arch/x86/kvm/vmx.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "irq.h"
  19 #include "mmu.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/module.h>
  23 #include <linux/kernel.h>
  24 #include <linux/mm.h>
  25 #include <linux/highmem.h>
  26 #include <linux/sched.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/ftrace_event.h>
  29 #include "kvm_cache_regs.h"
  30 #include "x86.h"
  31
  32 #include <asm/io.h>
  33 #include <asm/desc.h>
  34 #include <asm/vmx.h>
  35 #include <asm/virtext.h>
  36 #include <asm/mce.h>
  37
  38 #include "trace.h"
  39
  40 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  41
  42 MODULE_AUTHOR("Qumranet");
  43 MODULE_LICENSE("GPL");
  44
  45 static int __read_mostly bypass_guest_pf = 1;
  46 module_param(bypass_guest_pf, bool, S_IRUGO);
  47
  48 static int __read_mostly enable_vpid = 1;
  49 module_param_named(vpid, enable_vpid, bool, 0444);
  50
  51 static int __read_mostly flexpriority_enabled = 1;
  52 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  53
  54 static int __read_mostly enable_ept = 1;
  55 module_param_named(ept, enable_ept, bool, S_IRUGO);
  56
  57 static int __read_mostly enable_unrestricted_guest = 1;
  58 module_param_named(unrestricted_guest,
  59                         enable_unrestricted_guest, bool, S_IRUGO);
  60
  61 static int __read_mostly emulate_invalid_guest_state = 0;
  62 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  63
  64 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST                           \
  65         (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
  66 #define KVM_GUEST_CR0_MASK                                              \
  67         (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  68 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST                         \
  69         (X86_CR0_WP | X86_CR0_NE | X86_CR0_TS | X86_CR0_MP)
  70 #define KVM_VM_CR0_ALWAYS_ON                                            \
  71         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  72 #define KVM_CR4_GUEST_OWNED_BITS                                      \
  73         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
  74          | X86_CR4_OSXMMEXCPT)
  75
  76 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
  77 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
  78
  79 /*
  80  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
  81  * ple_gap:    upper bound on the amount of time between two successive
  82  *             executions of PAUSE in a loop. Also indicate if ple enabled.
  83  *             According to test, this time is usually small than 41 cycles.
  84  * ple_window: upper bound on the amount of time a guest is allowed to execute
  85  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
  86  *             less than 2^12 cycles
  87  * Time is measured based on a counter that runs at the same rate as the TSC,
  88  * refer SDM volume 3b section 21.6.13 & 22.1.3.
  89  */
  90 #define KVM_VMX_DEFAULT_PLE_GAP    41
  91 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
  92 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
  93 module_param(ple_gap, int, S_IRUGO);
  94
  95 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
  96 module_param(ple_window, int, S_IRUGO);
  97
  98 struct vmcs {
  99         u32 revision_id;
 100         u32 abort;
 101         char data[0];
 102 };
 103
 104 struct shared_msr_entry {
 105         unsigned index;
 106         u64 data;
 107         u64 mask;
 108 };
 109
 110 struct vcpu_vmx {
 111         struct kvm_vcpu       vcpu;
 112         struct list_head      local_vcpus_link;
 113         unsigned long         host_rsp;
 114         int                   launched;
 115         u8                    fail;
 116         u32                   idt_vectoring_info;
 117         struct shared_msr_entry *guest_msrs;
 118         int                   nmsrs;
 119         int                   save_nmsrs;
 120 #ifdef CONFIG_X86_64
 121         u64                   msr_host_kernel_gs_base;
 122         u64                   msr_guest_kernel_gs_base;
 123 #endif
 124         struct vmcs          *vmcs;
 125         struct {
 126                 int           loaded;
 127                 u16           fs_sel, gs_sel, ldt_sel;
 128                 int           gs_ldt_reload_needed;
 129                 int           fs_reload_needed;
 130         } host_state;
 131         struct {
 132                 int vm86_active;
 133                 u8 save_iopl;
 134                 struct kvm_save_segment {
 135                         u16 selector;
 136                         unsigned long base;
 137                         u32 limit;
 138                         u32 ar;
 139                 } tr, es, ds, fs, gs;
 140                 struct {
 141                         bool pending;
 142                         u8 vector;
 143                         unsigned rip;
 144                 } irq;
 145         } rmode;
 146         int vpid;
 147         bool emulation_required;
 148
 149         /* Support for vnmi-less CPUs */
 150         int soft_vnmi_blocked;
 151         ktime_t entry_time;
 152         s64 vnmi_blocked_time;
 153         u32 exit_reason;
 154
 155         bool rdtscp_enabled;
 156 };
 157
 158 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 159 {
 160         return container_of(vcpu, struct vcpu_vmx, vcpu);
 161 }
 162
 163 static int init_rmode(struct kvm *kvm);
 164 static u64 construct_eptp(unsigned long root_hpa);
 165
 166 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 167 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 168 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
 169
 170 static unsigned long *vmx_io_bitmap_a;
 171 static unsigned long *vmx_io_bitmap_b;
 172 static unsigned long *vmx_msr_bitmap_legacy;
 173 static unsigned long *vmx_msr_bitmap_longmode;
 174
 175 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 176 static DEFINE_SPINLOCK(vmx_vpid_lock);
 177
 178 static struct vmcs_config {
 179         int size;
 180         int order;
 181         u32 revision_id;
 182         u32 pin_based_exec_ctrl;
 183         u32 cpu_based_exec_ctrl;
 184         u32 cpu_based_2nd_exec_ctrl;
 185         u32 vmexit_ctrl;
 186         u32 vmentry_ctrl;
 187 } vmcs_config;
 188
 189 static struct vmx_capability {
 190         u32 ept;
 191         u32 vpid;
 192 } vmx_capability;
 193
 194 #define VMX_SEGMENT_FIELD(seg)                                  \
 195         [VCPU_SREG_##seg] = {                                   \
 196                 .selector = GUEST_##seg##_SELECTOR,             \
 197                 .base = GUEST_##seg##_BASE,                     \
 198                 .limit = GUEST_##seg##_LIMIT,                   \
 199                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 200         }
 201
 202 static struct kvm_vmx_segment_field {
 203         unsigned selector;
 204         unsigned base;
 205         unsigned limit;
 206         unsigned ar_bytes;
 207 } kvm_vmx_segment_fields[] = {
 208         VMX_SEGMENT_FIELD(CS),
 209         VMX_SEGMENT_FIELD(DS),
 210         VMX_SEGMENT_FIELD(ES),
 211         VMX_SEGMENT_FIELD(FS),
 212         VMX_SEGMENT_FIELD(GS),
 213         VMX_SEGMENT_FIELD(SS),
 214         VMX_SEGMENT_FIELD(TR),
 215         VMX_SEGMENT_FIELD(LDTR),
 216 };
 217
 218 static u64 host_efer;
 219
 220 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 221
 222 /*
 223  * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
 224  * away by decrementing the array size.
 225  */
 226 static const u32 vmx_msr_index[] = {
 227 #ifdef CONFIG_X86_64
 228         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 229 #endif
 230         MSR_EFER, MSR_TSC_AUX, MSR_K6_STAR,
 231 };
 232 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 233
 234 static inline int is_page_fault(u32 intr_info)
 235 {
 236         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 237                              INTR_INFO_VALID_MASK)) ==
 238                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 239 }
 240
 241 static inline int is_no_device(u32 intr_info)
 242 {
 243         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 244                              INTR_INFO_VALID_MASK)) ==
 245                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 246 }
 247
 248 static inline int is_invalid_opcode(u32 intr_info)
 249 {
 250         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 251                              INTR_INFO_VALID_MASK)) ==
 252                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 253 }
 254
 255 static inline int is_external_interrupt(u32 intr_info)
 256 {
 257         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 258                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 259 }
 260
 261 static inline int is_machine_check(u32 intr_info)
 262 {
 263         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 264                              INTR_INFO_VALID_MASK)) ==
 265                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
 266 }
 267
 268 static inline int cpu_has_vmx_msr_bitmap(void)
 269 {
 270         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
 271 }
 272
 273 static inline int cpu_has_vmx_tpr_shadow(void)
 274 {
 275         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
 276 }
 277
 278 static inline int vm_need_tpr_shadow(struct kvm *kvm)
 279 {
 280         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
 281 }
 282
 283 static inline int cpu_has_secondary_exec_ctrls(void)
 284 {
 285         return vmcs_config.cpu_based_exec_ctrl &
 286                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
 287 }
 288
 289 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 290 {
 291         return vmcs_config.cpu_based_2nd_exec_ctrl &
 292                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
 293 }
 294
 295 static inline bool cpu_has_vmx_flexpriority(void)
 296 {
 297         return cpu_has_vmx_tpr_shadow() &&
 298                 cpu_has_vmx_virtualize_apic_accesses();
 299 }
 300
 301 static inline bool cpu_has_vmx_ept_execute_only(void)
 302 {
 303         return !!(vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT);
 304 }
 305
 306 static inline bool cpu_has_vmx_eptp_uncacheable(void)
 307 {
 308         return !!(vmx_capability.ept & VMX_EPTP_UC_BIT);
 309 }
 310
 311 static inline bool cpu_has_vmx_eptp_writeback(void)
 312 {
 313         return !!(vmx_capability.ept & VMX_EPTP_WB_BIT);
 314 }
 315
 316 static inline bool cpu_has_vmx_ept_2m_page(void)
 317 {
 318         return !!(vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT);
 319 }
 320
 321 static inline int cpu_has_vmx_invept_individual_addr(void)
 322 {
 323         return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
 324 }
 325
 326 static inline int cpu_has_vmx_invept_context(void)
 327 {
 328         return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
 329 }
 330
 331 static inline int cpu_has_vmx_invept_global(void)
 332 {
 333         return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
 334 }
 335
 336 static inline int cpu_has_vmx_ept(void)
 337 {
 338         return vmcs_config.cpu_based_2nd_exec_ctrl &
 339                 SECONDARY_EXEC_ENABLE_EPT;
 340 }
 341
 342 static inline int cpu_has_vmx_unrestricted_guest(void)
 343 {
 344         return vmcs_config.cpu_based_2nd_exec_ctrl &
 345                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
 346 }
 347
 348 static inline int cpu_has_vmx_ple(void)
 349 {
 350         return vmcs_config.cpu_based_2nd_exec_ctrl &
 351                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 352 }
 353
 354 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
 355 {
 356         return flexpriority_enabled &&
 357                 (cpu_has_vmx_virtualize_apic_accesses()) &&
 358                 (irqchip_in_kernel(kvm));
 359 }
 360
 361 static inline int cpu_has_vmx_vpid(void)
 362 {
 363         return vmcs_config.cpu_based_2nd_exec_ctrl &
 364                 SECONDARY_EXEC_ENABLE_VPID;
 365 }
 366
 367 static inline int cpu_has_vmx_rdtscp(void)
 368 {
 369         return vmcs_config.cpu_based_2nd_exec_ctrl &
 370                 SECONDARY_EXEC_RDTSCP;
 371 }
 372
 373 static inline int cpu_has_virtual_nmis(void)
 374 {
 375         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
 376 }
 377
 378 static inline bool report_flexpriority(void)
 379 {
 380         return flexpriority_enabled;
 381 }
 382
 383 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
 384 {
 385         int i;
 386
 387         for (i = 0; i < vmx->nmsrs; ++i)
 388                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
 389                         return i;
 390         return -1;
 391 }
 392
 393 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
 394 {
 395     struct {
 396         u64 vpid : 16;
 397         u64 rsvd : 48;
 398         u64 gva;
 399     } operand = { vpid, 0, gva };
 400
 401     asm volatile (__ex(ASM_VMX_INVVPID)
 402                   /* CF==1 or ZF==1 --> rc = -1 */
 403                   "; ja 1f ; ud2 ; 1:"
 404                   : : "a"(&operand), "c"(ext) : "cc", "memory");
 405 }
 406
 407 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
 408 {
 409         struct {
 410                 u64 eptp, gpa;
 411         } operand = {eptp, gpa};
 412
 413         asm volatile (__ex(ASM_VMX_INVEPT)
 414                         /* CF==1 or ZF==1 --> rc = -1 */
 415                         "; ja 1f ; ud2 ; 1:\n"
 416                         : : "a" (&operand), "c" (ext) : "cc", "memory");
 417 }
 418
 419 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
 420 {
 421         int i;
 422
 423         i = __find_msr_index(vmx, msr);
 424         if (i >= 0)
 425                 return &vmx->guest_msrs[i];
 426         return NULL;
 427 }
 428
 429 static void vmcs_clear(struct vmcs *vmcs)
 430 {
 431         u64 phys_addr = __pa(vmcs);
 432         u8 error;
 433
 434         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
 435                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 436                       : "cc", "memory");
 437         if (error)
 438                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 439                        vmcs, phys_addr);
 440 }
 441
 442 static void __vcpu_clear(void *arg)
 443 {
 444         struct vcpu_vmx *vmx = arg;
 445         int cpu = raw_smp_processor_id();
 446
 447         if (vmx->vcpu.cpu == cpu)
 448                 vmcs_clear(vmx->vmcs);
 449         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
 450                 per_cpu(current_vmcs, cpu) = NULL;
 451         rdtscll(vmx->vcpu.arch.host_tsc);
 452         list_del(&vmx->local_vcpus_link);
 453         vmx->vcpu.cpu = -1;
 454         vmx->launched = 0;
 455 }
 456
 457 static void vcpu_clear(struct vcpu_vmx *vmx)
 458 {
 459         if (vmx->vcpu.cpu == -1)
 460                 return;
 461         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
 462 }
 463
 464 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
 465 {
 466         if (vmx->vpid == 0)
 467                 return;
 468
 469         __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
 470 }
 471
 472 static inline void ept_sync_global(void)
 473 {
 474         if (cpu_has_vmx_invept_global())
 475                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
 476 }
 477
 478 static inline void ept_sync_context(u64 eptp)
 479 {
 480         if (enable_ept) {
 481                 if (cpu_has_vmx_invept_context())
 482                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
 483                 else
 484                         ept_sync_global();
 485         }
 486 }
 487
 488 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
 489 {
 490         if (enable_ept) {
 491                 if (cpu_has_vmx_invept_individual_addr())
 492                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
 493                                         eptp, gpa);
 494                 else
 495                         ept_sync_context(eptp);
 496         }
 497 }
 498
 499 static unsigned long vmcs_readl(unsigned long field)
 500 {
 501         unsigned long value;
 502
 503         asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
 504                       : "=a"(value) : "d"(field) : "cc");
 505         return value;
 506 }
 507
 508 static u16 vmcs_read16(unsigned long field)
 509 {
 510         return vmcs_readl(field);
 511 }
 512
 513 static u32 vmcs_read32(unsigned long field)
 514 {
 515         return vmcs_readl(field);
 516 }
 517
 518 static u64 vmcs_read64(unsigned long field)
 519 {
 520 #ifdef CONFIG_X86_64
 521         return vmcs_readl(field);
 522 #else
 523         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
 524 #endif
 525 }
 526
 527 static noinline void vmwrite_error(unsigned long field, unsigned long value)
 528 {
 529         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
 530                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
 531         dump_stack();
 532 }
 533
 534 static void vmcs_writel(unsigned long field, unsigned long value)
 535 {
 536         u8 error;
 537
 538         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
 539                        : "=q"(error) : "a"(value), "d"(field) : "cc");
 540         if (unlikely(error))
 541                 vmwrite_error(field, value);
 542 }
 543
 544 static void vmcs_write16(unsigned long field, u16 value)
 545 {
 546         vmcs_writel(field, value);
 547 }
 548
 549 static void vmcs_write32(unsigned long field, u32 value)
 550 {
 551         vmcs_writel(field, value);
 552 }
 553
 554 static void vmcs_write64(unsigned long field, u64 value)
 555 {
 556         vmcs_writel(field, value);
 557 #ifndef CONFIG_X86_64
 558         asm volatile ("");
 559         vmcs_writel(field+1, value >> 32);
 560 #endif
 561 }
 562
 563 static void vmcs_clear_bits(unsigned long field, u32 mask)
 564 {
 565         vmcs_writel(field, vmcs_readl(field) & ~mask);
 566 }
 567
 568 static void vmcs_set_bits(unsigned long field, u32 mask)
 569 {
 570         vmcs_writel(field, vmcs_readl(field) | mask);
 571 }
 572
 573 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
 574 {
 575         u32 eb;
 576
 577         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR);
 578         if (!vcpu->fpu_active)
 579                 eb |= 1u << NM_VECTOR;
 580         /*
 581          * Unconditionally intercept #DB so we can maintain dr6 without
 582          * reading it every exit.
 583          */
 584         eb |= 1u << DB_VECTOR;
 585         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
 586                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
 587                         eb |= 1u << BP_VECTOR;
 588         }
 589         if (to_vmx(vcpu)->rmode.vm86_active)
 590                 eb = ~0;
 591         if (enable_ept)
 592                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
 593         vmcs_write32(EXCEPTION_BITMAP, eb);
 594 }
 595
 596 static void reload_tss(void)
 597 {
 598         /*
 599          * VT restores TR but not its size.  Useless.
 600          */
 601         struct descriptor_table gdt;
 602         struct desc_struct *descs;
 603
 604         kvm_get_gdt(&gdt);
 605         descs = (void *)gdt.base;
 606         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
 607         load_TR_desc();
 608 }
 609
 610 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
 611 {
 612         u64 guest_efer;
 613         u64 ignore_bits;
 614
 615         guest_efer = vmx->vcpu.arch.shadow_efer;
 616
 617         /*
 618          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
 619          * outside long mode
 620          */
 621         ignore_bits = EFER_NX | EFER_SCE;
 622 #ifdef CONFIG_X86_64
 623         ignore_bits |= EFER_LMA | EFER_LME;
 624         /* SCE is meaningful only in long mode on Intel */
 625         if (guest_efer & EFER_LMA)
 626                 ignore_bits &= ~(u64)EFER_SCE;
 627 #endif
 628         guest_efer &= ~ignore_bits;
 629         guest_efer |= host_efer & ignore_bits;
 630         vmx->guest_msrs[efer_offset].data = guest_efer;
 631         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
 632         return true;
 633 }
 634
 635 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
 636 {
 637         struct vcpu_vmx *vmx = to_vmx(vcpu);
 638         int i;
 639
 640         if (vmx->host_state.loaded)
 641                 return;
 642
 643         vmx->host_state.loaded = 1;
 644         /*
 645          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
 646          * allow segment selectors with cpl > 0 or ti == 1.
 647          */
 648         vmx->host_state.ldt_sel = kvm_read_ldt();
 649         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
 650         vmx->host_state.fs_sel = kvm_read_fs();
 651         if (!(vmx->host_state.fs_sel & 7)) {
 652                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
 653                 vmx->host_state.fs_reload_needed = 0;
 654         } else {
 655                 vmcs_write16(HOST_FS_SELECTOR, 0);
 656                 vmx->host_state.fs_reload_needed = 1;
 657         }
 658         vmx->host_state.gs_sel = kvm_read_gs();
 659         if (!(vmx->host_state.gs_sel & 7))
 660                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
 661         else {
 662                 vmcs_write16(HOST_GS_SELECTOR, 0);
 663                 vmx->host_state.gs_ldt_reload_needed = 1;
 664         }
 665
 666 #ifdef CONFIG_X86_64
 667         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
 668         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
 669 #else
 670         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
 671         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
 672 #endif
 673
 674 #ifdef CONFIG_X86_64
 675         if (is_long_mode(&vmx->vcpu)) {
 676                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
 677                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
 678         }
 679 #endif
 680         for (i = 0; i < vmx->save_nmsrs; ++i)
 681                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
 682                                    vmx->guest_msrs[i].data,
 683                                    vmx->guest_msrs[i].mask);
 684 }
 685
 686 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
 687 {
 688         unsigned long flags;
 689
 690         if (!vmx->host_state.loaded)
 691                 return;
 692
 693         ++vmx->vcpu.stat.host_state_reload;
 694         vmx->host_state.loaded = 0;
 695         if (vmx->host_state.fs_reload_needed)
 696                 kvm_load_fs(vmx->host_state.fs_sel);
 697         if (vmx->host_state.gs_ldt_reload_needed) {
 698                 kvm_load_ldt(vmx->host_state.ldt_sel);
 699                 /*
 700                  * If we have to reload gs, we must take care to
 701                  * preserve our gs base.
 702                  */
 703                 local_irq_save(flags);
 704                 kvm_load_gs(vmx->host_state.gs_sel);
 705 #ifdef CONFIG_X86_64
 706                 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
 707 #endif
 708                 local_irq_restore(flags);
 709         }
 710         reload_tss();
 711 #ifdef CONFIG_X86_64
 712         if (is_long_mode(&vmx->vcpu)) {
 713                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
 714                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
 715         }
 716 #endif
 717 }
 718
 719 static void vmx_load_host_state(struct vcpu_vmx *vmx)
 720 {
 721         preempt_disable();
 722         __vmx_load_host_state(vmx);
 723         preempt_enable();
 724 }
 725
 726 /*
 727  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
 728  * vcpu mutex is already taken.
 729  */
 730 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 731 {
 732         struct vcpu_vmx *vmx = to_vmx(vcpu);
 733         u64 phys_addr = __pa(vmx->vmcs);
 734         u64 tsc_this, delta, new_offset;
 735
 736         if (vcpu->cpu != cpu) {
 737                 vcpu_clear(vmx);
 738                 kvm_migrate_timers(vcpu);
 739                 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
 740                 local_irq_disable();
 741                 list_add(&vmx->local_vcpus_link,
 742                          &per_cpu(vcpus_on_cpu, cpu));
 743                 local_irq_enable();
 744         }
 745
 746         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
 747                 u8 error;
 748
 749                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
 750                 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
 751                               : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
 752                               : "cc");
 753                 if (error)
 754                         printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
 755                                vmx->vmcs, phys_addr);
 756         }
 757
 758         if (vcpu->cpu != cpu) {
 759                 struct descriptor_table dt;
 760                 unsigned long sysenter_esp;
 761
 762                 vcpu->cpu = cpu;
 763                 /*
 764                  * Linux uses per-cpu TSS and GDT, so set these when switching
 765                  * processors.
 766                  */
 767                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
 768                 kvm_get_gdt(&dt);
 769                 vmcs_writel(HOST_GDTR_BASE, dt.base);   /* 22.2.4 */
 770
 771                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
 772                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
 773
 774                 /*
 775                  * Make sure the time stamp counter is monotonous.
 776                  */
 777                 rdtscll(tsc_this);
 778                 if (tsc_this < vcpu->arch.host_tsc) {
 779                         delta = vcpu->arch.host_tsc - tsc_this;
 780                         new_offset = vmcs_read64(TSC_OFFSET) + delta;
 781                         vmcs_write64(TSC_OFFSET, new_offset);
 782                 }
 783         }
 784 }
 785
 786 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
 787 {
 788         __vmx_load_host_state(to_vmx(vcpu));
 789 }
 790
 791 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
 792 {
 793         if (vcpu->fpu_active)
 794                 return;
 795         vcpu->fpu_active = 1;
 796         vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
 797         if (vcpu->arch.cr0 & X86_CR0_TS)
 798                 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
 799         update_exception_bitmap(vcpu);
 800 }
 801
 802 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
 803 {
 804         if (!vcpu->fpu_active)
 805                 return;
 806         vcpu->fpu_active = 0;
 807         vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
 808         update_exception_bitmap(vcpu);
 809 }
 810
 811 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
 812 {
 813         unsigned long rflags;
 814
 815         rflags = vmcs_readl(GUEST_RFLAGS);
 816         if (to_vmx(vcpu)->rmode.vm86_active)
 817                 rflags &= ~(unsigned long)(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
 818         return rflags;
 819 }
 820
 821 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 822 {
 823         if (to_vmx(vcpu)->rmode.vm86_active)
 824                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
 825         vmcs_writel(GUEST_RFLAGS, rflags);
 826 }
 827
 828 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 829 {
 830         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
 831         int ret = 0;
 832
 833         if (interruptibility & GUEST_INTR_STATE_STI)
 834                 ret |= X86_SHADOW_INT_STI;
 835         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
 836                 ret |= X86_SHADOW_INT_MOV_SS;
 837
 838         return ret & mask;
 839 }
 840
 841 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 842 {
 843         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
 844         u32 interruptibility = interruptibility_old;
 845
 846         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
 847
 848         if (mask & X86_SHADOW_INT_MOV_SS)
 849                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
 850         if (mask & X86_SHADOW_INT_STI)
 851                 interruptibility |= GUEST_INTR_STATE_STI;
 852
 853         if ((interruptibility != interruptibility_old))
 854                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
 855 }
 856
 857 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 858 {
 859         unsigned long rip;
 860
 861         rip = kvm_rip_read(vcpu);
 862         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
 863         kvm_rip_write(vcpu, rip);
 864
 865         /* skipping an emulated instruction also counts */
 866         vmx_set_interrupt_shadow(vcpu, 0);
 867 }
 868
 869 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 870                                 bool has_error_code, u32 error_code)
 871 {
 872         struct vcpu_vmx *vmx = to_vmx(vcpu);
 873         u32 intr_info = nr | INTR_INFO_VALID_MASK;
 874
 875         if (has_error_code) {
 876                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
 877                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
 878         }
 879
 880         if (vmx->rmode.vm86_active) {
 881                 vmx->rmode.irq.pending = true;
 882                 vmx->rmode.irq.vector = nr;
 883                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
 884                 if (kvm_exception_is_soft(nr))
 885                         vmx->rmode.irq.rip +=
 886                                 vmx->vcpu.arch.event_exit_inst_len;
 887                 intr_info |= INTR_TYPE_SOFT_INTR;
 888                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
 889                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
 890                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
 891                 return;
 892         }
 893
 894         if (kvm_exception_is_soft(nr)) {
 895                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
 896                              vmx->vcpu.arch.event_exit_inst_len);
 897                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
 898         } else
 899                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
 900
 901         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
 902 }
 903
 904 static bool vmx_rdtscp_supported(void)
 905 {
 906         return cpu_has_vmx_rdtscp();
 907 }
 908
 909 /*
 910  * Swap MSR entry in host/guest MSR entry array.
 911  */
 912 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
 913 {
 914         struct shared_msr_entry tmp;
 915
 916         tmp = vmx->guest_msrs[to];
 917         vmx->guest_msrs[to] = vmx->guest_msrs[from];
 918         vmx->guest_msrs[from] = tmp;
 919 }
 920
 921 /*
 922  * Set up the vmcs to automatically save and restore system
 923  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
 924  * mode, as fiddling with msrs is very expensive.
 925  */
 926 static void setup_msrs(struct vcpu_vmx *vmx)
 927 {
 928         int save_nmsrs, index;
 929         unsigned long *msr_bitmap;
 930
 931         vmx_load_host_state(vmx);
 932         save_nmsrs = 0;
 933 #ifdef CONFIG_X86_64
 934         if (is_long_mode(&vmx->vcpu)) {
 935                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
 936                 if (index >= 0)
 937                         move_msr_up(vmx, index, save_nmsrs++);
 938                 index = __find_msr_index(vmx, MSR_LSTAR);
 939                 if (index >= 0)
 940                         move_msr_up(vmx, index, save_nmsrs++);
 941                 index = __find_msr_index(vmx, MSR_CSTAR);
 942                 if (index >= 0)
 943                         move_msr_up(vmx, index, save_nmsrs++);
 944                 index = __find_msr_index(vmx, MSR_TSC_AUX);
 945                 if (index >= 0 && vmx->rdtscp_enabled)
 946                         move_msr_up(vmx, index, save_nmsrs++);
 947                 /*
 948                  * MSR_K6_STAR is only needed on long mode guests, and only
 949                  * if efer.sce is enabled.
 950                  */
 951                 index = __find_msr_index(vmx, MSR_K6_STAR);
 952                 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
 953                         move_msr_up(vmx, index, save_nmsrs++);
 954         }
 955 #endif
 956         index = __find_msr_index(vmx, MSR_EFER);
 957         if (index >= 0 && update_transition_efer(vmx, index))
 958                 move_msr_up(vmx, index, save_nmsrs++);
 959
 960         vmx->save_nmsrs = save_nmsrs;
 961
 962         if (cpu_has_vmx_msr_bitmap()) {
 963                 if (is_long_mode(&vmx->vcpu))
 964                         msr_bitmap = vmx_msr_bitmap_longmode;
 965                 else
 966                         msr_bitmap = vmx_msr_bitmap_legacy;
 967
 968                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
 969         }
 970 }
 971
 972 /*
 973  * reads and returns guest's timestamp counter "register"
 974  * guest_tsc = host_tsc + tsc_offset    -- 21.3
 975  */
 976 static u64 guest_read_tsc(void)
 977 {
 978         u64 host_tsc, tsc_offset;
 979
 980         rdtscll(host_tsc);
 981         tsc_offset = vmcs_read64(TSC_OFFSET);
 982         return host_tsc + tsc_offset;
 983 }
 984
 985 /*
 986  * writes 'guest_tsc' into guest's timestamp counter "register"
 987  * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
 988  */
 989 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
 990 {
 991         vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
 992 }
 993
 994 /*
 995  * Reads an msr value (of 'msr_index') into 'pdata'.
 996  * Returns 0 on success, non-0 otherwise.
 997  * Assumes vcpu_load() was already called.
 998  */
 999 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1000 {
1001         u64 data;
1002         struct shared_msr_entry *msr;
1003
1004         if (!pdata) {
1005                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1006                 return -EINVAL;
1007         }
1008
1009         switch (msr_index) {
1010 #ifdef CONFIG_X86_64
1011         case MSR_FS_BASE:
1012                 data = vmcs_readl(GUEST_FS_BASE);
1013                 break;
1014         case MSR_GS_BASE:
1015                 data = vmcs_readl(GUEST_GS_BASE);
1016                 break;
1017         case MSR_KERNEL_GS_BASE:
1018                 vmx_load_host_state(to_vmx(vcpu));
1019                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1020                 break;
1021 #endif
1022         case MSR_EFER:
1023                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1024         case MSR_IA32_TSC:
1025                 data = guest_read_tsc();
1026                 break;
1027         case MSR_IA32_SYSENTER_CS:
1028                 data = vmcs_read32(GUEST_SYSENTER_CS);
1029                 break;
1030         case MSR_IA32_SYSENTER_EIP:
1031                 data = vmcs_readl(GUEST_SYSENTER_EIP);
1032                 break;
1033         case MSR_IA32_SYSENTER_ESP:
1034                 data = vmcs_readl(GUEST_SYSENTER_ESP);
1035                 break;
1036         case MSR_TSC_AUX:
1037                 if (!to_vmx(vcpu)->rdtscp_enabled)
1038                         return 1;
1039                 /* Otherwise falls through */
1040         default:
1041                 vmx_load_host_state(to_vmx(vcpu));
1042                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1043                 if (msr) {
1044                         vmx_load_host_state(to_vmx(vcpu));
1045                         data = msr->data;
1046                         break;
1047                 }
1048                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1049         }
1050
1051         *pdata = data;
1052         return 0;
1053 }
1054
1055 /*
1056  * Writes msr value into into the appropriate "register".
1057  * Returns 0 on success, non-0 otherwise.
1058  * Assumes vcpu_load() was already called.
1059  */
1060 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1061 {
1062         struct vcpu_vmx *vmx = to_vmx(vcpu);
1063         struct shared_msr_entry *msr;
1064         u64 host_tsc;
1065         int ret = 0;
1066
1067         switch (msr_index) {
1068         case MSR_EFER:
1069                 vmx_load_host_state(vmx);
1070                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1071                 break;
1072 #ifdef CONFIG_X86_64
1073         case MSR_FS_BASE:
1074                 vmcs_writel(GUEST_FS_BASE, data);
1075                 break;
1076         case MSR_GS_BASE:
1077                 vmcs_writel(GUEST_GS_BASE, data);
1078                 break;
1079         case MSR_KERNEL_GS_BASE:
1080                 vmx_load_host_state(vmx);
1081                 vmx->msr_guest_kernel_gs_base = data;
1082                 break;
1083 #endif
1084         case MSR_IA32_SYSENTER_CS:
1085                 vmcs_write32(GUEST_SYSENTER_CS, data);
1086                 break;
1087         case MSR_IA32_SYSENTER_EIP:
1088                 vmcs_writel(GUEST_SYSENTER_EIP, data);
1089                 break;
1090         case MSR_IA32_SYSENTER_ESP:
1091                 vmcs_writel(GUEST_SYSENTER_ESP, data);
1092                 break;
1093         case MSR_IA32_TSC:
1094                 rdtscll(host_tsc);
1095                 guest_write_tsc(data, host_tsc);
1096                 break;
1097         case MSR_IA32_CR_PAT:
1098                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1099                         vmcs_write64(GUEST_IA32_PAT, data);
1100                         vcpu->arch.pat = data;
1101                         break;
1102                 }
1103                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1104                 break;
1105         case MSR_TSC_AUX:
1106                 if (!vmx->rdtscp_enabled)
1107                         return 1;
1108                 /* Check reserved bit, higher 32 bits should be zero */
1109                 if ((data >> 32) != 0)
1110                         return 1;
1111                 /* Otherwise falls through */
1112         default:
1113                 msr = find_msr_entry(vmx, msr_index);
1114                 if (msr) {
1115                         vmx_load_host_state(vmx);
1116                         msr->data = data;
1117                         break;
1118                 }
1119                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1120         }
1121
1122         return ret;
1123 }
1124
1125 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1126 {
1127         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1128         switch (reg) {
1129         case VCPU_REGS_RSP:
1130                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1131                 break;
1132         case VCPU_REGS_RIP:
1133                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1134                 break;
1135         case VCPU_EXREG_PDPTR:
1136                 if (enable_ept)
1137                         ept_save_pdptrs(vcpu);
1138                 break;
1139         default:
1140                 break;
1141         }
1142 }
1143
1144 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1145 {
1146         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1147                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1148         else
1149                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1150
1151         update_exception_bitmap(vcpu);
1152 }
1153
1154 static __init int cpu_has_kvm_support(void)
1155 {
1156         return cpu_has_vmx();
1157 }
1158
1159 static __init int vmx_disabled_by_bios(void)
1160 {
1161         u64 msr;
1162
1163         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1164         return (msr & (FEATURE_CONTROL_LOCKED |
1165                        FEATURE_CONTROL_VMXON_ENABLED))
1166             == FEATURE_CONTROL_LOCKED;
1167         /* locked but not enabled */
1168 }
1169
1170 static int hardware_enable(void *garbage)
1171 {
1172         int cpu = raw_smp_processor_id();
1173         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1174         u64 old;
1175
1176         if (read_cr4() & X86_CR4_VMXE)
1177                 return -EBUSY;
1178
1179         INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1180         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1181         if ((old & (FEATURE_CONTROL_LOCKED |
1182                     FEATURE_CONTROL_VMXON_ENABLED))
1183             != (FEATURE_CONTROL_LOCKED |
1184                 FEATURE_CONTROL_VMXON_ENABLED))
1185                 /* enable and lock */
1186                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1187                        FEATURE_CONTROL_LOCKED |
1188                        FEATURE_CONTROL_VMXON_ENABLED);
1189         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1190         asm volatile (ASM_VMX_VMXON_RAX
1191                       : : "a"(&phys_addr), "m"(phys_addr)
1192                       : "memory", "cc");
1193
1194         ept_sync_global();
1195
1196         return 0;
1197 }
1198
1199 static void vmclear_local_vcpus(void)
1200 {
1201         int cpu = raw_smp_processor_id();
1202         struct vcpu_vmx *vmx, *n;
1203
1204         list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1205                                  local_vcpus_link)
1206                 __vcpu_clear(vmx);
1207 }
1208
1209
1210 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1211  * tricks.
1212  */
1213 static void kvm_cpu_vmxoff(void)
1214 {
1215         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1216         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1217 }
1218
1219 static void hardware_disable(void *garbage)
1220 {
1221         vmclear_local_vcpus();
1222         kvm_cpu_vmxoff();
1223 }
1224
1225 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1226                                       u32 msr, u32 *result)
1227 {
1228         u32 vmx_msr_low, vmx_msr_high;
1229         u32 ctl = ctl_min | ctl_opt;
1230
1231         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1232
1233         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1234         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1235
1236         /* Ensure minimum (required) set of control bits are supported. */
1237         if (ctl_min & ~ctl)
1238                 return -EIO;
1239
1240         *result = ctl;
1241         return 0;
1242 }
1243
1244 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1245 {
1246         u32 vmx_msr_low, vmx_msr_high;
1247         u32 min, opt, min2, opt2;
1248         u32 _pin_based_exec_control = 0;
1249         u32 _cpu_based_exec_control = 0;
1250         u32 _cpu_based_2nd_exec_control = 0;
1251         u32 _vmexit_control = 0;
1252         u32 _vmentry_control = 0;
1253
1254         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1255         opt = PIN_BASED_VIRTUAL_NMIS;
1256         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1257                                 &_pin_based_exec_control) < 0)
1258                 return -EIO;
1259
1260         min = CPU_BASED_HLT_EXITING |
1261 #ifdef CONFIG_X86_64
1262               CPU_BASED_CR8_LOAD_EXITING |
1263               CPU_BASED_CR8_STORE_EXITING |
1264 #endif
1265               CPU_BASED_CR3_LOAD_EXITING |
1266               CPU_BASED_CR3_STORE_EXITING |
1267               CPU_BASED_USE_IO_BITMAPS |
1268               CPU_BASED_MOV_DR_EXITING |
1269               CPU_BASED_USE_TSC_OFFSETING |
1270               CPU_BASED_MWAIT_EXITING |
1271               CPU_BASED_MONITOR_EXITING |
1272               CPU_BASED_INVLPG_EXITING;
1273         opt = CPU_BASED_TPR_SHADOW |
1274               CPU_BASED_USE_MSR_BITMAPS |
1275               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1276         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1277                                 &_cpu_based_exec_control) < 0)
1278                 return -EIO;
1279 #ifdef CONFIG_X86_64
1280         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1281                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1282                                            ~CPU_BASED_CR8_STORE_EXITING;
1283 #endif
1284         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1285                 min2 = 0;
1286                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1287                         SECONDARY_EXEC_WBINVD_EXITING |
1288                         SECONDARY_EXEC_ENABLE_VPID |
1289                         SECONDARY_EXEC_ENABLE_EPT |
1290                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
1291                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1292                         SECONDARY_EXEC_RDTSCP;
1293                 if (adjust_vmx_controls(min2, opt2,
1294                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1295                                         &_cpu_based_2nd_exec_control) < 0)
1296                         return -EIO;
1297         }
1298 #ifndef CONFIG_X86_64
1299         if (!(_cpu_based_2nd_exec_control &
1300                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1301                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1302 #endif
1303         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1304                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1305                    enabled */
1306                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1307                                              CPU_BASED_CR3_STORE_EXITING |
1308                                              CPU_BASED_INVLPG_EXITING);
1309                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1310                       vmx_capability.ept, vmx_capability.vpid);
1311         }
1312
1313         min = 0;
1314 #ifdef CONFIG_X86_64
1315         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1316 #endif
1317         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1318         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1319                                 &_vmexit_control) < 0)
1320                 return -EIO;
1321
1322         min = 0;
1323         opt = VM_ENTRY_LOAD_IA32_PAT;
1324         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1325                                 &_vmentry_control) < 0)
1326                 return -EIO;
1327
1328         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1329
1330         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1331         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1332                 return -EIO;
1333
1334 #ifdef CONFIG_X86_64
1335         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1336         if (vmx_msr_high & (1u<<16))
1337                 return -EIO;
1338 #endif
1339
1340         /* Require Write-Back (WB) memory type for VMCS accesses. */
1341         if (((vmx_msr_high >> 18) & 15) != 6)
1342                 return -EIO;
1343
1344         vmcs_conf->size = vmx_msr_high & 0x1fff;
1345         vmcs_conf->order = get_order(vmcs_config.size);
1346         vmcs_conf->revision_id = vmx_msr_low;
1347
1348         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1349         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1350         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1351         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1352         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1353
1354         return 0;
1355 }
1356
1357 static struct vmcs *alloc_vmcs_cpu(int cpu)
1358 {
1359         int node = cpu_to_node(cpu);
1360         struct page *pages;
1361         struct vmcs *vmcs;
1362
1363         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1364         if (!pages)
1365                 return NULL;
1366         vmcs = page_address(pages);
1367         memset(vmcs, 0, vmcs_config.size);
1368         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1369         return vmcs;
1370 }
1371
1372 static struct vmcs *alloc_vmcs(void)
1373 {
1374         return alloc_vmcs_cpu(raw_smp_processor_id());
1375 }
1376
1377 static void free_vmcs(struct vmcs *vmcs)
1378 {
1379         free_pages((unsigned long)vmcs, vmcs_config.order);
1380 }
1381
1382 static void free_kvm_area(void)
1383 {
1384         int cpu;
1385
1386         for_each_possible_cpu(cpu) {
1387                 free_vmcs(per_cpu(vmxarea, cpu));
1388                 per_cpu(vmxarea, cpu) = NULL;
1389         }
1390 }
1391
1392 static __init int alloc_kvm_area(void)
1393 {
1394         int cpu;
1395
1396         for_each_possible_cpu(cpu) {
1397                 struct vmcs *vmcs;
1398
1399                 vmcs = alloc_vmcs_cpu(cpu);
1400                 if (!vmcs) {
1401                         free_kvm_area();
1402                         return -ENOMEM;
1403                 }
1404
1405                 per_cpu(vmxarea, cpu) = vmcs;
1406         }
1407         return 0;
1408 }
1409
1410 static __init int hardware_setup(void)
1411 {
1412         if (setup_vmcs_config(&vmcs_config) < 0)
1413                 return -EIO;
1414
1415         if (boot_cpu_has(X86_FEATURE_NX))
1416                 kvm_enable_efer_bits(EFER_NX);
1417
1418         if (!cpu_has_vmx_vpid())
1419                 enable_vpid = 0;
1420
1421         if (!cpu_has_vmx_ept()) {
1422                 enable_ept = 0;
1423                 enable_unrestricted_guest = 0;
1424         }
1425
1426         if (!cpu_has_vmx_unrestricted_guest())
1427                 enable_unrestricted_guest = 0;
1428
1429         if (!cpu_has_vmx_flexpriority())
1430                 flexpriority_enabled = 0;
1431
1432         if (!cpu_has_vmx_tpr_shadow())
1433                 kvm_x86_ops->update_cr8_intercept = NULL;
1434
1435         if (enable_ept && !cpu_has_vmx_ept_2m_page())
1436                 kvm_disable_largepages();
1437
1438         if (!cpu_has_vmx_ple())
1439                 ple_gap = 0;
1440
1441         return alloc_kvm_area();
1442 }
1443
1444 static __exit void hardware_unsetup(void)
1445 {
1446         free_kvm_area();
1447 }
1448
1449 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1450 {
1451         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1452
1453         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1454                 vmcs_write16(sf->selector, save->selector);
1455                 vmcs_writel(sf->base, save->base);
1456                 vmcs_write32(sf->limit, save->limit);
1457                 vmcs_write32(sf->ar_bytes, save->ar);
1458         } else {
1459                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1460                         << AR_DPL_SHIFT;
1461                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1462         }
1463 }
1464
1465 static void enter_pmode(struct kvm_vcpu *vcpu)
1466 {
1467         unsigned long flags;
1468         struct vcpu_vmx *vmx = to_vmx(vcpu);
1469
1470         vmx->emulation_required = 1;
1471         vmx->rmode.vm86_active = 0;
1472
1473         vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1474         vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1475         vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1476
1477         flags = vmcs_readl(GUEST_RFLAGS);
1478         flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1479         flags |= (vmx->rmode.save_iopl << IOPL_SHIFT);
1480         vmcs_writel(GUEST_RFLAGS, flags);
1481
1482         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1483                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1484
1485         update_exception_bitmap(vcpu);
1486
1487         if (emulate_invalid_guest_state)
1488                 return;
1489
1490         fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1491         fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1492         fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1493         fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1494
1495         vmcs_write16(GUEST_SS_SELECTOR, 0);
1496         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1497
1498         vmcs_write16(GUEST_CS_SELECTOR,
1499                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1500         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1501 }
1502
1503 static gva_t rmode_tss_base(struct kvm *kvm)
1504 {
1505         if (!kvm->arch.tss_addr) {
1506                 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1507                                  kvm->memslots[0].npages - 3;
1508                 return base_gfn << PAGE_SHIFT;
1509         }
1510         return kvm->arch.tss_addr;
1511 }
1512
1513 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1514 {
1515         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1516
1517         save->selector = vmcs_read16(sf->selector);
1518         save->base = vmcs_readl(sf->base);
1519         save->limit = vmcs_read32(sf->limit);
1520         save->ar = vmcs_read32(sf->ar_bytes);
1521         vmcs_write16(sf->selector, save->base >> 4);
1522         vmcs_write32(sf->base, save->base & 0xfffff);
1523         vmcs_write32(sf->limit, 0xffff);
1524         vmcs_write32(sf->ar_bytes, 0xf3);
1525 }
1526
1527 static void enter_rmode(struct kvm_vcpu *vcpu)
1528 {
1529         unsigned long flags;
1530         struct vcpu_vmx *vmx = to_vmx(vcpu);
1531
1532         if (enable_unrestricted_guest)
1533                 return;
1534
1535         vmx->emulation_required = 1;
1536         vmx->rmode.vm86_active = 1;
1537
1538         vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1539         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1540
1541         vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1542         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1543
1544         vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1545         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1546
1547         flags = vmcs_readl(GUEST_RFLAGS);
1548         vmx->rmode.save_iopl
1549                 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1550
1551         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1552
1553         vmcs_writel(GUEST_RFLAGS, flags);
1554         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1555         update_exception_bitmap(vcpu);
1556
1557         if (emulate_invalid_guest_state)
1558                 goto continue_rmode;
1559
1560         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1561         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1562         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1563
1564         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1565         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1566         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1567                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1568         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1569
1570         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1571         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1572         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1573         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1574
1575 continue_rmode:
1576         kvm_mmu_reset_context(vcpu);
1577         init_rmode(vcpu->kvm);
1578 }
1579
1580 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1581 {
1582         struct vcpu_vmx *vmx = to_vmx(vcpu);
1583         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1584
1585         if (!msr)
1586                 return;
1587
1588         /*
1589          * Force kernel_gs_base reloading before EFER changes, as control
1590          * of this msr depends on is_long_mode().
1591          */
1592         vmx_load_host_state(to_vmx(vcpu));
1593         vcpu->arch.shadow_efer = efer;
1594         if (!msr)
1595                 return;
1596         if (efer & EFER_LMA) {
1597                 vmcs_write32(VM_ENTRY_CONTROLS,
1598                              vmcs_read32(VM_ENTRY_CONTROLS) |
1599                              VM_ENTRY_IA32E_MODE);
1600                 msr->data = efer;
1601         } else {
1602                 vmcs_write32(VM_ENTRY_CONTROLS,
1603                              vmcs_read32(VM_ENTRY_CONTROLS) &
1604                              ~VM_ENTRY_IA32E_MODE);
1605
1606                 msr->data = efer & ~EFER_LME;
1607         }
1608         setup_msrs(vmx);
1609 }
1610
1611 #ifdef CONFIG_X86_64
1612
1613 static void enter_lmode(struct kvm_vcpu *vcpu)
1614 {
1615         u32 guest_tr_ar;
1616
1617         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1618         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1619                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1620                        __func__);
1621                 vmcs_write32(GUEST_TR_AR_BYTES,
1622                              (guest_tr_ar & ~AR_TYPE_MASK)
1623                              | AR_TYPE_BUSY_64_TSS);
1624         }
1625         vcpu->arch.shadow_efer |= EFER_LMA;
1626         vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
1627 }
1628
1629 static void exit_lmode(struct kvm_vcpu *vcpu)
1630 {
1631         vcpu->arch.shadow_efer &= ~EFER_LMA;
1632
1633         vmcs_write32(VM_ENTRY_CONTROLS,
1634                      vmcs_read32(VM_ENTRY_CONTROLS)
1635                      & ~VM_ENTRY_IA32E_MODE);
1636 }
1637
1638 #endif
1639
1640 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1641 {
1642         vpid_sync_vcpu_all(to_vmx(vcpu));
1643         if (enable_ept)
1644                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1645 }
1646
1647 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1648 {
1649         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1650
1651         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1652         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1653 }
1654
1655 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1656 {
1657         if (!test_bit(VCPU_EXREG_PDPTR,
1658                       (unsigned long *)&vcpu->arch.regs_dirty))
1659                 return;
1660
1661         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1662                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1663                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1664                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1665                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1666         }
1667 }
1668
1669 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1670 {
1671         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1672                 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1673                 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1674                 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1675                 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1676         }
1677
1678         __set_bit(VCPU_EXREG_PDPTR,
1679                   (unsigned long *)&vcpu->arch.regs_avail);
1680         __set_bit(VCPU_EXREG_PDPTR,
1681                   (unsigned long *)&vcpu->arch.regs_dirty);
1682 }
1683
1684 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1685
1686 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1687                                         unsigned long cr0,
1688                                         struct kvm_vcpu *vcpu)
1689 {
1690         if (!(cr0 & X86_CR0_PG)) {
1691                 /* From paging/starting to nonpaging */
1692                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1693                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1694                              (CPU_BASED_CR3_LOAD_EXITING |
1695                               CPU_BASED_CR3_STORE_EXITING));
1696                 vcpu->arch.cr0 = cr0;
1697                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1698         } else if (!is_paging(vcpu)) {
1699                 /* From nonpaging to paging */
1700                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1701                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1702                              ~(CPU_BASED_CR3_LOAD_EXITING |
1703                                CPU_BASED_CR3_STORE_EXITING));
1704                 vcpu->arch.cr0 = cr0;
1705                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1706         }
1707
1708         if (!(cr0 & X86_CR0_WP))
1709                 *hw_cr0 &= ~X86_CR0_WP;
1710 }
1711
1712 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1713 {
1714         struct vcpu_vmx *vmx = to_vmx(vcpu);
1715         unsigned long hw_cr0;
1716
1717         if (enable_unrestricted_guest)
1718                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1719                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1720         else
1721                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1722
1723         vmx_fpu_deactivate(vcpu);
1724
1725         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1726                 enter_pmode(vcpu);
1727
1728         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1729                 enter_rmode(vcpu);
1730
1731 #ifdef CONFIG_X86_64
1732         if (vcpu->arch.shadow_efer & EFER_LME) {
1733                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1734                         enter_lmode(vcpu);
1735                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1736                         exit_lmode(vcpu);
1737         }
1738 #endif
1739
1740         if (enable_ept)
1741                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1742
1743         vmcs_writel(CR0_READ_SHADOW, cr0);
1744         vmcs_writel(GUEST_CR0, hw_cr0);
1745         vcpu->arch.cr0 = cr0;
1746
1747         if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1748                 vmx_fpu_activate(vcpu);
1749 }
1750
1751 static u64 construct_eptp(unsigned long root_hpa)
1752 {
1753         u64 eptp;
1754
1755         /* TODO write the value reading from MSR */
1756         eptp = VMX_EPT_DEFAULT_MT |
1757                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1758         eptp |= (root_hpa & PAGE_MASK);
1759
1760         return eptp;
1761 }
1762
1763 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1764 {
1765         unsigned long guest_cr3;
1766         u64 eptp;
1767
1768         guest_cr3 = cr3;
1769         if (enable_ept) {
1770                 eptp = construct_eptp(cr3);
1771                 vmcs_write64(EPT_POINTER, eptp);
1772                 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1773                         vcpu->kvm->arch.ept_identity_map_addr;
1774                 ept_load_pdptrs(vcpu);
1775         }
1776
1777         vmx_flush_tlb(vcpu);
1778         vmcs_writel(GUEST_CR3, guest_cr3);
1779         if (vcpu->arch.cr0 & X86_CR0_PE)
1780                 vmx_fpu_deactivate(vcpu);
1781 }
1782
1783 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1784 {
1785         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1786                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1787
1788         vcpu->arch.cr4 = cr4;
1789         if (enable_ept) {
1790                 if (!is_paging(vcpu)) {
1791                         hw_cr4 &= ~X86_CR4_PAE;
1792                         hw_cr4 |= X86_CR4_PSE;
1793                 } else if (!(cr4 & X86_CR4_PAE)) {
1794                         hw_cr4 &= ~X86_CR4_PAE;
1795                 }
1796         }
1797
1798         vmcs_writel(CR4_READ_SHADOW, cr4);
1799         vmcs_writel(GUEST_CR4, hw_cr4);
1800 }
1801
1802 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1803 {
1804         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1805
1806         return vmcs_readl(sf->base);
1807 }
1808
1809 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1810                             struct kvm_segment *var, int seg)
1811 {
1812         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1813         u32 ar;
1814
1815         var->base = vmcs_readl(sf->base);
1816         var->limit = vmcs_read32(sf->limit);
1817         var->selector = vmcs_read16(sf->selector);
1818         ar = vmcs_read32(sf->ar_bytes);
1819         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1820                 ar = 0;
1821         var->type = ar & 15;
1822         var->s = (ar >> 4) & 1;
1823         var->dpl = (ar >> 5) & 3;
1824         var->present = (ar >> 7) & 1;
1825         var->avl = (ar >> 12) & 1;
1826         var->l = (ar >> 13) & 1;
1827         var->db = (ar >> 14) & 1;
1828         var->g = (ar >> 15) & 1;
1829         var->unusable = (ar >> 16) & 1;
1830 }
1831
1832 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1833 {
1834         if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1835                 return 0;
1836
1837         if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1838                 return 3;
1839
1840         return vmcs_read16(GUEST_CS_SELECTOR) & 3;
1841 }
1842
1843 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1844 {
1845         u32 ar;
1846
1847         if (var->unusable)
1848                 ar = 1 << 16;
1849         else {
1850                 ar = var->type & 15;
1851                 ar |= (var->s & 1) << 4;
1852                 ar |= (var->dpl & 3) << 5;
1853                 ar |= (var->present & 1) << 7;
1854                 ar |= (var->avl & 1) << 12;
1855                 ar |= (var->l & 1) << 13;
1856                 ar |= (var->db & 1) << 14;
1857                 ar |= (var->g & 1) << 15;
1858         }
1859         if (ar == 0) /* a 0 value means unusable */
1860                 ar = AR_UNUSABLE_MASK;
1861
1862         return ar;
1863 }
1864
1865 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1866                             struct kvm_segment *var, int seg)
1867 {
1868         struct vcpu_vmx *vmx = to_vmx(vcpu);
1869         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1870         u32 ar;
1871
1872         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
1873                 vmx->rmode.tr.selector = var->selector;
1874                 vmx->rmode.tr.base = var->base;
1875                 vmx->rmode.tr.limit = var->limit;
1876                 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
1877                 return;
1878         }
1879         vmcs_writel(sf->base, var->base);
1880         vmcs_write32(sf->limit, var->limit);
1881         vmcs_write16(sf->selector, var->selector);
1882         if (vmx->rmode.vm86_active && var->s) {
1883                 /*
1884                  * Hack real-mode segments into vm86 compatibility.
1885                  */
1886                 if (var->base == 0xffff0000 && var->selector == 0xf000)
1887                         vmcs_writel(sf->base, 0xf0000);
1888                 ar = 0xf3;
1889         } else
1890                 ar = vmx_segment_access_rights(var);
1891
1892         /*
1893          *   Fix the "Accessed" bit in AR field of segment registers for older
1894          * qemu binaries.
1895          *   IA32 arch specifies that at the time of processor reset the
1896          * "Accessed" bit in the AR field of segment registers is 1. And qemu
1897          * is setting it to 0 in the usedland code. This causes invalid guest
1898          * state vmexit when "unrestricted guest" mode is turned on.
1899          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
1900          * tree. Newer qemu binaries with that qemu fix would not need this
1901          * kvm hack.
1902          */
1903         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
1904                 ar |= 0x1; /* Accessed */
1905
1906         vmcs_write32(sf->ar_bytes, ar);
1907 }
1908
1909 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1910 {
1911         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1912
1913         *db = (ar >> 14) & 1;
1914         *l = (ar >> 13) & 1;
1915 }
1916
1917 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1918 {
1919         dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1920         dt->base = vmcs_readl(GUEST_IDTR_BASE);
1921 }
1922
1923 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1924 {
1925         vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1926         vmcs_writel(GUEST_IDTR_BASE, dt->base);
1927 }
1928
1929 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1930 {
1931         dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1932         dt->base = vmcs_readl(GUEST_GDTR_BASE);
1933 }
1934
1935 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1936 {
1937         vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1938         vmcs_writel(GUEST_GDTR_BASE, dt->base);
1939 }
1940
1941 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1942 {
1943         struct kvm_segment var;
1944         u32 ar;
1945
1946         vmx_get_segment(vcpu, &var, seg);
1947         ar = vmx_segment_access_rights(&var);
1948
1949         if (var.base != (var.selector << 4))
1950                 return false;
1951         if (var.limit != 0xffff)
1952                 return false;
1953         if (ar != 0xf3)
1954                 return false;
1955
1956         return true;
1957 }
1958
1959 static bool code_segment_valid(struct kvm_vcpu *vcpu)
1960 {
1961         struct kvm_segment cs;
1962         unsigned int cs_rpl;
1963
1964         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1965         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1966
1967         if (cs.unusable)
1968                 return false;
1969         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1970                 return false;
1971         if (!cs.s)
1972                 return false;
1973         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
1974                 if (cs.dpl > cs_rpl)
1975                         return false;
1976         } else {
1977                 if (cs.dpl != cs_rpl)
1978                         return false;
1979         }
1980         if (!cs.present)
1981                 return false;
1982
1983         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1984         return true;
1985 }
1986
1987 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1988 {
1989         struct kvm_segment ss;
1990         unsigned int ss_rpl;
1991
1992         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1993         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1994
1995         if (ss.unusable)
1996                 return true;
1997         if (ss.type != 3 && ss.type != 7)
1998                 return false;
1999         if (!ss.s)
2000                 return false;
2001         if (ss.dpl != ss_rpl) /* DPL != RPL */
2002                 return false;
2003         if (!ss.present)
2004                 return false;
2005
2006         return true;
2007 }
2008
2009 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2010 {
2011         struct kvm_segment var;
2012         unsigned int rpl;
2013
2014         vmx_get_segment(vcpu, &var, seg);
2015         rpl = var.selector & SELECTOR_RPL_MASK;
2016
2017         if (var.unusable)
2018                 return true;
2019         if (!var.s)
2020                 return false;
2021         if (!var.present)
2022                 return false;
2023         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2024                 if (var.dpl < rpl) /* DPL < RPL */
2025                         return false;
2026         }
2027
2028         /* TODO: Add other members to kvm_segment_field to allow checking for other access
2029          * rights flags
2030          */
2031         return true;
2032 }
2033
2034 static bool tr_valid(struct kvm_vcpu *vcpu)
2035 {
2036         struct kvm_segment tr;
2037
2038         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2039
2040         if (tr.unusable)
2041                 return false;
2042         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
2043                 return false;
2044         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2045                 return false;
2046         if (!tr.present)
2047                 return false;
2048
2049         return true;
2050 }
2051
2052 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2053 {
2054         struct kvm_segment ldtr;
2055
2056         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2057
2058         if (ldtr.unusable)
2059                 return true;
2060         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
2061                 return false;
2062         if (ldtr.type != 2)
2063                 return false;
2064         if (!ldtr.present)
2065                 return false;
2066
2067         return true;
2068 }
2069
2070 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2071 {
2072         struct kvm_segment cs, ss;
2073
2074         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2075         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2076
2077         return ((cs.selector & SELECTOR_RPL_MASK) ==
2078                  (ss.selector & SELECTOR_RPL_MASK));
2079 }
2080
2081 /*
2082  * Check if guest state is valid. Returns true if valid, false if
2083  * not.
2084  * We assume that registers are always usable
2085  */
2086 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2087 {
2088         /* real mode guest state checks */
2089         if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
2090                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2091                         return false;
2092                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2093                         return false;
2094                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2095                         return false;
2096                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2097                         return false;
2098                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2099                         return false;
2100                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2101                         return false;
2102         } else {
2103         /* protected mode guest state checks */
2104                 if (!cs_ss_rpl_check(vcpu))
2105                         return false;
2106                 if (!code_segment_valid(vcpu))
2107                         return false;
2108                 if (!stack_segment_valid(vcpu))
2109                         return false;
2110                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2111                         return false;
2112                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2113                         return false;
2114                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2115                         return false;
2116                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2117                         return false;
2118                 if (!tr_valid(vcpu))
2119                         return false;
2120                 if (!ldtr_valid(vcpu))
2121                         return false;
2122         }
2123         /* TODO:
2124          * - Add checks on RIP
2125          * - Add checks on RFLAGS
2126          */
2127
2128         return true;
2129 }
2130
2131 static int init_rmode_tss(struct kvm *kvm)
2132 {
2133         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2134         u16 data = 0;
2135         int ret = 0;
2136         int r;
2137
2138         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2139         if (r < 0)
2140                 goto out;
2141         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2142         r = kvm_write_guest_page(kvm, fn++, &data,
2143                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
2144         if (r < 0)
2145                 goto out;
2146         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2147         if (r < 0)
2148                 goto out;
2149         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2150         if (r < 0)
2151                 goto out;
2152         data = ~0;
2153         r = kvm_write_guest_page(kvm, fn, &data,
2154                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2155                                  sizeof(u8));
2156         if (r < 0)
2157                 goto out;
2158
2159         ret = 1;
2160 out:
2161         return ret;
2162 }
2163
2164 static int init_rmode_identity_map(struct kvm *kvm)
2165 {
2166         int i, r, ret;
2167         pfn_t identity_map_pfn;
2168         u32 tmp;
2169
2170         if (!enable_ept)
2171                 return 1;
2172         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2173                 printk(KERN_ERR "EPT: identity-mapping pagetable "
2174                         "haven't been allocated!\n");
2175                 return 0;
2176         }
2177         if (likely(kvm->arch.ept_identity_pagetable_done))
2178                 return 1;
2179         ret = 0;
2180         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2181         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2182         if (r < 0)
2183                 goto out;
2184         /* Set up identity-mapping pagetable for EPT in real mode */
2185         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2186                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2187                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2188                 r = kvm_write_guest_page(kvm, identity_map_pfn,
2189                                 &tmp, i * sizeof(tmp), sizeof(tmp));
2190                 if (r < 0)
2191                         goto out;
2192         }
2193         kvm->arch.ept_identity_pagetable_done = true;
2194         ret = 1;
2195 out:
2196         return ret;
2197 }
2198
2199 static void seg_setup(int seg)
2200 {
2201         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2202         unsigned int ar;
2203
2204         vmcs_write16(sf->selector, 0);
2205         vmcs_writel(sf->base, 0);
2206         vmcs_write32(sf->limit, 0xffff);
2207         if (enable_unrestricted_guest) {
2208                 ar = 0x93;
2209                 if (seg == VCPU_SREG_CS)
2210                         ar |= 0x08; /* code segment */
2211         } else
2212                 ar = 0xf3;
2213
2214         vmcs_write32(sf->ar_bytes, ar);
2215 }
2216
2217 static int alloc_apic_access_page(struct kvm *kvm)
2218 {
2219         struct kvm_userspace_memory_region kvm_userspace_mem;
2220         int r = 0;
2221
2222         down_write(&kvm->slots_lock);
2223         if (kvm->arch.apic_access_page)
2224                 goto out;
2225         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2226         kvm_userspace_mem.flags = 0;
2227         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2228         kvm_userspace_mem.memory_size = PAGE_SIZE;
2229         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2230         if (r)
2231                 goto out;
2232
2233         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2234 out:
2235         up_write(&kvm->slots_lock);
2236         return r;
2237 }
2238
2239 static int alloc_identity_pagetable(struct kvm *kvm)
2240 {
2241         struct kvm_userspace_memory_region kvm_userspace_mem;
2242         int r = 0;
2243
2244         down_write(&kvm->slots_lock);
2245         if (kvm->arch.ept_identity_pagetable)
2246                 goto out;
2247         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2248         kvm_userspace_mem.flags = 0;
2249         kvm_userspace_mem.guest_phys_addr =
2250                 kvm->arch.ept_identity_map_addr;
2251         kvm_userspace_mem.memory_size = PAGE_SIZE;
2252         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2253         if (r)
2254                 goto out;
2255
2256         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2257                         kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2258 out:
2259         up_write(&kvm->slots_lock);
2260         return r;
2261 }
2262
2263 static void allocate_vpid(struct vcpu_vmx *vmx)
2264 {
2265         int vpid;
2266
2267         vmx->vpid = 0;
2268         if (!enable_vpid)
2269                 return;
2270         spin_lock(&vmx_vpid_lock);
2271         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2272         if (vpid < VMX_NR_VPIDS) {
2273                 vmx->vpid = vpid;
2274                 __set_bit(vpid, vmx_vpid_bitmap);
2275         }
2276         spin_unlock(&vmx_vpid_lock);
2277 }
2278
2279 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2280 {
2281         int f = sizeof(unsigned long);
2282
2283         if (!cpu_has_vmx_msr_bitmap())
2284                 return;
2285
2286         /*
2287          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2288          * have the write-low and read-high bitmap offsets the wrong way round.
2289          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2290          */
2291         if (msr <= 0x1fff) {
2292                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2293                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2294         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2295                 msr &= 0x1fff;
2296                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2297                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2298         }
2299 }
2300
2301 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2302 {
2303         if (!longmode_only)
2304                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2305         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2306 }
2307
2308 /*
2309  * Sets up the vmcs for emulated real mode.
2310  */
2311 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2312 {
2313         u32 host_sysenter_cs, msr_low, msr_high;
2314         u32 junk;
2315         u64 host_pat, tsc_this, tsc_base;
2316         unsigned long a;
2317         struct descriptor_table dt;
2318         int i;
2319         unsigned long kvm_vmx_return;
2320         u32 exec_control;
2321
2322         /* I/O */
2323         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2324         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2325
2326         if (cpu_has_vmx_msr_bitmap())
2327                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2328
2329         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2330
2331         /* Control */
2332         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2333                 vmcs_config.pin_based_exec_ctrl);
2334
2335         exec_control = vmcs_config.cpu_based_exec_ctrl;
2336         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2337                 exec_control &= ~CPU_BASED_TPR_SHADOW;
2338 #ifdef CONFIG_X86_64
2339                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2340                                 CPU_BASED_CR8_LOAD_EXITING;
2341 #endif
2342         }
2343         if (!enable_ept)
2344                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2345                                 CPU_BASED_CR3_LOAD_EXITING  |
2346                                 CPU_BASED_INVLPG_EXITING;
2347         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2348
2349         if (cpu_has_secondary_exec_ctrls()) {
2350                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2351                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2352                         exec_control &=
2353                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2354                 if (vmx->vpid == 0)
2355                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2356                 if (!enable_ept) {
2357                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2358                         enable_unrestricted_guest = 0;
2359                 }
2360                 if (!enable_unrestricted_guest)
2361                         exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2362                 if (!ple_gap)
2363                         exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2364                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2365         }
2366
2367         if (ple_gap) {
2368                 vmcs_write32(PLE_GAP, ple_gap);
2369                 vmcs_write32(PLE_WINDOW, ple_window);
2370         }
2371
2372         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2373         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2374         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
2375
2376         vmcs_writel(HOST_CR0, read_cr0());  /* 22.2.3 */
2377         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
2378         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
2379
2380         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
2381         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2382         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2383         vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs());    /* 22.2.4 */
2384         vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs());    /* 22.2.4 */
2385         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2386 #ifdef CONFIG_X86_64
2387         rdmsrl(MSR_FS_BASE, a);
2388         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2389         rdmsrl(MSR_GS_BASE, a);
2390         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2391 #else
2392         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2393         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2394 #endif
2395
2396         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
2397
2398         kvm_get_idt(&dt);
2399         vmcs_writel(HOST_IDTR_BASE, dt.base);   /* 22.2.4 */
2400
2401         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2402         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2403         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2404         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2405         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2406
2407         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2408         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2409         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2410         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
2411         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2412         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
2413
2414         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2415                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2416                 host_pat = msr_low | ((u64) msr_high << 32);
2417                 vmcs_write64(HOST_IA32_PAT, host_pat);
2418         }
2419         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2420                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2421                 host_pat = msr_low | ((u64) msr_high << 32);
2422                 /* Write the default value follow host pat */
2423                 vmcs_write64(GUEST_IA32_PAT, host_pat);
2424                 /* Keep arch.pat sync with GUEST_IA32_PAT */
2425                 vmx->vcpu.arch.pat = host_pat;
2426         }
2427
2428         for (i = 0; i < NR_VMX_MSR; ++i) {
2429                 u32 index = vmx_msr_index[i];
2430                 u32 data_low, data_high;
2431                 int j = vmx->nmsrs;
2432
2433                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2434                         continue;
2435                 if (wrmsr_safe(index, data_low, data_high) < 0)
2436                         continue;
2437                 vmx->guest_msrs[j].index = i;
2438                 vmx->guest_msrs[j].data = 0;
2439                 vmx->guest_msrs[j].mask = -1ull;
2440                 ++vmx->nmsrs;
2441         }
2442
2443         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2444
2445         /* 22.2.1, 20.8.1 */
2446         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2447
2448         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2449         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2450         if (enable_ept)
2451                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2452         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2453
2454         tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2455         rdtscll(tsc_this);
2456         if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2457                 tsc_base = tsc_this;
2458
2459         guest_write_tsc(0, tsc_base);
2460
2461         return 0;
2462 }
2463
2464 static int init_rmode(struct kvm *kvm)
2465 {
2466         if (!init_rmode_tss(kvm))
2467                 return 0;
2468         if (!init_rmode_identity_map(kvm))
2469                 return 0;
2470         return 1;
2471 }
2472
2473 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2474 {
2475         struct vcpu_vmx *vmx = to_vmx(vcpu);
2476         u64 msr;
2477         int ret;
2478
2479         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2480         down_read(&vcpu->kvm->slots_lock);
2481         if (!init_rmode(vmx->vcpu.kvm)) {
2482                 ret = -ENOMEM;
2483                 goto out;
2484         }
2485
2486         vmx->rmode.vm86_active = 0;
2487
2488         vmx->soft_vnmi_blocked = 0;
2489
2490         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2491         kvm_set_cr8(&vmx->vcpu, 0);
2492         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2493         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2494                 msr |= MSR_IA32_APICBASE_BSP;
2495         kvm_set_apic_base(&vmx->vcpu, msr);
2496
2497         fx_init(&vmx->vcpu);
2498
2499         seg_setup(VCPU_SREG_CS);
2500         /*
2501          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2502          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2503          */
2504         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2505                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2506                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2507         } else {
2508                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2509                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2510         }
2511
2512         seg_setup(VCPU_SREG_DS);
2513         seg_setup(VCPU_SREG_ES);
2514         seg_setup(VCPU_SREG_FS);
2515         seg_setup(VCPU_SREG_GS);
2516         seg_setup(VCPU_SREG_SS);
2517
2518         vmcs_write16(GUEST_TR_SELECTOR, 0);
2519         vmcs_writel(GUEST_TR_BASE, 0);
2520         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2521         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2522
2523         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2524         vmcs_writel(GUEST_LDTR_BASE, 0);
2525         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2526         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2527
2528         vmcs_write32(GUEST_SYSENTER_CS, 0);
2529         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2530         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2531
2532         vmcs_writel(GUEST_RFLAGS, 0x02);
2533         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2534                 kvm_rip_write(vcpu, 0xfff0);
2535         else
2536                 kvm_rip_write(vcpu, 0);
2537         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2538
2539         vmcs_writel(GUEST_DR7, 0x400);
2540
2541         vmcs_writel(GUEST_GDTR_BASE, 0);
2542         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2543
2544         vmcs_writel(GUEST_IDTR_BASE, 0);
2545         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2546
2547         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2548         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2549         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2550
2551         /* Special registers */
2552         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2553
2554         setup_msrs(vmx);
2555
2556         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2557
2558         if (cpu_has_vmx_tpr_shadow()) {
2559                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2560                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2561                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2562                                 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2563                 vmcs_write32(TPR_THRESHOLD, 0);
2564         }
2565
2566         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2567                 vmcs_write64(APIC_ACCESS_ADDR,
2568                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2569
2570         if (vmx->vpid != 0)
2571                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2572
2573         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2574         vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2575         vmx_set_cr4(&vmx->vcpu, 0);
2576         vmx_set_efer(&vmx->vcpu, 0);
2577         vmx_fpu_activate(&vmx->vcpu);
2578         update_exception_bitmap(&vmx->vcpu);
2579
2580         vpid_sync_vcpu_all(vmx);
2581
2582         ret = 0;
2583
2584         /* HACK: Don't enable emulation on guest boot/reset */
2585         vmx->emulation_required = 0;
2586
2587 out:
2588         up_read(&vcpu->kvm->slots_lock);
2589         return ret;
2590 }
2591
2592 static void enable_irq_window(struct kvm_vcpu *vcpu)
2593 {
2594         u32 cpu_based_vm_exec_control;
2595
2596         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2597         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2598         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2599 }
2600
2601 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2602 {
2603         u32 cpu_based_vm_exec_control;
2604
2605         if (!cpu_has_virtual_nmis()) {
2606                 enable_irq_window(vcpu);
2607                 return;
2608         }
2609
2610         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2611         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2612         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2613 }
2614
2615 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2616 {
2617         struct vcpu_vmx *vmx = to_vmx(vcpu);
2618         uint32_t intr;
2619         int irq = vcpu->arch.interrupt.nr;
2620
2621         trace_kvm_inj_virq(irq);
2622
2623         ++vcpu->stat.irq_injections;
2624         if (vmx->rmode.vm86_active) {
2625                 vmx->rmode.irq.pending = true;
2626                 vmx->rmode.irq.vector = irq;
2627                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2628                 if (vcpu->arch.interrupt.soft)
2629                         vmx->rmode.irq.rip +=
2630                                 vmx->vcpu.arch.event_exit_inst_len;
2631                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2632                              irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2633                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2634                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2635                 return;
2636         }
2637         intr = irq | INTR_INFO_VALID_MASK;
2638         if (vcpu->arch.interrupt.soft) {
2639                 intr |= INTR_TYPE_SOFT_INTR;
2640                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2641                              vmx->vcpu.arch.event_exit_inst_len);
2642         } else
2643                 intr |= INTR_TYPE_EXT_INTR;
2644         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2645 }
2646
2647 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2648 {
2649         struct vcpu_vmx *vmx = to_vmx(vcpu);
2650
2651         if (!cpu_has_virtual_nmis()) {
2652                 /*
2653                  * Tracking the NMI-blocked state in software is built upon
2654                  * finding the next open IRQ window. This, in turn, depends on
2655                  * well-behaving guests: They have to keep IRQs disabled at
2656                  * least as long as the NMI handler runs. Otherwise we may
2657                  * cause NMI nesting, maybe breaking the guest. But as this is
2658                  * highly unlikely, we can live with the residual risk.
2659                  */
2660                 vmx->soft_vnmi_blocked = 1;
2661                 vmx->vnmi_blocked_time = 0;
2662         }
2663
2664         ++vcpu->stat.nmi_injections;
2665         if (vmx->rmode.vm86_active) {
2666                 vmx->rmode.irq.pending = true;
2667                 vmx->rmode.irq.vector = NMI_VECTOR;
2668                 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2669                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2670                              NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2671                              INTR_INFO_VALID_MASK);
2672                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2673                 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2674                 return;
2675         }
2676         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2677                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2678 }
2679
2680 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2681 {
2682         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2683                 return 0;
2684
2685         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2686                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS |
2687                                 GUEST_INTR_STATE_NMI));
2688 }
2689
2690 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2691 {
2692         if (!cpu_has_virtual_nmis())
2693                 return to_vmx(vcpu)->soft_vnmi_blocked;
2694         else
2695                 return !!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2696                           GUEST_INTR_STATE_NMI);
2697 }
2698
2699 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2700 {
2701         struct vcpu_vmx *vmx = to_vmx(vcpu);
2702
2703         if (!cpu_has_virtual_nmis()) {
2704                 if (vmx->soft_vnmi_blocked != masked) {
2705                         vmx->soft_vnmi_blocked = masked;
2706                         vmx->vnmi_blocked_time = 0;
2707                 }
2708         } else {
2709                 if (masked)
2710                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2711                                       GUEST_INTR_STATE_NMI);
2712                 else
2713                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2714                                         GUEST_INTR_STATE_NMI);
2715         }
2716 }
2717
2718 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2719 {
2720         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2721                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2722                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2723 }
2724
2725 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2726 {
2727         int ret;
2728         struct kvm_userspace_memory_region tss_mem = {
2729                 .slot = TSS_PRIVATE_MEMSLOT,
2730                 .guest_phys_addr = addr,
2731                 .memory_size = PAGE_SIZE * 3,
2732                 .flags = 0,
2733         };
2734
2735         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2736         if (ret)
2737                 return ret;
2738         kvm->arch.tss_addr = addr;
2739         return 0;
2740 }
2741
2742 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2743                                   int vec, u32 err_code)
2744 {
2745         /*
2746          * Instruction with address size override prefix opcode 0x67
2747          * Cause the #SS fault with 0 error code in VM86 mode.
2748          */
2749         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2750                 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
2751                         return 1;
2752         /*
2753          * Forward all other exceptions that are valid in real mode.
2754          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2755          *        the required debugging infrastructure rework.
2756          */
2757         switch (vec) {
2758         case DB_VECTOR:
2759                 if (vcpu->guest_debug &
2760                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2761                         return 0;
2762                 kvm_queue_exception(vcpu, vec);
2763                 return 1;
2764         case BP_VECTOR:
2765                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2766                         return 0;
2767                 /* fall through */
2768         case DE_VECTOR:
2769         case OF_VECTOR:
2770         case BR_VECTOR:
2771         case UD_VECTOR:
2772         case DF_VECTOR:
2773         case SS_VECTOR:
2774         case GP_VECTOR:
2775         case MF_VECTOR:
2776                 kvm_queue_exception(vcpu, vec);
2777                 return 1;
2778         }
2779         return 0;
2780 }
2781
2782 /*
2783  * Trigger machine check on the host. We assume all the MSRs are already set up
2784  * by the CPU and that we still run on the same CPU as the MCE occurred on.
2785  * We pass a fake environment to the machine check handler because we want
2786  * the guest to be always treated like user space, no matter what context
2787  * it used internally.
2788  */
2789 static void kvm_machine_check(void)
2790 {
2791 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2792         struct pt_regs regs = {
2793                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2794                 .flags = X86_EFLAGS_IF,
2795         };
2796
2797         do_machine_check(&regs, 0);
2798 #endif
2799 }
2800
2801 static int handle_machine_check(struct kvm_vcpu *vcpu)
2802 {
2803         /* already handled by vcpu_run */
2804         return 1;
2805 }
2806
2807 static int handle_exception(struct kvm_vcpu *vcpu)
2808 {
2809         struct vcpu_vmx *vmx = to_vmx(vcpu);
2810         struct kvm_run *kvm_run = vcpu->run;
2811         u32 intr_info, ex_no, error_code;
2812         unsigned long cr2, rip, dr6;
2813         u32 vect_info;
2814         enum emulation_result er;
2815
2816         vect_info = vmx->idt_vectoring_info;
2817         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2818
2819         if (is_machine_check(intr_info))
2820                 return handle_machine_check(vcpu);
2821
2822         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2823             !is_page_fault(intr_info)) {
2824                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2825                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
2826                 vcpu->run->internal.ndata = 2;
2827                 vcpu->run->internal.data[0] = vect_info;
2828                 vcpu->run->internal.data[1] = intr_info;
2829                 return 0;
2830         }
2831
2832         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2833                 return 1;  /* already handled by vmx_vcpu_run() */
2834
2835         if (is_no_device(intr_info)) {
2836                 vmx_fpu_activate(vcpu);
2837                 return 1;
2838         }
2839
2840         if (is_invalid_opcode(intr_info)) {
2841                 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
2842                 if (er != EMULATE_DONE)
2843                         kvm_queue_exception(vcpu, UD_VECTOR);
2844                 return 1;
2845         }
2846
2847         error_code = 0;
2848         rip = kvm_rip_read(vcpu);
2849         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2850                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2851         if (is_page_fault(intr_info)) {
2852                 /* EPT won't cause page fault directly */
2853                 if (enable_ept)
2854                         BUG();
2855                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2856                 trace_kvm_page_fault(cr2, error_code);
2857
2858                 if (kvm_event_needs_reinjection(vcpu))
2859                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
2860                 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2861         }
2862
2863         if (vmx->rmode.vm86_active &&
2864             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2865                                                                 error_code)) {
2866                 if (vcpu->arch.halt_request) {
2867                         vcpu->arch.halt_request = 0;
2868                         return kvm_emulate_halt(vcpu);
2869                 }
2870                 return 1;
2871         }
2872
2873         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
2874         switch (ex_no) {
2875         case DB_VECTOR:
2876                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2877                 if (!(vcpu->guest_debug &
2878                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2879                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2880                         kvm_queue_exception(vcpu, DB_VECTOR);
2881                         return 1;
2882                 }
2883                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2884                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2885                 /* fall through */
2886         case BP_VECTOR:
2887                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2888                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2889                 kvm_run->debug.arch.exception = ex_no;
2890                 break;
2891         default:
2892                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2893                 kvm_run->ex.exception = ex_no;
2894                 kvm_run->ex.error_code = error_code;
2895                 break;
2896         }
2897         return 0;
2898 }
2899
2900 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
2901 {
2902         ++vcpu->stat.irq_exits;
2903         return 1;
2904 }
2905
2906 static int handle_triple_fault(struct kvm_vcpu *vcpu)
2907 {
2908         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
2909         return 0;
2910 }
2911
2912 static int handle_io(struct kvm_vcpu *vcpu)
2913 {
2914         unsigned long exit_qualification;
2915         int size, in, string;
2916         unsigned port;
2917
2918         ++vcpu->stat.io_exits;
2919         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2920         string = (exit_qualification & 16) != 0;
2921
2922         if (string) {
2923                 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DO_MMIO)
2924                         return 0;
2925                 return 1;
2926         }
2927
2928         size = (exit_qualification & 7) + 1;
2929         in = (exit_qualification & 8) != 0;
2930         port = exit_qualification >> 16;
2931
2932         skip_emulated_instruction(vcpu);
2933         return kvm_emulate_pio(vcpu, in, size, port);
2934 }
2935
2936 static void
2937 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2938 {
2939         /*
2940          * Patch in the VMCALL instruction:
2941          */
2942         hypercall[0] = 0x0f;
2943         hypercall[1] = 0x01;
2944         hypercall[2] = 0xc1;
2945 }
2946
2947 static int handle_cr(struct kvm_vcpu *vcpu)
2948 {
2949         unsigned long exit_qualification, val;
2950         int cr;
2951         int reg;
2952
2953         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2954         cr = exit_qualification & 15;
2955         reg = (exit_qualification >> 8) & 15;
2956         switch ((exit_qualification >> 4) & 3) {
2957         case 0: /* mov to cr */
2958                 val = kvm_register_read(vcpu, reg);
2959                 trace_kvm_cr_write(cr, val);
2960                 switch (cr) {
2961                 case 0:
2962                         kvm_set_cr0(vcpu, val);
2963                         skip_emulated_instruction(vcpu);
2964                         return 1;
2965                 case 3:
2966                         kvm_set_cr3(vcpu, val);
2967                         skip_emulated_instruction(vcpu);
2968                         return 1;
2969                 case 4:
2970                         kvm_set_cr4(vcpu, val);
2971                         skip_emulated_instruction(vcpu);
2972                         return 1;
2973                 case 8: {
2974                                 u8 cr8_prev = kvm_get_cr8(vcpu);
2975                                 u8 cr8 = kvm_register_read(vcpu, reg);
2976                                 kvm_set_cr8(vcpu, cr8);
2977                                 skip_emulated_instruction(vcpu);
2978                                 if (irqchip_in_kernel(vcpu->kvm))
2979                                         return 1;
2980                                 if (cr8_prev <= cr8)
2981                                         return 1;
2982                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
2983                                 return 0;
2984                         }
2985                 };
2986                 break;
2987         case 2: /* clts */
2988                 vmx_fpu_deactivate(vcpu);
2989                 vcpu->arch.cr0 &= ~X86_CR0_TS;
2990                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2991                 vmx_fpu_activate(vcpu);
2992                 skip_emulated_instruction(vcpu);
2993                 return 1;
2994         case 1: /*mov from cr*/
2995                 switch (cr) {
2996                 case 3:
2997                         kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2998                         trace_kvm_cr_read(cr, vcpu->arch.cr3);
2999                         skip_emulated_instruction(vcpu);
3000                         return 1;
3001                 case 8:
3002                         val = kvm_get_cr8(vcpu);
3003                         kvm_register_write(vcpu, reg, val);
3004                         trace_kvm_cr_read(cr, val);
3005                         skip_emulated_instruction(vcpu);
3006                         return 1;
3007                 }
3008                 break;
3009         case 3: /* lmsw */
3010                 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
3011
3012                 skip_emulated_instruction(vcpu);
3013                 return 1;
3014         default:
3015                 break;
3016         }
3017         vcpu->run->exit_reason = 0;
3018         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3019                (int)(exit_qualification >> 4) & 3, cr);
3020         return 0;
3021 }
3022
3023 static int handle_dr(struct kvm_vcpu *vcpu)
3024 {
3025         unsigned long exit_qualification;
3026         unsigned long val;
3027         int dr, reg;
3028
3029         if (!kvm_require_cpl(vcpu, 0))
3030                 return 1;
3031         dr = vmcs_readl(GUEST_DR7);
3032         if (dr & DR7_GD) {
3033                 /*
3034                  * As the vm-exit takes precedence over the debug trap, we
3035                  * need to emulate the latter, either for the host or the
3036                  * guest debugging itself.
3037                  */
3038                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3039                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3040                         vcpu->run->debug.arch.dr7 = dr;
3041                         vcpu->run->debug.arch.pc =
3042                                 vmcs_readl(GUEST_CS_BASE) +
3043                                 vmcs_readl(GUEST_RIP);
3044                         vcpu->run->debug.arch.exception = DB_VECTOR;
3045                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3046                         return 0;
3047                 } else {
3048                         vcpu->arch.dr7 &= ~DR7_GD;
3049                         vcpu->arch.dr6 |= DR6_BD;
3050                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3051                         kvm_queue_exception(vcpu, DB_VECTOR);
3052                         return 1;
3053                 }
3054         }
3055
3056         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3057         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3058         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3059         if (exit_qualification & TYPE_MOV_FROM_DR) {
3060                 switch (dr) {
3061                 case 0 ... 3:
3062                         val = vcpu->arch.db[dr];
3063                         break;
3064                 case 6:
3065                         val = vcpu->arch.dr6;
3066                         break;
3067                 case 7:
3068                         val = vcpu->arch.dr7;
3069                         break;
3070                 default:
3071                         val = 0;
3072                 }
3073                 kvm_register_write(vcpu, reg, val);
3074         } else {
3075                 val = vcpu->arch.regs[reg];
3076                 switch (dr) {
3077                 case 0 ... 3:
3078                         vcpu->arch.db[dr] = val;
3079                         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
3080                                 vcpu->arch.eff_db[dr] = val;
3081                         break;
3082                 case 4 ... 5:
3083                         if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
3084                                 kvm_queue_exception(vcpu, UD_VECTOR);
3085                         break;
3086                 case 6:
3087                         if (val & 0xffffffff00000000ULL) {
3088                                 kvm_queue_exception(vcpu, GP_VECTOR);
3089                                 break;
3090                         }
3091                         vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
3092                         break;
3093                 case 7:
3094                         if (val & 0xffffffff00000000ULL) {
3095                                 kvm_queue_exception(vcpu, GP_VECTOR);
3096                                 break;
3097                         }
3098                         vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
3099                         if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
3100                                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3101                                 vcpu->arch.switch_db_regs =
3102                                         (val & DR7_BP_EN_MASK);
3103                         }
3104                         break;
3105                 }
3106         }
3107         skip_emulated_instruction(vcpu);
3108         return 1;
3109 }
3110
3111 static int handle_cpuid(struct kvm_vcpu *vcpu)
3112 {
3113         kvm_emulate_cpuid(vcpu);
3114         return 1;
3115 }
3116
3117 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3118 {
3119         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3120         u64 data;
3121
3122         if (vmx_get_msr(vcpu, ecx, &data)) {
3123                 kvm_inject_gp(vcpu, 0);
3124                 return 1;
3125         }
3126
3127         trace_kvm_msr_read(ecx, data);
3128
3129         /* FIXME: handling of bits 32:63 of rax, rdx */
3130         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3131         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3132         skip_emulated_instruction(vcpu);
3133         return 1;
3134 }
3135
3136 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3137 {
3138         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3139         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3140                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3141
3142         trace_kvm_msr_write(ecx, data);
3143
3144         if (vmx_set_msr(vcpu, ecx, data) != 0) {
3145                 kvm_inject_gp(vcpu, 0);
3146                 return 1;
3147         }
3148
3149         skip_emulated_instruction(vcpu);
3150         return 1;
3151 }
3152
3153 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3154 {
3155         return 1;
3156 }
3157
3158 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3159 {
3160         u32 cpu_based_vm_exec_control;
3161
3162         /* clear pending irq */
3163         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3164         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3165         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3166
3167         ++vcpu->stat.irq_window_exits;
3168
3169         /*
3170          * If the user space waits to inject interrupts, exit as soon as
3171          * possible
3172          */
3173         if (!irqchip_in_kernel(vcpu->kvm) &&
3174             vcpu->run->request_interrupt_window &&
3175             !kvm_cpu_has_interrupt(vcpu)) {
3176                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3177                 return 0;
3178         }
3179         return 1;
3180 }
3181
3182 static int handle_halt(struct kvm_vcpu *vcpu)
3183 {
3184         skip_emulated_instruction(vcpu);
3185         return kvm_emulate_halt(vcpu);
3186 }
3187
3188 static int handle_vmcall(struct kvm_vcpu *vcpu)
3189 {
3190         skip_emulated_instruction(vcpu);
3191         kvm_emulate_hypercall(vcpu);
3192         return 1;
3193 }
3194
3195 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3196 {
3197         kvm_queue_exception(vcpu, UD_VECTOR);
3198         return 1;
3199 }
3200
3201 static int handle_invlpg(struct kvm_vcpu *vcpu)
3202 {
3203         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3204
3205         kvm_mmu_invlpg(vcpu, exit_qualification);
3206         skip_emulated_instruction(vcpu);
3207         return 1;
3208 }
3209
3210 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3211 {
3212         skip_emulated_instruction(vcpu);
3213         /* TODO: Add support for VT-d/pass-through device */
3214         return 1;
3215 }
3216
3217 static int handle_apic_access(struct kvm_vcpu *vcpu)
3218 {
3219         unsigned long exit_qualification;
3220         enum emulation_result er;
3221         unsigned long offset;
3222
3223         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3224         offset = exit_qualification & 0xffful;
3225
3226         er = emulate_instruction(vcpu, 0, 0, 0);
3227
3228         if (er !=  EMULATE_DONE) {
3229                 printk(KERN_ERR
3230                        "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3231                        offset);
3232                 return -ENOEXEC;
3233         }
3234         return 1;
3235 }
3236
3237 static int handle_task_switch(struct kvm_vcpu *vcpu)
3238 {
3239         struct vcpu_vmx *vmx = to_vmx(vcpu);
3240         unsigned long exit_qualification;
3241         u16 tss_selector;
3242         int reason, type, idt_v;
3243
3244         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3245         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3246
3247         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3248
3249         reason = (u32)exit_qualification >> 30;
3250         if (reason == TASK_SWITCH_GATE && idt_v) {
3251                 switch (type) {
3252                 case INTR_TYPE_NMI_INTR:
3253                         vcpu->arch.nmi_injected = false;
3254                         if (cpu_has_virtual_nmis())
3255                                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3256                                               GUEST_INTR_STATE_NMI);
3257                         break;
3258                 case INTR_TYPE_EXT_INTR:
3259                 case INTR_TYPE_SOFT_INTR:
3260                         kvm_clear_interrupt_queue(vcpu);
3261                         break;
3262                 case INTR_TYPE_HARD_EXCEPTION:
3263                 case INTR_TYPE_SOFT_EXCEPTION:
3264                         kvm_clear_exception_queue(vcpu);
3265                         break;
3266                 default:
3267                         break;
3268                 }
3269         }
3270         tss_selector = exit_qualification;
3271
3272         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3273                        type != INTR_TYPE_EXT_INTR &&
3274                        type != INTR_TYPE_NMI_INTR))
3275                 skip_emulated_instruction(vcpu);
3276
3277         if (!kvm_task_switch(vcpu, tss_selector, reason))
3278                 return 0;
3279
3280         /* clear all local breakpoint enable flags */
3281         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3282
3283         /*
3284          * TODO: What about debug traps on tss switch?
3285          *       Are we supposed to inject them and update dr6?
3286          */
3287
3288         return 1;
3289 }
3290
3291 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3292 {
3293         unsigned long exit_qualification;
3294         gpa_t gpa;
3295         int gla_validity;
3296
3297         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3298
3299         if (exit_qualification & (1 << 6)) {
3300                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3301                 return -EINVAL;
3302         }
3303
3304         gla_validity = (exit_qualification >> 7) & 0x3;
3305         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3306                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3307                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3308                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3309                         vmcs_readl(GUEST_LINEAR_ADDRESS));
3310                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3311                         (long unsigned int)exit_qualification);
3312                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3313                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3314                 return 0;
3315         }
3316
3317         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3318         trace_kvm_page_fault(gpa, exit_qualification);
3319         return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3320 }
3321
3322 static u64 ept_rsvd_mask(u64 spte, int level)
3323 {
3324         int i;
3325         u64 mask = 0;
3326
3327         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3328                 mask |= (1ULL << i);
3329
3330         if (level > 2)
3331                 /* bits 7:3 reserved */
3332                 mask |= 0xf8;
3333         else if (level == 2) {
3334                 if (spte & (1ULL << 7))
3335                         /* 2MB ref, bits 20:12 reserved */
3336                         mask |= 0x1ff000;
3337                 else
3338                         /* bits 6:3 reserved */
3339                         mask |= 0x78;
3340         }
3341
3342         return mask;
3343 }
3344
3345 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3346                                        int level)
3347 {
3348         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3349
3350         /* 010b (write-only) */
3351         WARN_ON((spte & 0x7) == 0x2);
3352
3353         /* 110b (write/execute) */
3354         WARN_ON((spte & 0x7) == 0x6);
3355
3356         /* 100b (execute-only) and value not supported by logical processor */
3357         if (!cpu_has_vmx_ept_execute_only())
3358                 WARN_ON((spte & 0x7) == 0x4);
3359
3360         /* not 000b */
3361         if ((spte & 0x7)) {
3362                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3363
3364                 if (rsvd_bits != 0) {
3365                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3366                                          __func__, rsvd_bits);
3367                         WARN_ON(1);
3368                 }
3369
3370                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3371                         u64 ept_mem_type = (spte & 0x38) >> 3;
3372
3373                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
3374                             ept_mem_type == 7) {
3375                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3376                                                 __func__, ept_mem_type);
3377                                 WARN_ON(1);
3378                         }
3379                 }
3380         }
3381 }
3382
3383 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3384 {
3385         u64 sptes[4];
3386         int nr_sptes, i;
3387         gpa_t gpa;
3388
3389         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3390
3391         printk(KERN_ERR "EPT: Misconfiguration.\n");
3392         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3393
3394         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3395
3396         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3397                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3398
3399         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3400         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3401
3402         return 0;
3403 }
3404
3405 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3406 {
3407         u32 cpu_based_vm_exec_control;
3408
3409         /* clear pending NMI */
3410         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3411         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3412         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3413         ++vcpu->stat.nmi_window_exits;
3414
3415         return 1;
3416 }
3417
3418 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3419 {
3420         struct vcpu_vmx *vmx = to_vmx(vcpu);
3421         enum emulation_result err = EMULATE_DONE;
3422         int ret = 1;
3423
3424         while (!guest_state_valid(vcpu)) {
3425                 err = emulate_instruction(vcpu, 0, 0, 0);
3426
3427                 if (err == EMULATE_DO_MMIO) {
3428                         ret = 0;
3429                         goto out;
3430                 }
3431
3432                 if (err != EMULATE_DONE) {
3433                         kvm_report_emulation_failure(vcpu, "emulation failure");
3434                         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3435                         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3436                         vcpu->run->internal.ndata = 0;
3437                         ret = 0;
3438                         goto out;
3439                 }
3440
3441                 if (signal_pending(current))
3442                         goto out;
3443                 if (need_resched())
3444                         schedule();
3445         }
3446
3447         vmx->emulation_required = 0;
3448 out:
3449         return ret;
3450 }
3451
3452 /*
3453  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3454  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3455  */
3456 static int handle_pause(struct kvm_vcpu *vcpu)
3457 {
3458         skip_emulated_instruction(vcpu);
3459         kvm_vcpu_on_spin(vcpu);
3460
3461         return 1;
3462 }
3463
3464 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3465 {
3466         kvm_queue_exception(vcpu, UD_VECTOR);
3467         return 1;
3468 }
3469
3470 /*
3471  * The exit handlers return 1 if the exit was handled fully and guest execution
3472  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
3473  * to be done to userspace and return 0.
3474  */
3475 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3476         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
3477         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
3478         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
3479         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
3480         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
3481         [EXIT_REASON_CR_ACCESS]               = handle_cr,
3482         [EXIT_REASON_DR_ACCESS]               = handle_dr,
3483         [EXIT_REASON_CPUID]                   = handle_cpuid,
3484         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
3485         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
3486         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
3487         [EXIT_REASON_HLT]                     = handle_halt,
3488         [EXIT_REASON_INVLPG]                  = handle_invlpg,
3489         [EXIT_REASON_VMCALL]                  = handle_vmcall,
3490         [EXIT_REASON_VMCLEAR]                 = handle_vmx_insn,
3491         [EXIT_REASON_VMLAUNCH]                = handle_vmx_insn,
3492         [EXIT_REASON_VMPTRLD]                 = handle_vmx_insn,
3493         [EXIT_REASON_VMPTRST]                 = handle_vmx_insn,
3494         [EXIT_REASON_VMREAD]                  = handle_vmx_insn,
3495         [EXIT_REASON_VMRESUME]                = handle_vmx_insn,
3496         [EXIT_REASON_VMWRITE]                 = handle_vmx_insn,
3497         [EXIT_REASON_VMOFF]                   = handle_vmx_insn,
3498         [EXIT_REASON_VMON]                    = handle_vmx_insn,
3499         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
3500         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
3501         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
3502         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
3503         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
3504         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
3505         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
3506         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
3507         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
3508         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
3509 };
3510
3511 static const int kvm_vmx_max_exit_handlers =
3512         ARRAY_SIZE(kvm_vmx_exit_handlers);
3513
3514 /*
3515  * The guest has exited.  See if we can fix it or if we need userspace
3516  * assistance.
3517  */
3518 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3519 {
3520         struct vcpu_vmx *vmx = to_vmx(vcpu);
3521         u32 exit_reason = vmx->exit_reason;
3522         u32 vectoring_info = vmx->idt_vectoring_info;
3523
3524         trace_kvm_exit(exit_reason, kvm_rip_read(vcpu));
3525
3526         /* If guest state is invalid, start emulating */
3527         if (vmx->emulation_required && emulate_invalid_guest_state)
3528                 return handle_invalid_guest_state(vcpu);
3529
3530         /* Access CR3 don't cause VMExit in paging mode, so we need
3531          * to sync with guest real CR3. */
3532         if (enable_ept && is_paging(vcpu))
3533                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3534
3535         if (unlikely(vmx->fail)) {
3536                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3537                 vcpu->run->fail_entry.hardware_entry_failure_reason
3538                         = vmcs_read32(VM_INSTRUCTION_ERROR);
3539                 return 0;
3540         }
3541
3542         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3543                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3544                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
3545                         exit_reason != EXIT_REASON_TASK_SWITCH))
3546                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3547                        "(0x%x) and exit reason is 0x%x\n",
3548                        __func__, vectoring_info, exit_reason);
3549
3550         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3551                 if (vmx_interrupt_allowed(vcpu)) {
3552                         vmx->soft_vnmi_blocked = 0;
3553                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3554                            vcpu->arch.nmi_pending) {
3555                         /*
3556                          * This CPU don't support us in finding the end of an
3557                          * NMI-blocked window if the guest runs with IRQs
3558                          * disabled. So we pull the trigger after 1 s of
3559                          * futile waiting, but inform the user about this.
3560                          */
3561                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3562                                "state on VCPU %d after 1 s timeout\n",
3563                                __func__, vcpu->vcpu_id);
3564                         vmx->soft_vnmi_blocked = 0;
3565                 }
3566         }
3567
3568         if (exit_reason < kvm_vmx_max_exit_handlers
3569             && kvm_vmx_exit_handlers[exit_reason])
3570                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3571         else {
3572                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3573                 vcpu->run->hw.hardware_exit_reason = exit_reason;
3574         }
3575         return 0;
3576 }
3577
3578 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3579 {
3580         if (irr == -1 || tpr < irr) {
3581                 vmcs_write32(TPR_THRESHOLD, 0);
3582                 return;
3583         }
3584
3585         vmcs_write32(TPR_THRESHOLD, irr);
3586 }
3587
3588 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3589 {
3590         u32 exit_intr_info;
3591         u32 idt_vectoring_info = vmx->idt_vectoring_info;
3592         bool unblock_nmi;
3593         u8 vector;
3594         int type;
3595         bool idtv_info_valid;
3596
3597         exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3598
3599         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3600
3601         /* Handle machine checks before interrupts are enabled */
3602         if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3603             || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3604                 && is_machine_check(exit_intr_info)))
3605                 kvm_machine_check();
3606
3607         /* We need to handle NMIs before interrupts are enabled */
3608         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3609             (exit_intr_info & INTR_INFO_VALID_MASK))
3610                 asm("int $2");
3611
3612         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3613
3614         if (cpu_has_virtual_nmis()) {
3615                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3616                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3617                 /*
3618                  * SDM 3: 27.7.1.2 (September 2008)
3619                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
3620                  * a guest IRET fault.
3621                  * SDM 3: 23.2.2 (September 2008)
3622                  * Bit 12 is undefined in any of the following cases:
3623                  *  If the VM exit sets the valid bit in the IDT-vectoring
3624                  *   information field.
3625                  *  If the VM exit is due to a double fault.
3626                  */
3627                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3628                     vector != DF_VECTOR && !idtv_info_valid)
3629                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3630                                       GUEST_INTR_STATE_NMI);
3631         } else if (unlikely(vmx->soft_vnmi_blocked))
3632                 vmx->vnmi_blocked_time +=
3633                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3634
3635         vmx->vcpu.arch.nmi_injected = false;
3636         kvm_clear_exception_queue(&vmx->vcpu);
3637         kvm_clear_interrupt_queue(&vmx->vcpu);
3638
3639         if (!idtv_info_valid)
3640                 return;
3641
3642         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3643         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3644
3645         switch (type) {
3646         case INTR_TYPE_NMI_INTR:
3647                 vmx->vcpu.arch.nmi_injected = true;
3648                 /*
3649                  * SDM 3: 27.7.1.2 (September 2008)
3650                  * Clear bit "block by NMI" before VM entry if a NMI
3651                  * delivery faulted.
3652                  */
3653                 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3654                                 GUEST_INTR_STATE_NMI);
3655                 break;
3656         case INTR_TYPE_SOFT_EXCEPTION:
3657                 vmx->vcpu.arch.event_exit_inst_len =
3658                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3659                 /* fall through */
3660         case INTR_TYPE_HARD_EXCEPTION:
3661                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3662                         u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3663                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
3664                 } else
3665                         kvm_queue_exception(&vmx->vcpu, vector);
3666                 break;
3667         case INTR_TYPE_SOFT_INTR:
3668                 vmx->vcpu.arch.event_exit_inst_len =
3669                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3670                 /* fall through */
3671         case INTR_TYPE_EXT_INTR:
3672                 kvm_queue_interrupt(&vmx->vcpu, vector,
3673                         type == INTR_TYPE_SOFT_INTR);
3674                 break;
3675         default:
3676                 break;
3677         }
3678 }
3679
3680 /*
3681  * Failure to inject an interrupt should give us the information
3682  * in IDT_VECTORING_INFO_FIELD.  However, if the failure occurs
3683  * when fetching the interrupt redirection bitmap in the real-mode
3684  * tss, this doesn't happen.  So we do it ourselves.
3685  */
3686 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3687 {
3688         vmx->rmode.irq.pending = 0;
3689         if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3690                 return;
3691         kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3692         if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3693                 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3694                 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3695                 return;
3696         }
3697         vmx->idt_vectoring_info =
3698                 VECTORING_INFO_VALID_MASK
3699                 | INTR_TYPE_EXT_INTR
3700                 | vmx->rmode.irq.vector;
3701 }
3702
3703 #ifdef CONFIG_X86_64
3704 #define R "r"
3705 #define Q "q"
3706 #else
3707 #define R "e"
3708 #define Q "l"
3709 #endif
3710
3711 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3712 {
3713         struct vcpu_vmx *vmx = to_vmx(vcpu);
3714
3715         /* Record the guest's net vcpu time for enforced NMI injections. */
3716         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3717                 vmx->entry_time = ktime_get();
3718
3719         /* Don't enter VMX if guest state is invalid, let the exit handler
3720            start emulation until we arrive back to a valid state */
3721         if (vmx->emulation_required && emulate_invalid_guest_state)
3722                 return;
3723
3724         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3725                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3726         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3727                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3728
3729         /* When single-stepping over STI and MOV SS, we must clear the
3730          * corresponding interruptibility bits in the guest state. Otherwise
3731          * vmentry fails as it then expects bit 14 (BS) in pending debug
3732          * exceptions being set, but that's not correct for the guest debugging
3733          * case. */
3734         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3735                 vmx_set_interrupt_shadow(vcpu, 0);
3736
3737         /*
3738          * Loading guest fpu may have cleared host cr0.ts
3739          */
3740         vmcs_writel(HOST_CR0, read_cr0());
3741
3742         if (vcpu->arch.switch_db_regs)
3743                 set_debugreg(vcpu->arch.dr6, 6);
3744
3745         asm(
3746                 /* Store host registers */
3747                 "push %%"R"dx; push %%"R"bp;"
3748                 "push %%"R"cx \n\t"
3749                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3750                 "je 1f \n\t"
3751                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3752                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3753                 "1: \n\t"
3754                 /* Reload cr2 if changed */
3755                 "mov %c[cr2](%0), %%"R"ax \n\t"
3756                 "mov %%cr2, %%"R"dx \n\t"
3757                 "cmp %%"R"ax, %%"R"dx \n\t"
3758                 "je 2f \n\t"
3759                 "mov %%"R"ax, %%cr2 \n\t"
3760                 "2: \n\t"
3761                 /* Check if vmlaunch of vmresume is needed */
3762                 "cmpl $0, %c[launched](%0) \n\t"
3763                 /* Load guest registers.  Don't clobber flags. */
3764                 "mov %c[rax](%0), %%"R"ax \n\t"
3765                 "mov %c[rbx](%0), %%"R"bx \n\t"
3766                 "mov %c[rdx](%0), %%"R"dx \n\t"
3767                 "mov %c[rsi](%0), %%"R"si \n\t"
3768                 "mov %c[rdi](%0), %%"R"di \n\t"
3769                 "mov %c[rbp](%0), %%"R"bp \n\t"
3770 #ifdef CONFIG_X86_64
3771                 "mov %c[r8](%0),  %%r8  \n\t"
3772                 "mov %c[r9](%0),  %%r9  \n\t"
3773                 "mov %c[r10](%0), %%r10 \n\t"
3774                 "mov %c[r11](%0), %%r11 \n\t"
3775                 "mov %c[r12](%0), %%r12 \n\t"
3776                 "mov %c[r13](%0), %%r13 \n\t"
3777                 "mov %c[r14](%0), %%r14 \n\t"
3778                 "mov %c[r15](%0), %%r15 \n\t"
3779 #endif
3780                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3781
3782                 /* Enter guest mode */
3783                 "jne .Llaunched \n\t"
3784                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3785                 "jmp .Lkvm_vmx_return \n\t"
3786                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3787                 ".Lkvm_vmx_return: "
3788                 /* Save guest registers, load host registers, keep flags */
3789                 "xchg %0,     (%%"R"sp) \n\t"
3790                 "mov %%"R"ax, %c[rax](%0) \n\t"
3791                 "mov %%"R"bx, %c[rbx](%0) \n\t"
3792                 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3793                 "mov %%"R"dx, %c[rdx](%0) \n\t"
3794                 "mov %%"R"si, %c[rsi](%0) \n\t"
3795                 "mov %%"R"di, %c[rdi](%0) \n\t"
3796                 "mov %%"R"bp, %c[rbp](%0) \n\t"
3797 #ifdef CONFIG_X86_64
3798                 "mov %%r8,  %c[r8](%0) \n\t"
3799                 "mov %%r9,  %c[r9](%0) \n\t"
3800                 "mov %%r10, %c[r10](%0) \n\t"
3801                 "mov %%r11, %c[r11](%0) \n\t"
3802                 "mov %%r12, %c[r12](%0) \n\t"
3803                 "mov %%r13, %c[r13](%0) \n\t"
3804                 "mov %%r14, %c[r14](%0) \n\t"
3805                 "mov %%r15, %c[r15](%0) \n\t"
3806 #endif
3807                 "mov %%cr2, %%"R"ax   \n\t"
3808                 "mov %%"R"ax, %c[cr2](%0) \n\t"
3809
3810                 "pop  %%"R"bp; pop  %%"R"bp; pop  %%"R"dx \n\t"
3811                 "setbe %c[fail](%0) \n\t"
3812               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3813                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3814                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3815                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3816                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3817                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3818                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3819                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3820                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3821                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3822                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3823 #ifdef CONFIG_X86_64
3824                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3825                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3826                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3827                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3828                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3829                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3830                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3831                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
3832 #endif
3833                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3834               : "cc", "memory"
3835                 , R"bx", R"di", R"si"
3836 #ifdef CONFIG_X86_64
3837                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3838 #endif
3839               );
3840
3841         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
3842                                   | (1 << VCPU_EXREG_PDPTR));
3843         vcpu->arch.regs_dirty = 0;
3844
3845         if (vcpu->arch.switch_db_regs)
3846                 get_debugreg(vcpu->arch.dr6, 6);
3847
3848         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3849         if (vmx->rmode.irq.pending)
3850                 fixup_rmode_irq(vmx);
3851
3852         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3853         vmx->launched = 1;
3854
3855         vmx_complete_interrupts(vmx);
3856 }
3857
3858 #undef R
3859 #undef Q
3860
3861 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3862 {
3863         struct vcpu_vmx *vmx = to_vmx(vcpu);
3864
3865         if (vmx->vmcs) {
3866                 vcpu_clear(vmx);
3867                 free_vmcs(vmx->vmcs);
3868                 vmx->vmcs = NULL;
3869         }
3870 }
3871
3872 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3873 {
3874         struct vcpu_vmx *vmx = to_vmx(vcpu);
3875
3876         spin_lock(&vmx_vpid_lock);
3877         if (vmx->vpid != 0)
3878                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3879         spin_unlock(&vmx_vpid_lock);
3880         vmx_free_vmcs(vcpu);
3881         kfree(vmx->guest_msrs);
3882         kvm_vcpu_uninit(vcpu);
3883         kmem_cache_free(kvm_vcpu_cache, vmx);
3884 }
3885
3886 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
3887 {
3888         int err;
3889         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3890         int cpu;
3891
3892         if (!vmx)
3893                 return ERR_PTR(-ENOMEM);
3894
3895         allocate_vpid(vmx);
3896
3897         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3898         if (err)
3899                 goto free_vcpu;
3900
3901         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3902         if (!vmx->guest_msrs) {
3903                 err = -ENOMEM;
3904                 goto uninit_vcpu;
3905         }
3906
3907         vmx->vmcs = alloc_vmcs();
3908         if (!vmx->vmcs)
3909                 goto free_msrs;
3910
3911         vmcs_clear(vmx->vmcs);
3912
3913         cpu = get_cpu();
3914         vmx_vcpu_load(&vmx->vcpu, cpu);
3915         err = vmx_vcpu_setup(vmx);
3916         vmx_vcpu_put(&vmx->vcpu);
3917         put_cpu();
3918         if (err)
3919                 goto free_vmcs;
3920         if (vm_need_virtualize_apic_accesses(kvm))
3921                 if (alloc_apic_access_page(kvm) != 0)
3922                         goto free_vmcs;
3923
3924         if (enable_ept) {
3925                 if (!kvm->arch.ept_identity_map_addr)
3926                         kvm->arch.ept_identity_map_addr =
3927                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3928                 if (alloc_identity_pagetable(kvm) != 0)
3929                         goto free_vmcs;
3930         }
3931
3932         return &vmx->vcpu;
3933
3934 free_vmcs:
3935         free_vmcs(vmx->vmcs);
3936 free_msrs:
3937         kfree(vmx->guest_msrs);
3938 uninit_vcpu:
3939         kvm_vcpu_uninit(&vmx->vcpu);
3940 free_vcpu:
3941         kmem_cache_free(kvm_vcpu_cache, vmx);
3942         return ERR_PTR(err);
3943 }
3944
3945 static void __init vmx_check_processor_compat(void *rtn)
3946 {
3947         struct vmcs_config vmcs_conf;
3948
3949         *(int *)rtn = 0;
3950         if (setup_vmcs_config(&vmcs_conf) < 0)
3951                 *(int *)rtn = -EIO;
3952         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3953                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3954                                 smp_processor_id());
3955                 *(int *)rtn = -EIO;
3956         }
3957 }
3958
3959 static int get_ept_level(void)
3960 {
3961         return VMX_EPT_DEFAULT_GAW + 1;
3962 }
3963
3964 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3965 {
3966         u64 ret;
3967
3968         /* For VT-d and EPT combination
3969          * 1. MMIO: always map as UC
3970          * 2. EPT with VT-d:
3971          *   a. VT-d without snooping control feature: can't guarantee the
3972          *      result, try to trust guest.
3973          *   b. VT-d with snooping control feature: snooping control feature of
3974          *      VT-d engine can guarantee the cache correctness. Just set it
3975          *      to WB to keep consistent with host. So the same as item 3.
3976          * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep
3977          *    consistent with host MTRR
3978          */
3979         if (is_mmio)
3980                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
3981         else if (vcpu->kvm->arch.iommu_domain &&
3982                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
3983                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
3984                       VMX_EPT_MT_EPTE_SHIFT;
3985         else
3986                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
3987                         | VMX_EPT_IGMT_BIT;
3988
3989         return ret;
3990 }
3991
3992 static const struct trace_print_flags vmx_exit_reasons_str[] = {
3993         { EXIT_REASON_EXCEPTION_NMI,           "exception" },
3994         { EXIT_REASON_EXTERNAL_INTERRUPT,      "ext_irq" },
3995         { EXIT_REASON_TRIPLE_FAULT,            "triple_fault" },
3996         { EXIT_REASON_NMI_WINDOW,              "nmi_window" },
3997         { EXIT_REASON_IO_INSTRUCTION,          "io_instruction" },
3998         { EXIT_REASON_CR_ACCESS,               "cr_access" },
3999         { EXIT_REASON_DR_ACCESS,               "dr_access" },
4000         { EXIT_REASON_CPUID,                   "cpuid" },
4001         { EXIT_REASON_MSR_READ,                "rdmsr" },
4002         { EXIT_REASON_MSR_WRITE,               "wrmsr" },
4003         { EXIT_REASON_PENDING_INTERRUPT,       "interrupt_window" },
4004         { EXIT_REASON_HLT,                     "halt" },
4005         { EXIT_REASON_INVLPG,                  "invlpg" },
4006         { EXIT_REASON_VMCALL,                  "hypercall" },
4007         { EXIT_REASON_TPR_BELOW_THRESHOLD,     "tpr_below_thres" },
4008         { EXIT_REASON_APIC_ACCESS,             "apic_access" },
4009         { EXIT_REASON_WBINVD,                  "wbinvd" },
4010         { EXIT_REASON_TASK_SWITCH,             "task_switch" },
4011         { EXIT_REASON_EPT_VIOLATION,           "ept_violation" },
4012         { -1, NULL }
4013 };
4014
4015 static bool vmx_gb_page_enable(void)
4016 {
4017         return false;
4018 }
4019
4020 static inline u32 bit(int bitno)
4021 {
4022         return 1 << (bitno & 31);
4023 }
4024
4025 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4026 {
4027         struct kvm_cpuid_entry2 *best;
4028         struct vcpu_vmx *vmx = to_vmx(vcpu);
4029         u32 exec_control;
4030
4031         vmx->rdtscp_enabled = false;
4032         if (vmx_rdtscp_supported()) {
4033                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4034                 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4035                         best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4036                         if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4037                                 vmx->rdtscp_enabled = true;
4038                         else {
4039                                 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4040                                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4041                                                 exec_control);
4042                         }
4043                 }
4044         }
4045 }
4046
4047 static struct kvm_x86_ops vmx_x86_ops = {
4048         .cpu_has_kvm_support = cpu_has_kvm_support,
4049         .disabled_by_bios = vmx_disabled_by_bios,
4050         .hardware_setup = hardware_setup,
4051         .hardware_unsetup = hardware_unsetup,
4052         .check_processor_compatibility = vmx_check_processor_compat,
4053         .hardware_enable = hardware_enable,
4054         .hardware_disable = hardware_disable,
4055         .cpu_has_accelerated_tpr = report_flexpriority,
4056
4057         .vcpu_create = vmx_create_vcpu,
4058         .vcpu_free = vmx_free_vcpu,
4059         .vcpu_reset = vmx_vcpu_reset,
4060
4061         .prepare_guest_switch = vmx_save_host_state,
4062         .vcpu_load = vmx_vcpu_load,
4063         .vcpu_put = vmx_vcpu_put,
4064
4065         .set_guest_debug = set_guest_debug,
4066         .get_msr = vmx_get_msr,
4067         .set_msr = vmx_set_msr,
4068         .get_segment_base = vmx_get_segment_base,
4069         .get_segment = vmx_get_segment,
4070         .set_segment = vmx_set_segment,
4071         .get_cpl = vmx_get_cpl,
4072         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4073         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4074         .set_cr0 = vmx_set_cr0,
4075         .set_cr3 = vmx_set_cr3,
4076         .set_cr4 = vmx_set_cr4,
4077         .set_efer = vmx_set_efer,
4078         .get_idt = vmx_get_idt,
4079         .set_idt = vmx_set_idt,
4080         .get_gdt = vmx_get_gdt,
4081         .set_gdt = vmx_set_gdt,
4082         .cache_reg = vmx_cache_reg,
4083         .get_rflags = vmx_get_rflags,
4084         .set_rflags = vmx_set_rflags,
4085
4086         .tlb_flush = vmx_flush_tlb,
4087
4088         .run = vmx_vcpu_run,
4089         .handle_exit = vmx_handle_exit,
4090         .skip_emulated_instruction = skip_emulated_instruction,
4091         .set_interrupt_shadow = vmx_set_interrupt_shadow,
4092         .get_interrupt_shadow = vmx_get_interrupt_shadow,
4093         .patch_hypercall = vmx_patch_hypercall,
4094         .set_irq = vmx_inject_irq,
4095         .set_nmi = vmx_inject_nmi,
4096         .queue_exception = vmx_queue_exception,
4097         .interrupt_allowed = vmx_interrupt_allowed,
4098         .nmi_allowed = vmx_nmi_allowed,
4099         .get_nmi_mask = vmx_get_nmi_mask,
4100         .set_nmi_mask = vmx_set_nmi_mask,
4101         .enable_nmi_window = enable_nmi_window,
4102         .enable_irq_window = enable_irq_window,
4103         .update_cr8_intercept = update_cr8_intercept,
4104
4105         .set_tss_addr = vmx_set_tss_addr,
4106         .get_tdp_level = get_ept_level,
4107         .get_mt_mask = vmx_get_mt_mask,
4108
4109         .exit_reasons_str = vmx_exit_reasons_str,
4110         .gb_page_enable = vmx_gb_page_enable,
4111
4112         .cpuid_update = vmx_cpuid_update,
4113
4114         .rdtscp_supported = vmx_rdtscp_supported,
4115 };
4116
4117 static int __init vmx_init(void)
4118 {
4119         int r, i;
4120
4121         rdmsrl_safe(MSR_EFER, &host_efer);
4122
4123         for (i = 0; i < NR_VMX_MSR; ++i)
4124                 kvm_define_shared_msr(i, vmx_msr_index[i]);
4125
4126         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4127         if (!vmx_io_bitmap_a)
4128                 return -ENOMEM;
4129
4130         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4131         if (!vmx_io_bitmap_b) {
4132                 r = -ENOMEM;
4133                 goto out;
4134         }
4135
4136         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4137         if (!vmx_msr_bitmap_legacy) {
4138                 r = -ENOMEM;
4139                 goto out1;
4140         }
4141
4142         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4143         if (!vmx_msr_bitmap_longmode) {
4144                 r = -ENOMEM;
4145                 goto out2;
4146         }
4147
4148         /*
4149          * Allow direct access to the PC debug port (it is often used for I/O
4150          * delays, but the vmexits simply slow things down).
4151          */
4152         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4153         clear_bit(0x80, vmx_io_bitmap_a);
4154
4155         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4156
4157         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4158         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4159
4160         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4161
4162         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
4163         if (r)
4164                 goto out3;
4165
4166         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4167         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4168         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4169         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4170         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4171         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4172
4173         if (enable_ept) {
4174                 bypass_guest_pf = 0;
4175                 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
4176                         VMX_EPT_WRITABLE_MASK);
4177                 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4178                                 VMX_EPT_EXECUTABLE_MASK);
4179                 kvm_enable_tdp();
4180         } else
4181                 kvm_disable_tdp();
4182
4183         if (bypass_guest_pf)
4184                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4185
4186         return 0;
4187
4188 out3:
4189         free_page((unsigned long)vmx_msr_bitmap_longmode);
4190 out2:
4191         free_page((unsigned long)vmx_msr_bitmap_legacy);
4192 out1:
4193         free_page((unsigned long)vmx_io_bitmap_b);
4194 out:
4195         free_page((unsigned long)vmx_io_bitmap_a);
4196         return r;
4197 }
4198
4199 static void __exit vmx_exit(void)
4200 {
4201         free_page((unsigned long)vmx_msr_bitmap_legacy);
4202         free_page((unsigned long)vmx_msr_bitmap_longmode);
4203         free_page((unsigned long)vmx_io_bitmap_b);
4204         free_page((unsigned long)vmx_io_bitmap_a);
4205
4206         kvm_exit();
4207 }
4208
4209 module_init(vmx_init)
4210 module_exit(vmx_exit)