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         if (cr0 & CR0_RESERVED_BITS) {
 434                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 435                        cr0, kvm_read_cr0(vcpu));
 436                 kvm_inject_gp(vcpu, 0);
 437                 return;
 438         }
 439
 440         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 441                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 442                 kvm_inject_gp(vcpu, 0);
 443                 return;
 444         }
 445
 446         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 447                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 448                        "and a clear PE flag\n");
 449                 kvm_inject_gp(vcpu, 0);
 450                 return;
 451         }
 452
 453         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 454 #ifdef CONFIG_X86_64
 455                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 456                         int cs_db, cs_l;
 457
 458                         if (!is_pae(vcpu)) {
 459                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 460                                        "in long mode while PAE is disabled\n");
 461                                 kvm_inject_gp(vcpu, 0);
 462                                 return;
 463                         }
 464                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 465                         if (cs_l) {
 466                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 467                                        "in long mode while CS.L == 1\n");
 468                                 kvm_inject_gp(vcpu, 0);
 469                                 return;
 470
 471                         }
 472                 } else
 473 #endif
 474                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 475                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 476                                "reserved bits\n");
 477                         kvm_inject_gp(vcpu, 0);
 478                         return;
 479                 }
 480
 481         }
 482
 483         kvm_x86_ops->set_cr0(vcpu, cr0);
 484         vcpu->arch.cr0 = cr0;
 485
 486         kvm_mmu_reset_context(vcpu);
 487         return;
 488 }
 489 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 490
 491 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 492 {
 493         kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
 494 }
 495 EXPORT_SYMBOL_GPL(kvm_lmsw);
 496
 497 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 498 {
 499         unsigned long old_cr4 = kvm_read_cr4(vcpu);
 500         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 501
 502         if (cr4 & CR4_RESERVED_BITS) {
 503                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 504                 kvm_inject_gp(vcpu, 0);
 505                 return;
 506         }
 507
 508         if (is_long_mode(vcpu)) {
 509                 if (!(cr4 & X86_CR4_PAE)) {
 510                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 511                                "in long mode\n");
 512                         kvm_inject_gp(vcpu, 0);
 513                         return;
 514                 }
 515         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 516                    && ((cr4 ^ old_cr4) & pdptr_bits)
 517                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 518                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 519                 kvm_inject_gp(vcpu, 0);
 520                 return;
 521         }
 522
 523         if (cr4 & X86_CR4_VMXE) {
 524                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 525                 kvm_inject_gp(vcpu, 0);
 526                 return;
 527         }
 528         kvm_x86_ops->set_cr4(vcpu, cr4);
 529         vcpu->arch.cr4 = cr4;
 530         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 531         kvm_mmu_reset_context(vcpu);
 532 }
 533 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 534
 535 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 536 {
 537         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 538                 kvm_mmu_sync_roots(vcpu);
 539                 kvm_mmu_flush_tlb(vcpu);
 540                 return;
 541         }
 542
 543         if (is_long_mode(vcpu)) {
 544                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 545                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 546                         kvm_inject_gp(vcpu, 0);
 547                         return;
 548                 }
 549         } else {
 550                 if (is_pae(vcpu)) {
 551                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 552                                 printk(KERN_DEBUG
 553                                        "set_cr3: #GP, reserved bits\n");
 554                                 kvm_inject_gp(vcpu, 0);
 555                                 return;
 556                         }
 557                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 558                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 559                                        "reserved bits\n");
 560                                 kvm_inject_gp(vcpu, 0);
 561                                 return;
 562                         }
 563                 }
 564                 /*
 565                  * We don't check reserved bits in nonpae mode, because
 566                  * this isn't enforced, and VMware depends on this.
 567                  */
 568         }
 569
 570         /*
 571          * Does the new cr3 value map to physical memory? (Note, we
 572          * catch an invalid cr3 even in real-mode, because it would
 573          * cause trouble later on when we turn on paging anyway.)
 574          *
 575          * A real CPU would silently accept an invalid cr3 and would
 576          * attempt to use it - with largely undefined (and often hard
 577          * to debug) behavior on the guest side.
 578          */
 579         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 580                 kvm_inject_gp(vcpu, 0);
 581         else {
 582                 vcpu->arch.cr3 = cr3;
 583                 vcpu->arch.mmu.new_cr3(vcpu);
 584         }
 585 }
 586 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 587
 588 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 589 {
 590         if (cr8 & CR8_RESERVED_BITS) {
 591                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 592                 kvm_inject_gp(vcpu, 0);
 593                 return;
 594         }
 595         if (irqchip_in_kernel(vcpu->kvm))
 596                 kvm_lapic_set_tpr(vcpu, cr8);
 597         else
 598                 vcpu->arch.cr8 = cr8;
 599 }
 600 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 601
 602 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 603 {
 604         if (irqchip_in_kernel(vcpu->kvm))
 605                 return kvm_lapic_get_cr8(vcpu);
 606         else
 607                 return vcpu->arch.cr8;
 608 }
 609 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 610
 611 static inline u32 bit(int bitno)
 612 {
 613         return 1 << (bitno & 31);
 614 }
 615
 616 /*
 617  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 618  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 619  *
 620  * This list is modified at module load time to reflect the
 621  * capabilities of the host cpu. This capabilities test skips MSRs that are
 622  * kvm-specific. Those are put in the beginning of the list.
 623  */
 624
 625 #define KVM_SAVE_MSRS_BEGIN     2
 626 static u32 msrs_to_save[] = {
 627         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 628         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 629         MSR_K6_STAR,
 630 #ifdef CONFIG_X86_64
 631         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 632 #endif
 633         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 634 };
 635
 636 static unsigned num_msrs_to_save;
 637
 638 static u32 emulated_msrs[] = {
 639         MSR_IA32_MISC_ENABLE,
 640 };
 641
 642 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 643 {
 644         if (efer & efer_reserved_bits) {
 645                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 646                        efer);
 647                 kvm_inject_gp(vcpu, 0);
 648                 return;
 649         }
 650
 651         if (is_paging(vcpu)
 652             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 653                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 654                 kvm_inject_gp(vcpu, 0);
 655                 return;
 656         }
 657
 658         if (efer & EFER_FFXSR) {
 659                 struct kvm_cpuid_entry2 *feat;
 660
 661                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 662                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 663                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 664                         kvm_inject_gp(vcpu, 0);
 665                         return;
 666                 }
 667         }
 668
 669         if (efer & EFER_SVME) {
 670                 struct kvm_cpuid_entry2 *feat;
 671
 672                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 673                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 674                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 675                         kvm_inject_gp(vcpu, 0);
 676                         return;
 677                 }
 678         }
 679
 680         kvm_x86_ops->set_efer(vcpu, efer);
 681
 682         efer &= ~EFER_LMA;
 683         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 684
 685         vcpu->arch.shadow_efer = efer;
 686
 687         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 688         kvm_mmu_reset_context(vcpu);
 689 }
 690
 691 void kvm_enable_efer_bits(u64 mask)
 692 {
 693        efer_reserved_bits &= ~mask;
 694 }
 695 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 696
 697
 698 /*
 699  * Writes msr value into into the appropriate "register".
 700  * Returns 0 on success, non-0 otherwise.
 701  * Assumes vcpu_load() was already called.
 702  */
 703 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 704 {
 705         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 706 }
 707
 708 /*
 709  * Adapt set_msr() to msr_io()'s calling convention
 710  */
 711 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 712 {
 713         return kvm_set_msr(vcpu, index, *data);
 714 }
 715
 716 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 717 {
 718         static int version;
 719         struct pvclock_wall_clock wc;
 720         struct timespec boot;
 721
 722         if (!wall_clock)
 723                 return;
 724
 725         version++;
 726
 727         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 728
 729         /*
 730          * The guest calculates current wall clock time by adding
 731          * system time (updated by kvm_write_guest_time below) to the
 732          * wall clock specified here.  guest system time equals host
 733          * system time for us, thus we must fill in host boot time here.
 734          */
 735         getboottime(&boot);
 736
 737         wc.sec = boot.tv_sec;
 738         wc.nsec = boot.tv_nsec;
 739         wc.version = version;
 740
 741         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 742
 743         version++;
 744         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 745 }
 746
 747 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 748 {
 749         uint32_t quotient, remainder;
 750
 751         /* Don't try to replace with do_div(), this one calculates
 752          * "(dividend << 32) / divisor" */
 753         __asm__ ( "divl %4"
 754                   : "=a" (quotient), "=d" (remainder)
 755                   : "0" (0), "1" (dividend), "r" (divisor) );
 756         return quotient;
 757 }
 758
 759 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 760 {
 761         uint64_t nsecs = 1000000000LL;
 762         int32_t  shift = 0;
 763         uint64_t tps64;
 764         uint32_t tps32;
 765
 766         tps64 = tsc_khz * 1000LL;
 767         while (tps64 > nsecs*2) {
 768                 tps64 >>= 1;
 769                 shift--;
 770         }
 771
 772         tps32 = (uint32_t)tps64;
 773         while (tps32 <= (uint32_t)nsecs) {
 774                 tps32 <<= 1;
 775                 shift++;
 776         }
 777
 778         hv_clock->tsc_shift = shift;
 779         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 780
 781         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 782                  __func__, tsc_khz, hv_clock->tsc_shift,
 783                  hv_clock->tsc_to_system_mul);
 784 }
 785
 786 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 787
 788 static void kvm_write_guest_time(struct kvm_vcpu *v)
 789 {
 790         struct timespec ts;
 791         unsigned long flags;
 792         struct kvm_vcpu_arch *vcpu = &v->arch;
 793         void *shared_kaddr;
 794         unsigned long this_tsc_khz;
 795
 796         if ((!vcpu->time_page))
 797                 return;
 798
 799         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 800         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 801                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 802                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 803         }
 804         put_cpu_var(cpu_tsc_khz);
 805
 806         /* Keep irq disabled to prevent changes to the clock */
 807         local_irq_save(flags);
 808         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 809         ktime_get_ts(&ts);
 810         monotonic_to_bootbased(&ts);
 811         local_irq_restore(flags);
 812
 813         /* With all the info we got, fill in the values */
 814
 815         vcpu->hv_clock.system_time = ts.tv_nsec +
 816                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
 817
 818         /*
 819          * The interface expects us to write an even number signaling that the
 820          * update is finished. Since the guest won't see the intermediate
 821          * state, we just increase by 2 at the end.
 822          */
 823         vcpu->hv_clock.version += 2;
 824
 825         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 826
 827         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 828                sizeof(vcpu->hv_clock));
 829
 830         kunmap_atomic(shared_kaddr, KM_USER0);
 831
 832         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 833 }
 834
 835 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 836 {
 837         struct kvm_vcpu_arch *vcpu = &v->arch;
 838
 839         if (!vcpu->time_page)
 840                 return 0;
 841         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 842         return 1;
 843 }
 844
 845 static bool msr_mtrr_valid(unsigned msr)
 846 {
 847         switch (msr) {
 848         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 849         case MSR_MTRRfix64K_00000:
 850         case MSR_MTRRfix16K_80000:
 851         case MSR_MTRRfix16K_A0000:
 852         case MSR_MTRRfix4K_C0000:
 853         case MSR_MTRRfix4K_C8000:
 854         case MSR_MTRRfix4K_D0000:
 855         case MSR_MTRRfix4K_D8000:
 856         case MSR_MTRRfix4K_E0000:
 857         case MSR_MTRRfix4K_E8000:
 858         case MSR_MTRRfix4K_F0000:
 859         case MSR_MTRRfix4K_F8000:
 860         case MSR_MTRRdefType:
 861         case MSR_IA32_CR_PAT:
 862                 return true;
 863         case 0x2f8:
 864                 return true;
 865         }
 866         return false;
 867 }
 868
 869 static bool valid_pat_type(unsigned t)
 870 {
 871         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 872 }
 873
 874 static bool valid_mtrr_type(unsigned t)
 875 {
 876         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 877 }
 878
 879 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 880 {
 881         int i;
 882
 883         if (!msr_mtrr_valid(msr))
 884                 return false;
 885
 886         if (msr == MSR_IA32_CR_PAT) {
 887                 for (i = 0; i < 8; i++)
 888                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 889                                 return false;
 890                 return true;
 891         } else if (msr == MSR_MTRRdefType) {
 892                 if (data & ~0xcff)
 893                         return false;
 894                 return valid_mtrr_type(data & 0xff);
 895         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 896                 for (i = 0; i < 8 ; i++)
 897                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 898                                 return false;
 899                 return true;
 900         }
 901
 902         /* variable MTRRs */
 903         return valid_mtrr_type(data & 0xff);
 904 }
 905
 906 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 907 {
 908         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 909
 910         if (!mtrr_valid(vcpu, msr, data))
 911                 return 1;
 912
 913         if (msr == MSR_MTRRdefType) {
 914                 vcpu->arch.mtrr_state.def_type = data;
 915                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 916         } else if (msr == MSR_MTRRfix64K_00000)
 917                 p[0] = data;
 918         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 919                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 920         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 921                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 922         else if (msr == MSR_IA32_CR_PAT)
 923                 vcpu->arch.pat = data;
 924         else {  /* Variable MTRRs */
 925                 int idx, is_mtrr_mask;
 926                 u64 *pt;
 927
 928                 idx = (msr - 0x200) / 2;
 929                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 930                 if (!is_mtrr_mask)
 931                         pt =
 932                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 933                 else
 934                         pt =
 935                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 936                 *pt = data;
 937         }
 938
 939         kvm_mmu_reset_context(vcpu);
 940         return 0;
 941 }
 942
 943 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 944 {
 945         u64 mcg_cap = vcpu->arch.mcg_cap;
 946         unsigned bank_num = mcg_cap & 0xff;
 947
 948         switch (msr) {
 949         case MSR_IA32_MCG_STATUS:
 950                 vcpu->arch.mcg_status = data;
 951                 break;
 952         case MSR_IA32_MCG_CTL:
 953                 if (!(mcg_cap & MCG_CTL_P))
 954                         return 1;
 955                 if (data != 0 && data != ~(u64)0)
 956                         return -1;
 957                 vcpu->arch.mcg_ctl = data;
 958                 break;
 959         default:
 960                 if (msr >= MSR_IA32_MC0_CTL &&
 961                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 962                         u32 offset = msr - MSR_IA32_MC0_CTL;
 963                         /* only 0 or all 1s can be written to IA32_MCi_CTL */
 964                         if ((offset & 0x3) == 0 &&
 965                             data != 0 && data != ~(u64)0)
 966                                 return -1;
 967                         vcpu->arch.mce_banks[offset] = data;
 968                         break;
 969                 }
 970                 return 1;
 971         }
 972         return 0;
 973 }
 974
 975 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
 976 {
 977         struct kvm *kvm = vcpu->kvm;
 978         int lm = is_long_mode(vcpu);
 979         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
 980                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
 981         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
 982                 : kvm->arch.xen_hvm_config.blob_size_32;
 983         u32 page_num = data & ~PAGE_MASK;
 984         u64 page_addr = data & PAGE_MASK;
 985         u8 *page;
 986         int r;
 987
 988         r = -E2BIG;
 989         if (page_num >= blob_size)
 990                 goto out;
 991         r = -ENOMEM;
 992         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 993         if (!page)
 994                 goto out;
 995         r = -EFAULT;
 996         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
 997                 goto out_free;
 998         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
 999                 goto out_free;
1000         r = 0;
1001 out_free:
1002         kfree(page);
1003 out:
1004         return r;
1005 }
1006
1007 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1008 {
1009         switch (msr) {
1010         case MSR_EFER:
1011                 set_efer(vcpu, data);
1012                 break;
1013         case MSR_K7_HWCR:
1014                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1015                 if (data != 0) {
1016                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1017                                 data);
1018                         return 1;
1019                 }
1020                 break;
1021         case MSR_FAM10H_MMIO_CONF_BASE:
1022                 if (data != 0) {
1023                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1024                                 "0x%llx\n", data);
1025                         return 1;
1026                 }
1027                 break;
1028         case MSR_AMD64_NB_CFG:
1029                 break;
1030         case MSR_IA32_DEBUGCTLMSR:
1031                 if (!data) {
1032                         /* We support the non-activated case already */
1033                         break;
1034                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1035                         /* Values other than LBR and BTF are vendor-specific,
1036                            thus reserved and should throw a #GP */
1037                         return 1;
1038                 }
1039                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1040                         __func__, data);
1041                 break;
1042         case MSR_IA32_UCODE_REV:
1043         case MSR_IA32_UCODE_WRITE:
1044         case MSR_VM_HSAVE_PA:
1045         case MSR_AMD64_PATCH_LOADER:
1046                 break;
1047         case 0x200 ... 0x2ff:
1048                 return set_msr_mtrr(vcpu, msr, data);
1049         case MSR_IA32_APICBASE:
1050                 kvm_set_apic_base(vcpu, data);
1051                 break;
1052         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1053                 return kvm_x2apic_msr_write(vcpu, msr, data);
1054         case MSR_IA32_MISC_ENABLE:
1055                 vcpu->arch.ia32_misc_enable_msr = data;
1056                 break;
1057         case MSR_KVM_WALL_CLOCK:
1058                 vcpu->kvm->arch.wall_clock = data;
1059                 kvm_write_wall_clock(vcpu->kvm, data);
1060                 break;
1061         case MSR_KVM_SYSTEM_TIME: {
1062                 if (vcpu->arch.time_page) {
1063                         kvm_release_page_dirty(vcpu->arch.time_page);
1064                         vcpu->arch.time_page = NULL;
1065                 }
1066
1067                 vcpu->arch.time = data;
1068
1069                 /* we verify if the enable bit is set... */
1070                 if (!(data & 1))
1071                         break;
1072
1073                 /* ...but clean it before doing the actual write */
1074                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1075
1076                 vcpu->arch.time_page =
1077                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1078
1079                 if (is_error_page(vcpu->arch.time_page)) {
1080                         kvm_release_page_clean(vcpu->arch.time_page);
1081                         vcpu->arch.time_page = NULL;
1082                 }
1083
1084                 kvm_request_guest_time_update(vcpu);
1085                 break;
1086         }
1087         case MSR_IA32_MCG_CTL:
1088         case MSR_IA32_MCG_STATUS:
1089         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1090                 return set_msr_mce(vcpu, msr, data);
1091
1092         /* Performance counters are not protected by a CPUID bit,
1093          * so we should check all of them in the generic path for the sake of
1094          * cross vendor migration.
1095          * Writing a zero into the event select MSRs disables them,
1096          * which we perfectly emulate ;-). Any other value should be at least
1097          * reported, some guests depend on them.
1098          */
1099         case MSR_P6_EVNTSEL0:
1100         case MSR_P6_EVNTSEL1:
1101         case MSR_K7_EVNTSEL0:
1102         case MSR_K7_EVNTSEL1:
1103         case MSR_K7_EVNTSEL2:
1104         case MSR_K7_EVNTSEL3:
1105                 if (data != 0)
1106                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1107                                 "0x%x data 0x%llx\n", msr, data);
1108                 break;
1109         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1110          * so we ignore writes to make it happy.
1111          */
1112         case MSR_P6_PERFCTR0:
1113         case MSR_P6_PERFCTR1:
1114         case MSR_K7_PERFCTR0:
1115         case MSR_K7_PERFCTR1:
1116         case MSR_K7_PERFCTR2:
1117         case MSR_K7_PERFCTR3:
1118                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1119                         "0x%x data 0x%llx\n", msr, data);
1120                 break;
1121         default:
1122                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1123                         return xen_hvm_config(vcpu, data);
1124                 if (!ignore_msrs) {
1125                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1126                                 msr, data);
1127                         return 1;
1128                 } else {
1129                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1130                                 msr, data);
1131                         break;
1132                 }
1133         }
1134         return 0;
1135 }
1136 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1137
1138
1139 /*
1140  * Reads an msr value (of 'msr_index') into 'pdata'.
1141  * Returns 0 on success, non-0 otherwise.
1142  * Assumes vcpu_load() was already called.
1143  */
1144 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1145 {
1146         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1147 }
1148
1149 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1150 {
1151         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1152
1153         if (!msr_mtrr_valid(msr))
1154                 return 1;
1155
1156         if (msr == MSR_MTRRdefType)
1157                 *pdata = vcpu->arch.mtrr_state.def_type +
1158                          (vcpu->arch.mtrr_state.enabled << 10);
1159         else if (msr == MSR_MTRRfix64K_00000)
1160                 *pdata = p[0];
1161         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1162                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1163         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1164                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1165         else if (msr == MSR_IA32_CR_PAT)
1166                 *pdata = vcpu->arch.pat;
1167         else {  /* Variable MTRRs */
1168                 int idx, is_mtrr_mask;
1169                 u64 *pt;
1170
1171                 idx = (msr - 0x200) / 2;
1172                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1173                 if (!is_mtrr_mask)
1174                         pt =
1175                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1176                 else
1177                         pt =
1178                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1179                 *pdata = *pt;
1180         }
1181
1182         return 0;
1183 }
1184
1185 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1186 {
1187         u64 data;
1188         u64 mcg_cap = vcpu->arch.mcg_cap;
1189         unsigned bank_num = mcg_cap & 0xff;
1190
1191         switch (msr) {
1192         case MSR_IA32_P5_MC_ADDR:
1193         case MSR_IA32_P5_MC_TYPE:
1194                 data = 0;
1195                 break;
1196         case MSR_IA32_MCG_CAP:
1197                 data = vcpu->arch.mcg_cap;
1198                 break;
1199         case MSR_IA32_MCG_CTL:
1200                 if (!(mcg_cap & MCG_CTL_P))
1201                         return 1;
1202                 data = vcpu->arch.mcg_ctl;
1203                 break;
1204         case MSR_IA32_MCG_STATUS:
1205                 data = vcpu->arch.mcg_status;
1206                 break;
1207         default:
1208                 if (msr >= MSR_IA32_MC0_CTL &&
1209                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1210                         u32 offset = msr - MSR_IA32_MC0_CTL;
1211                         data = vcpu->arch.mce_banks[offset];
1212                         break;
1213                 }
1214                 return 1;
1215         }
1216         *pdata = data;
1217         return 0;
1218 }
1219
1220 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1221 {
1222         u64 data;
1223
1224         switch (msr) {
1225         case MSR_IA32_PLATFORM_ID:
1226         case MSR_IA32_UCODE_REV:
1227         case MSR_IA32_EBL_CR_POWERON:
1228         case MSR_IA32_DEBUGCTLMSR:
1229         case MSR_IA32_LASTBRANCHFROMIP:
1230         case MSR_IA32_LASTBRANCHTOIP:
1231         case MSR_IA32_LASTINTFROMIP:
1232         case MSR_IA32_LASTINTTOIP:
1233         case MSR_K8_SYSCFG:
1234         case MSR_K7_HWCR:
1235         case MSR_VM_HSAVE_PA:
1236         case MSR_P6_PERFCTR0:
1237         case MSR_P6_PERFCTR1:
1238         case MSR_P6_EVNTSEL0:
1239         case MSR_P6_EVNTSEL1:
1240         case MSR_K7_EVNTSEL0:
1241         case MSR_K7_PERFCTR0:
1242         case MSR_K8_INT_PENDING_MSG:
1243         case MSR_AMD64_NB_CFG:
1244         case MSR_FAM10H_MMIO_CONF_BASE:
1245                 data = 0;
1246                 break;
1247         case MSR_MTRRcap:
1248                 data = 0x500 | KVM_NR_VAR_MTRR;
1249                 break;
1250         case 0x200 ... 0x2ff:
1251                 return get_msr_mtrr(vcpu, msr, pdata);
1252         case 0xcd: /* fsb frequency */
1253                 data = 3;
1254                 break;
1255         case MSR_IA32_APICBASE:
1256                 data = kvm_get_apic_base(vcpu);
1257                 break;
1258         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1259                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1260                 break;
1261         case MSR_IA32_MISC_ENABLE:
1262                 data = vcpu->arch.ia32_misc_enable_msr;
1263                 break;
1264         case MSR_IA32_PERF_STATUS:
1265                 /* TSC increment by tick */
1266                 data = 1000ULL;
1267                 /* CPU multiplier */
1268                 data |= (((uint64_t)4ULL) << 40);
1269                 break;
1270         case MSR_EFER:
1271                 data = vcpu->arch.shadow_efer;
1272                 break;
1273         case MSR_KVM_WALL_CLOCK:
1274                 data = vcpu->kvm->arch.wall_clock;
1275                 break;
1276         case MSR_KVM_SYSTEM_TIME:
1277                 data = vcpu->arch.time;
1278                 break;
1279         case MSR_IA32_P5_MC_ADDR:
1280         case MSR_IA32_P5_MC_TYPE:
1281         case MSR_IA32_MCG_CAP:
1282         case MSR_IA32_MCG_CTL:
1283         case MSR_IA32_MCG_STATUS:
1284         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1285                 return get_msr_mce(vcpu, msr, pdata);
1286         default:
1287                 if (!ignore_msrs) {
1288                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1289                         return 1;
1290                 } else {
1291                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1292                         data = 0;
1293                 }
1294                 break;
1295         }
1296         *pdata = data;
1297         return 0;
1298 }
1299 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1300
1301 /*
1302  * Read or write a bunch of msrs. All parameters are kernel addresses.
1303  *
1304  * @return number of msrs set successfully.
1305  */
1306 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1307                     struct kvm_msr_entry *entries,
1308                     int (*do_msr)(struct kvm_vcpu *vcpu,
1309                                   unsigned index, u64 *data))
1310 {
1311         int i, idx;
1312
1313         vcpu_load(vcpu);
1314
1315         idx = srcu_read_lock(&vcpu->kvm->srcu);
1316         for (i = 0; i < msrs->nmsrs; ++i)
1317                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1318                         break;
1319         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1320
1321         vcpu_put(vcpu);
1322
1323         return i;
1324 }
1325
1326 /*
1327  * Read or write a bunch of msrs. Parameters are user addresses.
1328  *
1329  * @return number of msrs set successfully.
1330  */
1331 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1332                   int (*do_msr)(struct kvm_vcpu *vcpu,
1333                                 unsigned index, u64 *data),
1334                   int writeback)
1335 {
1336         struct kvm_msrs msrs;
1337         struct kvm_msr_entry *entries;
1338         int r, n;
1339         unsigned size;
1340
1341         r = -EFAULT;
1342         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1343                 goto out;
1344
1345         r = -E2BIG;
1346         if (msrs.nmsrs >= MAX_IO_MSRS)
1347                 goto out;
1348
1349         r = -ENOMEM;
1350         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1351         entries = vmalloc(size);
1352         if (!entries)
1353                 goto out;
1354
1355         r = -EFAULT;
1356         if (copy_from_user(entries, user_msrs->entries, size))
1357                 goto out_free;
1358
1359         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1360         if (r < 0)
1361                 goto out_free;
1362
1363         r = -EFAULT;
1364         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1365                 goto out_free;
1366
1367         r = n;
1368
1369 out_free:
1370         vfree(entries);
1371 out:
1372         return r;
1373 }
1374
1375 int kvm_dev_ioctl_check_extension(long ext)
1376 {
1377         int r;
1378
1379         switch (ext) {
1380         case KVM_CAP_IRQCHIP:
1381         case KVM_CAP_HLT:
1382         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1383         case KVM_CAP_SET_TSS_ADDR:
1384         case KVM_CAP_EXT_CPUID:
1385         case KVM_CAP_CLOCKSOURCE:
1386         case KVM_CAP_PIT:
1387         case KVM_CAP_NOP_IO_DELAY:
1388         case KVM_CAP_MP_STATE:
1389         case KVM_CAP_SYNC_MMU:
1390         case KVM_CAP_REINJECT_CONTROL:
1391         case KVM_CAP_IRQ_INJECT_STATUS:
1392         case KVM_CAP_ASSIGN_DEV_IRQ:
1393         case KVM_CAP_IRQFD:
1394         case KVM_CAP_IOEVENTFD:
1395         case KVM_CAP_PIT2:
1396         case KVM_CAP_PIT_STATE2:
1397         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1398         case KVM_CAP_XEN_HVM:
1399         case KVM_CAP_ADJUST_CLOCK:
1400         case KVM_CAP_VCPU_EVENTS:
1401                 r = 1;
1402                 break;
1403         case KVM_CAP_COALESCED_MMIO:
1404                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1405                 break;
1406         case KVM_CAP_VAPIC:
1407                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1408                 break;
1409         case KVM_CAP_NR_VCPUS:
1410                 r = KVM_MAX_VCPUS;
1411                 break;
1412         case KVM_CAP_NR_MEMSLOTS:
1413                 r = KVM_MEMORY_SLOTS;
1414                 break;
1415         case KVM_CAP_PV_MMU:    /* obsolete */
1416                 r = 0;
1417                 break;
1418         case KVM_CAP_IOMMU:
1419                 r = iommu_found();
1420                 break;
1421         case KVM_CAP_MCE:
1422                 r = KVM_MAX_MCE_BANKS;
1423                 break;
1424         default:
1425                 r = 0;
1426                 break;
1427         }
1428         return r;
1429
1430 }
1431
1432 long kvm_arch_dev_ioctl(struct file *filp,
1433                         unsigned int ioctl, unsigned long arg)
1434 {
1435         void __user *argp = (void __user *)arg;
1436         long r;
1437
1438         switch (ioctl) {
1439         case KVM_GET_MSR_INDEX_LIST: {
1440                 struct kvm_msr_list __user *user_msr_list = argp;
1441                 struct kvm_msr_list msr_list;
1442                 unsigned n;
1443
1444                 r = -EFAULT;
1445                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1446                         goto out;
1447                 n = msr_list.nmsrs;
1448                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1449                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1450                         goto out;
1451                 r = -E2BIG;
1452                 if (n < msr_list.nmsrs)
1453                         goto out;
1454                 r = -EFAULT;
1455                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1456                                  num_msrs_to_save * sizeof(u32)))
1457                         goto out;
1458                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1459                                  &emulated_msrs,
1460                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1461                         goto out;
1462                 r = 0;
1463                 break;
1464         }
1465         case KVM_GET_SUPPORTED_CPUID: {
1466                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1467                 struct kvm_cpuid2 cpuid;
1468
1469                 r = -EFAULT;
1470                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1471                         goto out;
1472                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1473                                                       cpuid_arg->entries);
1474                 if (r)
1475                         goto out;
1476
1477                 r = -EFAULT;
1478                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1479                         goto out;
1480                 r = 0;
1481                 break;
1482         }
1483         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1484                 u64 mce_cap;
1485
1486                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1487                 r = -EFAULT;
1488                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1489                         goto out;
1490                 r = 0;
1491                 break;
1492         }
1493         default:
1494                 r = -EINVAL;
1495         }
1496 out:
1497         return r;
1498 }
1499
1500 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1501 {
1502         kvm_x86_ops->vcpu_load(vcpu, cpu);
1503         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1504                 unsigned long khz = cpufreq_quick_get(cpu);
1505                 if (!khz)
1506                         khz = tsc_khz;
1507                 per_cpu(cpu_tsc_khz, cpu) = khz;
1508         }
1509         kvm_request_guest_time_update(vcpu);
1510 }
1511
1512 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1513 {
1514         kvm_put_guest_fpu(vcpu);
1515         kvm_x86_ops->vcpu_put(vcpu);
1516 }
1517
1518 static int is_efer_nx(void)
1519 {
1520         unsigned long long efer = 0;
1521
1522         rdmsrl_safe(MSR_EFER, &efer);
1523         return efer & EFER_NX;
1524 }
1525
1526 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1527 {
1528         int i;
1529         struct kvm_cpuid_entry2 *e, *entry;
1530
1531         entry = NULL;
1532         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1533                 e = &vcpu->arch.cpuid_entries[i];
1534                 if (e->function == 0x80000001) {
1535                         entry = e;
1536                         break;
1537                 }
1538         }
1539         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1540                 entry->edx &= ~(1 << 20);
1541                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1542         }
1543 }
1544
1545 /* when an old userspace process fills a new kernel module */
1546 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1547                                     struct kvm_cpuid *cpuid,
1548                                     struct kvm_cpuid_entry __user *entries)
1549 {
1550         int r, i;
1551         struct kvm_cpuid_entry *cpuid_entries;
1552
1553         r = -E2BIG;
1554         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1555                 goto out;
1556         r = -ENOMEM;
1557         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1558         if (!cpuid_entries)
1559                 goto out;
1560         r = -EFAULT;
1561         if (copy_from_user(cpuid_entries, entries,
1562                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1563                 goto out_free;
1564         for (i = 0; i < cpuid->nent; i++) {
1565                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1566                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1567                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1568                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1569                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1570                 vcpu->arch.cpuid_entries[i].index = 0;
1571                 vcpu->arch.cpuid_entries[i].flags = 0;
1572                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1573                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1574                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1575         }
1576         vcpu->arch.cpuid_nent = cpuid->nent;
1577         cpuid_fix_nx_cap(vcpu);
1578         r = 0;
1579         kvm_apic_set_version(vcpu);
1580         kvm_x86_ops->cpuid_update(vcpu);
1581
1582 out_free:
1583         vfree(cpuid_entries);
1584 out:
1585         return r;
1586 }
1587
1588 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1589                                      struct kvm_cpuid2 *cpuid,
1590                                      struct kvm_cpuid_entry2 __user *entries)
1591 {
1592         int r;
1593
1594         r = -E2BIG;
1595         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1596                 goto out;
1597         r = -EFAULT;
1598         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1599                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1600                 goto out;
1601         vcpu->arch.cpuid_nent = cpuid->nent;
1602         kvm_apic_set_version(vcpu);
1603         kvm_x86_ops->cpuid_update(vcpu);
1604         return 0;
1605
1606 out:
1607         return r;
1608 }
1609
1610 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1611                                      struct kvm_cpuid2 *cpuid,
1612                                      struct kvm_cpuid_entry2 __user *entries)
1613 {
1614         int r;
1615
1616         r = -E2BIG;
1617         if (cpuid->nent < vcpu->arch.cpuid_nent)
1618                 goto out;
1619         r = -EFAULT;
1620         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1621                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1622                 goto out;
1623         return 0;
1624
1625 out:
1626         cpuid->nent = vcpu->arch.cpuid_nent;
1627         return r;
1628 }
1629
1630 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1631                            u32 index)
1632 {
1633         entry->function = function;
1634         entry->index = index;
1635         cpuid_count(entry->function, entry->index,
1636                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1637         entry->flags = 0;
1638 }
1639
1640 #define F(x) bit(X86_FEATURE_##x)
1641
1642 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1643                          u32 index, int *nent, int maxnent)
1644 {
1645         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1646 #ifdef CONFIG_X86_64
1647         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1648                                 ? F(GBPAGES) : 0;
1649         unsigned f_lm = F(LM);
1650 #else
1651         unsigned f_gbpages = 0;
1652         unsigned f_lm = 0;
1653 #endif
1654         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1655
1656         /* cpuid 1.edx */
1657         const u32 kvm_supported_word0_x86_features =
1658                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1659                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1660                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1661                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1662                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1663                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1664                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1665                 0 /* HTT, TM, Reserved, PBE */;
1666         /* cpuid 0x80000001.edx */
1667         const u32 kvm_supported_word1_x86_features =
1668                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1669                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1670                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1671                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1672                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1673                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1674                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1675                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1676         /* cpuid 1.ecx */
1677         const u32 kvm_supported_word4_x86_features =
1678                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1679                 0 /* DS-CPL, VMX, SMX, EST */ |
1680                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1681                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1682                 0 /* Reserved, DCA */ | F(XMM4_1) |
1683                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1684                 0 /* Reserved, XSAVE, OSXSAVE */;
1685         /* cpuid 0x80000001.ecx */
1686         const u32 kvm_supported_word6_x86_features =
1687                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1688                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1689                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1690                 0 /* SKINIT */ | 0 /* WDT */;
1691
1692         /* all calls to cpuid_count() should be made on the same cpu */
1693         get_cpu();
1694         do_cpuid_1_ent(entry, function, index);
1695         ++*nent;
1696
1697         switch (function) {
1698         case 0:
1699                 entry->eax = min(entry->eax, (u32)0xb);
1700                 break;
1701         case 1:
1702                 entry->edx &= kvm_supported_word0_x86_features;
1703                 entry->ecx &= kvm_supported_word4_x86_features;
1704                 /* we support x2apic emulation even if host does not support
1705                  * it since we emulate x2apic in software */
1706                 entry->ecx |= F(X2APIC);
1707                 break;
1708         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1709          * may return different values. This forces us to get_cpu() before
1710          * issuing the first command, and also to emulate this annoying behavior
1711          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1712         case 2: {
1713                 int t, times = entry->eax & 0xff;
1714
1715                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1716                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1717                 for (t = 1; t < times && *nent < maxnent; ++t) {
1718                         do_cpuid_1_ent(&entry[t], function, 0);
1719                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1720                         ++*nent;
1721                 }
1722                 break;
1723         }
1724         /* function 4 and 0xb have additional index. */
1725         case 4: {
1726                 int i, cache_type;
1727
1728                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1729                 /* read more entries until cache_type is zero */
1730                 for (i = 1; *nent < maxnent; ++i) {
1731                         cache_type = entry[i - 1].eax & 0x1f;
1732                         if (!cache_type)
1733                                 break;
1734                         do_cpuid_1_ent(&entry[i], function, i);
1735                         entry[i].flags |=
1736                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1737                         ++*nent;
1738                 }
1739                 break;
1740         }
1741         case 0xb: {
1742                 int i, level_type;
1743
1744                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1745                 /* read more entries until level_type is zero */
1746                 for (i = 1; *nent < maxnent; ++i) {
1747                         level_type = entry[i - 1].ecx & 0xff00;
1748                         if (!level_type)
1749                                 break;
1750                         do_cpuid_1_ent(&entry[i], function, i);
1751                         entry[i].flags |=
1752                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1753                         ++*nent;
1754                 }
1755                 break;
1756         }
1757         case 0x80000000:
1758                 entry->eax = min(entry->eax, 0x8000001a);
1759                 break;
1760         case 0x80000001:
1761                 entry->edx &= kvm_supported_word1_x86_features;
1762                 entry->ecx &= kvm_supported_word6_x86_features;
1763                 break;
1764         }
1765         put_cpu();
1766 }
1767
1768 #undef F
1769
1770 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1771                                      struct kvm_cpuid_entry2 __user *entries)
1772 {
1773         struct kvm_cpuid_entry2 *cpuid_entries;
1774         int limit, nent = 0, r = -E2BIG;
1775         u32 func;
1776
1777         if (cpuid->nent < 1)
1778                 goto out;
1779         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1780                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1781         r = -ENOMEM;
1782         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1783         if (!cpuid_entries)
1784                 goto out;
1785
1786         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1787         limit = cpuid_entries[0].eax;
1788         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1789                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1790                              &nent, cpuid->nent);
1791         r = -E2BIG;
1792         if (nent >= cpuid->nent)
1793                 goto out_free;
1794
1795         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1796         limit = cpuid_entries[nent - 1].eax;
1797         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1798                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1799                              &nent, cpuid->nent);
1800         r = -E2BIG;
1801         if (nent >= cpuid->nent)
1802                 goto out_free;
1803
1804         r = -EFAULT;
1805         if (copy_to_user(entries, cpuid_entries,
1806                          nent * sizeof(struct kvm_cpuid_entry2)))
1807                 goto out_free;
1808         cpuid->nent = nent;
1809         r = 0;
1810
1811 out_free:
1812         vfree(cpuid_entries);
1813 out:
1814         return r;
1815 }
1816
1817 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1818                                     struct kvm_lapic_state *s)
1819 {
1820         vcpu_load(vcpu);
1821         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1822         vcpu_put(vcpu);
1823
1824         return 0;
1825 }
1826
1827 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1828                                     struct kvm_lapic_state *s)
1829 {
1830         vcpu_load(vcpu);
1831         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1832         kvm_apic_post_state_restore(vcpu);
1833         update_cr8_intercept(vcpu);
1834         vcpu_put(vcpu);
1835
1836         return 0;
1837 }
1838
1839 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1840                                     struct kvm_interrupt *irq)
1841 {
1842         if (irq->irq < 0 || irq->irq >= 256)
1843                 return -EINVAL;
1844         if (irqchip_in_kernel(vcpu->kvm))
1845                 return -ENXIO;
1846         vcpu_load(vcpu);
1847
1848         kvm_queue_interrupt(vcpu, irq->irq, false);
1849
1850         vcpu_put(vcpu);
1851
1852         return 0;
1853 }
1854
1855 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1856 {
1857         vcpu_load(vcpu);
1858         kvm_inject_nmi(vcpu);
1859         vcpu_put(vcpu);
1860
1861         return 0;
1862 }
1863
1864 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1865                                            struct kvm_tpr_access_ctl *tac)
1866 {
1867         if (tac->flags)
1868                 return -EINVAL;
1869         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1870         return 0;
1871 }
1872
1873 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1874                                         u64 mcg_cap)
1875 {
1876         int r;
1877         unsigned bank_num = mcg_cap & 0xff, bank;
1878
1879         r = -EINVAL;
1880         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1881                 goto out;
1882         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1883                 goto out;
1884         r = 0;
1885         vcpu->arch.mcg_cap = mcg_cap;
1886         /* Init IA32_MCG_CTL to all 1s */
1887         if (mcg_cap & MCG_CTL_P)
1888                 vcpu->arch.mcg_ctl = ~(u64)0;
1889         /* Init IA32_MCi_CTL to all 1s */
1890         for (bank = 0; bank < bank_num; bank++)
1891                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1892 out:
1893         return r;
1894 }
1895
1896 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1897                                       struct kvm_x86_mce *mce)
1898 {
1899         u64 mcg_cap = vcpu->arch.mcg_cap;
1900         unsigned bank_num = mcg_cap & 0xff;
1901         u64 *banks = vcpu->arch.mce_banks;
1902
1903         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1904                 return -EINVAL;
1905         /*
1906          * if IA32_MCG_CTL is not all 1s, the uncorrected error
1907          * reporting is disabled
1908          */
1909         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1910             vcpu->arch.mcg_ctl != ~(u64)0)
1911                 return 0;
1912         banks += 4 * mce->bank;
1913         /*
1914          * if IA32_MCi_CTL is not all 1s, the uncorrected error
1915          * reporting is disabled for the bank
1916          */
1917         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1918                 return 0;
1919         if (mce->status & MCI_STATUS_UC) {
1920                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1921                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
1922                         printk(KERN_DEBUG "kvm: set_mce: "
1923                                "injects mce exception while "
1924                                "previous one is in progress!\n");
1925                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1926                         return 0;
1927                 }
1928                 if (banks[1] & MCI_STATUS_VAL)
1929                         mce->status |= MCI_STATUS_OVER;
1930                 banks[2] = mce->addr;
1931                 banks[3] = mce->misc;
1932                 vcpu->arch.mcg_status = mce->mcg_status;
1933                 banks[1] = mce->status;
1934                 kvm_queue_exception(vcpu, MC_VECTOR);
1935         } else if (!(banks[1] & MCI_STATUS_VAL)
1936                    || !(banks[1] & MCI_STATUS_UC)) {
1937                 if (banks[1] & MCI_STATUS_VAL)
1938                         mce->status |= MCI_STATUS_OVER;
1939                 banks[2] = mce->addr;
1940                 banks[3] = mce->misc;
1941                 banks[1] = mce->status;
1942         } else
1943                 banks[1] |= MCI_STATUS_OVER;
1944         return 0;
1945 }
1946
1947 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
1948                                                struct kvm_vcpu_events *events)
1949 {
1950         vcpu_load(vcpu);
1951
1952         events->exception.injected = vcpu->arch.exception.pending;
1953         events->exception.nr = vcpu->arch.exception.nr;
1954         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
1955         events->exception.error_code = vcpu->arch.exception.error_code;
1956
1957         events->interrupt.injected = vcpu->arch.interrupt.pending;
1958         events->interrupt.nr = vcpu->arch.interrupt.nr;
1959         events->interrupt.soft = vcpu->arch.interrupt.soft;
1960
1961         events->nmi.injected = vcpu->arch.nmi_injected;
1962         events->nmi.pending = vcpu->arch.nmi_pending;
1963         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
1964
1965         events->sipi_vector = vcpu->arch.sipi_vector;
1966
1967         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
1968                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
1969
1970         vcpu_put(vcpu);
1971 }
1972
1973 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
1974                                               struct kvm_vcpu_events *events)
1975 {
1976         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
1977                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
1978                 return -EINVAL;
1979
1980         vcpu_load(vcpu);
1981
1982         vcpu->arch.exception.pending = events->exception.injected;
1983         vcpu->arch.exception.nr = events->exception.nr;
1984         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
1985         vcpu->arch.exception.error_code = events->exception.error_code;
1986
1987         vcpu->arch.interrupt.pending = events->interrupt.injected;
1988         vcpu->arch.interrupt.nr = events->interrupt.nr;
1989         vcpu->arch.interrupt.soft = events->interrupt.soft;
1990         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
1991                 kvm_pic_clear_isr_ack(vcpu->kvm);
1992
1993         vcpu->arch.nmi_injected = events->nmi.injected;
1994         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
1995                 vcpu->arch.nmi_pending = events->nmi.pending;
1996         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
1997
1998         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
1999                 vcpu->arch.sipi_vector = events->sipi_vector;
2000
2001         vcpu_put(vcpu);
2002
2003         return 0;
2004 }
2005
2006 long kvm_arch_vcpu_ioctl(struct file *filp,
2007                          unsigned int ioctl, unsigned long arg)
2008 {
2009         struct kvm_vcpu *vcpu = filp->private_data;
2010         void __user *argp = (void __user *)arg;
2011         int r;
2012         struct kvm_lapic_state *lapic = NULL;
2013
2014         switch (ioctl) {
2015         case KVM_GET_LAPIC: {
2016                 r = -EINVAL;
2017                 if (!vcpu->arch.apic)
2018                         goto out;
2019                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2020
2021                 r = -ENOMEM;
2022                 if (!lapic)
2023                         goto out;
2024                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2025                 if (r)
2026                         goto out;
2027                 r = -EFAULT;
2028                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2029                         goto out;
2030                 r = 0;
2031                 break;
2032         }
2033         case KVM_SET_LAPIC: {
2034                 r = -EINVAL;
2035                 if (!vcpu->arch.apic)
2036                         goto out;
2037                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2038                 r = -ENOMEM;
2039                 if (!lapic)
2040                         goto out;
2041                 r = -EFAULT;
2042                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2043                         goto out;
2044                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2045                 if (r)
2046                         goto out;
2047                 r = 0;
2048                 break;
2049         }
2050         case KVM_INTERRUPT: {
2051                 struct kvm_interrupt irq;
2052
2053                 r = -EFAULT;
2054                 if (copy_from_user(&irq, argp, sizeof irq))
2055                         goto out;
2056                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2057                 if (r)
2058                         goto out;
2059                 r = 0;
2060                 break;
2061         }
2062         case KVM_NMI: {
2063                 r = kvm_vcpu_ioctl_nmi(vcpu);
2064                 if (r)
2065                         goto out;
2066                 r = 0;
2067                 break;
2068         }
2069         case KVM_SET_CPUID: {
2070                 struct kvm_cpuid __user *cpuid_arg = argp;
2071                 struct kvm_cpuid cpuid;
2072
2073                 r = -EFAULT;
2074                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2075                         goto out;
2076                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2077                 if (r)
2078                         goto out;
2079                 break;
2080         }
2081         case KVM_SET_CPUID2: {
2082                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2083                 struct kvm_cpuid2 cpuid;
2084
2085                 r = -EFAULT;
2086                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2087                         goto out;
2088                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2089                                               cpuid_arg->entries);
2090                 if (r)
2091                         goto out;
2092                 break;
2093         }
2094         case KVM_GET_CPUID2: {
2095                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2096                 struct kvm_cpuid2 cpuid;
2097
2098                 r = -EFAULT;
2099                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2100                         goto out;
2101                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2102                                               cpuid_arg->entries);
2103                 if (r)
2104                         goto out;
2105                 r = -EFAULT;
2106                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2107                         goto out;
2108                 r = 0;
2109                 break;
2110         }
2111         case KVM_GET_MSRS:
2112                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2113                 break;
2114         case KVM_SET_MSRS:
2115                 r = msr_io(vcpu, argp, do_set_msr, 0);
2116                 break;
2117         case KVM_TPR_ACCESS_REPORTING: {
2118                 struct kvm_tpr_access_ctl tac;
2119
2120                 r = -EFAULT;
2121                 if (copy_from_user(&tac, argp, sizeof tac))
2122                         goto out;
2123                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2124                 if (r)
2125                         goto out;
2126                 r = -EFAULT;
2127                 if (copy_to_user(argp, &tac, sizeof tac))
2128                         goto out;
2129                 r = 0;
2130                 break;
2131         };
2132         case KVM_SET_VAPIC_ADDR: {
2133                 struct kvm_vapic_addr va;
2134
2135                 r = -EINVAL;
2136                 if (!irqchip_in_kernel(vcpu->kvm))
2137                         goto out;
2138                 r = -EFAULT;
2139                 if (copy_from_user(&va, argp, sizeof va))
2140                         goto out;
2141                 r = 0;
2142                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2143                 break;
2144         }
2145         case KVM_X86_SETUP_MCE: {
2146                 u64 mcg_cap;
2147
2148                 r = -EFAULT;
2149                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2150                         goto out;
2151                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2152                 break;
2153         }
2154         case KVM_X86_SET_MCE: {
2155                 struct kvm_x86_mce mce;
2156
2157                 r = -EFAULT;
2158                 if (copy_from_user(&mce, argp, sizeof mce))
2159                         goto out;
2160                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2161                 break;
2162         }
2163         case KVM_GET_VCPU_EVENTS: {
2164                 struct kvm_vcpu_events events;
2165
2166                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2167
2168                 r = -EFAULT;
2169                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2170                         break;
2171                 r = 0;
2172                 break;
2173         }
2174         case KVM_SET_VCPU_EVENTS: {
2175                 struct kvm_vcpu_events events;
2176
2177                 r = -EFAULT;
2178                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2179                         break;
2180
2181                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2182                 break;
2183         }
2184         default:
2185                 r = -EINVAL;
2186         }
2187 out:
2188         kfree(lapic);
2189         return r;
2190 }
2191
2192 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2193 {
2194         int ret;
2195
2196         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2197                 return -1;
2198         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2199         return ret;
2200 }
2201
2202 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2203                                               u64 ident_addr)
2204 {
2205         kvm->arch.ept_identity_map_addr = ident_addr;
2206         return 0;
2207 }
2208
2209 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2210                                           u32 kvm_nr_mmu_pages)
2211 {
2212         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2213                 return -EINVAL;
2214
2215         mutex_lock(&kvm->slots_lock);
2216         spin_lock(&kvm->mmu_lock);
2217
2218         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2219         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2220
2221         spin_unlock(&kvm->mmu_lock);
2222         mutex_unlock(&kvm->slots_lock);
2223         return 0;
2224 }
2225
2226 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2227 {
2228         return kvm->arch.n_alloc_mmu_pages;
2229 }
2230
2231 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2232 {
2233         int i;
2234         struct kvm_mem_alias *alias;
2235         struct kvm_mem_aliases *aliases;
2236
2237         aliases = rcu_dereference(kvm->arch.aliases);
2238
2239         for (i = 0; i < aliases->naliases; ++i) {
2240                 alias = &aliases->aliases[i];
2241                 if (alias->flags & KVM_ALIAS_INVALID)
2242                         continue;
2243                 if (gfn >= alias->base_gfn
2244                     && gfn < alias->base_gfn + alias->npages)
2245                         return alias->target_gfn + gfn - alias->base_gfn;
2246         }
2247         return gfn;
2248 }
2249
2250 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2251 {
2252         int i;
2253         struct kvm_mem_alias *alias;
2254         struct kvm_mem_aliases *aliases;
2255
2256         aliases = rcu_dereference(kvm->arch.aliases);
2257
2258         for (i = 0; i < aliases->naliases; ++i) {
2259                 alias = &aliases->aliases[i];
2260                 if (gfn >= alias->base_gfn
2261                     && gfn < alias->base_gfn + alias->npages)
2262                         return alias->target_gfn + gfn - alias->base_gfn;
2263         }
2264         return gfn;
2265 }
2266
2267 /*
2268  * Set a new alias region.  Aliases map a portion of physical memory into
2269  * another portion.  This is useful for memory windows, for example the PC
2270  * VGA region.
2271  */
2272 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2273                                          struct kvm_memory_alias *alias)
2274 {
2275         int r, n;
2276         struct kvm_mem_alias *p;
2277         struct kvm_mem_aliases *aliases, *old_aliases;
2278
2279         r = -EINVAL;
2280         /* General sanity checks */
2281         if (alias->memory_size & (PAGE_SIZE - 1))
2282                 goto out;
2283         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2284                 goto out;
2285         if (alias->slot >= KVM_ALIAS_SLOTS)
2286                 goto out;
2287         if (alias->guest_phys_addr + alias->memory_size
2288             < alias->guest_phys_addr)
2289                 goto out;
2290         if (alias->target_phys_addr + alias->memory_size
2291             < alias->target_phys_addr)
2292                 goto out;
2293
2294         r = -ENOMEM;
2295         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2296         if (!aliases)
2297                 goto out;
2298
2299         mutex_lock(&kvm->slots_lock);
2300
2301         /* invalidate any gfn reference in case of deletion/shrinking */
2302         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2303         aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2304         old_aliases = kvm->arch.aliases;
2305         rcu_assign_pointer(kvm->arch.aliases, aliases);
2306         synchronize_srcu_expedited(&kvm->srcu);
2307         kvm_mmu_zap_all(kvm);
2308         kfree(old_aliases);
2309
2310         r = -ENOMEM;
2311         aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2312         if (!aliases)
2313                 goto out_unlock;
2314
2315         memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2316
2317         p = &aliases->aliases[alias->slot];
2318         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2319         p->npages = alias->memory_size >> PAGE_SHIFT;
2320         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2321         p->flags &= ~(KVM_ALIAS_INVALID);
2322
2323         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2324                 if (aliases->aliases[n - 1].npages)
2325                         break;
2326         aliases->naliases = n;
2327
2328         old_aliases = kvm->arch.aliases;
2329         rcu_assign_pointer(kvm->arch.aliases, aliases);
2330         synchronize_srcu_expedited(&kvm->srcu);
2331         kfree(old_aliases);
2332         r = 0;
2333
2334 out_unlock:
2335         mutex_unlock(&kvm->slots_lock);
2336 out:
2337         return r;
2338 }
2339
2340 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2341 {
2342         int r;
2343
2344         r = 0;
2345         switch (chip->chip_id) {
2346         case KVM_IRQCHIP_PIC_MASTER:
2347                 memcpy(&chip->chip.pic,
2348                         &pic_irqchip(kvm)->pics[0],
2349                         sizeof(struct kvm_pic_state));
2350                 break;
2351         case KVM_IRQCHIP_PIC_SLAVE:
2352                 memcpy(&chip->chip.pic,
2353                         &pic_irqchip(kvm)->pics[1],
2354                         sizeof(struct kvm_pic_state));
2355                 break;
2356         case KVM_IRQCHIP_IOAPIC:
2357                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2358                 break;
2359         default:
2360                 r = -EINVAL;
2361                 break;
2362         }
2363         return r;
2364 }
2365
2366 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2367 {
2368         int r;
2369
2370         r = 0;
2371         switch (chip->chip_id) {
2372         case KVM_IRQCHIP_PIC_MASTER:
2373                 spin_lock(&pic_irqchip(kvm)->lock);
2374                 memcpy(&pic_irqchip(kvm)->pics[0],
2375                         &chip->chip.pic,
2376                         sizeof(struct kvm_pic_state));
2377                 spin_unlock(&pic_irqchip(kvm)->lock);
2378                 break;
2379         case KVM_IRQCHIP_PIC_SLAVE:
2380                 spin_lock(&pic_irqchip(kvm)->lock);
2381                 memcpy(&pic_irqchip(kvm)->pics[1],
2382                         &chip->chip.pic,
2383                         sizeof(struct kvm_pic_state));
2384                 spin_unlock(&pic_irqchip(kvm)->lock);
2385                 break;
2386         case KVM_IRQCHIP_IOAPIC:
2387                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2388                 break;
2389         default:
2390                 r = -EINVAL;
2391                 break;
2392         }
2393         kvm_pic_update_irq(pic_irqchip(kvm));
2394         return r;
2395 }
2396
2397 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2398 {
2399         int r = 0;
2400
2401         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2402         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2403         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2404         return r;
2405 }
2406
2407 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2408 {
2409         int r = 0;
2410
2411         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2412         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2413         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2414         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2415         return r;
2416 }
2417
2418 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2419 {
2420         int r = 0;
2421
2422         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2423         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2424                 sizeof(ps->channels));
2425         ps->flags = kvm->arch.vpit->pit_state.flags;
2426         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2427         return r;
2428 }
2429
2430 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2431 {
2432         int r = 0, start = 0;
2433         u32 prev_legacy, cur_legacy;
2434         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2435         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2436         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2437         if (!prev_legacy && cur_legacy)
2438                 start = 1;
2439         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2440                sizeof(kvm->arch.vpit->pit_state.channels));
2441         kvm->arch.vpit->pit_state.flags = ps->flags;
2442         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2443         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2444         return r;
2445 }
2446
2447 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2448                                  struct kvm_reinject_control *control)
2449 {
2450         if (!kvm->arch.vpit)
2451                 return -ENXIO;
2452         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2453         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2454         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2455         return 0;
2456 }
2457
2458 /*
2459  * Get (and clear) the dirty memory log for a memory slot.
2460  */
2461 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2462                                       struct kvm_dirty_log *log)
2463 {
2464         int r, n, i;
2465         struct kvm_memory_slot *memslot;
2466         unsigned long is_dirty = 0;
2467         unsigned long *dirty_bitmap = NULL;
2468
2469         mutex_lock(&kvm->slots_lock);
2470
2471         r = -EINVAL;
2472         if (log->slot >= KVM_MEMORY_SLOTS)
2473                 goto out;
2474
2475         memslot = &kvm->memslots->memslots[log->slot];
2476         r = -ENOENT;
2477         if (!memslot->dirty_bitmap)
2478                 goto out;
2479
2480         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2481
2482         r = -ENOMEM;
2483         dirty_bitmap = vmalloc(n);
2484         if (!dirty_bitmap)
2485                 goto out;
2486         memset(dirty_bitmap, 0, n);
2487
2488         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2489                 is_dirty = memslot->dirty_bitmap[i];
2490
2491         /* If nothing is dirty, don't bother messing with page tables. */
2492         if (is_dirty) {
2493                 struct kvm_memslots *slots, *old_slots;
2494
2495                 spin_lock(&kvm->mmu_lock);
2496                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2497                 spin_unlock(&kvm->mmu_lock);
2498
2499                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2500                 if (!slots)
2501                         goto out_free;
2502
2503                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2504                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2505
2506                 old_slots = kvm->memslots;
2507                 rcu_assign_pointer(kvm->memslots, slots);
2508                 synchronize_srcu_expedited(&kvm->srcu);
2509                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2510                 kfree(old_slots);
2511         }
2512
2513         r = 0;
2514         if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2515                 r = -EFAULT;
2516 out_free:
2517         vfree(dirty_bitmap);
2518 out:
2519         mutex_unlock(&kvm->slots_lock);
2520         return r;
2521 }
2522
2523 long kvm_arch_vm_ioctl(struct file *filp,
2524                        unsigned int ioctl, unsigned long arg)
2525 {
2526         struct kvm *kvm = filp->private_data;
2527         void __user *argp = (void __user *)arg;
2528         int r = -ENOTTY;
2529         /*
2530          * This union makes it completely explicit to gcc-3.x
2531          * that these two variables' stack usage should be
2532          * combined, not added together.
2533          */
2534         union {
2535                 struct kvm_pit_state ps;
2536                 struct kvm_pit_state2 ps2;
2537                 struct kvm_memory_alias alias;
2538                 struct kvm_pit_config pit_config;
2539         } u;
2540
2541         switch (ioctl) {
2542         case KVM_SET_TSS_ADDR:
2543                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2544                 if (r < 0)
2545                         goto out;
2546                 break;
2547         case KVM_SET_IDENTITY_MAP_ADDR: {
2548                 u64 ident_addr;
2549
2550                 r = -EFAULT;
2551                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2552                         goto out;
2553                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2554                 if (r < 0)
2555                         goto out;
2556                 break;
2557         }
2558         case KVM_SET_MEMORY_REGION: {
2559                 struct kvm_memory_region kvm_mem;
2560                 struct kvm_userspace_memory_region kvm_userspace_mem;
2561
2562                 r = -EFAULT;
2563                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2564                         goto out;
2565                 kvm_userspace_mem.slot = kvm_mem.slot;
2566                 kvm_userspace_mem.flags = kvm_mem.flags;
2567                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2568                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2569                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2570                 if (r)
2571                         goto out;
2572                 break;
2573         }
2574         case KVM_SET_NR_MMU_PAGES:
2575                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2576                 if (r)
2577                         goto out;
2578                 break;
2579         case KVM_GET_NR_MMU_PAGES:
2580                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2581                 break;
2582         case KVM_SET_MEMORY_ALIAS:
2583                 r = -EFAULT;
2584                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2585                         goto out;
2586                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2587                 if (r)
2588                         goto out;
2589                 break;
2590         case KVM_CREATE_IRQCHIP: {
2591                 struct kvm_pic *vpic;
2592
2593                 mutex_lock(&kvm->lock);
2594                 r = -EEXIST;
2595                 if (kvm->arch.vpic)
2596                         goto create_irqchip_unlock;
2597                 r = -ENOMEM;
2598                 vpic = kvm_create_pic(kvm);
2599                 if (vpic) {
2600                         r = kvm_ioapic_init(kvm);
2601                         if (r) {
2602                                 kfree(vpic);
2603                                 goto create_irqchip_unlock;
2604                         }
2605                 } else
2606                         goto create_irqchip_unlock;
2607                 smp_wmb();
2608                 kvm->arch.vpic = vpic;
2609                 smp_wmb();
2610                 r = kvm_setup_default_irq_routing(kvm);
2611                 if (r) {
2612                         mutex_lock(&kvm->irq_lock);
2613                         kfree(kvm->arch.vpic);
2614                         kfree(kvm->arch.vioapic);
2615                         kvm->arch.vpic = NULL;
2616                         kvm->arch.vioapic = NULL;
2617                         mutex_unlock(&kvm->irq_lock);
2618                 }
2619         create_irqchip_unlock:
2620                 mutex_unlock(&kvm->lock);
2621                 break;
2622         }
2623         case KVM_CREATE_PIT:
2624                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2625                 goto create_pit;
2626         case KVM_CREATE_PIT2:
2627                 r = -EFAULT;
2628                 if (copy_from_user(&u.pit_config, argp,
2629                                    sizeof(struct kvm_pit_config)))
2630                         goto out;
2631         create_pit:
2632                 mutex_lock(&kvm->slots_lock);
2633                 r = -EEXIST;
2634                 if (kvm->arch.vpit)
2635                         goto create_pit_unlock;
2636                 r = -ENOMEM;
2637                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2638                 if (kvm->arch.vpit)
2639                         r = 0;
2640         create_pit_unlock:
2641                 mutex_unlock(&kvm->slots_lock);
2642                 break;
2643         case KVM_IRQ_LINE_STATUS:
2644         case KVM_IRQ_LINE: {
2645                 struct kvm_irq_level irq_event;
2646
2647                 r = -EFAULT;
2648                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2649                         goto out;
2650                 if (irqchip_in_kernel(kvm)) {
2651                         __s32 status;
2652                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2653                                         irq_event.irq, irq_event.level);
2654                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2655                                 irq_event.status = status;
2656                                 if (copy_to_user(argp, &irq_event,
2657                                                         sizeof irq_event))
2658                                         goto out;
2659                         }
2660                         r = 0;
2661                 }
2662                 break;
2663         }
2664         case KVM_GET_IRQCHIP: {
2665                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2666                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2667
2668                 r = -ENOMEM;
2669                 if (!chip)
2670                         goto out;
2671                 r = -EFAULT;
2672                 if (copy_from_user(chip, argp, sizeof *chip))
2673                         goto get_irqchip_out;
2674                 r = -ENXIO;
2675                 if (!irqchip_in_kernel(kvm))
2676                         goto get_irqchip_out;
2677                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2678                 if (r)
2679                         goto get_irqchip_out;
2680                 r = -EFAULT;
2681                 if (copy_to_user(argp, chip, sizeof *chip))
2682                         goto get_irqchip_out;
2683                 r = 0;
2684         get_irqchip_out:
2685                 kfree(chip);
2686                 if (r)
2687                         goto out;
2688                 break;
2689         }
2690         case KVM_SET_IRQCHIP: {
2691                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2692                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2693
2694                 r = -ENOMEM;
2695                 if (!chip)
2696                         goto out;
2697                 r = -EFAULT;
2698                 if (copy_from_user(chip, argp, sizeof *chip))
2699                         goto set_irqchip_out;
2700                 r = -ENXIO;
2701                 if (!irqchip_in_kernel(kvm))
2702                         goto set_irqchip_out;
2703                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2704                 if (r)
2705                         goto set_irqchip_out;
2706                 r = 0;
2707         set_irqchip_out:
2708                 kfree(chip);
2709                 if (r)
2710                         goto out;
2711                 break;
2712         }
2713         case KVM_GET_PIT: {
2714                 r = -EFAULT;
2715                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2716                         goto out;
2717                 r = -ENXIO;
2718                 if (!kvm->arch.vpit)
2719                         goto out;
2720                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2721                 if (r)
2722                         goto out;
2723                 r = -EFAULT;
2724                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2725                         goto out;
2726                 r = 0;
2727                 break;
2728         }
2729         case KVM_SET_PIT: {
2730                 r = -EFAULT;
2731                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2732                         goto out;
2733                 r = -ENXIO;
2734                 if (!kvm->arch.vpit)
2735                         goto out;
2736                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2737                 if (r)
2738                         goto out;
2739                 r = 0;
2740                 break;
2741         }
2742         case KVM_GET_PIT2: {
2743                 r = -ENXIO;
2744                 if (!kvm->arch.vpit)
2745                         goto out;
2746                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2747                 if (r)
2748                         goto out;
2749                 r = -EFAULT;
2750                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2751                         goto out;
2752                 r = 0;
2753                 break;
2754         }
2755         case KVM_SET_PIT2: {
2756                 r = -EFAULT;
2757                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2758                         goto out;
2759                 r = -ENXIO;
2760                 if (!kvm->arch.vpit)
2761                         goto out;
2762                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2763                 if (r)
2764                         goto out;
2765                 r = 0;
2766                 break;
2767         }
2768         case KVM_REINJECT_CONTROL: {
2769                 struct kvm_reinject_control control;
2770                 r =  -EFAULT;
2771                 if (copy_from_user(&control, argp, sizeof(control)))
2772                         goto out;
2773                 r = kvm_vm_ioctl_reinject(kvm, &control);
2774                 if (r)
2775                         goto out;
2776                 r = 0;
2777                 break;
2778         }
2779         case KVM_XEN_HVM_CONFIG: {
2780                 r = -EFAULT;
2781                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2782                                    sizeof(struct kvm_xen_hvm_config)))
2783                         goto out;
2784                 r = -EINVAL;
2785                 if (kvm->arch.xen_hvm_config.flags)
2786                         goto out;
2787                 r = 0;
2788                 break;
2789         }
2790         case KVM_SET_CLOCK: {
2791                 struct timespec now;
2792                 struct kvm_clock_data user_ns;
2793                 u64 now_ns;
2794                 s64 delta;
2795
2796                 r = -EFAULT;
2797                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2798                         goto out;
2799
2800                 r = -EINVAL;
2801                 if (user_ns.flags)
2802                         goto out;
2803
2804                 r = 0;
2805                 ktime_get_ts(&now);
2806                 now_ns = timespec_to_ns(&now);
2807                 delta = user_ns.clock - now_ns;
2808                 kvm->arch.kvmclock_offset = delta;
2809                 break;
2810         }
2811         case KVM_GET_CLOCK: {
2812                 struct timespec now;
2813                 struct kvm_clock_data user_ns;
2814                 u64 now_ns;
2815
2816                 ktime_get_ts(&now);
2817                 now_ns = timespec_to_ns(&now);
2818                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2819                 user_ns.flags = 0;
2820
2821                 r = -EFAULT;
2822                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2823                         goto out;
2824                 r = 0;
2825                 break;
2826         }
2827
2828         default:
2829                 ;
2830         }
2831 out:
2832         return r;
2833 }
2834
2835 static void kvm_init_msr_list(void)
2836 {
2837         u32 dummy[2];
2838         unsigned i, j;
2839
2840         /* skip the first msrs in the list. KVM-specific */
2841         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2842                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2843                         continue;
2844                 if (j < i)
2845                         msrs_to_save[j] = msrs_to_save[i];
2846                 j++;
2847         }
2848         num_msrs_to_save = j;
2849 }
2850
2851 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2852                            const void *v)
2853 {
2854         if (vcpu->arch.apic &&
2855             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2856                 return 0;
2857
2858         return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
2859 }
2860
2861 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2862 {
2863         if (vcpu->arch.apic &&
2864             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2865                 return 0;
2866
2867         return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
2868 }
2869
2870 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2871                                struct kvm_vcpu *vcpu)
2872 {
2873         void *data = val;
2874         int r = X86EMUL_CONTINUE;
2875
2876         while (bytes) {
2877                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2878                 unsigned offset = addr & (PAGE_SIZE-1);
2879                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2880                 int ret;
2881
2882                 if (gpa == UNMAPPED_GVA) {
2883                         r = X86EMUL_PROPAGATE_FAULT;
2884                         goto out;
2885                 }
2886                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2887                 if (ret < 0) {
2888                         r = X86EMUL_UNHANDLEABLE;
2889                         goto out;
2890                 }
2891
2892                 bytes -= toread;
2893                 data += toread;
2894                 addr += toread;
2895         }
2896 out:
2897         return r;
2898 }
2899
2900 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2901                                 struct kvm_vcpu *vcpu)
2902 {
2903         void *data = val;
2904         int r = X86EMUL_CONTINUE;
2905
2906         while (bytes) {
2907                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2908                 unsigned offset = addr & (PAGE_SIZE-1);
2909                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2910                 int ret;
2911
2912                 if (gpa == UNMAPPED_GVA) {
2913                         r = X86EMUL_PROPAGATE_FAULT;
2914                         goto out;
2915                 }
2916                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2917                 if (ret < 0) {
2918                         r = X86EMUL_UNHANDLEABLE;
2919                         goto out;
2920                 }
2921
2922                 bytes -= towrite;
2923                 data += towrite;
2924                 addr += towrite;
2925         }
2926 out:
2927         return r;
2928 }
2929
2930
2931 static int emulator_read_emulated(unsigned long addr,
2932                                   void *val,
2933                                   unsigned int bytes,
2934                                   struct kvm_vcpu *vcpu)
2935 {
2936         gpa_t                 gpa;
2937
2938         if (vcpu->mmio_read_completed) {
2939                 memcpy(val, vcpu->mmio_data, bytes);
2940                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2941                                vcpu->mmio_phys_addr, *(u64 *)val);
2942                 vcpu->mmio_read_completed = 0;
2943                 return X86EMUL_CONTINUE;
2944         }
2945
2946         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2947
2948         /* For APIC access vmexit */
2949         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2950                 goto mmio;
2951
2952         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2953                                 == X86EMUL_CONTINUE)
2954                 return X86EMUL_CONTINUE;
2955         if (gpa == UNMAPPED_GVA)
2956                 return X86EMUL_PROPAGATE_FAULT;
2957
2958 mmio:
2959         /*
2960          * Is this MMIO handled locally?
2961          */
2962         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2963                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2964                 return X86EMUL_CONTINUE;
2965         }
2966
2967         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2968
2969         vcpu->mmio_needed = 1;
2970         vcpu->mmio_phys_addr = gpa;
2971         vcpu->mmio_size = bytes;
2972         vcpu->mmio_is_write = 0;
2973
2974         return X86EMUL_UNHANDLEABLE;
2975 }
2976
2977 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2978                           const void *val, int bytes)
2979 {
2980         int ret;
2981
2982         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2983         if (ret < 0)
2984                 return 0;
2985         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2986         return 1;
2987 }
2988
2989 static int emulator_write_emulated_onepage(unsigned long addr,
2990                                            const void *val,
2991                                            unsigned int bytes,
2992                                            struct kvm_vcpu *vcpu)
2993 {
2994         gpa_t                 gpa;
2995
2996         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2997
2998         if (gpa == UNMAPPED_GVA) {
2999                 kvm_inject_page_fault(vcpu, addr, 2);
3000                 return X86EMUL_PROPAGATE_FAULT;
3001         }
3002
3003         /* For APIC access vmexit */
3004         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3005                 goto mmio;
3006
3007         if (emulator_write_phys(vcpu, gpa, val, bytes))
3008                 return X86EMUL_CONTINUE;
3009
3010 mmio:
3011         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3012         /*
3013          * Is this MMIO handled locally?
3014          */
3015         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3016                 return X86EMUL_CONTINUE;
3017
3018         vcpu->mmio_needed = 1;
3019         vcpu->mmio_phys_addr = gpa;
3020         vcpu->mmio_size = bytes;
3021         vcpu->mmio_is_write = 1;
3022         memcpy(vcpu->mmio_data, val, bytes);
3023
3024         return X86EMUL_CONTINUE;
3025 }
3026
3027 int emulator_write_emulated(unsigned long addr,
3028                                    const void *val,
3029                                    unsigned int bytes,
3030                                    struct kvm_vcpu *vcpu)
3031 {
3032         /* Crossing a page boundary? */
3033         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3034                 int rc, now;
3035
3036                 now = -addr & ~PAGE_MASK;
3037                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3038                 if (rc != X86EMUL_CONTINUE)
3039                         return rc;
3040                 addr += now;
3041                 val += now;
3042                 bytes -= now;
3043         }
3044         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3045 }
3046 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3047
3048 static int emulator_cmpxchg_emulated(unsigned long addr,
3049                                      const void *old,
3050                                      const void *new,
3051                                      unsigned int bytes,
3052                                      struct kvm_vcpu *vcpu)
3053 {
3054         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3055 #ifndef CONFIG_X86_64
3056         /* guests cmpxchg8b have to be emulated atomically */
3057         if (bytes == 8) {
3058                 gpa_t gpa;
3059                 struct page *page;
3060                 char *kaddr;
3061                 u64 val;
3062
3063                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
3064
3065                 if (gpa == UNMAPPED_GVA ||
3066                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3067                         goto emul_write;
3068
3069                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3070                         goto emul_write;
3071
3072                 val = *(u64 *)new;
3073
3074                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3075
3076                 kaddr = kmap_atomic(page, KM_USER0);
3077                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3078                 kunmap_atomic(kaddr, KM_USER0);
3079                 kvm_release_page_dirty(page);
3080         }
3081 emul_write:
3082 #endif
3083
3084         return emulator_write_emulated(addr, new, bytes, vcpu);
3085 }
3086
3087 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3088 {
3089         return kvm_x86_ops->get_segment_base(vcpu, seg);
3090 }
3091
3092 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3093 {
3094         kvm_mmu_invlpg(vcpu, address);
3095         return X86EMUL_CONTINUE;
3096 }
3097
3098 int emulate_clts(struct kvm_vcpu *vcpu)
3099 {
3100         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3101         return X86EMUL_CONTINUE;
3102 }
3103
3104 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3105 {
3106         struct kvm_vcpu *vcpu = ctxt->vcpu;
3107
3108         switch (dr) {
3109         case 0 ... 3:
3110                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
3111                 return X86EMUL_CONTINUE;
3112         default:
3113                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
3114                 return X86EMUL_UNHANDLEABLE;
3115         }
3116 }
3117
3118 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3119 {
3120         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3121         int exception;
3122
3123         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
3124         if (exception) {
3125                 /* FIXME: better handling */
3126                 return X86EMUL_UNHANDLEABLE;
3127         }
3128         return X86EMUL_CONTINUE;
3129 }
3130
3131 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3132 {
3133         u8 opcodes[4];
3134         unsigned long rip = kvm_rip_read(vcpu);
3135         unsigned long rip_linear;
3136
3137         if (!printk_ratelimit())
3138                 return;
3139
3140         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3141
3142         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
3143
3144         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3145                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3146 }
3147 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3148
3149 static struct x86_emulate_ops emulate_ops = {
3150         .read_std            = kvm_read_guest_virt,
3151         .read_emulated       = emulator_read_emulated,
3152         .write_emulated      = emulator_write_emulated,
3153         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3154 };
3155
3156 static void cache_all_regs(struct kvm_vcpu *vcpu)
3157 {
3158         kvm_register_read(vcpu, VCPU_REGS_RAX);
3159         kvm_register_read(vcpu, VCPU_REGS_RSP);
3160         kvm_register_read(vcpu, VCPU_REGS_RIP);
3161         vcpu->arch.regs_dirty = ~0;
3162 }
3163
3164 int emulate_instruction(struct kvm_vcpu *vcpu,
3165                         unsigned long cr2,
3166                         u16 error_code,
3167                         int emulation_type)
3168 {
3169         int r, shadow_mask;
3170         struct decode_cache *c;
3171         struct kvm_run *run = vcpu->run;
3172
3173         kvm_clear_exception_queue(vcpu);
3174         vcpu->arch.mmio_fault_cr2 = cr2;
3175         /*
3176          * TODO: fix emulate.c to use guest_read/write_register
3177          * instead of direct ->regs accesses, can save hundred cycles
3178          * on Intel for instructions that don't read/change RSP, for
3179          * for example.
3180          */
3181         cache_all_regs(vcpu);
3182
3183         vcpu->mmio_is_write = 0;
3184         vcpu->arch.pio.string = 0;
3185
3186         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3187                 int cs_db, cs_l;
3188                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3189
3190                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3191                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3192                 vcpu->arch.emulate_ctxt.mode =
3193                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3194                         ? X86EMUL_MODE_REAL : cs_l
3195                         ? X86EMUL_MODE_PROT64 : cs_db
3196                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3197
3198                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3199
3200                 /* Only allow emulation of specific instructions on #UD
3201                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3202                 c = &vcpu->arch.emulate_ctxt.decode;
3203                 if (emulation_type & EMULTYPE_TRAP_UD) {
3204                         if (!c->twobyte)
3205                                 return EMULATE_FAIL;
3206                         switch (c->b) {
3207                         case 0x01: /* VMMCALL */
3208                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3209                                         return EMULATE_FAIL;
3210                                 break;
3211                         case 0x34: /* sysenter */
3212                         case 0x35: /* sysexit */
3213                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3214                                         return EMULATE_FAIL;
3215                                 break;
3216                         case 0x05: /* syscall */
3217                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3218                                         return EMULATE_FAIL;
3219                                 break;
3220                         default:
3221                                 return EMULATE_FAIL;
3222                         }
3223
3224                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3225                                 return EMULATE_FAIL;
3226                 }
3227
3228                 ++vcpu->stat.insn_emulation;
3229                 if (r)  {
3230                         ++vcpu->stat.insn_emulation_fail;
3231                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3232                                 return EMULATE_DONE;
3233                         return EMULATE_FAIL;
3234                 }
3235         }
3236
3237         if (emulation_type & EMULTYPE_SKIP) {
3238                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3239                 return EMULATE_DONE;
3240         }
3241
3242         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3243         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3244
3245         if (r == 0)
3246                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3247
3248         if (vcpu->arch.pio.string)
3249                 return EMULATE_DO_MMIO;
3250
3251         if ((r || vcpu->mmio_is_write) && run) {
3252                 run->exit_reason = KVM_EXIT_MMIO;
3253                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3254                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3255                 run->mmio.len = vcpu->mmio_size;
3256                 run->mmio.is_write = vcpu->mmio_is_write;
3257         }
3258
3259         if (r) {
3260                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3261                         return EMULATE_DONE;
3262                 if (!vcpu->mmio_needed) {
3263                         kvm_report_emulation_failure(vcpu, "mmio");
3264                         return EMULATE_FAIL;
3265                 }
3266                 return EMULATE_DO_MMIO;
3267         }
3268
3269         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3270
3271         if (vcpu->mmio_is_write) {
3272                 vcpu->mmio_needed = 0;
3273                 return EMULATE_DO_MMIO;
3274         }
3275
3276         return EMULATE_DONE;
3277 }
3278 EXPORT_SYMBOL_GPL(emulate_instruction);
3279
3280 static int pio_copy_data(struct kvm_vcpu *vcpu)
3281 {
3282         void *p = vcpu->arch.pio_data;
3283         gva_t q = vcpu->arch.pio.guest_gva;
3284         unsigned bytes;
3285         int ret;
3286
3287         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3288         if (vcpu->arch.pio.in)
3289                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
3290         else
3291                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
3292         return ret;
3293 }
3294
3295 int complete_pio(struct kvm_vcpu *vcpu)
3296 {
3297         struct kvm_pio_request *io = &vcpu->arch.pio;
3298         long delta;
3299         int r;
3300         unsigned long val;
3301
3302         if (!io->string) {
3303                 if (io->in) {
3304                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3305                         memcpy(&val, vcpu->arch.pio_data, io->size);
3306                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3307                 }
3308         } else {
3309                 if (io->in) {
3310                         r = pio_copy_data(vcpu);
3311                         if (r)
3312                                 return r;
3313                 }
3314
3315                 delta = 1;
3316                 if (io->rep) {
3317                         delta *= io->cur_count;
3318                         /*
3319                          * The size of the register should really depend on
3320                          * current address size.
3321                          */
3322                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3323                         val -= delta;
3324                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3325                 }
3326                 if (io->down)
3327                         delta = -delta;
3328                 delta *= io->size;
3329                 if (io->in) {
3330                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3331                         val += delta;
3332                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3333                 } else {
3334                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3335                         val += delta;
3336                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3337                 }
3338         }
3339
3340         io->count -= io->cur_count;
3341         io->cur_count = 0;
3342
3343         return 0;
3344 }
3345
3346 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3347 {
3348         /* TODO: String I/O for in kernel device */
3349         int r;
3350
3351         if (vcpu->arch.pio.in)
3352                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3353                                     vcpu->arch.pio.size, pd);
3354         else
3355                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3356                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
3357                                      pd);
3358         return r;
3359 }
3360
3361 static int pio_string_write(struct kvm_vcpu *vcpu)
3362 {
3363         struct kvm_pio_request *io = &vcpu->arch.pio;
3364         void *pd = vcpu->arch.pio_data;
3365         int i, r = 0;
3366
3367         for (i = 0; i < io->cur_count; i++) {
3368                 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3369                                      io->port, io->size, pd)) {
3370                         r = -EOPNOTSUPP;
3371                         break;
3372                 }
3373                 pd += io->size;
3374         }
3375         return r;
3376 }
3377
3378 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3379 {
3380         unsigned long val;
3381
3382         vcpu->run->exit_reason = KVM_EXIT_IO;
3383         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3384         vcpu->run->io.size = vcpu->arch.pio.size = size;
3385         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3386         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3387         vcpu->run->io.port = vcpu->arch.pio.port = port;
3388         vcpu->arch.pio.in = in;
3389         vcpu->arch.pio.string = 0;
3390         vcpu->arch.pio.down = 0;
3391         vcpu->arch.pio.rep = 0;
3392
3393         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3394                       size, 1);
3395
3396         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3397         memcpy(vcpu->arch.pio_data, &val, 4);
3398
3399         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3400                 complete_pio(vcpu);
3401                 return 1;
3402         }
3403         return 0;
3404 }
3405 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3406
3407 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3408                   int size, unsigned long count, int down,
3409                   gva_t address, int rep, unsigned port)
3410 {
3411         unsigned now, in_page;
3412         int ret = 0;
3413
3414         vcpu->run->exit_reason = KVM_EXIT_IO;
3415         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3416         vcpu->run->io.size = vcpu->arch.pio.size = size;
3417         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3418         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3419         vcpu->run->io.port = vcpu->arch.pio.port = port;
3420         vcpu->arch.pio.in = in;
3421         vcpu->arch.pio.string = 1;
3422         vcpu->arch.pio.down = down;
3423         vcpu->arch.pio.rep = rep;
3424
3425         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3426                       size, count);
3427
3428         if (!count) {
3429                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3430                 return 1;
3431         }
3432
3433         if (!down)
3434                 in_page = PAGE_SIZE - offset_in_page(address);
3435         else
3436                 in_page = offset_in_page(address) + size;
3437         now = min(count, (unsigned long)in_page / size);
3438         if (!now)
3439                 now = 1;
3440         if (down) {
3441                 /*
3442                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3443                  */
3444                 pr_unimpl(vcpu, "guest string pio down\n");
3445                 kvm_inject_gp(vcpu, 0);
3446                 return 1;
3447         }
3448         vcpu->run->io.count = now;
3449         vcpu->arch.pio.cur_count = now;
3450
3451         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3452                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3453
3454         vcpu->arch.pio.guest_gva = address;
3455
3456         if (!vcpu->arch.pio.in) {
3457                 /* string PIO write */
3458                 ret = pio_copy_data(vcpu);
3459                 if (ret == X86EMUL_PROPAGATE_FAULT) {
3460                         kvm_inject_gp(vcpu, 0);
3461                         return 1;
3462                 }
3463                 if (ret == 0 && !pio_string_write(vcpu)) {
3464                         complete_pio(vcpu);
3465                         if (vcpu->arch.pio.count == 0)
3466                                 ret = 1;
3467                 }
3468         }
3469         /* no string PIO read support yet */
3470
3471         return ret;
3472 }
3473 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3474
3475 static void bounce_off(void *info)
3476 {
3477         /* nothing */
3478 }
3479
3480 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3481                                      void *data)
3482 {
3483         struct cpufreq_freqs *freq = data;
3484         struct kvm *kvm;
3485         struct kvm_vcpu *vcpu;
3486         int i, send_ipi = 0;
3487
3488         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3489                 return 0;
3490         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3491                 return 0;
3492         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3493
3494         spin_lock(&kvm_lock);
3495         list_for_each_entry(kvm, &vm_list, vm_list) {
3496                 kvm_for_each_vcpu(i, vcpu, kvm) {
3497                         if (vcpu->cpu != freq->cpu)
3498                                 continue;
3499                         if (!kvm_request_guest_time_update(vcpu))
3500                                 continue;
3501                         if (vcpu->cpu != smp_processor_id())
3502                                 send_ipi++;
3503                 }
3504         }
3505         spin_unlock(&kvm_lock);
3506
3507         if (freq->old < freq->new && send_ipi) {
3508                 /*
3509                  * We upscale the frequency.  Must make the guest
3510                  * doesn't see old kvmclock values while running with
3511                  * the new frequency, otherwise we risk the guest sees
3512                  * time go backwards.
3513                  *
3514                  * In case we update the frequency for another cpu
3515                  * (which might be in guest context) send an interrupt
3516                  * to kick the cpu out of guest context.  Next time
3517                  * guest context is entered kvmclock will be updated,
3518                  * so the guest will not see stale values.
3519                  */
3520                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3521         }
3522         return 0;
3523 }
3524
3525 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3526         .notifier_call  = kvmclock_cpufreq_notifier
3527 };
3528
3529 static void kvm_timer_init(void)
3530 {
3531         int cpu;
3532
3533         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3534                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3535                                           CPUFREQ_TRANSITION_NOTIFIER);
3536                 for_each_online_cpu(cpu) {
3537                         unsigned long khz = cpufreq_get(cpu);
3538                         if (!khz)
3539                                 khz = tsc_khz;
3540                         per_cpu(cpu_tsc_khz, cpu) = khz;
3541                 }
3542         } else {
3543                 for_each_possible_cpu(cpu)
3544                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3545         }
3546 }
3547
3548 int kvm_arch_init(void *opaque)
3549 {
3550         int r;
3551         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3552
3553         if (kvm_x86_ops) {
3554                 printk(KERN_ERR "kvm: already loaded the other module\n");
3555                 r = -EEXIST;
3556                 goto out;
3557         }
3558
3559         if (!ops->cpu_has_kvm_support()) {
3560                 printk(KERN_ERR "kvm: no hardware support\n");
3561                 r = -EOPNOTSUPP;
3562                 goto out;
3563         }
3564         if (ops->disabled_by_bios()) {
3565                 printk(KERN_ERR "kvm: disabled by bios\n");
3566                 r = -EOPNOTSUPP;
3567                 goto out;
3568         }
3569
3570         r = kvm_mmu_module_init();
3571         if (r)
3572                 goto out;
3573
3574         kvm_init_msr_list();
3575
3576         kvm_x86_ops = ops;
3577         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3578         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3579         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3580                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3581
3582         kvm_timer_init();
3583
3584         return 0;
3585
3586 out:
3587         return r;
3588 }
3589
3590 void kvm_arch_exit(void)
3591 {
3592         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3593                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3594                                             CPUFREQ_TRANSITION_NOTIFIER);
3595         kvm_x86_ops = NULL;
3596         kvm_mmu_module_exit();
3597 }
3598
3599 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3600 {
3601         ++vcpu->stat.halt_exits;
3602         if (irqchip_in_kernel(vcpu->kvm)) {
3603                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3604                 return 1;
3605         } else {
3606                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3607                 return 0;
3608         }
3609 }
3610 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3611
3612 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3613                            unsigned long a1)
3614 {
3615         if (is_long_mode(vcpu))
3616                 return a0;
3617         else
3618                 return a0 | ((gpa_t)a1 << 32);
3619 }
3620
3621 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3622 {
3623         unsigned long nr, a0, a1, a2, a3, ret;
3624         int r = 1;
3625
3626         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3627         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3628         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3629         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3630         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3631
3632         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3633
3634         if (!is_long_mode(vcpu)) {
3635                 nr &= 0xFFFFFFFF;
3636                 a0 &= 0xFFFFFFFF;
3637                 a1 &= 0xFFFFFFFF;
3638                 a2 &= 0xFFFFFFFF;
3639                 a3 &= 0xFFFFFFFF;
3640         }
3641
3642         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3643                 ret = -KVM_EPERM;
3644                 goto out;
3645         }
3646
3647         switch (nr) {
3648         case KVM_HC_VAPIC_POLL_IRQ:
3649                 ret = 0;
3650                 break;
3651         case KVM_HC_MMU_OP:
3652                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3653                 break;
3654         default:
3655                 ret = -KVM_ENOSYS;
3656                 break;
3657         }
3658 out:
3659         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3660         ++vcpu->stat.hypercalls;
3661         return r;
3662 }
3663 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3664
3665 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3666 {
3667         char instruction[3];
3668         int ret = 0;
3669         unsigned long rip = kvm_rip_read(vcpu);
3670
3671
3672         /*
3673          * Blow out the MMU to ensure that no other VCPU has an active mapping
3674          * to ensure that the updated hypercall appears atomically across all
3675          * VCPUs.
3676          */
3677         kvm_mmu_zap_all(vcpu->kvm);
3678
3679         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3680         if (emulator_write_emulated(rip, instruction, 3, vcpu)
3681             != X86EMUL_CONTINUE)
3682                 ret = -EFAULT;
3683
3684         return ret;
3685 }
3686
3687 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3688 {
3689         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3690 }
3691
3692 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3693 {
3694         struct descriptor_table dt = { limit, base };
3695
3696         kvm_x86_ops->set_gdt(vcpu, &dt);
3697 }
3698
3699 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3700 {
3701         struct descriptor_table dt = { limit, base };
3702
3703         kvm_x86_ops->set_idt(vcpu, &dt);
3704 }
3705
3706 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3707                    unsigned long *rflags)
3708 {
3709         kvm_lmsw(vcpu, msw);
3710         *rflags = kvm_get_rflags(vcpu);
3711 }
3712
3713 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3714 {
3715         unsigned long value;
3716
3717         switch (cr) {
3718         case 0:
3719                 value = kvm_read_cr0(vcpu);
3720                 break;
3721         case 2:
3722                 value = vcpu->arch.cr2;
3723                 break;
3724         case 3:
3725                 value = vcpu->arch.cr3;
3726                 break;
3727         case 4:
3728                 value = kvm_read_cr4(vcpu);
3729                 break;
3730         case 8:
3731                 value = kvm_get_cr8(vcpu);
3732                 break;
3733         default:
3734                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3735                 return 0;
3736         }
3737
3738         return value;
3739 }
3740
3741 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3742                      unsigned long *rflags)
3743 {
3744         switch (cr) {
3745         case 0:
3746                 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3747                 *rflags = kvm_get_rflags(vcpu);
3748                 break;
3749         case 2:
3750                 vcpu->arch.cr2 = val;
3751                 break;
3752         case 3:
3753                 kvm_set_cr3(vcpu, val);
3754                 break;
3755         case 4:
3756                 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3757                 break;
3758         case 8:
3759                 kvm_set_cr8(vcpu, val & 0xfUL);
3760                 break;
3761         default:
3762                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3763         }
3764 }
3765
3766 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3767 {
3768         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3769         int j, nent = vcpu->arch.cpuid_nent;
3770
3771         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3772         /* when no next entry is found, the current entry[i] is reselected */
3773         for (j = i + 1; ; j = (j + 1) % nent) {
3774                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3775                 if (ej->function == e->function) {
3776                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3777                         return j;
3778                 }
3779         }
3780         return 0; /* silence gcc, even though control never reaches here */
3781 }
3782
3783 /* find an entry with matching function, matching index (if needed), and that
3784  * should be read next (if it's stateful) */
3785 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3786         u32 function, u32 index)
3787 {
3788         if (e->function != function)
3789                 return 0;
3790         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3791                 return 0;
3792         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3793             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3794                 return 0;
3795         return 1;
3796 }
3797
3798 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3799                                               u32 function, u32 index)
3800 {
3801         int i;
3802         struct kvm_cpuid_entry2 *best = NULL;
3803
3804         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3805                 struct kvm_cpuid_entry2 *e;
3806
3807                 e = &vcpu->arch.cpuid_entries[i];
3808                 if (is_matching_cpuid_entry(e, function, index)) {
3809                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3810                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3811                         best = e;
3812                         break;
3813                 }
3814                 /*
3815                  * Both basic or both extended?
3816                  */
3817                 if (((e->function ^ function) & 0x80000000) == 0)
3818                         if (!best || e->function > best->function)
3819                                 best = e;
3820         }
3821         return best;
3822 }
3823 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
3824
3825 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3826 {
3827         struct kvm_cpuid_entry2 *best;
3828
3829         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3830         if (best)
3831                 return best->eax & 0xff;
3832         return 36;
3833 }
3834
3835 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3836 {
3837         u32 function, index;
3838         struct kvm_cpuid_entry2 *best;
3839
3840         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3841         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3842         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3843         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3844         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3845         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3846         best = kvm_find_cpuid_entry(vcpu, function, index);
3847         if (best) {
3848                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3849                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3850                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3851                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3852         }
3853         kvm_x86_ops->skip_emulated_instruction(vcpu);
3854         trace_kvm_cpuid(function,
3855                         kvm_register_read(vcpu, VCPU_REGS_RAX),
3856                         kvm_register_read(vcpu, VCPU_REGS_RBX),
3857                         kvm_register_read(vcpu, VCPU_REGS_RCX),
3858                         kvm_register_read(vcpu, VCPU_REGS_RDX));
3859 }
3860 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3861
3862 /*
3863  * Check if userspace requested an interrupt window, and that the
3864  * interrupt window is open.
3865  *
3866  * No need to exit to userspace if we already have an interrupt queued.
3867  */
3868 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
3869 {
3870         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3871                 vcpu->run->request_interrupt_window &&
3872                 kvm_arch_interrupt_allowed(vcpu));
3873 }
3874
3875 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
3876 {
3877         struct kvm_run *kvm_run = vcpu->run;
3878
3879         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3880         kvm_run->cr8 = kvm_get_cr8(vcpu);
3881         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3882         if (irqchip_in_kernel(vcpu->kvm))
3883                 kvm_run->ready_for_interrupt_injection = 1;
3884         else
3885                 kvm_run->ready_for_interrupt_injection =
3886                         kvm_arch_interrupt_allowed(vcpu) &&
3887                         !kvm_cpu_has_interrupt(vcpu) &&
3888                         !kvm_event_needs_reinjection(vcpu);
3889 }
3890
3891 static void vapic_enter(struct kvm_vcpu *vcpu)
3892 {
3893         struct kvm_lapic *apic = vcpu->arch.apic;
3894         struct page *page;
3895
3896         if (!apic || !apic->vapic_addr)
3897                 return;
3898
3899         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3900
3901         vcpu->arch.apic->vapic_page = page;
3902 }
3903
3904 static void vapic_exit(struct kvm_vcpu *vcpu)
3905 {
3906         struct kvm_lapic *apic = vcpu->arch.apic;
3907         int idx;
3908
3909         if (!apic || !apic->vapic_addr)
3910                 return;
3911
3912         idx = srcu_read_lock(&vcpu->kvm->srcu);
3913         kvm_release_page_dirty(apic->vapic_page);
3914         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3915         srcu_read_unlock(&vcpu->kvm->srcu, idx);
3916 }
3917
3918 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3919 {
3920         int max_irr, tpr;
3921
3922         if (!kvm_x86_ops->update_cr8_intercept)
3923                 return;
3924
3925         if (!vcpu->arch.apic)
3926                 return;
3927
3928         if (!vcpu->arch.apic->vapic_addr)
3929                 max_irr = kvm_lapic_find_highest_irr(vcpu);
3930         else
3931                 max_irr = -1;
3932
3933         if (max_irr != -1)
3934                 max_irr >>= 4;
3935
3936         tpr = kvm_lapic_get_cr8(vcpu);
3937
3938         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3939 }
3940
3941 static void inject_pending_event(struct kvm_vcpu *vcpu)
3942 {
3943         /* try to reinject previous events if any */
3944         if (vcpu->arch.exception.pending) {
3945                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3946                                           vcpu->arch.exception.has_error_code,
3947                                           vcpu->arch.exception.error_code);
3948                 return;
3949         }
3950
3951         if (vcpu->arch.nmi_injected) {
3952                 kvm_x86_ops->set_nmi(vcpu);
3953                 return;
3954         }
3955
3956         if (vcpu->arch.interrupt.pending) {
3957                 kvm_x86_ops->set_irq(vcpu);
3958                 return;
3959         }
3960
3961         /* try to inject new event if pending */
3962         if (vcpu->arch.nmi_pending) {
3963                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3964                         vcpu->arch.nmi_pending = false;
3965                         vcpu->arch.nmi_injected = true;
3966                         kvm_x86_ops->set_nmi(vcpu);
3967                 }
3968         } else if (kvm_cpu_has_interrupt(vcpu)) {
3969                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3970                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3971                                             false);
3972                         kvm_x86_ops->set_irq(vcpu);
3973                 }
3974         }
3975 }
3976
3977 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
3978 {
3979         int r;
3980         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3981                 vcpu->run->request_interrupt_window;
3982
3983         if (vcpu->requests)
3984                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3985                         kvm_mmu_unload(vcpu);
3986
3987         r = kvm_mmu_reload(vcpu);
3988         if (unlikely(r))
3989                 goto out;
3990
3991         if (vcpu->requests) {
3992                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3993                         __kvm_migrate_timers(vcpu);
3994                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3995                         kvm_write_guest_time(vcpu);
3996                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3997                         kvm_mmu_sync_roots(vcpu);
3998                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3999                         kvm_x86_ops->tlb_flush(vcpu);
4000                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4001                                        &vcpu->requests)) {
4002                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4003                         r = 0;
4004                         goto out;
4005                 }
4006                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4007                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4008                         r = 0;
4009                         goto out;
4010                 }
4011                 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4012                         vcpu->fpu_active = 0;
4013                         kvm_x86_ops->fpu_deactivate(vcpu);
4014                 }
4015         }
4016
4017         preempt_disable();
4018
4019         kvm_x86_ops->prepare_guest_switch(vcpu);
4020         kvm_load_guest_fpu(vcpu);
4021
4022         local_irq_disable();
4023
4024         clear_bit(KVM_REQ_KICK, &vcpu->requests);
4025         smp_mb__after_clear_bit();
4026
4027         if (vcpu->requests || need_resched() || signal_pending(current)) {
4028                 set_bit(KVM_REQ_KICK, &vcpu->requests);
4029                 local_irq_enable();
4030                 preempt_enable();
4031                 r = 1;
4032                 goto out;
4033         }
4034
4035         inject_pending_event(vcpu);
4036
4037         /* enable NMI/IRQ window open exits if needed */
4038         if (vcpu->arch.nmi_pending)
4039                 kvm_x86_ops->enable_nmi_window(vcpu);
4040         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4041                 kvm_x86_ops->enable_irq_window(vcpu);
4042
4043         if (kvm_lapic_enabled(vcpu)) {
4044                 update_cr8_intercept(vcpu);
4045                 kvm_lapic_sync_to_vapic(vcpu);
4046         }
4047
4048         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4049
4050         kvm_guest_enter();
4051
4052         if (unlikely(vcpu->arch.switch_db_regs)) {
4053                 set_debugreg(0, 7);
4054                 set_debugreg(vcpu->arch.eff_db[0], 0);
4055                 set_debugreg(vcpu->arch.eff_db[1], 1);
4056                 set_debugreg(vcpu->arch.eff_db[2], 2);
4057                 set_debugreg(vcpu->arch.eff_db[3], 3);
4058         }
4059
4060         trace_kvm_entry(vcpu->vcpu_id);
4061         kvm_x86_ops->run(vcpu);
4062
4063         /*
4064          * If the guest has used debug registers, at least dr7
4065          * will be disabled while returning to the host.
4066          * If we don't have active breakpoints in the host, we don't
4067          * care about the messed up debug address registers. But if
4068          * we have some of them active, restore the old state.
4069          */
4070         if (hw_breakpoint_active())
4071                 hw_breakpoint_restore();
4072
4073         set_bit(KVM_REQ_KICK, &vcpu->requests);
4074         local_irq_enable();
4075
4076         ++vcpu->stat.exits;
4077
4078         /*
4079          * We must have an instruction between local_irq_enable() and
4080          * kvm_guest_exit(), so the timer interrupt isn't delayed by
4081          * the interrupt shadow.  The stat.exits increment will do nicely.
4082          * But we need to prevent reordering, hence this barrier():
4083          */
4084         barrier();
4085
4086         kvm_guest_exit();
4087
4088         preempt_enable();
4089
4090         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4091
4092         /*
4093          * Profile KVM exit RIPs:
4094          */
4095         if (unlikely(prof_on == KVM_PROFILING)) {
4096                 unsigned long rip = kvm_rip_read(vcpu);
4097                 profile_hit(KVM_PROFILING, (void *)rip);
4098         }
4099
4100
4101         kvm_lapic_sync_from_vapic(vcpu);
4102
4103         r = kvm_x86_ops->handle_exit(vcpu);
4104 out:
4105         return r;
4106 }
4107
4108
4109 static int __vcpu_run(struct kvm_vcpu *vcpu)
4110 {
4111         int r;
4112         struct kvm *kvm = vcpu->kvm;
4113
4114         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4115                 pr_debug("vcpu %d received sipi with vector # %x\n",
4116                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
4117                 kvm_lapic_reset(vcpu);
4118                 r = kvm_arch_vcpu_reset(vcpu);
4119                 if (r)
4120                         return r;
4121                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4122         }
4123
4124         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4125         vapic_enter(vcpu);
4126
4127         r = 1;
4128         while (r > 0) {
4129                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4130                         r = vcpu_enter_guest(vcpu);
4131                 else {
4132                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4133                         kvm_vcpu_block(vcpu);
4134                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4135                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4136                         {
4137                                 switch(vcpu->arch.mp_state) {
4138                                 case KVM_MP_STATE_HALTED:
4139                                         vcpu->arch.mp_state =
4140                                                 KVM_MP_STATE_RUNNABLE;
4141                                 case KVM_MP_STATE_RUNNABLE:
4142                                         break;
4143                                 case KVM_MP_STATE_SIPI_RECEIVED:
4144                                 default:
4145                                         r = -EINTR;
4146                                         break;
4147                                 }
4148                         }
4149                 }
4150
4151                 if (r <= 0)
4152                         break;
4153
4154                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4155                 if (kvm_cpu_has_pending_timer(vcpu))
4156                         kvm_inject_pending_timer_irqs(vcpu);
4157
4158                 if (dm_request_for_irq_injection(vcpu)) {
4159                         r = -EINTR;
4160                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4161                         ++vcpu->stat.request_irq_exits;
4162                 }
4163                 if (signal_pending(current)) {
4164                         r = -EINTR;
4165                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4166                         ++vcpu->stat.signal_exits;
4167                 }
4168                 if (need_resched()) {
4169                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4170                         kvm_resched(vcpu);
4171                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4172                 }
4173         }
4174
4175         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4176         post_kvm_run_save(vcpu);
4177
4178         vapic_exit(vcpu);
4179
4180         return r;
4181 }
4182
4183 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4184 {
4185         int r;
4186         sigset_t sigsaved;
4187
4188         vcpu_load(vcpu);
4189
4190         if (vcpu->sigset_active)
4191                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4192
4193         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4194                 kvm_vcpu_block(vcpu);
4195                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4196                 r = -EAGAIN;
4197                 goto out;
4198         }
4199
4200         /* re-sync apic's tpr */
4201         if (!irqchip_in_kernel(vcpu->kvm))
4202                 kvm_set_cr8(vcpu, kvm_run->cr8);
4203
4204         if (vcpu->arch.pio.cur_count) {
4205                 r = complete_pio(vcpu);
4206                 if (r)
4207                         goto out;
4208         }
4209         if (vcpu->mmio_needed) {
4210                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4211                 vcpu->mmio_read_completed = 1;
4212                 vcpu->mmio_needed = 0;
4213
4214                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4215                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4216                                         EMULTYPE_NO_DECODE);
4217                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4218                 if (r == EMULATE_DO_MMIO) {
4219                         /*
4220                          * Read-modify-write.  Back to userspace.
4221                          */
4222                         r = 0;
4223                         goto out;
4224                 }
4225         }
4226         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4227                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4228                                      kvm_run->hypercall.ret);
4229
4230         r = __vcpu_run(vcpu);
4231
4232 out:
4233         if (vcpu->sigset_active)
4234                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4235
4236         vcpu_put(vcpu);
4237         return r;
4238 }
4239
4240 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4241 {
4242         vcpu_load(vcpu);
4243
4244         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4245         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4246         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4247         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4248         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4249         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4250         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4251         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4252 #ifdef CONFIG_X86_64
4253         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4254         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4255         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4256         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4257         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4258         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4259         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4260         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4261 #endif
4262
4263         regs->rip = kvm_rip_read(vcpu);
4264         regs->rflags = kvm_get_rflags(vcpu);
4265
4266         vcpu_put(vcpu);
4267
4268         return 0;
4269 }
4270
4271 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4272 {
4273         vcpu_load(vcpu);
4274
4275         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4276         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4277         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4278         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4279         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4280         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4281         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4282         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4283 #ifdef CONFIG_X86_64
4284         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4285         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4286         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4287         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4288         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4289         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4290         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4291         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4292 #endif
4293
4294         kvm_rip_write(vcpu, regs->rip);
4295         kvm_set_rflags(vcpu, regs->rflags);
4296
4297         vcpu->arch.exception.pending = false;
4298
4299         vcpu_put(vcpu);
4300
4301         return 0;
4302 }
4303
4304 void kvm_get_segment(struct kvm_vcpu *vcpu,
4305                      struct kvm_segment *var, int seg)
4306 {
4307         kvm_x86_ops->get_segment(vcpu, var, seg);
4308 }
4309
4310 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4311 {
4312         struct kvm_segment cs;
4313
4314         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4315         *db = cs.db;
4316         *l = cs.l;
4317 }
4318 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4319
4320 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4321                                   struct kvm_sregs *sregs)
4322 {
4323         struct descriptor_table dt;
4324
4325         vcpu_load(vcpu);
4326
4327         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4328         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4329         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4330         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4331         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4332         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4333
4334         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4335         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4336
4337         kvm_x86_ops->get_idt(vcpu, &dt);
4338         sregs->idt.limit = dt.limit;
4339         sregs->idt.base = dt.base;
4340         kvm_x86_ops->get_gdt(vcpu, &dt);
4341         sregs->gdt.limit = dt.limit;
4342         sregs->gdt.base = dt.base;
4343
4344         sregs->cr0 = kvm_read_cr0(vcpu);
4345         sregs->cr2 = vcpu->arch.cr2;
4346         sregs->cr3 = vcpu->arch.cr3;
4347         sregs->cr4 = kvm_read_cr4(vcpu);
4348         sregs->cr8 = kvm_get_cr8(vcpu);
4349         sregs->efer = vcpu->arch.shadow_efer;
4350         sregs->apic_base = kvm_get_apic_base(vcpu);
4351
4352         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4353
4354         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4355                 set_bit(vcpu->arch.interrupt.nr,
4356                         (unsigned long *)sregs->interrupt_bitmap);
4357
4358         vcpu_put(vcpu);
4359
4360         return 0;
4361 }
4362
4363 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4364                                     struct kvm_mp_state *mp_state)
4365 {
4366         vcpu_load(vcpu);
4367         mp_state->mp_state = vcpu->arch.mp_state;
4368         vcpu_put(vcpu);
4369         return 0;
4370 }
4371
4372 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4373                                     struct kvm_mp_state *mp_state)
4374 {
4375         vcpu_load(vcpu);
4376         vcpu->arch.mp_state = mp_state->mp_state;
4377         vcpu_put(vcpu);
4378         return 0;
4379 }
4380
4381 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4382                         struct kvm_segment *var, int seg)
4383 {
4384         kvm_x86_ops->set_segment(vcpu, var, seg);
4385 }
4386
4387 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4388                                    struct kvm_segment *kvm_desct)
4389 {
4390         kvm_desct->base = get_desc_base(seg_desc);
4391         kvm_desct->limit = get_desc_limit(seg_desc);
4392         if (seg_desc->g) {
4393                 kvm_desct->limit <<= 12;
4394                 kvm_desct->limit |= 0xfff;
4395         }
4396         kvm_desct->selector = selector;
4397         kvm_desct->type = seg_desc->type;
4398         kvm_desct->present = seg_desc->p;
4399         kvm_desct->dpl = seg_desc->dpl;
4400         kvm_desct->db = seg_desc->d;
4401         kvm_desct->s = seg_desc->s;
4402         kvm_desct->l = seg_desc->l;
4403         kvm_desct->g = seg_desc->g;
4404         kvm_desct->avl = seg_desc->avl;
4405         if (!selector)
4406                 kvm_desct->unusable = 1;
4407         else
4408                 kvm_desct->unusable = 0;
4409         kvm_desct->padding = 0;
4410 }
4411
4412 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4413                                           u16 selector,
4414                                           struct descriptor_table *dtable)
4415 {
4416         if (selector & 1 << 2) {
4417                 struct kvm_segment kvm_seg;
4418
4419                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4420
4421                 if (kvm_seg.unusable)
4422                         dtable->limit = 0;
4423                 else
4424                         dtable->limit = kvm_seg.limit;
4425                 dtable->base = kvm_seg.base;
4426         }
4427         else
4428                 kvm_x86_ops->get_gdt(vcpu, dtable);
4429 }
4430
4431 /* allowed just for 8 bytes segments */
4432 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4433                                          struct desc_struct *seg_desc)
4434 {
4435         struct descriptor_table dtable;
4436         u16 index = selector >> 3;
4437
4438         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4439
4440         if (dtable.limit < index * 8 + 7) {
4441                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4442                 return 1;
4443         }
4444         return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4445 }
4446
4447 /* allowed just for 8 bytes segments */
4448 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4449                                          struct desc_struct *seg_desc)
4450 {
4451         struct descriptor_table dtable;
4452         u16 index = selector >> 3;
4453
4454         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4455
4456         if (dtable.limit < index * 8 + 7)
4457                 return 1;
4458         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4459 }
4460
4461 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4462                              struct desc_struct *seg_desc)
4463 {
4464         u32 base_addr = get_desc_base(seg_desc);
4465
4466         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4467 }
4468
4469 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4470 {
4471         struct kvm_segment kvm_seg;
4472
4473         kvm_get_segment(vcpu, &kvm_seg, seg);
4474         return kvm_seg.selector;
4475 }
4476
4477 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4478                                                 u16 selector,
4479                                                 struct kvm_segment *kvm_seg)
4480 {
4481         struct desc_struct seg_desc;
4482
4483         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4484                 return 1;
4485         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4486         return 0;
4487 }
4488
4489 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4490 {
4491         struct kvm_segment segvar = {
4492                 .base = selector << 4,
4493                 .limit = 0xffff,
4494                 .selector = selector,
4495                 .type = 3,
4496                 .present = 1,
4497                 .dpl = 3,
4498                 .db = 0,
4499                 .s = 1,
4500                 .l = 0,
4501                 .g = 0,
4502                 .avl = 0,
4503                 .unusable = 0,
4504         };
4505         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4506         return 0;
4507 }
4508
4509 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4510 {
4511         return (seg != VCPU_SREG_LDTR) &&
4512                 (seg != VCPU_SREG_TR) &&
4513                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4514 }
4515
4516 static void kvm_check_segment_descriptor(struct kvm_vcpu *vcpu, int seg,
4517                                          u16 selector)
4518 {
4519         /* NULL selector is not valid for CS and SS */
4520         if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4521                 if (!selector)
4522                         kvm_queue_exception_e(vcpu, TS_VECTOR, selector >> 3);
4523 }
4524
4525 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4526                                 int type_bits, int seg)
4527 {
4528         struct kvm_segment kvm_seg;
4529
4530         if (is_vm86_segment(vcpu, seg) || !(kvm_read_cr0_bits(vcpu, X86_CR0_PE)))
4531                 return kvm_load_realmode_segment(vcpu, selector, seg);
4532         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4533                 return 1;
4534
4535         kvm_check_segment_descriptor(vcpu, seg, selector);
4536         kvm_seg.type |= type_bits;
4537
4538         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4539             seg != VCPU_SREG_LDTR)
4540                 if (!kvm_seg.s)
4541                         kvm_seg.unusable = 1;
4542
4543         kvm_set_segment(vcpu, &kvm_seg, seg);
4544         return 0;
4545 }
4546
4547 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4548                                 struct tss_segment_32 *tss)
4549 {
4550         tss->cr3 = vcpu->arch.cr3;
4551         tss->eip = kvm_rip_read(vcpu);
4552         tss->eflags = kvm_get_rflags(vcpu);
4553         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4554         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4555         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4556         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4557         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4558         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4559         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4560         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4561         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4562         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4563         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4564         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4565         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4566         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4567         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4568 }
4569
4570 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4571                                   struct tss_segment_32 *tss)
4572 {
4573         kvm_set_cr3(vcpu, tss->cr3);
4574
4575         kvm_rip_write(vcpu, tss->eip);
4576         kvm_set_rflags(vcpu, tss->eflags | 2);
4577
4578         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4579         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4580         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4581         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4582         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4583         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4584         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4585         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4586
4587         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4588                 return 1;
4589
4590         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4591                 return 1;
4592
4593         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4594                 return 1;
4595
4596         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4597                 return 1;
4598
4599         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4600                 return 1;
4601
4602         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4603                 return 1;
4604
4605         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4606                 return 1;
4607         return 0;
4608 }
4609
4610 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4611                                 struct tss_segment_16 *tss)
4612 {
4613         tss->ip = kvm_rip_read(vcpu);
4614         tss->flag = kvm_get_rflags(vcpu);
4615         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4616         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4617         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4618         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4619         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4620         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4621         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4622         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4623
4624         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4625         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4626         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4627         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4628         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4629 }
4630
4631 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4632                                  struct tss_segment_16 *tss)
4633 {
4634         kvm_rip_write(vcpu, tss->ip);
4635         kvm_set_rflags(vcpu, tss->flag | 2);
4636         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4637         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4638         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4639         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4640         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4641         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4642         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4643         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4644
4645         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4646                 return 1;
4647
4648         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4649                 return 1;
4650
4651         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4652                 return 1;
4653
4654         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4655                 return 1;
4656
4657         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4658                 return 1;
4659         return 0;
4660 }
4661
4662 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4663                               u16 old_tss_sel, u32 old_tss_base,
4664                               struct desc_struct *nseg_desc)
4665 {
4666         struct tss_segment_16 tss_segment_16;
4667         int ret = 0;
4668
4669         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4670                            sizeof tss_segment_16))
4671                 goto out;
4672
4673         save_state_to_tss16(vcpu, &tss_segment_16);
4674
4675         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4676                             sizeof tss_segment_16))
4677                 goto out;
4678
4679         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4680                            &tss_segment_16, sizeof tss_segment_16))
4681                 goto out;
4682
4683         if (old_tss_sel != 0xffff) {
4684                 tss_segment_16.prev_task_link = old_tss_sel;
4685
4686                 if (kvm_write_guest(vcpu->kvm,
4687                                     get_tss_base_addr(vcpu, nseg_desc),
4688                                     &tss_segment_16.prev_task_link,
4689                                     sizeof tss_segment_16.prev_task_link))
4690                         goto out;
4691         }
4692
4693         if (load_state_from_tss16(vcpu, &tss_segment_16))
4694                 goto out;
4695
4696         ret = 1;
4697 out:
4698         return ret;
4699 }
4700
4701 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4702                        u16 old_tss_sel, u32 old_tss_base,
4703                        struct desc_struct *nseg_desc)
4704 {
4705         struct tss_segment_32 tss_segment_32;
4706         int ret = 0;
4707
4708         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4709                            sizeof tss_segment_32))
4710                 goto out;
4711
4712         save_state_to_tss32(vcpu, &tss_segment_32);
4713
4714         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4715                             sizeof tss_segment_32))
4716                 goto out;
4717
4718         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4719                            &tss_segment_32, sizeof tss_segment_32))
4720                 goto out;
4721
4722         if (old_tss_sel != 0xffff) {
4723                 tss_segment_32.prev_task_link = old_tss_sel;
4724
4725                 if (kvm_write_guest(vcpu->kvm,
4726                                     get_tss_base_addr(vcpu, nseg_desc),
4727                                     &tss_segment_32.prev_task_link,
4728                                     sizeof tss_segment_32.prev_task_link))
4729                         goto out;
4730         }
4731
4732         if (load_state_from_tss32(vcpu, &tss_segment_32))
4733                 goto out;
4734
4735         ret = 1;
4736 out:
4737         return ret;
4738 }
4739
4740 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4741 {
4742         struct kvm_segment tr_seg;
4743         struct desc_struct cseg_desc;
4744         struct desc_struct nseg_desc;
4745         int ret = 0;
4746         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4747         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4748
4749         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4750
4751         /* FIXME: Handle errors. Failure to read either TSS or their
4752          * descriptors should generate a pagefault.
4753          */
4754         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4755                 goto out;
4756
4757         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4758                 goto out;
4759
4760         if (reason != TASK_SWITCH_IRET) {
4761                 int cpl;
4762
4763                 cpl = kvm_x86_ops->get_cpl(vcpu);
4764                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4765                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4766                         return 1;
4767                 }
4768         }
4769
4770         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4771                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4772                 return 1;
4773         }
4774
4775         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4776                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4777                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4778         }
4779
4780         if (reason == TASK_SWITCH_IRET) {
4781                 u32 eflags = kvm_get_rflags(vcpu);
4782                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4783         }
4784
4785         /* set back link to prev task only if NT bit is set in eflags
4786            note that old_tss_sel is not used afetr this point */
4787         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4788                 old_tss_sel = 0xffff;
4789
4790         if (nseg_desc.type & 8)
4791                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4792                                          old_tss_base, &nseg_desc);
4793         else
4794                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4795                                          old_tss_base, &nseg_desc);
4796
4797         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4798                 u32 eflags = kvm_get_rflags(vcpu);
4799                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4800         }
4801
4802         if (reason != TASK_SWITCH_IRET) {
4803                 nseg_desc.type |= (1 << 1);
4804                 save_guest_segment_descriptor(vcpu, tss_selector,
4805                                               &nseg_desc);
4806         }
4807
4808         kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
4809         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4810         tr_seg.type = 11;
4811         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4812 out:
4813         return ret;
4814 }
4815 EXPORT_SYMBOL_GPL(kvm_task_switch);
4816
4817 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4818                                   struct kvm_sregs *sregs)
4819 {
4820         int mmu_reset_needed = 0;
4821         int pending_vec, max_bits;
4822         struct descriptor_table dt;
4823
4824         vcpu_load(vcpu);
4825
4826         dt.limit = sregs->idt.limit;
4827         dt.base = sregs->idt.base;
4828         kvm_x86_ops->set_idt(vcpu, &dt);
4829         dt.limit = sregs->gdt.limit;
4830         dt.base = sregs->gdt.base;
4831         kvm_x86_ops->set_gdt(vcpu, &dt);
4832
4833         vcpu->arch.cr2 = sregs->cr2;
4834         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4835         vcpu->arch.cr3 = sregs->cr3;
4836
4837         kvm_set_cr8(vcpu, sregs->cr8);
4838
4839         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4840         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4841         kvm_set_apic_base(vcpu, sregs->apic_base);
4842
4843         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
4844         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4845         vcpu->arch.cr0 = sregs->cr0;
4846
4847         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
4848         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4849         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
4850                 load_pdptrs(vcpu, vcpu->arch.cr3);
4851                 mmu_reset_needed = 1;
4852         }
4853
4854         if (mmu_reset_needed)
4855                 kvm_mmu_reset_context(vcpu);
4856
4857         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4858         pending_vec = find_first_bit(
4859                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4860         if (pending_vec < max_bits) {
4861                 kvm_queue_interrupt(vcpu, pending_vec, false);
4862                 pr_debug("Set back pending irq %d\n", pending_vec);
4863                 if (irqchip_in_kernel(vcpu->kvm))
4864                         kvm_pic_clear_isr_ack(vcpu->kvm);
4865         }
4866
4867         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4868         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4869         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4870         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4871         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4872         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4873
4874         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4875         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4876
4877         update_cr8_intercept(vcpu);
4878
4879         /* Older userspace won't unhalt the vcpu on reset. */
4880         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4881             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4882             !(kvm_read_cr0_bits(vcpu, X86_CR0_PE)))
4883                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4884
4885         vcpu_put(vcpu);
4886
4887         return 0;
4888 }
4889
4890 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4891                                         struct kvm_guest_debug *dbg)
4892 {
4893         unsigned long rflags;
4894         int i, r;
4895
4896         vcpu_load(vcpu);
4897
4898         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
4899                 r = -EBUSY;
4900                 if (vcpu->arch.exception.pending)
4901                         goto unlock_out;
4902                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4903                         kvm_queue_exception(vcpu, DB_VECTOR);
4904                 else
4905                         kvm_queue_exception(vcpu, BP_VECTOR);
4906         }
4907
4908         /*
4909          * Read rflags as long as potentially injected trace flags are still
4910          * filtered out.
4911          */
4912         rflags = kvm_get_rflags(vcpu);
4913
4914         vcpu->guest_debug = dbg->control;
4915         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
4916                 vcpu->guest_debug = 0;
4917
4918         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4919                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4920                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4921                 vcpu->arch.switch_db_regs =
4922                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4923         } else {
4924                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4925                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4926                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4927         }
4928
4929         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
4930                 vcpu->arch.singlestep_cs =
4931                         get_segment_selector(vcpu, VCPU_SREG_CS);
4932                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
4933         }
4934
4935         /*
4936          * Trigger an rflags update that will inject or remove the trace
4937          * flags.
4938          */
4939         kvm_set_rflags(vcpu, rflags);
4940
4941         kvm_x86_ops->set_guest_debug(vcpu, dbg);
4942
4943         r = 0;
4944
4945 unlock_out:
4946         vcpu_put(vcpu);
4947
4948         return r;
4949 }
4950
4951 /*
4952  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4953  * we have asm/x86/processor.h
4954  */
4955 struct fxsave {
4956         u16     cwd;
4957         u16     swd;
4958         u16     twd;
4959         u16     fop;
4960         u64     rip;
4961         u64     rdp;
4962         u32     mxcsr;
4963         u32     mxcsr_mask;
4964         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4965 #ifdef CONFIG_X86_64
4966         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4967 #else
4968         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4969 #endif
4970 };
4971
4972 /*
4973  * Translate a guest virtual address to a guest physical address.
4974  */
4975 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4976                                     struct kvm_translation *tr)
4977 {
4978         unsigned long vaddr = tr->linear_address;
4979         gpa_t gpa;
4980         int idx;
4981
4982         vcpu_load(vcpu);
4983         idx = srcu_read_lock(&vcpu->kvm->srcu);
4984         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4985         srcu_read_unlock(&vcpu->kvm->srcu, idx);
4986         tr->physical_address = gpa;
4987         tr->valid = gpa != UNMAPPED_GVA;
4988         tr->writeable = 1;
4989         tr->usermode = 0;
4990         vcpu_put(vcpu);
4991
4992         return 0;
4993 }
4994
4995 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4996 {
4997         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4998
4999         vcpu_load(vcpu);
5000
5001         memcpy(fpu->fpr, fxsave->st_space, 128);
5002         fpu->fcw = fxsave->cwd;
5003         fpu->fsw = fxsave->swd;
5004         fpu->ftwx = fxsave->twd;
5005         fpu->last_opcode = fxsave->fop;
5006         fpu->last_ip = fxsave->rip;
5007         fpu->last_dp = fxsave->rdp;
5008         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5009
5010         vcpu_put(vcpu);
5011
5012         return 0;
5013 }
5014
5015 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5016 {
5017         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5018
5019         vcpu_load(vcpu);
5020
5021         memcpy(fxsave->st_space, fpu->fpr, 128);
5022         fxsave->cwd = fpu->fcw;
5023         fxsave->swd = fpu->fsw;
5024         fxsave->twd = fpu->ftwx;
5025         fxsave->fop = fpu->last_opcode;
5026         fxsave->rip = fpu->last_ip;
5027         fxsave->rdp = fpu->last_dp;
5028         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5029
5030         vcpu_put(vcpu);
5031
5032         return 0;
5033 }
5034
5035 void fx_init(struct kvm_vcpu *vcpu)
5036 {
5037         unsigned after_mxcsr_mask;
5038
5039         /*
5040          * Touch the fpu the first time in non atomic context as if
5041          * this is the first fpu instruction the exception handler
5042          * will fire before the instruction returns and it'll have to
5043          * allocate ram with GFP_KERNEL.
5044          */
5045         if (!used_math())
5046                 kvm_fx_save(&vcpu->arch.host_fx_image);
5047
5048         /* Initialize guest FPU by resetting ours and saving into guest's */
5049         preempt_disable();
5050         kvm_fx_save(&vcpu->arch.host_fx_image);
5051         kvm_fx_finit();
5052         kvm_fx_save(&vcpu->arch.guest_fx_image);
5053         kvm_fx_restore(&vcpu->arch.host_fx_image);
5054         preempt_enable();
5055
5056         vcpu->arch.cr0 |= X86_CR0_ET;
5057         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5058         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5059         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5060                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5061 }
5062 EXPORT_SYMBOL_GPL(fx_init);
5063
5064 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5065 {
5066         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
5067                 return;
5068
5069         vcpu->guest_fpu_loaded = 1;
5070         kvm_fx_save(&vcpu->arch.host_fx_image);
5071         kvm_fx_restore(&vcpu->arch.guest_fx_image);
5072 }
5073 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
5074
5075 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5076 {
5077         if (!vcpu->guest_fpu_loaded)
5078                 return;
5079
5080         vcpu->guest_fpu_loaded = 0;
5081         kvm_fx_save(&vcpu->arch.guest_fx_image);
5082         kvm_fx_restore(&vcpu->arch.host_fx_image);
5083         ++vcpu->stat.fpu_reload;
5084         set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5085 }
5086 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
5087
5088 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5089 {
5090         if (vcpu->arch.time_page) {
5091                 kvm_release_page_dirty(vcpu->arch.time_page);
5092                 vcpu->arch.time_page = NULL;
5093         }
5094
5095         kvm_x86_ops->vcpu_free(vcpu);
5096 }
5097
5098 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5099                                                 unsigned int id)
5100 {
5101         return kvm_x86_ops->vcpu_create(kvm, id);
5102 }
5103
5104 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5105 {
5106         int r;
5107
5108         /* We do fxsave: this must be aligned. */
5109         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5110
5111         vcpu->arch.mtrr_state.have_fixed = 1;
5112         vcpu_load(vcpu);
5113         r = kvm_arch_vcpu_reset(vcpu);
5114         if (r == 0)
5115                 r = kvm_mmu_setup(vcpu);
5116         vcpu_put(vcpu);
5117         if (r < 0)
5118                 goto free_vcpu;
5119
5120         return 0;
5121 free_vcpu:
5122         kvm_x86_ops->vcpu_free(vcpu);
5123         return r;
5124 }
5125
5126 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5127 {
5128         vcpu_load(vcpu);
5129         kvm_mmu_unload(vcpu);
5130         vcpu_put(vcpu);
5131
5132         kvm_x86_ops->vcpu_free(vcpu);
5133 }
5134
5135 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5136 {
5137         vcpu->arch.nmi_pending = false;
5138         vcpu->arch.nmi_injected = false;
5139
5140         vcpu->arch.switch_db_regs = 0;
5141         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5142         vcpu->arch.dr6 = DR6_FIXED_1;
5143         vcpu->arch.dr7 = DR7_FIXED_1;
5144
5145         return kvm_x86_ops->vcpu_reset(vcpu);
5146 }
5147
5148 int kvm_arch_hardware_enable(void *garbage)
5149 {
5150         /*
5151          * Since this may be called from a hotplug notifcation,
5152          * we can't get the CPU frequency directly.
5153          */
5154         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5155                 int cpu = raw_smp_processor_id();
5156                 per_cpu(cpu_tsc_khz, cpu) = 0;
5157         }
5158
5159         kvm_shared_msr_cpu_online();
5160
5161         return kvm_x86_ops->hardware_enable(garbage);
5162 }
5163
5164 void kvm_arch_hardware_disable(void *garbage)
5165 {
5166         kvm_x86_ops->hardware_disable(garbage);
5167         drop_user_return_notifiers(garbage);
5168 }
5169
5170 int kvm_arch_hardware_setup(void)
5171 {
5172         return kvm_x86_ops->hardware_setup();
5173 }
5174
5175 void kvm_arch_hardware_unsetup(void)
5176 {
5177         kvm_x86_ops->hardware_unsetup();
5178 }
5179
5180 void kvm_arch_check_processor_compat(void *rtn)
5181 {
5182         kvm_x86_ops->check_processor_compatibility(rtn);
5183 }
5184
5185 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5186 {
5187         struct page *page;
5188         struct kvm *kvm;
5189         int r;
5190
5191         BUG_ON(vcpu->kvm == NULL);
5192         kvm = vcpu->kvm;
5193
5194         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5195         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5196                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5197         else
5198                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5199
5200         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5201         if (!page) {
5202                 r = -ENOMEM;
5203                 goto fail;
5204         }
5205         vcpu->arch.pio_data = page_address(page);
5206
5207         r = kvm_mmu_create(vcpu);
5208         if (r < 0)
5209                 goto fail_free_pio_data;
5210
5211         if (irqchip_in_kernel(kvm)) {
5212                 r = kvm_create_lapic(vcpu);
5213                 if (r < 0)
5214                         goto fail_mmu_destroy;
5215         }
5216
5217         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5218                                        GFP_KERNEL);
5219         if (!vcpu->arch.mce_banks) {
5220                 r = -ENOMEM;
5221                 goto fail_free_lapic;
5222         }
5223         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5224
5225         return 0;
5226 fail_free_lapic:
5227         kvm_free_lapic(vcpu);
5228 fail_mmu_destroy:
5229         kvm_mmu_destroy(vcpu);
5230 fail_free_pio_data:
5231         free_page((unsigned long)vcpu->arch.pio_data);
5232 fail:
5233         return r;
5234 }
5235
5236 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5237 {
5238         int idx;
5239
5240         kfree(vcpu->arch.mce_banks);
5241         kvm_free_lapic(vcpu);
5242         idx = srcu_read_lock(&vcpu->kvm->srcu);
5243         kvm_mmu_destroy(vcpu);
5244         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5245         free_page((unsigned long)vcpu->arch.pio_data);
5246 }
5247
5248 struct  kvm *kvm_arch_create_vm(void)
5249 {
5250         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5251
5252         if (!kvm)
5253                 return ERR_PTR(-ENOMEM);
5254
5255         kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5256         if (!kvm->arch.aliases) {
5257                 kfree(kvm);
5258                 return ERR_PTR(-ENOMEM);
5259         }
5260
5261         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5262         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5263
5264         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5265         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5266
5267         rdtscll(kvm->arch.vm_init_tsc);
5268
5269         return kvm;
5270 }
5271
5272 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5273 {
5274         vcpu_load(vcpu);
5275         kvm_mmu_unload(vcpu);
5276         vcpu_put(vcpu);
5277 }
5278
5279 static void kvm_free_vcpus(struct kvm *kvm)
5280 {
5281         unsigned int i;
5282         struct kvm_vcpu *vcpu;
5283
5284         /*
5285          * Unpin any mmu pages first.
5286          */
5287         kvm_for_each_vcpu(i, vcpu, kvm)
5288                 kvm_unload_vcpu_mmu(vcpu);
5289         kvm_for_each_vcpu(i, vcpu, kvm)
5290                 kvm_arch_vcpu_free(vcpu);
5291
5292         mutex_lock(&kvm->lock);
5293         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5294                 kvm->vcpus[i] = NULL;
5295
5296         atomic_set(&kvm->online_vcpus, 0);
5297         mutex_unlock(&kvm->lock);
5298 }
5299
5300 void kvm_arch_sync_events(struct kvm *kvm)
5301 {
5302         kvm_free_all_assigned_devices(kvm);
5303 }
5304
5305 void kvm_arch_destroy_vm(struct kvm *kvm)
5306 {
5307         kvm_iommu_unmap_guest(kvm);
5308         kvm_free_pit(kvm);
5309         kfree(kvm->arch.vpic);
5310         kfree(kvm->arch.vioapic);
5311         kvm_free_vcpus(kvm);
5312         kvm_free_physmem(kvm);
5313         if (kvm->arch.apic_access_page)
5314                 put_page(kvm->arch.apic_access_page);
5315         if (kvm->arch.ept_identity_pagetable)
5316                 put_page(kvm->arch.ept_identity_pagetable);
5317         kfree(kvm->arch.aliases);
5318         kfree(kvm);
5319 }
5320
5321 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5322                                 struct kvm_memory_slot *memslot,
5323                                 struct kvm_memory_slot old,
5324                                 struct kvm_userspace_memory_region *mem,
5325                                 int user_alloc)
5326 {
5327         int npages = memslot->npages;
5328
5329         /*To keep backward compatibility with older userspace,
5330          *x86 needs to hanlde !user_alloc case.
5331          */
5332         if (!user_alloc) {
5333                 if (npages && !old.rmap) {
5334                         unsigned long userspace_addr;
5335
5336                         down_write(&current->mm->mmap_sem);
5337                         userspace_addr = do_mmap(NULL, 0,
5338                                                  npages * PAGE_SIZE,
5339                                                  PROT_READ | PROT_WRITE,
5340                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5341                                                  0);
5342                         up_write(&current->mm->mmap_sem);
5343
5344                         if (IS_ERR((void *)userspace_addr))
5345                                 return PTR_ERR((void *)userspace_addr);
5346
5347                         memslot->userspace_addr = userspace_addr;
5348                 }
5349         }
5350
5351
5352         return 0;
5353 }
5354
5355 void kvm_arch_commit_memory_region(struct kvm *kvm,
5356                                 struct kvm_userspace_memory_region *mem,
5357                                 struct kvm_memory_slot old,
5358                                 int user_alloc)
5359 {
5360
5361         int npages = mem->memory_size >> PAGE_SHIFT;
5362
5363         if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5364                 int ret;
5365
5366                 down_write(&current->mm->mmap_sem);
5367                 ret = do_munmap(current->mm, old.userspace_addr,
5368                                 old.npages * PAGE_SIZE);
5369                 up_write(&current->mm->mmap_sem);
5370                 if (ret < 0)
5371                         printk(KERN_WARNING
5372                                "kvm_vm_ioctl_set_memory_region: "
5373                                "failed to munmap memory\n");
5374         }
5375
5376         spin_lock(&kvm->mmu_lock);
5377         if (!kvm->arch.n_requested_mmu_pages) {
5378                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5379                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5380         }
5381
5382         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5383         spin_unlock(&kvm->mmu_lock);
5384 }
5385
5386 void kvm_arch_flush_shadow(struct kvm *kvm)
5387 {
5388         kvm_mmu_zap_all(kvm);
5389         kvm_reload_remote_mmus(kvm);
5390 }
5391
5392 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5393 {
5394         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5395                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5396                 || vcpu->arch.nmi_pending ||
5397                 (kvm_arch_interrupt_allowed(vcpu) &&
5398                  kvm_cpu_has_interrupt(vcpu));
5399 }
5400
5401 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5402 {
5403         int me;
5404         int cpu = vcpu->cpu;
5405
5406         if (waitqueue_active(&vcpu->wq)) {
5407                 wake_up_interruptible(&vcpu->wq);
5408                 ++vcpu->stat.halt_wakeup;
5409         }
5410
5411         me = get_cpu();
5412         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5413                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5414                         smp_send_reschedule(cpu);
5415         put_cpu();
5416 }
5417
5418 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5419 {
5420         return kvm_x86_ops->interrupt_allowed(vcpu);
5421 }
5422
5423 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5424 {
5425         unsigned long rflags;
5426
5427         rflags = kvm_x86_ops->get_rflags(vcpu);
5428         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5429                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5430         return rflags;
5431 }
5432 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5433
5434 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5435 {
5436         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5437             vcpu->arch.singlestep_cs ==
5438                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5439             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5440                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5441         kvm_x86_ops->set_rflags(vcpu, rflags);
5442 }
5443 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5444
5445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5450 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5451 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5452 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5453 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5454 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5455 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);