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  *
   8  * Authors:
   9  *   Avi Kivity   <avi@qumranet.com>
  10  *   Yaniv Kamay  <yaniv@qumranet.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2.  See
  13  * the COPYING file in the top-level directory.
  14  *
  15  */
  16
  17 #include <linux/kvm_host.h>
  18 #include "segment_descriptor.h"
  19 #include "irq.h"
  20 #include "mmu.h"
  21
  22 #include <linux/kvm.h>
  23 #include <linux/fs.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/module.h>
  26 #include <linux/mman.h>
  27 #include <linux/highmem.h>
  28
  29 #include <asm/uaccess.h>
  30 #include <asm/msr.h>
  31
  32 #define MAX_IO_MSRS 256
  33 #define CR0_RESERVED_BITS                                               \
  34         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  35                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  36                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  37 #define CR4_RESERVED_BITS                                               \
  38         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  39                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  40                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  41                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  42
  43 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  44 /* EFER defaults:
  45  * - enable syscall per default because its emulated by KVM
  46  * - enable LME and LMA per default on 64 bit KVM
  47  */
  48 #ifdef CONFIG_X86_64
  49 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
  50 #else
  51 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
  52 #endif
  53
  54 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  55 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  56
  57 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
  58                                     struct kvm_cpuid_entry2 __user *entries);
  59
  60 struct kvm_x86_ops *kvm_x86_ops;
  61
  62 struct kvm_stats_debugfs_item debugfs_entries[] = {
  63         { "pf_fixed", VCPU_STAT(pf_fixed) },
  64         { "pf_guest", VCPU_STAT(pf_guest) },
  65         { "tlb_flush", VCPU_STAT(tlb_flush) },
  66         { "invlpg", VCPU_STAT(invlpg) },
  67         { "exits", VCPU_STAT(exits) },
  68         { "io_exits", VCPU_STAT(io_exits) },
  69         { "mmio_exits", VCPU_STAT(mmio_exits) },
  70         { "signal_exits", VCPU_STAT(signal_exits) },
  71         { "irq_window", VCPU_STAT(irq_window_exits) },
  72         { "halt_exits", VCPU_STAT(halt_exits) },
  73         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  74         { "request_irq", VCPU_STAT(request_irq_exits) },
  75         { "irq_exits", VCPU_STAT(irq_exits) },
  76         { "host_state_reload", VCPU_STAT(host_state_reload) },
  77         { "efer_reload", VCPU_STAT(efer_reload) },
  78         { "fpu_reload", VCPU_STAT(fpu_reload) },
  79         { "insn_emulation", VCPU_STAT(insn_emulation) },
  80         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  81         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
  82         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
  83         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
  84         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
  85         { "mmu_flooded", VM_STAT(mmu_flooded) },
  86         { "mmu_recycled", VM_STAT(mmu_recycled) },
  87         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
  88         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
  89         { NULL }
  90 };
  91
  92
  93 unsigned long segment_base(u16 selector)
  94 {
  95         struct descriptor_table gdt;
  96         struct segment_descriptor *d;
  97         unsigned long table_base;
  98         unsigned long v;
  99
 100         if (selector == 0)
 101                 return 0;
 102
 103         asm("sgdt %0" : "=m"(gdt));
 104         table_base = gdt.base;
 105
 106         if (selector & 4) {           /* from ldt */
 107                 u16 ldt_selector;
 108
 109                 asm("sldt %0" : "=g"(ldt_selector));
 110                 table_base = segment_base(ldt_selector);
 111         }
 112         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 113         v = d->base_low | ((unsigned long)d->base_mid << 16) |
 114                 ((unsigned long)d->base_high << 24);
 115 #ifdef CONFIG_X86_64
 116         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 117                 v |= ((unsigned long) \
 118                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
 119 #endif
 120         return v;
 121 }
 122 EXPORT_SYMBOL_GPL(segment_base);
 123
 124 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 125 {
 126         if (irqchip_in_kernel(vcpu->kvm))
 127                 return vcpu->arch.apic_base;
 128         else
 129                 return vcpu->arch.apic_base;
 130 }
 131 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 132
 133 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 134 {
 135         /* TODO: reserve bits check */
 136         if (irqchip_in_kernel(vcpu->kvm))
 137                 kvm_lapic_set_base(vcpu, data);
 138         else
 139                 vcpu->arch.apic_base = data;
 140 }
 141 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 142
 143 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 144 {
 145         WARN_ON(vcpu->arch.exception.pending);
 146         vcpu->arch.exception.pending = true;
 147         vcpu->arch.exception.has_error_code = false;
 148         vcpu->arch.exception.nr = nr;
 149 }
 150 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 151
 152 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 153                            u32 error_code)
 154 {
 155         ++vcpu->stat.pf_guest;
 156         if (vcpu->arch.exception.pending && vcpu->arch.exception.nr == PF_VECTOR) {
 157                 printk(KERN_DEBUG "kvm: inject_page_fault:"
 158                        " double fault 0x%lx\n", addr);
 159                 vcpu->arch.exception.nr = DF_VECTOR;
 160                 vcpu->arch.exception.error_code = 0;
 161                 return;
 162         }
 163         vcpu->arch.cr2 = addr;
 164         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 165 }
 166
 167 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 168 {
 169         WARN_ON(vcpu->arch.exception.pending);
 170         vcpu->arch.exception.pending = true;
 171         vcpu->arch.exception.has_error_code = true;
 172         vcpu->arch.exception.nr = nr;
 173         vcpu->arch.exception.error_code = error_code;
 174 }
 175 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 176
 177 static void __queue_exception(struct kvm_vcpu *vcpu)
 178 {
 179         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 180                                      vcpu->arch.exception.has_error_code,
 181                                      vcpu->arch.exception.error_code);
 182 }
 183
 184 /*
 185  * Load the pae pdptrs.  Return true is they are all valid.
 186  */
 187 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 188 {
 189         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 190         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 191         int i;
 192         int ret;
 193         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 194
 195         down_read(&vcpu->kvm->slots_lock);
 196         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 197                                   offset * sizeof(u64), sizeof(pdpte));
 198         if (ret < 0) {
 199                 ret = 0;
 200                 goto out;
 201         }
 202         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 203                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 204                         ret = 0;
 205                         goto out;
 206                 }
 207         }
 208         ret = 1;
 209
 210         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 211 out:
 212         up_read(&vcpu->kvm->slots_lock);
 213
 214         return ret;
 215 }
 216 EXPORT_SYMBOL_GPL(load_pdptrs);
 217
 218 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 219 {
 220         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 221         bool changed = true;
 222         int r;
 223
 224         if (is_long_mode(vcpu) || !is_pae(vcpu))
 225                 return false;
 226
 227         down_read(&vcpu->kvm->slots_lock);
 228         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 229         if (r < 0)
 230                 goto out;
 231         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 232 out:
 233         up_read(&vcpu->kvm->slots_lock);
 234
 235         return changed;
 236 }
 237
 238 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 239 {
 240         if (cr0 & CR0_RESERVED_BITS) {
 241                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 242                        cr0, vcpu->arch.cr0);
 243                 kvm_inject_gp(vcpu, 0);
 244                 return;
 245         }
 246
 247         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 248                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 249                 kvm_inject_gp(vcpu, 0);
 250                 return;
 251         }
 252
 253         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 254                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 255                        "and a clear PE flag\n");
 256                 kvm_inject_gp(vcpu, 0);
 257                 return;
 258         }
 259
 260         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 261 #ifdef CONFIG_X86_64
 262                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 263                         int cs_db, cs_l;
 264
 265                         if (!is_pae(vcpu)) {
 266                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 267                                        "in long mode while PAE is disabled\n");
 268                                 kvm_inject_gp(vcpu, 0);
 269                                 return;
 270                         }
 271                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 272                         if (cs_l) {
 273                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 274                                        "in long mode while CS.L == 1\n");
 275                                 kvm_inject_gp(vcpu, 0);
 276                                 return;
 277
 278                         }
 279                 } else
 280 #endif
 281                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 282                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 283                                "reserved bits\n");
 284                         kvm_inject_gp(vcpu, 0);
 285                         return;
 286                 }
 287
 288         }
 289
 290         kvm_x86_ops->set_cr0(vcpu, cr0);
 291         vcpu->arch.cr0 = cr0;
 292
 293         kvm_mmu_reset_context(vcpu);
 294         return;
 295 }
 296 EXPORT_SYMBOL_GPL(set_cr0);
 297
 298 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 299 {
 300         set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 301 }
 302 EXPORT_SYMBOL_GPL(lmsw);
 303
 304 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 305 {
 306         if (cr4 & CR4_RESERVED_BITS) {
 307                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 308                 kvm_inject_gp(vcpu, 0);
 309                 return;
 310         }
 311
 312         if (is_long_mode(vcpu)) {
 313                 if (!(cr4 & X86_CR4_PAE)) {
 314                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 315                                "in long mode\n");
 316                         kvm_inject_gp(vcpu, 0);
 317                         return;
 318                 }
 319         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 320                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 321                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 322                 kvm_inject_gp(vcpu, 0);
 323                 return;
 324         }
 325
 326         if (cr4 & X86_CR4_VMXE) {
 327                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 328                 kvm_inject_gp(vcpu, 0);
 329                 return;
 330         }
 331         kvm_x86_ops->set_cr4(vcpu, cr4);
 332         vcpu->arch.cr4 = cr4;
 333         kvm_mmu_reset_context(vcpu);
 334 }
 335 EXPORT_SYMBOL_GPL(set_cr4);
 336
 337 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 338 {
 339         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 340                 kvm_mmu_flush_tlb(vcpu);
 341                 return;
 342         }
 343
 344         if (is_long_mode(vcpu)) {
 345                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 346                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 347                         kvm_inject_gp(vcpu, 0);
 348                         return;
 349                 }
 350         } else {
 351                 if (is_pae(vcpu)) {
 352                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 353                                 printk(KERN_DEBUG
 354                                        "set_cr3: #GP, reserved bits\n");
 355                                 kvm_inject_gp(vcpu, 0);
 356                                 return;
 357                         }
 358                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 359                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 360                                        "reserved bits\n");
 361                                 kvm_inject_gp(vcpu, 0);
 362                                 return;
 363                         }
 364                 }
 365                 /*
 366                  * We don't check reserved bits in nonpae mode, because
 367                  * this isn't enforced, and VMware depends on this.
 368                  */
 369         }
 370
 371         down_read(&vcpu->kvm->slots_lock);
 372         /*
 373          * Does the new cr3 value map to physical memory? (Note, we
 374          * catch an invalid cr3 even in real-mode, because it would
 375          * cause trouble later on when we turn on paging anyway.)
 376          *
 377          * A real CPU would silently accept an invalid cr3 and would
 378          * attempt to use it - with largely undefined (and often hard
 379          * to debug) behavior on the guest side.
 380          */
 381         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 382                 kvm_inject_gp(vcpu, 0);
 383         else {
 384                 vcpu->arch.cr3 = cr3;
 385                 vcpu->arch.mmu.new_cr3(vcpu);
 386         }
 387         up_read(&vcpu->kvm->slots_lock);
 388 }
 389 EXPORT_SYMBOL_GPL(set_cr3);
 390
 391 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 392 {
 393         if (cr8 & CR8_RESERVED_BITS) {
 394                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 395                 kvm_inject_gp(vcpu, 0);
 396                 return;
 397         }
 398         if (irqchip_in_kernel(vcpu->kvm))
 399                 kvm_lapic_set_tpr(vcpu, cr8);
 400         else
 401                 vcpu->arch.cr8 = cr8;
 402 }
 403 EXPORT_SYMBOL_GPL(set_cr8);
 404
 405 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 406 {
 407         if (irqchip_in_kernel(vcpu->kvm))
 408                 return kvm_lapic_get_cr8(vcpu);
 409         else
 410                 return vcpu->arch.cr8;
 411 }
 412 EXPORT_SYMBOL_GPL(get_cr8);
 413
 414 /*
 415  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 416  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 417  *
 418  * This list is modified at module load time to reflect the
 419  * capabilities of the host cpu.
 420  */
 421 static u32 msrs_to_save[] = {
 422         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 423         MSR_K6_STAR,
 424 #ifdef CONFIG_X86_64
 425         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 426 #endif
 427         MSR_IA32_TIME_STAMP_COUNTER,
 428 };
 429
 430 static unsigned num_msrs_to_save;
 431
 432 static u32 emulated_msrs[] = {
 433         MSR_IA32_MISC_ENABLE,
 434 };
 435
 436 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 437 {
 438         if (efer & efer_reserved_bits) {
 439                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 440                        efer);
 441                 kvm_inject_gp(vcpu, 0);
 442                 return;
 443         }
 444
 445         if (is_paging(vcpu)
 446             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 447                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 448                 kvm_inject_gp(vcpu, 0);
 449                 return;
 450         }
 451
 452         kvm_x86_ops->set_efer(vcpu, efer);
 453
 454         efer &= ~EFER_LMA;
 455         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 456
 457         vcpu->arch.shadow_efer = efer;
 458 }
 459
 460 void kvm_enable_efer_bits(u64 mask)
 461 {
 462        efer_reserved_bits &= ~mask;
 463 }
 464 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
 465
 466
 467 /*
 468  * Writes msr value into into the appropriate "register".
 469  * Returns 0 on success, non-0 otherwise.
 470  * Assumes vcpu_load() was already called.
 471  */
 472 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 473 {
 474         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 475 }
 476
 477 /*
 478  * Adapt set_msr() to msr_io()'s calling convention
 479  */
 480 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 481 {
 482         return kvm_set_msr(vcpu, index, *data);
 483 }
 484
 485
 486 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 487 {
 488         switch (msr) {
 489         case MSR_EFER:
 490                 set_efer(vcpu, data);
 491                 break;
 492         case MSR_IA32_MC0_STATUS:
 493                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 494                        __FUNCTION__, data);
 495                 break;
 496         case MSR_IA32_MCG_STATUS:
 497                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 498                         __FUNCTION__, data);
 499                 break;
 500         case MSR_IA32_MCG_CTL:
 501                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
 502                         __FUNCTION__, data);
 503                 break;
 504         case MSR_IA32_UCODE_REV:
 505         case MSR_IA32_UCODE_WRITE:
 506         case 0x200 ... 0x2ff: /* MTRRs */
 507                 break;
 508         case MSR_IA32_APICBASE:
 509                 kvm_set_apic_base(vcpu, data);
 510                 break;
 511         case MSR_IA32_MISC_ENABLE:
 512                 vcpu->arch.ia32_misc_enable_msr = data;
 513                 break;
 514         default:
 515                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 516                 return 1;
 517         }
 518         return 0;
 519 }
 520 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 521
 522
 523 /*
 524  * Reads an msr value (of 'msr_index') into 'pdata'.
 525  * Returns 0 on success, non-0 otherwise.
 526  * Assumes vcpu_load() was already called.
 527  */
 528 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 529 {
 530         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 531 }
 532
 533 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 534 {
 535         u64 data;
 536
 537         switch (msr) {
 538         case 0xc0010010: /* SYSCFG */
 539         case 0xc0010015: /* HWCR */
 540         case MSR_IA32_PLATFORM_ID:
 541         case MSR_IA32_P5_MC_ADDR:
 542         case MSR_IA32_P5_MC_TYPE:
 543         case MSR_IA32_MC0_CTL:
 544         case MSR_IA32_MCG_STATUS:
 545         case MSR_IA32_MCG_CAP:
 546         case MSR_IA32_MCG_CTL:
 547         case MSR_IA32_MC0_MISC:
 548         case MSR_IA32_MC0_MISC+4:
 549         case MSR_IA32_MC0_MISC+8:
 550         case MSR_IA32_MC0_MISC+12:
 551         case MSR_IA32_MC0_MISC+16:
 552         case MSR_IA32_UCODE_REV:
 553         case MSR_IA32_PERF_STATUS:
 554         case MSR_IA32_EBL_CR_POWERON:
 555                 /* MTRR registers */
 556         case 0xfe:
 557         case 0x200 ... 0x2ff:
 558                 data = 0;
 559                 break;
 560         case 0xcd: /* fsb frequency */
 561                 data = 3;
 562                 break;
 563         case MSR_IA32_APICBASE:
 564                 data = kvm_get_apic_base(vcpu);
 565                 break;
 566         case MSR_IA32_MISC_ENABLE:
 567                 data = vcpu->arch.ia32_misc_enable_msr;
 568                 break;
 569         case MSR_EFER:
 570                 data = vcpu->arch.shadow_efer;
 571                 break;
 572         default:
 573                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 574                 return 1;
 575         }
 576         *pdata = data;
 577         return 0;
 578 }
 579 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 580
 581 /*
 582  * Read or write a bunch of msrs. All parameters are kernel addresses.
 583  *
 584  * @return number of msrs set successfully.
 585  */
 586 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 587                     struct kvm_msr_entry *entries,
 588                     int (*do_msr)(struct kvm_vcpu *vcpu,
 589                                   unsigned index, u64 *data))
 590 {
 591         int i;
 592
 593         vcpu_load(vcpu);
 594
 595         for (i = 0; i < msrs->nmsrs; ++i)
 596                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 597                         break;
 598
 599         vcpu_put(vcpu);
 600
 601         return i;
 602 }
 603
 604 /*
 605  * Read or write a bunch of msrs. Parameters are user addresses.
 606  *
 607  * @return number of msrs set successfully.
 608  */
 609 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 610                   int (*do_msr)(struct kvm_vcpu *vcpu,
 611                                 unsigned index, u64 *data),
 612                   int writeback)
 613 {
 614         struct kvm_msrs msrs;
 615         struct kvm_msr_entry *entries;
 616         int r, n;
 617         unsigned size;
 618
 619         r = -EFAULT;
 620         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 621                 goto out;
 622
 623         r = -E2BIG;
 624         if (msrs.nmsrs >= MAX_IO_MSRS)
 625                 goto out;
 626
 627         r = -ENOMEM;
 628         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 629         entries = vmalloc(size);
 630         if (!entries)
 631                 goto out;
 632
 633         r = -EFAULT;
 634         if (copy_from_user(entries, user_msrs->entries, size))
 635                 goto out_free;
 636
 637         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 638         if (r < 0)
 639                 goto out_free;
 640
 641         r = -EFAULT;
 642         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 643                 goto out_free;
 644
 645         r = n;
 646
 647 out_free:
 648         vfree(entries);
 649 out:
 650         return r;
 651 }
 652
 653 /*
 654  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
 655  * cached on it.
 656  */
 657 void decache_vcpus_on_cpu(int cpu)
 658 {
 659         struct kvm *vm;
 660         struct kvm_vcpu *vcpu;
 661         int i;
 662
 663         spin_lock(&kvm_lock);
 664         list_for_each_entry(vm, &vm_list, vm_list)
 665                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 666                         vcpu = vm->vcpus[i];
 667                         if (!vcpu)
 668                                 continue;
 669                         /*
 670                          * If the vcpu is locked, then it is running on some
 671                          * other cpu and therefore it is not cached on the
 672                          * cpu in question.
 673                          *
 674                          * If it's not locked, check the last cpu it executed
 675                          * on.
 676                          */
 677                         if (mutex_trylock(&vcpu->mutex)) {
 678                                 if (vcpu->cpu == cpu) {
 679                                         kvm_x86_ops->vcpu_decache(vcpu);
 680                                         vcpu->cpu = -1;
 681                                 }
 682                                 mutex_unlock(&vcpu->mutex);
 683                         }
 684                 }
 685         spin_unlock(&kvm_lock);
 686 }
 687
 688 int kvm_dev_ioctl_check_extension(long ext)
 689 {
 690         int r;
 691
 692         switch (ext) {
 693         case KVM_CAP_IRQCHIP:
 694         case KVM_CAP_HLT:
 695         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 696         case KVM_CAP_USER_MEMORY:
 697         case KVM_CAP_SET_TSS_ADDR:
 698         case KVM_CAP_EXT_CPUID:
 699                 r = 1;
 700                 break;
 701         case KVM_CAP_VAPIC:
 702                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 703                 break;
 704         default:
 705                 r = 0;
 706                 break;
 707         }
 708         return r;
 709
 710 }
 711
 712 long kvm_arch_dev_ioctl(struct file *filp,
 713                         unsigned int ioctl, unsigned long arg)
 714 {
 715         void __user *argp = (void __user *)arg;
 716         long r;
 717
 718         switch (ioctl) {
 719         case KVM_GET_MSR_INDEX_LIST: {
 720                 struct kvm_msr_list __user *user_msr_list = argp;
 721                 struct kvm_msr_list msr_list;
 722                 unsigned n;
 723
 724                 r = -EFAULT;
 725                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 726                         goto out;
 727                 n = msr_list.nmsrs;
 728                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 729                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 730                         goto out;
 731                 r = -E2BIG;
 732                 if (n < num_msrs_to_save)
 733                         goto out;
 734                 r = -EFAULT;
 735                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 736                                  num_msrs_to_save * sizeof(u32)))
 737                         goto out;
 738                 if (copy_to_user(user_msr_list->indices
 739                                  + num_msrs_to_save * sizeof(u32),
 740                                  &emulated_msrs,
 741                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 742                         goto out;
 743                 r = 0;
 744                 break;
 745         }
 746         case KVM_GET_SUPPORTED_CPUID: {
 747                 struct kvm_cpuid2 __user *cpuid_arg = argp;
 748                 struct kvm_cpuid2 cpuid;
 749
 750                 r = -EFAULT;
 751                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
 752                         goto out;
 753                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
 754                         cpuid_arg->entries);
 755                 if (r)
 756                         goto out;
 757
 758                 r = -EFAULT;
 759                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
 760                         goto out;
 761                 r = 0;
 762                 break;
 763         }
 764         default:
 765                 r = -EINVAL;
 766         }
 767 out:
 768         return r;
 769 }
 770
 771 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 772 {
 773         kvm_x86_ops->vcpu_load(vcpu, cpu);
 774 }
 775
 776 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 777 {
 778         kvm_x86_ops->vcpu_put(vcpu);
 779         kvm_put_guest_fpu(vcpu);
 780 }
 781
 782 static int is_efer_nx(void)
 783 {
 784         u64 efer;
 785
 786         rdmsrl(MSR_EFER, efer);
 787         return efer & EFER_NX;
 788 }
 789
 790 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 791 {
 792         int i;
 793         struct kvm_cpuid_entry2 *e, *entry;
 794
 795         entry = NULL;
 796         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 797                 e = &vcpu->arch.cpuid_entries[i];
 798                 if (e->function == 0x80000001) {
 799                         entry = e;
 800                         break;
 801                 }
 802         }
 803         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
 804                 entry->edx &= ~(1 << 20);
 805                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 806         }
 807 }
 808
 809 /* when an old userspace process fills a new kernel module */
 810 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 811                                     struct kvm_cpuid *cpuid,
 812                                     struct kvm_cpuid_entry __user *entries)
 813 {
 814         int r, i;
 815         struct kvm_cpuid_entry *cpuid_entries;
 816
 817         r = -E2BIG;
 818         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 819                 goto out;
 820         r = -ENOMEM;
 821         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 822         if (!cpuid_entries)
 823                 goto out;
 824         r = -EFAULT;
 825         if (copy_from_user(cpuid_entries, entries,
 826                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 827                 goto out_free;
 828         for (i = 0; i < cpuid->nent; i++) {
 829                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 830                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 831                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 832                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 833                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 834                 vcpu->arch.cpuid_entries[i].index = 0;
 835                 vcpu->arch.cpuid_entries[i].flags = 0;
 836                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 837                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 838                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 839         }
 840         vcpu->arch.cpuid_nent = cpuid->nent;
 841         cpuid_fix_nx_cap(vcpu);
 842         r = 0;
 843
 844 out_free:
 845         vfree(cpuid_entries);
 846 out:
 847         return r;
 848 }
 849
 850 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 851                                     struct kvm_cpuid2 *cpuid,
 852                                     struct kvm_cpuid_entry2 __user *entries)
 853 {
 854         int r;
 855
 856         r = -E2BIG;
 857         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 858                 goto out;
 859         r = -EFAULT;
 860         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 861                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 862                 goto out;
 863         vcpu->arch.cpuid_nent = cpuid->nent;
 864         return 0;
 865
 866 out:
 867         return r;
 868 }
 869
 870 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 871                                     struct kvm_cpuid2 *cpuid,
 872                                     struct kvm_cpuid_entry2 __user *entries)
 873 {
 874         int r;
 875
 876         r = -E2BIG;
 877         if (cpuid->nent < vcpu->arch.cpuid_nent)
 878                 goto out;
 879         r = -EFAULT;
 880         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 881                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 882                 goto out;
 883         return 0;
 884
 885 out:
 886         cpuid->nent = vcpu->arch.cpuid_nent;
 887         return r;
 888 }
 889
 890 static inline u32 bit(int bitno)
 891 {
 892         return 1 << (bitno & 31);
 893 }
 894
 895 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 896                           u32 index)
 897 {
 898         entry->function = function;
 899         entry->index = index;
 900         cpuid_count(entry->function, entry->index,
 901                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 902         entry->flags = 0;
 903 }
 904
 905 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 906                          u32 index, int *nent, int maxnent)
 907 {
 908         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
 909                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
 910                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
 911                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
 912                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
 913                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
 914                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
 915                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
 916                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
 917                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
 918         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
 919                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
 920                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
 921                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
 922                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
 923                 bit(X86_FEATURE_PGE) |
 924                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
 925                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
 926                 bit(X86_FEATURE_SYSCALL) |
 927                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
 928 #ifdef CONFIG_X86_64
 929                 bit(X86_FEATURE_LM) |
 930 #endif
 931                 bit(X86_FEATURE_MMXEXT) |
 932                 bit(X86_FEATURE_3DNOWEXT) |
 933                 bit(X86_FEATURE_3DNOW);
 934         const u32 kvm_supported_word3_x86_features =
 935                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
 936         const u32 kvm_supported_word6_x86_features =
 937                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
 938
 939         /* all func 2 cpuid_count() should be called on the same cpu */
 940         get_cpu();
 941         do_cpuid_1_ent(entry, function, index);
 942         ++*nent;
 943
 944         switch (function) {
 945         case 0:
 946                 entry->eax = min(entry->eax, (u32)0xb);
 947                 break;
 948         case 1:
 949                 entry->edx &= kvm_supported_word0_x86_features;
 950                 entry->ecx &= kvm_supported_word3_x86_features;
 951                 break;
 952         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 953          * may return different values. This forces us to get_cpu() before
 954          * issuing the first command, and also to emulate this annoying behavior
 955          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 956         case 2: {
 957                 int t, times = entry->eax & 0xff;
 958
 959                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 960                 for (t = 1; t < times && *nent < maxnent; ++t) {
 961                         do_cpuid_1_ent(&entry[t], function, 0);
 962                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 963                         ++*nent;
 964                 }
 965                 break;
 966         }
 967         /* function 4 and 0xb have additional index. */
 968         case 4: {
 969                 int index, cache_type;
 970
 971                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 972                 /* read more entries until cache_type is zero */
 973                 for (index = 1; *nent < maxnent; ++index) {
 974                         cache_type = entry[index - 1].eax & 0x1f;
 975                         if (!cache_type)
 976                                 break;
 977                         do_cpuid_1_ent(&entry[index], function, index);
 978                         entry[index].flags |=
 979                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 980                         ++*nent;
 981                 }
 982                 break;
 983         }
 984         case 0xb: {
 985                 int index, level_type;
 986
 987                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 988                 /* read more entries until level_type is zero */
 989                 for (index = 1; *nent < maxnent; ++index) {
 990                         level_type = entry[index - 1].ecx & 0xff;
 991                         if (!level_type)
 992                                 break;
 993                         do_cpuid_1_ent(&entry[index], function, index);
 994                         entry[index].flags |=
 995                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 996                         ++*nent;
 997                 }
 998                 break;
 999         }
1000         case 0x80000000:
1001                 entry->eax = min(entry->eax, 0x8000001a);
1002                 break;
1003         case 0x80000001:
1004                 entry->edx &= kvm_supported_word1_x86_features;
1005                 entry->ecx &= kvm_supported_word6_x86_features;
1006                 break;
1007         }
1008         put_cpu();
1009 }
1010
1011 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1012                                     struct kvm_cpuid_entry2 __user *entries)
1013 {
1014         struct kvm_cpuid_entry2 *cpuid_entries;
1015         int limit, nent = 0, r = -E2BIG;
1016         u32 func;
1017
1018         if (cpuid->nent < 1)
1019                 goto out;
1020         r = -ENOMEM;
1021         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1022         if (!cpuid_entries)
1023                 goto out;
1024
1025         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1026         limit = cpuid_entries[0].eax;
1027         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1028                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1029                                 &nent, cpuid->nent);
1030         r = -E2BIG;
1031         if (nent >= cpuid->nent)
1032                 goto out_free;
1033
1034         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1035         limit = cpuid_entries[nent - 1].eax;
1036         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1037                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1038                                &nent, cpuid->nent);
1039         r = -EFAULT;
1040         if (copy_to_user(entries, cpuid_entries,
1041                         nent * sizeof(struct kvm_cpuid_entry2)))
1042                 goto out_free;
1043         cpuid->nent = nent;
1044         r = 0;
1045
1046 out_free:
1047         vfree(cpuid_entries);
1048 out:
1049         return r;
1050 }
1051
1052 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1053                                     struct kvm_lapic_state *s)
1054 {
1055         vcpu_load(vcpu);
1056         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1057         vcpu_put(vcpu);
1058
1059         return 0;
1060 }
1061
1062 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1063                                     struct kvm_lapic_state *s)
1064 {
1065         vcpu_load(vcpu);
1066         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1067         kvm_apic_post_state_restore(vcpu);
1068         vcpu_put(vcpu);
1069
1070         return 0;
1071 }
1072
1073 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1074                                     struct kvm_interrupt *irq)
1075 {
1076         if (irq->irq < 0 || irq->irq >= 256)
1077                 return -EINVAL;
1078         if (irqchip_in_kernel(vcpu->kvm))
1079                 return -ENXIO;
1080         vcpu_load(vcpu);
1081
1082         set_bit(irq->irq, vcpu->arch.irq_pending);
1083         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1084
1085         vcpu_put(vcpu);
1086
1087         return 0;
1088 }
1089
1090 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1091                                            struct kvm_tpr_access_ctl *tac)
1092 {
1093         if (tac->flags)
1094                 return -EINVAL;
1095         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1096         return 0;
1097 }
1098
1099 long kvm_arch_vcpu_ioctl(struct file *filp,
1100                          unsigned int ioctl, unsigned long arg)
1101 {
1102         struct kvm_vcpu *vcpu = filp->private_data;
1103         void __user *argp = (void __user *)arg;
1104         int r;
1105
1106         switch (ioctl) {
1107         case KVM_GET_LAPIC: {
1108                 struct kvm_lapic_state lapic;
1109
1110                 memset(&lapic, 0, sizeof lapic);
1111                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1112                 if (r)
1113                         goto out;
1114                 r = -EFAULT;
1115                 if (copy_to_user(argp, &lapic, sizeof lapic))
1116                         goto out;
1117                 r = 0;
1118                 break;
1119         }
1120         case KVM_SET_LAPIC: {
1121                 struct kvm_lapic_state lapic;
1122
1123                 r = -EFAULT;
1124                 if (copy_from_user(&lapic, argp, sizeof lapic))
1125                         goto out;
1126                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1127                 if (r)
1128                         goto out;
1129                 r = 0;
1130                 break;
1131         }
1132         case KVM_INTERRUPT: {
1133                 struct kvm_interrupt irq;
1134
1135                 r = -EFAULT;
1136                 if (copy_from_user(&irq, argp, sizeof irq))
1137                         goto out;
1138                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1139                 if (r)
1140                         goto out;
1141                 r = 0;
1142                 break;
1143         }
1144         case KVM_SET_CPUID: {
1145                 struct kvm_cpuid __user *cpuid_arg = argp;
1146                 struct kvm_cpuid cpuid;
1147
1148                 r = -EFAULT;
1149                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1150                         goto out;
1151                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1152                 if (r)
1153                         goto out;
1154                 break;
1155         }
1156         case KVM_SET_CPUID2: {
1157                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1158                 struct kvm_cpuid2 cpuid;
1159
1160                 r = -EFAULT;
1161                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1162                         goto out;
1163                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1164                                 cpuid_arg->entries);
1165                 if (r)
1166                         goto out;
1167                 break;
1168         }
1169         case KVM_GET_CPUID2: {
1170                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1171                 struct kvm_cpuid2 cpuid;
1172
1173                 r = -EFAULT;
1174                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1175                         goto out;
1176                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1177                                 cpuid_arg->entries);
1178                 if (r)
1179                         goto out;
1180                 r = -EFAULT;
1181                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1182                         goto out;
1183                 r = 0;
1184                 break;
1185         }
1186         case KVM_GET_MSRS:
1187                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1188                 break;
1189         case KVM_SET_MSRS:
1190                 r = msr_io(vcpu, argp, do_set_msr, 0);
1191                 break;
1192         case KVM_TPR_ACCESS_REPORTING: {
1193                 struct kvm_tpr_access_ctl tac;
1194
1195                 r = -EFAULT;
1196                 if (copy_from_user(&tac, argp, sizeof tac))
1197                         goto out;
1198                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1199                 if (r)
1200                         goto out;
1201                 r = -EFAULT;
1202                 if (copy_to_user(argp, &tac, sizeof tac))
1203                         goto out;
1204                 r = 0;
1205                 break;
1206         };
1207         case KVM_SET_VAPIC_ADDR: {
1208                 struct kvm_vapic_addr va;
1209
1210                 r = -EINVAL;
1211                 if (!irqchip_in_kernel(vcpu->kvm))
1212                         goto out;
1213                 r = -EFAULT;
1214                 if (copy_from_user(&va, argp, sizeof va))
1215                         goto out;
1216                 r = 0;
1217                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1218                 break;
1219         }
1220         default:
1221                 r = -EINVAL;
1222         }
1223 out:
1224         return r;
1225 }
1226
1227 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1228 {
1229         int ret;
1230
1231         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1232                 return -1;
1233         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1234         return ret;
1235 }
1236
1237 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1238                                           u32 kvm_nr_mmu_pages)
1239 {
1240         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1241                 return -EINVAL;
1242
1243         down_write(&kvm->slots_lock);
1244
1245         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1246         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1247
1248         up_write(&kvm->slots_lock);
1249         return 0;
1250 }
1251
1252 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1253 {
1254         return kvm->arch.n_alloc_mmu_pages;
1255 }
1256
1257 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1258 {
1259         int i;
1260         struct kvm_mem_alias *alias;
1261
1262         for (i = 0; i < kvm->arch.naliases; ++i) {
1263                 alias = &kvm->arch.aliases[i];
1264                 if (gfn >= alias->base_gfn
1265                     && gfn < alias->base_gfn + alias->npages)
1266                         return alias->target_gfn + gfn - alias->base_gfn;
1267         }
1268         return gfn;
1269 }
1270
1271 /*
1272  * Set a new alias region.  Aliases map a portion of physical memory into
1273  * another portion.  This is useful for memory windows, for example the PC
1274  * VGA region.
1275  */
1276 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1277                                          struct kvm_memory_alias *alias)
1278 {
1279         int r, n;
1280         struct kvm_mem_alias *p;
1281
1282         r = -EINVAL;
1283         /* General sanity checks */
1284         if (alias->memory_size & (PAGE_SIZE - 1))
1285                 goto out;
1286         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1287                 goto out;
1288         if (alias->slot >= KVM_ALIAS_SLOTS)
1289                 goto out;
1290         if (alias->guest_phys_addr + alias->memory_size
1291             < alias->guest_phys_addr)
1292                 goto out;
1293         if (alias->target_phys_addr + alias->memory_size
1294             < alias->target_phys_addr)
1295                 goto out;
1296
1297         down_write(&kvm->slots_lock);
1298
1299         p = &kvm->arch.aliases[alias->slot];
1300         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1301         p->npages = alias->memory_size >> PAGE_SHIFT;
1302         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1303
1304         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1305                 if (kvm->arch.aliases[n - 1].npages)
1306                         break;
1307         kvm->arch.naliases = n;
1308
1309         kvm_mmu_zap_all(kvm);
1310
1311         up_write(&kvm->slots_lock);
1312
1313         return 0;
1314
1315 out:
1316         return r;
1317 }
1318
1319 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1320 {
1321         int r;
1322
1323         r = 0;
1324         switch (chip->chip_id) {
1325         case KVM_IRQCHIP_PIC_MASTER:
1326                 memcpy(&chip->chip.pic,
1327                         &pic_irqchip(kvm)->pics[0],
1328                         sizeof(struct kvm_pic_state));
1329                 break;
1330         case KVM_IRQCHIP_PIC_SLAVE:
1331                 memcpy(&chip->chip.pic,
1332                         &pic_irqchip(kvm)->pics[1],
1333                         sizeof(struct kvm_pic_state));
1334                 break;
1335         case KVM_IRQCHIP_IOAPIC:
1336                 memcpy(&chip->chip.ioapic,
1337                         ioapic_irqchip(kvm),
1338                         sizeof(struct kvm_ioapic_state));
1339                 break;
1340         default:
1341                 r = -EINVAL;
1342                 break;
1343         }
1344         return r;
1345 }
1346
1347 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1348 {
1349         int r;
1350
1351         r = 0;
1352         switch (chip->chip_id) {
1353         case KVM_IRQCHIP_PIC_MASTER:
1354                 memcpy(&pic_irqchip(kvm)->pics[0],
1355                         &chip->chip.pic,
1356                         sizeof(struct kvm_pic_state));
1357                 break;
1358         case KVM_IRQCHIP_PIC_SLAVE:
1359                 memcpy(&pic_irqchip(kvm)->pics[1],
1360                         &chip->chip.pic,
1361                         sizeof(struct kvm_pic_state));
1362                 break;
1363         case KVM_IRQCHIP_IOAPIC:
1364                 memcpy(ioapic_irqchip(kvm),
1365                         &chip->chip.ioapic,
1366                         sizeof(struct kvm_ioapic_state));
1367                 break;
1368         default:
1369                 r = -EINVAL;
1370                 break;
1371         }
1372         kvm_pic_update_irq(pic_irqchip(kvm));
1373         return r;
1374 }
1375
1376 /*
1377  * Get (and clear) the dirty memory log for a memory slot.
1378  */
1379 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1380                                       struct kvm_dirty_log *log)
1381 {
1382         int r;
1383         int n;
1384         struct kvm_memory_slot *memslot;
1385         int is_dirty = 0;
1386
1387         down_write(&kvm->slots_lock);
1388
1389         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1390         if (r)
1391                 goto out;
1392
1393         /* If nothing is dirty, don't bother messing with page tables. */
1394         if (is_dirty) {
1395                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1396                 kvm_flush_remote_tlbs(kvm);
1397                 memslot = &kvm->memslots[log->slot];
1398                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1399                 memset(memslot->dirty_bitmap, 0, n);
1400         }
1401         r = 0;
1402 out:
1403         up_write(&kvm->slots_lock);
1404         return r;
1405 }
1406
1407 long kvm_arch_vm_ioctl(struct file *filp,
1408                        unsigned int ioctl, unsigned long arg)
1409 {
1410         struct kvm *kvm = filp->private_data;
1411         void __user *argp = (void __user *)arg;
1412         int r = -EINVAL;
1413
1414         switch (ioctl) {
1415         case KVM_SET_TSS_ADDR:
1416                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1417                 if (r < 0)
1418                         goto out;
1419                 break;
1420         case KVM_SET_MEMORY_REGION: {
1421                 struct kvm_memory_region kvm_mem;
1422                 struct kvm_userspace_memory_region kvm_userspace_mem;
1423
1424                 r = -EFAULT;
1425                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1426                         goto out;
1427                 kvm_userspace_mem.slot = kvm_mem.slot;
1428                 kvm_userspace_mem.flags = kvm_mem.flags;
1429                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1430                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1431                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1432                 if (r)
1433                         goto out;
1434                 break;
1435         }
1436         case KVM_SET_NR_MMU_PAGES:
1437                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1438                 if (r)
1439                         goto out;
1440                 break;
1441         case KVM_GET_NR_MMU_PAGES:
1442                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1443                 break;
1444         case KVM_SET_MEMORY_ALIAS: {
1445                 struct kvm_memory_alias alias;
1446
1447                 r = -EFAULT;
1448                 if (copy_from_user(&alias, argp, sizeof alias))
1449                         goto out;
1450                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1451                 if (r)
1452                         goto out;
1453                 break;
1454         }
1455         case KVM_CREATE_IRQCHIP:
1456                 r = -ENOMEM;
1457                 kvm->arch.vpic = kvm_create_pic(kvm);
1458                 if (kvm->arch.vpic) {
1459                         r = kvm_ioapic_init(kvm);
1460                         if (r) {
1461                                 kfree(kvm->arch.vpic);
1462                                 kvm->arch.vpic = NULL;
1463                                 goto out;
1464                         }
1465                 } else
1466                         goto out;
1467                 break;
1468         case KVM_IRQ_LINE: {
1469                 struct kvm_irq_level irq_event;
1470
1471                 r = -EFAULT;
1472                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1473                         goto out;
1474                 if (irqchip_in_kernel(kvm)) {
1475                         mutex_lock(&kvm->lock);
1476                         if (irq_event.irq < 16)
1477                                 kvm_pic_set_irq(pic_irqchip(kvm),
1478                                         irq_event.irq,
1479                                         irq_event.level);
1480                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1481                                         irq_event.irq,
1482                                         irq_event.level);
1483                         mutex_unlock(&kvm->lock);
1484                         r = 0;
1485                 }
1486                 break;
1487         }
1488         case KVM_GET_IRQCHIP: {
1489                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1490                 struct kvm_irqchip chip;
1491
1492                 r = -EFAULT;
1493                 if (copy_from_user(&chip, argp, sizeof chip))
1494                         goto out;
1495                 r = -ENXIO;
1496                 if (!irqchip_in_kernel(kvm))
1497                         goto out;
1498                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1499                 if (r)
1500                         goto out;
1501                 r = -EFAULT;
1502                 if (copy_to_user(argp, &chip, sizeof chip))
1503                         goto out;
1504                 r = 0;
1505                 break;
1506         }
1507         case KVM_SET_IRQCHIP: {
1508                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1509                 struct kvm_irqchip chip;
1510
1511                 r = -EFAULT;
1512                 if (copy_from_user(&chip, argp, sizeof chip))
1513                         goto out;
1514                 r = -ENXIO;
1515                 if (!irqchip_in_kernel(kvm))
1516                         goto out;
1517                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1518                 if (r)
1519                         goto out;
1520                 r = 0;
1521                 break;
1522         }
1523         default:
1524                 ;
1525         }
1526 out:
1527         return r;
1528 }
1529
1530 static void kvm_init_msr_list(void)
1531 {
1532         u32 dummy[2];
1533         unsigned i, j;
1534
1535         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1536                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1537                         continue;
1538                 if (j < i)
1539                         msrs_to_save[j] = msrs_to_save[i];
1540                 j++;
1541         }
1542         num_msrs_to_save = j;
1543 }
1544
1545 /*
1546  * Only apic need an MMIO device hook, so shortcut now..
1547  */
1548 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1549                                                 gpa_t addr)
1550 {
1551         struct kvm_io_device *dev;
1552
1553         if (vcpu->arch.apic) {
1554                 dev = &vcpu->arch.apic->dev;
1555                 if (dev->in_range(dev, addr))
1556                         return dev;
1557         }
1558         return NULL;
1559 }
1560
1561
1562 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1563                                                 gpa_t addr)
1564 {
1565         struct kvm_io_device *dev;
1566
1567         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1568         if (dev == NULL)
1569                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1570         return dev;
1571 }
1572
1573 int emulator_read_std(unsigned long addr,
1574                              void *val,
1575                              unsigned int bytes,
1576                              struct kvm_vcpu *vcpu)
1577 {
1578         void *data = val;
1579         int r = X86EMUL_CONTINUE;
1580
1581         down_read(&vcpu->kvm->slots_lock);
1582         while (bytes) {
1583                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1584                 unsigned offset = addr & (PAGE_SIZE-1);
1585                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1586                 int ret;
1587
1588                 if (gpa == UNMAPPED_GVA) {
1589                         r = X86EMUL_PROPAGATE_FAULT;
1590                         goto out;
1591                 }
1592                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1593                 if (ret < 0) {
1594                         r = X86EMUL_UNHANDLEABLE;
1595                         goto out;
1596                 }
1597
1598                 bytes -= tocopy;
1599                 data += tocopy;
1600                 addr += tocopy;
1601         }
1602 out:
1603         up_read(&vcpu->kvm->slots_lock);
1604         return r;
1605 }
1606 EXPORT_SYMBOL_GPL(emulator_read_std);
1607
1608 static int emulator_read_emulated(unsigned long addr,
1609                                   void *val,
1610                                   unsigned int bytes,
1611                                   struct kvm_vcpu *vcpu)
1612 {
1613         struct kvm_io_device *mmio_dev;
1614         gpa_t                 gpa;
1615
1616         if (vcpu->mmio_read_completed) {
1617                 memcpy(val, vcpu->mmio_data, bytes);
1618                 vcpu->mmio_read_completed = 0;
1619                 return X86EMUL_CONTINUE;
1620         }
1621
1622         down_read(&vcpu->kvm->slots_lock);
1623         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1624         up_read(&vcpu->kvm->slots_lock);
1625
1626         /* For APIC access vmexit */
1627         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1628                 goto mmio;
1629
1630         if (emulator_read_std(addr, val, bytes, vcpu)
1631                         == X86EMUL_CONTINUE)
1632                 return X86EMUL_CONTINUE;
1633         if (gpa == UNMAPPED_GVA)
1634                 return X86EMUL_PROPAGATE_FAULT;
1635
1636 mmio:
1637         /*
1638          * Is this MMIO handled locally?
1639          */
1640         mutex_lock(&vcpu->kvm->lock);
1641         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1642         if (mmio_dev) {
1643                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1644                 mutex_unlock(&vcpu->kvm->lock);
1645                 return X86EMUL_CONTINUE;
1646         }
1647         mutex_unlock(&vcpu->kvm->lock);
1648
1649         vcpu->mmio_needed = 1;
1650         vcpu->mmio_phys_addr = gpa;
1651         vcpu->mmio_size = bytes;
1652         vcpu->mmio_is_write = 0;
1653
1654         return X86EMUL_UNHANDLEABLE;
1655 }
1656
1657 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1658                                const void *val, int bytes)
1659 {
1660         int ret;
1661
1662         down_read(&vcpu->kvm->slots_lock);
1663         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1664         if (ret < 0) {
1665                 up_read(&vcpu->kvm->slots_lock);
1666                 return 0;
1667         }
1668         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1669         up_read(&vcpu->kvm->slots_lock);
1670         return 1;
1671 }
1672
1673 static int emulator_write_emulated_onepage(unsigned long addr,
1674                                            const void *val,
1675                                            unsigned int bytes,
1676                                            struct kvm_vcpu *vcpu)
1677 {
1678         struct kvm_io_device *mmio_dev;
1679         gpa_t                 gpa;
1680
1681         down_read(&vcpu->kvm->slots_lock);
1682         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1683         up_read(&vcpu->kvm->slots_lock);
1684
1685         if (gpa == UNMAPPED_GVA) {
1686                 kvm_inject_page_fault(vcpu, addr, 2);
1687                 return X86EMUL_PROPAGATE_FAULT;
1688         }
1689
1690         /* For APIC access vmexit */
1691         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1692                 goto mmio;
1693
1694         if (emulator_write_phys(vcpu, gpa, val, bytes))
1695                 return X86EMUL_CONTINUE;
1696
1697 mmio:
1698         /*
1699          * Is this MMIO handled locally?
1700          */
1701         mutex_lock(&vcpu->kvm->lock);
1702         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1703         if (mmio_dev) {
1704                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1705                 mutex_unlock(&vcpu->kvm->lock);
1706                 return X86EMUL_CONTINUE;
1707         }
1708         mutex_unlock(&vcpu->kvm->lock);
1709
1710         vcpu->mmio_needed = 1;
1711         vcpu->mmio_phys_addr = gpa;
1712         vcpu->mmio_size = bytes;
1713         vcpu->mmio_is_write = 1;
1714         memcpy(vcpu->mmio_data, val, bytes);
1715
1716         return X86EMUL_CONTINUE;
1717 }
1718
1719 int emulator_write_emulated(unsigned long addr,
1720                                    const void *val,
1721                                    unsigned int bytes,
1722                                    struct kvm_vcpu *vcpu)
1723 {
1724         /* Crossing a page boundary? */
1725         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1726                 int rc, now;
1727
1728                 now = -addr & ~PAGE_MASK;
1729                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1730                 if (rc != X86EMUL_CONTINUE)
1731                         return rc;
1732                 addr += now;
1733                 val += now;
1734                 bytes -= now;
1735         }
1736         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1737 }
1738 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1739
1740 static int emulator_cmpxchg_emulated(unsigned long addr,
1741                                      const void *old,
1742                                      const void *new,
1743                                      unsigned int bytes,
1744                                      struct kvm_vcpu *vcpu)
1745 {
1746         static int reported;
1747
1748         if (!reported) {
1749                 reported = 1;
1750                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1751         }
1752 #ifndef CONFIG_X86_64
1753         /* guests cmpxchg8b have to be emulated atomically */
1754         if (bytes == 8) {
1755                 gpa_t gpa;
1756                 struct page *page;
1757                 char *kaddr;
1758                 u64 val;
1759
1760                 down_read(&vcpu->kvm->slots_lock);
1761                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1762
1763                 if (gpa == UNMAPPED_GVA ||
1764                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1765                         goto emul_write;
1766
1767                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1768                         goto emul_write;
1769
1770                 val = *(u64 *)new;
1771
1772                 down_read(&current->mm->mmap_sem);
1773                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1774                 up_read(&current->mm->mmap_sem);
1775
1776                 kaddr = kmap_atomic(page, KM_USER0);
1777                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1778                 kunmap_atomic(kaddr, KM_USER0);
1779                 kvm_release_page_dirty(page);
1780         emul_write:
1781                 up_read(&vcpu->kvm->slots_lock);
1782         }
1783 #endif
1784
1785         return emulator_write_emulated(addr, new, bytes, vcpu);
1786 }
1787
1788 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1789 {
1790         return kvm_x86_ops->get_segment_base(vcpu, seg);
1791 }
1792
1793 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1794 {
1795         return X86EMUL_CONTINUE;
1796 }
1797
1798 int emulate_clts(struct kvm_vcpu *vcpu)
1799 {
1800         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1801         return X86EMUL_CONTINUE;
1802 }
1803
1804 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1805 {
1806         struct kvm_vcpu *vcpu = ctxt->vcpu;
1807
1808         switch (dr) {
1809         case 0 ... 3:
1810                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1811                 return X86EMUL_CONTINUE;
1812         default:
1813                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1814                 return X86EMUL_UNHANDLEABLE;
1815         }
1816 }
1817
1818 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1819 {
1820         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1821         int exception;
1822
1823         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1824         if (exception) {
1825                 /* FIXME: better handling */
1826                 return X86EMUL_UNHANDLEABLE;
1827         }
1828         return X86EMUL_CONTINUE;
1829 }
1830
1831 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1832 {
1833         static int reported;
1834         u8 opcodes[4];
1835         unsigned long rip = vcpu->arch.rip;
1836         unsigned long rip_linear;
1837
1838         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1839
1840         if (reported)
1841                 return;
1842
1843         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1844
1845         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1846                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1847         reported = 1;
1848 }
1849 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1850
1851 struct x86_emulate_ops emulate_ops = {
1852         .read_std            = emulator_read_std,
1853         .read_emulated       = emulator_read_emulated,
1854         .write_emulated      = emulator_write_emulated,
1855         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1856 };
1857
1858 int emulate_instruction(struct kvm_vcpu *vcpu,
1859                         struct kvm_run *run,
1860                         unsigned long cr2,
1861                         u16 error_code,
1862                         int emulation_type)
1863 {
1864         int r;
1865         struct decode_cache *c;
1866
1867         vcpu->arch.mmio_fault_cr2 = cr2;
1868         kvm_x86_ops->cache_regs(vcpu);
1869
1870         vcpu->mmio_is_write = 0;
1871         vcpu->arch.pio.string = 0;
1872
1873         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
1874                 int cs_db, cs_l;
1875                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1876
1877                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
1878                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1879                 vcpu->arch.emulate_ctxt.mode =
1880                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
1881                         ? X86EMUL_MODE_REAL : cs_l
1882                         ? X86EMUL_MODE_PROT64 : cs_db
1883                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1884
1885                 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1886                         vcpu->arch.emulate_ctxt.cs_base = 0;
1887                         vcpu->arch.emulate_ctxt.ds_base = 0;
1888                         vcpu->arch.emulate_ctxt.es_base = 0;
1889                         vcpu->arch.emulate_ctxt.ss_base = 0;
1890                 } else {
1891                         vcpu->arch.emulate_ctxt.cs_base =
1892                                         get_segment_base(vcpu, VCPU_SREG_CS);
1893                         vcpu->arch.emulate_ctxt.ds_base =
1894                                         get_segment_base(vcpu, VCPU_SREG_DS);
1895                         vcpu->arch.emulate_ctxt.es_base =
1896                                         get_segment_base(vcpu, VCPU_SREG_ES);
1897                         vcpu->arch.emulate_ctxt.ss_base =
1898                                         get_segment_base(vcpu, VCPU_SREG_SS);
1899                 }
1900
1901                 vcpu->arch.emulate_ctxt.gs_base =
1902                                         get_segment_base(vcpu, VCPU_SREG_GS);
1903                 vcpu->arch.emulate_ctxt.fs_base =
1904                                         get_segment_base(vcpu, VCPU_SREG_FS);
1905
1906                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
1907
1908                 /* Reject the instructions other than VMCALL/VMMCALL when
1909                  * try to emulate invalid opcode */
1910                 c = &vcpu->arch.emulate_ctxt.decode;
1911                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
1912                     (!(c->twobyte && c->b == 0x01 &&
1913                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
1914                        c->modrm_mod == 3 && c->modrm_rm == 1)))
1915                         return EMULATE_FAIL;
1916
1917                 ++vcpu->stat.insn_emulation;
1918                 if (r)  {
1919                         ++vcpu->stat.insn_emulation_fail;
1920                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1921                                 return EMULATE_DONE;
1922                         return EMULATE_FAIL;
1923                 }
1924         }
1925
1926         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
1927
1928         if (vcpu->arch.pio.string)
1929                 return EMULATE_DO_MMIO;
1930
1931         if ((r || vcpu->mmio_is_write) && run) {
1932                 run->exit_reason = KVM_EXIT_MMIO;
1933                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1934                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1935                 run->mmio.len = vcpu->mmio_size;
1936                 run->mmio.is_write = vcpu->mmio_is_write;
1937         }
1938
1939         if (r) {
1940                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1941                         return EMULATE_DONE;
1942                 if (!vcpu->mmio_needed) {
1943                         kvm_report_emulation_failure(vcpu, "mmio");
1944                         return EMULATE_FAIL;
1945                 }
1946                 return EMULATE_DO_MMIO;
1947         }
1948
1949         kvm_x86_ops->decache_regs(vcpu);
1950         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
1951
1952         if (vcpu->mmio_is_write) {
1953                 vcpu->mmio_needed = 0;
1954                 return EMULATE_DO_MMIO;
1955         }
1956
1957         return EMULATE_DONE;
1958 }
1959 EXPORT_SYMBOL_GPL(emulate_instruction);
1960
1961 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1962 {
1963         int i;
1964
1965         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
1966                 if (vcpu->arch.pio.guest_pages[i]) {
1967                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
1968                         vcpu->arch.pio.guest_pages[i] = NULL;
1969                 }
1970 }
1971
1972 static int pio_copy_data(struct kvm_vcpu *vcpu)
1973 {
1974         void *p = vcpu->arch.pio_data;
1975         void *q;
1976         unsigned bytes;
1977         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
1978
1979         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1980                  PAGE_KERNEL);
1981         if (!q) {
1982                 free_pio_guest_pages(vcpu);
1983                 return -ENOMEM;
1984         }
1985         q += vcpu->arch.pio.guest_page_offset;
1986         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
1987         if (vcpu->arch.pio.in)
1988                 memcpy(q, p, bytes);
1989         else
1990                 memcpy(p, q, bytes);
1991         q -= vcpu->arch.pio.guest_page_offset;
1992         vunmap(q);
1993         free_pio_guest_pages(vcpu);
1994         return 0;
1995 }
1996
1997 int complete_pio(struct kvm_vcpu *vcpu)
1998 {
1999         struct kvm_pio_request *io = &vcpu->arch.pio;
2000         long delta;
2001         int r;
2002
2003         kvm_x86_ops->cache_regs(vcpu);
2004
2005         if (!io->string) {
2006                 if (io->in)
2007                         memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2008                                io->size);
2009         } else {
2010                 if (io->in) {
2011                         r = pio_copy_data(vcpu);
2012                         if (r) {
2013                                 kvm_x86_ops->cache_regs(vcpu);
2014                                 return r;
2015                         }
2016                 }
2017
2018                 delta = 1;
2019                 if (io->rep) {
2020                         delta *= io->cur_count;
2021                         /*
2022                          * The size of the register should really depend on
2023                          * current address size.
2024                          */
2025                         vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2026                 }
2027                 if (io->down)
2028                         delta = -delta;
2029                 delta *= io->size;
2030                 if (io->in)
2031                         vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2032                 else
2033                         vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2034         }
2035
2036         kvm_x86_ops->decache_regs(vcpu);
2037
2038         io->count -= io->cur_count;
2039         io->cur_count = 0;
2040
2041         return 0;
2042 }
2043
2044 static void kernel_pio(struct kvm_io_device *pio_dev,
2045                        struct kvm_vcpu *vcpu,
2046                        void *pd)
2047 {
2048         /* TODO: String I/O for in kernel device */
2049
2050         mutex_lock(&vcpu->kvm->lock);
2051         if (vcpu->arch.pio.in)
2052                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2053                                   vcpu->arch.pio.size,
2054                                   pd);
2055         else
2056                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2057                                    vcpu->arch.pio.size,
2058                                    pd);
2059         mutex_unlock(&vcpu->kvm->lock);
2060 }
2061
2062 static void pio_string_write(struct kvm_io_device *pio_dev,
2063                              struct kvm_vcpu *vcpu)
2064 {
2065         struct kvm_pio_request *io = &vcpu->arch.pio;
2066         void *pd = vcpu->arch.pio_data;
2067         int i;
2068
2069         mutex_lock(&vcpu->kvm->lock);
2070         for (i = 0; i < io->cur_count; i++) {
2071                 kvm_iodevice_write(pio_dev, io->port,
2072                                    io->size,
2073                                    pd);
2074                 pd += io->size;
2075         }
2076         mutex_unlock(&vcpu->kvm->lock);
2077 }
2078
2079 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2080                                                gpa_t addr)
2081 {
2082         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2083 }
2084
2085 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2086                   int size, unsigned port)
2087 {
2088         struct kvm_io_device *pio_dev;
2089
2090         vcpu->run->exit_reason = KVM_EXIT_IO;
2091         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2092         vcpu->run->io.size = vcpu->arch.pio.size = size;
2093         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2094         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2095         vcpu->run->io.port = vcpu->arch.pio.port = port;
2096         vcpu->arch.pio.in = in;
2097         vcpu->arch.pio.string = 0;
2098         vcpu->arch.pio.down = 0;
2099         vcpu->arch.pio.guest_page_offset = 0;
2100         vcpu->arch.pio.rep = 0;
2101
2102         kvm_x86_ops->cache_regs(vcpu);
2103         memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2104         kvm_x86_ops->decache_regs(vcpu);
2105
2106         kvm_x86_ops->skip_emulated_instruction(vcpu);
2107
2108         pio_dev = vcpu_find_pio_dev(vcpu, port);
2109         if (pio_dev) {
2110                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2111                 complete_pio(vcpu);
2112                 return 1;
2113         }
2114         return 0;
2115 }
2116 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2117
2118 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2119                   int size, unsigned long count, int down,
2120                   gva_t address, int rep, unsigned port)
2121 {
2122         unsigned now, in_page;
2123         int i, ret = 0;
2124         int nr_pages = 1;
2125         struct page *page;
2126         struct kvm_io_device *pio_dev;
2127
2128         vcpu->run->exit_reason = KVM_EXIT_IO;
2129         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2130         vcpu->run->io.size = vcpu->arch.pio.size = size;
2131         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2132         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2133         vcpu->run->io.port = vcpu->arch.pio.port = port;
2134         vcpu->arch.pio.in = in;
2135         vcpu->arch.pio.string = 1;
2136         vcpu->arch.pio.down = down;
2137         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2138         vcpu->arch.pio.rep = rep;
2139
2140         if (!count) {
2141                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2142                 return 1;
2143         }
2144
2145         if (!down)
2146                 in_page = PAGE_SIZE - offset_in_page(address);
2147         else
2148                 in_page = offset_in_page(address) + size;
2149         now = min(count, (unsigned long)in_page / size);
2150         if (!now) {
2151                 /*
2152                  * String I/O straddles page boundary.  Pin two guest pages
2153                  * so that we satisfy atomicity constraints.  Do just one
2154                  * transaction to avoid complexity.
2155                  */
2156                 nr_pages = 2;
2157                 now = 1;
2158         }
2159         if (down) {
2160                 /*
2161                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2162                  */
2163                 pr_unimpl(vcpu, "guest string pio down\n");
2164                 kvm_inject_gp(vcpu, 0);
2165                 return 1;
2166         }
2167         vcpu->run->io.count = now;
2168         vcpu->arch.pio.cur_count = now;
2169
2170         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2171                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2172
2173         for (i = 0; i < nr_pages; ++i) {
2174                 down_read(&vcpu->kvm->slots_lock);
2175                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2176                 vcpu->arch.pio.guest_pages[i] = page;
2177                 up_read(&vcpu->kvm->slots_lock);
2178                 if (!page) {
2179                         kvm_inject_gp(vcpu, 0);
2180                         free_pio_guest_pages(vcpu);
2181                         return 1;
2182                 }
2183         }
2184
2185         pio_dev = vcpu_find_pio_dev(vcpu, port);
2186         if (!vcpu->arch.pio.in) {
2187                 /* string PIO write */
2188                 ret = pio_copy_data(vcpu);
2189                 if (ret >= 0 && pio_dev) {
2190                         pio_string_write(pio_dev, vcpu);
2191                         complete_pio(vcpu);
2192                         if (vcpu->arch.pio.count == 0)
2193                                 ret = 1;
2194                 }
2195         } else if (pio_dev)
2196                 pr_unimpl(vcpu, "no string pio read support yet, "
2197                        "port %x size %d count %ld\n",
2198                         port, size, count);
2199
2200         return ret;
2201 }
2202 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2203
2204 int kvm_arch_init(void *opaque)
2205 {
2206         int r;
2207         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2208
2209         if (kvm_x86_ops) {
2210                 printk(KERN_ERR "kvm: already loaded the other module\n");
2211                 r = -EEXIST;
2212                 goto out;
2213         }
2214
2215         if (!ops->cpu_has_kvm_support()) {
2216                 printk(KERN_ERR "kvm: no hardware support\n");
2217                 r = -EOPNOTSUPP;
2218                 goto out;
2219         }
2220         if (ops->disabled_by_bios()) {
2221                 printk(KERN_ERR "kvm: disabled by bios\n");
2222                 r = -EOPNOTSUPP;
2223                 goto out;
2224         }
2225
2226         r = kvm_mmu_module_init();
2227         if (r)
2228                 goto out;
2229
2230         kvm_init_msr_list();
2231
2232         kvm_x86_ops = ops;
2233         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2234         return 0;
2235
2236 out:
2237         return r;
2238 }
2239
2240 void kvm_arch_exit(void)
2241 {
2242         kvm_x86_ops = NULL;
2243         kvm_mmu_module_exit();
2244 }
2245
2246 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2247 {
2248         ++vcpu->stat.halt_exits;
2249         if (irqchip_in_kernel(vcpu->kvm)) {
2250                 vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
2251                 kvm_vcpu_block(vcpu);
2252                 if (vcpu->arch.mp_state != VCPU_MP_STATE_RUNNABLE)
2253                         return -EINTR;
2254                 return 1;
2255         } else {
2256                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2257                 return 0;
2258         }
2259 }
2260 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2261
2262 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2263 {
2264         unsigned long nr, a0, a1, a2, a3, ret;
2265
2266         kvm_x86_ops->cache_regs(vcpu);
2267
2268         nr = vcpu->arch.regs[VCPU_REGS_RAX];
2269         a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2270         a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2271         a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2272         a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2273
2274         if (!is_long_mode(vcpu)) {
2275                 nr &= 0xFFFFFFFF;
2276                 a0 &= 0xFFFFFFFF;
2277                 a1 &= 0xFFFFFFFF;
2278                 a2 &= 0xFFFFFFFF;
2279                 a3 &= 0xFFFFFFFF;
2280         }
2281
2282         switch (nr) {
2283         case KVM_HC_VAPIC_POLL_IRQ:
2284                 ret = 0;
2285                 break;
2286         default:
2287                 ret = -KVM_ENOSYS;
2288                 break;
2289         }
2290         vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2291         kvm_x86_ops->decache_regs(vcpu);
2292         return 0;
2293 }
2294 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2295
2296 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2297 {
2298         char instruction[3];
2299         int ret = 0;
2300
2301
2302         /*
2303          * Blow out the MMU to ensure that no other VCPU has an active mapping
2304          * to ensure that the updated hypercall appears atomically across all
2305          * VCPUs.
2306          */
2307         kvm_mmu_zap_all(vcpu->kvm);
2308
2309         kvm_x86_ops->cache_regs(vcpu);
2310         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2311         if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2312             != X86EMUL_CONTINUE)
2313                 ret = -EFAULT;
2314
2315         return ret;
2316 }
2317
2318 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2319 {
2320         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2321 }
2322
2323 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2324 {
2325         struct descriptor_table dt = { limit, base };
2326
2327         kvm_x86_ops->set_gdt(vcpu, &dt);
2328 }
2329
2330 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2331 {
2332         struct descriptor_table dt = { limit, base };
2333
2334         kvm_x86_ops->set_idt(vcpu, &dt);
2335 }
2336
2337 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2338                    unsigned long *rflags)
2339 {
2340         lmsw(vcpu, msw);
2341         *rflags = kvm_x86_ops->get_rflags(vcpu);
2342 }
2343
2344 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2345 {
2346         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2347         switch (cr) {
2348         case 0:
2349                 return vcpu->arch.cr0;
2350         case 2:
2351                 return vcpu->arch.cr2;
2352         case 3:
2353                 return vcpu->arch.cr3;
2354         case 4:
2355                 return vcpu->arch.cr4;
2356         case 8:
2357                 return get_cr8(vcpu);
2358         default:
2359                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2360                 return 0;
2361         }
2362 }
2363
2364 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2365                      unsigned long *rflags)
2366 {
2367         switch (cr) {
2368         case 0:
2369                 set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2370                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2371                 break;
2372         case 2:
2373                 vcpu->arch.cr2 = val;
2374                 break;
2375         case 3:
2376                 set_cr3(vcpu, val);
2377                 break;
2378         case 4:
2379                 set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2380                 break;
2381         case 8:
2382                 set_cr8(vcpu, val & 0xfUL);
2383                 break;
2384         default:
2385                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2386         }
2387 }
2388
2389 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2390 {
2391         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2392         int j, nent = vcpu->arch.cpuid_nent;
2393
2394         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2395         /* when no next entry is found, the current entry[i] is reselected */
2396         for (j = i + 1; j == i; j = (j + 1) % nent) {
2397                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2398                 if (ej->function == e->function) {
2399                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2400                         return j;
2401                 }
2402         }
2403         return 0; /* silence gcc, even though control never reaches here */
2404 }
2405
2406 /* find an entry with matching function, matching index (if needed), and that
2407  * should be read next (if it's stateful) */
2408 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2409         u32 function, u32 index)
2410 {
2411         if (e->function != function)
2412                 return 0;
2413         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2414                 return 0;
2415         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2416                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2417                 return 0;
2418         return 1;
2419 }
2420
2421 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2422 {
2423         int i;
2424         u32 function, index;
2425         struct kvm_cpuid_entry2 *e, *best;
2426
2427         kvm_x86_ops->cache_regs(vcpu);
2428         function = vcpu->arch.regs[VCPU_REGS_RAX];
2429         index = vcpu->arch.regs[VCPU_REGS_RCX];
2430         vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2431         vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2432         vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2433         vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2434         best = NULL;
2435         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2436                 e = &vcpu->arch.cpuid_entries[i];
2437                 if (is_matching_cpuid_entry(e, function, index)) {
2438                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2439                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2440                         best = e;
2441                         break;
2442                 }
2443                 /*
2444                  * Both basic or both extended?
2445                  */
2446                 if (((e->function ^ function) & 0x80000000) == 0)
2447                         if (!best || e->function > best->function)
2448                                 best = e;
2449         }
2450         if (best) {
2451                 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2452                 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2453                 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2454                 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2455         }
2456         kvm_x86_ops->decache_regs(vcpu);
2457         kvm_x86_ops->skip_emulated_instruction(vcpu);
2458 }
2459 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2460
2461 /*
2462  * Check if userspace requested an interrupt window, and that the
2463  * interrupt window is open.
2464  *
2465  * No need to exit to userspace if we already have an interrupt queued.
2466  */
2467 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2468                                           struct kvm_run *kvm_run)
2469 {
2470         return (!vcpu->arch.irq_summary &&
2471                 kvm_run->request_interrupt_window &&
2472                 vcpu->arch.interrupt_window_open &&
2473                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2474 }
2475
2476 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2477                               struct kvm_run *kvm_run)
2478 {
2479         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2480         kvm_run->cr8 = get_cr8(vcpu);
2481         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2482         if (irqchip_in_kernel(vcpu->kvm))
2483                 kvm_run->ready_for_interrupt_injection = 1;
2484         else
2485                 kvm_run->ready_for_interrupt_injection =
2486                                         (vcpu->arch.interrupt_window_open &&
2487                                          vcpu->arch.irq_summary == 0);
2488 }
2489
2490 static void vapic_enter(struct kvm_vcpu *vcpu)
2491 {
2492         struct kvm_lapic *apic = vcpu->arch.apic;
2493         struct page *page;
2494
2495         if (!apic || !apic->vapic_addr)
2496                 return;
2497
2498         down_read(&current->mm->mmap_sem);
2499         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2500         up_read(&current->mm->mmap_sem);
2501
2502         vcpu->arch.apic->vapic_page = page;
2503 }
2504
2505 static void vapic_exit(struct kvm_vcpu *vcpu)
2506 {
2507         struct kvm_lapic *apic = vcpu->arch.apic;
2508
2509         if (!apic || !apic->vapic_addr)
2510                 return;
2511
2512         kvm_release_page_dirty(apic->vapic_page);
2513         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2514 }
2515
2516 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2517 {
2518         int r;
2519
2520         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2521                 pr_debug("vcpu %d received sipi with vector # %x\n",
2522                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
2523                 kvm_lapic_reset(vcpu);
2524                 r = kvm_x86_ops->vcpu_reset(vcpu);
2525                 if (r)
2526                         return r;
2527                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
2528         }
2529
2530         vapic_enter(vcpu);
2531
2532 preempted:
2533         if (vcpu->guest_debug.enabled)
2534                 kvm_x86_ops->guest_debug_pre(vcpu);
2535
2536 again:
2537         r = kvm_mmu_reload(vcpu);
2538         if (unlikely(r))
2539                 goto out;
2540
2541         if (vcpu->requests) {
2542                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2543                         __kvm_migrate_apic_timer(vcpu);
2544                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2545                                        &vcpu->requests)) {
2546                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2547                         r = 0;
2548                         goto out;
2549                 }
2550         }
2551
2552         kvm_inject_pending_timer_irqs(vcpu);
2553
2554         preempt_disable();
2555
2556         kvm_x86_ops->prepare_guest_switch(vcpu);
2557         kvm_load_guest_fpu(vcpu);
2558
2559         local_irq_disable();
2560
2561         if (need_resched()) {
2562                 local_irq_enable();
2563                 preempt_enable();
2564                 r = 1;
2565                 goto out;
2566         }
2567
2568         if (signal_pending(current)) {
2569                 local_irq_enable();
2570                 preempt_enable();
2571                 r = -EINTR;
2572                 kvm_run->exit_reason = KVM_EXIT_INTR;
2573                 ++vcpu->stat.signal_exits;
2574                 goto out;
2575         }
2576
2577         if (vcpu->arch.exception.pending)
2578                 __queue_exception(vcpu);
2579         else if (irqchip_in_kernel(vcpu->kvm))
2580                 kvm_x86_ops->inject_pending_irq(vcpu);
2581         else
2582                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2583
2584         kvm_lapic_sync_to_vapic(vcpu);
2585
2586         vcpu->guest_mode = 1;
2587         kvm_guest_enter();
2588
2589         if (vcpu->requests)
2590                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2591                         kvm_x86_ops->tlb_flush(vcpu);
2592
2593         kvm_x86_ops->run(vcpu, kvm_run);
2594
2595         vcpu->guest_mode = 0;
2596         local_irq_enable();
2597
2598         ++vcpu->stat.exits;
2599
2600         /*
2601          * We must have an instruction between local_irq_enable() and
2602          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2603          * the interrupt shadow.  The stat.exits increment will do nicely.
2604          * But we need to prevent reordering, hence this barrier():
2605          */
2606         barrier();
2607
2608         kvm_guest_exit();
2609
2610         preempt_enable();
2611
2612         /*
2613          * Profile KVM exit RIPs:
2614          */
2615         if (unlikely(prof_on == KVM_PROFILING)) {
2616                 kvm_x86_ops->cache_regs(vcpu);
2617                 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2618         }
2619
2620         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2621                 vcpu->arch.exception.pending = false;
2622
2623         kvm_lapic_sync_from_vapic(vcpu);
2624
2625         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2626
2627         if (r > 0) {
2628                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2629                         r = -EINTR;
2630                         kvm_run->exit_reason = KVM_EXIT_INTR;
2631                         ++vcpu->stat.request_irq_exits;
2632                         goto out;
2633                 }
2634                 if (!need_resched())
2635                         goto again;
2636         }
2637
2638 out:
2639         if (r > 0) {
2640                 kvm_resched(vcpu);
2641                 goto preempted;
2642         }
2643
2644         post_kvm_run_save(vcpu, kvm_run);
2645
2646         vapic_exit(vcpu);
2647
2648         return r;
2649 }
2650
2651 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2652 {
2653         int r;
2654         sigset_t sigsaved;
2655
2656         vcpu_load(vcpu);
2657
2658         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2659                 kvm_vcpu_block(vcpu);
2660                 vcpu_put(vcpu);
2661                 return -EAGAIN;
2662         }
2663
2664         if (vcpu->sigset_active)
2665                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2666
2667         /* re-sync apic's tpr */
2668         if (!irqchip_in_kernel(vcpu->kvm))
2669                 set_cr8(vcpu, kvm_run->cr8);
2670
2671         if (vcpu->arch.pio.cur_count) {
2672                 r = complete_pio(vcpu);
2673                 if (r)
2674                         goto out;
2675         }
2676 #if CONFIG_HAS_IOMEM
2677         if (vcpu->mmio_needed) {
2678                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2679                 vcpu->mmio_read_completed = 1;
2680                 vcpu->mmio_needed = 0;
2681                 r = emulate_instruction(vcpu, kvm_run,
2682                                         vcpu->arch.mmio_fault_cr2, 0,
2683                                         EMULTYPE_NO_DECODE);
2684                 if (r == EMULATE_DO_MMIO) {
2685                         /*
2686                          * Read-modify-write.  Back to userspace.
2687                          */
2688                         r = 0;
2689                         goto out;
2690                 }
2691         }
2692 #endif
2693         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2694                 kvm_x86_ops->cache_regs(vcpu);
2695                 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2696                 kvm_x86_ops->decache_regs(vcpu);
2697         }
2698
2699         r = __vcpu_run(vcpu, kvm_run);
2700
2701 out:
2702         if (vcpu->sigset_active)
2703                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2704
2705         vcpu_put(vcpu);
2706         return r;
2707 }
2708
2709 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2710 {
2711         vcpu_load(vcpu);
2712
2713         kvm_x86_ops->cache_regs(vcpu);
2714
2715         regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2716         regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2717         regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2718         regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2719         regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2720         regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2721         regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2722         regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2723 #ifdef CONFIG_X86_64
2724         regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2725         regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2726         regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2727         regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2728         regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2729         regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
2730         regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
2731         regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
2732 #endif
2733
2734         regs->rip = vcpu->arch.rip;
2735         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2736
2737         /*
2738          * Don't leak debug flags in case they were set for guest debugging
2739          */
2740         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2741                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2742
2743         vcpu_put(vcpu);
2744
2745         return 0;
2746 }
2747
2748 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2749 {
2750         vcpu_load(vcpu);
2751
2752         vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
2753         vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
2754         vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
2755         vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
2756         vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
2757         vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
2758         vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
2759         vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
2760 #ifdef CONFIG_X86_64
2761         vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
2762         vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
2763         vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
2764         vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
2765         vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
2766         vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
2767         vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
2768         vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
2769 #endif
2770
2771         vcpu->arch.rip = regs->rip;
2772         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2773
2774         kvm_x86_ops->decache_regs(vcpu);
2775
2776         vcpu_put(vcpu);
2777
2778         return 0;
2779 }
2780
2781 static void get_segment(struct kvm_vcpu *vcpu,
2782                         struct kvm_segment *var, int seg)
2783 {
2784         return kvm_x86_ops->get_segment(vcpu, var, seg);
2785 }
2786
2787 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2788 {
2789         struct kvm_segment cs;
2790
2791         get_segment(vcpu, &cs, VCPU_SREG_CS);
2792         *db = cs.db;
2793         *l = cs.l;
2794 }
2795 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2796
2797 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2798                                   struct kvm_sregs *sregs)
2799 {
2800         struct descriptor_table dt;
2801         int pending_vec;
2802
2803         vcpu_load(vcpu);
2804
2805         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2806         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2807         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2808         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2809         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2810         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2811
2812         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2813         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2814
2815         kvm_x86_ops->get_idt(vcpu, &dt);
2816         sregs->idt.limit = dt.limit;
2817         sregs->idt.base = dt.base;
2818         kvm_x86_ops->get_gdt(vcpu, &dt);
2819         sregs->gdt.limit = dt.limit;
2820         sregs->gdt.base = dt.base;
2821
2822         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2823         sregs->cr0 = vcpu->arch.cr0;
2824         sregs->cr2 = vcpu->arch.cr2;
2825         sregs->cr3 = vcpu->arch.cr3;
2826         sregs->cr4 = vcpu->arch.cr4;
2827         sregs->cr8 = get_cr8(vcpu);
2828         sregs->efer = vcpu->arch.shadow_efer;
2829         sregs->apic_base = kvm_get_apic_base(vcpu);
2830
2831         if (irqchip_in_kernel(vcpu->kvm)) {
2832                 memset(sregs->interrupt_bitmap, 0,
2833                        sizeof sregs->interrupt_bitmap);
2834                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2835                 if (pending_vec >= 0)
2836                         set_bit(pending_vec,
2837                                 (unsigned long *)sregs->interrupt_bitmap);
2838         } else
2839                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
2840                        sizeof sregs->interrupt_bitmap);
2841
2842         vcpu_put(vcpu);
2843
2844         return 0;
2845 }
2846
2847 static void set_segment(struct kvm_vcpu *vcpu,
2848                         struct kvm_segment *var, int seg)
2849 {
2850         return kvm_x86_ops->set_segment(vcpu, var, seg);
2851 }
2852
2853 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2854                                   struct kvm_sregs *sregs)
2855 {
2856         int mmu_reset_needed = 0;
2857         int i, pending_vec, max_bits;
2858         struct descriptor_table dt;
2859
2860         vcpu_load(vcpu);
2861
2862         dt.limit = sregs->idt.limit;
2863         dt.base = sregs->idt.base;
2864         kvm_x86_ops->set_idt(vcpu, &dt);
2865         dt.limit = sregs->gdt.limit;
2866         dt.base = sregs->gdt.base;
2867         kvm_x86_ops->set_gdt(vcpu, &dt);
2868
2869         vcpu->arch.cr2 = sregs->cr2;
2870         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
2871         vcpu->arch.cr3 = sregs->cr3;
2872
2873         set_cr8(vcpu, sregs->cr8);
2874
2875         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
2876         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2877         kvm_set_apic_base(vcpu, sregs->apic_base);
2878
2879         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2880
2881         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
2882         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2883         vcpu->arch.cr0 = sregs->cr0;
2884
2885         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
2886         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2887         if (!is_long_mode(vcpu) && is_pae(vcpu))
2888                 load_pdptrs(vcpu, vcpu->arch.cr3);
2889
2890         if (mmu_reset_needed)
2891                 kvm_mmu_reset_context(vcpu);
2892
2893         if (!irqchip_in_kernel(vcpu->kvm)) {
2894                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
2895                        sizeof vcpu->arch.irq_pending);
2896                 vcpu->arch.irq_summary = 0;
2897                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
2898                         if (vcpu->arch.irq_pending[i])
2899                                 __set_bit(i, &vcpu->arch.irq_summary);
2900         } else {
2901                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2902                 pending_vec = find_first_bit(
2903                         (const unsigned long *)sregs->interrupt_bitmap,
2904                         max_bits);
2905                 /* Only pending external irq is handled here */
2906                 if (pending_vec < max_bits) {
2907                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2908                         pr_debug("Set back pending irq %d\n",
2909                                  pending_vec);
2910                 }
2911         }
2912
2913         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2914         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2915         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2916         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2917         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2918         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2919
2920         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2921         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2922
2923         vcpu_put(vcpu);
2924
2925         return 0;
2926 }
2927
2928 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2929                                     struct kvm_debug_guest *dbg)
2930 {
2931         int r;
2932
2933         vcpu_load(vcpu);
2934
2935         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2936
2937         vcpu_put(vcpu);
2938
2939         return r;
2940 }
2941
2942 /*
2943  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2944  * we have asm/x86/processor.h
2945  */
2946 struct fxsave {
2947         u16     cwd;
2948         u16     swd;
2949         u16     twd;
2950         u16     fop;
2951         u64     rip;
2952         u64     rdp;
2953         u32     mxcsr;
2954         u32     mxcsr_mask;
2955         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2956 #ifdef CONFIG_X86_64
2957         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2958 #else
2959         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2960 #endif
2961 };
2962
2963 /*
2964  * Translate a guest virtual address to a guest physical address.
2965  */
2966 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2967                                     struct kvm_translation *tr)
2968 {
2969         unsigned long vaddr = tr->linear_address;
2970         gpa_t gpa;
2971
2972         vcpu_load(vcpu);
2973         down_read(&vcpu->kvm->slots_lock);
2974         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
2975         up_read(&vcpu->kvm->slots_lock);
2976         tr->physical_address = gpa;
2977         tr->valid = gpa != UNMAPPED_GVA;
2978         tr->writeable = 1;
2979         tr->usermode = 0;
2980         vcpu_put(vcpu);
2981
2982         return 0;
2983 }
2984
2985 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2986 {
2987         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
2988
2989         vcpu_load(vcpu);
2990
2991         memcpy(fpu->fpr, fxsave->st_space, 128);
2992         fpu->fcw = fxsave->cwd;
2993         fpu->fsw = fxsave->swd;
2994         fpu->ftwx = fxsave->twd;
2995         fpu->last_opcode = fxsave->fop;
2996         fpu->last_ip = fxsave->rip;
2997         fpu->last_dp = fxsave->rdp;
2998         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2999
3000         vcpu_put(vcpu);
3001
3002         return 0;
3003 }
3004
3005 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3006 {
3007         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3008
3009         vcpu_load(vcpu);
3010
3011         memcpy(fxsave->st_space, fpu->fpr, 128);
3012         fxsave->cwd = fpu->fcw;
3013         fxsave->swd = fpu->fsw;
3014         fxsave->twd = fpu->ftwx;
3015         fxsave->fop = fpu->last_opcode;
3016         fxsave->rip = fpu->last_ip;
3017         fxsave->rdp = fpu->last_dp;
3018         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3019
3020         vcpu_put(vcpu);
3021
3022         return 0;
3023 }
3024
3025 void fx_init(struct kvm_vcpu *vcpu)
3026 {
3027         unsigned after_mxcsr_mask;
3028
3029         /* Initialize guest FPU by resetting ours and saving into guest's */
3030         preempt_disable();
3031         fx_save(&vcpu->arch.host_fx_image);
3032         fpu_init();
3033         fx_save(&vcpu->arch.guest_fx_image);
3034         fx_restore(&vcpu->arch.host_fx_image);
3035         preempt_enable();
3036
3037         vcpu->arch.cr0 |= X86_CR0_ET;
3038         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3039         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3040         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3041                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3042 }
3043 EXPORT_SYMBOL_GPL(fx_init);
3044
3045 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3046 {
3047         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3048                 return;
3049
3050         vcpu->guest_fpu_loaded = 1;
3051         fx_save(&vcpu->arch.host_fx_image);
3052         fx_restore(&vcpu->arch.guest_fx_image);
3053 }
3054 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3055
3056 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3057 {
3058         if (!vcpu->guest_fpu_loaded)
3059                 return;
3060
3061         vcpu->guest_fpu_loaded = 0;
3062         fx_save(&vcpu->arch.guest_fx_image);
3063         fx_restore(&vcpu->arch.host_fx_image);
3064         ++vcpu->stat.fpu_reload;
3065 }
3066 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3067
3068 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3069 {
3070         kvm_x86_ops->vcpu_free(vcpu);
3071 }
3072
3073 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3074                                                 unsigned int id)
3075 {
3076         return kvm_x86_ops->vcpu_create(kvm, id);
3077 }
3078
3079 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3080 {
3081         int r;
3082
3083         /* We do fxsave: this must be aligned. */
3084         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3085
3086         vcpu_load(vcpu);
3087         r = kvm_arch_vcpu_reset(vcpu);
3088         if (r == 0)
3089                 r = kvm_mmu_setup(vcpu);
3090         vcpu_put(vcpu);
3091         if (r < 0)
3092                 goto free_vcpu;
3093
3094         return 0;
3095 free_vcpu:
3096         kvm_x86_ops->vcpu_free(vcpu);
3097         return r;
3098 }
3099
3100 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3101 {
3102         vcpu_load(vcpu);
3103         kvm_mmu_unload(vcpu);
3104         vcpu_put(vcpu);
3105
3106         kvm_x86_ops->vcpu_free(vcpu);
3107 }
3108
3109 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3110 {
3111         return kvm_x86_ops->vcpu_reset(vcpu);
3112 }
3113
3114 void kvm_arch_hardware_enable(void *garbage)
3115 {
3116         kvm_x86_ops->hardware_enable(garbage);
3117 }
3118
3119 void kvm_arch_hardware_disable(void *garbage)
3120 {
3121         kvm_x86_ops->hardware_disable(garbage);
3122 }
3123
3124 int kvm_arch_hardware_setup(void)
3125 {
3126         return kvm_x86_ops->hardware_setup();
3127 }
3128
3129 void kvm_arch_hardware_unsetup(void)
3130 {
3131         kvm_x86_ops->hardware_unsetup();
3132 }
3133
3134 void kvm_arch_check_processor_compat(void *rtn)
3135 {
3136         kvm_x86_ops->check_processor_compatibility(rtn);
3137 }
3138
3139 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3140 {
3141         struct page *page;
3142         struct kvm *kvm;
3143         int r;
3144
3145         BUG_ON(vcpu->kvm == NULL);
3146         kvm = vcpu->kvm;
3147
3148         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3149         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3150                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
3151         else
3152                 vcpu->arch.mp_state = VCPU_MP_STATE_UNINITIALIZED;
3153
3154         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3155         if (!page) {
3156                 r = -ENOMEM;
3157                 goto fail;
3158         }
3159         vcpu->arch.pio_data = page_address(page);
3160
3161         r = kvm_mmu_create(vcpu);
3162         if (r < 0)
3163                 goto fail_free_pio_data;
3164
3165         if (irqchip_in_kernel(kvm)) {
3166                 r = kvm_create_lapic(vcpu);
3167                 if (r < 0)
3168                         goto fail_mmu_destroy;
3169         }
3170
3171         return 0;
3172
3173 fail_mmu_destroy:
3174         kvm_mmu_destroy(vcpu);
3175 fail_free_pio_data:
3176         free_page((unsigned long)vcpu->arch.pio_data);
3177 fail:
3178         return r;
3179 }
3180
3181 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3182 {
3183         kvm_free_lapic(vcpu);
3184         kvm_mmu_destroy(vcpu);
3185         free_page((unsigned long)vcpu->arch.pio_data);
3186 }
3187
3188 struct  kvm *kvm_arch_create_vm(void)
3189 {
3190         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3191
3192         if (!kvm)
3193                 return ERR_PTR(-ENOMEM);
3194
3195         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3196
3197         return kvm;
3198 }
3199
3200 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3201 {
3202         vcpu_load(vcpu);
3203         kvm_mmu_unload(vcpu);
3204         vcpu_put(vcpu);
3205 }
3206
3207 static void kvm_free_vcpus(struct kvm *kvm)
3208 {
3209         unsigned int i;
3210
3211         /*
3212          * Unpin any mmu pages first.
3213          */
3214         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3215                 if (kvm->vcpus[i])
3216                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3217         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3218                 if (kvm->vcpus[i]) {
3219                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3220                         kvm->vcpus[i] = NULL;
3221                 }
3222         }
3223
3224 }
3225
3226 void kvm_arch_destroy_vm(struct kvm *kvm)
3227 {
3228         kfree(kvm->arch.vpic);
3229         kfree(kvm->arch.vioapic);
3230         kvm_free_vcpus(kvm);
3231         kvm_free_physmem(kvm);
3232         kfree(kvm);
3233 }
3234
3235 int kvm_arch_set_memory_region(struct kvm *kvm,
3236                                 struct kvm_userspace_memory_region *mem,
3237                                 struct kvm_memory_slot old,
3238                                 int user_alloc)
3239 {
3240         int npages = mem->memory_size >> PAGE_SHIFT;
3241         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3242
3243         /*To keep backward compatibility with older userspace,
3244          *x86 needs to hanlde !user_alloc case.
3245          */
3246         if (!user_alloc) {
3247                 if (npages && !old.rmap) {
3248                         down_write(&current->mm->mmap_sem);
3249                         memslot->userspace_addr = do_mmap(NULL, 0,
3250                                                      npages * PAGE_SIZE,
3251                                                      PROT_READ | PROT_WRITE,
3252                                                      MAP_SHARED | MAP_ANONYMOUS,
3253                                                      0);
3254                         up_write(&current->mm->mmap_sem);
3255
3256                         if (IS_ERR((void *)memslot->userspace_addr))
3257                                 return PTR_ERR((void *)memslot->userspace_addr);
3258                 } else {
3259                         if (!old.user_alloc && old.rmap) {
3260                                 int ret;
3261
3262                                 down_write(&current->mm->mmap_sem);
3263                                 ret = do_munmap(current->mm, old.userspace_addr,
3264                                                 old.npages * PAGE_SIZE);
3265                                 up_write(&current->mm->mmap_sem);
3266                                 if (ret < 0)
3267                                         printk(KERN_WARNING
3268                                        "kvm_vm_ioctl_set_memory_region: "
3269                                        "failed to munmap memory\n");
3270                         }
3271                 }
3272         }
3273
3274         if (!kvm->arch.n_requested_mmu_pages) {
3275                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3276                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3277         }
3278
3279         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3280         kvm_flush_remote_tlbs(kvm);
3281
3282         return 0;
3283 }
3284
3285 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3286 {
3287         return vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE
3288                || vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED;
3289 }
3290
3291 static void vcpu_kick_intr(void *info)
3292 {
3293 #ifdef DEBUG
3294         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
3295         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
3296 #endif
3297 }
3298
3299 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3300 {
3301         int ipi_pcpu = vcpu->cpu;
3302
3303         if (waitqueue_active(&vcpu->wq)) {
3304                 wake_up_interruptible(&vcpu->wq);
3305                 ++vcpu->stat.halt_wakeup;
3306         }
3307         if (vcpu->guest_mode)
3308                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);
3309 }