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 <trace/events/kvm.h>
  43 #undef TRACE_INCLUDE_FILE
  44 #define CREATE_TRACE_POINTS
  45 #include "trace.h"
  46
  47 #include <asm/debugreg.h>
  48 #include <asm/uaccess.h>
  49 #include <asm/msr.h>
  50 #include <asm/desc.h>
  51 #include <asm/mtrr.h>
  52 #include <asm/mce.h>
  53
  54 #define MAX_IO_MSRS 256
  55 #define CR0_RESERVED_BITS                                               \
  56         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  57                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  58                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  59 #define CR4_RESERVED_BITS                                               \
  60         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  61                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  62                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  63                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  64
  65 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  66
  67 #define KVM_MAX_MCE_BANKS 32
  68 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  69
  70 /* EFER defaults:
  71  * - enable syscall per default because its emulated by KVM
  72  * - enable LME and LMA per default on 64 bit KVM
  73  */
  74 #ifdef CONFIG_X86_64
  75 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  76 #else
  77 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  78 #endif
  79
  80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  82
  83 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
  84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  85                                     struct kvm_cpuid_entry2 __user *entries);
  86
  87 struct kvm_x86_ops *kvm_x86_ops;
  88 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  89
  90 int ignore_msrs = 0;
  91 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  92
  93 #define KVM_NR_SHARED_MSRS 16
  94
  95 struct kvm_shared_msrs_global {
  96         int nr;
  97         u32 msrs[KVM_NR_SHARED_MSRS];
  98 };
  99
 100 struct kvm_shared_msrs {
 101         struct user_return_notifier urn;
 102         bool registered;
 103         struct kvm_shared_msr_values {
 104                 u64 host;
 105                 u64 curr;
 106         } values[KVM_NR_SHARED_MSRS];
 107 };
 108
 109 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
 110 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
 111
 112 struct kvm_stats_debugfs_item debugfs_entries[] = {
 113         { "pf_fixed", VCPU_STAT(pf_fixed) },
 114         { "pf_guest", VCPU_STAT(pf_guest) },
 115         { "tlb_flush", VCPU_STAT(tlb_flush) },
 116         { "invlpg", VCPU_STAT(invlpg) },
 117         { "exits", VCPU_STAT(exits) },
 118         { "io_exits", VCPU_STAT(io_exits) },
 119         { "mmio_exits", VCPU_STAT(mmio_exits) },
 120         { "signal_exits", VCPU_STAT(signal_exits) },
 121         { "irq_window", VCPU_STAT(irq_window_exits) },
 122         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 123         { "halt_exits", VCPU_STAT(halt_exits) },
 124         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 125         { "hypercalls", VCPU_STAT(hypercalls) },
 126         { "request_irq", VCPU_STAT(request_irq_exits) },
 127         { "irq_exits", VCPU_STAT(irq_exits) },
 128         { "host_state_reload", VCPU_STAT(host_state_reload) },
 129         { "efer_reload", VCPU_STAT(efer_reload) },
 130         { "fpu_reload", VCPU_STAT(fpu_reload) },
 131         { "insn_emulation", VCPU_STAT(insn_emulation) },
 132         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 133         { "irq_injections", VCPU_STAT(irq_injections) },
 134         { "nmi_injections", VCPU_STAT(nmi_injections) },
 135         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 136         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 137         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 138         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 139         { "mmu_flooded", VM_STAT(mmu_flooded) },
 140         { "mmu_recycled", VM_STAT(mmu_recycled) },
 141         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 142         { "mmu_unsync", VM_STAT(mmu_unsync) },
 143         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 144         { "largepages", VM_STAT(lpages) },
 145         { NULL }
 146 };
 147
 148 static void kvm_on_user_return(struct user_return_notifier *urn)
 149 {
 150         unsigned slot;
 151         struct kvm_shared_msrs *locals
 152                 = container_of(urn, struct kvm_shared_msrs, urn);
 153         struct kvm_shared_msr_values *values;
 154
 155         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
 156                 values = &locals->values[slot];
 157                 if (values->host != values->curr) {
 158                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
 159                         values->curr = values->host;
 160                 }
 161         }
 162         locals->registered = false;
 163         user_return_notifier_unregister(urn);
 164 }
 165
 166 static void shared_msr_update(unsigned slot, u32 msr)
 167 {
 168         struct kvm_shared_msrs *smsr;
 169         u64 value;
 170
 171         smsr = &__get_cpu_var(shared_msrs);
 172         /* only read, and nobody should modify it at this time,
 173          * so don't need lock */
 174         if (slot >= shared_msrs_global.nr) {
 175                 printk(KERN_ERR "kvm: invalid MSR slot!");
 176                 return;
 177         }
 178         rdmsrl_safe(msr, &value);
 179         smsr->values[slot].host = value;
 180         smsr->values[slot].curr = value;
 181 }
 182
 183 void kvm_define_shared_msr(unsigned slot, u32 msr)
 184 {
 185         if (slot >= shared_msrs_global.nr)
 186                 shared_msrs_global.nr = slot + 1;
 187         shared_msrs_global.msrs[slot] = msr;
 188         /* we need ensured the shared_msr_global have been updated */
 189         smp_wmb();
 190 }
 191 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
 192
 193 static void kvm_shared_msr_cpu_online(void)
 194 {
 195         unsigned i;
 196
 197         for (i = 0; i < shared_msrs_global.nr; ++i)
 198                 shared_msr_update(i, shared_msrs_global.msrs[i]);
 199 }
 200
 201 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
 202 {
 203         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 204
 205         if (((value ^ smsr->values[slot].curr) & mask) == 0)
 206                 return;
 207         smsr->values[slot].curr = value;
 208         wrmsrl(shared_msrs_global.msrs[slot], value);
 209         if (!smsr->registered) {
 210                 smsr->urn.on_user_return = kvm_on_user_return;
 211                 user_return_notifier_register(&smsr->urn);
 212                 smsr->registered = true;
 213         }
 214 }
 215 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
 216
 217 static void drop_user_return_notifiers(void *ignore)
 218 {
 219         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
 220
 221         if (smsr->registered)
 222                 kvm_on_user_return(&smsr->urn);
 223 }
 224
 225 unsigned long segment_base(u16 selector)
 226 {
 227         struct descriptor_table gdt;
 228         struct desc_struct *d;
 229         unsigned long table_base;
 230         unsigned long v;
 231
 232         if (selector == 0)
 233                 return 0;
 234
 235         kvm_get_gdt(&gdt);
 236         table_base = gdt.base;
 237
 238         if (selector & 4) {           /* from ldt */
 239                 u16 ldt_selector = kvm_read_ldt();
 240
 241                 table_base = segment_base(ldt_selector);
 242         }
 243         d = (struct desc_struct *)(table_base + (selector & ~7));
 244         v = get_desc_base(d);
 245 #ifdef CONFIG_X86_64
 246         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 247                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 248 #endif
 249         return v;
 250 }
 251 EXPORT_SYMBOL_GPL(segment_base);
 252
 253 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 254 {
 255         if (irqchip_in_kernel(vcpu->kvm))
 256                 return vcpu->arch.apic_base;
 257         else
 258                 return vcpu->arch.apic_base;
 259 }
 260 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 261
 262 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 263 {
 264         /* TODO: reserve bits check */
 265         if (irqchip_in_kernel(vcpu->kvm))
 266                 kvm_lapic_set_base(vcpu, data);
 267         else
 268                 vcpu->arch.apic_base = data;
 269 }
 270 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 271
 272 #define EXCPT_BENIGN            0
 273 #define EXCPT_CONTRIBUTORY      1
 274 #define EXCPT_PF                2
 275
 276 static int exception_class(int vector)
 277 {
 278         switch (vector) {
 279         case PF_VECTOR:
 280                 return EXCPT_PF;
 281         case DE_VECTOR:
 282         case TS_VECTOR:
 283         case NP_VECTOR:
 284         case SS_VECTOR:
 285         case GP_VECTOR:
 286                 return EXCPT_CONTRIBUTORY;
 287         default:
 288                 break;
 289         }
 290         return EXCPT_BENIGN;
 291 }
 292
 293 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
 294                 unsigned nr, bool has_error, u32 error_code)
 295 {
 296         u32 prev_nr;
 297         int class1, class2;
 298
 299         if (!vcpu->arch.exception.pending) {
 300         queue:
 301                 vcpu->arch.exception.pending = true;
 302                 vcpu->arch.exception.has_error_code = has_error;
 303                 vcpu->arch.exception.nr = nr;
 304                 vcpu->arch.exception.error_code = error_code;
 305                 return;
 306         }
 307
 308         /* to check exception */
 309         prev_nr = vcpu->arch.exception.nr;
 310         if (prev_nr == DF_VECTOR) {
 311                 /* triple fault -> shutdown */
 312                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 313                 return;
 314         }
 315         class1 = exception_class(prev_nr);
 316         class2 = exception_class(nr);
 317         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
 318                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
 319                 /* generate double fault per SDM Table 5-5 */
 320                 vcpu->arch.exception.pending = true;
 321                 vcpu->arch.exception.has_error_code = true;
 322                 vcpu->arch.exception.nr = DF_VECTOR;
 323                 vcpu->arch.exception.error_code = 0;
 324         } else
 325                 /* replace previous exception with a new one in a hope
 326                    that instruction re-execution will regenerate lost
 327                    exception */
 328                 goto queue;
 329 }
 330
 331 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 332 {
 333         kvm_multiple_exception(vcpu, nr, false, 0);
 334 }
 335 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 336
 337 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 338                            u32 error_code)
 339 {
 340         ++vcpu->stat.pf_guest;
 341         vcpu->arch.cr2 = addr;
 342         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 343 }
 344
 345 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 346 {
 347         vcpu->arch.nmi_pending = 1;
 348 }
 349 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 350
 351 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 352 {
 353         kvm_multiple_exception(vcpu, nr, true, error_code);
 354 }
 355 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 356
 357 /*
 358  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
 359  * a #GP and return false.
 360  */
 361 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
 362 {
 363         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
 364                 return true;
 365         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
 366         return false;
 367 }
 368 EXPORT_SYMBOL_GPL(kvm_require_cpl);
 369
 370 /*
 371  * Load the pae pdptrs.  Return true is they are all valid.
 372  */
 373 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 374 {
 375         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 376         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 377         int i;
 378         int ret;
 379         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 380
 381         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 382                                   offset * sizeof(u64), sizeof(pdpte));
 383         if (ret < 0) {
 384                 ret = 0;
 385                 goto out;
 386         }
 387         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 388                 if (is_present_gpte(pdpte[i]) &&
 389                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 390                         ret = 0;
 391                         goto out;
 392                 }
 393         }
 394         ret = 1;
 395
 396         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 397         __set_bit(VCPU_EXREG_PDPTR,
 398                   (unsigned long *)&vcpu->arch.regs_avail);
 399         __set_bit(VCPU_EXREG_PDPTR,
 400                   (unsigned long *)&vcpu->arch.regs_dirty);
 401 out:
 402
 403         return ret;
 404 }
 405 EXPORT_SYMBOL_GPL(load_pdptrs);
 406
 407 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 408 {
 409         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 410         bool changed = true;
 411         int r;
 412
 413         if (is_long_mode(vcpu) || !is_pae(vcpu))
 414                 return false;
 415
 416         if (!test_bit(VCPU_EXREG_PDPTR,
 417                       (unsigned long *)&vcpu->arch.regs_avail))
 418                 return true;
 419
 420         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 421         if (r < 0)
 422                 goto out;
 423         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 424 out:
 425
 426         return changed;
 427 }
 428
 429 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 430 {
 431         cr0 |= X86_CR0_ET;
 432
 433 #ifdef CONFIG_X86_64
 434         if (cr0 & 0xffffffff00000000UL) {
 435                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 436                        cr0, kvm_read_cr0(vcpu));
 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                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 446                 kvm_inject_gp(vcpu, 0);
 447                 return;
 448         }
 449
 450         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 451                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 452                        "and a clear PE flag\n");
 453                 kvm_inject_gp(vcpu, 0);
 454                 return;
 455         }
 456
 457         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 458 #ifdef CONFIG_X86_64
 459                 if ((vcpu->arch.efer & EFER_LME)) {
 460                         int cs_db, cs_l;
 461
 462                         if (!is_pae(vcpu)) {
 463                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 464                                        "in long mode while PAE is disabled\n");
 465                                 kvm_inject_gp(vcpu, 0);
 466                                 return;
 467                         }
 468                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 469                         if (cs_l) {
 470                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 471                                        "in long mode while CS.L == 1\n");
 472                                 kvm_inject_gp(vcpu, 0);
 473                                 return;
 474
 475                         }
 476                 } else
 477 #endif
 478                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 479                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 480                                "reserved bits\n");
 481                         kvm_inject_gp(vcpu, 0);
 482                         return;
 483                 }
 484
 485         }
 486
 487         kvm_x86_ops->set_cr0(vcpu, cr0);
 488         vcpu->arch.cr0 = cr0;
 489
 490         kvm_mmu_reset_context(vcpu);
 491         return;
 492 }
 493 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 494
 495 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 496 {
 497         kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
 498 }
 499 EXPORT_SYMBOL_GPL(kvm_lmsw);
 500
 501 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 502 {
 503         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 504         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 505
 506         if (cr4 & CR4_RESERVED_BITS) {
 507                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 508                 kvm_inject_gp(vcpu, 0);
 509                 return;
 510         }
 511
 512         if (is_long_mode(vcpu)) {
 513                 if (!(cr4 & X86_CR4_PAE)) {
 514                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 515                                "in long mode\n");
 516                         kvm_inject_gp(vcpu, 0);
 517                         return;
 518                 }
 519         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 520                    && ((cr4 ^ old_cr4) & pdptr_bits)
 521                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 522                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 523                 kvm_inject_gp(vcpu, 0);
 524                 return;
 525         }
 526
 527         if (cr4 & X86_CR4_VMXE) {
 528                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 529                 kvm_inject_gp(vcpu, 0);
 530                 return;
 531         }
 532         kvm_x86_ops->set_cr4(vcpu, cr4);
 533         vcpu->arch.cr4 = cr4;
 534         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 535         kvm_mmu_reset_context(vcpu);
 536 }
 537 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 538
 539 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 540 {
 541         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 542                 kvm_mmu_sync_roots(vcpu);
 543                 kvm_mmu_flush_tlb(vcpu);
 544                 return;
 545         }
 546
 547         if (is_long_mode(vcpu)) {
 548                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 549                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 550                         kvm_inject_gp(vcpu, 0);
 551                         return;
 552                 }
 553         } else {
 554                 if (is_pae(vcpu)) {
 555                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 556                                 printk(KERN_DEBUG
 557                                        "set_cr3: #GP, reserved bits\n");
 558                                 kvm_inject_gp(vcpu, 0);
 559                                 return;
 560                         }
 561                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 562                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 563                                        "reserved bits\n");
 564                                 kvm_inject_gp(vcpu, 0);
 565                                 return;
 566                         }
 567                 }
 568                 /*
 569                  * We don't check reserved bits in nonpae mode, because
 570                  * this isn't enforced, and VMware depends on this.
 571                  */
 572         }
 573
 574         /*
 575          * Does the new cr3 value map to physical memory? (Note, we
 576          * catch an invalid cr3 even in real-mode, because it would
 577          * cause trouble later on when we turn on paging anyway.)
 578          *
 579          * A real CPU would silently accept an invalid cr3 and would
 580          * attempt to use it - with largely undefined (and often hard
 581          * to debug) behavior on the guest side.
 582          */
 583         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 584                 kvm_inject_gp(vcpu, 0);
 585         else {
 586                 vcpu->arch.cr3 = cr3;
 587                 vcpu->arch.mmu.new_cr3(vcpu);
 588         }
 589 }
 590 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 591
 592 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 593 {
 594         if (cr8 & CR8_RESERVED_BITS) {
 595                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 596                 kvm_inject_gp(vcpu, 0);
 597                 return;
 598         }
 599         if (irqchip_in_kernel(vcpu->kvm))
 600                 kvm_lapic_set_tpr(vcpu, cr8);
 601         else
 602                 vcpu->arch.cr8 = cr8;
 603 }
 604 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 605
 606 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 607 {
 608         if (irqchip_in_kernel(vcpu->kvm))
 609                 return kvm_lapic_get_cr8(vcpu);
 610         else
 611                 return vcpu->arch.cr8;
 612 }
 613 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 614
 615 static inline u32 bit(int bitno)
 616 {
 617         return 1 << (bitno & 31);
 618 }
 619
 620 /*
 621  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 622  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 623  *
 624  * This list is modified at module load time to reflect the
 625  * capabilities of the host cpu. This capabilities test skips MSRs that are
 626  * kvm-specific. Those are put in the beginning of the list.
 627  */
 628
 629 #define KVM_SAVE_MSRS_BEGIN     5
 630 static u32 msrs_to_save[] = {
 631         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 632         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
 633         HV_X64_MSR_APIC_ASSIST_PAGE,
 634         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 635         MSR_K6_STAR,
 636 #ifdef CONFIG_X86_64
 637         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 638 #endif
 639         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 640 };
 641
 642 static unsigned num_msrs_to_save;
 643
 644 static u32 emulated_msrs[] = {
 645         MSR_IA32_MISC_ENABLE,
 646 };
 647
 648 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 649 {
 650         if (efer & efer_reserved_bits) {
 651                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 652                        efer);
 653                 kvm_inject_gp(vcpu, 0);
 654                 return;
 655         }
 656
 657         if (is_paging(vcpu)
 658             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) {
 659                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 660                 kvm_inject_gp(vcpu, 0);
 661                 return;
 662         }
 663
 664         if (efer & EFER_FFXSR) {
 665                 struct kvm_cpuid_entry2 *feat;
 666
 667                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 668                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 669                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 670                         kvm_inject_gp(vcpu, 0);
 671                         return;
 672                 }
 673         }
 674
 675         if (efer & EFER_SVME) {
 676                 struct kvm_cpuid_entry2 *feat;
 677
 678                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 679                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 680                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 681                         kvm_inject_gp(vcpu, 0);
 682                         return;
 683                 }
 684         }
 685
 686         kvm_x86_ops->set_efer(vcpu, efer);
 687
 688         efer &= ~EFER_LMA;
 689         efer |= vcpu->arch.efer & EFER_LMA;
 690
 691         vcpu->arch.efer = efer;
 692
 693         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 694         kvm_mmu_reset_context(vcpu);
 695 }
 696
 697 void kvm_enable_efer_bits(u64 mask)
 698 {
 699        efer_reserved_bits &= ~mask;
 700 }
 701 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 702
 703
 704 /*
 705  * Writes msr value into into the appropriate "register".
 706  * Returns 0 on success, non-0 otherwise.
 707  * Assumes vcpu_load() was already called.
 708  */
 709 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 710 {
 711         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 712 }
 713
 714 /*
 715  * Adapt set_msr() to msr_io()'s calling convention
 716  */
 717 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 718 {
 719         return kvm_set_msr(vcpu, index, *data);
 720 }
 721
 722 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 723 {
 724         static int version;
 725         struct pvclock_wall_clock wc;
 726         struct timespec boot;
 727
 728         if (!wall_clock)
 729                 return;
 730
 731         version++;
 732
 733         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 734
 735         /*
 736          * The guest calculates current wall clock time by adding
 737          * system time (updated by kvm_write_guest_time below) to the
 738          * wall clock specified here.  guest system time equals host
 739          * system time for us, thus we must fill in host boot time here.
 740          */
 741         getboottime(&boot);
 742
 743         wc.sec = boot.tv_sec;
 744         wc.nsec = boot.tv_nsec;
 745         wc.version = version;
 746
 747         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 748
 749         version++;
 750         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 751 }
 752
 753 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 754 {
 755         uint32_t quotient, remainder;
 756
 757         /* Don't try to replace with do_div(), this one calculates
 758          * "(dividend << 32) / divisor" */
 759         __asm__ ( "divl %4"
 760                   : "=a" (quotient), "=d" (remainder)
 761                   : "0" (0), "1" (dividend), "r" (divisor) );
 762         return quotient;
 763 }
 764
 765 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 766 {
 767         uint64_t nsecs = 1000000000LL;
 768         int32_t  shift = 0;
 769         uint64_t tps64;
 770         uint32_t tps32;
 771
 772         tps64 = tsc_khz * 1000LL;
 773         while (tps64 > nsecs*2) {
 774                 tps64 >>= 1;
 775                 shift--;
 776         }
 777
 778         tps32 = (uint32_t)tps64;
 779         while (tps32 <= (uint32_t)nsecs) {
 780                 tps32 <<= 1;
 781                 shift++;
 782         }
 783
 784         hv_clock->tsc_shift = shift;
 785         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 786
 787         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 788                  __func__, tsc_khz, hv_clock->tsc_shift,
 789                  hv_clock->tsc_to_system_mul);
 790 }
 791
 792 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 793
 794 static void kvm_write_guest_time(struct kvm_vcpu *v)
 795 {
 796         struct timespec ts;
 797         unsigned long flags;
 798         struct kvm_vcpu_arch *vcpu = &v->arch;
 799         void *shared_kaddr;
 800         unsigned long this_tsc_khz;
 801
 802         if ((!vcpu->time_page))
 803                 return;
 804
 805         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 806         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 807                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 808                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 809         }
 810         put_cpu_var(cpu_tsc_khz);
 811
 812         /* Keep irq disabled to prevent changes to the clock */
 813         local_irq_save(flags);
 814         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 815         ktime_get_ts(&ts);
 816         monotonic_to_bootbased(&ts);
 817         local_irq_restore(flags);
 818
 819         /* With all the info we got, fill in the values */
 820
 821         vcpu->hv_clock.system_time = ts.tv_nsec +
 822                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 823
 824         /*
 825          * The interface expects us to write an even number signaling that the
 826          * update is finished. Since the guest won't see the intermediate
 827          * state, we just increase by 2 at the end.
 828          */
 829         vcpu->hv_clock.version += 2;
 830
 831         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 832
 833         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 834                sizeof(vcpu->hv_clock));
 835
 836         kunmap_atomic(shared_kaddr, KM_USER0);
 837
 838         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 839 }
 840
 841 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 842 {
 843         struct kvm_vcpu_arch *vcpu = &v->arch;
 844
 845         if (!vcpu->time_page)
 846                 return 0;
 847         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 848         return 1;
 849 }
 850
 851 static bool msr_mtrr_valid(unsigned msr)
 852 {
 853         switch (msr) {
 854         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 855         case MSR_MTRRfix64K_00000:
 856         case MSR_MTRRfix16K_80000:
 857         case MSR_MTRRfix16K_A0000:
 858         case MSR_MTRRfix4K_C0000:
 859         case MSR_MTRRfix4K_C8000:
 860         case MSR_MTRRfix4K_D0000:
 861         case MSR_MTRRfix4K_D8000:
 862         case MSR_MTRRfix4K_E0000:
 863         case MSR_MTRRfix4K_E8000:
 864         case MSR_MTRRfix4K_F0000:
 865         case MSR_MTRRfix4K_F8000:
 866         case MSR_MTRRdefType:
 867         case MSR_IA32_CR_PAT:
 868                 return true;
 869         case 0x2f8:
 870                 return true;
 871         }
 872         return false;
 873 }
 874
 875 static bool valid_pat_type(unsigned t)
 876 {
 877         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 878 }
 879
 880 static bool valid_mtrr_type(unsigned t)
 881 {
 882         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 883 }
 884
 885 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 886 {
 887         int i;
 888
 889         if (!msr_mtrr_valid(msr))
 890                 return false;
 891
 892         if (msr == MSR_IA32_CR_PAT) {
 893                 for (i = 0; i < 8; i++)
 894                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 895                                 return false;
 896                 return true;
 897         } else if (msr == MSR_MTRRdefType) {
 898                 if (data & ~0xcff)
 899                         return false;
 900                 return valid_mtrr_type(data & 0xff);
 901         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 902                 for (i = 0; i < 8 ; i++)
 903                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 904                                 return false;
 905                 return true;
 906         }
 907
 908         /* variable MTRRs */
 909         return valid_mtrr_type(data & 0xff);
 910 }
 911
 912 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 913 {
 914         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 915
 916         if (!mtrr_valid(vcpu, msr, data))
 917                 return 1;
 918
 919         if (msr == MSR_MTRRdefType) {
 920                 vcpu->arch.mtrr_state.def_type = data;
 921                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 922         } else if (msr == MSR_MTRRfix64K_00000)
 923                 p[0] = data;
 924         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 925                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 926         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 927                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 928         else if (msr == MSR_IA32_CR_PAT)
 929                 vcpu->arch.pat = data;
 930         else {  /* Variable MTRRs */
 931                 int idx, is_mtrr_mask;
 932                 u64 *pt;
 933
 934                 idx = (msr - 0x200) / 2;
 935                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 936                 if (!is_mtrr_mask)
 937                         pt =
 938                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 939                 else
 940                         pt =
 941                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 942                 *pt = data;
 943         }
 944
 945         kvm_mmu_reset_context(vcpu);
 946         return 0;
 947 }
 948
 949 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 950 {
 951         u64 mcg_cap = vcpu->arch.mcg_cap;
 952         unsigned bank_num = mcg_cap & 0xff;
 953
 954         switch (msr) {
 955         case MSR_IA32_MCG_STATUS:
 956                 vcpu->arch.mcg_status = data;
 957                 break;
 958         case MSR_IA32_MCG_CTL:
 959                 if (!(mcg_cap & MCG_CTL_P))
 960                         return 1;
 961                 if (data != 0 && data != ~(u64)0)
 962                         return -1;
 963                 vcpu->arch.mcg_ctl = data;
 964                 break;
 965         default:
 966                 if (msr >= MSR_IA32_MC0_CTL &&
 967                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 968                         u32 offset = msr - MSR_IA32_MC0_CTL;
 969                         /* only 0 or all 1s can be written to IA32_MCi_CTL */
 970                         if ((offset & 0x3) == 0 &&
 971                             data != 0 && data != ~(u64)0)
 972                                 return -1;
 973                         vcpu->arch.mce_banks[offset] = data;
 974                         break;
 975                 }
 976                 return 1;
 977         }
 978         return 0;
 979 }
 980
 981 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 982 {
 983         struct kvm *kvm = vcpu->kvm;
 984         int lm = is_long_mode(vcpu);
 985         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 986                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 987         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 988                 : kvm->arch.xen_hvm_config.blob_size_32;
 989         u32 page_num = data & ~PAGE_MASK;
 990         u64 page_addr = data & PAGE_MASK;
 991         u8 *page;
 992         int r;
 993
 994         r = -E2BIG;
 995         if (page_num >= blob_size)
 996                 goto out;
 997         r = -ENOMEM;
 998         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 999         if (!page)
1000                 goto out;
1001         r = -EFAULT;
1002         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1003                 goto out_free;
1004         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1005                 goto out_free;
1006         r = 0;
1007 out_free:
1008         kfree(page);
1009 out:
1010         return r;
1011 }
1012
1013 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1014 {
1015         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1016 }
1017
1018 static bool kvm_hv_msr_partition_wide(u32 msr)
1019 {
1020         bool r = false;
1021         switch (msr) {
1022         case HV_X64_MSR_GUEST_OS_ID:
1023         case HV_X64_MSR_HYPERCALL:
1024                 r = true;
1025                 break;
1026         }
1027
1028         return r;
1029 }
1030
1031 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1032 {
1033         struct kvm *kvm = vcpu->kvm;
1034
1035         switch (msr) {
1036         case HV_X64_MSR_GUEST_OS_ID:
1037                 kvm->arch.hv_guest_os_id = data;
1038                 /* setting guest os id to zero disables hypercall page */
1039                 if (!kvm->arch.hv_guest_os_id)
1040                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1041                 break;
1042         case HV_X64_MSR_HYPERCALL: {
1043                 u64 gfn;
1044                 unsigned long addr;
1045                 u8 instructions[4];
1046
1047                 /* if guest os id is not set hypercall should remain disabled */
1048                 if (!kvm->arch.hv_guest_os_id)
1049                         break;
1050                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1051                         kvm->arch.hv_hypercall = data;
1052                         break;
1053                 }
1054                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1055                 addr = gfn_to_hva(kvm, gfn);
1056                 if (kvm_is_error_hva(addr))
1057                         return 1;
1058                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1059                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1060                 if (copy_to_user((void __user *)addr, instructions, 4))
1061                         return 1;
1062                 kvm->arch.hv_hypercall = data;
1063                 break;
1064         }
1065         default:
1066                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1067                           "data 0x%llx\n", msr, data);
1068                 return 1;
1069         }
1070         return 0;
1071 }
1072
1073 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1074 {
1075         switch (msr) {
1076         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1077                 unsigned long addr;
1078
1079                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1080                         vcpu->arch.hv_vapic = data;
1081                         break;
1082                 }
1083                 addr = gfn_to_hva(vcpu->kvm, data >>
1084                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1085                 if (kvm_is_error_hva(addr))
1086                         return 1;
1087                 if (clear_user((void __user *)addr, PAGE_SIZE))
1088                         return 1;
1089                 vcpu->arch.hv_vapic = data;
1090                 break;
1091         }
1092         case HV_X64_MSR_EOI:
1093                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1094         case HV_X64_MSR_ICR:
1095                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1096         case HV_X64_MSR_TPR:
1097                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1098         default:
1099                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1100                           "data 0x%llx\n", msr, data);
1101                 return 1;
1102         }
1103
1104         return 0;
1105 }
1106
1107 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1108 {
1109         switch (msr) {
1110         case MSR_EFER:
1111                 set_efer(vcpu, data);
1112                 break;
1113         case MSR_K7_HWCR:
1114                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1115                 if (data != 0) {
1116                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1117                                 data);
1118                         return 1;
1119                 }
1120                 break;
1121         case MSR_FAM10H_MMIO_CONF_BASE:
1122                 if (data != 0) {
1123                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1124                                 "0x%llx\n", data);
1125                         return 1;
1126                 }
1127                 break;
1128         case MSR_AMD64_NB_CFG:
1129                 break;
1130         case MSR_IA32_DEBUGCTLMSR:
1131                 if (!data) {
1132                         /* We support the non-activated case already */
1133                         break;
1134                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1135                         /* Values other than LBR and BTF are vendor-specific,
1136                            thus reserved and should throw a #GP */
1137                         return 1;
1138                 }
1139                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1140                         __func__, data);
1141                 break;
1142         case MSR_IA32_UCODE_REV:
1143         case MSR_IA32_UCODE_WRITE:
1144         case MSR_VM_HSAVE_PA:
1145         case MSR_AMD64_PATCH_LOADER:
1146                 break;
1147         case 0x200 ... 0x2ff:
1148                 return set_msr_mtrr(vcpu, msr, data);
1149         case MSR_IA32_APICBASE:
1150                 kvm_set_apic_base(vcpu, data);
1151                 break;
1152         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1153                 return kvm_x2apic_msr_write(vcpu, msr, data);
1154         case MSR_IA32_MISC_ENABLE:
1155                 vcpu->arch.ia32_misc_enable_msr = data;
1156                 break;
1157         case MSR_KVM_WALL_CLOCK:
1158                 vcpu->kvm->arch.wall_clock = data;
1159                 kvm_write_wall_clock(vcpu->kvm, data);
1160                 break;
1161         case MSR_KVM_SYSTEM_TIME: {
1162                 if (vcpu->arch.time_page) {
1163                         kvm_release_page_dirty(vcpu->arch.time_page);
1164                         vcpu->arch.time_page = NULL;
1165                 }
1166
1167                 vcpu->arch.time = data;
1168
1169                 /* we verify if the enable bit is set... */
1170                 if (!(data & 1))
1171                         break;
1172
1173                 /* ...but clean it before doing the actual write */
1174                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1175
1176                 vcpu->arch.time_page =
1177                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1178
1179                 if (is_error_page(vcpu->arch.time_page)) {
1180                         kvm_release_page_clean(vcpu->arch.time_page);
1181                         vcpu->arch.time_page = NULL;
1182                 }
1183
1184                 kvm_request_guest_time_update(vcpu);
1185                 break;
1186         }
1187         case MSR_IA32_MCG_CTL:
1188         case MSR_IA32_MCG_STATUS:
1189         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1190                 return set_msr_mce(vcpu, msr, data);
1191
1192         /* Performance counters are not protected by a CPUID bit,
1193          * so we should check all of them in the generic path for the sake of
1194          * cross vendor migration.
1195          * Writing a zero into the event select MSRs disables them,
1196          * which we perfectly emulate ;-). Any other value should be at least
1197          * reported, some guests depend on them.
1198          */
1199         case MSR_P6_EVNTSEL0:
1200         case MSR_P6_EVNTSEL1:
1201         case MSR_K7_EVNTSEL0:
1202         case MSR_K7_EVNTSEL1:
1203         case MSR_K7_EVNTSEL2:
1204         case MSR_K7_EVNTSEL3:
1205                 if (data != 0)
1206                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1207                                 "0x%x data 0x%llx\n", msr, data);
1208                 break;
1209         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1210          * so we ignore writes to make it happy.
1211          */
1212         case MSR_P6_PERFCTR0:
1213         case MSR_P6_PERFCTR1:
1214         case MSR_K7_PERFCTR0:
1215         case MSR_K7_PERFCTR1:
1216         case MSR_K7_PERFCTR2:
1217         case MSR_K7_PERFCTR3:
1218                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1219                         "0x%x data 0x%llx\n", msr, data);
1220                 break;
1221         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1222                 if (kvm_hv_msr_partition_wide(msr)) {
1223                         int r;
1224                         mutex_lock(&vcpu->kvm->lock);
1225                         r = set_msr_hyperv_pw(vcpu, msr, data);
1226                         mutex_unlock(&vcpu->kvm->lock);
1227                         return r;
1228                 } else
1229                         return set_msr_hyperv(vcpu, msr, data);
1230                 break;
1231         default:
1232                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1233                         return xen_hvm_config(vcpu, data);
1234                 if (!ignore_msrs) {
1235                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1236                                 msr, data);
1237                         return 1;
1238                 } else {
1239                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1240                                 msr, data);
1241                         break;
1242                 }
1243         }
1244         return 0;
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1247
1248
1249 /*
1250  * Reads an msr value (of 'msr_index') into 'pdata'.
1251  * Returns 0 on success, non-0 otherwise.
1252  * Assumes vcpu_load() was already called.
1253  */
1254 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1255 {
1256         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1257 }
1258
1259 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1260 {
1261         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1262
1263         if (!msr_mtrr_valid(msr))
1264                 return 1;
1265
1266         if (msr == MSR_MTRRdefType)
1267                 *pdata = vcpu->arch.mtrr_state.def_type +
1268                          (vcpu->arch.mtrr_state.enabled << 10);
1269         else if (msr == MSR_MTRRfix64K_00000)
1270                 *pdata = p[0];
1271         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1272                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1273         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1274                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1275         else if (msr == MSR_IA32_CR_PAT)
1276                 *pdata = vcpu->arch.pat;
1277         else {  /* Variable MTRRs */
1278                 int idx, is_mtrr_mask;
1279                 u64 *pt;
1280
1281                 idx = (msr - 0x200) / 2;
1282                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1283                 if (!is_mtrr_mask)
1284                         pt =
1285                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1286                 else
1287                         pt =
1288                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1289                 *pdata = *pt;
1290         }
1291
1292         return 0;
1293 }
1294
1295 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1296 {
1297         u64 data;
1298         u64 mcg_cap = vcpu->arch.mcg_cap;
1299         unsigned bank_num = mcg_cap & 0xff;
1300
1301         switch (msr) {
1302         case MSR_IA32_P5_MC_ADDR:
1303         case MSR_IA32_P5_MC_TYPE:
1304                 data = 0;
1305                 break;
1306         case MSR_IA32_MCG_CAP:
1307                 data = vcpu->arch.mcg_cap;
1308                 break;
1309         case MSR_IA32_MCG_CTL:
1310                 if (!(mcg_cap & MCG_CTL_P))
1311                         return 1;
1312                 data = vcpu->arch.mcg_ctl;
1313                 break;
1314         case MSR_IA32_MCG_STATUS:
1315                 data = vcpu->arch.mcg_status;
1316                 break;
1317         default:
1318                 if (msr >= MSR_IA32_MC0_CTL &&
1319                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1320                         u32 offset = msr - MSR_IA32_MC0_CTL;
1321                         data = vcpu->arch.mce_banks[offset];
1322                         break;
1323                 }
1324                 return 1;
1325         }
1326         *pdata = data;
1327         return 0;
1328 }
1329
1330 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1331 {
1332         u64 data = 0;
1333         struct kvm *kvm = vcpu->kvm;
1334
1335         switch (msr) {
1336         case HV_X64_MSR_GUEST_OS_ID:
1337                 data = kvm->arch.hv_guest_os_id;
1338                 break;
1339         case HV_X64_MSR_HYPERCALL:
1340                 data = kvm->arch.hv_hypercall;
1341                 break;
1342         default:
1343                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1344                 return 1;
1345         }
1346
1347         *pdata = data;
1348         return 0;
1349 }
1350
1351 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1352 {
1353         u64 data = 0;
1354
1355         switch (msr) {
1356         case HV_X64_MSR_VP_INDEX: {
1357                 int r;
1358                 struct kvm_vcpu *v;
1359                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1360                         if (v == vcpu)
1361                                 data = r;
1362                 break;
1363         }
1364         case HV_X64_MSR_EOI:
1365                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1366         case HV_X64_MSR_ICR:
1367                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1368         case HV_X64_MSR_TPR:
1369                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1370         default:
1371                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1372                 return 1;
1373         }
1374         *pdata = data;
1375         return 0;
1376 }
1377
1378 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1379 {
1380         u64 data;
1381
1382         switch (msr) {
1383         case MSR_IA32_PLATFORM_ID:
1384         case MSR_IA32_UCODE_REV:
1385         case MSR_IA32_EBL_CR_POWERON:
1386         case MSR_IA32_DEBUGCTLMSR:
1387         case MSR_IA32_LASTBRANCHFROMIP:
1388         case MSR_IA32_LASTBRANCHTOIP:
1389         case MSR_IA32_LASTINTFROMIP:
1390         case MSR_IA32_LASTINTTOIP:
1391         case MSR_K8_SYSCFG:
1392         case MSR_K7_HWCR:
1393         case MSR_VM_HSAVE_PA:
1394         case MSR_P6_PERFCTR0:
1395         case MSR_P6_PERFCTR1:
1396         case MSR_P6_EVNTSEL0:
1397         case MSR_P6_EVNTSEL1:
1398         case MSR_K7_EVNTSEL0:
1399         case MSR_K7_PERFCTR0:
1400         case MSR_K8_INT_PENDING_MSG:
1401         case MSR_AMD64_NB_CFG:
1402         case MSR_FAM10H_MMIO_CONF_BASE:
1403                 data = 0;
1404                 break;
1405         case MSR_MTRRcap:
1406                 data = 0x500 | KVM_NR_VAR_MTRR;
1407                 break;
1408         case 0x200 ... 0x2ff:
1409                 return get_msr_mtrr(vcpu, msr, pdata);
1410         case 0xcd: /* fsb frequency */
1411                 data = 3;
1412                 break;
1413         case MSR_IA32_APICBASE:
1414                 data = kvm_get_apic_base(vcpu);
1415                 break;
1416         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1417                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1418                 break;
1419         case MSR_IA32_MISC_ENABLE:
1420                 data = vcpu->arch.ia32_misc_enable_msr;
1421                 break;
1422         case MSR_IA32_PERF_STATUS:
1423                 /* TSC increment by tick */
1424                 data = 1000ULL;
1425                 /* CPU multiplier */
1426                 data |= (((uint64_t)4ULL) << 40);
1427                 break;
1428         case MSR_EFER:
1429                 data = vcpu->arch.efer;
1430                 break;
1431         case MSR_KVM_WALL_CLOCK:
1432                 data = vcpu->kvm->arch.wall_clock;
1433                 break;
1434         case MSR_KVM_SYSTEM_TIME:
1435                 data = vcpu->arch.time;
1436                 break;
1437         case MSR_IA32_P5_MC_ADDR:
1438         case MSR_IA32_P5_MC_TYPE:
1439         case MSR_IA32_MCG_CAP:
1440         case MSR_IA32_MCG_CTL:
1441         case MSR_IA32_MCG_STATUS:
1442         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1443                 return get_msr_mce(vcpu, msr, pdata);
1444         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1445                 if (kvm_hv_msr_partition_wide(msr)) {
1446                         int r;
1447                         mutex_lock(&vcpu->kvm->lock);
1448                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1449                         mutex_unlock(&vcpu->kvm->lock);
1450                         return r;
1451                 } else
1452                         return get_msr_hyperv(vcpu, msr, pdata);
1453                 break;
1454         default:
1455                 if (!ignore_msrs) {
1456                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1457                         return 1;
1458                 } else {
1459                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1460                         data = 0;
1461                 }
1462                 break;
1463         }
1464         *pdata = data;
1465         return 0;
1466 }
1467 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1468
1469 /*
1470  * Read or write a bunch of msrs. All parameters are kernel addresses.
1471  *
1472  * @return number of msrs set successfully.
1473  */
1474 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1475                     struct kvm_msr_entry *entries,
1476                     int (*do_msr)(struct kvm_vcpu *vcpu,
1477                                   unsigned index, u64 *data))
1478 {
1479         int i, idx;
1480
1481         vcpu_load(vcpu);
1482
1483         idx = srcu_read_lock(&vcpu->kvm->srcu);
1484         for (i = 0; i < msrs->nmsrs; ++i)
1485                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1486                         break;
1487         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1488
1489         vcpu_put(vcpu);
1490
1491         return i;
1492 }
1493
1494 /*
1495  * Read or write a bunch of msrs. Parameters are user addresses.
1496  *
1497  * @return number of msrs set successfully.
1498  */
1499 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1500                   int (*do_msr)(struct kvm_vcpu *vcpu,
1501                                 unsigned index, u64 *data),
1502                   int writeback)
1503 {
1504         struct kvm_msrs msrs;
1505         struct kvm_msr_entry *entries;
1506         int r, n;
1507         unsigned size;
1508
1509         r = -EFAULT;
1510         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1511                 goto out;
1512
1513         r = -E2BIG;
1514         if (msrs.nmsrs >= MAX_IO_MSRS)
1515                 goto out;
1516
1517         r = -ENOMEM;
1518         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1519         entries = vmalloc(size);
1520         if (!entries)
1521                 goto out;
1522
1523         r = -EFAULT;
1524         if (copy_from_user(entries, user_msrs->entries, size))
1525                 goto out_free;
1526
1527         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1528         if (r < 0)
1529                 goto out_free;
1530
1531         r = -EFAULT;
1532         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1533                 goto out_free;
1534
1535         r = n;
1536
1537 out_free:
1538         vfree(entries);
1539 out:
1540         return r;
1541 }
1542
1543 int kvm_dev_ioctl_check_extension(long ext)
1544 {
1545         int r;
1546
1547         switch (ext) {
1548         case KVM_CAP_IRQCHIP:
1549         case KVM_CAP_HLT:
1550         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1551         case KVM_CAP_SET_TSS_ADDR:
1552         case KVM_CAP_EXT_CPUID:
1553         case KVM_CAP_CLOCKSOURCE:
1554         case KVM_CAP_PIT:
1555         case KVM_CAP_NOP_IO_DELAY:
1556         case KVM_CAP_MP_STATE:
1557         case KVM_CAP_SYNC_MMU:
1558         case KVM_CAP_REINJECT_CONTROL:
1559         case KVM_CAP_IRQ_INJECT_STATUS:
1560         case KVM_CAP_ASSIGN_DEV_IRQ:
1561         case KVM_CAP_IRQFD:
1562         case KVM_CAP_IOEVENTFD:
1563         case KVM_CAP_PIT2:
1564         case KVM_CAP_PIT_STATE2:
1565         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1566         case KVM_CAP_XEN_HVM:
1567         case KVM_CAP_ADJUST_CLOCK:
1568         case KVM_CAP_VCPU_EVENTS:
1569         case KVM_CAP_HYPERV:
1570         case KVM_CAP_HYPERV_VAPIC:
1571         case KVM_CAP_HYPERV_SPIN:
1572         case KVM_CAP_PCI_SEGMENT:
1573                 r = 1;
1574                 break;
1575         case KVM_CAP_COALESCED_MMIO:
1576                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1577                 break;
1578         case KVM_CAP_VAPIC:
1579                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1580                 break;
1581         case KVM_CAP_NR_VCPUS:
1582                 r = KVM_MAX_VCPUS;
1583                 break;
1584         case KVM_CAP_NR_MEMSLOTS:
1585                 r = KVM_MEMORY_SLOTS;
1586                 break;
1587         case KVM_CAP_PV_MMU:    /* obsolete */
1588                 r = 0;
1589                 break;
1590         case KVM_CAP_IOMMU:
1591                 r = iommu_found();
1592                 break;
1593         case KVM_CAP_MCE:
1594                 r = KVM_MAX_MCE_BANKS;
1595                 break;
1596         default:
1597                 r = 0;
1598                 break;
1599         }
1600         return r;
1601
1602 }
1603
1604 long kvm_arch_dev_ioctl(struct file *filp,
1605                         unsigned int ioctl, unsigned long arg)
1606 {
1607         void __user *argp = (void __user *)arg;
1608         long r;
1609
1610         switch (ioctl) {
1611         case KVM_GET_MSR_INDEX_LIST: {
1612                 struct kvm_msr_list __user *user_msr_list = argp;
1613                 struct kvm_msr_list msr_list;
1614                 unsigned n;
1615
1616                 r = -EFAULT;
1617                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1618                         goto out;
1619                 n = msr_list.nmsrs;
1620                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1621                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1622                         goto out;
1623                 r = -E2BIG;
1624                 if (n < msr_list.nmsrs)
1625                         goto out;
1626                 r = -EFAULT;
1627                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1628                                  num_msrs_to_save * sizeof(u32)))
1629                         goto out;
1630                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1631                                  &emulated_msrs,
1632                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1633                         goto out;
1634                 r = 0;
1635                 break;
1636         }
1637         case KVM_GET_SUPPORTED_CPUID: {
1638                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1639                 struct kvm_cpuid2 cpuid;
1640
1641                 r = -EFAULT;
1642                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1643                         goto out;
1644                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1645                                                       cpuid_arg->entries);
1646                 if (r)
1647                         goto out;
1648
1649                 r = -EFAULT;
1650                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1651                         goto out;
1652                 r = 0;
1653                 break;
1654         }
1655         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1656                 u64 mce_cap;
1657
1658                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1659                 r = -EFAULT;
1660                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1661                         goto out;
1662                 r = 0;
1663                 break;
1664         }
1665         default:
1666                 r = -EINVAL;
1667         }
1668 out:
1669         return r;
1670 }
1671
1672 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1673 {
1674         kvm_x86_ops->vcpu_load(vcpu, cpu);
1675         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1676                 unsigned long khz = cpufreq_quick_get(cpu);
1677                 if (!khz)
1678                         khz = tsc_khz;
1679                 per_cpu(cpu_tsc_khz, cpu) = khz;
1680         }
1681         kvm_request_guest_time_update(vcpu);
1682 }
1683
1684 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1685 {
1686         kvm_put_guest_fpu(vcpu);
1687         kvm_x86_ops->vcpu_put(vcpu);
1688 }
1689
1690 static int is_efer_nx(void)
1691 {
1692         unsigned long long efer = 0;
1693
1694         rdmsrl_safe(MSR_EFER, &efer);
1695         return efer & EFER_NX;
1696 }
1697
1698 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1699 {
1700         int i;
1701         struct kvm_cpuid_entry2 *e, *entry;
1702
1703         entry = NULL;
1704         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1705                 e = &vcpu->arch.cpuid_entries[i];
1706                 if (e->function == 0x80000001) {
1707                         entry = e;
1708                         break;
1709                 }
1710         }
1711         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1712                 entry->edx &= ~(1 << 20);
1713                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1714         }
1715 }
1716
1717 /* when an old userspace process fills a new kernel module */
1718 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1719                                     struct kvm_cpuid *cpuid,
1720                                     struct kvm_cpuid_entry __user *entries)
1721 {
1722         int r, i;
1723         struct kvm_cpuid_entry *cpuid_entries;
1724
1725         r = -E2BIG;
1726         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1727                 goto out;
1728         r = -ENOMEM;
1729         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1730         if (!cpuid_entries)
1731                 goto out;
1732         r = -EFAULT;
1733         if (copy_from_user(cpuid_entries, entries,
1734                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1735                 goto out_free;
1736         for (i = 0; i < cpuid->nent; i++) {
1737                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1738                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1739                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1740                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1741                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1742                 vcpu->arch.cpuid_entries[i].index = 0;
1743                 vcpu->arch.cpuid_entries[i].flags = 0;
1744                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1745                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1746                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1747         }
1748         vcpu->arch.cpuid_nent = cpuid->nent;
1749         cpuid_fix_nx_cap(vcpu);
1750         r = 0;
1751         kvm_apic_set_version(vcpu);
1752         kvm_x86_ops->cpuid_update(vcpu);
1753
1754 out_free:
1755         vfree(cpuid_entries);
1756 out:
1757         return r;
1758 }
1759
1760 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1761                                      struct kvm_cpuid2 *cpuid,
1762                                      struct kvm_cpuid_entry2 __user *entries)
1763 {
1764         int r;
1765
1766         r = -E2BIG;
1767         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1768                 goto out;
1769         r = -EFAULT;
1770         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1771                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1772                 goto out;
1773         vcpu->arch.cpuid_nent = cpuid->nent;
1774         kvm_apic_set_version(vcpu);
1775         kvm_x86_ops->cpuid_update(vcpu);
1776         return 0;
1777
1778 out:
1779         return r;
1780 }
1781
1782 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1783                                      struct kvm_cpuid2 *cpuid,
1784                                      struct kvm_cpuid_entry2 __user *entries)
1785 {
1786         int r;
1787
1788         r = -E2BIG;
1789         if (cpuid->nent < vcpu->arch.cpuid_nent)
1790                 goto out;
1791         r = -EFAULT;
1792         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1793                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1794                 goto out;
1795         return 0;
1796
1797 out:
1798         cpuid->nent = vcpu->arch.cpuid_nent;
1799         return r;
1800 }
1801
1802 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1803                            u32 index)
1804 {
1805         entry->function = function;
1806         entry->index = index;
1807         cpuid_count(entry->function, entry->index,
1808                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1809         entry->flags = 0;
1810 }
1811
1812 #define F(x) bit(X86_FEATURE_##x)
1813
1814 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1815                          u32 index, int *nent, int maxnent)
1816 {
1817         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1818 #ifdef CONFIG_X86_64
1819         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1820                                 ? F(GBPAGES) : 0;
1821         unsigned f_lm = F(LM);
1822 #else
1823         unsigned f_gbpages = 0;
1824         unsigned f_lm = 0;
1825 #endif
1826         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1827
1828         /* cpuid 1.edx */
1829         const u32 kvm_supported_word0_x86_features =
1830                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1831                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1832                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1833                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1834                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1835                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1836                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1837                 0 /* HTT, TM, Reserved, PBE */;
1838         /* cpuid 0x80000001.edx */
1839         const u32 kvm_supported_word1_x86_features =
1840                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1841                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1842                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1843                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1844                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1845                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1846                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1847                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1848         /* cpuid 1.ecx */
1849         const u32 kvm_supported_word4_x86_features =
1850                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1851                 0 /* DS-CPL, VMX, SMX, EST */ |
1852                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1853                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1854                 0 /* Reserved, DCA */ | F(XMM4_1) |
1855                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1856                 0 /* Reserved, XSAVE, OSXSAVE */;
1857         /* cpuid 0x80000001.ecx */
1858         const u32 kvm_supported_word6_x86_features =
1859                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1860                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1861                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1862                 0 /* SKINIT */ | 0 /* WDT */;
1863
1864         /* all calls to cpuid_count() should be made on the same cpu */
1865         get_cpu();
1866         do_cpuid_1_ent(entry, function, index);
1867         ++*nent;
1868
1869         switch (function) {
1870         case 0:
1871                 entry->eax = min(entry->eax, (u32)0xb);
1872                 break;
1873         case 1:
1874                 entry->edx &= kvm_supported_word0_x86_features;
1875                 entry->ecx &= kvm_supported_word4_x86_features;
1876                 /* we support x2apic emulation even if host does not support
1877                  * it since we emulate x2apic in software */
1878                 entry->ecx |= F(X2APIC);
1879                 break;
1880         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1881          * may return different values. This forces us to get_cpu() before
1882          * issuing the first command, and also to emulate this annoying behavior
1883          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1884         case 2: {
1885                 int t, times = entry->eax & 0xff;
1886
1887                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1888                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1889                 for (t = 1; t < times && *nent < maxnent; ++t) {
1890                         do_cpuid_1_ent(&entry[t], function, 0);
1891                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1892                         ++*nent;
1893                 }
1894                 break;
1895         }
1896         /* function 4 and 0xb have additional index. */
1897         case 4: {
1898                 int i, cache_type;
1899
1900                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1901                 /* read more entries until cache_type is zero */
1902                 for (i = 1; *nent < maxnent; ++i) {
1903                         cache_type = entry[i - 1].eax & 0x1f;
1904                         if (!cache_type)
1905                                 break;
1906                         do_cpuid_1_ent(&entry[i], function, i);
1907                         entry[i].flags |=
1908                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1909                         ++*nent;
1910                 }
1911                 break;
1912         }
1913         case 0xb: {
1914                 int i, level_type;
1915
1916                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1917                 /* read more entries until level_type is zero */
1918                 for (i = 1; *nent < maxnent; ++i) {
1919                         level_type = entry[i - 1].ecx & 0xff00;
1920                         if (!level_type)
1921                                 break;
1922                         do_cpuid_1_ent(&entry[i], function, i);
1923                         entry[i].flags |=
1924                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1925                         ++*nent;
1926                 }
1927                 break;
1928         }
1929         case 0x80000000:
1930                 entry->eax = min(entry->eax, 0x8000001a);
1931                 break;
1932         case 0x80000001:
1933                 entry->edx &= kvm_supported_word1_x86_features;
1934                 entry->ecx &= kvm_supported_word6_x86_features;
1935                 break;
1936         }
1937         put_cpu();
1938 }
1939
1940 #undef F
1941
1942 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1943                                      struct kvm_cpuid_entry2 __user *entries)
1944 {
1945         struct kvm_cpuid_entry2 *cpuid_entries;
1946         int limit, nent = 0, r = -E2BIG;
1947         u32 func;
1948
1949         if (cpuid->nent < 1)
1950                 goto out;
1951         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1952                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1953         r = -ENOMEM;
1954         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1955         if (!cpuid_entries)
1956                 goto out;
1957
1958         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1959         limit = cpuid_entries[0].eax;
1960         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1961                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1962                              &nent, cpuid->nent);
1963         r = -E2BIG;
1964         if (nent >= cpuid->nent)
1965                 goto out_free;
1966
1967         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1968         limit = cpuid_entries[nent - 1].eax;
1969         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1970                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1971                              &nent, cpuid->nent);
1972         r = -E2BIG;
1973         if (nent >= cpuid->nent)
1974                 goto out_free;
1975
1976         r = -EFAULT;
1977         if (copy_to_user(entries, cpuid_entries,
1978                          nent * sizeof(struct kvm_cpuid_entry2)))
1979                 goto out_free;
1980         cpuid->nent = nent;
1981         r = 0;
1982
1983 out_free:
1984         vfree(cpuid_entries);
1985 out:
1986         return r;
1987 }
1988
1989 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1990                                     struct kvm_lapic_state *s)
1991 {
1992         vcpu_load(vcpu);
1993         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1994         vcpu_put(vcpu);
1995
1996         return 0;
1997 }
1998
1999 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2000                                     struct kvm_lapic_state *s)
2001 {
2002         vcpu_load(vcpu);
2003         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2004         kvm_apic_post_state_restore(vcpu);
2005         update_cr8_intercept(vcpu);
2006         vcpu_put(vcpu);
2007
2008         return 0;
2009 }
2010
2011 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2012                                     struct kvm_interrupt *irq)
2013 {
2014         if (irq->irq < 0 || irq->irq >= 256)
2015                 return -EINVAL;
2016         if (irqchip_in_kernel(vcpu->kvm))
2017                 return -ENXIO;
2018         vcpu_load(vcpu);
2019
2020         kvm_queue_interrupt(vcpu, irq->irq, false);
2021
2022         vcpu_put(vcpu);
2023
2024         return 0;
2025 }
2026
2027 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2028 {
2029         vcpu_load(vcpu);
2030         kvm_inject_nmi(vcpu);
2031         vcpu_put(vcpu);
2032
2033         return 0;
2034 }
2035
2036 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2037                                            struct kvm_tpr_access_ctl *tac)
2038 {
2039         if (tac->flags)
2040                 return -EINVAL;
2041         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2042         return 0;
2043 }
2044
2045 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2046                                         u64 mcg_cap)
2047 {
2048         int r;
2049         unsigned bank_num = mcg_cap & 0xff, bank;
2050
2051         r = -EINVAL;
2052         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2053                 goto out;
2054         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2055                 goto out;
2056         r = 0;
2057         vcpu->arch.mcg_cap = mcg_cap;
2058         /* Init IA32_MCG_CTL to all 1s */
2059         if (mcg_cap & MCG_CTL_P)
2060                 vcpu->arch.mcg_ctl = ~(u64)0;
2061         /* Init IA32_MCi_CTL to all 1s */
2062         for (bank = 0; bank < bank_num; bank++)
2063                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2064 out:
2065         return r;
2066 }
2067
2068 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2069                                       struct kvm_x86_mce *mce)
2070 {
2071         u64 mcg_cap = vcpu->arch.mcg_cap;
2072         unsigned bank_num = mcg_cap & 0xff;
2073         u64 *banks = vcpu->arch.mce_banks;
2074
2075         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2076                 return -EINVAL;
2077         /*
2078          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2079          * reporting is disabled
2080          */
2081         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2082             vcpu->arch.mcg_ctl != ~(u64)0)
2083                 return 0;
2084         banks += 4 * mce->bank;
2085         /*
2086          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2087          * reporting is disabled for the bank
2088          */
2089         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2090                 return 0;
2091         if (mce->status & MCI_STATUS_UC) {
2092                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2093                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2094                         printk(KERN_DEBUG "kvm: set_mce: "
2095                                "injects mce exception while "
2096                                "previous one is in progress!\n");
2097                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2098                         return 0;
2099                 }
2100                 if (banks[1] & MCI_STATUS_VAL)
2101                         mce->status |= MCI_STATUS_OVER;
2102                 banks[2] = mce->addr;
2103                 banks[3] = mce->misc;
2104                 vcpu->arch.mcg_status = mce->mcg_status;
2105                 banks[1] = mce->status;
2106                 kvm_queue_exception(vcpu, MC_VECTOR);
2107         } else if (!(banks[1] & MCI_STATUS_VAL)
2108                    || !(banks[1] & MCI_STATUS_UC)) {
2109                 if (banks[1] & MCI_STATUS_VAL)
2110                         mce->status |= MCI_STATUS_OVER;
2111                 banks[2] = mce->addr;
2112                 banks[3] = mce->misc;
2113                 banks[1] = mce->status;
2114         } else
2115                 banks[1] |= MCI_STATUS_OVER;
2116         return 0;
2117 }
2118
2119 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2120                                                struct kvm_vcpu_events *events)
2121 {
2122         vcpu_load(vcpu);
2123
2124         events->exception.injected = vcpu->arch.exception.pending;
2125         events->exception.nr = vcpu->arch.exception.nr;
2126         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2127         events->exception.error_code = vcpu->arch.exception.error_code;
2128
2129         events->interrupt.injected = vcpu->arch.interrupt.pending;
2130         events->interrupt.nr = vcpu->arch.interrupt.nr;
2131         events->interrupt.soft = vcpu->arch.interrupt.soft;
2132
2133         events->nmi.injected = vcpu->arch.nmi_injected;
2134         events->nmi.pending = vcpu->arch.nmi_pending;
2135         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2136
2137         events->sipi_vector = vcpu->arch.sipi_vector;
2138
2139         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2140                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
2141
2142         vcpu_put(vcpu);
2143 }
2144
2145 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2146                                               struct kvm_vcpu_events *events)
2147 {
2148         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2149                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
2150                 return -EINVAL;
2151
2152         vcpu_load(vcpu);
2153
2154         vcpu->arch.exception.pending = events->exception.injected;
2155         vcpu->arch.exception.nr = events->exception.nr;
2156         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2157         vcpu->arch.exception.error_code = events->exception.error_code;
2158
2159         vcpu->arch.interrupt.pending = events->interrupt.injected;
2160         vcpu->arch.interrupt.nr = events->interrupt.nr;
2161         vcpu->arch.interrupt.soft = events->interrupt.soft;
2162         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2163                 kvm_pic_clear_isr_ack(vcpu->kvm);
2164
2165         vcpu->arch.nmi_injected = events->nmi.injected;
2166         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2167                 vcpu->arch.nmi_pending = events->nmi.pending;
2168         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2169
2170         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2171                 vcpu->arch.sipi_vector = events->sipi_vector;
2172
2173         vcpu_put(vcpu);
2174
2175         return 0;
2176 }
2177
2178 long kvm_arch_vcpu_ioctl(struct file *filp,
2179                          unsigned int ioctl, unsigned long arg)
2180 {
2181         struct kvm_vcpu *vcpu = filp->private_data;
2182         void __user *argp = (void __user *)arg;
2183         int r;
2184         struct kvm_lapic_state *lapic = NULL;
2185
2186         switch (ioctl) {
2187         case KVM_GET_LAPIC: {
2188                 r = -EINVAL;
2189                 if (!vcpu->arch.apic)
2190                         goto out;
2191                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2192
2193                 r = -ENOMEM;
2194                 if (!lapic)
2195                         goto out;
2196                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2197                 if (r)
2198                         goto out;
2199                 r = -EFAULT;
2200                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2201                         goto out;
2202                 r = 0;
2203                 break;
2204         }
2205         case KVM_SET_LAPIC: {
2206                 r = -EINVAL;
2207                 if (!vcpu->arch.apic)
2208                         goto out;
2209                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2210                 r = -ENOMEM;
2211                 if (!lapic)
2212                         goto out;
2213                 r = -EFAULT;
2214                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2215                         goto out;
2216                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2217                 if (r)
2218                         goto out;
2219                 r = 0;
2220                 break;
2221         }
2222         case KVM_INTERRUPT: {
2223                 struct kvm_interrupt irq;
2224
2225                 r = -EFAULT;
2226                 if (copy_from_user(&irq, argp, sizeof irq))
2227                         goto out;
2228                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2229                 if (r)
2230                         goto out;
2231                 r = 0;
2232                 break;
2233         }
2234         case KVM_NMI: {
2235                 r = kvm_vcpu_ioctl_nmi(vcpu);
2236                 if (r)
2237                         goto out;
2238                 r = 0;
2239                 break;
2240         }
2241         case KVM_SET_CPUID: {
2242                 struct kvm_cpuid __user *cpuid_arg = argp;
2243                 struct kvm_cpuid cpuid;
2244
2245                 r = -EFAULT;
2246                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2247                         goto out;
2248                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2249                 if (r)
2250                         goto out;
2251                 break;
2252         }
2253         case KVM_SET_CPUID2: {
2254                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2255                 struct kvm_cpuid2 cpuid;
2256
2257                 r = -EFAULT;
2258                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2259                         goto out;
2260                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2261                                               cpuid_arg->entries);
2262                 if (r)
2263                         goto out;
2264                 break;
2265         }
2266         case KVM_GET_CPUID2: {
2267                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2268                 struct kvm_cpuid2 cpuid;
2269
2270                 r = -EFAULT;
2271                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2272                         goto out;
2273                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2274                                               cpuid_arg->entries);
2275                 if (r)
2276                         goto out;
2277                 r = -EFAULT;
2278                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2279                         goto out;
2280                 r = 0;
2281                 break;
2282         }
2283         case KVM_GET_MSRS:
2284                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2285                 break;
2286         case KVM_SET_MSRS:
2287                 r = msr_io(vcpu, argp, do_set_msr, 0);
2288                 break;
2289         case KVM_TPR_ACCESS_REPORTING: {
2290                 struct kvm_tpr_access_ctl tac;
2291
2292                 r = -EFAULT;
2293                 if (copy_from_user(&tac, argp, sizeof tac))
2294                         goto out;
2295                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2296                 if (r)
2297                         goto out;
2298                 r = -EFAULT;
2299                 if (copy_to_user(argp, &tac, sizeof tac))
2300                         goto out;
2301                 r = 0;
2302                 break;
2303         };
2304         case KVM_SET_VAPIC_ADDR: {
2305                 struct kvm_vapic_addr va;
2306
2307                 r = -EINVAL;
2308                 if (!irqchip_in_kernel(vcpu->kvm))
2309                         goto out;
2310                 r = -EFAULT;
2311                 if (copy_from_user(&va, argp, sizeof va))
2312                         goto out;
2313                 r = 0;
2314                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2315                 break;
2316         }
2317         case KVM_X86_SETUP_MCE: {
2318                 u64 mcg_cap;
2319
2320                 r = -EFAULT;
2321                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2322                         goto out;
2323                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2324                 break;
2325         }
2326         case KVM_X86_SET_MCE: {
2327                 struct kvm_x86_mce mce;
2328
2329                 r = -EFAULT;
2330                 if (copy_from_user(&mce, argp, sizeof mce))
2331                         goto out;
2332                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2333                 break;
2334         }
2335         case KVM_GET_VCPU_EVENTS: {
2336                 struct kvm_vcpu_events events;
2337
2338                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2339
2340                 r = -EFAULT;
2341                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2342                         break;
2343                 r = 0;
2344                 break;
2345         }
2346         case KVM_SET_VCPU_EVENTS: {
2347                 struct kvm_vcpu_events events;
2348
2349                 r = -EFAULT;
2350                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2351                         break;
2352
2353                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2354                 break;
2355         }
2356         default:
2357                 r = -EINVAL;
2358         }
2359 out:
2360         kfree(lapic);
2361         return r;
2362 }
2363
2364 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2365 {
2366         int ret;
2367
2368         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2369                 return -1;
2370         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2371         return ret;
2372 }
2373
2374 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2375                                               u64 ident_addr)
2376 {
2377         kvm->arch.ept_identity_map_addr = ident_addr;
2378         return 0;
2379 }
2380
2381 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2382                                           u32 kvm_nr_mmu_pages)
2383 {
2384         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2385                 return -EINVAL;
2386
2387         mutex_lock(&kvm->slots_lock);
2388         spin_lock(&kvm->mmu_lock);
2389
2390         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2391         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2392
2393         spin_unlock(&kvm->mmu_lock);
2394         mutex_unlock(&kvm->slots_lock);
2395         return 0;
2396 }
2397
2398 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2399 {
2400         return kvm->arch.n_alloc_mmu_pages;
2401 }
2402
2403 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2404 {
2405         int i;
2406         struct kvm_mem_alias *alias;
2407         struct kvm_mem_aliases *aliases;
2408
2409         aliases = rcu_dereference(kvm->arch.aliases);
2410
2411         for (i = 0; i < aliases->naliases; ++i) {
2412                 alias = &aliases->aliases[i];
2413                 if (alias->flags & KVM_ALIAS_INVALID)
2414                         continue;
2415                 if (gfn >= alias->base_gfn
2416                     && gfn < alias->base_gfn + alias->npages)
2417                         return alias->target_gfn + gfn - alias->base_gfn;
2418         }
2419         return gfn;
2420 }
2421
2422 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2423 {
2424         int i;
2425         struct kvm_mem_alias *alias;
2426         struct kvm_mem_aliases *aliases;
2427
2428         aliases = rcu_dereference(kvm->arch.aliases);
2429
2430         for (i = 0; i < aliases->naliases; ++i) {
2431                 alias = &aliases->aliases[i];
2432                 if (gfn >= alias->base_gfn
2433                     && gfn < alias->base_gfn + alias->npages)
2434                         return alias->target_gfn + gfn - alias->base_gfn;
2435         }
2436         return gfn;
2437 }
2438
2439 /*
2440  * Set a new alias region.  Aliases map a portion of physical memory into
2441  * another portion.  This is useful for memory windows, for example the PC
2442  * VGA region.
2443  */
2444 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2445                                          struct kvm_memory_alias *alias)
2446 {
2447         int r, n;
2448         struct kvm_mem_alias *p;
2449         struct kvm_mem_aliases *aliases, *old_aliases;
2450
2451         r = -EINVAL;
2452         /* General sanity checks */
2453         if (alias->memory_size & (PAGE_SIZE - 1))
2454                 goto out;
2455         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2456                 goto out;
2457         if (alias->slot >= KVM_ALIAS_SLOTS)
2458                 goto out;
2459         if (alias->guest_phys_addr + alias->memory_size
2460             < alias->guest_phys_addr)
2461                 goto out;
2462         if (alias->target_phys_addr + alias->memory_size
2463             < alias->target_phys_addr)
2464                 goto out;
2465
2466         r = -ENOMEM;
2467         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2468         if (!aliases)
2469                 goto out;
2470
2471         mutex_lock(&kvm->slots_lock);
2472
2473         /* invalidate any gfn reference in case of deletion/shrinking */
2474         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2475         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2476         old_aliases = kvm->arch.aliases;
2477         rcu_assign_pointer(kvm->arch.aliases, aliases);
2478         synchronize_srcu_expedited(&kvm->srcu);
2479         kvm_mmu_zap_all(kvm);
2480         kfree(old_aliases);
2481
2482         r = -ENOMEM;
2483         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2484         if (!aliases)
2485                 goto out_unlock;
2486
2487         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2488
2489         p = &aliases->aliases[alias->slot];
2490         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2491         p->npages = alias->memory_size >> PAGE_SHIFT;
2492         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2493         p->flags &= ~(KVM_ALIAS_INVALID);
2494
2495         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2496                 if (aliases->aliases[n - 1].npages)
2497                         break;
2498         aliases->naliases = n;
2499
2500         old_aliases = kvm->arch.aliases;
2501         rcu_assign_pointer(kvm->arch.aliases, aliases);
2502         synchronize_srcu_expedited(&kvm->srcu);
2503         kfree(old_aliases);
2504         r = 0;
2505
2506 out_unlock:
2507         mutex_unlock(&kvm->slots_lock);
2508 out:
2509         return r;
2510 }
2511
2512 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2513 {
2514         int r;
2515
2516         r = 0;
2517         switch (chip->chip_id) {
2518         case KVM_IRQCHIP_PIC_MASTER:
2519                 memcpy(&chip->chip.pic,
2520                         &pic_irqchip(kvm)->pics[0],
2521                         sizeof(struct kvm_pic_state));
2522                 break;
2523         case KVM_IRQCHIP_PIC_SLAVE:
2524                 memcpy(&chip->chip.pic,
2525                         &pic_irqchip(kvm)->pics[1],
2526                         sizeof(struct kvm_pic_state));
2527                 break;
2528         case KVM_IRQCHIP_IOAPIC:
2529                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2530                 break;
2531         default:
2532                 r = -EINVAL;
2533                 break;
2534         }
2535         return r;
2536 }
2537
2538 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2539 {
2540         int r;
2541
2542         r = 0;
2543         switch (chip->chip_id) {
2544         case KVM_IRQCHIP_PIC_MASTER:
2545                 spin_lock(&pic_irqchip(kvm)->lock);
2546                 memcpy(&pic_irqchip(kvm)->pics[0],
2547                         &chip->chip.pic,
2548                         sizeof(struct kvm_pic_state));
2549                 spin_unlock(&pic_irqchip(kvm)->lock);
2550                 break;
2551         case KVM_IRQCHIP_PIC_SLAVE:
2552                 spin_lock(&pic_irqchip(kvm)->lock);
2553                 memcpy(&pic_irqchip(kvm)->pics[1],
2554                         &chip->chip.pic,
2555                         sizeof(struct kvm_pic_state));
2556                 spin_unlock(&pic_irqchip(kvm)->lock);
2557                 break;
2558         case KVM_IRQCHIP_IOAPIC:
2559                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2560                 break;
2561         default:
2562                 r = -EINVAL;
2563                 break;
2564         }
2565         kvm_pic_update_irq(pic_irqchip(kvm));
2566         return r;
2567 }
2568
2569 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2570 {
2571         int r = 0;
2572
2573         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2574         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2575         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2576         return r;
2577 }
2578
2579 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2580 {
2581         int r = 0;
2582
2583         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2584         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2585         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2586         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2587         return r;
2588 }
2589
2590 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2591 {
2592         int r = 0;
2593
2594         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2595         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2596                 sizeof(ps->channels));
2597         ps->flags = kvm->arch.vpit->pit_state.flags;
2598         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2599         return r;
2600 }
2601
2602 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2603 {
2604         int r = 0, start = 0;
2605         u32 prev_legacy, cur_legacy;
2606         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2607         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2608         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2609         if (!prev_legacy && cur_legacy)
2610                 start = 1;
2611         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2612                sizeof(kvm->arch.vpit->pit_state.channels));
2613         kvm->arch.vpit->pit_state.flags = ps->flags;
2614         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2615         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2616         return r;
2617 }
2618
2619 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2620                                  struct kvm_reinject_control *control)
2621 {
2622         if (!kvm->arch.vpit)
2623                 return -ENXIO;
2624         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2625         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2626         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2627         return 0;
2628 }
2629
2630 /*
2631  * Get (and clear) the dirty memory log for a memory slot.
2632  */
2633 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2634                                       struct kvm_dirty_log *log)
2635 {
2636         int r, n, i;
2637         struct kvm_memory_slot *memslot;
2638         unsigned long is_dirty = 0;
2639         unsigned long *dirty_bitmap = NULL;
2640
2641         mutex_lock(&kvm->slots_lock);
2642
2643         r = -EINVAL;
2644         if (log->slot >= KVM_MEMORY_SLOTS)
2645                 goto out;
2646
2647         memslot = &kvm->memslots->memslots[log->slot];
2648         r = -ENOENT;
2649         if (!memslot->dirty_bitmap)
2650                 goto out;
2651
2652         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2653
2654         r = -ENOMEM;
2655         dirty_bitmap = vmalloc(n);
2656         if (!dirty_bitmap)
2657                 goto out;
2658         memset(dirty_bitmap, 0, n);
2659
2660         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2661                 is_dirty = memslot->dirty_bitmap[i];
2662
2663         /* If nothing is dirty, don't bother messing with page tables. */
2664         if (is_dirty) {
2665                 struct kvm_memslots *slots, *old_slots;
2666
2667                 spin_lock(&kvm->mmu_lock);
2668                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2669                 spin_unlock(&kvm->mmu_lock);
2670
2671                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2672                 if (!slots)
2673                         goto out_free;
2674
2675                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2676                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2677
2678                 old_slots = kvm->memslots;
2679                 rcu_assign_pointer(kvm->memslots, slots);
2680                 synchronize_srcu_expedited(&kvm->srcu);
2681                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2682                 kfree(old_slots);
2683         }
2684
2685         r = 0;
2686         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2687                 r = -EFAULT;
2688 out_free:
2689         vfree(dirty_bitmap);
2690 out:
2691         mutex_unlock(&kvm->slots_lock);
2692         return r;
2693 }
2694
2695 long kvm_arch_vm_ioctl(struct file *filp,
2696                        unsigned int ioctl, unsigned long arg)
2697 {
2698         struct kvm *kvm = filp->private_data;
2699         void __user *argp = (void __user *)arg;
2700         int r = -ENOTTY;
2701         /*
2702          * This union makes it completely explicit to gcc-3.x
2703          * that these two variables' stack usage should be
2704          * combined, not added together.
2705          */
2706         union {
2707                 struct kvm_pit_state ps;
2708                 struct kvm_pit_state2 ps2;
2709                 struct kvm_memory_alias alias;
2710                 struct kvm_pit_config pit_config;
2711         } u;
2712
2713         switch (ioctl) {
2714         case KVM_SET_TSS_ADDR:
2715                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2716                 if (r < 0)
2717                         goto out;
2718                 break;
2719         case KVM_SET_IDENTITY_MAP_ADDR: {
2720                 u64 ident_addr;
2721
2722                 r = -EFAULT;
2723                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2724                         goto out;
2725                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2726                 if (r < 0)
2727                         goto out;
2728                 break;
2729         }
2730         case KVM_SET_MEMORY_REGION: {
2731                 struct kvm_memory_region kvm_mem;
2732                 struct kvm_userspace_memory_region kvm_userspace_mem;
2733
2734                 r = -EFAULT;
2735                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2736                         goto out;
2737                 kvm_userspace_mem.slot = kvm_mem.slot;
2738                 kvm_userspace_mem.flags = kvm_mem.flags;
2739                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2740                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2741                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2742                 if (r)
2743                         goto out;
2744                 break;
2745         }
2746         case KVM_SET_NR_MMU_PAGES:
2747                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2748                 if (r)
2749                         goto out;
2750                 break;
2751         case KVM_GET_NR_MMU_PAGES:
2752                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2753                 break;
2754         case KVM_SET_MEMORY_ALIAS:
2755                 r = -EFAULT;
2756                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2757                         goto out;
2758                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2759                 if (r)
2760                         goto out;
2761                 break;
2762         case KVM_CREATE_IRQCHIP: {
2763                 struct kvm_pic *vpic;
2764
2765                 mutex_lock(&kvm->lock);
2766                 r = -EEXIST;
2767                 if (kvm->arch.vpic)
2768                         goto create_irqchip_unlock;
2769                 r = -ENOMEM;
2770                 vpic = kvm_create_pic(kvm);
2771                 if (vpic) {
2772                         r = kvm_ioapic_init(kvm);
2773                         if (r) {
2774                                 kfree(vpic);
2775                                 goto create_irqchip_unlock;
2776                         }
2777                 } else
2778                         goto create_irqchip_unlock;
2779                 smp_wmb();
2780                 kvm->arch.vpic = vpic;
2781                 smp_wmb();
2782                 r = kvm_setup_default_irq_routing(kvm);
2783                 if (r) {
2784                         mutex_lock(&kvm->irq_lock);
2785                         kfree(kvm->arch.vpic);
2786                         kfree(kvm->arch.vioapic);
2787                         kvm->arch.vpic = NULL;
2788                         kvm->arch.vioapic = NULL;
2789                         mutex_unlock(&kvm->irq_lock);
2790                 }
2791         create_irqchip_unlock:
2792                 mutex_unlock(&kvm->lock);
2793                 break;
2794         }
2795         case KVM_CREATE_PIT:
2796                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2797                 goto create_pit;
2798         case KVM_CREATE_PIT2:
2799                 r = -EFAULT;
2800                 if (copy_from_user(&u.pit_config, argp,
2801                                    sizeof(struct kvm_pit_config)))
2802                         goto out;
2803         create_pit:
2804                 mutex_lock(&kvm->slots_lock);
2805                 r = -EEXIST;
2806                 if (kvm->arch.vpit)
2807                         goto create_pit_unlock;
2808                 r = -ENOMEM;
2809                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2810                 if (kvm->arch.vpit)
2811                         r = 0;
2812         create_pit_unlock:
2813                 mutex_unlock(&kvm->slots_lock);
2814                 break;
2815         case KVM_IRQ_LINE_STATUS:
2816         case KVM_IRQ_LINE: {
2817                 struct kvm_irq_level irq_event;
2818
2819                 r = -EFAULT;
2820                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2821                         goto out;
2822                 if (irqchip_in_kernel(kvm)) {
2823                         __s32 status;
2824                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2825                                         irq_event.irq, irq_event.level);
2826                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2827                                 irq_event.status = status;
2828                                 if (copy_to_user(argp, &irq_event,
2829                                                         sizeof irq_event))
2830                                         goto out;
2831                         }
2832                         r = 0;
2833                 }
2834                 break;
2835         }
2836         case KVM_GET_IRQCHIP: {
2837                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2838                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2839
2840                 r = -ENOMEM;
2841                 if (!chip)
2842                         goto out;
2843                 r = -EFAULT;
2844                 if (copy_from_user(chip, argp, sizeof *chip))
2845                         goto get_irqchip_out;
2846                 r = -ENXIO;
2847                 if (!irqchip_in_kernel(kvm))
2848                         goto get_irqchip_out;
2849                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2850                 if (r)
2851                         goto get_irqchip_out;
2852                 r = -EFAULT;
2853                 if (copy_to_user(argp, chip, sizeof *chip))
2854                         goto get_irqchip_out;
2855                 r = 0;
2856         get_irqchip_out:
2857                 kfree(chip);
2858                 if (r)
2859                         goto out;
2860                 break;
2861         }
2862         case KVM_SET_IRQCHIP: {
2863                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2864                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2865
2866                 r = -ENOMEM;
2867                 if (!chip)
2868                         goto out;
2869                 r = -EFAULT;
2870                 if (copy_from_user(chip, argp, sizeof *chip))
2871                         goto set_irqchip_out;
2872                 r = -ENXIO;
2873                 if (!irqchip_in_kernel(kvm))
2874                         goto set_irqchip_out;
2875                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2876                 if (r)
2877                         goto set_irqchip_out;
2878                 r = 0;
2879         set_irqchip_out:
2880                 kfree(chip);
2881                 if (r)
2882                         goto out;
2883                 break;
2884         }
2885         case KVM_GET_PIT: {
2886                 r = -EFAULT;
2887                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2888                         goto out;
2889                 r = -ENXIO;
2890                 if (!kvm->arch.vpit)
2891                         goto out;
2892                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2893                 if (r)
2894                         goto out;
2895                 r = -EFAULT;
2896                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2897                         goto out;
2898                 r = 0;
2899                 break;
2900         }
2901         case KVM_SET_PIT: {
2902                 r = -EFAULT;
2903                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2904                         goto out;
2905                 r = -ENXIO;
2906                 if (!kvm->arch.vpit)
2907                         goto out;
2908                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2909                 if (r)
2910                         goto out;
2911                 r = 0;
2912                 break;
2913         }
2914         case KVM_GET_PIT2: {
2915                 r = -ENXIO;
2916                 if (!kvm->arch.vpit)
2917                         goto out;
2918                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2919                 if (r)
2920                         goto out;
2921                 r = -EFAULT;
2922                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2923                         goto out;
2924                 r = 0;
2925                 break;
2926         }
2927         case KVM_SET_PIT2: {
2928                 r = -EFAULT;
2929                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2930                         goto out;
2931                 r = -ENXIO;
2932                 if (!kvm->arch.vpit)
2933                         goto out;
2934                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2935                 if (r)
2936                         goto out;
2937                 r = 0;
2938                 break;
2939         }
2940         case KVM_REINJECT_CONTROL: {
2941                 struct kvm_reinject_control control;
2942                 r =  -EFAULT;
2943                 if (copy_from_user(&control, argp, sizeof(control)))
2944                         goto out;
2945                 r = kvm_vm_ioctl_reinject(kvm, &control);
2946                 if (r)
2947                         goto out;
2948                 r = 0;
2949                 break;
2950         }
2951         case KVM_XEN_HVM_CONFIG: {
2952                 r = -EFAULT;
2953                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2954                                    sizeof(struct kvm_xen_hvm_config)))
2955                         goto out;
2956                 r = -EINVAL;
2957                 if (kvm->arch.xen_hvm_config.flags)
2958                         goto out;
2959                 r = 0;
2960                 break;
2961         }
2962         case KVM_SET_CLOCK: {
2963                 struct timespec now;
2964                 struct kvm_clock_data user_ns;
2965                 u64 now_ns;
2966                 s64 delta;
2967
2968                 r = -EFAULT;
2969                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2970                         goto out;
2971
2972                 r = -EINVAL;
2973                 if (user_ns.flags)
2974                         goto out;
2975
2976                 r = 0;
2977                 ktime_get_ts(&now);
2978                 now_ns = timespec_to_ns(&now);
2979                 delta = user_ns.clock - now_ns;
2980                 kvm->arch.kvmclock_offset = delta;
2981                 break;
2982         }
2983         case KVM_GET_CLOCK: {
2984                 struct timespec now;
2985                 struct kvm_clock_data user_ns;
2986                 u64 now_ns;
2987
2988                 ktime_get_ts(&now);
2989                 now_ns = timespec_to_ns(&now);
2990                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2991                 user_ns.flags = 0;
2992
2993                 r = -EFAULT;
2994                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2995                         goto out;
2996                 r = 0;
2997                 break;
2998         }
2999
3000         default:
3001                 ;
3002         }
3003 out:
3004         return r;
3005 }
3006
3007 static void kvm_init_msr_list(void)
3008 {
3009         u32 dummy[2];
3010         unsigned i, j;
3011
3012         /* skip the first msrs in the list. KVM-specific */
3013         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3014                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3015                         continue;
3016                 if (j < i)
3017                         msrs_to_save[j] = msrs_to_save[i];
3018                 j++;
3019         }
3020         num_msrs_to_save = j;
3021 }
3022
3023 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3024                            const void *v)
3025 {
3026         if (vcpu->arch.apic &&
3027             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3028                 return 0;
3029
3030         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3031 }
3032
3033 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3034 {
3035         if (vcpu->arch.apic &&
3036             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3037                 return 0;
3038
3039         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3040 }
3041
3042 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3043                                struct kvm_vcpu *vcpu)
3044 {
3045         void *data = val;
3046         int r = X86EMUL_CONTINUE;
3047
3048         while (bytes) {
3049                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3050                 unsigned offset = addr & (PAGE_SIZE-1);
3051                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3052                 int ret;
3053
3054                 if (gpa == UNMAPPED_GVA) {
3055                         r = X86EMUL_PROPAGATE_FAULT;
3056                         goto out;
3057                 }
3058                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3059                 if (ret < 0) {
3060                         r = X86EMUL_UNHANDLEABLE;
3061                         goto out;
3062                 }
3063
3064                 bytes -= toread;
3065                 data += toread;
3066                 addr += toread;
3067         }
3068 out:
3069         return r;
3070 }
3071
3072 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3073                                 struct kvm_vcpu *vcpu)
3074 {
3075         void *data = val;
3076         int r = X86EMUL_CONTINUE;
3077
3078         while (bytes) {
3079                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3080                 unsigned offset = addr & (PAGE_SIZE-1);
3081                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3082                 int ret;
3083
3084                 if (gpa == UNMAPPED_GVA) {
3085                         r = X86EMUL_PROPAGATE_FAULT;
3086                         goto out;
3087                 }
3088                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3089                 if (ret < 0) {
3090                         r = X86EMUL_UNHANDLEABLE;
3091                         goto out;
3092                 }
3093
3094                 bytes -= towrite;
3095                 data += towrite;
3096                 addr += towrite;
3097         }
3098 out:
3099         return r;
3100 }
3101
3102
3103 static int emulator_read_emulated(unsigned long addr,
3104                                   void *val,
3105                                   unsigned int bytes,
3106                                   struct kvm_vcpu *vcpu)
3107 {
3108         gpa_t                 gpa;
3109
3110         if (vcpu->mmio_read_completed) {
3111                 memcpy(val, vcpu->mmio_data, bytes);
3112                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3113                                vcpu->mmio_phys_addr, *(u64 *)val);
3114                 vcpu->mmio_read_completed = 0;
3115                 return X86EMUL_CONTINUE;
3116         }
3117
3118         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3119
3120         /* For APIC access vmexit */
3121         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3122                 goto mmio;
3123
3124         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
3125                                 == X86EMUL_CONTINUE)
3126                 return X86EMUL_CONTINUE;
3127         if (gpa == UNMAPPED_GVA)
3128                 return X86EMUL_PROPAGATE_FAULT;
3129
3130 mmio:
3131         /*
3132          * Is this MMIO handled locally?
3133          */
3134         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3135                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3136                 return X86EMUL_CONTINUE;
3137         }
3138
3139         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3140
3141         vcpu->mmio_needed = 1;
3142         vcpu->mmio_phys_addr = gpa;
3143         vcpu->mmio_size = bytes;
3144         vcpu->mmio_is_write = 0;
3145
3146         return X86EMUL_UNHANDLEABLE;
3147 }
3148
3149 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3150                           const void *val, int bytes)
3151 {
3152         int ret;
3153
3154         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3155         if (ret < 0)
3156                 return 0;
3157         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3158         return 1;
3159 }
3160
3161 static int emulator_write_emulated_onepage(unsigned long addr,
3162                                            const void *val,
3163                                            unsigned int bytes,
3164                                            struct kvm_vcpu *vcpu)
3165 {
3166         gpa_t                 gpa;
3167
3168         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3169
3170         if (gpa == UNMAPPED_GVA) {
3171                 kvm_inject_page_fault(vcpu, addr, 2);
3172                 return X86EMUL_PROPAGATE_FAULT;
3173         }
3174
3175         /* For APIC access vmexit */
3176         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3177                 goto mmio;
3178
3179         if (emulator_write_phys(vcpu, gpa, val, bytes))
3180                 return X86EMUL_CONTINUE;
3181
3182 mmio:
3183         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3184         /*
3185          * Is this MMIO handled locally?
3186          */
3187         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3188                 return X86EMUL_CONTINUE;
3189
3190         vcpu->mmio_needed = 1;
3191         vcpu->mmio_phys_addr = gpa;
3192         vcpu->mmio_size = bytes;
3193         vcpu->mmio_is_write = 1;
3194         memcpy(vcpu->mmio_data, val, bytes);
3195
3196         return X86EMUL_CONTINUE;
3197 }
3198
3199 int emulator_write_emulated(unsigned long addr,
3200                                    const void *val,
3201                                    unsigned int bytes,
3202                                    struct kvm_vcpu *vcpu)
3203 {
3204         /* Crossing a page boundary? */
3205         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3206                 int rc, now;
3207
3208                 now = -addr & ~PAGE_MASK;
3209                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3210                 if (rc != X86EMUL_CONTINUE)
3211                         return rc;
3212                 addr += now;
3213                 val += now;
3214                 bytes -= now;
3215         }
3216         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3217 }
3218 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3219
3220 static int emulator_cmpxchg_emulated(unsigned long addr,
3221                                      const void *old,
3222                                      const void *new,
3223                                      unsigned int bytes,
3224                                      struct kvm_vcpu *vcpu)
3225 {
3226         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3227 #ifndef CONFIG_X86_64
3228         /* guests cmpxchg8b have to be emulated atomically */
3229         if (bytes == 8) {
3230                 gpa_t gpa;
3231                 struct page *page;
3232                 char *kaddr;
3233                 u64 val;
3234
3235                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3236
3237                 if (gpa == UNMAPPED_GVA ||
3238                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3239                         goto emul_write;
3240
3241                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3242                         goto emul_write;
3243
3244                 val = *(u64 *)new;
3245
3246                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3247
3248                 kaddr = kmap_atomic(page, KM_USER0);
3249                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3250                 kunmap_atomic(kaddr, KM_USER0);
3251                 kvm_release_page_dirty(page);
3252         }
3253 emul_write:
3254 #endif
3255
3256         return emulator_write_emulated(addr, new, bytes, vcpu);
3257 }
3258
3259 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3260 {
3261         return kvm_x86_ops->get_segment_base(vcpu, seg);
3262 }
3263
3264 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3265 {
3266         kvm_mmu_invlpg(vcpu, address);
3267         return X86EMUL_CONTINUE;
3268 }
3269
3270 int emulate_clts(struct kvm_vcpu *vcpu)
3271 {
3272         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3273         kvm_x86_ops->fpu_activate(vcpu);
3274         return X86EMUL_CONTINUE;
3275 }
3276
3277 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3278 {
3279         return kvm_x86_ops->get_dr(ctxt->vcpu, dr, dest);
3280 }
3281
3282 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3283 {
3284         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3285
3286         return kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask);
3287 }
3288
3289 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3290 {
3291         u8 opcodes[4];
3292         unsigned long rip = kvm_rip_read(vcpu);
3293         unsigned long rip_linear;
3294
3295         if (!printk_ratelimit())
3296                 return;
3297
3298         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3299
3300         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
3301
3302         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3303                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3304 }
3305 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3306
3307 static struct x86_emulate_ops emulate_ops = {
3308         .read_std            = kvm_read_guest_virt,
3309         .read_emulated       = emulator_read_emulated,
3310         .write_emulated      = emulator_write_emulated,
3311         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3312 };
3313
3314 static void cache_all_regs(struct kvm_vcpu *vcpu)
3315 {
3316         kvm_register_read(vcpu, VCPU_REGS_RAX);
3317         kvm_register_read(vcpu, VCPU_REGS_RSP);
3318         kvm_register_read(vcpu, VCPU_REGS_RIP);
3319         vcpu->arch.regs_dirty = ~0;
3320 }
3321
3322 int emulate_instruction(struct kvm_vcpu *vcpu,
3323                         unsigned long cr2,
3324                         u16 error_code,
3325                         int emulation_type)
3326 {
3327         int r, shadow_mask;
3328         struct decode_cache *c;
3329         struct kvm_run *run = vcpu->run;
3330
3331         kvm_clear_exception_queue(vcpu);
3332         vcpu->arch.mmio_fault_cr2 = cr2;
3333         /*
3334          * TODO: fix emulate.c to use guest_read/write_register
3335          * instead of direct ->regs accesses, can save hundred cycles
3336          * on Intel for instructions that don't read/change RSP, for
3337          * for example.
3338          */
3339         cache_all_regs(vcpu);
3340
3341         vcpu->mmio_is_write = 0;
3342         vcpu->arch.pio.string = 0;
3343
3344         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3345                 int cs_db, cs_l;
3346                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3347
3348                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3349                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3350                 vcpu->arch.emulate_ctxt.mode =
3351                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3352                         ? X86EMUL_MODE_REAL : cs_l
3353                         ? X86EMUL_MODE_PROT64 : cs_db
3354                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3355
3356                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3357
3358                 /* Only allow emulation of specific instructions on #UD
3359                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3360                 c = &vcpu->arch.emulate_ctxt.decode;
3361                 if (emulation_type & EMULTYPE_TRAP_UD) {
3362                         if (!c->twobyte)
3363                                 return EMULATE_FAIL;
3364                         switch (c->b) {
3365                         case 0x01: /* VMMCALL */
3366                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3367                                         return EMULATE_FAIL;
3368                                 break;
3369                         case 0x34: /* sysenter */
3370                         case 0x35: /* sysexit */
3371                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3372                                         return EMULATE_FAIL;
3373                                 break;
3374                         case 0x05: /* syscall */
3375                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3376                                         return EMULATE_FAIL;
3377                                 break;
3378                         default:
3379                                 return EMULATE_FAIL;
3380                         }
3381
3382                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3383                                 return EMULATE_FAIL;
3384                 }
3385
3386                 ++vcpu->stat.insn_emulation;
3387                 if (r)  {
3388                         ++vcpu->stat.insn_emulation_fail;
3389                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3390                                 return EMULATE_DONE;
3391                         return EMULATE_FAIL;
3392                 }
3393         }
3394
3395         if (emulation_type & EMULTYPE_SKIP) {
3396                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3397                 return EMULATE_DONE;
3398         }
3399
3400         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3401         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3402
3403         if (r == 0)
3404                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3405
3406         if (vcpu->arch.pio.string)
3407                 return EMULATE_DO_MMIO;
3408
3409         if ((r || vcpu->mmio_is_write) && run) {
3410                 run->exit_reason = KVM_EXIT_MMIO;
3411                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3412                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3413                 run->mmio.len = vcpu->mmio_size;
3414                 run->mmio.is_write = vcpu->mmio_is_write;
3415         }
3416
3417         if (r) {
3418                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3419                         return EMULATE_DONE;
3420                 if (!vcpu->mmio_needed) {
3421                         kvm_report_emulation_failure(vcpu, "mmio");
3422                         return EMULATE_FAIL;
3423                 }
3424                 return EMULATE_DO_MMIO;
3425         }
3426
3427         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3428
3429         if (vcpu->mmio_is_write) {
3430                 vcpu->mmio_needed = 0;
3431                 return EMULATE_DO_MMIO;
3432         }
3433
3434         return EMULATE_DONE;
3435 }
3436 EXPORT_SYMBOL_GPL(emulate_instruction);
3437
3438 static int pio_copy_data(struct kvm_vcpu *vcpu)
3439 {
3440         void *p = vcpu->arch.pio_data;
3441         gva_t q = vcpu->arch.pio.guest_gva;
3442         unsigned bytes;
3443         int ret;
3444
3445         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3446         if (vcpu->arch.pio.in)
3447                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
3448         else
3449                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
3450         return ret;
3451 }
3452
3453 int complete_pio(struct kvm_vcpu *vcpu)
3454 {
3455         struct kvm_pio_request *io = &vcpu->arch.pio;
3456         long delta;
3457         int r;
3458         unsigned long val;
3459
3460         if (!io->string) {
3461                 if (io->in) {
3462                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3463                         memcpy(&val, vcpu->arch.pio_data, io->size);
3464                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3465                 }
3466         } else {
3467                 if (io->in) {
3468                         r = pio_copy_data(vcpu);
3469                         if (r)
3470                                 return r;
3471                 }
3472
3473                 delta = 1;
3474                 if (io->rep) {
3475                         delta *= io->cur_count;
3476                         /*
3477                          * The size of the register should really depend on
3478                          * current address size.
3479                          */
3480                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3481                         val -= delta;
3482                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3483                 }
3484                 if (io->down)
3485                         delta = -delta;
3486                 delta *= io->size;
3487                 if (io->in) {
3488                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3489                         val += delta;
3490                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3491                 } else {
3492                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3493                         val += delta;
3494                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3495                 }
3496         }
3497
3498         io->count -= io->cur_count;
3499         io->cur_count = 0;
3500
3501         return 0;
3502 }
3503
3504 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3505 {
3506         /* TODO: String I/O for in kernel device */
3507         int r;
3508
3509         if (vcpu->arch.pio.in)
3510                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3511                                     vcpu->arch.pio.size, pd);
3512         else
3513                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3514                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3515                                      pd);
3516         return r;
3517 }
3518
3519 static int pio_string_write(struct kvm_vcpu *vcpu)
3520 {
3521         struct kvm_pio_request *io = &vcpu->arch.pio;
3522         void *pd = vcpu->arch.pio_data;
3523         int i, r = 0;
3524
3525         for (i = 0; i < io->cur_count; i++) {
3526                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3527                                      io->port, io->size, pd)) {
3528                         r = -EOPNOTSUPP;
3529                         break;
3530                 }
3531                 pd += io->size;
3532         }
3533         return r;
3534 }
3535
3536 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3537 {
3538         unsigned long val;
3539
3540         vcpu->run->exit_reason = KVM_EXIT_IO;
3541         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3542         vcpu->run->io.size = vcpu->arch.pio.size = size;
3543         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3544         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3545         vcpu->run->io.port = vcpu->arch.pio.port = port;
3546         vcpu->arch.pio.in = in;
3547         vcpu->arch.pio.string = 0;
3548         vcpu->arch.pio.down = 0;
3549         vcpu->arch.pio.rep = 0;
3550
3551         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3552                       size, 1);
3553
3554         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3555         memcpy(vcpu->arch.pio_data, &val, 4);
3556
3557         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3558                 complete_pio(vcpu);
3559                 return 1;
3560         }
3561         return 0;
3562 }
3563 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3564
3565 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3566                   int size, unsigned long count, int down,
3567                   gva_t address, int rep, unsigned port)
3568 {
3569         unsigned now, in_page;
3570         int ret = 0;
3571
3572         vcpu->run->exit_reason = KVM_EXIT_IO;
3573         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3574         vcpu->run->io.size = vcpu->arch.pio.size = size;
3575         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3576         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3577         vcpu->run->io.port = vcpu->arch.pio.port = port;
3578         vcpu->arch.pio.in = in;
3579         vcpu->arch.pio.string = 1;
3580         vcpu->arch.pio.down = down;
3581         vcpu->arch.pio.rep = rep;
3582
3583         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3584                       size, count);
3585
3586         if (!count) {
3587                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3588                 return 1;
3589         }
3590
3591         if (!down)
3592                 in_page = PAGE_SIZE - offset_in_page(address);
3593         else
3594                 in_page = offset_in_page(address) + size;
3595         now = min(count, (unsigned long)in_page / size);
3596         if (!now)
3597                 now = 1;
3598         if (down) {
3599                 /*
3600                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3601                  */
3602                 pr_unimpl(vcpu, "guest string pio down\n");
3603                 kvm_inject_gp(vcpu, 0);
3604                 return 1;
3605         }
3606         vcpu->run->io.count = now;
3607         vcpu->arch.pio.cur_count = now;
3608
3609         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3610                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3611
3612         vcpu->arch.pio.guest_gva = address;
3613
3614         if (!vcpu->arch.pio.in) {
3615                 /* string PIO write */
3616                 ret = pio_copy_data(vcpu);
3617                 if (ret == X86EMUL_PROPAGATE_FAULT) {
3618                         kvm_inject_gp(vcpu, 0);
3619                         return 1;
3620                 }
3621                 if (ret == 0 && !pio_string_write(vcpu)) {
3622                         complete_pio(vcpu);
3623                         if (vcpu->arch.pio.count == 0)
3624                                 ret = 1;
3625                 }
3626         }
3627         /* no string PIO read support yet */
3628
3629         return ret;
3630 }
3631 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3632
3633 static void bounce_off(void *info)
3634 {
3635         /* nothing */
3636 }
3637
3638 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3639                                      void *data)
3640 {
3641         struct cpufreq_freqs *freq = data;
3642         struct kvm *kvm;
3643         struct kvm_vcpu *vcpu;
3644         int i, send_ipi = 0;
3645
3646         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3647                 return 0;
3648         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3649                 return 0;
3650         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3651
3652         spin_lock(&kvm_lock);
3653         list_for_each_entry(kvm, &vm_list, vm_list) {
3654                 kvm_for_each_vcpu(i, vcpu, kvm) {
3655                         if (vcpu->cpu != freq->cpu)
3656                                 continue;
3657                         if (!kvm_request_guest_time_update(vcpu))
3658                                 continue;
3659                         if (vcpu->cpu != smp_processor_id())
3660                                 send_ipi++;
3661                 }
3662         }
3663         spin_unlock(&kvm_lock);
3664
3665         if (freq->old < freq->new && send_ipi) {
3666                 /*
3667                  * We upscale the frequency.  Must make the guest
3668                  * doesn't see old kvmclock values while running with
3669                  * the new frequency, otherwise we risk the guest sees
3670                  * time go backwards.
3671                  *
3672                  * In case we update the frequency for another cpu
3673                  * (which might be in guest context) send an interrupt
3674                  * to kick the cpu out of guest context.  Next time
3675                  * guest context is entered kvmclock will be updated,
3676                  * so the guest will not see stale values.
3677                  */
3678                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3679         }
3680         return 0;
3681 }
3682
3683 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3684         .notifier_call  = kvmclock_cpufreq_notifier
3685 };
3686
3687 static void kvm_timer_init(void)
3688 {
3689         int cpu;
3690
3691         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3692                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3693                                           CPUFREQ_TRANSITION_NOTIFIER);
3694                 for_each_online_cpu(cpu) {
3695                         unsigned long khz = cpufreq_get(cpu);
3696                         if (!khz)
3697                                 khz = tsc_khz;
3698                         per_cpu(cpu_tsc_khz, cpu) = khz;
3699                 }
3700         } else {
3701                 for_each_possible_cpu(cpu)
3702                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3703         }
3704 }
3705
3706 int kvm_arch_init(void *opaque)
3707 {
3708         int r;
3709         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3710
3711         if (kvm_x86_ops) {
3712                 printk(KERN_ERR "kvm: already loaded the other module\n");
3713                 r = -EEXIST;
3714                 goto out;
3715         }
3716
3717         if (!ops->cpu_has_kvm_support()) {
3718                 printk(KERN_ERR "kvm: no hardware support\n");
3719                 r = -EOPNOTSUPP;
3720                 goto out;
3721         }
3722         if (ops->disabled_by_bios()) {
3723                 printk(KERN_ERR "kvm: disabled by bios\n");
3724                 r = -EOPNOTSUPP;
3725                 goto out;
3726         }
3727
3728         r = kvm_mmu_module_init();
3729         if (r)
3730                 goto out;
3731
3732         kvm_init_msr_list();
3733
3734         kvm_x86_ops = ops;
3735         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3736         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3737         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3738                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3739
3740         kvm_timer_init();
3741
3742         return 0;
3743
3744 out:
3745         return r;
3746 }
3747
3748 void kvm_arch_exit(void)
3749 {
3750         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3751                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3752                                             CPUFREQ_TRANSITION_NOTIFIER);
3753         kvm_x86_ops = NULL;
3754         kvm_mmu_module_exit();
3755 }
3756
3757 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3758 {
3759         ++vcpu->stat.halt_exits;
3760         if (irqchip_in_kernel(vcpu->kvm)) {
3761                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3762                 return 1;
3763         } else {
3764                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3765                 return 0;
3766         }
3767 }
3768 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3769
3770 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3771                            unsigned long a1)
3772 {
3773         if (is_long_mode(vcpu))
3774                 return a0;
3775         else
3776                 return a0 | ((gpa_t)a1 << 32);
3777 }
3778
3779 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3780 {
3781         u64 param, ingpa, outgpa, ret;
3782         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3783         bool fast, longmode;
3784         int cs_db, cs_l;
3785
3786         /*
3787          * hypercall generates UD from non zero cpl and real mode
3788          * per HYPER-V spec
3789          */
3790         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
3791                 kvm_queue_exception(vcpu, UD_VECTOR);
3792                 return 0;
3793         }
3794
3795         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3796         longmode = is_long_mode(vcpu) && cs_l == 1;
3797
3798         if (!longmode) {
3799                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3800                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
3801                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3802                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
3803                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3804                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
3805         }
3806 #ifdef CONFIG_X86_64
3807         else {
3808                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3809                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3810                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3811         }
3812 #endif
3813
3814         code = param & 0xffff;
3815         fast = (param >> 16) & 0x1;
3816         rep_cnt = (param >> 32) & 0xfff;
3817         rep_idx = (param >> 48) & 0xfff;
3818
3819         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3820
3821         switch (code) {
3822         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
3823                 kvm_vcpu_on_spin(vcpu);
3824                 break;
3825         default:
3826                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
3827                 break;
3828         }
3829
3830         ret = res | (((u64)rep_done & 0xfff) << 32);
3831         if (longmode) {
3832                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3833         } else {
3834                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3835                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3836         }
3837
3838         return 1;
3839 }
3840
3841 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3842 {
3843         unsigned long nr, a0, a1, a2, a3, ret;
3844         int r = 1;
3845
3846         if (kvm_hv_hypercall_enabled(vcpu->kvm))
3847                 return kvm_hv_hypercall(vcpu);
3848
3849         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3850         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3851         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3852         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3853         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3854
3855         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3856
3857         if (!is_long_mode(vcpu)) {
3858                 nr &= 0xFFFFFFFF;
3859                 a0 &= 0xFFFFFFFF;
3860                 a1 &= 0xFFFFFFFF;
3861                 a2 &= 0xFFFFFFFF;
3862                 a3 &= 0xFFFFFFFF;
3863         }
3864
3865         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3866                 ret = -KVM_EPERM;
3867                 goto out;
3868         }
3869
3870         switch (nr) {
3871         case KVM_HC_VAPIC_POLL_IRQ:
3872                 ret = 0;
3873                 break;
3874         case KVM_HC_MMU_OP:
3875                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3876                 break;
3877         default:
3878                 ret = -KVM_ENOSYS;
3879                 break;
3880         }
3881 out:
3882         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3883         ++vcpu->stat.hypercalls;
3884         return r;
3885 }
3886 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3887
3888 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3889 {
3890         char instruction[3];
3891         unsigned long rip = kvm_rip_read(vcpu);
3892
3893         /*
3894          * Blow out the MMU to ensure that no other VCPU has an active mapping
3895          * to ensure that the updated hypercall appears atomically across all
3896          * VCPUs.
3897          */
3898         kvm_mmu_zap_all(vcpu->kvm);
3899
3900         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3901
3902         return emulator_write_emulated(rip, instruction, 3, vcpu);
3903 }
3904
3905 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3906 {
3907         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3908 }
3909
3910 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3911 {
3912         struct descriptor_table dt = { limit, base };
3913
3914         kvm_x86_ops->set_gdt(vcpu, &dt);
3915 }
3916
3917 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3918 {
3919         struct descriptor_table dt = { limit, base };
3920
3921         kvm_x86_ops->set_idt(vcpu, &dt);
3922 }
3923
3924 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3925                    unsigned long *rflags)
3926 {
3927         kvm_lmsw(vcpu, msw);
3928         *rflags = kvm_get_rflags(vcpu);
3929 }
3930
3931 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3932 {
3933         unsigned long value;
3934
3935         switch (cr) {
3936         case 0:
3937                 value = kvm_read_cr0(vcpu);
3938                 break;
3939         case 2:
3940                 value = vcpu->arch.cr2;
3941                 break;
3942         case 3:
3943                 value = vcpu->arch.cr3;
3944                 break;
3945         case 4:
3946                 value = kvm_read_cr4(vcpu);
3947                 break;
3948         case 8:
3949                 value = kvm_get_cr8(vcpu);
3950                 break;
3951         default:
3952                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3953                 return 0;
3954         }
3955
3956         return value;
3957 }
3958
3959 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3960                      unsigned long *rflags)
3961 {
3962         switch (cr) {
3963         case 0:
3964                 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3965                 *rflags = kvm_get_rflags(vcpu);
3966                 break;
3967         case 2:
3968                 vcpu->arch.cr2 = val;
3969                 break;
3970         case 3:
3971                 kvm_set_cr3(vcpu, val);
3972                 break;
3973         case 4:
3974                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3975                 break;
3976         case 8:
3977                 kvm_set_cr8(vcpu, val & 0xfUL);
3978                 break;
3979         default:
3980                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3981         }
3982 }
3983
3984 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3985 {
3986         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3987         int j, nent = vcpu->arch.cpuid_nent;
3988
3989         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3990         /* when no next entry is found, the current entry[i] is reselected */
3991         for (j = i + 1; ; j = (j + 1) % nent) {
3992                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3993                 if (ej->function == e->function) {
3994                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3995                         return j;
3996                 }
3997         }
3998         return 0; /* silence gcc, even though control never reaches here */
3999 }
4000
4001 /* find an entry with matching function, matching index (if needed), and that
4002  * should be read next (if it's stateful) */
4003 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4004         u32 function, u32 index)
4005 {
4006         if (e->function != function)
4007                 return 0;
4008         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4009                 return 0;
4010         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4011             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4012                 return 0;
4013         return 1;
4014 }
4015
4016 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4017                                               u32 function, u32 index)
4018 {
4019         int i;
4020         struct kvm_cpuid_entry2 *best = NULL;
4021
4022         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4023                 struct kvm_cpuid_entry2 *e;
4024
4025                 e = &vcpu->arch.cpuid_entries[i];
4026                 if (is_matching_cpuid_entry(e, function, index)) {
4027                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4028                                 move_to_next_stateful_cpuid_entry(vcpu, i);
4029                         best = e;
4030                         break;
4031                 }
4032                 /*
4033                  * Both basic or both extended?
4034                  */
4035                 if (((e->function ^ function) & 0x80000000) == 0)
4036                         if (!best || e->function > best->function)
4037                                 best = e;
4038         }
4039         return best;
4040 }
4041 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4042
4043 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4044 {
4045         struct kvm_cpuid_entry2 *best;
4046
4047         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4048         if (best)
4049                 return best->eax & 0xff;
4050         return 36;
4051 }
4052
4053 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4054 {
4055         u32 function, index;
4056         struct kvm_cpuid_entry2 *best;
4057
4058         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4059         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4060         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4061         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4062         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4063         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4064         best = kvm_find_cpuid_entry(vcpu, function, index);
4065         if (best) {
4066                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4067                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4068                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4069                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4070         }
4071         kvm_x86_ops->skip_emulated_instruction(vcpu);
4072         trace_kvm_cpuid(function,
4073                         kvm_register_read(vcpu, VCPU_REGS_RAX),
4074                         kvm_register_read(vcpu, VCPU_REGS_RBX),
4075                         kvm_register_read(vcpu, VCPU_REGS_RCX),
4076                         kvm_register_read(vcpu, VCPU_REGS_RDX));
4077 }
4078 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4079
4080 /*
4081  * Check if userspace requested an interrupt window, and that the
4082  * interrupt window is open.
4083  *
4084  * No need to exit to userspace if we already have an interrupt queued.
4085  */
4086 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4087 {
4088         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4089                 vcpu->run->request_interrupt_window &&
4090                 kvm_arch_interrupt_allowed(vcpu));
4091 }
4092
4093 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4094 {
4095         struct kvm_run *kvm_run = vcpu->run;
4096
4097         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4098         kvm_run->cr8 = kvm_get_cr8(vcpu);
4099         kvm_run->apic_base = kvm_get_apic_base(vcpu);
4100         if (irqchip_in_kernel(vcpu->kvm))
4101                 kvm_run->ready_for_interrupt_injection = 1;
4102         else
4103                 kvm_run->ready_for_interrupt_injection =
4104                         kvm_arch_interrupt_allowed(vcpu) &&
4105                         !kvm_cpu_has_interrupt(vcpu) &&
4106                         !kvm_event_needs_reinjection(vcpu);
4107 }
4108
4109 static void vapic_enter(struct kvm_vcpu *vcpu)
4110 {
4111         struct kvm_lapic *apic = vcpu->arch.apic;
4112         struct page *page;
4113
4114         if (!apic || !apic->vapic_addr)
4115                 return;
4116
4117         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4118
4119         vcpu->arch.apic->vapic_page = page;
4120 }
4121
4122 static void vapic_exit(struct kvm_vcpu *vcpu)
4123 {
4124         struct kvm_lapic *apic = vcpu->arch.apic;
4125         int idx;
4126
4127         if (!apic || !apic->vapic_addr)
4128                 return;
4129
4130         idx = srcu_read_lock(&vcpu->kvm->srcu);
4131         kvm_release_page_dirty(apic->vapic_page);
4132         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4133         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4134 }
4135
4136 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4137 {
4138         int max_irr, tpr;
4139
4140         if (!kvm_x86_ops->update_cr8_intercept)
4141                 return;
4142
4143         if (!vcpu->arch.apic)
4144                 return;
4145
4146         if (!vcpu->arch.apic->vapic_addr)
4147                 max_irr = kvm_lapic_find_highest_irr(vcpu);
4148         else
4149                 max_irr = -1;
4150
4151         if (max_irr != -1)
4152                 max_irr >>= 4;
4153
4154         tpr = kvm_lapic_get_cr8(vcpu);
4155
4156         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4157 }
4158
4159 static void inject_pending_event(struct kvm_vcpu *vcpu)
4160 {
4161         /* try to reinject previous events if any */
4162         if (vcpu->arch.exception.pending) {
4163                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4164                                           vcpu->arch.exception.has_error_code,
4165                                           vcpu->arch.exception.error_code);
4166                 return;
4167         }
4168
4169         if (vcpu->arch.nmi_injected) {
4170                 kvm_x86_ops->set_nmi(vcpu);
4171                 return;
4172         }
4173
4174         if (vcpu->arch.interrupt.pending) {
4175                 kvm_x86_ops->set_irq(vcpu);
4176                 return;
4177         }
4178
4179         /* try to inject new event if pending */
4180         if (vcpu->arch.nmi_pending) {
4181                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4182                         vcpu->arch.nmi_pending = false;
4183                         vcpu->arch.nmi_injected = true;
4184                         kvm_x86_ops->set_nmi(vcpu);
4185                 }
4186         } else if (kvm_cpu_has_interrupt(vcpu)) {
4187                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4188                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4189                                             false);
4190                         kvm_x86_ops->set_irq(vcpu);
4191                 }
4192         }
4193 }
4194
4195 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4196 {
4197         int r;
4198         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4199                 vcpu->run->request_interrupt_window;
4200
4201         if (vcpu->requests)
4202                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4203                         kvm_mmu_unload(vcpu);
4204
4205         r = kvm_mmu_reload(vcpu);
4206         if (unlikely(r))
4207                 goto out;
4208
4209         if (vcpu->requests) {
4210                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4211                         __kvm_migrate_timers(vcpu);
4212                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4213                         kvm_write_guest_time(vcpu);
4214                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4215                         kvm_mmu_sync_roots(vcpu);
4216                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4217                         kvm_x86_ops->tlb_flush(vcpu);
4218                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4219                                        &vcpu->requests)) {
4220                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4221                         r = 0;
4222                         goto out;
4223                 }
4224                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4225                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4226                         r = 0;
4227                         goto out;
4228                 }
4229                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4230                         vcpu->fpu_active = 0;
4231                         kvm_x86_ops->fpu_deactivate(vcpu);
4232                 }
4233         }
4234
4235         preempt_disable();
4236
4237         kvm_x86_ops->prepare_guest_switch(vcpu);
4238         if (vcpu->fpu_active)
4239                 kvm_load_guest_fpu(vcpu);
4240
4241         local_irq_disable();
4242
4243         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4244         smp_mb__after_clear_bit();
4245
4246         if (vcpu->requests || need_resched() || signal_pending(current)) {
4247                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4248                 local_irq_enable();
4249                 preempt_enable();
4250                 r = 1;
4251                 goto out;
4252         }
4253
4254         inject_pending_event(vcpu);
4255
4256         /* enable NMI/IRQ window open exits if needed */
4257         if (vcpu->arch.nmi_pending)
4258                 kvm_x86_ops->enable_nmi_window(vcpu);
4259         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4260                 kvm_x86_ops->enable_irq_window(vcpu);
4261
4262         if (kvm_lapic_enabled(vcpu)) {
4263                 update_cr8_intercept(vcpu);
4264                 kvm_lapic_sync_to_vapic(vcpu);
4265         }
4266
4267         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4268
4269         kvm_guest_enter();
4270
4271         if (unlikely(vcpu->arch.switch_db_regs)) {
4272                 set_debugreg(0, 7);
4273                 set_debugreg(vcpu->arch.eff_db[0], 0);
4274                 set_debugreg(vcpu->arch.eff_db[1], 1);
4275                 set_debugreg(vcpu->arch.eff_db[2], 2);
4276                 set_debugreg(vcpu->arch.eff_db[3], 3);
4277         }
4278
4279         trace_kvm_entry(vcpu->vcpu_id);
4280         kvm_x86_ops->run(vcpu);
4281
4282         /*
4283          * If the guest has used debug registers, at least dr7
4284          * will be disabled while returning to the host.
4285          * If we don't have active breakpoints in the host, we don't
4286          * care about the messed up debug address registers. But if
4287          * we have some of them active, restore the old state.
4288          */
4289         if (hw_breakpoint_active())
4290                 hw_breakpoint_restore();
4291
4292         set_bit(KVM_REQ_KICK, &vcpu->requests);
4293         local_irq_enable();
4294
4295         ++vcpu->stat.exits;
4296
4297         /*
4298          * We must have an instruction between local_irq_enable() and
4299          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4300          * the interrupt shadow.  The stat.exits increment will do nicely.
4301          * But we need to prevent reordering, hence this barrier():
4302          */
4303         barrier();
4304
4305         kvm_guest_exit();
4306
4307         preempt_enable();
4308
4309         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4310
4311         /*
4312          * Profile KVM exit RIPs:
4313          */
4314         if (unlikely(prof_on == KVM_PROFILING)) {
4315                 unsigned long rip = kvm_rip_read(vcpu);
4316                 profile_hit(KVM_PROFILING, (void *)rip);
4317         }
4318
4319
4320         kvm_lapic_sync_from_vapic(vcpu);
4321
4322         r = kvm_x86_ops->handle_exit(vcpu);
4323 out:
4324         return r;
4325 }
4326
4327
4328 static int __vcpu_run(struct kvm_vcpu *vcpu)
4329 {
4330         int r;
4331         struct kvm *kvm = vcpu->kvm;
4332
4333         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4334                 pr_debug("vcpu %d received sipi with vector # %x\n",
4335                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4336                 kvm_lapic_reset(vcpu);
4337                 r = kvm_arch_vcpu_reset(vcpu);
4338                 if (r)
4339                         return r;
4340                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4341         }
4342
4343         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4344         vapic_enter(vcpu);
4345
4346         r = 1;
4347         while (r > 0) {
4348                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4349                         r = vcpu_enter_guest(vcpu);
4350                 else {
4351                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4352                         kvm_vcpu_block(vcpu);
4353                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4354                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4355                         {
4356                                 switch(vcpu->arch.mp_state) {
4357                                 case KVM_MP_STATE_HALTED:
4358                                         vcpu->arch.mp_state =
4359                                                 KVM_MP_STATE_RUNNABLE;
4360                                 case KVM_MP_STATE_RUNNABLE:
4361                                         break;
4362                                 case KVM_MP_STATE_SIPI_RECEIVED:
4363                                 default:
4364                                         r = -EINTR;
4365                                         break;
4366                                 }
4367                         }
4368                 }
4369
4370                 if (r <= 0)
4371                         break;
4372
4373                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4374                 if (kvm_cpu_has_pending_timer(vcpu))
4375                         kvm_inject_pending_timer_irqs(vcpu);
4376
4377                 if (dm_request_for_irq_injection(vcpu)) {
4378                         r = -EINTR;
4379                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4380                         ++vcpu->stat.request_irq_exits;
4381                 }
4382                 if (signal_pending(current)) {
4383                         r = -EINTR;
4384                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4385                         ++vcpu->stat.signal_exits;
4386                 }
4387                 if (need_resched()) {
4388                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4389                         kvm_resched(vcpu);
4390                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4391                 }
4392         }
4393
4394         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4395         post_kvm_run_save(vcpu);
4396
4397         vapic_exit(vcpu);
4398
4399         return r;
4400 }
4401
4402 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4403 {
4404         int r;
4405         sigset_t sigsaved;
4406
4407         vcpu_load(vcpu);
4408
4409         if (vcpu->sigset_active)
4410                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4411
4412         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4413                 kvm_vcpu_block(vcpu);
4414                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4415                 r = -EAGAIN;
4416                 goto out;
4417         }
4418
4419         /* re-sync apic's tpr */
4420         if (!irqchip_in_kernel(vcpu->kvm))
4421                 kvm_set_cr8(vcpu, kvm_run->cr8);
4422
4423         if (vcpu->arch.pio.cur_count) {
4424                 r = complete_pio(vcpu);
4425                 if (r)
4426                         goto out;
4427         }
4428         if (vcpu->mmio_needed) {
4429                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4430                 vcpu->mmio_read_completed = 1;
4431                 vcpu->mmio_needed = 0;
4432
4433                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4434                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4435                                         EMULTYPE_NO_DECODE);
4436                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4437                 if (r == EMULATE_DO_MMIO) {
4438                         /*
4439                          * Read-modify-write.  Back to userspace.
4440                          */
4441                         r = 0;
4442                         goto out;
4443                 }
4444         }
4445         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4446                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4447                                      kvm_run->hypercall.ret);
4448
4449         r = __vcpu_run(vcpu);
4450
4451 out:
4452         if (vcpu->sigset_active)
4453                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4454
4455         vcpu_put(vcpu);
4456         return r;
4457 }
4458
4459 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4460 {
4461         vcpu_load(vcpu);
4462
4463         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4464         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4465         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4466         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4467         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4468         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4469         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4470         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4471 #ifdef CONFIG_X86_64
4472         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4473         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4474         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4475         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4476         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4477         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4478         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4479         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4480 #endif
4481
4482         regs->rip = kvm_rip_read(vcpu);
4483         regs->rflags = kvm_get_rflags(vcpu);
4484
4485         vcpu_put(vcpu);
4486
4487         return 0;
4488 }
4489
4490 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4491 {
4492         vcpu_load(vcpu);
4493
4494         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4495         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4496         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4497         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4498         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4499         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4500         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4501         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4502 #ifdef CONFIG_X86_64
4503         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4504         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4505         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4506         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4507         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4508         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4509         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4510         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4511 #endif
4512
4513         kvm_rip_write(vcpu, regs->rip);
4514         kvm_set_rflags(vcpu, regs->rflags);
4515
4516         vcpu->arch.exception.pending = false;
4517
4518         vcpu_put(vcpu);
4519
4520         return 0;
4521 }
4522
4523 void kvm_get_segment(struct kvm_vcpu *vcpu,
4524                      struct kvm_segment *var, int seg)
4525 {
4526         kvm_x86_ops->get_segment(vcpu, var, seg);
4527 }
4528
4529 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4530 {
4531         struct kvm_segment cs;
4532
4533         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4534         *db = cs.db;
4535         *l = cs.l;
4536 }
4537 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4538
4539 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4540                                   struct kvm_sregs *sregs)
4541 {
4542         struct descriptor_table dt;
4543
4544         vcpu_load(vcpu);
4545
4546         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4547         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4548         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4549         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4550         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4551         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4552
4553         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4554         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4555
4556         kvm_x86_ops->get_idt(vcpu, &dt);
4557         sregs->idt.limit = dt.limit;
4558         sregs->idt.base = dt.base;
4559         kvm_x86_ops->get_gdt(vcpu, &dt);
4560         sregs->gdt.limit = dt.limit;
4561         sregs->gdt.base = dt.base;
4562
4563         sregs->cr0 = kvm_read_cr0(vcpu);
4564         sregs->cr2 = vcpu->arch.cr2;
4565         sregs->cr3 = vcpu->arch.cr3;
4566         sregs->cr4 = kvm_read_cr4(vcpu);
4567         sregs->cr8 = kvm_get_cr8(vcpu);
4568         sregs->efer = vcpu->arch.efer;
4569         sregs->apic_base = kvm_get_apic_base(vcpu);
4570
4571         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4572
4573         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4574                 set_bit(vcpu->arch.interrupt.nr,
4575                         (unsigned long *)sregs->interrupt_bitmap);
4576
4577         vcpu_put(vcpu);
4578
4579         return 0;
4580 }
4581
4582 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4583                                     struct kvm_mp_state *mp_state)
4584 {
4585         vcpu_load(vcpu);
4586         mp_state->mp_state = vcpu->arch.mp_state;
4587         vcpu_put(vcpu);
4588         return 0;
4589 }
4590
4591 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4592                                     struct kvm_mp_state *mp_state)
4593 {
4594         vcpu_load(vcpu);
4595         vcpu->arch.mp_state = mp_state->mp_state;
4596         vcpu_put(vcpu);
4597         return 0;
4598 }
4599
4600 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4601                         struct kvm_segment *var, int seg)
4602 {
4603         kvm_x86_ops->set_segment(vcpu, var, seg);
4604 }
4605
4606 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4607                                    struct kvm_segment *kvm_desct)
4608 {
4609         kvm_desct->base = get_desc_base(seg_desc);
4610         kvm_desct->limit = get_desc_limit(seg_desc);
4611         if (seg_desc->g) {
4612                 kvm_desct->limit <<= 12;
4613                 kvm_desct->limit |= 0xfff;
4614         }
4615         kvm_desct->selector = selector;
4616         kvm_desct->type = seg_desc->type;
4617         kvm_desct->present = seg_desc->p;
4618         kvm_desct->dpl = seg_desc->dpl;
4619         kvm_desct->db = seg_desc->d;
4620         kvm_desct->s = seg_desc->s;
4621         kvm_desct->l = seg_desc->l;
4622         kvm_desct->g = seg_desc->g;
4623         kvm_desct->avl = seg_desc->avl;
4624         if (!selector)
4625                 kvm_desct->unusable = 1;
4626         else
4627                 kvm_desct->unusable = 0;
4628         kvm_desct->padding = 0;
4629 }
4630
4631 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4632                                           u16 selector,
4633                                           struct descriptor_table *dtable)
4634 {
4635         if (selector & 1 << 2) {
4636                 struct kvm_segment kvm_seg;
4637
4638                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4639
4640                 if (kvm_seg.unusable)
4641                         dtable->limit = 0;
4642                 else
4643                         dtable->limit = kvm_seg.limit;
4644                 dtable->base = kvm_seg.base;
4645         }
4646         else
4647                 kvm_x86_ops->get_gdt(vcpu, dtable);
4648 }
4649
4650 /* allowed just for 8 bytes segments */
4651 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4652                                          struct desc_struct *seg_desc)
4653 {
4654         struct descriptor_table dtable;
4655         u16 index = selector >> 3;
4656
4657         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4658
4659         if (dtable.limit < index * 8 + 7) {
4660                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4661                 return X86EMUL_PROPAGATE_FAULT;
4662         }
4663         return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4664 }
4665
4666 /* allowed just for 8 bytes segments */
4667 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4668                                          struct desc_struct *seg_desc)
4669 {
4670         struct descriptor_table dtable;
4671         u16 index = selector >> 3;
4672
4673         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4674
4675         if (dtable.limit < index * 8 + 7)
4676                 return 1;
4677         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4678 }
4679
4680 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4681                              struct desc_struct *seg_desc)
4682 {
4683         u32 base_addr = get_desc_base(seg_desc);
4684
4685         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4686 }
4687
4688 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4689 {
4690         struct kvm_segment kvm_seg;
4691
4692         kvm_get_segment(vcpu, &kvm_seg, seg);
4693         return kvm_seg.selector;
4694 }
4695
4696 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4697 {
4698         struct kvm_segment segvar = {
4699                 .base = selector << 4,
4700                 .limit = 0xffff,
4701                 .selector = selector,
4702                 .type = 3,
4703                 .present = 1,
4704                 .dpl = 3,
4705                 .db = 0,
4706                 .s = 1,
4707                 .l = 0,
4708                 .g = 0,
4709                 .avl = 0,
4710                 .unusable = 0,
4711         };
4712         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4713         return 0;
4714 }
4715
4716 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4717 {
4718         return (seg != VCPU_SREG_LDTR) &&
4719                 (seg != VCPU_SREG_TR) &&
4720                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4721 }
4722
4723 static void kvm_check_segment_descriptor(struct kvm_vcpu *vcpu, int seg,
4724                                          u16 selector)
4725 {
4726         /* NULL selector is not valid for CS and SS */
4727         if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4728                 if (!selector)
4729                         kvm_queue_exception_e(vcpu, TS_VECTOR, selector >> 3);
4730 }
4731
4732 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4733                                 int type_bits, int seg)
4734 {
4735         struct kvm_segment kvm_seg;
4736         struct desc_struct seg_desc;
4737
4738         if (is_vm86_segment(vcpu, seg) || !is_protmode(vcpu))
4739                 return kvm_load_realmode_segment(vcpu, selector, seg);
4740
4741         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4742                 return 1;
4743         seg_desct_to_kvm_desct(&seg_desc, selector, &kvm_seg);
4744
4745         kvm_check_segment_descriptor(vcpu, seg, selector);
4746         kvm_seg.type |= type_bits;
4747
4748         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4749             seg != VCPU_SREG_LDTR)
4750                 if (!kvm_seg.s)
4751                         kvm_seg.unusable = 1;
4752
4753         kvm_set_segment(vcpu, &kvm_seg, seg);
4754         if (selector && !kvm_seg.unusable && kvm_seg.s) {
4755                 /* mark segment as accessed */
4756                 seg_desc.type |= 1;
4757                 save_guest_segment_descriptor(vcpu, selector, &seg_desc);
4758         }
4759         return 0;
4760 }
4761
4762 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4763                                 struct tss_segment_32 *tss)
4764 {
4765         tss->cr3 = vcpu->arch.cr3;
4766         tss->eip = kvm_rip_read(vcpu);
4767         tss->eflags = kvm_get_rflags(vcpu);
4768         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4769         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4770         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4771         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4772         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4773         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4774         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4775         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4776         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4777         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4778         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4779         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4780         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4781         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4782         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4783 }
4784
4785 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4786                                   struct tss_segment_32 *tss)
4787 {
4788         kvm_set_cr3(vcpu, tss->cr3);
4789
4790         kvm_rip_write(vcpu, tss->eip);
4791         kvm_set_rflags(vcpu, tss->eflags | 2);
4792
4793         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4794         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4795         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4796         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4797         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4798         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4799         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4800         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4801
4802         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4803                 return 1;
4804
4805         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4806                 return 1;
4807
4808         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4809                 return 1;
4810
4811         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4812                 return 1;
4813
4814         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4815                 return 1;
4816
4817         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4818                 return 1;
4819
4820         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4821                 return 1;
4822         return 0;
4823 }
4824
4825 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4826                                 struct tss_segment_16 *tss)
4827 {
4828         tss->ip = kvm_rip_read(vcpu);
4829         tss->flag = kvm_get_rflags(vcpu);
4830         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4831         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4832         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4833         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4834         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4835         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4836         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4837         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4838
4839         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4840         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4841         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4842         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4843         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4844 }
4845
4846 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4847                                  struct tss_segment_16 *tss)
4848 {
4849         kvm_rip_write(vcpu, tss->ip);
4850         kvm_set_rflags(vcpu, tss->flag | 2);
4851         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4852         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4853         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4854         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4855         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4856         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4857         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4858         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4859
4860         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4861                 return 1;
4862
4863         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4864                 return 1;
4865
4866         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4867                 return 1;
4868
4869         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4870                 return 1;
4871
4872         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4873                 return 1;
4874         return 0;
4875 }
4876
4877 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4878                               u16 old_tss_sel, u32 old_tss_base,
4879                               struct desc_struct *nseg_desc)
4880 {
4881         struct tss_segment_16 tss_segment_16;
4882         int ret = 0;
4883
4884         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4885                            sizeof tss_segment_16))
4886                 goto out;
4887
4888         save_state_to_tss16(vcpu, &tss_segment_16);
4889
4890         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4891                             sizeof tss_segment_16))
4892                 goto out;
4893
4894         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4895                            &tss_segment_16, sizeof tss_segment_16))
4896                 goto out;
4897
4898         if (old_tss_sel != 0xffff) {
4899                 tss_segment_16.prev_task_link = old_tss_sel;
4900
4901                 if (kvm_write_guest(vcpu->kvm,
4902                                     get_tss_base_addr(vcpu, nseg_desc),
4903                                     &tss_segment_16.prev_task_link,
4904                                     sizeof tss_segment_16.prev_task_link))
4905                         goto out;
4906         }
4907
4908         if (load_state_from_tss16(vcpu, &tss_segment_16))
4909                 goto out;
4910
4911         ret = 1;
4912 out:
4913         return ret;
4914 }
4915
4916 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4917                        u16 old_tss_sel, u32 old_tss_base,
4918                        struct desc_struct *nseg_desc)
4919 {
4920         struct tss_segment_32 tss_segment_32;
4921         int ret = 0;
4922
4923         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4924                            sizeof tss_segment_32))
4925                 goto out;
4926
4927         save_state_to_tss32(vcpu, &tss_segment_32);
4928
4929         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4930                             sizeof tss_segment_32))
4931                 goto out;
4932
4933         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4934                            &tss_segment_32, sizeof tss_segment_32))
4935                 goto out;
4936
4937         if (old_tss_sel != 0xffff) {
4938                 tss_segment_32.prev_task_link = old_tss_sel;
4939
4940                 if (kvm_write_guest(vcpu->kvm,
4941                                     get_tss_base_addr(vcpu, nseg_desc),
4942                                     &tss_segment_32.prev_task_link,
4943                                     sizeof tss_segment_32.prev_task_link))
4944                         goto out;
4945         }
4946
4947         if (load_state_from_tss32(vcpu, &tss_segment_32))
4948                 goto out;
4949
4950         ret = 1;
4951 out:
4952         return ret;
4953 }
4954
4955 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4956 {
4957         struct kvm_segment tr_seg;
4958         struct desc_struct cseg_desc;
4959         struct desc_struct nseg_desc;
4960         int ret = 0;
4961         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4962         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4963
4964         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4965
4966         /* FIXME: Handle errors. Failure to read either TSS or their
4967          * descriptors should generate a pagefault.
4968          */
4969         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4970                 goto out;
4971
4972         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4973                 goto out;
4974
4975         if (reason != TASK_SWITCH_IRET) {
4976                 int cpl;
4977
4978                 cpl = kvm_x86_ops->get_cpl(vcpu);
4979                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4980                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4981                         return 1;
4982                 }
4983         }
4984
4985         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4986                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4987                 return 1;
4988         }
4989
4990         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4991                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4992                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4993         }
4994
4995         if (reason == TASK_SWITCH_IRET) {
4996                 u32 eflags = kvm_get_rflags(vcpu);
4997                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4998         }
4999
5000         /* set back link to prev task only if NT bit is set in eflags
5001            note that old_tss_sel is not used afetr this point */
5002         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
5003                 old_tss_sel = 0xffff;
5004
5005         if (nseg_desc.type & 8)
5006                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
5007                                          old_tss_base, &nseg_desc);
5008         else
5009                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
5010                                          old_tss_base, &nseg_desc);
5011
5012         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
5013                 u32 eflags = kvm_get_rflags(vcpu);
5014                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
5015         }
5016
5017         if (reason != TASK_SWITCH_IRET) {
5018                 nseg_desc.type |= (1 << 1);
5019                 save_guest_segment_descriptor(vcpu, tss_selector,
5020                                               &nseg_desc);
5021         }
5022
5023         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5024         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5025         tr_seg.type = 11;
5026         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5027 out:
5028         return ret;
5029 }
5030 EXPORT_SYMBOL_GPL(kvm_task_switch);
5031
5032 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5033                                   struct kvm_sregs *sregs)
5034 {
5035         int mmu_reset_needed = 0;
5036         int pending_vec, max_bits;
5037         struct descriptor_table dt;
5038
5039         vcpu_load(vcpu);
5040
5041         dt.limit = sregs->idt.limit;
5042         dt.base = sregs->idt.base;
5043         kvm_x86_ops->set_idt(vcpu, &dt);
5044         dt.limit = sregs->gdt.limit;
5045         dt.base = sregs->gdt.base;
5046         kvm_x86_ops->set_gdt(vcpu, &dt);
5047
5048         vcpu->arch.cr2 = sregs->cr2;
5049         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5050         vcpu->arch.cr3 = sregs->cr3;
5051
5052         kvm_set_cr8(vcpu, sregs->cr8);
5053
5054         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5055         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5056         kvm_set_apic_base(vcpu, sregs->apic_base);
5057
5058         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5059         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5060         vcpu->arch.cr0 = sregs->cr0;
5061
5062         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5063         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5064         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5065                 load_pdptrs(vcpu, vcpu->arch.cr3);
5066                 mmu_reset_needed = 1;
5067         }
5068
5069         if (mmu_reset_needed)
5070                 kvm_mmu_reset_context(vcpu);
5071
5072         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5073         pending_vec = find_first_bit(
5074                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5075         if (pending_vec < max_bits) {
5076                 kvm_queue_interrupt(vcpu, pending_vec, false);
5077                 pr_debug("Set back pending irq %d\n", pending_vec);
5078                 if (irqchip_in_kernel(vcpu->kvm))
5079                         kvm_pic_clear_isr_ack(vcpu->kvm);
5080         }
5081
5082         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5083         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5084         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5085         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5086         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5087         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5088
5089         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5090         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5091
5092         update_cr8_intercept(vcpu);
5093
5094         /* Older userspace won't unhalt the vcpu on reset. */
5095         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5096             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5097             !is_protmode(vcpu))
5098                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5099
5100         vcpu_put(vcpu);
5101
5102         return 0;
5103 }
5104
5105 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5106                                         struct kvm_guest_debug *dbg)
5107 {
5108         unsigned long rflags;
5109         int i, r;
5110
5111         vcpu_load(vcpu);
5112
5113         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5114                 r = -EBUSY;
5115                 if (vcpu->arch.exception.pending)
5116                         goto unlock_out;
5117                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5118                         kvm_queue_exception(vcpu, DB_VECTOR);
5119                 else
5120                         kvm_queue_exception(vcpu, BP_VECTOR);
5121         }
5122
5123         /*
5124          * Read rflags as long as potentially injected trace flags are still
5125          * filtered out.
5126          */
5127         rflags = kvm_get_rflags(vcpu);
5128
5129         vcpu->guest_debug = dbg->control;
5130         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5131                 vcpu->guest_debug = 0;
5132
5133         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5134                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5135                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5136                 vcpu->arch.switch_db_regs =
5137                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5138         } else {
5139                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5140                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5141                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5142         }
5143
5144         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5145                 vcpu->arch.singlestep_cs =
5146                         get_segment_selector(vcpu, VCPU_SREG_CS);
5147                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
5148         }
5149
5150         /*
5151          * Trigger an rflags update that will inject or remove the trace
5152          * flags.
5153          */
5154         kvm_set_rflags(vcpu, rflags);
5155
5156         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5157
5158         r = 0;
5159
5160 unlock_out:
5161         vcpu_put(vcpu);
5162
5163         return r;
5164 }
5165
5166 /*
5167  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
5168  * we have asm/x86/processor.h
5169  */
5170 struct fxsave {
5171         u16     cwd;
5172         u16     swd;
5173         u16     twd;
5174         u16     fop;
5175         u64     rip;
5176         u64     rdp;
5177         u32     mxcsr;
5178         u32     mxcsr_mask;
5179         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
5180 #ifdef CONFIG_X86_64
5181         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
5182 #else
5183         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
5184 #endif
5185 };
5186
5187 /*
5188  * Translate a guest virtual address to a guest physical address.
5189  */
5190 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5191                                     struct kvm_translation *tr)
5192 {
5193         unsigned long vaddr = tr->linear_address;
5194         gpa_t gpa;
5195         int idx;
5196
5197         vcpu_load(vcpu);
5198         idx = srcu_read_lock(&vcpu->kvm->srcu);
5199         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
5200         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5201         tr->physical_address = gpa;
5202         tr->valid = gpa != UNMAPPED_GVA;
5203         tr->writeable = 1;
5204         tr->usermode = 0;
5205         vcpu_put(vcpu);
5206
5207         return 0;
5208 }
5209
5210 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5211 {
5212         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5213
5214         vcpu_load(vcpu);
5215
5216         memcpy(fpu->fpr, fxsave->st_space, 128);
5217         fpu->fcw = fxsave->cwd;
5218         fpu->fsw = fxsave->swd;
5219         fpu->ftwx = fxsave->twd;
5220         fpu->last_opcode = fxsave->fop;
5221         fpu->last_ip = fxsave->rip;
5222         fpu->last_dp = fxsave->rdp;
5223         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5224
5225         vcpu_put(vcpu);
5226
5227         return 0;
5228 }
5229
5230 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5231 {
5232         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5233
5234         vcpu_load(vcpu);
5235
5236         memcpy(fxsave->st_space, fpu->fpr, 128);
5237         fxsave->cwd = fpu->fcw;
5238         fxsave->swd = fpu->fsw;
5239         fxsave->twd = fpu->ftwx;
5240         fxsave->fop = fpu->last_opcode;
5241         fxsave->rip = fpu->last_ip;
5242         fxsave->rdp = fpu->last_dp;
5243         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5244
5245         vcpu_put(vcpu);
5246
5247         return 0;
5248 }
5249
5250 void fx_init(struct kvm_vcpu *vcpu)
5251 {
5252         unsigned after_mxcsr_mask;
5253
5254         /*
5255          * Touch the fpu the first time in non atomic context as if
5256          * this is the first fpu instruction the exception handler
5257          * will fire before the instruction returns and it'll have to
5258          * allocate ram with GFP_KERNEL.
5259          */
5260         if (!used_math())
5261                 kvm_fx_save(&vcpu->arch.host_fx_image);
5262
5263         /* Initialize guest FPU by resetting ours and saving into guest's */
5264         preempt_disable();
5265         kvm_fx_save(&vcpu->arch.host_fx_image);
5266         kvm_fx_finit();
5267         kvm_fx_save(&vcpu->arch.guest_fx_image);
5268         kvm_fx_restore(&vcpu->arch.host_fx_image);
5269         preempt_enable();
5270
5271         vcpu->arch.cr0 |= X86_CR0_ET;
5272         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5273         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5274         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5275                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5276 }
5277 EXPORT_SYMBOL_GPL(fx_init);
5278
5279 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5280 {
5281         if (vcpu->guest_fpu_loaded)
5282                 return;
5283
5284         vcpu->guest_fpu_loaded = 1;
5285         kvm_fx_save(&vcpu->arch.host_fx_image);
5286         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5287         trace_kvm_fpu(1);
5288 }
5289
5290 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5291 {
5292         if (!vcpu->guest_fpu_loaded)
5293                 return;
5294
5295         vcpu->guest_fpu_loaded = 0;
5296         kvm_fx_save(&vcpu->arch.guest_fx_image);
5297         kvm_fx_restore(&vcpu->arch.host_fx_image);
5298         ++vcpu->stat.fpu_reload;
5299         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5300         trace_kvm_fpu(0);
5301 }
5302
5303 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5304 {
5305         if (vcpu->arch.time_page) {
5306                 kvm_release_page_dirty(vcpu->arch.time_page);
5307                 vcpu->arch.time_page = NULL;
5308         }
5309
5310         kvm_x86_ops->vcpu_free(vcpu);
5311 }
5312
5313 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5314                                                 unsigned int id)
5315 {
5316         return kvm_x86_ops->vcpu_create(kvm, id);
5317 }
5318
5319 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5320 {
5321         int r;
5322
5323         /* We do fxsave: this must be aligned. */
5324         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5325
5326         vcpu->arch.mtrr_state.have_fixed = 1;
5327         vcpu_load(vcpu);
5328         r = kvm_arch_vcpu_reset(vcpu);
5329         if (r == 0)
5330                 r = kvm_mmu_setup(vcpu);
5331         vcpu_put(vcpu);
5332         if (r < 0)
5333                 goto free_vcpu;
5334
5335         return 0;
5336 free_vcpu:
5337         kvm_x86_ops->vcpu_free(vcpu);
5338         return r;
5339 }
5340
5341 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5342 {
5343         vcpu_load(vcpu);
5344         kvm_mmu_unload(vcpu);
5345         vcpu_put(vcpu);
5346
5347         kvm_x86_ops->vcpu_free(vcpu);
5348 }
5349
5350 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5351 {
5352         vcpu->arch.nmi_pending = false;
5353         vcpu->arch.nmi_injected = false;
5354
5355         vcpu->arch.switch_db_regs = 0;
5356         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5357         vcpu->arch.dr6 = DR6_FIXED_1;
5358         vcpu->arch.dr7 = DR7_FIXED_1;
5359
5360         return kvm_x86_ops->vcpu_reset(vcpu);
5361 }
5362
5363 int kvm_arch_hardware_enable(void *garbage)
5364 {
5365         /*
5366          * Since this may be called from a hotplug notifcation,
5367          * we can't get the CPU frequency directly.
5368          */
5369         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5370                 int cpu = raw_smp_processor_id();
5371                 per_cpu(cpu_tsc_khz, cpu) = 0;
5372         }
5373
5374         kvm_shared_msr_cpu_online();
5375
5376         return kvm_x86_ops->hardware_enable(garbage);
5377 }
5378
5379 void kvm_arch_hardware_disable(void *garbage)
5380 {
5381         kvm_x86_ops->hardware_disable(garbage);
5382         drop_user_return_notifiers(garbage);
5383 }
5384
5385 int kvm_arch_hardware_setup(void)
5386 {
5387         return kvm_x86_ops->hardware_setup();
5388 }
5389
5390 void kvm_arch_hardware_unsetup(void)
5391 {
5392         kvm_x86_ops->hardware_unsetup();
5393 }
5394
5395 void kvm_arch_check_processor_compat(void *rtn)
5396 {
5397         kvm_x86_ops->check_processor_compatibility(rtn);
5398 }
5399
5400 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5401 {
5402         struct page *page;
5403         struct kvm *kvm;
5404         int r;
5405
5406         BUG_ON(vcpu->kvm == NULL);
5407         kvm = vcpu->kvm;
5408
5409         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5410         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5411                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5412         else
5413                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5414
5415         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5416         if (!page) {
5417                 r = -ENOMEM;
5418                 goto fail;
5419         }
5420         vcpu->arch.pio_data = page_address(page);
5421
5422         r = kvm_mmu_create(vcpu);
5423         if (r < 0)
5424                 goto fail_free_pio_data;
5425
5426         if (irqchip_in_kernel(kvm)) {
5427                 r = kvm_create_lapic(vcpu);
5428                 if (r < 0)
5429                         goto fail_mmu_destroy;
5430         }
5431
5432         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5433                                        GFP_KERNEL);
5434         if (!vcpu->arch.mce_banks) {
5435                 r = -ENOMEM;
5436                 goto fail_free_lapic;
5437         }
5438         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5439
5440         return 0;
5441 fail_free_lapic:
5442         kvm_free_lapic(vcpu);
5443 fail_mmu_destroy:
5444         kvm_mmu_destroy(vcpu);
5445 fail_free_pio_data:
5446         free_page((unsigned long)vcpu->arch.pio_data);
5447 fail:
5448         return r;
5449 }
5450
5451 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5452 {
5453         int idx;
5454
5455         kfree(vcpu->arch.mce_banks);
5456         kvm_free_lapic(vcpu);
5457         idx = srcu_read_lock(&vcpu->kvm->srcu);
5458         kvm_mmu_destroy(vcpu);
5459         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5460         free_page((unsigned long)vcpu->arch.pio_data);
5461 }
5462
5463 struct  kvm *kvm_arch_create_vm(void)
5464 {
5465         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5466
5467         if (!kvm)
5468                 return ERR_PTR(-ENOMEM);
5469
5470         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5471         if (!kvm->arch.aliases) {
5472                 kfree(kvm);
5473                 return ERR_PTR(-ENOMEM);
5474         }
5475
5476         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5477         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5478
5479         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5480         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5481
5482         rdtscll(kvm->arch.vm_init_tsc);
5483
5484         return kvm;
5485 }
5486
5487 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5488 {
5489         vcpu_load(vcpu);
5490         kvm_mmu_unload(vcpu);
5491         vcpu_put(vcpu);
5492 }
5493
5494 static void kvm_free_vcpus(struct kvm *kvm)
5495 {
5496         unsigned int i;
5497         struct kvm_vcpu *vcpu;
5498
5499         /*
5500          * Unpin any mmu pages first.
5501          */
5502         kvm_for_each_vcpu(i, vcpu, kvm)
5503                 kvm_unload_vcpu_mmu(vcpu);
5504         kvm_for_each_vcpu(i, vcpu, kvm)
5505                 kvm_arch_vcpu_free(vcpu);
5506
5507         mutex_lock(&kvm->lock);
5508         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5509                 kvm->vcpus[i] = NULL;
5510
5511         atomic_set(&kvm->online_vcpus, 0);
5512         mutex_unlock(&kvm->lock);
5513 }
5514
5515 void kvm_arch_sync_events(struct kvm *kvm)
5516 {
5517         kvm_free_all_assigned_devices(kvm);
5518 }
5519
5520 void kvm_arch_destroy_vm(struct kvm *kvm)
5521 {
5522         kvm_iommu_unmap_guest(kvm);
5523         kvm_free_pit(kvm);
5524         kfree(kvm->arch.vpic);
5525         kfree(kvm->arch.vioapic);
5526         kvm_free_vcpus(kvm);
5527         kvm_free_physmem(kvm);
5528         if (kvm->arch.apic_access_page)
5529                 put_page(kvm->arch.apic_access_page);
5530         if (kvm->arch.ept_identity_pagetable)
5531                 put_page(kvm->arch.ept_identity_pagetable);
5532         cleanup_srcu_struct(&kvm->srcu);
5533         kfree(kvm->arch.aliases);
5534         kfree(kvm);
5535 }
5536
5537 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5538                                 struct kvm_memory_slot *memslot,
5539                                 struct kvm_memory_slot old,
5540                                 struct kvm_userspace_memory_region *mem,
5541                                 int user_alloc)
5542 {
5543         int npages = memslot->npages;
5544
5545         /*To keep backward compatibility with older userspace,
5546          *x86 needs to hanlde !user_alloc case.
5547          */
5548         if (!user_alloc) {
5549                 if (npages && !old.rmap) {
5550                         unsigned long userspace_addr;
5551
5552                         down_write(&current->mm->mmap_sem);
5553                         userspace_addr = do_mmap(NULL, 0,
5554                                                  npages * PAGE_SIZE,
5555                                                  PROT_READ | PROT_WRITE,
5556                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5557                                                  0);
5558                         up_write(&current->mm->mmap_sem);
5559
5560                         if (IS_ERR((void *)userspace_addr))
5561                                 return PTR_ERR((void *)userspace_addr);
5562
5563                         memslot->userspace_addr = userspace_addr;
5564                 }
5565         }
5566
5567
5568         return 0;
5569 }
5570
5571 void kvm_arch_commit_memory_region(struct kvm *kvm,
5572                                 struct kvm_userspace_memory_region *mem,
5573                                 struct kvm_memory_slot old,
5574                                 int user_alloc)
5575 {
5576
5577         int npages = mem->memory_size >> PAGE_SHIFT;
5578
5579         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5580                 int ret;
5581
5582                 down_write(&current->mm->mmap_sem);
5583                 ret = do_munmap(current->mm, old.userspace_addr,
5584                                 old.npages * PAGE_SIZE);
5585                 up_write(&current->mm->mmap_sem);
5586                 if (ret < 0)
5587                         printk(KERN_WARNING
5588                                "kvm_vm_ioctl_set_memory_region: "
5589                                "failed to munmap memory\n");
5590         }
5591
5592         spin_lock(&kvm->mmu_lock);
5593         if (!kvm->arch.n_requested_mmu_pages) {
5594                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5595                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5596         }
5597
5598         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5599         spin_unlock(&kvm->mmu_lock);
5600 }
5601
5602 void kvm_arch_flush_shadow(struct kvm *kvm)
5603 {
5604         kvm_mmu_zap_all(kvm);
5605         kvm_reload_remote_mmus(kvm);
5606 }
5607
5608 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5609 {
5610         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5611                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5612                 || vcpu->arch.nmi_pending ||
5613                 (kvm_arch_interrupt_allowed(vcpu) &&
5614                  kvm_cpu_has_interrupt(vcpu));
5615 }
5616
5617 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5618 {
5619         int me;
5620         int cpu = vcpu->cpu;
5621
5622         if (waitqueue_active(&vcpu->wq)) {
5623                 wake_up_interruptible(&vcpu->wq);
5624                 ++vcpu->stat.halt_wakeup;
5625         }
5626
5627         me = get_cpu();
5628         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5629                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5630                         smp_send_reschedule(cpu);
5631         put_cpu();
5632 }
5633
5634 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5635 {
5636         return kvm_x86_ops->interrupt_allowed(vcpu);
5637 }
5638
5639 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5640 {
5641         unsigned long rflags;
5642
5643         rflags = kvm_x86_ops->get_rflags(vcpu);
5644         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5645                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5646         return rflags;
5647 }
5648 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5649
5650 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5651 {
5652         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5653             vcpu->arch.singlestep_cs ==
5654                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5655             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5656                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5657         kvm_x86_ops->set_rflags(vcpu, rflags);
5658 }
5659 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5660
5661 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5662 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5663 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5664 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5665 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5666 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5667 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5668 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5669 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5670 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5671 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);