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 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  45
  46 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  47 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  48
  49 struct kvm_x86_ops *kvm_x86_ops;
  50
  51 struct kvm_stats_debugfs_item debugfs_entries[] = {
  52         { "pf_fixed", VCPU_STAT(pf_fixed) },
  53         { "pf_guest", VCPU_STAT(pf_guest) },
  54         { "tlb_flush", VCPU_STAT(tlb_flush) },
  55         { "invlpg", VCPU_STAT(invlpg) },
  56         { "exits", VCPU_STAT(exits) },
  57         { "io_exits", VCPU_STAT(io_exits) },
  58         { "mmio_exits", VCPU_STAT(mmio_exits) },
  59         { "signal_exits", VCPU_STAT(signal_exits) },
  60         { "irq_window", VCPU_STAT(irq_window_exits) },
  61         { "halt_exits", VCPU_STAT(halt_exits) },
  62         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  63         { "request_irq", VCPU_STAT(request_irq_exits) },
  64         { "irq_exits", VCPU_STAT(irq_exits) },
  65         { "host_state_reload", VCPU_STAT(host_state_reload) },
  66         { "efer_reload", VCPU_STAT(efer_reload) },
  67         { "fpu_reload", VCPU_STAT(fpu_reload) },
  68         { "insn_emulation", VCPU_STAT(insn_emulation) },
  69         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
  70         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
  71         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
  72         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
  73         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
  74         { "mmu_flooded", VM_STAT(mmu_flooded) },
  75         { "mmu_recycled", VM_STAT(mmu_recycled) },
  76         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
  77         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
  78         { NULL }
  79 };
  80
  81
  82 unsigned long segment_base(u16 selector)
  83 {
  84         struct descriptor_table gdt;
  85         struct segment_descriptor *d;
  86         unsigned long table_base;
  87         unsigned long v;
  88
  89         if (selector == 0)
  90                 return 0;
  91
  92         asm("sgdt %0" : "=m"(gdt));
  93         table_base = gdt.base;
  94
  95         if (selector & 4) {           /* from ldt */
  96                 u16 ldt_selector;
  97
  98                 asm("sldt %0" : "=g"(ldt_selector));
  99                 table_base = segment_base(ldt_selector);
 100         }
 101         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 102         v = d->base_low | ((unsigned long)d->base_mid << 16) |
 103                 ((unsigned long)d->base_high << 24);
 104 #ifdef CONFIG_X86_64
 105         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
 106                 v |= ((unsigned long) \
 107                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
 108 #endif
 109         return v;
 110 }
 111 EXPORT_SYMBOL_GPL(segment_base);
 112
 113 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 114 {
 115         if (irqchip_in_kernel(vcpu->kvm))
 116                 return vcpu->arch.apic_base;
 117         else
 118                 return vcpu->arch.apic_base;
 119 }
 120 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 121
 122 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 123 {
 124         /* TODO: reserve bits check */
 125         if (irqchip_in_kernel(vcpu->kvm))
 126                 kvm_lapic_set_base(vcpu, data);
 127         else
 128                 vcpu->arch.apic_base = data;
 129 }
 130 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 131
 132 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
 133 {
 134         WARN_ON(vcpu->arch.exception.pending);
 135         vcpu->arch.exception.pending = true;
 136         vcpu->arch.exception.has_error_code = false;
 137         vcpu->arch.exception.nr = nr;
 138 }
 139 EXPORT_SYMBOL_GPL(kvm_queue_exception);
 140
 141 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
 142                            u32 error_code)
 143 {
 144         ++vcpu->stat.pf_guest;
 145         if (vcpu->arch.exception.pending && vcpu->arch.exception.nr == PF_VECTOR) {
 146                 printk(KERN_DEBUG "kvm: inject_page_fault:"
 147                        " double fault 0x%lx\n", addr);
 148                 vcpu->arch.exception.nr = DF_VECTOR;
 149                 vcpu->arch.exception.error_code = 0;
 150                 return;
 151         }
 152         vcpu->arch.cr2 = addr;
 153         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
 154 }
 155
 156 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
 157 {
 158         WARN_ON(vcpu->arch.exception.pending);
 159         vcpu->arch.exception.pending = true;
 160         vcpu->arch.exception.has_error_code = true;
 161         vcpu->arch.exception.nr = nr;
 162         vcpu->arch.exception.error_code = error_code;
 163 }
 164 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
 165
 166 static void __queue_exception(struct kvm_vcpu *vcpu)
 167 {
 168         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
 169                                      vcpu->arch.exception.has_error_code,
 170                                      vcpu->arch.exception.error_code);
 171 }
 172
 173 /*
 174  * Load the pae pdptrs.  Return true is they are all valid.
 175  */
 176 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 177 {
 178         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 179         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 180         int i;
 181         int ret;
 182         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 183
 184         down_read(&current->mm->mmap_sem);
 185         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 186                                   offset * sizeof(u64), sizeof(pdpte));
 187         if (ret < 0) {
 188                 ret = 0;
 189                 goto out;
 190         }
 191         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 192                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 193                         ret = 0;
 194                         goto out;
 195                 }
 196         }
 197         ret = 1;
 198
 199         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 200 out:
 201         up_read(&current->mm->mmap_sem);
 202
 203         return ret;
 204 }
 205
 206 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
 207 {
 208         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 209         bool changed = true;
 210         int r;
 211
 212         if (is_long_mode(vcpu) || !is_pae(vcpu))
 213                 return false;
 214
 215         down_read(&current->mm->mmap_sem);
 216         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
 217         if (r < 0)
 218                 goto out;
 219         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 220 out:
 221         up_read(&current->mm->mmap_sem);
 222
 223         return changed;
 224 }
 225
 226 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 227 {
 228         if (cr0 & CR0_RESERVED_BITS) {
 229                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 230                        cr0, vcpu->arch.cr0);
 231                 kvm_inject_gp(vcpu, 0);
 232                 return;
 233         }
 234
 235         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 236                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 237                 kvm_inject_gp(vcpu, 0);
 238                 return;
 239         }
 240
 241         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 242                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 243                        "and a clear PE flag\n");
 244                 kvm_inject_gp(vcpu, 0);
 245                 return;
 246         }
 247
 248         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 249 #ifdef CONFIG_X86_64
 250                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
 251                         int cs_db, cs_l;
 252
 253                         if (!is_pae(vcpu)) {
 254                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 255                                        "in long mode while PAE is disabled\n");
 256                                 kvm_inject_gp(vcpu, 0);
 257                                 return;
 258                         }
 259                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 260                         if (cs_l) {
 261                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 262                                        "in long mode while CS.L == 1\n");
 263                                 kvm_inject_gp(vcpu, 0);
 264                                 return;
 265
 266                         }
 267                 } else
 268 #endif
 269                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 270                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 271                                "reserved bits\n");
 272                         kvm_inject_gp(vcpu, 0);
 273                         return;
 274                 }
 275
 276         }
 277
 278         kvm_x86_ops->set_cr0(vcpu, cr0);
 279         vcpu->arch.cr0 = cr0;
 280
 281         kvm_mmu_reset_context(vcpu);
 282         return;
 283 }
 284 EXPORT_SYMBOL_GPL(set_cr0);
 285
 286 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 287 {
 288         set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
 289 }
 290 EXPORT_SYMBOL_GPL(lmsw);
 291
 292 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 293 {
 294         if (cr4 & CR4_RESERVED_BITS) {
 295                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 296                 kvm_inject_gp(vcpu, 0);
 297                 return;
 298         }
 299
 300         if (is_long_mode(vcpu)) {
 301                 if (!(cr4 & X86_CR4_PAE)) {
 302                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 303                                "in long mode\n");
 304                         kvm_inject_gp(vcpu, 0);
 305                         return;
 306                 }
 307         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 308                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
 309                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 310                 kvm_inject_gp(vcpu, 0);
 311                 return;
 312         }
 313
 314         if (cr4 & X86_CR4_VMXE) {
 315                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 316                 kvm_inject_gp(vcpu, 0);
 317                 return;
 318         }
 319         kvm_x86_ops->set_cr4(vcpu, cr4);
 320         vcpu->arch.cr4 = cr4;
 321         kvm_mmu_reset_context(vcpu);
 322 }
 323 EXPORT_SYMBOL_GPL(set_cr4);
 324
 325 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 326 {
 327         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
 328                 kvm_mmu_flush_tlb(vcpu);
 329                 return;
 330         }
 331
 332         if (is_long_mode(vcpu)) {
 333                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 334                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 335                         kvm_inject_gp(vcpu, 0);
 336                         return;
 337                 }
 338         } else {
 339                 if (is_pae(vcpu)) {
 340                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 341                                 printk(KERN_DEBUG
 342                                        "set_cr3: #GP, reserved bits\n");
 343                                 kvm_inject_gp(vcpu, 0);
 344                                 return;
 345                         }
 346                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 347                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 348                                        "reserved bits\n");
 349                                 kvm_inject_gp(vcpu, 0);
 350                                 return;
 351                         }
 352                 }
 353                 /*
 354                  * We don't check reserved bits in nonpae mode, because
 355                  * this isn't enforced, and VMware depends on this.
 356                  */
 357         }
 358
 359         down_read(&current->mm->mmap_sem);
 360         /*
 361          * Does the new cr3 value map to physical memory? (Note, we
 362          * catch an invalid cr3 even in real-mode, because it would
 363          * cause trouble later on when we turn on paging anyway.)
 364          *
 365          * A real CPU would silently accept an invalid cr3 and would
 366          * attempt to use it - with largely undefined (and often hard
 367          * to debug) behavior on the guest side.
 368          */
 369         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 370                 kvm_inject_gp(vcpu, 0);
 371         else {
 372                 vcpu->arch.cr3 = cr3;
 373                 vcpu->arch.mmu.new_cr3(vcpu);
 374         }
 375         up_read(&current->mm->mmap_sem);
 376 }
 377 EXPORT_SYMBOL_GPL(set_cr3);
 378
 379 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 380 {
 381         if (cr8 & CR8_RESERVED_BITS) {
 382                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 383                 kvm_inject_gp(vcpu, 0);
 384                 return;
 385         }
 386         if (irqchip_in_kernel(vcpu->kvm))
 387                 kvm_lapic_set_tpr(vcpu, cr8);
 388         else
 389                 vcpu->arch.cr8 = cr8;
 390 }
 391 EXPORT_SYMBOL_GPL(set_cr8);
 392
 393 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 394 {
 395         if (irqchip_in_kernel(vcpu->kvm))
 396                 return kvm_lapic_get_cr8(vcpu);
 397         else
 398                 return vcpu->arch.cr8;
 399 }
 400 EXPORT_SYMBOL_GPL(get_cr8);
 401
 402 /*
 403  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
 404  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
 405  *
 406  * This list is modified at module load time to reflect the
 407  * capabilities of the host cpu.
 408  */
 409 static u32 msrs_to_save[] = {
 410         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 411         MSR_K6_STAR,
 412 #ifdef CONFIG_X86_64
 413         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
 414 #endif
 415         MSR_IA32_TIME_STAMP_COUNTER,
 416 };
 417
 418 static unsigned num_msrs_to_save;
 419
 420 static u32 emulated_msrs[] = {
 421         MSR_IA32_MISC_ENABLE,
 422 };
 423
 424 #ifdef CONFIG_X86_64
 425
 426 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
 427 {
 428         if (efer & EFER_RESERVED_BITS) {
 429                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
 430                        efer);
 431                 kvm_inject_gp(vcpu, 0);
 432                 return;
 433         }
 434
 435         if (is_paging(vcpu)
 436             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
 437                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
 438                 kvm_inject_gp(vcpu, 0);
 439                 return;
 440         }
 441
 442         kvm_x86_ops->set_efer(vcpu, efer);
 443
 444         efer &= ~EFER_LMA;
 445         efer |= vcpu->arch.shadow_efer & EFER_LMA;
 446
 447         vcpu->arch.shadow_efer = efer;
 448 }
 449
 450 #endif
 451
 452 /*
 453  * Writes msr value into into the appropriate "register".
 454  * Returns 0 on success, non-0 otherwise.
 455  * Assumes vcpu_load() was already called.
 456  */
 457 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
 458 {
 459         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
 460 }
 461
 462 /*
 463  * Adapt set_msr() to msr_io()'s calling convention
 464  */
 465 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 466 {
 467         return kvm_set_msr(vcpu, index, *data);
 468 }
 469
 470
 471 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
 472 {
 473         switch (msr) {
 474 #ifdef CONFIG_X86_64
 475         case MSR_EFER:
 476                 set_efer(vcpu, data);
 477                 break;
 478 #endif
 479         case MSR_IA32_MC0_STATUS:
 480                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
 481                        __FUNCTION__, data);
 482                 break;
 483         case MSR_IA32_MCG_STATUS:
 484                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
 485                         __FUNCTION__, data);
 486                 break;
 487         case MSR_IA32_UCODE_REV:
 488         case MSR_IA32_UCODE_WRITE:
 489         case 0x200 ... 0x2ff: /* MTRRs */
 490                 break;
 491         case MSR_IA32_APICBASE:
 492                 kvm_set_apic_base(vcpu, data);
 493                 break;
 494         case MSR_IA32_MISC_ENABLE:
 495                 vcpu->arch.ia32_misc_enable_msr = data;
 496                 break;
 497         default:
 498                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
 499                 return 1;
 500         }
 501         return 0;
 502 }
 503 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
 504
 505
 506 /*
 507  * Reads an msr value (of 'msr_index') into 'pdata'.
 508  * Returns 0 on success, non-0 otherwise.
 509  * Assumes vcpu_load() was already called.
 510  */
 511 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 512 {
 513         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
 514 }
 515
 516 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 517 {
 518         u64 data;
 519
 520         switch (msr) {
 521         case 0xc0010010: /* SYSCFG */
 522         case 0xc0010015: /* HWCR */
 523         case MSR_IA32_PLATFORM_ID:
 524         case MSR_IA32_P5_MC_ADDR:
 525         case MSR_IA32_P5_MC_TYPE:
 526         case MSR_IA32_MC0_CTL:
 527         case MSR_IA32_MCG_STATUS:
 528         case MSR_IA32_MCG_CAP:
 529         case MSR_IA32_MC0_MISC:
 530         case MSR_IA32_MC0_MISC+4:
 531         case MSR_IA32_MC0_MISC+8:
 532         case MSR_IA32_MC0_MISC+12:
 533         case MSR_IA32_MC0_MISC+16:
 534         case MSR_IA32_UCODE_REV:
 535         case MSR_IA32_PERF_STATUS:
 536         case MSR_IA32_EBL_CR_POWERON:
 537                 /* MTRR registers */
 538         case 0xfe:
 539         case 0x200 ... 0x2ff:
 540                 data = 0;
 541                 break;
 542         case 0xcd: /* fsb frequency */
 543                 data = 3;
 544                 break;
 545         case MSR_IA32_APICBASE:
 546                 data = kvm_get_apic_base(vcpu);
 547                 break;
 548         case MSR_IA32_MISC_ENABLE:
 549                 data = vcpu->arch.ia32_misc_enable_msr;
 550                 break;
 551 #ifdef CONFIG_X86_64
 552         case MSR_EFER:
 553                 data = vcpu->arch.shadow_efer;
 554                 break;
 555 #endif
 556         default:
 557                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
 558                 return 1;
 559         }
 560         *pdata = data;
 561         return 0;
 562 }
 563 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
 564
 565 /*
 566  * Read or write a bunch of msrs. All parameters are kernel addresses.
 567  *
 568  * @return number of msrs set successfully.
 569  */
 570 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 571                     struct kvm_msr_entry *entries,
 572                     int (*do_msr)(struct kvm_vcpu *vcpu,
 573                                   unsigned index, u64 *data))
 574 {
 575         int i;
 576
 577         vcpu_load(vcpu);
 578
 579         for (i = 0; i < msrs->nmsrs; ++i)
 580                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
 581                         break;
 582
 583         vcpu_put(vcpu);
 584
 585         return i;
 586 }
 587
 588 /*
 589  * Read or write a bunch of msrs. Parameters are user addresses.
 590  *
 591  * @return number of msrs set successfully.
 592  */
 593 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 594                   int (*do_msr)(struct kvm_vcpu *vcpu,
 595                                 unsigned index, u64 *data),
 596                   int writeback)
 597 {
 598         struct kvm_msrs msrs;
 599         struct kvm_msr_entry *entries;
 600         int r, n;
 601         unsigned size;
 602
 603         r = -EFAULT;
 604         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
 605                 goto out;
 606
 607         r = -E2BIG;
 608         if (msrs.nmsrs >= MAX_IO_MSRS)
 609                 goto out;
 610
 611         r = -ENOMEM;
 612         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
 613         entries = vmalloc(size);
 614         if (!entries)
 615                 goto out;
 616
 617         r = -EFAULT;
 618         if (copy_from_user(entries, user_msrs->entries, size))
 619                 goto out_free;
 620
 621         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
 622         if (r < 0)
 623                 goto out_free;
 624
 625         r = -EFAULT;
 626         if (writeback && copy_to_user(user_msrs->entries, entries, size))
 627                 goto out_free;
 628
 629         r = n;
 630
 631 out_free:
 632         vfree(entries);
 633 out:
 634         return r;
 635 }
 636
 637 /*
 638  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
 639  * cached on it.
 640  */
 641 void decache_vcpus_on_cpu(int cpu)
 642 {
 643         struct kvm *vm;
 644         struct kvm_vcpu *vcpu;
 645         int i;
 646
 647         spin_lock(&kvm_lock);
 648         list_for_each_entry(vm, &vm_list, vm_list)
 649                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 650                         vcpu = vm->vcpus[i];
 651                         if (!vcpu)
 652                                 continue;
 653                         /*
 654                          * If the vcpu is locked, then it is running on some
 655                          * other cpu and therefore it is not cached on the
 656                          * cpu in question.
 657                          *
 658                          * If it's not locked, check the last cpu it executed
 659                          * on.
 660                          */
 661                         if (mutex_trylock(&vcpu->mutex)) {
 662                                 if (vcpu->cpu == cpu) {
 663                                         kvm_x86_ops->vcpu_decache(vcpu);
 664                                         vcpu->cpu = -1;
 665                                 }
 666                                 mutex_unlock(&vcpu->mutex);
 667                         }
 668                 }
 669         spin_unlock(&kvm_lock);
 670 }
 671
 672 int kvm_dev_ioctl_check_extension(long ext)
 673 {
 674         int r;
 675
 676         switch (ext) {
 677         case KVM_CAP_IRQCHIP:
 678         case KVM_CAP_HLT:
 679         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
 680         case KVM_CAP_USER_MEMORY:
 681         case KVM_CAP_SET_TSS_ADDR:
 682         case KVM_CAP_EXT_CPUID:
 683                 r = 1;
 684                 break;
 685         case KVM_CAP_VAPIC:
 686                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
 687                 break;
 688         default:
 689                 r = 0;
 690                 break;
 691         }
 692         return r;
 693
 694 }
 695
 696 long kvm_arch_dev_ioctl(struct file *filp,
 697                         unsigned int ioctl, unsigned long arg)
 698 {
 699         void __user *argp = (void __user *)arg;
 700         long r;
 701
 702         switch (ioctl) {
 703         case KVM_GET_MSR_INDEX_LIST: {
 704                 struct kvm_msr_list __user *user_msr_list = argp;
 705                 struct kvm_msr_list msr_list;
 706                 unsigned n;
 707
 708                 r = -EFAULT;
 709                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
 710                         goto out;
 711                 n = msr_list.nmsrs;
 712                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
 713                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
 714                         goto out;
 715                 r = -E2BIG;
 716                 if (n < num_msrs_to_save)
 717                         goto out;
 718                 r = -EFAULT;
 719                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
 720                                  num_msrs_to_save * sizeof(u32)))
 721                         goto out;
 722                 if (copy_to_user(user_msr_list->indices
 723                                  + num_msrs_to_save * sizeof(u32),
 724                                  &emulated_msrs,
 725                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
 726                         goto out;
 727                 r = 0;
 728                 break;
 729         }
 730         default:
 731                 r = -EINVAL;
 732         }
 733 out:
 734         return r;
 735 }
 736
 737 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 738 {
 739         kvm_x86_ops->vcpu_load(vcpu, cpu);
 740 }
 741
 742 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 743 {
 744         kvm_x86_ops->vcpu_put(vcpu);
 745         kvm_put_guest_fpu(vcpu);
 746 }
 747
 748 static int is_efer_nx(void)
 749 {
 750         u64 efer;
 751
 752         rdmsrl(MSR_EFER, efer);
 753         return efer & EFER_NX;
 754 }
 755
 756 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 757 {
 758         int i;
 759         struct kvm_cpuid_entry2 *e, *entry;
 760
 761         entry = NULL;
 762         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 763                 e = &vcpu->arch.cpuid_entries[i];
 764                 if (e->function == 0x80000001) {
 765                         entry = e;
 766                         break;
 767                 }
 768         }
 769         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
 770                 entry->edx &= ~(1 << 20);
 771                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 772         }
 773 }
 774
 775 /* when an old userspace process fills a new kernel module */
 776 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 777                                     struct kvm_cpuid *cpuid,
 778                                     struct kvm_cpuid_entry __user *entries)
 779 {
 780         int r, i;
 781         struct kvm_cpuid_entry *cpuid_entries;
 782
 783         r = -E2BIG;
 784         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 785                 goto out;
 786         r = -ENOMEM;
 787         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 788         if (!cpuid_entries)
 789                 goto out;
 790         r = -EFAULT;
 791         if (copy_from_user(cpuid_entries, entries,
 792                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 793                 goto out_free;
 794         for (i = 0; i < cpuid->nent; i++) {
 795                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 796                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 797                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 798                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 799                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 800                 vcpu->arch.cpuid_entries[i].index = 0;
 801                 vcpu->arch.cpuid_entries[i].flags = 0;
 802                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 803                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 804                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 805         }
 806         vcpu->arch.cpuid_nent = cpuid->nent;
 807         cpuid_fix_nx_cap(vcpu);
 808         r = 0;
 809
 810 out_free:
 811         vfree(cpuid_entries);
 812 out:
 813         return r;
 814 }
 815
 816 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 817                                     struct kvm_cpuid2 *cpuid,
 818                                     struct kvm_cpuid_entry2 __user *entries)
 819 {
 820         int r;
 821
 822         r = -E2BIG;
 823         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 824                 goto out;
 825         r = -EFAULT;
 826         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 827                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 828                 goto out;
 829         vcpu->arch.cpuid_nent = cpuid->nent;
 830         return 0;
 831
 832 out:
 833         return r;
 834 }
 835
 836 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 837                                     struct kvm_cpuid2 *cpuid,
 838                                     struct kvm_cpuid_entry2 __user *entries)
 839 {
 840         int r;
 841
 842         r = -E2BIG;
 843         if (cpuid->nent < vcpu->arch.cpuid_nent)
 844                 goto out;
 845         r = -EFAULT;
 846         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 847                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 848                 goto out;
 849         return 0;
 850
 851 out:
 852         cpuid->nent = vcpu->arch.cpuid_nent;
 853         return r;
 854 }
 855
 856 static inline u32 bit(int bitno)
 857 {
 858         return 1 << (bitno & 31);
 859 }
 860
 861 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 862                           u32 index)
 863 {
 864         entry->function = function;
 865         entry->index = index;
 866         cpuid_count(entry->function, entry->index,
 867                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 868         entry->flags = 0;
 869 }
 870
 871 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 872                          u32 index, int *nent, int maxnent)
 873 {
 874         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
 875                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
 876                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
 877                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
 878                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
 879                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
 880                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
 881                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
 882                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
 883                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
 884         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
 885                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
 886                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
 887                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
 888                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
 889                 bit(X86_FEATURE_PGE) |
 890                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
 891                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
 892                 bit(X86_FEATURE_SYSCALL) |
 893                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
 894 #ifdef CONFIG_X86_64
 895                 bit(X86_FEATURE_LM) |
 896 #endif
 897                 bit(X86_FEATURE_MMXEXT) |
 898                 bit(X86_FEATURE_3DNOWEXT) |
 899                 bit(X86_FEATURE_3DNOW);
 900         const u32 kvm_supported_word3_x86_features =
 901                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
 902         const u32 kvm_supported_word6_x86_features =
 903                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
 904
 905         /* all func 2 cpuid_count() should be called on the same cpu */
 906         get_cpu();
 907         do_cpuid_1_ent(entry, function, index);
 908         ++*nent;
 909
 910         switch (function) {
 911         case 0:
 912                 entry->eax = min(entry->eax, (u32)0xb);
 913                 break;
 914         case 1:
 915                 entry->edx &= kvm_supported_word0_x86_features;
 916                 entry->ecx &= kvm_supported_word3_x86_features;
 917                 break;
 918         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 919          * may return different values. This forces us to get_cpu() before
 920          * issuing the first command, and also to emulate this annoying behavior
 921          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 922         case 2: {
 923                 int t, times = entry->eax & 0xff;
 924
 925                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 926                 for (t = 1; t < times && *nent < maxnent; ++t) {
 927                         do_cpuid_1_ent(&entry[t], function, 0);
 928                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 929                         ++*nent;
 930                 }
 931                 break;
 932         }
 933         /* function 4 and 0xb have additional index. */
 934         case 4: {
 935                 int index, cache_type;
 936
 937                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 938                 /* read more entries until cache_type is zero */
 939                 for (index = 1; *nent < maxnent; ++index) {
 940                         cache_type = entry[index - 1].eax & 0x1f;
 941                         if (!cache_type)
 942                                 break;
 943                         do_cpuid_1_ent(&entry[index], function, index);
 944                         entry[index].flags |=
 945                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 946                         ++*nent;
 947                 }
 948                 break;
 949         }
 950         case 0xb: {
 951                 int index, level_type;
 952
 953                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 954                 /* read more entries until level_type is zero */
 955                 for (index = 1; *nent < maxnent; ++index) {
 956                         level_type = entry[index - 1].ecx & 0xff;
 957                         if (!level_type)
 958                                 break;
 959                         do_cpuid_1_ent(&entry[index], function, index);
 960                         entry[index].flags |=
 961                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 962                         ++*nent;
 963                 }
 964                 break;
 965         }
 966         case 0x80000000:
 967                 entry->eax = min(entry->eax, 0x8000001a);
 968                 break;
 969         case 0x80000001:
 970                 entry->edx &= kvm_supported_word1_x86_features;
 971                 entry->ecx &= kvm_supported_word6_x86_features;
 972                 break;
 973         }
 974         put_cpu();
 975 }
 976
 977 static int kvm_vm_ioctl_get_supported_cpuid(struct kvm *kvm,
 978                                     struct kvm_cpuid2 *cpuid,
 979                                     struct kvm_cpuid_entry2 __user *entries)
 980 {
 981         struct kvm_cpuid_entry2 *cpuid_entries;
 982         int limit, nent = 0, r = -E2BIG;
 983         u32 func;
 984
 985         if (cpuid->nent < 1)
 986                 goto out;
 987         r = -ENOMEM;
 988         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 989         if (!cpuid_entries)
 990                 goto out;
 991
 992         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
 993         limit = cpuid_entries[0].eax;
 994         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
 995                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
 996                                 &nent, cpuid->nent);
 997         r = -E2BIG;
 998         if (nent >= cpuid->nent)
 999                 goto out_free;
1000
1001         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1002         limit = cpuid_entries[nent - 1].eax;
1003         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1004                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1005                                &nent, cpuid->nent);
1006         r = -EFAULT;
1007         if (copy_to_user(entries, cpuid_entries,
1008                         nent * sizeof(struct kvm_cpuid_entry2)))
1009                 goto out_free;
1010         cpuid->nent = nent;
1011         r = 0;
1012
1013 out_free:
1014         vfree(cpuid_entries);
1015 out:
1016         return r;
1017 }
1018
1019 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1020                                     struct kvm_lapic_state *s)
1021 {
1022         vcpu_load(vcpu);
1023         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1024         vcpu_put(vcpu);
1025
1026         return 0;
1027 }
1028
1029 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1030                                     struct kvm_lapic_state *s)
1031 {
1032         vcpu_load(vcpu);
1033         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1034         kvm_apic_post_state_restore(vcpu);
1035         vcpu_put(vcpu);
1036
1037         return 0;
1038 }
1039
1040 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1041                                     struct kvm_interrupt *irq)
1042 {
1043         if (irq->irq < 0 || irq->irq >= 256)
1044                 return -EINVAL;
1045         if (irqchip_in_kernel(vcpu->kvm))
1046                 return -ENXIO;
1047         vcpu_load(vcpu);
1048
1049         set_bit(irq->irq, vcpu->arch.irq_pending);
1050         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1051
1052         vcpu_put(vcpu);
1053
1054         return 0;
1055 }
1056
1057 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1058                                            struct kvm_tpr_access_ctl *tac)
1059 {
1060         if (tac->flags)
1061                 return -EINVAL;
1062         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1063         return 0;
1064 }
1065
1066 long kvm_arch_vcpu_ioctl(struct file *filp,
1067                          unsigned int ioctl, unsigned long arg)
1068 {
1069         struct kvm_vcpu *vcpu = filp->private_data;
1070         void __user *argp = (void __user *)arg;
1071         int r;
1072
1073         switch (ioctl) {
1074         case KVM_GET_LAPIC: {
1075                 struct kvm_lapic_state lapic;
1076
1077                 memset(&lapic, 0, sizeof lapic);
1078                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1079                 if (r)
1080                         goto out;
1081                 r = -EFAULT;
1082                 if (copy_to_user(argp, &lapic, sizeof lapic))
1083                         goto out;
1084                 r = 0;
1085                 break;
1086         }
1087         case KVM_SET_LAPIC: {
1088                 struct kvm_lapic_state lapic;
1089
1090                 r = -EFAULT;
1091                 if (copy_from_user(&lapic, argp, sizeof lapic))
1092                         goto out;
1093                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1094                 if (r)
1095                         goto out;
1096                 r = 0;
1097                 break;
1098         }
1099         case KVM_INTERRUPT: {
1100                 struct kvm_interrupt irq;
1101
1102                 r = -EFAULT;
1103                 if (copy_from_user(&irq, argp, sizeof irq))
1104                         goto out;
1105                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1106                 if (r)
1107                         goto out;
1108                 r = 0;
1109                 break;
1110         }
1111         case KVM_SET_CPUID: {
1112                 struct kvm_cpuid __user *cpuid_arg = argp;
1113                 struct kvm_cpuid cpuid;
1114
1115                 r = -EFAULT;
1116                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1117                         goto out;
1118                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1119                 if (r)
1120                         goto out;
1121                 break;
1122         }
1123         case KVM_SET_CPUID2: {
1124                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1125                 struct kvm_cpuid2 cpuid;
1126
1127                 r = -EFAULT;
1128                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1129                         goto out;
1130                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1131                                 cpuid_arg->entries);
1132                 if (r)
1133                         goto out;
1134                 break;
1135         }
1136         case KVM_GET_CPUID2: {
1137                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1138                 struct kvm_cpuid2 cpuid;
1139
1140                 r = -EFAULT;
1141                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1142                         goto out;
1143                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1144                                 cpuid_arg->entries);
1145                 if (r)
1146                         goto out;
1147                 r = -EFAULT;
1148                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1149                         goto out;
1150                 r = 0;
1151                 break;
1152         }
1153         case KVM_GET_MSRS:
1154                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1155                 break;
1156         case KVM_SET_MSRS:
1157                 r = msr_io(vcpu, argp, do_set_msr, 0);
1158                 break;
1159         case KVM_TPR_ACCESS_REPORTING: {
1160                 struct kvm_tpr_access_ctl tac;
1161
1162                 r = -EFAULT;
1163                 if (copy_from_user(&tac, argp, sizeof tac))
1164                         goto out;
1165                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1166                 if (r)
1167                         goto out;
1168                 r = -EFAULT;
1169                 if (copy_to_user(argp, &tac, sizeof tac))
1170                         goto out;
1171                 r = 0;
1172                 break;
1173         };
1174         case KVM_SET_VAPIC_ADDR: {
1175                 struct kvm_vapic_addr va;
1176
1177                 r = -EINVAL;
1178                 if (!irqchip_in_kernel(vcpu->kvm))
1179                         goto out;
1180                 r = -EFAULT;
1181                 if (copy_from_user(&va, argp, sizeof va))
1182                         goto out;
1183                 r = 0;
1184                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1185                 break;
1186         }
1187         default:
1188                 r = -EINVAL;
1189         }
1190 out:
1191         return r;
1192 }
1193
1194 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1195 {
1196         int ret;
1197
1198         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1199                 return -1;
1200         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1201         return ret;
1202 }
1203
1204 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1205                                           u32 kvm_nr_mmu_pages)
1206 {
1207         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1208                 return -EINVAL;
1209
1210         down_write(&current->mm->mmap_sem);
1211
1212         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1213         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1214
1215         up_write(&current->mm->mmap_sem);
1216         return 0;
1217 }
1218
1219 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1220 {
1221         return kvm->arch.n_alloc_mmu_pages;
1222 }
1223
1224 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1225 {
1226         int i;
1227         struct kvm_mem_alias *alias;
1228
1229         for (i = 0; i < kvm->arch.naliases; ++i) {
1230                 alias = &kvm->arch.aliases[i];
1231                 if (gfn >= alias->base_gfn
1232                     && gfn < alias->base_gfn + alias->npages)
1233                         return alias->target_gfn + gfn - alias->base_gfn;
1234         }
1235         return gfn;
1236 }
1237
1238 /*
1239  * Set a new alias region.  Aliases map a portion of physical memory into
1240  * another portion.  This is useful for memory windows, for example the PC
1241  * VGA region.
1242  */
1243 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1244                                          struct kvm_memory_alias *alias)
1245 {
1246         int r, n;
1247         struct kvm_mem_alias *p;
1248
1249         r = -EINVAL;
1250         /* General sanity checks */
1251         if (alias->memory_size & (PAGE_SIZE - 1))
1252                 goto out;
1253         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1254                 goto out;
1255         if (alias->slot >= KVM_ALIAS_SLOTS)
1256                 goto out;
1257         if (alias->guest_phys_addr + alias->memory_size
1258             < alias->guest_phys_addr)
1259                 goto out;
1260         if (alias->target_phys_addr + alias->memory_size
1261             < alias->target_phys_addr)
1262                 goto out;
1263
1264         down_write(&current->mm->mmap_sem);
1265
1266         p = &kvm->arch.aliases[alias->slot];
1267         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1268         p->npages = alias->memory_size >> PAGE_SHIFT;
1269         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1270
1271         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1272                 if (kvm->arch.aliases[n - 1].npages)
1273                         break;
1274         kvm->arch.naliases = n;
1275
1276         kvm_mmu_zap_all(kvm);
1277
1278         up_write(&current->mm->mmap_sem);
1279
1280         return 0;
1281
1282 out:
1283         return r;
1284 }
1285
1286 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1287 {
1288         int r;
1289
1290         r = 0;
1291         switch (chip->chip_id) {
1292         case KVM_IRQCHIP_PIC_MASTER:
1293                 memcpy(&chip->chip.pic,
1294                         &pic_irqchip(kvm)->pics[0],
1295                         sizeof(struct kvm_pic_state));
1296                 break;
1297         case KVM_IRQCHIP_PIC_SLAVE:
1298                 memcpy(&chip->chip.pic,
1299                         &pic_irqchip(kvm)->pics[1],
1300                         sizeof(struct kvm_pic_state));
1301                 break;
1302         case KVM_IRQCHIP_IOAPIC:
1303                 memcpy(&chip->chip.ioapic,
1304                         ioapic_irqchip(kvm),
1305                         sizeof(struct kvm_ioapic_state));
1306                 break;
1307         default:
1308                 r = -EINVAL;
1309                 break;
1310         }
1311         return r;
1312 }
1313
1314 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1315 {
1316         int r;
1317
1318         r = 0;
1319         switch (chip->chip_id) {
1320         case KVM_IRQCHIP_PIC_MASTER:
1321                 memcpy(&pic_irqchip(kvm)->pics[0],
1322                         &chip->chip.pic,
1323                         sizeof(struct kvm_pic_state));
1324                 break;
1325         case KVM_IRQCHIP_PIC_SLAVE:
1326                 memcpy(&pic_irqchip(kvm)->pics[1],
1327                         &chip->chip.pic,
1328                         sizeof(struct kvm_pic_state));
1329                 break;
1330         case KVM_IRQCHIP_IOAPIC:
1331                 memcpy(ioapic_irqchip(kvm),
1332                         &chip->chip.ioapic,
1333                         sizeof(struct kvm_ioapic_state));
1334                 break;
1335         default:
1336                 r = -EINVAL;
1337                 break;
1338         }
1339         kvm_pic_update_irq(pic_irqchip(kvm));
1340         return r;
1341 }
1342
1343 /*
1344  * Get (and clear) the dirty memory log for a memory slot.
1345  */
1346 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1347                                       struct kvm_dirty_log *log)
1348 {
1349         int r;
1350         int n;
1351         struct kvm_memory_slot *memslot;
1352         int is_dirty = 0;
1353
1354         down_write(&current->mm->mmap_sem);
1355
1356         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1357         if (r)
1358                 goto out;
1359
1360         /* If nothing is dirty, don't bother messing with page tables. */
1361         if (is_dirty) {
1362                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1363                 kvm_flush_remote_tlbs(kvm);
1364                 memslot = &kvm->memslots[log->slot];
1365                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1366                 memset(memslot->dirty_bitmap, 0, n);
1367         }
1368         r = 0;
1369 out:
1370         up_write(&current->mm->mmap_sem);
1371         return r;
1372 }
1373
1374 long kvm_arch_vm_ioctl(struct file *filp,
1375                        unsigned int ioctl, unsigned long arg)
1376 {
1377         struct kvm *kvm = filp->private_data;
1378         void __user *argp = (void __user *)arg;
1379         int r = -EINVAL;
1380
1381         switch (ioctl) {
1382         case KVM_SET_TSS_ADDR:
1383                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1384                 if (r < 0)
1385                         goto out;
1386                 break;
1387         case KVM_SET_MEMORY_REGION: {
1388                 struct kvm_memory_region kvm_mem;
1389                 struct kvm_userspace_memory_region kvm_userspace_mem;
1390
1391                 r = -EFAULT;
1392                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1393                         goto out;
1394                 kvm_userspace_mem.slot = kvm_mem.slot;
1395                 kvm_userspace_mem.flags = kvm_mem.flags;
1396                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1397                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1398                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1399                 if (r)
1400                         goto out;
1401                 break;
1402         }
1403         case KVM_SET_NR_MMU_PAGES:
1404                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1405                 if (r)
1406                         goto out;
1407                 break;
1408         case KVM_GET_NR_MMU_PAGES:
1409                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1410                 break;
1411         case KVM_SET_MEMORY_ALIAS: {
1412                 struct kvm_memory_alias alias;
1413
1414                 r = -EFAULT;
1415                 if (copy_from_user(&alias, argp, sizeof alias))
1416                         goto out;
1417                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1418                 if (r)
1419                         goto out;
1420                 break;
1421         }
1422         case KVM_CREATE_IRQCHIP:
1423                 r = -ENOMEM;
1424                 kvm->arch.vpic = kvm_create_pic(kvm);
1425                 if (kvm->arch.vpic) {
1426                         r = kvm_ioapic_init(kvm);
1427                         if (r) {
1428                                 kfree(kvm->arch.vpic);
1429                                 kvm->arch.vpic = NULL;
1430                                 goto out;
1431                         }
1432                 } else
1433                         goto out;
1434                 break;
1435         case KVM_IRQ_LINE: {
1436                 struct kvm_irq_level irq_event;
1437
1438                 r = -EFAULT;
1439                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1440                         goto out;
1441                 if (irqchip_in_kernel(kvm)) {
1442                         mutex_lock(&kvm->lock);
1443                         if (irq_event.irq < 16)
1444                                 kvm_pic_set_irq(pic_irqchip(kvm),
1445                                         irq_event.irq,
1446                                         irq_event.level);
1447                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1448                                         irq_event.irq,
1449                                         irq_event.level);
1450                         mutex_unlock(&kvm->lock);
1451                         r = 0;
1452                 }
1453                 break;
1454         }
1455         case KVM_GET_IRQCHIP: {
1456                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1457                 struct kvm_irqchip chip;
1458
1459                 r = -EFAULT;
1460                 if (copy_from_user(&chip, argp, sizeof chip))
1461                         goto out;
1462                 r = -ENXIO;
1463                 if (!irqchip_in_kernel(kvm))
1464                         goto out;
1465                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1466                 if (r)
1467                         goto out;
1468                 r = -EFAULT;
1469                 if (copy_to_user(argp, &chip, sizeof chip))
1470                         goto out;
1471                 r = 0;
1472                 break;
1473         }
1474         case KVM_SET_IRQCHIP: {
1475                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1476                 struct kvm_irqchip chip;
1477
1478                 r = -EFAULT;
1479                 if (copy_from_user(&chip, argp, sizeof chip))
1480                         goto out;
1481                 r = -ENXIO;
1482                 if (!irqchip_in_kernel(kvm))
1483                         goto out;
1484                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1485                 if (r)
1486                         goto out;
1487                 r = 0;
1488                 break;
1489         }
1490         case KVM_GET_SUPPORTED_CPUID: {
1491                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1492                 struct kvm_cpuid2 cpuid;
1493
1494                 r = -EFAULT;
1495                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1496                         goto out;
1497                 r = kvm_vm_ioctl_get_supported_cpuid(kvm, &cpuid,
1498                         cpuid_arg->entries);
1499                 if (r)
1500                         goto out;
1501
1502                 r = -EFAULT;
1503                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1504                         goto out;
1505                 r = 0;
1506                 break;
1507         }
1508         default:
1509                 ;
1510         }
1511 out:
1512         return r;
1513 }
1514
1515 static void kvm_init_msr_list(void)
1516 {
1517         u32 dummy[2];
1518         unsigned i, j;
1519
1520         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1521                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1522                         continue;
1523                 if (j < i)
1524                         msrs_to_save[j] = msrs_to_save[i];
1525                 j++;
1526         }
1527         num_msrs_to_save = j;
1528 }
1529
1530 /*
1531  * Only apic need an MMIO device hook, so shortcut now..
1532  */
1533 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1534                                                 gpa_t addr)
1535 {
1536         struct kvm_io_device *dev;
1537
1538         if (vcpu->arch.apic) {
1539                 dev = &vcpu->arch.apic->dev;
1540                 if (dev->in_range(dev, addr))
1541                         return dev;
1542         }
1543         return NULL;
1544 }
1545
1546
1547 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1548                                                 gpa_t addr)
1549 {
1550         struct kvm_io_device *dev;
1551
1552         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1553         if (dev == NULL)
1554                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1555         return dev;
1556 }
1557
1558 int emulator_read_std(unsigned long addr,
1559                              void *val,
1560                              unsigned int bytes,
1561                              struct kvm_vcpu *vcpu)
1562 {
1563         void *data = val;
1564         int r = X86EMUL_CONTINUE;
1565
1566         down_read(&current->mm->mmap_sem);
1567         while (bytes) {
1568                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1569                 unsigned offset = addr & (PAGE_SIZE-1);
1570                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1571                 int ret;
1572
1573                 if (gpa == UNMAPPED_GVA) {
1574                         r = X86EMUL_PROPAGATE_FAULT;
1575                         goto out;
1576                 }
1577                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1578                 if (ret < 0) {
1579                         r = X86EMUL_UNHANDLEABLE;
1580                         goto out;
1581                 }
1582
1583                 bytes -= tocopy;
1584                 data += tocopy;
1585                 addr += tocopy;
1586         }
1587 out:
1588         up_read(&current->mm->mmap_sem);
1589         return r;
1590 }
1591 EXPORT_SYMBOL_GPL(emulator_read_std);
1592
1593 static int emulator_read_emulated(unsigned long addr,
1594                                   void *val,
1595                                   unsigned int bytes,
1596                                   struct kvm_vcpu *vcpu)
1597 {
1598         struct kvm_io_device *mmio_dev;
1599         gpa_t                 gpa;
1600
1601         if (vcpu->mmio_read_completed) {
1602                 memcpy(val, vcpu->mmio_data, bytes);
1603                 vcpu->mmio_read_completed = 0;
1604                 return X86EMUL_CONTINUE;
1605         }
1606
1607         down_read(&current->mm->mmap_sem);
1608         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1609         up_read(&current->mm->mmap_sem);
1610
1611         /* For APIC access vmexit */
1612         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1613                 goto mmio;
1614
1615         if (emulator_read_std(addr, val, bytes, vcpu)
1616                         == X86EMUL_CONTINUE)
1617                 return X86EMUL_CONTINUE;
1618         if (gpa == UNMAPPED_GVA)
1619                 return X86EMUL_PROPAGATE_FAULT;
1620
1621 mmio:
1622         /*
1623          * Is this MMIO handled locally?
1624          */
1625         mutex_lock(&vcpu->kvm->lock);
1626         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1627         if (mmio_dev) {
1628                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1629                 mutex_unlock(&vcpu->kvm->lock);
1630                 return X86EMUL_CONTINUE;
1631         }
1632         mutex_unlock(&vcpu->kvm->lock);
1633
1634         vcpu->mmio_needed = 1;
1635         vcpu->mmio_phys_addr = gpa;
1636         vcpu->mmio_size = bytes;
1637         vcpu->mmio_is_write = 0;
1638
1639         return X86EMUL_UNHANDLEABLE;
1640 }
1641
1642 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1643                                const void *val, int bytes)
1644 {
1645         int ret;
1646
1647         down_read(&current->mm->mmap_sem);
1648         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1649         if (ret < 0) {
1650                 up_read(&current->mm->mmap_sem);
1651                 return 0;
1652         }
1653         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1654         up_read(&current->mm->mmap_sem);
1655         return 1;
1656 }
1657
1658 static int emulator_write_emulated_onepage(unsigned long addr,
1659                                            const void *val,
1660                                            unsigned int bytes,
1661                                            struct kvm_vcpu *vcpu)
1662 {
1663         struct kvm_io_device *mmio_dev;
1664         gpa_t                 gpa;
1665
1666         down_read(&current->mm->mmap_sem);
1667         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1668         up_read(&current->mm->mmap_sem);
1669
1670         if (gpa == UNMAPPED_GVA) {
1671                 kvm_inject_page_fault(vcpu, addr, 2);
1672                 return X86EMUL_PROPAGATE_FAULT;
1673         }
1674
1675         /* For APIC access vmexit */
1676         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1677                 goto mmio;
1678
1679         if (emulator_write_phys(vcpu, gpa, val, bytes))
1680                 return X86EMUL_CONTINUE;
1681
1682 mmio:
1683         /*
1684          * Is this MMIO handled locally?
1685          */
1686         mutex_lock(&vcpu->kvm->lock);
1687         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1688         if (mmio_dev) {
1689                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1690                 mutex_unlock(&vcpu->kvm->lock);
1691                 return X86EMUL_CONTINUE;
1692         }
1693         mutex_unlock(&vcpu->kvm->lock);
1694
1695         vcpu->mmio_needed = 1;
1696         vcpu->mmio_phys_addr = gpa;
1697         vcpu->mmio_size = bytes;
1698         vcpu->mmio_is_write = 1;
1699         memcpy(vcpu->mmio_data, val, bytes);
1700
1701         return X86EMUL_CONTINUE;
1702 }
1703
1704 int emulator_write_emulated(unsigned long addr,
1705                                    const void *val,
1706                                    unsigned int bytes,
1707                                    struct kvm_vcpu *vcpu)
1708 {
1709         /* Crossing a page boundary? */
1710         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1711                 int rc, now;
1712
1713                 now = -addr & ~PAGE_MASK;
1714                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1715                 if (rc != X86EMUL_CONTINUE)
1716                         return rc;
1717                 addr += now;
1718                 val += now;
1719                 bytes -= now;
1720         }
1721         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1722 }
1723 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1724
1725 static int emulator_cmpxchg_emulated(unsigned long addr,
1726                                      const void *old,
1727                                      const void *new,
1728                                      unsigned int bytes,
1729                                      struct kvm_vcpu *vcpu)
1730 {
1731         static int reported;
1732
1733         if (!reported) {
1734                 reported = 1;
1735                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1736         }
1737 #ifndef CONFIG_X86_64
1738         /* guests cmpxchg8b have to be emulated atomically */
1739         if (bytes == 8) {
1740                 gpa_t gpa;
1741                 struct page *page;
1742                 char *kaddr;
1743                 u64 val;
1744
1745                 down_read(&current->mm->mmap_sem);
1746                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1747
1748                 if (gpa == UNMAPPED_GVA ||
1749                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1750                         goto emul_write;
1751
1752                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1753                         goto emul_write;
1754
1755                 val = *(u64 *)new;
1756                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1757                 kaddr = kmap_atomic(page, KM_USER0);
1758                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1759                 kunmap_atomic(kaddr, KM_USER0);
1760                 kvm_release_page_dirty(page);
1761         emul_write:
1762                 up_read(&current->mm->mmap_sem);
1763         }
1764 #endif
1765
1766         return emulator_write_emulated(addr, new, bytes, vcpu);
1767 }
1768
1769 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1770 {
1771         return kvm_x86_ops->get_segment_base(vcpu, seg);
1772 }
1773
1774 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1775 {
1776         return X86EMUL_CONTINUE;
1777 }
1778
1779 int emulate_clts(struct kvm_vcpu *vcpu)
1780 {
1781         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1782         return X86EMUL_CONTINUE;
1783 }
1784
1785 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1786 {
1787         struct kvm_vcpu *vcpu = ctxt->vcpu;
1788
1789         switch (dr) {
1790         case 0 ... 3:
1791                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1792                 return X86EMUL_CONTINUE;
1793         default:
1794                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1795                 return X86EMUL_UNHANDLEABLE;
1796         }
1797 }
1798
1799 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1800 {
1801         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1802         int exception;
1803
1804         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1805         if (exception) {
1806                 /* FIXME: better handling */
1807                 return X86EMUL_UNHANDLEABLE;
1808         }
1809         return X86EMUL_CONTINUE;
1810 }
1811
1812 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1813 {
1814         static int reported;
1815         u8 opcodes[4];
1816         unsigned long rip = vcpu->arch.rip;
1817         unsigned long rip_linear;
1818
1819         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1820
1821         if (reported)
1822                 return;
1823
1824         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1825
1826         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1827                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1828         reported = 1;
1829 }
1830 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1831
1832 struct x86_emulate_ops emulate_ops = {
1833         .read_std            = emulator_read_std,
1834         .read_emulated       = emulator_read_emulated,
1835         .write_emulated      = emulator_write_emulated,
1836         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1837 };
1838
1839 int emulate_instruction(struct kvm_vcpu *vcpu,
1840                         struct kvm_run *run,
1841                         unsigned long cr2,
1842                         u16 error_code,
1843                         int emulation_type)
1844 {
1845         int r;
1846         struct decode_cache *c;
1847
1848         vcpu->arch.mmio_fault_cr2 = cr2;
1849         kvm_x86_ops->cache_regs(vcpu);
1850
1851         vcpu->mmio_is_write = 0;
1852         vcpu->arch.pio.string = 0;
1853
1854         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
1855                 int cs_db, cs_l;
1856                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1857
1858                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
1859                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1860                 vcpu->arch.emulate_ctxt.mode =
1861                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
1862                         ? X86EMUL_MODE_REAL : cs_l
1863                         ? X86EMUL_MODE_PROT64 : cs_db
1864                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1865
1866                 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1867                         vcpu->arch.emulate_ctxt.cs_base = 0;
1868                         vcpu->arch.emulate_ctxt.ds_base = 0;
1869                         vcpu->arch.emulate_ctxt.es_base = 0;
1870                         vcpu->arch.emulate_ctxt.ss_base = 0;
1871                 } else {
1872                         vcpu->arch.emulate_ctxt.cs_base =
1873                                         get_segment_base(vcpu, VCPU_SREG_CS);
1874                         vcpu->arch.emulate_ctxt.ds_base =
1875                                         get_segment_base(vcpu, VCPU_SREG_DS);
1876                         vcpu->arch.emulate_ctxt.es_base =
1877                                         get_segment_base(vcpu, VCPU_SREG_ES);
1878                         vcpu->arch.emulate_ctxt.ss_base =
1879                                         get_segment_base(vcpu, VCPU_SREG_SS);
1880                 }
1881
1882                 vcpu->arch.emulate_ctxt.gs_base =
1883                                         get_segment_base(vcpu, VCPU_SREG_GS);
1884                 vcpu->arch.emulate_ctxt.fs_base =
1885                                         get_segment_base(vcpu, VCPU_SREG_FS);
1886
1887                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
1888
1889                 /* Reject the instructions other than VMCALL/VMMCALL when
1890                  * try to emulate invalid opcode */
1891                 c = &vcpu->arch.emulate_ctxt.decode;
1892                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
1893                     (!(c->twobyte && c->b == 0x01 &&
1894                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
1895                        c->modrm_mod == 3 && c->modrm_rm == 1)))
1896                         return EMULATE_FAIL;
1897
1898                 ++vcpu->stat.insn_emulation;
1899                 if (r)  {
1900                         ++vcpu->stat.insn_emulation_fail;
1901                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1902                                 return EMULATE_DONE;
1903                         return EMULATE_FAIL;
1904                 }
1905         }
1906
1907         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
1908
1909         if (vcpu->arch.pio.string)
1910                 return EMULATE_DO_MMIO;
1911
1912         if ((r || vcpu->mmio_is_write) && run) {
1913                 run->exit_reason = KVM_EXIT_MMIO;
1914                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1915                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1916                 run->mmio.len = vcpu->mmio_size;
1917                 run->mmio.is_write = vcpu->mmio_is_write;
1918         }
1919
1920         if (r) {
1921                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1922                         return EMULATE_DONE;
1923                 if (!vcpu->mmio_needed) {
1924                         kvm_report_emulation_failure(vcpu, "mmio");
1925                         return EMULATE_FAIL;
1926                 }
1927                 return EMULATE_DO_MMIO;
1928         }
1929
1930         kvm_x86_ops->decache_regs(vcpu);
1931         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
1932
1933         if (vcpu->mmio_is_write) {
1934                 vcpu->mmio_needed = 0;
1935                 return EMULATE_DO_MMIO;
1936         }
1937
1938         return EMULATE_DONE;
1939 }
1940 EXPORT_SYMBOL_GPL(emulate_instruction);
1941
1942 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1943 {
1944         int i;
1945
1946         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
1947                 if (vcpu->arch.pio.guest_pages[i]) {
1948                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
1949                         vcpu->arch.pio.guest_pages[i] = NULL;
1950                 }
1951 }
1952
1953 static int pio_copy_data(struct kvm_vcpu *vcpu)
1954 {
1955         void *p = vcpu->arch.pio_data;
1956         void *q;
1957         unsigned bytes;
1958         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
1959
1960         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1961                  PAGE_KERNEL);
1962         if (!q) {
1963                 free_pio_guest_pages(vcpu);
1964                 return -ENOMEM;
1965         }
1966         q += vcpu->arch.pio.guest_page_offset;
1967         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
1968         if (vcpu->arch.pio.in)
1969                 memcpy(q, p, bytes);
1970         else
1971                 memcpy(p, q, bytes);
1972         q -= vcpu->arch.pio.guest_page_offset;
1973         vunmap(q);
1974         free_pio_guest_pages(vcpu);
1975         return 0;
1976 }
1977
1978 int complete_pio(struct kvm_vcpu *vcpu)
1979 {
1980         struct kvm_pio_request *io = &vcpu->arch.pio;
1981         long delta;
1982         int r;
1983
1984         kvm_x86_ops->cache_regs(vcpu);
1985
1986         if (!io->string) {
1987                 if (io->in)
1988                         memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
1989                                io->size);
1990         } else {
1991                 if (io->in) {
1992                         r = pio_copy_data(vcpu);
1993                         if (r) {
1994                                 kvm_x86_ops->cache_regs(vcpu);
1995                                 return r;
1996                         }
1997                 }
1998
1999                 delta = 1;
2000                 if (io->rep) {
2001                         delta *= io->cur_count;
2002                         /*
2003                          * The size of the register should really depend on
2004                          * current address size.
2005                          */
2006                         vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2007                 }
2008                 if (io->down)
2009                         delta = -delta;
2010                 delta *= io->size;
2011                 if (io->in)
2012                         vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2013                 else
2014                         vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2015         }
2016
2017         kvm_x86_ops->decache_regs(vcpu);
2018
2019         io->count -= io->cur_count;
2020         io->cur_count = 0;
2021
2022         return 0;
2023 }
2024
2025 static void kernel_pio(struct kvm_io_device *pio_dev,
2026                        struct kvm_vcpu *vcpu,
2027                        void *pd)
2028 {
2029         /* TODO: String I/O for in kernel device */
2030
2031         mutex_lock(&vcpu->kvm->lock);
2032         if (vcpu->arch.pio.in)
2033                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2034                                   vcpu->arch.pio.size,
2035                                   pd);
2036         else
2037                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2038                                    vcpu->arch.pio.size,
2039                                    pd);
2040         mutex_unlock(&vcpu->kvm->lock);
2041 }
2042
2043 static void pio_string_write(struct kvm_io_device *pio_dev,
2044                              struct kvm_vcpu *vcpu)
2045 {
2046         struct kvm_pio_request *io = &vcpu->arch.pio;
2047         void *pd = vcpu->arch.pio_data;
2048         int i;
2049
2050         mutex_lock(&vcpu->kvm->lock);
2051         for (i = 0; i < io->cur_count; i++) {
2052                 kvm_iodevice_write(pio_dev, io->port,
2053                                    io->size,
2054                                    pd);
2055                 pd += io->size;
2056         }
2057         mutex_unlock(&vcpu->kvm->lock);
2058 }
2059
2060 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2061                                                gpa_t addr)
2062 {
2063         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2064 }
2065
2066 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2067                   int size, unsigned port)
2068 {
2069         struct kvm_io_device *pio_dev;
2070
2071         vcpu->run->exit_reason = KVM_EXIT_IO;
2072         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2073         vcpu->run->io.size = vcpu->arch.pio.size = size;
2074         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2075         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2076         vcpu->run->io.port = vcpu->arch.pio.port = port;
2077         vcpu->arch.pio.in = in;
2078         vcpu->arch.pio.string = 0;
2079         vcpu->arch.pio.down = 0;
2080         vcpu->arch.pio.guest_page_offset = 0;
2081         vcpu->arch.pio.rep = 0;
2082
2083         kvm_x86_ops->cache_regs(vcpu);
2084         memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2085         kvm_x86_ops->decache_regs(vcpu);
2086
2087         kvm_x86_ops->skip_emulated_instruction(vcpu);
2088
2089         pio_dev = vcpu_find_pio_dev(vcpu, port);
2090         if (pio_dev) {
2091                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2092                 complete_pio(vcpu);
2093                 return 1;
2094         }
2095         return 0;
2096 }
2097 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2098
2099 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2100                   int size, unsigned long count, int down,
2101                   gva_t address, int rep, unsigned port)
2102 {
2103         unsigned now, in_page;
2104         int i, ret = 0;
2105         int nr_pages = 1;
2106         struct page *page;
2107         struct kvm_io_device *pio_dev;
2108
2109         vcpu->run->exit_reason = KVM_EXIT_IO;
2110         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2111         vcpu->run->io.size = vcpu->arch.pio.size = size;
2112         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2113         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2114         vcpu->run->io.port = vcpu->arch.pio.port = port;
2115         vcpu->arch.pio.in = in;
2116         vcpu->arch.pio.string = 1;
2117         vcpu->arch.pio.down = down;
2118         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2119         vcpu->arch.pio.rep = rep;
2120
2121         if (!count) {
2122                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2123                 return 1;
2124         }
2125
2126         if (!down)
2127                 in_page = PAGE_SIZE - offset_in_page(address);
2128         else
2129                 in_page = offset_in_page(address) + size;
2130         now = min(count, (unsigned long)in_page / size);
2131         if (!now) {
2132                 /*
2133                  * String I/O straddles page boundary.  Pin two guest pages
2134                  * so that we satisfy atomicity constraints.  Do just one
2135                  * transaction to avoid complexity.
2136                  */
2137                 nr_pages = 2;
2138                 now = 1;
2139         }
2140         if (down) {
2141                 /*
2142                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2143                  */
2144                 pr_unimpl(vcpu, "guest string pio down\n");
2145                 kvm_inject_gp(vcpu, 0);
2146                 return 1;
2147         }
2148         vcpu->run->io.count = now;
2149         vcpu->arch.pio.cur_count = now;
2150
2151         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2152                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2153
2154         for (i = 0; i < nr_pages; ++i) {
2155                 down_read(&current->mm->mmap_sem);
2156                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2157                 vcpu->arch.pio.guest_pages[i] = page;
2158                 up_read(&current->mm->mmap_sem);
2159                 if (!page) {
2160                         kvm_inject_gp(vcpu, 0);
2161                         free_pio_guest_pages(vcpu);
2162                         return 1;
2163                 }
2164         }
2165
2166         pio_dev = vcpu_find_pio_dev(vcpu, port);
2167         if (!vcpu->arch.pio.in) {
2168                 /* string PIO write */
2169                 ret = pio_copy_data(vcpu);
2170                 if (ret >= 0 && pio_dev) {
2171                         pio_string_write(pio_dev, vcpu);
2172                         complete_pio(vcpu);
2173                         if (vcpu->arch.pio.count == 0)
2174                                 ret = 1;
2175                 }
2176         } else if (pio_dev)
2177                 pr_unimpl(vcpu, "no string pio read support yet, "
2178                        "port %x size %d count %ld\n",
2179                         port, size, count);
2180
2181         return ret;
2182 }
2183 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2184
2185 int kvm_arch_init(void *opaque)
2186 {
2187         int r;
2188         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2189
2190         if (kvm_x86_ops) {
2191                 printk(KERN_ERR "kvm: already loaded the other module\n");
2192                 r = -EEXIST;
2193                 goto out;
2194         }
2195
2196         if (!ops->cpu_has_kvm_support()) {
2197                 printk(KERN_ERR "kvm: no hardware support\n");
2198                 r = -EOPNOTSUPP;
2199                 goto out;
2200         }
2201         if (ops->disabled_by_bios()) {
2202                 printk(KERN_ERR "kvm: disabled by bios\n");
2203                 r = -EOPNOTSUPP;
2204                 goto out;
2205         }
2206
2207         r = kvm_mmu_module_init();
2208         if (r)
2209                 goto out;
2210
2211         kvm_init_msr_list();
2212
2213         kvm_x86_ops = ops;
2214         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2215         return 0;
2216
2217 out:
2218         return r;
2219 }
2220
2221 void kvm_arch_exit(void)
2222 {
2223         kvm_x86_ops = NULL;
2224         kvm_mmu_module_exit();
2225 }
2226
2227 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2228 {
2229         ++vcpu->stat.halt_exits;
2230         if (irqchip_in_kernel(vcpu->kvm)) {
2231                 vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
2232                 kvm_vcpu_block(vcpu);
2233                 if (vcpu->arch.mp_state != VCPU_MP_STATE_RUNNABLE)
2234                         return -EINTR;
2235                 return 1;
2236         } else {
2237                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2238                 return 0;
2239         }
2240 }
2241 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2242
2243 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2244 {
2245         unsigned long nr, a0, a1, a2, a3, ret;
2246
2247         kvm_x86_ops->cache_regs(vcpu);
2248
2249         nr = vcpu->arch.regs[VCPU_REGS_RAX];
2250         a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2251         a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2252         a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2253         a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2254
2255         if (!is_long_mode(vcpu)) {
2256                 nr &= 0xFFFFFFFF;
2257                 a0 &= 0xFFFFFFFF;
2258                 a1 &= 0xFFFFFFFF;
2259                 a2 &= 0xFFFFFFFF;
2260                 a3 &= 0xFFFFFFFF;
2261         }
2262
2263         switch (nr) {
2264         case KVM_HC_VAPIC_POLL_IRQ:
2265                 ret = 0;
2266                 break;
2267         default:
2268                 ret = -KVM_ENOSYS;
2269                 break;
2270         }
2271         vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2272         kvm_x86_ops->decache_regs(vcpu);
2273         return 0;
2274 }
2275 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2276
2277 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2278 {
2279         char instruction[3];
2280         int ret = 0;
2281
2282
2283         /*
2284          * Blow out the MMU to ensure that no other VCPU has an active mapping
2285          * to ensure that the updated hypercall appears atomically across all
2286          * VCPUs.
2287          */
2288         kvm_mmu_zap_all(vcpu->kvm);
2289
2290         kvm_x86_ops->cache_regs(vcpu);
2291         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2292         if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2293             != X86EMUL_CONTINUE)
2294                 ret = -EFAULT;
2295
2296         return ret;
2297 }
2298
2299 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2300 {
2301         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2302 }
2303
2304 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2305 {
2306         struct descriptor_table dt = { limit, base };
2307
2308         kvm_x86_ops->set_gdt(vcpu, &dt);
2309 }
2310
2311 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2312 {
2313         struct descriptor_table dt = { limit, base };
2314
2315         kvm_x86_ops->set_idt(vcpu, &dt);
2316 }
2317
2318 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2319                    unsigned long *rflags)
2320 {
2321         lmsw(vcpu, msw);
2322         *rflags = kvm_x86_ops->get_rflags(vcpu);
2323 }
2324
2325 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2326 {
2327         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2328         switch (cr) {
2329         case 0:
2330                 return vcpu->arch.cr0;
2331         case 2:
2332                 return vcpu->arch.cr2;
2333         case 3:
2334                 return vcpu->arch.cr3;
2335         case 4:
2336                 return vcpu->arch.cr4;
2337         case 8:
2338                 return get_cr8(vcpu);
2339         default:
2340                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2341                 return 0;
2342         }
2343 }
2344
2345 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2346                      unsigned long *rflags)
2347 {
2348         switch (cr) {
2349         case 0:
2350                 set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2351                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2352                 break;
2353         case 2:
2354                 vcpu->arch.cr2 = val;
2355                 break;
2356         case 3:
2357                 set_cr3(vcpu, val);
2358                 break;
2359         case 4:
2360                 set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2361                 break;
2362         case 8:
2363                 set_cr8(vcpu, val & 0xfUL);
2364                 break;
2365         default:
2366                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2367         }
2368 }
2369
2370 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2371 {
2372         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2373         int j, nent = vcpu->arch.cpuid_nent;
2374
2375         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2376         /* when no next entry is found, the current entry[i] is reselected */
2377         for (j = i + 1; j == i; j = (j + 1) % nent) {
2378                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2379                 if (ej->function == e->function) {
2380                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2381                         return j;
2382                 }
2383         }
2384         return 0; /* silence gcc, even though control never reaches here */
2385 }
2386
2387 /* find an entry with matching function, matching index (if needed), and that
2388  * should be read next (if it's stateful) */
2389 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2390         u32 function, u32 index)
2391 {
2392         if (e->function != function)
2393                 return 0;
2394         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2395                 return 0;
2396         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2397                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2398                 return 0;
2399         return 1;
2400 }
2401
2402 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2403 {
2404         int i;
2405         u32 function, index;
2406         struct kvm_cpuid_entry2 *e, *best;
2407
2408         kvm_x86_ops->cache_regs(vcpu);
2409         function = vcpu->arch.regs[VCPU_REGS_RAX];
2410         index = vcpu->arch.regs[VCPU_REGS_RCX];
2411         vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2412         vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2413         vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2414         vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2415         best = NULL;
2416         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2417                 e = &vcpu->arch.cpuid_entries[i];
2418                 if (is_matching_cpuid_entry(e, function, index)) {
2419                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2420                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2421                         best = e;
2422                         break;
2423                 }
2424                 /*
2425                  * Both basic or both extended?
2426                  */
2427                 if (((e->function ^ function) & 0x80000000) == 0)
2428                         if (!best || e->function > best->function)
2429                                 best = e;
2430         }
2431         if (best) {
2432                 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2433                 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2434                 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2435                 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2436         }
2437         kvm_x86_ops->decache_regs(vcpu);
2438         kvm_x86_ops->skip_emulated_instruction(vcpu);
2439 }
2440 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2441
2442 /*
2443  * Check if userspace requested an interrupt window, and that the
2444  * interrupt window is open.
2445  *
2446  * No need to exit to userspace if we already have an interrupt queued.
2447  */
2448 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2449                                           struct kvm_run *kvm_run)
2450 {
2451         return (!vcpu->arch.irq_summary &&
2452                 kvm_run->request_interrupt_window &&
2453                 vcpu->arch.interrupt_window_open &&
2454                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2455 }
2456
2457 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2458                               struct kvm_run *kvm_run)
2459 {
2460         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2461         kvm_run->cr8 = get_cr8(vcpu);
2462         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2463         if (irqchip_in_kernel(vcpu->kvm))
2464                 kvm_run->ready_for_interrupt_injection = 1;
2465         else
2466                 kvm_run->ready_for_interrupt_injection =
2467                                         (vcpu->arch.interrupt_window_open &&
2468                                          vcpu->arch.irq_summary == 0);
2469 }
2470
2471 static void vapic_enter(struct kvm_vcpu *vcpu)
2472 {
2473         struct kvm_lapic *apic = vcpu->arch.apic;
2474         struct page *page;
2475
2476         if (!apic || !apic->vapic_addr)
2477                 return;
2478
2479         down_read(&current->mm->mmap_sem);
2480         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2481         vcpu->arch.apic->vapic_page = page;
2482         up_read(&current->mm->mmap_sem);
2483 }
2484
2485 static void vapic_exit(struct kvm_vcpu *vcpu)
2486 {
2487         struct kvm_lapic *apic = vcpu->arch.apic;
2488
2489         if (!apic || !apic->vapic_addr)
2490                 return;
2491
2492         kvm_release_page_dirty(apic->vapic_page);
2493         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2494 }
2495
2496 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2497 {
2498         int r;
2499
2500         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2501                 pr_debug("vcpu %d received sipi with vector # %x\n",
2502                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
2503                 kvm_lapic_reset(vcpu);
2504                 r = kvm_x86_ops->vcpu_reset(vcpu);
2505                 if (r)
2506                         return r;
2507                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
2508         }
2509
2510         vapic_enter(vcpu);
2511
2512 preempted:
2513         if (vcpu->guest_debug.enabled)
2514                 kvm_x86_ops->guest_debug_pre(vcpu);
2515
2516 again:
2517         r = kvm_mmu_reload(vcpu);
2518         if (unlikely(r))
2519                 goto out;
2520
2521         if (vcpu->requests) {
2522                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2523                         __kvm_migrate_apic_timer(vcpu);
2524                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2525                                        &vcpu->requests)) {
2526                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2527                         r = 0;
2528                         goto out;
2529                 }
2530         }
2531
2532         kvm_inject_pending_timer_irqs(vcpu);
2533
2534         preempt_disable();
2535
2536         kvm_x86_ops->prepare_guest_switch(vcpu);
2537         kvm_load_guest_fpu(vcpu);
2538
2539         local_irq_disable();
2540
2541         if (need_resched()) {
2542                 local_irq_enable();
2543                 preempt_enable();
2544                 r = 1;
2545                 goto out;
2546         }
2547
2548         if (signal_pending(current)) {
2549                 local_irq_enable();
2550                 preempt_enable();
2551                 r = -EINTR;
2552                 kvm_run->exit_reason = KVM_EXIT_INTR;
2553                 ++vcpu->stat.signal_exits;
2554                 goto out;
2555         }
2556
2557         if (vcpu->arch.exception.pending)
2558                 __queue_exception(vcpu);
2559         else if (irqchip_in_kernel(vcpu->kvm))
2560                 kvm_x86_ops->inject_pending_irq(vcpu);
2561         else
2562                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2563
2564         kvm_lapic_sync_to_vapic(vcpu);
2565
2566         vcpu->guest_mode = 1;
2567         kvm_guest_enter();
2568
2569         if (vcpu->requests)
2570                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2571                         kvm_x86_ops->tlb_flush(vcpu);
2572
2573         kvm_x86_ops->run(vcpu, kvm_run);
2574
2575         vcpu->guest_mode = 0;
2576         local_irq_enable();
2577
2578         ++vcpu->stat.exits;
2579
2580         /*
2581          * We must have an instruction between local_irq_enable() and
2582          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2583          * the interrupt shadow.  The stat.exits increment will do nicely.
2584          * But we need to prevent reordering, hence this barrier():
2585          */
2586         barrier();
2587
2588         kvm_guest_exit();
2589
2590         preempt_enable();
2591
2592         /*
2593          * Profile KVM exit RIPs:
2594          */
2595         if (unlikely(prof_on == KVM_PROFILING)) {
2596                 kvm_x86_ops->cache_regs(vcpu);
2597                 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2598         }
2599
2600         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2601                 vcpu->arch.exception.pending = false;
2602
2603         kvm_lapic_sync_from_vapic(vcpu);
2604
2605         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2606
2607         if (r > 0) {
2608                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2609                         r = -EINTR;
2610                         kvm_run->exit_reason = KVM_EXIT_INTR;
2611                         ++vcpu->stat.request_irq_exits;
2612                         goto out;
2613                 }
2614                 if (!need_resched())
2615                         goto again;
2616         }
2617
2618 out:
2619         if (r > 0) {
2620                 kvm_resched(vcpu);
2621                 goto preempted;
2622         }
2623
2624         post_kvm_run_save(vcpu, kvm_run);
2625
2626         vapic_exit(vcpu);
2627
2628         return r;
2629 }
2630
2631 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2632 {
2633         int r;
2634         sigset_t sigsaved;
2635
2636         vcpu_load(vcpu);
2637
2638         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2639                 kvm_vcpu_block(vcpu);
2640                 vcpu_put(vcpu);
2641                 return -EAGAIN;
2642         }
2643
2644         if (vcpu->sigset_active)
2645                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2646
2647         /* re-sync apic's tpr */
2648         if (!irqchip_in_kernel(vcpu->kvm))
2649                 set_cr8(vcpu, kvm_run->cr8);
2650
2651         if (vcpu->arch.pio.cur_count) {
2652                 r = complete_pio(vcpu);
2653                 if (r)
2654                         goto out;
2655         }
2656 #if CONFIG_HAS_IOMEM
2657         if (vcpu->mmio_needed) {
2658                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2659                 vcpu->mmio_read_completed = 1;
2660                 vcpu->mmio_needed = 0;
2661                 r = emulate_instruction(vcpu, kvm_run,
2662                                         vcpu->arch.mmio_fault_cr2, 0,
2663                                         EMULTYPE_NO_DECODE);
2664                 if (r == EMULATE_DO_MMIO) {
2665                         /*
2666                          * Read-modify-write.  Back to userspace.
2667                          */
2668                         r = 0;
2669                         goto out;
2670                 }
2671         }
2672 #endif
2673         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2674                 kvm_x86_ops->cache_regs(vcpu);
2675                 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2676                 kvm_x86_ops->decache_regs(vcpu);
2677         }
2678
2679         r = __vcpu_run(vcpu, kvm_run);
2680
2681 out:
2682         if (vcpu->sigset_active)
2683                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2684
2685         vcpu_put(vcpu);
2686         return r;
2687 }
2688
2689 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2690 {
2691         vcpu_load(vcpu);
2692
2693         kvm_x86_ops->cache_regs(vcpu);
2694
2695         regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2696         regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2697         regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2698         regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2699         regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2700         regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2701         regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2702         regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2703 #ifdef CONFIG_X86_64
2704         regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2705         regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2706         regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2707         regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2708         regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2709         regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
2710         regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
2711         regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
2712 #endif
2713
2714         regs->rip = vcpu->arch.rip;
2715         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2716
2717         /*
2718          * Don't leak debug flags in case they were set for guest debugging
2719          */
2720         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2721                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2722
2723         vcpu_put(vcpu);
2724
2725         return 0;
2726 }
2727
2728 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2729 {
2730         vcpu_load(vcpu);
2731
2732         vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
2733         vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
2734         vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
2735         vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
2736         vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
2737         vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
2738         vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
2739         vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
2740 #ifdef CONFIG_X86_64
2741         vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
2742         vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
2743         vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
2744         vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
2745         vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
2746         vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
2747         vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
2748         vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
2749 #endif
2750
2751         vcpu->arch.rip = regs->rip;
2752         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2753
2754         kvm_x86_ops->decache_regs(vcpu);
2755
2756         vcpu_put(vcpu);
2757
2758         return 0;
2759 }
2760
2761 static void get_segment(struct kvm_vcpu *vcpu,
2762                         struct kvm_segment *var, int seg)
2763 {
2764         return kvm_x86_ops->get_segment(vcpu, var, seg);
2765 }
2766
2767 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2768 {
2769         struct kvm_segment cs;
2770
2771         get_segment(vcpu, &cs, VCPU_SREG_CS);
2772         *db = cs.db;
2773         *l = cs.l;
2774 }
2775 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2776
2777 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2778                                   struct kvm_sregs *sregs)
2779 {
2780         struct descriptor_table dt;
2781         int pending_vec;
2782
2783         vcpu_load(vcpu);
2784
2785         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2786         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2787         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2788         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2789         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2790         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2791
2792         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2793         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2794
2795         kvm_x86_ops->get_idt(vcpu, &dt);
2796         sregs->idt.limit = dt.limit;
2797         sregs->idt.base = dt.base;
2798         kvm_x86_ops->get_gdt(vcpu, &dt);
2799         sregs->gdt.limit = dt.limit;
2800         sregs->gdt.base = dt.base;
2801
2802         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2803         sregs->cr0 = vcpu->arch.cr0;
2804         sregs->cr2 = vcpu->arch.cr2;
2805         sregs->cr3 = vcpu->arch.cr3;
2806         sregs->cr4 = vcpu->arch.cr4;
2807         sregs->cr8 = get_cr8(vcpu);
2808         sregs->efer = vcpu->arch.shadow_efer;
2809         sregs->apic_base = kvm_get_apic_base(vcpu);
2810
2811         if (irqchip_in_kernel(vcpu->kvm)) {
2812                 memset(sregs->interrupt_bitmap, 0,
2813                        sizeof sregs->interrupt_bitmap);
2814                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2815                 if (pending_vec >= 0)
2816                         set_bit(pending_vec,
2817                                 (unsigned long *)sregs->interrupt_bitmap);
2818         } else
2819                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
2820                        sizeof sregs->interrupt_bitmap);
2821
2822         vcpu_put(vcpu);
2823
2824         return 0;
2825 }
2826
2827 static void set_segment(struct kvm_vcpu *vcpu,
2828                         struct kvm_segment *var, int seg)
2829 {
2830         return kvm_x86_ops->set_segment(vcpu, var, seg);
2831 }
2832
2833 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2834                                   struct kvm_sregs *sregs)
2835 {
2836         int mmu_reset_needed = 0;
2837         int i, pending_vec, max_bits;
2838         struct descriptor_table dt;
2839
2840         vcpu_load(vcpu);
2841
2842         dt.limit = sregs->idt.limit;
2843         dt.base = sregs->idt.base;
2844         kvm_x86_ops->set_idt(vcpu, &dt);
2845         dt.limit = sregs->gdt.limit;
2846         dt.base = sregs->gdt.base;
2847         kvm_x86_ops->set_gdt(vcpu, &dt);
2848
2849         vcpu->arch.cr2 = sregs->cr2;
2850         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
2851         vcpu->arch.cr3 = sregs->cr3;
2852
2853         set_cr8(vcpu, sregs->cr8);
2854
2855         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
2856 #ifdef CONFIG_X86_64
2857         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2858 #endif
2859         kvm_set_apic_base(vcpu, sregs->apic_base);
2860
2861         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2862
2863         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
2864         vcpu->arch.cr0 = sregs->cr0;
2865         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2866
2867         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
2868         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2869         if (!is_long_mode(vcpu) && is_pae(vcpu))
2870                 load_pdptrs(vcpu, vcpu->arch.cr3);
2871
2872         if (mmu_reset_needed)
2873                 kvm_mmu_reset_context(vcpu);
2874
2875         if (!irqchip_in_kernel(vcpu->kvm)) {
2876                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
2877                        sizeof vcpu->arch.irq_pending);
2878                 vcpu->arch.irq_summary = 0;
2879                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
2880                         if (vcpu->arch.irq_pending[i])
2881                                 __set_bit(i, &vcpu->arch.irq_summary);
2882         } else {
2883                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2884                 pending_vec = find_first_bit(
2885                         (const unsigned long *)sregs->interrupt_bitmap,
2886                         max_bits);
2887                 /* Only pending external irq is handled here */
2888                 if (pending_vec < max_bits) {
2889                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2890                         pr_debug("Set back pending irq %d\n",
2891                                  pending_vec);
2892                 }
2893         }
2894
2895         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2896         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2897         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2898         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2899         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2900         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2901
2902         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2903         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2904
2905         vcpu_put(vcpu);
2906
2907         return 0;
2908 }
2909
2910 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2911                                     struct kvm_debug_guest *dbg)
2912 {
2913         int r;
2914
2915         vcpu_load(vcpu);
2916
2917         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2918
2919         vcpu_put(vcpu);
2920
2921         return r;
2922 }
2923
2924 /*
2925  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2926  * we have asm/x86/processor.h
2927  */
2928 struct fxsave {
2929         u16     cwd;
2930         u16     swd;
2931         u16     twd;
2932         u16     fop;
2933         u64     rip;
2934         u64     rdp;
2935         u32     mxcsr;
2936         u32     mxcsr_mask;
2937         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2938 #ifdef CONFIG_X86_64
2939         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2940 #else
2941         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2942 #endif
2943 };
2944
2945 /*
2946  * Translate a guest virtual address to a guest physical address.
2947  */
2948 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2949                                     struct kvm_translation *tr)
2950 {
2951         unsigned long vaddr = tr->linear_address;
2952         gpa_t gpa;
2953
2954         vcpu_load(vcpu);
2955         down_read(&current->mm->mmap_sem);
2956         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
2957         up_read(&current->mm->mmap_sem);
2958         tr->physical_address = gpa;
2959         tr->valid = gpa != UNMAPPED_GVA;
2960         tr->writeable = 1;
2961         tr->usermode = 0;
2962         vcpu_put(vcpu);
2963
2964         return 0;
2965 }
2966
2967 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2968 {
2969         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
2970
2971         vcpu_load(vcpu);
2972
2973         memcpy(fpu->fpr, fxsave->st_space, 128);
2974         fpu->fcw = fxsave->cwd;
2975         fpu->fsw = fxsave->swd;
2976         fpu->ftwx = fxsave->twd;
2977         fpu->last_opcode = fxsave->fop;
2978         fpu->last_ip = fxsave->rip;
2979         fpu->last_dp = fxsave->rdp;
2980         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2981
2982         vcpu_put(vcpu);
2983
2984         return 0;
2985 }
2986
2987 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2988 {
2989         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
2990
2991         vcpu_load(vcpu);
2992
2993         memcpy(fxsave->st_space, fpu->fpr, 128);
2994         fxsave->cwd = fpu->fcw;
2995         fxsave->swd = fpu->fsw;
2996         fxsave->twd = fpu->ftwx;
2997         fxsave->fop = fpu->last_opcode;
2998         fxsave->rip = fpu->last_ip;
2999         fxsave->rdp = fpu->last_dp;
3000         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3001
3002         vcpu_put(vcpu);
3003
3004         return 0;
3005 }
3006
3007 void fx_init(struct kvm_vcpu *vcpu)
3008 {
3009         unsigned after_mxcsr_mask;
3010
3011         /* Initialize guest FPU by resetting ours and saving into guest's */
3012         preempt_disable();
3013         fx_save(&vcpu->arch.host_fx_image);
3014         fpu_init();
3015         fx_save(&vcpu->arch.guest_fx_image);
3016         fx_restore(&vcpu->arch.host_fx_image);
3017         preempt_enable();
3018
3019         vcpu->arch.cr0 |= X86_CR0_ET;
3020         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3021         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3022         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3023                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3024 }
3025 EXPORT_SYMBOL_GPL(fx_init);
3026
3027 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3028 {
3029         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3030                 return;
3031
3032         vcpu->guest_fpu_loaded = 1;
3033         fx_save(&vcpu->arch.host_fx_image);
3034         fx_restore(&vcpu->arch.guest_fx_image);
3035 }
3036 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3037
3038 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3039 {
3040         if (!vcpu->guest_fpu_loaded)
3041                 return;
3042
3043         vcpu->guest_fpu_loaded = 0;
3044         fx_save(&vcpu->arch.guest_fx_image);
3045         fx_restore(&vcpu->arch.host_fx_image);
3046         ++vcpu->stat.fpu_reload;
3047 }
3048 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3049
3050 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3051 {
3052         kvm_x86_ops->vcpu_free(vcpu);
3053 }
3054
3055 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3056                                                 unsigned int id)
3057 {
3058         return kvm_x86_ops->vcpu_create(kvm, id);
3059 }
3060
3061 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3062 {
3063         int r;
3064
3065         /* We do fxsave: this must be aligned. */
3066         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3067
3068         vcpu_load(vcpu);
3069         r = kvm_arch_vcpu_reset(vcpu);
3070         if (r == 0)
3071                 r = kvm_mmu_setup(vcpu);
3072         vcpu_put(vcpu);
3073         if (r < 0)
3074                 goto free_vcpu;
3075
3076         return 0;
3077 free_vcpu:
3078         kvm_x86_ops->vcpu_free(vcpu);
3079         return r;
3080 }
3081
3082 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3083 {
3084         vcpu_load(vcpu);
3085         kvm_mmu_unload(vcpu);
3086         vcpu_put(vcpu);
3087
3088         kvm_x86_ops->vcpu_free(vcpu);
3089 }
3090
3091 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3092 {
3093         return kvm_x86_ops->vcpu_reset(vcpu);
3094 }
3095
3096 void kvm_arch_hardware_enable(void *garbage)
3097 {
3098         kvm_x86_ops->hardware_enable(garbage);
3099 }
3100
3101 void kvm_arch_hardware_disable(void *garbage)
3102 {
3103         kvm_x86_ops->hardware_disable(garbage);
3104 }
3105
3106 int kvm_arch_hardware_setup(void)
3107 {
3108         return kvm_x86_ops->hardware_setup();
3109 }
3110
3111 void kvm_arch_hardware_unsetup(void)
3112 {
3113         kvm_x86_ops->hardware_unsetup();
3114 }
3115
3116 void kvm_arch_check_processor_compat(void *rtn)
3117 {
3118         kvm_x86_ops->check_processor_compatibility(rtn);
3119 }
3120
3121 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3122 {
3123         struct page *page;
3124         struct kvm *kvm;
3125         int r;
3126
3127         BUG_ON(vcpu->kvm == NULL);
3128         kvm = vcpu->kvm;
3129
3130         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3131         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3132                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
3133         else
3134                 vcpu->arch.mp_state = VCPU_MP_STATE_UNINITIALIZED;
3135
3136         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3137         if (!page) {
3138                 r = -ENOMEM;
3139                 goto fail;
3140         }
3141         vcpu->arch.pio_data = page_address(page);
3142
3143         r = kvm_mmu_create(vcpu);
3144         if (r < 0)
3145                 goto fail_free_pio_data;
3146
3147         if (irqchip_in_kernel(kvm)) {
3148                 r = kvm_create_lapic(vcpu);
3149                 if (r < 0)
3150                         goto fail_mmu_destroy;
3151         }
3152
3153         return 0;
3154
3155 fail_mmu_destroy:
3156         kvm_mmu_destroy(vcpu);
3157 fail_free_pio_data:
3158         free_page((unsigned long)vcpu->arch.pio_data);
3159 fail:
3160         return r;
3161 }
3162
3163 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3164 {
3165         kvm_free_lapic(vcpu);
3166         kvm_mmu_destroy(vcpu);
3167         free_page((unsigned long)vcpu->arch.pio_data);
3168 }
3169
3170 struct  kvm *kvm_arch_create_vm(void)
3171 {
3172         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3173
3174         if (!kvm)
3175                 return ERR_PTR(-ENOMEM);
3176
3177         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3178
3179         return kvm;
3180 }
3181
3182 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3183 {
3184         vcpu_load(vcpu);
3185         kvm_mmu_unload(vcpu);
3186         vcpu_put(vcpu);
3187 }
3188
3189 static void kvm_free_vcpus(struct kvm *kvm)
3190 {
3191         unsigned int i;
3192
3193         /*
3194          * Unpin any mmu pages first.
3195          */
3196         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3197                 if (kvm->vcpus[i])
3198                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3199         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3200                 if (kvm->vcpus[i]) {
3201                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3202                         kvm->vcpus[i] = NULL;
3203                 }
3204         }
3205
3206 }
3207
3208 void kvm_arch_destroy_vm(struct kvm *kvm)
3209 {
3210         kfree(kvm->arch.vpic);
3211         kfree(kvm->arch.vioapic);
3212         kvm_free_vcpus(kvm);
3213         kvm_free_physmem(kvm);
3214         kfree(kvm);
3215 }
3216
3217 int kvm_arch_set_memory_region(struct kvm *kvm,
3218                                 struct kvm_userspace_memory_region *mem,
3219                                 struct kvm_memory_slot old,
3220                                 int user_alloc)
3221 {
3222         int npages = mem->memory_size >> PAGE_SHIFT;
3223         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3224
3225         /*To keep backward compatibility with older userspace,
3226          *x86 needs to hanlde !user_alloc case.
3227          */
3228         if (!user_alloc) {
3229                 if (npages && !old.rmap) {
3230                         memslot->userspace_addr = do_mmap(NULL, 0,
3231                                                      npages * PAGE_SIZE,
3232                                                      PROT_READ | PROT_WRITE,
3233                                                      MAP_SHARED | MAP_ANONYMOUS,
3234                                                      0);
3235
3236                         if (IS_ERR((void *)memslot->userspace_addr))
3237                                 return PTR_ERR((void *)memslot->userspace_addr);
3238                 } else {
3239                         if (!old.user_alloc && old.rmap) {
3240                                 int ret;
3241
3242                                 ret = do_munmap(current->mm, old.userspace_addr,
3243                                                 old.npages * PAGE_SIZE);
3244                                 if (ret < 0)
3245                                         printk(KERN_WARNING
3246                                        "kvm_vm_ioctl_set_memory_region: "
3247                                        "failed to munmap memory\n");
3248                         }
3249                 }
3250         }
3251
3252         if (!kvm->arch.n_requested_mmu_pages) {
3253                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3254                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3255         }
3256
3257         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3258         kvm_flush_remote_tlbs(kvm);
3259
3260         return 0;
3261 }
3262
3263 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3264 {
3265         return vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE
3266                || vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED;
3267 }
3268
3269 static void vcpu_kick_intr(void *info)
3270 {
3271 #ifdef DEBUG
3272         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
3273         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
3274 #endif
3275 }
3276
3277 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3278 {
3279         int ipi_pcpu = vcpu->cpu;
3280
3281         if (waitqueue_active(&vcpu->wq)) {
3282                 wake_up_interruptible(&vcpu->wq);
3283                 ++vcpu->stat.halt_wakeup;
3284         }
3285         if (vcpu->guest_mode)
3286                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);
3287 }