvm_seg.type & 0xa) != 0x2 || dpl != cpl)
4904                         goto exception;
4905                 break;
4906         case VCPU_SREG_CS:
4907                 if (!(kvm_seg.type & 8))
4908                         goto exception;
4909
4910                 if (kvm_seg.type & 4) {
4911                         /* conforming */
4912                         if (dpl > cpl)
4913                                 goto exception;
4914                 } else {
4915                         /* nonconforming */
4916                         if (rpl > cpl || dpl != cpl)
4917                                 goto exception;
4918                 }
4919                 /* CS(RPL) <- CPL */
4920                 selector = (selector & 0xfffc) | cpl;
4921             break;
4922         case VCPU_SREG_TR:
4923                 if (kvm_seg.s || (kvm_seg.type != 1 && kvm_seg.type != 9))
4924                         goto exception;
4925                 break;
4926         case VCPU_SREG_LDTR:
4927                 if (kvm_seg.s || kvm_seg.type != 2)
4928                         goto exception;
4929                 break;
4930         default: /*  DS, ES, FS, or GS */
4931                 /*
4932                  * segment is not a data or readable code segment or
4933                  * ((segment is a data or nonconforming code segment)
4934                  * and (both RPL and CPL > DPL))
4935                  */
4936                 if ((kvm_seg.type & 0xa) == 0x8 ||
4937                     (((kvm_seg.type & 0xc) != 0xc) && (rpl > dpl && cpl > dpl)))
4938                         goto exception;
4939                 break;
4940         }
4941
4942         if (!kvm_seg.unusable && kvm_seg.s) {
4943                 /* mark segment as accessed */
4944                 kvm_seg.type |= 1;
4945                 seg_desc.type |= 1;
4946                 save_guest_segment_descriptor(vcpu, selector, &seg_desc);
4947         }
4948 load:
4949         kvm_set_segment(vcpu, &kvm_seg, seg);
4950         return X86EMUL_CONTINUE;
4951 exception:
4952         kvm_queue_exception_e(vcpu, err_vec, err_code);
4953         return X86EMUL_PROPAGATE_FAULT;
4954 }
4955
4956 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4957                                 struct tss_segment_32 *tss)
4958 {
4959         tss->cr3 = vcpu->arch.cr3;
4960         tss->eip = kvm_rip_read(vcpu);
4961         tss->eflags = kvm_get_rflags(vcpu);
4962         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4963         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4964         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4965         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4966         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4967         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4968         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4969         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4970         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4971         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4972         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4973         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4974         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4975         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4976         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4977 }
4978
4979 static void kvm_load_segment_selector(struct kvm_vcpu *vcpu, u16 sel, int seg)
4980 {
4981         struct kvm_segment kvm_seg;
4982         kvm_get_segment(vcpu, &kvm_seg, seg);
4983         kvm_seg.selector = sel;
4984         kvm_set_segment(vcpu, &kvm_seg, seg);
4985 }
4986
4987 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4988                                   struct tss_segment_32 *tss)
4989 {
4990         kvm_set_cr3(vcpu, tss->cr3);
4991
4992         kvm_rip_write(vcpu, tss->eip);
4993         kvm_set_rflags(vcpu, tss->eflags | 2);
4994
4995         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4996         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4997         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4998         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4999         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
5000         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
5001         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
5002         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
5003
5004         /*
5005          * SDM says that segment selectors are loaded before segment
5006          * descriptors
5007          */
5008         kvm_load_segment_selector(vcpu, tss->ldt_selector, VCPU_SREG_LDTR);
5009         kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
5010         kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
5011         kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5012         kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5013         kvm_load_segment_selector(vcpu, tss->fs, VCPU_SREG_FS);
5014         kvm_load_segment_selector(vcpu, tss->gs, VCPU_SREG_GS);
5015
5016         /*
5017          * Now load segment descriptors. If fault happenes at this stage
5018          * it is handled in a context of new task
5019          */
5020         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, VCPU_SREG_LDTR))
5021                 return 1;
5022
5023         if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5024                 return 1;
5025
5026         if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5027                 return 1;
5028
5029         if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5030                 return 1;
5031
5032         if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5033                 return 1;
5034
5035         if (kvm_load_segment_descriptor(vcpu, tss->fs, VCPU_SREG_FS))
5036                 return 1;
5037
5038         if (kvm_load_segment_descriptor(vcpu, tss->gs, VCPU_SREG_GS))
5039                 return 1;
5040         return 0;
5041 }
5042
5043 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
5044                                 struct tss_segment_16 *tss)
5045 {
5046         tss->ip = kvm_rip_read(vcpu);
5047         tss->flag = kvm_get_rflags(vcpu);
5048         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5049         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5050         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5051         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5052         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5053         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5054         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
5055         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
5056
5057         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
5058         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
5059         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
5060         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
5061         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
5062 }
5063
5064 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
5065                                  struct tss_segment_16 *tss)
5066 {
5067         kvm_rip_write(vcpu, tss->ip);
5068         kvm_set_rflags(vcpu, tss->flag | 2);
5069         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
5070         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
5071         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
5072         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
5073         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
5074         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
5075         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
5076         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
5077
5078         /*
5079          * SDM says that segment selectors are loaded before segment
5080          * descriptors
5081          */
5082         kvm_load_segment_selector(vcpu, tss->ldt, VCPU_SREG_LDTR);
5083         kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
5084         kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
5085         kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5086         kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5087
5088         /*
5089          * Now load segment descriptors. If fault happenes at this stage
5090          * it is handled in a context of new task
5091          */
5092         if (kvm_load_segment_descriptor(vcpu, tss->ldt, VCPU_SREG_LDTR))
5093                 return 1;
5094
5095         if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5096                 return 1;
5097
5098         if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5099                 return 1;
5100
5101         if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5102                 return 1;
5103
5104         if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5105                 return 1;
5106         return 0;
5107 }
5108
5109 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
5110                               u16 old_tss_sel, u32 old_tss_base,
5111                               struct desc_struct *nseg_desc)
5112 {
5113         struct tss_segment_16 tss_segment_16;
5114         int ret = 0;
5115
5116         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5117                            sizeof tss_segment_16))
5118                 goto out;
5119
5120         save_state_to_tss16(vcpu, &tss_segment_16);
5121
5122         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5123                             sizeof tss_segment_16))
5124                 goto out;
5125
5126         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5127                            &tss_segment_16, sizeof tss_segment_16))
5128                 goto out;
5129
5130         if (old_tss_sel != 0xffff) {
5131                 tss_segment_16.prev_task_link = old_tss_sel;
5132
5133                 if (kvm_write_guest(vcpu->kvm,
5134                                     get_tss_base_addr_write(vcpu, nseg_desc),
5135                                     &tss_segment_16.prev_task_link,
5136                                     sizeof tss_segment_16.prev_task_link))
5137                         goto out;
5138         }
5139
5140         if (load_state_from_tss16(vcpu, &tss_segment_16))
5141                 goto out;
5142
5143         ret = 1;
5144 out:
5145         return ret;
5146 }
5147
5148 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
5149                        u16 old_tss_sel, u32 old_tss_base,
5150                        struct desc_struct *nseg_desc)
5151 {
5152         struct tss_segment_32 tss_segment_32;
5153         int ret = 0;
5154
5155         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5156                            sizeof tss_segment_32))
5157                 goto out;
5158
5159         save_state_to_tss32(vcpu, &tss_segment_32);
5160
5161         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5162                             sizeof tss_segment_32))
5163                 goto out;
5164
5165         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5166                            &tss_segment_32, sizeof tss_segment_32))
5167                 goto out;
5168
5169         if (old_tss_sel != 0xffff) {
5170                 tss_segment_32.prev_task_link = old_tss_sel;
5171
5172                 if (kvm_write_guest(vcpu->kvm,
5173                                     get_tss_base_addr_write(vcpu, nseg_desc),
5174                                     &tss_segment_32.prev_task_link,
5175                                     sizeof tss_segment_32.prev_task_link))
5176                         goto out;
5177         }
5178
5179         if (load_state_from_tss32(vcpu, &tss_segment_32))
5180                 goto out;
5181
5182         ret = 1;
5183 out:
5184         return ret;
5185 }
5186
5187 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
5188 {
5189         struct kvm_segment tr_seg;
5190         struct desc_struct cseg_desc;
5191         struct desc_struct nseg_desc;
5192         int ret = 0;
5193         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
5194         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
5195         u32 desc_limit;
5196
5197         old_tss_base = kvm_mmu_gva_to_gpa_write(vcpu, old_tss_base, NULL);
5198
5199         /* FIXME: Handle errors. Failure to read either TSS or their
5200          * descriptors should generate a pagefault.
5201          */
5202         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
5203                 goto out;
5204
5205         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
5206                 goto out;
5207
5208         if (reason != TASK_SWITCH_IRET) {
5209                 int cpl;
5210
5211                 cpl = kvm_x86_ops->get_cpl(vcpu);
5212                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
5213                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5214                         return 1;
5215                 }
5216         }
5217
5218         desc_limit = get_desc_limit(&nseg_desc);
5219         if (!nseg_desc.p ||
5220             ((desc_limit < 0x67 && (nseg_desc.type & 8)) ||
5221              desc_limit < 0x2b)) {
5222                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
5223                 return 1;
5224         }
5225
5226         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
5227                 cseg_desc.type &= ~(1 << 1); //clear the B flag
5228                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
5229         }
5230
5231         if (reason == TASK_SWITCH_IRET) {
5232                 u32 eflags = kvm_get_rflags(vcpu);
5233                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
5234         }
5235
5236         /* set back link to prev task only if NT bit is set in eflags
5237            note that old_tss_sel is not used afetr this point */
5238         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
5239                 old_tss_sel = 0xffff;
5240
5241         if (nseg_desc.type & 8)
5242                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
5243                                          old_tss_base, &nseg_desc);
5244         else
5245                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
5246                                          old_tss_base, &nseg_desc);
5247
5248         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
5249                 u32 eflags = kvm_get_rflags(vcpu);
5250                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
5251         }
5252
5253         if (reason != TASK_SWITCH_IRET) {
5254                 nseg_desc.type |= (1 << 1);
5255                 save_guest_segment_descriptor(vcpu, tss_selector,
5256                                               &nseg_desc);
5257         }
5258
5259         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5260         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5261         tr_seg.type = 11;
5262         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5263 out:
5264         return ret;
5265 }
5266 EXPORT_SYMBOL_GPL(kvm_task_switch);
5267
5268 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5269                                   struct kvm_sregs *sregs)
5270 {
5271         int mmu_reset_needed = 0;
5272         int pending_vec, max_bits;
5273         struct desc_ptr dt;
5274
5275         vcpu_load(vcpu);
5276
5277         dt.size = sregs->idt.limit;
5278         dt.address = sregs->idt.base;
5279         kvm_x86_ops->set_idt(vcpu, &dt);
5280         dt.size = sregs->gdt.limit;
5281         dt.address = sregs->gdt.base;
5282         kvm_x86_ops->set_gdt(vcpu, &dt);
5283
5284         vcpu->arch.cr2 = sregs->cr2;
5285         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5286         vcpu->arch.cr3 = sregs->cr3;
5287
5288         kvm_set_cr8(vcpu, sregs->cr8);
5289
5290         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5291         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5292         kvm_set_apic_base(vcpu, sregs->apic_base);
5293
5294         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5295         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5296         vcpu->arch.cr0 = sregs->cr0;
5297
5298         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5299         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5300         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5301                 load_pdptrs(vcpu, vcpu->arch.cr3);
5302                 mmu_reset_needed = 1;
5303         }
5304
5305         if (mmu_reset_needed)
5306                 kvm_mmu_reset_context(vcpu);
5307
5308         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5309         pending_vec = find_first_bit(
5310                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5311         if (pending_vec < max_bits) {
5312                 kvm_queue_interrupt(vcpu, pending_vec, false);
5313                 pr_debug("Set back pending irq %d\n", pending_vec);
5314                 if (irqchip_in_kernel(vcpu->kvm))
5315                         kvm_pic_clear_isr_ack(vcpu->kvm);
5316         }
5317
5318         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5319         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5320         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5321         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5322         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5323         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5324
5325         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5326         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5327
5328         update_cr8_intercept(vcpu);
5329
5330         /* Older userspace won't unhalt the vcpu on reset. */
5331         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5332             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5333             !is_protmode(vcpu))
5334                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5335
5336         vcpu_put(vcpu);
5337
5338         return 0;
5339 }
5340
5341 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5342                                         struct kvm_guest_debug *dbg)
5343 {
5344         unsigned long rflags;
5345         int i, r;
5346
5347         vcpu_load(vcpu);
5348
5349         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5350                 r = -EBUSY;
5351                 if (vcpu->arch.exception.pending)
5352                         goto unlock_out;
5353                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5354                         kvm_queue_exception(vcpu, DB_VECTOR);
5355                 else
5356                         kvm_queue_exception(vcpu, BP_VECTOR);
5357         }
5358
5359         /*
5360          * Read rflags as long as potentially injected trace flags are still
5361          * filtered out.
5362          */
5363         rflags = kvm_get_rflags(vcpu);
5364
5365         vcpu->guest_debug = dbg->control;
5366         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5367                 vcpu->guest_debug = 0;
5368
5369         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5370                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5371                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5372                 vcpu->arch.switch_db_regs =
5373                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5374         } else {
5375                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5376                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5377                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5378         }
5379
5380         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5381                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5382                         get_segment_base(vcpu, VCPU_SREG_CS);
5383
5384         /*
5385          * Trigger an rflags update that will inject or remove the trace
5386          * flags.
5387          */
5388         kvm_set_rflags(vcpu, rflags);
5389
5390         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5391
5392         r = 0;
5393
5394 unlock_out:
5395         vcpu_put(vcpu);
5396
5397         return r;
5398 }
5399
5400 /*
5401  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
5402  * we have asm/x86/processor.h
5403  */
5404 struct fxsave {
5405         u16     cwd;
5406         u16     swd;
5407         u16     twd;
5408         u16     fop;
5409         u64     rip;
5410         u64     rdp;
5411         u32     mxcsr;
5412         u32     mxcsr_mask;
5413         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
5414 #ifdef CONFIG_X86_64
5415         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
5416 #else
5417         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
5418 #endif
5419 };
5420
5421 /*
5422  * Translate a guest virtual address to a guest physical address.
5423  */
5424 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5425                                     struct kvm_translation *tr)
5426 {
5427         unsigned long vaddr = tr->linear_address;
5428         gpa_t gpa;
5429         int idx;
5430
5431         vcpu_load(vcpu);
5432         idx = srcu_read_lock(&vcpu->kvm->srcu);
5433         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5434         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5435         tr->physical_address = gpa;
5436         tr->valid = gpa != UNMAPPED_GVA;
5437         tr->writeable = 1;
5438         tr->usermode = 0;
5439         vcpu_put(vcpu);
5440
5441         return 0;
5442 }
5443
5444 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5445 {
5446         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5447
5448         vcpu_load(vcpu);
5449
5450         memcpy(fpu->fpr, fxsave->st_space, 128);
5451         fpu->fcw = fxsave->cwd;
5452         fpu->fsw = fxsave->swd;
5453         fpu->ftwx = fxsave->twd;
5454         fpu->last_opcode = fxsave->fop;
5455         fpu->last_ip = fxsave->rip;
5456         fpu->last_dp = fxsave->rdp;
5457         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5458
5459         vcpu_put(vcpu);
5460
5461         return 0;
5462 }
5463
5464 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5465 {
5466         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5467
5468         vcpu_load(vcpu);
5469
5470         memcpy(fxsave->st_space, fpu->fpr, 128);
5471         fxsave->cwd = fpu->fcw;
5472         fxsave->swd = fpu->fsw;
5473         fxsave->twd = fpu->ftwx;
5474         fxsave->fop = fpu->last_opcode;
5475         fxsave->rip = fpu->last_ip;
5476         fxsave->rdp = fpu->last_dp;
5477         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5478
5479         vcpu_put(vcpu);
5480
5481         return 0;
5482 }
5483
5484 void fx_init(struct kvm_vcpu *vcpu)
5485 {
5486         unsigned after_mxcsr_mask;
5487
5488         /*
5489          * Touch the fpu the first time in non atomic context as if
5490          * this is the first fpu instruction the exception handler
5491          * will fire before the instruction returns and it'll have to
5492          * allocate ram with GFP_KERNEL.
5493          */
5494         if (!used_math())
5495                 kvm_fx_save(&vcpu->arch.host_fx_image);
5496
5497         /* Initialize guest FPU by resetting ours and saving into guest's */
5498         preempt_disable();
5499         kvm_fx_save(&vcpu->arch.host_fx_image);
5500         kvm_fx_finit();
5501         kvm_fx_save(&vcpu->arch.guest_fx_image);
5502         kvm_fx_restore(&vcpu->arch.host_fx_image);
5503         preempt_enable();
5504
5505         vcpu->arch.cr0 |= X86_CR0_ET;
5506         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5507         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5508         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5509                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5510 }
5511 EXPORT_SYMBOL_GPL(fx_init);
5512
5513 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5514 {
5515         if (vcpu->guest_fpu_loaded)
5516                 return;
5517
5518         vcpu->guest_fpu_loaded = 1;
5519         kvm_fx_save(&vcpu->arch.host_fx_image);
5520         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5521         trace_kvm_fpu(1);
5522 }
5523
5524 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5525 {
5526         if (!vcpu->guest_fpu_loaded)
5527                 return;
5528
5529         vcpu->guest_fpu_loaded = 0;
5530         kvm_fx_save(&vcpu->arch.guest_fx_image);
5531         kvm_fx_restore(&vcpu->arch.host_fx_image);
5532         ++vcpu->stat.fpu_reload;
5533         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5534         trace_kvm_fpu(0);
5535 }
5536
5537 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5538 {
5539         if (vcpu->arch.time_page) {
5540                 kvm_release_page_dirty(vcpu->arch.time_page);
5541                 vcpu->arch.time_page = NULL;
5542         }
5543
5544         kvm_x86_ops->vcpu_free(vcpu);
5545 }
5546
5547 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5548                                                 unsigned int id)
5549 {
5550         return kvm_x86_ops->vcpu_create(kvm, id);
5551 }
5552
5553 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5554 {
5555         int r;
5556
5557         /* We do fxsave: this must be aligned. */
5558         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5559
5560         vcpu->arch.mtrr_state.have_fixed = 1;
5561         vcpu_load(vcpu);
5562         r = kvm_arch_vcpu_reset(vcpu);
5563         if (r == 0)
5564                 r = kvm_mmu_setup(vcpu);
5565         vcpu_put(vcpu);
5566         if (r < 0)
5567                 goto free_vcpu;
5568
5569         return 0;
5570 free_vcpu:
5571         kvm_x86_ops->vcpu_free(vcpu);
5572         return r;
5573 }
5574
5575 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5576 {
5577         vcpu_load(vcpu);
5578         kvm_mmu_unload(vcpu);
5579         vcpu_put(vcpu);
5580
5581         kvm_x86_ops->vcpu_free(vcpu);
5582 }
5583
5584 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5585 {
5586         vcpu->arch.nmi_pending = false;
5587         vcpu->arch.nmi_injected = false;
5588
5589         vcpu->arch.switch_db_regs = 0;
5590         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5591         vcpu->arch.dr6 = DR6_FIXED_1;
5592         vcpu->arch.dr7 = DR7_FIXED_1;
5593
5594         return kvm_x86_ops->vcpu_reset(vcpu);
5595 }
5596
5597 int kvm_arch_hardware_enable(void *garbage)
5598 {
5599         /*
5600          * Since this may be called from a hotplug notifcation,
5601          * we can't get the CPU frequency directly.
5602          */
5603         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5604                 int cpu = raw_smp_processor_id();
5605                 per_cpu(cpu_tsc_khz, cpu) = 0;
5606         }
5607
5608         kvm_shared_msr_cpu_online();
5609
5610         return kvm_x86_ops->hardware_enable(garbage);
5611 }
5612
5613 void kvm_arch_hardware_disable(void *garbage)
5614 {
5615         kvm_x86_ops->hardware_disable(garbage);
5616         drop_user_return_notifiers(garbage);
5617 }
5618
5619 int kvm_arch_hardware_setup(void)
5620 {
5621         return kvm_x86_ops->hardware_setup();
5622 }
5623
5624 void kvm_arch_hardware_unsetup(void)
5625 {
5626         kvm_x86_ops->hardware_unsetup();
5627 }
5628
5629 void kvm_arch_check_processor_compat(void *rtn)
5630 {
5631         kvm_x86_ops->check_processor_compatibility(rtn);
5632 }
5633
5634 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5635 {
5636         struct page *page;
5637         struct kvm *kvm;
5638         int r;
5639
5640         BUG_ON(vcpu->kvm == NULL);
5641         kvm = vcpu->kvm;
5642
5643         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5644         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5645                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5646         else
5647                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5648
5649         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5650         if (!page) {
5651                 r = -ENOMEM;
5652                 goto fail;
5653         }
5654         vcpu->arch.pio_data = page_address(page);
5655
5656         r = kvm_mmu_create(vcpu);
5657         if (r < 0)
5658                 goto fail_free_pio_data;
5659
5660         if (irqchip_in_kernel(kvm)) {
5661                 r = kvm_create_lapic(vcpu);
5662                 if (r < 0)
5663                         goto fail_mmu_destroy;
5664         }
5665
5666         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5667                                        GFP_KERNEL);
5668         if (!vcpu->arch.mce_banks) {
5669                 r = -ENOMEM;
5670                 goto fail_free_lapic;
5671         }
5672         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5673
5674         return 0;
5675 fail_free_lapic:
5676         kvm_free_lapic(vcpu);
5677 fail_mmu_destroy:
5678         kvm_mmu_destroy(vcpu);
5679 fail_free_pio_data:
5680         free_page((unsigned long)vcpu->arch.pio_data);
5681 fail:
5682         return r;
5683 }
5684
5685 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5686 {
5687         int idx;
5688
5689         kfree(vcpu->arch.mce_banks);
5690         kvm_free_lapic(vcpu);
5691         idx = srcu_read_lock(&vcpu->kvm->srcu);
5692         kvm_mmu_destroy(vcpu);
5693         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5694         free_page((unsigned long)vcpu->arch.pio_data);
5695 }
5696
5697 struct  kvm *kvm_arch_create_vm(void)
5698 {
5699         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5700
5701         if (!kvm)
5702                 return ERR_PTR(-ENOMEM);
5703
5704         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5705         if (!kvm->arch.aliases) {
5706                 kfree(kvm);
5707                 return ERR_PTR(-ENOMEM);
5708         }
5709
5710         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5711         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5712
5713         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5714         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5715
5716         rdtscll(kvm->arch.vm_init_tsc);
5717
5718         return kvm;
5719 }
5720
5721 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5722 {
5723         vcpu_load(vcpu);
5724         kvm_mmu_unload(vcpu);
5725         vcpu_put(vcpu);
5726 }
5727
5728 static void kvm_free_vcpus(struct kvm *kvm)
5729 {
5730         unsigned int i;
5731         struct kvm_vcpu *vcpu;
5732
5733         /*
5734          * Unpin any mmu pages first.
5735          */
5736         kvm_for_each_vcpu(i, vcpu, kvm)
5737                 kvm_unload_vcpu_mmu(vcpu);
5738         kvm_for_each_vcpu(i, vcpu, kvm)
5739                 kvm_arch_vcpu_free(vcpu);
5740
5741         mutex_lock(&kvm->lock);
5742         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5743                 kvm->vcpus[i] = NULL;
5744
5745         atomic_set(&kvm->online_vcpus, 0);
5746         mutex_unlock(&kvm->lock);
5747 }
5748
5749 void kvm_arch_sync_events(struct kvm *kvm)
5750 {
5751         kvm_free_all_assigned_devices(kvm);
5752 }
5753
5754 void kvm_arch_destroy_vm(struct kvm *kvm)
5755 {
5756         kvm_iommu_unmap_guest(kvm);
5757         kvm_free_pit(kvm);
5758         kfree(kvm->arch.vpic);
5759         kfree(kvm->arch.vioapic);
5760         kvm_free_vcpus(kvm);
5761         kvm_free_physmem(kvm);
5762         if (kvm->arch.apic_access_page)
5763                 put_page(kvm->arch.apic_access_page);
5764         if (kvm->arch.ept_identity_pagetable)
5765                 put_page(kvm->arch.ept_identity_pagetable);
5766         cleanup_srcu_struct(&kvm->srcu);
5767         kfree(kvm->arch.aliases);
5768         kfree(kvm);
5769 }
5770
5771 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5772                                 struct kvm_memory_slot *memslot,
5773                                 struct kvm_memory_slot old,
5774                                 struct kvm_userspace_memory_region *mem,
5775                                 int user_alloc)
5776 {
5777         int npages = memslot->npages;
5778
5779         /*To keep backward compatibility with older userspace,
5780          *x86 needs to hanlde !user_alloc case.
5781          */
5782         if (!user_alloc) {
5783                 if (npages && !old.rmap) {
5784                         unsigned long userspace_addr;
5785
5786                         down_write(&current->mm->mmap_sem);
5787                         userspace_addr = do_mmap(NULL, 0,
5788                                                  npages * PAGE_SIZE,
5789                                                  PROT_READ | PROT_WRITE,
5790                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5791                                                  0);
5792                         up_write(&current->mm->mmap_sem);
5793
5794                         if (IS_ERR((void *)userspace_addr))
5795                                 return PTR_ERR((void *)userspace_addr);
5796
5797                         memslot->userspace_addr = userspace_addr;
5798                 }
5799         }
5800
5801
5802         return 0;
5803 }
5804
5805 void kvm_arch_commit_memory_region(struct kvm *kvm,
5806                                 struct kvm_userspace_memory_region *mem,
5807                                 struct kvm_memory_slot old,
5808                                 int user_alloc)
5809 {
5810
5811         int npages = mem->memory_size >> PAGE_SHIFT;
5812
5813         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5814                 int ret;
5815
5816                 down_write(&current->mm->mmap_sem);
5817                 ret = do_munmap(current->mm, old.userspace_addr,
5818                                 old.npages * PAGE_SIZE);
5819                 up_write(&current->mm->mmap_sem);
5820                 if (ret < 0)
5821                         printk(KERN_WARNING
5822                                "kvm_vm_ioctl_set_memory_region: "
5823                                "failed to munmap memory\n");
5824         }
5825
5826         spin_lock(&kvm->mmu_lock);
5827         if (!kvm->arch.n_requested_mmu_pages) {
5828                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5829                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5830         }
5831
5832         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5833         spin_unlock(&kvm->mmu_lock);
5834 }
5835
5836 void kvm_arch_flush_shadow(struct kvm *kvm)
5837 {
5838         kvm_mmu_zap_all(kvm);
5839         kvm_reload_remote_mmus(kvm);
5840 }
5841
5842 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5843 {
5844         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5845                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5846                 || vcpu->arch.nmi_pending ||
5847                 (kvm_arch_interrupt_allowed(vcpu) &&
5848                  kvm_cpu_has_interrupt(vcpu));
5849 }
5850
5851 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5852 {
5853         int me;
5854         int cpu = vcpu->cpu;
5855
5856         if (waitqueue_active(&vcpu->wq)) {
5857                 wake_up_interruptible(&vcpu->wq);
5858                 ++vcpu->stat.halt_wakeup;
5859         }
5860
5861         me = get_cpu();
5862         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5863                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5864                         smp_send_reschedule(cpu);
5865         put_cpu();
5866 }
5867
5868 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5869 {
5870         return kvm_x86_ops->interrupt_allowed(vcpu);
5871 }
5872
5873 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5874 {
5875         unsigned long current_rip = kvm_rip_read(vcpu) +
5876                 get_segment_base(vcpu, VCPU_SREG_CS);
5877
5878         return current_rip == linear_rip;
5879 }
5880 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5881
5882 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5883 {
5884         unsigned long rflags;
5885
5886         rflags = kvm_x86_ops->get_rflags(vcpu);
5887         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5888                 rflags &= ~X86_EFLAGS_TF;
5889         return rflags;
5890 }
5891 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5892
5893 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5894 {
5895         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5896             kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5897                 rflags |= X86_EFLAGS_TF;
5898         kvm_x86_ops->set_rflags(vcpu, rflags);
5899 }
5900 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5901
5902 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5903 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5904 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5905 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5906 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5907 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5908 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5909 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5910 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5911 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5912 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5913 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);