SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/x86/kvm/x86.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *   Amit Shah    <amit.shah@qumranet.com>
  14  *   Ben-Ami Yassour <benami@il.ibm.com>
  15  *
  16  * This work is licensed under the terms of the GNU GPL, version 2.  See
  17  * the COPYING file in the top-level directory.
  18  *
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "i8254.h"
  25 #include "tss.h"
  26 #include "kvm_cache_regs.h"
  27 #include "x86.h"
  28
  29 #include <linux/clocksource.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/kvm.h>
  32 #include <linux/fs.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/module.h>
  35 #include <linux/mman.h>
  36 #include <linux/highmem.h>
  37 #include <linux/iommu.h>
  38 #include <linux/intel-iommu.h>
  39 #include <linux/cpufreq.h>
  40 #include <linux/user-return-notifier.h>
  41 #include <linux/srcu.h>
  42 #include <linux/slab.h>
  43 #include <linux/perf_event.h>
  44 #include <trace/events/kvm.h>
  45 #undef TRACE_INCLUDE_FILE
  46 #define CREATE_TRACE_POINTS
  47 #include "trace.h"
  48
  49 #include <asm/debugreg.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/msr.h>
  52 #include <asm/desc.h>
  53 #include <asm/mtrr.h>
  54 #include <asm/mce.h>
  55
  56 #define MAX_IO_MSRS 256
  57 #define CR0_RESERVED_BITS                                               \
  58         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  59                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  60                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  61 #define CR4_RESERVED_BITS                                               \
  62         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  63                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  64                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  65                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  66
  67 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  68
  69 #define KVM_MAX_MCE_BANKS 32
  70 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  71
  72 /* EFER defaults:
  73  * - enable syscall per default because its emulated by KVM
  74  * - enable LME and LMA per default on 64 bit KVM
  75  */
  76 #ifdef CONFIG_X86_64
  77 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  78 #else
  79 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  80 #endif
  81
  82 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  83 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  84
  85 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  86 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  87                                     struct kvm_cpuid_entry2 __user *entries);
  88
  89 struct kvm_x86_ops *kvm_x86_ops;
  90 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  91
  92 int ignore_msrs = 0;
  93 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  94
  95 #define KVM_NR_SHARED_MSRS 16
  96
  97 struct kvm_shared_msrs_global {
  98         int nr;
  99         u32 msrs[KVM_NR_SHARED_MSRS];
 100 };
 101
 102 struct kvm_shared_msrs {
 103         struct user_return_notifier urn;
 104         bool registered;
 105         struct kvm_shared_msr_values {
 106                 u64 host;
 107                 u64 curr;
 108         } values[KVM_NR_SHARED_MSRS];
 109 };
 110
 111 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 112 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 113
 114 struct kvm_stats_debugfs_item debugfs_entries[] = {
 115         { "pf_fixed", VCPU_STAT(pf_fixed) },
 116         { "pf_guest", VCPU_STAT(pf_guest) },
 117         { "tlb_flush", VCPU_STAT(tlb_flush) },
 118         { "invlpg", VCPU_STAT(invlpg) },
 119         { "exits", VCPU_STAT(exits) },
 120         { "io_exits", VCPU_STAT(io_exits) },
 121         { "mmio_exits", VCPU_STAT(mmio_exits) },
 122         { "signal_exits", VCPU_STAT(signal_exits) },
 123         { "irq_window", VCPU_STAT(irq_window_exits) },
 124         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 125         { "halt_exits", VCPU_STAT(halt_exits) },
 126         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 127         { "hypercalls", VCPU_STAT(hypercalls) },
 128         { "request_irq", VCPU_STAT(request_irq_exits) },
 129         { "irq_exits", VCPU_STAT(irq_exits) },
 130         { "host_state_reload", VCPU_STAT(host_state_reload) },
 131         { "efer_reload", VCPU_STAT(efer_reload) },
 132         { "fpu_reload", VCPU_STAT(fpu_reload) },
 133         { "insn_emulation", VCPU_STAT(insn_emulation) },
 134         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 135         { "irq_injections", VCPU_STAT(irq_injections) },
 136         { "nmi_injections", VCPU_STAT(nmi_injections) },
 137         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 138         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 139         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 140         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 141         { "mmu_flooded", VM_STAT(mmu_flooded) },
 142         { "mmu_recycled", VM_STAT(mmu_recycled) },
 143         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 144         { "mmu_unsync", VM_STAT(mmu_unsync) },
 145         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 146         { "largepages", VM_STAT(lpages) },
 147         { NULL }
 148 };
 149
 150 static void kvm_on_user_return(struct user_return_notifier *urn)
 151 {
 152         unsigned slot;
 153         struct kvm_shared_msrs *locals
 154                 = container_of(urn, struct kvm_shared_msrs, urn);
 155         struct kvm_shared_msr_values *values;
 156
 157         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 158                 values = &locals->values[slot];
 159                 if (values->host != values->curr) {
 160                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 161                         values->curr = values->host;
 162                 }
 163         }
 164         locals->registered = false;
 165         user_return_notifier_unregister(urn);
 166 }
 167
 168 static void shared_msr_update(unsigned slot, u32 msr)
 169 {
 170         struct kvm_shared_msrs *smsr;
 171         u64 value;
 172
 173         smsr = &__get_cpu_var(shared_msrs);
 174         /* only read, and nobody should modify it at this time,
 175          * so don't need lock */
 176         if (slot >= shared_msrs_global.nr) {
 177                 printk(KERN_ERR "kvm: invalid MSR slot!");
 178                 return;
 179         }
 180         rdmsrl_safe(msr, &value);
 181         smsr->values[slot].host = value;
 182         smsr->values[slot].curr = value;
 183 }
 184
 185 void kvm_define_shared_msr(unsigned slot, u32 msr)
 186 {
 187         if (slot >= shared_msrs_global.nr)
 188                 shared_msrs_global.nr = slot + 1;
 189         shared_msrs_global.msrs[slot] = msr;
 190         /* we need ensured the shared_msr_global have been updated */
 191         smp_wmb();
 192 }
 193 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 194
 195 static void kvm_shared_msr_cpu_online(void)
 196 {
 197         unsigned i;
 198
 199         for (i = 0; i < shared_msrs_global.nr; ++i)
 200                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 201 }
 202
 203 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 204 {
 205         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 206
 207         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 208                 return;
 209         smsr->values[slot].curr = value;
 210         wrmsrl(shared_msrs_global.msrs[slot], value);
 211         if (!smsr->registered) {
 212                 smsr->urn.on_user_return = kvm_on_user_return;
 213                 user_return_notifier_register(&smsr->urn);
 214                 smsr->registered = true;
 215         }
 216 }
 217 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 218
 219 static void drop_user_return_notifiers(void *ignore)
 220 {
 221         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 222
 223         if (smsr->registered)
 224                 kvm_on_user_return(&smsr->urn);
 225 }
 226
 227 unsigned long segment_base(u16 selector)
 228 {
 229         struct descriptor_table gdt;
 230         struct desc_struct *d;
 231         unsigned long table_base;
 232         unsigned long v;
 233
 234         if (selector == 0)
 235                 return 0;
 236
 237         kvm_get_gdt(&gdt);
 238         table_base = gdt.base;
 239
 240         if (selector & 4) {           /* from ldt */
 241                 u16 ldt_selector = kvm_read_ldt();
 242
 243                 table_base = segment_base(ldt_selector);
 244         }
 245         d = (struct desc_struct *)(table_base + (selector & ~7));
 246         v = get_desc_base(d);
 247 #ifdef CONFIG_X86_64
 248         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 249                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 250 #endif
 251         return v;
 252 }
 253 EXPORT_SYMBOL_GPL(segment_base);
 254
 255 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 256 {
 257         if (irqchip_in_kernel(vcpu->kvm))
 258                 return vcpu->arch.apic_base;
 259         else
 260                 return vcpu->arch.apic_base;
 261 }
 262 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 263
 264 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 265 {
 266         /* TODO: reserve bits check */
 267         if (irqchip_in_kernel(vcpu->kvm))
 268                 kvm_lapic_set_base(vcpu, data);
 269         else
 270                 vcpu->arch.apic_base = data;
 271 }
 272 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 273
 274 #define EXCPT_BENIGN            0
 275 #define EXCPT_CONTRIBUTORY      1
 276 #define EXCPT_PF                2
 277
 278 static int exception_class(int vector)
 279 {
 280         switch (vector) {
 281         case PF_VECTOR:
 282                 return EXCPT_PF;
 283         case DE_VECTOR:
 284         case TS_VECTOR:
 285         case NP_VECTOR:
 286         case SS_VECTOR:
 287         case GP_VECTOR:
 288                 return EXCPT_CONTRIBUTORY;
 289         default:
 290                 break;
 291         }
 292         return EXCPT_BENIGN;
 293 }
 294
 295 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 296                 unsigned nr, bool has_error, u32 error_code)
 297 {
 298         u32 prev_nr;
 299         int class1, class2;
 300
 301         if (!vcpu->arch.exception.pending) {
 302         queue:
 303                 vcpu->arch.exception.pending = true;
 304                 vcpu->arch.exception.has_error_code = has_error;
 305                 vcpu->arch.exception.nr = nr;
 306                 vcpu->arch.exception.error_code = error_code;
 307                 return;
 308         }
 309
 310         /* to check exception */
 311         prev_nr = vcpu->arch.exception.nr;
 312         if (prev_nr == DF_VECTOR) {
 313                 /* triple fault -> shutdown */
 314                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 315                 return;
 316         }
 317         class1 = exception_class(prev_nr);
 318         class2 = exception_class(nr);
 319         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 320                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 321                 /* generate double fault per SDM Table 5-5 */
 322                 vcpu->arch.exception.pending = true;
 323                 vcpu->arch.exception.has_error_code = true;
 324                 vcpu->arch.exception.nr = DF_VECTOR;
 325                 vcpu->arch.exception.error_code = 0;
 326         } else
 327                 /* replace previous exception with a new one in a hope
 328                    that instruction re-execution will regenerate lost
 329                    exception */
 330                 goto queue;
 331 }
 332
 333 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 334 {
 335         kvm_multiple_exception(vcpu, nr, false, 0);
 336 }
 337 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 338
 339 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 340                            u32 error_code)
 341 {
 342         ++vcpu->stat.pf_guest;
 343         vcpu->arch.cr2 = addr;
 344         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 345 }
 346
 347 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 348 {
 349         vcpu->arch.nmi_pending = 1;
 350 }
 351 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 352
 353 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 354 {
 355         kvm_multiple_exception(vcpu, nr, true, error_code);
 356 }
 357 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 358
 359 /*
 360  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 361  * a #GP and return false.
 362  */
 363 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 364 {
 365         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 366                 return true;
 367         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 368         return false;
 369 }
 370 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 371
 372 /*
 373  * Load the pae pdptrs.  Return true is they are all valid.
 374  */
 375 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 376 {
 377         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 378         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 379         int i;
 380         int ret;
 381         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 382
 383         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 384                                   offset * sizeof(u64), sizeof(pdpte));
 385         if (ret < 0) {
 386                 ret = 0;
 387                 goto out;
 388         }
 389         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 390                 if (is_present_gpte(pdpte[i]) &&
 391                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 392                         ret = 0;
 393                         goto out;
 394                 }
 395         }
 396         ret = 1;
 397
 398         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 399         __set_bit(VCPU_EXREG_PDPTR,
 400                   (unsigned long *)&vcpu->arch.regs_avail);
 401         __set_bit(VCPU_EXREG_PDPTR,
 402                   (unsigned long *)&vcpu->arch.regs_dirty);
 403 out:
 404
 405         return ret;
 406 }
 407 EXPORT_SYMBOL_GPL(load_pdptrs);
 408
 409 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 410 {
 411         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 412         bool changed = true;
 413         int r;
 414
 415         if (is_long_mode(vcpu) || !is_pae(vcpu))
 416                 return false;
 417
 418         if (!test_bit(VCPU_EXREG_PDPTR,
 419                       (unsigned long *)&vcpu->arch.regs_avail))
 420                 return true;
 421
 422         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 423         if (r < 0)
 424                 goto out;
 425         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 426 out:
 427
 428         return changed;
 429 }
 430
 431 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 432 {
 433         cr0 |= X86_CR0_ET;
 434
 435 #ifdef CONFIG_X86_64
 436         if (cr0 & 0xffffffff00000000UL) {
 437                 kvm_inject_gp(vcpu, 0);
 438                 return;
 439         }
 440 #endif
 441
 442         cr0 &= ~CR0_RESERVED_BITS;
 443
 444         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 445                 kvm_inject_gp(vcpu, 0);
 446                 return;
 447         }
 448
 449         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 450                 kvm_inject_gp(vcpu, 0);
 451                 return;
 452         }
 453
 454         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 455 #ifdef CONFIG_X86_64
 456                 if ((vcpu->arch.efer & EFER_LME)) {
 457                         int cs_db, cs_l;
 458
 459                         if (!is_pae(vcpu)) {
 460                                 kvm_inject_gp(vcpu, 0);
 461                                 return;
 462                         }
 463                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 464                         if (cs_l) {
 465                                 kvm_inject_gp(vcpu, 0);
 466                                 return;
 467
 468                         }
 469                 } else
 470 #endif
 471                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 472                         kvm_inject_gp(vcpu, 0);
 473                         return;
 474                 }
 475
 476         }
 477
 478         kvm_x86_ops->set_cr0(vcpu, cr0);
 479         vcpu->arch.cr0 = cr0;
 480
 481         kvm_mmu_reset_context(vcpu);
 482         return;
 483 }
 484 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 485
 486 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 487 {
 488         kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
 489 }
 490 EXPORT_SYMBOL_GPL(kvm_lmsw);
 491
 492 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 493 {
 494         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 495         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 496
 497         if (cr4 & CR4_RESERVED_BITS) {
 498                 kvm_inject_gp(vcpu, 0);
 499                 return;
 500         }
 501
 502         if (is_long_mode(vcpu)) {
 503                 if (!(cr4 & X86_CR4_PAE)) {
 504                         kvm_inject_gp(vcpu, 0);
 505                         return;
 506                 }
 507         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 508                    && ((cr4 ^ old_cr4) & pdptr_bits)
 509                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 510                 kvm_inject_gp(vcpu, 0);
 511                 return;
 512         }
 513
 514         if (cr4 & X86_CR4_VMXE) {
 515                 kvm_inject_gp(vcpu, 0);
 516                 return;
 517         }
 518         kvm_x86_ops->set_cr4(vcpu, cr4);
 519         vcpu->arch.cr4 = cr4;
 520         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 521         kvm_mmu_reset_context(vcpu);
 522 }
 523 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 524
 525 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 526 {
 527         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 528                 kvm_mmu_sync_roots(vcpu);
 529                 kvm_mmu_flush_tlb(vcpu);
 530                 return;
 531         }
 532
 533         if (is_long_mode(vcpu)) {
 534                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 535                         kvm_inject_gp(vcpu, 0);
 536                         return;
 537                 }
 538         } else {
 539                 if (is_pae(vcpu)) {
 540                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 541                                 kvm_inject_gp(vcpu, 0);
 542                                 return;
 543                         }
 544                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 545                                 kvm_inject_gp(vcpu, 0);
 546                                 return;
 547                         }
 548                 }
 549                 /*
 550                  * We don't check reserved bits in nonpae mode, because
 551                  * this isn't enforced, and VMware depends on this.
 552                  */
 553         }
 554
 555         /*
 556          * Does the new cr3 value map to physical memory? (Note, we
 557          * catch an invalid cr3 even in real-mode, because it would
 558          * cause trouble later on when we turn on paging anyway.)
 559          *
 560          * A real CPU would silently accept an invalid cr3 and would
 561          * attempt to use it - with largely undefined (and often hard
 562          * to debug) behavior on the guest side.
 563          */
 564         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 565                 kvm_inject_gp(vcpu, 0);
 566         else {
 567                 vcpu->arch.cr3 = cr3;
 568                 vcpu->arch.mmu.new_cr3(vcpu);
 569         }
 570 }
 571 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 572
 573 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 574 {
 575         if (cr8 & CR8_RESERVED_BITS) {
 576                 kvm_inject_gp(vcpu, 0);
 577                 return;
 578         }
 579         if (irqchip_in_kernel(vcpu->kvm))
 580                 kvm_lapic_set_tpr(vcpu, cr8);
 581         else
 582                 vcpu->arch.cr8 = cr8;
 583 }
 584 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 585
 586 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 587 {
 588         if (irqchip_in_kernel(vcpu->kvm))
 589                 return kvm_lapic_get_cr8(vcpu);
 590         else
 591                 return vcpu->arch.cr8;
 592 }
 593 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 594
 595 static inline u32 bit(int bitno)
 596 {
 597         return 1 << (bitno & 31);
 598 }
 599
 600 /*
 601  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 602  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 603  *
 604  * This list is modified at module load time to reflect the
 605  * capabilities of the host cpu. This capabilities test skips MSRs that are
 606  * kvm-specific. Those are put in the beginning of the list.
 607  */
 608
 609 #define KVM_SAVE_MSRS_BEGIN     5
 610 static u32 msrs_to_save[] = {
 611         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 612         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 613         HV_X64_MSR_APIC_ASSIST_PAGE,
 614         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 615         MSR_K6_STAR,
 616 #ifdef CONFIG_X86_64
 617         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 618 #endif
 619         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 620 };
 621
 622 static unsigned num_msrs_to_save;
 623
 624 static u32 emulated_msrs[] = {
 625         MSR_IA32_MISC_ENABLE,
 626 };
 627
 628 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 629 {
 630         if (efer & efer_reserved_bits) {
 631                 kvm_inject_gp(vcpu, 0);
 632                 return;
 633         }
 634
 635         if (is_paging(vcpu)
 636             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) {
 637                 kvm_inject_gp(vcpu, 0);
 638                 return;
 639         }
 640
 641         if (efer & EFER_FFXSR) {
 642                 struct kvm_cpuid_entry2 *feat;
 643
 644                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 645                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 646                         kvm_inject_gp(vcpu, 0);
 647                         return;
 648                 }
 649         }
 650
 651         if (efer & EFER_SVME) {
 652                 struct kvm_cpuid_entry2 *feat;
 653
 654                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 655                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 656                         kvm_inject_gp(vcpu, 0);
 657                         return;
 658                 }
 659         }
 660
 661         kvm_x86_ops->set_efer(vcpu, efer);
 662
 663         efer &= ~EFER_LMA;
 664         efer |= vcpu->arch.efer & EFER_LMA;
 665
 666         vcpu->arch.efer = efer;
 667
 668         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 669         kvm_mmu_reset_context(vcpu);
 670 }
 671
 672 void kvm_enable_efer_bits(u64 mask)
 673 {
 674        efer_reserved_bits &= ~mask;
 675 }
 676 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 677
 678
 679 /*
 680  * Writes msr value into into the appropriate "register".
 681  * Returns 0 on success, non-0 otherwise.
 682  * Assumes vcpu_load() was already called.
 683  */
 684 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 685 {
 686         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 687 }
 688
 689 /*
 690  * Adapt set_msr() to msr_io()'s calling convention
 691  */
 692 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 693 {
 694         return kvm_set_msr(vcpu, index, *data);
 695 }
 696
 697 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 698 {
 699         static int version;
 700         struct pvclock_wall_clock wc;
 701         struct timespec boot;
 702
 703         if (!wall_clock)
 704                 return;
 705
 706         version++;
 707
 708         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 709
 710         /*
 711          * The guest calculates current wall clock time by adding
 712          * system time (updated by kvm_write_guest_time below) to the
 713          * wall clock specified here.  guest system time equals host
 714          * system time for us, thus we must fill in host boot time here.
 715          */
 716         getboottime(&boot);
 717
 718         wc.sec = boot.tv_sec;
 719         wc.nsec = boot.tv_nsec;
 720         wc.version = version;
 721
 722         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 723
 724         version++;
 725         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 726 }
 727
 728 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 729 {
 730         uint32_t quotient, remainder;
 731
 732         /* Don't try to replace with do_div(), this one calculates
 733          * "(dividend << 32) / divisor" */
 734         __asm__ ( "divl %4"
 735                   : "=a" (quotient), "=d" (remainder)
 736                   : "0" (0), "1" (dividend), "r" (divisor) );
 737         return quotient;
 738 }
 739
 740 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 741 {
 742         uint64_t nsecs = 1000000000LL;
 743         int32_t  shift = 0;
 744         uint64_t tps64;
 745         uint32_t tps32;
 746
 747         tps64 = tsc_khz * 1000LL;
 748         while (tps64 > nsecs*2) {
 749                 tps64 >>= 1;
 750                 shift--;
 751         }
 752
 753         tps32 = (uint32_t)tps64;
 754         while (tps32 <= (uint32_t)nsecs) {
 755                 tps32 <<= 1;
 756                 shift++;
 757         }
 758
 759         hv_clock->tsc_shift = shift;
 760         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 761
 762         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 763                  __func__, tsc_khz, hv_clock->tsc_shift,
 764                  hv_clock->tsc_to_system_mul);
 765 }
 766
 767 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 768
 769 static void kvm_write_guest_time(struct kvm_vcpu *v)
 770 {
 771         struct timespec ts;
 772         unsigned long flags;
 773         struct kvm_vcpu_arch *vcpu = &v->arch;
 774         void *shared_kaddr;
 775         unsigned long this_tsc_khz;
 776
 777         if ((!vcpu->time_page))
 778                 return;
 779
 780         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 781         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 782                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 783                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 784         }
 785         put_cpu_var(cpu_tsc_khz);
 786
 787         /* Keep irq disabled to prevent changes to the clock */
 788         local_irq_save(flags);
 789         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 790         ktime_get_ts(&ts);
 791         monotonic_to_bootbased(&ts);
 792         local_irq_restore(flags);
 793
 794         /* With all the info we got, fill in the values */
 795
 796         vcpu->hv_clock.system_time = ts.tv_nsec +
 797                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 798
 799         /*
 800          * The interface expects us to write an even number signaling that the
 801          * update is finished. Since the guest won't see the intermediate
 802          * state, we just increase by 2 at the end.
 803          */
 804         vcpu->hv_clock.version += 2;
 805
 806         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 807
 808         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 809                sizeof(vcpu->hv_clock));
 810
 811         kunmap_atomic(shared_kaddr, KM_USER0);
 812
 813         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 814 }
 815
 816 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 817 {
 818         struct kvm_vcpu_arch *vcpu = &v->arch;
 819
 820         if (!vcpu->time_page)
 821                 return 0;
 822         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 823         return 1;
 824 }
 825
 826 static bool msr_mtrr_valid(unsigned msr)
 827 {
 828         switch (msr) {
 829         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 830         case MSR_MTRRfix64K_00000:
 831         case MSR_MTRRfix16K_80000:
 832         case MSR_MTRRfix16K_A0000:
 833         case MSR_MTRRfix4K_C0000:
 834         case MSR_MTRRfix4K_C8000:
 835         case MSR_MTRRfix4K_D0000:
 836         case MSR_MTRRfix4K_D8000:
 837         case MSR_MTRRfix4K_E0000:
 838         case MSR_MTRRfix4K_E8000:
 839         case MSR_MTRRfix4K_F0000:
 840         case MSR_MTRRfix4K_F8000:
 841         case MSR_MTRRdefType:
 842         case MSR_IA32_CR_PAT:
 843                 return true;
 844         case 0x2f8:
 845                 return true;
 846         }
 847         return false;
 848 }
 849
 850 static bool valid_pat_type(unsigned t)
 851 {
 852         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 853 }
 854
 855 static bool valid_mtrr_type(unsigned t)
 856 {
 857         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 858 }
 859
 860 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 861 {
 862         int i;
 863
 864         if (!msr_mtrr_valid(msr))
 865                 return false;
 866
 867         if (msr == MSR_IA32_CR_PAT) {
 868                 for (i = 0; i < 8; i++)
 869                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 870                                 return false;
 871                 return true;
 872         } else if (msr == MSR_MTRRdefType) {
 873                 if (data & ~0xcff)
 874                         return false;
 875                 return valid_mtrr_type(data & 0xff);
 876         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 877                 for (i = 0; i < 8 ; i++)
 878                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 879                                 return false;
 880                 return true;
 881         }
 882
 883         /* variable MTRRs */
 884         return valid_mtrr_type(data & 0xff);
 885 }
 886
 887 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 888 {
 889         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 890
 891         if (!mtrr_valid(vcpu, msr, data))
 892                 return 1;
 893
 894         if (msr == MSR_MTRRdefType) {
 895                 vcpu->arch.mtrr_state.def_type = data;
 896                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 897         } else if (msr == MSR_MTRRfix64K_00000)
 898                 p[0] = data;
 899         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 900                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 901         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 902                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 903         else if (msr == MSR_IA32_CR_PAT)
 904                 vcpu->arch.pat = data;
 905         else {  /* Variable MTRRs */
 906                 int idx, is_mtrr_mask;
 907                 u64 *pt;
 908
 909                 idx = (msr - 0x200) / 2;
 910                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 911                 if (!is_mtrr_mask)
 912                         pt =
 913                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 914                 else
 915                         pt =
 916                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 917                 *pt = data;
 918         }
 919
 920         kvm_mmu_reset_context(vcpu);
 921         return 0;
 922 }
 923
 924 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 925 {
 926         u64 mcg_cap = vcpu->arch.mcg_cap;
 927         unsigned bank_num = mcg_cap & 0xff;
 928
 929         switch (msr) {
 930         case MSR_IA32_MCG_STATUS:
 931                 vcpu->arch.mcg_status = data;
 932                 break;
 933         case MSR_IA32_MCG_CTL:
 934                 if (!(mcg_cap & MCG_CTL_P))
 935                         return 1;
 936                 if (data != 0 && data != ~(u64)0)
 937                         return -1;
 938                 vcpu->arch.mcg_ctl = data;
 939                 break;
 940         default:
 941                 if (msr >= MSR_IA32_MC0_CTL &&
 942                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 943                         u32 offset = msr - MSR_IA32_MC0_CTL;
 944                         /* only 0 or all 1s can be written to IA32_MCi_CTL
 945                          * some Linux kernels though clear bit 10 in bank 4 to
 946                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
 947                          * this to avoid an uncatched #GP in the guest
 948                          */
 949                         if ((offset & 0x3) == 0 &&
 950                             data != 0 && (data | (1 << 10)) != ~(u64)0)
 951                                 return -1;
 952                         vcpu->arch.mce_banks[offset] = data;
 953                         break;
 954                 }
 955                 return 1;
 956         }
 957         return 0;
 958 }
 959
 960 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 961 {
 962         struct kvm *kvm = vcpu->kvm;
 963         int lm = is_long_mode(vcpu);
 964         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 965                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 966         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 967                 : kvm->arch.xen_hvm_config.blob_size_32;
 968         u32 page_num = data & ~PAGE_MASK;
 969         u64 page_addr = data & PAGE_MASK;
 970         u8 *page;
 971         int r;
 972
 973         r = -E2BIG;
 974         if (page_num >= blob_size)
 975                 goto out;
 976         r = -ENOMEM;
 977         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 978         if (!page)
 979                 goto out;
 980         r = -EFAULT;
 981         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
 982                 goto out_free;
 983         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
 984                 goto out_free;
 985         r = 0;
 986 out_free:
 987         kfree(page);
 988 out:
 989         return r;
 990 }
 991
 992 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
 993 {
 994         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
 995 }
 996
 997 static bool kvm_hv_msr_partition_wide(u32 msr)
 998 {
 999         bool r = false;
1000         switch (msr) {
1001         case HV_X64_MSR_GUEST_OS_ID:
1002         case HV_X64_MSR_HYPERCALL:
1003                 r = true;
1004                 break;
1005         }
1006
1007         return r;
1008 }
1009
1010 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1011 {
1012         struct kvm *kvm = vcpu->kvm;
1013
1014         switch (msr) {
1015         case HV_X64_MSR_GUEST_OS_ID:
1016                 kvm->arch.hv_guest_os_id = data;
1017                 /* setting guest os id to zero disables hypercall page */
1018                 if (!kvm->arch.hv_guest_os_id)
1019                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1020                 break;
1021         case HV_X64_MSR_HYPERCALL: {
1022                 u64 gfn;
1023                 unsigned long addr;
1024                 u8 instructions[4];
1025
1026                 /* if guest os id is not set hypercall should remain disabled */
1027                 if (!kvm->arch.hv_guest_os_id)
1028                         break;
1029                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1030                         kvm->arch.hv_hypercall = data;
1031                         break;
1032                 }
1033                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1034                 addr = gfn_to_hva(kvm, gfn);
1035                 if (kvm_is_error_hva(addr))
1036                         return 1;
1037                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1038                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1039                 if (copy_to_user((void __user *)addr, instructions, 4))
1040                         return 1;
1041                 kvm->arch.hv_hypercall = data;
1042                 break;
1043         }
1044         default:
1045                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1046                           "data 0x%llx\n", msr, data);
1047                 return 1;
1048         }
1049         return 0;
1050 }
1051
1052 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1053 {
1054         switch (msr) {
1055         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1056                 unsigned long addr;
1057
1058                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1059                         vcpu->arch.hv_vapic = data;
1060                         break;
1061                 }
1062                 addr = gfn_to_hva(vcpu->kvm, data >>
1063                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1064                 if (kvm_is_error_hva(addr))
1065                         return 1;
1066                 if (clear_user((void __user *)addr, PAGE_SIZE))
1067                         return 1;
1068                 vcpu->arch.hv_vapic = data;
1069                 break;
1070         }
1071         case HV_X64_MSR_EOI:
1072                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1073         case HV_X64_MSR_ICR:
1074                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1075         case HV_X64_MSR_TPR:
1076                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1077         default:
1078                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1079                           "data 0x%llx\n", msr, data);
1080                 return 1;
1081         }
1082
1083         return 0;
1084 }
1085
1086 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1087 {
1088         switch (msr) {
1089         case MSR_EFER:
1090                 set_efer(vcpu, data);
1091                 break;
1092         case MSR_K7_HWCR:
1093                 data &= ~(u64)0x40;     /* ignore flush filter disable */
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_X86_ROBUST_SINGLESTEP:
1553                 r = 1;
1554                 break;
1555         case KVM_CAP_COALESCED_MMIO:
1556                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1557                 break;
1558         case KVM_CAP_VAPIC:
1559                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1560                 break;
1561         case KVM_CAP_NR_VCPUS:
1562                 r = KVM_MAX_VCPUS;
1563                 break;
1564         case KVM_CAP_NR_MEMSLOTS:
1565                 r = KVM_MEMORY_SLOTS;
1566                 break;
1567         case KVM_CAP_PV_MMU:    /* obsolete */
1568                 r = 0;
1569                 break;
1570         case KVM_CAP_IOMMU:
1571                 r = iommu_found();
1572                 break;
1573         case KVM_CAP_MCE:
1574                 r = KVM_MAX_MCE_BANKS;
1575                 break;
1576         default:
1577                 r = 0;
1578                 break;
1579         }
1580         return r;
1581
1582 }
1583
1584 long kvm_arch_dev_ioctl(struct file *filp,
1585                         unsigned int ioctl, unsigned long arg)
1586 {
1587         void __user *argp = (void __user *)arg;
1588         long r;
1589
1590         switch (ioctl) {
1591         case KVM_GET_MSR_INDEX_LIST: {
1592                 struct kvm_msr_list __user *user_msr_list = argp;
1593                 struct kvm_msr_list msr_list;
1594                 unsigned n;
1595
1596                 r = -EFAULT;
1597                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1598                         goto out;
1599                 n = msr_list.nmsrs;
1600                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1601                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1602                         goto out;
1603                 r = -E2BIG;
1604                 if (n < msr_list.nmsrs)
1605                         goto out;
1606                 r = -EFAULT;
1607                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1608                                  num_msrs_to_save * sizeof(u32)))
1609                         goto out;
1610                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1611                                  &emulated_msrs,
1612                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1613                         goto out;
1614                 r = 0;
1615                 break;
1616         }
1617         case KVM_GET_SUPPORTED_CPUID: {
1618                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1619                 struct kvm_cpuid2 cpuid;
1620
1621                 r = -EFAULT;
1622                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1623                         goto out;
1624                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1625                                                       cpuid_arg->entries);
1626                 if (r)
1627                         goto out;
1628
1629                 r = -EFAULT;
1630                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1631                         goto out;
1632                 r = 0;
1633                 break;
1634         }
1635         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1636                 u64 mce_cap;
1637
1638                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1639                 r = -EFAULT;
1640                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1641                         goto out;
1642                 r = 0;
1643                 break;
1644         }
1645         default:
1646                 r = -EINVAL;
1647         }
1648 out:
1649         return r;
1650 }
1651
1652 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1653 {
1654         kvm_x86_ops->vcpu_load(vcpu, cpu);
1655         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1656                 unsigned long khz = cpufreq_quick_get(cpu);
1657                 if (!khz)
1658                         khz = tsc_khz;
1659                 per_cpu(cpu_tsc_khz, cpu) = khz;
1660         }
1661         kvm_request_guest_time_update(vcpu);
1662 }
1663
1664 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1665 {
1666         kvm_put_guest_fpu(vcpu);
1667         kvm_x86_ops->vcpu_put(vcpu);
1668 }
1669
1670 static int is_efer_nx(void)
1671 {
1672         unsigned long long efer = 0;
1673
1674         rdmsrl_safe(MSR_EFER, &efer);
1675         return efer & EFER_NX;
1676 }
1677
1678 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1679 {
1680         int i;
1681         struct kvm_cpuid_entry2 *e, *entry;
1682
1683         entry = NULL;
1684         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1685                 e = &vcpu->arch.cpuid_entries[i];
1686                 if (e->function == 0x80000001) {
1687                         entry = e;
1688                         break;
1689                 }
1690         }
1691         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1692                 entry->edx &= ~(1 << 20);
1693                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1694         }
1695 }
1696
1697 /* when an old userspace process fills a new kernel module */
1698 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1699                                     struct kvm_cpuid *cpuid,
1700                                     struct kvm_cpuid_entry __user *entries)
1701 {
1702         int r, i;
1703         struct kvm_cpuid_entry *cpuid_entries;
1704
1705         r = -E2BIG;
1706         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1707                 goto out;
1708         r = -ENOMEM;
1709         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1710         if (!cpuid_entries)
1711                 goto out;
1712         r = -EFAULT;
1713         if (copy_from_user(cpuid_entries, entries,
1714                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1715                 goto out_free;
1716         for (i = 0; i < cpuid->nent; i++) {
1717                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1718                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1719                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1720                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1721                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1722                 vcpu->arch.cpuid_entries[i].index = 0;
1723                 vcpu->arch.cpuid_entries[i].flags = 0;
1724                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1725                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1726                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1727         }
1728         vcpu->arch.cpuid_nent = cpuid->nent;
1729         cpuid_fix_nx_cap(vcpu);
1730         r = 0;
1731         kvm_apic_set_version(vcpu);
1732         kvm_x86_ops->cpuid_update(vcpu);
1733
1734 out_free:
1735         vfree(cpuid_entries);
1736 out:
1737         return r;
1738 }
1739
1740 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1741                                      struct kvm_cpuid2 *cpuid,
1742                                      struct kvm_cpuid_entry2 __user *entries)
1743 {
1744         int r;
1745
1746         r = -E2BIG;
1747         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1748                 goto out;
1749         r = -EFAULT;
1750         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1751                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1752                 goto out;
1753         vcpu->arch.cpuid_nent = cpuid->nent;
1754         kvm_apic_set_version(vcpu);
1755         kvm_x86_ops->cpuid_update(vcpu);
1756         return 0;
1757
1758 out:
1759         return r;
1760 }
1761
1762 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1763                                      struct kvm_cpuid2 *cpuid,
1764                                      struct kvm_cpuid_entry2 __user *entries)
1765 {
1766         int r;
1767
1768         r = -E2BIG;
1769         if (cpuid->nent < vcpu->arch.cpuid_nent)
1770                 goto out;
1771         r = -EFAULT;
1772         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1773                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1774                 goto out;
1775         return 0;
1776
1777 out:
1778         cpuid->nent = vcpu->arch.cpuid_nent;
1779         return r;
1780 }
1781
1782 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1783                            u32 index)
1784 {
1785         entry->function = function;
1786         entry->index = index;
1787         cpuid_count(entry->function, entry->index,
1788                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1789         entry->flags = 0;
1790 }
1791
1792 #define F(x) bit(X86_FEATURE_##x)
1793
1794 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1795                          u32 index, int *nent, int maxnent)
1796 {
1797         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1798 #ifdef CONFIG_X86_64
1799         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1800                                 ? F(GBPAGES) : 0;
1801         unsigned f_lm = F(LM);
1802 #else
1803         unsigned f_gbpages = 0;
1804         unsigned f_lm = 0;
1805 #endif
1806         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1807
1808         /* cpuid 1.edx */
1809         const u32 kvm_supported_word0_x86_features =
1810                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1811                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1812                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1813                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1814                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1815                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1816                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1817                 0 /* HTT, TM, Reserved, PBE */;
1818         /* cpuid 0x80000001.edx */
1819         const u32 kvm_supported_word1_x86_features =
1820                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1821                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1822                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1823                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1824                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1825                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1826                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1827                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1828         /* cpuid 1.ecx */
1829         const u32 kvm_supported_word4_x86_features =
1830                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1831                 0 /* DS-CPL, VMX, SMX, EST */ |
1832                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1833                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1834                 0 /* Reserved, DCA */ | F(XMM4_1) |
1835                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1836                 0 /* Reserved, XSAVE, OSXSAVE */;
1837         /* cpuid 0x80000001.ecx */
1838         const u32 kvm_supported_word6_x86_features =
1839                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1840                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1841                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1842                 0 /* SKINIT */ | 0 /* WDT */;
1843
1844         /* all calls to cpuid_count() should be made on the same cpu */
1845         get_cpu();
1846         do_cpuid_1_ent(entry, function, index);
1847         ++*nent;
1848
1849         switch (function) {
1850         case 0:
1851                 entry->eax = min(entry->eax, (u32)0xb);
1852                 break;
1853         case 1:
1854                 entry->edx &= kvm_supported_word0_x86_features;
1855                 entry->ecx &= kvm_supported_word4_x86_features;
1856                 /* we support x2apic emulation even if host does not support
1857                  * it since we emulate x2apic in software */
1858                 entry->ecx |= F(X2APIC);
1859                 break;
1860         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1861          * may return different values. This forces us to get_cpu() before
1862          * issuing the first command, and also to emulate this annoying behavior
1863          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1864         case 2: {
1865                 int t, times = entry->eax & 0xff;
1866
1867                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1868                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1869                 for (t = 1; t < times && *nent < maxnent; ++t) {
1870                         do_cpuid_1_ent(&entry[t], function, 0);
1871                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1872                         ++*nent;
1873                 }
1874                 break;
1875         }
1876         /* function 4 and 0xb have additional index. */
1877         case 4: {
1878                 int i, cache_type;
1879
1880                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1881                 /* read more entries until cache_type is zero */
1882                 for (i = 1; *nent < maxnent; ++i) {
1883                         cache_type = entry[i - 1].eax & 0x1f;
1884                         if (!cache_type)
1885                                 break;
1886                         do_cpuid_1_ent(&entry[i], function, i);
1887                         entry[i].flags |=
1888                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1889                         ++*nent;
1890                 }
1891                 break;
1892         }
1893         case 0xb: {
1894                 int i, level_type;
1895
1896                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1897                 /* read more entries until level_type is zero */
1898                 for (i = 1; *nent < maxnent; ++i) {
1899                         level_type = entry[i - 1].ecx & 0xff00;
1900                         if (!level_type)
1901                                 break;
1902                         do_cpuid_1_ent(&entry[i], function, i);
1903                         entry[i].flags |=
1904                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1905                         ++*nent;
1906                 }
1907                 break;
1908         }
1909         case 0x80000000:
1910                 entry->eax = min(entry->eax, 0x8000001a);
1911                 break;
1912         case 0x80000001:
1913                 entry->edx &= kvm_supported_word1_x86_features;
1914                 entry->ecx &= kvm_supported_word6_x86_features;
1915                 break;
1916         }
1917         put_cpu();
1918 }
1919
1920 #undef F
1921
1922 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1923                                      struct kvm_cpuid_entry2 __user *entries)
1924 {
1925         struct kvm_cpuid_entry2 *cpuid_entries;
1926         int limit, nent = 0, r = -E2BIG;
1927         u32 func;
1928
1929         if (cpuid->nent < 1)
1930                 goto out;
1931         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1932                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1933         r = -ENOMEM;
1934         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1935         if (!cpuid_entries)
1936                 goto out;
1937
1938         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1939         limit = cpuid_entries[0].eax;
1940         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1941                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1942                              &nent, cpuid->nent);
1943         r = -E2BIG;
1944         if (nent >= cpuid->nent)
1945                 goto out_free;
1946
1947         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1948         limit = cpuid_entries[nent - 1].eax;
1949         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1950                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1951                              &nent, cpuid->nent);
1952         r = -E2BIG;
1953         if (nent >= cpuid->nent)
1954                 goto out_free;
1955
1956         r = -EFAULT;
1957         if (copy_to_user(entries, cpuid_entries,
1958                          nent * sizeof(struct kvm_cpuid_entry2)))
1959                 goto out_free;
1960         cpuid->nent = nent;
1961         r = 0;
1962
1963 out_free:
1964         vfree(cpuid_entries);
1965 out:
1966         return r;
1967 }
1968
1969 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1970                                     struct kvm_lapic_state *s)
1971 {
1972         vcpu_load(vcpu);
1973         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1974         vcpu_put(vcpu);
1975
1976         return 0;
1977 }
1978
1979 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1980                                     struct kvm_lapic_state *s)
1981 {
1982         vcpu_load(vcpu);
1983         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1984         kvm_apic_post_state_restore(vcpu);
1985         update_cr8_intercept(vcpu);
1986         vcpu_put(vcpu);
1987
1988         return 0;
1989 }
1990
1991 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1992                                     struct kvm_interrupt *irq)
1993 {
1994         if (irq->irq < 0 || irq->irq >= 256)
1995                 return -EINVAL;
1996         if (irqchip_in_kernel(vcpu->kvm))
1997                 return -ENXIO;
1998         vcpu_load(vcpu);
1999
2000         kvm_queue_interrupt(vcpu, irq->irq, false);
2001
2002         vcpu_put(vcpu);
2003
2004         return 0;
2005 }
2006
2007 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2008 {
2009         vcpu_load(vcpu);
2010         kvm_inject_nmi(vcpu);
2011         vcpu_put(vcpu);
2012
2013         return 0;
2014 }
2015
2016 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2017                                            struct kvm_tpr_access_ctl *tac)
2018 {
2019         if (tac->flags)
2020                 return -EINVAL;
2021         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2022         return 0;
2023 }
2024
2025 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2026                                         u64 mcg_cap)
2027 {
2028         int r;
2029         unsigned bank_num = mcg_cap & 0xff, bank;
2030
2031         r = -EINVAL;
2032         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2033                 goto out;
2034         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2035                 goto out;
2036         r = 0;
2037         vcpu->arch.mcg_cap = mcg_cap;
2038         /* Init IA32_MCG_CTL to all 1s */
2039         if (mcg_cap & MCG_CTL_P)
2040                 vcpu->arch.mcg_ctl = ~(u64)0;
2041         /* Init IA32_MCi_CTL to all 1s */
2042         for (bank = 0; bank < bank_num; bank++)
2043                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2044 out:
2045         return r;
2046 }
2047
2048 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2049                                       struct kvm_x86_mce *mce)
2050 {
2051         u64 mcg_cap = vcpu->arch.mcg_cap;
2052         unsigned bank_num = mcg_cap & 0xff;
2053         u64 *banks = vcpu->arch.mce_banks;
2054
2055         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2056                 return -EINVAL;
2057         /*
2058          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2059          * reporting is disabled
2060          */
2061         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2062             vcpu->arch.mcg_ctl != ~(u64)0)
2063                 return 0;
2064         banks += 4 * mce->bank;
2065         /*
2066          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2067          * reporting is disabled for the bank
2068          */
2069         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2070                 return 0;
2071         if (mce->status & MCI_STATUS_UC) {
2072                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2073                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2074                         printk(KERN_DEBUG "kvm: set_mce: "
2075                                "injects mce exception while "
2076                                "previous one is in progress!\n");
2077                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2078                         return 0;
2079                 }
2080                 if (banks[1] & MCI_STATUS_VAL)
2081                         mce->status |= MCI_STATUS_OVER;
2082                 banks[2] = mce->addr;
2083                 banks[3] = mce->misc;
2084                 vcpu->arch.mcg_status = mce->mcg_status;
2085                 banks[1] = mce->status;
2086                 kvm_queue_exception(vcpu, MC_VECTOR);
2087         } else if (!(banks[1] & MCI_STATUS_VAL)
2088                    || !(banks[1] & MCI_STATUS_UC)) {
2089                 if (banks[1] & MCI_STATUS_VAL)
2090                         mce->status |= MCI_STATUS_OVER;
2091                 banks[2] = mce->addr;
2092                 banks[3] = mce->misc;
2093                 banks[1] = mce->status;
2094         } else
2095                 banks[1] |= MCI_STATUS_OVER;
2096         return 0;
2097 }
2098
2099 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2100                                                struct kvm_vcpu_events *events)
2101 {
2102         vcpu_load(vcpu);
2103
2104         events->exception.injected = vcpu->arch.exception.pending;
2105         events->exception.nr = vcpu->arch.exception.nr;
2106         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2107         events->exception.error_code = vcpu->arch.exception.error_code;
2108
2109         events->interrupt.injected = vcpu->arch.interrupt.pending;
2110         events->interrupt.nr = vcpu->arch.interrupt.nr;
2111         events->interrupt.soft = vcpu->arch.interrupt.soft;
2112
2113         events->nmi.injected = vcpu->arch.nmi_injected;
2114         events->nmi.pending = vcpu->arch.nmi_pending;
2115         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2116
2117         events->sipi_vector = vcpu->arch.sipi_vector;
2118
2119         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2120                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
2121
2122         vcpu_put(vcpu);
2123 }
2124
2125 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2126                                               struct kvm_vcpu_events *events)
2127 {
2128         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2129                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
2130                 return -EINVAL;
2131
2132         vcpu_load(vcpu);
2133
2134         vcpu->arch.exception.pending = events->exception.injected;
2135         vcpu->arch.exception.nr = events->exception.nr;
2136         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2137         vcpu->arch.exception.error_code = events->exception.error_code;
2138
2139         vcpu->arch.interrupt.pending = events->interrupt.injected;
2140         vcpu->arch.interrupt.nr = events->interrupt.nr;
2141         vcpu->arch.interrupt.soft = events->interrupt.soft;
2142         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2143                 kvm_pic_clear_isr_ack(vcpu->kvm);
2144
2145         vcpu->arch.nmi_injected = events->nmi.injected;
2146         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2147                 vcpu->arch.nmi_pending = events->nmi.pending;
2148         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2149
2150         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2151                 vcpu->arch.sipi_vector = events->sipi_vector;
2152
2153         vcpu_put(vcpu);
2154
2155         return 0;
2156 }
2157
2158 long kvm_arch_vcpu_ioctl(struct file *filp,
2159                          unsigned int ioctl, unsigned long arg)
2160 {
2161         struct kvm_vcpu *vcpu = filp->private_data;
2162         void __user *argp = (void __user *)arg;
2163         int r;
2164         struct kvm_lapic_state *lapic = NULL;
2165
2166         switch (ioctl) {
2167         case KVM_GET_LAPIC: {
2168                 r = -EINVAL;
2169                 if (!vcpu->arch.apic)
2170                         goto out;
2171                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2172
2173                 r = -ENOMEM;
2174                 if (!lapic)
2175                         goto out;
2176                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2177                 if (r)
2178                         goto out;
2179                 r = -EFAULT;
2180                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2181                         goto out;
2182                 r = 0;
2183                 break;
2184         }
2185         case KVM_SET_LAPIC: {
2186                 r = -EINVAL;
2187                 if (!vcpu->arch.apic)
2188                         goto out;
2189                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2190                 r = -ENOMEM;
2191                 if (!lapic)
2192                         goto out;
2193                 r = -EFAULT;
2194                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2195                         goto out;
2196                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2197                 if (r)
2198                         goto out;
2199                 r = 0;
2200                 break;
2201         }
2202         case KVM_INTERRUPT: {
2203                 struct kvm_interrupt irq;
2204
2205                 r = -EFAULT;
2206                 if (copy_from_user(&irq, argp, sizeof irq))
2207                         goto out;
2208                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2209                 if (r)
2210                         goto out;
2211                 r = 0;
2212                 break;
2213         }
2214         case KVM_NMI: {
2215                 r = kvm_vcpu_ioctl_nmi(vcpu);
2216                 if (r)
2217                         goto out;
2218                 r = 0;
2219                 break;
2220         }
2221         case KVM_SET_CPUID: {
2222                 struct kvm_cpuid __user *cpuid_arg = argp;
2223                 struct kvm_cpuid cpuid;
2224
2225                 r = -EFAULT;
2226                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2227                         goto out;
2228                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2229                 if (r)
2230                         goto out;
2231                 break;
2232         }
2233         case KVM_SET_CPUID2: {
2234                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2235                 struct kvm_cpuid2 cpuid;
2236
2237                 r = -EFAULT;
2238                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2239                         goto out;
2240                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2241                                               cpuid_arg->entries);
2242                 if (r)
2243                         goto out;
2244                 break;
2245         }
2246         case KVM_GET_CPUID2: {
2247                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2248                 struct kvm_cpuid2 cpuid;
2249
2250                 r = -EFAULT;
2251                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2252                         goto out;
2253                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2254                                               cpuid_arg->entries);
2255                 if (r)
2256                         goto out;
2257                 r = -EFAULT;
2258                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2259                         goto out;
2260                 r = 0;
2261                 break;
2262         }
2263         case KVM_GET_MSRS:
2264                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2265                 break;
2266         case KVM_SET_MSRS:
2267                 r = msr_io(vcpu, argp, do_set_msr, 0);
2268                 break;
2269         case KVM_TPR_ACCESS_REPORTING: {
2270                 struct kvm_tpr_access_ctl tac;
2271
2272                 r = -EFAULT;
2273                 if (copy_from_user(&tac, argp, sizeof tac))
2274                         goto out;
2275                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2276                 if (r)
2277                         goto out;
2278                 r = -EFAULT;
2279                 if (copy_to_user(argp, &tac, sizeof tac))
2280                         goto out;
2281                 r = 0;
2282                 break;
2283         };
2284         case KVM_SET_VAPIC_ADDR: {
2285                 struct kvm_vapic_addr va;
2286
2287                 r = -EINVAL;
2288                 if (!irqchip_in_kernel(vcpu->kvm))
2289                         goto out;
2290                 r = -EFAULT;
2291                 if (copy_from_user(&va, argp, sizeof va))
2292                         goto out;
2293                 r = 0;
2294                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2295                 break;
2296         }
2297         case KVM_X86_SETUP_MCE: {
2298                 u64 mcg_cap;
2299
2300                 r = -EFAULT;
2301                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2302                         goto out;
2303                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2304                 break;
2305         }
2306         case KVM_X86_SET_MCE: {
2307                 struct kvm_x86_mce mce;
2308
2309                 r = -EFAULT;
2310                 if (copy_from_user(&mce, argp, sizeof mce))
2311                         goto out;
2312                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2313                 break;
2314         }
2315         case KVM_GET_VCPU_EVENTS: {
2316                 struct kvm_vcpu_events events;
2317
2318                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2319
2320                 r = -EFAULT;
2321                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2322                         break;
2323                 r = 0;
2324                 break;
2325         }
2326         case KVM_SET_VCPU_EVENTS: {
2327                 struct kvm_vcpu_events events;
2328
2329                 r = -EFAULT;
2330                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2331                         break;
2332
2333                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2334                 break;
2335         }
2336         default:
2337                 r = -EINVAL;
2338         }
2339 out:
2340         kfree(lapic);
2341         return r;
2342 }
2343
2344 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2345 {
2346         int ret;
2347
2348         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2349                 return -1;
2350         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2351         return ret;
2352 }
2353
2354 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2355                                               u64 ident_addr)
2356 {
2357         kvm->arch.ept_identity_map_addr = ident_addr;
2358         return 0;
2359 }
2360
2361 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2362                                           u32 kvm_nr_mmu_pages)
2363 {
2364         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2365                 return -EINVAL;
2366
2367         mutex_lock(&kvm->slots_lock);
2368         spin_lock(&kvm->mmu_lock);
2369
2370         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2371         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2372
2373         spin_unlock(&kvm->mmu_lock);
2374         mutex_unlock(&kvm->slots_lock);
2375         return 0;
2376 }
2377
2378 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2379 {
2380         return kvm->arch.n_alloc_mmu_pages;
2381 }
2382
2383 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2384 {
2385         int i;
2386         struct kvm_mem_alias *alias;
2387         struct kvm_mem_aliases *aliases;
2388
2389         aliases = rcu_dereference(kvm->arch.aliases);
2390
2391         for (i = 0; i < aliases->naliases; ++i) {
2392                 alias = &aliases->aliases[i];
2393                 if (alias->flags & KVM_ALIAS_INVALID)
2394                         continue;
2395                 if (gfn >= alias->base_gfn
2396                     && gfn < alias->base_gfn + alias->npages)
2397                         return alias->target_gfn + gfn - alias->base_gfn;
2398         }
2399         return gfn;
2400 }
2401
2402 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2403 {
2404         int i;
2405         struct kvm_mem_alias *alias;
2406         struct kvm_mem_aliases *aliases;
2407
2408         aliases = rcu_dereference(kvm->arch.aliases);
2409
2410         for (i = 0; i < aliases->naliases; ++i) {
2411                 alias = &aliases->aliases[i];
2412                 if (gfn >= alias->base_gfn
2413                     && gfn < alias->base_gfn + alias->npages)
2414                         return alias->target_gfn + gfn - alias->base_gfn;
2415         }
2416         return gfn;
2417 }
2418
2419 /*
2420  * Set a new alias region.  Aliases map a portion of physical memory into
2421  * another portion.  This is useful for memory windows, for example the PC
2422  * VGA region.
2423  */
2424 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2425                                          struct kvm_memory_alias *alias)
2426 {
2427         int r, n;
2428         struct kvm_mem_alias *p;
2429         struct kvm_mem_aliases *aliases, *old_aliases;
2430
2431         r = -EINVAL;
2432         /* General sanity checks */
2433         if (alias->memory_size & (PAGE_SIZE - 1))
2434                 goto out;
2435         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2436                 goto out;
2437         if (alias->slot >= KVM_ALIAS_SLOTS)
2438                 goto out;
2439         if (alias->guest_phys_addr + alias->memory_size
2440             < alias->guest_phys_addr)
2441                 goto out;
2442         if (alias->target_phys_addr + alias->memory_size
2443             < alias->target_phys_addr)
2444                 goto out;
2445
2446         r = -ENOMEM;
2447         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2448         if (!aliases)
2449                 goto out;
2450
2451         mutex_lock(&kvm->slots_lock);
2452
2453         /* invalidate any gfn reference in case of deletion/shrinking */
2454         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2455         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2456         old_aliases = kvm->arch.aliases;
2457         rcu_assign_pointer(kvm->arch.aliases, aliases);
2458         synchronize_srcu_expedited(&kvm->srcu);
2459         kvm_mmu_zap_all(kvm);
2460         kfree(old_aliases);
2461
2462         r = -ENOMEM;
2463         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2464         if (!aliases)
2465                 goto out_unlock;
2466
2467         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2468
2469         p = &aliases->aliases[alias->slot];
2470         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2471         p->npages = alias->memory_size >> PAGE_SHIFT;
2472         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2473         p->flags &= ~(KVM_ALIAS_INVALID);
2474
2475         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2476                 if (aliases->aliases[n - 1].npages)
2477                         break;
2478         aliases->naliases = n;
2479
2480         old_aliases = kvm->arch.aliases;
2481         rcu_assign_pointer(kvm->arch.aliases, aliases);
2482         synchronize_srcu_expedited(&kvm->srcu);
2483         kfree(old_aliases);
2484         r = 0;
2485
2486 out_unlock:
2487         mutex_unlock(&kvm->slots_lock);
2488 out:
2489         return r;
2490 }
2491
2492 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2493 {
2494         int r;
2495
2496         r = 0;
2497         switch (chip->chip_id) {
2498         case KVM_IRQCHIP_PIC_MASTER:
2499                 memcpy(&chip->chip.pic,
2500                         &pic_irqchip(kvm)->pics[0],
2501                         sizeof(struct kvm_pic_state));
2502                 break;
2503         case KVM_IRQCHIP_PIC_SLAVE:
2504                 memcpy(&chip->chip.pic,
2505                         &pic_irqchip(kvm)->pics[1],
2506                         sizeof(struct kvm_pic_state));
2507                 break;
2508         case KVM_IRQCHIP_IOAPIC:
2509                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2510                 break;
2511         default:
2512                 r = -EINVAL;
2513                 break;
2514         }
2515         return r;
2516 }
2517
2518 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2519 {
2520         int r;
2521
2522         r = 0;
2523         switch (chip->chip_id) {
2524         case KVM_IRQCHIP_PIC_MASTER:
2525                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2526                 memcpy(&pic_irqchip(kvm)->pics[0],
2527                         &chip->chip.pic,
2528                         sizeof(struct kvm_pic_state));
2529                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2530                 break;
2531         case KVM_IRQCHIP_PIC_SLAVE:
2532                 raw_spin_lock(&pic_irqchip(kvm)->lock);
2533                 memcpy(&pic_irqchip(kvm)->pics[1],
2534                         &chip->chip.pic,
2535                         sizeof(struct kvm_pic_state));
2536                 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2537                 break;
2538         case KVM_IRQCHIP_IOAPIC:
2539                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2540                 break;
2541         default:
2542                 r = -EINVAL;
2543                 break;
2544         }
2545         kvm_pic_update_irq(pic_irqchip(kvm));
2546         return r;
2547 }
2548
2549 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2550 {
2551         int r = 0;
2552
2553         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2554         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2555         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2556         return r;
2557 }
2558
2559 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2560 {
2561         int r = 0;
2562
2563         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2564         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2565         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2566         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2567         return r;
2568 }
2569
2570 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2571 {
2572         int r = 0;
2573
2574         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2575         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2576                 sizeof(ps->channels));
2577         ps->flags = kvm->arch.vpit->pit_state.flags;
2578         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2579         return r;
2580 }
2581
2582 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2583 {
2584         int r = 0, start = 0;
2585         u32 prev_legacy, cur_legacy;
2586         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2587         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2588         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2589         if (!prev_legacy && cur_legacy)
2590                 start = 1;
2591         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2592                sizeof(kvm->arch.vpit->pit_state.channels));
2593         kvm->arch.vpit->pit_state.flags = ps->flags;
2594         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2595         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2596         return r;
2597 }
2598
2599 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2600                                  struct kvm_reinject_control *control)
2601 {
2602         if (!kvm->arch.vpit)
2603                 return -ENXIO;
2604         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2605         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2606         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2607         return 0;
2608 }
2609
2610 /*
2611  * Get (and clear) the dirty memory log for a memory slot.
2612  */
2613 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2614                                       struct kvm_dirty_log *log)
2615 {
2616         int r, i;
2617         struct kvm_memory_slot *memslot;
2618         unsigned long n;
2619         unsigned long is_dirty = 0;
2620         unsigned long *dirty_bitmap = NULL;
2621
2622         mutex_lock(&kvm->slots_lock);
2623
2624         r = -EINVAL;
2625         if (log->slot >= KVM_MEMORY_SLOTS)
2626                 goto out;
2627
2628         memslot = &kvm->memslots->memslots[log->slot];
2629         r = -ENOENT;
2630         if (!memslot->dirty_bitmap)
2631                 goto out;
2632
2633         n = kvm_dirty_bitmap_bytes(memslot);
2634
2635         r = -ENOMEM;
2636         dirty_bitmap = vmalloc(n);
2637         if (!dirty_bitmap)
2638                 goto out;
2639         memset(dirty_bitmap, 0, n);
2640
2641         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2642                 is_dirty = memslot->dirty_bitmap[i];
2643
2644         /* If nothing is dirty, don't bother messing with page tables. */
2645         if (is_dirty) {
2646                 struct kvm_memslots *slots, *old_slots;
2647
2648                 spin_lock(&kvm->mmu_lock);
2649                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2650                 spin_unlock(&kvm->mmu_lock);
2651
2652                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2653                 if (!slots)
2654                         goto out_free;
2655
2656                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2657                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2658
2659                 old_slots = kvm->memslots;
2660                 rcu_assign_pointer(kvm->memslots, slots);
2661                 synchronize_srcu_expedited(&kvm->srcu);
2662                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2663                 kfree(old_slots);
2664         }
2665
2666         r = 0;
2667         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2668                 r = -EFAULT;
2669 out_free:
2670         vfree(dirty_bitmap);
2671 out:
2672         mutex_unlock(&kvm->slots_lock);
2673         return r;
2674 }
2675
2676 long kvm_arch_vm_ioctl(struct file *filp,
2677                        unsigned int ioctl, unsigned long arg)
2678 {
2679         struct kvm *kvm = filp->private_data;
2680         void __user *argp = (void __user *)arg;
2681         int r = -ENOTTY;
2682         /*
2683          * This union makes it completely explicit to gcc-3.x
2684          * that these two variables' stack usage should be
2685          * combined, not added together.
2686          */
2687         union {
2688                 struct kvm_pit_state ps;
2689                 struct kvm_pit_state2 ps2;
2690                 struct kvm_memory_alias alias;
2691                 struct kvm_pit_config pit_config;
2692         } u;
2693
2694         switch (ioctl) {
2695         case KVM_SET_TSS_ADDR:
2696                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2697                 if (r < 0)
2698                         goto out;
2699                 break;
2700         case KVM_SET_IDENTITY_MAP_ADDR: {
2701                 u64 ident_addr;
2702
2703                 r = -EFAULT;
2704                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2705                         goto out;
2706                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2707                 if (r < 0)
2708                         goto out;
2709                 break;
2710         }
2711         case KVM_SET_MEMORY_REGION: {
2712                 struct kvm_memory_region kvm_mem;
2713                 struct kvm_userspace_memory_region kvm_userspace_mem;
2714
2715                 r = -EFAULT;
2716                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2717                         goto out;
2718                 kvm_userspace_mem.slot = kvm_mem.slot;
2719                 kvm_userspace_mem.flags = kvm_mem.flags;
2720                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2721                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2722                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2723                 if (r)
2724                         goto out;
2725                 break;
2726         }
2727         case KVM_SET_NR_MMU_PAGES:
2728                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2729                 if (r)
2730                         goto out;
2731                 break;
2732         case KVM_GET_NR_MMU_PAGES:
2733                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2734                 break;
2735         case KVM_SET_MEMORY_ALIAS:
2736                 r = -EFAULT;
2737                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2738                         goto out;
2739                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2740                 if (r)
2741                         goto out;
2742                 break;
2743         case KVM_CREATE_IRQCHIP: {
2744                 struct kvm_pic *vpic;
2745
2746                 mutex_lock(&kvm->lock);
2747                 r = -EEXIST;
2748                 if (kvm->arch.vpic)
2749                         goto create_irqchip_unlock;
2750                 r = -ENOMEM;
2751                 vpic = kvm_create_pic(kvm);
2752                 if (vpic) {
2753                         r = kvm_ioapic_init(kvm);
2754                         if (r) {
2755                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2756                                                           &vpic->dev);
2757                                 kfree(vpic);
2758                                 goto create_irqchip_unlock;
2759                         }
2760                 } else
2761                         goto create_irqchip_unlock;
2762                 smp_wmb();
2763                 kvm->arch.vpic = vpic;
2764                 smp_wmb();
2765                 r = kvm_setup_default_irq_routing(kvm);
2766                 if (r) {
2767                         mutex_lock(&kvm->irq_lock);
2768                         kvm_ioapic_destroy(kvm);
2769                         kvm_destroy_pic(kvm);
2770                         mutex_unlock(&kvm->irq_lock);
2771                 }
2772         create_irqchip_unlock:
2773                 mutex_unlock(&kvm->lock);
2774                 break;
2775         }
2776         case KVM_CREATE_PIT:
2777                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2778                 goto create_pit;
2779         case KVM_CREATE_PIT2:
2780                 r = -EFAULT;
2781                 if (copy_from_user(&u.pit_config, argp,
2782                                    sizeof(struct kvm_pit_config)))
2783                         goto out;
2784         create_pit:
2785                 mutex_lock(&kvm->slots_lock);
2786                 r = -EEXIST;
2787                 if (kvm->arch.vpit)
2788                         goto create_pit_unlock;
2789                 r = -ENOMEM;
2790                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2791                 if (kvm->arch.vpit)
2792                         r = 0;
2793         create_pit_unlock:
2794                 mutex_unlock(&kvm->slots_lock);
2795                 break;
2796         case KVM_IRQ_LINE_STATUS:
2797         case KVM_IRQ_LINE: {
2798                 struct kvm_irq_level irq_event;
2799
2800                 r = -EFAULT;
2801                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2802                         goto out;
2803                 if (irqchip_in_kernel(kvm)) {
2804                         __s32 status;
2805                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2806                                         irq_event.irq, irq_event.level);
2807                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2808                                 irq_event.status = status;
2809                                 if (copy_to_user(argp, &irq_event,
2810                                                         sizeof irq_event))
2811                                         goto out;
2812                         }
2813                         r = 0;
2814                 }
2815                 break;
2816         }
2817         case KVM_GET_IRQCHIP: {
2818                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2819                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2820
2821                 r = -ENOMEM;
2822                 if (!chip)
2823                         goto out;
2824                 r = -EFAULT;
2825                 if (copy_from_user(chip, argp, sizeof *chip))
2826                         goto get_irqchip_out;
2827                 r = -ENXIO;
2828                 if (!irqchip_in_kernel(kvm))
2829                         goto get_irqchip_out;
2830                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2831                 if (r)
2832                         goto get_irqchip_out;
2833                 r = -EFAULT;
2834                 if (copy_to_user(argp, chip, sizeof *chip))
2835                         goto get_irqchip_out;
2836                 r = 0;
2837         get_irqchip_out:
2838                 kfree(chip);
2839                 if (r)
2840                         goto out;
2841                 break;
2842         }
2843         case KVM_SET_IRQCHIP: {
2844                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2845                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2846
2847                 r = -ENOMEM;
2848                 if (!chip)
2849                         goto out;
2850                 r = -EFAULT;
2851                 if (copy_from_user(chip, argp, sizeof *chip))
2852                         goto set_irqchip_out;
2853                 r = -ENXIO;
2854                 if (!irqchip_in_kernel(kvm))
2855                         goto set_irqchip_out;
2856                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2857                 if (r)
2858                         goto set_irqchip_out;
2859                 r = 0;
2860         set_irqchip_out:
2861                 kfree(chip);
2862                 if (r)
2863                         goto out;
2864                 break;
2865         }
2866         case KVM_GET_PIT: {
2867                 r = -EFAULT;
2868                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2869                         goto out;
2870                 r = -ENXIO;
2871                 if (!kvm->arch.vpit)
2872                         goto out;
2873                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2874                 if (r)
2875                         goto out;
2876                 r = -EFAULT;
2877                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2878                         goto out;
2879                 r = 0;
2880                 break;
2881         }
2882         case KVM_SET_PIT: {
2883                 r = -EFAULT;
2884                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2885                         goto out;
2886                 r = -ENXIO;
2887                 if (!kvm->arch.vpit)
2888                         goto out;
2889                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2890                 if (r)
2891                         goto out;
2892                 r = 0;
2893                 break;
2894         }
2895         case KVM_GET_PIT2: {
2896                 r = -ENXIO;
2897                 if (!kvm->arch.vpit)
2898                         goto out;
2899                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2900                 if (r)
2901                         goto out;
2902                 r = -EFAULT;
2903                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2904                         goto out;
2905                 r = 0;
2906                 break;
2907         }
2908         case KVM_SET_PIT2: {
2909                 r = -EFAULT;
2910                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2911                         goto out;
2912                 r = -ENXIO;
2913                 if (!kvm->arch.vpit)
2914                         goto out;
2915                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2916                 if (r)
2917                         goto out;
2918                 r = 0;
2919                 break;
2920         }
2921         case KVM_REINJECT_CONTROL: {
2922                 struct kvm_reinject_control control;
2923                 r =  -EFAULT;
2924                 if (copy_from_user(&control, argp, sizeof(control)))
2925                         goto out;
2926                 r = kvm_vm_ioctl_reinject(kvm, &control);
2927                 if (r)
2928                         goto out;
2929                 r = 0;
2930                 break;
2931         }
2932         case KVM_XEN_HVM_CONFIG: {
2933                 r = -EFAULT;
2934                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2935                                    sizeof(struct kvm_xen_hvm_config)))
2936                         goto out;
2937                 r = -EINVAL;
2938                 if (kvm->arch.xen_hvm_config.flags)
2939                         goto out;
2940                 r = 0;
2941                 break;
2942         }
2943         case KVM_SET_CLOCK: {
2944                 struct timespec now;
2945                 struct kvm_clock_data user_ns;
2946                 u64 now_ns;
2947                 s64 delta;
2948
2949                 r = -EFAULT;
2950                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2951                         goto out;
2952
2953                 r = -EINVAL;
2954                 if (user_ns.flags)
2955                         goto out;
2956
2957                 r = 0;
2958                 ktime_get_ts(&now);
2959                 now_ns = timespec_to_ns(&now);
2960                 delta = user_ns.clock - now_ns;
2961                 kvm->arch.kvmclock_offset = delta;
2962                 break;
2963         }
2964         case KVM_GET_CLOCK: {
2965                 struct timespec now;
2966                 struct kvm_clock_data user_ns;
2967                 u64 now_ns;
2968
2969                 ktime_get_ts(&now);
2970                 now_ns = timespec_to_ns(&now);
2971                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2972                 user_ns.flags = 0;
2973
2974                 r = -EFAULT;
2975                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2976                         goto out;
2977                 r = 0;
2978                 break;
2979         }
2980
2981         default:
2982                 ;
2983         }
2984 out:
2985         return r;
2986 }
2987
2988 static void kvm_init_msr_list(void)
2989 {
2990         u32 dummy[2];
2991         unsigned i, j;
2992
2993         /* skip the first msrs in the list. KVM-specific */
2994         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2995                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2996                         continue;
2997                 if (j < i)
2998                         msrs_to_save[j] = msrs_to_save[i];
2999                 j++;
3000         }
3001         num_msrs_to_save = j;
3002 }
3003
3004 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3005                            const void *v)
3006 {
3007         if (vcpu->arch.apic &&
3008             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3009                 return 0;
3010
3011         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3012 }
3013
3014 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3015 {
3016         if (vcpu->arch.apic &&
3017             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3018                 return 0;
3019
3020         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3021 }
3022
3023 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3024 {
3025         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3026         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3027 }
3028
3029  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3030 {
3031         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3032         access |= PFERR_FETCH_MASK;
3033         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3034 }
3035
3036 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3037 {
3038         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3039         access |= PFERR_WRITE_MASK;
3040         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3041 }
3042
3043 /* uses this to access any guest's mapped memory without checking CPL */
3044 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3045 {
3046         return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3047 }
3048
3049 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3050                                       struct kvm_vcpu *vcpu, u32 access,
3051                                       u32 *error)
3052 {
3053         void *data = val;
3054         int r = X86EMUL_CONTINUE;
3055
3056         while (bytes) {
3057                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3058                 unsigned offset = addr & (PAGE_SIZE-1);
3059                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3060                 int ret;
3061
3062                 if (gpa == UNMAPPED_GVA) {
3063                         r = X86EMUL_PROPAGATE_FAULT;
3064                         goto out;
3065                 }
3066                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3067                 if (ret < 0) {
3068                         r = X86EMUL_UNHANDLEABLE;
3069                         goto out;
3070                 }
3071
3072                 bytes -= toread;
3073                 data += toread;
3074                 addr += toread;
3075         }
3076 out:
3077         return r;
3078 }
3079
3080 /* used for instruction fetching */
3081 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3082                                 struct kvm_vcpu *vcpu, u32 *error)
3083 {
3084         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3085         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3086                                           access | PFERR_FETCH_MASK, error);
3087 }
3088
3089 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3090                                struct kvm_vcpu *vcpu, u32 *error)
3091 {
3092         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3093         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3094                                           error);
3095 }
3096
3097 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3098                                struct kvm_vcpu *vcpu, u32 *error)
3099 {
3100         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3101 }
3102
3103 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3104                                 struct kvm_vcpu *vcpu, u32 *error)
3105 {
3106         void *data = val;
3107         int r = X86EMUL_CONTINUE;
3108
3109         while (bytes) {
3110                 gpa_t gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error);
3111                 unsigned offset = addr & (PAGE_SIZE-1);
3112                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3113                 int ret;
3114
3115                 if (gpa == UNMAPPED_GVA) {
3116                         r = X86EMUL_PROPAGATE_FAULT;
3117                         goto out;
3118                 }
3119                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3120                 if (ret < 0) {
3121                         r = X86EMUL_UNHANDLEABLE;
3122                         goto out;
3123                 }
3124
3125                 bytes -= towrite;
3126                 data += towrite;
3127                 addr += towrite;
3128         }
3129 out:
3130         return r;
3131 }
3132
3133
3134 static int emulator_read_emulated(unsigned long addr,
3135                                   void *val,
3136                                   unsigned int bytes,
3137                                   struct kvm_vcpu *vcpu)
3138 {
3139         gpa_t                 gpa;
3140         u32 error_code;
3141
3142         if (vcpu->mmio_read_completed) {
3143                 memcpy(val, vcpu->mmio_data, bytes);
3144                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3145                                vcpu->mmio_phys_addr, *(u64 *)val);
3146                 vcpu->mmio_read_completed = 0;
3147                 return X86EMUL_CONTINUE;
3148         }
3149
3150         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, &error_code);
3151
3152         if (gpa == UNMAPPED_GVA) {
3153                 kvm_inject_page_fault(vcpu, addr, error_code);
3154                 return X86EMUL_PROPAGATE_FAULT;
3155         }
3156
3157         /* For APIC access vmexit */
3158         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3159                 goto mmio;
3160
3161         if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3162                                 == X86EMUL_CONTINUE)
3163                 return X86EMUL_CONTINUE;
3164
3165 mmio:
3166         /*
3167          * Is this MMIO handled locally?
3168          */
3169         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3170                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3171                 return X86EMUL_CONTINUE;
3172         }
3173
3174         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3175
3176         vcpu->mmio_needed = 1;
3177         vcpu->mmio_phys_addr = gpa;
3178         vcpu->mmio_size = bytes;
3179         vcpu->mmio_is_write = 0;
3180
3181         return X86EMUL_UNHANDLEABLE;
3182 }
3183
3184 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3185                           const void *val, int bytes)
3186 {
3187         int ret;
3188
3189         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3190         if (ret < 0)
3191                 return 0;
3192         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3193         return 1;
3194 }
3195
3196 static int emulator_write_emulated_onepage(unsigned long addr,
3197                                            const void *val,
3198                                            unsigned int bytes,
3199                                            struct kvm_vcpu *vcpu)
3200 {
3201         gpa_t                 gpa;
3202         u32 error_code;
3203
3204         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, &error_code);
3205
3206         if (gpa == UNMAPPED_GVA) {
3207                 kvm_inject_page_fault(vcpu, addr, error_code);
3208                 return X86EMUL_PROPAGATE_FAULT;
3209         }
3210
3211         /* For APIC access vmexit */
3212         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3213                 goto mmio;
3214
3215         if (emulator_write_phys(vcpu, gpa, val, bytes))
3216                 return X86EMUL_CONTINUE;
3217
3218 mmio:
3219         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3220         /*
3221          * Is this MMIO handled locally?
3222          */
3223         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3224                 return X86EMUL_CONTINUE;
3225
3226         vcpu->mmio_needed = 1;
3227         vcpu->mmio_phys_addr = gpa;
3228         vcpu->mmio_size = bytes;
3229         vcpu->mmio_is_write = 1;
3230         memcpy(vcpu->mmio_data, val, bytes);
3231
3232         return X86EMUL_CONTINUE;
3233 }
3234
3235 int emulator_write_emulated(unsigned long addr,
3236                                    const void *val,
3237                                    unsigned int bytes,
3238                                    struct kvm_vcpu *vcpu)
3239 {
3240         /* Crossing a page boundary? */
3241         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3242                 int rc, now;
3243
3244                 now = -addr & ~PAGE_MASK;
3245                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3246                 if (rc != X86EMUL_CONTINUE)
3247                         return rc;
3248                 addr += now;
3249                 val += now;
3250                 bytes -= now;
3251         }
3252         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3253 }
3254 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3255
3256 static int emulator_cmpxchg_emulated(unsigned long addr,
3257                                      const void *old,
3258                                      const void *new,
3259                                      unsigned int bytes,
3260                                      struct kvm_vcpu *vcpu)
3261 {
3262         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3263 #ifndef CONFIG_X86_64
3264         /* guests cmpxchg8b have to be emulated atomically */
3265         if (bytes == 8) {
3266                 gpa_t gpa;
3267                 struct page *page;
3268                 char *kaddr;
3269                 u64 val;
3270
3271                 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3272
3273                 if (gpa == UNMAPPED_GVA ||
3274                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3275                         goto emul_write;
3276
3277                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3278                         goto emul_write;
3279
3280                 val = *(u64 *)new;
3281
3282                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3283
3284                 kaddr = kmap_atomic(page, KM_USER0);
3285                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3286                 kunmap_atomic(kaddr, KM_USER0);
3287                 kvm_release_page_dirty(page);
3288         }
3289 emul_write:
3290 #endif
3291
3292         return emulator_write_emulated(addr, new, bytes, vcpu);
3293 }
3294
3295 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3296 {
3297         return kvm_x86_ops->get_segment_base(vcpu, seg);
3298 }
3299
3300 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3301 {
3302         kvm_mmu_invlpg(vcpu, address);
3303         return X86EMUL_CONTINUE;
3304 }
3305
3306 int emulate_clts(struct kvm_vcpu *vcpu)
3307 {
3308         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3309         kvm_x86_ops->fpu_activate(vcpu);
3310         return X86EMUL_CONTINUE;
3311 }
3312
3313 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3314 {
3315         return kvm_x86_ops->get_dr(ctxt->vcpu, dr, dest);
3316 }
3317
3318 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3319 {
3320         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3321
3322         return kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask);
3323 }
3324
3325 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3326 {
3327         u8 opcodes[4];
3328         unsigned long rip = kvm_rip_read(vcpu);
3329         unsigned long rip_linear;
3330
3331         if (!printk_ratelimit())
3332                 return;
3333
3334         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3335
3336         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu, NULL);
3337
3338         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3339                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3340 }
3341 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3342
3343 static struct x86_emulate_ops emulate_ops = {
3344         .read_std            = kvm_read_guest_virt_system,
3345         .fetch               = kvm_fetch_guest_virt,
3346         .read_emulated       = emulator_read_emulated,
3347         .write_emulated      = emulator_write_emulated,
3348         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3349 };
3350
3351 static void cache_all_regs(struct kvm_vcpu *vcpu)
3352 {
3353         kvm_register_read(vcpu, VCPU_REGS_RAX);
3354         kvm_register_read(vcpu, VCPU_REGS_RSP);
3355         kvm_register_read(vcpu, VCPU_REGS_RIP);
3356         vcpu->arch.regs_dirty = ~0;
3357 }
3358
3359 int emulate_instruction(struct kvm_vcpu *vcpu,
3360                         unsigned long cr2,
3361                         u16 error_code,
3362                         int emulation_type)
3363 {
3364         int r, shadow_mask;
3365         struct decode_cache *c;
3366         struct kvm_run *run = vcpu->run;
3367
3368         kvm_clear_exception_queue(vcpu);
3369         vcpu->arch.mmio_fault_cr2 = cr2;
3370         /*
3371          * TODO: fix emulate.c to use guest_read/write_register
3372          * instead of direct ->regs accesses, can save hundred cycles
3373          * on Intel for instructions that don't read/change RSP, for
3374          * for example.
3375          */
3376         cache_all_regs(vcpu);
3377
3378         vcpu->mmio_is_write = 0;
3379         vcpu->arch.pio.string = 0;
3380
3381         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3382                 int cs_db, cs_l;
3383                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3384
3385                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3386                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3387                 vcpu->arch.emulate_ctxt.mode =
3388                         (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3389                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3390                         ? X86EMUL_MODE_VM86 : cs_l
3391                         ? X86EMUL_MODE_PROT64 : cs_db
3392                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3393
3394                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3395
3396                 /* Only allow emulation of specific instructions on #UD
3397                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3398                 c = &vcpu->arch.emulate_ctxt.decode;
3399                 if (emulation_type & EMULTYPE_TRAP_UD) {
3400                         if (!c->twobyte)
3401                                 return EMULATE_FAIL;
3402                         switch (c->b) {
3403                         case 0x01: /* VMMCALL */
3404                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3405                                         return EMULATE_FAIL;
3406                                 break;
3407                         case 0x34: /* sysenter */
3408                         case 0x35: /* sysexit */
3409                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3410                                         return EMULATE_FAIL;
3411                                 break;
3412                         case 0x05: /* syscall */
3413                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3414                                         return EMULATE_FAIL;
3415                                 break;
3416                         default:
3417                                 return EMULATE_FAIL;
3418                         }
3419
3420                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3421                                 return EMULATE_FAIL;
3422                 }
3423
3424                 ++vcpu->stat.insn_emulation;
3425                 if (r)  {
3426                         ++vcpu->stat.insn_emulation_fail;
3427                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3428                                 return EMULATE_DONE;
3429                         return EMULATE_FAIL;
3430                 }
3431         }
3432
3433         if (emulation_type & EMULTYPE_SKIP) {
3434                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3435                 return EMULATE_DONE;
3436         }
3437
3438         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3439         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3440
3441         if (r == 0)
3442                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3443
3444         if (vcpu->arch.pio.string)
3445                 return EMULATE_DO_MMIO;
3446
3447         if ((r || vcpu->mmio_is_write) && run) {
3448                 run->exit_reason = KVM_EXIT_MMIO;
3449                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3450                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3451                 run->mmio.len = vcpu->mmio_size;
3452                 run->mmio.is_write = vcpu->mmio_is_write;
3453         }
3454
3455         if (r) {
3456                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3457                         return EMULATE_DONE;
3458                 if (!vcpu->mmio_needed) {
3459                         kvm_report_emulation_failure(vcpu, "mmio");
3460                         return EMULATE_FAIL;
3461                 }
3462                 return EMULATE_DO_MMIO;
3463         }
3464
3465         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3466
3467         if (vcpu->mmio_is_write) {
3468                 vcpu->mmio_needed = 0;
3469                 return EMULATE_DO_MMIO;
3470         }
3471
3472         return EMULATE_DONE;
3473 }
3474 EXPORT_SYMBOL_GPL(emulate_instruction);
3475
3476 static int pio_copy_data(struct kvm_vcpu *vcpu)
3477 {
3478         void *p = vcpu->arch.pio_data;
3479         gva_t q = vcpu->arch.pio.guest_gva;
3480         unsigned bytes;
3481         int ret;
3482         u32 error_code;
3483
3484         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3485         if (vcpu->arch.pio.in)
3486                 ret = kvm_write_guest_virt(q, p, bytes, vcpu, &error_code);
3487         else
3488                 ret = kvm_read_guest_virt(q, p, bytes, vcpu, &error_code);
3489
3490         if (ret == X86EMUL_PROPAGATE_FAULT)
3491                 kvm_inject_page_fault(vcpu, q, error_code);
3492
3493         return ret;
3494 }
3495
3496 int complete_pio(struct kvm_vcpu *vcpu)
3497 {
3498         struct kvm_pio_request *io = &vcpu->arch.pio;
3499         long delta;
3500         int r;
3501         unsigned long val;
3502
3503         if (!io->string) {
3504                 if (io->in) {
3505                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3506                         memcpy(&val, vcpu->arch.pio_data, io->size);
3507                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3508                 }
3509         } else {
3510                 if (io->in) {
3511                         r = pio_copy_data(vcpu);
3512                         if (r)
3513                                 goto out;
3514                 }
3515
3516                 delta = 1;
3517                 if (io->rep) {
3518                         delta *= io->cur_count;
3519                         /*
3520                          * The size of the register should really depend on
3521                          * current address size.
3522                          */
3523                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3524                         val -= delta;
3525                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3526                 }
3527                 if (io->down)
3528                         delta = -delta;
3529                 delta *= io->size;
3530                 if (io->in) {
3531                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3532                         val += delta;
3533                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3534                 } else {
3535                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3536                         val += delta;
3537                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3538                 }
3539         }
3540 out:
3541         io->count -= io->cur_count;
3542         io->cur_count = 0;
3543
3544         return 0;
3545 }
3546
3547 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3548 {
3549         /* TODO: String I/O for in kernel device */
3550         int r;
3551
3552         if (vcpu->arch.pio.in)
3553                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3554                                     vcpu->arch.pio.size, pd);
3555         else
3556                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3557                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3558                                      pd);
3559         return r;
3560 }
3561
3562 static int pio_string_write(struct kvm_vcpu *vcpu)
3563 {
3564         struct kvm_pio_request *io = &vcpu->arch.pio;
3565         void *pd = vcpu->arch.pio_data;
3566         int i, r = 0;
3567
3568         for (i = 0; i < io->cur_count; i++) {
3569                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3570                                      io->port, io->size, pd)) {
3571                         r = -EOPNOTSUPP;
3572                         break;
3573                 }
3574                 pd += io->size;
3575         }
3576         return r;
3577 }
3578
3579 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3580 {
3581         unsigned long val;
3582
3583         trace_kvm_pio(!in, port, size, 1);
3584
3585         vcpu->run->exit_reason = KVM_EXIT_IO;
3586         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3587         vcpu->run->io.size = vcpu->arch.pio.size = size;
3588         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3589         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3590         vcpu->run->io.port = vcpu->arch.pio.port = port;
3591         vcpu->arch.pio.in = in;
3592         vcpu->arch.pio.string = 0;
3593         vcpu->arch.pio.down = 0;
3594         vcpu->arch.pio.rep = 0;
3595
3596         if (!vcpu->arch.pio.in) {
3597                 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3598                 memcpy(vcpu->arch.pio_data, &val, 4);
3599         }
3600
3601         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3602                 complete_pio(vcpu);
3603                 return 1;
3604         }
3605         return 0;
3606 }
3607 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3608
3609 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3610                   int size, unsigned long count, int down,
3611                   gva_t address, int rep, unsigned port)
3612 {
3613         unsigned now, in_page;
3614         int ret = 0;
3615
3616         trace_kvm_pio(!in, port, size, count);
3617
3618         vcpu->run->exit_reason = KVM_EXIT_IO;
3619         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3620         vcpu->run->io.size = vcpu->arch.pio.size = size;
3621         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3622         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3623         vcpu->run->io.port = vcpu->arch.pio.port = port;
3624         vcpu->arch.pio.in = in;
3625         vcpu->arch.pio.string = 1;
3626         vcpu->arch.pio.down = down;
3627         vcpu->arch.pio.rep = rep;
3628
3629         if (!count) {
3630                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3631                 return 1;
3632         }
3633
3634         if (!down)
3635                 in_page = PAGE_SIZE - offset_in_page(address);
3636         else
3637                 in_page = offset_in_page(address) + size;
3638         now = min(count, (unsigned long)in_page / size);
3639         if (!now)
3640                 now = 1;
3641         if (down) {
3642                 /*
3643                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3644                  */
3645                 pr_unimpl(vcpu, "guest string pio down\n");
3646                 kvm_inject_gp(vcpu, 0);
3647                 return 1;
3648         }
3649         vcpu->run->io.count = now;
3650         vcpu->arch.pio.cur_count = now;
3651
3652         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3653                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3654
3655         vcpu->arch.pio.guest_gva = address;
3656
3657         if (!vcpu->arch.pio.in) {
3658                 /* string PIO write */
3659                 ret = pio_copy_data(vcpu);
3660                 if (ret == X86EMUL_PROPAGATE_FAULT)
3661                         return 1;
3662                 if (ret == 0 && !pio_string_write(vcpu)) {
3663                         complete_pio(vcpu);
3664                         if (vcpu->arch.pio.count == 0)
3665                                 ret = 1;
3666                 }
3667         }
3668         /* no string PIO read support yet */
3669
3670         return ret;
3671 }
3672 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3673
3674 static void bounce_off(void *info)
3675 {
3676         /* nothing */
3677 }
3678
3679 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3680                                      void *data)
3681 {
3682         struct cpufreq_freqs *freq = data;
3683         struct kvm *kvm;
3684         struct kvm_vcpu *vcpu;
3685         int i, send_ipi = 0;
3686
3687         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3688                 return 0;
3689         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3690                 return 0;
3691         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3692
3693         spin_lock(&kvm_lock);
3694         list_for_each_entry(kvm, &vm_list, vm_list) {
3695                 kvm_for_each_vcpu(i, vcpu, kvm) {
3696                         if (vcpu->cpu != freq->cpu)
3697                                 continue;
3698                         if (!kvm_request_guest_time_update(vcpu))
3699                                 continue;
3700                         if (vcpu->cpu != smp_processor_id())
3701                                 send_ipi++;
3702                 }
3703         }
3704         spin_unlock(&kvm_lock);
3705
3706         if (freq->old < freq->new && send_ipi) {
3707                 /*
3708                  * We upscale the frequency.  Must make the guest
3709                  * doesn't see old kvmclock values while running with
3710                  * the new frequency, otherwise we risk the guest sees
3711                  * time go backwards.
3712                  *
3713                  * In case we update the frequency for another cpu
3714                  * (which might be in guest context) send an interrupt
3715                  * to kick the cpu out of guest context.  Next time
3716                  * guest context is entered kvmclock will be updated,
3717                  * so the guest will not see stale values.
3718                  */
3719                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3720         }
3721         return 0;
3722 }
3723
3724 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3725         .notifier_call  = kvmclock_cpufreq_notifier
3726 };
3727
3728 static void kvm_timer_init(void)
3729 {
3730         int cpu;
3731
3732         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3733                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3734                                           CPUFREQ_TRANSITION_NOTIFIER);
3735                 for_each_online_cpu(cpu) {
3736                         unsigned long khz = cpufreq_get(cpu);
3737                         if (!khz)
3738                                 khz = tsc_khz;
3739                         per_cpu(cpu_tsc_khz, cpu) = khz;
3740                 }
3741         } else {
3742                 for_each_possible_cpu(cpu)
3743                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3744         }
3745 }
3746
3747 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
3748
3749 static int kvm_is_in_guest(void)
3750 {
3751         return percpu_read(current_vcpu) != NULL;
3752 }
3753
3754 static int kvm_is_user_mode(void)
3755 {
3756         int user_mode = 3;
3757
3758         if (percpu_read(current_vcpu))
3759                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
3760
3761         return user_mode != 0;
3762 }
3763
3764 static unsigned long kvm_get_guest_ip(void)
3765 {
3766         unsigned long ip = 0;
3767
3768         if (percpu_read(current_vcpu))
3769                 ip = kvm_rip_read(percpu_read(current_vcpu));
3770
3771         return ip;
3772 }
3773
3774 static struct perf_guest_info_callbacks kvm_guest_cbs = {
3775         .is_in_guest            = kvm_is_in_guest,
3776         .is_user_mode           = kvm_is_user_mode,
3777         .get_guest_ip           = kvm_get_guest_ip,
3778 };
3779
3780 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
3781 {
3782         percpu_write(current_vcpu, vcpu);
3783 }
3784 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
3785
3786 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
3787 {
3788         percpu_write(current_vcpu, NULL);
3789 }
3790 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
3791
3792 int kvm_arch_init(void *opaque)
3793 {
3794         int r;
3795         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3796
3797         if (kvm_x86_ops) {
3798                 printk(KERN_ERR "kvm: already loaded the other module\n");
3799                 r = -EEXIST;
3800                 goto out;
3801         }
3802
3803         if (!ops->cpu_has_kvm_support()) {
3804                 printk(KERN_ERR "kvm: no hardware support\n");
3805                 r = -EOPNOTSUPP;
3806                 goto out;
3807         }
3808         if (ops->disabled_by_bios()) {
3809                 printk(KERN_ERR "kvm: disabled by bios\n");
3810                 r = -EOPNOTSUPP;
3811                 goto out;
3812         }
3813
3814         r = kvm_mmu_module_init();
3815         if (r)
3816                 goto out;
3817
3818         kvm_init_msr_list();
3819
3820         kvm_x86_ops = ops;
3821         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3822         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3823         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3824                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3825
3826         kvm_timer_init();
3827
3828         perf_register_guest_info_callbacks(&kvm_guest_cbs);
3829
3830         return 0;
3831
3832 out:
3833         return r;
3834 }
3835
3836 void kvm_arch_exit(void)
3837 {
3838         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
3839
3840         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3841                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3842                                             CPUFREQ_TRANSITION_NOTIFIER);
3843         kvm_x86_ops = NULL;
3844         kvm_mmu_module_exit();
3845 }
3846
3847 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3848 {
3849         ++vcpu->stat.halt_exits;
3850         if (irqchip_in_kernel(vcpu->kvm)) {
3851                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3852                 return 1;
3853         } else {
3854                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3855                 return 0;
3856         }
3857 }
3858 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3859
3860 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3861                            unsigned long a1)
3862 {
3863         if (is_long_mode(vcpu))
3864                 return a0;
3865         else
3866                 return a0 | ((gpa_t)a1 << 32);
3867 }
3868
3869 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3870 {
3871         u64 param, ingpa, outgpa, ret;
3872         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3873         bool fast, longmode;
3874         int cs_db, cs_l;
3875
3876         /*
3877          * hypercall generates UD from non zero cpl and real mode
3878          * per HYPER-V spec
3879          */
3880         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
3881                 kvm_queue_exception(vcpu, UD_VECTOR);
3882                 return 0;
3883         }
3884
3885         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3886         longmode = is_long_mode(vcpu) && cs_l == 1;
3887
3888         if (!longmode) {
3889                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3890                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
3891                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3892                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
3893                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3894                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
3895         }
3896 #ifdef CONFIG_X86_64
3897         else {
3898                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3899                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3900                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3901         }
3902 #endif
3903
3904         code = param & 0xffff;
3905         fast = (param >> 16) & 0x1;
3906         rep_cnt = (param >> 32) & 0xfff;
3907         rep_idx = (param >> 48) & 0xfff;
3908
3909         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3910
3911         switch (code) {
3912         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
3913                 kvm_vcpu_on_spin(vcpu);
3914                 break;
3915         default:
3916                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
3917                 break;
3918         }
3919
3920         ret = res | (((u64)rep_done & 0xfff) << 32);
3921         if (longmode) {
3922                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3923         } else {
3924                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3925                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3926         }
3927
3928         return 1;
3929 }
3930
3931 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3932 {
3933         unsigned long nr, a0, a1, a2, a3, ret;
3934         int r = 1;
3935
3936         if (kvm_hv_hypercall_enabled(vcpu->kvm))
3937                 return kvm_hv_hypercall(vcpu);
3938
3939         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3940         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3941         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3942         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3943         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3944
3945         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3946
3947         if (!is_long_mode(vcpu)) {
3948                 nr &= 0xFFFFFFFF;
3949                 a0 &= 0xFFFFFFFF;
3950                 a1 &= 0xFFFFFFFF;
3951                 a2 &= 0xFFFFFFFF;
3952                 a3 &= 0xFFFFFFFF;
3953         }
3954
3955         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3956                 ret = -KVM_EPERM;
3957                 goto out;
3958         }
3959
3960         switch (nr) {
3961         case KVM_HC_VAPIC_POLL_IRQ:
3962                 ret = 0;
3963                 break;
3964         case KVM_HC_MMU_OP:
3965                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3966                 break;
3967         default:
3968                 ret = -KVM_ENOSYS;
3969                 break;
3970         }
3971 out:
3972         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3973         ++vcpu->stat.hypercalls;
3974         return r;
3975 }
3976 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3977
3978 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3979 {
3980         char instruction[3];
3981         unsigned long rip = kvm_rip_read(vcpu);
3982
3983         /*
3984          * Blow out the MMU to ensure that no other VCPU has an active mapping
3985          * to ensure that the updated hypercall appears atomically across all
3986          * VCPUs.
3987          */
3988         kvm_mmu_zap_all(vcpu->kvm);
3989
3990         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3991
3992         return emulator_write_emulated(rip, instruction, 3, vcpu);
3993 }
3994
3995 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3996 {
3997         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3998 }
3999
4000 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4001 {
4002         struct descriptor_table dt = { limit, base };
4003
4004         kvm_x86_ops->set_gdt(vcpu, &dt);
4005 }
4006
4007 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4008 {
4009         struct descriptor_table dt = { limit, base };
4010
4011         kvm_x86_ops->set_idt(vcpu, &dt);
4012 }
4013
4014 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
4015                    unsigned long *rflags)
4016 {
4017         kvm_lmsw(vcpu, msw);
4018         *rflags = kvm_get_rflags(vcpu);
4019 }
4020
4021 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
4022 {
4023         unsigned long value;
4024
4025         switch (cr) {
4026         case 0:
4027                 value = kvm_read_cr0(vcpu);
4028                 break;
4029         case 2:
4030                 value = vcpu->arch.cr2;
4031                 break;
4032         case 3:
4033                 value = vcpu->arch.cr3;
4034                 break;
4035         case 4:
4036                 value = kvm_read_cr4(vcpu);
4037                 break;
4038         case 8:
4039                 value = kvm_get_cr8(vcpu);
4040                 break;
4041         default:
4042                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4043                 return 0;
4044         }
4045
4046         return value;
4047 }
4048
4049 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
4050                      unsigned long *rflags)
4051 {
4052         switch (cr) {
4053         case 0:
4054                 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4055                 *rflags = kvm_get_rflags(vcpu);
4056                 break;
4057         case 2:
4058                 vcpu->arch.cr2 = val;
4059                 break;
4060         case 3:
4061                 kvm_set_cr3(vcpu, val);
4062                 break;
4063         case 4:
4064                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4065                 break;
4066         case 8:
4067                 kvm_set_cr8(vcpu, val & 0xfUL);
4068                 break;
4069         default:
4070                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4071         }
4072 }
4073
4074 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4075 {
4076         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4077         int j, nent = vcpu->arch.cpuid_nent;
4078
4079         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4080         /* when no next entry is found, the current entry[i] is reselected */
4081         for (j = i + 1; ; j = (j + 1) % nent) {
4082                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4083                 if (ej->function == e->function) {
4084                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4085                         return j;
4086                 }
4087         }
4088         return 0; /* silence gcc, even though control never reaches here */
4089 }
4090
4091 /* find an entry with matching function, matching index (if needed), and that
4092  * should be read next (if it's stateful) */
4093 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4094         u32 function, u32 index)
4095 {
4096         if (e->function != function)
4097                 return 0;
4098         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4099                 return 0;
4100         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4101             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4102                 return 0;
4103         return 1;
4104 }
4105
4106 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4107                                               u32 function, u32 index)
4108 {
4109         int i;
4110         struct kvm_cpuid_entry2 *best = NULL;
4111
4112         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4113                 struct kvm_cpuid_entry2 *e;
4114
4115                 e = &vcpu->arch.cpuid_entries[i];
4116                 if (is_matching_cpuid_entry(e, function, index)) {
4117                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4118                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4119                         best = e;
4120                         break;
4121                 }
4122                 /*
4123                  * Both basic or both extended?
4124                  */
4125                 if (((e->function ^ function) & 0x80000000) == 0)
4126                         if (!best || e->function > best->function)
4127                                 best = e;
4128         }
4129         return best;
4130 }
4131 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4132
4133 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4134 {
4135         struct kvm_cpuid_entry2 *best;
4136
4137         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4138         if (best)
4139                 return best->eax & 0xff;
4140         return 36;
4141 }
4142
4143 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4144 {
4145         u32 function, index;
4146         struct kvm_cpuid_entry2 *best;
4147
4148         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4149         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4150         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4151         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4152         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4153         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4154         best = kvm_find_cpuid_entry(vcpu, function, index);
4155         if (best) {
4156                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4157                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4158                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4159                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4160         }
4161         kvm_x86_ops->skip_emulated_instruction(vcpu);
4162         trace_kvm_cpuid(function,
4163                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4164                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4165                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4166                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4167 }
4168 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4169
4170 /*
4171  * Check if userspace requested an interrupt window, and that the
4172  * interrupt window is open.
4173  *
4174  * No need to exit to userspace if we already have an interrupt queued.
4175  */
4176 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4177 {
4178         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4179                 vcpu->run->request_interrupt_window &&
4180                 kvm_arch_interrupt_allowed(vcpu));
4181 }
4182
4183 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4184 {
4185         struct kvm_run *kvm_run = vcpu->run;
4186
4187         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4188         kvm_run->cr8 = kvm_get_cr8(vcpu);
4189         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4190         if (irqchip_in_kernel(vcpu->kvm))
4191                 kvm_run->ready_for_interrupt_injection = 1;
4192         else
4193                 kvm_run->ready_for_interrupt_injection =
4194                         kvm_arch_interrupt_allowed(vcpu) &&
4195                         !kvm_cpu_has_interrupt(vcpu) &&
4196                         !kvm_event_needs_reinjection(vcpu);
4197 }
4198
4199 static void vapic_enter(struct kvm_vcpu *vcpu)
4200 {
4201         struct kvm_lapic *apic = vcpu->arch.apic;
4202         struct page *page;
4203
4204         if (!apic || !apic->vapic_addr)
4205                 return;
4206
4207         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4208
4209         vcpu->arch.apic->vapic_page = page;
4210 }
4211
4212 static void vapic_exit(struct kvm_vcpu *vcpu)
4213 {
4214         struct kvm_lapic *apic = vcpu->arch.apic;
4215         int idx;
4216
4217         if (!apic || !apic->vapic_addr)
4218                 return;
4219
4220         idx = srcu_read_lock(&vcpu->kvm->srcu);
4221         kvm_release_page_dirty(apic->vapic_page);
4222         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4223         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4224 }
4225
4226 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4227 {
4228         int max_irr, tpr;
4229
4230         if (!kvm_x86_ops->update_cr8_intercept)
4231                 return;
4232
4233         if (!vcpu->arch.apic)
4234                 return;
4235
4236         if (!vcpu->arch.apic->vapic_addr)
4237                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4238         else
4239                 max_irr = -1;
4240
4241         if (max_irr != -1)
4242                 max_irr >>= 4;
4243
4244         tpr = kvm_lapic_get_cr8(vcpu);
4245
4246         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4247 }
4248
4249 static void inject_pending_event(struct kvm_vcpu *vcpu)
4250 {
4251         /* try to reinject previous events if any */
4252         if (vcpu->arch.exception.pending) {
4253                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4254                                           vcpu->arch.exception.has_error_code,
4255                                           vcpu->arch.exception.error_code);
4256                 return;
4257         }
4258
4259         if (vcpu->arch.nmi_injected) {
4260                 kvm_x86_ops->set_nmi(vcpu);
4261                 return;
4262         }
4263
4264         if (vcpu->arch.interrupt.pending) {
4265                 kvm_x86_ops->set_irq(vcpu);
4266                 return;
4267         }
4268
4269         /* try to inject new event if pending */
4270         if (vcpu->arch.nmi_pending) {
4271                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4272                         vcpu->arch.nmi_pending = false;
4273                         vcpu->arch.nmi_injected = true;
4274                         kvm_x86_ops->set_nmi(vcpu);
4275                 }
4276         } else if (kvm_cpu_has_interrupt(vcpu)) {
4277                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4278                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4279                                             false);
4280                         kvm_x86_ops->set_irq(vcpu);
4281                 }
4282         }
4283 }
4284
4285 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4286 {
4287         int r;
4288         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4289                 vcpu->run->request_interrupt_window;
4290
4291         if (vcpu->requests)
4292                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4293                         kvm_mmu_unload(vcpu);
4294
4295         r = kvm_mmu_reload(vcpu);
4296         if (unlikely(r))
4297                 goto out;
4298
4299         if (vcpu->requests) {
4300                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4301                         __kvm_migrate_timers(vcpu);
4302                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4303                         kvm_write_guest_time(vcpu);
4304                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4305                         kvm_mmu_sync_roots(vcpu);
4306                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4307                         kvm_x86_ops->tlb_flush(vcpu);
4308                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4309                                        &vcpu->requests)) {
4310                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4311                         r = 0;
4312                         goto out;
4313                 }
4314                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4315                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4316                         r = 0;
4317                         goto out;
4318                 }
4319                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4320                         vcpu->fpu_active = 0;
4321                         kvm_x86_ops->fpu_deactivate(vcpu);
4322                 }
4323         }
4324
4325         preempt_disable();
4326
4327         kvm_x86_ops->prepare_guest_switch(vcpu);
4328         if (vcpu->fpu_active)
4329                 kvm_load_guest_fpu(vcpu);
4330
4331         local_irq_disable();
4332
4333         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4334         smp_mb__after_clear_bit();
4335
4336         if (vcpu->requests || need_resched() || signal_pending(current)) {
4337                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4338                 local_irq_enable();
4339                 preempt_enable();
4340                 r = 1;
4341                 goto out;
4342         }
4343
4344         inject_pending_event(vcpu);
4345
4346         /* enable NMI/IRQ window open exits if needed */
4347         if (vcpu->arch.nmi_pending)
4348                 kvm_x86_ops->enable_nmi_window(vcpu);
4349         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4350                 kvm_x86_ops->enable_irq_window(vcpu);
4351
4352         if (kvm_lapic_enabled(vcpu)) {
4353                 update_cr8_intercept(vcpu);
4354                 kvm_lapic_sync_to_vapic(vcpu);
4355         }
4356
4357         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4358
4359         kvm_guest_enter();
4360
4361         if (unlikely(vcpu->arch.switch_db_regs)) {
4362                 set_debugreg(0, 7);
4363                 set_debugreg(vcpu->arch.eff_db[0], 0);
4364                 set_debugreg(vcpu->arch.eff_db[1], 1);
4365                 set_debugreg(vcpu->arch.eff_db[2], 2);
4366                 set_debugreg(vcpu->arch.eff_db[3], 3);
4367         }
4368
4369         trace_kvm_entry(vcpu->vcpu_id);
4370         kvm_x86_ops->run(vcpu);
4371
4372         /*
4373          * If the guest has used debug registers, at least dr7
4374          * will be disabled while returning to the host.
4375          * If we don't have active breakpoints in the host, we don't
4376          * care about the messed up debug address registers. But if
4377          * we have some of them active, restore the old state.
4378          */
4379         if (hw_breakpoint_active())
4380                 hw_breakpoint_restore();
4381
4382         set_bit(KVM_REQ_KICK, &vcpu->requests);
4383         local_irq_enable();
4384
4385         ++vcpu->stat.exits;
4386
4387         /*
4388          * We must have an instruction between local_irq_enable() and
4389          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4390          * the interrupt shadow.  The stat.exits increment will do nicely.
4391          * But we need to prevent reordering, hence this barrier():
4392          */
4393         barrier();
4394
4395         kvm_guest_exit();
4396
4397         preempt_enable();
4398
4399         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4400
4401         /*
4402          * Profile KVM exit RIPs:
4403          */
4404         if (unlikely(prof_on == KVM_PROFILING)) {
4405                 unsigned long rip = kvm_rip_read(vcpu);
4406                 profile_hit(KVM_PROFILING, (void *)rip);
4407         }
4408
4409
4410         kvm_lapic_sync_from_vapic(vcpu);
4411
4412         r = kvm_x86_ops->handle_exit(vcpu);
4413 out:
4414         return r;
4415 }
4416
4417
4418 static int __vcpu_run(struct kvm_vcpu *vcpu)
4419 {
4420         int r;
4421         struct kvm *kvm = vcpu->kvm;
4422
4423         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4424                 pr_debug("vcpu %d received sipi with vector # %x\n",
4425                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4426                 kvm_lapic_reset(vcpu);
4427                 r = kvm_arch_vcpu_reset(vcpu);
4428                 if (r)
4429                         return r;
4430                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4431         }
4432
4433         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4434         vapic_enter(vcpu);
4435
4436         r = 1;
4437         while (r > 0) {
4438                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4439                         r = vcpu_enter_guest(vcpu);
4440                 else {
4441                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4442                         kvm_vcpu_block(vcpu);
4443                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4444                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4445                         {
4446                                 switch(vcpu->arch.mp_state) {
4447                                 case KVM_MP_STATE_HALTED:
4448                                         vcpu->arch.mp_state =
4449                                                 KVM_MP_STATE_RUNNABLE;
4450                                 case KVM_MP_STATE_RUNNABLE:
4451                                         break;
4452                                 case KVM_MP_STATE_SIPI_RECEIVED:
4453                                 default:
4454                                         r = -EINTR;
4455                                         break;
4456                                 }
4457                         }
4458                 }
4459
4460                 if (r <= 0)
4461                         break;
4462
4463                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4464                 if (kvm_cpu_has_pending_timer(vcpu))
4465                         kvm_inject_pending_timer_irqs(vcpu);
4466
4467                 if (dm_request_for_irq_injection(vcpu)) {
4468                         r = -EINTR;
4469                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4470                         ++vcpu->stat.request_irq_exits;
4471                 }
4472                 if (signal_pending(current)) {
4473                         r = -EINTR;
4474                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4475                         ++vcpu->stat.signal_exits;
4476                 }
4477                 if (need_resched()) {
4478                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4479                         kvm_resched(vcpu);
4480                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4481                 }
4482         }
4483
4484         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4485         post_kvm_run_save(vcpu);
4486
4487         vapic_exit(vcpu);
4488
4489         return r;
4490 }
4491
4492 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4493 {
4494         int r;
4495         sigset_t sigsaved;
4496
4497         vcpu_load(vcpu);
4498
4499         if (vcpu->sigset_active)
4500                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4501
4502         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4503                 kvm_vcpu_block(vcpu);
4504                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4505                 r = -EAGAIN;
4506                 goto out;
4507         }
4508
4509         /* re-sync apic's tpr */
4510         if (!irqchip_in_kernel(vcpu->kvm))
4511                 kvm_set_cr8(vcpu, kvm_run->cr8);
4512
4513         if (vcpu->arch.pio.cur_count) {
4514                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4515                 r = complete_pio(vcpu);
4516                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4517                 if (r)
4518                         goto out;
4519         }
4520         if (vcpu->mmio_needed) {
4521                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4522                 vcpu->mmio_read_completed = 1;
4523                 vcpu->mmio_needed = 0;
4524
4525                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4526                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4527                                         EMULTYPE_NO_DECODE);
4528                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4529                 if (r == EMULATE_DO_MMIO) {
4530                         /*
4531                          * Read-modify-write.  Back to userspace.
4532                          */
4533                         r = 0;
4534                         goto out;
4535                 }
4536         }
4537         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4538                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4539                                      kvm_run->hypercall.ret);
4540
4541         r = __vcpu_run(vcpu);
4542
4543 out:
4544         if (vcpu->sigset_active)
4545                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4546
4547         vcpu_put(vcpu);
4548         return r;
4549 }
4550
4551 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4552 {
4553         vcpu_load(vcpu);
4554
4555         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4556         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4557         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4558         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4559         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4560         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4561         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4562         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4563 #ifdef CONFIG_X86_64
4564         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4565         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4566         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4567         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4568         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4569         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4570         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4571         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4572 #endif
4573
4574         regs->rip = kvm_rip_read(vcpu);
4575         regs->rflags = kvm_get_rflags(vcpu);
4576
4577         vcpu_put(vcpu);
4578
4579         return 0;
4580 }
4581
4582 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4583 {
4584         vcpu_load(vcpu);
4585
4586         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4587         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4588         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4589         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4590         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4591         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4592         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4593         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4594 #ifdef CONFIG_X86_64
4595         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4596         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4597         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4598         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4599         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4600         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4601         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4602         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4603 #endif
4604
4605         kvm_rip_write(vcpu, regs->rip);
4606         kvm_set_rflags(vcpu, regs->rflags);
4607
4608         vcpu->arch.exception.pending = false;
4609
4610         vcpu_put(vcpu);
4611
4612         return 0;
4613 }
4614
4615 void kvm_get_segment(struct kvm_vcpu *vcpu,
4616                      struct kvm_segment *var, int seg)
4617 {
4618         kvm_x86_ops->get_segment(vcpu, var, seg);
4619 }
4620
4621 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4622 {
4623         struct kvm_segment cs;
4624
4625         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4626         *db = cs.db;
4627         *l = cs.l;
4628 }
4629 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4630
4631 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4632                                   struct kvm_sregs *sregs)
4633 {
4634         struct descriptor_table dt;
4635
4636         vcpu_load(vcpu);
4637
4638         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4639         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4640         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4641         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4642         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4643         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4644
4645         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4646         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4647
4648         kvm_x86_ops->get_idt(vcpu, &dt);
4649         sregs->idt.limit = dt.limit;
4650         sregs->idt.base = dt.base;
4651         kvm_x86_ops->get_gdt(vcpu, &dt);
4652         sregs->gdt.limit = dt.limit;
4653         sregs->gdt.base = dt.base;
4654
4655         sregs->cr0 = kvm_read_cr0(vcpu);
4656         sregs->cr2 = vcpu->arch.cr2;
4657         sregs->cr3 = vcpu->arch.cr3;
4658         sregs->cr4 = kvm_read_cr4(vcpu);
4659         sregs->cr8 = kvm_get_cr8(vcpu);
4660         sregs->efer = vcpu->arch.efer;
4661         sregs->apic_base = kvm_get_apic_base(vcpu);
4662
4663         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4664
4665         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4666                 set_bit(vcpu->arch.interrupt.nr,
4667                         (unsigned long *)sregs->interrupt_bitmap);
4668
4669         vcpu_put(vcpu);
4670
4671         return 0;
4672 }
4673
4674 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4675                                     struct kvm_mp_state *mp_state)
4676 {
4677         vcpu_load(vcpu);
4678         mp_state->mp_state = vcpu->arch.mp_state;
4679         vcpu_put(vcpu);
4680         return 0;
4681 }
4682
4683 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4684                                     struct kvm_mp_state *mp_state)
4685 {
4686         vcpu_load(vcpu);
4687         vcpu->arch.mp_state = mp_state->mp_state;
4688         vcpu_put(vcpu);
4689         return 0;
4690 }
4691
4692 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4693                         struct kvm_segment *var, int seg)
4694 {
4695         kvm_x86_ops->set_segment(vcpu, var, seg);
4696 }
4697
4698 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4699                                    struct kvm_segment *kvm_desct)
4700 {
4701         kvm_desct->base = get_desc_base(seg_desc);
4702         kvm_desct->limit = get_desc_limit(seg_desc);
4703         if (seg_desc->g) {
4704                 kvm_desct->limit <<= 12;
4705                 kvm_desct->limit |= 0xfff;
4706         }
4707         kvm_desct->selector = selector;
4708         kvm_desct->type = seg_desc->type;
4709         kvm_desct->present = seg_desc->p;
4710         kvm_desct->dpl = seg_desc->dpl;
4711         kvm_desct->db = seg_desc->d;
4712         kvm_desct->s = seg_desc->s;
4713         kvm_desct->l = seg_desc->l;
4714         kvm_desct->g = seg_desc->g;
4715         kvm_desct->avl = seg_desc->avl;
4716         if (!selector)
4717                 kvm_desct->unusable = 1;
4718         else
4719                 kvm_desct->unusable = 0;
4720         kvm_desct->padding = 0;
4721 }
4722
4723 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4724                                           u16 selector,
4725                                           struct descriptor_table *dtable)
4726 {
4727         if (selector & 1 << 2) {
4728                 struct kvm_segment kvm_seg;
4729
4730                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4731
4732                 if (kvm_seg.unusable)
4733                         dtable->limit = 0;
4734                 else
4735                         dtable->limit = kvm_seg.limit;
4736                 dtable->base = kvm_seg.base;
4737         }
4738         else
4739                 kvm_x86_ops->get_gdt(vcpu, dtable);
4740 }
4741
4742 /* allowed just for 8 bytes segments */
4743 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4744                                          struct desc_struct *seg_desc)
4745 {
4746         struct descriptor_table dtable;
4747         u16 index = selector >> 3;
4748         int ret;
4749         u32 err;
4750         gva_t addr;
4751
4752         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4753
4754         if (dtable.limit < index * 8 + 7) {
4755                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4756                 return X86EMUL_PROPAGATE_FAULT;
4757         }
4758         addr = dtable.base + index * 8;
4759         ret = kvm_read_guest_virt_system(addr, seg_desc, sizeof(*seg_desc),
4760                                          vcpu,  &err);
4761         if (ret == X86EMUL_PROPAGATE_FAULT)
4762                 kvm_inject_page_fault(vcpu, addr, err);
4763
4764        return ret;
4765 }
4766
4767 /* allowed just for 8 bytes segments */
4768 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4769                                          struct desc_struct *seg_desc)
4770 {
4771         struct descriptor_table dtable;
4772         u16 index = selector >> 3;
4773
4774         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4775
4776         if (dtable.limit < index * 8 + 7)
4777                 return 1;
4778         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu, NULL);
4779 }
4780
4781 static gpa_t get_tss_base_addr_write(struct kvm_vcpu *vcpu,
4782                                struct desc_struct *seg_desc)
4783 {
4784         u32 base_addr = get_desc_base(seg_desc);
4785
4786         return kvm_mmu_gva_to_gpa_write(vcpu, base_addr, NULL);
4787 }
4788
4789 static gpa_t get_tss_base_addr_read(struct kvm_vcpu *vcpu,
4790                              struct desc_struct *seg_desc)
4791 {
4792         u32 base_addr = get_desc_base(seg_desc);
4793
4794         return kvm_mmu_gva_to_gpa_read(vcpu, base_addr, NULL);
4795 }
4796
4797 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4798 {
4799         struct kvm_segment kvm_seg;
4800
4801         kvm_get_segment(vcpu, &kvm_seg, seg);
4802         return kvm_seg.selector;
4803 }
4804
4805 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4806 {
4807         struct kvm_segment segvar = {
4808                 .base = selector << 4,
4809                 .limit = 0xffff,
4810                 .selector = selector,
4811                 .type = 3,
4812                 .present = 1,
4813                 .dpl = 3,
4814                 .db = 0,
4815                 .s = 1,
4816                 .l = 0,
4817                 .g = 0,
4818                 .avl = 0,
4819                 .unusable = 0,
4820         };
4821         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4822         return X86EMUL_CONTINUE;
4823 }
4824
4825 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4826 {
4827         return (seg != VCPU_SREG_LDTR) &&
4828                 (seg != VCPU_SREG_TR) &&
4829                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4830 }
4831
4832 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector, int seg)
4833 {
4834         struct kvm_segment kvm_seg;
4835         struct desc_struct seg_desc;
4836         u8 dpl, rpl, cpl;
4837         unsigned err_vec = GP_VECTOR;
4838         u32 err_code = 0;
4839         bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
4840         int ret;
4841
4842         if (is_vm86_segment(vcpu, seg) || !is_protmode(vcpu))
4843                 return kvm_load_realmode_segment(vcpu, selector, seg);
4844
4845         /* NULL selector is not valid for TR, CS and SS */
4846         if ((seg == VCPU_SREG_CS || seg == VCPU_SREG_SS || seg == VCPU_SREG_TR)
4847             && null_selector)
4848                 goto exception;
4849
4850         /* TR should be in GDT only */
4851         if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
4852                 goto exception;
4853
4854         ret = load_guest_segment_descriptor(vcpu, selector, &seg_desc);
4855         if (ret)
4856                 return ret;
4857
4858         seg_desct_to_kvm_desct(&seg_desc, selector, &kvm_seg);
4859
4860         if (null_selector) { /* for NULL selector skip all following checks */
4861                 kvm_seg.unusable = 1;
4862                 goto load;
4863         }
4864
4865         err_code = selector & 0xfffc;
4866         err_vec = GP_VECTOR;
4867
4868         /* can't load system descriptor into segment selecor */
4869         if (seg <= VCPU_SREG_GS && !kvm_seg.s)
4870                 goto exception;
4871
4872         if (!kvm_seg.present) {
4873                 err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
4874                 goto exception;
4875         }
4876
4877         rpl = selector & 3;
4878         dpl = kvm_seg.dpl;
4879         cpl = kvm_x86_ops->get_cpl(vcpu);
4880
4881         switch (seg) {
4882         case VCPU_SREG_SS:
4883                 /*
4884                  * segment is not a writable data segment or segment
4885                  * selector's RPL != CPL or segment selector's RPL != CPL
4886                  */
4887                 if (rpl != cpl || (kvm_seg.type & 0xa) != 0x2 || dpl != cpl)
4888                         goto exception;
4889                 break;
4890         case VCPU_SREG_CS:
4891                 if (!(kvm_seg.type & 8))
4892                         goto exception;
4893
4894                 if (kvm_seg.type & 4) {
4895                         /* conforming */
4896                         if (dpl > cpl)
4897                                 goto exception;
4898                 } else {
4899                         /* nonconforming */
4900                         if (rpl > cpl || dpl != cpl)
4901                                 goto exception;
4902                 }
4903                 /* CS(RPL) <- CPL */
4904                 selector = (selector & 0xfffc) | cpl;
4905             break;
4906         case VCPU_SREG_TR:
4907                 if (kvm_seg.s || (kvm_seg.type != 1 && kvm_seg.type != 9))
4908                         goto exception;
4909                 break;
4910         case VCPU_SREG_LDTR:
4911                 if (kvm_seg.s || kvm_seg.type != 2)
4912                         goto exception;
4913                 break;
4914         default: /*  DS, ES, FS, or GS */
4915                 /*
4916                  * segment is not a data or readable code segment or
4917                  * ((segment is a data or nonconforming code segment)
4918                  * and (both RPL and CPL > DPL))
4919                  */
4920                 if ((kvm_seg.type & 0xa) == 0x8 ||
4921                     (((kvm_seg.type & 0xc) != 0xc) && (rpl > dpl && cpl > dpl)))
4922                         goto exception;
4923                 break;
4924         }
4925
4926         if (!kvm_seg.unusable && kvm_seg.s) {
4927                 /* mark segment as accessed */
4928                 kvm_seg.type |= 1;
4929                 seg_desc.type |= 1;
4930                 save_guest_segment_descriptor(vcpu, selector, &seg_desc);
4931         }
4932 load:
4933         kvm_set_segment(vcpu, &kvm_seg, seg);
4934         return X86EMUL_CONTINUE;
4935 exception:
4936         kvm_queue_exception_e(vcpu, err_vec, err_code);
4937         return X86EMUL_PROPAGATE_FAULT;
4938 }
4939
4940 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4941                                 struct tss_segment_32 *tss)
4942 {
4943         tss->cr3 = vcpu->arch.cr3;
4944         tss->eip = kvm_rip_read(vcpu);
4945         tss->eflags = kvm_get_rflags(vcpu);
4946         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4947         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4948         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4949         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4950         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4951         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4952         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4953         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4954         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4955         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4956         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4957         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4958         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4959         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4960         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4961 }
4962
4963 static void kvm_load_segment_selector(struct kvm_vcpu *vcpu, u16 sel, int seg)
4964 {
4965         struct kvm_segment kvm_seg;
4966         kvm_get_segment(vcpu, &kvm_seg, seg);
4967         kvm_seg.selector = sel;
4968         kvm_set_segment(vcpu, &kvm_seg, seg);
4969 }
4970
4971 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4972                                   struct tss_segment_32 *tss)
4973 {
4974         kvm_set_cr3(vcpu, tss->cr3);
4975
4976         kvm_rip_write(vcpu, tss->eip);
4977         kvm_set_rflags(vcpu, tss->eflags | 2);
4978
4979         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4980         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4981         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4982         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4983         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4984         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4985         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4986         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4987
4988         /*
4989          * SDM says that segment selectors are loaded before segment
4990          * descriptors
4991          */
4992         kvm_load_segment_selector(vcpu, tss->ldt_selector, VCPU_SREG_LDTR);
4993         kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
4994         kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
4995         kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
4996         kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
4997         kvm_load_segment_selector(vcpu, tss->fs, VCPU_SREG_FS);
4998         kvm_load_segment_selector(vcpu, tss->gs, VCPU_SREG_GS);
4999
5000         /*
5001          * Now load segment descriptors. If fault happenes at this stage
5002          * it is handled in a context of new task
5003          */
5004         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, VCPU_SREG_LDTR))
5005                 return 1;
5006
5007         if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5008                 return 1;
5009
5010         if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5011                 return 1;
5012
5013         if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5014                 return 1;
5015
5016         if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5017                 return 1;
5018
5019         if (kvm_load_segment_descriptor(vcpu, tss->fs, VCPU_SREG_FS))
5020                 return 1;
5021
5022         if (kvm_load_segment_descriptor(vcpu, tss->gs, VCPU_SREG_GS))
5023                 return 1;
5024         return 0;
5025 }
5026
5027 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
5028                                 struct tss_segment_16 *tss)
5029 {
5030         tss->ip = kvm_rip_read(vcpu);
5031         tss->flag = kvm_get_rflags(vcpu);
5032         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5033         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5034         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5035         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5036         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5037         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5038         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
5039         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
5040
5041         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
5042         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
5043         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
5044         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
5045         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
5046 }
5047
5048 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
5049                                  struct tss_segment_16 *tss)
5050 {
5051         kvm_rip_write(vcpu, tss->ip);
5052         kvm_set_rflags(vcpu, tss->flag | 2);
5053         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
5054         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
5055         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
5056         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
5057         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
5058         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
5059         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
5060         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
5061
5062         /*
5063          * SDM says that segment selectors are loaded before segment
5064          * descriptors
5065          */
5066         kvm_load_segment_selector(vcpu, tss->ldt, VCPU_SREG_LDTR);
5067         kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
5068         kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
5069         kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5070         kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5071
5072         /*
5073          * Now load segment descriptors. If fault happenes at this stage
5074          * it is handled in a context of new task
5075          */
5076         if (kvm_load_segment_descriptor(vcpu, tss->ldt, VCPU_SREG_LDTR))
5077                 return 1;
5078
5079         if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5080                 return 1;
5081
5082         if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5083                 return 1;
5084
5085         if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5086                 return 1;
5087
5088         if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5089                 return 1;
5090         return 0;
5091 }
5092
5093 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
5094                               u16 old_tss_sel, u32 old_tss_base,
5095                               struct desc_struct *nseg_desc)
5096 {
5097         struct tss_segment_16 tss_segment_16;
5098         int ret = 0;
5099
5100         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5101                            sizeof tss_segment_16))
5102                 goto out;
5103
5104         save_state_to_tss16(vcpu, &tss_segment_16);
5105
5106         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5107                             sizeof tss_segment_16))
5108                 goto out;
5109
5110         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5111                            &tss_segment_16, sizeof tss_segment_16))
5112                 goto out;
5113
5114         if (old_tss_sel != 0xffff) {
5115                 tss_segment_16.prev_task_link = old_tss_sel;
5116
5117                 if (kvm_write_guest(vcpu->kvm,
5118                                     get_tss_base_addr_write(vcpu, nseg_desc),
5119                                     &tss_segment_16.prev_task_link,
5120                                     sizeof tss_segment_16.prev_task_link))
5121                         goto out;
5122         }
5123
5124         if (load_state_from_tss16(vcpu, &tss_segment_16))
5125                 goto out;
5126
5127         ret = 1;
5128 out:
5129         return ret;
5130 }
5131
5132 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
5133                        u16 old_tss_sel, u32 old_tss_base,
5134                        struct desc_struct *nseg_desc)
5135 {
5136         struct tss_segment_32 tss_segment_32;
5137         int ret = 0;
5138
5139         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5140                            sizeof tss_segment_32))
5141                 goto out;
5142
5143         save_state_to_tss32(vcpu, &tss_segment_32);
5144
5145         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5146                             sizeof tss_segment_32))
5147                 goto out;
5148
5149         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5150                            &tss_segment_32, sizeof tss_segment_32))
5151                 goto out;
5152
5153         if (old_tss_sel != 0xffff) {
5154                 tss_segment_32.prev_task_link = old_tss_sel;
5155
5156                 if (kvm_write_guest(vcpu->kvm,
5157                                     get_tss_base_addr_write(vcpu, nseg_desc),
5158                                     &tss_segment_32.prev_task_link,
5159                                     sizeof tss_segment_32.prev_task_link))
5160                         goto out;
5161         }
5162
5163         if (load_state_from_tss32(vcpu, &tss_segment_32))
5164                 goto out;
5165
5166         ret = 1;
5167 out:
5168         return ret;
5169 }
5170
5171 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
5172 {
5173         struct kvm_segment tr_seg;
5174         struct desc_struct cseg_desc;
5175         struct desc_struct nseg_desc;
5176         int ret = 0;
5177         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
5178         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
5179         u32 desc_limit;
5180
5181         old_tss_base = kvm_mmu_gva_to_gpa_write(vcpu, old_tss_base, NULL);
5182
5183         /* FIXME: Handle errors. Failure to read either TSS or their
5184          * descriptors should generate a pagefault.
5185          */
5186         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
5187                 goto out;
5188
5189         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
5190                 goto out;
5191
5192         if (reason != TASK_SWITCH_IRET) {
5193                 int cpl;
5194
5195                 cpl = kvm_x86_ops->get_cpl(vcpu);
5196                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
5197                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5198                         return 1;
5199                 }
5200         }
5201
5202         desc_limit = get_desc_limit(&nseg_desc);
5203         if (!nseg_desc.p ||
5204             ((desc_limit < 0x67 && (nseg_desc.type & 8)) ||
5205              desc_limit < 0x2b)) {
5206                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
5207                 return 1;
5208         }
5209
5210         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
5211                 cseg_desc.type &= ~(1 << 1); //clear the B flag
5212                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
5213         }
5214
5215         if (reason == TASK_SWITCH_IRET) {
5216                 u32 eflags = kvm_get_rflags(vcpu);
5217                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
5218         }
5219
5220         /* set back link to prev task only if NT bit is set in eflags
5221            note that old_tss_sel is not used afetr this point */
5222         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
5223                 old_tss_sel = 0xffff;
5224
5225         if (nseg_desc.type & 8)
5226                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
5227                                          old_tss_base, &nseg_desc);
5228         else
5229                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
5230                                          old_tss_base, &nseg_desc);
5231
5232         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
5233                 u32 eflags = kvm_get_rflags(vcpu);
5234                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
5235         }
5236
5237         if (reason != TASK_SWITCH_IRET) {
5238                 nseg_desc.type |= (1 << 1);
5239                 save_guest_segment_descriptor(vcpu, tss_selector,
5240                                               &nseg_desc);
5241         }
5242
5243         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5244         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5245         tr_seg.type = 11;
5246         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5247 out:
5248         return ret;
5249 }
5250 EXPORT_SYMBOL_GPL(kvm_task_switch);
5251
5252 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5253                                   struct kvm_sregs *sregs)
5254 {
5255         int mmu_reset_needed = 0;
5256         int pending_vec, max_bits;
5257         struct descriptor_table dt;
5258
5259         vcpu_load(vcpu);
5260
5261         dt.limit = sregs->idt.limit;
5262         dt.base = sregs->idt.base;
5263         kvm_x86_ops->set_idt(vcpu, &dt);
5264         dt.limit = sregs->gdt.limit;
5265         dt.base = sregs->gdt.base;
5266         kvm_x86_ops->set_gdt(vcpu, &dt);
5267
5268         vcpu->arch.cr2 = sregs->cr2;
5269         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5270         vcpu->arch.cr3 = sregs->cr3;
5271
5272         kvm_set_cr8(vcpu, sregs->cr8);
5273
5274         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5275         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5276         kvm_set_apic_base(vcpu, sregs->apic_base);
5277
5278         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5279         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5280         vcpu->arch.cr0 = sregs->cr0;
5281
5282         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5283         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5284         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5285                 load_pdptrs(vcpu, vcpu->arch.cr3);
5286                 mmu_reset_needed = 1;
5287         }
5288
5289         if (mmu_reset_needed)
5290                 kvm_mmu_reset_context(vcpu);
5291
5292         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5293         pending_vec = find_first_bit(
5294                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5295         if (pending_vec < max_bits) {
5296                 kvm_queue_interrupt(vcpu, pending_vec, false);
5297                 pr_debug("Set back pending irq %d\n", pending_vec);
5298                 if (irqchip_in_kernel(vcpu->kvm))
5299                         kvm_pic_clear_isr_ack(vcpu->kvm);
5300         }
5301
5302         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5303         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5304         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5305         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5306         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5307         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5308
5309         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5310         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5311
5312         update_cr8_intercept(vcpu);
5313
5314         /* Older userspace won't unhalt the vcpu on reset. */
5315         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5316             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5317             !is_protmode(vcpu))
5318                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5319
5320         vcpu_put(vcpu);
5321
5322         return 0;
5323 }
5324
5325 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5326                                         struct kvm_guest_debug *dbg)
5327 {
5328         unsigned long rflags;
5329         int i, r;
5330
5331         vcpu_load(vcpu);
5332
5333         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5334                 r = -EBUSY;
5335                 if (vcpu->arch.exception.pending)
5336                         goto unlock_out;
5337                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5338                         kvm_queue_exception(vcpu, DB_VECTOR);
5339                 else
5340                         kvm_queue_exception(vcpu, BP_VECTOR);
5341         }
5342
5343         /*
5344          * Read rflags as long as potentially injected trace flags are still
5345          * filtered out.
5346          */
5347         rflags = kvm_get_rflags(vcpu);
5348
5349         vcpu->guest_debug = dbg->control;
5350         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5351                 vcpu->guest_debug = 0;
5352
5353         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5354                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5355                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5356                 vcpu->arch.switch_db_regs =
5357                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5358         } else {
5359                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5360                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5361                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5362         }
5363
5364         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5365                 vcpu->arch.singlestep_cs =
5366                         get_segment_selector(vcpu, VCPU_SREG_CS);
5367                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
5368         }
5369
5370         /*
5371          * Trigger an rflags update that will inject or remove the trace
5372          * flags.
5373          */
5374         kvm_set_rflags(vcpu, rflags);
5375
5376         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5377
5378         r = 0;
5379
5380 unlock_out:
5381         vcpu_put(vcpu);
5382
5383         return r;
5384 }
5385
5386 /*
5387  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
5388  * we have asm/x86/processor.h
5389  */
5390 struct fxsave {
5391         u16     cwd;
5392         u16     swd;
5393         u16     twd;
5394         u16     fop;
5395         u64     rip;
5396         u64     rdp;
5397         u32     mxcsr;
5398         u32     mxcsr_mask;
5399         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
5400 #ifdef CONFIG_X86_64
5401         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
5402 #else
5403         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
5404 #endif
5405 };
5406
5407 /*
5408  * Translate a guest virtual address to a guest physical address.
5409  */
5410 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5411                                     struct kvm_translation *tr)
5412 {
5413         unsigned long vaddr = tr->linear_address;
5414         gpa_t gpa;
5415         int idx;
5416
5417         vcpu_load(vcpu);
5418         idx = srcu_read_lock(&vcpu->kvm->srcu);
5419         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5420         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5421         tr->physical_address = gpa;
5422         tr->valid = gpa != UNMAPPED_GVA;
5423         tr->writeable = 1;
5424         tr->usermode = 0;
5425         vcpu_put(vcpu);
5426
5427         return 0;
5428 }
5429
5430 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5431 {
5432         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5433
5434         vcpu_load(vcpu);
5435
5436         memcpy(fpu->fpr, fxsave->st_space, 128);
5437         fpu->fcw = fxsave->cwd;
5438         fpu->fsw = fxsave->swd;
5439         fpu->ftwx = fxsave->twd;
5440         fpu->last_opcode = fxsave->fop;
5441         fpu->last_ip = fxsave->rip;
5442         fpu->last_dp = fxsave->rdp;
5443         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5444
5445         vcpu_put(vcpu);
5446
5447         return 0;
5448 }
5449
5450 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5451 {
5452         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5453
5454         vcpu_load(vcpu);
5455
5456         memcpy(fxsave->st_space, fpu->fpr, 128);
5457         fxsave->cwd = fpu->fcw;
5458         fxsave->swd = fpu->fsw;
5459         fxsave->twd = fpu->ftwx;
5460         fxsave->fop = fpu->last_opcode;
5461         fxsave->rip = fpu->last_ip;
5462         fxsave->rdp = fpu->last_dp;
5463         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5464
5465         vcpu_put(vcpu);
5466
5467         return 0;
5468 }
5469
5470 void fx_init(struct kvm_vcpu *vcpu)
5471 {
5472         unsigned after_mxcsr_mask;
5473
5474         /*
5475          * Touch the fpu the first time in non atomic context as if
5476          * this is the first fpu instruction the exception handler
5477          * will fire before the instruction returns and it'll have to
5478          * allocate ram with GFP_KERNEL.
5479          */
5480         if (!used_math())
5481                 kvm_fx_save(&vcpu->arch.host_fx_image);
5482
5483         /* Initialize guest FPU by resetting ours and saving into guest's */
5484         preempt_disable();
5485         kvm_fx_save(&vcpu->arch.host_fx_image);
5486         kvm_fx_finit();
5487         kvm_fx_save(&vcpu->arch.guest_fx_image);
5488         kvm_fx_restore(&vcpu->arch.host_fx_image);
5489         preempt_enable();
5490
5491         vcpu->arch.cr0 |= X86_CR0_ET;
5492         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5493         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5494         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5495                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5496 }
5497 EXPORT_SYMBOL_GPL(fx_init);
5498
5499 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5500 {
5501         if (vcpu->guest_fpu_loaded)
5502                 return;
5503
5504         vcpu->guest_fpu_loaded = 1;
5505         kvm_fx_save(&vcpu->arch.host_fx_image);
5506         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5507         trace_kvm_fpu(1);
5508 }
5509
5510 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5511 {
5512         if (!vcpu->guest_fpu_loaded)
5513                 return;
5514
5515         vcpu->guest_fpu_loaded = 0;
5516         kvm_fx_save(&vcpu->arch.guest_fx_image);
5517         kvm_fx_restore(&vcpu->arch.host_fx_image);
5518         ++vcpu->stat.fpu_reload;
5519         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5520         trace_kvm_fpu(0);
5521 }
5522
5523 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5524 {
5525         if (vcpu->arch.time_page) {
5526                 kvm_release_page_dirty(vcpu->arch.time_page);
5527                 vcpu->arch.time_page = NULL;
5528         }
5529
5530         kvm_x86_ops->vcpu_free(vcpu);
5531 }
5532
5533 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5534                                                 unsigned int id)
5535 {
5536         return kvm_x86_ops->vcpu_create(kvm, id);
5537 }
5538
5539 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5540 {
5541         int r;
5542
5543         /* We do fxsave: this must be aligned. */
5544         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5545
5546         vcpu->arch.mtrr_state.have_fixed = 1;
5547         vcpu_load(vcpu);
5548         r = kvm_arch_vcpu_reset(vcpu);
5549         if (r == 0)
5550                 r = kvm_mmu_setup(vcpu);
5551         vcpu_put(vcpu);
5552         if (r < 0)
5553                 goto free_vcpu;
5554
5555         return 0;
5556 free_vcpu:
5557         kvm_x86_ops->vcpu_free(vcpu);
5558         return r;
5559 }
5560
5561 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5562 {
5563         vcpu_load(vcpu);
5564         kvm_mmu_unload(vcpu);
5565         vcpu_put(vcpu);
5566
5567         kvm_x86_ops->vcpu_free(vcpu);
5568 }
5569
5570 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5571 {
5572         vcpu->arch.nmi_pending = false;
5573         vcpu->arch.nmi_injected = false;
5574
5575         vcpu->arch.switch_db_regs = 0;
5576         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5577         vcpu->arch.dr6 = DR6_FIXED_1;
5578         vcpu->arch.dr7 = DR7_FIXED_1;
5579
5580         return kvm_x86_ops->vcpu_reset(vcpu);
5581 }
5582
5583 int kvm_arch_hardware_enable(void *garbage)
5584 {
5585         /*
5586          * Since this may be called from a hotplug notifcation,
5587          * we can't get the CPU frequency directly.
5588          */
5589         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5590                 int cpu = raw_smp_processor_id();
5591                 per_cpu(cpu_tsc_khz, cpu) = 0;
5592         }
5593
5594         kvm_shared_msr_cpu_online();
5595
5596         return kvm_x86_ops->hardware_enable(garbage);
5597 }
5598
5599 void kvm_arch_hardware_disable(void *garbage)
5600 {
5601         kvm_x86_ops->hardware_disable(garbage);
5602         drop_user_return_notifiers(garbage);
5603 }
5604
5605 int kvm_arch_hardware_setup(void)
5606 {
5607         return kvm_x86_ops->hardware_setup();
5608 }
5609
5610 void kvm_arch_hardware_unsetup(void)
5611 {
5612         kvm_x86_ops->hardware_unsetup();
5613 }
5614
5615 void kvm_arch_check_processor_compat(void *rtn)
5616 {
5617         kvm_x86_ops->check_processor_compatibility(rtn);
5618 }
5619
5620 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5621 {
5622         struct page *page;
5623         struct kvm *kvm;
5624         int r;
5625
5626         BUG_ON(vcpu->kvm == NULL);
5627         kvm = vcpu->kvm;
5628
5629         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5630         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5631                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5632         else
5633                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5634
5635         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5636         if (!page) {
5637                 r = -ENOMEM;
5638                 goto fail;
5639         }
5640         vcpu->arch.pio_data = page_address(page);
5641
5642         r = kvm_mmu_create(vcpu);
5643         if (r < 0)
5644                 goto fail_free_pio_data;
5645
5646         if (irqchip_in_kernel(kvm)) {
5647                 r = kvm_create_lapic(vcpu);
5648                 if (r < 0)
5649                         goto fail_mmu_destroy;
5650         }
5651
5652         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5653                                        GFP_KERNEL);
5654         if (!vcpu->arch.mce_banks) {
5655                 r = -ENOMEM;
5656                 goto fail_free_lapic;
5657         }
5658         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5659
5660         return 0;
5661 fail_free_lapic:
5662         kvm_free_lapic(vcpu);
5663 fail_mmu_destroy:
5664         kvm_mmu_destroy(vcpu);
5665 fail_free_pio_data:
5666         free_page((unsigned long)vcpu->arch.pio_data);
5667 fail:
5668         return r;
5669 }
5670
5671 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5672 {
5673         int idx;
5674
5675         kfree(vcpu->arch.mce_banks);
5676         kvm_free_lapic(vcpu);
5677         idx = srcu_read_lock(&vcpu->kvm->srcu);
5678         kvm_mmu_destroy(vcpu);
5679         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5680         free_page((unsigned long)vcpu->arch.pio_data);
5681 }
5682
5683 struct  kvm *kvm_arch_create_vm(void)
5684 {
5685         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5686
5687         if (!kvm)
5688                 return ERR_PTR(-ENOMEM);
5689
5690         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5691         if (!kvm->arch.aliases) {
5692                 kfree(kvm);
5693                 return ERR_PTR(-ENOMEM);
5694         }
5695
5696         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5697         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5698
5699         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5700         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5701
5702         rdtscll(kvm->arch.vm_init_tsc);
5703
5704         return kvm;
5705 }
5706
5707 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5708 {
5709         vcpu_load(vcpu);
5710         kvm_mmu_unload(vcpu);
5711         vcpu_put(vcpu);
5712 }
5713
5714 static void kvm_free_vcpus(struct kvm *kvm)
5715 {
5716         unsigned int i;
5717         struct kvm_vcpu *vcpu;
5718
5719         /*
5720          * Unpin any mmu pages first.
5721          */
5722         kvm_for_each_vcpu(i, vcpu, kvm)
5723                 kvm_unload_vcpu_mmu(vcpu);
5724         kvm_for_each_vcpu(i, vcpu, kvm)
5725                 kvm_arch_vcpu_free(vcpu);
5726
5727         mutex_lock(&kvm->lock);
5728         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5729                 kvm->vcpus[i] = NULL;
5730
5731         atomic_set(&kvm->online_vcpus, 0);
5732         mutex_unlock(&kvm->lock);
5733 }
5734
5735 void kvm_arch_sync_events(struct kvm *kvm)
5736 {
5737         kvm_free_all_assigned_devices(kvm);
5738 }
5739
5740 void kvm_arch_destroy_vm(struct kvm *kvm)
5741 {
5742         kvm_iommu_unmap_guest(kvm);
5743         kvm_free_pit(kvm);
5744         kfree(kvm->arch.vpic);
5745         kfree(kvm->arch.vioapic);
5746         kvm_free_vcpus(kvm);
5747         kvm_free_physmem(kvm);
5748         if (kvm->arch.apic_access_page)
5749                 put_page(kvm->arch.apic_access_page);
5750         if (kvm->arch.ept_identity_pagetable)
5751                 put_page(kvm->arch.ept_identity_pagetable);
5752         cleanup_srcu_struct(&kvm->srcu);
5753         kfree(kvm->arch.aliases);
5754         kfree(kvm);
5755 }
5756
5757 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5758                                 struct kvm_memory_slot *memslot,
5759                                 struct kvm_memory_slot old,
5760                                 struct kvm_userspace_memory_region *mem,
5761                                 int user_alloc)
5762 {
5763         int npages = memslot->npages;
5764
5765         /*To keep backward compatibility with older userspace,
5766          *x86 needs to hanlde !user_alloc case.
5767          */
5768         if (!user_alloc) {
5769                 if (npages && !old.rmap) {
5770                         unsigned long userspace_addr;
5771
5772                         down_write(&current->mm->mmap_sem);
5773                         userspace_addr = do_mmap(NULL, 0,
5774                                                  npages * PAGE_SIZE,
5775                                                  PROT_READ | PROT_WRITE,
5776                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5777                                                  0);
5778                         up_write(&current->mm->mmap_sem);
5779
5780                         if (IS_ERR((void *)userspace_addr))
5781                                 return PTR_ERR((void *)userspace_addr);
5782
5783                         memslot->userspace_addr = userspace_addr;
5784                 }
5785         }
5786
5787
5788         return 0;
5789 }
5790
5791 void kvm_arch_commit_memory_region(struct kvm *kvm,
5792                                 struct kvm_userspace_memory_region *mem,
5793                                 struct kvm_memory_slot old,
5794                                 int user_alloc)
5795 {
5796
5797         int npages = mem->memory_size >> PAGE_SHIFT;
5798
5799         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5800                 int ret;
5801
5802                 down_write(&current->mm->mmap_sem);
5803                 ret = do_munmap(current->mm, old.userspace_addr,
5804                                 old.npages * PAGE_SIZE);
5805                 up_write(&current->mm->mmap_sem);
5806                 if (ret < 0)
5807                         printk(KERN_WARNING
5808                                "kvm_vm_ioctl_set_memory_region: "
5809                                "failed to munmap memory\n");
5810         }
5811
5812         spin_lock(&kvm->mmu_lock);
5813         if (!kvm->arch.n_requested_mmu_pages) {
5814                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5815                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5816         }
5817
5818         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5819         spin_unlock(&kvm->mmu_lock);
5820 }
5821
5822 void kvm_arch_flush_shadow(struct kvm *kvm)
5823 {
5824         kvm_mmu_zap_all(kvm);
5825         kvm_reload_remote_mmus(kvm);
5826 }
5827
5828 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5829 {
5830         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5831                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5832                 || vcpu->arch.nmi_pending ||
5833                 (kvm_arch_interrupt_allowed(vcpu) &&
5834                  kvm_cpu_has_interrupt(vcpu));
5835 }
5836
5837 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5838 {
5839         int me;
5840         int cpu = vcpu->cpu;
5841
5842         if (waitqueue_active(&vcpu->wq)) {
5843                 wake_up_interruptible(&vcpu->wq);
5844                 ++vcpu->stat.halt_wakeup;
5845         }
5846
5847         me = get_cpu();
5848         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5849                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5850                         smp_send_reschedule(cpu);
5851         put_cpu();
5852 }
5853
5854 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5855 {
5856         return kvm_x86_ops->interrupt_allowed(vcpu);
5857 }
5858
5859 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5860 {
5861         unsigned long rflags;
5862
5863         rflags = kvm_x86_ops->get_rflags(vcpu);
5864         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5865                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5866         return rflags;
5867 }
5868 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5869
5870 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5871 {
5872         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5873             vcpu->arch.singlestep_cs ==
5874                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5875             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5876                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5877         kvm_x86_ops->set_rflags(vcpu, rflags);
5878 }
5879 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5880
5881 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5882 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5883 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5884 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5885 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5886 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5887 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5888 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5889 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5890 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5891 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);