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  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "irq.h"
  20 #include "mmu.h"
  21 #include "cpuid.h"
  22
  23 #include <linux/kvm_host.h>
  24 #include <linux/module.h>
  25 #include <linux/kernel.h>
  26 #include <linux/mm.h>
  27 #include <linux/highmem.h>
  28 #include <linux/sched.h>
  29 #include <linux/moduleparam.h>
  30 #include <linux/mod_devicetable.h>
  31 #include <linux/ftrace_event.h>
  32 #include <linux/slab.h>
  33 #include <linux/tboot.h>
  34 #include "kvm_cache_regs.h"
  35 #include "x86.h"
  36
  37 #include <asm/io.h>
  38 #include <asm/desc.h>
  39 #include <asm/vmx.h>
  40 #include <asm/virtext.h>
  41 #include <asm/mce.h>
  42 #include <asm/i387.h>
  43 #include <asm/xcr.h>
  44 #include <asm/perf_event.h>
  45 #include <asm/kexec.h>
  46
  47 #include "trace.h"
  48
  49 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  50 #define __ex_clear(x, reg) \
  51         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
  52
  53 MODULE_AUTHOR("Qumranet");
  54 MODULE_LICENSE("GPL");
  55
  56 static const struct x86_cpu_id vmx_cpu_id[] = {
  57         X86_FEATURE_MATCH(X86_FEATURE_VMX),
  58         {}
  59 };
  60 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
  61
  62 static bool __read_mostly enable_vpid = 1;
  63 module_param_named(vpid, enable_vpid, bool, 0444);
  64
  65 static bool __read_mostly flexpriority_enabled = 1;
  66 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  67
  68 static bool __read_mostly enable_ept = 1;
  69 module_param_named(ept, enable_ept, bool, S_IRUGO);
  70
  71 static bool __read_mostly enable_unrestricted_guest = 1;
  72 module_param_named(unrestricted_guest,
  73                         enable_unrestricted_guest, bool, S_IRUGO);
  74
  75 static bool __read_mostly enable_ept_ad_bits = 1;
  76 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
  77
  78 static bool __read_mostly emulate_invalid_guest_state = true;
  79 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  80
  81 static bool __read_mostly vmm_exclusive = 1;
  82 module_param(vmm_exclusive, bool, S_IRUGO);
  83
  84 static bool __read_mostly fasteoi = 1;
  85 module_param(fasteoi, bool, S_IRUGO);
  86
  87 static bool __read_mostly enable_apicv = 1;
  88 module_param(enable_apicv, bool, S_IRUGO);
  89
  90 static bool __read_mostly enable_shadow_vmcs = 1;
  91 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
  92 /*
  93  * If nested=1, nested virtualization is supported, i.e., guests may use
  94  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
  95  * use VMX instructions.
  96  */
  97 static bool __read_mostly nested = 0;
  98 module_param(nested, bool, S_IRUGO);
  99
 100 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
 101 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
 102 #define KVM_VM_CR0_ALWAYS_ON                                            \
 103         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
 104 #define KVM_CR4_GUEST_OWNED_BITS                                      \
 105         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
 106          | X86_CR4_OSXMMEXCPT)
 107
 108 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
 109 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
 110
 111 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
 112
 113 /*
 114  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 115  * ple_gap:    upper bound on the amount of time between two successive
 116  *             executions of PAUSE in a loop. Also indicate if ple enabled.
 117  *             According to test, this time is usually smaller than 128 cycles.
 118  * ple_window: upper bound on the amount of time a guest is allowed to execute
 119  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
 120  *             less than 2^12 cycles
 121  * Time is measured based on a counter that runs at the same rate as the TSC,
 122  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 123  */
 124 #define KVM_VMX_DEFAULT_PLE_GAP    128
 125 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
 126 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 127 module_param(ple_gap, int, S_IRUGO);
 128
 129 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 130 module_param(ple_window, int, S_IRUGO);
 131
 132 extern const ulong vmx_return;
 133
 134 #define NR_AUTOLOAD_MSRS 8
 135 #define VMCS02_POOL_SIZE 1
 136
 137 struct vmcs {
 138         u32 revision_id;
 139         u32 abort;
 140         char data[0];
 141 };
 142
 143 /*
 144  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
 145  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
 146  * loaded on this CPU (so we can clear them if the CPU goes down).
 147  */
 148 struct loaded_vmcs {
 149         struct vmcs *vmcs;
 150         int cpu;
 151         int launched;
 152         struct list_head loaded_vmcss_on_cpu_link;
 153 };
 154
 155 struct shared_msr_entry {
 156         unsigned index;
 157         u64 data;
 158         u64 mask;
 159 };
 160
 161 /*
 162  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
 163  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
 164  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
 165  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
 166  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
 167  * More than one of these structures may exist, if L1 runs multiple L2 guests.
 168  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
 169  * underlying hardware which will be used to run L2.
 170  * This structure is packed to ensure that its layout is identical across
 171  * machines (necessary for live migration).
 172  * If there are changes in this struct, VMCS12_REVISION must be changed.
 173  */
 174 typedef u64 natural_width;
 175 struct __packed vmcs12 {
 176         /* According to the Intel spec, a VMCS region must start with the
 177          * following two fields. Then follow implementation-specific data.
 178          */
 179         u32 revision_id;
 180         u32 abort;
 181
 182         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
 183         u32 padding[7]; /* room for future expansion */
 184
 185         u64 io_bitmap_a;
 186         u64 io_bitmap_b;
 187         u64 msr_bitmap;
 188         u64 vm_exit_msr_store_addr;
 189         u64 vm_exit_msr_load_addr;
 190         u64 vm_entry_msr_load_addr;
 191         u64 tsc_offset;
 192         u64 virtual_apic_page_addr;
 193         u64 apic_access_addr;
 194         u64 ept_pointer;
 195         u64 guest_physical_address;
 196         u64 vmcs_link_pointer;
 197         u64 guest_ia32_debugctl;
 198         u64 guest_ia32_pat;
 199         u64 guest_ia32_efer;
 200         u64 guest_ia32_perf_global_ctrl;
 201         u64 guest_pdptr0;
 202         u64 guest_pdptr1;
 203         u64 guest_pdptr2;
 204         u64 guest_pdptr3;
 205         u64 host_ia32_pat;
 206         u64 host_ia32_efer;
 207         u64 host_ia32_perf_global_ctrl;
 208         u64 padding64[8]; /* room for future expansion */
 209         /*
 210          * To allow migration of L1 (complete with its L2 guests) between
 211          * machines of different natural widths (32 or 64 bit), we cannot have
 212          * unsigned long fields with no explict size. We use u64 (aliased
 213          * natural_width) instead. Luckily, x86 is little-endian.
 214          */
 215         natural_width cr0_guest_host_mask;
 216         natural_width cr4_guest_host_mask;
 217         natural_width cr0_read_shadow;
 218         natural_width cr4_read_shadow;
 219         natural_width cr3_target_value0;
 220         natural_width cr3_target_value1;
 221         natural_width cr3_target_value2;
 222         natural_width cr3_target_value3;
 223         natural_width exit_qualification;
 224         natural_width guest_linear_address;
 225         natural_width guest_cr0;
 226         natural_width guest_cr3;
 227         natural_width guest_cr4;
 228         natural_width guest_es_base;
 229         natural_width guest_cs_base;
 230         natural_width guest_ss_base;
 231         natural_width guest_ds_base;
 232         natural_width guest_fs_base;
 233         natural_width guest_gs_base;
 234         natural_width guest_ldtr_base;
 235         natural_width guest_tr_base;
 236         natural_width guest_gdtr_base;
 237         natural_width guest_idtr_base;
 238         natural_width guest_dr7;
 239         natural_width guest_rsp;
 240         natural_width guest_rip;
 241         natural_width guest_rflags;
 242         natural_width guest_pending_dbg_exceptions;
 243         natural_width guest_sysenter_esp;
 244         natural_width guest_sysenter_eip;
 245         natural_width host_cr0;
 246         natural_width host_cr3;
 247         natural_width host_cr4;
 248         natural_width host_fs_base;
 249         natural_width host_gs_base;
 250         natural_width host_tr_base;
 251         natural_width host_gdtr_base;
 252         natural_width host_idtr_base;
 253         natural_width host_ia32_sysenter_esp;
 254         natural_width host_ia32_sysenter_eip;
 255         natural_width host_rsp;
 256         natural_width host_rip;
 257         natural_width paddingl[8]; /* room for future expansion */
 258         u32 pin_based_vm_exec_control;
 259         u32 cpu_based_vm_exec_control;
 260         u32 exception_bitmap;
 261         u32 page_fault_error_code_mask;
 262         u32 page_fault_error_code_match;
 263         u32 cr3_target_count;
 264         u32 vm_exit_controls;
 265         u32 vm_exit_msr_store_count;
 266         u32 vm_exit_msr_load_count;
 267         u32 vm_entry_controls;
 268         u32 vm_entry_msr_load_count;
 269         u32 vm_entry_intr_info_field;
 270         u32 vm_entry_exception_error_code;
 271         u32 vm_entry_instruction_len;
 272         u32 tpr_threshold;
 273         u32 secondary_vm_exec_control;
 274         u32 vm_instruction_error;
 275         u32 vm_exit_reason;
 276         u32 vm_exit_intr_info;
 277         u32 vm_exit_intr_error_code;
 278         u32 idt_vectoring_info_field;
 279         u32 idt_vectoring_error_code;
 280         u32 vm_exit_instruction_len;
 281         u32 vmx_instruction_info;
 282         u32 guest_es_limit;
 283         u32 guest_cs_limit;
 284         u32 guest_ss_limit;
 285         u32 guest_ds_limit;
 286         u32 guest_fs_limit;
 287         u32 guest_gs_limit;
 288         u32 guest_ldtr_limit;
 289         u32 guest_tr_limit;
 290         u32 guest_gdtr_limit;
 291         u32 guest_idtr_limit;
 292         u32 guest_es_ar_bytes;
 293         u32 guest_cs_ar_bytes;
 294         u32 guest_ss_ar_bytes;
 295         u32 guest_ds_ar_bytes;
 296         u32 guest_fs_ar_bytes;
 297         u32 guest_gs_ar_bytes;
 298         u32 guest_ldtr_ar_bytes;
 299         u32 guest_tr_ar_bytes;
 300         u32 guest_interruptibility_info;
 301         u32 guest_activity_state;
 302         u32 guest_sysenter_cs;
 303         u32 host_ia32_sysenter_cs;
 304         u32 vmx_preemption_timer_value;
 305         u32 padding32[7]; /* room for future expansion */
 306         u16 virtual_processor_id;
 307         u16 guest_es_selector;
 308         u16 guest_cs_selector;
 309         u16 guest_ss_selector;
 310         u16 guest_ds_selector;
 311         u16 guest_fs_selector;
 312         u16 guest_gs_selector;
 313         u16 guest_ldtr_selector;
 314         u16 guest_tr_selector;
 315         u16 host_es_selector;
 316         u16 host_cs_selector;
 317         u16 host_ss_selector;
 318         u16 host_ds_selector;
 319         u16 host_fs_selector;
 320         u16 host_gs_selector;
 321         u16 host_tr_selector;
 322 };
 323
 324 /*
 325  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
 326  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
 327  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
 328  */
 329 #define VMCS12_REVISION 0x11e57ed0
 330
 331 /*
 332  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
 333  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
 334  * current implementation, 4K are reserved to avoid future complications.
 335  */
 336 #define VMCS12_SIZE 0x1000
 337
 338 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
 339 struct vmcs02_list {
 340         struct list_head list;
 341         gpa_t vmptr;
 342         struct loaded_vmcs vmcs02;
 343 };
 344
 345 /*
 346  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 347  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 348  */
 349 struct nested_vmx {
 350         /* Has the level1 guest done vmxon? */
 351         bool vmxon;
 352
 353         /* The guest-physical address of the current VMCS L1 keeps for L2 */
 354         gpa_t current_vmptr;
 355         /* The host-usable pointer to the above */
 356         struct page *current_vmcs12_page;
 357         struct vmcs12 *current_vmcs12;
 358         struct vmcs *current_shadow_vmcs;
 359         /*
 360          * Indicates if the shadow vmcs must be updated with the
 361          * data hold by vmcs12
 362          */
 363         bool sync_shadow_vmcs;
 364
 365         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
 366         struct list_head vmcs02_pool;
 367         int vmcs02_num;
 368         u64 vmcs01_tsc_offset;
 369         /* L2 must run next, and mustn't decide to exit to L1. */
 370         bool nested_run_pending;
 371         /*
 372          * Guest pages referred to in vmcs02 with host-physical pointers, so
 373          * we must keep them pinned while L2 runs.
 374          */
 375         struct page *apic_access_page;
 376         u64 msr_ia32_feature_control;
 377 };
 378
 379 #define POSTED_INTR_ON  0
 380 /* Posted-Interrupt Descriptor */
 381 struct pi_desc {
 382         u32 pir[8];     /* Posted interrupt requested */
 383         u32 control;    /* bit 0 of control is outstanding notification bit */
 384         u32 rsvd[7];
 385 } __aligned(64);
 386
 387 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
 388 {
 389         return test_and_set_bit(POSTED_INTR_ON,
 390                         (unsigned long *)&pi_desc->control);
 391 }
 392
 393 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
 394 {
 395         return test_and_clear_bit(POSTED_INTR_ON,
 396                         (unsigned long *)&pi_desc->control);
 397 }
 398
 399 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
 400 {
 401         return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
 402 }
 403
 404 struct vcpu_vmx {
 405         struct kvm_vcpu       vcpu;
 406         unsigned long         host_rsp;
 407         u8                    fail;
 408         u8                    cpl;
 409         bool                  nmi_known_unmasked;
 410         u32                   exit_intr_info;
 411         u32                   idt_vectoring_info;
 412         ulong                 rflags;
 413         struct shared_msr_entry *guest_msrs;
 414         int                   nmsrs;
 415         int                   save_nmsrs;
 416         unsigned long         host_idt_base;
 417 #ifdef CONFIG_X86_64
 418         u64                   msr_host_kernel_gs_base;
 419         u64                   msr_guest_kernel_gs_base;
 420 #endif
 421         /*
 422          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 423          * non-nested (L1) guest, it always points to vmcs01. For a nested
 424          * guest (L2), it points to a different VMCS.
 425          */
 426         struct loaded_vmcs    vmcs01;
 427         struct loaded_vmcs   *loaded_vmcs;
 428         bool                  __launched; /* temporary, used in vmx_vcpu_run */
 429         struct msr_autoload {
 430                 unsigned nr;
 431                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 432                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 433         } msr_autoload;
 434         struct {
 435                 int           loaded;
 436                 u16           fs_sel, gs_sel, ldt_sel;
 437 #ifdef CONFIG_X86_64
 438                 u16           ds_sel, es_sel;
 439 #endif
 440                 int           gs_ldt_reload_needed;
 441                 int           fs_reload_needed;
 442         } host_state;
 443         struct {
 444                 int vm86_active;
 445                 ulong save_rflags;
 446                 struct kvm_segment segs[8];
 447         } rmode;
 448         struct {
 449                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 450                 struct kvm_save_segment {
 451                         u16 selector;
 452                         unsigned long base;
 453                         u32 limit;
 454                         u32 ar;
 455                 } seg[8];
 456         } segment_cache;
 457         int vpid;
 458         bool emulation_required;
 459
 460         /* Support for vnmi-less CPUs */
 461         int soft_vnmi_blocked;
 462         ktime_t entry_time;
 463         s64 vnmi_blocked_time;
 464         u32 exit_reason;
 465
 466         bool rdtscp_enabled;
 467
 468         /* Posted interrupt descriptor */
 469         struct pi_desc pi_desc;
 470
 471         /* Support for a guest hypervisor (nested VMX) */
 472         struct nested_vmx nested;
 473 };
 474
 475 enum segment_cache_field {
 476         SEG_FIELD_SEL = 0,
 477         SEG_FIELD_BASE = 1,
 478         SEG_FIELD_LIMIT = 2,
 479         SEG_FIELD_AR = 3,
 480
 481         SEG_FIELD_NR = 4
 482 };
 483
 484 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 485 {
 486         return container_of(vcpu, struct vcpu_vmx, vcpu);
 487 }
 488
 489 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
 490 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
 491 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
 492                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
 493
 494
 495 static const unsigned long shadow_read_only_fields[] = {
 496         /*
 497          * We do NOT shadow fields that are modified when L0
 498          * traps and emulates any vmx instruction (e.g. VMPTRLD,
 499          * VMXON...) executed by L1.
 500          * For example, VM_INSTRUCTION_ERROR is read
 501          * by L1 if a vmx instruction fails (part of the error path).
 502          * Note the code assumes this logic. If for some reason
 503          * we start shadowing these fields then we need to
 504          * force a shadow sync when L0 emulates vmx instructions
 505          * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
 506          * by nested_vmx_failValid)
 507          */
 508         VM_EXIT_REASON,
 509         VM_EXIT_INTR_INFO,
 510         VM_EXIT_INSTRUCTION_LEN,
 511         IDT_VECTORING_INFO_FIELD,
 512         IDT_VECTORING_ERROR_CODE,
 513         VM_EXIT_INTR_ERROR_CODE,
 514         EXIT_QUALIFICATION,
 515         GUEST_LINEAR_ADDRESS,
 516         GUEST_PHYSICAL_ADDRESS
 517 };
 518 static const int max_shadow_read_only_fields =
 519         ARRAY_SIZE(shadow_read_only_fields);
 520
 521 static const unsigned long shadow_read_write_fields[] = {
 522         GUEST_RIP,
 523         GUEST_RSP,
 524         GUEST_CR0,
 525         GUEST_CR3,
 526         GUEST_CR4,
 527         GUEST_INTERRUPTIBILITY_INFO,
 528         GUEST_RFLAGS,
 529         GUEST_CS_SELECTOR,
 530         GUEST_CS_AR_BYTES,
 531         GUEST_CS_LIMIT,
 532         GUEST_CS_BASE,
 533         GUEST_ES_BASE,
 534         CR0_GUEST_HOST_MASK,
 535         CR0_READ_SHADOW,
 536         CR4_READ_SHADOW,
 537         TSC_OFFSET,
 538         EXCEPTION_BITMAP,
 539         CPU_BASED_VM_EXEC_CONTROL,
 540         VM_ENTRY_EXCEPTION_ERROR_CODE,
 541         VM_ENTRY_INTR_INFO_FIELD,
 542         VM_ENTRY_INSTRUCTION_LEN,
 543         VM_ENTRY_EXCEPTION_ERROR_CODE,
 544         HOST_FS_BASE,
 545         HOST_GS_BASE,
 546         HOST_FS_SELECTOR,
 547         HOST_GS_SELECTOR
 548 };
 549 static const int max_shadow_read_write_fields =
 550         ARRAY_SIZE(shadow_read_write_fields);
 551
 552 static const unsigned short vmcs_field_to_offset_table[] = {
 553         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 554         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
 555         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
 556         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
 557         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
 558         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
 559         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
 560         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
 561         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
 562         FIELD(HOST_ES_SELECTOR, host_es_selector),
 563         FIELD(HOST_CS_SELECTOR, host_cs_selector),
 564         FIELD(HOST_SS_SELECTOR, host_ss_selector),
 565         FIELD(HOST_DS_SELECTOR, host_ds_selector),
 566         FIELD(HOST_FS_SELECTOR, host_fs_selector),
 567         FIELD(HOST_GS_SELECTOR, host_gs_selector),
 568         FIELD(HOST_TR_SELECTOR, host_tr_selector),
 569         FIELD64(IO_BITMAP_A, io_bitmap_a),
 570         FIELD64(IO_BITMAP_B, io_bitmap_b),
 571         FIELD64(MSR_BITMAP, msr_bitmap),
 572         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
 573         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
 574         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
 575         FIELD64(TSC_OFFSET, tsc_offset),
 576         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
 577         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
 578         FIELD64(EPT_POINTER, ept_pointer),
 579         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
 580         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
 581         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
 582         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
 583         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
 584         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
 585         FIELD64(GUEST_PDPTR0, guest_pdptr0),
 586         FIELD64(GUEST_PDPTR1, guest_pdptr1),
 587         FIELD64(GUEST_PDPTR2, guest_pdptr2),
 588         FIELD64(GUEST_PDPTR3, guest_pdptr3),
 589         FIELD64(HOST_IA32_PAT, host_ia32_pat),
 590         FIELD64(HOST_IA32_EFER, host_ia32_efer),
 591         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
 592         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
 593         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
 594         FIELD(EXCEPTION_BITMAP, exception_bitmap),
 595         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
 596         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
 597         FIELD(CR3_TARGET_COUNT, cr3_target_count),
 598         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
 599         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
 600         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
 601         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
 602         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
 603         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
 604         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
 605         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
 606         FIELD(TPR_THRESHOLD, tpr_threshold),
 607         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
 608         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
 609         FIELD(VM_EXIT_REASON, vm_exit_reason),
 610         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
 611         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
 612         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
 613         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
 614         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
 615         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
 616         FIELD(GUEST_ES_LIMIT, guest_es_limit),
 617         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
 618         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
 619         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
 620         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
 621         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
 622         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
 623         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
 624         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
 625         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
 626         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
 627         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
 628         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
 629         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
 630         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
 631         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
 632         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
 633         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
 634         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
 635         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
 636         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
 637         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
 638         FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
 639         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
 640         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
 641         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
 642         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
 643         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
 644         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
 645         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
 646         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
 647         FIELD(EXIT_QUALIFICATION, exit_qualification),
 648         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
 649         FIELD(GUEST_CR0, guest_cr0),
 650         FIELD(GUEST_CR3, guest_cr3),
 651         FIELD(GUEST_CR4, guest_cr4),
 652         FIELD(GUEST_ES_BASE, guest_es_base),
 653         FIELD(GUEST_CS_BASE, guest_cs_base),
 654         FIELD(GUEST_SS_BASE, guest_ss_base),
 655         FIELD(GUEST_DS_BASE, guest_ds_base),
 656         FIELD(GUEST_FS_BASE, guest_fs_base),
 657         FIELD(GUEST_GS_BASE, guest_gs_base),
 658         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
 659         FIELD(GUEST_TR_BASE, guest_tr_base),
 660         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
 661         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
 662         FIELD(GUEST_DR7, guest_dr7),
 663         FIELD(GUEST_RSP, guest_rsp),
 664         FIELD(GUEST_RIP, guest_rip),
 665         FIELD(GUEST_RFLAGS, guest_rflags),
 666         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
 667         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
 668         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
 669         FIELD(HOST_CR0, host_cr0),
 670         FIELD(HOST_CR3, host_cr3),
 671         FIELD(HOST_CR4, host_cr4),
 672         FIELD(HOST_FS_BASE, host_fs_base),
 673         FIELD(HOST_GS_BASE, host_gs_base),
 674         FIELD(HOST_TR_BASE, host_tr_base),
 675         FIELD(HOST_GDTR_BASE, host_gdtr_base),
 676         FIELD(HOST_IDTR_BASE, host_idtr_base),
 677         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
 678         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
 679         FIELD(HOST_RSP, host_rsp),
 680         FIELD(HOST_RIP, host_rip),
 681 };
 682 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
 683
 684 static inline short vmcs_field_to_offset(unsigned long field)
 685 {
 686         if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
 687                 return -1;
 688         return vmcs_field_to_offset_table[field];
 689 }
 690
 691 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
 692 {
 693         return to_vmx(vcpu)->nested.current_vmcs12;
 694 }
 695
 696 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
 697 {
 698         struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
 699         if (is_error_page(page))
 700                 return NULL;
 701
 702         return page;
 703 }
 704
 705 static void nested_release_page(struct page *page)
 706 {
 707         kvm_release_page_dirty(page);
 708 }
 709
 710 static void nested_release_page_clean(struct page *page)
 711 {
 712         kvm_release_page_clean(page);
 713 }
 714
 715 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
 716 static u64 construct_eptp(unsigned long root_hpa);
 717 static void kvm_cpu_vmxon(u64 addr);
 718 static void kvm_cpu_vmxoff(void);
 719 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 720 static void vmx_set_segment(struct kvm_vcpu *vcpu,
 721                             struct kvm_segment *var, int seg);
 722 static void vmx_get_segment(struct kvm_vcpu *vcpu,
 723                             struct kvm_segment *var, int seg);
 724 static bool guest_state_valid(struct kvm_vcpu *vcpu);
 725 static u32 vmx_segment_access_rights(struct kvm_segment *var);
 726 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
 727 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
 728 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
 729
 730 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 731 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 732 /*
 733  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
 734  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
 735  */
 736 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
 737 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 738
 739 static unsigned long *vmx_io_bitmap_a;
 740 static unsigned long *vmx_io_bitmap_b;
 741 static unsigned long *vmx_msr_bitmap_legacy;
 742 static unsigned long *vmx_msr_bitmap_longmode;
 743 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
 744 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
 745 static unsigned long *vmx_vmread_bitmap;
 746 static unsigned long *vmx_vmwrite_bitmap;
 747
 748 static bool cpu_has_load_ia32_efer;
 749 static bool cpu_has_load_perf_global_ctrl;
 750
 751 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 752 static DEFINE_SPINLOCK(vmx_vpid_lock);
 753
 754 static struct vmcs_config {
 755         int size;
 756         int order;
 757         u32 revision_id;
 758         u32 pin_based_exec_ctrl;
 759         u32 cpu_based_exec_ctrl;
 760         u32 cpu_based_2nd_exec_ctrl;
 761         u32 vmexit_ctrl;
 762         u32 vmentry_ctrl;
 763 } vmcs_config;
 764
 765 static struct vmx_capability {
 766         u32 ept;
 767         u32 vpid;
 768 } vmx_capability;
 769
 770 #define VMX_SEGMENT_FIELD(seg)                                  \
 771         [VCPU_SREG_##seg] = {                                   \
 772                 .selector = GUEST_##seg##_SELECTOR,             \
 773                 .base = GUEST_##seg##_BASE,                     \
 774                 .limit = GUEST_##seg##_LIMIT,                   \
 775                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 776         }
 777
 778 static const struct kvm_vmx_segment_field {
 779         unsigned selector;
 780         unsigned base;
 781         unsigned limit;
 782         unsigned ar_bytes;
 783 } kvm_vmx_segment_fields[] = {
 784         VMX_SEGMENT_FIELD(CS),
 785         VMX_SEGMENT_FIELD(DS),
 786         VMX_SEGMENT_FIELD(ES),
 787         VMX_SEGMENT_FIELD(FS),
 788         VMX_SEGMENT_FIELD(GS),
 789         VMX_SEGMENT_FIELD(SS),
 790         VMX_SEGMENT_FIELD(TR),
 791         VMX_SEGMENT_FIELD(LDTR),
 792 };
 793
 794 static u64 host_efer;
 795
 796 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 797
 798 /*
 799  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
 800  * away by decrementing the array size.
 801  */
 802 static const u32 vmx_msr_index[] = {
 803 #ifdef CONFIG_X86_64
 804         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 805 #endif
 806         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
 807 };
 808 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 809
 810 static inline bool is_page_fault(u32 intr_info)
 811 {
 812         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 813                              INTR_INFO_VALID_MASK)) ==
 814                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 815 }
 816
 817 static inline bool is_no_device(u32 intr_info)
 818 {
 819         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 820                              INTR_INFO_VALID_MASK)) ==
 821                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 822 }
 823
 824 static inline bool is_invalid_opcode(u32 intr_info)
 825 {
 826         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 827                              INTR_INFO_VALID_MASK)) ==
 828                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 829 }
 830
 831 static inline bool is_external_interrupt(u32 intr_info)
 832 {
 833         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 834                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 835 }
 836
 837 static inline bool is_machine_check(u32 intr_info)
 838 {
 839         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 840                              INTR_INFO_VALID_MASK)) ==
 841                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
 842 }
 843
 844 static inline bool cpu_has_vmx_msr_bitmap(void)
 845 {
 846         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
 847 }
 848
 849 static inline bool cpu_has_vmx_tpr_shadow(void)
 850 {
 851         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
 852 }
 853
 854 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
 855 {
 856         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
 857 }
 858
 859 static inline bool cpu_has_secondary_exec_ctrls(void)
 860 {
 861         return vmcs_config.cpu_based_exec_ctrl &
 862                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
 863 }
 864
 865 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 866 {
 867         return vmcs_config.cpu_based_2nd_exec_ctrl &
 868                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
 869 }
 870
 871 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
 872 {
 873         return vmcs_config.cpu_based_2nd_exec_ctrl &
 874                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
 875 }
 876
 877 static inline bool cpu_has_vmx_apic_register_virt(void)
 878 {
 879         return vmcs_config.cpu_based_2nd_exec_ctrl &
 880                 SECONDARY_EXEC_APIC_REGISTER_VIRT;
 881 }
 882
 883 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
 884 {
 885         return vmcs_config.cpu_based_2nd_exec_ctrl &
 886                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
 887 }
 888
 889 static inline bool cpu_has_vmx_posted_intr(void)
 890 {
 891         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
 892 }
 893
 894 static inline bool cpu_has_vmx_apicv(void)
 895 {
 896         return cpu_has_vmx_apic_register_virt() &&
 897                 cpu_has_vmx_virtual_intr_delivery() &&
 898                 cpu_has_vmx_posted_intr();
 899 }
 900
 901 static inline bool cpu_has_vmx_flexpriority(void)
 902 {
 903         return cpu_has_vmx_tpr_shadow() &&
 904                 cpu_has_vmx_virtualize_apic_accesses();
 905 }
 906
 907 static inline bool cpu_has_vmx_ept_execute_only(void)
 908 {
 909         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
 910 }
 911
 912 static inline bool cpu_has_vmx_eptp_uncacheable(void)
 913 {
 914         return vmx_capability.ept & VMX_EPTP_UC_BIT;
 915 }
 916
 917 static inline bool cpu_has_vmx_eptp_writeback(void)
 918 {
 919         return vmx_capability.ept & VMX_EPTP_WB_BIT;
 920 }
 921
 922 static inline bool cpu_has_vmx_ept_2m_page(void)
 923 {
 924         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
 925 }
 926
 927 static inline bool cpu_has_vmx_ept_1g_page(void)
 928 {
 929         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
 930 }
 931
 932 static inline bool cpu_has_vmx_ept_4levels(void)
 933 {
 934         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
 935 }
 936
 937 static inline bool cpu_has_vmx_ept_ad_bits(void)
 938 {
 939         return vmx_capability.ept & VMX_EPT_AD_BIT;
 940 }
 941
 942 static inline bool cpu_has_vmx_invept_context(void)
 943 {
 944         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
 945 }
 946
 947 static inline bool cpu_has_vmx_invept_global(void)
 948 {
 949         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
 950 }
 951
 952 static inline bool cpu_has_vmx_invvpid_single(void)
 953 {
 954         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
 955 }
 956
 957 static inline bool cpu_has_vmx_invvpid_global(void)
 958 {
 959         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
 960 }
 961
 962 static inline bool cpu_has_vmx_ept(void)
 963 {
 964         return vmcs_config.cpu_based_2nd_exec_ctrl &
 965                 SECONDARY_EXEC_ENABLE_EPT;
 966 }
 967
 968 static inline bool cpu_has_vmx_unrestricted_guest(void)
 969 {
 970         return vmcs_config.cpu_based_2nd_exec_ctrl &
 971                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
 972 }
 973
 974 static inline bool cpu_has_vmx_ple(void)
 975 {
 976         return vmcs_config.cpu_based_2nd_exec_ctrl &
 977                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 978 }
 979
 980 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
 981 {
 982         return flexpriority_enabled && irqchip_in_kernel(kvm);
 983 }
 984
 985 static inline bool cpu_has_vmx_vpid(void)
 986 {
 987         return vmcs_config.cpu_based_2nd_exec_ctrl &
 988                 SECONDARY_EXEC_ENABLE_VPID;
 989 }
 990
 991 static inline bool cpu_has_vmx_rdtscp(void)
 992 {
 993         return vmcs_config.cpu_based_2nd_exec_ctrl &
 994                 SECONDARY_EXEC_RDTSCP;
 995 }
 996
 997 static inline bool cpu_has_vmx_invpcid(void)
 998 {
 999         return vmcs_config.cpu_based_2nd_exec_ctrl &
1000                 SECONDARY_EXEC_ENABLE_INVPCID;
1001 }
1002
1003 static inline bool cpu_has_virtual_nmis(void)
1004 {
1005         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1006 }
1007
1008 static inline bool cpu_has_vmx_wbinvd_exit(void)
1009 {
1010         return vmcs_config.cpu_based_2nd_exec_ctrl &
1011                 SECONDARY_EXEC_WBINVD_EXITING;
1012 }
1013
1014 static inline bool cpu_has_vmx_shadow_vmcs(void)
1015 {
1016         u64 vmx_msr;
1017         rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1018         /* check if the cpu supports writing r/o exit information fields */
1019         if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1020                 return false;
1021
1022         return vmcs_config.cpu_based_2nd_exec_ctrl &
1023                 SECONDARY_EXEC_SHADOW_VMCS;
1024 }
1025
1026 static inline bool report_flexpriority(void)
1027 {
1028         return flexpriority_enabled;
1029 }
1030
1031 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1032 {
1033         return vmcs12->cpu_based_vm_exec_control & bit;
1034 }
1035
1036 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1037 {
1038         return (vmcs12->cpu_based_vm_exec_control &
1039                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1040                 (vmcs12->secondary_vm_exec_control & bit);
1041 }
1042
1043 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1044 {
1045         return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1046 }
1047
1048 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1049 {
1050         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1051 }
1052
1053 static inline bool is_exception(u32 intr_info)
1054 {
1055         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1056                 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1057 }
1058
1059 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
1060 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1061                         struct vmcs12 *vmcs12,
1062                         u32 reason, unsigned long qualification);
1063
1064 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1065 {
1066         int i;
1067
1068         for (i = 0; i < vmx->nmsrs; ++i)
1069                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1070                         return i;
1071         return -1;
1072 }
1073
1074 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1075 {
1076     struct {
1077         u64 vpid : 16;
1078         u64 rsvd : 48;
1079         u64 gva;
1080     } operand = { vpid, 0, gva };
1081
1082     asm volatile (__ex(ASM_VMX_INVVPID)
1083                   /* CF==1 or ZF==1 --> rc = -1 */
1084                   "; ja 1f ; ud2 ; 1:"
1085                   : : "a"(&operand), "c"(ext) : "cc", "memory");
1086 }
1087
1088 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1089 {
1090         struct {
1091                 u64 eptp, gpa;
1092         } operand = {eptp, gpa};
1093
1094         asm volatile (__ex(ASM_VMX_INVEPT)
1095                         /* CF==1 or ZF==1 --> rc = -1 */
1096                         "; ja 1f ; ud2 ; 1:\n"
1097                         : : "a" (&operand), "c" (ext) : "cc", "memory");
1098 }
1099
1100 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1101 {
1102         int i;
1103
1104         i = __find_msr_index(vmx, msr);
1105         if (i >= 0)
1106                 return &vmx->guest_msrs[i];
1107         return NULL;
1108 }
1109
1110 static void vmcs_clear(struct vmcs *vmcs)
1111 {
1112         u64 phys_addr = __pa(vmcs);
1113         u8 error;
1114
1115         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1116                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1117                       : "cc", "memory");
1118         if (error)
1119                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1120                        vmcs, phys_addr);
1121 }
1122
1123 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1124 {
1125         vmcs_clear(loaded_vmcs->vmcs);
1126         loaded_vmcs->cpu = -1;
1127         loaded_vmcs->launched = 0;
1128 }
1129
1130 static void vmcs_load(struct vmcs *vmcs)
1131 {
1132         u64 phys_addr = __pa(vmcs);
1133         u8 error;
1134
1135         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1136                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1137                         : "cc", "memory");
1138         if (error)
1139                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1140                        vmcs, phys_addr);
1141 }
1142
1143 #ifdef CONFIG_KEXEC
1144 /*
1145  * This bitmap is used to indicate whether the vmclear
1146  * operation is enabled on all cpus. All disabled by
1147  * default.
1148  */
1149 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1150
1151 static inline void crash_enable_local_vmclear(int cpu)
1152 {
1153         cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1154 }
1155
1156 static inline void crash_disable_local_vmclear(int cpu)
1157 {
1158         cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1159 }
1160
1161 static inline int crash_local_vmclear_enabled(int cpu)
1162 {
1163         return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1164 }
1165
1166 static void crash_vmclear_local_loaded_vmcss(void)
1167 {
1168         int cpu = raw_smp_processor_id();
1169         struct loaded_vmcs *v;
1170
1171         if (!crash_local_vmclear_enabled(cpu))
1172                 return;
1173
1174         list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1175                             loaded_vmcss_on_cpu_link)
1176                 vmcs_clear(v->vmcs);
1177 }
1178 #else
1179 static inline void crash_enable_local_vmclear(int cpu) { }
1180 static inline void crash_disable_local_vmclear(int cpu) { }
1181 #endif /* CONFIG_KEXEC */
1182
1183 static void __loaded_vmcs_clear(void *arg)
1184 {
1185         struct loaded_vmcs *loaded_vmcs = arg;
1186         int cpu = raw_smp_processor_id();
1187
1188         if (loaded_vmcs->cpu != cpu)
1189                 return; /* vcpu migration can race with cpu offline */
1190         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1191                 per_cpu(current_vmcs, cpu) = NULL;
1192         crash_disable_local_vmclear(cpu);
1193         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1194
1195         /*
1196          * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1197          * is before setting loaded_vmcs->vcpu to -1 which is done in
1198          * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1199          * then adds the vmcs into percpu list before it is deleted.
1200          */
1201         smp_wmb();
1202
1203         loaded_vmcs_init(loaded_vmcs);
1204         crash_enable_local_vmclear(cpu);
1205 }
1206
1207 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1208 {
1209         int cpu = loaded_vmcs->cpu;
1210
1211         if (cpu != -1)
1212                 smp_call_function_single(cpu,
1213                          __loaded_vmcs_clear, loaded_vmcs, 1);
1214 }
1215
1216 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1217 {
1218         if (vmx->vpid == 0)
1219                 return;
1220
1221         if (cpu_has_vmx_invvpid_single())
1222                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1223 }
1224
1225 static inline void vpid_sync_vcpu_global(void)
1226 {
1227         if (cpu_has_vmx_invvpid_global())
1228                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1229 }
1230
1231 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1232 {
1233         if (cpu_has_vmx_invvpid_single())
1234                 vpid_sync_vcpu_single(vmx);
1235         else
1236                 vpid_sync_vcpu_global();
1237 }
1238
1239 static inline void ept_sync_global(void)
1240 {
1241         if (cpu_has_vmx_invept_global())
1242                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1243 }
1244
1245 static inline void ept_sync_context(u64 eptp)
1246 {
1247         if (enable_ept) {
1248                 if (cpu_has_vmx_invept_context())
1249                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1250                 else
1251                         ept_sync_global();
1252         }
1253 }
1254
1255 static __always_inline unsigned long vmcs_readl(unsigned long field)
1256 {
1257         unsigned long value;
1258
1259         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1260                       : "=a"(value) : "d"(field) : "cc");
1261         return value;
1262 }
1263
1264 static __always_inline u16 vmcs_read16(unsigned long field)
1265 {
1266         return vmcs_readl(field);
1267 }
1268
1269 static __always_inline u32 vmcs_read32(unsigned long field)
1270 {
1271         return vmcs_readl(field);
1272 }
1273
1274 static __always_inline u64 vmcs_read64(unsigned long field)
1275 {
1276 #ifdef CONFIG_X86_64
1277         return vmcs_readl(field);
1278 #else
1279         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1280 #endif
1281 }
1282
1283 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1284 {
1285         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1286                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1287         dump_stack();
1288 }
1289
1290 static void vmcs_writel(unsigned long field, unsigned long value)
1291 {
1292         u8 error;
1293
1294         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1295                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1296         if (unlikely(error))
1297                 vmwrite_error(field, value);
1298 }
1299
1300 static void vmcs_write16(unsigned long field, u16 value)
1301 {
1302         vmcs_writel(field, value);
1303 }
1304
1305 static void vmcs_write32(unsigned long field, u32 value)
1306 {
1307         vmcs_writel(field, value);
1308 }
1309
1310 static void vmcs_write64(unsigned long field, u64 value)
1311 {
1312         vmcs_writel(field, value);
1313 #ifndef CONFIG_X86_64
1314         asm volatile ("");
1315         vmcs_writel(field+1, value >> 32);
1316 #endif
1317 }
1318
1319 static void vmcs_clear_bits(unsigned long field, u32 mask)
1320 {
1321         vmcs_writel(field, vmcs_readl(field) & ~mask);
1322 }
1323
1324 static void vmcs_set_bits(unsigned long field, u32 mask)
1325 {
1326         vmcs_writel(field, vmcs_readl(field) | mask);
1327 }
1328
1329 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1330 {
1331         vmx->segment_cache.bitmask = 0;
1332 }
1333
1334 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1335                                        unsigned field)
1336 {
1337         bool ret;
1338         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1339
1340         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1341                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1342                 vmx->segment_cache.bitmask = 0;
1343         }
1344         ret = vmx->segment_cache.bitmask & mask;
1345         vmx->segment_cache.bitmask |= mask;
1346         return ret;
1347 }
1348
1349 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1350 {
1351         u16 *p = &vmx->segment_cache.seg[seg].selector;
1352
1353         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1354                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1355         return *p;
1356 }
1357
1358 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1359 {
1360         ulong *p = &vmx->segment_cache.seg[seg].base;
1361
1362         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1363                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1364         return *p;
1365 }
1366
1367 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1368 {
1369         u32 *p = &vmx->segment_cache.seg[seg].limit;
1370
1371         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1372                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1373         return *p;
1374 }
1375
1376 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1377 {
1378         u32 *p = &vmx->segment_cache.seg[seg].ar;
1379
1380         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1381                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1382         return *p;
1383 }
1384
1385 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1386 {
1387         u32 eb;
1388
1389         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1390              (1u << NM_VECTOR) | (1u << DB_VECTOR);
1391         if ((vcpu->guest_debug &
1392              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1393             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1394                 eb |= 1u << BP_VECTOR;
1395         if (to_vmx(vcpu)->rmode.vm86_active)
1396                 eb = ~0;
1397         if (enable_ept)
1398                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1399         if (vcpu->fpu_active)
1400                 eb &= ~(1u << NM_VECTOR);
1401
1402         /* When we are running a nested L2 guest and L1 specified for it a
1403          * certain exception bitmap, we must trap the same exceptions and pass
1404          * them to L1. When running L2, we will only handle the exceptions
1405          * specified above if L1 did not want them.
1406          */
1407         if (is_guest_mode(vcpu))
1408                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1409
1410         vmcs_write32(EXCEPTION_BITMAP, eb);
1411 }
1412
1413 static void clear_atomic_switch_msr_special(unsigned long entry,
1414                 unsigned long exit)
1415 {
1416         vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1417         vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1418 }
1419
1420 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1421 {
1422         unsigned i;
1423         struct msr_autoload *m = &vmx->msr_autoload;
1424
1425         switch (msr) {
1426         case MSR_EFER:
1427                 if (cpu_has_load_ia32_efer) {
1428                         clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1429                                         VM_EXIT_LOAD_IA32_EFER);
1430                         return;
1431                 }
1432                 break;
1433         case MSR_CORE_PERF_GLOBAL_CTRL:
1434                 if (cpu_has_load_perf_global_ctrl) {
1435                         clear_atomic_switch_msr_special(
1436                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1437                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1438                         return;
1439                 }
1440                 break;
1441         }
1442
1443         for (i = 0; i < m->nr; ++i)
1444                 if (m->guest[i].index == msr)
1445                         break;
1446
1447         if (i == m->nr)
1448                 return;
1449         --m->nr;
1450         m->guest[i] = m->guest[m->nr];
1451         m->host[i] = m->host[m->nr];
1452         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1453         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1454 }
1455
1456 static void add_atomic_switch_msr_special(unsigned long entry,
1457                 unsigned long exit, unsigned long guest_val_vmcs,
1458                 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1459 {
1460         vmcs_write64(guest_val_vmcs, guest_val);
1461         vmcs_write64(host_val_vmcs, host_val);
1462         vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1463         vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1464 }
1465
1466 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1467                                   u64 guest_val, u64 host_val)
1468 {
1469         unsigned i;
1470         struct msr_autoload *m = &vmx->msr_autoload;
1471
1472         switch (msr) {
1473         case MSR_EFER:
1474                 if (cpu_has_load_ia32_efer) {
1475                         add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1476                                         VM_EXIT_LOAD_IA32_EFER,
1477                                         GUEST_IA32_EFER,
1478                                         HOST_IA32_EFER,
1479                                         guest_val, host_val);
1480                         return;
1481                 }
1482                 break;
1483         case MSR_CORE_PERF_GLOBAL_CTRL:
1484                 if (cpu_has_load_perf_global_ctrl) {
1485                         add_atomic_switch_msr_special(
1486                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1487                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1488                                         GUEST_IA32_PERF_GLOBAL_CTRL,
1489                                         HOST_IA32_PERF_GLOBAL_CTRL,
1490                                         guest_val, host_val);
1491                         return;
1492                 }
1493                 break;
1494         }
1495
1496         for (i = 0; i < m->nr; ++i)
1497                 if (m->guest[i].index == msr)
1498                         break;
1499
1500         if (i == NR_AUTOLOAD_MSRS) {
1501                 printk_once(KERN_WARNING"Not enough mst switch entries. "
1502                                 "Can't add msr %x\n", msr);
1503                 return;
1504         } else if (i == m->nr) {
1505                 ++m->nr;
1506                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1507                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1508         }
1509
1510         m->guest[i].index = msr;
1511         m->guest[i].value = guest_val;
1512         m->host[i].index = msr;
1513         m->host[i].value = host_val;
1514 }
1515
1516 static void reload_tss(void)
1517 {
1518         /*
1519          * VT restores TR but not its size.  Useless.
1520          */
1521         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1522         struct desc_struct *descs;
1523
1524         descs = (void *)gdt->address;
1525         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1526         load_TR_desc();
1527 }
1528
1529 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1530 {
1531         u64 guest_efer;
1532         u64 ignore_bits;
1533
1534         guest_efer = vmx->vcpu.arch.efer;
1535
1536         /*
1537          * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1538          * outside long mode
1539          */
1540         ignore_bits = EFER_NX | EFER_SCE;
1541 #ifdef CONFIG_X86_64
1542         ignore_bits |= EFER_LMA | EFER_LME;
1543         /* SCE is meaningful only in long mode on Intel */
1544         if (guest_efer & EFER_LMA)
1545                 ignore_bits &= ~(u64)EFER_SCE;
1546 #endif
1547         guest_efer &= ~ignore_bits;
1548         guest_efer |= host_efer & ignore_bits;
1549         vmx->guest_msrs[efer_offset].data = guest_efer;
1550         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1551
1552         clear_atomic_switch_msr(vmx, MSR_EFER);
1553         /* On ept, can't emulate nx, and must switch nx atomically */
1554         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1555                 guest_efer = vmx->vcpu.arch.efer;
1556                 if (!(guest_efer & EFER_LMA))
1557                         guest_efer &= ~EFER_LME;
1558                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1559                 return false;
1560         }
1561
1562         return true;
1563 }
1564
1565 static unsigned long segment_base(u16 selector)
1566 {
1567         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1568         struct desc_struct *d;
1569         unsigned long table_base;
1570         unsigned long v;
1571
1572         if (!(selector & ~3))
1573                 return 0;
1574
1575         table_base = gdt->address;
1576
1577         if (selector & 4) {           /* from ldt */
1578                 u16 ldt_selector = kvm_read_ldt();
1579
1580                 if (!(ldt_selector & ~3))
1581                         return 0;
1582
1583                 table_base = segment_base(ldt_selector);
1584         }
1585         d = (struct desc_struct *)(table_base + (selector & ~7));
1586         v = get_desc_base(d);
1587 #ifdef CONFIG_X86_64
1588        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1589                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1590 #endif
1591         return v;
1592 }
1593
1594 static inline unsigned long kvm_read_tr_base(void)
1595 {
1596         u16 tr;
1597         asm("str %0" : "=g"(tr));
1598         return segment_base(tr);
1599 }
1600
1601 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1602 {
1603         struct vcpu_vmx *vmx = to_vmx(vcpu);
1604         int i;
1605
1606         if (vmx->host_state.loaded)
1607                 return;
1608
1609         vmx->host_state.loaded = 1;
1610         /*
1611          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1612          * allow segment selectors with cpl > 0 or ti == 1.
1613          */
1614         vmx->host_state.ldt_sel = kvm_read_ldt();
1615         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1616         savesegment(fs, vmx->host_state.fs_sel);
1617         if (!(vmx->host_state.fs_sel & 7)) {
1618                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1619                 vmx->host_state.fs_reload_needed = 0;
1620         } else {
1621                 vmcs_write16(HOST_FS_SELECTOR, 0);
1622                 vmx->host_state.fs_reload_needed = 1;
1623         }
1624         savesegment(gs, vmx->host_state.gs_sel);
1625         if (!(vmx->host_state.gs_sel & 7))
1626                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1627         else {
1628                 vmcs_write16(HOST_GS_SELECTOR, 0);
1629                 vmx->host_state.gs_ldt_reload_needed = 1;
1630         }
1631
1632 #ifdef CONFIG_X86_64
1633         savesegment(ds, vmx->host_state.ds_sel);
1634         savesegment(es, vmx->host_state.es_sel);
1635 #endif
1636
1637 #ifdef CONFIG_X86_64
1638         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1639         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1640 #else
1641         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1642         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1643 #endif
1644
1645 #ifdef CONFIG_X86_64
1646         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1647         if (is_long_mode(&vmx->vcpu))
1648                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1649 #endif
1650         for (i = 0; i < vmx->save_nmsrs; ++i)
1651                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1652                                    vmx->guest_msrs[i].data,
1653                                    vmx->guest_msrs[i].mask);
1654 }
1655
1656 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1657 {
1658         if (!vmx->host_state.loaded)
1659                 return;
1660
1661         ++vmx->vcpu.stat.host_state_reload;
1662         vmx->host_state.loaded = 0;
1663 #ifdef CONFIG_X86_64
1664         if (is_long_mode(&vmx->vcpu))
1665                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1666 #endif
1667         if (vmx->host_state.gs_ldt_reload_needed) {
1668                 kvm_load_ldt(vmx->host_state.ldt_sel);
1669 #ifdef CONFIG_X86_64
1670                 load_gs_index(vmx->host_state.gs_sel);
1671 #else
1672                 loadsegment(gs, vmx->host_state.gs_sel);
1673 #endif
1674         }
1675         if (vmx->host_state.fs_reload_needed)
1676                 loadsegment(fs, vmx->host_state.fs_sel);
1677 #ifdef CONFIG_X86_64
1678         if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1679                 loadsegment(ds, vmx->host_state.ds_sel);
1680                 loadsegment(es, vmx->host_state.es_sel);
1681         }
1682 #endif
1683         reload_tss();
1684 #ifdef CONFIG_X86_64
1685         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1686 #endif
1687         /*
1688          * If the FPU is not active (through the host task or
1689          * the guest vcpu), then restore the cr0.TS bit.
1690          */
1691         if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1692                 stts();
1693         load_gdt(&__get_cpu_var(host_gdt));
1694 }
1695
1696 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1697 {
1698         preempt_disable();
1699         __vmx_load_host_state(vmx);
1700         preempt_enable();
1701 }
1702
1703 /*
1704  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1705  * vcpu mutex is already taken.
1706  */
1707 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1708 {
1709         struct vcpu_vmx *vmx = to_vmx(vcpu);
1710         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1711
1712         if (!vmm_exclusive)
1713                 kvm_cpu_vmxon(phys_addr);
1714         else if (vmx->loaded_vmcs->cpu != cpu)
1715                 loaded_vmcs_clear(vmx->loaded_vmcs);
1716
1717         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1718                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1719                 vmcs_load(vmx->loaded_vmcs->vmcs);
1720         }
1721
1722         if (vmx->loaded_vmcs->cpu != cpu) {
1723                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1724                 unsigned long sysenter_esp;
1725
1726                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1727                 local_irq_disable();
1728                 crash_disable_local_vmclear(cpu);
1729
1730                 /*
1731                  * Read loaded_vmcs->cpu should be before fetching
1732                  * loaded_vmcs->loaded_vmcss_on_cpu_link.
1733                  * See the comments in __loaded_vmcs_clear().
1734                  */
1735                 smp_rmb();
1736
1737                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1738                          &per_cpu(loaded_vmcss_on_cpu, cpu));
1739                 crash_enable_local_vmclear(cpu);
1740                 local_irq_enable();
1741
1742                 /*
1743                  * Linux uses per-cpu TSS and GDT, so set these when switching
1744                  * processors.
1745                  */
1746                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1747                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
1748
1749                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1750                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1751                 vmx->loaded_vmcs->cpu = cpu;
1752         }
1753 }
1754
1755 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1756 {
1757         __vmx_load_host_state(to_vmx(vcpu));
1758         if (!vmm_exclusive) {
1759                 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1760                 vcpu->cpu = -1;
1761                 kvm_cpu_vmxoff();
1762         }
1763 }
1764
1765 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1766 {
1767         ulong cr0;
1768
1769         if (vcpu->fpu_active)
1770                 return;
1771         vcpu->fpu_active = 1;
1772         cr0 = vmcs_readl(GUEST_CR0);
1773         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1774         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1775         vmcs_writel(GUEST_CR0, cr0);
1776         update_exception_bitmap(vcpu);
1777         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1778         if (is_guest_mode(vcpu))
1779                 vcpu->arch.cr0_guest_owned_bits &=
1780                         ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1781         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1782 }
1783
1784 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1785
1786 /*
1787  * Return the cr0 value that a nested guest would read. This is a combination
1788  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1789  * its hypervisor (cr0_read_shadow).
1790  */
1791 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1792 {
1793         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1794                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1795 }
1796 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1797 {
1798         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1799                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1800 }
1801
1802 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1803 {
1804         /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1805          * set this *before* calling this function.
1806          */
1807         vmx_decache_cr0_guest_bits(vcpu);
1808         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1809         update_exception_bitmap(vcpu);
1810         vcpu->arch.cr0_guest_owned_bits = 0;
1811         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1812         if (is_guest_mode(vcpu)) {
1813                 /*
1814                  * L1's specified read shadow might not contain the TS bit,
1815                  * so now that we turned on shadowing of this bit, we need to
1816                  * set this bit of the shadow. Like in nested_vmx_run we need
1817                  * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1818                  * up-to-date here because we just decached cr0.TS (and we'll
1819                  * only update vmcs12->guest_cr0 on nested exit).
1820                  */
1821                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1822                 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1823                         (vcpu->arch.cr0 & X86_CR0_TS);
1824                 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1825         } else
1826                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1827 }
1828
1829 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1830 {
1831         unsigned long rflags, save_rflags;
1832
1833         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1834                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1835                 rflags = vmcs_readl(GUEST_RFLAGS);
1836                 if (to_vmx(vcpu)->rmode.vm86_active) {
1837                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1838                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1839                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1840                 }
1841                 to_vmx(vcpu)->rflags = rflags;
1842         }
1843         return to_vmx(vcpu)->rflags;
1844 }
1845
1846 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1847 {
1848         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1849         to_vmx(vcpu)->rflags = rflags;
1850         if (to_vmx(vcpu)->rmode.vm86_active) {
1851                 to_vmx(vcpu)->rmode.save_rflags = rflags;
1852                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1853         }
1854         vmcs_writel(GUEST_RFLAGS, rflags);
1855 }
1856
1857 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1858 {
1859         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1860         int ret = 0;
1861
1862         if (interruptibility & GUEST_INTR_STATE_STI)
1863                 ret |= KVM_X86_SHADOW_INT_STI;
1864         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1865                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1866
1867         return ret & mask;
1868 }
1869
1870 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1871 {
1872         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1873         u32 interruptibility = interruptibility_old;
1874
1875         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1876
1877         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1878                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1879         else if (mask & KVM_X86_SHADOW_INT_STI)
1880                 interruptibility |= GUEST_INTR_STATE_STI;
1881
1882         if ((interruptibility != interruptibility_old))
1883                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1884 }
1885
1886 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1887 {
1888         unsigned long rip;
1889
1890         rip = kvm_rip_read(vcpu);
1891         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1892         kvm_rip_write(vcpu, rip);
1893
1894         /* skipping an emulated instruction also counts */
1895         vmx_set_interrupt_shadow(vcpu, 0);
1896 }
1897
1898 /*
1899  * KVM wants to inject page-faults which it got to the guest. This function
1900  * checks whether in a nested guest, we need to inject them to L1 or L2.
1901  * This function assumes it is called with the exit reason in vmcs02 being
1902  * a #PF exception (this is the only case in which KVM injects a #PF when L2
1903  * is running).
1904  */
1905 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1906 {
1907         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1908
1909         /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1910         if (!(vmcs12->exception_bitmap & (1u << PF_VECTOR)))
1911                 return 0;
1912
1913         nested_vmx_vmexit(vcpu);
1914         return 1;
1915 }
1916
1917 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1918                                 bool has_error_code, u32 error_code,
1919                                 bool reinject)
1920 {
1921         struct vcpu_vmx *vmx = to_vmx(vcpu);
1922         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1923
1924         if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1925             !vmx->nested.nested_run_pending && nested_pf_handled(vcpu))
1926                 return;
1927
1928         if (has_error_code) {
1929                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1930                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1931         }
1932
1933         if (vmx->rmode.vm86_active) {
1934                 int inc_eip = 0;
1935                 if (kvm_exception_is_soft(nr))
1936                         inc_eip = vcpu->arch.event_exit_inst_len;
1937                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1938                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1939                 return;
1940         }
1941
1942         if (kvm_exception_is_soft(nr)) {
1943                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1944                              vmx->vcpu.arch.event_exit_inst_len);
1945                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1946         } else
1947                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1948
1949         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1950 }
1951
1952 static bool vmx_rdtscp_supported(void)
1953 {
1954         return cpu_has_vmx_rdtscp();
1955 }
1956
1957 static bool vmx_invpcid_supported(void)
1958 {
1959         return cpu_has_vmx_invpcid() && enable_ept;
1960 }
1961
1962 /*
1963  * Swap MSR entry in host/guest MSR entry array.
1964  */
1965 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1966 {
1967         struct shared_msr_entry tmp;
1968
1969         tmp = vmx->guest_msrs[to];
1970         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1971         vmx->guest_msrs[from] = tmp;
1972 }
1973
1974 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
1975 {
1976         unsigned long *msr_bitmap;
1977
1978         if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
1979                 if (is_long_mode(vcpu))
1980                         msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
1981                 else
1982                         msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
1983         } else {
1984                 if (is_long_mode(vcpu))
1985                         msr_bitmap = vmx_msr_bitmap_longmode;
1986                 else
1987                         msr_bitmap = vmx_msr_bitmap_legacy;
1988         }
1989
1990         vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1991 }
1992
1993 /*
1994  * Set up the vmcs to automatically save and restore system
1995  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1996  * mode, as fiddling with msrs is very expensive.
1997  */
1998 static void setup_msrs(struct vcpu_vmx *vmx)
1999 {
2000         int save_nmsrs, index;
2001
2002         save_nmsrs = 0;
2003 #ifdef CONFIG_X86_64
2004         if (is_long_mode(&vmx->vcpu)) {
2005                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2006                 if (index >= 0)
2007                         move_msr_up(vmx, index, save_nmsrs++);
2008                 index = __find_msr_index(vmx, MSR_LSTAR);
2009                 if (index >= 0)
2010                         move_msr_up(vmx, index, save_nmsrs++);
2011                 index = __find_msr_index(vmx, MSR_CSTAR);
2012                 if (index >= 0)
2013                         move_msr_up(vmx, index, save_nmsrs++);
2014                 index = __find_msr_index(vmx, MSR_TSC_AUX);
2015                 if (index >= 0 && vmx->rdtscp_enabled)
2016                         move_msr_up(vmx, index, save_nmsrs++);
2017                 /*
2018                  * MSR_STAR is only needed on long mode guests, and only
2019                  * if efer.sce is enabled.
2020                  */
2021                 index = __find_msr_index(vmx, MSR_STAR);
2022                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2023                         move_msr_up(vmx, index, save_nmsrs++);
2024         }
2025 #endif
2026         index = __find_msr_index(vmx, MSR_EFER);
2027         if (index >= 0 && update_transition_efer(vmx, index))
2028                 move_msr_up(vmx, index, save_nmsrs++);
2029
2030         vmx->save_nmsrs = save_nmsrs;
2031
2032         if (cpu_has_vmx_msr_bitmap())
2033                 vmx_set_msr_bitmap(&vmx->vcpu);
2034 }
2035
2036 /*
2037  * reads and returns guest's timestamp counter "register"
2038  * guest_tsc = host_tsc + tsc_offset    -- 21.3
2039  */
2040 static u64 guest_read_tsc(void)
2041 {
2042         u64 host_tsc, tsc_offset;
2043
2044         rdtscll(host_tsc);
2045         tsc_offset = vmcs_read64(TSC_OFFSET);
2046         return host_tsc + tsc_offset;
2047 }
2048
2049 /*
2050  * Like guest_read_tsc, but always returns L1's notion of the timestamp
2051  * counter, even if a nested guest (L2) is currently running.
2052  */
2053 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2054 {
2055         u64 tsc_offset;
2056
2057         tsc_offset = is_guest_mode(vcpu) ?
2058                 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2059                 vmcs_read64(TSC_OFFSET);
2060         return host_tsc + tsc_offset;
2061 }
2062
2063 /*
2064  * Engage any workarounds for mis-matched TSC rates.  Currently limited to
2065  * software catchup for faster rates on slower CPUs.
2066  */
2067 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2068 {
2069         if (!scale)
2070                 return;
2071
2072         if (user_tsc_khz > tsc_khz) {
2073                 vcpu->arch.tsc_catchup = 1;
2074                 vcpu->arch.tsc_always_catchup = 1;
2075         } else
2076                 WARN(1, "user requested TSC rate below hardware speed\n");
2077 }
2078
2079 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2080 {
2081         return vmcs_read64(TSC_OFFSET);
2082 }
2083
2084 /*
2085  * writes 'offset' into guest's timestamp counter offset register
2086  */
2087 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2088 {
2089         if (is_guest_mode(vcpu)) {
2090                 /*
2091                  * We're here if L1 chose not to trap WRMSR to TSC. According
2092                  * to the spec, this should set L1's TSC; The offset that L1
2093                  * set for L2 remains unchanged, and still needs to be added
2094                  * to the newly set TSC to get L2's TSC.
2095                  */
2096                 struct vmcs12 *vmcs12;
2097                 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2098                 /* recalculate vmcs02.TSC_OFFSET: */
2099                 vmcs12 = get_vmcs12(vcpu);
2100                 vmcs_write64(TSC_OFFSET, offset +
2101                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2102                          vmcs12->tsc_offset : 0));
2103         } else {
2104                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2105                                            vmcs_read64(TSC_OFFSET), offset);
2106                 vmcs_write64(TSC_OFFSET, offset);
2107         }
2108 }
2109
2110 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2111 {
2112         u64 offset = vmcs_read64(TSC_OFFSET);
2113
2114         vmcs_write64(TSC_OFFSET, offset + adjustment);
2115         if (is_guest_mode(vcpu)) {
2116                 /* Even when running L2, the adjustment needs to apply to L1 */
2117                 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2118         } else
2119                 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2120                                            offset + adjustment);
2121 }
2122
2123 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2124 {
2125         return target_tsc - native_read_tsc();
2126 }
2127
2128 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2129 {
2130         struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2131         return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2132 }
2133
2134 /*
2135  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2136  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2137  * all guests if the "nested" module option is off, and can also be disabled
2138  * for a single guest by disabling its VMX cpuid bit.
2139  */
2140 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2141 {
2142         return nested && guest_cpuid_has_vmx(vcpu);
2143 }
2144
2145 /*
2146  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2147  * returned for the various VMX controls MSRs when nested VMX is enabled.
2148  * The same values should also be used to verify that vmcs12 control fields are
2149  * valid during nested entry from L1 to L2.
2150  * Each of these control msrs has a low and high 32-bit half: A low bit is on
2151  * if the corresponding bit in the (32-bit) control field *must* be on, and a
2152  * bit in the high half is on if the corresponding bit in the control field
2153  * may be on. See also vmx_control_verify().
2154  * TODO: allow these variables to be modified (downgraded) by module options
2155  * or other means.
2156  */
2157 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
2158 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
2159 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
2160 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
2161 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
2162 static u32 nested_vmx_misc_low, nested_vmx_misc_high;
2163 static u32 nested_vmx_ept_caps;
2164 static __init void nested_vmx_setup_ctls_msrs(void)
2165 {
2166         /*
2167          * Note that as a general rule, the high half of the MSRs (bits in
2168          * the control fields which may be 1) should be initialized by the
2169          * intersection of the underlying hardware's MSR (i.e., features which
2170          * can be supported) and the list of features we want to expose -
2171          * because they are known to be properly supported in our code.
2172          * Also, usually, the low half of the MSRs (bits which must be 1) can
2173          * be set to 0, meaning that L1 may turn off any of these bits. The
2174          * reason is that if one of these bits is necessary, it will appear
2175          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2176          * fields of vmcs01 and vmcs02, will turn these bits off - and
2177          * nested_vmx_exit_handled() will not pass related exits to L1.
2178          * These rules have exceptions below.
2179          */
2180
2181         /* pin-based controls */
2182         rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2183               nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high);
2184         /*
2185          * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
2186          * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
2187          */
2188         nested_vmx_pinbased_ctls_low |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2189         nested_vmx_pinbased_ctls_high &= PIN_BASED_EXT_INTR_MASK |
2190                 PIN_BASED_NMI_EXITING | PIN_BASED_VIRTUAL_NMIS |
2191                 PIN_BASED_VMX_PREEMPTION_TIMER;
2192         nested_vmx_pinbased_ctls_high |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2193
2194         /*
2195          * Exit controls
2196          * If bit 55 of VMX_BASIC is off, bits 0-8 and 10, 11, 13, 14, 16 and
2197          * 17 must be 1.
2198          */
2199         rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2200                 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high);
2201         nested_vmx_exit_ctls_low = VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2202         /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
2203         nested_vmx_exit_ctls_high &=
2204 #ifdef CONFIG_X86_64
2205                 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2206 #endif
2207                 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2208         nested_vmx_exit_ctls_high |= (VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2209                                       VM_EXIT_LOAD_IA32_EFER);
2210
2211         /* entry controls */
2212         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2213                 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
2214         /* If bit 55 of VMX_BASIC is off, bits 0-8 and 12 must be 1. */
2215         nested_vmx_entry_ctls_low = VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2216         nested_vmx_entry_ctls_high &=
2217 #ifdef CONFIG_X86_64
2218                 VM_ENTRY_IA32E_MODE |
2219 #endif
2220                 VM_ENTRY_LOAD_IA32_PAT;
2221         nested_vmx_entry_ctls_high |= (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR |
2222                                        VM_ENTRY_LOAD_IA32_EFER);
2223
2224         /* cpu-based controls */
2225         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2226                 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
2227         nested_vmx_procbased_ctls_low = 0;
2228         nested_vmx_procbased_ctls_high &=
2229                 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2230                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2231                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2232                 CPU_BASED_CR3_STORE_EXITING |
2233 #ifdef CONFIG_X86_64
2234                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2235 #endif
2236                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2237                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2238                 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2239                 CPU_BASED_PAUSE_EXITING |
2240                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2241         /*
2242          * We can allow some features even when not supported by the
2243          * hardware. For example, L1 can specify an MSR bitmap - and we
2244          * can use it to avoid exits to L1 - even when L0 runs L2
2245          * without MSR bitmaps.
2246          */
2247         nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
2248
2249         /* secondary cpu-based controls */
2250         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2251                 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2252         nested_vmx_secondary_ctls_low = 0;
2253         nested_vmx_secondary_ctls_high &=
2254                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2255                 SECONDARY_EXEC_WBINVD_EXITING;
2256
2257         if (enable_ept) {
2258                 /* nested EPT: emulate EPT also to L1 */
2259                 nested_vmx_secondary_ctls_high |= SECONDARY_EXEC_ENABLE_EPT;
2260                 nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2261                          VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2262                 nested_vmx_ept_caps &= vmx_capability.ept;
2263                 /*
2264                  * Since invept is completely emulated we support both global
2265                  * and context invalidation independent of what host cpu
2266                  * supports
2267                  */
2268                 nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2269                         VMX_EPT_EXTENT_CONTEXT_BIT;
2270         } else
2271                 nested_vmx_ept_caps = 0;
2272
2273         /* miscellaneous data */
2274         rdmsr(MSR_IA32_VMX_MISC, nested_vmx_misc_low, nested_vmx_misc_high);
2275         nested_vmx_misc_low &= VMX_MISC_PREEMPTION_TIMER_RATE_MASK |
2276                 VMX_MISC_SAVE_EFER_LMA;
2277         nested_vmx_misc_high = 0;
2278 }
2279
2280 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2281 {
2282         /*
2283          * Bits 0 in high must be 0, and bits 1 in low must be 1.
2284          */
2285         return ((control & high) | low) == control;
2286 }
2287
2288 static inline u64 vmx_control_msr(u32 low, u32 high)
2289 {
2290         return low | ((u64)high << 32);
2291 }
2292
2293 /*
2294  * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
2295  * also let it use VMX-specific MSRs.
2296  * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
2297  * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
2298  * like all other MSRs).
2299  */
2300 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2301 {
2302         if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2303                      msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2304                 /*
2305                  * According to the spec, processors which do not support VMX
2306                  * should throw a #GP(0) when VMX capability MSRs are read.
2307                  */
2308                 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2309                 return 1;
2310         }
2311
2312         switch (msr_index) {
2313         case MSR_IA32_FEATURE_CONTROL:
2314                 if (nested_vmx_allowed(vcpu)) {
2315                         *pdata = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2316                         break;
2317                 }
2318                 return 0;
2319         case MSR_IA32_VMX_BASIC:
2320                 /*
2321                  * This MSR reports some information about VMX support. We
2322                  * should return information about the VMX we emulate for the
2323                  * guest, and the VMCS structure we give it - not about the
2324                  * VMX support of the underlying hardware.
2325                  */
2326                 *pdata = VMCS12_REVISION |
2327                            ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2328                            (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2329                 break;
2330         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2331         case MSR_IA32_VMX_PINBASED_CTLS:
2332                 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2333                                         nested_vmx_pinbased_ctls_high);
2334                 break;
2335         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2336         case MSR_IA32_VMX_PROCBASED_CTLS:
2337                 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2338                                         nested_vmx_procbased_ctls_high);
2339                 break;
2340         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2341         case MSR_IA32_VMX_EXIT_CTLS:
2342                 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2343                                         nested_vmx_exit_ctls_high);
2344                 break;
2345         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2346         case MSR_IA32_VMX_ENTRY_CTLS:
2347                 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2348                                         nested_vmx_entry_ctls_high);
2349                 break;
2350         case MSR_IA32_VMX_MISC:
2351                 *pdata = vmx_control_msr(nested_vmx_misc_low,
2352                                          nested_vmx_misc_high);
2353                 break;
2354         /*
2355          * These MSRs specify bits which the guest must keep fixed (on or off)
2356          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2357          * We picked the standard core2 setting.
2358          */
2359 #define VMXON_CR0_ALWAYSON      (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2360 #define VMXON_CR4_ALWAYSON      X86_CR4_VMXE
2361         case MSR_IA32_VMX_CR0_FIXED0:
2362                 *pdata = VMXON_CR0_ALWAYSON;
2363                 break;
2364         case MSR_IA32_VMX_CR0_FIXED1:
2365                 *pdata = -1ULL;
2366                 break;
2367         case MSR_IA32_VMX_CR4_FIXED0:
2368                 *pdata = VMXON_CR4_ALWAYSON;
2369                 break;
2370         case MSR_IA32_VMX_CR4_FIXED1:
2371                 *pdata = -1ULL;
2372                 break;
2373         case MSR_IA32_VMX_VMCS_ENUM:
2374                 *pdata = 0x1f;
2375                 break;
2376         case MSR_IA32_VMX_PROCBASED_CTLS2:
2377                 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2378                                         nested_vmx_secondary_ctls_high);
2379                 break;
2380         case MSR_IA32_VMX_EPT_VPID_CAP:
2381                 /* Currently, no nested vpid support */
2382                 *pdata = nested_vmx_ept_caps;
2383                 break;
2384         default:
2385                 return 0;
2386         }
2387
2388         return 1;
2389 }
2390
2391 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2392 {
2393         u32 msr_index = msr_info->index;
2394         u64 data = msr_info->data;
2395         bool host_initialized = msr_info->host_initiated;
2396
2397         if (!nested_vmx_allowed(vcpu))
2398                 return 0;
2399
2400         if (msr_index == MSR_IA32_FEATURE_CONTROL) {
2401                 if (!host_initialized &&
2402                                 to_vmx(vcpu)->nested.msr_ia32_feature_control
2403                                 & FEATURE_CONTROL_LOCKED)
2404                         return 0;
2405                 to_vmx(vcpu)->nested.msr_ia32_feature_control = data;
2406                 return 1;
2407         }
2408
2409         /*
2410          * No need to treat VMX capability MSRs specially: If we don't handle
2411          * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2412          */
2413         return 0;
2414 }
2415
2416 /*
2417  * Reads an msr value (of 'msr_index') into 'pdata'.
2418  * Returns 0 on success, non-0 otherwise.
2419  * Assumes vcpu_load() was already called.
2420  */
2421 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2422 {
2423         u64 data;
2424         struct shared_msr_entry *msr;
2425
2426         if (!pdata) {
2427                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2428                 return -EINVAL;
2429         }
2430
2431         switch (msr_index) {
2432 #ifdef CONFIG_X86_64
2433         case MSR_FS_BASE:
2434                 data = vmcs_readl(GUEST_FS_BASE);
2435                 break;
2436         case MSR_GS_BASE:
2437                 data = vmcs_readl(GUEST_GS_BASE);
2438                 break;
2439         case MSR_KERNEL_GS_BASE:
2440                 vmx_load_host_state(to_vmx(vcpu));
2441                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2442                 break;
2443 #endif
2444         case MSR_EFER:
2445                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2446         case MSR_IA32_TSC:
2447                 data = guest_read_tsc();
2448                 break;
2449         case MSR_IA32_SYSENTER_CS:
2450                 data = vmcs_read32(GUEST_SYSENTER_CS);
2451                 break;
2452         case MSR_IA32_SYSENTER_EIP:
2453                 data = vmcs_readl(GUEST_SYSENTER_EIP);
2454                 break;
2455         case MSR_IA32_SYSENTER_ESP:
2456                 data = vmcs_readl(GUEST_SYSENTER_ESP);
2457                 break;
2458         case MSR_TSC_AUX:
2459                 if (!to_vmx(vcpu)->rdtscp_enabled)
2460                         return 1;
2461                 /* Otherwise falls through */
2462         default:
2463                 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2464                         return 0;
2465                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2466                 if (msr) {
2467                         data = msr->data;
2468                         break;
2469                 }
2470                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2471         }
2472
2473         *pdata = data;
2474         return 0;
2475 }
2476
2477 /*
2478  * Writes msr value into into the appropriate "register".
2479  * Returns 0 on success, non-0 otherwise.
2480  * Assumes vcpu_load() was already called.
2481  */
2482 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2483 {
2484         struct vcpu_vmx *vmx = to_vmx(vcpu);
2485         struct shared_msr_entry *msr;
2486         int ret = 0;
2487         u32 msr_index = msr_info->index;
2488         u64 data = msr_info->data;
2489
2490         switch (msr_index) {
2491         case MSR_EFER:
2492                 ret = kvm_set_msr_common(vcpu, msr_info);
2493                 break;
2494 #ifdef CONFIG_X86_64
2495         case MSR_FS_BASE:
2496                 vmx_segment_cache_clear(vmx);
2497                 vmcs_writel(GUEST_FS_BASE, data);
2498                 break;
2499         case MSR_GS_BASE:
2500                 vmx_segment_cache_clear(vmx);
2501                 vmcs_writel(GUEST_GS_BASE, data);
2502                 break;
2503         case MSR_KERNEL_GS_BASE:
2504                 vmx_load_host_state(vmx);
2505                 vmx->msr_guest_kernel_gs_base = data;
2506                 break;
2507 #endif
2508         case MSR_IA32_SYSENTER_CS:
2509                 vmcs_write32(GUEST_SYSENTER_CS, data);
2510                 break;
2511         case MSR_IA32_SYSENTER_EIP:
2512                 vmcs_writel(GUEST_SYSENTER_EIP, data);
2513                 break;
2514         case MSR_IA32_SYSENTER_ESP:
2515                 vmcs_writel(GUEST_SYSENTER_ESP, data);
2516                 break;
2517         case MSR_IA32_TSC:
2518                 kvm_write_tsc(vcpu, msr_info);
2519                 break;
2520         case MSR_IA32_CR_PAT:
2521                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2522                         vmcs_write64(GUEST_IA32_PAT, data);
2523                         vcpu->arch.pat = data;
2524                         break;
2525                 }
2526                 ret = kvm_set_msr_common(vcpu, msr_info);
2527                 break;
2528         case MSR_IA32_TSC_ADJUST:
2529                 ret = kvm_set_msr_common(vcpu, msr_info);
2530                 break;
2531         case MSR_TSC_AUX:
2532                 if (!vmx->rdtscp_enabled)
2533                         return 1;
2534                 /* Check reserved bit, higher 32 bits should be zero */
2535                 if ((data >> 32) != 0)
2536                         return 1;
2537                 /* Otherwise falls through */
2538         default:
2539                 if (vmx_set_vmx_msr(vcpu, msr_info))
2540                         break;
2541                 msr = find_msr_entry(vmx, msr_index);
2542                 if (msr) {
2543                         msr->data = data;
2544                         if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2545                                 preempt_disable();
2546                                 kvm_set_shared_msr(msr->index, msr->data,
2547                                                    msr->mask);
2548                                 preempt_enable();
2549                         }
2550                         break;
2551                 }
2552                 ret = kvm_set_msr_common(vcpu, msr_info);
2553         }
2554
2555         return ret;
2556 }
2557
2558 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2559 {
2560         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2561         switch (reg) {
2562         case VCPU_REGS_RSP:
2563                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2564                 break;
2565         case VCPU_REGS_RIP:
2566                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2567                 break;
2568         case VCPU_EXREG_PDPTR:
2569                 if (enable_ept)
2570                         ept_save_pdptrs(vcpu);
2571                 break;
2572         default:
2573                 break;
2574         }
2575 }
2576
2577 static __init int cpu_has_kvm_support(void)
2578 {
2579         return cpu_has_vmx();
2580 }
2581
2582 static __init int vmx_disabled_by_bios(void)
2583 {
2584         u64 msr;
2585
2586         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2587         if (msr & FEATURE_CONTROL_LOCKED) {
2588                 /* launched w/ TXT and VMX disabled */
2589                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2590                         && tboot_enabled())
2591                         return 1;
2592                 /* launched w/o TXT and VMX only enabled w/ TXT */
2593                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2594                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2595                         && !tboot_enabled()) {
2596                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2597                                 "activate TXT before enabling KVM\n");
2598                         return 1;
2599                 }
2600                 /* launched w/o TXT and VMX disabled */
2601                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2602                         && !tboot_enabled())
2603                         return 1;
2604         }
2605
2606         return 0;
2607 }
2608
2609 static void kvm_cpu_vmxon(u64 addr)
2610 {
2611         asm volatile (ASM_VMX_VMXON_RAX
2612                         : : "a"(&addr), "m"(addr)
2613                         : "memory", "cc");
2614 }
2615
2616 static int hardware_enable(void *garbage)
2617 {
2618         int cpu = raw_smp_processor_id();
2619         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2620         u64 old, test_bits;
2621
2622         if (read_cr4() & X86_CR4_VMXE)
2623                 return -EBUSY;
2624
2625         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2626
2627         /*
2628          * Now we can enable the vmclear operation in kdump
2629          * since the loaded_vmcss_on_cpu list on this cpu
2630          * has been initialized.
2631          *
2632          * Though the cpu is not in VMX operation now, there
2633          * is no problem to enable the vmclear operation
2634          * for the loaded_vmcss_on_cpu list is empty!
2635          */
2636         crash_enable_local_vmclear(cpu);
2637
2638         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2639
2640         test_bits = FEATURE_CONTROL_LOCKED;
2641         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2642         if (tboot_enabled())
2643                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2644
2645         if ((old & test_bits) != test_bits) {
2646                 /* enable and lock */
2647                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2648         }
2649         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2650
2651         if (vmm_exclusive) {
2652                 kvm_cpu_vmxon(phys_addr);
2653                 ept_sync_global();
2654         }
2655
2656         native_store_gdt(&__get_cpu_var(host_gdt));
2657
2658         return 0;
2659 }
2660
2661 static void vmclear_local_loaded_vmcss(void)
2662 {
2663         int cpu = raw_smp_processor_id();
2664         struct loaded_vmcs *v, *n;
2665
2666         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2667                                  loaded_vmcss_on_cpu_link)
2668                 __loaded_vmcs_clear(v);
2669 }
2670
2671
2672 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2673  * tricks.
2674  */
2675 static void kvm_cpu_vmxoff(void)
2676 {
2677         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2678 }
2679
2680 static void hardware_disable(void *garbage)
2681 {
2682         if (vmm_exclusive) {
2683                 vmclear_local_loaded_vmcss();
2684                 kvm_cpu_vmxoff();
2685         }
2686         write_cr4(read_cr4() & ~X86_CR4_VMXE);
2687 }
2688
2689 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2690                                       u32 msr, u32 *result)
2691 {
2692         u32 vmx_msr_low, vmx_msr_high;
2693         u32 ctl = ctl_min | ctl_opt;
2694
2695         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2696
2697         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2698         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2699
2700         /* Ensure minimum (required) set of control bits are supported. */
2701         if (ctl_min & ~ctl)
2702                 return -EIO;
2703
2704         *result = ctl;
2705         return 0;
2706 }
2707
2708 static __init bool allow_1_setting(u32 msr, u32 ctl)
2709 {
2710         u32 vmx_msr_low, vmx_msr_high;
2711
2712         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2713         return vmx_msr_high & ctl;
2714 }
2715
2716 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2717 {
2718         u32 vmx_msr_low, vmx_msr_high;
2719         u32 min, opt, min2, opt2;
2720         u32 _pin_based_exec_control = 0;
2721         u32 _cpu_based_exec_control = 0;
2722         u32 _cpu_based_2nd_exec_control = 0;
2723         u32 _vmexit_control = 0;
2724         u32 _vmentry_control = 0;
2725
2726         min = CPU_BASED_HLT_EXITING |
2727 #ifdef CONFIG_X86_64
2728               CPU_BASED_CR8_LOAD_EXITING |
2729               CPU_BASED_CR8_STORE_EXITING |
2730 #endif
2731               CPU_BASED_CR3_LOAD_EXITING |
2732               CPU_BASED_CR3_STORE_EXITING |
2733               CPU_BASED_USE_IO_BITMAPS |
2734               CPU_BASED_MOV_DR_EXITING |
2735               CPU_BASED_USE_TSC_OFFSETING |
2736               CPU_BASED_MWAIT_EXITING |
2737               CPU_BASED_MONITOR_EXITING |
2738               CPU_BASED_INVLPG_EXITING |
2739               CPU_BASED_RDPMC_EXITING;
2740
2741         opt = CPU_BASED_TPR_SHADOW |
2742               CPU_BASED_USE_MSR_BITMAPS |
2743               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2744         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2745                                 &_cpu_based_exec_control) < 0)
2746                 return -EIO;
2747 #ifdef CONFIG_X86_64
2748         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2749                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2750                                            ~CPU_BASED_CR8_STORE_EXITING;
2751 #endif
2752         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2753                 min2 = 0;
2754                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2755                         SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2756                         SECONDARY_EXEC_WBINVD_EXITING |
2757                         SECONDARY_EXEC_ENABLE_VPID |
2758                         SECONDARY_EXEC_ENABLE_EPT |
2759                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
2760                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2761                         SECONDARY_EXEC_RDTSCP |
2762                         SECONDARY_EXEC_ENABLE_INVPCID |
2763                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
2764                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2765                         SECONDARY_EXEC_SHADOW_VMCS;
2766                 if (adjust_vmx_controls(min2, opt2,
2767                                         MSR_IA32_VMX_PROCBASED_CTLS2,
2768                                         &_cpu_based_2nd_exec_control) < 0)
2769                         return -EIO;
2770         }
2771 #ifndef CONFIG_X86_64
2772         if (!(_cpu_based_2nd_exec_control &
2773                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2774                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2775 #endif
2776
2777         if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2778                 _cpu_based_2nd_exec_control &= ~(
2779                                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2780                                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2781                                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2782
2783         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2784                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2785                    enabled */
2786                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2787                                              CPU_BASED_CR3_STORE_EXITING |
2788                                              CPU_BASED_INVLPG_EXITING);
2789                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2790                       vmx_capability.ept, vmx_capability.vpid);
2791         }
2792
2793         min = 0;
2794 #ifdef CONFIG_X86_64
2795         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2796 #endif
2797         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
2798                 VM_EXIT_ACK_INTR_ON_EXIT;
2799         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2800                                 &_vmexit_control) < 0)
2801                 return -EIO;
2802
2803         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2804         opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
2805         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2806                                 &_pin_based_exec_control) < 0)
2807                 return -EIO;
2808
2809         if (!(_cpu_based_2nd_exec_control &
2810                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
2811                 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
2812                 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2813
2814         min = 0;
2815         opt = VM_ENTRY_LOAD_IA32_PAT;
2816         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2817                                 &_vmentry_control) < 0)
2818                 return -EIO;
2819
2820         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2821
2822         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2823         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2824                 return -EIO;
2825
2826 #ifdef CONFIG_X86_64
2827         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2828         if (vmx_msr_high & (1u<<16))
2829                 return -EIO;
2830 #endif
2831
2832         /* Require Write-Back (WB) memory type for VMCS accesses. */
2833         if (((vmx_msr_high >> 18) & 15) != 6)
2834                 return -EIO;
2835
2836         vmcs_conf->size = vmx_msr_high & 0x1fff;
2837         vmcs_conf->order = get_order(vmcs_config.size);
2838         vmcs_conf->revision_id = vmx_msr_low;
2839
2840         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2841         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2842         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2843         vmcs_conf->vmexit_ctrl         = _vmexit_control;
2844         vmcs_conf->vmentry_ctrl        = _vmentry_control;
2845
2846         cpu_has_load_ia32_efer =
2847                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2848                                 VM_ENTRY_LOAD_IA32_EFER)
2849                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2850                                    VM_EXIT_LOAD_IA32_EFER);
2851
2852         cpu_has_load_perf_global_ctrl =
2853                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2854                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2855                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2856                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2857
2858         /*
2859          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2860          * but due to arrata below it can't be used. Workaround is to use
2861          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2862          *
2863          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2864          *
2865          * AAK155             (model 26)
2866          * AAP115             (model 30)
2867          * AAT100             (model 37)
2868          * BC86,AAY89,BD102   (model 44)
2869          * BA97               (model 46)
2870          *
2871          */
2872         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2873                 switch (boot_cpu_data.x86_model) {
2874                 case 26:
2875                 case 30:
2876                 case 37:
2877                 case 44:
2878                 case 46:
2879                         cpu_has_load_perf_global_ctrl = false;
2880                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2881                                         "does not work properly. Using workaround\n");
2882                         break;
2883                 default:
2884                         break;
2885                 }
2886         }
2887
2888         return 0;
2889 }
2890
2891 static struct vmcs *alloc_vmcs_cpu(int cpu)
2892 {
2893         int node = cpu_to_node(cpu);
2894         struct page *pages;
2895         struct vmcs *vmcs;
2896
2897         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2898         if (!pages)
2899                 return NULL;
2900         vmcs = page_address(pages);
2901         memset(vmcs, 0, vmcs_config.size);
2902         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2903         return vmcs;
2904 }
2905
2906 static struct vmcs *alloc_vmcs(void)
2907 {
2908         return alloc_vmcs_cpu(raw_smp_processor_id());
2909 }
2910
2911 static void free_vmcs(struct vmcs *vmcs)
2912 {
2913         free_pages((unsigned long)vmcs, vmcs_config.order);
2914 }
2915
2916 /*
2917  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2918  */
2919 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2920 {
2921         if (!loaded_vmcs->vmcs)
2922                 return;
2923         loaded_vmcs_clear(loaded_vmcs);
2924         free_vmcs(loaded_vmcs->vmcs);
2925         loaded_vmcs->vmcs = NULL;
2926 }
2927
2928 static void free_kvm_area(void)
2929 {
2930         int cpu;
2931
2932         for_each_possible_cpu(cpu) {
2933                 free_vmcs(per_cpu(vmxarea, cpu));
2934                 per_cpu(vmxarea, cpu) = NULL;
2935         }
2936 }
2937
2938 static __init int alloc_kvm_area(void)
2939 {
2940         int cpu;
2941
2942         for_each_possible_cpu(cpu) {
2943                 struct vmcs *vmcs;
2944
2945                 vmcs = alloc_vmcs_cpu(cpu);
2946                 if (!vmcs) {
2947                         free_kvm_area();
2948                         return -ENOMEM;
2949                 }
2950
2951                 per_cpu(vmxarea, cpu) = vmcs;
2952         }
2953         return 0;
2954 }
2955
2956 static __init int hardware_setup(void)
2957 {
2958         if (setup_vmcs_config(&vmcs_config) < 0)
2959                 return -EIO;
2960
2961         if (boot_cpu_has(X86_FEATURE_NX))
2962                 kvm_enable_efer_bits(EFER_NX);
2963
2964         if (!cpu_has_vmx_vpid())
2965                 enable_vpid = 0;
2966         if (!cpu_has_vmx_shadow_vmcs())
2967                 enable_shadow_vmcs = 0;
2968
2969         if (!cpu_has_vmx_ept() ||
2970             !cpu_has_vmx_ept_4levels()) {
2971                 enable_ept = 0;
2972                 enable_unrestricted_guest = 0;
2973                 enable_ept_ad_bits = 0;
2974         }
2975
2976         if (!cpu_has_vmx_ept_ad_bits())
2977                 enable_ept_ad_bits = 0;
2978
2979         if (!cpu_has_vmx_unrestricted_guest())
2980                 enable_unrestricted_guest = 0;
2981
2982         if (!cpu_has_vmx_flexpriority())
2983                 flexpriority_enabled = 0;
2984
2985         if (!cpu_has_vmx_tpr_shadow())
2986                 kvm_x86_ops->update_cr8_intercept = NULL;
2987
2988         if (enable_ept && !cpu_has_vmx_ept_2m_page())
2989                 kvm_disable_largepages();
2990
2991         if (!cpu_has_vmx_ple())
2992                 ple_gap = 0;
2993
2994         if (!cpu_has_vmx_apicv())
2995                 enable_apicv = 0;
2996
2997         if (enable_apicv)
2998                 kvm_x86_ops->update_cr8_intercept = NULL;
2999         else {
3000                 kvm_x86_ops->hwapic_irr_update = NULL;
3001                 kvm_x86_ops->deliver_posted_interrupt = NULL;
3002                 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
3003         }
3004
3005         if (nested)
3006                 nested_vmx_setup_ctls_msrs();
3007
3008         return alloc_kvm_area();
3009 }
3010
3011 static __exit void hardware_unsetup(void)
3012 {
3013         free_kvm_area();
3014 }
3015
3016 static bool emulation_required(struct kvm_vcpu *vcpu)
3017 {
3018         return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3019 }
3020
3021 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3022                 struct kvm_segment *save)
3023 {
3024         if (!emulate_invalid_guest_state) {
3025                 /*
3026                  * CS and SS RPL should be equal during guest entry according
3027                  * to VMX spec, but in reality it is not always so. Since vcpu
3028                  * is in the middle of the transition from real mode to
3029                  * protected mode it is safe to assume that RPL 0 is a good
3030                  * default value.
3031                  */
3032                 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3033                         save->selector &= ~SELECTOR_RPL_MASK;
3034                 save->dpl = save->selector & SELECTOR_RPL_MASK;
3035                 save->s = 1;
3036         }
3037         vmx_set_segment(vcpu, save, seg);
3038 }
3039
3040 static void enter_pmode(struct kvm_vcpu *vcpu)
3041 {
3042         unsigned long flags;
3043         struct vcpu_vmx *vmx = to_vmx(vcpu);
3044
3045         /*
3046          * Update real mode segment cache. It may be not up-to-date if sement
3047          * register was written while vcpu was in a guest mode.
3048          */
3049         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3050         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3051         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3052         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3053         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3054         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3055
3056         vmx->rmode.vm86_active = 0;
3057
3058         vmx_segment_cache_clear(vmx);
3059
3060         vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3061
3062         flags = vmcs_readl(GUEST_RFLAGS);
3063         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3064         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3065         vmcs_writel(GUEST_RFLAGS, flags);
3066
3067         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3068                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3069
3070         update_exception_bitmap(vcpu);
3071
3072         fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3073         fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3074         fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3075         fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3076         fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3077         fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3078
3079         /* CPL is always 0 when CPU enters protected mode */
3080         __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3081         vmx->cpl = 0;
3082 }
3083
3084 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3085 {
3086         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3087         struct kvm_segment var = *save;
3088
3089         var.dpl = 0x3;
3090         if (seg == VCPU_SREG_CS)
3091                 var.type = 0x3;
3092
3093         if (!emulate_invalid_guest_state) {
3094                 var.selector = var.base >> 4;
3095                 var.base = var.base & 0xffff0;
3096                 var.limit = 0xffff;
3097                 var.g = 0;
3098                 var.db = 0;
3099                 var.present = 1;
3100                 var.s = 1;
3101                 var.l = 0;
3102                 var.unusable = 0;
3103                 var.type = 0x3;
3104                 var.avl = 0;
3105                 if (save->base & 0xf)
3106                         printk_once(KERN_WARNING "kvm: segment base is not "
3107                                         "paragraph aligned when entering "
3108                                         "protected mode (seg=%d)", seg);
3109         }
3110
3111         vmcs_write16(sf->selector, var.selector);
3112         vmcs_write32(sf->base, var.base);
3113         vmcs_write32(sf->limit, var.limit);
3114         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3115 }
3116
3117 static void enter_rmode(struct kvm_vcpu *vcpu)
3118 {
3119         unsigned long flags;
3120         struct vcpu_vmx *vmx = to_vmx(vcpu);
3121
3122         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3123         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3124         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3125         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3126         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3127         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3128         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3129
3130         vmx->rmode.vm86_active = 1;
3131
3132         /*
3133          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3134          * vcpu. Warn the user that an update is overdue.
3135          */
3136         if (!vcpu->kvm->arch.tss_addr)
3137                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3138                              "called before entering vcpu\n");
3139
3140         vmx_segment_cache_clear(vmx);
3141
3142         vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3143         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3144         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3145
3146         flags = vmcs_readl(GUEST_RFLAGS);
3147         vmx->rmode.save_rflags = flags;
3148
3149         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3150
3151         vmcs_writel(GUEST_RFLAGS, flags);
3152         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3153         update_exception_bitmap(vcpu);
3154
3155         fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3156         fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3157         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3158         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3159         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3160         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3161
3162         kvm_mmu_reset_context(vcpu);
3163 }
3164
3165 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3166 {
3167         struct vcpu_vmx *vmx = to_vmx(vcpu);
3168         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3169
3170         if (!msr)
3171                 return;
3172
3173         /*
3174          * Force kernel_gs_base reloading before EFER changes, as control
3175          * of this msr depends on is_long_mode().
3176          */
3177         vmx_load_host_state(to_vmx(vcpu));
3178         vcpu->arch.efer = efer;
3179         if (efer & EFER_LMA) {
3180                 vmcs_write32(VM_ENTRY_CONTROLS,
3181                              vmcs_read32(VM_ENTRY_CONTROLS) |
3182                              VM_ENTRY_IA32E_MODE);
3183                 msr->data = efer;
3184         } else {
3185                 vmcs_write32(VM_ENTRY_CONTROLS,
3186                              vmcs_read32(VM_ENTRY_CONTROLS) &
3187                              ~VM_ENTRY_IA32E_MODE);
3188
3189                 msr->data = efer & ~EFER_LME;
3190         }
3191         setup_msrs(vmx);
3192 }
3193
3194 #ifdef CONFIG_X86_64
3195
3196 static void enter_lmode(struct kvm_vcpu *vcpu)
3197 {
3198         u32 guest_tr_ar;
3199
3200         vmx_segment_cache_clear(to_vmx(vcpu));
3201
3202         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3203         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3204                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3205                                      __func__);
3206                 vmcs_write32(GUEST_TR_AR_BYTES,
3207                              (guest_tr_ar & ~AR_TYPE_MASK)
3208                              | AR_TYPE_BUSY_64_TSS);
3209         }
3210         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3211 }
3212
3213 static void exit_lmode(struct kvm_vcpu *vcpu)
3214 {
3215         vmcs_write32(VM_ENTRY_CONTROLS,
3216                      vmcs_read32(VM_ENTRY_CONTROLS)
3217                      & ~VM_ENTRY_IA32E_MODE);
3218         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3219 }
3220
3221 #endif
3222
3223 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3224 {
3225         vpid_sync_context(to_vmx(vcpu));
3226         if (enable_ept) {
3227                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3228                         return;
3229                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3230         }
3231 }
3232
3233 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3234 {
3235         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3236
3237         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3238         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3239 }
3240
3241 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3242 {
3243         if (enable_ept && is_paging(vcpu))
3244                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3245         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3246 }
3247
3248 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3249 {
3250         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3251
3252         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3253         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3254 }
3255
3256 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3257 {
3258         if (!test_bit(VCPU_EXREG_PDPTR,
3259                       (unsigned long *)&vcpu->arch.regs_dirty))
3260                 return;
3261
3262         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3263                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
3264                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
3265                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
3266                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
3267         }
3268 }
3269
3270 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3271 {
3272         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3273                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3274                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3275                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3276                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3277         }
3278
3279         __set_bit(VCPU_EXREG_PDPTR,
3280                   (unsigned long *)&vcpu->arch.regs_avail);
3281         __set_bit(VCPU_EXREG_PDPTR,
3282                   (unsigned long *)&vcpu->arch.regs_dirty);
3283 }
3284
3285 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3286
3287 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3288                                         unsigned long cr0,
3289                                         struct kvm_vcpu *vcpu)
3290 {
3291         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3292                 vmx_decache_cr3(vcpu);
3293         if (!(cr0 & X86_CR0_PG)) {
3294                 /* From paging/starting to nonpaging */
3295                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3296                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3297                              (CPU_BASED_CR3_LOAD_EXITING |
3298                               CPU_BASED_CR3_STORE_EXITING));
3299                 vcpu->arch.cr0 = cr0;
3300                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3301         } else if (!is_paging(vcpu)) {
3302                 /* From nonpaging to paging */
3303                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3304                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3305                              ~(CPU_BASED_CR3_LOAD_EXITING |
3306                                CPU_BASED_CR3_STORE_EXITING));
3307                 vcpu->arch.cr0 = cr0;
3308                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3309         }
3310
3311         if (!(cr0 & X86_CR0_WP))
3312                 *hw_cr0 &= ~X86_CR0_WP;
3313 }
3314
3315 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3316 {
3317         struct vcpu_vmx *vmx = to_vmx(vcpu);
3318         unsigned long hw_cr0;
3319
3320         hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3321         if (enable_unrestricted_guest)
3322                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3323         else {
3324                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3325
3326                 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3327                         enter_pmode(vcpu);
3328
3329                 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3330                         enter_rmode(vcpu);
3331         }
3332
3333 #ifdef CONFIG_X86_64
3334         if (vcpu->arch.efer & EFER_LME) {
3335                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3336                         enter_lmode(vcpu);
3337                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3338                         exit_lmode(vcpu);
3339         }
3340 #endif
3341
3342         if (enable_ept)
3343                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3344
3345         if (!vcpu->fpu_active)
3346                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3347
3348         vmcs_writel(CR0_READ_SHADOW, cr0);
3349         vmcs_writel(GUEST_CR0, hw_cr0);
3350         vcpu->arch.cr0 = cr0;
3351
3352         /* depends on vcpu->arch.cr0 to be set to a new value */
3353         vmx->emulation_required = emulation_required(vcpu);
3354 }
3355
3356 static u64 construct_eptp(unsigned long root_hpa)
3357 {
3358         u64 eptp;
3359
3360         /* TODO write the value reading from MSR */
3361         eptp = VMX_EPT_DEFAULT_MT |
3362                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3363         if (enable_ept_ad_bits)
3364                 eptp |= VMX_EPT_AD_ENABLE_BIT;
3365         eptp |= (root_hpa & PAGE_MASK);
3366
3367         return eptp;
3368 }
3369
3370 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3371 {
3372         unsigned long guest_cr3;
3373         u64 eptp;
3374
3375         guest_cr3 = cr3;
3376         if (enable_ept) {
3377                 eptp = construct_eptp(cr3);
3378                 vmcs_write64(EPT_POINTER, eptp);
3379                 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3380                         vcpu->kvm->arch.ept_identity_map_addr;
3381                 ept_load_pdptrs(vcpu);
3382         }
3383
3384         vmx_flush_tlb(vcpu);
3385         vmcs_writel(GUEST_CR3, guest_cr3);
3386 }
3387
3388 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3389 {
3390         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3391                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3392
3393         if (cr4 & X86_CR4_VMXE) {
3394                 /*
3395                  * To use VMXON (and later other VMX instructions), a guest
3396                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
3397                  * So basically the check on whether to allow nested VMX
3398                  * is here.
3399                  */
3400                 if (!nested_vmx_allowed(vcpu))
3401                         return 1;
3402         }
3403         if (to_vmx(vcpu)->nested.vmxon &&
3404             ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3405                 return 1;
3406
3407         vcpu->arch.cr4 = cr4;
3408         if (enable_ept) {
3409                 if (!is_paging(vcpu)) {
3410                         hw_cr4 &= ~X86_CR4_PAE;
3411                         hw_cr4 |= X86_CR4_PSE;
3412                         /*
3413                          * SMEP is disabled if CPU is in non-paging mode in
3414                          * hardware. However KVM always uses paging mode to
3415                          * emulate guest non-paging mode with TDP.
3416                          * To emulate this behavior, SMEP needs to be manually
3417                          * disabled when guest switches to non-paging mode.
3418                          */
3419                         hw_cr4 &= ~X86_CR4_SMEP;
3420                 } else if (!(cr4 & X86_CR4_PAE)) {
3421                         hw_cr4 &= ~X86_CR4_PAE;
3422                 }
3423         }
3424
3425         vmcs_writel(CR4_READ_SHADOW, cr4);
3426         vmcs_writel(GUEST_CR4, hw_cr4);
3427         return 0;
3428 }
3429
3430 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3431                             struct kvm_segment *var, int seg)
3432 {
3433         struct vcpu_vmx *vmx = to_vmx(vcpu);
3434         u32 ar;
3435
3436         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3437                 *var = vmx->rmode.segs[seg];
3438                 if (seg == VCPU_SREG_TR
3439                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3440                         return;
3441                 var->base = vmx_read_guest_seg_base(vmx, seg);
3442                 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3443                 return;
3444         }
3445         var->base = vmx_read_guest_seg_base(vmx, seg);
3446         var->limit = vmx_read_guest_seg_limit(vmx, seg);
3447         var->selector = vmx_read_guest_seg_selector(vmx, seg);
3448         ar = vmx_read_guest_seg_ar(vmx, seg);
3449         var->unusable = (ar >> 16) & 1;
3450         var->type = ar & 15;
3451         var->s = (ar >> 4) & 1;
3452         var->dpl = (ar >> 5) & 3;
3453         /*
3454          * Some userspaces do not preserve unusable property. Since usable
3455          * segment has to be present according to VMX spec we can use present
3456          * property to amend userspace bug by making unusable segment always
3457          * nonpresent. vmx_segment_access_rights() already marks nonpresent
3458          * segment as unusable.
3459          */
3460         var->present = !var->unusable;
3461         var->avl = (ar >> 12) & 1;
3462         var->l = (ar >> 13) & 1;
3463         var->db = (ar >> 14) & 1;
3464         var->g = (ar >> 15) & 1;
3465 }
3466
3467 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3468 {
3469         struct kvm_segment s;
3470
3471         if (to_vmx(vcpu)->rmode.vm86_active) {
3472                 vmx_get_segment(vcpu, &s, seg);
3473                 return s.base;
3474         }
3475         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3476 }
3477
3478 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3479 {
3480         struct vcpu_vmx *vmx = to_vmx(vcpu);
3481
3482         if (!is_protmode(vcpu))
3483                 return 0;
3484
3485         if (!is_long_mode(vcpu)
3486             && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3487                 return 3;
3488
3489         if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3490                 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3491                 vmx->cpl = vmx_read_guest_seg_selector(vmx, VCPU_SREG_CS) & 3;
3492         }
3493
3494         return vmx->cpl;
3495 }
3496
3497
3498 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3499 {
3500         u32 ar;
3501
3502         if (var->unusable || !var->present)
3503                 ar = 1 << 16;
3504         else {
3505                 ar = var->type & 15;
3506                 ar |= (var->s & 1) << 4;
3507                 ar |= (var->dpl & 3) << 5;
3508                 ar |= (var->present & 1) << 7;
3509                 ar |= (var->avl & 1) << 12;
3510                 ar |= (var->l & 1) << 13;
3511                 ar |= (var->db & 1) << 14;
3512                 ar |= (var->g & 1) << 15;
3513         }
3514
3515         return ar;
3516 }
3517
3518 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3519                             struct kvm_segment *var, int seg)
3520 {
3521         struct vcpu_vmx *vmx = to_vmx(vcpu);
3522         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3523
3524         vmx_segment_cache_clear(vmx);
3525         if (seg == VCPU_SREG_CS)
3526                 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3527
3528         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3529                 vmx->rmode.segs[seg] = *var;
3530                 if (seg == VCPU_SREG_TR)
3531                         vmcs_write16(sf->selector, var->selector);
3532                 else if (var->s)
3533                         fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3534                 goto out;
3535         }
3536
3537         vmcs_writel(sf->base, var->base);
3538         vmcs_write32(sf->limit, var->limit);
3539         vmcs_write16(sf->selector, var->selector);
3540
3541         /*
3542          *   Fix the "Accessed" bit in AR field of segment registers for older
3543          * qemu binaries.
3544          *   IA32 arch specifies that at the time of processor reset the
3545          * "Accessed" bit in the AR field of segment registers is 1. And qemu
3546          * is setting it to 0 in the userland code. This causes invalid guest
3547          * state vmexit when "unrestricted guest" mode is turned on.
3548          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3549          * tree. Newer qemu binaries with that qemu fix would not need this
3550          * kvm hack.
3551          */
3552         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3553                 var->type |= 0x1; /* Accessed */
3554
3555         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3556
3557 out:
3558         vmx->emulation_required |= emulation_required(vcpu);
3559 }
3560
3561 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3562 {
3563         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3564
3565         *db = (ar >> 14) & 1;
3566         *l = (ar >> 13) & 1;
3567 }
3568
3569 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3570 {
3571         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3572         dt->address = vmcs_readl(GUEST_IDTR_BASE);
3573 }
3574
3575 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3576 {
3577         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3578         vmcs_writel(GUEST_IDTR_BASE, dt->address);
3579 }
3580
3581 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3582 {
3583         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3584         dt->address = vmcs_readl(GUEST_GDTR_BASE);
3585 }
3586
3587 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3588 {
3589         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3590         vmcs_writel(GUEST_GDTR_BASE, dt->address);
3591 }
3592
3593 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3594 {
3595         struct kvm_segment var;
3596         u32 ar;
3597
3598         vmx_get_segment(vcpu, &var, seg);
3599         var.dpl = 0x3;
3600         if (seg == VCPU_SREG_CS)
3601                 var.type = 0x3;
3602         ar = vmx_segment_access_rights(&var);
3603
3604         if (var.base != (var.selector << 4))
3605                 return false;
3606         if (var.limit != 0xffff)
3607                 return false;
3608         if (ar != 0xf3)
3609                 return false;
3610
3611         return true;
3612 }
3613
3614 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3615 {
3616         struct kvm_segment cs;
3617         unsigned int cs_rpl;
3618
3619         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3620         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3621
3622         if (cs.unusable)
3623                 return false;
3624         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3625                 return false;
3626         if (!cs.s)
3627                 return false;
3628         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3629                 if (cs.dpl > cs_rpl)
3630                         return false;
3631         } else {
3632                 if (cs.dpl != cs_rpl)
3633                         return false;
3634         }
3635         if (!cs.present)
3636                 return false;
3637
3638         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3639         return true;
3640 }
3641
3642 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3643 {
3644         struct kvm_segment ss;
3645         unsigned int ss_rpl;
3646
3647         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3648         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3649
3650         if (ss.unusable)
3651                 return true;
3652         if (ss.type != 3 && ss.type != 7)
3653                 return false;
3654         if (!ss.s)
3655                 return false;
3656         if (ss.dpl != ss_rpl) /* DPL != RPL */
3657                 return false;
3658         if (!ss.present)
3659                 return false;
3660
3661         return true;
3662 }
3663
3664 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3665 {
3666         struct kvm_segment var;
3667         unsigned int rpl;
3668
3669         vmx_get_segment(vcpu, &var, seg);
3670         rpl = var.selector & SELECTOR_RPL_MASK;
3671
3672         if (var.unusable)
3673                 return true;
3674         if (!var.s)
3675                 return false;
3676         if (!var.present)
3677                 return false;
3678         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3679                 if (var.dpl < rpl) /* DPL < RPL */
3680                         return false;
3681         }
3682
3683         /* TODO: Add other members to kvm_segment_field to allow checking for other access
3684          * rights flags
3685          */
3686         return true;
3687 }
3688
3689 static bool tr_valid(struct kvm_vcpu *vcpu)
3690 {
3691         struct kvm_segment tr;
3692
3693         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3694
3695         if (tr.unusable)
3696                 return false;
3697         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
3698                 return false;
3699         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3700                 return false;
3701         if (!tr.present)
3702                 return false;
3703
3704         return true;
3705 }
3706
3707 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3708 {
3709         struct kvm_segment ldtr;
3710
3711         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3712
3713         if (ldtr.unusable)
3714                 return true;
3715         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
3716                 return false;
3717         if (ldtr.type != 2)
3718                 return false;
3719         if (!ldtr.present)
3720                 return false;
3721
3722         return true;
3723 }
3724
3725 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3726 {
3727         struct kvm_segment cs, ss;
3728
3729         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3730         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3731
3732         return ((cs.selector & SELECTOR_RPL_MASK) ==
3733                  (ss.selector & SELECTOR_RPL_MASK));
3734 }
3735
3736 /*
3737  * Check if guest state is valid. Returns true if valid, false if
3738  * not.
3739  * We assume that registers are always usable
3740  */
3741 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3742 {
3743         if (enable_unrestricted_guest)
3744                 return true;
3745
3746         /* real mode guest state checks */
3747         if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3748                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3749                         return false;
3750                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3751                         return false;
3752                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3753                         return false;
3754                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3755                         return false;
3756                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3757                         return false;
3758                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3759                         return false;
3760         } else {
3761         /* protected mode guest state checks */
3762                 if (!cs_ss_rpl_check(vcpu))
3763                         return false;
3764                 if (!code_segment_valid(vcpu))
3765                         return false;
3766                 if (!stack_segment_valid(vcpu))
3767                         return false;
3768                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3769                         return false;
3770                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3771                         return false;
3772                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3773                         return false;
3774                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3775                         return false;
3776                 if (!tr_valid(vcpu))
3777                         return false;
3778                 if (!ldtr_valid(vcpu))
3779                         return false;
3780         }
3781         /* TODO:
3782          * - Add checks on RIP
3783          * - Add checks on RFLAGS
3784          */
3785
3786         return true;
3787 }
3788
3789 static int init_rmode_tss(struct kvm *kvm)
3790 {
3791         gfn_t fn;
3792         u16 data = 0;
3793         int r, idx, ret = 0;
3794
3795         idx = srcu_read_lock(&kvm->srcu);
3796         fn = kvm->arch.tss_addr >> PAGE_SHIFT;
3797         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3798         if (r < 0)
3799                 goto out;
3800         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3801         r = kvm_write_guest_page(kvm, fn++, &data,
3802                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
3803         if (r < 0)
3804                 goto out;
3805         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3806         if (r < 0)
3807                 goto out;
3808         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3809         if (r < 0)
3810                 goto out;
3811         data = ~0;
3812         r = kvm_write_guest_page(kvm, fn, &data,
3813                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3814                                  sizeof(u8));
3815         if (r < 0)
3816                 goto out;
3817
3818         ret = 1;
3819 out:
3820         srcu_read_unlock(&kvm->srcu, idx);
3821         return ret;
3822 }
3823
3824 static int init_rmode_identity_map(struct kvm *kvm)
3825 {
3826         int i, idx, r, ret;
3827         pfn_t identity_map_pfn;
3828         u32 tmp;
3829
3830         if (!enable_ept)
3831                 return 1;
3832         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3833                 printk(KERN_ERR "EPT: identity-mapping pagetable "
3834                         "haven't been allocated!\n");
3835                 return 0;
3836         }
3837         if (likely(kvm->arch.ept_identity_pagetable_done))
3838                 return 1;
3839         ret = 0;
3840         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3841         idx = srcu_read_lock(&kvm->srcu);
3842         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3843         if (r < 0)
3844                 goto out;
3845         /* Set up identity-mapping pagetable for EPT in real mode */
3846         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3847                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3848                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3849                 r = kvm_write_guest_page(kvm, identity_map_pfn,
3850                                 &tmp, i * sizeof(tmp), sizeof(tmp));
3851                 if (r < 0)
3852                         goto out;
3853         }
3854         kvm->arch.ept_identity_pagetable_done = true;
3855         ret = 1;
3856 out:
3857         srcu_read_unlock(&kvm->srcu, idx);
3858         return ret;
3859 }
3860
3861 static void seg_setup(int seg)
3862 {
3863         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3864         unsigned int ar;
3865
3866         vmcs_write16(sf->selector, 0);
3867         vmcs_writel(sf->base, 0);
3868         vmcs_write32(sf->limit, 0xffff);
3869         ar = 0x93;
3870         if (seg == VCPU_SREG_CS)
3871                 ar |= 0x08; /* code segment */
3872
3873         vmcs_write32(sf->ar_bytes, ar);
3874 }
3875
3876 static int alloc_apic_access_page(struct kvm *kvm)
3877 {
3878         struct page *page;
3879         struct kvm_userspace_memory_region kvm_userspace_mem;
3880         int r = 0;
3881
3882         mutex_lock(&kvm->slots_lock);
3883         if (kvm->arch.apic_access_page)
3884                 goto out;
3885         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3886         kvm_userspace_mem.flags = 0;
3887         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3888         kvm_userspace_mem.memory_size = PAGE_SIZE;
3889         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3890         if (r)
3891                 goto out;
3892
3893         page = gfn_to_page(kvm, 0xfee00);
3894         if (is_error_page(page)) {
3895                 r = -EFAULT;
3896                 goto out;
3897         }
3898
3899         kvm->arch.apic_access_page = page;
3900 out:
3901         mutex_unlock(&kvm->slots_lock);
3902         return r;
3903 }
3904
3905 static int alloc_identity_pagetable(struct kvm *kvm)
3906 {
3907         struct page *page;
3908         struct kvm_userspace_memory_region kvm_userspace_mem;
3909         int r = 0;
3910
3911         mutex_lock(&kvm->slots_lock);
3912         if (kvm->arch.ept_identity_pagetable)
3913                 goto out;
3914         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3915         kvm_userspace_mem.flags = 0;
3916         kvm_userspace_mem.guest_phys_addr =
3917                 kvm->arch.ept_identity_map_addr;
3918         kvm_userspace_mem.memory_size = PAGE_SIZE;
3919         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3920         if (r)
3921                 goto out;
3922
3923         page = gfn_to_page(kvm, kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3924         if (is_error_page(page)) {
3925                 r = -EFAULT;
3926                 goto out;
3927         }
3928
3929         kvm->arch.ept_identity_pagetable = page;
3930 out:
3931         mutex_unlock(&kvm->slots_lock);
3932         return r;
3933 }
3934
3935 static void allocate_vpid(struct vcpu_vmx *vmx)
3936 {
3937         int vpid;
3938
3939         vmx->vpid = 0;
3940         if (!enable_vpid)
3941                 return;
3942         spin_lock(&vmx_vpid_lock);
3943         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3944         if (vpid < VMX_NR_VPIDS) {
3945                 vmx->vpid = vpid;
3946                 __set_bit(vpid, vmx_vpid_bitmap);
3947         }
3948         spin_unlock(&vmx_vpid_lock);
3949 }
3950
3951 static void free_vpid(struct vcpu_vmx *vmx)
3952 {
3953         if (!enable_vpid)
3954                 return;
3955         spin_lock(&vmx_vpid_lock);
3956         if (vmx->vpid != 0)
3957                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3958         spin_unlock(&vmx_vpid_lock);
3959 }
3960
3961 #define MSR_TYPE_R      1
3962 #define MSR_TYPE_W      2
3963 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3964                                                 u32 msr, int type)
3965 {
3966         int f = sizeof(unsigned long);
3967
3968         if (!cpu_has_vmx_msr_bitmap())
3969                 return;
3970
3971         /*
3972          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3973          * have the write-low and read-high bitmap offsets the wrong way round.
3974          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3975          */
3976         if (msr <= 0x1fff) {
3977                 if (type & MSR_TYPE_R)
3978                         /* read-low */
3979                         __clear_bit(msr, msr_bitmap + 0x000 / f);
3980
3981                 if (type & MSR_TYPE_W)
3982                         /* write-low */
3983                         __clear_bit(msr, msr_bitmap + 0x800 / f);
3984
3985         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3986                 msr &= 0x1fff;
3987                 if (type & MSR_TYPE_R)
3988                         /* read-high */
3989                         __clear_bit(msr, msr_bitmap + 0x400 / f);
3990
3991                 if (type & MSR_TYPE_W)
3992                         /* write-high */
3993                         __clear_bit(msr, msr_bitmap + 0xc00 / f);
3994
3995         }
3996 }
3997
3998 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3999                                                 u32 msr, int type)
4000 {
4001         int f = sizeof(unsigned long);
4002
4003         if (!cpu_has_vmx_msr_bitmap())
4004                 return;
4005
4006         /*
4007          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4008          * have the write-low and read-high bitmap offsets the wrong way round.
4009          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4010          */
4011         if (msr <= 0x1fff) {
4012                 if (type & MSR_TYPE_R)
4013                         /* read-low */
4014                         __set_bit(msr, msr_bitmap + 0x000 / f);
4015
4016                 if (type & MSR_TYPE_W)
4017                         /* write-low */
4018                         __set_bit(msr, msr_bitmap + 0x800 / f);
4019
4020         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4021                 msr &= 0x1fff;
4022                 if (type & MSR_TYPE_R)
4023                         /* read-high */
4024                         __set_bit(msr, msr_bitmap + 0x400 / f);
4025
4026                 if (type & MSR_TYPE_W)
4027                         /* write-high */
4028                         __set_bit(msr, msr_bitmap + 0xc00 / f);
4029
4030         }
4031 }
4032
4033 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4034 {
4035         if (!longmode_only)
4036                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4037                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
4038         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4039                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
4040 }
4041
4042 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4043 {
4044         __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4045                         msr, MSR_TYPE_R);
4046         __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4047                         msr, MSR_TYPE_R);
4048 }
4049
4050 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4051 {
4052         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4053                         msr, MSR_TYPE_R);
4054         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4055                         msr, MSR_TYPE_R);
4056 }
4057
4058 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4059 {
4060         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4061                         msr, MSR_TYPE_W);
4062         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4063                         msr, MSR_TYPE_W);
4064 }
4065
4066 static int vmx_vm_has_apicv(struct kvm *kvm)
4067 {
4068         return enable_apicv && irqchip_in_kernel(kvm);
4069 }
4070
4071 /*
4072  * Send interrupt to vcpu via posted interrupt way.
4073  * 1. If target vcpu is running(non-root mode), send posted interrupt
4074  * notification to vcpu and hardware will sync PIR to vIRR atomically.
4075  * 2. If target vcpu isn't running(root mode), kick it to pick up the
4076  * interrupt from PIR in next vmentry.
4077  */
4078 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4079 {
4080         struct vcpu_vmx *vmx = to_vmx(vcpu);
4081         int r;
4082
4083         if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4084                 return;
4085
4086         r = pi_test_and_set_on(&vmx->pi_desc);
4087         kvm_make_request(KVM_REQ_EVENT, vcpu);
4088 #ifdef CONFIG_SMP
4089         if (!r && (vcpu->mode == IN_GUEST_MODE))
4090                 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4091                                 POSTED_INTR_VECTOR);
4092         else
4093 #endif
4094                 kvm_vcpu_kick(vcpu);
4095 }
4096
4097 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4098 {
4099         struct vcpu_vmx *vmx = to_vmx(vcpu);
4100
4101         if (!pi_test_and_clear_on(&vmx->pi_desc))
4102                 return;
4103
4104         kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4105 }
4106
4107 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4108 {
4109         return;
4110 }
4111
4112 /*
4113  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4114  * will not change in the lifetime of the guest.
4115  * Note that host-state that does change is set elsewhere. E.g., host-state
4116  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4117  */
4118 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4119 {
4120         u32 low32, high32;
4121         unsigned long tmpl;
4122         struct desc_ptr dt;
4123
4124         vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS);  /* 22.2.3 */
4125         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
4126         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
4127
4128         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
4129 #ifdef CONFIG_X86_64
4130         /*
4131          * Load null selectors, so we can avoid reloading them in
4132          * __vmx_load_host_state(), in case userspace uses the null selectors
4133          * too (the expected case).
4134          */
4135         vmcs_write16(HOST_DS_SELECTOR, 0);
4136         vmcs_write16(HOST_ES_SELECTOR, 0);
4137 #else
4138         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4139         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4140 #endif
4141         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4142         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
4143
4144         native_store_idt(&dt);
4145         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
4146         vmx->host_idt_base = dt.address;
4147
4148         vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4149
4150         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4151         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4152         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4153         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
4154
4155         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4156                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4157                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4158         }
4159 }
4160
4161 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4162 {
4163         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4164         if (enable_ept)
4165                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4166         if (is_guest_mode(&vmx->vcpu))
4167                 vmx->vcpu.arch.cr4_guest_owned_bits &=
4168                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4169         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4170 }
4171
4172 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4173 {
4174         u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4175
4176         if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4177                 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4178         return pin_based_exec_ctrl;
4179 }
4180
4181 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4182 {
4183         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4184         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
4185                 exec_control &= ~CPU_BASED_TPR_SHADOW;
4186 #ifdef CONFIG_X86_64
4187                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4188                                 CPU_BASED_CR8_LOAD_EXITING;
4189 #endif
4190         }
4191         if (!enable_ept)
4192                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4193                                 CPU_BASED_CR3_LOAD_EXITING  |
4194                                 CPU_BASED_INVLPG_EXITING;
4195         return exec_control;
4196 }
4197
4198 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4199 {
4200         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4201         if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4202                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4203         if (vmx->vpid == 0)
4204                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4205         if (!enable_ept) {
4206                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4207                 enable_unrestricted_guest = 0;
4208                 /* Enable INVPCID for non-ept guests may cause performance regression. */
4209                 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4210         }
4211         if (!enable_unrestricted_guest)
4212                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4213         if (!ple_gap)
4214                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4215         if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4216                 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4217                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4218         exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4219         /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4220            (handle_vmptrld).
4221            We can NOT enable shadow_vmcs here because we don't have yet
4222            a current VMCS12
4223         */
4224         exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4225         return exec_control;
4226 }
4227
4228 static void ept_set_mmio_spte_mask(void)
4229 {
4230         /*
4231          * EPT Misconfigurations can be generated if the value of bits 2:0
4232          * of an EPT paging-structure entry is 110b (write/execute).
4233          * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4234          * spte.
4235          */
4236         kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4237 }
4238
4239 /*
4240  * Sets up the vmcs for emulated real mode.
4241  */
4242 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4243 {
4244 #ifdef CONFIG_X86_64
4245         unsigned long a;
4246 #endif
4247         int i;
4248
4249         /* I/O */
4250         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4251         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4252
4253         if (enable_shadow_vmcs) {
4254                 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4255                 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4256         }
4257         if (cpu_has_vmx_msr_bitmap())
4258                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4259
4260         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4261
4262         /* Control */
4263         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4264
4265         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4266
4267         if (cpu_has_secondary_exec_ctrls()) {
4268                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4269                                 vmx_secondary_exec_control(vmx));
4270         }
4271
4272         if (vmx_vm_has_apicv(vmx->vcpu.kvm)) {
4273                 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4274                 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4275                 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4276                 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4277
4278                 vmcs_write16(GUEST_INTR_STATUS, 0);
4279
4280                 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4281                 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4282         }
4283
4284         if (ple_gap) {
4285                 vmcs_write32(PLE_GAP, ple_gap);
4286                 vmcs_write32(PLE_WINDOW, ple_window);
4287         }
4288
4289         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4290         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4291         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4292
4293         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4294         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4295         vmx_set_constant_host_state(vmx);
4296 #ifdef CONFIG_X86_64
4297         rdmsrl(MSR_FS_BASE, a);
4298         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4299         rdmsrl(MSR_GS_BASE, a);
4300         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4301 #else
4302         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4303         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4304 #endif
4305
4306         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4307         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4308         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4309         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4310         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4311
4312         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
4313                 u32 msr_low, msr_high;
4314                 u64 host_pat;
4315                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
4316                 host_pat = msr_low | ((u64) msr_high << 32);
4317                 /* Write the default value follow host pat */
4318                 vmcs_write64(GUEST_IA32_PAT, host_pat);
4319                 /* Keep arch.pat sync with GUEST_IA32_PAT */
4320                 vmx->vcpu.arch.pat = host_pat;
4321         }
4322
4323         for (i = 0; i < NR_VMX_MSR; ++i) {
4324                 u32 index = vmx_msr_index[i];
4325                 u32 data_low, data_high;
4326                 int j = vmx->nmsrs;
4327
4328                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4329                         continue;
4330                 if (wrmsr_safe(index, data_low, data_high) < 0)
4331                         continue;
4332                 vmx->guest_msrs[j].index = i;
4333                 vmx->guest_msrs[j].data = 0;
4334                 vmx->guest_msrs[j].mask = -1ull;
4335                 ++vmx->nmsrs;
4336         }
4337
4338         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
4339
4340         /* 22.2.1, 20.8.1 */
4341         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
4342
4343         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4344         set_cr4_guest_host_mask(vmx);
4345
4346         return 0;
4347 }
4348
4349 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4350 {
4351         struct vcpu_vmx *vmx = to_vmx(vcpu);
4352         u64 msr;
4353
4354         vmx->rmode.vm86_active = 0;
4355
4356         vmx->soft_vnmi_blocked = 0;
4357
4358         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4359         kvm_set_cr8(&vmx->vcpu, 0);
4360         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
4361         if (kvm_vcpu_is_bsp(&vmx->vcpu))
4362                 msr |= MSR_IA32_APICBASE_BSP;
4363         kvm_set_apic_base(&vmx->vcpu, msr);
4364
4365         vmx_segment_cache_clear(vmx);
4366
4367         seg_setup(VCPU_SREG_CS);
4368         vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4369         vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4370
4371         seg_setup(VCPU_SREG_DS);
4372         seg_setup(VCPU_SREG_ES);
4373         seg_setup(VCPU_SREG_FS);
4374         seg_setup(VCPU_SREG_GS);
4375         seg_setup(VCPU_SREG_SS);
4376
4377         vmcs_write16(GUEST_TR_SELECTOR, 0);
4378         vmcs_writel(GUEST_TR_BASE, 0);
4379         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4380         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4381
4382         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4383         vmcs_writel(GUEST_LDTR_BASE, 0);
4384         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4385         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4386
4387         vmcs_write32(GUEST_SYSENTER_CS, 0);
4388         vmcs_writel(GUEST_SYSENTER_ESP, 0);
4389         vmcs_writel(GUEST_SYSENTER_EIP, 0);
4390
4391         vmcs_writel(GUEST_RFLAGS, 0x02);
4392         kvm_rip_write(vcpu, 0xfff0);
4393
4394         vmcs_writel(GUEST_GDTR_BASE, 0);
4395         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4396
4397         vmcs_writel(GUEST_IDTR_BASE, 0);
4398         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4399
4400         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4401         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4402         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4403
4404         /* Special registers */
4405         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4406
4407         setup_msrs(vmx);
4408
4409         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4410
4411         if (cpu_has_vmx_tpr_shadow()) {
4412                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4413                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
4414                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4415                                      __pa(vmx->vcpu.arch.apic->regs));
4416                 vmcs_write32(TPR_THRESHOLD, 0);
4417         }
4418
4419         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4420                 vmcs_write64(APIC_ACCESS_ADDR,
4421                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
4422
4423         if (vmx_vm_has_apicv(vcpu->kvm))
4424                 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4425
4426         if (vmx->vpid != 0)
4427                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4428
4429         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4430         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4431         vmx_set_cr4(&vmx->vcpu, 0);
4432         vmx_set_efer(&vmx->vcpu, 0);
4433         vmx_fpu_activate(&vmx->vcpu);
4434         update_exception_bitmap(&vmx->vcpu);
4435
4436         vpid_sync_context(vmx);
4437 }
4438
4439 /*
4440  * In nested virtualization, check if L1 asked to exit on external interrupts.
4441  * For most existing hypervisors, this will always return true.
4442  */
4443 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4444 {
4445         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4446                 PIN_BASED_EXT_INTR_MASK;
4447 }
4448
4449 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4450 {
4451         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4452                 PIN_BASED_NMI_EXITING;
4453 }
4454
4455 static int enable_irq_window(struct kvm_vcpu *vcpu)
4456 {
4457         u32 cpu_based_vm_exec_control;
4458
4459         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4460                 /*
4461                  * We get here if vmx_interrupt_allowed() said we can't
4462                  * inject to L1 now because L2 must run. The caller will have
4463                  * to make L2 exit right after entry, so we can inject to L1
4464                  * more promptly.
4465                  */
4466                 return -EBUSY;
4467
4468         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4469         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4470         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4471         return 0;
4472 }
4473
4474 static int enable_nmi_window(struct kvm_vcpu *vcpu)
4475 {
4476         u32 cpu_based_vm_exec_control;
4477
4478         if (!cpu_has_virtual_nmis())
4479                 return enable_irq_window(vcpu);
4480
4481         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI)
4482                 return enable_irq_window(vcpu);
4483
4484         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4485         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4486         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4487         return 0;
4488 }
4489
4490 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4491 {
4492         struct vcpu_vmx *vmx = to_vmx(vcpu);
4493         uint32_t intr;
4494         int irq = vcpu->arch.interrupt.nr;
4495
4496         trace_kvm_inj_virq(irq);
4497
4498         ++vcpu->stat.irq_injections;
4499         if (vmx->rmode.vm86_active) {
4500                 int inc_eip = 0;
4501                 if (vcpu->arch.interrupt.soft)
4502                         inc_eip = vcpu->arch.event_exit_inst_len;
4503                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4504                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4505                 return;
4506         }
4507         intr = irq | INTR_INFO_VALID_MASK;
4508         if (vcpu->arch.interrupt.soft) {
4509                 intr |= INTR_TYPE_SOFT_INTR;
4510                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4511                              vmx->vcpu.arch.event_exit_inst_len);
4512         } else
4513                 intr |= INTR_TYPE_EXT_INTR;
4514         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4515 }
4516
4517 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4518 {
4519         struct vcpu_vmx *vmx = to_vmx(vcpu);
4520
4521         if (is_guest_mode(vcpu))
4522                 return;
4523
4524         if (!cpu_has_virtual_nmis()) {
4525                 /*
4526                  * Tracking the NMI-blocked state in software is built upon
4527                  * finding the next open IRQ window. This, in turn, depends on
4528                  * well-behaving guests: They have to keep IRQs disabled at
4529                  * least as long as the NMI handler runs. Otherwise we may
4530                  * cause NMI nesting, maybe breaking the guest. But as this is
4531                  * highly unlikely, we can live with the residual risk.
4532                  */
4533                 vmx->soft_vnmi_blocked = 1;
4534                 vmx->vnmi_blocked_time = 0;
4535         }
4536
4537         ++vcpu->stat.nmi_injections;
4538         vmx->nmi_known_unmasked = false;
4539         if (vmx->rmode.vm86_active) {
4540                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4541                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4542                 return;
4543         }
4544         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4545                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4546 }
4547
4548 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4549 {
4550         if (!cpu_has_virtual_nmis())
4551                 return to_vmx(vcpu)->soft_vnmi_blocked;
4552         if (to_vmx(vcpu)->nmi_known_unmasked)
4553                 return false;
4554         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4555 }
4556
4557 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4558 {
4559         struct vcpu_vmx *vmx = to_vmx(vcpu);
4560
4561         if (!cpu_has_virtual_nmis()) {
4562                 if (vmx->soft_vnmi_blocked != masked) {
4563                         vmx->soft_vnmi_blocked = masked;
4564                         vmx->vnmi_blocked_time = 0;
4565                 }
4566         } else {
4567                 vmx->nmi_known_unmasked = !masked;
4568                 if (masked)
4569                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4570                                       GUEST_INTR_STATE_NMI);
4571                 else
4572                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4573                                         GUEST_INTR_STATE_NMI);
4574         }
4575 }
4576
4577 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4578 {
4579         if (is_guest_mode(vcpu)) {
4580                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4581
4582                 if (to_vmx(vcpu)->nested.nested_run_pending)
4583                         return 0;
4584                 if (nested_exit_on_nmi(vcpu)) {
4585                         nested_vmx_vmexit(vcpu);
4586                         vmcs12->vm_exit_reason = EXIT_REASON_EXCEPTION_NMI;
4587                         vmcs12->vm_exit_intr_info = NMI_VECTOR |
4588                                 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK;
4589                         /*
4590                          * The NMI-triggered VM exit counts as injection:
4591                          * clear this one and block further NMIs.
4592                          */
4593                         vcpu->arch.nmi_pending = 0;
4594                         vmx_set_nmi_mask(vcpu, true);
4595                         return 0;
4596                 }
4597         }
4598
4599         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4600                 return 0;
4601
4602         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4603                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4604                    | GUEST_INTR_STATE_NMI));
4605 }
4606
4607 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4608 {
4609         if (is_guest_mode(vcpu)) {
4610                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4611
4612                 if (to_vmx(vcpu)->nested.nested_run_pending)
4613                         return 0;
4614                 if (nested_exit_on_intr(vcpu)) {
4615                         nested_vmx_vmexit(vcpu);
4616                         vmcs12->vm_exit_reason =
4617                                 EXIT_REASON_EXTERNAL_INTERRUPT;
4618                         vmcs12->vm_exit_intr_info = 0;
4619                         /*
4620                          * fall through to normal code, but now in L1, not L2
4621                          */
4622                 }
4623         }
4624
4625         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4626                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4627                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4628 }
4629
4630 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4631 {
4632         int ret;
4633         struct kvm_userspace_memory_region tss_mem = {
4634                 .slot = TSS_PRIVATE_MEMSLOT,
4635                 .guest_phys_addr = addr,
4636                 .memory_size = PAGE_SIZE * 3,
4637                 .flags = 0,
4638         };
4639
4640         ret = kvm_set_memory_region(kvm, &tss_mem);
4641         if (ret)
4642                 return ret;
4643         kvm->arch.tss_addr = addr;
4644         if (!init_rmode_tss(kvm))
4645                 return  -ENOMEM;
4646
4647         return 0;
4648 }
4649
4650 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4651 {
4652         switch (vec) {
4653         case BP_VECTOR:
4654                 /*
4655                  * Update instruction length as we may reinject the exception
4656                  * from user space while in guest debugging mode.
4657                  */
4658                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4659                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4660                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4661                         return false;
4662                 /* fall through */
4663         case DB_VECTOR:
4664                 if (vcpu->guest_debug &
4665                         (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4666                         return false;
4667                 /* fall through */
4668         case DE_VECTOR:
4669         case OF_VECTOR:
4670         case BR_VECTOR:
4671         case UD_VECTOR:
4672         case DF_VECTOR:
4673         case SS_VECTOR:
4674         case GP_VECTOR:
4675         case MF_VECTOR:
4676                 return true;
4677         break;
4678         }
4679         return false;
4680 }
4681
4682 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4683                                   int vec, u32 err_code)
4684 {
4685         /*
4686          * Instruction with address size override prefix opcode 0x67
4687          * Cause the #SS fault with 0 error code in VM86 mode.
4688          */
4689         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4690                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4691                         if (vcpu->arch.halt_request) {
4692                                 vcpu->arch.halt_request = 0;
4693                                 return kvm_emulate_halt(vcpu);
4694                         }
4695                         return 1;
4696                 }
4697                 return 0;
4698         }
4699
4700         /*
4701          * Forward all other exceptions that are valid in real mode.
4702          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4703          *        the required debugging infrastructure rework.
4704          */
4705         kvm_queue_exception(vcpu, vec);
4706         return 1;
4707 }
4708
4709 /*
4710  * Trigger machine check on the host. We assume all the MSRs are already set up
4711  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4712  * We pass a fake environment to the machine check handler because we want
4713  * the guest to be always treated like user space, no matter what context
4714  * it used internally.
4715  */
4716 static void kvm_machine_check(void)
4717 {
4718 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4719         struct pt_regs regs = {
4720                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4721                 .flags = X86_EFLAGS_IF,
4722         };
4723
4724         do_machine_check(&regs, 0);
4725 #endif
4726 }
4727
4728 static int handle_machine_check(struct kvm_vcpu *vcpu)
4729 {
4730         /* already handled by vcpu_run */
4731         return 1;
4732 }
4733
4734 static int handle_exception(struct kvm_vcpu *vcpu)
4735 {
4736         struct vcpu_vmx *vmx = to_vmx(vcpu);
4737         struct kvm_run *kvm_run = vcpu->run;
4738         u32 intr_info, ex_no, error_code;
4739         unsigned long cr2, rip, dr6;
4740         u32 vect_info;
4741         enum emulation_result er;
4742
4743         vect_info = vmx->idt_vectoring_info;
4744         intr_info = vmx->exit_intr_info;
4745
4746         if (is_machine_check(intr_info))
4747                 return handle_machine_check(vcpu);
4748
4749         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4750                 return 1;  /* already handled by vmx_vcpu_run() */
4751
4752         if (is_no_device(intr_info)) {
4753                 vmx_fpu_activate(vcpu);
4754                 return 1;
4755         }
4756
4757         if (is_invalid_opcode(intr_info)) {
4758                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4759                 if (er != EMULATE_DONE)
4760                         kvm_queue_exception(vcpu, UD_VECTOR);
4761                 return 1;
4762         }
4763
4764         error_code = 0;
4765         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4766                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4767
4768         /*
4769          * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4770          * MMIO, it is better to report an internal error.
4771          * See the comments in vmx_handle_exit.
4772          */
4773         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4774             !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4775                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4776                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4777                 vcpu->run->internal.ndata = 2;
4778                 vcpu->run->internal.data[0] = vect_info;
4779                 vcpu->run->internal.data[1] = intr_info;
4780                 return 0;
4781         }
4782
4783         if (is_page_fault(intr_info)) {
4784                 /* EPT won't cause page fault directly */
4785                 BUG_ON(enable_ept);
4786                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4787                 trace_kvm_page_fault(cr2, error_code);
4788
4789                 if (kvm_event_needs_reinjection(vcpu))
4790                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
4791                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4792         }
4793
4794         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4795
4796         if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4797                 return handle_rmode_exception(vcpu, ex_no, error_code);
4798
4799         switch (ex_no) {
4800         case DB_VECTOR:
4801                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4802                 if (!(vcpu->guest_debug &
4803                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4804                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4805                         kvm_queue_exception(vcpu, DB_VECTOR);
4806                         return 1;
4807                 }
4808                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4809                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4810                 /* fall through */
4811         case BP_VECTOR:
4812                 /*
4813                  * Update instruction length as we may reinject #BP from
4814                  * user space while in guest debugging mode. Reading it for
4815                  * #DB as well causes no harm, it is not used in that case.
4816                  */
4817                 vmx->vcpu.arch.event_exit_inst_len =
4818                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4819                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4820                 rip = kvm_rip_read(vcpu);
4821                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4822                 kvm_run->debug.arch.exception = ex_no;
4823                 break;
4824         default:
4825                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4826                 kvm_run->ex.exception = ex_no;
4827                 kvm_run->ex.error_code = error_code;
4828                 break;
4829         }
4830         return 0;
4831 }
4832
4833 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4834 {
4835         ++vcpu->stat.irq_exits;
4836         return 1;
4837 }
4838
4839 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4840 {
4841         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4842         return 0;
4843 }
4844
4845 static int handle_io(struct kvm_vcpu *vcpu)
4846 {
4847         unsigned long exit_qualification;
4848         int size, in, string;
4849         unsigned port;
4850
4851         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4852         string = (exit_qualification & 16) != 0;
4853         in = (exit_qualification & 8) != 0;
4854
4855         ++vcpu->stat.io_exits;
4856
4857         if (string || in)
4858                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4859
4860         port = exit_qualification >> 16;
4861         size = (exit_qualification & 7) + 1;
4862         skip_emulated_instruction(vcpu);
4863
4864         return kvm_fast_pio_out(vcpu, size, port);
4865 }
4866
4867 static void
4868 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4869 {
4870         /*
4871          * Patch in the VMCALL instruction:
4872          */
4873         hypercall[0] = 0x0f;
4874         hypercall[1] = 0x01;
4875         hypercall[2] = 0xc1;
4876 }
4877
4878 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4879 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4880 {
4881         if (is_guest_mode(vcpu)) {
4882                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4883                 unsigned long orig_val = val;
4884
4885                 /*
4886                  * We get here when L2 changed cr0 in a way that did not change
4887                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4888                  * but did change L0 shadowed bits. So we first calculate the
4889                  * effective cr0 value that L1 would like to write into the
4890                  * hardware. It consists of the L2-owned bits from the new
4891                  * value combined with the L1-owned bits from L1's guest_cr0.
4892                  */
4893                 val = (val & ~vmcs12->cr0_guest_host_mask) |
4894                         (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4895
4896                 /* TODO: will have to take unrestricted guest mode into
4897                  * account */
4898                 if ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON)
4899                         return 1;
4900
4901                 if (kvm_set_cr0(vcpu, val))
4902                         return 1;
4903                 vmcs_writel(CR0_READ_SHADOW, orig_val);
4904                 return 0;
4905         } else {
4906                 if (to_vmx(vcpu)->nested.vmxon &&
4907                     ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4908                         return 1;
4909                 return kvm_set_cr0(vcpu, val);
4910         }
4911 }
4912
4913 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4914 {
4915         if (is_guest_mode(vcpu)) {
4916                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4917                 unsigned long orig_val = val;
4918
4919                 /* analogously to handle_set_cr0 */
4920                 val = (val & ~vmcs12->cr4_guest_host_mask) |
4921                         (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4922                 if (kvm_set_cr4(vcpu, val))
4923                         return 1;
4924                 vmcs_writel(CR4_READ_SHADOW, orig_val);
4925                 return 0;
4926         } else
4927                 return kvm_set_cr4(vcpu, val);
4928 }
4929
4930 /* called to set cr0 as approriate for clts instruction exit. */
4931 static void handle_clts(struct kvm_vcpu *vcpu)
4932 {
4933         if (is_guest_mode(vcpu)) {
4934                 /*
4935                  * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4936                  * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4937                  * just pretend it's off (also in arch.cr0 for fpu_activate).
4938                  */
4939                 vmcs_writel(CR0_READ_SHADOW,
4940                         vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4941                 vcpu->arch.cr0 &= ~X86_CR0_TS;
4942         } else
4943                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4944 }
4945
4946 static int handle_cr(struct kvm_vcpu *vcpu)
4947 {
4948         unsigned long exit_qualification, val;
4949         int cr;
4950         int reg;
4951         int err;
4952
4953         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4954         cr = exit_qualification & 15;
4955         reg = (exit_qualification >> 8) & 15;
4956         switch ((exit_qualification >> 4) & 3) {
4957         case 0: /* mov to cr */
4958                 val = kvm_register_read(vcpu, reg);
4959                 trace_kvm_cr_write(cr, val);
4960                 switch (cr) {
4961                 case 0:
4962                         err = handle_set_cr0(vcpu, val);
4963                         kvm_complete_insn_gp(vcpu, err);
4964                         return 1;
4965                 case 3:
4966                         err = kvm_set_cr3(vcpu, val);
4967                         kvm_complete_insn_gp(vcpu, err);
4968                         return 1;
4969                 case 4:
4970                         err = handle_set_cr4(vcpu, val);
4971                         kvm_complete_insn_gp(vcpu, err);
4972                         return 1;
4973                 case 8: {
4974                                 u8 cr8_prev = kvm_get_cr8(vcpu);
4975                                 u8 cr8 = kvm_register_read(vcpu, reg);
4976                                 err = kvm_set_cr8(vcpu, cr8);
4977                                 kvm_complete_insn_gp(vcpu, err);
4978                                 if (irqchip_in_kernel(vcpu->kvm))
4979                                         return 1;
4980                                 if (cr8_prev <= cr8)
4981                                         return 1;
4982                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4983                                 return 0;
4984                         }
4985                 }
4986                 break;
4987         case 2: /* clts */
4988                 handle_clts(vcpu);
4989                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4990                 skip_emulated_instruction(vcpu);
4991                 vmx_fpu_activate(vcpu);
4992                 return 1;
4993         case 1: /*mov from cr*/
4994                 switch (cr) {
4995                 case 3:
4996                         val = kvm_read_cr3(vcpu);
4997                         kvm_register_write(vcpu, reg, val);
4998                         trace_kvm_cr_read(cr, val);
4999                         skip_emulated_instruction(vcpu);
5000                         return 1;
5001                 case 8:
5002                         val = kvm_get_cr8(vcpu);
5003                         kvm_register_write(vcpu, reg, val);
5004                         trace_kvm_cr_read(cr, val);
5005                         skip_emulated_instruction(vcpu);
5006                         return 1;
5007                 }
5008                 break;
5009         case 3: /* lmsw */
5010                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5011                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5012                 kvm_lmsw(vcpu, val);
5013
5014                 skip_emulated_instruction(vcpu);
5015                 return 1;
5016         default:
5017                 break;
5018         }
5019         vcpu->run->exit_reason = 0;
5020         vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5021                (int)(exit_qualification >> 4) & 3, cr);
5022         return 0;
5023 }
5024
5025 static int handle_dr(struct kvm_vcpu *vcpu)
5026 {
5027         unsigned long exit_qualification;
5028         int dr, reg;
5029
5030         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5031         if (!kvm_require_cpl(vcpu, 0))
5032                 return 1;
5033         dr = vmcs_readl(GUEST_DR7);
5034         if (dr & DR7_GD) {
5035                 /*
5036                  * As the vm-exit takes precedence over the debug trap, we
5037                  * need to emulate the latter, either for the host or the
5038                  * guest debugging itself.
5039                  */
5040                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5041                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5042                         vcpu->run->debug.arch.dr7 = dr;
5043                         vcpu->run->debug.arch.pc =
5044                                 vmcs_readl(GUEST_CS_BASE) +
5045                                 vmcs_readl(GUEST_RIP);
5046                         vcpu->run->debug.arch.exception = DB_VECTOR;
5047                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5048                         return 0;
5049                 } else {
5050                         vcpu->arch.dr7 &= ~DR7_GD;
5051                         vcpu->arch.dr6 |= DR6_BD;
5052                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
5053                         kvm_queue_exception(vcpu, DB_VECTOR);
5054                         return 1;
5055                 }
5056         }
5057
5058         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5059         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5060         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5061         if (exit_qualification & TYPE_MOV_FROM_DR) {
5062                 unsigned long val;
5063                 if (!kvm_get_dr(vcpu, dr, &val))
5064                         kvm_register_write(vcpu, reg, val);
5065         } else
5066                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
5067         skip_emulated_instruction(vcpu);
5068         return 1;
5069 }
5070
5071 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5072 {
5073         vmcs_writel(GUEST_DR7, val);
5074 }
5075
5076 static int handle_cpuid(struct kvm_vcpu *vcpu)
5077 {
5078         kvm_emulate_cpuid(vcpu);
5079         return 1;
5080 }
5081
5082 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5083 {
5084         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5085         u64 data;
5086
5087         if (vmx_get_msr(vcpu, ecx, &data)) {
5088                 trace_kvm_msr_read_ex(ecx);
5089                 kvm_inject_gp(vcpu, 0);
5090                 return 1;
5091         }
5092
5093         trace_kvm_msr_read(ecx, data);
5094
5095         /* FIXME: handling of bits 32:63 of rax, rdx */
5096         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
5097         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
5098         skip_emulated_instruction(vcpu);
5099         return 1;
5100 }
5101
5102 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5103 {
5104         struct msr_data msr;
5105         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5106         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5107                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5108
5109         msr.data = data;
5110         msr.index = ecx;
5111         msr.host_initiated = false;
5112         if (vmx_set_msr(vcpu, &msr) != 0) {
5113                 trace_kvm_msr_write_ex(ecx, data);
5114                 kvm_inject_gp(vcpu, 0);
5115                 return 1;
5116         }
5117
5118         trace_kvm_msr_write(ecx, data);
5119         skip_emulated_instruction(vcpu);
5120         return 1;
5121 }
5122
5123 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5124 {
5125         kvm_make_request(KVM_REQ_EVENT, vcpu);
5126         return 1;
5127 }
5128
5129 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5130 {
5131         u32 cpu_based_vm_exec_control;
5132
5133         /* clear pending irq */
5134         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5135         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5136         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5137
5138         kvm_make_request(KVM_REQ_EVENT, vcpu);
5139
5140         ++vcpu->stat.irq_window_exits;
5141
5142         /*
5143          * If the user space waits to inject interrupts, exit as soon as
5144          * possible
5145          */
5146         if (!irqchip_in_kernel(vcpu->kvm) &&
5147             vcpu->run->request_interrupt_window &&
5148             !kvm_cpu_has_interrupt(vcpu)) {
5149                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
5150                 return 0;
5151         }
5152         return 1;
5153 }
5154
5155 static int handle_halt(struct kvm_vcpu *vcpu)
5156 {
5157         skip_emulated_instruction(vcpu);
5158         return kvm_emulate_halt(vcpu);
5159 }
5160
5161 static int handle_vmcall(struct kvm_vcpu *vcpu)
5162 {
5163         skip_emulated_instruction(vcpu);
5164         kvm_emulate_hypercall(vcpu);
5165         return 1;
5166 }
5167
5168 static int handle_invd(struct kvm_vcpu *vcpu)
5169 {
5170         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5171 }
5172
5173 static int handle_invlpg(struct kvm_vcpu *vcpu)
5174 {
5175         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5176
5177         kvm_mmu_invlpg(vcpu, exit_qualification);
5178         skip_emulated_instruction(vcpu);
5179         return 1;
5180 }
5181
5182 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5183 {
5184         int err;
5185
5186         err = kvm_rdpmc(vcpu);
5187         kvm_complete_insn_gp(vcpu, err);
5188
5189         return 1;
5190 }
5191
5192 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5193 {
5194         skip_emulated_instruction(vcpu);
5195         kvm_emulate_wbinvd(vcpu);
5196         return 1;
5197 }
5198
5199 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5200 {
5201         u64 new_bv = kvm_read_edx_eax(vcpu);
5202         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5203
5204         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5205                 skip_emulated_instruction(vcpu);
5206         return 1;
5207 }
5208
5209 static int handle_apic_access(struct kvm_vcpu *vcpu)
5210 {
5211         if (likely(fasteoi)) {
5212                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5213                 int access_type, offset;
5214
5215                 access_type = exit_qualification & APIC_ACCESS_TYPE;
5216                 offset = exit_qualification & APIC_ACCESS_OFFSET;
5217                 /*
5218                  * Sane guest uses MOV to write EOI, with written value
5219                  * not cared. So make a short-circuit here by avoiding
5220                  * heavy instruction emulation.
5221                  */
5222                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5223                     (offset == APIC_EOI)) {
5224                         kvm_lapic_set_eoi(vcpu);
5225                         skip_emulated_instruction(vcpu);
5226                         return 1;
5227                 }
5228         }
5229         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5230 }
5231
5232 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5233 {
5234         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5235         int vector = exit_qualification & 0xff;
5236
5237         /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5238         kvm_apic_set_eoi_accelerated(vcpu, vector);
5239         return 1;
5240 }
5241
5242 static int handle_apic_write(struct kvm_vcpu *vcpu)
5243 {
5244         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5245         u32 offset = exit_qualification & 0xfff;
5246
5247         /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5248         kvm_apic_write_nodecode(vcpu, offset);
5249         return 1;
5250 }
5251
5252 static int handle_task_switch(struct kvm_vcpu *vcpu)
5253 {
5254         struct vcpu_vmx *vmx = to_vmx(vcpu);
5255         unsigned long exit_qualification;
5256         bool has_error_code = false;
5257         u32 error_code = 0;
5258         u16 tss_selector;
5259         int reason, type, idt_v, idt_index;
5260
5261         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5262         idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5263         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5264
5265         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5266
5267         reason = (u32)exit_qualification >> 30;
5268         if (reason == TASK_SWITCH_GATE && idt_v) {
5269                 switch (type) {
5270                 case INTR_TYPE_NMI_INTR:
5271                         vcpu->arch.nmi_injected = false;
5272                         vmx_set_nmi_mask(vcpu, true);
5273                         break;
5274                 case INTR_TYPE_EXT_INTR:
5275                 case INTR_TYPE_SOFT_INTR:
5276                         kvm_clear_interrupt_queue(vcpu);
5277                         break;
5278                 case INTR_TYPE_HARD_EXCEPTION:
5279                         if (vmx->idt_vectoring_info &
5280                             VECTORING_INFO_DELIVER_CODE_MASK) {
5281                                 has_error_code = true;
5282                                 error_code =
5283                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
5284                         }
5285                         /* fall through */
5286                 case INTR_TYPE_SOFT_EXCEPTION:
5287                         kvm_clear_exception_queue(vcpu);
5288                         break;
5289                 default:
5290                         break;
5291                 }
5292         }
5293         tss_selector = exit_qualification;
5294
5295         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5296                        type != INTR_TYPE_EXT_INTR &&
5297                        type != INTR_TYPE_NMI_INTR))
5298                 skip_emulated_instruction(vcpu);
5299
5300         if (kvm_task_switch(vcpu, tss_selector,
5301                             type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5302                             has_error_code, error_code) == EMULATE_FAIL) {
5303                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5304                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5305                 vcpu->run->internal.ndata = 0;
5306                 return 0;
5307         }
5308
5309         /* clear all local breakpoint enable flags */
5310         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
5311
5312         /*
5313          * TODO: What about debug traps on tss switch?
5314          *       Are we supposed to inject them and update dr6?
5315          */
5316
5317         return 1;
5318 }
5319
5320 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5321 {
5322         unsigned long exit_qualification;
5323         gpa_t gpa;
5324         u32 error_code;
5325         int gla_validity;
5326
5327         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5328
5329         gla_validity = (exit_qualification >> 7) & 0x3;
5330         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5331                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5332                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5333                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5334                         vmcs_readl(GUEST_LINEAR_ADDRESS));
5335                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5336                         (long unsigned int)exit_qualification);
5337                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5338                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5339                 return 0;
5340         }
5341
5342         /*
5343          * EPT violation happened while executing iret from NMI,
5344          * "blocked by NMI" bit has to be set before next VM entry.
5345          * There are errata that may cause this bit to not be set:
5346          * AAK134, BY25.
5347          */
5348         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5349                         cpu_has_virtual_nmis() &&
5350                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5351                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5352
5353         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5354         trace_kvm_page_fault(gpa, exit_qualification);
5355
5356         /* It is a write fault? */
5357         error_code = exit_qualification & (1U << 1);
5358         /* It is a fetch fault? */
5359         error_code |= (exit_qualification & (1U << 2)) << 2;
5360         /* ept page table is present? */
5361         error_code |= (exit_qualification >> 3) & 0x1;
5362
5363         vcpu->arch.exit_qualification = exit_qualification;
5364
5365         return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5366 }
5367
5368 static u64 ept_rsvd_mask(u64 spte, int level)
5369 {
5370         int i;
5371         u64 mask = 0;
5372
5373         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5374                 mask |= (1ULL << i);
5375
5376         if (level > 2)
5377                 /* bits 7:3 reserved */
5378                 mask |= 0xf8;
5379         else if (level == 2) {
5380                 if (spte & (1ULL << 7))
5381                         /* 2MB ref, bits 20:12 reserved */
5382                         mask |= 0x1ff000;
5383                 else
5384                         /* bits 6:3 reserved */
5385                         mask |= 0x78;
5386         }
5387
5388         return mask;
5389 }
5390
5391 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5392                                        int level)
5393 {
5394         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5395
5396         /* 010b (write-only) */
5397         WARN_ON((spte & 0x7) == 0x2);
5398
5399         /* 110b (write/execute) */
5400         WARN_ON((spte & 0x7) == 0x6);
5401
5402         /* 100b (execute-only) and value not supported by logical processor */
5403         if (!cpu_has_vmx_ept_execute_only())
5404                 WARN_ON((spte & 0x7) == 0x4);
5405
5406         /* not 000b */
5407         if ((spte & 0x7)) {
5408                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5409
5410                 if (rsvd_bits != 0) {
5411                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5412                                          __func__, rsvd_bits);
5413                         WARN_ON(1);
5414                 }
5415
5416                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
5417                         u64 ept_mem_type = (spte & 0x38) >> 3;
5418
5419                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
5420                             ept_mem_type == 7) {
5421                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5422                                                 __func__, ept_mem_type);
5423                                 WARN_ON(1);
5424                         }
5425                 }
5426         }
5427 }
5428
5429 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5430 {
5431         u64 sptes[4];
5432         int nr_sptes, i, ret;
5433         gpa_t gpa;
5434
5435         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5436
5437         ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5438         if (likely(ret == RET_MMIO_PF_EMULATE))
5439                 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5440                                               EMULATE_DONE;
5441
5442         if (unlikely(ret == RET_MMIO_PF_INVALID))
5443                 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5444
5445         if (unlikely(ret == RET_MMIO_PF_RETRY))
5446                 return 1;
5447
5448         /* It is the real ept misconfig */
5449         printk(KERN_ERR "EPT: Misconfiguration.\n");
5450         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5451
5452         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5453
5454         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5455                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5456
5457         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5458         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5459
5460         return 0;
5461 }
5462
5463 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5464 {
5465         u32 cpu_based_vm_exec_control;
5466
5467         /* clear pending NMI */
5468         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5469         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5470         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5471         ++vcpu->stat.nmi_window_exits;
5472         kvm_make_request(KVM_REQ_EVENT, vcpu);
5473
5474         return 1;
5475 }
5476
5477 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5478 {
5479         struct vcpu_vmx *vmx = to_vmx(vcpu);
5480         enum emulation_result err = EMULATE_DONE;
5481         int ret = 1;
5482         u32 cpu_exec_ctrl;
5483         bool intr_window_requested;
5484         unsigned count = 130;
5485
5486         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5487         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5488
5489         while (!guest_state_valid(vcpu) && count-- != 0) {
5490                 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5491                         return handle_interrupt_window(&vmx->vcpu);
5492
5493                 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5494                         return 1;
5495
5496                 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5497
5498                 if (err == EMULATE_USER_EXIT) {
5499                         ++vcpu->stat.mmio_exits;
5500                         ret = 0;
5501                         goto out;
5502                 }
5503
5504                 if (err != EMULATE_DONE) {
5505                         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5506                         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5507                         vcpu->run->internal.ndata = 0;
5508                         return 0;
5509                 }
5510
5511                 if (vcpu->arch.halt_request) {
5512                         vcpu->arch.halt_request = 0;
5513                         ret = kvm_emulate_halt(vcpu);
5514                         goto out;
5515                 }
5516
5517                 if (signal_pending(current))
5518                         goto out;
5519                 if (need_resched())
5520                         schedule();
5521         }
5522
5523         vmx->emulation_required = emulation_required(vcpu);
5524 out:
5525         return ret;
5526 }
5527
5528 /*
5529  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5530  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5531  */
5532 static int handle_pause(struct kvm_vcpu *vcpu)
5533 {
5534         skip_emulated_instruction(vcpu);
5535         kvm_vcpu_on_spin(vcpu);
5536
5537         return 1;
5538 }
5539
5540 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5541 {
5542         kvm_queue_exception(vcpu, UD_VECTOR);
5543         return 1;
5544 }
5545
5546 /*
5547  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5548  * We could reuse a single VMCS for all the L2 guests, but we also want the
5549  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5550  * allows keeping them loaded on the processor, and in the future will allow
5551  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5552  * every entry if they never change.
5553  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5554  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5555  *
5556  * The following functions allocate and free a vmcs02 in this pool.
5557  */
5558
5559 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5560 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5561 {
5562         struct vmcs02_list *item;
5563         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5564                 if (item->vmptr == vmx->nested.current_vmptr) {
5565                         list_move(&item->list, &vmx->nested.vmcs02_pool);
5566                         return &item->vmcs02;
5567                 }
5568
5569         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5570                 /* Recycle the least recently used VMCS. */
5571                 item = list_entry(vmx->nested.vmcs02_pool.prev,
5572                         struct vmcs02_list, list);
5573                 item->vmptr = vmx->nested.current_vmptr;
5574                 list_move(&item->list, &vmx->nested.vmcs02_pool);
5575                 return &item->vmcs02;
5576         }
5577
5578         /* Create a new VMCS */
5579         item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5580         if (!item)
5581                 return NULL;
5582         item->vmcs02.vmcs = alloc_vmcs();
5583         if (!item->vmcs02.vmcs) {
5584                 kfree(item);
5585                 return NULL;
5586         }
5587         loaded_vmcs_init(&item->vmcs02);
5588         item->vmptr = vmx->nested.current_vmptr;
5589         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5590         vmx->nested.vmcs02_num++;
5591         return &item->vmcs02;
5592 }
5593
5594 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5595 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5596 {
5597         struct vmcs02_list *item;
5598         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5599                 if (item->vmptr == vmptr) {
5600                         free_loaded_vmcs(&item->vmcs02);
5601                         list_del(&item->list);
5602                         kfree(item);
5603                         vmx->nested.vmcs02_num--;
5604                         return;
5605                 }
5606 }
5607
5608 /*
5609  * Free all VMCSs saved for this vcpu, except the one pointed by
5610  * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5611  * currently used, if running L2), and vmcs01 when running L2.
5612  */
5613 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5614 {
5615         struct vmcs02_list *item, *n;
5616         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5617                 if (vmx->loaded_vmcs != &item->vmcs02)
5618                         free_loaded_vmcs(&item->vmcs02);
5619                 list_del(&item->list);
5620                 kfree(item);
5621         }
5622         vmx->nested.vmcs02_num = 0;
5623
5624         if (vmx->loaded_vmcs != &vmx->vmcs01)
5625                 free_loaded_vmcs(&vmx->vmcs01);
5626 }
5627
5628 /*
5629  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5630  * set the success or error code of an emulated VMX instruction, as specified
5631  * by Vol 2B, VMX Instruction Reference, "Conventions".
5632  */
5633 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5634 {
5635         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5636                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5637                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5638 }
5639
5640 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5641 {
5642         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5643                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5644                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5645                         | X86_EFLAGS_CF);
5646 }
5647
5648 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5649                                         u32 vm_instruction_error)
5650 {
5651         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5652                 /*
5653                  * failValid writes the error number to the current VMCS, which
5654                  * can't be done there isn't a current VMCS.
5655                  */
5656                 nested_vmx_failInvalid(vcpu);
5657                 return;
5658         }
5659         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5660                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5661                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5662                         | X86_EFLAGS_ZF);
5663         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5664         /*
5665          * We don't need to force a shadow sync because
5666          * VM_INSTRUCTION_ERROR is not shadowed
5667          */
5668 }
5669
5670 /*
5671  * Emulate the VMXON instruction.
5672  * Currently, we just remember that VMX is active, and do not save or even
5673  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5674  * do not currently need to store anything in that guest-allocated memory
5675  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5676  * argument is different from the VMXON pointer (which the spec says they do).
5677  */
5678 static int handle_vmon(struct kvm_vcpu *vcpu)
5679 {
5680         struct kvm_segment cs;
5681         struct vcpu_vmx *vmx = to_vmx(vcpu);
5682         struct vmcs *shadow_vmcs;
5683         const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
5684                 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
5685
5686         /* The Intel VMX Instruction Reference lists a bunch of bits that
5687          * are prerequisite to running VMXON, most notably cr4.VMXE must be
5688          * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5689          * Otherwise, we should fail with #UD. We test these now:
5690          */
5691         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5692             !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5693             (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5694                 kvm_queue_exception(vcpu, UD_VECTOR);
5695                 return 1;
5696         }
5697
5698         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5699         if (is_long_mode(vcpu) && !cs.l) {
5700                 kvm_queue_exception(vcpu, UD_VECTOR);
5701                 return 1;
5702         }
5703
5704         if (vmx_get_cpl(vcpu)) {
5705                 kvm_inject_gp(vcpu, 0);
5706                 return 1;
5707         }
5708         if (vmx->nested.vmxon) {
5709                 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
5710                 skip_emulated_instruction(vcpu);
5711                 return 1;
5712         }
5713
5714         if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
5715                         != VMXON_NEEDED_FEATURES) {
5716                 kvm_inject_gp(vcpu, 0);
5717                 return 1;
5718         }
5719
5720         if (enable_shadow_vmcs) {
5721                 shadow_vmcs = alloc_vmcs();
5722                 if (!shadow_vmcs)
5723                         return -ENOMEM;
5724                 /* mark vmcs as shadow */
5725                 shadow_vmcs->revision_id |= (1u << 31);
5726                 /* init shadow vmcs */
5727                 vmcs_clear(shadow_vmcs);
5728                 vmx->nested.current_shadow_vmcs = shadow_vmcs;
5729         }
5730
5731         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5732         vmx->nested.vmcs02_num = 0;
5733
5734         vmx->nested.vmxon = true;
5735
5736         skip_emulated_instruction(vcpu);
5737         nested_vmx_succeed(vcpu);
5738         return 1;
5739 }
5740
5741 /*
5742  * Intel's VMX Instruction Reference specifies a common set of prerequisites
5743  * for running VMX instructions (except VMXON, whose prerequisites are
5744  * slightly different). It also specifies what exception to inject otherwise.
5745  */
5746 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5747 {
5748         struct kvm_segment cs;
5749         struct vcpu_vmx *vmx = to_vmx(vcpu);
5750
5751         if (!vmx->nested.vmxon) {
5752                 kvm_queue_exception(vcpu, UD_VECTOR);
5753                 return 0;
5754         }
5755
5756         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5757         if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5758             (is_long_mode(vcpu) && !cs.l)) {
5759                 kvm_queue_exception(vcpu, UD_VECTOR);
5760                 return 0;
5761         }
5762
5763         if (vmx_get_cpl(vcpu)) {
5764                 kvm_inject_gp(vcpu, 0);
5765                 return 0;
5766         }
5767
5768         return 1;
5769 }
5770
5771 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
5772 {
5773         u32 exec_control;
5774         if (enable_shadow_vmcs) {
5775                 if (vmx->nested.current_vmcs12 != NULL) {
5776                         /* copy to memory all shadowed fields in case
5777                            they were modified */
5778                         copy_shadow_to_vmcs12(vmx);
5779                         vmx->nested.sync_shadow_vmcs = false;
5780                         exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5781                         exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5782                         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
5783                         vmcs_write64(VMCS_LINK_POINTER, -1ull);
5784                 }
5785         }
5786         kunmap(vmx->nested.current_vmcs12_page);
5787         nested_release_page(vmx->nested.current_vmcs12_page);
5788 }
5789
5790 /*
5791  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5792  * just stops using VMX.
5793  */
5794 static void free_nested(struct vcpu_vmx *vmx)
5795 {
5796         if (!vmx->nested.vmxon)
5797                 return;
5798         vmx->nested.vmxon = false;
5799         if (vmx->nested.current_vmptr != -1ull) {
5800                 nested_release_vmcs12(vmx);
5801                 vmx->nested.current_vmptr = -1ull;
5802                 vmx->nested.current_vmcs12 = NULL;
5803         }
5804         if (enable_shadow_vmcs)
5805                 free_vmcs(vmx->nested.current_shadow_vmcs);
5806         /* Unpin physical memory we referred to in current vmcs02 */
5807         if (vmx->nested.apic_access_page) {
5808                 nested_release_page(vmx->nested.apic_access_page);
5809                 vmx->nested.apic_access_page = 0;
5810         }
5811
5812         nested_free_all_saved_vmcss(vmx);
5813 }
5814
5815 /* Emulate the VMXOFF instruction */
5816 static int handle_vmoff(struct kvm_vcpu *vcpu)
5817 {
5818         if (!nested_vmx_check_permission(vcpu))
5819                 return 1;
5820         free_nested(to_vmx(vcpu));
5821         skip_emulated_instruction(vcpu);
5822         nested_vmx_succeed(vcpu);
5823         return 1;
5824 }
5825
5826 /*
5827  * Decode the memory-address operand of a vmx instruction, as recorded on an
5828  * exit caused by such an instruction (run by a guest hypervisor).
5829  * On success, returns 0. When the operand is invalid, returns 1 and throws
5830  * #UD or #GP.
5831  */
5832 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5833                                  unsigned long exit_qualification,
5834                                  u32 vmx_instruction_info, gva_t *ret)
5835 {
5836         /*
5837          * According to Vol. 3B, "Information for VM Exits Due to Instruction
5838          * Execution", on an exit, vmx_instruction_info holds most of the
5839          * addressing components of the operand. Only the displacement part
5840          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5841          * For how an actual address is calculated from all these components,
5842          * refer to Vol. 1, "Operand Addressing".
5843          */
5844         int  scaling = vmx_instruction_info & 3;
5845         int  addr_size = (vmx_instruction_info >> 7) & 7;
5846         bool is_reg = vmx_instruction_info & (1u << 10);
5847         int  seg_reg = (vmx_instruction_info >> 15) & 7;
5848         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
5849         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5850         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
5851         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
5852
5853         if (is_reg) {
5854                 kvm_queue_exception(vcpu, UD_VECTOR);
5855                 return 1;
5856         }
5857
5858         /* Addr = segment_base + offset */
5859         /* offset = base + [index * scale] + displacement */
5860         *ret = vmx_get_segment_base(vcpu, seg_reg);
5861         if (base_is_valid)
5862                 *ret += kvm_register_read(vcpu, base_reg);
5863         if (index_is_valid)
5864                 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5865         *ret += exit_qualification; /* holds the displacement */
5866
5867         if (addr_size == 1) /* 32 bit */
5868                 *ret &= 0xffffffff;
5869
5870         /*
5871          * TODO: throw #GP (and return 1) in various cases that the VM*
5872          * instructions require it - e.g., offset beyond segment limit,
5873          * unusable or unreadable/unwritable segment, non-canonical 64-bit
5874          * address, and so on. Currently these are not checked.
5875          */
5876         return 0;
5877 }
5878
5879 /* Emulate the VMCLEAR instruction */
5880 static int handle_vmclear(struct kvm_vcpu *vcpu)
5881 {
5882         struct vcpu_vmx *vmx = to_vmx(vcpu);
5883         gva_t gva;
5884         gpa_t vmptr;
5885         struct vmcs12 *vmcs12;
5886         struct page *page;
5887         struct x86_exception e;
5888
5889         if (!nested_vmx_check_permission(vcpu))
5890                 return 1;
5891
5892         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5893                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5894                 return 1;
5895
5896         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5897                                 sizeof(vmptr), &e)) {
5898                 kvm_inject_page_fault(vcpu, &e);
5899                 return 1;
5900         }
5901
5902         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5903                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5904                 skip_emulated_instruction(vcpu);
5905                 return 1;
5906         }
5907
5908         if (vmptr == vmx->nested.current_vmptr) {
5909                 nested_release_vmcs12(vmx);
5910                 vmx->nested.current_vmptr = -1ull;
5911                 vmx->nested.current_vmcs12 = NULL;
5912         }
5913
5914         page = nested_get_page(vcpu, vmptr);
5915         if (page == NULL) {
5916                 /*
5917                  * For accurate processor emulation, VMCLEAR beyond available
5918                  * physical memory should do nothing at all. However, it is
5919                  * possible that a nested vmx bug, not a guest hypervisor bug,
5920                  * resulted in this case, so let's shut down before doing any
5921                  * more damage:
5922                  */
5923                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5924                 return 1;
5925         }
5926         vmcs12 = kmap(page);
5927         vmcs12->launch_state = 0;
5928         kunmap(page);
5929         nested_release_page(page);
5930
5931         nested_free_vmcs02(vmx, vmptr);
5932
5933         skip_emulated_instruction(vcpu);
5934         nested_vmx_succeed(vcpu);
5935         return 1;
5936 }
5937
5938 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5939
5940 /* Emulate the VMLAUNCH instruction */
5941 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5942 {
5943         return nested_vmx_run(vcpu, true);
5944 }
5945
5946 /* Emulate the VMRESUME instruction */
5947 static int handle_vmresume(struct kvm_vcpu *vcpu)
5948 {
5949
5950         return nested_vmx_run(vcpu, false);
5951 }
5952
5953 enum vmcs_field_type {
5954         VMCS_FIELD_TYPE_U16 = 0,
5955         VMCS_FIELD_TYPE_U64 = 1,
5956         VMCS_FIELD_TYPE_U32 = 2,
5957         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5958 };
5959
5960 static inline int vmcs_field_type(unsigned long field)
5961 {
5962         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
5963                 return VMCS_FIELD_TYPE_U32;
5964         return (field >> 13) & 0x3 ;
5965 }
5966
5967 static inline int vmcs_field_readonly(unsigned long field)
5968 {
5969         return (((field >> 10) & 0x3) == 1);
5970 }
5971
5972 /*
5973  * Read a vmcs12 field. Since these can have varying lengths and we return
5974  * one type, we chose the biggest type (u64) and zero-extend the return value
5975  * to that size. Note that the caller, handle_vmread, might need to use only
5976  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5977  * 64-bit fields are to be returned).
5978  */
5979 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5980                                         unsigned long field, u64 *ret)
5981 {
5982         short offset = vmcs_field_to_offset(field);
5983         char *p;
5984
5985         if (offset < 0)
5986                 return 0;
5987
5988         p = ((char *)(get_vmcs12(vcpu))) + offset;
5989
5990         switch (vmcs_field_type(field)) {
5991         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5992                 *ret = *((natural_width *)p);
5993                 return 1;
5994         case VMCS_FIELD_TYPE_U16:
5995                 *ret = *((u16 *)p);
5996                 return 1;
5997         case VMCS_FIELD_TYPE_U32:
5998                 *ret = *((u32 *)p);
5999                 return 1;
6000         case VMCS_FIELD_TYPE_U64:
6001                 *ret = *((u64 *)p);
6002                 return 1;
6003         default:
6004                 return 0; /* can never happen. */
6005         }
6006 }
6007
6008
6009 static inline bool vmcs12_write_any(struct kvm_vcpu *vcpu,
6010                                     unsigned long field, u64 field_value){
6011         short offset = vmcs_field_to_offset(field);
6012         char *p = ((char *) get_vmcs12(vcpu)) + offset;
6013         if (offset < 0)
6014                 return false;
6015
6016         switch (vmcs_field_type(field)) {
6017         case VMCS_FIELD_TYPE_U16:
6018                 *(u16 *)p = field_value;
6019                 return true;
6020         case VMCS_FIELD_TYPE_U32:
6021                 *(u32 *)p = field_value;
6022                 return true;
6023         case VMCS_FIELD_TYPE_U64:
6024                 *(u64 *)p = field_value;
6025                 return true;
6026         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6027                 *(natural_width *)p = field_value;
6028                 return true;
6029         default:
6030                 return false; /* can never happen. */
6031         }
6032
6033 }
6034
6035 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6036 {
6037         int i;
6038         unsigned long field;
6039         u64 field_value;
6040         struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6041         const unsigned long *fields = shadow_read_write_fields;
6042         const int num_fields = max_shadow_read_write_fields;
6043
6044         vmcs_load(shadow_vmcs);
6045
6046         for (i = 0; i < num_fields; i++) {
6047                 field = fields[i];
6048                 switch (vmcs_field_type(field)) {
6049                 case VMCS_FIELD_TYPE_U16:
6050                         field_value = vmcs_read16(field);
6051                         break;
6052                 case VMCS_FIELD_TYPE_U32:
6053                         field_value = vmcs_read32(field);
6054                         break;
6055                 case VMCS_FIELD_TYPE_U64:
6056                         field_value = vmcs_read64(field);
6057                         break;
6058                 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6059                         field_value = vmcs_readl(field);
6060                         break;
6061                 }
6062                 vmcs12_write_any(&vmx->vcpu, field, field_value);
6063         }
6064
6065         vmcs_clear(shadow_vmcs);
6066         vmcs_load(vmx->loaded_vmcs->vmcs);
6067 }
6068
6069 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
6070 {
6071         const unsigned long *fields[] = {
6072                 shadow_read_write_fields,
6073                 shadow_read_only_fields
6074         };
6075         const int max_fields[] = {
6076                 max_shadow_read_write_fields,
6077                 max_shadow_read_only_fields
6078         };
6079         int i, q;
6080         unsigned long field;
6081         u64 field_value = 0;
6082         struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6083
6084         vmcs_load(shadow_vmcs);
6085
6086         for (q = 0; q < ARRAY_SIZE(fields); q++) {
6087                 for (i = 0; i < max_fields[q]; i++) {
6088                         field = fields[q][i];
6089                         vmcs12_read_any(&vmx->vcpu, field, &field_value);
6090
6091                         switch (vmcs_field_type(field)) {
6092                         case VMCS_FIELD_TYPE_U16:
6093                                 vmcs_write16(field, (u16)field_value);
6094                                 break;
6095                         case VMCS_FIELD_TYPE_U32:
6096                                 vmcs_write32(field, (u32)field_value);
6097                                 break;
6098                         case VMCS_FIELD_TYPE_U64:
6099                                 vmcs_write64(field, (u64)field_value);
6100                                 break;
6101                         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6102                                 vmcs_writel(field, (long)field_value);
6103                                 break;
6104                         }
6105                 }
6106         }
6107
6108         vmcs_clear(shadow_vmcs);
6109         vmcs_load(vmx->loaded_vmcs->vmcs);
6110 }
6111
6112 /*
6113  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
6114  * used before) all generate the same failure when it is missing.
6115  */
6116 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
6117 {
6118         struct vcpu_vmx *vmx = to_vmx(vcpu);
6119         if (vmx->nested.current_vmptr == -1ull) {
6120                 nested_vmx_failInvalid(vcpu);
6121                 skip_emulated_instruction(vcpu);
6122                 return 0;
6123         }
6124         return 1;
6125 }
6126
6127 static int handle_vmread(struct kvm_vcpu *vcpu)
6128 {
6129         unsigned long field;
6130         u64 field_value;
6131         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6132         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6133         gva_t gva = 0;
6134
6135         if (!nested_vmx_check_permission(vcpu) ||
6136             !nested_vmx_check_vmcs12(vcpu))
6137                 return 1;
6138
6139         /* Decode instruction info and find the field to read */
6140         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6141         /* Read the field, zero-extended to a u64 field_value */
6142         if (!vmcs12_read_any(vcpu, field, &field_value)) {
6143                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6144                 skip_emulated_instruction(vcpu);
6145                 return 1;
6146         }
6147         /*
6148          * Now copy part of this value to register or memory, as requested.
6149          * Note that the number of bits actually copied is 32 or 64 depending
6150          * on the guest's mode (32 or 64 bit), not on the given field's length.
6151          */
6152         if (vmx_instruction_info & (1u << 10)) {
6153                 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
6154                         field_value);
6155         } else {
6156                 if (get_vmx_mem_address(vcpu, exit_qualification,
6157                                 vmx_instruction_info, &gva))
6158                         return 1;
6159                 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
6160                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
6161                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
6162         }
6163
6164         nested_vmx_succeed(vcpu);
6165         skip_emulated_instruction(vcpu);
6166         return 1;
6167 }
6168
6169
6170 static int handle_vmwrite(struct kvm_vcpu *vcpu)
6171 {
6172         unsigned long field;
6173         gva_t gva;
6174         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6175         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6176         /* The value to write might be 32 or 64 bits, depending on L1's long
6177          * mode, and eventually we need to write that into a field of several
6178          * possible lengths. The code below first zero-extends the value to 64
6179          * bit (field_value), and then copies only the approriate number of
6180          * bits into the vmcs12 field.
6181          */
6182         u64 field_value = 0;
6183         struct x86_exception e;
6184
6185         if (!nested_vmx_check_permission(vcpu) ||
6186             !nested_vmx_check_vmcs12(vcpu))
6187                 return 1;
6188
6189         if (vmx_instruction_info & (1u << 10))
6190                 field_value = kvm_register_read(vcpu,
6191                         (((vmx_instruction_info) >> 3) & 0xf));
6192         else {
6193                 if (get_vmx_mem_address(vcpu, exit_qualification,
6194                                 vmx_instruction_info, &gva))
6195                         return 1;
6196                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
6197                            &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
6198                         kvm_inject_page_fault(vcpu, &e);
6199                         return 1;
6200                 }
6201         }
6202
6203
6204         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6205         if (vmcs_field_readonly(field)) {
6206                 nested_vmx_failValid(vcpu,
6207                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
6208                 skip_emulated_instruction(vcpu);
6209                 return 1;
6210         }
6211
6212         if (!vmcs12_write_any(vcpu, field, field_value)) {
6213                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6214                 skip_emulated_instruction(vcpu);
6215                 return 1;
6216         }
6217
6218         nested_vmx_succeed(vcpu);
6219         skip_emulated_instruction(vcpu);
6220         return 1;
6221 }
6222
6223 /* Emulate the VMPTRLD instruction */
6224 static int handle_vmptrld(struct kvm_vcpu *vcpu)
6225 {
6226         struct vcpu_vmx *vmx = to_vmx(vcpu);
6227         gva_t gva;
6228         gpa_t vmptr;
6229         struct x86_exception e;
6230         u32 exec_control;
6231
6232         if (!nested_vmx_check_permission(vcpu))
6233                 return 1;
6234
6235         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6236                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
6237                 return 1;
6238
6239         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6240                                 sizeof(vmptr), &e)) {
6241                 kvm_inject_page_fault(vcpu, &e);
6242                 return 1;
6243         }
6244
6245         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
6246                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
6247                 skip_emulated_instruction(vcpu);
6248                 return 1;
6249         }
6250
6251         if (vmx->nested.current_vmptr != vmptr) {
6252                 struct vmcs12 *new_vmcs12;
6253                 struct page *page;
6254                 page = nested_get_page(vcpu, vmptr);
6255                 if (page == NULL) {
6256                         nested_vmx_failInvalid(vcpu);
6257                         skip_emulated_instruction(vcpu);
6258                         return 1;
6259                 }
6260                 new_vmcs12 = kmap(page);
6261                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
6262                         kunmap(page);
6263                         nested_release_page_clean(page);
6264                         nested_vmx_failValid(vcpu,
6265                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
6266                         skip_emulated_instruction(vcpu);
6267                         return 1;
6268                 }
6269                 if (vmx->nested.current_vmptr != -1ull)
6270                         nested_release_vmcs12(vmx);
6271
6272                 vmx->nested.current_vmptr = vmptr;
6273                 vmx->nested.current_vmcs12 = new_vmcs12;
6274                 vmx->nested.current_vmcs12_page = page;
6275                 if (enable_shadow_vmcs) {
6276                         exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6277                         exec_control |= SECONDARY_EXEC_SHADOW_VMCS;
6278                         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6279                         vmcs_write64(VMCS_LINK_POINTER,
6280                                      __pa(vmx->nested.current_shadow_vmcs));
6281                         vmx->nested.sync_shadow_vmcs = true;
6282                 }
6283         }
6284
6285         nested_vmx_succeed(vcpu);
6286         skip_emulated_instruction(vcpu);
6287         return 1;
6288 }
6289
6290 /* Emulate the VMPTRST instruction */
6291 static int handle_vmptrst(struct kvm_vcpu *vcpu)
6292 {
6293         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6294         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6295         gva_t vmcs_gva;
6296         struct x86_exception e;
6297
6298         if (!nested_vmx_check_permission(vcpu))
6299                 return 1;
6300
6301         if (get_vmx_mem_address(vcpu, exit_qualification,
6302                         vmx_instruction_info, &vmcs_gva))
6303                 return 1;
6304         /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
6305         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
6306                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
6307                                  sizeof(u64), &e)) {
6308                 kvm_inject_page_fault(vcpu, &e);
6309                 return 1;
6310         }
6311         nested_vmx_succeed(vcpu);
6312         skip_emulated_instruction(vcpu);
6313         return 1;
6314 }
6315
6316 /* Emulate the INVEPT instruction */
6317 static int handle_invept(struct kvm_vcpu *vcpu)
6318 {
6319         u32 vmx_instruction_info, types;
6320         unsigned long type;
6321         gva_t gva;
6322         struct x86_exception e;
6323         struct {
6324                 u64 eptp, gpa;
6325         } operand;
6326         u64 eptp_mask = ((1ull << 51) - 1) & PAGE_MASK;
6327
6328         if (!(nested_vmx_secondary_ctls_high & SECONDARY_EXEC_ENABLE_EPT) ||
6329             !(nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
6330                 kvm_queue_exception(vcpu, UD_VECTOR);
6331                 return 1;
6332         }
6333
6334         if (!nested_vmx_check_permission(vcpu))
6335                 return 1;
6336
6337         if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
6338                 kvm_queue_exception(vcpu, UD_VECTOR);
6339                 return 1;
6340         }
6341
6342         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6343         type = kvm_register_read(vcpu, (vmx_instruction_info >> 28) & 0xf);
6344
6345         types = (nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
6346
6347         if (!(types & (1UL << type))) {
6348                 nested_vmx_failValid(vcpu,
6349                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6350                 return 1;
6351         }
6352
6353         /* According to the Intel VMX instruction reference, the memory
6354          * operand is read even if it isn't needed (e.g., for type==global)
6355          */
6356         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6357                         vmx_instruction_info, &gva))
6358                 return 1;
6359         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
6360                                 sizeof(operand), &e)) {
6361                 kvm_inject_page_fault(vcpu, &e);
6362                 return 1;
6363         }
6364
6365         switch (type) {
6366         case VMX_EPT_EXTENT_CONTEXT:
6367                 if ((operand.eptp & eptp_mask) !=
6368                                 (nested_ept_get_cr3(vcpu) & eptp_mask))
6369                         break;
6370         case VMX_EPT_EXTENT_GLOBAL:
6371                 kvm_mmu_sync_roots(vcpu);
6372                 kvm_mmu_flush_tlb(vcpu);
6373                 nested_vmx_succeed(vcpu);
6374                 break;
6375         default:
6376                 BUG_ON(1);
6377                 break;
6378         }
6379
6380         skip_emulated_instruction(vcpu);
6381         return 1;
6382 }
6383
6384 /*
6385  * The exit handlers return 1 if the exit was handled fully and guest execution
6386  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
6387  * to be done to userspace and return 0.
6388  */
6389 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6390         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
6391         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
6392         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
6393         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
6394         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
6395         [EXIT_REASON_CR_ACCESS]               = handle_cr,
6396         [EXIT_REASON_DR_ACCESS]               = handle_dr,
6397         [EXIT_REASON_CPUID]                   = handle_cpuid,
6398         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
6399         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
6400         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
6401         [EXIT_REASON_HLT]                     = handle_halt,
6402         [EXIT_REASON_INVD]                    = handle_invd,
6403         [EXIT_REASON_INVLPG]                  = handle_invlpg,
6404         [EXIT_REASON_RDPMC]                   = handle_rdpmc,
6405         [EXIT_REASON_VMCALL]                  = handle_vmcall,
6406         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
6407         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
6408         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
6409         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
6410         [EXIT_REASON_VMREAD]                  = handle_vmread,
6411         [EXIT_REASON_VMRESUME]                = handle_vmresume,
6412         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
6413         [EXIT_REASON_VMOFF]                   = handle_vmoff,
6414         [EXIT_REASON_VMON]                    = handle_vmon,
6415         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
6416         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
6417         [EXIT_REASON_APIC_WRITE]              = handle_apic_write,
6418         [EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
6419         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
6420         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
6421         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
6422         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
6423         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
6424         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
6425         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
6426         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
6427         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
6428         [EXIT_REASON_INVEPT]                  = handle_invept,
6429 };
6430
6431 static const int kvm_vmx_max_exit_handlers =
6432         ARRAY_SIZE(kvm_vmx_exit_handlers);
6433
6434 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6435                                        struct vmcs12 *vmcs12)
6436 {
6437         unsigned long exit_qualification;
6438         gpa_t bitmap, last_bitmap;
6439         unsigned int port;
6440         int size;
6441         u8 b;
6442
6443         if (nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING))
6444                 return 1;
6445
6446         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6447                 return 0;
6448
6449         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6450
6451         port = exit_qualification >> 16;
6452         size = (exit_qualification & 7) + 1;
6453
6454         last_bitmap = (gpa_t)-1;
6455         b = -1;
6456
6457         while (size > 0) {
6458                 if (port < 0x8000)
6459                         bitmap = vmcs12->io_bitmap_a;
6460                 else if (port < 0x10000)
6461                         bitmap = vmcs12->io_bitmap_b;
6462                 else
6463                         return 1;
6464                 bitmap += (port & 0x7fff) / 8;
6465
6466                 if (last_bitmap != bitmap)
6467                         if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
6468                                 return 1;
6469                 if (b & (1 << (port & 7)))
6470                         return 1;
6471
6472                 port++;
6473                 size--;
6474                 last_bitmap = bitmap;
6475         }
6476
6477         return 0;
6478 }
6479
6480 /*
6481  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
6482  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6483  * disinterest in the current event (read or write a specific MSR) by using an
6484  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6485  */
6486 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6487         struct vmcs12 *vmcs12, u32 exit_reason)
6488 {
6489         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
6490         gpa_t bitmap;
6491
6492         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6493                 return 1;
6494
6495         /*
6496          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6497          * for the four combinations of read/write and low/high MSR numbers.
6498          * First we need to figure out which of the four to use:
6499          */
6500         bitmap = vmcs12->msr_bitmap;
6501         if (exit_reason == EXIT_REASON_MSR_WRITE)
6502                 bitmap += 2048;
6503         if (msr_index >= 0xc0000000) {
6504                 msr_index -= 0xc0000000;
6505                 bitmap += 1024;
6506         }
6507
6508         /* Then read the msr_index'th bit from this bitmap: */
6509         if (msr_index < 1024*8) {
6510                 unsigned char b;
6511                 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
6512                         return 1;
6513                 return 1 & (b >> (msr_index & 7));
6514         } else
6515                 return 1; /* let L1 handle the wrong parameter */
6516 }
6517
6518 /*
6519  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6520  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6521  * intercept (via guest_host_mask etc.) the current event.
6522  */
6523 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6524         struct vmcs12 *vmcs12)
6525 {
6526         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6527         int cr = exit_qualification & 15;
6528         int reg = (exit_qualification >> 8) & 15;
6529         unsigned long val = kvm_register_read(vcpu, reg);
6530
6531         switch ((exit_qualification >> 4) & 3) {
6532         case 0: /* mov to cr */
6533                 switch (cr) {
6534                 case 0:
6535                         if (vmcs12->cr0_guest_host_mask &
6536                             (val ^ vmcs12->cr0_read_shadow))
6537                                 return 1;
6538                         break;
6539                 case 3:
6540                         if ((vmcs12->cr3_target_count >= 1 &&
6541                                         vmcs12->cr3_target_value0 == val) ||
6542                                 (vmcs12->cr3_target_count >= 2 &&
6543                                         vmcs12->cr3_target_value1 == val) ||
6544                                 (vmcs12->cr3_target_count >= 3 &&
6545                                         vmcs12->cr3_target_value2 == val) ||
6546                                 (vmcs12->cr3_target_count >= 4 &&
6547                                         vmcs12->cr3_target_value3 == val))
6548                                 return 0;
6549                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6550                                 return 1;
6551                         break;
6552                 case 4:
6553                         if (vmcs12->cr4_guest_host_mask &
6554                             (vmcs12->cr4_read_shadow ^ val))
6555                                 return 1;
6556                         break;
6557                 case 8:
6558                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6559                                 return 1;
6560                         break;
6561                 }
6562                 break;
6563         case 2: /* clts */
6564                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6565                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
6566                         return 1;
6567                 break;
6568         case 1: /* mov from cr */
6569                 switch (cr) {
6570                 case 3:
6571                         if (vmcs12->cpu_based_vm_exec_control &
6572                             CPU_BASED_CR3_STORE_EXITING)
6573                                 return 1;
6574                         break;
6575                 case 8:
6576                         if (vmcs12->cpu_based_vm_exec_control &
6577                             CPU_BASED_CR8_STORE_EXITING)
6578                                 return 1;
6579                         break;
6580                 }
6581                 break;
6582         case 3: /* lmsw */
6583                 /*
6584                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6585                  * cr0. Other attempted changes are ignored, with no exit.
6586                  */
6587                 if (vmcs12->cr0_guest_host_mask & 0xe &
6588                     (val ^ vmcs12->cr0_read_shadow))
6589                         return 1;
6590                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6591                     !(vmcs12->cr0_read_shadow & 0x1) &&
6592                     (val & 0x1))
6593                         return 1;
6594                 break;
6595         }
6596         return 0;
6597 }
6598
6599 /*
6600  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
6601  * should handle it ourselves in L0 (and then continue L2). Only call this
6602  * when in is_guest_mode (L2).
6603  */
6604 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
6605 {
6606         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6607         struct vcpu_vmx *vmx = to_vmx(vcpu);
6608         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6609         u32 exit_reason = vmx->exit_reason;
6610
6611         if (vmx->nested.nested_run_pending)
6612                 return 0;
6613
6614         if (unlikely(vmx->fail)) {
6615                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
6616                                     vmcs_read32(VM_INSTRUCTION_ERROR));
6617                 return 1;
6618         }
6619
6620         switch (exit_reason) {
6621         case EXIT_REASON_EXCEPTION_NMI:
6622                 if (!is_exception(intr_info))
6623                         return 0;
6624                 else if (is_page_fault(intr_info))
6625                         return enable_ept;
6626                 return vmcs12->exception_bitmap &
6627                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
6628         case EXIT_REASON_EXTERNAL_INTERRUPT:
6629                 return 0;
6630         case EXIT_REASON_TRIPLE_FAULT:
6631                 return 1;
6632         case EXIT_REASON_PENDING_INTERRUPT:
6633                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
6634         case EXIT_REASON_NMI_WINDOW:
6635                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
6636         case EXIT_REASON_TASK_SWITCH:
6637                 return 1;
6638         case EXIT_REASON_CPUID:
6639                 return 1;
6640         case EXIT_REASON_HLT:
6641                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6642         case EXIT_REASON_INVD:
6643                 return 1;
6644         case EXIT_REASON_INVLPG:
6645                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6646         case EXIT_REASON_RDPMC:
6647                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6648         case EXIT_REASON_RDTSC:
6649                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6650         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6651         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6652         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
6653         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
6654         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6655         case EXIT_REASON_INVEPT:
6656                 /*
6657                  * VMX instructions trap unconditionally. This allows L1 to
6658                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
6659                  */
6660                 return 1;
6661         case EXIT_REASON_CR_ACCESS:
6662                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6663         case EXIT_REASON_DR_ACCESS:
6664                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6665         case EXIT_REASON_IO_INSTRUCTION:
6666                 return nested_vmx_exit_handled_io(vcpu, vmcs12);
6667         case EXIT_REASON_MSR_READ:
6668         case EXIT_REASON_MSR_WRITE:
6669                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6670         case EXIT_REASON_INVALID_STATE:
6671                 return 1;
6672         case EXIT_REASON_MWAIT_INSTRUCTION:
6673                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6674         case EXIT_REASON_MONITOR_INSTRUCTION:
6675                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6676         case EXIT_REASON_PAUSE_INSTRUCTION:
6677                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6678                         nested_cpu_has2(vmcs12,
6679                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6680         case EXIT_REASON_MCE_DURING_VMENTRY:
6681                 return 0;
6682         case EXIT_REASON_TPR_BELOW_THRESHOLD:
6683                 return 1;
6684         case EXIT_REASON_APIC_ACCESS:
6685                 return nested_cpu_has2(vmcs12,
6686                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
6687         case EXIT_REASON_EPT_VIOLATION:
6688                 /*
6689                  * L0 always deals with the EPT violation. If nested EPT is
6690                  * used, and the nested mmu code discovers that the address is
6691                  * missing in the guest EPT table (EPT12), the EPT violation
6692                  * will be injected with nested_ept_inject_page_fault()
6693                  */
6694                 return 0;
6695         case EXIT_REASON_EPT_MISCONFIG:
6696                 /*
6697                  * L2 never uses directly L1's EPT, but rather L0's own EPT
6698                  * table (shadow on EPT) or a merged EPT table that L0 built
6699                  * (EPT on EPT). So any problems with the structure of the
6700                  * table is L0's fault.
6701                  */
6702                 return 0;
6703         case EXIT_REASON_PREEMPTION_TIMER:
6704                 return vmcs12->pin_based_vm_exec_control &
6705                         PIN_BASED_VMX_PREEMPTION_TIMER;
6706         case EXIT_REASON_WBINVD:
6707                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6708         case EXIT_REASON_XSETBV:
6709                 return 1;
6710         default:
6711                 return 1;
6712         }
6713 }
6714
6715 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
6716 {
6717         *info1 = vmcs_readl(EXIT_QUALIFICATION);
6718         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
6719 }
6720
6721 /*
6722  * The guest has exited.  See if we can fix it or if we need userspace
6723  * assistance.
6724  */
6725 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
6726 {
6727         struct vcpu_vmx *vmx = to_vmx(vcpu);
6728         u32 exit_reason = vmx->exit_reason;
6729         u32 vectoring_info = vmx->idt_vectoring_info;
6730
6731         /* If guest state is invalid, start emulating */
6732         if (vmx->emulation_required)
6733                 return handle_invalid_guest_state(vcpu);
6734
6735         /*
6736          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
6737          * we did not inject a still-pending event to L1 now because of
6738          * nested_run_pending, we need to re-enable this bit.
6739          */
6740         if (vmx->nested.nested_run_pending)
6741                 kvm_make_request(KVM_REQ_EVENT, vcpu);
6742
6743         if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
6744             exit_reason == EXIT_REASON_VMRESUME))
6745                 vmx->nested.nested_run_pending = 1;
6746         else
6747                 vmx->nested.nested_run_pending = 0;
6748
6749         if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
6750                 nested_vmx_vmexit(vcpu);
6751                 return 1;
6752         }
6753
6754         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
6755                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6756                 vcpu->run->fail_entry.hardware_entry_failure_reason
6757                         = exit_reason;
6758                 return 0;
6759         }
6760
6761         if (unlikely(vmx->fail)) {
6762                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6763                 vcpu->run->fail_entry.hardware_entry_failure_reason
6764                         = vmcs_read32(VM_INSTRUCTION_ERROR);
6765                 return 0;
6766         }
6767
6768         /*
6769          * Note:
6770          * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6771          * delivery event since it indicates guest is accessing MMIO.
6772          * The vm-exit can be triggered again after return to guest that
6773          * will cause infinite loop.
6774          */
6775         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6776                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
6777                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
6778                         exit_reason != EXIT_REASON_TASK_SWITCH)) {
6779                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6780                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6781                 vcpu->run->internal.ndata = 2;
6782                 vcpu->run->internal.data[0] = vectoring_info;
6783                 vcpu->run->internal.data[1] = exit_reason;
6784                 return 0;
6785         }
6786
6787         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
6788             !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
6789                                         get_vmcs12(vcpu))))) {
6790                 if (vmx_interrupt_allowed(vcpu)) {
6791                         vmx->soft_vnmi_blocked = 0;
6792                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
6793                            vcpu->arch.nmi_pending) {
6794                         /*
6795                          * This CPU don't support us in finding the end of an
6796                          * NMI-blocked window if the guest runs with IRQs
6797                          * disabled. So we pull the trigger after 1 s of
6798                          * futile waiting, but inform the user about this.
6799                          */
6800                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6801                                "state on VCPU %d after 1 s timeout\n",
6802                                __func__, vcpu->vcpu_id);
6803                         vmx->soft_vnmi_blocked = 0;
6804                 }
6805         }
6806
6807         if (exit_reason < kvm_vmx_max_exit_handlers
6808             && kvm_vmx_exit_handlers[exit_reason])
6809                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6810         else {
6811                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6812                 vcpu->run->hw.hardware_exit_reason = exit_reason;
6813         }
6814         return 0;
6815 }
6816
6817 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6818 {
6819         if (irr == -1 || tpr < irr) {
6820                 vmcs_write32(TPR_THRESHOLD, 0);
6821                 return;
6822         }
6823
6824         vmcs_write32(TPR_THRESHOLD, irr);
6825 }
6826
6827 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
6828 {
6829         u32 sec_exec_control;
6830
6831         /*
6832          * There is not point to enable virtualize x2apic without enable
6833          * apicv
6834          */
6835         if (!cpu_has_vmx_virtualize_x2apic_mode() ||
6836                                 !vmx_vm_has_apicv(vcpu->kvm))
6837                 return;
6838
6839         if (!vm_need_tpr_shadow(vcpu->kvm))
6840                 return;
6841
6842         sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6843
6844         if (set) {
6845                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6846                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6847         } else {
6848                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6849                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6850         }
6851         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
6852
6853         vmx_set_msr_bitmap(vcpu);
6854 }
6855
6856 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
6857 {
6858         u16 status;
6859         u8 old;
6860
6861         if (!vmx_vm_has_apicv(kvm))
6862                 return;
6863
6864         if (isr == -1)
6865                 isr = 0;
6866
6867         status = vmcs_read16(GUEST_INTR_STATUS);
6868         old = status >> 8;
6869         if (isr != old) {
6870                 status &= 0xff;
6871                 status |= isr << 8;
6872                 vmcs_write16(GUEST_INTR_STATUS, status);
6873         }
6874 }
6875
6876 static void vmx_set_rvi(int vector)
6877 {
6878         u16 status;
6879         u8 old;
6880
6881         status = vmcs_read16(GUEST_INTR_STATUS);
6882         old = (u8)status & 0xff;
6883         if ((u8)vector != old) {
6884                 status &= ~0xff;
6885                 status |= (u8)vector;
6886                 vmcs_write16(GUEST_INTR_STATUS, status);
6887         }
6888 }
6889
6890 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6891 {
6892         if (max_irr == -1)
6893                 return;
6894
6895         vmx_set_rvi(max_irr);
6896 }
6897
6898 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6899 {
6900         if (!vmx_vm_has_apicv(vcpu->kvm))
6901                 return;
6902
6903         vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6904         vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6905         vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6906         vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6907 }
6908
6909 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6910 {
6911         u32 exit_intr_info;
6912
6913         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6914               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
6915                 return;
6916
6917         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6918         exit_intr_info = vmx->exit_intr_info;
6919
6920         /* Handle machine checks before interrupts are enabled */
6921         if (is_machine_check(exit_intr_info))
6922                 kvm_machine_check();
6923
6924         /* We need to handle NMIs before interrupts are enabled */
6925         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
6926             (exit_intr_info & INTR_INFO_VALID_MASK)) {
6927                 kvm_before_handle_nmi(&vmx->vcpu);
6928                 asm("int $2");
6929                 kvm_after_handle_nmi(&vmx->vcpu);
6930         }
6931 }
6932
6933 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
6934 {
6935         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6936
6937         /*
6938          * If external interrupt exists, IF bit is set in rflags/eflags on the
6939          * interrupt stack frame, and interrupt will be enabled on a return
6940          * from interrupt handler.
6941          */
6942         if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
6943                         == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
6944                 unsigned int vector;
6945                 unsigned long entry;
6946                 gate_desc *desc;
6947                 struct vcpu_vmx *vmx = to_vmx(vcpu);
6948 #ifdef CONFIG_X86_64
6949                 unsigned long tmp;
6950 #endif
6951
6952                 vector =  exit_intr_info & INTR_INFO_VECTOR_MASK;
6953                 desc = (gate_desc *)vmx->host_idt_base + vector;
6954                 entry = gate_offset(*desc);
6955                 asm volatile(
6956 #ifdef CONFIG_X86_64
6957                         "mov %%" _ASM_SP ", %[sp]\n\t"
6958                         "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6959                         "push $%c[ss]\n\t"
6960                         "push %[sp]\n\t"
6961 #endif
6962                         "pushf\n\t"
6963                         "orl $0x200, (%%" _ASM_SP ")\n\t"
6964                         __ASM_SIZE(push) " $%c[cs]\n\t"
6965                         "call *%[entry]\n\t"
6966                         :
6967 #ifdef CONFIG_X86_64
6968                         [sp]"=&r"(tmp)
6969 #endif
6970                         :
6971                         [entry]"r"(entry),
6972                         [ss]"i"(__KERNEL_DS),
6973                         [cs]"i"(__KERNEL_CS)
6974                         );
6975         } else
6976                 local_irq_enable();
6977 }
6978
6979 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6980 {
6981         u32 exit_intr_info;
6982         bool unblock_nmi;
6983         u8 vector;
6984         bool idtv_info_valid;
6985
6986         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6987
6988         if (cpu_has_virtual_nmis()) {
6989                 if (vmx->nmi_known_unmasked)
6990                         return;
6991                 /*
6992                  * Can't use vmx->exit_intr_info since we're not sure what
6993                  * the exit reason is.
6994                  */
6995                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6996                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6997                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6998                 /*
6999                  * SDM 3: 27.7.1.2 (September 2008)
7000                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
7001                  * a guest IRET fault.
7002                  * SDM 3: 23.2.2 (September 2008)
7003                  * Bit 12 is undefined in any of the following cases:
7004                  *  If the VM exit sets the valid bit in the IDT-vectoring
7005                  *   information field.
7006                  *  If the VM exit is due to a double fault.
7007                  */
7008                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
7009                     vector != DF_VECTOR && !idtv_info_valid)
7010                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7011                                       GUEST_INTR_STATE_NMI);
7012                 else
7013                         vmx->nmi_known_unmasked =
7014                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
7015                                   & GUEST_INTR_STATE_NMI);
7016         } else if (unlikely(vmx->soft_vnmi_blocked))
7017                 vmx->vnmi_blocked_time +=
7018                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
7019 }
7020
7021 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
7022                                       u32 idt_vectoring_info,
7023                                       int instr_len_field,
7024                                       int error_code_field)
7025 {
7026         u8 vector;
7027         int type;
7028         bool idtv_info_valid;
7029
7030         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7031
7032         vcpu->arch.nmi_injected = false;
7033         kvm_clear_exception_queue(vcpu);
7034         kvm_clear_interrupt_queue(vcpu);
7035
7036         if (!idtv_info_valid)
7037                 return;
7038
7039         kvm_make_request(KVM_REQ_EVENT, vcpu);
7040
7041         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
7042         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7043
7044         switch (type) {
7045         case INTR_TYPE_NMI_INTR:
7046                 vcpu->arch.nmi_injected = true;
7047                 /*
7048                  * SDM 3: 27.7.1.2 (September 2008)
7049                  * Clear bit "block by NMI" before VM entry if a NMI
7050                  * delivery faulted.
7051                  */
7052                 vmx_set_nmi_mask(vcpu, false);
7053                 break;
7054         case INTR_TYPE_SOFT_EXCEPTION:
7055                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7056                 /* fall through */
7057         case INTR_TYPE_HARD_EXCEPTION:
7058                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7059                         u32 err = vmcs_read32(error_code_field);
7060                         kvm_queue_exception_e(vcpu, vector, err);
7061                 } else
7062                         kvm_queue_exception(vcpu, vector);
7063                 break;
7064         case INTR_TYPE_SOFT_INTR:
7065                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7066                 /* fall through */
7067         case INTR_TYPE_EXT_INTR:
7068                 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7069                 break;
7070         default:
7071                 break;
7072         }
7073 }
7074
7075 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7076 {
7077         __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7078                                   VM_EXIT_INSTRUCTION_LEN,
7079                                   IDT_VECTORING_ERROR_CODE);
7080 }
7081
7082 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7083 {
7084         __vmx_complete_interrupts(vcpu,
7085                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7086                                   VM_ENTRY_INSTRUCTION_LEN,
7087                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
7088
7089         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7090 }
7091
7092 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7093 {
7094         int i, nr_msrs;
7095         struct perf_guest_switch_msr *msrs;
7096
7097         msrs = perf_guest_get_msrs(&nr_msrs);
7098
7099         if (!msrs)
7100                 return;
7101
7102         for (i = 0; i < nr_msrs; i++)
7103                 if (msrs[i].host == msrs[i].guest)
7104                         clear_atomic_switch_msr(vmx, msrs[i].msr);
7105                 else
7106                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7107                                         msrs[i].host);
7108 }
7109
7110 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
7111 {
7112         struct vcpu_vmx *vmx = to_vmx(vcpu);
7113         unsigned long debugctlmsr;
7114
7115         /* Record the guest's net vcpu time for enforced NMI injections. */
7116         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
7117                 vmx->entry_time = ktime_get();
7118
7119         /* Don't enter VMX if guest state is invalid, let the exit handler
7120            start emulation until we arrive back to a valid state */
7121         if (vmx->emulation_required)
7122                 return;
7123
7124         if (vmx->nested.sync_shadow_vmcs) {
7125                 copy_vmcs12_to_shadow(vmx);
7126                 vmx->nested.sync_shadow_vmcs = false;
7127         }
7128
7129         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
7130                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7131         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
7132                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7133
7134         /* When single-stepping over STI and MOV SS, we must clear the
7135          * corresponding interruptibility bits in the guest state. Otherwise
7136          * vmentry fails as it then expects bit 14 (BS) in pending debug
7137          * exceptions being set, but that's not correct for the guest debugging
7138          * case. */
7139         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7140                 vmx_set_interrupt_shadow(vcpu, 0);
7141
7142         atomic_switch_perf_msrs(vmx);
7143         debugctlmsr = get_debugctlmsr();
7144
7145         vmx->__launched = vmx->loaded_vmcs->launched;
7146         asm(
7147                 /* Store host registers */
7148                 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
7149                 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
7150                 "push %%" _ASM_CX " \n\t"
7151                 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7152                 "je 1f \n\t"
7153                 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7154                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
7155                 "1: \n\t"
7156                 /* Reload cr2 if changed */
7157                 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
7158                 "mov %%cr2, %%" _ASM_DX " \n\t"
7159                 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
7160                 "je 2f \n\t"
7161                 "mov %%" _ASM_AX", %%cr2 \n\t"
7162                 "2: \n\t"
7163                 /* Check if vmlaunch of vmresume is needed */
7164                 "cmpl $0, %c[launched](%0) \n\t"
7165                 /* Load guest registers.  Don't clobber flags. */
7166                 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
7167                 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
7168                 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
7169                 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
7170                 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
7171                 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
7172 #ifdef CONFIG_X86_64
7173                 "mov %c[r8](%0),  %%r8  \n\t"
7174                 "mov %c[r9](%0),  %%r9  \n\t"
7175                 "mov %c[r10](%0), %%r10 \n\t"
7176                 "mov %c[r11](%0), %%r11 \n\t"
7177                 "mov %c[r12](%0), %%r12 \n\t"
7178                 "mov %c[r13](%0), %%r13 \n\t"
7179                 "mov %c[r14](%0), %%r14 \n\t"
7180                 "mov %c[r15](%0), %%r15 \n\t"
7181 #endif
7182                 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
7183
7184                 /* Enter guest mode */
7185                 "jne 1f \n\t"
7186                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
7187                 "jmp 2f \n\t"
7188                 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
7189                 "2: "
7190                 /* Save guest registers, load host registers, keep flags */
7191                 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
7192                 "pop %0 \n\t"
7193                 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
7194                 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
7195                 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
7196                 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
7197                 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
7198                 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
7199                 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
7200 #ifdef CONFIG_X86_64
7201                 "mov %%r8,  %c[r8](%0) \n\t"
7202                 "mov %%r9,  %c[r9](%0) \n\t"
7203                 "mov %%r10, %c[r10](%0) \n\t"
7204                 "mov %%r11, %c[r11](%0) \n\t"
7205                 "mov %%r12, %c[r12](%0) \n\t"
7206                 "mov %%r13, %c[r13](%0) \n\t"
7207                 "mov %%r14, %c[r14](%0) \n\t"
7208                 "mov %%r15, %c[r15](%0) \n\t"
7209 #endif
7210                 "mov %%cr2, %%" _ASM_AX "   \n\t"
7211                 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
7212
7213                 "pop  %%" _ASM_BP "; pop  %%" _ASM_DX " \n\t"
7214                 "setbe %c[fail](%0) \n\t"
7215                 ".pushsection .rodata \n\t"
7216                 ".global vmx_return \n\t"
7217                 "vmx_return: " _ASM_PTR " 2b \n\t"
7218                 ".popsection"
7219               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
7220                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
7221                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
7222                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
7223                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
7224                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
7225                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
7226                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
7227                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
7228                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
7229                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
7230 #ifdef CONFIG_X86_64
7231                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
7232                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
7233                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
7234                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
7235                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
7236                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
7237                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
7238                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
7239 #endif
7240                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
7241                 [wordsize]"i"(sizeof(ulong))
7242               : "cc", "memory"
7243 #ifdef CONFIG_X86_64
7244                 , "rax", "rbx", "rdi", "rsi"
7245                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
7246 #else
7247                 , "eax", "ebx", "edi", "esi"
7248 #endif
7249               );
7250
7251         /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7252         if (debugctlmsr)
7253                 update_debugctlmsr(debugctlmsr);
7254
7255 #ifndef CONFIG_X86_64
7256         /*
7257          * The sysexit path does not restore ds/es, so we must set them to
7258          * a reasonable value ourselves.
7259          *
7260          * We can't defer this to vmx_load_host_state() since that function
7261          * may be executed in interrupt context, which saves and restore segments
7262          * around it, nullifying its effect.
7263          */
7264         loadsegment(ds, __USER_DS);
7265         loadsegment(es, __USER_DS);
7266 #endif
7267
7268         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
7269                                   | (1 << VCPU_EXREG_RFLAGS)
7270                                   | (1 << VCPU_EXREG_CPL)
7271                                   | (1 << VCPU_EXREG_PDPTR)
7272                                   | (1 << VCPU_EXREG_SEGMENTS)
7273                                   | (1 << VCPU_EXREG_CR3));
7274         vcpu->arch.regs_dirty = 0;
7275
7276         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7277
7278         vmx->loaded_vmcs->launched = 1;
7279
7280         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
7281         trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
7282
7283         vmx_complete_atomic_exit(vmx);
7284         vmx_recover_nmi_blocking(vmx);
7285         vmx_complete_interrupts(vmx);
7286 }
7287
7288 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
7289 {
7290         struct vcpu_vmx *vmx = to_vmx(vcpu);
7291
7292         free_vpid(vmx);
7293         free_nested(vmx);
7294         free_loaded_vmcs(vmx->loaded_vmcs);
7295         kfree(vmx->guest_msrs);
7296         kvm_vcpu_uninit(vcpu);
7297         kmem_cache_free(kvm_vcpu_cache, vmx);
7298 }
7299
7300 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
7301 {
7302         int err;
7303         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
7304         int cpu;
7305
7306         if (!vmx)
7307                 return ERR_PTR(-ENOMEM);
7308
7309         allocate_vpid(vmx);
7310
7311         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
7312         if (err)
7313                 goto free_vcpu;
7314
7315         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
7316         err = -ENOMEM;
7317         if (!vmx->guest_msrs) {
7318                 goto uninit_vcpu;
7319         }
7320
7321         vmx->loaded_vmcs = &vmx->vmcs01;
7322         vmx->loaded_vmcs->vmcs = alloc_vmcs();
7323         if (!vmx->loaded_vmcs->vmcs)
7324                 goto free_msrs;
7325         if (!vmm_exclusive)
7326                 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
7327         loaded_vmcs_init(vmx->loaded_vmcs);
7328         if (!vmm_exclusive)
7329                 kvm_cpu_vmxoff();
7330
7331         cpu = get_cpu();
7332         vmx_vcpu_load(&vmx->vcpu, cpu);
7333         vmx->vcpu.cpu = cpu;
7334         err = vmx_vcpu_setup(vmx);
7335         vmx_vcpu_put(&vmx->vcpu);
7336         put_cpu();
7337         if (err)
7338                 goto free_vmcs;
7339         if (vm_need_virtualize_apic_accesses(kvm)) {
7340                 err = alloc_apic_access_page(kvm);
7341                 if (err)
7342                         goto free_vmcs;
7343         }
7344
7345         if (enable_ept) {
7346                 if (!kvm->arch.ept_identity_map_addr)
7347                         kvm->arch.ept_identity_map_addr =
7348                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
7349                 err = -ENOMEM;
7350                 if (alloc_identity_pagetable(kvm) != 0)
7351                         goto free_vmcs;
7352                 if (!init_rmode_identity_map(kvm))
7353                         goto free_vmcs;
7354         }
7355
7356         vmx->nested.current_vmptr = -1ull;
7357         vmx->nested.current_vmcs12 = NULL;
7358
7359         return &vmx->vcpu;
7360
7361 free_vmcs:
7362         free_loaded_vmcs(vmx->loaded_vmcs);
7363 free_msrs:
7364         kfree(vmx->guest_msrs);
7365 uninit_vcpu:
7366         kvm_vcpu_uninit(&vmx->vcpu);
7367 free_vcpu:
7368         free_vpid(vmx);
7369         kmem_cache_free(kvm_vcpu_cache, vmx);
7370         return ERR_PTR(err);
7371 }
7372
7373 static void __init vmx_check_processor_compat(void *rtn)
7374 {
7375         struct vmcs_config vmcs_conf;
7376
7377         *(int *)rtn = 0;
7378         if (setup_vmcs_config(&vmcs_conf) < 0)
7379                 *(int *)rtn = -EIO;
7380         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7381                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7382                                 smp_processor_id());
7383                 *(int *)rtn = -EIO;
7384         }
7385 }
7386
7387 static int get_ept_level(void)
7388 {
7389         return VMX_EPT_DEFAULT_GAW + 1;
7390 }
7391
7392 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7393 {
7394         u64 ret;
7395
7396         /* For VT-d and EPT combination
7397          * 1. MMIO: always map as UC
7398          * 2. EPT with VT-d:
7399          *   a. VT-d without snooping control feature: can't guarantee the
7400          *      result, try to trust guest.
7401          *   b. VT-d with snooping control feature: snooping control feature of
7402          *      VT-d engine can guarantee the cache correctness. Just set it
7403          *      to WB to keep consistent with host. So the same as item 3.
7404          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
7405          *    consistent with host MTRR
7406          */
7407         if (is_mmio)
7408                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7409         else if (vcpu->kvm->arch.iommu_domain &&
7410                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
7411                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
7412                       VMX_EPT_MT_EPTE_SHIFT;
7413         else
7414                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
7415                         | VMX_EPT_IPAT_BIT;
7416
7417         return ret;
7418 }
7419
7420 static int vmx_get_lpage_level(void)
7421 {
7422         if (enable_ept && !cpu_has_vmx_ept_1g_page())
7423                 return PT_DIRECTORY_LEVEL;
7424         else
7425                 /* For shadow and EPT supported 1GB page */
7426                 return PT_PDPE_LEVEL;
7427 }
7428
7429 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7430 {
7431         struct kvm_cpuid_entry2 *best;
7432         struct vcpu_vmx *vmx = to_vmx(vcpu);
7433         u32 exec_control;
7434
7435         vmx->rdtscp_enabled = false;
7436         if (vmx_rdtscp_supported()) {
7437                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7438                 if (exec_control & SECONDARY_EXEC_RDTSCP) {
7439                         best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
7440                         if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
7441                                 vmx->rdtscp_enabled = true;
7442                         else {
7443                                 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7444                                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7445                                                 exec_control);
7446                         }
7447                 }
7448         }
7449
7450         /* Exposing INVPCID only when PCID is exposed */
7451         best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7452         if (vmx_invpcid_supported() &&
7453             best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
7454             guest_cpuid_has_pcid(vcpu)) {
7455                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7456                 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
7457                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7458                              exec_control);
7459         } else {
7460                 if (cpu_has_secondary_exec_ctrls()) {
7461                         exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7462                         exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
7463                         vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7464                                      exec_control);
7465                 }
7466                 if (best)
7467                         best->ebx &= ~bit(X86_FEATURE_INVPCID);
7468         }
7469 }
7470
7471 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7472 {
7473         if (func == 1 && nested)
7474                 entry->ecx |= bit(X86_FEATURE_VMX);
7475 }
7476
7477 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
7478                 struct x86_exception *fault)
7479 {
7480         struct vmcs12 *vmcs12;
7481         nested_vmx_vmexit(vcpu);
7482         vmcs12 = get_vmcs12(vcpu);
7483
7484         if (fault->error_code & PFERR_RSVD_MASK)
7485                 vmcs12->vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
7486         else
7487                 vmcs12->vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
7488         vmcs12->exit_qualification = vcpu->arch.exit_qualification;
7489         vmcs12->guest_physical_address = fault->address;
7490 }
7491
7492 /* Callbacks for nested_ept_init_mmu_context: */
7493
7494 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
7495 {
7496         /* return the page table to be shadowed - in our case, EPT12 */
7497         return get_vmcs12(vcpu)->ept_pointer;
7498 }
7499
7500 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
7501 {
7502         int r = kvm_init_shadow_ept_mmu(vcpu, &vcpu->arch.mmu,
7503                         nested_vmx_ept_caps & VMX_EPT_EXECUTE_ONLY_BIT);
7504
7505         vcpu->arch.mmu.set_cr3           = vmx_set_cr3;
7506         vcpu->arch.mmu.get_cr3           = nested_ept_get_cr3;
7507         vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
7508
7509         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
7510
7511         return r;
7512 }
7513
7514 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
7515 {
7516         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
7517 }
7518
7519 /*
7520  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
7521  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
7522  * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
7523  * guest in a way that will both be appropriate to L1's requests, and our
7524  * needs. In addition to modifying the active vmcs (which is vmcs02), this
7525  * function also has additional necessary side-effects, like setting various
7526  * vcpu->arch fields.
7527  */
7528 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7529 {
7530         struct vcpu_vmx *vmx = to_vmx(vcpu);
7531         u32 exec_control;
7532
7533         vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
7534         vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
7535         vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
7536         vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
7537         vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
7538         vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
7539         vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
7540         vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
7541         vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
7542         vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
7543         vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
7544         vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
7545         vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
7546         vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
7547         vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
7548         vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
7549         vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
7550         vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
7551         vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
7552         vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
7553         vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
7554         vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
7555         vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
7556         vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
7557         vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
7558         vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
7559         vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
7560         vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
7561         vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
7562         vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
7563         vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
7564         vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
7565         vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
7566         vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
7567         vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
7568         vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
7569
7570         vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
7571         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
7572                 vmcs12->vm_entry_intr_info_field);
7573         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
7574                 vmcs12->vm_entry_exception_error_code);
7575         vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
7576                 vmcs12->vm_entry_instruction_len);
7577         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
7578                 vmcs12->guest_interruptibility_info);
7579         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
7580         kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
7581         vmx_set_rflags(vcpu, vmcs12->guest_rflags);
7582         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
7583                 vmcs12->guest_pending_dbg_exceptions);
7584         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
7585         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
7586
7587         vmcs_write64(VMCS_LINK_POINTER, -1ull);
7588
7589         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
7590                 (vmcs_config.pin_based_exec_ctrl |
7591                  vmcs12->pin_based_vm_exec_control));
7592
7593         if (vmcs12->pin_based_vm_exec_control & PIN_BASED_VMX_PREEMPTION_TIMER)
7594                 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE,
7595                              vmcs12->vmx_preemption_timer_value);
7596
7597         /*
7598          * Whether page-faults are trapped is determined by a combination of
7599          * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
7600          * If enable_ept, L0 doesn't care about page faults and we should
7601          * set all of these to L1's desires. However, if !enable_ept, L0 does
7602          * care about (at least some) page faults, and because it is not easy
7603          * (if at all possible?) to merge L0 and L1's desires, we simply ask
7604          * to exit on each and every L2 page fault. This is done by setting
7605          * MASK=MATCH=0 and (see below) EB.PF=1.
7606          * Note that below we don't need special code to set EB.PF beyond the
7607          * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
7608          * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
7609          * !enable_ept, EB.PF is 1, so the "or" will always be 1.
7610          *
7611          * A problem with this approach (when !enable_ept) is that L1 may be
7612          * injected with more page faults than it asked for. This could have
7613          * caused problems, but in practice existing hypervisors don't care.
7614          * To fix this, we will need to emulate the PFEC checking (on the L1
7615          * page tables), using walk_addr(), when injecting PFs to L1.
7616          */
7617         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
7618                 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
7619         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
7620                 enable_ept ? vmcs12->page_fault_error_code_match : 0);
7621
7622         if (cpu_has_secondary_exec_ctrls()) {
7623                 u32 exec_control = vmx_secondary_exec_control(vmx);
7624                 if (!vmx->rdtscp_enabled)
7625                         exec_control &= ~SECONDARY_EXEC_RDTSCP;
7626                 /* Take the following fields only from vmcs12 */
7627                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7628                 if (nested_cpu_has(vmcs12,
7629                                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
7630                         exec_control |= vmcs12->secondary_vm_exec_control;
7631
7632                 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
7633                         /*
7634                          * Translate L1 physical address to host physical
7635                          * address for vmcs02. Keep the page pinned, so this
7636                          * physical address remains valid. We keep a reference
7637                          * to it so we can release it later.
7638                          */
7639                         if (vmx->nested.apic_access_page) /* shouldn't happen */
7640                                 nested_release_page(vmx->nested.apic_access_page);
7641                         vmx->nested.apic_access_page =
7642                                 nested_get_page(vcpu, vmcs12->apic_access_addr);
7643                         /*
7644                          * If translation failed, no matter: This feature asks
7645                          * to exit when accessing the given address, and if it
7646                          * can never be accessed, this feature won't do
7647                          * anything anyway.
7648                          */
7649                         if (!vmx->nested.apic_access_page)
7650                                 exec_control &=
7651                                   ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7652                         else
7653                                 vmcs_write64(APIC_ACCESS_ADDR,
7654                                   page_to_phys(vmx->nested.apic_access_page));
7655                 }
7656
7657                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
7658         }
7659
7660
7661         /*
7662          * Set host-state according to L0's settings (vmcs12 is irrelevant here)
7663          * Some constant fields are set here by vmx_set_constant_host_state().
7664          * Other fields are different per CPU, and will be set later when
7665          * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
7666          */
7667         vmx_set_constant_host_state(vmx);
7668
7669         /*
7670          * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
7671          * entry, but only if the current (host) sp changed from the value
7672          * we wrote last (vmx->host_rsp). This cache is no longer relevant
7673          * if we switch vmcs, and rather than hold a separate cache per vmcs,
7674          * here we just force the write to happen on entry.
7675          */
7676         vmx->host_rsp = 0;
7677
7678         exec_control = vmx_exec_control(vmx); /* L0's desires */
7679         exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
7680         exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
7681         exec_control &= ~CPU_BASED_TPR_SHADOW;
7682         exec_control |= vmcs12->cpu_based_vm_exec_control;
7683         /*
7684          * Merging of IO and MSR bitmaps not currently supported.
7685          * Rather, exit every time.
7686          */
7687         exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
7688         exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
7689         exec_control |= CPU_BASED_UNCOND_IO_EXITING;
7690
7691         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
7692
7693         /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
7694          * bitwise-or of what L1 wants to trap for L2, and what we want to
7695          * trap. Note that CR0.TS also needs updating - we do this later.
7696          */
7697         update_exception_bitmap(vcpu);
7698         vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
7699         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7700
7701         /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
7702          * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
7703          * bits are further modified by vmx_set_efer() below.
7704          */
7705         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
7706
7707         /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
7708          * emulated by vmx_set_efer(), below.
7709          */
7710         vmcs_write32(VM_ENTRY_CONTROLS,
7711                 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
7712                         ~VM_ENTRY_IA32E_MODE) |
7713                 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
7714
7715         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
7716                 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
7717                 vcpu->arch.pat = vmcs12->guest_ia32_pat;
7718         } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
7719                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
7720
7721
7722         set_cr4_guest_host_mask(vmx);
7723
7724         if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
7725                 vmcs_write64(TSC_OFFSET,
7726                         vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
7727         else
7728                 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
7729
7730         if (enable_vpid) {
7731                 /*
7732                  * Trivially support vpid by letting L2s share their parent
7733                  * L1's vpid. TODO: move to a more elaborate solution, giving
7734                  * each L2 its own vpid and exposing the vpid feature to L1.
7735                  */
7736                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
7737                 vmx_flush_tlb(vcpu);
7738         }
7739
7740         if (nested_cpu_has_ept(vmcs12)) {
7741                 kvm_mmu_unload(vcpu);
7742                 nested_ept_init_mmu_context(vcpu);
7743         }
7744
7745         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
7746                 vcpu->arch.efer = vmcs12->guest_ia32_efer;
7747         else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
7748                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
7749         else
7750                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
7751         /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
7752         vmx_set_efer(vcpu, vcpu->arch.efer);
7753
7754         /*
7755          * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
7756          * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
7757          * The CR0_READ_SHADOW is what L2 should have expected to read given
7758          * the specifications by L1; It's not enough to take
7759          * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
7760          * have more bits than L1 expected.
7761          */
7762         vmx_set_cr0(vcpu, vmcs12->guest_cr0);
7763         vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
7764
7765         vmx_set_cr4(vcpu, vmcs12->guest_cr4);
7766         vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
7767
7768         /* shadow page tables on either EPT or shadow page tables */
7769         kvm_set_cr3(vcpu, vmcs12->guest_cr3);
7770         kvm_mmu_reset_context(vcpu);
7771
7772         /*
7773          * L1 may access the L2's PDPTR, so save them to construct vmcs12
7774          */
7775         if (enable_ept) {
7776                 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
7777                 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
7778                 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
7779                 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
7780                 __clear_bit(VCPU_EXREG_PDPTR,
7781                                 (unsigned long *)&vcpu->arch.regs_avail);
7782                 __clear_bit(VCPU_EXREG_PDPTR,
7783                                 (unsigned long *)&vcpu->arch.regs_dirty);
7784         }
7785
7786         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
7787         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
7788 }
7789
7790 /*
7791  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
7792  * for running an L2 nested guest.
7793  */
7794 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
7795 {
7796         struct vmcs12 *vmcs12;
7797         struct vcpu_vmx *vmx = to_vmx(vcpu);
7798         int cpu;
7799         struct loaded_vmcs *vmcs02;
7800         bool ia32e;
7801
7802         if (!nested_vmx_check_permission(vcpu) ||
7803             !nested_vmx_check_vmcs12(vcpu))
7804                 return 1;
7805
7806         skip_emulated_instruction(vcpu);
7807         vmcs12 = get_vmcs12(vcpu);
7808
7809         if (enable_shadow_vmcs)
7810                 copy_shadow_to_vmcs12(vmx);
7811
7812         /*
7813          * The nested entry process starts with enforcing various prerequisites
7814          * on vmcs12 as required by the Intel SDM, and act appropriately when
7815          * they fail: As the SDM explains, some conditions should cause the
7816          * instruction to fail, while others will cause the instruction to seem
7817          * to succeed, but return an EXIT_REASON_INVALID_STATE.
7818          * To speed up the normal (success) code path, we should avoid checking
7819          * for misconfigurations which will anyway be caught by the processor
7820          * when using the merged vmcs02.
7821          */
7822         if (vmcs12->launch_state == launch) {
7823                 nested_vmx_failValid(vcpu,
7824                         launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
7825                                : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
7826                 return 1;
7827         }
7828
7829         if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE) {
7830                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7831                 return 1;
7832         }
7833
7834         if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
7835                         !IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
7836                 /*TODO: Also verify bits beyond physical address width are 0*/
7837                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7838                 return 1;
7839         }
7840
7841         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
7842                         !IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
7843                 /*TODO: Also verify bits beyond physical address width are 0*/
7844                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7845                 return 1;
7846         }
7847
7848         if (vmcs12->vm_entry_msr_load_count > 0 ||
7849             vmcs12->vm_exit_msr_load_count > 0 ||
7850             vmcs12->vm_exit_msr_store_count > 0) {
7851                 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
7852                                     __func__);
7853                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7854                 return 1;
7855         }
7856
7857         if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
7858               nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
7859             !vmx_control_verify(vmcs12->secondary_vm_exec_control,
7860               nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
7861             !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
7862               nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
7863             !vmx_control_verify(vmcs12->vm_exit_controls,
7864               nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
7865             !vmx_control_verify(vmcs12->vm_entry_controls,
7866               nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
7867         {
7868                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7869                 return 1;
7870         }
7871
7872         if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
7873             ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
7874                 nested_vmx_failValid(vcpu,
7875                         VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7876                 return 1;
7877         }
7878
7879         if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
7880             ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
7881                 nested_vmx_entry_failure(vcpu, vmcs12,
7882                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7883                 return 1;
7884         }
7885         if (vmcs12->vmcs_link_pointer != -1ull) {
7886                 nested_vmx_entry_failure(vcpu, vmcs12,
7887                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
7888                 return 1;
7889         }
7890
7891         /*
7892          * If the load IA32_EFER VM-entry control is 1, the following checks
7893          * are performed on the field for the IA32_EFER MSR:
7894          * - Bits reserved in the IA32_EFER MSR must be 0.
7895          * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
7896          *   the IA-32e mode guest VM-exit control. It must also be identical
7897          *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
7898          *   CR0.PG) is 1.
7899          */
7900         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
7901                 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
7902                 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
7903                     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
7904                     ((vmcs12->guest_cr0 & X86_CR0_PG) &&
7905                      ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
7906                         nested_vmx_entry_failure(vcpu, vmcs12,
7907                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7908                         return 1;
7909                 }
7910         }
7911
7912         /*
7913          * If the load IA32_EFER VM-exit control is 1, bits reserved in the
7914          * IA32_EFER MSR must be 0 in the field for that register. In addition,
7915          * the values of the LMA and LME bits in the field must each be that of
7916          * the host address-space size VM-exit control.
7917          */
7918         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
7919                 ia32e = (vmcs12->vm_exit_controls &
7920                          VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
7921                 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
7922                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
7923                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
7924                         nested_vmx_entry_failure(vcpu, vmcs12,
7925                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7926                         return 1;
7927                 }
7928         }
7929
7930         /*
7931          * We're finally done with prerequisite checking, and can start with
7932          * the nested entry.
7933          */
7934
7935         vmcs02 = nested_get_current_vmcs02(vmx);
7936         if (!vmcs02)
7937                 return -ENOMEM;
7938
7939         enter_guest_mode(vcpu);
7940
7941         vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
7942
7943         cpu = get_cpu();
7944         vmx->loaded_vmcs = vmcs02;
7945         vmx_vcpu_put(vcpu);
7946         vmx_vcpu_load(vcpu, cpu);
7947         vcpu->cpu = cpu;
7948         put_cpu();
7949
7950         vmx_segment_cache_clear(vmx);
7951
7952         vmcs12->launch_state = 1;
7953
7954         prepare_vmcs02(vcpu, vmcs12);
7955
7956         /*
7957          * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
7958          * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
7959          * returned as far as L1 is concerned. It will only return (and set
7960          * the success flag) when L2 exits (see nested_vmx_vmexit()).
7961          */
7962         return 1;
7963 }
7964
7965 /*
7966  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
7967  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
7968  * This function returns the new value we should put in vmcs12.guest_cr0.
7969  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
7970  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
7971  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
7972  *     didn't trap the bit, because if L1 did, so would L0).
7973  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
7974  *     been modified by L2, and L1 knows it. So just leave the old value of
7975  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
7976  *     isn't relevant, because if L0 traps this bit it can set it to anything.
7977  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
7978  *     changed these bits, and therefore they need to be updated, but L0
7979  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
7980  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
7981  */
7982 static inline unsigned long
7983 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7984 {
7985         return
7986         /*1*/   (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
7987         /*2*/   (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
7988         /*3*/   (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
7989                         vcpu->arch.cr0_guest_owned_bits));
7990 }
7991
7992 static inline unsigned long
7993 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7994 {
7995         return
7996         /*1*/   (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
7997         /*2*/   (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
7998         /*3*/   (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
7999                         vcpu->arch.cr4_guest_owned_bits));
8000 }
8001
8002 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
8003                                        struct vmcs12 *vmcs12)
8004 {
8005         u32 idt_vectoring;
8006         unsigned int nr;
8007
8008         if (vcpu->arch.exception.pending) {
8009                 nr = vcpu->arch.exception.nr;
8010                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8011
8012                 if (kvm_exception_is_soft(nr)) {
8013                         vmcs12->vm_exit_instruction_len =
8014                                 vcpu->arch.event_exit_inst_len;
8015                         idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
8016                 } else
8017                         idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
8018
8019                 if (vcpu->arch.exception.has_error_code) {
8020                         idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
8021                         vmcs12->idt_vectoring_error_code =
8022                                 vcpu->arch.exception.error_code;
8023                 }
8024
8025                 vmcs12->idt_vectoring_info_field = idt_vectoring;
8026         } else if (vcpu->arch.nmi_pending) {
8027                 vmcs12->idt_vectoring_info_field =
8028                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
8029         } else if (vcpu->arch.interrupt.pending) {
8030                 nr = vcpu->arch.interrupt.nr;
8031                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8032
8033                 if (vcpu->arch.interrupt.soft) {
8034                         idt_vectoring |= INTR_TYPE_SOFT_INTR;
8035                         vmcs12->vm_entry_instruction_len =
8036                                 vcpu->arch.event_exit_inst_len;
8037                 } else
8038                         idt_vectoring |= INTR_TYPE_EXT_INTR;
8039
8040                 vmcs12->idt_vectoring_info_field = idt_vectoring;
8041         }
8042 }
8043
8044 /*
8045  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
8046  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
8047  * and this function updates it to reflect the changes to the guest state while
8048  * L2 was running (and perhaps made some exits which were handled directly by L0
8049  * without going back to L1), and to reflect the exit reason.
8050  * Note that we do not have to copy here all VMCS fields, just those that
8051  * could have changed by the L2 guest or the exit - i.e., the guest-state and
8052  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
8053  * which already writes to vmcs12 directly.
8054  */
8055 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8056 {
8057         /* update guest state fields: */
8058         vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
8059         vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
8060
8061         kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
8062         vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8063         vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
8064         vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
8065
8066         vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
8067         vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
8068         vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
8069         vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
8070         vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
8071         vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
8072         vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
8073         vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
8074         vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
8075         vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
8076         vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
8077         vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
8078         vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
8079         vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
8080         vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
8081         vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
8082         vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
8083         vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
8084         vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
8085         vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
8086         vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
8087         vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
8088         vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
8089         vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
8090         vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
8091         vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
8092         vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
8093         vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
8094         vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
8095         vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
8096         vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
8097         vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
8098         vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
8099         vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
8100         vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
8101         vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
8102
8103         vmcs12->guest_interruptibility_info =
8104                 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
8105         vmcs12->guest_pending_dbg_exceptions =
8106                 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
8107
8108         /*
8109          * In some cases (usually, nested EPT), L2 is allowed to change its
8110          * own CR3 without exiting. If it has changed it, we must keep it.
8111          * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
8112          * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
8113          *
8114          * Additionally, restore L2's PDPTR to vmcs12.
8115          */
8116         if (enable_ept) {
8117                 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
8118                 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
8119                 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
8120                 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
8121                 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
8122         }
8123
8124         vmcs12->vm_entry_controls =
8125                 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
8126                 (vmcs_read32(VM_ENTRY_CONTROLS) & VM_ENTRY_IA32E_MODE);
8127
8128         /* TODO: These cannot have changed unless we have MSR bitmaps and
8129          * the relevant bit asks not to trap the change */
8130         vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8131         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
8132                 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
8133         vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
8134         vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
8135         vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
8136
8137         /* update exit information fields: */
8138
8139         vmcs12->vm_exit_reason  = to_vmx(vcpu)->exit_reason;
8140         vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8141
8142         vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8143         if ((vmcs12->vm_exit_intr_info &
8144              (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8145             (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
8146                 vmcs12->vm_exit_intr_error_code =
8147                         vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8148         vmcs12->idt_vectoring_info_field = 0;
8149         vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
8150         vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8151
8152         if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
8153                 /* vm_entry_intr_info_field is cleared on exit. Emulate this
8154                  * instead of reading the real value. */
8155                 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
8156
8157                 /*
8158                  * Transfer the event that L0 or L1 may wanted to inject into
8159                  * L2 to IDT_VECTORING_INFO_FIELD.
8160                  */
8161                 vmcs12_save_pending_event(vcpu, vmcs12);
8162         }
8163
8164         /*
8165          * Drop what we picked up for L2 via vmx_complete_interrupts. It is
8166          * preserved above and would only end up incorrectly in L1.
8167          */
8168         vcpu->arch.nmi_injected = false;
8169         kvm_clear_exception_queue(vcpu);
8170         kvm_clear_interrupt_queue(vcpu);
8171 }
8172
8173 /*
8174  * A part of what we need to when the nested L2 guest exits and we want to
8175  * run its L1 parent, is to reset L1's guest state to the host state specified
8176  * in vmcs12.
8177  * This function is to be called not only on normal nested exit, but also on
8178  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
8179  * Failures During or After Loading Guest State").
8180  * This function should be called when the active VMCS is L1's (vmcs01).
8181  */
8182 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
8183                                    struct vmcs12 *vmcs12)
8184 {
8185         struct kvm_segment seg;
8186
8187         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
8188                 vcpu->arch.efer = vmcs12->host_ia32_efer;
8189         else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8190                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8191         else
8192                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8193         vmx_set_efer(vcpu, vcpu->arch.efer);
8194
8195         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
8196         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
8197         vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
8198         /*
8199          * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
8200          * actually changed, because it depends on the current state of
8201          * fpu_active (which may have changed).
8202          * Note that vmx_set_cr0 refers to efer set above.
8203          */
8204         kvm_set_cr0(vcpu, vmcs12->host_cr0);
8205         /*
8206          * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
8207          * to apply the same changes to L1's vmcs. We just set cr0 correctly,
8208          * but we also need to update cr0_guest_host_mask and exception_bitmap.
8209          */
8210         update_exception_bitmap(vcpu);
8211         vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
8212         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8213
8214         /*
8215          * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
8216          * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
8217          */
8218         vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
8219         kvm_set_cr4(vcpu, vmcs12->host_cr4);
8220
8221         if (nested_cpu_has_ept(vmcs12))
8222                 nested_ept_uninit_mmu_context(vcpu);
8223
8224         kvm_set_cr3(vcpu, vmcs12->host_cr3);
8225         kvm_mmu_reset_context(vcpu);
8226
8227         if (enable_vpid) {
8228                 /*
8229                  * Trivially support vpid by letting L2s share their parent
8230                  * L1's vpid. TODO: move to a more elaborate solution, giving
8231                  * each L2 its own vpid and exposing the vpid feature to L1.
8232                  */
8233                 vmx_flush_tlb(vcpu);
8234         }
8235
8236
8237         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
8238         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
8239         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
8240         vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
8241         vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
8242
8243         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
8244                 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
8245                 vcpu->arch.pat = vmcs12->host_ia32_pat;
8246         }
8247         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8248                 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
8249                         vmcs12->host_ia32_perf_global_ctrl);
8250
8251         /* Set L1 segment info according to Intel SDM
8252             27.5.2 Loading Host Segment and Descriptor-Table Registers */
8253         seg = (struct kvm_segment) {
8254                 .base = 0,
8255                 .limit = 0xFFFFFFFF,
8256                 .selector = vmcs12->host_cs_selector,
8257                 .type = 11,
8258                 .present = 1,
8259                 .s = 1,
8260                 .g = 1
8261         };
8262         if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8263                 seg.l = 1;
8264         else
8265                 seg.db = 1;
8266         vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
8267         seg = (struct kvm_segment) {
8268                 .base = 0,
8269                 .limit = 0xFFFFFFFF,
8270                 .type = 3,
8271                 .present = 1,
8272                 .s = 1,
8273                 .db = 1,
8274                 .g = 1
8275         };
8276         seg.selector = vmcs12->host_ds_selector;
8277         vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
8278         seg.selector = vmcs12->host_es_selector;
8279         vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
8280         seg.selector = vmcs12->host_ss_selector;
8281         vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
8282         seg.selector = vmcs12->host_fs_selector;
8283         seg.base = vmcs12->host_fs_base;
8284         vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
8285         seg.selector = vmcs12->host_gs_selector;
8286         seg.base = vmcs12->host_gs_base;
8287         vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
8288         seg = (struct kvm_segment) {
8289                 .base = vmcs12->host_tr_base,
8290                 .limit = 0x67,
8291                 .selector = vmcs12->host_tr_selector,
8292                 .type = 11,
8293                 .present = 1
8294         };
8295         vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
8296
8297         kvm_set_dr(vcpu, 7, 0x400);
8298         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
8299 }
8300
8301 /*
8302  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
8303  * and modify vmcs12 to make it see what it would expect to see there if
8304  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
8305  */
8306 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
8307 {
8308         struct vcpu_vmx *vmx = to_vmx(vcpu);
8309         int cpu;
8310         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8311
8312         /* trying to cancel vmlaunch/vmresume is a bug */
8313         WARN_ON_ONCE(vmx->nested.nested_run_pending);
8314
8315         leave_guest_mode(vcpu);
8316         prepare_vmcs12(vcpu, vmcs12);
8317
8318         cpu = get_cpu();
8319         vmx->loaded_vmcs = &vmx->vmcs01;
8320         vmx_vcpu_put(vcpu);
8321         vmx_vcpu_load(vcpu, cpu);
8322         vcpu->cpu = cpu;
8323         put_cpu();
8324
8325         vmx_segment_cache_clear(vmx);
8326
8327         /* if no vmcs02 cache requested, remove the one we used */
8328         if (VMCS02_POOL_SIZE == 0)
8329                 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
8330
8331         load_vmcs12_host_state(vcpu, vmcs12);
8332
8333         /* Update TSC_OFFSET if TSC was changed while L2 ran */
8334         vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8335
8336         /* This is needed for same reason as it was needed in prepare_vmcs02 */
8337         vmx->host_rsp = 0;
8338
8339         /* Unpin physical memory we referred to in vmcs02 */
8340         if (vmx->nested.apic_access_page) {
8341                 nested_release_page(vmx->nested.apic_access_page);
8342                 vmx->nested.apic_access_page = 0;
8343         }
8344
8345         /*
8346          * Exiting from L2 to L1, we're now back to L1 which thinks it just
8347          * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
8348          * success or failure flag accordingly.
8349          */
8350         if (unlikely(vmx->fail)) {
8351                 vmx->fail = 0;
8352                 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
8353         } else
8354                 nested_vmx_succeed(vcpu);
8355         if (enable_shadow_vmcs)
8356                 vmx->nested.sync_shadow_vmcs = true;
8357 }
8358
8359 /*
8360  * L1's failure to enter L2 is a subset of a normal exit, as explained in
8361  * 23.7 "VM-entry failures during or after loading guest state" (this also
8362  * lists the acceptable exit-reason and exit-qualification parameters).
8363  * It should only be called before L2 actually succeeded to run, and when
8364  * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
8365  */
8366 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
8367                         struct vmcs12 *vmcs12,
8368                         u32 reason, unsigned long qualification)
8369 {
8370         load_vmcs12_host_state(vcpu, vmcs12);
8371         vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
8372         vmcs12->exit_qualification = qualification;
8373         nested_vmx_succeed(vcpu);
8374         if (enable_shadow_vmcs)
8375                 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
8376 }
8377
8378 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
8379                                struct x86_instruction_info *info,
8380                                enum x86_intercept_stage stage)
8381 {
8382         return X86EMUL_CONTINUE;
8383 }
8384
8385 static struct kvm_x86_ops vmx_x86_ops = {
8386         .cpu_has_kvm_support = cpu_has_kvm_support,
8387         .disabled_by_bios = vmx_disabled_by_bios,
8388         .hardware_setup = hardware_setup,
8389         .hardware_unsetup = hardware_unsetup,
8390         .check_processor_compatibility = vmx_check_processor_compat,
8391         .hardware_enable = hardware_enable,
8392         .hardware_disable = hardware_disable,
8393         .cpu_has_accelerated_tpr = report_flexpriority,
8394
8395         .vcpu_create = vmx_create_vcpu,
8396         .vcpu_free = vmx_free_vcpu,
8397         .vcpu_reset = vmx_vcpu_reset,
8398
8399         .prepare_guest_switch = vmx_save_host_state,
8400         .vcpu_load = vmx_vcpu_load,
8401         .vcpu_put = vmx_vcpu_put,
8402
8403         .update_db_bp_intercept = update_exception_bitmap,
8404         .get_msr = vmx_get_msr,
8405         .set_msr = vmx_set_msr,
8406         .get_segment_base = vmx_get_segment_base,
8407         .get_segment = vmx_get_segment,
8408         .set_segment = vmx_set_segment,
8409         .get_cpl = vmx_get_cpl,
8410         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
8411         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
8412         .decache_cr3 = vmx_decache_cr3,
8413         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
8414         .set_cr0 = vmx_set_cr0,
8415         .set_cr3 = vmx_set_cr3,
8416         .set_cr4 = vmx_set_cr4,
8417         .set_efer = vmx_set_efer,
8418         .get_idt = vmx_get_idt,
8419         .set_idt = vmx_set_idt,
8420         .get_gdt = vmx_get_gdt,
8421         .set_gdt = vmx_set_gdt,
8422         .set_dr7 = vmx_set_dr7,
8423         .cache_reg = vmx_cache_reg,
8424         .get_rflags = vmx_get_rflags,
8425         .set_rflags = vmx_set_rflags,
8426         .fpu_activate = vmx_fpu_activate,
8427         .fpu_deactivate = vmx_fpu_deactivate,
8428
8429         .tlb_flush = vmx_flush_tlb,
8430
8431         .run = vmx_vcpu_run,
8432         .handle_exit = vmx_handle_exit,
8433         .skip_emulated_instruction = skip_emulated_instruction,
8434         .set_interrupt_shadow = vmx_set_interrupt_shadow,
8435         .get_interrupt_shadow = vmx_get_interrupt_shadow,
8436         .patch_hypercall = vmx_patch_hypercall,
8437         .set_irq = vmx_inject_irq,
8438         .set_nmi = vmx_inject_nmi,
8439         .queue_exception = vmx_queue_exception,
8440         .cancel_injection = vmx_cancel_injection,
8441         .interrupt_allowed = vmx_interrupt_allowed,
8442         .nmi_allowed = vmx_nmi_allowed,
8443         .get_nmi_mask = vmx_get_nmi_mask,
8444         .set_nmi_mask = vmx_set_nmi_mask,
8445         .enable_nmi_window = enable_nmi_window,
8446         .enable_irq_window = enable_irq_window,
8447         .update_cr8_intercept = update_cr8_intercept,
8448         .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
8449         .vm_has_apicv = vmx_vm_has_apicv,
8450         .load_eoi_exitmap = vmx_load_eoi_exitmap,
8451         .hwapic_irr_update = vmx_hwapic_irr_update,
8452         .hwapic_isr_update = vmx_hwapic_isr_update,
8453         .sync_pir_to_irr = vmx_sync_pir_to_irr,
8454         .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
8455
8456         .set_tss_addr = vmx_set_tss_addr,
8457         .get_tdp_level = get_ept_level,
8458         .get_mt_mask = vmx_get_mt_mask,
8459
8460         .get_exit_info = vmx_get_exit_info,
8461
8462         .get_lpage_level = vmx_get_lpage_level,
8463
8464         .cpuid_update = vmx_cpuid_update,
8465
8466         .rdtscp_supported = vmx_rdtscp_supported,
8467         .invpcid_supported = vmx_invpcid_supported,
8468
8469         .set_supported_cpuid = vmx_set_supported_cpuid,
8470
8471         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
8472
8473         .set_tsc_khz = vmx_set_tsc_khz,
8474         .read_tsc_offset = vmx_read_tsc_offset,
8475         .write_tsc_offset = vmx_write_tsc_offset,
8476         .adjust_tsc_offset = vmx_adjust_tsc_offset,
8477         .compute_tsc_offset = vmx_compute_tsc_offset,
8478         .read_l1_tsc = vmx_read_l1_tsc,
8479
8480         .set_tdp_cr3 = vmx_set_cr3,
8481
8482         .check_intercept = vmx_check_intercept,
8483         .handle_external_intr = vmx_handle_external_intr,
8484 };
8485
8486 static int __init vmx_init(void)
8487 {
8488         int r, i, msr;
8489
8490         rdmsrl_safe(MSR_EFER, &host_efer);
8491
8492         for (i = 0; i < NR_VMX_MSR; ++i)
8493                 kvm_define_shared_msr(i, vmx_msr_index[i]);
8494
8495         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
8496         if (!vmx_io_bitmap_a)
8497                 return -ENOMEM;
8498
8499         r = -ENOMEM;
8500
8501         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
8502         if (!vmx_io_bitmap_b)
8503                 goto out;
8504
8505         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
8506         if (!vmx_msr_bitmap_legacy)
8507                 goto out1;
8508
8509         vmx_msr_bitmap_legacy_x2apic =
8510                                 (unsigned long *)__get_free_page(GFP_KERNEL);
8511         if (!vmx_msr_bitmap_legacy_x2apic)
8512                 goto out2;
8513
8514         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
8515         if (!vmx_msr_bitmap_longmode)
8516                 goto out3;
8517
8518         vmx_msr_bitmap_longmode_x2apic =
8519                                 (unsigned long *)__get_free_page(GFP_KERNEL);
8520         if (!vmx_msr_bitmap_longmode_x2apic)
8521                 goto out4;
8522         vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
8523         if (!vmx_vmread_bitmap)
8524                 goto out5;
8525
8526         vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
8527         if (!vmx_vmwrite_bitmap)
8528                 goto out6;
8529
8530         memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
8531         memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
8532         /* shadowed read/write fields */
8533         for (i = 0; i < max_shadow_read_write_fields; i++) {
8534                 clear_bit(shadow_read_write_fields[i], vmx_vmwrite_bitmap);
8535                 clear_bit(shadow_read_write_fields[i], vmx_vmread_bitmap);
8536         }
8537         /* shadowed read only fields */
8538         for (i = 0; i < max_shadow_read_only_fields; i++)
8539                 clear_bit(shadow_read_only_fields[i], vmx_vmread_bitmap);
8540
8541         /*
8542          * Allow direct access to the PC debug port (it is often used for I/O
8543          * delays, but the vmexits simply slow things down).
8544          */
8545         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
8546         clear_bit(0x80, vmx_io_bitmap_a);
8547
8548         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
8549
8550         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
8551         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
8552
8553         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8554
8555         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
8556                      __alignof__(struct vcpu_vmx), THIS_MODULE);
8557         if (r)
8558                 goto out7;
8559
8560 #ifdef CONFIG_KEXEC
8561         rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8562                            crash_vmclear_local_loaded_vmcss);
8563 #endif
8564
8565         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
8566         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
8567         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
8568         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
8569         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
8570         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
8571         memcpy(vmx_msr_bitmap_legacy_x2apic,
8572                         vmx_msr_bitmap_legacy, PAGE_SIZE);
8573         memcpy(vmx_msr_bitmap_longmode_x2apic,
8574                         vmx_msr_bitmap_longmode, PAGE_SIZE);
8575
8576         if (enable_apicv) {
8577                 for (msr = 0x800; msr <= 0x8ff; msr++)
8578                         vmx_disable_intercept_msr_read_x2apic(msr);
8579
8580                 /* According SDM, in x2apic mode, the whole id reg is used.
8581                  * But in KVM, it only use the highest eight bits. Need to
8582                  * intercept it */
8583                 vmx_enable_intercept_msr_read_x2apic(0x802);
8584                 /* TMCCT */
8585                 vmx_enable_intercept_msr_read_x2apic(0x839);
8586                 /* TPR */
8587                 vmx_disable_intercept_msr_write_x2apic(0x808);
8588                 /* EOI */
8589                 vmx_disable_intercept_msr_write_x2apic(0x80b);
8590                 /* SELF-IPI */
8591                 vmx_disable_intercept_msr_write_x2apic(0x83f);
8592         }
8593
8594         if (enable_ept) {
8595                 kvm_mmu_set_mask_ptes(0ull,
8596                         (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
8597                         (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
8598                         0ull, VMX_EPT_EXECUTABLE_MASK);
8599                 ept_set_mmio_spte_mask();
8600                 kvm_enable_tdp();
8601         } else
8602                 kvm_disable_tdp();
8603
8604         return 0;
8605
8606 out7:
8607         free_page((unsigned long)vmx_vmwrite_bitmap);
8608 out6:
8609         free_page((unsigned long)vmx_vmread_bitmap);
8610 out5:
8611         free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
8612 out4:
8613         free_page((unsigned long)vmx_msr_bitmap_longmode);
8614 out3:
8615         free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
8616 out2:
8617         free_page((unsigned long)vmx_msr_bitmap_legacy);
8618 out1:
8619         free_page((unsigned long)vmx_io_bitmap_b);
8620 out:
8621         free_page((unsigned long)vmx_io_bitmap_a);
8622         return r;
8623 }
8624
8625 static void __exit vmx_exit(void)
8626 {
8627         free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
8628         free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
8629         free_page((unsigned long)vmx_msr_bitmap_legacy);
8630         free_page((unsigned long)vmx_msr_bitmap_longmode);
8631         free_page((unsigned long)vmx_io_bitmap_b);
8632         free_page((unsigned long)vmx_io_bitmap_a);
8633         free_page((unsigned long)vmx_vmwrite_bitmap);
8634         free_page((unsigned long)vmx_vmread_bitmap);
8635
8636 #ifdef CONFIG_KEXEC
8637         rcu_assign_pointer(crash_vmclear_loaded_vmcss, NULL);
8638         synchronize_rcu();
8639 #endif
8640
8641         kvm_exit();
8642 }
8643
8644 module_init(vmx_init)
8645 module_exit(vmx_exit)