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
  40 #include <asm/uaccess.h>
  41 #include <asm/msr.h>
  42 #include <asm/desc.h>
  43 #include <asm/mtrr.h>
  44
  45 #define MAX_IO_MSRS 256
  46 #define CR0_RESERVED_BITS                                               \
  47         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  48                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  49                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  50 #define CR4_RESERVED_BITS                                               \
  51         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  52                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  53                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  54                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  55
  56 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  57 /* EFER defaults:
  58  * - enable syscall per default because its emulated by KVM
  59  * - enable LME and LMA per default on 64 bit KVM
  60  */
  61 #ifdef CONFIG_X86_64
  62 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  63 #else
  64 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  65 #endif
  66
  67 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  68 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  69
  70 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  71                                     struct kvm_cpuid_entry2 __user *entries);
  72 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
  73                                               u32 function, u32 index);
  74
  75 struct kvm_x86_ops *kvm_x86_ops;
  76 EXPORT_SYMBOL_GPL(kvm_x86_ops);
  77
  78 struct kvm_stats_debugfs_item debugfs_entries[] = {
  79         { "pf_fixed", VCPU_STAT(pf_fixed) },
  80         { "pf_guest", VCPU_STAT(pf_guest) },
  81         { "tlb_flush", VCPU_STAT(tlb_flush) },
  82         { "invlpg", VCPU_STAT(invlpg) },
  83         { "exits", VCPU_STAT(exits) },
  84         { "io_exits", VCPU_STAT(io_exits) },
  85         { "mmio_exits", VCPU_STAT(mmio_exits) },
  86         { "signal_exits", VCPU_STAT(signal_exits) },
  87         { "irq_window", VCPU_STAT(irq_window_exits) },
  88         { "nmi_window", VCPU_STAT(nmi_window_exits) },
  89         { "halt_exits", VCPU_STAT(halt_exits) },
  90         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  91         { "hypercalls", VCPU_STAT(hypercalls) },
  92         { "request_irq", VCPU_STAT(request_irq_exits) },
  93         { "request_nmi", VCPU_STAT(request_nmi_exits) },
  94         { "irq_exits", VCPU_STAT(irq_exits) },
  95         { "host_state_reload", VCPU_STAT(host_state_reload) },
  96         { "efer_reload", VCPU_STAT(efer_reload) },
  97         { "fpu_reload", VCPU_STAT(fpu_reload) },
  98         { "insn_emulation", VCPU_STAT(insn_emulation) },
  99         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
 100         { "irq_injections", VCPU_STAT(irq_injections) },
 101         { "nmi_injections", VCPU_STAT(nmi_injections) },
 102         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
 103         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
 104         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
 105         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
 106         { "mmu_flooded", VM_STAT(mmu_flooded) },
 107         { "mmu_recycled", VM_STAT(mmu_recycled) },
 108         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
 109         { "mmu_unsync", VM_STAT(mmu_unsync) },
 110         { "mmu_unsync_global", VM_STAT(mmu_unsync_global) },
 111         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
 112         { "largepages", VM_STAT(lpages) },
 113         { NULL }
 114 };
 115
 116 unsigned long segment_base(u16 selector)
 117 {
 118         struct descriptor_table gdt;
 119         struct desc_struct *d;
 120         unsigned long table_base;
 121         unsigned long v;
 122
 123         if (selector == 0)
 124                 return 0;
 125
 126         asm("sgdt %0" : "=m"(gdt));
 127         table_base = gdt.base;
 128
 129         if (selector & 4) {           /* from ldt */
 130                 u16 ldt_selector;
 131
 132                 asm("sldt %0" : "=g"(ldt_selector));
 133                 table_base = segment_base(ldt_selector);
 134         }
 135         d = (struct desc_struct *)(table_base + (selector & ~7));
 136         v = d->base0 | ((unsigned long)d->base1 << 16) |
 137                 ((unsigned long)d->base2 << 24);
 138 #ifdef CONFIG_X86_64
 139         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 140                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
 141 #endif
 142         return v;
 143 }
 144 EXPORT_SYMBOL_GPL(segment_base);
 145
 146 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 147 {
 148         if (irqchip_in_kernel(vcpu->kvm))
 149                 return vcpu->arch.apic_base;
 150         else
 151                 return vcpu->arch.apic_base;
 152 }
 153 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 154
 155 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 156 {
 157         /* TODO: reserve bits check */
 158         if (irqchip_in_kernel(vcpu->kvm))
 159                 kvm_lapic_set_base(vcpu, data);
 160         else
 161                 vcpu->arch.apic_base = data;
 162 }
 163 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 164
 165 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 166 {
 167         WARN_ON(vcpu->arch.exception.pending);
 168         vcpu->arch.exception.pending = true;
 169         vcpu->arch.exception.has_error_code = false;
 170         vcpu->arch.exception.nr = nr;
 171 }
 172 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 173
 174 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 175                            u32 error_code)
 176 {
 177         ++vcpu->stat.pf_guest;
 178
 179         if (vcpu->arch.exception.pending) {
 180                 if (vcpu->arch.exception.nr == PF_VECTOR) {
 181                         printk(KERN_DEBUG "kvm: inject_page_fault:"
 182                                         " double fault 0x%lx\n", addr);
 183                         vcpu->arch.exception.nr = DF_VECTOR;
 184                         vcpu->arch.exception.error_code = 0;
 185                 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
 186                         /* triple fault -> shutdown */
 187                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
 188                 }
 189                 return;
 190         }
 191         vcpu->arch.cr2 = addr;
 192         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 193 }
 194
 195 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
 196 {
 197         vcpu->arch.nmi_pending = 1;
 198 }
 199 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
 200
 201 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 202 {
 203         WARN_ON(vcpu->arch.exception.pending);
 204         vcpu->arch.exception.pending = true;
 205         vcpu->arch.exception.has_error_code = true;
 206         vcpu->arch.exception.nr = nr;
 207         vcpu->arch.exception.error_code = error_code;
 208 }
 209 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 210
 211 static void __queue_exception(struct kvm_vcpu *vcpu)
 212 {
 213         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 214                                      vcpu->arch.exception.has_error_code,
 215                                      vcpu->arch.exception.error_code);
 216 }
 217
 218 /*
 219  * Load the pae pdptrs.  Return true is they are all valid.
 220  */
 221 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 222 {
 223         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 224         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 225         int i;
 226         int ret;
 227         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 228
 229         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 230                                   offset * sizeof(u64), sizeof(pdpte));
 231         if (ret < 0) {
 232                 ret = 0;
 233                 goto out;
 234         }
 235         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 236                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 237                         ret = 0;
 238                         goto out;
 239                 }
 240         }
 241         ret = 1;
 242
 243         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 244 out:
 245
 246         return ret;
 247 }
 248 EXPORT_SYMBOL_GPL(load_pdptrs);
 249
 250 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 251 {
 252         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 253         bool changed = true;
 254         int r;
 255
 256         if (is_long_mode(vcpu) || !is_pae(vcpu))
 257                 return false;
 258
 259         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 260         if (r < 0)
 261                 goto out;
 262         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 263 out:
 264
 265         return changed;
 266 }
 267
 268 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 269 {
 270         if (cr0 & CR0_RESERVED_BITS) {
 271                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 272                        cr0, vcpu->arch.cr0);
 273                 kvm_inject_gp(vcpu, 0);
 274                 return;
 275         }
 276
 277         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 278                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 279                 kvm_inject_gp(vcpu, 0);
 280                 return;
 281         }
 282
 283         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 284                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 285                        "and a clear PE flag\n");
 286                 kvm_inject_gp(vcpu, 0);
 287                 return;
 288         }
 289
 290         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 291 #ifdef CONFIG_X86_64
 292                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 293                         int cs_db, cs_l;
 294
 295                         if (!is_pae(vcpu)) {
 296                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 297                                        "in long mode while PAE is disabled\n");
 298                                 kvm_inject_gp(vcpu, 0);
 299                                 return;
 300                         }
 301                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 302                         if (cs_l) {
 303                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 304                                        "in long mode while CS.L == 1\n");
 305                                 kvm_inject_gp(vcpu, 0);
 306                                 return;
 307
 308                         }
 309                 } else
 310 #endif
 311                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 312                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 313                                "reserved bits\n");
 314                         kvm_inject_gp(vcpu, 0);
 315                         return;
 316                 }
 317
 318         }
 319
 320         kvm_x86_ops->set_cr0(vcpu, cr0);
 321         vcpu->arch.cr0 = cr0;
 322
 323         kvm_mmu_sync_global(vcpu);
 324         kvm_mmu_reset_context(vcpu);
 325         return;
 326 }
 327 EXPORT_SYMBOL_GPL(kvm_set_cr0);
 328
 329 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 330 {
 331         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 332         KVMTRACE_1D(LMSW, vcpu,
 333                     (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
 334                     handler);
 335 }
 336 EXPORT_SYMBOL_GPL(kvm_lmsw);
 337
 338 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 339 {
 340         if (cr4 & CR4_RESERVED_BITS) {
 341                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 342                 kvm_inject_gp(vcpu, 0);
 343                 return;
 344         }
 345
 346         if (is_long_mode(vcpu)) {
 347                 if (!(cr4 & X86_CR4_PAE)) {
 348                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 349                                "in long mode\n");
 350                         kvm_inject_gp(vcpu, 0);
 351                         return;
 352                 }
 353         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 354                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 355                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 356                 kvm_inject_gp(vcpu, 0);
 357                 return;
 358         }
 359
 360         if (cr4 & X86_CR4_VMXE) {
 361                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 362                 kvm_inject_gp(vcpu, 0);
 363                 return;
 364         }
 365         kvm_x86_ops->set_cr4(vcpu, cr4);
 366         vcpu->arch.cr4 = cr4;
 367         vcpu->arch.mmu.base_role.cr4_pge = !!(cr4 & X86_CR4_PGE);
 368         kvm_mmu_sync_global(vcpu);
 369         kvm_mmu_reset_context(vcpu);
 370 }
 371 EXPORT_SYMBOL_GPL(kvm_set_cr4);
 372
 373 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 374 {
 375         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 376                 kvm_mmu_sync_roots(vcpu);
 377                 kvm_mmu_flush_tlb(vcpu);
 378                 return;
 379         }
 380
 381         if (is_long_mode(vcpu)) {
 382                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 383                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 384                         kvm_inject_gp(vcpu, 0);
 385                         return;
 386                 }
 387         } else {
 388                 if (is_pae(vcpu)) {
 389                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 390                                 printk(KERN_DEBUG
 391                                        "set_cr3: #GP, reserved bits\n");
 392                                 kvm_inject_gp(vcpu, 0);
 393                                 return;
 394                         }
 395                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 396                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 397                                        "reserved bits\n");
 398                                 kvm_inject_gp(vcpu, 0);
 399                                 return;
 400                         }
 401                 }
 402                 /*
 403                  * We don't check reserved bits in nonpae mode, because
 404                  * this isn't enforced, and VMware depends on this.
 405                  */
 406         }
 407
 408         /*
 409          * Does the new cr3 value map to physical memory? (Note, we
 410          * catch an invalid cr3 even in real-mode, because it would
 411          * cause trouble later on when we turn on paging anyway.)
 412          *
 413          * A real CPU would silently accept an invalid cr3 and would
 414          * attempt to use it - with largely undefined (and often hard
 415          * to debug) behavior on the guest side.
 416          */
 417         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 418                 kvm_inject_gp(vcpu, 0);
 419         else {
 420                 vcpu->arch.cr3 = cr3;
 421                 vcpu->arch.mmu.new_cr3(vcpu);
 422         }
 423 }
 424 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 425
 426 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 427 {
 428         if (cr8 & CR8_RESERVED_BITS) {
 429                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 430                 kvm_inject_gp(vcpu, 0);
 431                 return;
 432         }
 433         if (irqchip_in_kernel(vcpu->kvm))
 434                 kvm_lapic_set_tpr(vcpu, cr8);
 435         else
 436                 vcpu->arch.cr8 = cr8;
 437 }
 438 EXPORT_SYMBOL_GPL(kvm_set_cr8);
 439
 440 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
 441 {
 442         if (irqchip_in_kernel(vcpu->kvm))
 443                 return kvm_lapic_get_cr8(vcpu);
 444         else
 445                 return vcpu->arch.cr8;
 446 }
 447 EXPORT_SYMBOL_GPL(kvm_get_cr8);
 448
 449 static inline u32 bit(int bitno)
 450 {
 451         return 1 << (bitno & 31);
 452 }
 453
 454 /*
 455  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 456  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 457  *
 458  * This list is modified at module load time to reflect the
 459  * capabilities of the host cpu.
 460  */
 461 static u32 msrs_to_save[] = {
 462         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 463         MSR_K6_STAR,
 464 #ifdef CONFIG_X86_64
 465         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 466 #endif
 467         MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
 468         MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
 469 };
 470
 471 static unsigned num_msrs_to_save;
 472
 473 static u32 emulated_msrs[] = {
 474         MSR_IA32_MISC_ENABLE,
 475 };
 476
 477 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 478 {
 479         if (efer & efer_reserved_bits) {
 480                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 481                        efer);
 482                 kvm_inject_gp(vcpu, 0);
 483                 return;
 484         }
 485
 486         if (is_paging(vcpu)
 487             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 488                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 489                 kvm_inject_gp(vcpu, 0);
 490                 return;
 491         }
 492
 493         if (efer & EFER_SVME) {
 494                 struct kvm_cpuid_entry2 *feat;
 495
 496                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
 497                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
 498                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
 499                         kvm_inject_gp(vcpu, 0);
 500                         return;
 501                 }
 502         }
 503
 504         kvm_x86_ops->set_efer(vcpu, efer);
 505
 506         efer &= ~EFER_LMA;
 507         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 508
 509         vcpu->arch.shadow_efer = efer;
 510 }
 511
 512 void kvm_enable_efer_bits(u64 mask)
 513 {
 514        efer_reserved_bits &= ~mask;
 515 }
 516 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 517
 518
 519 /*
 520  * Writes msr value into into the appropriate "register".
 521  * Returns 0 on success, non-0 otherwise.
 522  * Assumes vcpu_load() was already called.
 523  */
 524 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 525 {
 526         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 527 }
 528
 529 /*
 530  * Adapt set_msr() to msr_io()'s calling convention
 531  */
 532 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 533 {
 534         return kvm_set_msr(vcpu, index, *data);
 535 }
 536
 537 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 538 {
 539         static int version;
 540         struct pvclock_wall_clock wc;
 541         struct timespec now, sys, boot;
 542
 543         if (!wall_clock)
 544                 return;
 545
 546         version++;
 547
 548         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 549
 550         /*
 551          * The guest calculates current wall clock time by adding
 552          * system time (updated by kvm_write_guest_time below) to the
 553          * wall clock specified here.  guest system time equals host
 554          * system time for us, thus we must fill in host boot time here.
 555          */
 556         now = current_kernel_time();
 557         ktime_get_ts(&sys);
 558         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
 559
 560         wc.sec = boot.tv_sec;
 561         wc.nsec = boot.tv_nsec;
 562         wc.version = version;
 563
 564         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 565
 566         version++;
 567         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 568 }
 569
 570 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
 571 {
 572         uint32_t quotient, remainder;
 573
 574         /* Don't try to replace with do_div(), this one calculates
 575          * "(dividend << 32) / divisor" */
 576         __asm__ ( "divl %4"
 577                   : "=a" (quotient), "=d" (remainder)
 578                   : "0" (0), "1" (dividend), "r" (divisor) );
 579         return quotient;
 580 }
 581
 582 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
 583 {
 584         uint64_t nsecs = 1000000000LL;
 585         int32_t  shift = 0;
 586         uint64_t tps64;
 587         uint32_t tps32;
 588
 589         tps64 = tsc_khz * 1000LL;
 590         while (tps64 > nsecs*2) {
 591                 tps64 >>= 1;
 592                 shift--;
 593         }
 594
 595         tps32 = (uint32_t)tps64;
 596         while (tps32 <= (uint32_t)nsecs) {
 597                 tps32 <<= 1;
 598                 shift++;
 599         }
 600
 601         hv_clock->tsc_shift = shift;
 602         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
 603
 604         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
 605                  __func__, tsc_khz, hv_clock->tsc_shift,
 606                  hv_clock->tsc_to_system_mul);
 607 }
 608
 609 static void kvm_write_guest_time(struct kvm_vcpu *v)
 610 {
 611         struct timespec ts;
 612         unsigned long flags;
 613         struct kvm_vcpu_arch *vcpu = &v->arch;
 614         void *shared_kaddr;
 615
 616         if ((!vcpu->time_page))
 617                 return;
 618
 619         if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
 620                 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
 621                 vcpu->hv_clock_tsc_khz = tsc_khz;
 622         }
 623
 624         /* Keep irq disabled to prevent changes to the clock */
 625         local_irq_save(flags);
 626         kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
 627                           &vcpu->hv_clock.tsc_timestamp);
 628         ktime_get_ts(&ts);
 629         local_irq_restore(flags);
 630
 631         /* With all the info we got, fill in the values */
 632
 633         vcpu->hv_clock.system_time = ts.tv_nsec +
 634                                      (NSEC_PER_SEC * (u64)ts.tv_sec);
 635         /*
 636          * The interface expects us to write an even number signaling that the
 637          * update is finished. Since the guest won't see the intermediate
 638          * state, we just increase by 2 at the end.
 639          */
 640         vcpu->hv_clock.version += 2;
 641
 642         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 643
 644         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
 645                sizeof(vcpu->hv_clock));
 646
 647         kunmap_atomic(shared_kaddr, KM_USER0);
 648
 649         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
 650 }
 651
 652 static bool msr_mtrr_valid(unsigned msr)
 653 {
 654         switch (msr) {
 655         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
 656         case MSR_MTRRfix64K_00000:
 657         case MSR_MTRRfix16K_80000:
 658         case MSR_MTRRfix16K_A0000:
 659         case MSR_MTRRfix4K_C0000:
 660         case MSR_MTRRfix4K_C8000:
 661         case MSR_MTRRfix4K_D0000:
 662         case MSR_MTRRfix4K_D8000:
 663         case MSR_MTRRfix4K_E0000:
 664         case MSR_MTRRfix4K_E8000:
 665         case MSR_MTRRfix4K_F0000:
 666         case MSR_MTRRfix4K_F8000:
 667         case MSR_MTRRdefType:
 668         case MSR_IA32_CR_PAT:
 669                 return true;
 670         case 0x2f8:
 671                 return true;
 672         }
 673         return false;
 674 }
 675
 676 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 677 {
 678         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 679
 680         if (!msr_mtrr_valid(msr))
 681                 return 1;
 682
 683         if (msr == MSR_MTRRdefType) {
 684                 vcpu->arch.mtrr_state.def_type = data;
 685                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
 686         } else if (msr == MSR_MTRRfix64K_00000)
 687                 p[0] = data;
 688         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 689                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
 690         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 691                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
 692         else if (msr == MSR_IA32_CR_PAT)
 693                 vcpu->arch.pat = data;
 694         else {  /* Variable MTRRs */
 695                 int idx, is_mtrr_mask;
 696                 u64 *pt;
 697
 698                 idx = (msr - 0x200) / 2;
 699                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 700                 if (!is_mtrr_mask)
 701                         pt =
 702                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 703                 else
 704                         pt =
 705                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 706                 *pt = data;
 707         }
 708
 709         kvm_mmu_reset_context(vcpu);
 710         return 0;
 711 }
 712
 713 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 714 {
 715         switch (msr) {
 716         case MSR_EFER:
 717                 set_efer(vcpu, data);
 718                 break;
 719         case MSR_IA32_MC0_STATUS:
 720                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 721                        __func__, data);
 722                 break;
 723         case MSR_IA32_MCG_STATUS:
 724                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 725                         __func__, data);
 726                 break;
 727         case MSR_IA32_MCG_CTL:
 728                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 729                         __func__, data);
 730                 break;
 731         case MSR_IA32_DEBUGCTLMSR:
 732                 if (!data) {
 733                         /* We support the non-activated case already */
 734                         break;
 735                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
 736                         /* Values other than LBR and BTF are vendor-specific,
 737                            thus reserved and should throw a #GP */
 738                         return 1;
 739                 }
 740                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
 741                         __func__, data);
 742                 break;
 743         case MSR_IA32_UCODE_REV:
 744         case MSR_IA32_UCODE_WRITE:
 745         case MSR_VM_HSAVE_PA:
 746                 break;
 747         case 0x200 ... 0x2ff:
 748                 return set_msr_mtrr(vcpu, msr, data);
 749         case MSR_IA32_APICBASE:
 750                 kvm_set_apic_base(vcpu, data);
 751                 break;
 752         case MSR_IA32_MISC_ENABLE:
 753                 vcpu->arch.ia32_misc_enable_msr = data;
 754                 break;
 755         case MSR_KVM_WALL_CLOCK:
 756                 vcpu->kvm->arch.wall_clock = data;
 757                 kvm_write_wall_clock(vcpu->kvm, data);
 758                 break;
 759         case MSR_KVM_SYSTEM_TIME: {
 760                 if (vcpu->arch.time_page) {
 761                         kvm_release_page_dirty(vcpu->arch.time_page);
 762                         vcpu->arch.time_page = NULL;
 763                 }
 764
 765                 vcpu->arch.time = data;
 766
 767                 /* we verify if the enable bit is set... */
 768                 if (!(data & 1))
 769                         break;
 770
 771                 /* ...but clean it before doing the actual write */
 772                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 773
 774                 vcpu->arch.time_page =
 775                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
 776
 777                 if (is_error_page(vcpu->arch.time_page)) {
 778                         kvm_release_page_clean(vcpu->arch.time_page);
 779                         vcpu->arch.time_page = NULL;
 780                 }
 781
 782                 kvm_write_guest_time(vcpu);
 783                 break;
 784         }
 785         default:
 786                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 787                 return 1;
 788         }
 789         return 0;
 790 }
 791 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 792
 793
 794 /*
 795  * Reads an msr value (of 'msr_index') into 'pdata'.
 796  * Returns 0 on success, non-0 otherwise.
 797  * Assumes vcpu_load() was already called.
 798  */
 799 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 800 {
 801         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 802 }
 803
 804 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 805 {
 806         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
 807
 808         if (!msr_mtrr_valid(msr))
 809                 return 1;
 810
 811         if (msr == MSR_MTRRdefType)
 812                 *pdata = vcpu->arch.mtrr_state.def_type +
 813                          (vcpu->arch.mtrr_state.enabled << 10);
 814         else if (msr == MSR_MTRRfix64K_00000)
 815                 *pdata = p[0];
 816         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
 817                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
 818         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
 819                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
 820         else if (msr == MSR_IA32_CR_PAT)
 821                 *pdata = vcpu->arch.pat;
 822         else {  /* Variable MTRRs */
 823                 int idx, is_mtrr_mask;
 824                 u64 *pt;
 825
 826                 idx = (msr - 0x200) / 2;
 827                 is_mtrr_mask = msr - 0x200 - 2 * idx;
 828                 if (!is_mtrr_mask)
 829                         pt =
 830                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
 831                 else
 832                         pt =
 833                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
 834                 *pdata = *pt;
 835         }
 836
 837         return 0;
 838 }
 839
 840 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 841 {
 842         u64 data;
 843
 844         switch (msr) {
 845         case 0xc0010010: /* SYSCFG */
 846         case 0xc0010015: /* HWCR */
 847         case MSR_IA32_PLATFORM_ID:
 848         case MSR_IA32_P5_MC_ADDR:
 849         case MSR_IA32_P5_MC_TYPE:
 850         case MSR_IA32_MC0_CTL:
 851         case MSR_IA32_MCG_STATUS:
 852         case MSR_IA32_MCG_CAP:
 853         case MSR_IA32_MCG_CTL:
 854         case MSR_IA32_MC0_MISC:
 855         case MSR_IA32_MC0_MISC+4:
 856         case MSR_IA32_MC0_MISC+8:
 857         case MSR_IA32_MC0_MISC+12:
 858         case MSR_IA32_MC0_MISC+16:
 859         case MSR_IA32_MC0_MISC+20:
 860         case MSR_IA32_UCODE_REV:
 861         case MSR_IA32_EBL_CR_POWERON:
 862         case MSR_IA32_DEBUGCTLMSR:
 863         case MSR_IA32_LASTBRANCHFROMIP:
 864         case MSR_IA32_LASTBRANCHTOIP:
 865         case MSR_IA32_LASTINTFROMIP:
 866         case MSR_IA32_LASTINTTOIP:
 867         case MSR_VM_HSAVE_PA:
 868                 data = 0;
 869                 break;
 870         case MSR_MTRRcap:
 871                 data = 0x500 | KVM_NR_VAR_MTRR;
 872                 break;
 873         case 0x200 ... 0x2ff:
 874                 return get_msr_mtrr(vcpu, msr, pdata);
 875         case 0xcd: /* fsb frequency */
 876                 data = 3;
 877                 break;
 878         case MSR_IA32_APICBASE:
 879                 data = kvm_get_apic_base(vcpu);
 880                 break;
 881         case MSR_IA32_MISC_ENABLE:
 882                 data = vcpu->arch.ia32_misc_enable_msr;
 883                 break;
 884         case MSR_IA32_PERF_STATUS:
 885                 /* TSC increment by tick */
 886                 data = 1000ULL;
 887                 /* CPU multiplier */
 888                 data |= (((uint64_t)4ULL) << 40);
 889                 break;
 890         case MSR_EFER:
 891                 data = vcpu->arch.shadow_efer;
 892                 break;
 893         case MSR_KVM_WALL_CLOCK:
 894                 data = vcpu->kvm->arch.wall_clock;
 895                 break;
 896         case MSR_KVM_SYSTEM_TIME:
 897                 data = vcpu->arch.time;
 898                 break;
 899         default:
 900                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 901                 return 1;
 902         }
 903         *pdata = data;
 904         return 0;
 905 }
 906 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 907
 908 /*
 909  * Read or write a bunch of msrs. All parameters are kernel addresses.
 910  *
 911  * @return number of msrs set successfully.
 912  */
 913 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 914                     struct kvm_msr_entry *entries,
 915                     int (*do_msr)(struct kvm_vcpu *vcpu,
 916                                   unsigned index, u64 *data))
 917 {
 918         int i;
 919
 920         vcpu_load(vcpu);
 921
 922         down_read(&vcpu->kvm->slots_lock);
 923         for (i = 0; i < msrs->nmsrs; ++i)
 924                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 925                         break;
 926         up_read(&vcpu->kvm->slots_lock);
 927
 928         vcpu_put(vcpu);
 929
 930         return i;
 931 }
 932
 933 /*
 934  * Read or write a bunch of msrs. Parameters are user addresses.
 935  *
 936  * @return number of msrs set successfully.
 937  */
 938 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 939                   int (*do_msr)(struct kvm_vcpu *vcpu,
 940                                 unsigned index, u64 *data),
 941                   int writeback)
 942 {
 943         struct kvm_msrs msrs;
 944         struct kvm_msr_entry *entries;
 945         int r, n;
 946         unsigned size;
 947
 948         r = -EFAULT;
 949         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 950                 goto out;
 951
 952         r = -E2BIG;
 953         if (msrs.nmsrs >= MAX_IO_MSRS)
 954                 goto out;
 955
 956         r = -ENOMEM;
 957         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 958         entries = vmalloc(size);
 959         if (!entries)
 960                 goto out;
 961
 962         r = -EFAULT;
 963         if (copy_from_user(entries, user_msrs->entries, size))
 964                 goto out_free;
 965
 966         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 967         if (r < 0)
 968                 goto out_free;
 969
 970         r = -EFAULT;
 971         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 972                 goto out_free;
 973
 974         r = n;
 975
 976 out_free:
 977         vfree(entries);
 978 out:
 979         return r;
 980 }
 981
 982 int kvm_dev_ioctl_check_extension(long ext)
 983 {
 984         int r;
 985
 986         switch (ext) {
 987         case KVM_CAP_IRQCHIP:
 988         case KVM_CAP_HLT:
 989         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 990         case KVM_CAP_SET_TSS_ADDR:
 991         case KVM_CAP_EXT_CPUID:
 992         case KVM_CAP_PIT:
 993         case KVM_CAP_NOP_IO_DELAY:
 994         case KVM_CAP_MP_STATE:
 995         case KVM_CAP_SYNC_MMU:
 996         case KVM_CAP_REINJECT_CONTROL:
 997                 r = 1;
 998                 break;
 999         case KVM_CAP_COALESCED_MMIO:
1000                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1001                 break;
1002         case KVM_CAP_VAPIC:
1003                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1004                 break;
1005         case KVM_CAP_NR_VCPUS:
1006                 r = KVM_MAX_VCPUS;
1007                 break;
1008         case KVM_CAP_NR_MEMSLOTS:
1009                 r = KVM_MEMORY_SLOTS;
1010                 break;
1011         case KVM_CAP_PV_MMU:
1012                 r = !tdp_enabled;
1013                 break;
1014         case KVM_CAP_IOMMU:
1015                 r = iommu_found();
1016                 break;
1017         case KVM_CAP_CLOCKSOURCE:
1018                 r = boot_cpu_has(X86_FEATURE_CONSTANT_TSC);
1019                 break;
1020         default:
1021                 r = 0;
1022                 break;
1023         }
1024         return r;
1025
1026 }
1027
1028 long kvm_arch_dev_ioctl(struct file *filp,
1029                         unsigned int ioctl, unsigned long arg)
1030 {
1031         void __user *argp = (void __user *)arg;
1032         long r;
1033
1034         switch (ioctl) {
1035         case KVM_GET_MSR_INDEX_LIST: {
1036                 struct kvm_msr_list __user *user_msr_list = argp;
1037                 struct kvm_msr_list msr_list;
1038                 unsigned n;
1039
1040                 r = -EFAULT;
1041                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1042                         goto out;
1043                 n = msr_list.nmsrs;
1044                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1045                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1046                         goto out;
1047                 r = -E2BIG;
1048                 if (n < num_msrs_to_save)
1049                         goto out;
1050                 r = -EFAULT;
1051                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1052                                  num_msrs_to_save * sizeof(u32)))
1053                         goto out;
1054                 if (copy_to_user(user_msr_list->indices
1055                                  + num_msrs_to_save * sizeof(u32),
1056                                  &emulated_msrs,
1057                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1058                         goto out;
1059                 r = 0;
1060                 break;
1061         }
1062         case KVM_GET_SUPPORTED_CPUID: {
1063                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1064                 struct kvm_cpuid2 cpuid;
1065
1066                 r = -EFAULT;
1067                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1068                         goto out;
1069                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1070                         cpuid_arg->entries);
1071                 if (r)
1072                         goto out;
1073
1074                 r = -EFAULT;
1075                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1076                         goto out;
1077                 r = 0;
1078                 break;
1079         }
1080         default:
1081                 r = -EINVAL;
1082         }
1083 out:
1084         return r;
1085 }
1086
1087 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1088 {
1089         kvm_x86_ops->vcpu_load(vcpu, cpu);
1090         kvm_write_guest_time(vcpu);
1091 }
1092
1093 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1094 {
1095         kvm_x86_ops->vcpu_put(vcpu);
1096         kvm_put_guest_fpu(vcpu);
1097 }
1098
1099 static int is_efer_nx(void)
1100 {
1101         u64 efer;
1102
1103         rdmsrl(MSR_EFER, efer);
1104         return efer & EFER_NX;
1105 }
1106
1107 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1108 {
1109         int i;
1110         struct kvm_cpuid_entry2 *e, *entry;
1111
1112         entry = NULL;
1113         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1114                 e = &vcpu->arch.cpuid_entries[i];
1115                 if (e->function == 0x80000001) {
1116                         entry = e;
1117                         break;
1118                 }
1119         }
1120         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1121                 entry->edx &= ~(1 << 20);
1122                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1123         }
1124 }
1125
1126 /* when an old userspace process fills a new kernel module */
1127 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1128                                     struct kvm_cpuid *cpuid,
1129                                     struct kvm_cpuid_entry __user *entries)
1130 {
1131         int r, i;
1132         struct kvm_cpuid_entry *cpuid_entries;
1133
1134         r = -E2BIG;
1135         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1136                 goto out;
1137         r = -ENOMEM;
1138         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1139         if (!cpuid_entries)
1140                 goto out;
1141         r = -EFAULT;
1142         if (copy_from_user(cpuid_entries, entries,
1143                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1144                 goto out_free;
1145         for (i = 0; i < cpuid->nent; i++) {
1146                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1147                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1148                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1149                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1150                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1151                 vcpu->arch.cpuid_entries[i].index = 0;
1152                 vcpu->arch.cpuid_entries[i].flags = 0;
1153                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1154                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1155                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1156         }
1157         vcpu->arch.cpuid_nent = cpuid->nent;
1158         cpuid_fix_nx_cap(vcpu);
1159         r = 0;
1160
1161 out_free:
1162         vfree(cpuid_entries);
1163 out:
1164         return r;
1165 }
1166
1167 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1168                                     struct kvm_cpuid2 *cpuid,
1169                                     struct kvm_cpuid_entry2 __user *entries)
1170 {
1171         int r;
1172
1173         r = -E2BIG;
1174         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1175                 goto out;
1176         r = -EFAULT;
1177         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1178                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1179                 goto out;
1180         vcpu->arch.cpuid_nent = cpuid->nent;
1181         return 0;
1182
1183 out:
1184         return r;
1185 }
1186
1187 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1188                                     struct kvm_cpuid2 *cpuid,
1189                                     struct kvm_cpuid_entry2 __user *entries)
1190 {
1191         int r;
1192
1193         r = -E2BIG;
1194         if (cpuid->nent < vcpu->arch.cpuid_nent)
1195                 goto out;
1196         r = -EFAULT;
1197         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1198                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1199                 goto out;
1200         return 0;
1201
1202 out:
1203         cpuid->nent = vcpu->arch.cpuid_nent;
1204         return r;
1205 }
1206
1207 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1208                           u32 index)
1209 {
1210         entry->function = function;
1211         entry->index = index;
1212         cpuid_count(entry->function, entry->index,
1213                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1214         entry->flags = 0;
1215 }
1216
1217 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1218                          u32 index, int *nent, int maxnent)
1219 {
1220         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1221                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1222                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1223                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1224                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1225                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1226                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1227                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1228                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1229                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1230         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1231                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1232                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1233                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1234                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1235                 bit(X86_FEATURE_PGE) |
1236                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1237                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1238                 bit(X86_FEATURE_SYSCALL) |
1239                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1240 #ifdef CONFIG_X86_64
1241                 bit(X86_FEATURE_LM) |
1242 #endif
1243                 bit(X86_FEATURE_MMXEXT) |
1244                 bit(X86_FEATURE_3DNOWEXT) |
1245                 bit(X86_FEATURE_3DNOW);
1246         const u32 kvm_supported_word3_x86_features =
1247                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1248         const u32 kvm_supported_word6_x86_features =
1249                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY) |
1250                 bit(X86_FEATURE_SVM);
1251
1252         /* all func 2 cpuid_count() should be called on the same cpu */
1253         get_cpu();
1254         do_cpuid_1_ent(entry, function, index);
1255         ++*nent;
1256
1257         switch (function) {
1258         case 0:
1259                 entry->eax = min(entry->eax, (u32)0xb);
1260                 break;
1261         case 1:
1262                 entry->edx &= kvm_supported_word0_x86_features;
1263                 entry->ecx &= kvm_supported_word3_x86_features;
1264                 break;
1265         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1266          * may return different values. This forces us to get_cpu() before
1267          * issuing the first command, and also to emulate this annoying behavior
1268          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1269         case 2: {
1270                 int t, times = entry->eax & 0xff;
1271
1272                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1273                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1274                 for (t = 1; t < times && *nent < maxnent; ++t) {
1275                         do_cpuid_1_ent(&entry[t], function, 0);
1276                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1277                         ++*nent;
1278                 }
1279                 break;
1280         }
1281         /* function 4 and 0xb have additional index. */
1282         case 4: {
1283                 int i, cache_type;
1284
1285                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1286                 /* read more entries until cache_type is zero */
1287                 for (i = 1; *nent < maxnent; ++i) {
1288                         cache_type = entry[i - 1].eax & 0x1f;
1289                         if (!cache_type)
1290                                 break;
1291                         do_cpuid_1_ent(&entry[i], function, i);
1292                         entry[i].flags |=
1293                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1294                         ++*nent;
1295                 }
1296                 break;
1297         }
1298         case 0xb: {
1299                 int i, level_type;
1300
1301                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1302                 /* read more entries until level_type is zero */
1303                 for (i = 1; *nent < maxnent; ++i) {
1304                         level_type = entry[i - 1].ecx & 0xff00;
1305                         if (!level_type)
1306                                 break;
1307                         do_cpuid_1_ent(&entry[i], function, i);
1308                         entry[i].flags |=
1309                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1310                         ++*nent;
1311                 }
1312                 break;
1313         }
1314         case 0x80000000:
1315                 entry->eax = min(entry->eax, 0x8000001a);
1316                 break;
1317         case 0x80000001:
1318                 entry->edx &= kvm_supported_word1_x86_features;
1319                 entry->ecx &= kvm_supported_word6_x86_features;
1320                 break;
1321         }
1322         put_cpu();
1323 }
1324
1325 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1326                                     struct kvm_cpuid_entry2 __user *entries)
1327 {
1328         struct kvm_cpuid_entry2 *cpuid_entries;
1329         int limit, nent = 0, r = -E2BIG;
1330         u32 func;
1331
1332         if (cpuid->nent < 1)
1333                 goto out;
1334         r = -ENOMEM;
1335         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1336         if (!cpuid_entries)
1337                 goto out;
1338
1339         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1340         limit = cpuid_entries[0].eax;
1341         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1342                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1343                                 &nent, cpuid->nent);
1344         r = -E2BIG;
1345         if (nent >= cpuid->nent)
1346                 goto out_free;
1347
1348         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1349         limit = cpuid_entries[nent - 1].eax;
1350         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1351                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1352                                &nent, cpuid->nent);
1353         r = -EFAULT;
1354         if (copy_to_user(entries, cpuid_entries,
1355                         nent * sizeof(struct kvm_cpuid_entry2)))
1356                 goto out_free;
1357         cpuid->nent = nent;
1358         r = 0;
1359
1360 out_free:
1361         vfree(cpuid_entries);
1362 out:
1363         return r;
1364 }
1365
1366 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1367                                     struct kvm_lapic_state *s)
1368 {
1369         vcpu_load(vcpu);
1370         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1371         vcpu_put(vcpu);
1372
1373         return 0;
1374 }
1375
1376 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1377                                     struct kvm_lapic_state *s)
1378 {
1379         vcpu_load(vcpu);
1380         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1381         kvm_apic_post_state_restore(vcpu);
1382         vcpu_put(vcpu);
1383
1384         return 0;
1385 }
1386
1387 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1388                                     struct kvm_interrupt *irq)
1389 {
1390         if (irq->irq < 0 || irq->irq >= 256)
1391                 return -EINVAL;
1392         if (irqchip_in_kernel(vcpu->kvm))
1393                 return -ENXIO;
1394         vcpu_load(vcpu);
1395
1396         set_bit(irq->irq, vcpu->arch.irq_pending);
1397         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1398
1399         vcpu_put(vcpu);
1400
1401         return 0;
1402 }
1403
1404 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1405 {
1406         vcpu_load(vcpu);
1407         kvm_inject_nmi(vcpu);
1408         vcpu_put(vcpu);
1409
1410         return 0;
1411 }
1412
1413 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1414                                            struct kvm_tpr_access_ctl *tac)
1415 {
1416         if (tac->flags)
1417                 return -EINVAL;
1418         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1419         return 0;
1420 }
1421
1422 long kvm_arch_vcpu_ioctl(struct file *filp,
1423                          unsigned int ioctl, unsigned long arg)
1424 {
1425         struct kvm_vcpu *vcpu = filp->private_data;
1426         void __user *argp = (void __user *)arg;
1427         int r;
1428         struct kvm_lapic_state *lapic = NULL;
1429
1430         switch (ioctl) {
1431         case KVM_GET_LAPIC: {
1432                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1433
1434                 r = -ENOMEM;
1435                 if (!lapic)
1436                         goto out;
1437                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1438                 if (r)
1439                         goto out;
1440                 r = -EFAULT;
1441                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1442                         goto out;
1443                 r = 0;
1444                 break;
1445         }
1446         case KVM_SET_LAPIC: {
1447                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1448                 r = -ENOMEM;
1449                 if (!lapic)
1450                         goto out;
1451                 r = -EFAULT;
1452                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1453                         goto out;
1454                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1455                 if (r)
1456                         goto out;
1457                 r = 0;
1458                 break;
1459         }
1460         case KVM_INTERRUPT: {
1461                 struct kvm_interrupt irq;
1462
1463                 r = -EFAULT;
1464                 if (copy_from_user(&irq, argp, sizeof irq))
1465                         goto out;
1466                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1467                 if (r)
1468                         goto out;
1469                 r = 0;
1470                 break;
1471         }
1472         case KVM_NMI: {
1473                 r = kvm_vcpu_ioctl_nmi(vcpu);
1474                 if (r)
1475                         goto out;
1476                 r = 0;
1477                 break;
1478         }
1479         case KVM_SET_CPUID: {
1480                 struct kvm_cpuid __user *cpuid_arg = argp;
1481                 struct kvm_cpuid cpuid;
1482
1483                 r = -EFAULT;
1484                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1485                         goto out;
1486                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1487                 if (r)
1488                         goto out;
1489                 break;
1490         }
1491         case KVM_SET_CPUID2: {
1492                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1493                 struct kvm_cpuid2 cpuid;
1494
1495                 r = -EFAULT;
1496                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1497                         goto out;
1498                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1499                                 cpuid_arg->entries);
1500                 if (r)
1501                         goto out;
1502                 break;
1503         }
1504         case KVM_GET_CPUID2: {
1505                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1506                 struct kvm_cpuid2 cpuid;
1507
1508                 r = -EFAULT;
1509                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1510                         goto out;
1511                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1512                                 cpuid_arg->entries);
1513                 if (r)
1514                         goto out;
1515                 r = -EFAULT;
1516                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1517                         goto out;
1518                 r = 0;
1519                 break;
1520         }
1521         case KVM_GET_MSRS:
1522                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1523                 break;
1524         case KVM_SET_MSRS:
1525                 r = msr_io(vcpu, argp, do_set_msr, 0);
1526                 break;
1527         case KVM_TPR_ACCESS_REPORTING: {
1528                 struct kvm_tpr_access_ctl tac;
1529
1530                 r = -EFAULT;
1531                 if (copy_from_user(&tac, argp, sizeof tac))
1532                         goto out;
1533                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1534                 if (r)
1535                         goto out;
1536                 r = -EFAULT;
1537                 if (copy_to_user(argp, &tac, sizeof tac))
1538                         goto out;
1539                 r = 0;
1540                 break;
1541         };
1542         case KVM_SET_VAPIC_ADDR: {
1543                 struct kvm_vapic_addr va;
1544
1545                 r = -EINVAL;
1546                 if (!irqchip_in_kernel(vcpu->kvm))
1547                         goto out;
1548                 r = -EFAULT;
1549                 if (copy_from_user(&va, argp, sizeof va))
1550                         goto out;
1551                 r = 0;
1552                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1553                 break;
1554         }
1555         default:
1556                 r = -EINVAL;
1557         }
1558 out:
1559         if (lapic)
1560                 kfree(lapic);
1561         return r;
1562 }
1563
1564 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1565 {
1566         int ret;
1567
1568         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1569                 return -1;
1570         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1571         return ret;
1572 }
1573
1574 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1575                                           u32 kvm_nr_mmu_pages)
1576 {
1577         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1578                 return -EINVAL;
1579
1580         down_write(&kvm->slots_lock);
1581
1582         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1583         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1584
1585         up_write(&kvm->slots_lock);
1586         return 0;
1587 }
1588
1589 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1590 {
1591         return kvm->arch.n_alloc_mmu_pages;
1592 }
1593
1594 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1595 {
1596         int i;
1597         struct kvm_mem_alias *alias;
1598
1599         for (i = 0; i < kvm->arch.naliases; ++i) {
1600                 alias = &kvm->arch.aliases[i];
1601                 if (gfn >= alias->base_gfn
1602                     && gfn < alias->base_gfn + alias->npages)
1603                         return alias->target_gfn + gfn - alias->base_gfn;
1604         }
1605         return gfn;
1606 }
1607
1608 /*
1609  * Set a new alias region.  Aliases map a portion of physical memory into
1610  * another portion.  This is useful for memory windows, for example the PC
1611  * VGA region.
1612  */
1613 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1614                                          struct kvm_memory_alias *alias)
1615 {
1616         int r, n;
1617         struct kvm_mem_alias *p;
1618
1619         r = -EINVAL;
1620         /* General sanity checks */
1621         if (alias->memory_size & (PAGE_SIZE - 1))
1622                 goto out;
1623         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1624                 goto out;
1625         if (alias->slot >= KVM_ALIAS_SLOTS)
1626                 goto out;
1627         if (alias->guest_phys_addr + alias->memory_size
1628             < alias->guest_phys_addr)
1629                 goto out;
1630         if (alias->target_phys_addr + alias->memory_size
1631             < alias->target_phys_addr)
1632                 goto out;
1633
1634         down_write(&kvm->slots_lock);
1635         spin_lock(&kvm->mmu_lock);
1636
1637         p = &kvm->arch.aliases[alias->slot];
1638         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1639         p->npages = alias->memory_size >> PAGE_SHIFT;
1640         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1641
1642         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1643                 if (kvm->arch.aliases[n - 1].npages)
1644                         break;
1645         kvm->arch.naliases = n;
1646
1647         spin_unlock(&kvm->mmu_lock);
1648         kvm_mmu_zap_all(kvm);
1649
1650         up_write(&kvm->slots_lock);
1651
1652         return 0;
1653
1654 out:
1655         return r;
1656 }
1657
1658 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1659 {
1660         int r;
1661
1662         r = 0;
1663         switch (chip->chip_id) {
1664         case KVM_IRQCHIP_PIC_MASTER:
1665                 memcpy(&chip->chip.pic,
1666                         &pic_irqchip(kvm)->pics[0],
1667                         sizeof(struct kvm_pic_state));
1668                 break;
1669         case KVM_IRQCHIP_PIC_SLAVE:
1670                 memcpy(&chip->chip.pic,
1671                         &pic_irqchip(kvm)->pics[1],
1672                         sizeof(struct kvm_pic_state));
1673                 break;
1674         case KVM_IRQCHIP_IOAPIC:
1675                 memcpy(&chip->chip.ioapic,
1676                         ioapic_irqchip(kvm),
1677                         sizeof(struct kvm_ioapic_state));
1678                 break;
1679         default:
1680                 r = -EINVAL;
1681                 break;
1682         }
1683         return r;
1684 }
1685
1686 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1687 {
1688         int r;
1689
1690         r = 0;
1691         switch (chip->chip_id) {
1692         case KVM_IRQCHIP_PIC_MASTER:
1693                 memcpy(&pic_irqchip(kvm)->pics[0],
1694                         &chip->chip.pic,
1695                         sizeof(struct kvm_pic_state));
1696                 break;
1697         case KVM_IRQCHIP_PIC_SLAVE:
1698                 memcpy(&pic_irqchip(kvm)->pics[1],
1699                         &chip->chip.pic,
1700                         sizeof(struct kvm_pic_state));
1701                 break;
1702         case KVM_IRQCHIP_IOAPIC:
1703                 memcpy(ioapic_irqchip(kvm),
1704                         &chip->chip.ioapic,
1705                         sizeof(struct kvm_ioapic_state));
1706                 break;
1707         default:
1708                 r = -EINVAL;
1709                 break;
1710         }
1711         kvm_pic_update_irq(pic_irqchip(kvm));
1712         return r;
1713 }
1714
1715 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1716 {
1717         int r = 0;
1718
1719         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1720         return r;
1721 }
1722
1723 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1724 {
1725         int r = 0;
1726
1727         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1728         kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1729         return r;
1730 }
1731
1732 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1733                                  struct kvm_reinject_control *control)
1734 {
1735         if (!kvm->arch.vpit)
1736                 return -ENXIO;
1737         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1738         return 0;
1739 }
1740
1741 /*
1742  * Get (and clear) the dirty memory log for a memory slot.
1743  */
1744 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1745                                       struct kvm_dirty_log *log)
1746 {
1747         int r;
1748         int n;
1749         struct kvm_memory_slot *memslot;
1750         int is_dirty = 0;
1751
1752         down_write(&kvm->slots_lock);
1753
1754         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1755         if (r)
1756                 goto out;
1757
1758         /* If nothing is dirty, don't bother messing with page tables. */
1759         if (is_dirty) {
1760                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1761                 kvm_flush_remote_tlbs(kvm);
1762                 memslot = &kvm->memslots[log->slot];
1763                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1764                 memset(memslot->dirty_bitmap, 0, n);
1765         }
1766         r = 0;
1767 out:
1768         up_write(&kvm->slots_lock);
1769         return r;
1770 }
1771
1772 long kvm_arch_vm_ioctl(struct file *filp,
1773                        unsigned int ioctl, unsigned long arg)
1774 {
1775         struct kvm *kvm = filp->private_data;
1776         void __user *argp = (void __user *)arg;
1777         int r = -EINVAL;
1778         /*
1779          * This union makes it completely explicit to gcc-3.x
1780          * that these two variables' stack usage should be
1781          * combined, not added together.
1782          */
1783         union {
1784                 struct kvm_pit_state ps;
1785                 struct kvm_memory_alias alias;
1786         } u;
1787
1788         switch (ioctl) {
1789         case KVM_SET_TSS_ADDR:
1790                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1791                 if (r < 0)
1792                         goto out;
1793                 break;
1794         case KVM_SET_MEMORY_REGION: {
1795                 struct kvm_memory_region kvm_mem;
1796                 struct kvm_userspace_memory_region kvm_userspace_mem;
1797
1798                 r = -EFAULT;
1799                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1800                         goto out;
1801                 kvm_userspace_mem.slot = kvm_mem.slot;
1802                 kvm_userspace_mem.flags = kvm_mem.flags;
1803                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1804                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1805                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1806                 if (r)
1807                         goto out;
1808                 break;
1809         }
1810         case KVM_SET_NR_MMU_PAGES:
1811                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1812                 if (r)
1813                         goto out;
1814                 break;
1815         case KVM_GET_NR_MMU_PAGES:
1816                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1817                 break;
1818         case KVM_SET_MEMORY_ALIAS:
1819                 r = -EFAULT;
1820                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1821                         goto out;
1822                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1823                 if (r)
1824                         goto out;
1825                 break;
1826         case KVM_CREATE_IRQCHIP:
1827                 r = -ENOMEM;
1828                 kvm->arch.vpic = kvm_create_pic(kvm);
1829                 if (kvm->arch.vpic) {
1830                         r = kvm_ioapic_init(kvm);
1831                         if (r) {
1832                                 kfree(kvm->arch.vpic);
1833                                 kvm->arch.vpic = NULL;
1834                                 goto out;
1835                         }
1836                 } else
1837                         goto out;
1838                 break;
1839         case KVM_CREATE_PIT:
1840                 r = -ENOMEM;
1841                 kvm->arch.vpit = kvm_create_pit(kvm);
1842                 if (kvm->arch.vpit)
1843                         r = 0;
1844                 break;
1845         case KVM_IRQ_LINE: {
1846                 struct kvm_irq_level irq_event;
1847
1848                 r = -EFAULT;
1849                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1850                         goto out;
1851                 if (irqchip_in_kernel(kvm)) {
1852                         mutex_lock(&kvm->lock);
1853                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1854                                     irq_event.irq, irq_event.level);
1855                         mutex_unlock(&kvm->lock);
1856                         r = 0;
1857                 }
1858                 break;
1859         }
1860         case KVM_GET_IRQCHIP: {
1861                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1862                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1863
1864                 r = -ENOMEM;
1865                 if (!chip)
1866                         goto out;
1867                 r = -EFAULT;
1868                 if (copy_from_user(chip, argp, sizeof *chip))
1869                         goto get_irqchip_out;
1870                 r = -ENXIO;
1871                 if (!irqchip_in_kernel(kvm))
1872                         goto get_irqchip_out;
1873                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1874                 if (r)
1875                         goto get_irqchip_out;
1876                 r = -EFAULT;
1877                 if (copy_to_user(argp, chip, sizeof *chip))
1878                         goto get_irqchip_out;
1879                 r = 0;
1880         get_irqchip_out:
1881                 kfree(chip);
1882                 if (r)
1883                         goto out;
1884                 break;
1885         }
1886         case KVM_SET_IRQCHIP: {
1887                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1888                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1889
1890                 r = -ENOMEM;
1891                 if (!chip)
1892                         goto out;
1893                 r = -EFAULT;
1894                 if (copy_from_user(chip, argp, sizeof *chip))
1895                         goto set_irqchip_out;
1896                 r = -ENXIO;
1897                 if (!irqchip_in_kernel(kvm))
1898                         goto set_irqchip_out;
1899                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1900                 if (r)
1901                         goto set_irqchip_out;
1902                 r = 0;
1903         set_irqchip_out:
1904                 kfree(chip);
1905                 if (r)
1906                         goto out;
1907                 break;
1908         }
1909         case KVM_GET_PIT: {
1910                 r = -EFAULT;
1911                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1912                         goto out;
1913                 r = -ENXIO;
1914                 if (!kvm->arch.vpit)
1915                         goto out;
1916                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1917                 if (r)
1918                         goto out;
1919                 r = -EFAULT;
1920                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1921                         goto out;
1922                 r = 0;
1923                 break;
1924         }
1925         case KVM_SET_PIT: {
1926                 r = -EFAULT;
1927                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1928                         goto out;
1929                 r = -ENXIO;
1930                 if (!kvm->arch.vpit)
1931                         goto out;
1932                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1933                 if (r)
1934                         goto out;
1935                 r = 0;
1936                 break;
1937         }
1938         case KVM_REINJECT_CONTROL: {
1939                 struct kvm_reinject_control control;
1940                 r =  -EFAULT;
1941                 if (copy_from_user(&control, argp, sizeof(control)))
1942                         goto out;
1943                 r = kvm_vm_ioctl_reinject(kvm, &control);
1944                 if (r)
1945                         goto out;
1946                 r = 0;
1947                 break;
1948         }
1949         default:
1950                 ;
1951         }
1952 out:
1953         return r;
1954 }
1955
1956 static void kvm_init_msr_list(void)
1957 {
1958         u32 dummy[2];
1959         unsigned i, j;
1960
1961         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1962                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1963                         continue;
1964                 if (j < i)
1965                         msrs_to_save[j] = msrs_to_save[i];
1966                 j++;
1967         }
1968         num_msrs_to_save = j;
1969 }
1970
1971 /*
1972  * Only apic need an MMIO device hook, so shortcut now..
1973  */
1974 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1975                                                 gpa_t addr, int len,
1976                                                 int is_write)
1977 {
1978         struct kvm_io_device *dev;
1979
1980         if (vcpu->arch.apic) {
1981                 dev = &vcpu->arch.apic->dev;
1982                 if (dev->in_range(dev, addr, len, is_write))
1983                         return dev;
1984         }
1985         return NULL;
1986 }
1987
1988
1989 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1990                                                 gpa_t addr, int len,
1991                                                 int is_write)
1992 {
1993         struct kvm_io_device *dev;
1994
1995         dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1996         if (dev == NULL)
1997                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1998                                           is_write);
1999         return dev;
2000 }
2001
2002 int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2003                         struct kvm_vcpu *vcpu)
2004 {
2005         void *data = val;
2006         int r = X86EMUL_CONTINUE;
2007
2008         while (bytes) {
2009                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2010                 unsigned offset = addr & (PAGE_SIZE-1);
2011                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2012                 int ret;
2013
2014                 if (gpa == UNMAPPED_GVA) {
2015                         r = X86EMUL_PROPAGATE_FAULT;
2016                         goto out;
2017                 }
2018                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2019                 if (ret < 0) {
2020                         r = X86EMUL_UNHANDLEABLE;
2021                         goto out;
2022                 }
2023
2024                 bytes -= toread;
2025                 data += toread;
2026                 addr += toread;
2027         }
2028 out:
2029         return r;
2030 }
2031
2032 int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2033                          struct kvm_vcpu *vcpu)
2034 {
2035         void *data = val;
2036         int r = X86EMUL_CONTINUE;
2037
2038         while (bytes) {
2039                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2040                 unsigned offset = addr & (PAGE_SIZE-1);
2041                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2042                 int ret;
2043
2044                 if (gpa == UNMAPPED_GVA) {
2045                         r = X86EMUL_PROPAGATE_FAULT;
2046                         goto out;
2047                 }
2048                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2049                 if (ret < 0) {
2050                         r = X86EMUL_UNHANDLEABLE;
2051                         goto out;
2052                 }
2053
2054                 bytes -= towrite;
2055                 data += towrite;
2056                 addr += towrite;
2057         }
2058 out:
2059         return r;
2060 }
2061
2062
2063 static int emulator_read_emulated(unsigned long addr,
2064                                   void *val,
2065                                   unsigned int bytes,
2066                                   struct kvm_vcpu *vcpu)
2067 {
2068         struct kvm_io_device *mmio_dev;
2069         gpa_t                 gpa;
2070
2071         if (vcpu->mmio_read_completed) {
2072                 memcpy(val, vcpu->mmio_data, bytes);
2073                 vcpu->mmio_read_completed = 0;
2074                 return X86EMUL_CONTINUE;
2075         }
2076
2077         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2078
2079         /* For APIC access vmexit */
2080         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2081                 goto mmio;
2082
2083         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2084                                 == X86EMUL_CONTINUE)
2085                 return X86EMUL_CONTINUE;
2086         if (gpa == UNMAPPED_GVA)
2087                 return X86EMUL_PROPAGATE_FAULT;
2088
2089 mmio:
2090         /*
2091          * Is this MMIO handled locally?
2092          */
2093         mutex_lock(&vcpu->kvm->lock);
2094         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2095         if (mmio_dev) {
2096                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2097                 mutex_unlock(&vcpu->kvm->lock);
2098                 return X86EMUL_CONTINUE;
2099         }
2100         mutex_unlock(&vcpu->kvm->lock);
2101
2102         vcpu->mmio_needed = 1;
2103         vcpu->mmio_phys_addr = gpa;
2104         vcpu->mmio_size = bytes;
2105         vcpu->mmio_is_write = 0;
2106
2107         return X86EMUL_UNHANDLEABLE;
2108 }
2109
2110 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2111                           const void *val, int bytes)
2112 {
2113         int ret;
2114
2115         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2116         if (ret < 0)
2117                 return 0;
2118         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2119         return 1;
2120 }
2121
2122 static int emulator_write_emulated_onepage(unsigned long addr,
2123                                            const void *val,
2124                                            unsigned int bytes,
2125                                            struct kvm_vcpu *vcpu)
2126 {
2127         struct kvm_io_device *mmio_dev;
2128         gpa_t                 gpa;
2129
2130         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2131
2132         if (gpa == UNMAPPED_GVA) {
2133                 kvm_inject_page_fault(vcpu, addr, 2);
2134                 return X86EMUL_PROPAGATE_FAULT;
2135         }
2136
2137         /* For APIC access vmexit */
2138         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2139                 goto mmio;
2140
2141         if (emulator_write_phys(vcpu, gpa, val, bytes))
2142                 return X86EMUL_CONTINUE;
2143
2144 mmio:
2145         /*
2146          * Is this MMIO handled locally?
2147          */
2148         mutex_lock(&vcpu->kvm->lock);
2149         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2150         if (mmio_dev) {
2151                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2152                 mutex_unlock(&vcpu->kvm->lock);
2153                 return X86EMUL_CONTINUE;
2154         }
2155         mutex_unlock(&vcpu->kvm->lock);
2156
2157         vcpu->mmio_needed = 1;
2158         vcpu->mmio_phys_addr = gpa;
2159         vcpu->mmio_size = bytes;
2160         vcpu->mmio_is_write = 1;
2161         memcpy(vcpu->mmio_data, val, bytes);
2162
2163         return X86EMUL_CONTINUE;
2164 }
2165
2166 int emulator_write_emulated(unsigned long addr,
2167                                    const void *val,
2168                                    unsigned int bytes,
2169                                    struct kvm_vcpu *vcpu)
2170 {
2171         /* Crossing a page boundary? */
2172         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2173                 int rc, now;
2174
2175                 now = -addr & ~PAGE_MASK;
2176                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2177                 if (rc != X86EMUL_CONTINUE)
2178                         return rc;
2179                 addr += now;
2180                 val += now;
2181                 bytes -= now;
2182         }
2183         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2184 }
2185 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2186
2187 static int emulator_cmpxchg_emulated(unsigned long addr,
2188                                      const void *old,
2189                                      const void *new,
2190                                      unsigned int bytes,
2191                                      struct kvm_vcpu *vcpu)
2192 {
2193         static int reported;
2194
2195         if (!reported) {
2196                 reported = 1;
2197                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2198         }
2199 #ifndef CONFIG_X86_64
2200         /* guests cmpxchg8b have to be emulated atomically */
2201         if (bytes == 8) {
2202                 gpa_t gpa;
2203                 struct page *page;
2204                 char *kaddr;
2205                 u64 val;
2206
2207                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2208
2209                 if (gpa == UNMAPPED_GVA ||
2210                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2211                         goto emul_write;
2212
2213                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2214                         goto emul_write;
2215
2216                 val = *(u64 *)new;
2217
2218                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2219
2220                 kaddr = kmap_atomic(page, KM_USER0);
2221                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2222                 kunmap_atomic(kaddr, KM_USER0);
2223                 kvm_release_page_dirty(page);
2224         }
2225 emul_write:
2226 #endif
2227
2228         return emulator_write_emulated(addr, new, bytes, vcpu);
2229 }
2230
2231 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2232 {
2233         return kvm_x86_ops->get_segment_base(vcpu, seg);
2234 }
2235
2236 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2237 {
2238         kvm_mmu_invlpg(vcpu, address);
2239         return X86EMUL_CONTINUE;
2240 }
2241
2242 int emulate_clts(struct kvm_vcpu *vcpu)
2243 {
2244         KVMTRACE_0D(CLTS, vcpu, handler);
2245         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2246         return X86EMUL_CONTINUE;
2247 }
2248
2249 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2250 {
2251         struct kvm_vcpu *vcpu = ctxt->vcpu;
2252
2253         switch (dr) {
2254         case 0 ... 3:
2255                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2256                 return X86EMUL_CONTINUE;
2257         default:
2258                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2259                 return X86EMUL_UNHANDLEABLE;
2260         }
2261 }
2262
2263 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2264 {
2265         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2266         int exception;
2267
2268         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2269         if (exception) {
2270                 /* FIXME: better handling */
2271                 return X86EMUL_UNHANDLEABLE;
2272         }
2273         return X86EMUL_CONTINUE;
2274 }
2275
2276 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2277 {
2278         u8 opcodes[4];
2279         unsigned long rip = kvm_rip_read(vcpu);
2280         unsigned long rip_linear;
2281
2282         if (!printk_ratelimit())
2283                 return;
2284
2285         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2286
2287         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2288
2289         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2290                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2291 }
2292 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2293
2294 static struct x86_emulate_ops emulate_ops = {
2295         .read_std            = kvm_read_guest_virt,
2296         .read_emulated       = emulator_read_emulated,
2297         .write_emulated      = emulator_write_emulated,
2298         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
2299 };
2300
2301 static void cache_all_regs(struct kvm_vcpu *vcpu)
2302 {
2303         kvm_register_read(vcpu, VCPU_REGS_RAX);
2304         kvm_register_read(vcpu, VCPU_REGS_RSP);
2305         kvm_register_read(vcpu, VCPU_REGS_RIP);
2306         vcpu->arch.regs_dirty = ~0;
2307 }
2308
2309 int emulate_instruction(struct kvm_vcpu *vcpu,
2310                         struct kvm_run *run,
2311                         unsigned long cr2,
2312                         u16 error_code,
2313                         int emulation_type)
2314 {
2315         int r;
2316         struct decode_cache *c;
2317
2318         kvm_clear_exception_queue(vcpu);
2319         vcpu->arch.mmio_fault_cr2 = cr2;
2320         /*
2321          * TODO: fix x86_emulate.c to use guest_read/write_register
2322          * instead of direct ->regs accesses, can save hundred cycles
2323          * on Intel for instructions that don't read/change RSP, for
2324          * for example.
2325          */
2326         cache_all_regs(vcpu);
2327
2328         vcpu->mmio_is_write = 0;
2329         vcpu->arch.pio.string = 0;
2330
2331         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2332                 int cs_db, cs_l;
2333                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2334
2335                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2336                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2337                 vcpu->arch.emulate_ctxt.mode =
2338                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2339                         ? X86EMUL_MODE_REAL : cs_l
2340                         ? X86EMUL_MODE_PROT64 : cs_db
2341                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2342
2343                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2344
2345                 /* Reject the instructions other than VMCALL/VMMCALL when
2346                  * try to emulate invalid opcode */
2347                 c = &vcpu->arch.emulate_ctxt.decode;
2348                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2349                     (!(c->twobyte && c->b == 0x01 &&
2350                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2351                        c->modrm_mod == 3 && c->modrm_rm == 1)))
2352                         return EMULATE_FAIL;
2353
2354                 ++vcpu->stat.insn_emulation;
2355                 if (r)  {
2356                         ++vcpu->stat.insn_emulation_fail;
2357                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2358                                 return EMULATE_DONE;
2359                         return EMULATE_FAIL;
2360                 }
2361         }
2362
2363         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2364
2365         if (vcpu->arch.pio.string)
2366                 return EMULATE_DO_MMIO;
2367
2368         if ((r || vcpu->mmio_is_write) && run) {
2369                 run->exit_reason = KVM_EXIT_MMIO;
2370                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2371                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2372                 run->mmio.len = vcpu->mmio_size;
2373                 run->mmio.is_write = vcpu->mmio_is_write;
2374         }
2375
2376         if (r) {
2377                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2378                         return EMULATE_DONE;
2379                 if (!vcpu->mmio_needed) {
2380                         kvm_report_emulation_failure(vcpu, "mmio");
2381                         return EMULATE_FAIL;
2382                 }
2383                 return EMULATE_DO_MMIO;
2384         }
2385
2386         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2387
2388         if (vcpu->mmio_is_write) {
2389                 vcpu->mmio_needed = 0;
2390                 return EMULATE_DO_MMIO;
2391         }
2392
2393         return EMULATE_DONE;
2394 }
2395 EXPORT_SYMBOL_GPL(emulate_instruction);
2396
2397 static int pio_copy_data(struct kvm_vcpu *vcpu)
2398 {
2399         void *p = vcpu->arch.pio_data;
2400         gva_t q = vcpu->arch.pio.guest_gva;
2401         unsigned bytes;
2402         int ret;
2403
2404         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2405         if (vcpu->arch.pio.in)
2406                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2407         else
2408                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2409         return ret;
2410 }
2411
2412 int complete_pio(struct kvm_vcpu *vcpu)
2413 {
2414         struct kvm_pio_request *io = &vcpu->arch.pio;
2415         long delta;
2416         int r;
2417         unsigned long val;
2418
2419         if (!io->string) {
2420                 if (io->in) {
2421                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2422                         memcpy(&val, vcpu->arch.pio_data, io->size);
2423                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2424                 }
2425         } else {
2426                 if (io->in) {
2427                         r = pio_copy_data(vcpu);
2428                         if (r)
2429                                 return r;
2430                 }
2431
2432                 delta = 1;
2433                 if (io->rep) {
2434                         delta *= io->cur_count;
2435                         /*
2436                          * The size of the register should really depend on
2437                          * current address size.
2438                          */
2439                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2440                         val -= delta;
2441                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2442                 }
2443                 if (io->down)
2444                         delta = -delta;
2445                 delta *= io->size;
2446                 if (io->in) {
2447                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2448                         val += delta;
2449                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2450                 } else {
2451                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2452                         val += delta;
2453                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2454                 }
2455         }
2456
2457         io->count -= io->cur_count;
2458         io->cur_count = 0;
2459
2460         return 0;
2461 }
2462
2463 static void kernel_pio(struct kvm_io_device *pio_dev,
2464                        struct kvm_vcpu *vcpu,
2465                        void *pd)
2466 {
2467         /* TODO: String I/O for in kernel device */
2468
2469         mutex_lock(&vcpu->kvm->lock);
2470         if (vcpu->arch.pio.in)
2471                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2472                                   vcpu->arch.pio.size,
2473                                   pd);
2474         else
2475                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2476                                    vcpu->arch.pio.size,
2477                                    pd);
2478         mutex_unlock(&vcpu->kvm->lock);
2479 }
2480
2481 static void pio_string_write(struct kvm_io_device *pio_dev,
2482                              struct kvm_vcpu *vcpu)
2483 {
2484         struct kvm_pio_request *io = &vcpu->arch.pio;
2485         void *pd = vcpu->arch.pio_data;
2486         int i;
2487
2488         mutex_lock(&vcpu->kvm->lock);
2489         for (i = 0; i < io->cur_count; i++) {
2490                 kvm_iodevice_write(pio_dev, io->port,
2491                                    io->size,
2492                                    pd);
2493                 pd += io->size;
2494         }
2495         mutex_unlock(&vcpu->kvm->lock);
2496 }
2497
2498 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2499                                                gpa_t addr, int len,
2500                                                int is_write)
2501 {
2502         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2503 }
2504
2505 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2506                   int size, unsigned port)
2507 {
2508         struct kvm_io_device *pio_dev;
2509         unsigned long val;
2510
2511         vcpu->run->exit_reason = KVM_EXIT_IO;
2512         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2513         vcpu->run->io.size = vcpu->arch.pio.size = size;
2514         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2515         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2516         vcpu->run->io.port = vcpu->arch.pio.port = port;
2517         vcpu->arch.pio.in = in;
2518         vcpu->arch.pio.string = 0;
2519         vcpu->arch.pio.down = 0;
2520         vcpu->arch.pio.rep = 0;
2521
2522         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2523                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2524                             handler);
2525         else
2526                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2527                             handler);
2528
2529         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2530         memcpy(vcpu->arch.pio_data, &val, 4);
2531
2532         pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2533         if (pio_dev) {
2534                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2535                 complete_pio(vcpu);
2536                 return 1;
2537         }
2538         return 0;
2539 }
2540 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2541
2542 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2543                   int size, unsigned long count, int down,
2544                   gva_t address, int rep, unsigned port)
2545 {
2546         unsigned now, in_page;
2547         int ret = 0;
2548         struct kvm_io_device *pio_dev;
2549
2550         vcpu->run->exit_reason = KVM_EXIT_IO;
2551         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2552         vcpu->run->io.size = vcpu->arch.pio.size = size;
2553         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2554         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2555         vcpu->run->io.port = vcpu->arch.pio.port = port;
2556         vcpu->arch.pio.in = in;
2557         vcpu->arch.pio.string = 1;
2558         vcpu->arch.pio.down = down;
2559         vcpu->arch.pio.rep = rep;
2560
2561         if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2562                 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2563                             handler);
2564         else
2565                 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2566                             handler);
2567
2568         if (!count) {
2569                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2570                 return 1;
2571         }
2572
2573         if (!down)
2574                 in_page = PAGE_SIZE - offset_in_page(address);
2575         else
2576                 in_page = offset_in_page(address) + size;
2577         now = min(count, (unsigned long)in_page / size);
2578         if (!now)
2579                 now = 1;
2580         if (down) {
2581                 /*
2582                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2583                  */
2584                 pr_unimpl(vcpu, "guest string pio down\n");
2585                 kvm_inject_gp(vcpu, 0);
2586                 return 1;
2587         }
2588         vcpu->run->io.count = now;
2589         vcpu->arch.pio.cur_count = now;
2590
2591         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2592                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2593
2594         vcpu->arch.pio.guest_gva = address;
2595
2596         pio_dev = vcpu_find_pio_dev(vcpu, port,
2597                                     vcpu->arch.pio.cur_count,
2598                                     !vcpu->arch.pio.in);
2599         if (!vcpu->arch.pio.in) {
2600                 /* string PIO write */
2601                 ret = pio_copy_data(vcpu);
2602                 if (ret == X86EMUL_PROPAGATE_FAULT) {
2603                         kvm_inject_gp(vcpu, 0);
2604                         return 1;
2605                 }
2606                 if (ret == 0 && pio_dev) {
2607                         pio_string_write(pio_dev, vcpu);
2608                         complete_pio(vcpu);
2609                         if (vcpu->arch.pio.count == 0)
2610                                 ret = 1;
2611                 }
2612         } else if (pio_dev)
2613                 pr_unimpl(vcpu, "no string pio read support yet, "
2614                        "port %x size %d count %ld\n",
2615                         port, size, count);
2616
2617         return ret;
2618 }
2619 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2620
2621 int kvm_arch_init(void *opaque)
2622 {
2623         int r;
2624         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2625
2626         if (kvm_x86_ops) {
2627                 printk(KERN_ERR "kvm: already loaded the other module\n");
2628                 r = -EEXIST;
2629                 goto out;
2630         }
2631
2632         if (!ops->cpu_has_kvm_support()) {
2633                 printk(KERN_ERR "kvm: no hardware support\n");
2634                 r = -EOPNOTSUPP;
2635                 goto out;
2636         }
2637         if (ops->disabled_by_bios()) {
2638                 printk(KERN_ERR "kvm: disabled by bios\n");
2639                 r = -EOPNOTSUPP;
2640                 goto out;
2641         }
2642
2643         r = kvm_mmu_module_init();
2644         if (r)
2645                 goto out;
2646
2647         kvm_init_msr_list();
2648
2649         kvm_x86_ops = ops;
2650         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2651         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2652         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2653                         PT_DIRTY_MASK, PT64_NX_MASK, 0, 0);
2654         return 0;
2655
2656 out:
2657         return r;
2658 }
2659
2660 void kvm_arch_exit(void)
2661 {
2662         kvm_x86_ops = NULL;
2663         kvm_mmu_module_exit();
2664 }
2665
2666 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2667 {
2668         ++vcpu->stat.halt_exits;
2669         KVMTRACE_0D(HLT, vcpu, handler);
2670         if (irqchip_in_kernel(vcpu->kvm)) {
2671                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2672                 return 1;
2673         } else {
2674                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2675                 return 0;
2676         }
2677 }
2678 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2679
2680 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2681                            unsigned long a1)
2682 {
2683         if (is_long_mode(vcpu))
2684                 return a0;
2685         else
2686                 return a0 | ((gpa_t)a1 << 32);
2687 }
2688
2689 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2690 {
2691         unsigned long nr, a0, a1, a2, a3, ret;
2692         int r = 1;
2693
2694         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2695         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2696         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2697         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2698         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2699
2700         KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2701
2702         if (!is_long_mode(vcpu)) {
2703                 nr &= 0xFFFFFFFF;
2704                 a0 &= 0xFFFFFFFF;
2705                 a1 &= 0xFFFFFFFF;
2706                 a2 &= 0xFFFFFFFF;
2707                 a3 &= 0xFFFFFFFF;
2708         }
2709
2710         switch (nr) {
2711         case KVM_HC_VAPIC_POLL_IRQ:
2712                 ret = 0;
2713                 break;
2714         case KVM_HC_MMU_OP:
2715                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2716                 break;
2717         default:
2718                 ret = -KVM_ENOSYS;
2719                 break;
2720         }
2721         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2722         ++vcpu->stat.hypercalls;
2723         return r;
2724 }
2725 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2726
2727 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2728 {
2729         char instruction[3];
2730         int ret = 0;
2731         unsigned long rip = kvm_rip_read(vcpu);
2732
2733
2734         /*
2735          * Blow out the MMU to ensure that no other VCPU has an active mapping
2736          * to ensure that the updated hypercall appears atomically across all
2737          * VCPUs.
2738          */
2739         kvm_mmu_zap_all(vcpu->kvm);
2740
2741         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2742         if (emulator_write_emulated(rip, instruction, 3, vcpu)
2743             != X86EMUL_CONTINUE)
2744                 ret = -EFAULT;
2745
2746         return ret;
2747 }
2748
2749 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2750 {
2751         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2752 }
2753
2754 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2755 {
2756         struct descriptor_table dt = { limit, base };
2757
2758         kvm_x86_ops->set_gdt(vcpu, &dt);
2759 }
2760
2761 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2762 {
2763         struct descriptor_table dt = { limit, base };
2764
2765         kvm_x86_ops->set_idt(vcpu, &dt);
2766 }
2767
2768 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2769                    unsigned long *rflags)
2770 {
2771         kvm_lmsw(vcpu, msw);
2772         *rflags = kvm_x86_ops->get_rflags(vcpu);
2773 }
2774
2775 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2776 {
2777         unsigned long value;
2778
2779         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2780         switch (cr) {
2781         case 0:
2782                 value = vcpu->arch.cr0;
2783                 break;
2784         case 2:
2785                 value = vcpu->arch.cr2;
2786                 break;
2787         case 3:
2788                 value = vcpu->arch.cr3;
2789                 break;
2790         case 4:
2791                 value = vcpu->arch.cr4;
2792                 break;
2793         case 8:
2794                 value = kvm_get_cr8(vcpu);
2795                 break;
2796         default:
2797                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2798                 return 0;
2799         }
2800         KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2801                     (u32)((u64)value >> 32), handler);
2802
2803         return value;
2804 }
2805
2806 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2807                      unsigned long *rflags)
2808 {
2809         KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2810                     (u32)((u64)val >> 32), handler);
2811
2812         switch (cr) {
2813         case 0:
2814                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2815                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2816                 break;
2817         case 2:
2818                 vcpu->arch.cr2 = val;
2819                 break;
2820         case 3:
2821                 kvm_set_cr3(vcpu, val);
2822                 break;
2823         case 4:
2824                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2825                 break;
2826         case 8:
2827                 kvm_set_cr8(vcpu, val & 0xfUL);
2828                 break;
2829         default:
2830                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2831         }
2832 }
2833
2834 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2835 {
2836         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2837         int j, nent = vcpu->arch.cpuid_nent;
2838
2839         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2840         /* when no next entry is found, the current entry[i] is reselected */
2841         for (j = i + 1; ; j = (j + 1) % nent) {
2842                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2843                 if (ej->function == e->function) {
2844                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2845                         return j;
2846                 }
2847         }
2848         return 0; /* silence gcc, even though control never reaches here */
2849 }
2850
2851 /* find an entry with matching function, matching index (if needed), and that
2852  * should be read next (if it's stateful) */
2853 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2854         u32 function, u32 index)
2855 {
2856         if (e->function != function)
2857                 return 0;
2858         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2859                 return 0;
2860         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2861                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2862                 return 0;
2863         return 1;
2864 }
2865
2866 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
2867                                               u32 function, u32 index)
2868 {
2869         int i;
2870         struct kvm_cpuid_entry2 *best = NULL;
2871
2872         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2873                 struct kvm_cpuid_entry2 *e;
2874
2875                 e = &vcpu->arch.cpuid_entries[i];
2876                 if (is_matching_cpuid_entry(e, function, index)) {
2877                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2878                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2879                         best = e;
2880                         break;
2881                 }
2882                 /*
2883                  * Both basic or both extended?
2884                  */
2885                 if (((e->function ^ function) & 0x80000000) == 0)
2886                         if (!best || e->function > best->function)
2887                                 best = e;
2888         }
2889
2890         return best;
2891 }
2892
2893 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2894 {
2895         u32 function, index;
2896         struct kvm_cpuid_entry2 *best;
2897
2898         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
2899         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
2900         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
2901         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
2902         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
2903         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
2904         best = kvm_find_cpuid_entry(vcpu, function, index);
2905         if (best) {
2906                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
2907                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
2908                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
2909                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
2910         }
2911         kvm_x86_ops->skip_emulated_instruction(vcpu);
2912         KVMTRACE_5D(CPUID, vcpu, function,
2913                     (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
2914                     (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
2915                     (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
2916                     (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
2917 }
2918 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2919
2920 /*
2921  * Check if userspace requested an interrupt window, and that the
2922  * interrupt window is open.
2923  *
2924  * No need to exit to userspace if we already have an interrupt queued.
2925  */
2926 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2927                                           struct kvm_run *kvm_run)
2928 {
2929         return (!vcpu->arch.irq_summary &&
2930                 kvm_run->request_interrupt_window &&
2931                 vcpu->arch.interrupt_window_open &&
2932                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2933 }
2934
2935 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2936                               struct kvm_run *kvm_run)
2937 {
2938         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2939         kvm_run->cr8 = kvm_get_cr8(vcpu);
2940         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2941         if (irqchip_in_kernel(vcpu->kvm))
2942                 kvm_run->ready_for_interrupt_injection = 1;
2943         else
2944                 kvm_run->ready_for_interrupt_injection =
2945                                         (vcpu->arch.interrupt_window_open &&
2946                                          vcpu->arch.irq_summary == 0);
2947 }
2948
2949 static void vapic_enter(struct kvm_vcpu *vcpu)
2950 {
2951         struct kvm_lapic *apic = vcpu->arch.apic;
2952         struct page *page;
2953
2954         if (!apic || !apic->vapic_addr)
2955                 return;
2956
2957         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2958
2959         vcpu->arch.apic->vapic_page = page;
2960 }
2961
2962 static void vapic_exit(struct kvm_vcpu *vcpu)
2963 {
2964         struct kvm_lapic *apic = vcpu->arch.apic;
2965
2966         if (!apic || !apic->vapic_addr)
2967                 return;
2968
2969         down_read(&vcpu->kvm->slots_lock);
2970         kvm_release_page_dirty(apic->vapic_page);
2971         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2972         up_read(&vcpu->kvm->slots_lock);
2973 }
2974
2975 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2976 {
2977         int r;
2978
2979         if (vcpu->requests)
2980                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2981                         kvm_mmu_unload(vcpu);
2982
2983         r = kvm_mmu_reload(vcpu);
2984         if (unlikely(r))
2985                 goto out;
2986
2987         if (vcpu->requests) {
2988                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2989                         __kvm_migrate_timers(vcpu);
2990                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
2991                         kvm_mmu_sync_roots(vcpu);
2992                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2993                         kvm_x86_ops->tlb_flush(vcpu);
2994                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2995                                        &vcpu->requests)) {
2996                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2997                         r = 0;
2998                         goto out;
2999                 }
3000                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3001                         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3002                         r = 0;
3003                         goto out;
3004                 }
3005         }
3006
3007         clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3008         kvm_inject_pending_timer_irqs(vcpu);
3009
3010         preempt_disable();
3011
3012         kvm_x86_ops->prepare_guest_switch(vcpu);
3013         kvm_load_guest_fpu(vcpu);
3014
3015         local_irq_disable();
3016
3017         if (vcpu->requests || need_resched() || signal_pending(current)) {
3018                 local_irq_enable();
3019                 preempt_enable();
3020                 r = 1;
3021                 goto out;
3022         }
3023
3024         vcpu->guest_mode = 1;
3025         /*
3026          * Make sure that guest_mode assignment won't happen after
3027          * testing the pending IRQ vector bitmap.
3028          */
3029         smp_wmb();
3030
3031         if (vcpu->arch.exception.pending)
3032                 __queue_exception(vcpu);
3033         else if (irqchip_in_kernel(vcpu->kvm))
3034                 kvm_x86_ops->inject_pending_irq(vcpu);
3035         else
3036                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3037
3038         kvm_lapic_sync_to_vapic(vcpu);
3039
3040         up_read(&vcpu->kvm->slots_lock);
3041
3042         kvm_guest_enter();
3043
3044         get_debugreg(vcpu->arch.host_dr6, 6);
3045         get_debugreg(vcpu->arch.host_dr7, 7);
3046         if (unlikely(vcpu->arch.switch_db_regs)) {
3047                 get_debugreg(vcpu->arch.host_db[0], 0);
3048                 get_debugreg(vcpu->arch.host_db[1], 1);
3049                 get_debugreg(vcpu->arch.host_db[2], 2);
3050                 get_debugreg(vcpu->arch.host_db[3], 3);
3051
3052                 set_debugreg(0, 7);
3053                 set_debugreg(vcpu->arch.eff_db[0], 0);
3054                 set_debugreg(vcpu->arch.eff_db[1], 1);
3055                 set_debugreg(vcpu->arch.eff_db[2], 2);
3056                 set_debugreg(vcpu->arch.eff_db[3], 3);
3057         }
3058
3059         KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3060         kvm_x86_ops->run(vcpu, kvm_run);
3061
3062         if (unlikely(vcpu->arch.switch_db_regs)) {
3063                 set_debugreg(0, 7);
3064                 set_debugreg(vcpu->arch.host_db[0], 0);
3065                 set_debugreg(vcpu->arch.host_db[1], 1);
3066                 set_debugreg(vcpu->arch.host_db[2], 2);
3067                 set_debugreg(vcpu->arch.host_db[3], 3);
3068         }
3069         set_debugreg(vcpu->arch.host_dr6, 6);
3070         set_debugreg(vcpu->arch.host_dr7, 7);
3071
3072         vcpu->guest_mode = 0;
3073         local_irq_enable();
3074
3075         ++vcpu->stat.exits;
3076
3077         /*
3078          * We must have an instruction between local_irq_enable() and
3079          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3080          * the interrupt shadow.  The stat.exits increment will do nicely.
3081          * But we need to prevent reordering, hence this barrier():
3082          */
3083         barrier();
3084
3085         kvm_guest_exit();
3086
3087         preempt_enable();
3088
3089         down_read(&vcpu->kvm->slots_lock);
3090
3091         /*
3092          * Profile KVM exit RIPs:
3093          */
3094         if (unlikely(prof_on == KVM_PROFILING)) {
3095                 unsigned long rip = kvm_rip_read(vcpu);
3096                 profile_hit(KVM_PROFILING, (void *)rip);
3097         }
3098
3099         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3100                 vcpu->arch.exception.pending = false;
3101
3102         kvm_lapic_sync_from_vapic(vcpu);
3103
3104         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3105 out:
3106         return r;
3107 }
3108
3109 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3110 {
3111         int r;
3112
3113         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3114                 pr_debug("vcpu %d received sipi with vector # %x\n",
3115                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3116                 kvm_lapic_reset(vcpu);
3117                 r = kvm_arch_vcpu_reset(vcpu);
3118                 if (r)
3119                         return r;
3120                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3121         }
3122
3123         down_read(&vcpu->kvm->slots_lock);
3124         vapic_enter(vcpu);
3125
3126         r = 1;
3127         while (r > 0) {
3128                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3129                         r = vcpu_enter_guest(vcpu, kvm_run);
3130                 else {
3131                         up_read(&vcpu->kvm->slots_lock);
3132                         kvm_vcpu_block(vcpu);
3133                         down_read(&vcpu->kvm->slots_lock);
3134                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3135                                 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3136                                         vcpu->arch.mp_state =
3137                                                         KVM_MP_STATE_RUNNABLE;
3138                         if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3139                                 r = -EINTR;
3140                 }
3141
3142                 if (r > 0) {
3143                         if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3144                                 r = -EINTR;
3145                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3146                                 ++vcpu->stat.request_irq_exits;
3147                         }
3148                         if (signal_pending(current)) {
3149                                 r = -EINTR;
3150                                 kvm_run->exit_reason = KVM_EXIT_INTR;
3151                                 ++vcpu->stat.signal_exits;
3152                         }
3153                         if (need_resched()) {
3154                                 up_read(&vcpu->kvm->slots_lock);
3155                                 kvm_resched(vcpu);
3156                                 down_read(&vcpu->kvm->slots_lock);
3157                         }
3158                 }
3159         }
3160
3161         up_read(&vcpu->kvm->slots_lock);
3162         post_kvm_run_save(vcpu, kvm_run);
3163
3164         vapic_exit(vcpu);
3165
3166         return r;
3167 }
3168
3169 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3170 {
3171         int r;
3172         sigset_t sigsaved;
3173
3174         vcpu_load(vcpu);
3175
3176         if (vcpu->sigset_active)
3177                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3178
3179         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3180                 kvm_vcpu_block(vcpu);
3181                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3182                 r = -EAGAIN;
3183                 goto out;
3184         }
3185
3186         /* re-sync apic's tpr */
3187         if (!irqchip_in_kernel(vcpu->kvm))
3188                 kvm_set_cr8(vcpu, kvm_run->cr8);
3189
3190         if (vcpu->arch.pio.cur_count) {
3191                 r = complete_pio(vcpu);
3192                 if (r)
3193                         goto out;
3194         }
3195 #if CONFIG_HAS_IOMEM
3196         if (vcpu->mmio_needed) {
3197                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3198                 vcpu->mmio_read_completed = 1;
3199                 vcpu->mmio_needed = 0;
3200
3201                 down_read(&vcpu->kvm->slots_lock);
3202                 r = emulate_instruction(vcpu, kvm_run,
3203                                         vcpu->arch.mmio_fault_cr2, 0,
3204                                         EMULTYPE_NO_DECODE);
3205                 up_read(&vcpu->kvm->slots_lock);
3206                 if (r == EMULATE_DO_MMIO) {
3207                         /*
3208                          * Read-modify-write.  Back to userspace.
3209                          */
3210                         r = 0;
3211                         goto out;
3212                 }
3213         }
3214 #endif
3215         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3216                 kvm_register_write(vcpu, VCPU_REGS_RAX,
3217                                      kvm_run->hypercall.ret);
3218
3219         r = __vcpu_run(vcpu, kvm_run);
3220
3221 out:
3222         if (vcpu->sigset_active)
3223                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3224
3225         vcpu_put(vcpu);
3226         return r;
3227 }
3228
3229 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3230 {
3231         vcpu_load(vcpu);
3232
3233         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3234         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3235         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3236         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3237         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3238         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3239         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3240         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3241 #ifdef CONFIG_X86_64
3242         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3243         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3244         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3245         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3246         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3247         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3248         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3249         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3250 #endif
3251
3252         regs->rip = kvm_rip_read(vcpu);
3253         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3254
3255         /*
3256          * Don't leak debug flags in case they were set for guest debugging
3257          */
3258         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3259                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3260
3261         vcpu_put(vcpu);
3262
3263         return 0;
3264 }
3265
3266 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3267 {
3268         vcpu_load(vcpu);
3269
3270         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3271         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3272         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3273         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3274         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3275         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3276         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3277         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3278 #ifdef CONFIG_X86_64
3279         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3280         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3281         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3282         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3283         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3284         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3285         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3286         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3287
3288 #endif
3289
3290         kvm_rip_write(vcpu, regs->rip);
3291         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3292
3293
3294         vcpu->arch.exception.pending = false;
3295
3296         vcpu_put(vcpu);
3297
3298         return 0;
3299 }
3300
3301 void kvm_get_segment(struct kvm_vcpu *vcpu,
3302                      struct kvm_segment *var, int seg)
3303 {
3304         kvm_x86_ops->get_segment(vcpu, var, seg);
3305 }
3306
3307 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3308 {
3309         struct kvm_segment cs;
3310
3311         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3312         *db = cs.db;
3313         *l = cs.l;
3314 }
3315 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3316
3317 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3318                                   struct kvm_sregs *sregs)
3319 {
3320         struct descriptor_table dt;
3321         int pending_vec;
3322
3323         vcpu_load(vcpu);
3324
3325         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3326         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3327         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3328         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3329         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3330         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3331
3332         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3333         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3334
3335         kvm_x86_ops->get_idt(vcpu, &dt);
3336         sregs->idt.limit = dt.limit;
3337         sregs->idt.base = dt.base;
3338         kvm_x86_ops->get_gdt(vcpu, &dt);
3339         sregs->gdt.limit = dt.limit;
3340         sregs->gdt.base = dt.base;
3341
3342         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3343         sregs->cr0 = vcpu->arch.cr0;
3344         sregs->cr2 = vcpu->arch.cr2;
3345         sregs->cr3 = vcpu->arch.cr3;
3346         sregs->cr4 = vcpu->arch.cr4;
3347         sregs->cr8 = kvm_get_cr8(vcpu);
3348         sregs->efer = vcpu->arch.shadow_efer;
3349         sregs->apic_base = kvm_get_apic_base(vcpu);
3350
3351         if (irqchip_in_kernel(vcpu->kvm)) {
3352                 memset(sregs->interrupt_bitmap, 0,
3353                        sizeof sregs->interrupt_bitmap);
3354                 pending_vec = kvm_x86_ops->get_irq(vcpu);
3355                 if (pending_vec >= 0)
3356                         set_bit(pending_vec,
3357                                 (unsigned long *)sregs->interrupt_bitmap);
3358         } else
3359                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3360                        sizeof sregs->interrupt_bitmap);
3361
3362         vcpu_put(vcpu);
3363
3364         return 0;
3365 }
3366
3367 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3368                                     struct kvm_mp_state *mp_state)
3369 {
3370         vcpu_load(vcpu);
3371         mp_state->mp_state = vcpu->arch.mp_state;
3372         vcpu_put(vcpu);
3373         return 0;
3374 }
3375
3376 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3377                                     struct kvm_mp_state *mp_state)
3378 {
3379         vcpu_load(vcpu);
3380         vcpu->arch.mp_state = mp_state->mp_state;
3381         vcpu_put(vcpu);
3382         return 0;
3383 }
3384
3385 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3386                         struct kvm_segment *var, int seg)
3387 {
3388         kvm_x86_ops->set_segment(vcpu, var, seg);
3389 }
3390
3391 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3392                                    struct kvm_segment *kvm_desct)
3393 {
3394         kvm_desct->base = seg_desc->base0;
3395         kvm_desct->base |= seg_desc->base1 << 16;
3396         kvm_desct->base |= seg_desc->base2 << 24;
3397         kvm_desct->limit = seg_desc->limit0;
3398         kvm_desct->limit |= seg_desc->limit << 16;
3399         if (seg_desc->g) {
3400                 kvm_desct->limit <<= 12;
3401                 kvm_desct->limit |= 0xfff;
3402         }
3403         kvm_desct->selector = selector;
3404         kvm_desct->type = seg_desc->type;
3405         kvm_desct->present = seg_desc->p;
3406         kvm_desct->dpl = seg_desc->dpl;
3407         kvm_desct->db = seg_desc->d;
3408         kvm_desct->s = seg_desc->s;
3409         kvm_desct->l = seg_desc->l;
3410         kvm_desct->g = seg_desc->g;
3411         kvm_desct->avl = seg_desc->avl;
3412         if (!selector)
3413                 kvm_desct->unusable = 1;
3414         else
3415                 kvm_desct->unusable = 0;
3416         kvm_desct->padding = 0;
3417 }
3418
3419 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3420                                           u16 selector,
3421                                           struct descriptor_table *dtable)
3422 {
3423         if (selector & 1 << 2) {
3424                 struct kvm_segment kvm_seg;
3425
3426                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3427
3428                 if (kvm_seg.unusable)
3429                         dtable->limit = 0;
3430                 else
3431                         dtable->limit = kvm_seg.limit;
3432                 dtable->base = kvm_seg.base;
3433         }
3434         else
3435                 kvm_x86_ops->get_gdt(vcpu, dtable);
3436 }
3437
3438 /* allowed just for 8 bytes segments */
3439 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3440                                          struct desc_struct *seg_desc)
3441 {
3442         gpa_t gpa;
3443         struct descriptor_table dtable;
3444         u16 index = selector >> 3;
3445
3446         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3447
3448         if (dtable.limit < index * 8 + 7) {
3449                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3450                 return 1;
3451         }
3452         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3453         gpa += index * 8;
3454         return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3455 }
3456
3457 /* allowed just for 8 bytes segments */
3458 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3459                                          struct desc_struct *seg_desc)
3460 {
3461         gpa_t gpa;
3462         struct descriptor_table dtable;
3463         u16 index = selector >> 3;
3464
3465         get_segment_descriptor_dtable(vcpu, selector, &dtable);
3466
3467         if (dtable.limit < index * 8 + 7)
3468                 return 1;
3469         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3470         gpa += index * 8;
3471         return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3472 }
3473
3474 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3475                              struct desc_struct *seg_desc)
3476 {
3477         u32 base_addr;
3478
3479         base_addr = seg_desc->base0;
3480         base_addr |= (seg_desc->base1 << 16);
3481         base_addr |= (seg_desc->base2 << 24);
3482
3483         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3484 }
3485
3486 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3487 {
3488         struct kvm_segment kvm_seg;
3489
3490         kvm_get_segment(vcpu, &kvm_seg, seg);
3491         return kvm_seg.selector;
3492 }
3493
3494 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3495                                                 u16 selector,
3496                                                 struct kvm_segment *kvm_seg)
3497 {
3498         struct desc_struct seg_desc;
3499
3500         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3501                 return 1;
3502         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3503         return 0;
3504 }
3505
3506 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3507 {
3508         struct kvm_segment segvar = {
3509                 .base = selector << 4,
3510                 .limit = 0xffff,
3511                 .selector = selector,
3512                 .type = 3,
3513                 .present = 1,
3514                 .dpl = 3,
3515                 .db = 0,
3516                 .s = 1,
3517                 .l = 0,
3518                 .g = 0,
3519                 .avl = 0,
3520                 .unusable = 0,
3521         };
3522         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3523         return 0;
3524 }
3525
3526 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3527                                 int type_bits, int seg)
3528 {
3529         struct kvm_segment kvm_seg;
3530
3531         if (!(vcpu->arch.cr0 & X86_CR0_PE))
3532                 return kvm_load_realmode_segment(vcpu, selector, seg);
3533         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3534                 return 1;
3535         kvm_seg.type |= type_bits;
3536
3537         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3538             seg != VCPU_SREG_LDTR)
3539                 if (!kvm_seg.s)
3540                         kvm_seg.unusable = 1;
3541
3542         kvm_set_segment(vcpu, &kvm_seg, seg);
3543         return 0;
3544 }
3545
3546 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3547                                 struct tss_segment_32 *tss)
3548 {
3549         tss->cr3 = vcpu->arch.cr3;
3550         tss->eip = kvm_rip_read(vcpu);
3551         tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3552         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3553         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3554         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3555         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3556         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3557         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3558         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3559         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3560         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3561         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3562         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3563         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3564         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3565         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3566         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3567         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3568 }
3569
3570 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3571                                   struct tss_segment_32 *tss)
3572 {
3573         kvm_set_cr3(vcpu, tss->cr3);
3574
3575         kvm_rip_write(vcpu, tss->eip);
3576         kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3577
3578         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3579         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3580         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3581         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3582         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3583         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3584         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3585         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3586
3587         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3588                 return 1;
3589
3590         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3591                 return 1;
3592
3593         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3594                 return 1;
3595
3596         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3597                 return 1;
3598
3599         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3600                 return 1;
3601
3602         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3603                 return 1;
3604
3605         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3606                 return 1;
3607         return 0;
3608 }
3609
3610 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3611                                 struct tss_segment_16 *tss)
3612 {
3613         tss->ip = kvm_rip_read(vcpu);
3614         tss->flag = kvm_x86_ops->get_rflags(vcpu);
3615         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3616         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3617         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3618         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3619         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3620         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3621         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3622         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3623
3624         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3625         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3626         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3627         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3628         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3629         tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3630 }
3631
3632 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3633                                  struct tss_segment_16 *tss)
3634 {
3635         kvm_rip_write(vcpu, tss->ip);
3636         kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3637         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3638         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3639         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3640         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3641         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3642         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3643         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3644         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3645
3646         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3647                 return 1;
3648
3649         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3650                 return 1;
3651
3652         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3653                 return 1;
3654
3655         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3656                 return 1;
3657
3658         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3659                 return 1;
3660         return 0;
3661 }
3662
3663 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3664                        u32 old_tss_base,
3665                        struct desc_struct *nseg_desc)
3666 {
3667         struct tss_segment_16 tss_segment_16;
3668         int ret = 0;
3669
3670         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3671                            sizeof tss_segment_16))
3672                 goto out;
3673
3674         save_state_to_tss16(vcpu, &tss_segment_16);
3675
3676         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3677                             sizeof tss_segment_16))
3678                 goto out;
3679
3680         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3681                            &tss_segment_16, sizeof tss_segment_16))
3682                 goto out;
3683
3684         if (load_state_from_tss16(vcpu, &tss_segment_16))
3685                 goto out;
3686
3687         ret = 1;
3688 out:
3689         return ret;
3690 }
3691
3692 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3693                        u32 old_tss_base,
3694                        struct desc_struct *nseg_desc)
3695 {
3696         struct tss_segment_32 tss_segment_32;
3697         int ret = 0;
3698
3699         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3700                            sizeof tss_segment_32))
3701                 goto out;
3702
3703         save_state_to_tss32(vcpu, &tss_segment_32);
3704
3705         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3706                             sizeof tss_segment_32))
3707                 goto out;
3708
3709         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3710                            &tss_segment_32, sizeof tss_segment_32))
3711                 goto out;
3712
3713         if (load_state_from_tss32(vcpu, &tss_segment_32))
3714                 goto out;
3715
3716         ret = 1;
3717 out:
3718         return ret;
3719 }
3720
3721 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3722 {
3723         struct kvm_segment tr_seg;
3724         struct desc_struct cseg_desc;
3725         struct desc_struct nseg_desc;
3726         int ret = 0;
3727         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3728         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3729
3730         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3731
3732         /* FIXME: Handle errors. Failure to read either TSS or their
3733          * descriptors should generate a pagefault.
3734          */
3735         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3736                 goto out;
3737
3738         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3739                 goto out;
3740
3741         if (reason != TASK_SWITCH_IRET) {
3742                 int cpl;
3743
3744                 cpl = kvm_x86_ops->get_cpl(vcpu);
3745                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3746                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3747                         return 1;
3748                 }
3749         }
3750
3751         if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3752                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3753                 return 1;
3754         }
3755
3756         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3757                 cseg_desc.type &= ~(1 << 1); //clear the B flag
3758                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3759         }
3760
3761         if (reason == TASK_SWITCH_IRET) {
3762                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3763                 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3764         }
3765
3766         kvm_x86_ops->skip_emulated_instruction(vcpu);
3767
3768         if (nseg_desc.type & 8)
3769                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3770                                          &nseg_desc);
3771         else
3772                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3773                                          &nseg_desc);
3774
3775         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3776                 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3777                 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3778         }
3779
3780         if (reason != TASK_SWITCH_IRET) {
3781                 nseg_desc.type |= (1 << 1);
3782                 save_guest_segment_descriptor(vcpu, tss_selector,
3783                                               &nseg_desc);
3784         }
3785
3786         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3787         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3788         tr_seg.type = 11;
3789         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3790 out:
3791         return ret;
3792 }
3793 EXPORT_SYMBOL_GPL(kvm_task_switch);
3794
3795 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3796                                   struct kvm_sregs *sregs)
3797 {
3798         int mmu_reset_needed = 0;
3799         int i, pending_vec, max_bits;
3800         struct descriptor_table dt;
3801
3802         vcpu_load(vcpu);
3803
3804         dt.limit = sregs->idt.limit;
3805         dt.base = sregs->idt.base;
3806         kvm_x86_ops->set_idt(vcpu, &dt);
3807         dt.limit = sregs->gdt.limit;
3808         dt.base = sregs->gdt.base;
3809         kvm_x86_ops->set_gdt(vcpu, &dt);
3810
3811         vcpu->arch.cr2 = sregs->cr2;
3812         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3813         vcpu->arch.cr3 = sregs->cr3;
3814
3815         kvm_set_cr8(vcpu, sregs->cr8);
3816
3817         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3818         kvm_x86_ops->set_efer(vcpu, sregs->efer);
3819         kvm_set_apic_base(vcpu, sregs->apic_base);
3820
3821         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3822
3823         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3824         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3825         vcpu->arch.cr0 = sregs->cr0;
3826
3827         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3828         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3829         if (!is_long_mode(vcpu) && is_pae(vcpu))
3830                 load_pdptrs(vcpu, vcpu->arch.cr3);
3831
3832         if (mmu_reset_needed)
3833                 kvm_mmu_reset_context(vcpu);
3834
3835         if (!irqchip_in_kernel(vcpu->kvm)) {
3836                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3837                        sizeof vcpu->arch.irq_pending);
3838                 vcpu->arch.irq_summary = 0;
3839                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3840                         if (vcpu->arch.irq_pending[i])
3841                                 __set_bit(i, &vcpu->arch.irq_summary);
3842         } else {
3843                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3844                 pending_vec = find_first_bit(
3845                         (const unsigned long *)sregs->interrupt_bitmap,
3846                         max_bits);
3847                 /* Only pending external irq is handled here */
3848                 if (pending_vec < max_bits) {
3849                         kvm_x86_ops->set_irq(vcpu, pending_vec);
3850                         pr_debug("Set back pending irq %d\n",
3851                                  pending_vec);
3852                 }
3853                 kvm_pic_clear_isr_ack(vcpu->kvm);
3854         }
3855
3856         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3857         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3858         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3859         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3860         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3861         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3862
3863         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3864         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3865
3866         /* Older userspace won't unhalt the vcpu on reset. */
3867         if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3868             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3869             !(vcpu->arch.cr0 & X86_CR0_PE))
3870                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3871
3872         vcpu_put(vcpu);
3873
3874         return 0;
3875 }
3876
3877 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
3878                                         struct kvm_guest_debug *dbg)
3879 {
3880         int i, r;
3881
3882         vcpu_load(vcpu);
3883
3884         if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
3885             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
3886                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
3887                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
3888                 vcpu->arch.switch_db_regs =
3889                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
3890         } else {
3891                 for (i = 0; i < KVM_NR_DB_REGS; i++)
3892                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
3893                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
3894         }
3895
3896         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3897
3898         if (dbg->control & KVM_GUESTDBG_INJECT_DB)
3899                 kvm_queue_exception(vcpu, DB_VECTOR);
3900         else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
3901                 kvm_queue_exception(vcpu, BP_VECTOR);
3902
3903         vcpu_put(vcpu);
3904
3905         return r;
3906 }
3907
3908 /*
3909  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
3910  * we have asm/x86/processor.h
3911  */
3912 struct fxsave {
3913         u16     cwd;
3914         u16     swd;
3915         u16     twd;
3916         u16     fop;
3917         u64     rip;
3918         u64     rdp;
3919         u32     mxcsr;
3920         u32     mxcsr_mask;
3921         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
3922 #ifdef CONFIG_X86_64
3923         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
3924 #else
3925         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
3926 #endif
3927 };
3928
3929 /*
3930  * Translate a guest virtual address to a guest physical address.
3931  */
3932 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3933                                     struct kvm_translation *tr)
3934 {
3935         unsigned long vaddr = tr->linear_address;
3936         gpa_t gpa;
3937
3938         vcpu_load(vcpu);
3939         down_read(&vcpu->kvm->slots_lock);
3940         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3941         up_read(&vcpu->kvm->slots_lock);
3942         tr->physical_address = gpa;
3943         tr->valid = gpa != UNMAPPED_GVA;
3944         tr->writeable = 1;
3945         tr->usermode = 0;
3946         vcpu_put(vcpu);
3947
3948         return 0;
3949 }
3950
3951 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3952 {
3953         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3954
3955         vcpu_load(vcpu);
3956
3957         memcpy(fpu->fpr, fxsave->st_space, 128);
3958         fpu->fcw = fxsave->cwd;
3959         fpu->fsw = fxsave->swd;
3960         fpu->ftwx = fxsave->twd;
3961         fpu->last_opcode = fxsave->fop;
3962         fpu->last_ip = fxsave->rip;
3963         fpu->last_dp = fxsave->rdp;
3964         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3965
3966         vcpu_put(vcpu);
3967
3968         return 0;
3969 }
3970
3971 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3972 {
3973         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3974
3975         vcpu_load(vcpu);
3976
3977         memcpy(fxsave->st_space, fpu->fpr, 128);
3978         fxsave->cwd = fpu->fcw;
3979         fxsave->swd = fpu->fsw;
3980         fxsave->twd = fpu->ftwx;
3981         fxsave->fop = fpu->last_opcode;
3982         fxsave->rip = fpu->last_ip;
3983         fxsave->rdp = fpu->last_dp;
3984         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3985
3986         vcpu_put(vcpu);
3987
3988         return 0;
3989 }
3990
3991 void fx_init(struct kvm_vcpu *vcpu)
3992 {
3993         unsigned after_mxcsr_mask;
3994
3995         /*
3996          * Touch the fpu the first time in non atomic context as if
3997          * this is the first fpu instruction the exception handler
3998          * will fire before the instruction returns and it'll have to
3999          * allocate ram with GFP_KERNEL.
4000          */
4001         if (!used_math())
4002                 kvm_fx_save(&vcpu->arch.host_fx_image);
4003
4004         /* Initialize guest FPU by resetting ours and saving into guest's */
4005         preempt_disable();
4006         kvm_fx_save(&vcpu->arch.host_fx_image);
4007         kvm_fx_finit();
4008         kvm_fx_save(&vcpu->arch.guest_fx_image);
4009         kvm_fx_restore(&vcpu->arch.host_fx_image);
4010         preempt_enable();
4011
4012         vcpu->arch.cr0 |= X86_CR0_ET;
4013         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4014         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4015         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4016                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4017 }
4018 EXPORT_SYMBOL_GPL(fx_init);
4019
4020 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4021 {
4022         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4023                 return;
4024
4025         vcpu->guest_fpu_loaded = 1;
4026         kvm_fx_save(&vcpu->arch.host_fx_image);
4027         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4028 }
4029 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4030
4031 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4032 {
4033         if (!vcpu->guest_fpu_loaded)
4034                 return;
4035
4036         vcpu->guest_fpu_loaded = 0;
4037         kvm_fx_save(&vcpu->arch.guest_fx_image);
4038         kvm_fx_restore(&vcpu->arch.host_fx_image);
4039         ++vcpu->stat.fpu_reload;
4040 }
4041 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4042
4043 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4044 {
4045         kvm_x86_ops->vcpu_free(vcpu);
4046 }
4047
4048 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4049                                                 unsigned int id)
4050 {
4051         return kvm_x86_ops->vcpu_create(kvm, id);
4052 }
4053
4054 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4055 {
4056         int r;
4057
4058         /* We do fxsave: this must be aligned. */
4059         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4060
4061         vcpu->arch.mtrr_state.have_fixed = 1;
4062         vcpu_load(vcpu);
4063         r = kvm_arch_vcpu_reset(vcpu);
4064         if (r == 0)
4065                 r = kvm_mmu_setup(vcpu);
4066         vcpu_put(vcpu);
4067         if (r < 0)
4068                 goto free_vcpu;
4069
4070         return 0;
4071 free_vcpu:
4072         kvm_x86_ops->vcpu_free(vcpu);
4073         return r;
4074 }
4075
4076 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4077 {
4078         vcpu_load(vcpu);
4079         kvm_mmu_unload(vcpu);
4080         vcpu_put(vcpu);
4081
4082         kvm_x86_ops->vcpu_free(vcpu);
4083 }
4084
4085 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4086 {
4087         vcpu->arch.nmi_pending = false;
4088         vcpu->arch.nmi_injected = false;
4089
4090         vcpu->arch.switch_db_regs = 0;
4091         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4092         vcpu->arch.dr6 = DR6_FIXED_1;
4093         vcpu->arch.dr7 = DR7_FIXED_1;
4094
4095         return kvm_x86_ops->vcpu_reset(vcpu);
4096 }
4097
4098 void kvm_arch_hardware_enable(void *garbage)
4099 {
4100         kvm_x86_ops->hardware_enable(garbage);
4101 }
4102
4103 void kvm_arch_hardware_disable(void *garbage)
4104 {
4105         kvm_x86_ops->hardware_disable(garbage);
4106 }
4107
4108 int kvm_arch_hardware_setup(void)
4109 {
4110         return kvm_x86_ops->hardware_setup();
4111 }
4112
4113 void kvm_arch_hardware_unsetup(void)
4114 {
4115         kvm_x86_ops->hardware_unsetup();
4116 }
4117
4118 void kvm_arch_check_processor_compat(void *rtn)
4119 {
4120         kvm_x86_ops->check_processor_compatibility(rtn);
4121 }
4122
4123 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4124 {
4125         struct page *page;
4126         struct kvm *kvm;
4127         int r;
4128
4129         BUG_ON(vcpu->kvm == NULL);
4130         kvm = vcpu->kvm;
4131
4132         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4133         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4134                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4135         else
4136                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4137
4138         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4139         if (!page) {
4140                 r = -ENOMEM;
4141                 goto fail;
4142         }
4143         vcpu->arch.pio_data = page_address(page);
4144
4145         r = kvm_mmu_create(vcpu);
4146         if (r < 0)
4147                 goto fail_free_pio_data;
4148
4149         if (irqchip_in_kernel(kvm)) {
4150                 r = kvm_create_lapic(vcpu);
4151                 if (r < 0)
4152                         goto fail_mmu_destroy;
4153         }
4154
4155         return 0;
4156
4157 fail_mmu_destroy:
4158         kvm_mmu_destroy(vcpu);
4159 fail_free_pio_data:
4160         free_page((unsigned long)vcpu->arch.pio_data);
4161 fail:
4162         return r;
4163 }
4164
4165 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4166 {
4167         kvm_free_lapic(vcpu);
4168         down_read(&vcpu->kvm->slots_lock);
4169         kvm_mmu_destroy(vcpu);
4170         up_read(&vcpu->kvm->slots_lock);
4171         free_page((unsigned long)vcpu->arch.pio_data);
4172 }
4173
4174 struct  kvm *kvm_arch_create_vm(void)
4175 {
4176         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4177
4178         if (!kvm)
4179                 return ERR_PTR(-ENOMEM);
4180
4181         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4182         INIT_LIST_HEAD(&kvm->arch.oos_global_pages);
4183         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4184
4185         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4186         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4187
4188         rdtscll(kvm->arch.vm_init_tsc);
4189
4190         return kvm;
4191 }
4192
4193 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4194 {
4195         vcpu_load(vcpu);
4196         kvm_mmu_unload(vcpu);
4197         vcpu_put(vcpu);
4198 }
4199
4200 static void kvm_free_vcpus(struct kvm *kvm)
4201 {
4202         unsigned int i;
4203
4204         /*
4205          * Unpin any mmu pages first.
4206          */
4207         for (i = 0; i < KVM_MAX_VCPUS; ++i)
4208                 if (kvm->vcpus[i])
4209                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4210         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4211                 if (kvm->vcpus[i]) {
4212                         kvm_arch_vcpu_free(kvm->vcpus[i]);
4213                         kvm->vcpus[i] = NULL;
4214                 }
4215         }
4216
4217 }
4218
4219 void kvm_arch_sync_events(struct kvm *kvm)
4220 {
4221         kvm_free_all_assigned_devices(kvm);
4222 }
4223
4224 void kvm_arch_destroy_vm(struct kvm *kvm)
4225 {
4226         kvm_iommu_unmap_guest(kvm);
4227         kvm_free_pit(kvm);
4228         kfree(kvm->arch.vpic);
4229         kfree(kvm->arch.vioapic);
4230         kvm_free_vcpus(kvm);
4231         kvm_free_physmem(kvm);
4232         if (kvm->arch.apic_access_page)
4233                 put_page(kvm->arch.apic_access_page);
4234         if (kvm->arch.ept_identity_pagetable)
4235                 put_page(kvm->arch.ept_identity_pagetable);
4236         kfree(kvm);
4237 }
4238
4239 int kvm_arch_set_memory_region(struct kvm *kvm,
4240                                 struct kvm_userspace_memory_region *mem,
4241                                 struct kvm_memory_slot old,
4242                                 int user_alloc)
4243 {
4244         int npages = mem->memory_size >> PAGE_SHIFT;
4245         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4246
4247         /*To keep backward compatibility with older userspace,
4248          *x86 needs to hanlde !user_alloc case.
4249          */
4250         if (!user_alloc) {
4251                 if (npages && !old.rmap) {
4252                         unsigned long userspace_addr;
4253
4254                         down_write(&current->mm->mmap_sem);
4255                         userspace_addr = do_mmap(NULL, 0,
4256                                                  npages * PAGE_SIZE,
4257                                                  PROT_READ | PROT_WRITE,
4258                                                  MAP_PRIVATE | MAP_ANONYMOUS,
4259                                                  0);
4260                         up_write(&current->mm->mmap_sem);
4261
4262                         if (IS_ERR((void *)userspace_addr))
4263                                 return PTR_ERR((void *)userspace_addr);
4264
4265                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
4266                         spin_lock(&kvm->mmu_lock);
4267                         memslot->userspace_addr = userspace_addr;
4268                         spin_unlock(&kvm->mmu_lock);
4269                 } else {
4270                         if (!old.user_alloc && old.rmap) {
4271                                 int ret;
4272
4273                                 down_write(&current->mm->mmap_sem);
4274                                 ret = do_munmap(current->mm, old.userspace_addr,
4275                                                 old.npages * PAGE_SIZE);
4276                                 up_write(&current->mm->mmap_sem);
4277                                 if (ret < 0)
4278                                         printk(KERN_WARNING
4279                                        "kvm_vm_ioctl_set_memory_region: "
4280                                        "failed to munmap memory\n");
4281                         }
4282                 }
4283         }
4284
4285         if (!kvm->arch.n_requested_mmu_pages) {
4286                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4287                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4288         }
4289
4290         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4291         kvm_flush_remote_tlbs(kvm);
4292
4293         return 0;
4294 }
4295
4296 void kvm_arch_flush_shadow(struct kvm *kvm)
4297 {
4298         kvm_mmu_zap_all(kvm);
4299 }
4300
4301 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4302 {
4303         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4304                || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4305                || vcpu->arch.nmi_pending;
4306 }
4307
4308 static void vcpu_kick_intr(void *info)
4309 {
4310 #ifdef DEBUG
4311         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4312         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4313 #endif
4314 }
4315
4316 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4317 {
4318         int ipi_pcpu = vcpu->cpu;
4319         int cpu = get_cpu();
4320
4321         if (waitqueue_active(&vcpu->wq)) {
4322                 wake_up_interruptible(&vcpu->wq);
4323                 ++vcpu->stat.halt_wakeup;
4324         }
4325         /*
4326          * We may be called synchronously with irqs disabled in guest mode,
4327          * So need not to call smp_call_function_single() in that case.
4328          */
4329         if (vcpu->guest_mode && vcpu->cpu != cpu)
4330                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
4331         put_cpu();
4332 }