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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * derived from drivers/kvm/kvm_main.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *   Amit Shah    <amit.shah@qumranet.com>
  14  *   Ben-Ami Yassour <benami@il.ibm.com>
  15  *
  16  * This work is licensed under the terms of the GNU GPL, version 2.  See
  17  * the COPYING file in the top-level directory.
  18  *
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "i8254.h"
  25 #include "tss.h"
  26 #include "kvm_cache_regs.h"
  27 #include "x86.h"
  28
  29 #include <linux/clocksource.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/kvm.h>
  32 #include <linux/fs.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/module.h>
  35 #include <linux/mman.h>
  36 #include <linux/highmem.h>
  37 #include <linux/iommu.h>
  38 #include <linux/intel-iommu.h>
  39 #include <linux/cpufreq.h>
  40 #define CREATE_TRACE_POINTS
  41 #include "trace.h"
  42
  43 #include <asm/uaccess.h>
  44 #include <asm/msr.h>
  45 #include <asm/desc.h>
  46 #include <asm/mtrr.h>
  47 #include <asm/mce.h>
  48
  49 #define MAX_IO_MSRS 256
  50 #define CR0_RESERVED_BITS                                               \
  51         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  52                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  53                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  54 #define CR4_RESERVED_BITS                                               \
  55         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  56                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  57                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  58                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  59
  60 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  61
  62 #define KVM_MAX_MCE_BANKS 32
  63 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
  64
  65 /* EFER defaults:
  66  * - enable syscall per default because its emulated by KVM
  67  * - enable LME and LMA per default on 64 bit KVM
  68  */
  69 #ifdef CONFIG_X86_64
  70 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  71 #else
  72 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  73 #endif
  74
  75 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  76 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  77
  78 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  79                                     struct kvm_cpuid_entry2 __user *entries);
  80 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
  81                                               u32 function, u32 index);
  82
  83 struct kvm_x86_ops *kvm_x86_ops;
  84 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  85
  86 int ignore_msrs = 0;
  87 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
  88
  89 struct kvm_stats_debugfs_item debugfs_entries[] = {
  90         { "pf_fixed", VCPU_STAT(pf_fixed) },
  91         { "pf_guest", VCPU_STAT(pf_guest) },
  92         { "tlb_flush", VCPU_STAT(tlb_flush) },
  93         { "invlpg", VCPU_STAT(invlpg) },
  94         { "exits", VCPU_STAT(exits) },
  95         { "io_exits", VCPU_STAT(io_exits) },
  96         { "mmio_exits", VCPU_STAT(mmio_exits) },
  97         { "signal_exits", VCPU_STAT(signal_exits) },
  98         { "irq_window", VCPU_STAT(irq_window_exits) },
  99         { "nmi_window", VCPU_STAT(nmi_window_exits) },
 100         { "halt_exits", VCPU_STAT(halt_exits) },
 101         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
 102         { "hypercalls", VCPU_STAT(hypercalls) },
 103         { "request_irq", VCPU_STAT(request_irq_exits) },
 104         { "irq_exits", VCPU_STAT(irq_exits) },
 105         { "host_state_reload", VCPU_STAT(host_state_reload) },
 106         { "efer_reload", VCPU_STAT(efer_reload) },
 107         { "fpu_reload", VCPU_STAT(fpu_reload) },
 108         { "insn_emulation", VCPU_STAT(insn_emulation) },
 109         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 110         { "irq_injections", VCPU_STAT(irq_injections) },
 111         { "nmi_injections", VCPU_STAT(nmi_injections) },
 112         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 113         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 114         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 115         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 116         { "mmu_flooded", VM_STAT(mmu_flooded) },
 117         { "mmu_recycled", VM_STAT(mmu_recycled) },
 118         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 119         { "mmu_unsync", VM_STAT(mmu_unsync) },
 120         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 121         { "largepages", VM_STAT(lpages) },
 122         { NULL }
 123 };
 124
 125 unsigned long segment_base(u16 selector)
 126 {
 127         struct descriptor_table gdt;
 128         struct desc_struct *d;
 129         unsigned long table_base;
 130         unsigned long v;
 131
 132         if (selector == 0)
 133                 return 0;
 134
 135         asm("sgdt %0" : "=m"(gdt));
 136         table_base = gdt.base;
 137
 138         if (selector & 4) {           /* from ldt */
 139                 u16 ldt_selector;
 140
 141                 asm("sldt %0" : "=g"(ldt_selector));
 142                 table_base = segment_base(ldt_selector);
 143         }
 144         d = (struct desc_struct *)(table_base + (selector & ~7));
 145         v = d->base0 | ((unsigned long)d->base1 << 16) |
 146                 ((unsigned long)d->base2 << 24);
 147 #ifdef CONFIG_X86_64
 148         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 149                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 150 #endif
 151         return v;
 152 }
 153 EXPORT_SYMBOL_GPL(segment_base);
 154
 155 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 156 {
 157         if (irqchip_in_kernel(vcpu->kvm))
 158                 return vcpu->arch.apic_base;
 159         else
 160                 return vcpu->arch.apic_base;
 161 }
 162 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 163
 164 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 165 {
 166         /* TODO: reserve bits check */
 167         if (irqchip_in_kernel(vcpu->kvm))
 168                 kvm_lapic_set_base(vcpu, data);
 169         else
 170                 vcpu->arch.apic_base = data;
 171 }
 172 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 173
 174 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 175 {
 176         WARN_ON(vcpu->arch.exception.pending);
 177         vcpu->arch.exception.pending = true;
 178         vcpu->arch.exception.has_error_code = false;
 179         vcpu->arch.exception.nr = nr;
 180 }
 181 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 182
 183 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 184                            u32 error_code)
 185 {
 186         ++vcpu->stat.pf_guest;
 187
 188         if (vcpu->arch.exception.pending) {
 189                 switch(vcpu->arch.exception.nr) {
 190                 case DF_VECTOR:
 191                         /* triple fault -> shutdown */
 192                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 193                         return;
 194                 case PF_VECTOR:
 195                         vcpu->arch.exception.nr = DF_VECTOR;
 196                         vcpu->arch.exception.error_code = 0;
 197                         return;
 198                 default:
 199                         /* replace previous exception with a new one in a hope
 200                            that instruction re-execution will regenerate lost
 201                            exception */
 202                         vcpu->arch.exception.pending = false;
 203                         break;
 204                 }
 205         }
 206         vcpu->arch.cr2 = addr;
 207         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 208 }
 209
 210 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 211 {
 212         vcpu->arch.nmi_pending = 1;
 213 }
 214 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 215
 216 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 217 {
 218         WARN_ON(vcpu->arch.exception.pending);
 219         vcpu->arch.exception.pending = true;
 220         vcpu->arch.exception.has_error_code = true;
 221         vcpu->arch.exception.nr = nr;
 222         vcpu->arch.exception.error_code = error_code;
 223 }
 224 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 225
 226 static void __queue_exception(struct kvm_vcpu *vcpu)
 227 {
 228         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 229                                      vcpu->arch.exception.has_error_code,
 230                                      vcpu->arch.exception.error_code);
 231 }
 232
 233 /*
 234  * Load the pae pdptrs.  Return true is they are all valid.
 235  */
 236 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 237 {
 238         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 239         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 240         int i;
 241         int ret;
 242         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 243
 244         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 245                                   offset * sizeof(u64), sizeof(pdpte));
 246         if (ret < 0) {
 247                 ret = 0;
 248                 goto out;
 249         }
 250         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 251                 if (is_present_gpte(pdpte[i]) &&
 252                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
 253                         ret = 0;
 254                         goto out;
 255                 }
 256         }
 257         ret = 1;
 258
 259         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 260         __set_bit(VCPU_EXREG_PDPTR,
 261                   (unsigned long *)&vcpu->arch.regs_avail);
 262         __set_bit(VCPU_EXREG_PDPTR,
 263                   (unsigned long *)&vcpu->arch.regs_dirty);
 264 out:
 265
 266         return ret;
 267 }
 268 EXPORT_SYMBOL_GPL(load_pdptrs);
 269
 270 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 271 {
 272         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 273         bool changed = true;
 274         int r;
 275
 276         if (is_long_mode(vcpu) || !is_pae(vcpu))
 277                 return false;
 278
 279         if (!test_bit(VCPU_EXREG_PDPTR,
 280                       (unsigned long *)&vcpu->arch.regs_avail))
 281                 return true;
 282
 283         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 284         if (r < 0)
 285                 goto out;
 286         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 287 out:
 288
 289         return changed;
 290 }
 291
 292 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 293 {
 294         if (cr0 & CR0_RESERVED_BITS) {
 295                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 296                        cr0, vcpu->arch.cr0);
 297                 kvm_inject_gp(vcpu, 0);
 298                 return;
 299         }
 300
 301         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 302                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 303                 kvm_inject_gp(vcpu, 0);
 304                 return;
 305         }
 306
 307         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 308                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 309                        "and a clear PE flag\n");
 310                 kvm_inject_gp(vcpu, 0);
 311                 return;
 312         }
 313
 314         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 315 #ifdef CONFIG_X86_64
 316                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 317                         int cs_db, cs_l;
 318
 319                         if (!is_pae(vcpu)) {
 320                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 321                                        "in long mode while PAE is disabled\n");
 322                                 kvm_inject_gp(vcpu, 0);
 323                                 return;
 324                         }
 325                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 326                         if (cs_l) {
 327                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 328                                        "in long mode while CS.L == 1\n");
 329                                 kvm_inject_gp(vcpu, 0);
 330                                 return;
 331
 332                         }
 333                 } else
 334 #endif
 335                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 336                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 337                                "reserved bits\n");
 338                         kvm_inject_gp(vcpu, 0);
 339                         return;
 340                 }
 341
 342         }
 343
 344         kvm_x86_ops->set_cr0(vcpu, cr0);
 345         vcpu->arch.cr0 = cr0;
 346
 347         kvm_mmu_reset_context(vcpu);
 348         return;
 349 }
 350 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 351
 352 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 353 {
 354         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 355 }
 356 EXPORT_SYMBOL_GPL(kvm_lmsw);
 357
 358 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 359 {
 360         unsigned long old_cr4 = vcpu->arch.cr4;
 361         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
 362
 363         if (cr4 & CR4_RESERVED_BITS) {
 364                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 365                 kvm_inject_gp(vcpu, 0);
 366                 return;
 367         }
 368
 369         if (is_long_mode(vcpu)) {
 370                 if (!(cr4 & X86_CR4_PAE)) {
 371                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 372                                "in long mode\n");
 373                         kvm_inject_gp(vcpu, 0);
 374                         return;
 375                 }
 376         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
 377                    && ((cr4 ^ old_cr4) & pdptr_bits)
 378                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 379                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 380                 kvm_inject_gp(vcpu, 0);
 381                 return;
 382         }
 383
 384         if (cr4 & X86_CR4_VMXE) {
 385                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 386                 kvm_inject_gp(vcpu, 0);
 387                 return;
 388         }
 389         kvm_x86_ops->set_cr4(vcpu, cr4);
 390         vcpu->arch.cr4 = cr4;
 391         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
 392         kvm_mmu_reset_context(vcpu);
 393 }
 394 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 395
 396 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 397 {
 398         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 399                 kvm_mmu_sync_roots(vcpu);
 400                 kvm_mmu_flush_tlb(vcpu);
 401                 return;
 402         }
 403
 404         if (is_long_mode(vcpu)) {
 405                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 406                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 407                         kvm_inject_gp(vcpu, 0);
 408                         return;
 409                 }
 410         } else {
 411                 if (is_pae(vcpu)) {
 412                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 413                                 printk(KERN_DEBUG
 414                                        "set_cr3: #GP, reserved bits\n");
 415                                 kvm_inject_gp(vcpu, 0);
 416                                 return;
 417                         }
 418                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 419                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 420                                        "reserved bits\n");
 421                                 kvm_inject_gp(vcpu, 0);
 422                                 return;
 423                         }
 424                 }
 425                 /*
 426                  * We don't check reserved bits in nonpae mode, because
 427                  * this isn't enforced, and VMware depends on this.
 428                  */
 429         }
 430
 431         /*
 432          * Does the new cr3 value map to physical memory? (Note, we
 433          * catch an invalid cr3 even in real-mode, because it would
 434          * cause trouble later on when we turn on paging anyway.)
 435          *
 436          * A real CPU would silently accept an invalid cr3 and would
 437          * attempt to use it - with largely undefined (and often hard
 438          * to debug) behavior on the guest side.
 439          */
 440         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 441                 kvm_inject_gp(vcpu, 0);
 442         else {
 443                 vcpu->arch.cr3 = cr3;
 444                 vcpu->arch.mmu.new_cr3(vcpu);
 445         }
 446 }
 447 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 448
 449 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 450 {
 451         if (cr8 & CR8_RESERVED_BITS) {
 452                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 453                 kvm_inject_gp(vcpu, 0);
 454                 return;
 455         }
 456         if (irqchip_in_kernel(vcpu->kvm))
 457                 kvm_lapic_set_tpr(vcpu, cr8);
 458         else
 459                 vcpu->arch.cr8 = cr8;
 460 }
 461 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 462
 463 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 464 {
 465         if (irqchip_in_kernel(vcpu->kvm))
 466                 return kvm_lapic_get_cr8(vcpu);
 467         else
 468                 return vcpu->arch.cr8;
 469 }
 470 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 471
 472 static inline u32 bit(int bitno)
 473 {
 474         return 1 << (bitno & 31);
 475 }
 476
 477 /*
 478  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 479  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 480  *
 481  * This list is modified at module load time to reflect the
 482  * capabilities of the host cpu.
 483  */
 484 static u32 msrs_to_save[] = {
 485         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 486         MSR_K6_STAR,
 487 #ifdef CONFIG_X86_64
 488         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 489 #endif
 490         MSR_IA32_TSC, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 491         MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 492 };
 493
 494 static unsigned num_msrs_to_save;
 495
 496 static u32 emulated_msrs[] = {
 497         MSR_IA32_MISC_ENABLE,
 498 };
 499
 500 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 501 {
 502         if (efer & efer_reserved_bits) {
 503                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 504                        efer);
 505                 kvm_inject_gp(vcpu, 0);
 506                 return;
 507         }
 508
 509         if (is_paging(vcpu)
 510             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 511                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 512                 kvm_inject_gp(vcpu, 0);
 513                 return;
 514         }
 515
 516         if (efer & EFER_FFXSR) {
 517                 struct kvm_cpuid_entry2 *feat;
 518
 519                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 520                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
 521                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
 522                         kvm_inject_gp(vcpu, 0);
 523                         return;
 524                 }
 525         }
 526
 527         if (efer & EFER_SVME) {
 528                 struct kvm_cpuid_entry2 *feat;
 529
 530                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 531                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 532                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 533                         kvm_inject_gp(vcpu, 0);
 534                         return;
 535                 }
 536         }
 537
 538         kvm_x86_ops->set_efer(vcpu, efer);
 539
 540         efer &= ~EFER_LMA;
 541         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 542
 543         vcpu->arch.shadow_efer = efer;
 544
 545         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
 546         kvm_mmu_reset_context(vcpu);
 547 }
 548
 549 void kvm_enable_efer_bits(u64 mask)
 550 {
 551        efer_reserved_bits &= ~mask;
 552 }
 553 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 554
 555
 556 /*
 557  * Writes msr value into into the appropriate "register".
 558  * Returns 0 on success, non-0 otherwise.
 559  * Assumes vcpu_load() was already called.
 560  */
 561 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 562 {
 563         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 564 }
 565
 566 /*
 567  * Adapt set_msr() to msr_io()'s calling convention
 568  */
 569 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 570 {
 571         return kvm_set_msr(vcpu, index, *data);
 572 }
 573
 574 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 575 {
 576         static int version;
 577         struct pvclock_wall_clock wc;
 578         struct timespec now, sys, boot;
 579
 580         if (!wall_clock)
 581                 return;
 582
 583         version++;
 584
 585         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 586
 587         /*
 588          * The guest calculates current wall clock time by adding
 589          * system time (updated by kvm_write_guest_time below) to the
 590          * wall clock specified here.  guest system time equals host
 591          * system time for us, thus we must fill in host boot time here.
 592          */
 593         now = current_kernel_time();
 594         ktime_get_ts(&sys);
 595         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 596
 597         wc.sec = boot.tv_sec;
 598         wc.nsec = boot.tv_nsec;
 599         wc.version = version;
 600
 601         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 602
 603         version++;
 604         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 605 }
 606
 607 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 608 {
 609         uint32_t quotient, remainder;
 610
 611         /* Don't try to replace with do_div(), this one calculates
 612          * "(dividend << 32) / divisor" */
 613         __asm__ ( "divl %4"
 614                   : "=a" (quotient), "=d" (remainder)
 615                   : "0" (0), "1" (dividend), "r" (divisor) );
 616         return quotient;
 617 }
 618
 619 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 620 {
 621         uint64_t nsecs = 1000000000LL;
 622         int32_t  shift = 0;
 623         uint64_t tps64;
 624         uint32_t tps32;
 625
 626         tps64 = tsc_khz * 1000LL;
 627         while (tps64 > nsecs*2) {
 628                 tps64 >>= 1;
 629                 shift--;
 630         }
 631
 632         tps32 = (uint32_t)tps64;
 633         while (tps32 <= (uint32_t)nsecs) {
 634                 tps32 <<= 1;
 635                 shift++;
 636         }
 637
 638         hv_clock->tsc_shift = shift;
 639         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 640
 641         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 642                  __func__, tsc_khz, hv_clock->tsc_shift,
 643                  hv_clock->tsc_to_system_mul);
 644 }
 645
 646 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
 647
 648 static void kvm_write_guest_time(struct kvm_vcpu *v)
 649 {
 650         struct timespec ts;
 651         unsigned long flags;
 652         struct kvm_vcpu_arch *vcpu = &v->arch;
 653         void *shared_kaddr;
 654         unsigned long this_tsc_khz;
 655
 656         if ((!vcpu->time_page))
 657                 return;
 658
 659         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
 660         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
 661                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
 662                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
 663         }
 664         put_cpu_var(cpu_tsc_khz);
 665
 666         /* Keep irq disabled to prevent changes to the clock */
 667         local_irq_save(flags);
 668         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
 669         ktime_get_ts(&ts);
 670         local_irq_restore(flags);
 671
 672         /* With all the info we got, fill in the values */
 673
 674         vcpu->hv_clock.system_time = ts.tv_nsec +
 675                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 676         /*
 677          * The interface expects us to write an even number signaling that the
 678          * update is finished. Since the guest won't see the intermediate
 679          * state, we just increase by 2 at the end.
 680          */
 681         vcpu->hv_clock.version += 2;
 682
 683         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 684
 685         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 686                sizeof(vcpu->hv_clock));
 687
 688         kunmap_atomic(shared_kaddr, KM_USER0);
 689
 690         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 691 }
 692
 693 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
 694 {
 695         struct kvm_vcpu_arch *vcpu = &v->arch;
 696
 697         if (!vcpu->time_page)
 698                 return 0;
 699         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
 700         return 1;
 701 }
 702
 703 static bool msr_mtrr_valid(unsigned msr)
 704 {
 705         switch (msr) {
 706         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 707         case MSR_MTRRfix64K_00000:
 708         case MSR_MTRRfix16K_80000:
 709         case MSR_MTRRfix16K_A0000:
 710         case MSR_MTRRfix4K_C0000:
 711         case MSR_MTRRfix4K_C8000:
 712         case MSR_MTRRfix4K_D0000:
 713         case MSR_MTRRfix4K_D8000:
 714         case MSR_MTRRfix4K_E0000:
 715         case MSR_MTRRfix4K_E8000:
 716         case MSR_MTRRfix4K_F0000:
 717         case MSR_MTRRfix4K_F8000:
 718         case MSR_MTRRdefType:
 719         case MSR_IA32_CR_PAT:
 720                 return true;
 721         case 0x2f8:
 722                 return true;
 723         }
 724         return false;
 725 }
 726
 727 static bool valid_pat_type(unsigned t)
 728 {
 729         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
 730 }
 731
 732 static bool valid_mtrr_type(unsigned t)
 733 {
 734         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
 735 }
 736
 737 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 738 {
 739         int i;
 740
 741         if (!msr_mtrr_valid(msr))
 742                 return false;
 743
 744         if (msr == MSR_IA32_CR_PAT) {
 745                 for (i = 0; i < 8; i++)
 746                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
 747                                 return false;
 748                 return true;
 749         } else if (msr == MSR_MTRRdefType) {
 750                 if (data & ~0xcff)
 751                         return false;
 752                 return valid_mtrr_type(data & 0xff);
 753         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
 754                 for (i = 0; i < 8 ; i++)
 755                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
 756                                 return false;
 757                 return true;
 758         }
 759
 760         /* variable MTRRs */
 761         return valid_mtrr_type(data & 0xff);
 762 }
 763
 764 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 765 {
 766         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 767
 768         if (!mtrr_valid(vcpu, msr, data))
 769                 return 1;
 770
 771         if (msr == MSR_MTRRdefType) {
 772                 vcpu->arch.mtrr_state.def_type = data;
 773                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 774         } else if (msr == MSR_MTRRfix64K_00000)
 775                 p[0] = data;
 776         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 777                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 778         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 779                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 780         else if (msr == MSR_IA32_CR_PAT)
 781                 vcpu->arch.pat = data;
 782         else {  /* Variable MTRRs */
 783                 int idx, is_mtrr_mask;
 784                 u64 *pt;
 785
 786                 idx = (msr - 0x200) / 2;
 787                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 788                 if (!is_mtrr_mask)
 789                         pt =
 790                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 791                 else
 792                         pt =
 793                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 794                 *pt = data;
 795         }
 796
 797         kvm_mmu_reset_context(vcpu);
 798         return 0;
 799 }
 800
 801 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 802 {
 803         u64 mcg_cap = vcpu->arch.mcg_cap;
 804         unsigned bank_num = mcg_cap & 0xff;
 805
 806         switch (msr) {
 807         case MSR_IA32_MCG_STATUS:
 808                 vcpu->arch.mcg_status = data;
 809                 break;
 810         case MSR_IA32_MCG_CTL:
 811                 if (!(mcg_cap & MCG_CTL_P))
 812                         return 1;
 813                 if (data != 0 && data != ~(u64)0)
 814                         return -1;
 815                 vcpu->arch.mcg_ctl = data;
 816                 break;
 817         default:
 818                 if (msr >= MSR_IA32_MC0_CTL &&
 819                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
 820                         u32 offset = msr - MSR_IA32_MC0_CTL;
 821                         /* only 0 or all 1s can be written to IA32_MCi_CTL */
 822                         if ((offset & 0x3) == 0 &&
 823                             data != 0 && data != ~(u64)0)
 824                                 return -1;
 825                         vcpu->arch.mce_banks[offset] = data;
 826                         break;
 827                 }
 828                 return 1;
 829         }
 830         return 0;
 831 }
 832
 833 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 834 {
 835         switch (msr) {
 836         case MSR_EFER:
 837                 set_efer(vcpu, data);
 838                 break;
 839         case MSR_K7_HWCR:
 840                 data &= ~(u64)0x40;     /* ignore flush filter disable */
 841                 if (data != 0) {
 842                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
 843                                 data);
 844                         return 1;
 845                 }
 846                 break;
 847         case MSR_IA32_DEBUGCTLMSR:
 848                 if (!data) {
 849                         /* We support the non-activated case already */
 850                         break;
 851                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 852                         /* Values other than LBR and BTF are vendor-specific,
 853                            thus reserved and should throw a #GP */
 854                         return 1;
 855                 }
 856                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 857                         __func__, data);
 858                 break;
 859         case MSR_IA32_UCODE_REV:
 860         case MSR_IA32_UCODE_WRITE:
 861         case MSR_VM_HSAVE_PA:
 862                 break;
 863         case 0x200 ... 0x2ff:
 864                 return set_msr_mtrr(vcpu, msr, data);
 865         case MSR_IA32_APICBASE:
 866                 kvm_set_apic_base(vcpu, data);
 867                 break;
 868         case MSR_IA32_MISC_ENABLE:
 869                 vcpu->arch.ia32_misc_enable_msr = data;
 870                 break;
 871         case MSR_KVM_WALL_CLOCK:
 872                 vcpu->kvm->arch.wall_clock = data;
 873                 kvm_write_wall_clock(vcpu->kvm, data);
 874                 break;
 875         case MSR_KVM_SYSTEM_TIME: {
 876                 if (vcpu->arch.time_page) {
 877                         kvm_release_page_dirty(vcpu->arch.time_page);
 878                         vcpu->arch.time_page = NULL;
 879                 }
 880
 881                 vcpu->arch.time = data;
 882
 883                 /* we verify if the enable bit is set... */
 884                 if (!(data & 1))
 885                         break;
 886
 887                 /* ...but clean it before doing the actual write */
 888                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 889
 890                 vcpu->arch.time_page =
 891                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 892
 893                 if (is_error_page(vcpu->arch.time_page)) {
 894                         kvm_release_page_clean(vcpu->arch.time_page);
 895                         vcpu->arch.time_page = NULL;
 896                 }
 897
 898                 kvm_request_guest_time_update(vcpu);
 899                 break;
 900         }
 901         case MSR_IA32_MCG_CTL:
 902         case MSR_IA32_MCG_STATUS:
 903         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
 904                 return set_msr_mce(vcpu, msr, data);
 905
 906         /* Performance counters are not protected by a CPUID bit,
 907          * so we should check all of them in the generic path for the sake of
 908          * cross vendor migration.
 909          * Writing a zero into the event select MSRs disables them,
 910          * which we perfectly emulate ;-). Any other value should be at least
 911          * reported, some guests depend on them.
 912          */
 913         case MSR_P6_EVNTSEL0:
 914         case MSR_P6_EVNTSEL1:
 915         case MSR_K7_EVNTSEL0:
 916         case MSR_K7_EVNTSEL1:
 917         case MSR_K7_EVNTSEL2:
 918         case MSR_K7_EVNTSEL3:
 919                 if (data != 0)
 920                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
 921                                 "0x%x data 0x%llx\n", msr, data);
 922                 break;
 923         /* at least RHEL 4 unconditionally writes to the perfctr registers,
 924          * so we ignore writes to make it happy.
 925          */
 926         case MSR_P6_PERFCTR0:
 927         case MSR_P6_PERFCTR1:
 928         case MSR_K7_PERFCTR0:
 929         case MSR_K7_PERFCTR1:
 930         case MSR_K7_PERFCTR2:
 931         case MSR_K7_PERFCTR3:
 932                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
 933                         "0x%x data 0x%llx\n", msr, data);
 934                 break;
 935         default:
 936                 if (!ignore_msrs) {
 937                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
 938                                 msr, data);
 939                         return 1;
 940                 } else {
 941                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
 942                                 msr, data);
 943                         break;
 944                 }
 945         }
 946         return 0;
 947 }
 948 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 949
 950
 951 /*
 952  * Reads an msr value (of 'msr_index') into 'pdata'.
 953  * Returns 0 on success, non-0 otherwise.
 954  * Assumes vcpu_load() was already called.
 955  */
 956 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 957 {
 958         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 959 }
 960
 961 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 962 {
 963         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 964
 965         if (!msr_mtrr_valid(msr))
 966                 return 1;
 967
 968         if (msr == MSR_MTRRdefType)
 969                 *pdata = vcpu->arch.mtrr_state.def_type +
 970                          (vcpu->arch.mtrr_state.enabled << 10);
 971         else if (msr == MSR_MTRRfix64K_00000)
 972                 *pdata = p[0];
 973         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 974                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 975         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 976                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 977         else if (msr == MSR_IA32_CR_PAT)
 978                 *pdata = vcpu->arch.pat;
 979         else {  /* Variable MTRRs */
 980                 int idx, is_mtrr_mask;
 981                 u64 *pt;
 982
 983                 idx = (msr - 0x200) / 2;
 984                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 985                 if (!is_mtrr_mask)
 986                         pt =
 987                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 988                 else
 989                         pt =
 990                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 991                 *pdata = *pt;
 992         }
 993
 994         return 0;
 995 }
 996
 997 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 998 {
 999         u64 data;
1000         u64 mcg_cap = vcpu->arch.mcg_cap;
1001         unsigned bank_num = mcg_cap & 0xff;
1002
1003         switch (msr) {
1004         case MSR_IA32_P5_MC_ADDR:
1005         case MSR_IA32_P5_MC_TYPE:
1006                 data = 0;
1007                 break;
1008         case MSR_IA32_MCG_CAP:
1009                 data = vcpu->arch.mcg_cap;
1010                 break;
1011         case MSR_IA32_MCG_CTL:
1012                 if (!(mcg_cap & MCG_CTL_P))
1013                         return 1;
1014                 data = vcpu->arch.mcg_ctl;
1015                 break;
1016         case MSR_IA32_MCG_STATUS:
1017                 data = vcpu->arch.mcg_status;
1018                 break;
1019         default:
1020                 if (msr >= MSR_IA32_MC0_CTL &&
1021                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1022                         u32 offset = msr - MSR_IA32_MC0_CTL;
1023                         data = vcpu->arch.mce_banks[offset];
1024                         break;
1025                 }
1026                 return 1;
1027         }
1028         *pdata = data;
1029         return 0;
1030 }
1031
1032 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1033 {
1034         u64 data;
1035
1036         switch (msr) {
1037         case MSR_IA32_PLATFORM_ID:
1038         case MSR_IA32_UCODE_REV:
1039         case MSR_IA32_EBL_CR_POWERON:
1040         case MSR_IA32_DEBUGCTLMSR:
1041         case MSR_IA32_LASTBRANCHFROMIP:
1042         case MSR_IA32_LASTBRANCHTOIP:
1043         case MSR_IA32_LASTINTFROMIP:
1044         case MSR_IA32_LASTINTTOIP:
1045         case MSR_K8_SYSCFG:
1046         case MSR_K7_HWCR:
1047         case MSR_VM_HSAVE_PA:
1048         case MSR_P6_EVNTSEL0:
1049         case MSR_P6_EVNTSEL1:
1050         case MSR_K7_EVNTSEL0:
1051         case MSR_K8_INT_PENDING_MSG:
1052                 data = 0;
1053                 break;
1054         case MSR_MTRRcap:
1055                 data = 0x500 | KVM_NR_VAR_MTRR;
1056                 break;
1057         case 0x200 ... 0x2ff:
1058                 return get_msr_mtrr(vcpu, msr, pdata);
1059         case 0xcd: /* fsb frequency */
1060                 data = 3;
1061                 break;
1062         case MSR_IA32_APICBASE:
1063                 data = kvm_get_apic_base(vcpu);
1064                 break;
1065         case MSR_IA32_MISC_ENABLE:
1066                 data = vcpu->arch.ia32_misc_enable_msr;
1067                 break;
1068         case MSR_IA32_PERF_STATUS:
1069                 /* TSC increment by tick */
1070                 data = 1000ULL;
1071                 /* CPU multiplier */
1072                 data |= (((uint64_t)4ULL) << 40);
1073                 break;
1074         case MSR_EFER:
1075                 data = vcpu->arch.shadow_efer;
1076                 break;
1077         case MSR_KVM_WALL_CLOCK:
1078                 data = vcpu->kvm->arch.wall_clock;
1079                 break;
1080         case MSR_KVM_SYSTEM_TIME:
1081                 data = vcpu->arch.time;
1082                 break;
1083         case MSR_IA32_P5_MC_ADDR:
1084         case MSR_IA32_P5_MC_TYPE:
1085         case MSR_IA32_MCG_CAP:
1086         case MSR_IA32_MCG_CTL:
1087         case MSR_IA32_MCG_STATUS:
1088         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1089                 return get_msr_mce(vcpu, msr, pdata);
1090         default:
1091                 if (!ignore_msrs) {
1092                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1093                         return 1;
1094                 } else {
1095                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1096                         data = 0;
1097                 }
1098                 break;
1099         }
1100         *pdata = data;
1101         return 0;
1102 }
1103 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1104
1105 /*
1106  * Read or write a bunch of msrs. All parameters are kernel addresses.
1107  *
1108  * @return number of msrs set successfully.
1109  */
1110 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1111                     struct kvm_msr_entry *entries,
1112                     int (*do_msr)(struct kvm_vcpu *vcpu,
1113                                   unsigned index, u64 *data))
1114 {
1115         int i;
1116
1117         vcpu_load(vcpu);
1118
1119         down_read(&vcpu->kvm->slots_lock);
1120         for (i = 0; i < msrs->nmsrs; ++i)
1121                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1122                         break;
1123         up_read(&vcpu->kvm->slots_lock);
1124
1125         vcpu_put(vcpu);
1126
1127         return i;
1128 }
1129
1130 /*
1131  * Read or write a bunch of msrs. Parameters are user addresses.
1132  *
1133  * @return number of msrs set successfully.
1134  */
1135 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1136                   int (*do_msr)(struct kvm_vcpu *vcpu,
1137                                 unsigned index, u64 *data),
1138                   int writeback)
1139 {
1140         struct kvm_msrs msrs;
1141         struct kvm_msr_entry *entries;
1142         int r, n;
1143         unsigned size;
1144
1145         r = -EFAULT;
1146         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1147                 goto out;
1148
1149         r = -E2BIG;
1150         if (msrs.nmsrs >= MAX_IO_MSRS)
1151                 goto out;
1152
1153         r = -ENOMEM;
1154         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1155         entries = vmalloc(size);
1156         if (!entries)
1157                 goto out;
1158
1159         r = -EFAULT;
1160         if (copy_from_user(entries, user_msrs->entries, size))
1161                 goto out_free;
1162
1163         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1164         if (r < 0)
1165                 goto out_free;
1166
1167         r = -EFAULT;
1168         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1169                 goto out_free;
1170
1171         r = n;
1172
1173 out_free:
1174         vfree(entries);
1175 out:
1176         return r;
1177 }
1178
1179 int kvm_dev_ioctl_check_extension(long ext)
1180 {
1181         int r;
1182
1183         switch (ext) {
1184         case KVM_CAP_IRQCHIP:
1185         case KVM_CAP_HLT:
1186         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1187         case KVM_CAP_SET_TSS_ADDR:
1188         case KVM_CAP_EXT_CPUID:
1189         case KVM_CAP_CLOCKSOURCE:
1190         case KVM_CAP_PIT:
1191         case KVM_CAP_NOP_IO_DELAY:
1192         case KVM_CAP_MP_STATE:
1193         case KVM_CAP_SYNC_MMU:
1194         case KVM_CAP_REINJECT_CONTROL:
1195         case KVM_CAP_IRQ_INJECT_STATUS:
1196         case KVM_CAP_ASSIGN_DEV_IRQ:
1197         case KVM_CAP_IRQFD:
1198         case KVM_CAP_PIT2:
1199                 r = 1;
1200                 break;
1201         case KVM_CAP_COALESCED_MMIO:
1202                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1203                 break;
1204         case KVM_CAP_VAPIC:
1205                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1206                 break;
1207         case KVM_CAP_NR_VCPUS:
1208                 r = KVM_MAX_VCPUS;
1209                 break;
1210         case KVM_CAP_NR_MEMSLOTS:
1211                 r = KVM_MEMORY_SLOTS;
1212                 break;
1213         case KVM_CAP_PV_MMU:
1214                 r = !tdp_enabled;
1215                 break;
1216         case KVM_CAP_IOMMU:
1217                 r = iommu_found();
1218                 break;
1219         case KVM_CAP_MCE:
1220                 r = KVM_MAX_MCE_BANKS;
1221                 break;
1222         default:
1223                 r = 0;
1224                 break;
1225         }
1226         return r;
1227
1228 }
1229
1230 long kvm_arch_dev_ioctl(struct file *filp,
1231                         unsigned int ioctl, unsigned long arg)
1232 {
1233         void __user *argp = (void __user *)arg;
1234         long r;
1235
1236         switch (ioctl) {
1237         case KVM_GET_MSR_INDEX_LIST: {
1238                 struct kvm_msr_list __user *user_msr_list = argp;
1239                 struct kvm_msr_list msr_list;
1240                 unsigned n;
1241
1242                 r = -EFAULT;
1243                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1244                         goto out;
1245                 n = msr_list.nmsrs;
1246                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1247                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1248                         goto out;
1249                 r = -E2BIG;
1250                 if (n < msr_list.nmsrs)
1251                         goto out;
1252                 r = -EFAULT;
1253                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1254                                  num_msrs_to_save * sizeof(u32)))
1255                         goto out;
1256                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1257                                  &emulated_msrs,
1258                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1259                         goto out;
1260                 r = 0;
1261                 break;
1262         }
1263         case KVM_GET_SUPPORTED_CPUID: {
1264                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1265                 struct kvm_cpuid2 cpuid;
1266
1267                 r = -EFAULT;
1268                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1269                         goto out;
1270                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1271                                                       cpuid_arg->entries);
1272                 if (r)
1273                         goto out;
1274
1275                 r = -EFAULT;
1276                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1277                         goto out;
1278                 r = 0;
1279                 break;
1280         }
1281         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1282                 u64 mce_cap;
1283
1284                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1285                 r = -EFAULT;
1286                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1287                         goto out;
1288                 r = 0;
1289                 break;
1290         }
1291         default:
1292                 r = -EINVAL;
1293         }
1294 out:
1295         return r;
1296 }
1297
1298 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1299 {
1300         kvm_x86_ops->vcpu_load(vcpu, cpu);
1301         kvm_request_guest_time_update(vcpu);
1302 }
1303
1304 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1305 {
1306         kvm_x86_ops->vcpu_put(vcpu);
1307         kvm_put_guest_fpu(vcpu);
1308 }
1309
1310 static int is_efer_nx(void)
1311 {
1312         unsigned long long efer = 0;
1313
1314         rdmsrl_safe(MSR_EFER, &efer);
1315         return efer & EFER_NX;
1316 }
1317
1318 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1319 {
1320         int i;
1321         struct kvm_cpuid_entry2 *e, *entry;
1322
1323         entry = NULL;
1324         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1325                 e = &vcpu->arch.cpuid_entries[i];
1326                 if (e->function == 0x80000001) {
1327                         entry = e;
1328                         break;
1329                 }
1330         }
1331         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1332                 entry->edx &= ~(1 << 20);
1333                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1334         }
1335 }
1336
1337 /* when an old userspace process fills a new kernel module */
1338 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1339                                     struct kvm_cpuid *cpuid,
1340                                     struct kvm_cpuid_entry __user *entries)
1341 {
1342         int r, i;
1343         struct kvm_cpuid_entry *cpuid_entries;
1344
1345         r = -E2BIG;
1346         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1347                 goto out;
1348         r = -ENOMEM;
1349         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1350         if (!cpuid_entries)
1351                 goto out;
1352         r = -EFAULT;
1353         if (copy_from_user(cpuid_entries, entries,
1354                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1355                 goto out_free;
1356         for (i = 0; i < cpuid->nent; i++) {
1357                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1358                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1359                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1360                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1361                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1362                 vcpu->arch.cpuid_entries[i].index = 0;
1363                 vcpu->arch.cpuid_entries[i].flags = 0;
1364                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1365                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1366                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1367         }
1368         vcpu->arch.cpuid_nent = cpuid->nent;
1369         cpuid_fix_nx_cap(vcpu);
1370         r = 0;
1371
1372 out_free:
1373         vfree(cpuid_entries);
1374 out:
1375         return r;
1376 }
1377
1378 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1379                                      struct kvm_cpuid2 *cpuid,
1380                                      struct kvm_cpuid_entry2 __user *entries)
1381 {
1382         int r;
1383
1384         r = -E2BIG;
1385         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1386                 goto out;
1387         r = -EFAULT;
1388         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1389                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1390                 goto out;
1391         vcpu->arch.cpuid_nent = cpuid->nent;
1392         return 0;
1393
1394 out:
1395         return r;
1396 }
1397
1398 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1399                                      struct kvm_cpuid2 *cpuid,
1400                                      struct kvm_cpuid_entry2 __user *entries)
1401 {
1402         int r;
1403
1404         r = -E2BIG;
1405         if (cpuid->nent < vcpu->arch.cpuid_nent)
1406                 goto out;
1407         r = -EFAULT;
1408         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1409                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1410                 goto out;
1411         return 0;
1412
1413 out:
1414         cpuid->nent = vcpu->arch.cpuid_nent;
1415         return r;
1416 }
1417
1418 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1419                            u32 index)
1420 {
1421         entry->function = function;
1422         entry->index = index;
1423         cpuid_count(entry->function, entry->index,
1424                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1425         entry->flags = 0;
1426 }
1427
1428 #define F(x) bit(X86_FEATURE_##x)
1429
1430 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1431                          u32 index, int *nent, int maxnent)
1432 {
1433         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1434 #ifdef CONFIG_X86_64
1435         unsigned f_lm = F(LM);
1436 #else
1437         unsigned f_lm = 0;
1438 #endif
1439
1440         /* cpuid 1.edx */
1441         const u32 kvm_supported_word0_x86_features =
1442                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1443                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1444                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1445                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1446                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1447                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1448                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1449                 0 /* HTT, TM, Reserved, PBE */;
1450         /* cpuid 0x80000001.edx */
1451         const u32 kvm_supported_word1_x86_features =
1452                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1453                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1454                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1455                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1456                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1457                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1458                 F(FXSR) | F(FXSR_OPT) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1459                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1460         /* cpuid 1.ecx */
1461         const u32 kvm_supported_word4_x86_features =
1462                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1463                 0 /* DS-CPL, VMX, SMX, EST */ |
1464                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1465                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1466                 0 /* Reserved, DCA */ | F(XMM4_1) |
1467                 F(XMM4_2) | 0 /* x2APIC */ | F(MOVBE) | F(POPCNT) |
1468                 0 /* Reserved, XSAVE, OSXSAVE */;
1469         /* cpuid 0x80000001.ecx */
1470         const u32 kvm_supported_word6_x86_features =
1471                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1472                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1473                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1474                 0 /* SKINIT */ | 0 /* WDT */;
1475
1476         /* all calls to cpuid_count() should be made on the same cpu */
1477         get_cpu();
1478         do_cpuid_1_ent(entry, function, index);
1479         ++*nent;
1480
1481         switch (function) {
1482         case 0:
1483                 entry->eax = min(entry->eax, (u32)0xb);
1484                 break;
1485         case 1:
1486                 entry->edx &= kvm_supported_word0_x86_features;
1487                 entry->ecx &= kvm_supported_word4_x86_features;
1488                 break;
1489         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1490          * may return different values. This forces us to get_cpu() before
1491          * issuing the first command, and also to emulate this annoying behavior
1492          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1493         case 2: {
1494                 int t, times = entry->eax & 0xff;
1495
1496                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1497                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1498                 for (t = 1; t < times && *nent < maxnent; ++t) {
1499                         do_cpuid_1_ent(&entry[t], function, 0);
1500                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1501                         ++*nent;
1502                 }
1503                 break;
1504         }
1505         /* function 4 and 0xb have additional index. */
1506         case 4: {
1507                 int i, cache_type;
1508
1509                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1510                 /* read more entries until cache_type is zero */
1511                 for (i = 1; *nent < maxnent; ++i) {
1512                         cache_type = entry[i - 1].eax & 0x1f;
1513                         if (!cache_type)
1514                                 break;
1515                         do_cpuid_1_ent(&entry[i], function, i);
1516                         entry[i].flags |=
1517                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1518                         ++*nent;
1519                 }
1520                 break;
1521         }
1522         case 0xb: {
1523                 int i, level_type;
1524
1525                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1526                 /* read more entries until level_type is zero */
1527                 for (i = 1; *nent < maxnent; ++i) {
1528                         level_type = entry[i - 1].ecx & 0xff00;
1529                         if (!level_type)
1530                                 break;
1531                         do_cpuid_1_ent(&entry[i], function, i);
1532                         entry[i].flags |=
1533                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1534                         ++*nent;
1535                 }
1536                 break;
1537         }
1538         case 0x80000000:
1539                 entry->eax = min(entry->eax, 0x8000001a);
1540                 break;
1541         case 0x80000001:
1542                 entry->edx &= kvm_supported_word1_x86_features;
1543                 entry->ecx &= kvm_supported_word6_x86_features;
1544                 break;
1545         }
1546         put_cpu();
1547 }
1548
1549 #undef F
1550
1551 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1552                                      struct kvm_cpuid_entry2 __user *entries)
1553 {
1554         struct kvm_cpuid_entry2 *cpuid_entries;
1555         int limit, nent = 0, r = -E2BIG;
1556         u32 func;
1557
1558         if (cpuid->nent < 1)
1559                 goto out;
1560         r = -ENOMEM;
1561         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1562         if (!cpuid_entries)
1563                 goto out;
1564
1565         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1566         limit = cpuid_entries[0].eax;
1567         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1568                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1569                              &nent, cpuid->nent);
1570         r = -E2BIG;
1571         if (nent >= cpuid->nent)
1572                 goto out_free;
1573
1574         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1575         limit = cpuid_entries[nent - 1].eax;
1576         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1577                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1578                              &nent, cpuid->nent);
1579         r = -E2BIG;
1580         if (nent >= cpuid->nent)
1581                 goto out_free;
1582
1583         r = -EFAULT;
1584         if (copy_to_user(entries, cpuid_entries,
1585                          nent * sizeof(struct kvm_cpuid_entry2)))
1586                 goto out_free;
1587         cpuid->nent = nent;
1588         r = 0;
1589
1590 out_free:
1591         vfree(cpuid_entries);
1592 out:
1593         return r;
1594 }
1595
1596 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1597                                     struct kvm_lapic_state *s)
1598 {
1599         vcpu_load(vcpu);
1600         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1601         vcpu_put(vcpu);
1602
1603         return 0;
1604 }
1605
1606 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1607                                     struct kvm_lapic_state *s)
1608 {
1609         vcpu_load(vcpu);
1610         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1611         kvm_apic_post_state_restore(vcpu);
1612         vcpu_put(vcpu);
1613
1614         return 0;
1615 }
1616
1617 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1618                                     struct kvm_interrupt *irq)
1619 {
1620         if (irq->irq < 0 || irq->irq >= 256)
1621                 return -EINVAL;
1622         if (irqchip_in_kernel(vcpu->kvm))
1623                 return -ENXIO;
1624         vcpu_load(vcpu);
1625
1626         kvm_queue_interrupt(vcpu, irq->irq, false);
1627
1628         vcpu_put(vcpu);
1629
1630         return 0;
1631 }
1632
1633 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1634 {
1635         vcpu_load(vcpu);
1636         kvm_inject_nmi(vcpu);
1637         vcpu_put(vcpu);
1638
1639         return 0;
1640 }
1641
1642 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1643                                            struct kvm_tpr_access_ctl *tac)
1644 {
1645         if (tac->flags)
1646                 return -EINVAL;
1647         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1648         return 0;
1649 }
1650
1651 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1652                                         u64 mcg_cap)
1653 {
1654         int r;
1655         unsigned bank_num = mcg_cap & 0xff, bank;
1656
1657         r = -EINVAL;
1658         if (!bank_num)
1659                 goto out;
1660         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1661                 goto out;
1662         r = 0;
1663         vcpu->arch.mcg_cap = mcg_cap;
1664         /* Init IA32_MCG_CTL to all 1s */
1665         if (mcg_cap & MCG_CTL_P)
1666                 vcpu->arch.mcg_ctl = ~(u64)0;
1667         /* Init IA32_MCi_CTL to all 1s */
1668         for (bank = 0; bank < bank_num; bank++)
1669                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1670 out:
1671         return r;
1672 }
1673
1674 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1675                                       struct kvm_x86_mce *mce)
1676 {
1677         u64 mcg_cap = vcpu->arch.mcg_cap;
1678         unsigned bank_num = mcg_cap & 0xff;
1679         u64 *banks = vcpu->arch.mce_banks;
1680
1681         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1682                 return -EINVAL;
1683         /*
1684          * if IA32_MCG_CTL is not all 1s, the uncorrected error
1685          * reporting is disabled
1686          */
1687         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1688             vcpu->arch.mcg_ctl != ~(u64)0)
1689                 return 0;
1690         banks += 4 * mce->bank;
1691         /*
1692          * if IA32_MCi_CTL is not all 1s, the uncorrected error
1693          * reporting is disabled for the bank
1694          */
1695         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1696                 return 0;
1697         if (mce->status & MCI_STATUS_UC) {
1698                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1699                     !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1700                         printk(KERN_DEBUG "kvm: set_mce: "
1701                                "injects mce exception while "
1702                                "previous one is in progress!\n");
1703                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1704                         return 0;
1705                 }
1706                 if (banks[1] & MCI_STATUS_VAL)
1707                         mce->status |= MCI_STATUS_OVER;
1708                 banks[2] = mce->addr;
1709                 banks[3] = mce->misc;
1710                 vcpu->arch.mcg_status = mce->mcg_status;
1711                 banks[1] = mce->status;
1712                 kvm_queue_exception(vcpu, MC_VECTOR);
1713         } else if (!(banks[1] & MCI_STATUS_VAL)
1714                    || !(banks[1] & MCI_STATUS_UC)) {
1715                 if (banks[1] & MCI_STATUS_VAL)
1716                         mce->status |= MCI_STATUS_OVER;
1717                 banks[2] = mce->addr;
1718                 banks[3] = mce->misc;
1719                 banks[1] = mce->status;
1720         } else
1721                 banks[1] |= MCI_STATUS_OVER;
1722         return 0;
1723 }
1724
1725 long kvm_arch_vcpu_ioctl(struct file *filp,
1726                          unsigned int ioctl, unsigned long arg)
1727 {
1728         struct kvm_vcpu *vcpu = filp->private_data;
1729         void __user *argp = (void __user *)arg;
1730         int r;
1731         struct kvm_lapic_state *lapic = NULL;
1732
1733         switch (ioctl) {
1734         case KVM_GET_LAPIC: {
1735                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1736
1737                 r = -ENOMEM;
1738                 if (!lapic)
1739                         goto out;
1740                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1741                 if (r)
1742                         goto out;
1743                 r = -EFAULT;
1744                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1745                         goto out;
1746                 r = 0;
1747                 break;
1748         }
1749         case KVM_SET_LAPIC: {
1750                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1751                 r = -ENOMEM;
1752                 if (!lapic)
1753                         goto out;
1754                 r = -EFAULT;
1755                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1756                         goto out;
1757                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1758                 if (r)
1759                         goto out;
1760                 r = 0;
1761                 break;
1762         }
1763         case KVM_INTERRUPT: {
1764                 struct kvm_interrupt irq;
1765
1766                 r = -EFAULT;
1767                 if (copy_from_user(&irq, argp, sizeof irq))
1768                         goto out;
1769                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1770                 if (r)
1771                         goto out;
1772                 r = 0;
1773                 break;
1774         }
1775         case KVM_NMI: {
1776                 r = kvm_vcpu_ioctl_nmi(vcpu);
1777                 if (r)
1778                         goto out;
1779                 r = 0;
1780                 break;
1781         }
1782         case KVM_SET_CPUID: {
1783                 struct kvm_cpuid __user *cpuid_arg = argp;
1784                 struct kvm_cpuid cpuid;
1785
1786                 r = -EFAULT;
1787                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1788                         goto out;
1789                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1790                 if (r)
1791                         goto out;
1792                 break;
1793         }
1794         case KVM_SET_CPUID2: {
1795                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1796                 struct kvm_cpuid2 cpuid;
1797
1798                 r = -EFAULT;
1799                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1800                         goto out;
1801                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1802                                               cpuid_arg->entries);
1803                 if (r)
1804                         goto out;
1805                 break;
1806         }
1807         case KVM_GET_CPUID2: {
1808                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1809                 struct kvm_cpuid2 cpuid;
1810
1811                 r = -EFAULT;
1812                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1813                         goto out;
1814                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1815                                               cpuid_arg->entries);
1816                 if (r)
1817                         goto out;
1818                 r = -EFAULT;
1819                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1820                         goto out;
1821                 r = 0;
1822                 break;
1823         }
1824         case KVM_GET_MSRS:
1825                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1826                 break;
1827         case KVM_SET_MSRS:
1828                 r = msr_io(vcpu, argp, do_set_msr, 0);
1829                 break;
1830         case KVM_TPR_ACCESS_REPORTING: {
1831                 struct kvm_tpr_access_ctl tac;
1832
1833                 r = -EFAULT;
1834                 if (copy_from_user(&tac, argp, sizeof tac))
1835                         goto out;
1836                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1837                 if (r)
1838                         goto out;
1839                 r = -EFAULT;
1840                 if (copy_to_user(argp, &tac, sizeof tac))
1841                         goto out;
1842                 r = 0;
1843                 break;
1844         };
1845         case KVM_SET_VAPIC_ADDR: {
1846                 struct kvm_vapic_addr va;
1847
1848                 r = -EINVAL;
1849                 if (!irqchip_in_kernel(vcpu->kvm))
1850                         goto out;
1851                 r = -EFAULT;
1852                 if (copy_from_user(&va, argp, sizeof va))
1853                         goto out;
1854                 r = 0;
1855                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1856                 break;
1857         }
1858         case KVM_X86_SETUP_MCE: {
1859                 u64 mcg_cap;
1860
1861                 r = -EFAULT;
1862                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
1863                         goto out;
1864                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
1865                 break;
1866         }
1867         case KVM_X86_SET_MCE: {
1868                 struct kvm_x86_mce mce;
1869
1870                 r = -EFAULT;
1871                 if (copy_from_user(&mce, argp, sizeof mce))
1872                         goto out;
1873                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
1874                 break;
1875         }
1876         default:
1877                 r = -EINVAL;
1878         }
1879 out:
1880         kfree(lapic);
1881         return r;
1882 }
1883
1884 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1885 {
1886         int ret;
1887
1888         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1889                 return -1;
1890         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1891         return ret;
1892 }
1893
1894 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1895                                           u32 kvm_nr_mmu_pages)
1896 {
1897         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1898                 return -EINVAL;
1899
1900         down_write(&kvm->slots_lock);
1901         spin_lock(&kvm->mmu_lock);
1902
1903         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1904         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1905
1906         spin_unlock(&kvm->mmu_lock);
1907         up_write(&kvm->slots_lock);
1908         return 0;
1909 }
1910
1911 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1912 {
1913         return kvm->arch.n_alloc_mmu_pages;
1914 }
1915
1916 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1917 {
1918         int i;
1919         struct kvm_mem_alias *alias;
1920
1921         for (i = 0; i < kvm->arch.naliases; ++i) {
1922                 alias = &kvm->arch.aliases[i];
1923                 if (gfn >= alias->base_gfn
1924                     && gfn < alias->base_gfn + alias->npages)
1925                         return alias->target_gfn + gfn - alias->base_gfn;
1926         }
1927         return gfn;
1928 }
1929
1930 /*
1931  * Set a new alias region.  Aliases map a portion of physical memory into
1932  * another portion.  This is useful for memory windows, for example the PC
1933  * VGA region.
1934  */
1935 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1936                                          struct kvm_memory_alias *alias)
1937 {
1938         int r, n;
1939         struct kvm_mem_alias *p;
1940
1941         r = -EINVAL;
1942         /* General sanity checks */
1943         if (alias->memory_size & (PAGE_SIZE - 1))
1944                 goto out;
1945         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1946                 goto out;
1947         if (alias->slot >= KVM_ALIAS_SLOTS)
1948                 goto out;
1949         if (alias->guest_phys_addr + alias->memory_size
1950             < alias->guest_phys_addr)
1951                 goto out;
1952         if (alias->target_phys_addr + alias->memory_size
1953             < alias->target_phys_addr)
1954                 goto out;
1955
1956         down_write(&kvm->slots_lock);
1957         spin_lock(&kvm->mmu_lock);
1958
1959         p = &kvm->arch.aliases[alias->slot];
1960         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1961         p->npages = alias->memory_size >> PAGE_SHIFT;
1962         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1963
1964         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1965                 if (kvm->arch.aliases[n - 1].npages)
1966                         break;
1967         kvm->arch.naliases = n;
1968
1969         spin_unlock(&kvm->mmu_lock);
1970         kvm_mmu_zap_all(kvm);
1971
1972         up_write(&kvm->slots_lock);
1973
1974         return 0;
1975
1976 out:
1977         return r;
1978 }
1979
1980 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1981 {
1982         int r;
1983
1984         r = 0;
1985         switch (chip->chip_id) {
1986         case KVM_IRQCHIP_PIC_MASTER:
1987                 memcpy(&chip->chip.pic,
1988                         &pic_irqchip(kvm)->pics[0],
1989                         sizeof(struct kvm_pic_state));
1990                 break;
1991         case KVM_IRQCHIP_PIC_SLAVE:
1992                 memcpy(&chip->chip.pic,
1993                         &pic_irqchip(kvm)->pics[1],
1994                         sizeof(struct kvm_pic_state));
1995                 break;
1996         case KVM_IRQCHIP_IOAPIC:
1997                 memcpy(&chip->chip.ioapic,
1998                         ioapic_irqchip(kvm),
1999                         sizeof(struct kvm_ioapic_state));
2000                 break;
2001         default:
2002                 r = -EINVAL;
2003                 break;
2004         }
2005         return r;
2006 }
2007
2008 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2009 {
2010         int r;
2011
2012         r = 0;
2013         switch (chip->chip_id) {
2014         case KVM_IRQCHIP_PIC_MASTER:
2015                 spin_lock(&pic_irqchip(kvm)->lock);
2016                 memcpy(&pic_irqchip(kvm)->pics[0],
2017                         &chip->chip.pic,
2018                         sizeof(struct kvm_pic_state));
2019                 spin_unlock(&pic_irqchip(kvm)->lock);
2020                 break;
2021         case KVM_IRQCHIP_PIC_SLAVE:
2022                 spin_lock(&pic_irqchip(kvm)->lock);
2023                 memcpy(&pic_irqchip(kvm)->pics[1],
2024                         &chip->chip.pic,
2025                         sizeof(struct kvm_pic_state));
2026                 spin_unlock(&pic_irqchip(kvm)->lock);
2027                 break;
2028         case KVM_IRQCHIP_IOAPIC:
2029                 mutex_lock(&kvm->irq_lock);
2030                 memcpy(ioapic_irqchip(kvm),
2031                         &chip->chip.ioapic,
2032                         sizeof(struct kvm_ioapic_state));
2033                 mutex_unlock(&kvm->irq_lock);
2034                 break;
2035         default:
2036                 r = -EINVAL;
2037                 break;
2038         }
2039         kvm_pic_update_irq(pic_irqchip(kvm));
2040         return r;
2041 }
2042
2043 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2044 {
2045         int r = 0;
2046
2047         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2048         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2049         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2050         return r;
2051 }
2052
2053 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2054 {
2055         int r = 0;
2056
2057         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2058         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2059         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
2060         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2061         return r;
2062 }
2063
2064 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2065                                  struct kvm_reinject_control *control)
2066 {
2067         if (!kvm->arch.vpit)
2068                 return -ENXIO;
2069         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2070         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2071         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2072         return 0;
2073 }
2074
2075 /*
2076  * Get (and clear) the dirty memory log for a memory slot.
2077  */
2078 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2079                                       struct kvm_dirty_log *log)
2080 {
2081         int r;
2082         int n;
2083         struct kvm_memory_slot *memslot;
2084         int is_dirty = 0;
2085
2086         down_write(&kvm->slots_lock);
2087
2088         r = kvm_get_dirty_log(kvm, log, &is_dirty);
2089         if (r)
2090                 goto out;
2091
2092         /* If nothing is dirty, don't bother messing with page tables. */
2093         if (is_dirty) {
2094                 spin_lock(&kvm->mmu_lock);
2095                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2096                 spin_unlock(&kvm->mmu_lock);
2097                 kvm_flush_remote_tlbs(kvm);
2098                 memslot = &kvm->memslots[log->slot];
2099                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2100                 memset(memslot->dirty_bitmap, 0, n);
2101         }
2102         r = 0;
2103 out:
2104         up_write(&kvm->slots_lock);
2105         return r;
2106 }
2107
2108 long kvm_arch_vm_ioctl(struct file *filp,
2109                        unsigned int ioctl, unsigned long arg)
2110 {
2111         struct kvm *kvm = filp->private_data;
2112         void __user *argp = (void __user *)arg;
2113         int r = -EINVAL;
2114         /*
2115          * This union makes it completely explicit to gcc-3.x
2116          * that these two variables' stack usage should be
2117          * combined, not added together.
2118          */
2119         union {
2120                 struct kvm_pit_state ps;
2121                 struct kvm_memory_alias alias;
2122                 struct kvm_pit_config pit_config;
2123         } u;
2124
2125         switch (ioctl) {
2126         case KVM_SET_TSS_ADDR:
2127                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2128                 if (r < 0)
2129                         goto out;
2130                 break;
2131         case KVM_SET_MEMORY_REGION: {
2132                 struct kvm_memory_region kvm_mem;
2133                 struct kvm_userspace_memory_region kvm_userspace_mem;
2134
2135                 r = -EFAULT;
2136                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2137                         goto out;
2138                 kvm_userspace_mem.slot = kvm_mem.slot;
2139                 kvm_userspace_mem.flags = kvm_mem.flags;
2140                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2141                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2142                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2143                 if (r)
2144                         goto out;
2145                 break;
2146         }
2147         case KVM_SET_NR_MMU_PAGES:
2148                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2149                 if (r)
2150                         goto out;
2151                 break;
2152         case KVM_GET_NR_MMU_PAGES:
2153                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2154                 break;
2155         case KVM_SET_MEMORY_ALIAS:
2156                 r = -EFAULT;
2157                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2158                         goto out;
2159                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2160                 if (r)
2161                         goto out;
2162                 break;
2163         case KVM_CREATE_IRQCHIP:
2164                 r = -ENOMEM;
2165                 kvm->arch.vpic = kvm_create_pic(kvm);
2166                 if (kvm->arch.vpic) {
2167                         r = kvm_ioapic_init(kvm);
2168                         if (r) {
2169                                 kfree(kvm->arch.vpic);
2170                                 kvm->arch.vpic = NULL;
2171                                 goto out;
2172                         }
2173                 } else
2174                         goto out;
2175                 r = kvm_setup_default_irq_routing(kvm);
2176                 if (r) {
2177                         kfree(kvm->arch.vpic);
2178                         kfree(kvm->arch.vioapic);
2179                         goto out;
2180                 }
2181                 break;
2182         case KVM_CREATE_PIT:
2183                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2184                 goto create_pit;
2185         case KVM_CREATE_PIT2:
2186                 r = -EFAULT;
2187                 if (copy_from_user(&u.pit_config, argp,
2188                                    sizeof(struct kvm_pit_config)))
2189                         goto out;
2190         create_pit:
2191                 down_write(&kvm->slots_lock);
2192                 r = -EEXIST;
2193                 if (kvm->arch.vpit)
2194                         goto create_pit_unlock;
2195                 r = -ENOMEM;
2196                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2197                 if (kvm->arch.vpit)
2198                         r = 0;
2199         create_pit_unlock:
2200                 up_write(&kvm->slots_lock);
2201                 break;
2202         case KVM_IRQ_LINE_STATUS:
2203         case KVM_IRQ_LINE: {
2204                 struct kvm_irq_level irq_event;
2205
2206                 r = -EFAULT;
2207                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2208                         goto out;
2209                 if (irqchip_in_kernel(kvm)) {
2210                         __s32 status;
2211                         mutex_lock(&kvm->irq_lock);
2212                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2213                                         irq_event.irq, irq_event.level);
2214                         mutex_unlock(&kvm->irq_lock);
2215                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2216                                 irq_event.status = status;
2217                                 if (copy_to_user(argp, &irq_event,
2218                                                         sizeof irq_event))
2219                                         goto out;
2220                         }
2221                         r = 0;
2222                 }
2223                 break;
2224         }
2225         case KVM_GET_IRQCHIP: {
2226                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2227                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2228
2229                 r = -ENOMEM;
2230                 if (!chip)
2231                         goto out;
2232                 r = -EFAULT;
2233                 if (copy_from_user(chip, argp, sizeof *chip))
2234                         goto get_irqchip_out;
2235                 r = -ENXIO;
2236                 if (!irqchip_in_kernel(kvm))
2237                         goto get_irqchip_out;
2238                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2239                 if (r)
2240                         goto get_irqchip_out;
2241                 r = -EFAULT;
2242                 if (copy_to_user(argp, chip, sizeof *chip))
2243                         goto get_irqchip_out;
2244                 r = 0;
2245         get_irqchip_out:
2246                 kfree(chip);
2247                 if (r)
2248                         goto out;
2249                 break;
2250         }
2251         case KVM_SET_IRQCHIP: {
2252                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2253                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2254
2255                 r = -ENOMEM;
2256                 if (!chip)
2257                         goto out;
2258                 r = -EFAULT;
2259                 if (copy_from_user(chip, argp, sizeof *chip))
2260                         goto set_irqchip_out;
2261                 r = -ENXIO;
2262                 if (!irqchip_in_kernel(kvm))
2263                         goto set_irqchip_out;
2264                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2265                 if (r)
2266                         goto set_irqchip_out;
2267                 r = 0;
2268         set_irqchip_out:
2269                 kfree(chip);
2270                 if (r)
2271                         goto out;
2272                 break;
2273         }
2274         case KVM_GET_PIT: {
2275                 r = -EFAULT;
2276                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2277                         goto out;
2278                 r = -ENXIO;
2279                 if (!kvm->arch.vpit)
2280                         goto out;
2281                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2282                 if (r)
2283                         goto out;
2284                 r = -EFAULT;
2285                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2286                         goto out;
2287                 r = 0;
2288                 break;
2289         }
2290         case KVM_SET_PIT: {
2291                 r = -EFAULT;
2292                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2293                         goto out;
2294                 r = -ENXIO;
2295                 if (!kvm->arch.vpit)
2296                         goto out;
2297                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2298                 if (r)
2299                         goto out;
2300                 r = 0;
2301                 break;
2302         }
2303         case KVM_REINJECT_CONTROL: {
2304                 struct kvm_reinject_control control;
2305                 r =  -EFAULT;
2306                 if (copy_from_user(&control, argp, sizeof(control)))
2307                         goto out;
2308                 r = kvm_vm_ioctl_reinject(kvm, &control);
2309                 if (r)
2310                         goto out;
2311                 r = 0;
2312                 break;
2313         }
2314         default:
2315                 ;
2316         }
2317 out:
2318         return r;
2319 }
2320
2321 static void kvm_init_msr_list(void)
2322 {
2323         u32 dummy[2];
2324         unsigned i, j;
2325
2326         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2327                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2328                         continue;
2329                 if (j < i)
2330                         msrs_to_save[j] = msrs_to_save[i];
2331                 j++;
2332         }
2333         num_msrs_to_save = j;
2334 }
2335
2336 /*
2337  * Only apic need an MMIO device hook, so shortcut now..
2338  */
2339 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2340                                                 gpa_t addr, int len,
2341                                                 int is_write)
2342 {
2343         struct kvm_io_device *dev;
2344
2345         if (vcpu->arch.apic) {
2346                 dev = &vcpu->arch.apic->dev;
2347                 if (kvm_iodevice_in_range(dev, addr, len, is_write))
2348                         return dev;
2349         }
2350         return NULL;
2351 }
2352
2353
2354 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2355                                                 gpa_t addr, int len,
2356                                                 int is_write)
2357 {
2358         struct kvm_io_device *dev;
2359
2360         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2361         if (dev == NULL)
2362                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2363                                           is_write);
2364         return dev;
2365 }
2366
2367 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2368                                struct kvm_vcpu *vcpu)
2369 {
2370         void *data = val;
2371         int r = X86EMUL_CONTINUE;
2372
2373         while (bytes) {
2374                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2375                 unsigned offset = addr & (PAGE_SIZE-1);
2376                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2377                 int ret;
2378
2379                 if (gpa == UNMAPPED_GVA) {
2380                         r = X86EMUL_PROPAGATE_FAULT;
2381                         goto out;
2382                 }
2383                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2384                 if (ret < 0) {
2385                         r = X86EMUL_UNHANDLEABLE;
2386                         goto out;
2387                 }
2388
2389                 bytes -= toread;
2390                 data += toread;
2391                 addr += toread;
2392         }
2393 out:
2394         return r;
2395 }
2396
2397 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2398                                 struct kvm_vcpu *vcpu)
2399 {
2400         void *data = val;
2401         int r = X86EMUL_CONTINUE;
2402
2403         while (bytes) {
2404                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2405                 unsigned offset = addr & (PAGE_SIZE-1);
2406                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2407                 int ret;
2408
2409                 if (gpa == UNMAPPED_GVA) {
2410                         r = X86EMUL_PROPAGATE_FAULT;
2411                         goto out;
2412                 }
2413                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2414                 if (ret < 0) {
2415                         r = X86EMUL_UNHANDLEABLE;
2416                         goto out;
2417                 }
2418
2419                 bytes -= towrite;
2420                 data += towrite;
2421                 addr += towrite;
2422         }
2423 out:
2424         return r;
2425 }
2426
2427
2428 static int emulator_read_emulated(unsigned long addr,
2429                                   void *val,
2430                                   unsigned int bytes,
2431                                   struct kvm_vcpu *vcpu)
2432 {
2433         struct kvm_io_device *mmio_dev;
2434         gpa_t                 gpa;
2435
2436         if (vcpu->mmio_read_completed) {
2437                 memcpy(val, vcpu->mmio_data, bytes);
2438                 vcpu->mmio_read_completed = 0;
2439                 return X86EMUL_CONTINUE;
2440         }
2441
2442         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2443
2444         /* For APIC access vmexit */
2445         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2446                 goto mmio;
2447
2448         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2449                                 == X86EMUL_CONTINUE)
2450                 return X86EMUL_CONTINUE;
2451         if (gpa == UNMAPPED_GVA)
2452                 return X86EMUL_PROPAGATE_FAULT;
2453
2454 mmio:
2455         /*
2456          * Is this MMIO handled locally?
2457          */
2458         mutex_lock(&vcpu->kvm->lock);
2459         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2460         mutex_unlock(&vcpu->kvm->lock);
2461         if (mmio_dev) {
2462                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2463                 return X86EMUL_CONTINUE;
2464         }
2465
2466         vcpu->mmio_needed = 1;
2467         vcpu->mmio_phys_addr = gpa;
2468         vcpu->mmio_size = bytes;
2469         vcpu->mmio_is_write = 0;
2470
2471         return X86EMUL_UNHANDLEABLE;
2472 }
2473
2474 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2475                           const void *val, int bytes)
2476 {
2477         int ret;
2478
2479         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2480         if (ret < 0)
2481                 return 0;
2482         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2483         return 1;
2484 }
2485
2486 static int emulator_write_emulated_onepage(unsigned long addr,
2487                                            const void *val,
2488                                            unsigned int bytes,
2489                                            struct kvm_vcpu *vcpu)
2490 {
2491         struct kvm_io_device *mmio_dev;
2492         gpa_t                 gpa;
2493
2494         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2495
2496         if (gpa == UNMAPPED_GVA) {
2497                 kvm_inject_page_fault(vcpu, addr, 2);
2498                 return X86EMUL_PROPAGATE_FAULT;
2499         }
2500
2501         /* For APIC access vmexit */
2502         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2503                 goto mmio;
2504
2505         if (emulator_write_phys(vcpu, gpa, val, bytes))
2506                 return X86EMUL_CONTINUE;
2507
2508 mmio:
2509         /*
2510          * Is this MMIO handled locally?
2511          */
2512         mutex_lock(&vcpu->kvm->lock);
2513         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2514         mutex_unlock(&vcpu->kvm->lock);
2515         if (mmio_dev) {
2516                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2517                 return X86EMUL_CONTINUE;
2518         }
2519
2520         vcpu->mmio_needed = 1;
2521         vcpu->mmio_phys_addr = gpa;
2522         vcpu->mmio_size = bytes;
2523         vcpu->mmio_is_write = 1;
2524         memcpy(vcpu->mmio_data, val, bytes);
2525
2526         return X86EMUL_CONTINUE;
2527 }
2528
2529 int emulator_write_emulated(unsigned long addr,
2530                                    const void *val,
2531                                    unsigned int bytes,
2532                                    struct kvm_vcpu *vcpu)
2533 {
2534         /* Crossing a page boundary? */
2535         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2536                 int rc, now;
2537
2538                 now = -addr & ~PAGE_MASK;
2539                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2540                 if (rc != X86EMUL_CONTINUE)
2541                         return rc;
2542                 addr += now;
2543                 val += now;
2544                 bytes -= now;
2545         }
2546         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2547 }
2548 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2549
2550 static int emulator_cmpxchg_emulated(unsigned long addr,
2551                                      const void *old,
2552                                      const void *new,
2553                                      unsigned int bytes,
2554                                      struct kvm_vcpu *vcpu)
2555 {
2556         static int reported;
2557
2558         if (!reported) {
2559                 reported = 1;
2560                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2561         }
2562 #ifndef CONFIG_X86_64
2563         /* guests cmpxchg8b have to be emulated atomically */
2564         if (bytes == 8) {
2565                 gpa_t gpa;
2566                 struct page *page;
2567                 char *kaddr;
2568                 u64 val;
2569
2570                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2571
2572                 if (gpa == UNMAPPED_GVA ||
2573                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2574                         goto emul_write;
2575
2576                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2577                         goto emul_write;
2578
2579                 val = *(u64 *)new;
2580
2581                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2582
2583                 kaddr = kmap_atomic(page, KM_USER0);
2584                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2585                 kunmap_atomic(kaddr, KM_USER0);
2586                 kvm_release_page_dirty(page);
2587         }
2588 emul_write:
2589 #endif
2590
2591         return emulator_write_emulated(addr, new, bytes, vcpu);
2592 }
2593
2594 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2595 {
2596         return kvm_x86_ops->get_segment_base(vcpu, seg);
2597 }
2598
2599 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2600 {
2601         kvm_mmu_invlpg(vcpu, address);
2602         return X86EMUL_CONTINUE;
2603 }
2604
2605 int emulate_clts(struct kvm_vcpu *vcpu)
2606 {
2607         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2608         return X86EMUL_CONTINUE;
2609 }
2610
2611 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2612 {
2613         struct kvm_vcpu *vcpu = ctxt->vcpu;
2614
2615         switch (dr) {
2616         case 0 ... 3:
2617                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2618                 return X86EMUL_CONTINUE;
2619         default:
2620                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2621                 return X86EMUL_UNHANDLEABLE;
2622         }
2623 }
2624
2625 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2626 {
2627         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2628         int exception;
2629
2630         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2631         if (exception) {
2632                 /* FIXME: better handling */
2633                 return X86EMUL_UNHANDLEABLE;
2634         }
2635         return X86EMUL_CONTINUE;
2636 }
2637
2638 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2639 {
2640         u8 opcodes[4];
2641         unsigned long rip = kvm_rip_read(vcpu);
2642         unsigned long rip_linear;
2643
2644         if (!printk_ratelimit())
2645                 return;
2646
2647         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2648
2649         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2650
2651         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2652                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2653 }
2654 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2655
2656 static struct x86_emulate_ops emulate_ops = {
2657         .read_std            = kvm_read_guest_virt,
2658         .read_emulated       = emulator_read_emulated,
2659         .write_emulated      = emulator_write_emulated,
2660         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2661 };
2662
2663 static void cache_all_regs(struct kvm_vcpu *vcpu)
2664 {
2665         kvm_register_read(vcpu, VCPU_REGS_RAX);
2666         kvm_register_read(vcpu, VCPU_REGS_RSP);
2667         kvm_register_read(vcpu, VCPU_REGS_RIP);
2668         vcpu->arch.regs_dirty = ~0;
2669 }
2670
2671 int emulate_instruction(struct kvm_vcpu *vcpu,
2672                         struct kvm_run *run,
2673                         unsigned long cr2,
2674                         u16 error_code,
2675                         int emulation_type)
2676 {
2677         int r, shadow_mask;
2678         struct decode_cache *c;
2679
2680         kvm_clear_exception_queue(vcpu);
2681         vcpu->arch.mmio_fault_cr2 = cr2;
2682         /*
2683          * TODO: fix x86_emulate.c to use guest_read/write_register
2684          * instead of direct ->regs accesses, can save hundred cycles
2685          * on Intel for instructions that don't read/change RSP, for
2686          * for example.
2687          */
2688         cache_all_regs(vcpu);
2689
2690         vcpu->mmio_is_write = 0;
2691         vcpu->arch.pio.string = 0;
2692
2693         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2694                 int cs_db, cs_l;
2695                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2696
2697                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2698                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2699                 vcpu->arch.emulate_ctxt.mode =
2700                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2701                         ? X86EMUL_MODE_REAL : cs_l
2702                         ? X86EMUL_MODE_PROT64 : cs_db
2703                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2704
2705                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2706
2707                 /* Only allow emulation of specific instructions on #UD
2708                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
2709                 c = &vcpu->arch.emulate_ctxt.decode;
2710                 if (emulation_type & EMULTYPE_TRAP_UD) {
2711                         if (!c->twobyte)
2712                                 return EMULATE_FAIL;
2713                         switch (c->b) {
2714                         case 0x01: /* VMMCALL */
2715                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
2716                                         return EMULATE_FAIL;
2717                                 break;
2718                         case 0x34: /* sysenter */
2719                         case 0x35: /* sysexit */
2720                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2721                                         return EMULATE_FAIL;
2722                                 break;
2723                         case 0x05: /* syscall */
2724                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2725                                         return EMULATE_FAIL;
2726                                 break;
2727                         default:
2728                                 return EMULATE_FAIL;
2729                         }
2730
2731                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
2732                                 return EMULATE_FAIL;
2733                 }
2734
2735                 ++vcpu->stat.insn_emulation;
2736                 if (r)  {
2737                         ++vcpu->stat.insn_emulation_fail;
2738                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2739                                 return EMULATE_DONE;
2740                         return EMULATE_FAIL;
2741                 }
2742         }
2743
2744         if (emulation_type & EMULTYPE_SKIP) {
2745                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2746                 return EMULATE_DONE;
2747         }
2748
2749         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2750         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2751
2752         if (r == 0)
2753                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2754
2755         if (vcpu->arch.pio.string)
2756                 return EMULATE_DO_MMIO;
2757
2758         if ((r || vcpu->mmio_is_write) && run) {
2759                 run->exit_reason = KVM_EXIT_MMIO;
2760                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2761                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2762                 run->mmio.len = vcpu->mmio_size;
2763                 run->mmio.is_write = vcpu->mmio_is_write;
2764         }
2765
2766         if (r) {
2767                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2768                         return EMULATE_DONE;
2769                 if (!vcpu->mmio_needed) {
2770                         kvm_report_emulation_failure(vcpu, "mmio");
2771                         return EMULATE_FAIL;
2772                 }
2773                 return EMULATE_DO_MMIO;
2774         }
2775
2776         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2777
2778         if (vcpu->mmio_is_write) {
2779                 vcpu->mmio_needed = 0;
2780                 return EMULATE_DO_MMIO;
2781         }
2782
2783         return EMULATE_DONE;
2784 }
2785 EXPORT_SYMBOL_GPL(emulate_instruction);
2786
2787 static int pio_copy_data(struct kvm_vcpu *vcpu)
2788 {
2789         void *p = vcpu->arch.pio_data;
2790         gva_t q = vcpu->arch.pio.guest_gva;
2791         unsigned bytes;
2792         int ret;
2793
2794         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2795         if (vcpu->arch.pio.in)
2796                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2797         else
2798                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2799         return ret;
2800 }
2801
2802 int complete_pio(struct kvm_vcpu *vcpu)
2803 {
2804         struct kvm_pio_request *io = &vcpu->arch.pio;
2805         long delta;
2806         int r;
2807         unsigned long val;
2808
2809         if (!io->string) {
2810                 if (io->in) {
2811                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2812                         memcpy(&val, vcpu->arch.pio_data, io->size);
2813                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2814                 }
2815         } else {
2816                 if (io->in) {
2817                         r = pio_copy_data(vcpu);
2818                         if (r)
2819                                 return r;
2820                 }
2821
2822                 delta = 1;
2823                 if (io->rep) {
2824                         delta *= io->cur_count;
2825                         /*
2826                          * The size of the register should really depend on
2827                          * current address size.
2828                          */
2829                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2830                         val -= delta;
2831                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2832                 }
2833                 if (io->down)
2834                         delta = -delta;
2835                 delta *= io->size;
2836                 if (io->in) {
2837                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2838                         val += delta;
2839                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2840                 } else {
2841                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2842                         val += delta;
2843                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2844                 }
2845         }
2846
2847         io->count -= io->cur_count;
2848         io->cur_count = 0;
2849
2850         return 0;
2851 }
2852
2853 static void kernel_pio(struct kvm_io_device *pio_dev,
2854                        struct kvm_vcpu *vcpu,
2855                        void *pd)
2856 {
2857         /* TODO: String I/O for in kernel device */
2858
2859         if (vcpu->arch.pio.in)
2860                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2861                                   vcpu->arch.pio.size,
2862                                   pd);
2863         else
2864                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2865                                    vcpu->arch.pio.size,
2866                                    pd);
2867 }
2868
2869 static void pio_string_write(struct kvm_io_device *pio_dev,
2870                              struct kvm_vcpu *vcpu)
2871 {
2872         struct kvm_pio_request *io = &vcpu->arch.pio;
2873         void *pd = vcpu->arch.pio_data;
2874         int i;
2875
2876         for (i = 0; i < io->cur_count; i++) {
2877                 kvm_iodevice_write(pio_dev, io->port,
2878                                    io->size,
2879                                    pd);
2880                 pd += io->size;
2881         }
2882 }
2883
2884 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2885                                                gpa_t addr, int len,
2886                                                int is_write)
2887 {
2888         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2889 }
2890
2891 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2892                   int size, unsigned port)
2893 {
2894         struct kvm_io_device *pio_dev;
2895         unsigned long val;
2896
2897         vcpu->run->exit_reason = KVM_EXIT_IO;
2898         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2899         vcpu->run->io.size = vcpu->arch.pio.size = size;
2900         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2901         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2902         vcpu->run->io.port = vcpu->arch.pio.port = port;
2903         vcpu->arch.pio.in = in;
2904         vcpu->arch.pio.string = 0;
2905         vcpu->arch.pio.down = 0;
2906         vcpu->arch.pio.rep = 0;
2907
2908         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
2909                       size, 1);
2910
2911         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2912         memcpy(vcpu->arch.pio_data, &val, 4);
2913
2914         mutex_lock(&vcpu->kvm->lock);
2915         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2916         mutex_unlock(&vcpu->kvm->lock);
2917         if (pio_dev) {
2918                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2919                 complete_pio(vcpu);
2920                 return 1;
2921         }
2922         return 0;
2923 }
2924 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2925
2926 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2927                   int size, unsigned long count, int down,
2928                   gva_t address, int rep, unsigned port)
2929 {
2930         unsigned now, in_page;
2931         int ret = 0;
2932         struct kvm_io_device *pio_dev;
2933
2934         vcpu->run->exit_reason = KVM_EXIT_IO;
2935         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2936         vcpu->run->io.size = vcpu->arch.pio.size = size;
2937         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2938         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2939         vcpu->run->io.port = vcpu->arch.pio.port = port;
2940         vcpu->arch.pio.in = in;
2941         vcpu->arch.pio.string = 1;
2942         vcpu->arch.pio.down = down;
2943         vcpu->arch.pio.rep = rep;
2944
2945         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
2946                       size, count);
2947
2948         if (!count) {
2949                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2950                 return 1;
2951         }
2952
2953         if (!down)
2954                 in_page = PAGE_SIZE - offset_in_page(address);
2955         else
2956                 in_page = offset_in_page(address) + size;
2957         now = min(count, (unsigned long)in_page / size);
2958         if (!now)
2959                 now = 1;
2960         if (down) {
2961                 /*
2962                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2963                  */
2964                 pr_unimpl(vcpu, "guest string pio down\n");
2965                 kvm_inject_gp(vcpu, 0);
2966                 return 1;
2967         }
2968         vcpu->run->io.count = now;
2969         vcpu->arch.pio.cur_count = now;
2970
2971         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2972                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2973
2974         vcpu->arch.pio.guest_gva = address;
2975
2976         mutex_lock(&vcpu->kvm->lock);
2977         pio_dev = vcpu_find_pio_dev(vcpu, port,
2978                                     vcpu->arch.pio.cur_count,
2979                                     !vcpu->arch.pio.in);
2980         mutex_unlock(&vcpu->kvm->lock);
2981
2982         if (!vcpu->arch.pio.in) {
2983                 /* string PIO write */
2984                 ret = pio_copy_data(vcpu);
2985                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2986                         kvm_inject_gp(vcpu, 0);
2987                         return 1;
2988                 }
2989                 if (ret == 0 && pio_dev) {
2990                         pio_string_write(pio_dev, vcpu);
2991                         complete_pio(vcpu);
2992                         if (vcpu->arch.pio.count == 0)
2993                                 ret = 1;
2994                 }
2995         } else if (pio_dev)
2996                 pr_unimpl(vcpu, "no string pio read support yet, "
2997                        "port %x size %d count %ld\n",
2998                         port, size, count);
2999
3000         return ret;
3001 }
3002 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3003
3004 static void bounce_off(void *info)
3005 {
3006         /* nothing */
3007 }
3008
3009 static unsigned int  ref_freq;
3010 static unsigned long tsc_khz_ref;
3011
3012 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3013                                      void *data)
3014 {
3015         struct cpufreq_freqs *freq = data;
3016         struct kvm *kvm;
3017         struct kvm_vcpu *vcpu;
3018         int i, send_ipi = 0;
3019
3020         if (!ref_freq)
3021                 ref_freq = freq->old;
3022
3023         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3024                 return 0;
3025         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3026                 return 0;
3027         per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
3028
3029         spin_lock(&kvm_lock);
3030         list_for_each_entry(kvm, &vm_list, vm_list) {
3031                 kvm_for_each_vcpu(i, vcpu, kvm) {
3032                         if (vcpu->cpu != freq->cpu)
3033                                 continue;
3034                         if (!kvm_request_guest_time_update(vcpu))
3035                                 continue;
3036                         if (vcpu->cpu != smp_processor_id())
3037                                 send_ipi++;
3038                 }
3039         }
3040         spin_unlock(&kvm_lock);
3041
3042         if (freq->old < freq->new && send_ipi) {
3043                 /*
3044                  * We upscale the frequency.  Must make the guest
3045                  * doesn't see old kvmclock values while running with
3046                  * the new frequency, otherwise we risk the guest sees
3047                  * time go backwards.
3048                  *
3049                  * In case we update the frequency for another cpu
3050                  * (which might be in guest context) send an interrupt
3051                  * to kick the cpu out of guest context.  Next time
3052                  * guest context is entered kvmclock will be updated,
3053                  * so the guest will not see stale values.
3054                  */
3055                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3056         }
3057         return 0;
3058 }
3059
3060 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3061         .notifier_call  = kvmclock_cpufreq_notifier
3062 };
3063
3064 int kvm_arch_init(void *opaque)
3065 {
3066         int r, cpu;
3067         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3068
3069         if (kvm_x86_ops) {
3070                 printk(KERN_ERR "kvm: already loaded the other module\n");
3071                 r = -EEXIST;
3072                 goto out;
3073         }
3074
3075         if (!ops->cpu_has_kvm_support()) {
3076                 printk(KERN_ERR "kvm: no hardware support\n");
3077                 r = -EOPNOTSUPP;
3078                 goto out;
3079         }
3080         if (ops->disabled_by_bios()) {
3081                 printk(KERN_ERR "kvm: disabled by bios\n");
3082                 r = -EOPNOTSUPP;
3083                 goto out;
3084         }
3085
3086         r = kvm_mmu_module_init();
3087         if (r)
3088                 goto out;
3089
3090         kvm_init_msr_list();
3091
3092         kvm_x86_ops = ops;
3093         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3094         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3095         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3096                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3097
3098         for_each_possible_cpu(cpu)
3099                 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3100         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3101                 tsc_khz_ref = tsc_khz;
3102                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3103                                           CPUFREQ_TRANSITION_NOTIFIER);
3104         }
3105
3106         return 0;
3107
3108 out:
3109         return r;
3110 }
3111
3112 void kvm_arch_exit(void)
3113 {
3114         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3115                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3116                                             CPUFREQ_TRANSITION_NOTIFIER);
3117         kvm_x86_ops = NULL;
3118         kvm_mmu_module_exit();
3119 }
3120
3121 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3122 {
3123         ++vcpu->stat.halt_exits;
3124         if (irqchip_in_kernel(vcpu->kvm)) {
3125                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3126                 return 1;
3127         } else {
3128                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3129                 return 0;
3130         }
3131 }
3132 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3133
3134 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3135                            unsigned long a1)
3136 {
3137         if (is_long_mode(vcpu))
3138                 return a0;
3139         else
3140                 return a0 | ((gpa_t)a1 << 32);
3141 }
3142
3143 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3144 {
3145         unsigned long nr, a0, a1, a2, a3, ret;
3146         int r = 1;
3147
3148         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3149         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3150         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3151         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3152         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3153
3154         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3155
3156         if (!is_long_mode(vcpu)) {
3157                 nr &= 0xFFFFFFFF;
3158                 a0 &= 0xFFFFFFFF;
3159                 a1 &= 0xFFFFFFFF;
3160                 a2 &= 0xFFFFFFFF;
3161                 a3 &= 0xFFFFFFFF;
3162         }
3163
3164         switch (nr) {
3165         case KVM_HC_VAPIC_POLL_IRQ:
3166                 ret = 0;
3167                 break;
3168         case KVM_HC_MMU_OP:
3169                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3170                 break;
3171         default:
3172                 ret = -KVM_ENOSYS;
3173                 break;
3174         }
3175         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3176         ++vcpu->stat.hypercalls;
3177         return r;
3178 }
3179 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3180
3181 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3182 {
3183         char instruction[3];
3184         int ret = 0;
3185         unsigned long rip = kvm_rip_read(vcpu);
3186
3187
3188         /*
3189          * Blow out the MMU to ensure that no other VCPU has an active mapping
3190          * to ensure that the updated hypercall appears atomically across all
3191          * VCPUs.
3192          */
3193         kvm_mmu_zap_all(vcpu->kvm);
3194
3195         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3196         if (emulator_write_emulated(rip, instruction, 3, vcpu)
3197             != X86EMUL_CONTINUE)
3198                 ret = -EFAULT;
3199
3200         return ret;
3201 }
3202
3203 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3204 {
3205         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3206 }
3207
3208 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3209 {
3210         struct descriptor_table dt = { limit, base };
3211
3212         kvm_x86_ops->set_gdt(vcpu, &dt);
3213 }
3214
3215 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3216 {
3217         struct descriptor_table dt = { limit, base };
3218
3219         kvm_x86_ops->set_idt(vcpu, &dt);
3220 }
3221
3222 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3223                    unsigned long *rflags)
3224 {
3225         kvm_lmsw(vcpu, msw);
3226         *rflags = kvm_x86_ops->get_rflags(vcpu);
3227 }
3228
3229 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3230 {
3231         unsigned long value;
3232
3233         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3234         switch (cr) {
3235         case 0:
3236                 value = vcpu->arch.cr0;
3237                 break;
3238         case 2:
3239                 value = vcpu->arch.cr2;
3240                 break;
3241         case 3:
3242                 value = vcpu->arch.cr3;
3243                 break;
3244         case 4:
3245                 value = vcpu->arch.cr4;
3246                 break;
3247         case 8:
3248                 value = kvm_get_cr8(vcpu);
3249                 break;
3250         default:
3251                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3252                 return 0;
3253         }
3254
3255         return value;
3256 }
3257
3258 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3259                      unsigned long *rflags)
3260 {
3261         switch (cr) {
3262         case 0:
3263                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3264                 *rflags = kvm_x86_ops->get_rflags(vcpu);
3265                 break;
3266         case 2:
3267                 vcpu->arch.cr2 = val;
3268                 break;
3269         case 3:
3270                 kvm_set_cr3(vcpu, val);
3271                 break;
3272         case 4:
3273                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3274                 break;
3275         case 8:
3276                 kvm_set_cr8(vcpu, val & 0xfUL);
3277                 break;
3278         default:
3279                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3280         }
3281 }
3282
3283 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3284 {
3285         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3286         int j, nent = vcpu->arch.cpuid_nent;
3287
3288         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3289         /* when no next entry is found, the current entry[i] is reselected */
3290         for (j = i + 1; ; j = (j + 1) % nent) {
3291                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3292                 if (ej->function == e->function) {
3293                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3294                         return j;
3295                 }
3296         }
3297         return 0; /* silence gcc, even though control never reaches here */
3298 }
3299
3300 /* find an entry with matching function, matching index (if needed), and that
3301  * should be read next (if it's stateful) */
3302 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3303         u32 function, u32 index)
3304 {
3305         if (e->function != function)
3306                 return 0;
3307         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3308                 return 0;
3309         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3310             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3311                 return 0;
3312         return 1;
3313 }
3314
3315 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3316                                               u32 function, u32 index)
3317 {
3318         int i;
3319         struct kvm_cpuid_entry2 *best = NULL;
3320
3321         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3322                 struct kvm_cpuid_entry2 *e;
3323
3324                 e = &vcpu->arch.cpuid_entries[i];
3325                 if (is_matching_cpuid_entry(e, function, index)) {
3326                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3327                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3328                         best = e;
3329                         break;
3330                 }
3331                 /*
3332                  * Both basic or both extended?
3333                  */
3334                 if (((e->function ^ function) & 0x80000000) == 0)
3335                         if (!best || e->function > best->function)
3336                                 best = e;
3337         }
3338         return best;
3339 }
3340
3341 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3342 {
3343         struct kvm_cpuid_entry2 *best;
3344
3345         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3346         if (best)
3347                 return best->eax & 0xff;
3348         return 36;
3349 }
3350
3351 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3352 {
3353         u32 function, index;
3354         struct kvm_cpuid_entry2 *best;
3355
3356         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3357         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3358         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3359         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3360         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3361         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3362         best = kvm_find_cpuid_entry(vcpu, function, index);
3363         if (best) {
3364                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3365                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3366                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3367                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3368         }
3369         kvm_x86_ops->skip_emulated_instruction(vcpu);
3370         trace_kvm_cpuid(function,
3371                         kvm_register_read(vcpu, VCPU_REGS_RAX),
3372                         kvm_register_read(vcpu, VCPU_REGS_RBX),
3373                         kvm_register_read(vcpu, VCPU_REGS_RCX),
3374                         kvm_register_read(vcpu, VCPU_REGS_RDX));
3375 }
3376 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3377
3378 /*
3379  * Check if userspace requested an interrupt window, and that the
3380  * interrupt window is open.
3381  *
3382  * No need to exit to userspace if we already have an interrupt queued.
3383  */
3384 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3385                                           struct kvm_run *kvm_run)
3386 {
3387         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3388                 kvm_run->request_interrupt_window &&
3389                 kvm_arch_interrupt_allowed(vcpu));
3390 }
3391
3392 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3393                               struct kvm_run *kvm_run)
3394 {
3395         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3396         kvm_run->cr8 = kvm_get_cr8(vcpu);
3397         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3398         if (irqchip_in_kernel(vcpu->kvm))
3399                 kvm_run->ready_for_interrupt_injection = 1;
3400         else
3401                 kvm_run->ready_for_interrupt_injection =
3402                         kvm_arch_interrupt_allowed(vcpu) &&
3403                         !kvm_cpu_has_interrupt(vcpu) &&
3404                         !kvm_event_needs_reinjection(vcpu);
3405 }
3406
3407 static void vapic_enter(struct kvm_vcpu *vcpu)
3408 {
3409         struct kvm_lapic *apic = vcpu->arch.apic;
3410         struct page *page;
3411
3412         if (!apic || !apic->vapic_addr)
3413                 return;
3414
3415         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3416
3417         vcpu->arch.apic->vapic_page = page;
3418 }
3419
3420 static void vapic_exit(struct kvm_vcpu *vcpu)
3421 {
3422         struct kvm_lapic *apic = vcpu->arch.apic;
3423
3424         if (!apic || !apic->vapic_addr)
3425                 return;
3426
3427         down_read(&vcpu->kvm->slots_lock);
3428         kvm_release_page_dirty(apic->vapic_page);
3429         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3430         up_read(&vcpu->kvm->slots_lock);
3431 }
3432
3433 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3434 {
3435         int max_irr, tpr;
3436
3437         if (!kvm_x86_ops->update_cr8_intercept)
3438                 return;
3439
3440         if (!vcpu->arch.apic->vapic_addr)
3441                 max_irr = kvm_lapic_find_highest_irr(vcpu);
3442         else
3443                 max_irr = -1;
3444
3445         if (max_irr != -1)
3446                 max_irr >>= 4;
3447
3448         tpr = kvm_lapic_get_cr8(vcpu);
3449
3450         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3451 }
3452
3453 static void inject_pending_irq(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3454 {
3455         /* try to reinject previous events if any */
3456         if (vcpu->arch.nmi_injected) {
3457                 kvm_x86_ops->set_nmi(vcpu);
3458                 return;
3459         }
3460
3461         if (vcpu->arch.interrupt.pending) {
3462                 kvm_x86_ops->set_irq(vcpu);
3463                 return;
3464         }
3465
3466         /* try to inject new event if pending */
3467         if (vcpu->arch.nmi_pending) {
3468                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3469                         vcpu->arch.nmi_pending = false;
3470                         vcpu->arch.nmi_injected = true;
3471                         kvm_x86_ops->set_nmi(vcpu);
3472                 }
3473         } else if (kvm_cpu_has_interrupt(vcpu)) {
3474                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3475                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3476                                             false);
3477                         kvm_x86_ops->set_irq(vcpu);
3478                 }
3479         }
3480 }
3481
3482 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3483 {
3484         int r;
3485         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3486                 kvm_run->request_interrupt_window;
3487
3488         if (vcpu->requests)
3489                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3490                         kvm_mmu_unload(vcpu);
3491
3492         r = kvm_mmu_reload(vcpu);
3493         if (unlikely(r))
3494                 goto out;
3495
3496         if (vcpu->requests) {
3497                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3498                         __kvm_migrate_timers(vcpu);
3499                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3500                         kvm_write_guest_time(vcpu);
3501                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3502                         kvm_mmu_sync_roots(vcpu);
3503                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3504                         kvm_x86_ops->tlb_flush(vcpu);
3505                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3506                                        &vcpu->requests)) {
3507                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3508                         r = 0;
3509                         goto out;
3510                 }
3511                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3512                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3513                         r = 0;
3514                         goto out;
3515                 }
3516         }
3517
3518         preempt_disable();
3519
3520         kvm_x86_ops->prepare_guest_switch(vcpu);
3521         kvm_load_guest_fpu(vcpu);
3522
3523         local_irq_disable();
3524
3525         clear_bit(KVM_REQ_KICK, &vcpu->requests);
3526         smp_mb__after_clear_bit();
3527
3528         if (vcpu->requests || need_resched() || signal_pending(current)) {
3529                 local_irq_enable();
3530                 preempt_enable();
3531                 r = 1;
3532                 goto out;
3533         }
3534
3535         if (vcpu->arch.exception.pending)
3536                 __queue_exception(vcpu);
3537         else
3538                 inject_pending_irq(vcpu, kvm_run);
3539
3540         /* enable NMI/IRQ window open exits if needed */
3541         if (vcpu->arch.nmi_pending)
3542                 kvm_x86_ops->enable_nmi_window(vcpu);
3543         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3544                 kvm_x86_ops->enable_irq_window(vcpu);
3545
3546         if (kvm_lapic_enabled(vcpu)) {
3547                 update_cr8_intercept(vcpu);
3548                 kvm_lapic_sync_to_vapic(vcpu);
3549         }
3550
3551         up_read(&vcpu->kvm->slots_lock);
3552
3553         kvm_guest_enter();
3554
3555         get_debugreg(vcpu->arch.host_dr6, 6);
3556         get_debugreg(vcpu->arch.host_dr7, 7);
3557         if (unlikely(vcpu->arch.switch_db_regs)) {
3558                 get_debugreg(vcpu->arch.host_db[0], 0);
3559                 get_debugreg(vcpu->arch.host_db[1], 1);
3560                 get_debugreg(vcpu->arch.host_db[2], 2);
3561                 get_debugreg(vcpu->arch.host_db[3], 3);
3562
3563                 set_debugreg(0, 7);
3564                 set_debugreg(vcpu->arch.eff_db[0], 0);
3565                 set_debugreg(vcpu->arch.eff_db[1], 1);
3566                 set_debugreg(vcpu->arch.eff_db[2], 2);
3567                 set_debugreg(vcpu->arch.eff_db[3], 3);
3568         }
3569
3570         trace_kvm_entry(vcpu->vcpu_id);
3571         kvm_x86_ops->run(vcpu, kvm_run);
3572
3573         if (unlikely(vcpu->arch.switch_db_regs)) {
3574                 set_debugreg(0, 7);
3575                 set_debugreg(vcpu->arch.host_db[0], 0);
3576                 set_debugreg(vcpu->arch.host_db[1], 1);
3577                 set_debugreg(vcpu->arch.host_db[2], 2);
3578                 set_debugreg(vcpu->arch.host_db[3], 3);
3579         }
3580         set_debugreg(vcpu->arch.host_dr6, 6);
3581         set_debugreg(vcpu->arch.host_dr7, 7);
3582
3583         set_bit(KVM_REQ_KICK, &vcpu->requests);
3584         local_irq_enable();
3585
3586         ++vcpu->stat.exits;
3587
3588         /*
3589          * We must have an instruction between local_irq_enable() and
3590          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3591          * the interrupt shadow.  The stat.exits increment will do nicely.
3592          * But we need to prevent reordering, hence this barrier():
3593          */
3594         barrier();
3595
3596         kvm_guest_exit();
3597
3598         preempt_enable();
3599
3600         down_read(&vcpu->kvm->slots_lock);
3601
3602         /*
3603          * Profile KVM exit RIPs:
3604          */
3605         if (unlikely(prof_on == KVM_PROFILING)) {
3606                 unsigned long rip = kvm_rip_read(vcpu);
3607                 profile_hit(KVM_PROFILING, (void *)rip);
3608         }
3609
3610
3611         kvm_lapic_sync_from_vapic(vcpu);
3612
3613         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3614 out:
3615         return r;
3616 }
3617
3618
3619 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3620 {
3621         int r;
3622
3623         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3624                 pr_debug("vcpu %d received sipi with vector # %x\n",
3625                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3626                 kvm_lapic_reset(vcpu);
3627                 r = kvm_arch_vcpu_reset(vcpu);
3628                 if (r)
3629                         return r;
3630                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3631         }
3632
3633         down_read(&vcpu->kvm->slots_lock);
3634         vapic_enter(vcpu);
3635
3636         r = 1;
3637         while (r > 0) {
3638                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3639                         r = vcpu_enter_guest(vcpu, kvm_run);
3640                 else {
3641                         up_read(&vcpu->kvm->slots_lock);
3642                         kvm_vcpu_block(vcpu);
3643                         down_read(&vcpu->kvm->slots_lock);
3644                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3645                         {
3646                                 switch(vcpu->arch.mp_state) {
3647                                 case KVM_MP_STATE_HALTED:
3648                                         vcpu->arch.mp_state =
3649                                                 KVM_MP_STATE_RUNNABLE;
3650                                 case KVM_MP_STATE_RUNNABLE:
3651                                         break;
3652                                 case KVM_MP_STATE_SIPI_RECEIVED:
3653                                 default:
3654                                         r = -EINTR;
3655                                         break;
3656                                 }
3657                         }
3658                 }
3659
3660                 if (r <= 0)
3661                         break;
3662
3663                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3664                 if (kvm_cpu_has_pending_timer(vcpu))
3665                         kvm_inject_pending_timer_irqs(vcpu);
3666
3667                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3668                         r = -EINTR;
3669                         kvm_run->exit_reason = KVM_EXIT_INTR;
3670                         ++vcpu->stat.request_irq_exits;
3671                 }
3672                 if (signal_pending(current)) {
3673                         r = -EINTR;
3674                         kvm_run->exit_reason = KVM_EXIT_INTR;
3675                         ++vcpu->stat.signal_exits;
3676                 }
3677                 if (need_resched()) {
3678                         up_read(&vcpu->kvm->slots_lock);
3679                         kvm_resched(vcpu);
3680                         down_read(&vcpu->kvm->slots_lock);
3681                 }
3682         }
3683
3684         up_read(&vcpu->kvm->slots_lock);
3685         post_kvm_run_save(vcpu, kvm_run);
3686
3687         vapic_exit(vcpu);
3688
3689         return r;
3690 }
3691
3692 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3693 {
3694         int r;
3695         sigset_t sigsaved;
3696
3697         vcpu_load(vcpu);
3698
3699         if (vcpu->sigset_active)
3700                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3701
3702         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3703                 kvm_vcpu_block(vcpu);
3704                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3705                 r = -EAGAIN;
3706                 goto out;
3707         }
3708
3709         /* re-sync apic's tpr */
3710         if (!irqchip_in_kernel(vcpu->kvm))
3711                 kvm_set_cr8(vcpu, kvm_run->cr8);
3712
3713         if (vcpu->arch.pio.cur_count) {
3714                 r = complete_pio(vcpu);
3715                 if (r)
3716                         goto out;
3717         }
3718 #if CONFIG_HAS_IOMEM
3719         if (vcpu->mmio_needed) {
3720                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3721                 vcpu->mmio_read_completed = 1;
3722                 vcpu->mmio_needed = 0;
3723
3724                 down_read(&vcpu->kvm->slots_lock);
3725                 r = emulate_instruction(vcpu, kvm_run,
3726                                         vcpu->arch.mmio_fault_cr2, 0,
3727                                         EMULTYPE_NO_DECODE);
3728                 up_read(&vcpu->kvm->slots_lock);
3729                 if (r == EMULATE_DO_MMIO) {
3730                         /*
3731                          * Read-modify-write.  Back to userspace.
3732                          */
3733                         r = 0;
3734                         goto out;
3735                 }
3736         }
3737 #endif
3738         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3739                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3740                                      kvm_run->hypercall.ret);
3741
3742         r = __vcpu_run(vcpu, kvm_run);
3743
3744 out:
3745         if (vcpu->sigset_active)
3746                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3747
3748         vcpu_put(vcpu);
3749         return r;
3750 }
3751
3752 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3753 {
3754         vcpu_load(vcpu);
3755
3756         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3757         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3758         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3759         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3760         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3761         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3762         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3763         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3764 #ifdef CONFIG_X86_64
3765         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3766         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3767         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3768         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3769         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3770         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3771         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3772         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3773 #endif
3774
3775         regs->rip = kvm_rip_read(vcpu);
3776         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3777
3778         /*
3779          * Don't leak debug flags in case they were set for guest debugging
3780          */
3781         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3782                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3783
3784         vcpu_put(vcpu);
3785
3786         return 0;
3787 }
3788
3789 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3790 {
3791         vcpu_load(vcpu);
3792
3793         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3794         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3795         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3796         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3797         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3798         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3799         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3800         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3801 #ifdef CONFIG_X86_64
3802         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3803         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3804         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3805         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3806         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3807         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3808         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3809         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3810
3811 #endif
3812
3813         kvm_rip_write(vcpu, regs->rip);
3814         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3815
3816
3817         vcpu->arch.exception.pending = false;
3818
3819         vcpu_put(vcpu);
3820
3821         return 0;
3822 }
3823
3824 void kvm_get_segment(struct kvm_vcpu *vcpu,
3825                      struct kvm_segment *var, int seg)
3826 {
3827         kvm_x86_ops->get_segment(vcpu, var, seg);
3828 }
3829
3830 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3831 {
3832         struct kvm_segment cs;
3833
3834         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3835         *db = cs.db;
3836         *l = cs.l;
3837 }
3838 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3839
3840 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3841                                   struct kvm_sregs *sregs)
3842 {
3843         struct descriptor_table dt;
3844
3845         vcpu_load(vcpu);
3846
3847         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3848         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3849         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3850         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3851         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3852         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3853
3854         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3855         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3856
3857         kvm_x86_ops->get_idt(vcpu, &dt);
3858         sregs->idt.limit = dt.limit;
3859         sregs->idt.base = dt.base;
3860         kvm_x86_ops->get_gdt(vcpu, &dt);
3861         sregs->gdt.limit = dt.limit;
3862         sregs->gdt.base = dt.base;
3863
3864         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3865         sregs->cr0 = vcpu->arch.cr0;
3866         sregs->cr2 = vcpu->arch.cr2;
3867         sregs->cr3 = vcpu->arch.cr3;
3868         sregs->cr4 = vcpu->arch.cr4;
3869         sregs->cr8 = kvm_get_cr8(vcpu);
3870         sregs->efer = vcpu->arch.shadow_efer;
3871         sregs->apic_base = kvm_get_apic_base(vcpu);
3872
3873         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3874
3875         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3876                 set_bit(vcpu->arch.interrupt.nr,
3877                         (unsigned long *)sregs->interrupt_bitmap);
3878
3879         vcpu_put(vcpu);
3880
3881         return 0;
3882 }
3883
3884 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3885                                     struct kvm_mp_state *mp_state)
3886 {
3887         vcpu_load(vcpu);
3888         mp_state->mp_state = vcpu->arch.mp_state;
3889         vcpu_put(vcpu);
3890         return 0;
3891 }
3892
3893 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3894                                     struct kvm_mp_state *mp_state)
3895 {
3896         vcpu_load(vcpu);
3897         vcpu->arch.mp_state = mp_state->mp_state;
3898         vcpu_put(vcpu);
3899         return 0;
3900 }
3901
3902 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3903                         struct kvm_segment *var, int seg)
3904 {
3905         kvm_x86_ops->set_segment(vcpu, var, seg);
3906 }
3907
3908 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3909                                    struct kvm_segment *kvm_desct)
3910 {
3911         kvm_desct->base = seg_desc->base0;
3912         kvm_desct->base |= seg_desc->base1 << 16;
3913         kvm_desct->base |= seg_desc->base2 << 24;
3914         kvm_desct->limit = seg_desc->limit0;
3915         kvm_desct->limit |= seg_desc->limit << 16;
3916         if (seg_desc->g) {
3917                 kvm_desct->limit <<= 12;
3918                 kvm_desct->limit |= 0xfff;
3919         }
3920         kvm_desct->selector = selector;
3921         kvm_desct->type = seg_desc->type;
3922         kvm_desct->present = seg_desc->p;
3923         kvm_desct->dpl = seg_desc->dpl;
3924         kvm_desct->db = seg_desc->d;
3925         kvm_desct->s = seg_desc->s;
3926         kvm_desct->l = seg_desc->l;
3927         kvm_desct->g = seg_desc->g;
3928         kvm_desct->avl = seg_desc->avl;
3929         if (!selector)
3930                 kvm_desct->unusable = 1;
3931         else
3932                 kvm_desct->unusable = 0;
3933         kvm_desct->padding = 0;
3934 }
3935
3936 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3937                                           u16 selector,
3938                                           struct descriptor_table *dtable)
3939 {
3940         if (selector & 1 << 2) {
3941                 struct kvm_segment kvm_seg;
3942
3943                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3944
3945                 if (kvm_seg.unusable)
3946                         dtable->limit = 0;
3947                 else
3948                         dtable->limit = kvm_seg.limit;
3949                 dtable->base = kvm_seg.base;
3950         }
3951         else
3952                 kvm_x86_ops->get_gdt(vcpu, dtable);
3953 }
3954
3955 /* allowed just for 8 bytes segments */
3956 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3957                                          struct desc_struct *seg_desc)
3958 {
3959         gpa_t gpa;
3960         struct descriptor_table dtable;
3961         u16 index = selector >> 3;
3962
3963         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3964
3965         if (dtable.limit < index * 8 + 7) {
3966                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3967                 return 1;
3968         }
3969         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3970         gpa += index * 8;
3971         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3972 }
3973
3974 /* allowed just for 8 bytes segments */
3975 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3976                                          struct desc_struct *seg_desc)
3977 {
3978         gpa_t gpa;
3979         struct descriptor_table dtable;
3980         u16 index = selector >> 3;
3981
3982         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3983
3984         if (dtable.limit < index * 8 + 7)
3985                 return 1;
3986         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3987         gpa += index * 8;
3988         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3989 }
3990
3991 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3992                              struct desc_struct *seg_desc)
3993 {
3994         u32 base_addr;
3995
3996         base_addr = seg_desc->base0;
3997         base_addr |= (seg_desc->base1 << 16);
3998         base_addr |= (seg_desc->base2 << 24);
3999
4000         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4001 }
4002
4003 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4004 {
4005         struct kvm_segment kvm_seg;
4006
4007         kvm_get_segment(vcpu, &kvm_seg, seg);
4008         return kvm_seg.selector;
4009 }
4010
4011 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4012                                                 u16 selector,
4013                                                 struct kvm_segment *kvm_seg)
4014 {
4015         struct desc_struct seg_desc;
4016
4017         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4018                 return 1;
4019         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4020         return 0;
4021 }
4022
4023 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4024 {
4025         struct kvm_segment segvar = {
4026                 .base = selector << 4,
4027                 .limit = 0xffff,
4028                 .selector = selector,
4029                 .type = 3,
4030                 .present = 1,
4031                 .dpl = 3,
4032                 .db = 0,
4033                 .s = 1,
4034                 .l = 0,
4035                 .g = 0,
4036                 .avl = 0,
4037                 .unusable = 0,
4038         };
4039         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4040         return 0;
4041 }
4042
4043 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4044                                 int type_bits, int seg)
4045 {
4046         struct kvm_segment kvm_seg;
4047
4048         if (!(vcpu->arch.cr0 & X86_CR0_PE))
4049                 return kvm_load_realmode_segment(vcpu, selector, seg);
4050         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4051                 return 1;
4052         kvm_seg.type |= type_bits;
4053
4054         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4055             seg != VCPU_SREG_LDTR)
4056                 if (!kvm_seg.s)
4057                         kvm_seg.unusable = 1;
4058
4059         kvm_set_segment(vcpu, &kvm_seg, seg);
4060         return 0;
4061 }
4062
4063 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4064                                 struct tss_segment_32 *tss)
4065 {
4066         tss->cr3 = vcpu->arch.cr3;
4067         tss->eip = kvm_rip_read(vcpu);
4068         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
4069         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4070         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4071         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4072         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4073         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4074         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4075         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4076         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4077         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4078         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4079         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4080         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4081         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4082         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4083         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4084 }
4085
4086 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4087                                   struct tss_segment_32 *tss)
4088 {
4089         kvm_set_cr3(vcpu, tss->cr3);
4090
4091         kvm_rip_write(vcpu, tss->eip);
4092         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
4093
4094         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4095         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4096         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4097         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4098         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4099         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4100         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4101         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4102
4103         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4104                 return 1;
4105
4106         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4107                 return 1;
4108
4109         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4110                 return 1;
4111
4112         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4113                 return 1;
4114
4115         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4116                 return 1;
4117
4118         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4119                 return 1;
4120
4121         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4122                 return 1;
4123         return 0;
4124 }
4125
4126 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4127                                 struct tss_segment_16 *tss)
4128 {
4129         tss->ip = kvm_rip_read(vcpu);
4130         tss->flag = kvm_x86_ops->get_rflags(vcpu);
4131         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4132         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4133         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4134         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4135         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4136         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4137         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4138         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4139
4140         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4141         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4142         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4143         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4144         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4145         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
4146 }
4147
4148 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4149                                  struct tss_segment_16 *tss)
4150 {
4151         kvm_rip_write(vcpu, tss->ip);
4152         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
4153         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4154         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4155         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4156         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4157         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4158         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4159         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4160         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4161
4162         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4163                 return 1;
4164
4165         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4166                 return 1;
4167
4168         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4169                 return 1;
4170
4171         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4172                 return 1;
4173
4174         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4175                 return 1;
4176         return 0;
4177 }
4178
4179 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4180                               u16 old_tss_sel, u32 old_tss_base,
4181                               struct desc_struct *nseg_desc)
4182 {
4183         struct tss_segment_16 tss_segment_16;
4184         int ret = 0;
4185
4186         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4187                            sizeof tss_segment_16))
4188                 goto out;
4189
4190         save_state_to_tss16(vcpu, &tss_segment_16);
4191
4192         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4193                             sizeof tss_segment_16))
4194                 goto out;
4195
4196         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4197                            &tss_segment_16, sizeof tss_segment_16))
4198                 goto out;
4199
4200         if (old_tss_sel != 0xffff) {
4201                 tss_segment_16.prev_task_link = old_tss_sel;
4202
4203                 if (kvm_write_guest(vcpu->kvm,
4204                                     get_tss_base_addr(vcpu, nseg_desc),
4205                                     &tss_segment_16.prev_task_link,
4206                                     sizeof tss_segment_16.prev_task_link))
4207                         goto out;
4208         }
4209
4210         if (load_state_from_tss16(vcpu, &tss_segment_16))
4211                 goto out;
4212
4213         ret = 1;
4214 out:
4215         return ret;
4216 }
4217
4218 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4219                        u16 old_tss_sel, u32 old_tss_base,
4220                        struct desc_struct *nseg_desc)
4221 {
4222         struct tss_segment_32 tss_segment_32;
4223         int ret = 0;
4224
4225         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4226                            sizeof tss_segment_32))
4227                 goto out;
4228
4229         save_state_to_tss32(vcpu, &tss_segment_32);
4230
4231         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4232                             sizeof tss_segment_32))
4233                 goto out;
4234
4235         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4236                            &tss_segment_32, sizeof tss_segment_32))
4237                 goto out;
4238
4239         if (old_tss_sel != 0xffff) {
4240                 tss_segment_32.prev_task_link = old_tss_sel;
4241
4242                 if (kvm_write_guest(vcpu->kvm,
4243                                     get_tss_base_addr(vcpu, nseg_desc),
4244                                     &tss_segment_32.prev_task_link,
4245                                     sizeof tss_segment_32.prev_task_link))
4246                         goto out;
4247         }
4248
4249         if (load_state_from_tss32(vcpu, &tss_segment_32))
4250                 goto out;
4251
4252         ret = 1;
4253 out:
4254         return ret;
4255 }
4256
4257 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4258 {
4259         struct kvm_segment tr_seg;
4260         struct desc_struct cseg_desc;
4261         struct desc_struct nseg_desc;
4262         int ret = 0;
4263         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4264         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4265
4266         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4267
4268         /* FIXME: Handle errors. Failure to read either TSS or their
4269          * descriptors should generate a pagefault.
4270          */
4271         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4272                 goto out;
4273
4274         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4275                 goto out;
4276
4277         if (reason != TASK_SWITCH_IRET) {
4278                 int cpl;
4279
4280                 cpl = kvm_x86_ops->get_cpl(vcpu);
4281                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4282                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4283                         return 1;
4284                 }
4285         }
4286
4287         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
4288                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4289                 return 1;
4290         }
4291
4292         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4293                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4294                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4295         }
4296
4297         if (reason == TASK_SWITCH_IRET) {
4298                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4299                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4300         }
4301
4302         /* set back link to prev task only if NT bit is set in eflags
4303            note that old_tss_sel is not used afetr this point */
4304         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4305                 old_tss_sel = 0xffff;
4306
4307         /* set back link to prev task only if NT bit is set in eflags
4308            note that old_tss_sel is not used afetr this point */
4309         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4310                 old_tss_sel = 0xffff;
4311
4312         if (nseg_desc.type & 8)
4313                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4314                                          old_tss_base, &nseg_desc);
4315         else
4316                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4317                                          old_tss_base, &nseg_desc);
4318
4319         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4320                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4321                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4322         }
4323
4324         if (reason != TASK_SWITCH_IRET) {
4325                 nseg_desc.type |= (1 << 1);
4326                 save_guest_segment_descriptor(vcpu, tss_selector,
4327                                               &nseg_desc);
4328         }
4329
4330         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4331         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4332         tr_seg.type = 11;
4333         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4334 out:
4335         return ret;
4336 }
4337 EXPORT_SYMBOL_GPL(kvm_task_switch);
4338
4339 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4340                                   struct kvm_sregs *sregs)
4341 {
4342         int mmu_reset_needed = 0;
4343         int pending_vec, max_bits;
4344         struct descriptor_table dt;
4345
4346         vcpu_load(vcpu);
4347
4348         dt.limit = sregs->idt.limit;
4349         dt.base = sregs->idt.base;
4350         kvm_x86_ops->set_idt(vcpu, &dt);
4351         dt.limit = sregs->gdt.limit;
4352         dt.base = sregs->gdt.base;
4353         kvm_x86_ops->set_gdt(vcpu, &dt);
4354
4355         vcpu->arch.cr2 = sregs->cr2;
4356         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4357
4358         down_read(&vcpu->kvm->slots_lock);
4359         if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
4360                 vcpu->arch.cr3 = sregs->cr3;
4361         else
4362                 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
4363         up_read(&vcpu->kvm->slots_lock);
4364
4365         kvm_set_cr8(vcpu, sregs->cr8);
4366
4367         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4368         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4369         kvm_set_apic_base(vcpu, sregs->apic_base);
4370
4371         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4372
4373         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4374         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4375         vcpu->arch.cr0 = sregs->cr0;
4376
4377         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4378         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4379         if (!is_long_mode(vcpu) && is_pae(vcpu))
4380                 load_pdptrs(vcpu, vcpu->arch.cr3);
4381
4382         if (mmu_reset_needed)
4383                 kvm_mmu_reset_context(vcpu);
4384
4385         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4386         pending_vec = find_first_bit(
4387                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4388         if (pending_vec < max_bits) {
4389                 kvm_queue_interrupt(vcpu, pending_vec, false);
4390                 pr_debug("Set back pending irq %d\n", pending_vec);
4391                 if (irqchip_in_kernel(vcpu->kvm))
4392                         kvm_pic_clear_isr_ack(vcpu->kvm);
4393         }
4394
4395         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4396         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4397         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4398         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4399         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4400         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4401
4402         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4403         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4404
4405         /* Older userspace won't unhalt the vcpu on reset. */
4406         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4407             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4408             !(vcpu->arch.cr0 & X86_CR0_PE))
4409                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4410
4411         vcpu_put(vcpu);
4412
4413         return 0;
4414 }
4415
4416 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4417                                         struct kvm_guest_debug *dbg)
4418 {
4419         int i, r;
4420
4421         vcpu_load(vcpu);
4422
4423         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4424             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4425                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4426                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4427                 vcpu->arch.switch_db_regs =
4428                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4429         } else {
4430                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4431                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4432                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4433         }
4434
4435         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4436
4437         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4438                 kvm_queue_exception(vcpu, DB_VECTOR);
4439         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4440                 kvm_queue_exception(vcpu, BP_VECTOR);
4441
4442         vcpu_put(vcpu);
4443
4444         return r;
4445 }
4446
4447 /*
4448  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4449  * we have asm/x86/processor.h
4450  */
4451 struct fxsave {
4452         u16     cwd;
4453         u16     swd;
4454         u16     twd;
4455         u16     fop;
4456         u64     rip;
4457         u64     rdp;
4458         u32     mxcsr;
4459         u32     mxcsr_mask;
4460         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4461 #ifdef CONFIG_X86_64
4462         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4463 #else
4464         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4465 #endif
4466 };
4467
4468 /*
4469  * Translate a guest virtual address to a guest physical address.
4470  */
4471 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4472                                     struct kvm_translation *tr)
4473 {
4474         unsigned long vaddr = tr->linear_address;
4475         gpa_t gpa;
4476
4477         vcpu_load(vcpu);
4478         down_read(&vcpu->kvm->slots_lock);
4479         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4480         up_read(&vcpu->kvm->slots_lock);
4481         tr->physical_address = gpa;
4482         tr->valid = gpa != UNMAPPED_GVA;
4483         tr->writeable = 1;
4484         tr->usermode = 0;
4485         vcpu_put(vcpu);
4486
4487         return 0;
4488 }
4489
4490 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4491 {
4492         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4493
4494         vcpu_load(vcpu);
4495
4496         memcpy(fpu->fpr, fxsave->st_space, 128);
4497         fpu->fcw = fxsave->cwd;
4498         fpu->fsw = fxsave->swd;
4499         fpu->ftwx = fxsave->twd;
4500         fpu->last_opcode = fxsave->fop;
4501         fpu->last_ip = fxsave->rip;
4502         fpu->last_dp = fxsave->rdp;
4503         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4504
4505         vcpu_put(vcpu);
4506
4507         return 0;
4508 }
4509
4510 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4511 {
4512         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4513
4514         vcpu_load(vcpu);
4515
4516         memcpy(fxsave->st_space, fpu->fpr, 128);
4517         fxsave->cwd = fpu->fcw;
4518         fxsave->swd = fpu->fsw;
4519         fxsave->twd = fpu->ftwx;
4520         fxsave->fop = fpu->last_opcode;
4521         fxsave->rip = fpu->last_ip;
4522         fxsave->rdp = fpu->last_dp;
4523         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4524
4525         vcpu_put(vcpu);
4526
4527         return 0;
4528 }
4529
4530 void fx_init(struct kvm_vcpu *vcpu)
4531 {
4532         unsigned after_mxcsr_mask;
4533
4534         /*
4535          * Touch the fpu the first time in non atomic context as if
4536          * this is the first fpu instruction the exception handler
4537          * will fire before the instruction returns and it'll have to
4538          * allocate ram with GFP_KERNEL.
4539          */
4540         if (!used_math())
4541                 kvm_fx_save(&vcpu->arch.host_fx_image);
4542
4543         /* Initialize guest FPU by resetting ours and saving into guest's */
4544         preempt_disable();
4545         kvm_fx_save(&vcpu->arch.host_fx_image);
4546         kvm_fx_finit();
4547         kvm_fx_save(&vcpu->arch.guest_fx_image);
4548         kvm_fx_restore(&vcpu->arch.host_fx_image);
4549         preempt_enable();
4550
4551         vcpu->arch.cr0 |= X86_CR0_ET;
4552         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4553         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4554         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4555                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4556 }
4557 EXPORT_SYMBOL_GPL(fx_init);
4558
4559 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4560 {
4561         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4562                 return;
4563
4564         vcpu->guest_fpu_loaded = 1;
4565         kvm_fx_save(&vcpu->arch.host_fx_image);
4566         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4567 }
4568 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4569
4570 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4571 {
4572         if (!vcpu->guest_fpu_loaded)
4573                 return;
4574
4575         vcpu->guest_fpu_loaded = 0;
4576         kvm_fx_save(&vcpu->arch.guest_fx_image);
4577         kvm_fx_restore(&vcpu->arch.host_fx_image);
4578         ++vcpu->stat.fpu_reload;
4579 }
4580 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4581
4582 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4583 {
4584         if (vcpu->arch.time_page) {
4585                 kvm_release_page_dirty(vcpu->arch.time_page);
4586                 vcpu->arch.time_page = NULL;
4587         }
4588
4589         kvm_x86_ops->vcpu_free(vcpu);
4590 }
4591
4592 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4593                                                 unsigned int id)
4594 {
4595         return kvm_x86_ops->vcpu_create(kvm, id);
4596 }
4597
4598 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4599 {
4600         int r;
4601
4602         /* We do fxsave: this must be aligned. */
4603         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4604
4605         vcpu->arch.mtrr_state.have_fixed = 1;
4606         vcpu_load(vcpu);
4607         r = kvm_arch_vcpu_reset(vcpu);
4608         if (r == 0)
4609                 r = kvm_mmu_setup(vcpu);
4610         vcpu_put(vcpu);
4611         if (r < 0)
4612                 goto free_vcpu;
4613
4614         return 0;
4615 free_vcpu:
4616         kvm_x86_ops->vcpu_free(vcpu);
4617         return r;
4618 }
4619
4620 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4621 {
4622         vcpu_load(vcpu);
4623         kvm_mmu_unload(vcpu);
4624         vcpu_put(vcpu);
4625
4626         kvm_x86_ops->vcpu_free(vcpu);
4627 }
4628
4629 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4630 {
4631         vcpu->arch.nmi_pending = false;
4632         vcpu->arch.nmi_injected = false;
4633
4634         vcpu->arch.switch_db_regs = 0;
4635         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4636         vcpu->arch.dr6 = DR6_FIXED_1;
4637         vcpu->arch.dr7 = DR7_FIXED_1;
4638
4639         return kvm_x86_ops->vcpu_reset(vcpu);
4640 }
4641
4642 void kvm_arch_hardware_enable(void *garbage)
4643 {
4644         kvm_x86_ops->hardware_enable(garbage);
4645 }
4646
4647 void kvm_arch_hardware_disable(void *garbage)
4648 {
4649         kvm_x86_ops->hardware_disable(garbage);
4650 }
4651
4652 int kvm_arch_hardware_setup(void)
4653 {
4654         return kvm_x86_ops->hardware_setup();
4655 }
4656
4657 void kvm_arch_hardware_unsetup(void)
4658 {
4659         kvm_x86_ops->hardware_unsetup();
4660 }
4661
4662 void kvm_arch_check_processor_compat(void *rtn)
4663 {
4664         kvm_x86_ops->check_processor_compatibility(rtn);
4665 }
4666
4667 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4668 {
4669         struct page *page;
4670         struct kvm *kvm;
4671         int r;
4672
4673         BUG_ON(vcpu->kvm == NULL);
4674         kvm = vcpu->kvm;
4675
4676         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4677         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
4678                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4679         else
4680                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4681
4682         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4683         if (!page) {
4684                 r = -ENOMEM;
4685                 goto fail;
4686         }
4687         vcpu->arch.pio_data = page_address(page);
4688
4689         r = kvm_mmu_create(vcpu);
4690         if (r < 0)
4691                 goto fail_free_pio_data;
4692
4693         if (irqchip_in_kernel(kvm)) {
4694                 r = kvm_create_lapic(vcpu);
4695                 if (r < 0)
4696                         goto fail_mmu_destroy;
4697         }
4698
4699         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
4700                                        GFP_KERNEL);
4701         if (!vcpu->arch.mce_banks) {
4702                 r = -ENOMEM;
4703                 goto fail_mmu_destroy;
4704         }
4705         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
4706
4707         return 0;
4708
4709 fail_mmu_destroy:
4710         kvm_mmu_destroy(vcpu);
4711 fail_free_pio_data:
4712         free_page((unsigned long)vcpu->arch.pio_data);
4713 fail:
4714         return r;
4715 }
4716
4717 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4718 {
4719         kvm_free_lapic(vcpu);
4720         down_read(&vcpu->kvm->slots_lock);
4721         kvm_mmu_destroy(vcpu);
4722         up_read(&vcpu->kvm->slots_lock);
4723         free_page((unsigned long)vcpu->arch.pio_data);
4724 }
4725
4726 struct  kvm *kvm_arch_create_vm(void)
4727 {
4728         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4729
4730         if (!kvm)
4731                 return ERR_PTR(-ENOMEM);
4732
4733         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4734         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4735
4736         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4737         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4738
4739         rdtscll(kvm->arch.vm_init_tsc);
4740
4741         return kvm;
4742 }
4743
4744 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4745 {
4746         vcpu_load(vcpu);
4747         kvm_mmu_unload(vcpu);
4748         vcpu_put(vcpu);
4749 }
4750
4751 static void kvm_free_vcpus(struct kvm *kvm)
4752 {
4753         unsigned int i;
4754         struct kvm_vcpu *vcpu;
4755
4756         /*
4757          * Unpin any mmu pages first.
4758          */
4759         kvm_for_each_vcpu(i, vcpu, kvm)
4760                 kvm_unload_vcpu_mmu(vcpu);
4761         kvm_for_each_vcpu(i, vcpu, kvm)
4762                 kvm_arch_vcpu_free(vcpu);
4763
4764         mutex_lock(&kvm->lock);
4765         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
4766                 kvm->vcpus[i] = NULL;
4767
4768         atomic_set(&kvm->online_vcpus, 0);
4769         mutex_unlock(&kvm->lock);
4770 }
4771
4772 void kvm_arch_sync_events(struct kvm *kvm)
4773 {
4774         kvm_free_all_assigned_devices(kvm);
4775 }
4776
4777 void kvm_arch_destroy_vm(struct kvm *kvm)
4778 {
4779         kvm_iommu_unmap_guest(kvm);
4780         kvm_free_pit(kvm);
4781         kfree(kvm->arch.vpic);
4782         kfree(kvm->arch.vioapic);
4783         kvm_free_vcpus(kvm);
4784         kvm_free_physmem(kvm);
4785         if (kvm->arch.apic_access_page)
4786                 put_page(kvm->arch.apic_access_page);
4787         if (kvm->arch.ept_identity_pagetable)
4788                 put_page(kvm->arch.ept_identity_pagetable);
4789         kfree(kvm);
4790 }
4791
4792 int kvm_arch_set_memory_region(struct kvm *kvm,
4793                                 struct kvm_userspace_memory_region *mem,
4794                                 struct kvm_memory_slot old,
4795                                 int user_alloc)
4796 {
4797         int npages = mem->memory_size >> PAGE_SHIFT;
4798         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4799
4800         /*To keep backward compatibility with older userspace,
4801          *x86 needs to hanlde !user_alloc case.
4802          */
4803         if (!user_alloc) {
4804                 if (npages && !old.rmap) {
4805                         unsigned long userspace_addr;
4806
4807                         down_write(&current->mm->mmap_sem);
4808                         userspace_addr = do_mmap(NULL, 0,
4809                                                  npages * PAGE_SIZE,
4810                                                  PROT_READ | PROT_WRITE,
4811                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4812                                                  0);
4813                         up_write(&current->mm->mmap_sem);
4814
4815                         if (IS_ERR((void *)userspace_addr))
4816                                 return PTR_ERR((void *)userspace_addr);
4817
4818                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4819                         spin_lock(&kvm->mmu_lock);
4820                         memslot->userspace_addr = userspace_addr;
4821                         spin_unlock(&kvm->mmu_lock);
4822                 } else {
4823                         if (!old.user_alloc && old.rmap) {
4824                                 int ret;
4825
4826                                 down_write(&current->mm->mmap_sem);
4827                                 ret = do_munmap(current->mm, old.userspace_addr,
4828                                                 old.npages * PAGE_SIZE);
4829                                 up_write(&current->mm->mmap_sem);
4830                                 if (ret < 0)
4831                                         printk(KERN_WARNING
4832                                        "kvm_vm_ioctl_set_memory_region: "
4833                                        "failed to munmap memory\n");
4834                         }
4835                 }
4836         }
4837
4838         spin_lock(&kvm->mmu_lock);
4839         if (!kvm->arch.n_requested_mmu_pages) {
4840                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4841                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4842         }
4843
4844         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4845         spin_unlock(&kvm->mmu_lock);
4846         kvm_flush_remote_tlbs(kvm);
4847
4848         return 0;
4849 }
4850
4851 void kvm_arch_flush_shadow(struct kvm *kvm)
4852 {
4853         kvm_mmu_zap_all(kvm);
4854         kvm_reload_remote_mmus(kvm);
4855 }
4856
4857 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4858 {
4859         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4860                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4861                || vcpu->arch.nmi_pending;
4862 }
4863
4864 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4865 {
4866         int me;
4867         int cpu = vcpu->cpu;
4868
4869         if (waitqueue_active(&vcpu->wq)) {
4870                 wake_up_interruptible(&vcpu->wq);
4871                 ++vcpu->stat.halt_wakeup;
4872         }
4873
4874         me = get_cpu();
4875         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4876                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4877                         smp_send_reschedule(cpu);
4878         put_cpu();
4879 }
4880
4881 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4882 {
4883         return kvm_x86_ops->interrupt_allowed(vcpu);
4884 }
4885
4886 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
4887 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
4888 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
4889 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
4890 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);