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 <linux/hrtimer.h>
  35 #include "kvm_cache_regs.h"
  36 #include "x86.h"
  37
  38 #include <asm/io.h>
  39 #include <asm/desc.h>
  40 #include <asm/vmx.h>
  41 #include <asm/virtext.h>
  42 #include <asm/mce.h>
  43 #include <asm/i387.h>
  44 #include <asm/xcr.h>
  45 #include <asm/perf_event.h>
  46 #include <asm/debugreg.h>
  47 #include <asm/kexec.h>
  48
  49 #include "trace.h"
  50
  51 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  52 #define __ex_clear(x, reg) \
  53         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
  54
  55 MODULE_AUTHOR("Qumranet");
  56 MODULE_LICENSE("GPL");
  57
  58 static const struct x86_cpu_id vmx_cpu_id[] = {
  59         X86_FEATURE_MATCH(X86_FEATURE_VMX),
  60         {}
  61 };
  62 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
  63
  64 static bool __read_mostly enable_vpid = 1;
  65 module_param_named(vpid, enable_vpid, bool, 0444);
  66
  67 static bool __read_mostly flexpriority_enabled = 1;
  68 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  69
  70 static bool __read_mostly enable_ept = 1;
  71 module_param_named(ept, enable_ept, bool, S_IRUGO);
  72
  73 static bool __read_mostly enable_unrestricted_guest = 1;
  74 module_param_named(unrestricted_guest,
  75                         enable_unrestricted_guest, bool, S_IRUGO);
  76
  77 static bool __read_mostly enable_ept_ad_bits = 1;
  78 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
  79
  80 static bool __read_mostly emulate_invalid_guest_state = true;
  81 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  82
  83 static bool __read_mostly vmm_exclusive = 1;
  84 module_param(vmm_exclusive, bool, S_IRUGO);
  85
  86 static bool __read_mostly fasteoi = 1;
  87 module_param(fasteoi, bool, S_IRUGO);
  88
  89 static bool __read_mostly enable_apicv = 1;
  90 module_param(enable_apicv, bool, S_IRUGO);
  91
  92 static bool __read_mostly enable_shadow_vmcs = 1;
  93 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
  94 /*
  95  * If nested=1, nested virtualization is supported, i.e., guests may use
  96  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
  97  * use VMX instructions.
  98  */
  99 static bool __read_mostly nested = 0;
 100 module_param(nested, bool, S_IRUGO);
 101
 102 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
 103 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
 104 #define KVM_VM_CR0_ALWAYS_ON                                            \
 105         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
 106 #define KVM_CR4_GUEST_OWNED_BITS                                      \
 107         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
 108          | X86_CR4_OSXMMEXCPT)
 109
 110 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
 111 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
 112
 113 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
 114
 115 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
 116
 117 /*
 118  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 119  * ple_gap:    upper bound on the amount of time between two successive
 120  *             executions of PAUSE in a loop. Also indicate if ple enabled.
 121  *             According to test, this time is usually smaller than 128 cycles.
 122  * ple_window: upper bound on the amount of time a guest is allowed to execute
 123  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
 124  *             less than 2^12 cycles
 125  * Time is measured based on a counter that runs at the same rate as the TSC,
 126  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 127  */
 128 #define KVM_VMX_DEFAULT_PLE_GAP    128
 129 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
 130 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 131 module_param(ple_gap, int, S_IRUGO);
 132
 133 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 134 module_param(ple_window, int, S_IRUGO);
 135
 136 extern const ulong vmx_return;
 137
 138 #define NR_AUTOLOAD_MSRS 8
 139 #define VMCS02_POOL_SIZE 1
 140
 141 struct vmcs {
 142         u32 revision_id;
 143         u32 abort;
 144         char data[0];
 145 };
 146
 147 /*
 148  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
 149  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
 150  * loaded on this CPU (so we can clear them if the CPU goes down).
 151  */
 152 struct loaded_vmcs {
 153         struct vmcs *vmcs;
 154         int cpu;
 155         int launched;
 156         struct list_head loaded_vmcss_on_cpu_link;
 157 };
 158
 159 struct shared_msr_entry {
 160         unsigned index;
 161         u64 data;
 162         u64 mask;
 163 };
 164
 165 /*
 166  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
 167  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
 168  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
 169  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
 170  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
 171  * More than one of these structures may exist, if L1 runs multiple L2 guests.
 172  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
 173  * underlying hardware which will be used to run L2.
 174  * This structure is packed to ensure that its layout is identical across
 175  * machines (necessary for live migration).
 176  * If there are changes in this struct, VMCS12_REVISION must be changed.
 177  */
 178 typedef u64 natural_width;
 179 struct __packed vmcs12 {
 180         /* According to the Intel spec, a VMCS region must start with the
 181          * following two fields. Then follow implementation-specific data.
 182          */
 183         u32 revision_id;
 184         u32 abort;
 185
 186         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
 187         u32 padding[7]; /* room for future expansion */
 188
 189         u64 io_bitmap_a;
 190         u64 io_bitmap_b;
 191         u64 msr_bitmap;
 192         u64 vm_exit_msr_store_addr;
 193         u64 vm_exit_msr_load_addr;
 194         u64 vm_entry_msr_load_addr;
 195         u64 tsc_offset;
 196         u64 virtual_apic_page_addr;
 197         u64 apic_access_addr;
 198         u64 ept_pointer;
 199         u64 guest_physical_address;
 200         u64 vmcs_link_pointer;
 201         u64 guest_ia32_debugctl;
 202         u64 guest_ia32_pat;
 203         u64 guest_ia32_efer;
 204         u64 guest_ia32_perf_global_ctrl;
 205         u64 guest_pdptr0;
 206         u64 guest_pdptr1;
 207         u64 guest_pdptr2;
 208         u64 guest_pdptr3;
 209         u64 guest_bndcfgs;
 210         u64 host_ia32_pat;
 211         u64 host_ia32_efer;
 212         u64 host_ia32_perf_global_ctrl;
 213         u64 padding64[8]; /* room for future expansion */
 214         /*
 215          * To allow migration of L1 (complete with its L2 guests) between
 216          * machines of different natural widths (32 or 64 bit), we cannot have
 217          * unsigned long fields with no explict size. We use u64 (aliased
 218          * natural_width) instead. Luckily, x86 is little-endian.
 219          */
 220         natural_width cr0_guest_host_mask;
 221         natural_width cr4_guest_host_mask;
 222         natural_width cr0_read_shadow;
 223         natural_width cr4_read_shadow;
 224         natural_width cr3_target_value0;
 225         natural_width cr3_target_value1;
 226         natural_width cr3_target_value2;
 227         natural_width cr3_target_value3;
 228         natural_width exit_qualification;
 229         natural_width guest_linear_address;
 230         natural_width guest_cr0;
 231         natural_width guest_cr3;
 232         natural_width guest_cr4;
 233         natural_width guest_es_base;
 234         natural_width guest_cs_base;
 235         natural_width guest_ss_base;
 236         natural_width guest_ds_base;
 237         natural_width guest_fs_base;
 238         natural_width guest_gs_base;
 239         natural_width guest_ldtr_base;
 240         natural_width guest_tr_base;
 241         natural_width guest_gdtr_base;
 242         natural_width guest_idtr_base;
 243         natural_width guest_dr7;
 244         natural_width guest_rsp;
 245         natural_width guest_rip;
 246         natural_width guest_rflags;
 247         natural_width guest_pending_dbg_exceptions;
 248         natural_width guest_sysenter_esp;
 249         natural_width guest_sysenter_eip;
 250         natural_width host_cr0;
 251         natural_width host_cr3;
 252         natural_width host_cr4;
 253         natural_width host_fs_base;
 254         natural_width host_gs_base;
 255         natural_width host_tr_base;
 256         natural_width host_gdtr_base;
 257         natural_width host_idtr_base;
 258         natural_width host_ia32_sysenter_esp;
 259         natural_width host_ia32_sysenter_eip;
 260         natural_width host_rsp;
 261         natural_width host_rip;
 262         natural_width paddingl[8]; /* room for future expansion */
 263         u32 pin_based_vm_exec_control;
 264         u32 cpu_based_vm_exec_control;
 265         u32 exception_bitmap;
 266         u32 page_fault_error_code_mask;
 267         u32 page_fault_error_code_match;
 268         u32 cr3_target_count;
 269         u32 vm_exit_controls;
 270         u32 vm_exit_msr_store_count;
 271         u32 vm_exit_msr_load_count;
 272         u32 vm_entry_controls;
 273         u32 vm_entry_msr_load_count;
 274         u32 vm_entry_intr_info_field;
 275         u32 vm_entry_exception_error_code;
 276         u32 vm_entry_instruction_len;
 277         u32 tpr_threshold;
 278         u32 secondary_vm_exec_control;
 279         u32 vm_instruction_error;
 280         u32 vm_exit_reason;
 281         u32 vm_exit_intr_info;
 282         u32 vm_exit_intr_error_code;
 283         u32 idt_vectoring_info_field;
 284         u32 idt_vectoring_error_code;
 285         u32 vm_exit_instruction_len;
 286         u32 vmx_instruction_info;
 287         u32 guest_es_limit;
 288         u32 guest_cs_limit;
 289         u32 guest_ss_limit;
 290         u32 guest_ds_limit;
 291         u32 guest_fs_limit;
 292         u32 guest_gs_limit;
 293         u32 guest_ldtr_limit;
 294         u32 guest_tr_limit;
 295         u32 guest_gdtr_limit;
 296         u32 guest_idtr_limit;
 297         u32 guest_es_ar_bytes;
 298         u32 guest_cs_ar_bytes;
 299         u32 guest_ss_ar_bytes;
 300         u32 guest_ds_ar_bytes;
 301         u32 guest_fs_ar_bytes;
 302         u32 guest_gs_ar_bytes;
 303         u32 guest_ldtr_ar_bytes;
 304         u32 guest_tr_ar_bytes;
 305         u32 guest_interruptibility_info;
 306         u32 guest_activity_state;
 307         u32 guest_sysenter_cs;
 308         u32 host_ia32_sysenter_cs;
 309         u32 vmx_preemption_timer_value;
 310         u32 padding32[7]; /* room for future expansion */
 311         u16 virtual_processor_id;
 312         u16 guest_es_selector;
 313         u16 guest_cs_selector;
 314         u16 guest_ss_selector;
 315         u16 guest_ds_selector;
 316         u16 guest_fs_selector;
 317         u16 guest_gs_selector;
 318         u16 guest_ldtr_selector;
 319         u16 guest_tr_selector;
 320         u16 host_es_selector;
 321         u16 host_cs_selector;
 322         u16 host_ss_selector;
 323         u16 host_ds_selector;
 324         u16 host_fs_selector;
 325         u16 host_gs_selector;
 326         u16 host_tr_selector;
 327 };
 328
 329 /*
 330  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
 331  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
 332  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
 333  */
 334 #define VMCS12_REVISION 0x11e57ed0
 335
 336 /*
 337  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
 338  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
 339  * current implementation, 4K are reserved to avoid future complications.
 340  */
 341 #define VMCS12_SIZE 0x1000
 342
 343 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
 344 struct vmcs02_list {
 345         struct list_head list;
 346         gpa_t vmptr;
 347         struct loaded_vmcs vmcs02;
 348 };
 349
 350 /*
 351  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 352  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 353  */
 354 struct nested_vmx {
 355         /* Has the level1 guest done vmxon? */
 356         bool vmxon;
 357         gpa_t vmxon_ptr;
 358
 359         /* The guest-physical address of the current VMCS L1 keeps for L2 */
 360         gpa_t current_vmptr;
 361         /* The host-usable pointer to the above */
 362         struct page *current_vmcs12_page;
 363         struct vmcs12 *current_vmcs12;
 364         struct vmcs *current_shadow_vmcs;
 365         /*
 366          * Indicates if the shadow vmcs must be updated with the
 367          * data hold by vmcs12
 368          */
 369         bool sync_shadow_vmcs;
 370
 371         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
 372         struct list_head vmcs02_pool;
 373         int vmcs02_num;
 374         u64 vmcs01_tsc_offset;
 375         /* L2 must run next, and mustn't decide to exit to L1. */
 376         bool nested_run_pending;
 377         /*
 378          * Guest pages referred to in vmcs02 with host-physical pointers, so
 379          * we must keep them pinned while L2 runs.
 380          */
 381         struct page *apic_access_page;
 382         u64 msr_ia32_feature_control;
 383
 384         struct hrtimer preemption_timer;
 385         bool preemption_timer_expired;
 386 };
 387
 388 #define POSTED_INTR_ON  0
 389 /* Posted-Interrupt Descriptor */
 390 struct pi_desc {
 391         u32 pir[8];     /* Posted interrupt requested */
 392         u32 control;    /* bit 0 of control is outstanding notification bit */
 393         u32 rsvd[7];
 394 } __aligned(64);
 395
 396 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
 397 {
 398         return test_and_set_bit(POSTED_INTR_ON,
 399                         (unsigned long *)&pi_desc->control);
 400 }
 401
 402 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
 403 {
 404         return test_and_clear_bit(POSTED_INTR_ON,
 405                         (unsigned long *)&pi_desc->control);
 406 }
 407
 408 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
 409 {
 410         return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
 411 }
 412
 413 struct vcpu_vmx {
 414         struct kvm_vcpu       vcpu;
 415         unsigned long         host_rsp;
 416         u8                    fail;
 417         bool                  nmi_known_unmasked;
 418         u32                   exit_intr_info;
 419         u32                   idt_vectoring_info;
 420         ulong                 rflags;
 421         struct shared_msr_entry *guest_msrs;
 422         int                   nmsrs;
 423         int                   save_nmsrs;
 424         unsigned long         host_idt_base;
 425 #ifdef CONFIG_X86_64
 426         u64                   msr_host_kernel_gs_base;
 427         u64                   msr_guest_kernel_gs_base;
 428 #endif
 429         u32 vm_entry_controls_shadow;
 430         u32 vm_exit_controls_shadow;
 431         /*
 432          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 433          * non-nested (L1) guest, it always points to vmcs01. For a nested
 434          * guest (L2), it points to a different VMCS.
 435          */
 436         struct loaded_vmcs    vmcs01;
 437         struct loaded_vmcs   *loaded_vmcs;
 438         bool                  __launched; /* temporary, used in vmx_vcpu_run */
 439         struct msr_autoload {
 440                 unsigned nr;
 441                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 442                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 443         } msr_autoload;
 444         struct {
 445                 int           loaded;
 446                 u16           fs_sel, gs_sel, ldt_sel;
 447 #ifdef CONFIG_X86_64
 448                 u16           ds_sel, es_sel;
 449 #endif
 450                 int           gs_ldt_reload_needed;
 451                 int           fs_reload_needed;
 452                 u64           msr_host_bndcfgs;
 453         } host_state;
 454         struct {
 455                 int vm86_active;
 456                 ulong save_rflags;
 457                 struct kvm_segment segs[8];
 458         } rmode;
 459         struct {
 460                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 461                 struct kvm_save_segment {
 462                         u16 selector;
 463                         unsigned long base;
 464                         u32 limit;
 465                         u32 ar;
 466                 } seg[8];
 467         } segment_cache;
 468         int vpid;
 469         bool emulation_required;
 470
 471         /* Support for vnmi-less CPUs */
 472         int soft_vnmi_blocked;
 473         ktime_t entry_time;
 474         s64 vnmi_blocked_time;
 475         u32 exit_reason;
 476
 477         bool rdtscp_enabled;
 478
 479         /* Posted interrupt descriptor */
 480         struct pi_desc pi_desc;
 481
 482         /* Support for a guest hypervisor (nested VMX) */
 483         struct nested_vmx nested;
 484 };
 485
 486 enum segment_cache_field {
 487         SEG_FIELD_SEL = 0,
 488         SEG_FIELD_BASE = 1,
 489         SEG_FIELD_LIMIT = 2,
 490         SEG_FIELD_AR = 3,
 491
 492         SEG_FIELD_NR = 4
 493 };
 494
 495 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 496 {
 497         return container_of(vcpu, struct vcpu_vmx, vcpu);
 498 }
 499
 500 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
 501 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
 502 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
 503                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
 504
 505
 506 static unsigned long shadow_read_only_fields[] = {
 507         /*
 508          * We do NOT shadow fields that are modified when L0
 509          * traps and emulates any vmx instruction (e.g. VMPTRLD,
 510          * VMXON...) executed by L1.
 511          * For example, VM_INSTRUCTION_ERROR is read
 512          * by L1 if a vmx instruction fails (part of the error path).
 513          * Note the code assumes this logic. If for some reason
 514          * we start shadowing these fields then we need to
 515          * force a shadow sync when L0 emulates vmx instructions
 516          * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
 517          * by nested_vmx_failValid)
 518          */
 519         VM_EXIT_REASON,
 520         VM_EXIT_INTR_INFO,
 521         VM_EXIT_INSTRUCTION_LEN,
 522         IDT_VECTORING_INFO_FIELD,
 523         IDT_VECTORING_ERROR_CODE,
 524         VM_EXIT_INTR_ERROR_CODE,
 525         EXIT_QUALIFICATION,
 526         GUEST_LINEAR_ADDRESS,
 527         GUEST_PHYSICAL_ADDRESS
 528 };
 529 static int max_shadow_read_only_fields =
 530         ARRAY_SIZE(shadow_read_only_fields);
 531
 532 static unsigned long shadow_read_write_fields[] = {
 533         GUEST_RIP,
 534         GUEST_RSP,
 535         GUEST_CR0,
 536         GUEST_CR3,
 537         GUEST_CR4,
 538         GUEST_INTERRUPTIBILITY_INFO,
 539         GUEST_RFLAGS,
 540         GUEST_CS_SELECTOR,
 541         GUEST_CS_AR_BYTES,
 542         GUEST_CS_LIMIT,
 543         GUEST_CS_BASE,
 544         GUEST_ES_BASE,
 545         GUEST_BNDCFGS,
 546         CR0_GUEST_HOST_MASK,
 547         CR0_READ_SHADOW,
 548         CR4_READ_SHADOW,
 549         TSC_OFFSET,
 550         EXCEPTION_BITMAP,
 551         CPU_BASED_VM_EXEC_CONTROL,
 552         VM_ENTRY_EXCEPTION_ERROR_CODE,
 553         VM_ENTRY_INTR_INFO_FIELD,
 554         VM_ENTRY_INSTRUCTION_LEN,
 555         VM_ENTRY_EXCEPTION_ERROR_CODE,
 556         HOST_FS_BASE,
 557         HOST_GS_BASE,
 558         HOST_FS_SELECTOR,
 559         HOST_GS_SELECTOR
 560 };
 561 static int max_shadow_read_write_fields =
 562         ARRAY_SIZE(shadow_read_write_fields);
 563
 564 static const unsigned short vmcs_field_to_offset_table[] = {
 565         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 566         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
 567         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
 568         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
 569         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
 570         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
 571         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
 572         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
 573         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
 574         FIELD(HOST_ES_SELECTOR, host_es_selector),
 575         FIELD(HOST_CS_SELECTOR, host_cs_selector),
 576         FIELD(HOST_SS_SELECTOR, host_ss_selector),
 577         FIELD(HOST_DS_SELECTOR, host_ds_selector),
 578         FIELD(HOST_FS_SELECTOR, host_fs_selector),
 579         FIELD(HOST_GS_SELECTOR, host_gs_selector),
 580         FIELD(HOST_TR_SELECTOR, host_tr_selector),
 581         FIELD64(IO_BITMAP_A, io_bitmap_a),
 582         FIELD64(IO_BITMAP_B, io_bitmap_b),
 583         FIELD64(MSR_BITMAP, msr_bitmap),
 584         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
 585         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
 586         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
 587         FIELD64(TSC_OFFSET, tsc_offset),
 588         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
 589         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
 590         FIELD64(EPT_POINTER, ept_pointer),
 591         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
 592         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
 593         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
 594         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
 595         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
 596         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
 597         FIELD64(GUEST_PDPTR0, guest_pdptr0),
 598         FIELD64(GUEST_PDPTR1, guest_pdptr1),
 599         FIELD64(GUEST_PDPTR2, guest_pdptr2),
 600         FIELD64(GUEST_PDPTR3, guest_pdptr3),
 601         FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
 602         FIELD64(HOST_IA32_PAT, host_ia32_pat),
 603         FIELD64(HOST_IA32_EFER, host_ia32_efer),
 604         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
 605         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
 606         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
 607         FIELD(EXCEPTION_BITMAP, exception_bitmap),
 608         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
 609         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
 610         FIELD(CR3_TARGET_COUNT, cr3_target_count),
 611         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
 612         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
 613         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
 614         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
 615         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
 616         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
 617         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
 618         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
 619         FIELD(TPR_THRESHOLD, tpr_threshold),
 620         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
 621         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
 622         FIELD(VM_EXIT_REASON, vm_exit_reason),
 623         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
 624         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
 625         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
 626         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
 627         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
 628         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
 629         FIELD(GUEST_ES_LIMIT, guest_es_limit),
 630         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
 631         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
 632         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
 633         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
 634         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
 635         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
 636         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
 637         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
 638         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
 639         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
 640         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
 641         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
 642         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
 643         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
 644         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
 645         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
 646         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
 647         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
 648         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
 649         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
 650         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
 651         FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
 652         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
 653         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
 654         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
 655         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
 656         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
 657         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
 658         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
 659         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
 660         FIELD(EXIT_QUALIFICATION, exit_qualification),
 661         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
 662         FIELD(GUEST_CR0, guest_cr0),
 663         FIELD(GUEST_CR3, guest_cr3),
 664         FIELD(GUEST_CR4, guest_cr4),
 665         FIELD(GUEST_ES_BASE, guest_es_base),
 666         FIELD(GUEST_CS_BASE, guest_cs_base),
 667         FIELD(GUEST_SS_BASE, guest_ss_base),
 668         FIELD(GUEST_DS_BASE, guest_ds_base),
 669         FIELD(GUEST_FS_BASE, guest_fs_base),
 670         FIELD(GUEST_GS_BASE, guest_gs_base),
 671         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
 672         FIELD(GUEST_TR_BASE, guest_tr_base),
 673         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
 674         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
 675         FIELD(GUEST_DR7, guest_dr7),
 676         FIELD(GUEST_RSP, guest_rsp),
 677         FIELD(GUEST_RIP, guest_rip),
 678         FIELD(GUEST_RFLAGS, guest_rflags),
 679         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
 680         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
 681         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
 682         FIELD(HOST_CR0, host_cr0),
 683         FIELD(HOST_CR3, host_cr3),
 684         FIELD(HOST_CR4, host_cr4),
 685         FIELD(HOST_FS_BASE, host_fs_base),
 686         FIELD(HOST_GS_BASE, host_gs_base),
 687         FIELD(HOST_TR_BASE, host_tr_base),
 688         FIELD(HOST_GDTR_BASE, host_gdtr_base),
 689         FIELD(HOST_IDTR_BASE, host_idtr_base),
 690         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
 691         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
 692         FIELD(HOST_RSP, host_rsp),
 693         FIELD(HOST_RIP, host_rip),
 694 };
 695 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
 696
 697 static inline short vmcs_field_to_offset(unsigned long field)
 698 {
 699         if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
 700                 return -1;
 701         return vmcs_field_to_offset_table[field];
 702 }
 703
 704 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
 705 {
 706         return to_vmx(vcpu)->nested.current_vmcs12;
 707 }
 708
 709 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
 710 {
 711         struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
 712         if (is_error_page(page))
 713                 return NULL;
 714
 715         return page;
 716 }
 717
 718 static void nested_release_page(struct page *page)
 719 {
 720         kvm_release_page_dirty(page);
 721 }
 722
 723 static void nested_release_page_clean(struct page *page)
 724 {
 725         kvm_release_page_clean(page);
 726 }
 727
 728 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
 729 static u64 construct_eptp(unsigned long root_hpa);
 730 static void kvm_cpu_vmxon(u64 addr);
 731 static void kvm_cpu_vmxoff(void);
 732 static bool vmx_mpx_supported(void);
 733 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 734 static void vmx_set_segment(struct kvm_vcpu *vcpu,
 735                             struct kvm_segment *var, int seg);
 736 static void vmx_get_segment(struct kvm_vcpu *vcpu,
 737                             struct kvm_segment *var, int seg);
 738 static bool guest_state_valid(struct kvm_vcpu *vcpu);
 739 static u32 vmx_segment_access_rights(struct kvm_segment *var);
 740 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
 741 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
 742 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
 743 static bool vmx_mpx_supported(void);
 744
 745 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 746 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 747 /*
 748  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
 749  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
 750  */
 751 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
 752 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 753
 754 static unsigned long *vmx_io_bitmap_a;
 755 static unsigned long *vmx_io_bitmap_b;
 756 static unsigned long *vmx_msr_bitmap_legacy;
 757 static unsigned long *vmx_msr_bitmap_longmode;
 758 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
 759 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
 760 static unsigned long *vmx_vmread_bitmap;
 761 static unsigned long *vmx_vmwrite_bitmap;
 762
 763 static bool cpu_has_load_ia32_efer;
 764 static bool cpu_has_load_perf_global_ctrl;
 765
 766 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 767 static DEFINE_SPINLOCK(vmx_vpid_lock);
 768
 769 static struct vmcs_config {
 770         int size;
 771         int order;
 772         u32 revision_id;
 773         u32 pin_based_exec_ctrl;
 774         u32 cpu_based_exec_ctrl;
 775         u32 cpu_based_2nd_exec_ctrl;
 776         u32 vmexit_ctrl;
 777         u32 vmentry_ctrl;
 778 } vmcs_config;
 779
 780 static struct vmx_capability {
 781         u32 ept;
 782         u32 vpid;
 783 } vmx_capability;
 784
 785 #define VMX_SEGMENT_FIELD(seg)                                  \
 786         [VCPU_SREG_##seg] = {                                   \
 787                 .selector = GUEST_##seg##_SELECTOR,             \
 788                 .base = GUEST_##seg##_BASE,                     \
 789                 .limit = GUEST_##seg##_LIMIT,                   \
 790                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 791         }
 792
 793 static const struct kvm_vmx_segment_field {
 794         unsigned selector;
 795         unsigned base;
 796         unsigned limit;
 797         unsigned ar_bytes;
 798 } kvm_vmx_segment_fields[] = {
 799         VMX_SEGMENT_FIELD(CS),
 800         VMX_SEGMENT_FIELD(DS),
 801         VMX_SEGMENT_FIELD(ES),
 802         VMX_SEGMENT_FIELD(FS),
 803         VMX_SEGMENT_FIELD(GS),
 804         VMX_SEGMENT_FIELD(SS),
 805         VMX_SEGMENT_FIELD(TR),
 806         VMX_SEGMENT_FIELD(LDTR),
 807 };
 808
 809 static u64 host_efer;
 810
 811 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 812
 813 /*
 814  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
 815  * away by decrementing the array size.
 816  */
 817 static const u32 vmx_msr_index[] = {
 818 #ifdef CONFIG_X86_64
 819         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 820 #endif
 821         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
 822 };
 823 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 824
 825 static inline bool is_page_fault(u32 intr_info)
 826 {
 827         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 828                              INTR_INFO_VALID_MASK)) ==
 829                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 830 }
 831
 832 static inline bool is_no_device(u32 intr_info)
 833 {
 834         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 835                              INTR_INFO_VALID_MASK)) ==
 836                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 837 }
 838
 839 static inline bool is_invalid_opcode(u32 intr_info)
 840 {
 841         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 842                              INTR_INFO_VALID_MASK)) ==
 843                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 844 }
 845
 846 static inline bool is_external_interrupt(u32 intr_info)
 847 {
 848         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 849                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 850 }
 851
 852 static inline bool is_machine_check(u32 intr_info)
 853 {
 854         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 855                              INTR_INFO_VALID_MASK)) ==
 856                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
 857 }
 858
 859 static inline bool cpu_has_vmx_msr_bitmap(void)
 860 {
 861         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
 862 }
 863
 864 static inline bool cpu_has_vmx_tpr_shadow(void)
 865 {
 866         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
 867 }
 868
 869 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
 870 {
 871         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
 872 }
 873
 874 static inline bool cpu_has_secondary_exec_ctrls(void)
 875 {
 876         return vmcs_config.cpu_based_exec_ctrl &
 877                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
 878 }
 879
 880 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 881 {
 882         return vmcs_config.cpu_based_2nd_exec_ctrl &
 883                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
 884 }
 885
 886 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
 887 {
 888         return vmcs_config.cpu_based_2nd_exec_ctrl &
 889                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
 890 }
 891
 892 static inline bool cpu_has_vmx_apic_register_virt(void)
 893 {
 894         return vmcs_config.cpu_based_2nd_exec_ctrl &
 895                 SECONDARY_EXEC_APIC_REGISTER_VIRT;
 896 }
 897
 898 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
 899 {
 900         return vmcs_config.cpu_based_2nd_exec_ctrl &
 901                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
 902 }
 903
 904 static inline bool cpu_has_vmx_posted_intr(void)
 905 {
 906         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
 907 }
 908
 909 static inline bool cpu_has_vmx_apicv(void)
 910 {
 911         return cpu_has_vmx_apic_register_virt() &&
 912                 cpu_has_vmx_virtual_intr_delivery() &&
 913                 cpu_has_vmx_posted_intr();
 914 }
 915
 916 static inline bool cpu_has_vmx_flexpriority(void)
 917 {
 918         return cpu_has_vmx_tpr_shadow() &&
 919                 cpu_has_vmx_virtualize_apic_accesses();
 920 }
 921
 922 static inline bool cpu_has_vmx_ept_execute_only(void)
 923 {
 924         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
 925 }
 926
 927 static inline bool cpu_has_vmx_eptp_uncacheable(void)
 928 {
 929         return vmx_capability.ept & VMX_EPTP_UC_BIT;
 930 }
 931
 932 static inline bool cpu_has_vmx_eptp_writeback(void)
 933 {
 934         return vmx_capability.ept & VMX_EPTP_WB_BIT;
 935 }
 936
 937 static inline bool cpu_has_vmx_ept_2m_page(void)
 938 {
 939         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
 940 }
 941
 942 static inline bool cpu_has_vmx_ept_1g_page(void)
 943 {
 944         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
 945 }
 946
 947 static inline bool cpu_has_vmx_ept_4levels(void)
 948 {
 949         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
 950 }
 951
 952 static inline bool cpu_has_vmx_ept_ad_bits(void)
 953 {
 954         return vmx_capability.ept & VMX_EPT_AD_BIT;
 955 }
 956
 957 static inline bool cpu_has_vmx_invept_context(void)
 958 {
 959         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
 960 }
 961
 962 static inline bool cpu_has_vmx_invept_global(void)
 963 {
 964         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
 965 }
 966
 967 static inline bool cpu_has_vmx_invvpid_single(void)
 968 {
 969         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
 970 }
 971
 972 static inline bool cpu_has_vmx_invvpid_global(void)
 973 {
 974         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
 975 }
 976
 977 static inline bool cpu_has_vmx_ept(void)
 978 {
 979         return vmcs_config.cpu_based_2nd_exec_ctrl &
 980                 SECONDARY_EXEC_ENABLE_EPT;
 981 }
 982
 983 static inline bool cpu_has_vmx_unrestricted_guest(void)
 984 {
 985         return vmcs_config.cpu_based_2nd_exec_ctrl &
 986                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
 987 }
 988
 989 static inline bool cpu_has_vmx_ple(void)
 990 {
 991         return vmcs_config.cpu_based_2nd_exec_ctrl &
 992                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 993 }
 994
 995 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
 996 {
 997         return flexpriority_enabled && irqchip_in_kernel(kvm);
 998 }
 999
1000 static inline bool cpu_has_vmx_vpid(void)
1001 {
1002         return vmcs_config.cpu_based_2nd_exec_ctrl &
1003                 SECONDARY_EXEC_ENABLE_VPID;
1004 }
1005
1006 static inline bool cpu_has_vmx_rdtscp(void)
1007 {
1008         return vmcs_config.cpu_based_2nd_exec_ctrl &
1009                 SECONDARY_EXEC_RDTSCP;
1010 }
1011
1012 static inline bool cpu_has_vmx_invpcid(void)
1013 {
1014         return vmcs_config.cpu_based_2nd_exec_ctrl &
1015                 SECONDARY_EXEC_ENABLE_INVPCID;
1016 }
1017
1018 static inline bool cpu_has_virtual_nmis(void)
1019 {
1020         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1021 }
1022
1023 static inline bool cpu_has_vmx_wbinvd_exit(void)
1024 {
1025         return vmcs_config.cpu_based_2nd_exec_ctrl &
1026                 SECONDARY_EXEC_WBINVD_EXITING;
1027 }
1028
1029 static inline bool cpu_has_vmx_shadow_vmcs(void)
1030 {
1031         u64 vmx_msr;
1032         rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1033         /* check if the cpu supports writing r/o exit information fields */
1034         if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1035                 return false;
1036
1037         return vmcs_config.cpu_based_2nd_exec_ctrl &
1038                 SECONDARY_EXEC_SHADOW_VMCS;
1039 }
1040
1041 static inline bool report_flexpriority(void)
1042 {
1043         return flexpriority_enabled;
1044 }
1045
1046 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1047 {
1048         return vmcs12->cpu_based_vm_exec_control & bit;
1049 }
1050
1051 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1052 {
1053         return (vmcs12->cpu_based_vm_exec_control &
1054                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1055                 (vmcs12->secondary_vm_exec_control & bit);
1056 }
1057
1058 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1059 {
1060         return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1061 }
1062
1063 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1064 {
1065         return vmcs12->pin_based_vm_exec_control &
1066                 PIN_BASED_VMX_PREEMPTION_TIMER;
1067 }
1068
1069 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1070 {
1071         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1072 }
1073
1074 static inline bool is_exception(u32 intr_info)
1075 {
1076         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1077                 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1078 }
1079
1080 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1081                               u32 exit_intr_info,
1082                               unsigned long exit_qualification);
1083 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1084                         struct vmcs12 *vmcs12,
1085                         u32 reason, unsigned long qualification);
1086
1087 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1088 {
1089         int i;
1090
1091         for (i = 0; i < vmx->nmsrs; ++i)
1092                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1093                         return i;
1094         return -1;
1095 }
1096
1097 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1098 {
1099     struct {
1100         u64 vpid : 16;
1101         u64 rsvd : 48;
1102         u64 gva;
1103     } operand = { vpid, 0, gva };
1104
1105     asm volatile (__ex(ASM_VMX_INVVPID)
1106                   /* CF==1 or ZF==1 --> rc = -1 */
1107                   "; ja 1f ; ud2 ; 1:"
1108                   : : "a"(&operand), "c"(ext) : "cc", "memory");
1109 }
1110
1111 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1112 {
1113         struct {
1114                 u64 eptp, gpa;
1115         } operand = {eptp, gpa};
1116
1117         asm volatile (__ex(ASM_VMX_INVEPT)
1118                         /* CF==1 or ZF==1 --> rc = -1 */
1119                         "; ja 1f ; ud2 ; 1:\n"
1120                         : : "a" (&operand), "c" (ext) : "cc", "memory");
1121 }
1122
1123 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1124 {
1125         int i;
1126
1127         i = __find_msr_index(vmx, msr);
1128         if (i >= 0)
1129                 return &vmx->guest_msrs[i];
1130         return NULL;
1131 }
1132
1133 static void vmcs_clear(struct vmcs *vmcs)
1134 {
1135         u64 phys_addr = __pa(vmcs);
1136         u8 error;
1137
1138         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1139                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1140                       : "cc", "memory");
1141         if (error)
1142                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1143                        vmcs, phys_addr);
1144 }
1145
1146 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1147 {
1148         vmcs_clear(loaded_vmcs->vmcs);
1149         loaded_vmcs->cpu = -1;
1150         loaded_vmcs->launched = 0;
1151 }
1152
1153 static void vmcs_load(struct vmcs *vmcs)
1154 {
1155         u64 phys_addr = __pa(vmcs);
1156         u8 error;
1157
1158         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1159                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1160                         : "cc", "memory");
1161         if (error)
1162                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1163                        vmcs, phys_addr);
1164 }
1165
1166 #ifdef CONFIG_KEXEC
1167 /*
1168  * This bitmap is used to indicate whether the vmclear
1169  * operation is enabled on all cpus. All disabled by
1170  * default.
1171  */
1172 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1173
1174 static inline void crash_enable_local_vmclear(int cpu)
1175 {
1176         cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1177 }
1178
1179 static inline void crash_disable_local_vmclear(int cpu)
1180 {
1181         cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1182 }
1183
1184 static inline int crash_local_vmclear_enabled(int cpu)
1185 {
1186         return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1187 }
1188
1189 static void crash_vmclear_local_loaded_vmcss(void)
1190 {
1191         int cpu = raw_smp_processor_id();
1192         struct loaded_vmcs *v;
1193
1194         if (!crash_local_vmclear_enabled(cpu))
1195                 return;
1196
1197         list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1198                             loaded_vmcss_on_cpu_link)
1199                 vmcs_clear(v->vmcs);
1200 }
1201 #else
1202 static inline void crash_enable_local_vmclear(int cpu) { }
1203 static inline void crash_disable_local_vmclear(int cpu) { }
1204 #endif /* CONFIG_KEXEC */
1205
1206 static void __loaded_vmcs_clear(void *arg)
1207 {
1208         struct loaded_vmcs *loaded_vmcs = arg;
1209         int cpu = raw_smp_processor_id();
1210
1211         if (loaded_vmcs->cpu != cpu)
1212                 return; /* vcpu migration can race with cpu offline */
1213         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1214                 per_cpu(current_vmcs, cpu) = NULL;
1215         crash_disable_local_vmclear(cpu);
1216         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1217
1218         /*
1219          * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1220          * is before setting loaded_vmcs->vcpu to -1 which is done in
1221          * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1222          * then adds the vmcs into percpu list before it is deleted.
1223          */
1224         smp_wmb();
1225
1226         loaded_vmcs_init(loaded_vmcs);
1227         crash_enable_local_vmclear(cpu);
1228 }
1229
1230 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1231 {
1232         int cpu = loaded_vmcs->cpu;
1233
1234         if (cpu != -1)
1235                 smp_call_function_single(cpu,
1236                          __loaded_vmcs_clear, loaded_vmcs, 1);
1237 }
1238
1239 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1240 {
1241         if (vmx->vpid == 0)
1242                 return;
1243
1244         if (cpu_has_vmx_invvpid_single())
1245                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1246 }
1247
1248 static inline void vpid_sync_vcpu_global(void)
1249 {
1250         if (cpu_has_vmx_invvpid_global())
1251                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1252 }
1253
1254 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1255 {
1256         if (cpu_has_vmx_invvpid_single())
1257                 vpid_sync_vcpu_single(vmx);
1258         else
1259                 vpid_sync_vcpu_global();
1260 }
1261
1262 static inline void ept_sync_global(void)
1263 {
1264         if (cpu_has_vmx_invept_global())
1265                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1266 }
1267
1268 static inline void ept_sync_context(u64 eptp)
1269 {
1270         if (enable_ept) {
1271                 if (cpu_has_vmx_invept_context())
1272                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1273                 else
1274                         ept_sync_global();
1275         }
1276 }
1277
1278 static __always_inline unsigned long vmcs_readl(unsigned long field)
1279 {
1280         unsigned long value;
1281
1282         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1283                       : "=a"(value) : "d"(field) : "cc");
1284         return value;
1285 }
1286
1287 static __always_inline u16 vmcs_read16(unsigned long field)
1288 {
1289         return vmcs_readl(field);
1290 }
1291
1292 static __always_inline u32 vmcs_read32(unsigned long field)
1293 {
1294         return vmcs_readl(field);
1295 }
1296
1297 static __always_inline u64 vmcs_read64(unsigned long field)
1298 {
1299 #ifdef CONFIG_X86_64
1300         return vmcs_readl(field);
1301 #else
1302         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1303 #endif
1304 }
1305
1306 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1307 {
1308         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1309                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1310         dump_stack();
1311 }
1312
1313 static void vmcs_writel(unsigned long field, unsigned long value)
1314 {
1315         u8 error;
1316
1317         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1318                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1319         if (unlikely(error))
1320                 vmwrite_error(field, value);
1321 }
1322
1323 static void vmcs_write16(unsigned long field, u16 value)
1324 {
1325         vmcs_writel(field, value);
1326 }
1327
1328 static void vmcs_write32(unsigned long field, u32 value)
1329 {
1330         vmcs_writel(field, value);
1331 }
1332
1333 static void vmcs_write64(unsigned long field, u64 value)
1334 {
1335         vmcs_writel(field, value);
1336 #ifndef CONFIG_X86_64
1337         asm volatile ("");
1338         vmcs_writel(field+1, value >> 32);
1339 #endif
1340 }
1341
1342 static void vmcs_clear_bits(unsigned long field, u32 mask)
1343 {
1344         vmcs_writel(field, vmcs_readl(field) & ~mask);
1345 }
1346
1347 static void vmcs_set_bits(unsigned long field, u32 mask)
1348 {
1349         vmcs_writel(field, vmcs_readl(field) | mask);
1350 }
1351
1352 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1353 {
1354         vmcs_write32(VM_ENTRY_CONTROLS, val);
1355         vmx->vm_entry_controls_shadow = val;
1356 }
1357
1358 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1359 {
1360         if (vmx->vm_entry_controls_shadow != val)
1361                 vm_entry_controls_init(vmx, val);
1362 }
1363
1364 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1365 {
1366         return vmx->vm_entry_controls_shadow;
1367 }
1368
1369
1370 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1371 {
1372         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1373 }
1374
1375 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1376 {
1377         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1378 }
1379
1380 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1381 {
1382         vmcs_write32(VM_EXIT_CONTROLS, val);
1383         vmx->vm_exit_controls_shadow = val;
1384 }
1385
1386 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1387 {
1388         if (vmx->vm_exit_controls_shadow != val)
1389                 vm_exit_controls_init(vmx, val);
1390 }
1391
1392 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1393 {
1394         return vmx->vm_exit_controls_shadow;
1395 }
1396
1397
1398 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1399 {
1400         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1401 }
1402
1403 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1404 {
1405         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1406 }
1407
1408 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1409 {
1410         vmx->segment_cache.bitmask = 0;
1411 }
1412
1413 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1414                                        unsigned field)
1415 {
1416         bool ret;
1417         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1418
1419         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1420                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1421                 vmx->segment_cache.bitmask = 0;
1422         }
1423         ret = vmx->segment_cache.bitmask & mask;
1424         vmx->segment_cache.bitmask |= mask;
1425         return ret;
1426 }
1427
1428 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1429 {
1430         u16 *p = &vmx->segment_cache.seg[seg].selector;
1431
1432         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1433                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1434         return *p;
1435 }
1436
1437 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1438 {
1439         ulong *p = &vmx->segment_cache.seg[seg].base;
1440
1441         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1442                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1443         return *p;
1444 }
1445
1446 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1447 {
1448         u32 *p = &vmx->segment_cache.seg[seg].limit;
1449
1450         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1451                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1452         return *p;
1453 }
1454
1455 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1456 {
1457         u32 *p = &vmx->segment_cache.seg[seg].ar;
1458
1459         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1460                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1461         return *p;
1462 }
1463
1464 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1465 {
1466         u32 eb;
1467
1468         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1469              (1u << NM_VECTOR) | (1u << DB_VECTOR);
1470         if ((vcpu->guest_debug &
1471              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1472             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1473                 eb |= 1u << BP_VECTOR;
1474         if (to_vmx(vcpu)->rmode.vm86_active)
1475                 eb = ~0;
1476         if (enable_ept)
1477                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1478         if (vcpu->fpu_active)
1479                 eb &= ~(1u << NM_VECTOR);
1480
1481         /* When we are running a nested L2 guest and L1 specified for it a
1482          * certain exception bitmap, we must trap the same exceptions and pass
1483          * them to L1. When running L2, we will only handle the exceptions
1484          * specified above if L1 did not want them.
1485          */
1486         if (is_guest_mode(vcpu))
1487                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1488
1489         vmcs_write32(EXCEPTION_BITMAP, eb);
1490 }
1491
1492 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1493                 unsigned long entry, unsigned long exit)
1494 {
1495         vm_entry_controls_clearbit(vmx, entry);
1496         vm_exit_controls_clearbit(vmx, exit);
1497 }
1498
1499 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1500 {
1501         unsigned i;
1502         struct msr_autoload *m = &vmx->msr_autoload;
1503
1504         switch (msr) {
1505         case MSR_EFER:
1506                 if (cpu_has_load_ia32_efer) {
1507                         clear_atomic_switch_msr_special(vmx,
1508                                         VM_ENTRY_LOAD_IA32_EFER,
1509                                         VM_EXIT_LOAD_IA32_EFER);
1510                         return;
1511                 }
1512                 break;
1513         case MSR_CORE_PERF_GLOBAL_CTRL:
1514                 if (cpu_has_load_perf_global_ctrl) {
1515                         clear_atomic_switch_msr_special(vmx,
1516                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1517                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1518                         return;
1519                 }
1520                 break;
1521         }
1522
1523         for (i = 0; i < m->nr; ++i)
1524                 if (m->guest[i].index == msr)
1525                         break;
1526
1527         if (i == m->nr)
1528                 return;
1529         --m->nr;
1530         m->guest[i] = m->guest[m->nr];
1531         m->host[i] = m->host[m->nr];
1532         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1533         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1534 }
1535
1536 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1537                 unsigned long entry, unsigned long exit,
1538                 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1539                 u64 guest_val, u64 host_val)
1540 {
1541         vmcs_write64(guest_val_vmcs, guest_val);
1542         vmcs_write64(host_val_vmcs, host_val);
1543         vm_entry_controls_setbit(vmx, entry);
1544         vm_exit_controls_setbit(vmx, exit);
1545 }
1546
1547 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1548                                   u64 guest_val, u64 host_val)
1549 {
1550         unsigned i;
1551         struct msr_autoload *m = &vmx->msr_autoload;
1552
1553         switch (msr) {
1554         case MSR_EFER:
1555                 if (cpu_has_load_ia32_efer) {
1556                         add_atomic_switch_msr_special(vmx,
1557                                         VM_ENTRY_LOAD_IA32_EFER,
1558                                         VM_EXIT_LOAD_IA32_EFER,
1559                                         GUEST_IA32_EFER,
1560                                         HOST_IA32_EFER,
1561                                         guest_val, host_val);
1562                         return;
1563                 }
1564                 break;
1565         case MSR_CORE_PERF_GLOBAL_CTRL:
1566                 if (cpu_has_load_perf_global_ctrl) {
1567                         add_atomic_switch_msr_special(vmx,
1568                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1569                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1570                                         GUEST_IA32_PERF_GLOBAL_CTRL,
1571                                         HOST_IA32_PERF_GLOBAL_CTRL,
1572                                         guest_val, host_val);
1573                         return;
1574                 }
1575                 break;
1576         }
1577
1578         for (i = 0; i < m->nr; ++i)
1579                 if (m->guest[i].index == msr)
1580                         break;
1581
1582         if (i == NR_AUTOLOAD_MSRS) {
1583                 printk_once(KERN_WARNING "Not enough msr switch entries. "
1584                                 "Can't add msr %x\n", msr);
1585                 return;
1586         } else if (i == m->nr) {
1587                 ++m->nr;
1588                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1589                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1590         }
1591
1592         m->guest[i].index = msr;
1593         m->guest[i].value = guest_val;
1594         m->host[i].index = msr;
1595         m->host[i].value = host_val;
1596 }
1597
1598 static void reload_tss(void)
1599 {
1600         /*
1601          * VT restores TR but not its size.  Useless.
1602          */
1603         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1604         struct desc_struct *descs;
1605
1606         descs = (void *)gdt->address;
1607         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1608         load_TR_desc();
1609 }
1610
1611 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1612 {
1613         u64 guest_efer;
1614         u64 ignore_bits;
1615
1616         guest_efer = vmx->vcpu.arch.efer;
1617
1618         /*
1619          * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1620          * outside long mode
1621          */
1622         ignore_bits = EFER_NX | EFER_SCE;
1623 #ifdef CONFIG_X86_64
1624         ignore_bits |= EFER_LMA | EFER_LME;
1625         /* SCE is meaningful only in long mode on Intel */
1626         if (guest_efer & EFER_LMA)
1627                 ignore_bits &= ~(u64)EFER_SCE;
1628 #endif
1629         guest_efer &= ~ignore_bits;
1630         guest_efer |= host_efer & ignore_bits;
1631         vmx->guest_msrs[efer_offset].data = guest_efer;
1632         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1633
1634         clear_atomic_switch_msr(vmx, MSR_EFER);
1635         /* On ept, can't emulate nx, and must switch nx atomically */
1636         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1637                 guest_efer = vmx->vcpu.arch.efer;
1638                 if (!(guest_efer & EFER_LMA))
1639                         guest_efer &= ~EFER_LME;
1640                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1641                 return false;
1642         }
1643
1644         return true;
1645 }
1646
1647 static unsigned long segment_base(u16 selector)
1648 {
1649         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1650         struct desc_struct *d;
1651         unsigned long table_base;
1652         unsigned long v;
1653
1654         if (!(selector & ~3))
1655                 return 0;
1656
1657         table_base = gdt->address;
1658
1659         if (selector & 4) {           /* from ldt */
1660                 u16 ldt_selector = kvm_read_ldt();
1661
1662                 if (!(ldt_selector & ~3))
1663                         return 0;
1664
1665                 table_base = segment_base(ldt_selector);
1666         }
1667         d = (struct desc_struct *)(table_base + (selector & ~7));
1668         v = get_desc_base(d);
1669 #ifdef CONFIG_X86_64
1670        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1671                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1672 #endif
1673         return v;
1674 }
1675
1676 static inline unsigned long kvm_read_tr_base(void)
1677 {
1678         u16 tr;
1679         asm("str %0" : "=g"(tr));
1680         return segment_base(tr);
1681 }
1682
1683 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1684 {
1685         struct vcpu_vmx *vmx = to_vmx(vcpu);
1686         int i;
1687
1688         if (vmx->host_state.loaded)
1689                 return;
1690
1691         vmx->host_state.loaded = 1;
1692         /*
1693          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1694          * allow segment selectors with cpl > 0 or ti == 1.
1695          */
1696         vmx->host_state.ldt_sel = kvm_read_ldt();
1697         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1698         savesegment(fs, vmx->host_state.fs_sel);
1699         if (!(vmx->host_state.fs_sel & 7)) {
1700                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1701                 vmx->host_state.fs_reload_needed = 0;
1702         } else {
1703                 vmcs_write16(HOST_FS_SELECTOR, 0);
1704                 vmx->host_state.fs_reload_needed = 1;
1705         }
1706         savesegment(gs, vmx->host_state.gs_sel);
1707         if (!(vmx->host_state.gs_sel & 7))
1708                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1709         else {
1710                 vmcs_write16(HOST_GS_SELECTOR, 0);
1711                 vmx->host_state.gs_ldt_reload_needed = 1;
1712         }
1713
1714 #ifdef CONFIG_X86_64
1715         savesegment(ds, vmx->host_state.ds_sel);
1716         savesegment(es, vmx->host_state.es_sel);
1717 #endif
1718
1719 #ifdef CONFIG_X86_64
1720         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1721         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1722 #else
1723         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1724         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1725 #endif
1726
1727 #ifdef CONFIG_X86_64
1728         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1729         if (is_long_mode(&vmx->vcpu))
1730                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1731 #endif
1732         if (boot_cpu_has(X86_FEATURE_MPX))
1733                 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1734         for (i = 0; i < vmx->save_nmsrs; ++i)
1735                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1736                                    vmx->guest_msrs[i].data,
1737                                    vmx->guest_msrs[i].mask);
1738 }
1739
1740 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1741 {
1742         if (!vmx->host_state.loaded)
1743                 return;
1744
1745         ++vmx->vcpu.stat.host_state_reload;
1746         vmx->host_state.loaded = 0;
1747 #ifdef CONFIG_X86_64
1748         if (is_long_mode(&vmx->vcpu))
1749                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1750 #endif
1751         if (vmx->host_state.gs_ldt_reload_needed) {
1752                 kvm_load_ldt(vmx->host_state.ldt_sel);
1753 #ifdef CONFIG_X86_64
1754                 load_gs_index(vmx->host_state.gs_sel);
1755 #else
1756                 loadsegment(gs, vmx->host_state.gs_sel);
1757 #endif
1758         }
1759         if (vmx->host_state.fs_reload_needed)
1760                 loadsegment(fs, vmx->host_state.fs_sel);
1761 #ifdef CONFIG_X86_64
1762         if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1763                 loadsegment(ds, vmx->host_state.ds_sel);
1764                 loadsegment(es, vmx->host_state.es_sel);
1765         }
1766 #endif
1767         reload_tss();
1768 #ifdef CONFIG_X86_64
1769         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1770 #endif
1771         if (vmx->host_state.msr_host_bndcfgs)
1772                 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1773         /*
1774          * If the FPU is not active (through the host task or
1775          * the guest vcpu), then restore the cr0.TS bit.
1776          */
1777         if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1778                 stts();
1779         load_gdt(&__get_cpu_var(host_gdt));
1780 }
1781
1782 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1783 {
1784         preempt_disable();
1785         __vmx_load_host_state(vmx);
1786         preempt_enable();
1787 }
1788
1789 /*
1790  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1791  * vcpu mutex is already taken.
1792  */
1793 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1794 {
1795         struct vcpu_vmx *vmx = to_vmx(vcpu);
1796         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1797
1798         if (!vmm_exclusive)
1799                 kvm_cpu_vmxon(phys_addr);
1800         else if (vmx->loaded_vmcs->cpu != cpu)
1801                 loaded_vmcs_clear(vmx->loaded_vmcs);
1802
1803         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1804                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1805                 vmcs_load(vmx->loaded_vmcs->vmcs);
1806         }
1807
1808         if (vmx->loaded_vmcs->cpu != cpu) {
1809                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1810                 unsigned long sysenter_esp;
1811
1812                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1813                 local_irq_disable();
1814                 crash_disable_local_vmclear(cpu);
1815
1816                 /*
1817                  * Read loaded_vmcs->cpu should be before fetching
1818                  * loaded_vmcs->loaded_vmcss_on_cpu_link.
1819                  * See the comments in __loaded_vmcs_clear().
1820                  */
1821                 smp_rmb();
1822
1823                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1824                          &per_cpu(loaded_vmcss_on_cpu, cpu));
1825                 crash_enable_local_vmclear(cpu);
1826                 local_irq_enable();
1827
1828                 /*
1829                  * Linux uses per-cpu TSS and GDT, so set these when switching
1830                  * processors.
1831                  */
1832                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1833                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
1834
1835                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1836                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1837                 vmx->loaded_vmcs->cpu = cpu;
1838         }
1839 }
1840
1841 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1842 {
1843         __vmx_load_host_state(to_vmx(vcpu));
1844         if (!vmm_exclusive) {
1845                 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1846                 vcpu->cpu = -1;
1847                 kvm_cpu_vmxoff();
1848         }
1849 }
1850
1851 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1852 {
1853         ulong cr0;
1854
1855         if (vcpu->fpu_active)
1856                 return;
1857         vcpu->fpu_active = 1;
1858         cr0 = vmcs_readl(GUEST_CR0);
1859         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1860         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1861         vmcs_writel(GUEST_CR0, cr0);
1862         update_exception_bitmap(vcpu);
1863         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1864         if (is_guest_mode(vcpu))
1865                 vcpu->arch.cr0_guest_owned_bits &=
1866                         ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1867         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1868 }
1869
1870 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1871
1872 /*
1873  * Return the cr0 value that a nested guest would read. This is a combination
1874  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1875  * its hypervisor (cr0_read_shadow).
1876  */
1877 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1878 {
1879         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1880                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1881 }
1882 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1883 {
1884         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1885                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1886 }
1887
1888 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1889 {
1890         /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1891          * set this *before* calling this function.
1892          */
1893         vmx_decache_cr0_guest_bits(vcpu);
1894         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1895         update_exception_bitmap(vcpu);
1896         vcpu->arch.cr0_guest_owned_bits = 0;
1897         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1898         if (is_guest_mode(vcpu)) {
1899                 /*
1900                  * L1's specified read shadow might not contain the TS bit,
1901                  * so now that we turned on shadowing of this bit, we need to
1902                  * set this bit of the shadow. Like in nested_vmx_run we need
1903                  * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1904                  * up-to-date here because we just decached cr0.TS (and we'll
1905                  * only update vmcs12->guest_cr0 on nested exit).
1906                  */
1907                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1908                 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1909                         (vcpu->arch.cr0 & X86_CR0_TS);
1910                 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1911         } else
1912                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1913 }
1914
1915 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1916 {
1917         unsigned long rflags, save_rflags;
1918
1919         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1920                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1921                 rflags = vmcs_readl(GUEST_RFLAGS);
1922                 if (to_vmx(vcpu)->rmode.vm86_active) {
1923                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1924                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1925                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1926                 }
1927                 to_vmx(vcpu)->rflags = rflags;
1928         }
1929         return to_vmx(vcpu)->rflags;
1930 }
1931
1932 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1933 {
1934         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1935         to_vmx(vcpu)->rflags = rflags;
1936         if (to_vmx(vcpu)->rmode.vm86_active) {
1937                 to_vmx(vcpu)->rmode.save_rflags = rflags;
1938                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1939         }
1940         vmcs_writel(GUEST_RFLAGS, rflags);
1941 }
1942
1943 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1944 {
1945         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1946         int ret = 0;
1947
1948         if (interruptibility & GUEST_INTR_STATE_STI)
1949                 ret |= KVM_X86_SHADOW_INT_STI;
1950         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1951                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1952
1953         return ret & mask;
1954 }
1955
1956 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1957 {
1958         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1959         u32 interruptibility = interruptibility_old;
1960
1961         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1962
1963         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1964                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1965         else if (mask & KVM_X86_SHADOW_INT_STI)
1966                 interruptibility |= GUEST_INTR_STATE_STI;
1967
1968         if ((interruptibility != interruptibility_old))
1969                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1970 }
1971
1972 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1973 {
1974         unsigned long rip;
1975
1976         rip = kvm_rip_read(vcpu);
1977         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1978         kvm_rip_write(vcpu, rip);
1979
1980         /* skipping an emulated instruction also counts */
1981         vmx_set_interrupt_shadow(vcpu, 0);
1982 }
1983
1984 /*
1985  * KVM wants to inject page-faults which it got to the guest. This function
1986  * checks whether in a nested guest, we need to inject them to L1 or L2.
1987  */
1988 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
1989 {
1990         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1991
1992         if (!(vmcs12->exception_bitmap & (1u << nr)))
1993                 return 0;
1994
1995         nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
1996                           vmcs_read32(VM_EXIT_INTR_INFO),
1997                           vmcs_readl(EXIT_QUALIFICATION));
1998         return 1;
1999 }
2000
2001 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2002                                 bool has_error_code, u32 error_code,
2003                                 bool reinject)
2004 {
2005         struct vcpu_vmx *vmx = to_vmx(vcpu);
2006         u32 intr_info = nr | INTR_INFO_VALID_MASK;
2007
2008         if (!reinject && is_guest_mode(vcpu) &&
2009             nested_vmx_check_exception(vcpu, nr))
2010                 return;
2011
2012         if (has_error_code) {
2013                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2014                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2015         }
2016
2017         if (vmx->rmode.vm86_active) {
2018                 int inc_eip = 0;
2019                 if (kvm_exception_is_soft(nr))
2020                         inc_eip = vcpu->arch.event_exit_inst_len;
2021                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2022                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2023                 return;
2024         }
2025
2026         if (kvm_exception_is_soft(nr)) {
2027                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2028                              vmx->vcpu.arch.event_exit_inst_len);
2029                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2030         } else
2031                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2032
2033         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2034 }
2035
2036 static bool vmx_rdtscp_supported(void)
2037 {
2038         return cpu_has_vmx_rdtscp();
2039 }
2040
2041 static bool vmx_invpcid_supported(void)
2042 {
2043         return cpu_has_vmx_invpcid() && enable_ept;
2044 }
2045
2046 /*
2047  * Swap MSR entry in host/guest MSR entry array.
2048  */
2049 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2050 {
2051         struct shared_msr_entry tmp;
2052
2053         tmp = vmx->guest_msrs[to];
2054         vmx->guest_msrs[to] = vmx->guest_msrs[from];
2055         vmx->guest_msrs[from] = tmp;
2056 }
2057
2058 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2059 {
2060         unsigned long *msr_bitmap;
2061
2062         if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
2063                 if (is_long_mode(vcpu))
2064                         msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2065                 else
2066                         msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2067         } else {
2068                 if (is_long_mode(vcpu))
2069                         msr_bitmap = vmx_msr_bitmap_longmode;
2070                 else
2071                         msr_bitmap = vmx_msr_bitmap_legacy;
2072         }
2073
2074         vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2075 }
2076
2077 /*
2078  * Set up the vmcs to automatically save and restore system
2079  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
2080  * mode, as fiddling with msrs is very expensive.
2081  */
2082 static void setup_msrs(struct vcpu_vmx *vmx)
2083 {
2084         int save_nmsrs, index;
2085
2086         save_nmsrs = 0;
2087 #ifdef CONFIG_X86_64
2088         if (is_long_mode(&vmx->vcpu)) {
2089                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2090                 if (index >= 0)
2091                         move_msr_up(vmx, index, save_nmsrs++);
2092                 index = __find_msr_index(vmx, MSR_LSTAR);
2093                 if (index >= 0)
2094                         move_msr_up(vmx, index, save_nmsrs++);
2095                 index = __find_msr_index(vmx, MSR_CSTAR);
2096                 if (index >= 0)
2097                         move_msr_up(vmx, index, save_nmsrs++);
2098                 index = __find_msr_index(vmx, MSR_TSC_AUX);
2099                 if (index >= 0 && vmx->rdtscp_enabled)
2100                         move_msr_up(vmx, index, save_nmsrs++);
2101                 /*
2102                  * MSR_STAR is only needed on long mode guests, and only
2103                  * if efer.sce is enabled.
2104                  */
2105                 index = __find_msr_index(vmx, MSR_STAR);
2106                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2107                         move_msr_up(vmx, index, save_nmsrs++);
2108         }
2109 #endif
2110         index = __find_msr_index(vmx, MSR_EFER);
2111         if (index >= 0 && update_transition_efer(vmx, index))
2112                 move_msr_up(vmx, index, save_nmsrs++);
2113
2114         vmx->save_nmsrs = save_nmsrs;
2115
2116         if (cpu_has_vmx_msr_bitmap())
2117                 vmx_set_msr_bitmap(&vmx->vcpu);
2118 }
2119
2120 /*
2121  * reads and returns guest's timestamp counter "register"
2122  * guest_tsc = host_tsc + tsc_offset    -- 21.3
2123  */
2124 static u64 guest_read_tsc(void)
2125 {
2126         u64 host_tsc, tsc_offset;
2127
2128         rdtscll(host_tsc);
2129         tsc_offset = vmcs_read64(TSC_OFFSET);
2130         return host_tsc + tsc_offset;
2131 }
2132
2133 /*
2134  * Like guest_read_tsc, but always returns L1's notion of the timestamp
2135  * counter, even if a nested guest (L2) is currently running.
2136  */
2137 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2138 {
2139         u64 tsc_offset;
2140
2141         tsc_offset = is_guest_mode(vcpu) ?
2142                 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2143                 vmcs_read64(TSC_OFFSET);
2144         return host_tsc + tsc_offset;
2145 }
2146
2147 /*
2148  * Engage any workarounds for mis-matched TSC rates.  Currently limited to
2149  * software catchup for faster rates on slower CPUs.
2150  */
2151 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2152 {
2153         if (!scale)
2154                 return;
2155
2156         if (user_tsc_khz > tsc_khz) {
2157                 vcpu->arch.tsc_catchup = 1;
2158                 vcpu->arch.tsc_always_catchup = 1;
2159         } else
2160                 WARN(1, "user requested TSC rate below hardware speed\n");
2161 }
2162
2163 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2164 {
2165         return vmcs_read64(TSC_OFFSET);
2166 }
2167
2168 /*
2169  * writes 'offset' into guest's timestamp counter offset register
2170  */
2171 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2172 {
2173         if (is_guest_mode(vcpu)) {
2174                 /*
2175                  * We're here if L1 chose not to trap WRMSR to TSC. According
2176                  * to the spec, this should set L1's TSC; The offset that L1
2177                  * set for L2 remains unchanged, and still needs to be added
2178                  * to the newly set TSC to get L2's TSC.
2179                  */
2180                 struct vmcs12 *vmcs12;
2181                 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2182                 /* recalculate vmcs02.TSC_OFFSET: */
2183                 vmcs12 = get_vmcs12(vcpu);
2184                 vmcs_write64(TSC_OFFSET, offset +
2185                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2186                          vmcs12->tsc_offset : 0));
2187         } else {
2188                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2189                                            vmcs_read64(TSC_OFFSET), offset);
2190                 vmcs_write64(TSC_OFFSET, offset);
2191         }
2192 }
2193
2194 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2195 {
2196         u64 offset = vmcs_read64(TSC_OFFSET);
2197
2198         vmcs_write64(TSC_OFFSET, offset + adjustment);
2199         if (is_guest_mode(vcpu)) {
2200                 /* Even when running L2, the adjustment needs to apply to L1 */
2201                 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2202         } else
2203                 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2204                                            offset + adjustment);
2205 }
2206
2207 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2208 {
2209         return target_tsc - native_read_tsc();
2210 }
2211
2212 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2213 {
2214         struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2215         return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2216 }
2217
2218 /*
2219  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2220  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2221  * all guests if the "nested" module option is off, and can also be disabled
2222  * for a single guest by disabling its VMX cpuid bit.
2223  */
2224 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2225 {
2226         return nested && guest_cpuid_has_vmx(vcpu);
2227 }
2228
2229 /*
2230  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2231  * returned for the various VMX controls MSRs when nested VMX is enabled.
2232  * The same values should also be used to verify that vmcs12 control fields are
2233  * valid during nested entry from L1 to L2.
2234  * Each of these control msrs has a low and high 32-bit half: A low bit is on
2235  * if the corresponding bit in the (32-bit) control field *must* be on, and a
2236  * bit in the high half is on if the corresponding bit in the control field
2237  * may be on. See also vmx_control_verify().
2238  * TODO: allow these variables to be modified (downgraded) by module options
2239  * or other means.
2240  */
2241 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
2242 static u32 nested_vmx_true_procbased_ctls_low;
2243 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
2244 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
2245 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
2246 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
2247 static u32 nested_vmx_misc_low, nested_vmx_misc_high;
2248 static u32 nested_vmx_ept_caps;
2249 static __init void nested_vmx_setup_ctls_msrs(void)
2250 {
2251         /*
2252          * Note that as a general rule, the high half of the MSRs (bits in
2253          * the control fields which may be 1) should be initialized by the
2254          * intersection of the underlying hardware's MSR (i.e., features which
2255          * can be supported) and the list of features we want to expose -
2256          * because they are known to be properly supported in our code.
2257          * Also, usually, the low half of the MSRs (bits which must be 1) can
2258          * be set to 0, meaning that L1 may turn off any of these bits. The
2259          * reason is that if one of these bits is necessary, it will appear
2260          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2261          * fields of vmcs01 and vmcs02, will turn these bits off - and
2262          * nested_vmx_exit_handled() will not pass related exits to L1.
2263          * These rules have exceptions below.
2264          */
2265
2266         /* pin-based controls */
2267         rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2268               nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high);
2269         nested_vmx_pinbased_ctls_low |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2270         nested_vmx_pinbased_ctls_high &= PIN_BASED_EXT_INTR_MASK |
2271                 PIN_BASED_NMI_EXITING | PIN_BASED_VIRTUAL_NMIS;
2272         nested_vmx_pinbased_ctls_high |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2273                 PIN_BASED_VMX_PREEMPTION_TIMER;
2274
2275         /* exit controls */
2276         rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2277                 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high);
2278         nested_vmx_exit_ctls_low = VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2279
2280         nested_vmx_exit_ctls_high &=
2281 #ifdef CONFIG_X86_64
2282                 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2283 #endif
2284                 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2285         nested_vmx_exit_ctls_high |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2286                 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2287                 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2288
2289         if (vmx_mpx_supported())
2290                 nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2291
2292         /* entry controls */
2293         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2294                 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
2295         nested_vmx_entry_ctls_low = VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2296         nested_vmx_entry_ctls_high &=
2297 #ifdef CONFIG_X86_64
2298                 VM_ENTRY_IA32E_MODE |
2299 #endif
2300                 VM_ENTRY_LOAD_IA32_PAT;
2301         nested_vmx_entry_ctls_high |= (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR |
2302                                        VM_ENTRY_LOAD_IA32_EFER);
2303         if (vmx_mpx_supported())
2304                 nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2305
2306         /* cpu-based controls */
2307         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2308                 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
2309         nested_vmx_procbased_ctls_low = 0;
2310         nested_vmx_procbased_ctls_high &=
2311                 CPU_BASED_VIRTUAL_INTR_PENDING |
2312                 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2313                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2314                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2315                 CPU_BASED_CR3_STORE_EXITING |
2316 #ifdef CONFIG_X86_64
2317                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2318 #endif
2319                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2320                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2321                 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2322                 CPU_BASED_PAUSE_EXITING |
2323                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2324         /*
2325          * We can allow some features even when not supported by the
2326          * hardware. For example, L1 can specify an MSR bitmap - and we
2327          * can use it to avoid exits to L1 - even when L0 runs L2
2328          * without MSR bitmaps.
2329          */
2330         nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
2331
2332         /* We support free control of CR3 access interception. */
2333         nested_vmx_true_procbased_ctls_low = nested_vmx_procbased_ctls_low &
2334                 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2335
2336         /* secondary cpu-based controls */
2337         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2338                 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2339         nested_vmx_secondary_ctls_low = 0;
2340         nested_vmx_secondary_ctls_high &=
2341                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2342                 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2343                 SECONDARY_EXEC_WBINVD_EXITING;
2344
2345         if (enable_ept) {
2346                 /* nested EPT: emulate EPT also to L1 */
2347                 nested_vmx_secondary_ctls_high |= SECONDARY_EXEC_ENABLE_EPT;
2348                 nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2349                          VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2350                          VMX_EPT_INVEPT_BIT;
2351                 nested_vmx_ept_caps &= vmx_capability.ept;
2352                 /*
2353                  * For nested guests, we don't do anything specific
2354                  * for single context invalidation. Hence, only advertise
2355                  * support for global context invalidation.
2356                  */
2357                 nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2358         } else
2359                 nested_vmx_ept_caps = 0;
2360
2361         /* miscellaneous data */
2362         rdmsr(MSR_IA32_VMX_MISC, nested_vmx_misc_low, nested_vmx_misc_high);
2363         nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2364         nested_vmx_misc_low |= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2365                 VMX_MISC_ACTIVITY_HLT;
2366         nested_vmx_misc_high = 0;
2367 }
2368
2369 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2370 {
2371         /*
2372          * Bits 0 in high must be 0, and bits 1 in low must be 1.
2373          */
2374         return ((control & high) | low) == control;
2375 }
2376
2377 static inline u64 vmx_control_msr(u32 low, u32 high)
2378 {
2379         return low | ((u64)high << 32);
2380 }
2381
2382 /* Returns 0 on success, non-0 otherwise. */
2383 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2384 {
2385         switch (msr_index) {
2386         case MSR_IA32_VMX_BASIC:
2387                 /*
2388                  * This MSR reports some information about VMX support. We
2389                  * should return information about the VMX we emulate for the
2390                  * guest, and the VMCS structure we give it - not about the
2391                  * VMX support of the underlying hardware.
2392                  */
2393                 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2394                            ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2395                            (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2396                 break;
2397         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2398         case MSR_IA32_VMX_PINBASED_CTLS:
2399                 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2400                                         nested_vmx_pinbased_ctls_high);
2401                 break;
2402         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2403                 *pdata = vmx_control_msr(nested_vmx_true_procbased_ctls_low,
2404                                         nested_vmx_procbased_ctls_high);
2405                 break;
2406         case MSR_IA32_VMX_PROCBASED_CTLS:
2407                 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2408                                         nested_vmx_procbased_ctls_high);
2409                 break;
2410         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2411         case MSR_IA32_VMX_EXIT_CTLS:
2412                 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2413                                         nested_vmx_exit_ctls_high);
2414                 break;
2415         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2416         case MSR_IA32_VMX_ENTRY_CTLS:
2417                 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2418                                         nested_vmx_entry_ctls_high);
2419                 break;
2420         case MSR_IA32_VMX_MISC:
2421                 *pdata = vmx_control_msr(nested_vmx_misc_low,
2422                                          nested_vmx_misc_high);
2423                 break;
2424         /*
2425          * These MSRs specify bits which the guest must keep fixed (on or off)
2426          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2427          * We picked the standard core2 setting.
2428          */
2429 #define VMXON_CR0_ALWAYSON      (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2430 #define VMXON_CR4_ALWAYSON      X86_CR4_VMXE
2431         case MSR_IA32_VMX_CR0_FIXED0:
2432                 *pdata = VMXON_CR0_ALWAYSON;
2433                 break;
2434         case MSR_IA32_VMX_CR0_FIXED1:
2435                 *pdata = -1ULL;
2436                 break;
2437         case MSR_IA32_VMX_CR4_FIXED0:
2438                 *pdata = VMXON_CR4_ALWAYSON;
2439                 break;
2440         case MSR_IA32_VMX_CR4_FIXED1:
2441                 *pdata = -1ULL;
2442                 break;
2443         case MSR_IA32_VMX_VMCS_ENUM:
2444                 *pdata = 0x1f;
2445                 break;
2446         case MSR_IA32_VMX_PROCBASED_CTLS2:
2447                 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2448                                         nested_vmx_secondary_ctls_high);
2449                 break;
2450         case MSR_IA32_VMX_EPT_VPID_CAP:
2451                 /* Currently, no nested vpid support */
2452                 *pdata = nested_vmx_ept_caps;
2453                 break;
2454         default:
2455                 return 1;
2456         }
2457
2458         return 0;
2459 }
2460
2461 /*
2462  * Reads an msr value (of 'msr_index') into 'pdata'.
2463  * Returns 0 on success, non-0 otherwise.
2464  * Assumes vcpu_load() was already called.
2465  */
2466 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2467 {
2468         u64 data;
2469         struct shared_msr_entry *msr;
2470
2471         if (!pdata) {
2472                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2473                 return -EINVAL;
2474         }
2475
2476         switch (msr_index) {
2477 #ifdef CONFIG_X86_64
2478         case MSR_FS_BASE:
2479                 data = vmcs_readl(GUEST_FS_BASE);
2480                 break;
2481         case MSR_GS_BASE:
2482                 data = vmcs_readl(GUEST_GS_BASE);
2483                 break;
2484         case MSR_KERNEL_GS_BASE:
2485                 vmx_load_host_state(to_vmx(vcpu));
2486                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2487                 break;
2488 #endif
2489         case MSR_EFER:
2490                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2491         case MSR_IA32_TSC:
2492                 data = guest_read_tsc();
2493                 break;
2494         case MSR_IA32_SYSENTER_CS:
2495                 data = vmcs_read32(GUEST_SYSENTER_CS);
2496                 break;
2497         case MSR_IA32_SYSENTER_EIP:
2498                 data = vmcs_readl(GUEST_SYSENTER_EIP);
2499                 break;
2500         case MSR_IA32_SYSENTER_ESP:
2501                 data = vmcs_readl(GUEST_SYSENTER_ESP);
2502                 break;
2503         case MSR_IA32_BNDCFGS:
2504                 if (!vmx_mpx_supported())
2505                         return 1;
2506                 data = vmcs_read64(GUEST_BNDCFGS);
2507                 break;
2508         case MSR_IA32_FEATURE_CONTROL:
2509                 if (!nested_vmx_allowed(vcpu))
2510                         return 1;
2511                 data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2512                 break;
2513         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2514                 if (!nested_vmx_allowed(vcpu))
2515                         return 1;
2516                 return vmx_get_vmx_msr(vcpu, msr_index, pdata);
2517         case MSR_TSC_AUX:
2518                 if (!to_vmx(vcpu)->rdtscp_enabled)
2519                         return 1;
2520                 /* Otherwise falls through */
2521         default:
2522                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2523                 if (msr) {
2524                         data = msr->data;
2525                         break;
2526                 }
2527                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2528         }
2529
2530         *pdata = data;
2531         return 0;
2532 }
2533
2534 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2535
2536 /*
2537  * Writes msr value into into the appropriate "register".
2538  * Returns 0 on success, non-0 otherwise.
2539  * Assumes vcpu_load() was already called.
2540  */
2541 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2542 {
2543         struct vcpu_vmx *vmx = to_vmx(vcpu);
2544         struct shared_msr_entry *msr;
2545         int ret = 0;
2546         u32 msr_index = msr_info->index;
2547         u64 data = msr_info->data;
2548
2549         switch (msr_index) {
2550         case MSR_EFER:
2551                 ret = kvm_set_msr_common(vcpu, msr_info);
2552                 break;
2553 #ifdef CONFIG_X86_64
2554         case MSR_FS_BASE:
2555                 vmx_segment_cache_clear(vmx);
2556                 vmcs_writel(GUEST_FS_BASE, data);
2557                 break;
2558         case MSR_GS_BASE:
2559                 vmx_segment_cache_clear(vmx);
2560                 vmcs_writel(GUEST_GS_BASE, data);
2561                 break;
2562         case MSR_KERNEL_GS_BASE:
2563                 vmx_load_host_state(vmx);
2564                 vmx->msr_guest_kernel_gs_base = data;
2565                 break;
2566 #endif
2567         case MSR_IA32_SYSENTER_CS:
2568                 vmcs_write32(GUEST_SYSENTER_CS, data);
2569                 break;
2570         case MSR_IA32_SYSENTER_EIP:
2571                 vmcs_writel(GUEST_SYSENTER_EIP, data);
2572                 break;
2573         case MSR_IA32_SYSENTER_ESP:
2574                 vmcs_writel(GUEST_SYSENTER_ESP, data);
2575                 break;
2576         case MSR_IA32_BNDCFGS:
2577                 if (!vmx_mpx_supported())
2578                         return 1;
2579                 vmcs_write64(GUEST_BNDCFGS, data);
2580                 break;
2581         case MSR_IA32_TSC:
2582                 kvm_write_tsc(vcpu, msr_info);
2583                 break;
2584         case MSR_IA32_CR_PAT:
2585                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2586                         vmcs_write64(GUEST_IA32_PAT, data);
2587                         vcpu->arch.pat = data;
2588                         break;
2589                 }
2590                 ret = kvm_set_msr_common(vcpu, msr_info);
2591                 break;
2592         case MSR_IA32_TSC_ADJUST:
2593                 ret = kvm_set_msr_common(vcpu, msr_info);
2594                 break;
2595         case MSR_IA32_FEATURE_CONTROL:
2596                 if (!nested_vmx_allowed(vcpu) ||
2597                     (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2598                      FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2599                         return 1;
2600                 vmx->nested.msr_ia32_feature_control = data;
2601                 if (msr_info->host_initiated && data == 0)
2602                         vmx_leave_nested(vcpu);
2603                 break;
2604         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2605                 return 1; /* they are read-only */
2606         case MSR_TSC_AUX:
2607                 if (!vmx->rdtscp_enabled)
2608                         return 1;
2609                 /* Check reserved bit, higher 32 bits should be zero */
2610                 if ((data >> 32) != 0)
2611                         return 1;
2612                 /* Otherwise falls through */
2613         default:
2614                 msr = find_msr_entry(vmx, msr_index);
2615                 if (msr) {
2616                         msr->data = data;
2617                         if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2618                                 preempt_disable();
2619                                 kvm_set_shared_msr(msr->index, msr->data,
2620                                                    msr->mask);
2621                                 preempt_enable();
2622                         }
2623                         break;
2624                 }
2625                 ret = kvm_set_msr_common(vcpu, msr_info);
2626         }
2627
2628         return ret;
2629 }
2630
2631 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2632 {
2633         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2634         switch (reg) {
2635         case VCPU_REGS_RSP:
2636                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2637                 break;
2638         case VCPU_REGS_RIP:
2639                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2640                 break;
2641         case VCPU_EXREG_PDPTR:
2642                 if (enable_ept)
2643                         ept_save_pdptrs(vcpu);
2644                 break;
2645         default:
2646                 break;
2647         }
2648 }
2649
2650 static __init int cpu_has_kvm_support(void)
2651 {
2652         return cpu_has_vmx();
2653 }
2654
2655 static __init int vmx_disabled_by_bios(void)
2656 {
2657         u64 msr;
2658
2659         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2660         if (msr & FEATURE_CONTROL_LOCKED) {
2661                 /* launched w/ TXT and VMX disabled */
2662                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2663                         && tboot_enabled())
2664                         return 1;
2665                 /* launched w/o TXT and VMX only enabled w/ TXT */
2666                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2667                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2668                         && !tboot_enabled()) {
2669                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2670                                 "activate TXT before enabling KVM\n");
2671                         return 1;
2672                 }
2673                 /* launched w/o TXT and VMX disabled */
2674                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2675                         && !tboot_enabled())
2676                         return 1;
2677         }
2678
2679         return 0;
2680 }
2681
2682 static void kvm_cpu_vmxon(u64 addr)
2683 {
2684         asm volatile (ASM_VMX_VMXON_RAX
2685                         : : "a"(&addr), "m"(addr)
2686                         : "memory", "cc");
2687 }
2688
2689 static int hardware_enable(void *garbage)
2690 {
2691         int cpu = raw_smp_processor_id();
2692         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2693         u64 old, test_bits;
2694
2695         if (read_cr4() & X86_CR4_VMXE)
2696                 return -EBUSY;
2697
2698         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2699
2700         /*
2701          * Now we can enable the vmclear operation in kdump
2702          * since the loaded_vmcss_on_cpu list on this cpu
2703          * has been initialized.
2704          *
2705          * Though the cpu is not in VMX operation now, there
2706          * is no problem to enable the vmclear operation
2707          * for the loaded_vmcss_on_cpu list is empty!
2708          */
2709         crash_enable_local_vmclear(cpu);
2710
2711         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2712
2713         test_bits = FEATURE_CONTROL_LOCKED;
2714         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2715         if (tboot_enabled())
2716                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2717
2718         if ((old & test_bits) != test_bits) {
2719                 /* enable and lock */
2720                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2721         }
2722         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2723
2724         if (vmm_exclusive) {
2725                 kvm_cpu_vmxon(phys_addr);
2726                 ept_sync_global();
2727         }
2728
2729         native_store_gdt(&__get_cpu_var(host_gdt));
2730
2731         return 0;
2732 }
2733
2734 static void vmclear_local_loaded_vmcss(void)
2735 {
2736         int cpu = raw_smp_processor_id();
2737         struct loaded_vmcs *v, *n;
2738
2739         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2740                                  loaded_vmcss_on_cpu_link)
2741                 __loaded_vmcs_clear(v);
2742 }
2743
2744
2745 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2746  * tricks.
2747  */
2748 static void kvm_cpu_vmxoff(void)
2749 {
2750         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2751 }
2752
2753 static void hardware_disable(void *garbage)
2754 {
2755         if (vmm_exclusive) {
2756                 vmclear_local_loaded_vmcss();
2757                 kvm_cpu_vmxoff();
2758         }
2759         write_cr4(read_cr4() & ~X86_CR4_VMXE);
2760 }
2761
2762 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2763                                       u32 msr, u32 *result)
2764 {
2765         u32 vmx_msr_low, vmx_msr_high;
2766         u32 ctl = ctl_min | ctl_opt;
2767
2768         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2769
2770         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2771         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2772
2773         /* Ensure minimum (required) set of control bits are supported. */
2774         if (ctl_min & ~ctl)
2775                 return -EIO;
2776
2777         *result = ctl;
2778         return 0;
2779 }
2780
2781 static __init bool allow_1_setting(u32 msr, u32 ctl)
2782 {
2783         u32 vmx_msr_low, vmx_msr_high;
2784
2785         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2786         return vmx_msr_high & ctl;
2787 }
2788
2789 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2790 {
2791         u32 vmx_msr_low, vmx_msr_high;
2792         u32 min, opt, min2, opt2;
2793         u32 _pin_based_exec_control = 0;
2794         u32 _cpu_based_exec_control = 0;
2795         u32 _cpu_based_2nd_exec_control = 0;
2796         u32 _vmexit_control = 0;
2797         u32 _vmentry_control = 0;
2798
2799         min = CPU_BASED_HLT_EXITING |
2800 #ifdef CONFIG_X86_64
2801               CPU_BASED_CR8_LOAD_EXITING |
2802               CPU_BASED_CR8_STORE_EXITING |
2803 #endif
2804               CPU_BASED_CR3_LOAD_EXITING |
2805               CPU_BASED_CR3_STORE_EXITING |
2806               CPU_BASED_USE_IO_BITMAPS |
2807               CPU_BASED_MOV_DR_EXITING |
2808               CPU_BASED_USE_TSC_OFFSETING |
2809               CPU_BASED_MWAIT_EXITING |
2810               CPU_BASED_MONITOR_EXITING |
2811               CPU_BASED_INVLPG_EXITING |
2812               CPU_BASED_RDPMC_EXITING;
2813
2814         opt = CPU_BASED_TPR_SHADOW |
2815               CPU_BASED_USE_MSR_BITMAPS |
2816               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2817         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2818                                 &_cpu_based_exec_control) < 0)
2819                 return -EIO;
2820 #ifdef CONFIG_X86_64
2821         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2822                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2823                                            ~CPU_BASED_CR8_STORE_EXITING;
2824 #endif
2825         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2826                 min2 = 0;
2827                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2828                         SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2829                         SECONDARY_EXEC_WBINVD_EXITING |
2830                         SECONDARY_EXEC_ENABLE_VPID |
2831                         SECONDARY_EXEC_ENABLE_EPT |
2832                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
2833                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2834                         SECONDARY_EXEC_RDTSCP |
2835                         SECONDARY_EXEC_ENABLE_INVPCID |
2836                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
2837                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2838                         SECONDARY_EXEC_SHADOW_VMCS;
2839                 if (adjust_vmx_controls(min2, opt2,
2840                                         MSR_IA32_VMX_PROCBASED_CTLS2,
2841                                         &_cpu_based_2nd_exec_control) < 0)
2842                         return -EIO;
2843         }
2844 #ifndef CONFIG_X86_64
2845         if (!(_cpu_based_2nd_exec_control &
2846                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2847                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2848 #endif
2849
2850         if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2851                 _cpu_based_2nd_exec_control &= ~(
2852                                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2853                                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2854                                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2855
2856         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2857                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2858                    enabled */
2859                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2860                                              CPU_BASED_CR3_STORE_EXITING |
2861                                              CPU_BASED_INVLPG_EXITING);
2862                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2863                       vmx_capability.ept, vmx_capability.vpid);
2864         }
2865
2866         min = VM_EXIT_SAVE_DEBUG_CONTROLS;
2867 #ifdef CONFIG_X86_64
2868         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2869 #endif
2870         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
2871                 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
2872         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2873                                 &_vmexit_control) < 0)
2874                 return -EIO;
2875
2876         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2877         opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
2878         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2879                                 &_pin_based_exec_control) < 0)
2880                 return -EIO;
2881
2882         if (!(_cpu_based_2nd_exec_control &
2883                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
2884                 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
2885                 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2886
2887         min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2888         opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
2889         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2890                                 &_vmentry_control) < 0)
2891                 return -EIO;
2892
2893         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2894
2895         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2896         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2897                 return -EIO;
2898
2899 #ifdef CONFIG_X86_64
2900         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2901         if (vmx_msr_high & (1u<<16))
2902                 return -EIO;
2903 #endif
2904
2905         /* Require Write-Back (WB) memory type for VMCS accesses. */
2906         if (((vmx_msr_high >> 18) & 15) != 6)
2907                 return -EIO;
2908
2909         vmcs_conf->size = vmx_msr_high & 0x1fff;
2910         vmcs_conf->order = get_order(vmcs_config.size);
2911         vmcs_conf->revision_id = vmx_msr_low;
2912
2913         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2914         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2915         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2916         vmcs_conf->vmexit_ctrl         = _vmexit_control;
2917         vmcs_conf->vmentry_ctrl        = _vmentry_control;
2918
2919         cpu_has_load_ia32_efer =
2920                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2921                                 VM_ENTRY_LOAD_IA32_EFER)
2922                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2923                                    VM_EXIT_LOAD_IA32_EFER);
2924
2925         cpu_has_load_perf_global_ctrl =
2926                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2927                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2928                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2929                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2930
2931         /*
2932          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2933          * but due to arrata below it can't be used. Workaround is to use
2934          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2935          *
2936          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2937          *
2938          * AAK155             (model 26)
2939          * AAP115             (model 30)
2940          * AAT100             (model 37)
2941          * BC86,AAY89,BD102   (model 44)
2942          * BA97               (model 46)
2943          *
2944          */
2945         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2946                 switch (boot_cpu_data.x86_model) {
2947                 case 26:
2948                 case 30:
2949                 case 37:
2950                 case 44:
2951                 case 46:
2952                         cpu_has_load_perf_global_ctrl = false;
2953                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2954                                         "does not work properly. Using workaround\n");
2955                         break;
2956                 default:
2957                         break;
2958                 }
2959         }
2960
2961         return 0;
2962 }
2963
2964 static struct vmcs *alloc_vmcs_cpu(int cpu)
2965 {
2966         int node = cpu_to_node(cpu);
2967         struct page *pages;
2968         struct vmcs *vmcs;
2969
2970         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2971         if (!pages)
2972                 return NULL;
2973         vmcs = page_address(pages);
2974         memset(vmcs, 0, vmcs_config.size);
2975         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2976         return vmcs;
2977 }
2978
2979 static struct vmcs *alloc_vmcs(void)
2980 {
2981         return alloc_vmcs_cpu(raw_smp_processor_id());
2982 }
2983
2984 static void free_vmcs(struct vmcs *vmcs)
2985 {
2986         free_pages((unsigned long)vmcs, vmcs_config.order);
2987 }
2988
2989 /*
2990  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2991  */
2992 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2993 {
2994         if (!loaded_vmcs->vmcs)
2995                 return;
2996         loaded_vmcs_clear(loaded_vmcs);
2997         free_vmcs(loaded_vmcs->vmcs);
2998         loaded_vmcs->vmcs = NULL;
2999 }
3000
3001 static void free_kvm_area(void)
3002 {
3003         int cpu;
3004
3005         for_each_possible_cpu(cpu) {
3006                 free_vmcs(per_cpu(vmxarea, cpu));
3007                 per_cpu(vmxarea, cpu) = NULL;
3008         }
3009 }
3010
3011 static void init_vmcs_shadow_fields(void)
3012 {
3013         int i, j;
3014
3015         /* No checks for read only fields yet */
3016
3017         for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3018                 switch (shadow_read_write_fields[i]) {
3019                 case GUEST_BNDCFGS:
3020                         if (!vmx_mpx_supported())
3021                                 continue;
3022                         break;
3023                 default:
3024                         break;
3025                 }
3026
3027                 if (j < i)
3028                         shadow_read_write_fields[j] =
3029                                 shadow_read_write_fields[i];
3030                 j++;
3031         }
3032         max_shadow_read_write_fields = j;
3033
3034         /* shadowed fields guest access without vmexit */
3035         for (i = 0; i < max_shadow_read_write_fields; i++) {
3036                 clear_bit(shadow_read_write_fields[i],
3037                           vmx_vmwrite_bitmap);
3038                 clear_bit(shadow_read_write_fields[i],
3039                           vmx_vmread_bitmap);
3040         }
3041         for (i = 0; i < max_shadow_read_only_fields; i++)
3042                 clear_bit(shadow_read_only_fields[i],
3043                           vmx_vmread_bitmap);
3044 }
3045
3046 static __init int alloc_kvm_area(void)
3047 {
3048         int cpu;
3049
3050         for_each_possible_cpu(cpu) {
3051                 struct vmcs *vmcs;
3052
3053                 vmcs = alloc_vmcs_cpu(cpu);
3054                 if (!vmcs) {
3055                         free_kvm_area();
3056                         return -ENOMEM;
3057                 }
3058
3059                 per_cpu(vmxarea, cpu) = vmcs;
3060         }
3061         return 0;
3062 }
3063
3064 static __init int hardware_setup(void)
3065 {
3066         if (setup_vmcs_config(&vmcs_config) < 0)
3067                 return -EIO;
3068
3069         if (boot_cpu_has(X86_FEATURE_NX))
3070                 kvm_enable_efer_bits(EFER_NX);
3071
3072         if (!cpu_has_vmx_vpid())
3073                 enable_vpid = 0;
3074         if (!cpu_has_vmx_shadow_vmcs())
3075                 enable_shadow_vmcs = 0;
3076         if (enable_shadow_vmcs)
3077                 init_vmcs_shadow_fields();
3078
3079         if (!cpu_has_vmx_ept() ||
3080             !cpu_has_vmx_ept_4levels()) {
3081                 enable_ept = 0;
3082                 enable_unrestricted_guest = 0;
3083                 enable_ept_ad_bits = 0;
3084         }
3085
3086         if (!cpu_has_vmx_ept_ad_bits())
3087                 enable_ept_ad_bits = 0;
3088
3089         if (!cpu_has_vmx_unrestricted_guest())
3090                 enable_unrestricted_guest = 0;
3091
3092         if (!cpu_has_vmx_flexpriority())
3093                 flexpriority_enabled = 0;
3094
3095         if (!cpu_has_vmx_tpr_shadow())
3096                 kvm_x86_ops->update_cr8_intercept = NULL;
3097
3098         if (enable_ept && !cpu_has_vmx_ept_2m_page())
3099                 kvm_disable_largepages();
3100
3101         if (!cpu_has_vmx_ple())
3102                 ple_gap = 0;
3103
3104         if (!cpu_has_vmx_apicv())
3105                 enable_apicv = 0;
3106
3107         if (enable_apicv)
3108                 kvm_x86_ops->update_cr8_intercept = NULL;
3109         else {
3110                 kvm_x86_ops->hwapic_irr_update = NULL;
3111                 kvm_x86_ops->deliver_posted_interrupt = NULL;
3112                 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
3113         }
3114
3115         if (nested)
3116                 nested_vmx_setup_ctls_msrs();
3117
3118         return alloc_kvm_area();
3119 }
3120
3121 static __exit void hardware_unsetup(void)
3122 {
3123         free_kvm_area();
3124 }
3125
3126 static bool emulation_required(struct kvm_vcpu *vcpu)
3127 {
3128         return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3129 }
3130
3131 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3132                 struct kvm_segment *save)
3133 {
3134         if (!emulate_invalid_guest_state) {
3135                 /*
3136                  * CS and SS RPL should be equal during guest entry according
3137                  * to VMX spec, but in reality it is not always so. Since vcpu
3138                  * is in the middle of the transition from real mode to
3139                  * protected mode it is safe to assume that RPL 0 is a good
3140                  * default value.
3141                  */
3142                 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3143                         save->selector &= ~SELECTOR_RPL_MASK;
3144                 save->dpl = save->selector & SELECTOR_RPL_MASK;
3145                 save->s = 1;
3146         }
3147         vmx_set_segment(vcpu, save, seg);
3148 }
3149
3150 static void enter_pmode(struct kvm_vcpu *vcpu)
3151 {
3152         unsigned long flags;
3153         struct vcpu_vmx *vmx = to_vmx(vcpu);
3154
3155         /*
3156          * Update real mode segment cache. It may be not up-to-date if sement
3157          * register was written while vcpu was in a guest mode.
3158          */
3159         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3160         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3161         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3162         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3163         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3164         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3165
3166         vmx->rmode.vm86_active = 0;
3167
3168         vmx_segment_cache_clear(vmx);
3169
3170         vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3171
3172         flags = vmcs_readl(GUEST_RFLAGS);
3173         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3174         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3175         vmcs_writel(GUEST_RFLAGS, flags);
3176
3177         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3178                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3179
3180         update_exception_bitmap(vcpu);
3181
3182         fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3183         fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3184         fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3185         fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3186         fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3187         fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3188 }
3189
3190 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3191 {
3192         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3193         struct kvm_segment var = *save;
3194
3195         var.dpl = 0x3;
3196         if (seg == VCPU_SREG_CS)
3197                 var.type = 0x3;
3198
3199         if (!emulate_invalid_guest_state) {
3200                 var.selector = var.base >> 4;
3201                 var.base = var.base & 0xffff0;
3202                 var.limit = 0xffff;
3203                 var.g = 0;
3204                 var.db = 0;
3205                 var.present = 1;
3206                 var.s = 1;
3207                 var.l = 0;
3208                 var.unusable = 0;
3209                 var.type = 0x3;
3210                 var.avl = 0;
3211                 if (save->base & 0xf)
3212                         printk_once(KERN_WARNING "kvm: segment base is not "
3213                                         "paragraph aligned when entering "
3214                                         "protected mode (seg=%d)", seg);
3215         }
3216
3217         vmcs_write16(sf->selector, var.selector);
3218         vmcs_write32(sf->base, var.base);
3219         vmcs_write32(sf->limit, var.limit);
3220         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3221 }
3222
3223 static void enter_rmode(struct kvm_vcpu *vcpu)
3224 {
3225         unsigned long flags;
3226         struct vcpu_vmx *vmx = to_vmx(vcpu);
3227
3228         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3229         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3230         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3231         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3232         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3233         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3234         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3235
3236         vmx->rmode.vm86_active = 1;
3237
3238         /*
3239          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3240          * vcpu. Warn the user that an update is overdue.
3241          */
3242         if (!vcpu->kvm->arch.tss_addr)
3243                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3244                              "called before entering vcpu\n");
3245
3246         vmx_segment_cache_clear(vmx);
3247
3248         vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3249         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3250         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3251
3252         flags = vmcs_readl(GUEST_RFLAGS);
3253         vmx->rmode.save_rflags = flags;
3254
3255         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3256
3257         vmcs_writel(GUEST_RFLAGS, flags);
3258         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3259         update_exception_bitmap(vcpu);
3260
3261         fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3262         fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3263         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3264         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3265         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3266         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3267
3268         kvm_mmu_reset_context(vcpu);
3269 }
3270
3271 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3272 {
3273         struct vcpu_vmx *vmx = to_vmx(vcpu);
3274         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3275
3276         if (!msr)
3277                 return;
3278
3279         /*
3280          * Force kernel_gs_base reloading before EFER changes, as control
3281          * of this msr depends on is_long_mode().
3282          */
3283         vmx_load_host_state(to_vmx(vcpu));
3284         vcpu->arch.efer = efer;
3285         if (efer & EFER_LMA) {
3286                 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3287                 msr->data = efer;
3288         } else {
3289                 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3290
3291                 msr->data = efer & ~EFER_LME;
3292         }
3293         setup_msrs(vmx);
3294 }
3295
3296 #ifdef CONFIG_X86_64
3297
3298 static void enter_lmode(struct kvm_vcpu *vcpu)
3299 {
3300         u32 guest_tr_ar;
3301
3302         vmx_segment_cache_clear(to_vmx(vcpu));
3303
3304         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3305         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3306                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3307                                      __func__);
3308                 vmcs_write32(GUEST_TR_AR_BYTES,
3309                              (guest_tr_ar & ~AR_TYPE_MASK)
3310                              | AR_TYPE_BUSY_64_TSS);
3311         }
3312         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3313 }
3314
3315 static void exit_lmode(struct kvm_vcpu *vcpu)
3316 {
3317         vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3318         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3319 }
3320
3321 #endif
3322
3323 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3324 {
3325         vpid_sync_context(to_vmx(vcpu));
3326         if (enable_ept) {
3327                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3328                         return;
3329                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3330         }
3331 }
3332
3333 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3334 {
3335         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3336
3337         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3338         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3339 }
3340
3341 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3342 {
3343         if (enable_ept && is_paging(vcpu))
3344                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3345         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3346 }
3347
3348 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3349 {
3350         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3351
3352         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3353         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3354 }
3355
3356 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3357 {
3358         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3359
3360         if (!test_bit(VCPU_EXREG_PDPTR,
3361                       (unsigned long *)&vcpu->arch.regs_dirty))
3362                 return;
3363
3364         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3365                 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3366                 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3367                 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3368                 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3369         }
3370 }
3371
3372 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3373 {
3374         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3375
3376         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3377                 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3378                 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3379                 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3380                 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3381         }
3382
3383         __set_bit(VCPU_EXREG_PDPTR,
3384                   (unsigned long *)&vcpu->arch.regs_avail);
3385         __set_bit(VCPU_EXREG_PDPTR,
3386                   (unsigned long *)&vcpu->arch.regs_dirty);
3387 }
3388
3389 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3390
3391 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3392                                         unsigned long cr0,
3393                                         struct kvm_vcpu *vcpu)
3394 {
3395         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3396                 vmx_decache_cr3(vcpu);
3397         if (!(cr0 & X86_CR0_PG)) {
3398                 /* From paging/starting to nonpaging */
3399                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3400                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3401                              (CPU_BASED_CR3_LOAD_EXITING |
3402                               CPU_BASED_CR3_STORE_EXITING));
3403                 vcpu->arch.cr0 = cr0;
3404                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3405         } else if (!is_paging(vcpu)) {
3406                 /* From nonpaging to paging */
3407                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3408                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3409                              ~(CPU_BASED_CR3_LOAD_EXITING |
3410                                CPU_BASED_CR3_STORE_EXITING));
3411                 vcpu->arch.cr0 = cr0;
3412                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3413         }
3414
3415         if (!(cr0 & X86_CR0_WP))
3416                 *hw_cr0 &= ~X86_CR0_WP;
3417 }
3418
3419 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3420 {
3421         struct vcpu_vmx *vmx = to_vmx(vcpu);
3422         unsigned long hw_cr0;
3423
3424         hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3425         if (enable_unrestricted_guest)
3426                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3427         else {
3428                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3429
3430                 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3431                         enter_pmode(vcpu);
3432
3433                 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3434                         enter_rmode(vcpu);
3435         }
3436
3437 #ifdef CONFIG_X86_64
3438         if (vcpu->arch.efer & EFER_LME) {
3439                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3440                         enter_lmode(vcpu);
3441                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3442                         exit_lmode(vcpu);
3443         }
3444 #endif
3445
3446         if (enable_ept)
3447                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3448
3449         if (!vcpu->fpu_active)
3450                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3451
3452         vmcs_writel(CR0_READ_SHADOW, cr0);
3453         vmcs_writel(GUEST_CR0, hw_cr0);
3454         vcpu->arch.cr0 = cr0;
3455
3456         /* depends on vcpu->arch.cr0 to be set to a new value */
3457         vmx->emulation_required = emulation_required(vcpu);
3458 }
3459
3460 static u64 construct_eptp(unsigned long root_hpa)
3461 {
3462         u64 eptp;
3463
3464         /* TODO write the value reading from MSR */
3465         eptp = VMX_EPT_DEFAULT_MT |
3466                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3467         if (enable_ept_ad_bits)
3468                 eptp |= VMX_EPT_AD_ENABLE_BIT;
3469         eptp |= (root_hpa & PAGE_MASK);
3470
3471         return eptp;
3472 }
3473
3474 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3475 {
3476         unsigned long guest_cr3;
3477         u64 eptp;
3478
3479         guest_cr3 = cr3;
3480         if (enable_ept) {
3481                 eptp = construct_eptp(cr3);
3482                 vmcs_write64(EPT_POINTER, eptp);
3483                 if (is_paging(vcpu) || is_guest_mode(vcpu))
3484                         guest_cr3 = kvm_read_cr3(vcpu);
3485                 else
3486                         guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3487                 ept_load_pdptrs(vcpu);
3488         }
3489
3490         vmx_flush_tlb(vcpu);
3491         vmcs_writel(GUEST_CR3, guest_cr3);
3492 }
3493
3494 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3495 {
3496         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3497                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3498
3499         if (cr4 & X86_CR4_VMXE) {
3500                 /*
3501                  * To use VMXON (and later other VMX instructions), a guest
3502                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
3503                  * So basically the check on whether to allow nested VMX
3504                  * is here.
3505                  */
3506                 if (!nested_vmx_allowed(vcpu))
3507                         return 1;
3508         }
3509         if (to_vmx(vcpu)->nested.vmxon &&
3510             ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3511                 return 1;
3512
3513         vcpu->arch.cr4 = cr4;
3514         if (enable_ept) {
3515                 if (!is_paging(vcpu)) {
3516                         hw_cr4 &= ~X86_CR4_PAE;
3517                         hw_cr4 |= X86_CR4_PSE;
3518                         /*
3519                          * SMEP/SMAP is disabled if CPU is in non-paging mode
3520                          * in hardware. However KVM always uses paging mode to
3521                          * emulate guest non-paging mode with TDP.
3522                          * To emulate this behavior, SMEP/SMAP needs to be
3523                          * manually disabled when guest switches to non-paging
3524                          * mode.
3525                          */
3526                         hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3527                 } else if (!(cr4 & X86_CR4_PAE)) {
3528                         hw_cr4 &= ~X86_CR4_PAE;
3529                 }
3530         }
3531
3532         vmcs_writel(CR4_READ_SHADOW, cr4);
3533         vmcs_writel(GUEST_CR4, hw_cr4);
3534         return 0;
3535 }
3536
3537 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3538                             struct kvm_segment *var, int seg)
3539 {
3540         struct vcpu_vmx *vmx = to_vmx(vcpu);
3541         u32 ar;
3542
3543         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3544                 *var = vmx->rmode.segs[seg];
3545                 if (seg == VCPU_SREG_TR
3546                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3547                         return;
3548                 var->base = vmx_read_guest_seg_base(vmx, seg);
3549                 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3550                 return;
3551         }
3552         var->base = vmx_read_guest_seg_base(vmx, seg);
3553         var->limit = vmx_read_guest_seg_limit(vmx, seg);
3554         var->selector = vmx_read_guest_seg_selector(vmx, seg);
3555         ar = vmx_read_guest_seg_ar(vmx, seg);
3556         var->unusable = (ar >> 16) & 1;
3557         var->type = ar & 15;
3558         var->s = (ar >> 4) & 1;
3559         var->dpl = (ar >> 5) & 3;
3560         /*
3561          * Some userspaces do not preserve unusable property. Since usable
3562          * segment has to be present according to VMX spec we can use present
3563          * property to amend userspace bug by making unusable segment always
3564          * nonpresent. vmx_segment_access_rights() already marks nonpresent
3565          * segment as unusable.
3566          */
3567         var->present = !var->unusable;
3568         var->avl = (ar >> 12) & 1;
3569         var->l = (ar >> 13) & 1;
3570         var->db = (ar >> 14) & 1;
3571         var->g = (ar >> 15) & 1;
3572 }
3573
3574 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3575 {
3576         struct kvm_segment s;
3577
3578         if (to_vmx(vcpu)->rmode.vm86_active) {
3579                 vmx_get_segment(vcpu, &s, seg);
3580                 return s.base;
3581         }
3582         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3583 }
3584
3585 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3586 {
3587         struct vcpu_vmx *vmx = to_vmx(vcpu);
3588
3589         if (unlikely(vmx->rmode.vm86_active))
3590                 return 0;
3591         else {
3592                 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3593                 return AR_DPL(ar);
3594         }
3595 }
3596
3597 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3598 {
3599         u32 ar;
3600
3601         if (var->unusable || !var->present)
3602                 ar = 1 << 16;
3603         else {
3604                 ar = var->type & 15;
3605                 ar |= (var->s & 1) << 4;
3606                 ar |= (var->dpl & 3) << 5;
3607                 ar |= (var->present & 1) << 7;
3608                 ar |= (var->avl & 1) << 12;
3609                 ar |= (var->l & 1) << 13;
3610                 ar |= (var->db & 1) << 14;
3611                 ar |= (var->g & 1) << 15;
3612         }
3613
3614         return ar;
3615 }
3616
3617 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3618                             struct kvm_segment *var, int seg)
3619 {
3620         struct vcpu_vmx *vmx = to_vmx(vcpu);
3621         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3622
3623         vmx_segment_cache_clear(vmx);
3624
3625         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3626                 vmx->rmode.segs[seg] = *var;
3627                 if (seg == VCPU_SREG_TR)
3628                         vmcs_write16(sf->selector, var->selector);
3629                 else if (var->s)
3630                         fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3631                 goto out;
3632         }
3633
3634         vmcs_writel(sf->base, var->base);
3635         vmcs_write32(sf->limit, var->limit);
3636         vmcs_write16(sf->selector, var->selector);
3637
3638         /*
3639          *   Fix the "Accessed" bit in AR field of segment registers for older
3640          * qemu binaries.
3641          *   IA32 arch specifies that at the time of processor reset the
3642          * "Accessed" bit in the AR field of segment registers is 1. And qemu
3643          * is setting it to 0 in the userland code. This causes invalid guest
3644          * state vmexit when "unrestricted guest" mode is turned on.
3645          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3646          * tree. Newer qemu binaries with that qemu fix would not need this
3647          * kvm hack.
3648          */
3649         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3650                 var->type |= 0x1; /* Accessed */
3651
3652         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3653
3654 out:
3655         vmx->emulation_required |= emulation_required(vcpu);
3656 }
3657
3658 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3659 {
3660         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3661
3662         *db = (ar >> 14) & 1;
3663         *l = (ar >> 13) & 1;
3664 }
3665
3666 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3667 {
3668         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3669         dt->address = vmcs_readl(GUEST_IDTR_BASE);
3670 }
3671
3672 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3673 {
3674         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3675         vmcs_writel(GUEST_IDTR_BASE, dt->address);
3676 }
3677
3678 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3679 {
3680         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3681         dt->address = vmcs_readl(GUEST_GDTR_BASE);
3682 }
3683
3684 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3685 {
3686         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3687         vmcs_writel(GUEST_GDTR_BASE, dt->address);
3688 }
3689
3690 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3691 {
3692         struct kvm_segment var;
3693         u32 ar;
3694
3695         vmx_get_segment(vcpu, &var, seg);
3696         var.dpl = 0x3;
3697         if (seg == VCPU_SREG_CS)
3698                 var.type = 0x3;
3699         ar = vmx_segment_access_rights(&var);
3700
3701         if (var.base != (var.selector << 4))
3702                 return false;
3703         if (var.limit != 0xffff)
3704                 return false;
3705         if (ar != 0xf3)
3706                 return false;
3707
3708         return true;
3709 }
3710
3711 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3712 {
3713         struct kvm_segment cs;
3714         unsigned int cs_rpl;
3715
3716         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3717         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3718
3719         if (cs.unusable)
3720                 return false;
3721         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3722                 return false;
3723         if (!cs.s)
3724                 return false;
3725         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3726                 if (cs.dpl > cs_rpl)
3727                         return false;
3728         } else {
3729                 if (cs.dpl != cs_rpl)
3730                         return false;
3731         }
3732         if (!cs.present)
3733                 return false;
3734
3735         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3736         return true;
3737 }
3738
3739 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3740 {
3741         struct kvm_segment ss;
3742         unsigned int ss_rpl;
3743
3744         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3745         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3746
3747         if (ss.unusable)
3748                 return true;
3749         if (ss.type != 3 && ss.type != 7)
3750                 return false;
3751         if (!ss.s)
3752                 return false;
3753         if (ss.dpl != ss_rpl) /* DPL != RPL */
3754                 return false;
3755         if (!ss.present)
3756                 return false;
3757
3758         return true;
3759 }
3760
3761 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3762 {
3763         struct kvm_segment var;
3764         unsigned int rpl;
3765
3766         vmx_get_segment(vcpu, &var, seg);
3767         rpl = var.selector & SELECTOR_RPL_MASK;
3768
3769         if (var.unusable)
3770                 return true;
3771         if (!var.s)
3772                 return false;
3773         if (!var.present)
3774                 return false;
3775         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3776                 if (var.dpl < rpl) /* DPL < RPL */
3777                         return false;
3778         }
3779
3780         /* TODO: Add other members to kvm_segment_field to allow checking for other access
3781          * rights flags
3782          */
3783         return true;
3784 }
3785
3786 static bool tr_valid(struct kvm_vcpu *vcpu)
3787 {
3788         struct kvm_segment tr;
3789
3790         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3791
3792         if (tr.unusable)
3793                 return false;
3794         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
3795                 return false;
3796         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3797                 return false;
3798         if (!tr.present)
3799                 return false;
3800
3801         return true;
3802 }
3803
3804 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3805 {
3806         struct kvm_segment ldtr;
3807
3808         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3809
3810         if (ldtr.unusable)
3811                 return true;
3812         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
3813                 return false;
3814         if (ldtr.type != 2)
3815                 return false;
3816         if (!ldtr.present)
3817                 return false;
3818
3819         return true;
3820 }
3821
3822 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3823 {
3824         struct kvm_segment cs, ss;
3825
3826         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3827         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3828
3829         return ((cs.selector & SELECTOR_RPL_MASK) ==
3830                  (ss.selector & SELECTOR_RPL_MASK));
3831 }
3832
3833 /*
3834  * Check if guest state is valid. Returns true if valid, false if
3835  * not.
3836  * We assume that registers are always usable
3837  */
3838 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3839 {
3840         if (enable_unrestricted_guest)
3841                 return true;
3842
3843         /* real mode guest state checks */
3844         if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3845                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3846                         return false;
3847                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3848                         return false;
3849                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3850                         return false;
3851                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3852                         return false;
3853                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3854                         return false;
3855                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3856                         return false;
3857         } else {
3858         /* protected mode guest state checks */
3859                 if (!cs_ss_rpl_check(vcpu))
3860                         return false;
3861                 if (!code_segment_valid(vcpu))
3862                         return false;
3863                 if (!stack_segment_valid(vcpu))
3864                         return false;
3865                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3866                         return false;
3867                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3868                         return false;
3869                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3870                         return false;
3871                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3872                         return false;
3873                 if (!tr_valid(vcpu))
3874                         return false;
3875                 if (!ldtr_valid(vcpu))
3876                         return false;
3877         }
3878         /* TODO:
3879          * - Add checks on RIP
3880          * - Add checks on RFLAGS
3881          */
3882
3883         return true;
3884 }
3885
3886 static int init_rmode_tss(struct kvm *kvm)
3887 {
3888         gfn_t fn;
3889         u16 data = 0;
3890         int r, idx, ret = 0;
3891
3892         idx = srcu_read_lock(&kvm->srcu);
3893         fn = kvm->arch.tss_addr >> PAGE_SHIFT;
3894         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3895         if (r < 0)
3896                 goto out;
3897         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3898         r = kvm_write_guest_page(kvm, fn++, &data,
3899                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
3900         if (r < 0)
3901                 goto out;
3902         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3903         if (r < 0)
3904                 goto out;
3905         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3906         if (r < 0)
3907                 goto out;
3908         data = ~0;
3909         r = kvm_write_guest_page(kvm, fn, &data,
3910                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3911                                  sizeof(u8));
3912         if (r < 0)
3913                 goto out;
3914
3915         ret = 1;
3916 out:
3917         srcu_read_unlock(&kvm->srcu, idx);
3918         return ret;
3919 }
3920
3921 static int init_rmode_identity_map(struct kvm *kvm)
3922 {
3923         int i, idx, r, ret;
3924         pfn_t identity_map_pfn;
3925         u32 tmp;
3926
3927         if (!enable_ept)
3928                 return 1;
3929         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3930                 printk(KERN_ERR "EPT: identity-mapping pagetable "
3931                         "haven't been allocated!\n");
3932                 return 0;
3933         }
3934         if (likely(kvm->arch.ept_identity_pagetable_done))
3935                 return 1;
3936         ret = 0;
3937         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3938         idx = srcu_read_lock(&kvm->srcu);
3939         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3940         if (r < 0)
3941                 goto out;
3942         /* Set up identity-mapping pagetable for EPT in real mode */
3943         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3944                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3945                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3946                 r = kvm_write_guest_page(kvm, identity_map_pfn,
3947                                 &tmp, i * sizeof(tmp), sizeof(tmp));
3948                 if (r < 0)
3949                         goto out;
3950         }
3951         kvm->arch.ept_identity_pagetable_done = true;
3952         ret = 1;
3953 out:
3954         srcu_read_unlock(&kvm->srcu, idx);
3955         return ret;
3956 }
3957
3958 static void seg_setup(int seg)
3959 {
3960         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3961         unsigned int ar;
3962
3963         vmcs_write16(sf->selector, 0);
3964         vmcs_writel(sf->base, 0);
3965         vmcs_write32(sf->limit, 0xffff);
3966         ar = 0x93;
3967         if (seg == VCPU_SREG_CS)
3968                 ar |= 0x08; /* code segment */
3969
3970         vmcs_write32(sf->ar_bytes, ar);
3971 }
3972
3973 static int alloc_apic_access_page(struct kvm *kvm)
3974 {
3975         struct page *page;
3976         struct kvm_userspace_memory_region kvm_userspace_mem;
3977         int r = 0;
3978
3979         mutex_lock(&kvm->slots_lock);
3980         if (kvm->arch.apic_access_page)
3981                 goto out;
3982         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3983         kvm_userspace_mem.flags = 0;
3984         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3985         kvm_userspace_mem.memory_size = PAGE_SIZE;
3986         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3987         if (r)
3988                 goto out;
3989
3990         page = gfn_to_page(kvm, 0xfee00);
3991         if (is_error_page(page)) {
3992                 r = -EFAULT;
3993                 goto out;
3994         }
3995
3996         kvm->arch.apic_access_page = page;
3997 out:
3998         mutex_unlock(&kvm->slots_lock);
3999         return r;
4000 }
4001
4002 static int alloc_identity_pagetable(struct kvm *kvm)
4003 {
4004         struct page *page;
4005         struct kvm_userspace_memory_region kvm_userspace_mem;
4006         int r = 0;
4007
4008         mutex_lock(&kvm->slots_lock);
4009         if (kvm->arch.ept_identity_pagetable)
4010                 goto out;
4011         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
4012         kvm_userspace_mem.flags = 0;
4013         kvm_userspace_mem.guest_phys_addr =
4014                 kvm->arch.ept_identity_map_addr;
4015         kvm_userspace_mem.memory_size = PAGE_SIZE;
4016         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
4017         if (r)
4018                 goto out;
4019
4020         page = gfn_to_page(kvm, kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
4021         if (is_error_page(page)) {
4022                 r = -EFAULT;
4023                 goto out;
4024         }
4025
4026         kvm->arch.ept_identity_pagetable = page;
4027 out:
4028         mutex_unlock(&kvm->slots_lock);
4029         return r;
4030 }
4031
4032 static void allocate_vpid(struct vcpu_vmx *vmx)
4033 {
4034         int vpid;
4035
4036         vmx->vpid = 0;
4037         if (!enable_vpid)
4038                 return;
4039         spin_lock(&vmx_vpid_lock);
4040         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4041         if (vpid < VMX_NR_VPIDS) {
4042                 vmx->vpid = vpid;
4043                 __set_bit(vpid, vmx_vpid_bitmap);
4044         }
4045         spin_unlock(&vmx_vpid_lock);
4046 }
4047
4048 static void free_vpid(struct vcpu_vmx *vmx)
4049 {
4050         if (!enable_vpid)
4051                 return;
4052         spin_lock(&vmx_vpid_lock);
4053         if (vmx->vpid != 0)
4054                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
4055         spin_unlock(&vmx_vpid_lock);
4056 }
4057
4058 #define MSR_TYPE_R      1
4059 #define MSR_TYPE_W      2
4060 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4061                                                 u32 msr, int type)
4062 {
4063         int f = sizeof(unsigned long);
4064
4065         if (!cpu_has_vmx_msr_bitmap())
4066                 return;
4067
4068         /*
4069          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4070          * have the write-low and read-high bitmap offsets the wrong way round.
4071          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4072          */
4073         if (msr <= 0x1fff) {
4074                 if (type & MSR_TYPE_R)
4075                         /* read-low */
4076                         __clear_bit(msr, msr_bitmap + 0x000 / f);
4077
4078                 if (type & MSR_TYPE_W)
4079                         /* write-low */
4080                         __clear_bit(msr, msr_bitmap + 0x800 / f);
4081
4082         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4083                 msr &= 0x1fff;
4084                 if (type & MSR_TYPE_R)
4085                         /* read-high */
4086                         __clear_bit(msr, msr_bitmap + 0x400 / f);
4087
4088                 if (type & MSR_TYPE_W)
4089                         /* write-high */
4090                         __clear_bit(msr, msr_bitmap + 0xc00 / f);
4091
4092         }
4093 }
4094
4095 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4096                                                 u32 msr, int type)
4097 {
4098         int f = sizeof(unsigned long);
4099
4100         if (!cpu_has_vmx_msr_bitmap())
4101                 return;
4102
4103         /*
4104          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4105          * have the write-low and read-high bitmap offsets the wrong way round.
4106          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4107          */
4108         if (msr <= 0x1fff) {
4109                 if (type & MSR_TYPE_R)
4110                         /* read-low */
4111                         __set_bit(msr, msr_bitmap + 0x000 / f);
4112
4113                 if (type & MSR_TYPE_W)
4114                         /* write-low */
4115                         __set_bit(msr, msr_bitmap + 0x800 / f);
4116
4117         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4118                 msr &= 0x1fff;
4119                 if (type & MSR_TYPE_R)
4120                         /* read-high */
4121                         __set_bit(msr, msr_bitmap + 0x400 / f);
4122
4123                 if (type & MSR_TYPE_W)
4124                         /* write-high */
4125                         __set_bit(msr, msr_bitmap + 0xc00 / f);
4126
4127         }
4128 }
4129
4130 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4131 {
4132         if (!longmode_only)
4133                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4134                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
4135         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4136                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
4137 }
4138
4139 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4140 {
4141         __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4142                         msr, MSR_TYPE_R);
4143         __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4144                         msr, MSR_TYPE_R);
4145 }
4146
4147 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4148 {
4149         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4150                         msr, MSR_TYPE_R);
4151         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4152                         msr, MSR_TYPE_R);
4153 }
4154
4155 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4156 {
4157         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4158                         msr, MSR_TYPE_W);
4159         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4160                         msr, MSR_TYPE_W);
4161 }
4162
4163 static int vmx_vm_has_apicv(struct kvm *kvm)
4164 {
4165         return enable_apicv && irqchip_in_kernel(kvm);
4166 }
4167
4168 /*
4169  * Send interrupt to vcpu via posted interrupt way.
4170  * 1. If target vcpu is running(non-root mode), send posted interrupt
4171  * notification to vcpu and hardware will sync PIR to vIRR atomically.
4172  * 2. If target vcpu isn't running(root mode), kick it to pick up the
4173  * interrupt from PIR in next vmentry.
4174  */
4175 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4176 {
4177         struct vcpu_vmx *vmx = to_vmx(vcpu);
4178         int r;
4179
4180         if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4181                 return;
4182
4183         r = pi_test_and_set_on(&vmx->pi_desc);
4184         kvm_make_request(KVM_REQ_EVENT, vcpu);
4185 #ifdef CONFIG_SMP
4186         if (!r && (vcpu->mode == IN_GUEST_MODE))
4187                 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4188                                 POSTED_INTR_VECTOR);
4189         else
4190 #endif
4191                 kvm_vcpu_kick(vcpu);
4192 }
4193
4194 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4195 {
4196         struct vcpu_vmx *vmx = to_vmx(vcpu);
4197
4198         if (!pi_test_and_clear_on(&vmx->pi_desc))
4199                 return;
4200
4201         kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4202 }
4203
4204 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4205 {
4206         return;
4207 }
4208
4209 /*
4210  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4211  * will not change in the lifetime of the guest.
4212  * Note that host-state that does change is set elsewhere. E.g., host-state
4213  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4214  */
4215 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4216 {
4217         u32 low32, high32;
4218         unsigned long tmpl;
4219         struct desc_ptr dt;
4220
4221         vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS);  /* 22.2.3 */
4222         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
4223         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
4224
4225         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
4226 #ifdef CONFIG_X86_64
4227         /*
4228          * Load null selectors, so we can avoid reloading them in
4229          * __vmx_load_host_state(), in case userspace uses the null selectors
4230          * too (the expected case).
4231          */
4232         vmcs_write16(HOST_DS_SELECTOR, 0);
4233         vmcs_write16(HOST_ES_SELECTOR, 0);
4234 #else
4235         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4236         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4237 #endif
4238         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4239         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
4240
4241         native_store_idt(&dt);
4242         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
4243         vmx->host_idt_base = dt.address;
4244
4245         vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4246
4247         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4248         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4249         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4250         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
4251
4252         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4253                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4254                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4255         }
4256 }
4257
4258 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4259 {
4260         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4261         if (enable_ept)
4262                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4263         if (is_guest_mode(&vmx->vcpu))
4264                 vmx->vcpu.arch.cr4_guest_owned_bits &=
4265                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4266         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4267 }
4268
4269 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4270 {
4271         u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4272
4273         if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4274                 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4275         return pin_based_exec_ctrl;
4276 }
4277
4278 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4279 {
4280         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4281
4282         if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4283                 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4284
4285         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
4286                 exec_control &= ~CPU_BASED_TPR_SHADOW;
4287 #ifdef CONFIG_X86_64
4288                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4289                                 CPU_BASED_CR8_LOAD_EXITING;
4290 #endif
4291         }
4292         if (!enable_ept)
4293                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4294                                 CPU_BASED_CR3_LOAD_EXITING  |
4295                                 CPU_BASED_INVLPG_EXITING;
4296         return exec_control;
4297 }
4298
4299 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4300 {
4301         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4302         if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4303                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4304         if (vmx->vpid == 0)
4305                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4306         if (!enable_ept) {
4307                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4308                 enable_unrestricted_guest = 0;
4309                 /* Enable INVPCID for non-ept guests may cause performance regression. */
4310                 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4311         }
4312         if (!enable_unrestricted_guest)
4313                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4314         if (!ple_gap)
4315                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4316         if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4317                 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4318                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4319         exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4320         /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4321            (handle_vmptrld).
4322            We can NOT enable shadow_vmcs here because we don't have yet
4323            a current VMCS12
4324         */
4325         exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4326         return exec_control;
4327 }
4328
4329 static void ept_set_mmio_spte_mask(void)
4330 {
4331         /*
4332          * EPT Misconfigurations can be generated if the value of bits 2:0
4333          * of an EPT paging-structure entry is 110b (write/execute).
4334          * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4335          * spte.
4336          */
4337         kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4338 }
4339
4340 /*
4341  * Sets up the vmcs for emulated real mode.
4342  */
4343 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4344 {
4345 #ifdef CONFIG_X86_64
4346         unsigned long a;
4347 #endif
4348         int i;
4349
4350         /* I/O */
4351         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4352         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4353
4354         if (enable_shadow_vmcs) {
4355                 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4356                 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4357         }
4358         if (cpu_has_vmx_msr_bitmap())
4359                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4360
4361         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4362
4363         /* Control */
4364         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4365
4366         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4367
4368         if (cpu_has_secondary_exec_ctrls()) {
4369                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4370                                 vmx_secondary_exec_control(vmx));
4371         }
4372
4373         if (vmx_vm_has_apicv(vmx->vcpu.kvm)) {
4374                 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4375                 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4376                 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4377                 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4378
4379                 vmcs_write16(GUEST_INTR_STATUS, 0);
4380
4381                 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4382                 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4383         }
4384
4385         if (ple_gap) {
4386                 vmcs_write32(PLE_GAP, ple_gap);
4387                 vmcs_write32(PLE_WINDOW, ple_window);
4388         }
4389
4390         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4391         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4392         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4393
4394         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4395         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4396         vmx_set_constant_host_state(vmx);
4397 #ifdef CONFIG_X86_64
4398         rdmsrl(MSR_FS_BASE, a);
4399         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4400         rdmsrl(MSR_GS_BASE, a);
4401         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4402 #else
4403         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4404         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4405 #endif
4406
4407         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4408         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4409         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4410         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4411         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4412
4413         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
4414                 u32 msr_low, msr_high;
4415                 u64 host_pat;
4416                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
4417                 host_pat = msr_low | ((u64) msr_high << 32);
4418                 /* Write the default value follow host pat */
4419                 vmcs_write64(GUEST_IA32_PAT, host_pat);
4420                 /* Keep arch.pat sync with GUEST_IA32_PAT */
4421                 vmx->vcpu.arch.pat = host_pat;
4422         }
4423
4424         for (i = 0; i < NR_VMX_MSR; ++i) {
4425                 u32 index = vmx_msr_index[i];
4426                 u32 data_low, data_high;
4427                 int j = vmx->nmsrs;
4428
4429                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4430                         continue;
4431                 if (wrmsr_safe(index, data_low, data_high) < 0)
4432                         continue;
4433                 vmx->guest_msrs[j].index = i;
4434                 vmx->guest_msrs[j].data = 0;
4435                 vmx->guest_msrs[j].mask = -1ull;
4436                 ++vmx->nmsrs;
4437         }
4438
4439
4440         vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4441
4442         /* 22.2.1, 20.8.1 */
4443         vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4444
4445         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4446         set_cr4_guest_host_mask(vmx);
4447
4448         return 0;
4449 }
4450
4451 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4452 {
4453         struct vcpu_vmx *vmx = to_vmx(vcpu);
4454         struct msr_data apic_base_msr;
4455
4456         vmx->rmode.vm86_active = 0;
4457
4458         vmx->soft_vnmi_blocked = 0;
4459
4460         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4461         kvm_set_cr8(&vmx->vcpu, 0);
4462         apic_base_msr.data = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
4463         if (kvm_vcpu_is_bsp(&vmx->vcpu))
4464                 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4465         apic_base_msr.host_initiated = true;
4466         kvm_set_apic_base(&vmx->vcpu, &apic_base_msr);
4467
4468         vmx_segment_cache_clear(vmx);
4469
4470         seg_setup(VCPU_SREG_CS);
4471         vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4472         vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4473
4474         seg_setup(VCPU_SREG_DS);
4475         seg_setup(VCPU_SREG_ES);
4476         seg_setup(VCPU_SREG_FS);
4477         seg_setup(VCPU_SREG_GS);
4478         seg_setup(VCPU_SREG_SS);
4479
4480         vmcs_write16(GUEST_TR_SELECTOR, 0);
4481         vmcs_writel(GUEST_TR_BASE, 0);
4482         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4483         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4484
4485         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4486         vmcs_writel(GUEST_LDTR_BASE, 0);
4487         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4488         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4489
4490         vmcs_write32(GUEST_SYSENTER_CS, 0);
4491         vmcs_writel(GUEST_SYSENTER_ESP, 0);
4492         vmcs_writel(GUEST_SYSENTER_EIP, 0);
4493
4494         vmcs_writel(GUEST_RFLAGS, 0x02);
4495         kvm_rip_write(vcpu, 0xfff0);
4496
4497         vmcs_writel(GUEST_GDTR_BASE, 0);
4498         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4499
4500         vmcs_writel(GUEST_IDTR_BASE, 0);
4501         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4502
4503         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4504         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4505         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4506
4507         /* Special registers */
4508         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4509
4510         setup_msrs(vmx);
4511
4512         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4513
4514         if (cpu_has_vmx_tpr_shadow()) {
4515                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4516                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
4517                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4518                                      __pa(vmx->vcpu.arch.apic->regs));
4519                 vmcs_write32(TPR_THRESHOLD, 0);
4520         }
4521
4522         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4523                 vmcs_write64(APIC_ACCESS_ADDR,
4524                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
4525
4526         if (vmx_vm_has_apicv(vcpu->kvm))
4527                 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4528
4529         if (vmx->vpid != 0)
4530                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4531
4532         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4533         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4534         vmx_set_cr4(&vmx->vcpu, 0);
4535         vmx_set_efer(&vmx->vcpu, 0);
4536         vmx_fpu_activate(&vmx->vcpu);
4537         update_exception_bitmap(&vmx->vcpu);
4538
4539         vpid_sync_context(vmx);
4540 }
4541
4542 /*
4543  * In nested virtualization, check if L1 asked to exit on external interrupts.
4544  * For most existing hypervisors, this will always return true.
4545  */
4546 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4547 {
4548         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4549                 PIN_BASED_EXT_INTR_MASK;
4550 }
4551
4552 /*
4553  * In nested virtualization, check if L1 has set
4554  * VM_EXIT_ACK_INTR_ON_EXIT
4555  */
4556 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
4557 {
4558         return get_vmcs12(vcpu)->vm_exit_controls &
4559                 VM_EXIT_ACK_INTR_ON_EXIT;
4560 }
4561
4562 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4563 {
4564         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4565                 PIN_BASED_NMI_EXITING;
4566 }
4567
4568 static void enable_irq_window(struct kvm_vcpu *vcpu)
4569 {
4570         u32 cpu_based_vm_exec_control;
4571
4572         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4573         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4574         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4575 }
4576
4577 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4578 {
4579         u32 cpu_based_vm_exec_control;
4580
4581         if (!cpu_has_virtual_nmis() ||
4582             vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4583                 enable_irq_window(vcpu);
4584                 return;
4585         }
4586
4587         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4588         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4589         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4590 }
4591
4592 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4593 {
4594         struct vcpu_vmx *vmx = to_vmx(vcpu);
4595         uint32_t intr;
4596         int irq = vcpu->arch.interrupt.nr;
4597
4598         trace_kvm_inj_virq(irq);
4599
4600         ++vcpu->stat.irq_injections;
4601         if (vmx->rmode.vm86_active) {
4602                 int inc_eip = 0;
4603                 if (vcpu->arch.interrupt.soft)
4604                         inc_eip = vcpu->arch.event_exit_inst_len;
4605                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4606                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4607                 return;
4608         }
4609         intr = irq | INTR_INFO_VALID_MASK;
4610         if (vcpu->arch.interrupt.soft) {
4611                 intr |= INTR_TYPE_SOFT_INTR;
4612                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4613                              vmx->vcpu.arch.event_exit_inst_len);
4614         } else
4615                 intr |= INTR_TYPE_EXT_INTR;
4616         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4617 }
4618
4619 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4620 {
4621         struct vcpu_vmx *vmx = to_vmx(vcpu);
4622
4623         if (is_guest_mode(vcpu))
4624                 return;
4625
4626         if (!cpu_has_virtual_nmis()) {
4627                 /*
4628                  * Tracking the NMI-blocked state in software is built upon
4629                  * finding the next open IRQ window. This, in turn, depends on
4630                  * well-behaving guests: They have to keep IRQs disabled at
4631                  * least as long as the NMI handler runs. Otherwise we may
4632                  * cause NMI nesting, maybe breaking the guest. But as this is
4633                  * highly unlikely, we can live with the residual risk.
4634                  */
4635                 vmx->soft_vnmi_blocked = 1;
4636                 vmx->vnmi_blocked_time = 0;
4637         }
4638
4639         ++vcpu->stat.nmi_injections;
4640         vmx->nmi_known_unmasked = false;
4641         if (vmx->rmode.vm86_active) {
4642                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4643                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4644                 return;
4645         }
4646         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4647                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4648 }
4649
4650 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4651 {
4652         if (!cpu_has_virtual_nmis())
4653                 return to_vmx(vcpu)->soft_vnmi_blocked;
4654         if (to_vmx(vcpu)->nmi_known_unmasked)
4655                 return false;
4656         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4657 }
4658
4659 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4660 {
4661         struct vcpu_vmx *vmx = to_vmx(vcpu);
4662
4663         if (!cpu_has_virtual_nmis()) {
4664                 if (vmx->soft_vnmi_blocked != masked) {
4665                         vmx->soft_vnmi_blocked = masked;
4666                         vmx->vnmi_blocked_time = 0;
4667                 }
4668         } else {
4669                 vmx->nmi_known_unmasked = !masked;
4670                 if (masked)
4671                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4672                                       GUEST_INTR_STATE_NMI);
4673                 else
4674                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4675                                         GUEST_INTR_STATE_NMI);
4676         }
4677 }
4678
4679 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4680 {
4681         if (to_vmx(vcpu)->nested.nested_run_pending)
4682                 return 0;
4683
4684         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4685                 return 0;
4686
4687         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4688                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4689                    | GUEST_INTR_STATE_NMI));
4690 }
4691
4692 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4693 {
4694         return (!to_vmx(vcpu)->nested.nested_run_pending &&
4695                 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4696                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4697                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4698 }
4699
4700 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4701 {
4702         int ret;
4703         struct kvm_userspace_memory_region tss_mem = {
4704                 .slot = TSS_PRIVATE_MEMSLOT,
4705                 .guest_phys_addr = addr,
4706                 .memory_size = PAGE_SIZE * 3,
4707                 .flags = 0,
4708         };
4709
4710         ret = kvm_set_memory_region(kvm, &tss_mem);
4711         if (ret)
4712                 return ret;
4713         kvm->arch.tss_addr = addr;
4714         if (!init_rmode_tss(kvm))
4715                 return  -ENOMEM;
4716
4717         return 0;
4718 }
4719
4720 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4721 {
4722         switch (vec) {
4723         case BP_VECTOR:
4724                 /*
4725                  * Update instruction length as we may reinject the exception
4726                  * from user space while in guest debugging mode.
4727                  */
4728                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4729                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4730                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4731                         return false;
4732                 /* fall through */
4733         case DB_VECTOR:
4734                 if (vcpu->guest_debug &
4735                         (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4736                         return false;
4737                 /* fall through */
4738         case DE_VECTOR:
4739         case OF_VECTOR:
4740         case BR_VECTOR:
4741         case UD_VECTOR:
4742         case DF_VECTOR:
4743         case SS_VECTOR:
4744         case GP_VECTOR:
4745         case MF_VECTOR:
4746                 return true;
4747         break;
4748         }
4749         return false;
4750 }
4751
4752 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4753                                   int vec, u32 err_code)
4754 {
4755         /*
4756          * Instruction with address size override prefix opcode 0x67
4757          * Cause the #SS fault with 0 error code in VM86 mode.
4758          */
4759         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4760                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4761                         if (vcpu->arch.halt_request) {
4762                                 vcpu->arch.halt_request = 0;
4763                                 return kvm_emulate_halt(vcpu);
4764                         }
4765                         return 1;
4766                 }
4767                 return 0;
4768         }
4769
4770         /*
4771          * Forward all other exceptions that are valid in real mode.
4772          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4773          *        the required debugging infrastructure rework.
4774          */
4775         kvm_queue_exception(vcpu, vec);
4776         return 1;
4777 }
4778
4779 /*
4780  * Trigger machine check on the host. We assume all the MSRs are already set up
4781  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4782  * We pass a fake environment to the machine check handler because we want
4783  * the guest to be always treated like user space, no matter what context
4784  * it used internally.
4785  */
4786 static void kvm_machine_check(void)
4787 {
4788 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4789         struct pt_regs regs = {
4790                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4791                 .flags = X86_EFLAGS_IF,
4792         };
4793
4794         do_machine_check(&regs, 0);
4795 #endif
4796 }
4797
4798 static int handle_machine_check(struct kvm_vcpu *vcpu)
4799 {
4800         /* already handled by vcpu_run */
4801         return 1;
4802 }
4803
4804 static int handle_exception(struct kvm_vcpu *vcpu)
4805 {
4806         struct vcpu_vmx *vmx = to_vmx(vcpu);
4807         struct kvm_run *kvm_run = vcpu->run;
4808         u32 intr_info, ex_no, error_code;
4809         unsigned long cr2, rip, dr6;
4810         u32 vect_info;
4811         enum emulation_result er;
4812
4813         vect_info = vmx->idt_vectoring_info;
4814         intr_info = vmx->exit_intr_info;
4815
4816         if (is_machine_check(intr_info))
4817                 return handle_machine_check(vcpu);
4818
4819         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4820                 return 1;  /* already handled by vmx_vcpu_run() */
4821
4822         if (is_no_device(intr_info)) {
4823                 vmx_fpu_activate(vcpu);
4824                 return 1;
4825         }
4826
4827         if (is_invalid_opcode(intr_info)) {
4828                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4829                 if (er != EMULATE_DONE)
4830                         kvm_queue_exception(vcpu, UD_VECTOR);
4831                 return 1;
4832         }
4833
4834         error_code = 0;
4835         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4836                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4837
4838         /*
4839          * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4840          * MMIO, it is better to report an internal error.
4841          * See the comments in vmx_handle_exit.
4842          */
4843         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4844             !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4845                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4846                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4847                 vcpu->run->internal.ndata = 2;
4848                 vcpu->run->internal.data[0] = vect_info;
4849                 vcpu->run->internal.data[1] = intr_info;
4850                 return 0;
4851         }
4852
4853         if (is_page_fault(intr_info)) {
4854                 /* EPT won't cause page fault directly */
4855                 BUG_ON(enable_ept);
4856                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4857                 trace_kvm_page_fault(cr2, error_code);
4858
4859                 if (kvm_event_needs_reinjection(vcpu))
4860                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
4861                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4862         }
4863
4864         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4865
4866         if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4867                 return handle_rmode_exception(vcpu, ex_no, error_code);
4868
4869         switch (ex_no) {
4870         case DB_VECTOR:
4871                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4872                 if (!(vcpu->guest_debug &
4873                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4874                         vcpu->arch.dr6 &= ~15;
4875                         vcpu->arch.dr6 |= dr6;
4876                         if (!(dr6 & ~DR6_RESERVED)) /* icebp */
4877                                 skip_emulated_instruction(vcpu);
4878
4879                         kvm_queue_exception(vcpu, DB_VECTOR);
4880                         return 1;
4881                 }
4882                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4883                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4884                 /* fall through */
4885         case BP_VECTOR:
4886                 /*
4887                  * Update instruction length as we may reinject #BP from
4888                  * user space while in guest debugging mode. Reading it for
4889                  * #DB as well causes no harm, it is not used in that case.
4890                  */
4891                 vmx->vcpu.arch.event_exit_inst_len =
4892                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4893                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4894                 rip = kvm_rip_read(vcpu);
4895                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4896                 kvm_run->debug.arch.exception = ex_no;
4897                 break;
4898         default:
4899                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4900                 kvm_run->ex.exception = ex_no;
4901                 kvm_run->ex.error_code = error_code;
4902                 break;
4903         }
4904         return 0;
4905 }
4906
4907 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4908 {
4909         ++vcpu->stat.irq_exits;
4910         return 1;
4911 }
4912
4913 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4914 {
4915         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4916         return 0;
4917 }
4918
4919 static int handle_io(struct kvm_vcpu *vcpu)
4920 {
4921         unsigned long exit_qualification;
4922         int size, in, string;
4923         unsigned port;
4924
4925         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4926         string = (exit_qualification & 16) != 0;
4927         in = (exit_qualification & 8) != 0;
4928
4929         ++vcpu->stat.io_exits;
4930
4931         if (string || in)
4932                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4933
4934         port = exit_qualification >> 16;
4935         size = (exit_qualification & 7) + 1;
4936         skip_emulated_instruction(vcpu);
4937
4938         return kvm_fast_pio_out(vcpu, size, port);
4939 }
4940
4941 static void
4942 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4943 {
4944         /*
4945          * Patch in the VMCALL instruction:
4946          */
4947         hypercall[0] = 0x0f;
4948         hypercall[1] = 0x01;
4949         hypercall[2] = 0xc1;
4950 }
4951
4952 static bool nested_cr0_valid(struct vmcs12 *vmcs12, unsigned long val)
4953 {
4954         unsigned long always_on = VMXON_CR0_ALWAYSON;
4955
4956         if (nested_vmx_secondary_ctls_high &
4957                 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4958             nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4959                 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
4960         return (val & always_on) == always_on;
4961 }
4962
4963 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4964 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4965 {
4966         if (is_guest_mode(vcpu)) {
4967                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4968                 unsigned long orig_val = val;
4969
4970                 /*
4971                  * We get here when L2 changed cr0 in a way that did not change
4972                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4973                  * but did change L0 shadowed bits. So we first calculate the
4974                  * effective cr0 value that L1 would like to write into the
4975                  * hardware. It consists of the L2-owned bits from the new
4976                  * value combined with the L1-owned bits from L1's guest_cr0.
4977                  */
4978                 val = (val & ~vmcs12->cr0_guest_host_mask) |
4979                         (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4980
4981                 if (!nested_cr0_valid(vmcs12, val))
4982                         return 1;
4983
4984                 if (kvm_set_cr0(vcpu, val))
4985                         return 1;
4986                 vmcs_writel(CR0_READ_SHADOW, orig_val);
4987                 return 0;
4988         } else {
4989                 if (to_vmx(vcpu)->nested.vmxon &&
4990                     ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4991                         return 1;
4992                 return kvm_set_cr0(vcpu, val);
4993         }
4994 }
4995
4996 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4997 {
4998         if (is_guest_mode(vcpu)) {
4999                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5000                 unsigned long orig_val = val;
5001
5002                 /* analogously to handle_set_cr0 */
5003                 val = (val & ~vmcs12->cr4_guest_host_mask) |
5004                         (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5005                 if (kvm_set_cr4(vcpu, val))
5006                         return 1;
5007                 vmcs_writel(CR4_READ_SHADOW, orig_val);
5008                 return 0;
5009         } else
5010                 return kvm_set_cr4(vcpu, val);
5011 }
5012
5013 /* called to set cr0 as approriate for clts instruction exit. */
5014 static void handle_clts(struct kvm_vcpu *vcpu)
5015 {
5016         if (is_guest_mode(vcpu)) {
5017                 /*
5018                  * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5019                  * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5020                  * just pretend it's off (also in arch.cr0 for fpu_activate).
5021                  */
5022                 vmcs_writel(CR0_READ_SHADOW,
5023                         vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5024                 vcpu->arch.cr0 &= ~X86_CR0_TS;
5025         } else
5026                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5027 }
5028
5029 static int handle_cr(struct kvm_vcpu *vcpu)
5030 {
5031         unsigned long exit_qualification, val;
5032         int cr;
5033         int reg;
5034         int err;
5035
5036         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5037         cr = exit_qualification & 15;
5038         reg = (exit_qualification >> 8) & 15;
5039         switch ((exit_qualification >> 4) & 3) {
5040         case 0: /* mov to cr */
5041                 val = kvm_register_read(vcpu, reg);
5042                 trace_kvm_cr_write(cr, val);
5043                 switch (cr) {
5044                 case 0:
5045                         err = handle_set_cr0(vcpu, val);
5046                         kvm_complete_insn_gp(vcpu, err);
5047                         return 1;
5048                 case 3:
5049                         err = kvm_set_cr3(vcpu, val);
5050                         kvm_complete_insn_gp(vcpu, err);
5051                         return 1;
5052                 case 4:
5053                         err = handle_set_cr4(vcpu, val);
5054                         kvm_complete_insn_gp(vcpu, err);
5055                         return 1;
5056                 case 8: {
5057                                 u8 cr8_prev = kvm_get_cr8(vcpu);
5058                                 u8 cr8 = kvm_register_read(vcpu, reg);
5059                                 err = kvm_set_cr8(vcpu, cr8);
5060                                 kvm_complete_insn_gp(vcpu, err);
5061                                 if (irqchip_in_kernel(vcpu->kvm))
5062                                         return 1;
5063                                 if (cr8_prev <= cr8)
5064                                         return 1;
5065                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5066                                 return 0;
5067                         }
5068                 }
5069                 break;
5070         case 2: /* clts */
5071                 handle_clts(vcpu);
5072                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5073                 skip_emulated_instruction(vcpu);
5074                 vmx_fpu_activate(vcpu);
5075                 return 1;
5076         case 1: /*mov from cr*/
5077                 switch (cr) {
5078                 case 3:
5079                         val = kvm_read_cr3(vcpu);
5080                         kvm_register_write(vcpu, reg, val);
5081                         trace_kvm_cr_read(cr, val);
5082                         skip_emulated_instruction(vcpu);
5083                         return 1;
5084                 case 8:
5085                         val = kvm_get_cr8(vcpu);
5086                         kvm_register_write(vcpu, reg, val);
5087                         trace_kvm_cr_read(cr, val);
5088                         skip_emulated_instruction(vcpu);
5089                         return 1;
5090                 }
5091                 break;
5092         case 3: /* lmsw */
5093                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5094                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5095                 kvm_lmsw(vcpu, val);
5096
5097                 skip_emulated_instruction(vcpu);
5098                 return 1;
5099         default:
5100                 break;
5101         }
5102         vcpu->run->exit_reason = 0;
5103         vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5104                (int)(exit_qualification >> 4) & 3, cr);
5105         return 0;
5106 }
5107
5108 static int handle_dr(struct kvm_vcpu *vcpu)
5109 {
5110         unsigned long exit_qualification;
5111         int dr, reg;
5112
5113         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5114         if (!kvm_require_cpl(vcpu, 0))
5115                 return 1;
5116         dr = vmcs_readl(GUEST_DR7);
5117         if (dr & DR7_GD) {
5118                 /*
5119                  * As the vm-exit takes precedence over the debug trap, we
5120                  * need to emulate the latter, either for the host or the
5121                  * guest debugging itself.
5122                  */
5123                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5124                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5125                         vcpu->run->debug.arch.dr7 = dr;
5126                         vcpu->run->debug.arch.pc =
5127                                 vmcs_readl(GUEST_CS_BASE) +
5128                                 vmcs_readl(GUEST_RIP);
5129                         vcpu->run->debug.arch.exception = DB_VECTOR;
5130                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5131                         return 0;
5132                 } else {
5133                         vcpu->arch.dr7 &= ~DR7_GD;
5134                         vcpu->arch.dr6 |= DR6_BD;
5135                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
5136                         kvm_queue_exception(vcpu, DB_VECTOR);
5137                         return 1;
5138                 }
5139         }
5140
5141         if (vcpu->guest_debug == 0) {
5142                 u32 cpu_based_vm_exec_control;
5143
5144                 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5145                 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5146                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5147
5148                 /*
5149                  * No more DR vmexits; force a reload of the debug registers
5150                  * and reenter on this instruction.  The next vmexit will
5151                  * retrieve the full state of the debug registers.
5152                  */
5153                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5154                 return 1;
5155         }
5156
5157         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5158         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5159         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5160         if (exit_qualification & TYPE_MOV_FROM_DR) {
5161                 unsigned long val;
5162
5163                 if (kvm_get_dr(vcpu, dr, &val))
5164                         return 1;
5165                 kvm_register_write(vcpu, reg, val);
5166         } else
5167                 if (kvm_set_dr(vcpu, dr, kvm_register_read(vcpu, reg)))
5168                         return 1;
5169
5170         skip_emulated_instruction(vcpu);
5171         return 1;
5172 }
5173
5174 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5175 {
5176         return vcpu->arch.dr6;
5177 }
5178
5179 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5180 {
5181 }
5182
5183 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5184 {
5185         u32 cpu_based_vm_exec_control;
5186
5187         get_debugreg(vcpu->arch.db[0], 0);
5188         get_debugreg(vcpu->arch.db[1], 1);
5189         get_debugreg(vcpu->arch.db[2], 2);
5190         get_debugreg(vcpu->arch.db[3], 3);
5191         get_debugreg(vcpu->arch.dr6, 6);
5192         vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5193
5194         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5195
5196         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5197         cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5198         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5199 }
5200
5201 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5202 {
5203         vmcs_writel(GUEST_DR7, val);
5204 }
5205
5206 static int handle_cpuid(struct kvm_vcpu *vcpu)
5207 {
5208         kvm_emulate_cpuid(vcpu);
5209         return 1;
5210 }
5211
5212 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5213 {
5214         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5215         u64 data;
5216
5217         if (vmx_get_msr(vcpu, ecx, &data)) {
5218                 trace_kvm_msr_read_ex(ecx);
5219                 kvm_inject_gp(vcpu, 0);
5220                 return 1;
5221         }
5222
5223         trace_kvm_msr_read(ecx, data);
5224
5225         /* FIXME: handling of bits 32:63 of rax, rdx */
5226         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
5227         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
5228         skip_emulated_instruction(vcpu);
5229         return 1;
5230 }
5231
5232 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5233 {
5234         struct msr_data msr;
5235         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5236         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5237                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5238
5239         msr.data = data;
5240         msr.index = ecx;
5241         msr.host_initiated = false;
5242         if (vmx_set_msr(vcpu, &msr) != 0) {
5243                 trace_kvm_msr_write_ex(ecx, data);
5244                 kvm_inject_gp(vcpu, 0);
5245                 return 1;
5246         }
5247
5248         trace_kvm_msr_write(ecx, data);
5249         skip_emulated_instruction(vcpu);
5250         return 1;
5251 }
5252
5253 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5254 {
5255         kvm_make_request(KVM_REQ_EVENT, vcpu);
5256         return 1;
5257 }
5258
5259 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5260 {
5261         u32 cpu_based_vm_exec_control;
5262
5263         /* clear pending irq */
5264         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5265         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5266         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5267
5268         kvm_make_request(KVM_REQ_EVENT, vcpu);
5269
5270         ++vcpu->stat.irq_window_exits;
5271
5272         /*
5273          * If the user space waits to inject interrupts, exit as soon as
5274          * possible
5275          */
5276         if (!irqchip_in_kernel(vcpu->kvm) &&
5277             vcpu->run->request_interrupt_window &&
5278             !kvm_cpu_has_interrupt(vcpu)) {
5279                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
5280                 return 0;
5281         }
5282         return 1;
5283 }
5284
5285 static int handle_halt(struct kvm_vcpu *vcpu)
5286 {
5287         skip_emulated_instruction(vcpu);
5288         return kvm_emulate_halt(vcpu);
5289 }
5290
5291 static int handle_vmcall(struct kvm_vcpu *vcpu)
5292 {
5293         skip_emulated_instruction(vcpu);
5294         kvm_emulate_hypercall(vcpu);
5295         return 1;
5296 }
5297
5298 static int handle_invd(struct kvm_vcpu *vcpu)
5299 {
5300         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5301 }
5302
5303 static int handle_invlpg(struct kvm_vcpu *vcpu)
5304 {
5305         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5306
5307         kvm_mmu_invlpg(vcpu, exit_qualification);
5308         skip_emulated_instruction(vcpu);
5309         return 1;
5310 }
5311
5312 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5313 {
5314         int err;
5315
5316         err = kvm_rdpmc(vcpu);
5317         kvm_complete_insn_gp(vcpu, err);
5318
5319         return 1;
5320 }
5321
5322 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5323 {
5324         skip_emulated_instruction(vcpu);
5325         kvm_emulate_wbinvd(vcpu);
5326         return 1;
5327 }
5328
5329 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5330 {
5331         u64 new_bv = kvm_read_edx_eax(vcpu);
5332         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5333
5334         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5335                 skip_emulated_instruction(vcpu);
5336         return 1;
5337 }
5338
5339 static int handle_apic_access(struct kvm_vcpu *vcpu)
5340 {
5341         if (likely(fasteoi)) {
5342                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5343                 int access_type, offset;
5344
5345                 access_type = exit_qualification & APIC_ACCESS_TYPE;
5346                 offset = exit_qualification & APIC_ACCESS_OFFSET;
5347                 /*
5348                  * Sane guest uses MOV to write EOI, with written value
5349                  * not cared. So make a short-circuit here by avoiding
5350                  * heavy instruction emulation.
5351                  */
5352                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5353                     (offset == APIC_EOI)) {
5354                         kvm_lapic_set_eoi(vcpu);
5355                         skip_emulated_instruction(vcpu);
5356                         return 1;
5357                 }
5358         }
5359         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5360 }
5361
5362 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5363 {
5364         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5365         int vector = exit_qualification & 0xff;
5366
5367         /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5368         kvm_apic_set_eoi_accelerated(vcpu, vector);
5369         return 1;
5370 }
5371
5372 static int handle_apic_write(struct kvm_vcpu *vcpu)
5373 {
5374         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5375         u32 offset = exit_qualification & 0xfff;
5376
5377         /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5378         kvm_apic_write_nodecode(vcpu, offset);
5379         return 1;
5380 }
5381
5382 static int handle_task_switch(struct kvm_vcpu *vcpu)
5383 {
5384         struct vcpu_vmx *vmx = to_vmx(vcpu);
5385         unsigned long exit_qualification;
5386         bool has_error_code = false;
5387         u32 error_code = 0;
5388         u16 tss_selector;
5389         int reason, type, idt_v, idt_index;
5390
5391         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5392         idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5393         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5394
5395         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5396
5397         reason = (u32)exit_qualification >> 30;
5398         if (reason == TASK_SWITCH_GATE && idt_v) {
5399                 switch (type) {
5400                 case INTR_TYPE_NMI_INTR:
5401                         vcpu->arch.nmi_injected = false;
5402                         vmx_set_nmi_mask(vcpu, true);
5403                         break;
5404                 case INTR_TYPE_EXT_INTR:
5405                 case INTR_TYPE_SOFT_INTR:
5406                         kvm_clear_interrupt_queue(vcpu);
5407                         break;
5408                 case INTR_TYPE_HARD_EXCEPTION:
5409                         if (vmx->idt_vectoring_info &
5410                             VECTORING_INFO_DELIVER_CODE_MASK) {
5411                                 has_error_code = true;
5412                                 error_code =
5413                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
5414                         }
5415                         /* fall through */
5416                 case INTR_TYPE_SOFT_EXCEPTION:
5417                         kvm_clear_exception_queue(vcpu);
5418                         break;
5419                 default:
5420                         break;
5421                 }
5422         }
5423         tss_selector = exit_qualification;
5424
5425         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5426                        type != INTR_TYPE_EXT_INTR &&
5427                        type != INTR_TYPE_NMI_INTR))
5428                 skip_emulated_instruction(vcpu);
5429
5430         if (kvm_task_switch(vcpu, tss_selector,
5431                             type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5432                             has_error_code, error_code) == EMULATE_FAIL) {
5433                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5434                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5435                 vcpu->run->internal.ndata = 0;
5436                 return 0;
5437         }
5438
5439         /* clear all local breakpoint enable flags */
5440         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~0x55);
5441
5442         /*
5443          * TODO: What about debug traps on tss switch?
5444          *       Are we supposed to inject them and update dr6?
5445          */
5446
5447         return 1;
5448 }
5449
5450 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5451 {
5452         unsigned long exit_qualification;
5453         gpa_t gpa;
5454         u32 error_code;
5455         int gla_validity;
5456
5457         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5458
5459         gla_validity = (exit_qualification >> 7) & 0x3;
5460         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5461                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5462                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5463                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5464                         vmcs_readl(GUEST_LINEAR_ADDRESS));
5465                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5466                         (long unsigned int)exit_qualification);
5467                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5468                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5469                 return 0;
5470         }
5471
5472         /*
5473          * EPT violation happened while executing iret from NMI,
5474          * "blocked by NMI" bit has to be set before next VM entry.
5475          * There are errata that may cause this bit to not be set:
5476          * AAK134, BY25.
5477          */
5478         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5479                         cpu_has_virtual_nmis() &&
5480                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5481                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5482
5483         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5484         trace_kvm_page_fault(gpa, exit_qualification);
5485
5486         /* It is a write fault? */
5487         error_code = exit_qualification & (1U << 1);
5488         /* It is a fetch fault? */
5489         error_code |= (exit_qualification & (1U << 2)) << 2;
5490         /* ept page table is present? */
5491         error_code |= (exit_qualification >> 3) & 0x1;
5492
5493         vcpu->arch.exit_qualification = exit_qualification;
5494
5495         return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5496 }
5497
5498 static u64 ept_rsvd_mask(u64 spte, int level)
5499 {
5500         int i;
5501         u64 mask = 0;
5502
5503         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5504                 mask |= (1ULL << i);
5505
5506         if (level > 2)
5507                 /* bits 7:3 reserved */
5508                 mask |= 0xf8;
5509         else if (level == 2) {
5510                 if (spte & (1ULL << 7))
5511                         /* 2MB ref, bits 20:12 reserved */
5512                         mask |= 0x1ff000;
5513                 else
5514                         /* bits 6:3 reserved */
5515                         mask |= 0x78;
5516         }
5517
5518         return mask;
5519 }
5520
5521 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5522                                        int level)
5523 {
5524         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5525
5526         /* 010b (write-only) */
5527         WARN_ON((spte & 0x7) == 0x2);
5528
5529         /* 110b (write/execute) */
5530         WARN_ON((spte & 0x7) == 0x6);
5531
5532         /* 100b (execute-only) and value not supported by logical processor */
5533         if (!cpu_has_vmx_ept_execute_only())
5534                 WARN_ON((spte & 0x7) == 0x4);
5535
5536         /* not 000b */
5537         if ((spte & 0x7)) {
5538                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5539
5540                 if (rsvd_bits != 0) {
5541                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5542                                          __func__, rsvd_bits);
5543                         WARN_ON(1);
5544                 }
5545
5546                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
5547                         u64 ept_mem_type = (spte & 0x38) >> 3;
5548
5549                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
5550                             ept_mem_type == 7) {
5551                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5552                                                 __func__, ept_mem_type);
5553                                 WARN_ON(1);
5554                         }
5555                 }
5556         }
5557 }
5558
5559 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5560 {
5561         u64 sptes[4];
5562         int nr_sptes, i, ret;
5563         gpa_t gpa;
5564
5565         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5566         if (!kvm_io_bus_write(vcpu->kvm, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5567                 skip_emulated_instruction(vcpu);
5568                 return 1;
5569         }
5570
5571         ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5572         if (likely(ret == RET_MMIO_PF_EMULATE))
5573                 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5574                                               EMULATE_DONE;
5575
5576         if (unlikely(ret == RET_MMIO_PF_INVALID))
5577                 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5578
5579         if (unlikely(ret == RET_MMIO_PF_RETRY))
5580                 return 1;
5581
5582         /* It is the real ept misconfig */
5583         printk(KERN_ERR "EPT: Misconfiguration.\n");
5584         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5585
5586         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5587
5588         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5589                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5590
5591         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5592         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5593
5594         return 0;
5595 }
5596
5597 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5598 {
5599         u32 cpu_based_vm_exec_control;
5600
5601         /* clear pending NMI */
5602         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5603         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5604         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5605         ++vcpu->stat.nmi_window_exits;
5606         kvm_make_request(KVM_REQ_EVENT, vcpu);
5607
5608         return 1;
5609 }
5610
5611 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5612 {
5613         struct vcpu_vmx *vmx = to_vmx(vcpu);
5614         enum emulation_result err = EMULATE_DONE;
5615         int ret = 1;
5616         u32 cpu_exec_ctrl;
5617         bool intr_window_requested;
5618         unsigned count = 130;
5619
5620         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5621         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5622
5623         while (!guest_state_valid(vcpu) && count-- != 0) {
5624                 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5625                         return handle_interrupt_window(&vmx->vcpu);
5626
5627                 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5628                         return 1;
5629
5630                 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5631
5632                 if (err == EMULATE_USER_EXIT) {
5633                         ++vcpu->stat.mmio_exits;
5634                         ret = 0;
5635                         goto out;
5636                 }
5637
5638                 if (err != EMULATE_DONE) {
5639                         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5640                         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5641                         vcpu->run->internal.ndata = 0;
5642                         return 0;
5643                 }
5644
5645                 if (vcpu->arch.halt_request) {
5646                         vcpu->arch.halt_request = 0;
5647                         ret = kvm_emulate_halt(vcpu);
5648                         goto out;
5649                 }
5650
5651                 if (signal_pending(current))
5652                         goto out;
5653                 if (need_resched())
5654                         schedule();
5655         }
5656
5657         vmx->emulation_required = emulation_required(vcpu);
5658 out:
5659         return ret;
5660 }
5661
5662 /*
5663  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5664  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5665  */
5666 static int handle_pause(struct kvm_vcpu *vcpu)
5667 {
5668         skip_emulated_instruction(vcpu);
5669         kvm_vcpu_on_spin(vcpu);
5670
5671         return 1;
5672 }
5673
5674 static int handle_nop(struct kvm_vcpu *vcpu)
5675 {
5676         skip_emulated_instruction(vcpu);
5677         return 1;
5678 }
5679
5680 static int handle_mwait(struct kvm_vcpu *vcpu)
5681 {
5682         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5683         return handle_nop(vcpu);
5684 }
5685
5686 static int handle_monitor(struct kvm_vcpu *vcpu)
5687 {
5688         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5689         return handle_nop(vcpu);
5690 }
5691
5692 /*
5693  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5694  * We could reuse a single VMCS for all the L2 guests, but we also want the
5695  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5696  * allows keeping them loaded on the processor, and in the future will allow
5697  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5698  * every entry if they never change.
5699  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5700  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5701  *
5702  * The following functions allocate and free a vmcs02 in this pool.
5703  */
5704
5705 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5706 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5707 {
5708         struct vmcs02_list *item;
5709         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5710                 if (item->vmptr == vmx->nested.current_vmptr) {
5711                         list_move(&item->list, &vmx->nested.vmcs02_pool);
5712                         return &item->vmcs02;
5713                 }
5714
5715         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5716                 /* Recycle the least recently used VMCS. */
5717                 item = list_entry(vmx->nested.vmcs02_pool.prev,
5718                         struct vmcs02_list, list);
5719                 item->vmptr = vmx->nested.current_vmptr;
5720                 list_move(&item->list, &vmx->nested.vmcs02_pool);
5721                 return &item->vmcs02;
5722         }
5723
5724         /* Create a new VMCS */
5725         item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5726         if (!item)
5727                 return NULL;
5728         item->vmcs02.vmcs = alloc_vmcs();
5729         if (!item->vmcs02.vmcs) {
5730                 kfree(item);
5731                 return NULL;
5732         }
5733         loaded_vmcs_init(&item->vmcs02);
5734         item->vmptr = vmx->nested.current_vmptr;
5735         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5736         vmx->nested.vmcs02_num++;
5737         return &item->vmcs02;
5738 }
5739
5740 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5741 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5742 {
5743         struct vmcs02_list *item;
5744         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5745                 if (item->vmptr == vmptr) {
5746                         free_loaded_vmcs(&item->vmcs02);
5747                         list_del(&item->list);
5748                         kfree(item);
5749                         vmx->nested.vmcs02_num--;
5750                         return;
5751                 }
5752 }
5753
5754 /*
5755  * Free all VMCSs saved for this vcpu, except the one pointed by
5756  * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5757  * currently used, if running L2), and vmcs01 when running L2.
5758  */
5759 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5760 {
5761         struct vmcs02_list *item, *n;
5762         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5763                 if (vmx->loaded_vmcs != &item->vmcs02)
5764                         free_loaded_vmcs(&item->vmcs02);
5765                 list_del(&item->list);
5766                 kfree(item);
5767         }
5768         vmx->nested.vmcs02_num = 0;
5769
5770         if (vmx->loaded_vmcs != &vmx->vmcs01)
5771                 free_loaded_vmcs(&vmx->vmcs01);
5772 }
5773
5774 /*
5775  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5776  * set the success or error code of an emulated VMX instruction, as specified
5777  * by Vol 2B, VMX Instruction Reference, "Conventions".
5778  */
5779 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5780 {
5781         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5782                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5783                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5784 }
5785
5786 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5787 {
5788         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5789                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5790                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5791                         | X86_EFLAGS_CF);
5792 }
5793
5794 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5795                                         u32 vm_instruction_error)
5796 {
5797         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5798                 /*
5799                  * failValid writes the error number to the current VMCS, which
5800                  * can't be done there isn't a current VMCS.
5801                  */
5802                 nested_vmx_failInvalid(vcpu);
5803                 return;
5804         }
5805         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5806                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5807                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5808                         | X86_EFLAGS_ZF);
5809         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5810         /*
5811          * We don't need to force a shadow sync because
5812          * VM_INSTRUCTION_ERROR is not shadowed
5813          */
5814 }
5815
5816 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
5817 {
5818         struct vcpu_vmx *vmx =
5819                 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
5820
5821         vmx->nested.preemption_timer_expired = true;
5822         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
5823         kvm_vcpu_kick(&vmx->vcpu);
5824
5825         return HRTIMER_NORESTART;
5826 }
5827
5828 /*
5829  * Decode the memory-address operand of a vmx instruction, as recorded on an
5830  * exit caused by such an instruction (run by a guest hypervisor).
5831  * On success, returns 0. When the operand is invalid, returns 1 and throws
5832  * #UD or #GP.
5833  */
5834 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5835                                  unsigned long exit_qualification,
5836                                  u32 vmx_instruction_info, gva_t *ret)
5837 {
5838         /*
5839          * According to Vol. 3B, "Information for VM Exits Due to Instruction
5840          * Execution", on an exit, vmx_instruction_info holds most of the
5841          * addressing components of the operand. Only the displacement part
5842          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5843          * For how an actual address is calculated from all these components,
5844          * refer to Vol. 1, "Operand Addressing".
5845          */
5846         int  scaling = vmx_instruction_info & 3;
5847         int  addr_size = (vmx_instruction_info >> 7) & 7;
5848         bool is_reg = vmx_instruction_info & (1u << 10);
5849         int  seg_reg = (vmx_instruction_info >> 15) & 7;
5850         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
5851         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5852         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
5853         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
5854
5855         if (is_reg) {
5856                 kvm_queue_exception(vcpu, UD_VECTOR);
5857                 return 1;
5858         }
5859
5860         /* Addr = segment_base + offset */
5861         /* offset = base + [index * scale] + displacement */
5862         *ret = vmx_get_segment_base(vcpu, seg_reg);
5863         if (base_is_valid)
5864                 *ret += kvm_register_read(vcpu, base_reg);
5865         if (index_is_valid)
5866                 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5867         *ret += exit_qualification; /* holds the displacement */
5868
5869         if (addr_size == 1) /* 32 bit */
5870                 *ret &= 0xffffffff;
5871
5872         /*
5873          * TODO: throw #GP (and return 1) in various cases that the VM*
5874          * instructions require it - e.g., offset beyond segment limit,
5875          * unusable or unreadable/unwritable segment, non-canonical 64-bit
5876          * address, and so on. Currently these are not checked.
5877          */
5878         return 0;
5879 }
5880
5881 /*
5882  * This function performs the various checks including
5883  * - if it's 4KB aligned
5884  * - No bits beyond the physical address width are set
5885  * - Returns 0 on success or else 1
5886  * (Intel SDM Section 30.3)
5887  */
5888 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
5889                                   gpa_t *vmpointer)
5890 {
5891         gva_t gva;
5892         gpa_t vmptr;
5893         struct x86_exception e;
5894         struct page *page;
5895         struct vcpu_vmx *vmx = to_vmx(vcpu);
5896         int maxphyaddr = cpuid_maxphyaddr(vcpu);
5897
5898         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5899                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5900                 return 1;
5901
5902         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5903                                 sizeof(vmptr), &e)) {
5904                 kvm_inject_page_fault(vcpu, &e);
5905                 return 1;
5906         }
5907
5908         switch (exit_reason) {
5909         case EXIT_REASON_VMON:
5910                 /*
5911                  * SDM 3: 24.11.5
5912                  * The first 4 bytes of VMXON region contain the supported
5913                  * VMCS revision identifier
5914                  *
5915                  * Note - IA32_VMX_BASIC[48] will never be 1
5916                  * for the nested case;
5917                  * which replaces physical address width with 32
5918                  *
5919                  */
5920                 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
5921                         nested_vmx_failInvalid(vcpu);
5922                         skip_emulated_instruction(vcpu);
5923                         return 1;
5924                 }
5925
5926                 page = nested_get_page(vcpu, vmptr);
5927                 if (page == NULL ||
5928                     *(u32 *)kmap(page) != VMCS12_REVISION) {
5929                         nested_vmx_failInvalid(vcpu);
5930                         kunmap(page);
5931                         skip_emulated_instruction(vcpu);
5932                         return 1;
5933                 }
5934                 kunmap(page);
5935                 vmx->nested.vmxon_ptr = vmptr;
5936                 break;
5937         case EXIT_REASON_VMCLEAR:
5938                 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
5939                         nested_vmx_failValid(vcpu,
5940                                              VMXERR_VMCLEAR_INVALID_ADDRESS);
5941                         skip_emulated_instruction(vcpu);
5942                         return 1;
5943                 }
5944
5945                 if (vmptr == vmx->nested.vmxon_ptr) {
5946                         nested_vmx_failValid(vcpu,
5947                                              VMXERR_VMCLEAR_VMXON_POINTER);
5948                         skip_emulated_instruction(vcpu);
5949                         return 1;
5950                 }
5951                 break;
5952         case EXIT_REASON_VMPTRLD:
5953                 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
5954                         nested_vmx_failValid(vcpu,
5955                                              VMXERR_VMPTRLD_INVALID_ADDRESS);
5956                         skip_emulated_instruction(vcpu);
5957                         return 1;
5958                 }
5959
5960                 if (vmptr == vmx->nested.vmxon_ptr) {
5961                         nested_vmx_failValid(vcpu,
5962                                              VMXERR_VMCLEAR_VMXON_POINTER);
5963                         skip_emulated_instruction(vcpu);
5964                         return 1;
5965                 }
5966                 break;
5967         default:
5968                 return 1; /* shouldn't happen */
5969         }
5970
5971         if (vmpointer)
5972                 *vmpointer = vmptr;
5973         return 0;
5974 }
5975
5976 /*
5977  * Emulate the VMXON instruction.
5978  * Currently, we just remember that VMX is active, and do not save or even
5979  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5980  * do not currently need to store anything in that guest-allocated memory
5981  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5982  * argument is different from the VMXON pointer (which the spec says they do).
5983  */
5984 static int handle_vmon(struct kvm_vcpu *vcpu)
5985 {
5986         struct kvm_segment cs;
5987         struct vcpu_vmx *vmx = to_vmx(vcpu);
5988         struct vmcs *shadow_vmcs;
5989         const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
5990                 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
5991
5992         /* The Intel VMX Instruction Reference lists a bunch of bits that
5993          * are prerequisite to running VMXON, most notably cr4.VMXE must be
5994          * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5995          * Otherwise, we should fail with #UD. We test these now:
5996          */
5997         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5998             !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5999             (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6000                 kvm_queue_exception(vcpu, UD_VECTOR);
6001                 return 1;
6002         }
6003
6004         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6005         if (is_long_mode(vcpu) && !cs.l) {
6006                 kvm_queue_exception(vcpu, UD_VECTOR);
6007                 return 1;
6008         }
6009
6010         if (vmx_get_cpl(vcpu)) {
6011                 kvm_inject_gp(vcpu, 0);
6012                 return 1;
6013         }
6014
6015         if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6016                 return 1;
6017
6018         if (vmx->nested.vmxon) {
6019                 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6020                 skip_emulated_instruction(vcpu);
6021                 return 1;
6022         }
6023
6024         if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6025                         != VMXON_NEEDED_FEATURES) {
6026                 kvm_inject_gp(vcpu, 0);
6027                 return 1;
6028         }
6029
6030         if (enable_shadow_vmcs) {
6031                 shadow_vmcs = alloc_vmcs();
6032                 if (!shadow_vmcs)
6033                         return -ENOMEM;
6034                 /* mark vmcs as shadow */
6035                 shadow_vmcs->revision_id |= (1u << 31);
6036                 /* init shadow vmcs */
6037                 vmcs_clear(shadow_vmcs);
6038                 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6039         }
6040
6041         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6042         vmx->nested.vmcs02_num = 0;
6043
6044         hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6045                      HRTIMER_MODE_REL);
6046         vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6047
6048         vmx->nested.vmxon = true;
6049
6050         skip_emulated_instruction(vcpu);
6051         nested_vmx_succeed(vcpu);
6052         return 1;
6053 }
6054
6055 /*
6056  * Intel's VMX Instruction Reference specifies a common set of prerequisites
6057  * for running VMX instructions (except VMXON, whose prerequisites are
6058  * slightly different). It also specifies what exception to inject otherwise.
6059  */
6060 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6061 {
6062         struct kvm_segment cs;
6063         struct vcpu_vmx *vmx = to_vmx(vcpu);
6064
6065         if (!vmx->nested.vmxon) {
6066                 kvm_queue_exception(vcpu, UD_VECTOR);
6067                 return 0;
6068         }
6069
6070         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6071         if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6072             (is_long_mode(vcpu) && !cs.l)) {
6073                 kvm_queue_exception(vcpu, UD_VECTOR);
6074                 return 0;
6075         }
6076
6077         if (vmx_get_cpl(vcpu)) {
6078                 kvm_inject_gp(vcpu, 0);
6079                 return 0;
6080         }
6081
6082         return 1;
6083 }
6084
6085 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6086 {
6087         u32 exec_control;
6088         if (enable_shadow_vmcs) {
6089                 if (vmx->nested.current_vmcs12 != NULL) {
6090                         /* copy to memory all shadowed fields in case
6091                            they were modified */
6092                         copy_shadow_to_vmcs12(vmx);
6093                         vmx->nested.sync_shadow_vmcs = false;
6094                         exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6095                         exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
6096                         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6097                         vmcs_write64(VMCS_LINK_POINTER, -1ull);
6098                 }
6099         }
6100         kunmap(vmx->nested.current_vmcs12_page);
6101         nested_release_page(vmx->nested.current_vmcs12_page);
6102 }
6103
6104 /*
6105  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6106  * just stops using VMX.
6107  */
6108 static void free_nested(struct vcpu_vmx *vmx)
6109 {
6110         if (!vmx->nested.vmxon)
6111                 return;
6112         vmx->nested.vmxon = false;
6113         if (vmx->nested.current_vmptr != -1ull) {
6114                 nested_release_vmcs12(vmx);
6115                 vmx->nested.current_vmptr = -1ull;
6116                 vmx->nested.current_vmcs12 = NULL;
6117         }
6118         if (enable_shadow_vmcs)
6119                 free_vmcs(vmx->nested.current_shadow_vmcs);
6120         /* Unpin physical memory we referred to in current vmcs02 */
6121         if (vmx->nested.apic_access_page) {
6122                 nested_release_page(vmx->nested.apic_access_page);
6123                 vmx->nested.apic_access_page = 0;
6124         }
6125
6126         nested_free_all_saved_vmcss(vmx);
6127 }
6128
6129 /* Emulate the VMXOFF instruction */
6130 static int handle_vmoff(struct kvm_vcpu *vcpu)
6131 {
6132         if (!nested_vmx_check_permission(vcpu))
6133                 return 1;
6134         free_nested(to_vmx(vcpu));
6135         skip_emulated_instruction(vcpu);
6136         nested_vmx_succeed(vcpu);
6137         return 1;
6138 }
6139
6140 /* Emulate the VMCLEAR instruction */
6141 static int handle_vmclear(struct kvm_vcpu *vcpu)
6142 {
6143         struct vcpu_vmx *vmx = to_vmx(vcpu);
6144         gpa_t vmptr;
6145         struct vmcs12 *vmcs12;
6146         struct page *page;
6147
6148         if (!nested_vmx_check_permission(vcpu))
6149                 return 1;
6150
6151         if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6152                 return 1;
6153
6154         if (vmptr == vmx->nested.current_vmptr) {
6155                 nested_release_vmcs12(vmx);
6156                 vmx->nested.current_vmptr = -1ull;
6157                 vmx->nested.current_vmcs12 = NULL;
6158         }
6159
6160         page = nested_get_page(vcpu, vmptr);
6161         if (page == NULL) {
6162                 /*
6163                  * For accurate processor emulation, VMCLEAR beyond available
6164                  * physical memory should do nothing at all. However, it is
6165                  * possible that a nested vmx bug, not a guest hypervisor bug,
6166                  * resulted in this case, so let's shut down before doing any
6167                  * more damage:
6168                  */
6169                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6170                 return 1;
6171         }
6172         vmcs12 = kmap(page);
6173         vmcs12->launch_state = 0;
6174         kunmap(page);
6175         nested_release_page(page);
6176
6177         nested_free_vmcs02(vmx, vmptr);
6178
6179         skip_emulated_instruction(vcpu);
6180         nested_vmx_succeed(vcpu);
6181         return 1;
6182 }
6183
6184 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6185
6186 /* Emulate the VMLAUNCH instruction */
6187 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6188 {
6189         return nested_vmx_run(vcpu, true);
6190 }
6191
6192 /* Emulate the VMRESUME instruction */
6193 static int handle_vmresume(struct kvm_vcpu *vcpu)
6194 {
6195
6196         return nested_vmx_run(vcpu, false);
6197 }
6198
6199 enum vmcs_field_type {
6200         VMCS_FIELD_TYPE_U16 = 0,
6201         VMCS_FIELD_TYPE_U64 = 1,
6202         VMCS_FIELD_TYPE_U32 = 2,
6203         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6204 };
6205
6206 static inline int vmcs_field_type(unsigned long field)
6207 {
6208         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
6209                 return VMCS_FIELD_TYPE_U32;
6210         return (field >> 13) & 0x3 ;
6211 }
6212
6213 static inline int vmcs_field_readonly(unsigned long field)
6214 {
6215         return (((field >> 10) & 0x3) == 1);
6216 }
6217
6218 /*
6219  * Read a vmcs12 field. Since these can have varying lengths and we return
6220  * one type, we chose the biggest type (u64) and zero-extend the return value
6221  * to that size. Note that the caller, handle_vmread, might need to use only
6222  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
6223  * 64-bit fields are to be returned).
6224  */
6225 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
6226                                         unsigned long field, u64 *ret)
6227 {
6228         short offset = vmcs_field_to_offset(field);
6229         char *p;
6230
6231         if (offset < 0)
6232                 return 0;
6233
6234         p = ((char *)(get_vmcs12(vcpu))) + offset;
6235
6236         switch (vmcs_field_type(field)) {
6237         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6238                 *ret = *((natural_width *)p);
6239                 return 1;
6240         case VMCS_FIELD_TYPE_U16:
6241                 *ret = *((u16 *)p);
6242                 return 1;
6243         case VMCS_FIELD_TYPE_U32:
6244                 *ret = *((u32 *)p);
6245                 return 1;
6246         case VMCS_FIELD_TYPE_U64:
6247                 *ret = *((u64 *)p);
6248                 return 1;
6249         default:
6250                 return 0; /* can never happen. */
6251         }
6252 }
6253
6254
6255 static inline bool vmcs12_write_any(struct kvm_vcpu *vcpu,
6256                                     unsigned long field, u64 field_value){
6257         short offset = vmcs_field_to_offset(field);
6258         char *p = ((char *) get_vmcs12(vcpu)) + offset;
6259         if (offset < 0)
6260                 return false;
6261
6262         switch (vmcs_field_type(field)) {
6263         case VMCS_FIELD_TYPE_U16:
6264                 *(u16 *)p = field_value;
6265                 return true;
6266         case VMCS_FIELD_TYPE_U32:
6267                 *(u32 *)p = field_value;
6268                 return true;
6269         case VMCS_FIELD_TYPE_U64:
6270                 *(u64 *)p = field_value;
6271                 return true;
6272         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6273                 *(natural_width *)p = field_value;
6274                 return true;
6275         default:
6276                 return false; /* can never happen. */
6277         }
6278
6279 }
6280
6281 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6282 {
6283         int i;
6284         unsigned long field;
6285         u64 field_value;
6286         struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6287         const unsigned long *fields = shadow_read_write_fields;
6288         const int num_fields = max_shadow_read_write_fields;
6289
6290         vmcs_load(shadow_vmcs);
6291
6292         for (i = 0; i < num_fields; i++) {
6293                 field = fields[i];
6294                 switch (vmcs_field_type(field)) {
6295                 case VMCS_FIELD_TYPE_U16:
6296                         field_value = vmcs_read16(field);
6297                         break;
6298                 case VMCS_FIELD_TYPE_U32:
6299                         field_value = vmcs_read32(field);
6300                         break;
6301                 case VMCS_FIELD_TYPE_U64:
6302                         field_value = vmcs_read64(field);
6303                         break;
6304                 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6305                         field_value = vmcs_readl(field);
6306                         break;
6307                 }
6308                 vmcs12_write_any(&vmx->vcpu, field, field_value);
6309         }
6310
6311         vmcs_clear(shadow_vmcs);
6312         vmcs_load(vmx->loaded_vmcs->vmcs);
6313 }
6314
6315 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
6316 {
6317         const unsigned long *fields[] = {
6318                 shadow_read_write_fields,
6319                 shadow_read_only_fields
6320         };
6321         const int max_fields[] = {
6322                 max_shadow_read_write_fields,
6323                 max_shadow_read_only_fields
6324         };
6325         int i, q;
6326         unsigned long field;
6327         u64 field_value = 0;
6328         struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6329
6330         vmcs_load(shadow_vmcs);
6331
6332         for (q = 0; q < ARRAY_SIZE(fields); q++) {
6333                 for (i = 0; i < max_fields[q]; i++) {
6334                         field = fields[q][i];
6335                         vmcs12_read_any(&vmx->vcpu, field, &field_value);
6336
6337                         switch (vmcs_field_type(field)) {
6338                         case VMCS_FIELD_TYPE_U16:
6339                                 vmcs_write16(field, (u16)field_value);
6340                                 break;
6341                         case VMCS_FIELD_TYPE_U32:
6342                                 vmcs_write32(field, (u32)field_value);
6343                                 break;
6344                         case VMCS_FIELD_TYPE_U64:
6345                                 vmcs_write64(field, (u64)field_value);
6346                                 break;
6347                         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6348                                 vmcs_writel(field, (long)field_value);
6349                                 break;
6350                         }
6351                 }
6352         }
6353
6354         vmcs_clear(shadow_vmcs);
6355         vmcs_load(vmx->loaded_vmcs->vmcs);
6356 }
6357
6358 /*
6359  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
6360  * used before) all generate the same failure when it is missing.
6361  */
6362 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
6363 {
6364         struct vcpu_vmx *vmx = to_vmx(vcpu);
6365         if (vmx->nested.current_vmptr == -1ull) {
6366                 nested_vmx_failInvalid(vcpu);
6367                 skip_emulated_instruction(vcpu);
6368                 return 0;
6369         }
6370         return 1;
6371 }
6372
6373 static int handle_vmread(struct kvm_vcpu *vcpu)
6374 {
6375         unsigned long field;
6376         u64 field_value;
6377         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6378         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6379         gva_t gva = 0;
6380
6381         if (!nested_vmx_check_permission(vcpu) ||
6382             !nested_vmx_check_vmcs12(vcpu))
6383                 return 1;
6384
6385         /* Decode instruction info and find the field to read */
6386         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6387         /* Read the field, zero-extended to a u64 field_value */
6388         if (!vmcs12_read_any(vcpu, field, &field_value)) {
6389                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6390                 skip_emulated_instruction(vcpu);
6391                 return 1;
6392         }
6393         /*
6394          * Now copy part of this value to register or memory, as requested.
6395          * Note that the number of bits actually copied is 32 or 64 depending
6396          * on the guest's mode (32 or 64 bit), not on the given field's length.
6397          */
6398         if (vmx_instruction_info & (1u << 10)) {
6399                 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
6400                         field_value);
6401         } else {
6402                 if (get_vmx_mem_address(vcpu, exit_qualification,
6403                                 vmx_instruction_info, &gva))
6404                         return 1;
6405                 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
6406                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
6407                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
6408         }
6409
6410         nested_vmx_succeed(vcpu);
6411         skip_emulated_instruction(vcpu);
6412         return 1;
6413 }
6414
6415
6416 static int handle_vmwrite(struct kvm_vcpu *vcpu)
6417 {
6418         unsigned long field;
6419         gva_t gva;
6420         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6421         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6422         /* The value to write might be 32 or 64 bits, depending on L1's long
6423          * mode, and eventually we need to write that into a field of several
6424          * possible lengths. The code below first zero-extends the value to 64
6425          * bit (field_value), and then copies only the approriate number of
6426          * bits into the vmcs12 field.
6427          */
6428         u64 field_value = 0;
6429         struct x86_exception e;
6430
6431         if (!nested_vmx_check_permission(vcpu) ||
6432             !nested_vmx_check_vmcs12(vcpu))
6433                 return 1;
6434
6435         if (vmx_instruction_info & (1u << 10))
6436                 field_value = kvm_register_read(vcpu,
6437                         (((vmx_instruction_info) >> 3) & 0xf));
6438         else {
6439                 if (get_vmx_mem_address(vcpu, exit_qualification,
6440                                 vmx_instruction_info, &gva))
6441                         return 1;
6442                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
6443                            &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
6444                         kvm_inject_page_fault(vcpu, &e);
6445                         return 1;
6446                 }
6447         }
6448
6449
6450         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6451         if (vmcs_field_readonly(field)) {
6452                 nested_vmx_failValid(vcpu,
6453                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
6454                 skip_emulated_instruction(vcpu);
6455                 return 1;
6456         }
6457
6458         if (!vmcs12_write_any(vcpu, field, field_value)) {
6459                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6460                 skip_emulated_instruction(vcpu);
6461                 return 1;
6462         }
6463
6464         nested_vmx_succeed(vcpu);
6465         skip_emulated_instruction(vcpu);
6466         return 1;
6467 }
6468
6469 /* Emulate the VMPTRLD instruction */
6470 static int handle_vmptrld(struct kvm_vcpu *vcpu)
6471 {
6472         struct vcpu_vmx *vmx = to_vmx(vcpu);
6473         gpa_t vmptr;
6474         u32 exec_control;
6475
6476         if (!nested_vmx_check_permission(vcpu))
6477                 return 1;
6478
6479         if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
6480                 return 1;
6481
6482         if (vmx->nested.current_vmptr != vmptr) {
6483                 struct vmcs12 *new_vmcs12;
6484                 struct page *page;
6485                 page = nested_get_page(vcpu, vmptr);
6486                 if (page == NULL) {
6487                         nested_vmx_failInvalid(vcpu);
6488                         skip_emulated_instruction(vcpu);
6489                         return 1;
6490                 }
6491                 new_vmcs12 = kmap(page);
6492                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
6493                         kunmap(page);
6494                         nested_release_page_clean(page);
6495                         nested_vmx_failValid(vcpu,
6496                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
6497                         skip_emulated_instruction(vcpu);
6498                         return 1;
6499                 }
6500                 if (vmx->nested.current_vmptr != -1ull)
6501                         nested_release_vmcs12(vmx);
6502
6503                 vmx->nested.current_vmptr = vmptr;
6504                 vmx->nested.current_vmcs12 = new_vmcs12;
6505                 vmx->nested.current_vmcs12_page = page;
6506                 if (enable_shadow_vmcs) {
6507                         exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6508                         exec_control |= SECONDARY_EXEC_SHADOW_VMCS;
6509                         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6510                         vmcs_write64(VMCS_LINK_POINTER,
6511                                      __pa(vmx->nested.current_shadow_vmcs));
6512                         vmx->nested.sync_shadow_vmcs = true;
6513                 }
6514         }
6515
6516         nested_vmx_succeed(vcpu);
6517         skip_emulated_instruction(vcpu);
6518         return 1;
6519 }
6520
6521 /* Emulate the VMPTRST instruction */
6522 static int handle_vmptrst(struct kvm_vcpu *vcpu)
6523 {
6524         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6525         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6526         gva_t vmcs_gva;
6527         struct x86_exception e;
6528
6529         if (!nested_vmx_check_permission(vcpu))
6530                 return 1;
6531
6532         if (get_vmx_mem_address(vcpu, exit_qualification,
6533                         vmx_instruction_info, &vmcs_gva))
6534                 return 1;
6535         /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
6536         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
6537                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
6538                                  sizeof(u64), &e)) {
6539                 kvm_inject_page_fault(vcpu, &e);
6540                 return 1;
6541         }
6542         nested_vmx_succeed(vcpu);
6543         skip_emulated_instruction(vcpu);
6544         return 1;
6545 }
6546
6547 /* Emulate the INVEPT instruction */
6548 static int handle_invept(struct kvm_vcpu *vcpu)
6549 {
6550         u32 vmx_instruction_info, types;
6551         unsigned long type;
6552         gva_t gva;
6553         struct x86_exception e;
6554         struct {
6555                 u64 eptp, gpa;
6556         } operand;
6557
6558         if (!(nested_vmx_secondary_ctls_high & SECONDARY_EXEC_ENABLE_EPT) ||
6559             !(nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
6560                 kvm_queue_exception(vcpu, UD_VECTOR);
6561                 return 1;
6562         }
6563
6564         if (!nested_vmx_check_permission(vcpu))
6565                 return 1;
6566
6567         if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
6568                 kvm_queue_exception(vcpu, UD_VECTOR);
6569                 return 1;
6570         }
6571
6572         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6573         type = kvm_register_read(vcpu, (vmx_instruction_info >> 28) & 0xf);
6574
6575         types = (nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
6576
6577         if (!(types & (1UL << type))) {
6578                 nested_vmx_failValid(vcpu,
6579                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6580                 return 1;
6581         }
6582
6583         /* According to the Intel VMX instruction reference, the memory
6584          * operand is read even if it isn't needed (e.g., for type==global)
6585          */
6586         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6587                         vmx_instruction_info, &gva))
6588                 return 1;
6589         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
6590                                 sizeof(operand), &e)) {
6591                 kvm_inject_page_fault(vcpu, &e);
6592                 return 1;
6593         }
6594
6595         switch (type) {
6596         case VMX_EPT_EXTENT_GLOBAL:
6597                 kvm_mmu_sync_roots(vcpu);
6598                 kvm_mmu_flush_tlb(vcpu);
6599                 nested_vmx_succeed(vcpu);
6600                 break;
6601         default:
6602                 /* Trap single context invalidation invept calls */
6603                 BUG_ON(1);
6604                 break;
6605         }
6606
6607         skip_emulated_instruction(vcpu);
6608         return 1;
6609 }
6610
6611 /*
6612  * The exit handlers return 1 if the exit was handled fully and guest execution
6613  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
6614  * to be done to userspace and return 0.
6615  */
6616 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6617         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
6618         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
6619         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
6620         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
6621         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
6622         [EXIT_REASON_CR_ACCESS]               = handle_cr,
6623         [EXIT_REASON_DR_ACCESS]               = handle_dr,
6624         [EXIT_REASON_CPUID]                   = handle_cpuid,
6625         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
6626         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
6627         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
6628         [EXIT_REASON_HLT]                     = handle_halt,
6629         [EXIT_REASON_INVD]                    = handle_invd,
6630         [EXIT_REASON_INVLPG]                  = handle_invlpg,
6631         [EXIT_REASON_RDPMC]                   = handle_rdpmc,
6632         [EXIT_REASON_VMCALL]                  = handle_vmcall,
6633         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
6634         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
6635         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
6636         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
6637         [EXIT_REASON_VMREAD]                  = handle_vmread,
6638         [EXIT_REASON_VMRESUME]                = handle_vmresume,
6639         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
6640         [EXIT_REASON_VMOFF]                   = handle_vmoff,
6641         [EXIT_REASON_VMON]                    = handle_vmon,
6642         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
6643         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
6644         [EXIT_REASON_APIC_WRITE]              = handle_apic_write,
6645         [EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
6646         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
6647         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
6648         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
6649         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
6650         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
6651         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
6652         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
6653         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_mwait,
6654         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
6655         [EXIT_REASON_INVEPT]                  = handle_invept,
6656 };
6657
6658 static const int kvm_vmx_max_exit_handlers =
6659         ARRAY_SIZE(kvm_vmx_exit_handlers);
6660
6661 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6662                                        struct vmcs12 *vmcs12)
6663 {
6664         unsigned long exit_qualification;
6665         gpa_t bitmap, last_bitmap;
6666         unsigned int port;
6667         int size;
6668         u8 b;
6669
6670         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6671                 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
6672
6673         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6674
6675         port = exit_qualification >> 16;
6676         size = (exit_qualification & 7) + 1;
6677
6678         last_bitmap = (gpa_t)-1;
6679         b = -1;
6680
6681         while (size > 0) {
6682                 if (port < 0x8000)
6683                         bitmap = vmcs12->io_bitmap_a;
6684                 else if (port < 0x10000)
6685                         bitmap = vmcs12->io_bitmap_b;
6686                 else
6687                         return 1;
6688                 bitmap += (port & 0x7fff) / 8;
6689
6690                 if (last_bitmap != bitmap)
6691                         if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
6692                                 return 1;
6693                 if (b & (1 << (port & 7)))
6694                         return 1;
6695
6696                 port++;
6697                 size--;
6698                 last_bitmap = bitmap;
6699         }
6700
6701         return 0;
6702 }
6703
6704 /*
6705  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
6706  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6707  * disinterest in the current event (read or write a specific MSR) by using an
6708  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6709  */
6710 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6711         struct vmcs12 *vmcs12, u32 exit_reason)
6712 {
6713         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
6714         gpa_t bitmap;
6715
6716         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6717                 return 1;
6718
6719         /*
6720          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6721          * for the four combinations of read/write and low/high MSR numbers.
6722          * First we need to figure out which of the four to use:
6723          */
6724         bitmap = vmcs12->msr_bitmap;
6725         if (exit_reason == EXIT_REASON_MSR_WRITE)
6726                 bitmap += 2048;
6727         if (msr_index >= 0xc0000000) {
6728                 msr_index -= 0xc0000000;
6729                 bitmap += 1024;
6730         }
6731
6732         /* Then read the msr_index'th bit from this bitmap: */
6733         if (msr_index < 1024*8) {
6734                 unsigned char b;
6735                 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
6736                         return 1;
6737                 return 1 & (b >> (msr_index & 7));
6738         } else
6739                 return 1; /* let L1 handle the wrong parameter */
6740 }
6741
6742 /*
6743  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6744  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6745  * intercept (via guest_host_mask etc.) the current event.
6746  */
6747 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6748         struct vmcs12 *vmcs12)
6749 {
6750         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6751         int cr = exit_qualification & 15;
6752         int reg = (exit_qualification >> 8) & 15;
6753         unsigned long val = kvm_register_read(vcpu, reg);
6754
6755         switch ((exit_qualification >> 4) & 3) {
6756         case 0: /* mov to cr */
6757                 switch (cr) {
6758                 case 0:
6759                         if (vmcs12->cr0_guest_host_mask &
6760                             (val ^ vmcs12->cr0_read_shadow))
6761                                 return 1;
6762                         break;
6763                 case 3:
6764                         if ((vmcs12->cr3_target_count >= 1 &&
6765                                         vmcs12->cr3_target_value0 == val) ||
6766                                 (vmcs12->cr3_target_count >= 2 &&
6767                                         vmcs12->cr3_target_value1 == val) ||
6768                                 (vmcs12->cr3_target_count >= 3 &&
6769                                         vmcs12->cr3_target_value2 == val) ||
6770                                 (vmcs12->cr3_target_count >= 4 &&
6771                                         vmcs12->cr3_target_value3 == val))
6772                                 return 0;
6773                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6774                                 return 1;
6775                         break;
6776                 case 4:
6777                         if (vmcs12->cr4_guest_host_mask &
6778                             (vmcs12->cr4_read_shadow ^ val))
6779                                 return 1;
6780                         break;
6781                 case 8:
6782                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6783                                 return 1;
6784                         break;
6785                 }
6786                 break;
6787         case 2: /* clts */
6788                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6789                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
6790                         return 1;
6791                 break;
6792         case 1: /* mov from cr */
6793                 switch (cr) {
6794                 case 3:
6795                         if (vmcs12->cpu_based_vm_exec_control &
6796                             CPU_BASED_CR3_STORE_EXITING)
6797                                 return 1;
6798                         break;
6799                 case 8:
6800                         if (vmcs12->cpu_based_vm_exec_control &
6801                             CPU_BASED_CR8_STORE_EXITING)
6802                                 return 1;
6803                         break;
6804                 }
6805                 break;
6806         case 3: /* lmsw */
6807                 /*
6808                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6809                  * cr0. Other attempted changes are ignored, with no exit.
6810                  */
6811                 if (vmcs12->cr0_guest_host_mask & 0xe &
6812                     (val ^ vmcs12->cr0_read_shadow))
6813                         return 1;
6814                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6815                     !(vmcs12->cr0_read_shadow & 0x1) &&
6816                     (val & 0x1))
6817                         return 1;
6818                 break;
6819         }
6820         return 0;
6821 }
6822
6823 /*
6824  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
6825  * should handle it ourselves in L0 (and then continue L2). Only call this
6826  * when in is_guest_mode (L2).
6827  */
6828 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
6829 {
6830         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6831         struct vcpu_vmx *vmx = to_vmx(vcpu);
6832         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6833         u32 exit_reason = vmx->exit_reason;
6834
6835         trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
6836                                 vmcs_readl(EXIT_QUALIFICATION),
6837                                 vmx->idt_vectoring_info,
6838                                 intr_info,
6839                                 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
6840                                 KVM_ISA_VMX);
6841
6842         if (vmx->nested.nested_run_pending)
6843                 return 0;
6844
6845         if (unlikely(vmx->fail)) {
6846                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
6847                                     vmcs_read32(VM_INSTRUCTION_ERROR));
6848                 return 1;
6849         }
6850
6851         switch (exit_reason) {
6852         case EXIT_REASON_EXCEPTION_NMI:
6853                 if (!is_exception(intr_info))
6854                         return 0;
6855                 else if (is_page_fault(intr_info))
6856                         return enable_ept;
6857                 else if (is_no_device(intr_info) &&
6858                          !(vmcs12->guest_cr0 & X86_CR0_TS))
6859                         return 0;
6860                 return vmcs12->exception_bitmap &
6861                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
6862         case EXIT_REASON_EXTERNAL_INTERRUPT:
6863                 return 0;
6864         case EXIT_REASON_TRIPLE_FAULT:
6865                 return 1;
6866         case EXIT_REASON_PENDING_INTERRUPT:
6867                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
6868         case EXIT_REASON_NMI_WINDOW:
6869                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
6870         case EXIT_REASON_TASK_SWITCH:
6871                 return 1;
6872         case EXIT_REASON_CPUID:
6873                 return 1;
6874         case EXIT_REASON_HLT:
6875                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6876         case EXIT_REASON_INVD:
6877                 return 1;
6878         case EXIT_REASON_INVLPG:
6879                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6880         case EXIT_REASON_RDPMC:
6881                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6882         case EXIT_REASON_RDTSC:
6883                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6884         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6885         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6886         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
6887         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
6888         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6889         case EXIT_REASON_INVEPT:
6890                 /*
6891                  * VMX instructions trap unconditionally. This allows L1 to
6892                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
6893                  */
6894                 return 1;
6895         case EXIT_REASON_CR_ACCESS:
6896                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6897         case EXIT_REASON_DR_ACCESS:
6898                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6899         case EXIT_REASON_IO_INSTRUCTION:
6900                 return nested_vmx_exit_handled_io(vcpu, vmcs12);
6901         case EXIT_REASON_MSR_READ:
6902         case EXIT_REASON_MSR_WRITE:
6903                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6904         case EXIT_REASON_INVALID_STATE:
6905                 return 1;
6906         case EXIT_REASON_MWAIT_INSTRUCTION:
6907                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6908         case EXIT_REASON_MONITOR_INSTRUCTION:
6909                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6910         case EXIT_REASON_PAUSE_INSTRUCTION:
6911                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6912                         nested_cpu_has2(vmcs12,
6913                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6914         case EXIT_REASON_MCE_DURING_VMENTRY:
6915                 return 0;
6916         case EXIT_REASON_TPR_BELOW_THRESHOLD:
6917                 return 1;
6918         case EXIT_REASON_APIC_ACCESS:
6919                 return nested_cpu_has2(vmcs12,
6920                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
6921         case EXIT_REASON_EPT_VIOLATION:
6922                 /*
6923                  * L0 always deals with the EPT violation. If nested EPT is
6924                  * used, and the nested mmu code discovers that the address is
6925                  * missing in the guest EPT table (EPT12), the EPT violation
6926                  * will be injected with nested_ept_inject_page_fault()
6927                  */
6928                 return 0;
6929         case EXIT_REASON_EPT_MISCONFIG:
6930                 /*
6931                  * L2 never uses directly L1's EPT, but rather L0's own EPT
6932                  * table (shadow on EPT) or a merged EPT table that L0 built
6933                  * (EPT on EPT). So any problems with the structure of the
6934                  * table is L0's fault.
6935                  */
6936                 return 0;
6937         case EXIT_REASON_WBINVD:
6938                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6939         case EXIT_REASON_XSETBV:
6940                 return 1;
6941         default:
6942                 return 1;
6943         }
6944 }
6945
6946 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
6947 {
6948         *info1 = vmcs_readl(EXIT_QUALIFICATION);
6949         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
6950 }
6951
6952 /*
6953  * The guest has exited.  See if we can fix it or if we need userspace
6954  * assistance.
6955  */
6956 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
6957 {
6958         struct vcpu_vmx *vmx = to_vmx(vcpu);
6959         u32 exit_reason = vmx->exit_reason;
6960         u32 vectoring_info = vmx->idt_vectoring_info;
6961
6962         /* If guest state is invalid, start emulating */
6963         if (vmx->emulation_required)
6964                 return handle_invalid_guest_state(vcpu);
6965
6966         if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
6967                 nested_vmx_vmexit(vcpu, exit_reason,
6968                                   vmcs_read32(VM_EXIT_INTR_INFO),
6969                                   vmcs_readl(EXIT_QUALIFICATION));
6970                 return 1;
6971         }
6972
6973         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
6974                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6975                 vcpu->run->fail_entry.hardware_entry_failure_reason
6976                         = exit_reason;
6977                 return 0;
6978         }
6979
6980         if (unlikely(vmx->fail)) {
6981                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6982                 vcpu->run->fail_entry.hardware_entry_failure_reason
6983                         = vmcs_read32(VM_INSTRUCTION_ERROR);
6984                 return 0;
6985         }
6986
6987         /*
6988          * Note:
6989          * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6990          * delivery event since it indicates guest is accessing MMIO.
6991          * The vm-exit can be triggered again after return to guest that
6992          * will cause infinite loop.
6993          */
6994         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6995                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
6996                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
6997                         exit_reason != EXIT_REASON_TASK_SWITCH)) {
6998                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6999                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
7000                 vcpu->run->internal.ndata = 2;
7001                 vcpu->run->internal.data[0] = vectoring_info;
7002                 vcpu->run->internal.data[1] = exit_reason;
7003                 return 0;
7004         }
7005
7006         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
7007             !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
7008                                         get_vmcs12(vcpu))))) {
7009                 if (vmx_interrupt_allowed(vcpu)) {
7010                         vmx->soft_vnmi_blocked = 0;
7011                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
7012                            vcpu->arch.nmi_pending) {
7013                         /*
7014                          * This CPU don't support us in finding the end of an
7015                          * NMI-blocked window if the guest runs with IRQs
7016                          * disabled. So we pull the trigger after 1 s of
7017                          * futile waiting, but inform the user about this.
7018                          */
7019                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
7020                                "state on VCPU %d after 1 s timeout\n",
7021                                __func__, vcpu->vcpu_id);
7022                         vmx->soft_vnmi_blocked = 0;
7023                 }
7024         }
7025
7026         if (exit_reason < kvm_vmx_max_exit_handlers
7027             && kvm_vmx_exit_handlers[exit_reason])
7028                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
7029         else {
7030                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
7031                 vcpu->run->hw.hardware_exit_reason = exit_reason;
7032         }
7033         return 0;
7034 }
7035
7036 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
7037 {
7038         if (irr == -1 || tpr < irr) {
7039                 vmcs_write32(TPR_THRESHOLD, 0);
7040                 return;
7041         }
7042
7043         vmcs_write32(TPR_THRESHOLD, irr);
7044 }
7045
7046 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
7047 {
7048         u32 sec_exec_control;
7049
7050         /*
7051          * There is not point to enable virtualize x2apic without enable
7052          * apicv
7053          */
7054         if (!cpu_has_vmx_virtualize_x2apic_mode() ||
7055                                 !vmx_vm_has_apicv(vcpu->kvm))
7056                 return;
7057
7058         if (!vm_need_tpr_shadow(vcpu->kvm))
7059                 return;
7060
7061         sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7062
7063         if (set) {
7064                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7065                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7066         } else {
7067                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7068                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7069         }
7070         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
7071
7072         vmx_set_msr_bitmap(vcpu);
7073 }
7074
7075 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
7076 {
7077         u16 status;
7078         u8 old;
7079
7080         if (!vmx_vm_has_apicv(kvm))
7081                 return;
7082
7083         if (isr == -1)
7084                 isr = 0;
7085
7086         status = vmcs_read16(GUEST_INTR_STATUS);
7087         old = status >> 8;
7088         if (isr != old) {
7089                 status &= 0xff;
7090                 status |= isr << 8;
7091                 vmcs_write16(GUEST_INTR_STATUS, status);
7092         }
7093 }
7094
7095 static void vmx_set_rvi(int vector)
7096 {
7097         u16 status;
7098         u8 old;
7099
7100         status = vmcs_read16(GUEST_INTR_STATUS);
7101         old = (u8)status & 0xff;
7102         if ((u8)vector != old) {
7103                 status &= ~0xff;
7104                 status |= (u8)vector;
7105                 vmcs_write16(GUEST_INTR_STATUS, status);
7106         }
7107 }
7108
7109 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
7110 {
7111         if (max_irr == -1)
7112                 return;
7113
7114         vmx_set_rvi(max_irr);
7115 }
7116
7117 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
7118 {
7119         if (!vmx_vm_has_apicv(vcpu->kvm))
7120                 return;
7121
7122         vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
7123         vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
7124         vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
7125         vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
7126 }
7127
7128 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
7129 {
7130         u32 exit_intr_info;
7131
7132         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
7133               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
7134                 return;
7135
7136         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7137         exit_intr_info = vmx->exit_intr_info;
7138
7139         /* Handle machine checks before interrupts are enabled */
7140         if (is_machine_check(exit_intr_info))
7141                 kvm_machine_check();
7142
7143         /* We need to handle NMIs before interrupts are enabled */
7144         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
7145             (exit_intr_info & INTR_INFO_VALID_MASK)) {
7146                 kvm_before_handle_nmi(&vmx->vcpu);
7147                 asm("int $2");
7148                 kvm_after_handle_nmi(&vmx->vcpu);
7149         }
7150 }
7151
7152 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
7153 {
7154         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7155
7156         /*
7157          * If external interrupt exists, IF bit is set in rflags/eflags on the
7158          * interrupt stack frame, and interrupt will be enabled on a return
7159          * from interrupt handler.
7160          */
7161         if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
7162                         == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
7163                 unsigned int vector;
7164                 unsigned long entry;
7165                 gate_desc *desc;
7166                 struct vcpu_vmx *vmx = to_vmx(vcpu);
7167 #ifdef CONFIG_X86_64
7168                 unsigned long tmp;
7169 #endif
7170
7171                 vector =  exit_intr_info & INTR_INFO_VECTOR_MASK;
7172                 desc = (gate_desc *)vmx->host_idt_base + vector;
7173                 entry = gate_offset(*desc);
7174                 asm volatile(
7175 #ifdef CONFIG_X86_64
7176                         "mov %%" _ASM_SP ", %[sp]\n\t"
7177                         "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
7178                         "push $%c[ss]\n\t"
7179                         "push %[sp]\n\t"
7180 #endif
7181                         "pushf\n\t"
7182                         "orl $0x200, (%%" _ASM_SP ")\n\t"
7183                         __ASM_SIZE(push) " $%c[cs]\n\t"
7184                         "call *%[entry]\n\t"
7185                         :
7186 #ifdef CONFIG_X86_64
7187                         [sp]"=&r"(tmp)
7188 #endif
7189                         :
7190                         [entry]"r"(entry),
7191                         [ss]"i"(__KERNEL_DS),
7192                         [cs]"i"(__KERNEL_CS)
7193                         );
7194         } else
7195                 local_irq_enable();
7196 }
7197
7198 static bool vmx_mpx_supported(void)
7199 {
7200         return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
7201                 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
7202 }
7203
7204 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
7205 {
7206         u32 exit_intr_info;
7207         bool unblock_nmi;
7208         u8 vector;
7209         bool idtv_info_valid;
7210
7211         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7212
7213         if (cpu_has_virtual_nmis()) {
7214                 if (vmx->nmi_known_unmasked)
7215                         return;
7216                 /*
7217                  * Can't use vmx->exit_intr_info since we're not sure what
7218                  * the exit reason is.
7219                  */
7220                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7221                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
7222                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7223                 /*
7224                  * SDM 3: 27.7.1.2 (September 2008)
7225                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
7226                  * a guest IRET fault.
7227                  * SDM 3: 23.2.2 (September 2008)
7228                  * Bit 12 is undefined in any of the following cases:
7229                  *  If the VM exit sets the valid bit in the IDT-vectoring
7230                  *   information field.
7231                  *  If the VM exit is due to a double fault.
7232                  */
7233                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
7234                     vector != DF_VECTOR && !idtv_info_valid)
7235                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7236                                       GUEST_INTR_STATE_NMI);
7237                 else
7238                         vmx->nmi_known_unmasked =
7239                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
7240                                   & GUEST_INTR_STATE_NMI);
7241         } else if (unlikely(vmx->soft_vnmi_blocked))
7242                 vmx->vnmi_blocked_time +=
7243                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
7244 }
7245
7246 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
7247                                       u32 idt_vectoring_info,
7248                                       int instr_len_field,
7249                                       int error_code_field)
7250 {
7251         u8 vector;
7252         int type;
7253         bool idtv_info_valid;
7254
7255         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7256
7257         vcpu->arch.nmi_injected = false;
7258         kvm_clear_exception_queue(vcpu);
7259         kvm_clear_interrupt_queue(vcpu);
7260
7261         if (!idtv_info_valid)
7262                 return;
7263
7264         kvm_make_request(KVM_REQ_EVENT, vcpu);
7265
7266         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
7267         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7268
7269         switch (type) {
7270         case INTR_TYPE_NMI_INTR:
7271                 vcpu->arch.nmi_injected = true;
7272                 /*
7273                  * SDM 3: 27.7.1.2 (September 2008)
7274                  * Clear bit "block by NMI" before VM entry if a NMI
7275                  * delivery faulted.
7276                  */
7277                 vmx_set_nmi_mask(vcpu, false);
7278                 break;
7279         case INTR_TYPE_SOFT_EXCEPTION:
7280                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7281                 /* fall through */
7282         case INTR_TYPE_HARD_EXCEPTION:
7283                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7284                         u32 err = vmcs_read32(error_code_field);
7285                         kvm_requeue_exception_e(vcpu, vector, err);
7286                 } else
7287                         kvm_requeue_exception(vcpu, vector);
7288                 break;
7289         case INTR_TYPE_SOFT_INTR:
7290                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7291                 /* fall through */
7292         case INTR_TYPE_EXT_INTR:
7293                 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7294                 break;
7295         default:
7296                 break;
7297         }
7298 }
7299
7300 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7301 {
7302         __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7303                                   VM_EXIT_INSTRUCTION_LEN,
7304                                   IDT_VECTORING_ERROR_CODE);
7305 }
7306
7307 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7308 {
7309         __vmx_complete_interrupts(vcpu,
7310                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7311                                   VM_ENTRY_INSTRUCTION_LEN,
7312                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
7313
7314         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7315 }
7316
7317 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7318 {
7319         int i, nr_msrs;
7320         struct perf_guest_switch_msr *msrs;
7321
7322         msrs = perf_guest_get_msrs(&nr_msrs);
7323
7324         if (!msrs)
7325                 return;
7326
7327         for (i = 0; i < nr_msrs; i++)
7328                 if (msrs[i].host == msrs[i].guest)
7329                         clear_atomic_switch_msr(vmx, msrs[i].msr);
7330                 else
7331                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7332                                         msrs[i].host);
7333 }
7334
7335 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
7336 {
7337         struct vcpu_vmx *vmx = to_vmx(vcpu);
7338         unsigned long debugctlmsr;
7339
7340         /* Record the guest's net vcpu time for enforced NMI injections. */
7341         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
7342                 vmx->entry_time = ktime_get();
7343
7344         /* Don't enter VMX if guest state is invalid, let the exit handler
7345            start emulation until we arrive back to a valid state */
7346         if (vmx->emulation_required)
7347                 return;
7348
7349         if (vmx->nested.sync_shadow_vmcs) {
7350                 copy_vmcs12_to_shadow(vmx);
7351                 vmx->nested.sync_shadow_vmcs = false;
7352         }
7353
7354         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
7355                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7356         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
7357                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7358
7359         /* When single-stepping over STI and MOV SS, we must clear the
7360          * corresponding interruptibility bits in the guest state. Otherwise
7361          * vmentry fails as it then expects bit 14 (BS) in pending debug
7362          * exceptions being set, but that's not correct for the guest debugging
7363          * case. */
7364         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7365                 vmx_set_interrupt_shadow(vcpu, 0);
7366
7367         atomic_switch_perf_msrs(vmx);
7368         debugctlmsr = get_debugctlmsr();
7369
7370         vmx->__launched = vmx->loaded_vmcs->launched;
7371         asm(
7372                 /* Store host registers */
7373                 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
7374                 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
7375                 "push %%" _ASM_CX " \n\t"
7376                 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7377                 "je 1f \n\t"
7378                 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7379                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
7380                 "1: \n\t"
7381                 /* Reload cr2 if changed */
7382                 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
7383                 "mov %%cr2, %%" _ASM_DX " \n\t"
7384                 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
7385                 "je 2f \n\t"
7386                 "mov %%" _ASM_AX", %%cr2 \n\t"
7387                 "2: \n\t"
7388                 /* Check if vmlaunch of vmresume is needed */
7389                 "cmpl $0, %c[launched](%0) \n\t"
7390                 /* Load guest registers.  Don't clobber flags. */
7391                 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
7392                 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
7393                 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
7394                 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
7395                 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
7396                 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
7397 #ifdef CONFIG_X86_64
7398                 "mov %c[r8](%0),  %%r8  \n\t"
7399                 "mov %c[r9](%0),  %%r9  \n\t"
7400                 "mov %c[r10](%0), %%r10 \n\t"
7401                 "mov %c[r11](%0), %%r11 \n\t"
7402                 "mov %c[r12](%0), %%r12 \n\t"
7403                 "mov %c[r13](%0), %%r13 \n\t"
7404                 "mov %c[r14](%0), %%r14 \n\t"
7405                 "mov %c[r15](%0), %%r15 \n\t"
7406 #endif
7407                 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
7408
7409                 /* Enter guest mode */
7410                 "jne 1f \n\t"
7411                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
7412                 "jmp 2f \n\t"
7413                 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
7414                 "2: "
7415                 /* Save guest registers, load host registers, keep flags */
7416                 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
7417                 "pop %0 \n\t"
7418                 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
7419                 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
7420                 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
7421                 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
7422                 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
7423                 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
7424                 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
7425 #ifdef CONFIG_X86_64
7426                 "mov %%r8,  %c[r8](%0) \n\t"
7427                 "mov %%r9,  %c[r9](%0) \n\t"
7428                 "mov %%r10, %c[r10](%0) \n\t"
7429                 "mov %%r11, %c[r11](%0) \n\t"
7430                 "mov %%r12, %c[r12](%0) \n\t"
7431                 "mov %%r13, %c[r13](%0) \n\t"
7432                 "mov %%r14, %c[r14](%0) \n\t"
7433                 "mov %%r15, %c[r15](%0) \n\t"
7434 #endif
7435                 "mov %%cr2, %%" _ASM_AX "   \n\t"
7436                 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
7437
7438                 "pop  %%" _ASM_BP "; pop  %%" _ASM_DX " \n\t"
7439                 "setbe %c[fail](%0) \n\t"
7440                 ".pushsection .rodata \n\t"
7441                 ".global vmx_return \n\t"
7442                 "vmx_return: " _ASM_PTR " 2b \n\t"
7443                 ".popsection"
7444               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
7445                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
7446                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
7447                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
7448                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
7449                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
7450                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
7451                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
7452                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
7453                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
7454                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
7455 #ifdef CONFIG_X86_64
7456                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
7457                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
7458                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
7459                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
7460                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
7461                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
7462                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
7463                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
7464 #endif
7465                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
7466                 [wordsize]"i"(sizeof(ulong))
7467               : "cc", "memory"
7468 #ifdef CONFIG_X86_64
7469                 , "rax", "rbx", "rdi", "rsi"
7470                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
7471 #else
7472                 , "eax", "ebx", "edi", "esi"
7473 #endif
7474               );
7475
7476         /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7477         if (debugctlmsr)
7478                 update_debugctlmsr(debugctlmsr);
7479
7480 #ifndef CONFIG_X86_64
7481         /*
7482          * The sysexit path does not restore ds/es, so we must set them to
7483          * a reasonable value ourselves.
7484          *
7485          * We can't defer this to vmx_load_host_state() since that function
7486          * may be executed in interrupt context, which saves and restore segments
7487          * around it, nullifying its effect.
7488          */
7489         loadsegment(ds, __USER_DS);
7490         loadsegment(es, __USER_DS);
7491 #endif
7492
7493         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
7494                                   | (1 << VCPU_EXREG_RFLAGS)
7495                                   | (1 << VCPU_EXREG_PDPTR)
7496                                   | (1 << VCPU_EXREG_SEGMENTS)
7497                                   | (1 << VCPU_EXREG_CR3));
7498         vcpu->arch.regs_dirty = 0;
7499
7500         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7501
7502         vmx->loaded_vmcs->launched = 1;
7503
7504         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
7505         trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
7506
7507         /*
7508          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
7509          * we did not inject a still-pending event to L1 now because of
7510          * nested_run_pending, we need to re-enable this bit.
7511          */
7512         if (vmx->nested.nested_run_pending)
7513                 kvm_make_request(KVM_REQ_EVENT, vcpu);
7514
7515         vmx->nested.nested_run_pending = 0;
7516
7517         vmx_complete_atomic_exit(vmx);
7518         vmx_recover_nmi_blocking(vmx);
7519         vmx_complete_interrupts(vmx);
7520 }
7521
7522 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
7523 {
7524         struct vcpu_vmx *vmx = to_vmx(vcpu);
7525
7526         free_vpid(vmx);
7527         free_loaded_vmcs(vmx->loaded_vmcs);
7528         free_nested(vmx);
7529         kfree(vmx->guest_msrs);
7530         kvm_vcpu_uninit(vcpu);
7531         kmem_cache_free(kvm_vcpu_cache, vmx);
7532 }
7533
7534 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
7535 {
7536         int err;
7537         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
7538         int cpu;
7539
7540         if (!vmx)
7541                 return ERR_PTR(-ENOMEM);
7542
7543         allocate_vpid(vmx);
7544
7545         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
7546         if (err)
7547                 goto free_vcpu;
7548
7549         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
7550         err = -ENOMEM;
7551         if (!vmx->guest_msrs) {
7552                 goto uninit_vcpu;
7553         }
7554
7555         vmx->loaded_vmcs = &vmx->vmcs01;
7556         vmx->loaded_vmcs->vmcs = alloc_vmcs();
7557         if (!vmx->loaded_vmcs->vmcs)
7558                 goto free_msrs;
7559         if (!vmm_exclusive)
7560                 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
7561         loaded_vmcs_init(vmx->loaded_vmcs);
7562         if (!vmm_exclusive)
7563                 kvm_cpu_vmxoff();
7564
7565         cpu = get_cpu();
7566         vmx_vcpu_load(&vmx->vcpu, cpu);
7567         vmx->vcpu.cpu = cpu;
7568         err = vmx_vcpu_setup(vmx);
7569         vmx_vcpu_put(&vmx->vcpu);
7570         put_cpu();
7571         if (err)
7572                 goto free_vmcs;
7573         if (vm_need_virtualize_apic_accesses(kvm)) {
7574                 err = alloc_apic_access_page(kvm);
7575                 if (err)
7576                         goto free_vmcs;
7577         }
7578
7579         if (enable_ept) {
7580                 if (!kvm->arch.ept_identity_map_addr)
7581                         kvm->arch.ept_identity_map_addr =
7582                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
7583                 err = -ENOMEM;
7584                 if (alloc_identity_pagetable(kvm) != 0)
7585                         goto free_vmcs;
7586                 if (!init_rmode_identity_map(kvm))
7587                         goto free_vmcs;
7588         }
7589
7590         vmx->nested.current_vmptr = -1ull;
7591         vmx->nested.current_vmcs12 = NULL;
7592
7593         return &vmx->vcpu;
7594
7595 free_vmcs:
7596         free_loaded_vmcs(vmx->loaded_vmcs);
7597 free_msrs:
7598         kfree(vmx->guest_msrs);
7599 uninit_vcpu:
7600         kvm_vcpu_uninit(&vmx->vcpu);
7601 free_vcpu:
7602         free_vpid(vmx);
7603         kmem_cache_free(kvm_vcpu_cache, vmx);
7604         return ERR_PTR(err);
7605 }
7606
7607 static void __init vmx_check_processor_compat(void *rtn)
7608 {
7609         struct vmcs_config vmcs_conf;
7610
7611         *(int *)rtn = 0;
7612         if (setup_vmcs_config(&vmcs_conf) < 0)
7613                 *(int *)rtn = -EIO;
7614         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7615                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7616                                 smp_processor_id());
7617                 *(int *)rtn = -EIO;
7618         }
7619 }
7620
7621 static int get_ept_level(void)
7622 {
7623         return VMX_EPT_DEFAULT_GAW + 1;
7624 }
7625
7626 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7627 {
7628         u64 ret;
7629
7630         /* For VT-d and EPT combination
7631          * 1. MMIO: always map as UC
7632          * 2. EPT with VT-d:
7633          *   a. VT-d without snooping control feature: can't guarantee the
7634          *      result, try to trust guest.
7635          *   b. VT-d with snooping control feature: snooping control feature of
7636          *      VT-d engine can guarantee the cache correctness. Just set it
7637          *      to WB to keep consistent with host. So the same as item 3.
7638          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
7639          *    consistent with host MTRR
7640          */
7641         if (is_mmio)
7642                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7643         else if (kvm_arch_has_noncoherent_dma(vcpu->kvm))
7644                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
7645                       VMX_EPT_MT_EPTE_SHIFT;
7646         else
7647                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
7648                         | VMX_EPT_IPAT_BIT;
7649
7650         return ret;
7651 }
7652
7653 static int vmx_get_lpage_level(void)
7654 {
7655         if (enable_ept && !cpu_has_vmx_ept_1g_page())
7656                 return PT_DIRECTORY_LEVEL;
7657         else
7658                 /* For shadow and EPT supported 1GB page */
7659                 return PT_PDPE_LEVEL;
7660 }
7661
7662 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7663 {
7664         struct kvm_cpuid_entry2 *best;
7665         struct vcpu_vmx *vmx = to_vmx(vcpu);
7666         u32 exec_control;
7667
7668         vmx->rdtscp_enabled = false;
7669         if (vmx_rdtscp_supported()) {
7670                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7671                 if (exec_control & SECONDARY_EXEC_RDTSCP) {
7672                         best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
7673                         if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
7674                                 vmx->rdtscp_enabled = true;
7675                         else {
7676                                 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7677                                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7678                                                 exec_control);
7679                         }
7680                 }
7681         }
7682
7683         /* Exposing INVPCID only when PCID is exposed */
7684         best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7685         if (vmx_invpcid_supported() &&
7686             best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
7687             guest_cpuid_has_pcid(vcpu)) {
7688                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7689                 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
7690                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7691                              exec_control);
7692         } else {
7693                 if (cpu_has_secondary_exec_ctrls()) {
7694                         exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7695                         exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
7696                         vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7697                                      exec_control);
7698                 }
7699                 if (best)
7700                         best->ebx &= ~bit(X86_FEATURE_INVPCID);
7701         }
7702 }
7703
7704 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7705 {
7706         if (func == 1 && nested)
7707                 entry->ecx |= bit(X86_FEATURE_VMX);
7708 }
7709
7710 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
7711                 struct x86_exception *fault)
7712 {
7713         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7714         u32 exit_reason;
7715
7716         if (fault->error_code & PFERR_RSVD_MASK)
7717                 exit_reason = EXIT_REASON_EPT_MISCONFIG;
7718         else
7719                 exit_reason = EXIT_REASON_EPT_VIOLATION;
7720         nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
7721         vmcs12->guest_physical_address = fault->address;
7722 }
7723
7724 /* Callbacks for nested_ept_init_mmu_context: */
7725
7726 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
7727 {
7728         /* return the page table to be shadowed - in our case, EPT12 */
7729         return get_vmcs12(vcpu)->ept_pointer;
7730 }
7731
7732 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
7733 {
7734         kvm_init_shadow_ept_mmu(vcpu, &vcpu->arch.mmu,
7735                         nested_vmx_ept_caps & VMX_EPT_EXECUTE_ONLY_BIT);
7736
7737         vcpu->arch.mmu.set_cr3           = vmx_set_cr3;
7738         vcpu->arch.mmu.get_cr3           = nested_ept_get_cr3;
7739         vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
7740
7741         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
7742 }
7743
7744 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
7745 {
7746         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
7747 }
7748
7749 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
7750                 struct x86_exception *fault)
7751 {
7752         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7753
7754         WARN_ON(!is_guest_mode(vcpu));
7755
7756         /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
7757         if (vmcs12->exception_bitmap & (1u << PF_VECTOR))
7758                 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
7759                                   vmcs_read32(VM_EXIT_INTR_INFO),
7760                                   vmcs_readl(EXIT_QUALIFICATION));
7761         else
7762                 kvm_inject_page_fault(vcpu, fault);
7763 }
7764
7765 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
7766 {
7767         u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
7768         struct vcpu_vmx *vmx = to_vmx(vcpu);
7769
7770         if (vcpu->arch.virtual_tsc_khz == 0)
7771                 return;
7772
7773         /* Make sure short timeouts reliably trigger an immediate vmexit.
7774          * hrtimer_start does not guarantee this. */
7775         if (preemption_timeout <= 1) {
7776                 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
7777                 return;
7778         }
7779
7780         preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
7781         preemption_timeout *= 1000000;
7782         do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
7783         hrtimer_start(&vmx->nested.preemption_timer,
7784                       ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
7785 }
7786
7787 /*
7788  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
7789  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
7790  * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
7791  * guest in a way that will both be appropriate to L1's requests, and our
7792  * needs. In addition to modifying the active vmcs (which is vmcs02), this
7793  * function also has additional necessary side-effects, like setting various
7794  * vcpu->arch fields.
7795  */
7796 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7797 {
7798         struct vcpu_vmx *vmx = to_vmx(vcpu);
7799         u32 exec_control;
7800
7801         vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
7802         vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
7803         vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
7804         vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
7805         vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
7806         vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
7807         vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
7808         vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
7809         vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
7810         vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
7811         vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
7812         vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
7813         vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
7814         vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
7815         vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
7816         vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
7817         vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
7818         vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
7819         vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
7820         vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
7821         vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
7822         vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
7823         vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
7824         vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
7825         vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
7826         vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
7827         vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
7828         vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
7829         vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
7830         vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
7831         vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
7832         vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
7833         vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
7834         vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
7835         vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
7836         vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
7837
7838         vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
7839         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
7840                 vmcs12->vm_entry_intr_info_field);
7841         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
7842                 vmcs12->vm_entry_exception_error_code);
7843         vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
7844                 vmcs12->vm_entry_instruction_len);
7845         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
7846                 vmcs12->guest_interruptibility_info);
7847         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
7848         kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
7849         vmx_set_rflags(vcpu, vmcs12->guest_rflags);
7850         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
7851                 vmcs12->guest_pending_dbg_exceptions);
7852         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
7853         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
7854
7855         vmcs_write64(VMCS_LINK_POINTER, -1ull);
7856
7857         exec_control = vmcs12->pin_based_vm_exec_control;
7858         exec_control |= vmcs_config.pin_based_exec_ctrl;
7859         exec_control &= ~(PIN_BASED_VMX_PREEMPTION_TIMER |
7860                           PIN_BASED_POSTED_INTR);
7861         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
7862
7863         vmx->nested.preemption_timer_expired = false;
7864         if (nested_cpu_has_preemption_timer(vmcs12))
7865                 vmx_start_preemption_timer(vcpu);
7866
7867         /*
7868          * Whether page-faults are trapped is determined by a combination of
7869          * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
7870          * If enable_ept, L0 doesn't care about page faults and we should
7871          * set all of these to L1's desires. However, if !enable_ept, L0 does
7872          * care about (at least some) page faults, and because it is not easy
7873          * (if at all possible?) to merge L0 and L1's desires, we simply ask
7874          * to exit on each and every L2 page fault. This is done by setting
7875          * MASK=MATCH=0 and (see below) EB.PF=1.
7876          * Note that below we don't need special code to set EB.PF beyond the
7877          * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
7878          * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
7879          * !enable_ept, EB.PF is 1, so the "or" will always be 1.
7880          *
7881          * A problem with this approach (when !enable_ept) is that L1 may be
7882          * injected with more page faults than it asked for. This could have
7883          * caused problems, but in practice existing hypervisors don't care.
7884          * To fix this, we will need to emulate the PFEC checking (on the L1
7885          * page tables), using walk_addr(), when injecting PFs to L1.
7886          */
7887         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
7888                 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
7889         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
7890                 enable_ept ? vmcs12->page_fault_error_code_match : 0);
7891
7892         if (cpu_has_secondary_exec_ctrls()) {
7893                 exec_control = vmx_secondary_exec_control(vmx);
7894                 if (!vmx->rdtscp_enabled)
7895                         exec_control &= ~SECONDARY_EXEC_RDTSCP;
7896                 /* Take the following fields only from vmcs12 */
7897                 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7898                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
7899                                   SECONDARY_EXEC_APIC_REGISTER_VIRT);
7900                 if (nested_cpu_has(vmcs12,
7901                                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
7902                         exec_control |= vmcs12->secondary_vm_exec_control;
7903
7904                 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
7905                         /*
7906                          * Translate L1 physical address to host physical
7907                          * address for vmcs02. Keep the page pinned, so this
7908                          * physical address remains valid. We keep a reference
7909                          * to it so we can release it later.
7910                          */
7911                         if (vmx->nested.apic_access_page) /* shouldn't happen */
7912                                 nested_release_page(vmx->nested.apic_access_page);
7913                         vmx->nested.apic_access_page =
7914                                 nested_get_page(vcpu, vmcs12->apic_access_addr);
7915                         /*
7916                          * If translation failed, no matter: This feature asks
7917                          * to exit when accessing the given address, and if it
7918                          * can never be accessed, this feature won't do
7919                          * anything anyway.
7920                          */
7921                         if (!vmx->nested.apic_access_page)
7922                                 exec_control &=
7923                                   ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7924                         else
7925                                 vmcs_write64(APIC_ACCESS_ADDR,
7926                                   page_to_phys(vmx->nested.apic_access_page));
7927                 } else if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm)) {
7928                         exec_control |=
7929                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7930                         vmcs_write64(APIC_ACCESS_ADDR,
7931                                 page_to_phys(vcpu->kvm->arch.apic_access_page));
7932                 }
7933
7934                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
7935         }
7936
7937
7938         /*
7939          * Set host-state according to L0's settings (vmcs12 is irrelevant here)
7940          * Some constant fields are set here by vmx_set_constant_host_state().
7941          * Other fields are different per CPU, and will be set later when
7942          * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
7943          */
7944         vmx_set_constant_host_state(vmx);
7945
7946         /*
7947          * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
7948          * entry, but only if the current (host) sp changed from the value
7949          * we wrote last (vmx->host_rsp). This cache is no longer relevant
7950          * if we switch vmcs, and rather than hold a separate cache per vmcs,
7951          * here we just force the write to happen on entry.
7952          */
7953         vmx->host_rsp = 0;
7954
7955         exec_control = vmx_exec_control(vmx); /* L0's desires */
7956         exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
7957         exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
7958         exec_control &= ~CPU_BASED_TPR_SHADOW;
7959         exec_control |= vmcs12->cpu_based_vm_exec_control;
7960         /*
7961          * Merging of IO and MSR bitmaps not currently supported.
7962          * Rather, exit every time.
7963          */
7964         exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
7965         exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
7966         exec_control |= CPU_BASED_UNCOND_IO_EXITING;
7967
7968         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
7969
7970         /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
7971          * bitwise-or of what L1 wants to trap for L2, and what we want to
7972          * trap. Note that CR0.TS also needs updating - we do this later.
7973          */
7974         update_exception_bitmap(vcpu);
7975         vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
7976         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7977
7978         /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
7979          * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
7980          * bits are further modified by vmx_set_efer() below.
7981          */
7982         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
7983
7984         /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
7985          * emulated by vmx_set_efer(), below.
7986          */
7987         vm_entry_controls_init(vmx,
7988                 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
7989                         ~VM_ENTRY_IA32E_MODE) |
7990                 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
7991
7992         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
7993                 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
7994                 vcpu->arch.pat = vmcs12->guest_ia32_pat;
7995         } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
7996                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
7997
7998
7999         set_cr4_guest_host_mask(vmx);
8000
8001         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
8002                 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
8003
8004         if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
8005                 vmcs_write64(TSC_OFFSET,
8006                         vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
8007         else
8008                 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8009
8010         if (enable_vpid) {
8011                 /*
8012                  * Trivially support vpid by letting L2s share their parent
8013                  * L1's vpid. TODO: move to a more elaborate solution, giving
8014                  * each L2 its own vpid and exposing the vpid feature to L1.
8015                  */
8016                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
8017                 vmx_flush_tlb(vcpu);
8018         }
8019
8020         if (nested_cpu_has_ept(vmcs12)) {
8021                 kvm_mmu_unload(vcpu);
8022                 nested_ept_init_mmu_context(vcpu);
8023         }
8024
8025         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
8026                 vcpu->arch.efer = vmcs12->guest_ia32_efer;
8027         else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
8028                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8029         else
8030                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8031         /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
8032         vmx_set_efer(vcpu, vcpu->arch.efer);
8033
8034         /*
8035          * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
8036          * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
8037          * The CR0_READ_SHADOW is what L2 should have expected to read given
8038          * the specifications by L1; It's not enough to take
8039          * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
8040          * have more bits than L1 expected.
8041          */
8042         vmx_set_cr0(vcpu, vmcs12->guest_cr0);
8043         vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
8044
8045         vmx_set_cr4(vcpu, vmcs12->guest_cr4);
8046         vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
8047
8048         /* shadow page tables on either EPT or shadow page tables */
8049         kvm_set_cr3(vcpu, vmcs12->guest_cr3);
8050         kvm_mmu_reset_context(vcpu);
8051
8052         if (!enable_ept)
8053                 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
8054
8055         /*
8056          * L1 may access the L2's PDPTR, so save them to construct vmcs12
8057          */
8058         if (enable_ept) {
8059                 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
8060                 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
8061                 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
8062                 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
8063         }
8064
8065         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
8066         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
8067 }
8068
8069 /*
8070  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
8071  * for running an L2 nested guest.
8072  */
8073 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
8074 {
8075         struct vmcs12 *vmcs12;
8076         struct vcpu_vmx *vmx = to_vmx(vcpu);
8077         int cpu;
8078         struct loaded_vmcs *vmcs02;
8079         bool ia32e;
8080
8081         if (!nested_vmx_check_permission(vcpu) ||
8082             !nested_vmx_check_vmcs12(vcpu))
8083                 return 1;
8084
8085         skip_emulated_instruction(vcpu);
8086         vmcs12 = get_vmcs12(vcpu);
8087
8088         if (enable_shadow_vmcs)
8089                 copy_shadow_to_vmcs12(vmx);
8090
8091         /*
8092          * The nested entry process starts with enforcing various prerequisites
8093          * on vmcs12 as required by the Intel SDM, and act appropriately when
8094          * they fail: As the SDM explains, some conditions should cause the
8095          * instruction to fail, while others will cause the instruction to seem
8096          * to succeed, but return an EXIT_REASON_INVALID_STATE.
8097          * To speed up the normal (success) code path, we should avoid checking
8098          * for misconfigurations which will anyway be caught by the processor
8099          * when using the merged vmcs02.
8100          */
8101         if (vmcs12->launch_state == launch) {
8102                 nested_vmx_failValid(vcpu,
8103                         launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
8104                                : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
8105                 return 1;
8106         }
8107
8108         if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
8109             vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
8110                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8111                 return 1;
8112         }
8113
8114         if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
8115                         !PAGE_ALIGNED(vmcs12->msr_bitmap)) {
8116                 /*TODO: Also verify bits beyond physical address width are 0*/
8117                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8118                 return 1;
8119         }
8120
8121         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
8122                         !PAGE_ALIGNED(vmcs12->apic_access_addr)) {
8123                 /*TODO: Also verify bits beyond physical address width are 0*/
8124                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8125                 return 1;
8126         }
8127
8128         if (vmcs12->vm_entry_msr_load_count > 0 ||
8129             vmcs12->vm_exit_msr_load_count > 0 ||
8130             vmcs12->vm_exit_msr_store_count > 0) {
8131                 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
8132                                     __func__);
8133                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8134                 return 1;
8135         }
8136
8137         if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
8138                                 nested_vmx_true_procbased_ctls_low,
8139                                 nested_vmx_procbased_ctls_high) ||
8140             !vmx_control_verify(vmcs12->secondary_vm_exec_control,
8141               nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
8142             !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
8143               nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
8144             !vmx_control_verify(vmcs12->vm_exit_controls,
8145               nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
8146             !vmx_control_verify(vmcs12->vm_entry_controls,
8147               nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
8148         {
8149                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8150                 return 1;
8151         }
8152
8153         if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
8154             ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
8155                 nested_vmx_failValid(vcpu,
8156                         VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
8157                 return 1;
8158         }
8159
8160         if (!nested_cr0_valid(vmcs12, vmcs12->guest_cr0) ||
8161             ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
8162                 nested_vmx_entry_failure(vcpu, vmcs12,
8163                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8164                 return 1;
8165         }
8166         if (vmcs12->vmcs_link_pointer != -1ull) {
8167                 nested_vmx_entry_failure(vcpu, vmcs12,
8168                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
8169                 return 1;
8170         }
8171
8172         /*
8173          * If the load IA32_EFER VM-entry control is 1, the following checks
8174          * are performed on the field for the IA32_EFER MSR:
8175          * - Bits reserved in the IA32_EFER MSR must be 0.
8176          * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
8177          *   the IA-32e mode guest VM-exit control. It must also be identical
8178          *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
8179          *   CR0.PG) is 1.
8180          */
8181         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
8182                 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
8183                 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
8184                     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
8185                     ((vmcs12->guest_cr0 & X86_CR0_PG) &&
8186                      ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
8187                         nested_vmx_entry_failure(vcpu, vmcs12,
8188                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8189                         return 1;
8190                 }
8191         }
8192
8193         /*
8194          * If the load IA32_EFER VM-exit control is 1, bits reserved in the
8195          * IA32_EFER MSR must be 0 in the field for that register. In addition,
8196          * the values of the LMA and LME bits in the field must each be that of
8197          * the host address-space size VM-exit control.
8198          */
8199         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
8200                 ia32e = (vmcs12->vm_exit_controls &
8201                          VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
8202                 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
8203                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
8204                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
8205                         nested_vmx_entry_failure(vcpu, vmcs12,
8206                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8207                         return 1;
8208                 }
8209         }
8210
8211         /*
8212          * We're finally done with prerequisite checking, and can start with
8213          * the nested entry.
8214          */
8215
8216         vmcs02 = nested_get_current_vmcs02(vmx);
8217         if (!vmcs02)
8218                 return -ENOMEM;
8219
8220         enter_guest_mode(vcpu);
8221
8222         vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
8223
8224         cpu = get_cpu();
8225         vmx->loaded_vmcs = vmcs02;
8226         vmx_vcpu_put(vcpu);
8227         vmx_vcpu_load(vcpu, cpu);
8228         vcpu->cpu = cpu;
8229         put_cpu();
8230
8231         vmx_segment_cache_clear(vmx);
8232
8233         vmcs12->launch_state = 1;
8234
8235         prepare_vmcs02(vcpu, vmcs12);
8236
8237         if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
8238                 return kvm_emulate_halt(vcpu);
8239
8240         vmx->nested.nested_run_pending = 1;
8241
8242         /*
8243          * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
8244          * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
8245          * returned as far as L1 is concerned. It will only return (and set
8246          * the success flag) when L2 exits (see nested_vmx_vmexit()).
8247          */
8248         return 1;
8249 }
8250
8251 /*
8252  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
8253  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
8254  * This function returns the new value we should put in vmcs12.guest_cr0.
8255  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
8256  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
8257  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
8258  *     didn't trap the bit, because if L1 did, so would L0).
8259  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
8260  *     been modified by L2, and L1 knows it. So just leave the old value of
8261  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
8262  *     isn't relevant, because if L0 traps this bit it can set it to anything.
8263  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
8264  *     changed these bits, and therefore they need to be updated, but L0
8265  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
8266  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
8267  */
8268 static inline unsigned long
8269 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8270 {
8271         return
8272         /*1*/   (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
8273         /*2*/   (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
8274         /*3*/   (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
8275                         vcpu->arch.cr0_guest_owned_bits));
8276 }
8277
8278 static inline unsigned long
8279 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8280 {
8281         return
8282         /*1*/   (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
8283         /*2*/   (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
8284         /*3*/   (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
8285                         vcpu->arch.cr4_guest_owned_bits));
8286 }
8287
8288 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
8289                                        struct vmcs12 *vmcs12)
8290 {
8291         u32 idt_vectoring;
8292         unsigned int nr;
8293
8294         if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
8295                 nr = vcpu->arch.exception.nr;
8296                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8297
8298                 if (kvm_exception_is_soft(nr)) {
8299                         vmcs12->vm_exit_instruction_len =
8300                                 vcpu->arch.event_exit_inst_len;
8301                         idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
8302                 } else
8303                         idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
8304
8305                 if (vcpu->arch.exception.has_error_code) {
8306                         idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
8307                         vmcs12->idt_vectoring_error_code =
8308                                 vcpu->arch.exception.error_code;
8309                 }
8310
8311                 vmcs12->idt_vectoring_info_field = idt_vectoring;
8312         } else if (vcpu->arch.nmi_injected) {
8313                 vmcs12->idt_vectoring_info_field =
8314                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
8315         } else if (vcpu->arch.interrupt.pending) {
8316                 nr = vcpu->arch.interrupt.nr;
8317                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8318
8319                 if (vcpu->arch.interrupt.soft) {
8320                         idt_vectoring |= INTR_TYPE_SOFT_INTR;
8321                         vmcs12->vm_entry_instruction_len =
8322                                 vcpu->arch.event_exit_inst_len;
8323                 } else
8324                         idt_vectoring |= INTR_TYPE_EXT_INTR;
8325
8326                 vmcs12->idt_vectoring_info_field = idt_vectoring;
8327         }
8328 }
8329
8330 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
8331 {
8332         struct vcpu_vmx *vmx = to_vmx(vcpu);
8333
8334         if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
8335             vmx->nested.preemption_timer_expired) {
8336                 if (vmx->nested.nested_run_pending)
8337                         return -EBUSY;
8338                 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
8339                 return 0;
8340         }
8341
8342         if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
8343                 if (vmx->nested.nested_run_pending ||
8344                     vcpu->arch.interrupt.pending)
8345                         return -EBUSY;
8346                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
8347                                   NMI_VECTOR | INTR_TYPE_NMI_INTR |
8348                                   INTR_INFO_VALID_MASK, 0);
8349                 /*
8350                  * The NMI-triggered VM exit counts as injection:
8351                  * clear this one and block further NMIs.
8352                  */
8353                 vcpu->arch.nmi_pending = 0;
8354                 vmx_set_nmi_mask(vcpu, true);
8355                 return 0;
8356         }
8357
8358         if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
8359             nested_exit_on_intr(vcpu)) {
8360                 if (vmx->nested.nested_run_pending)
8361                         return -EBUSY;
8362                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
8363         }
8364
8365         return 0;
8366 }
8367
8368 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
8369 {
8370         ktime_t remaining =
8371                 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
8372         u64 value;
8373
8374         if (ktime_to_ns(remaining) <= 0)
8375                 return 0;
8376
8377         value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
8378         do_div(value, 1000000);
8379         return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
8380 }
8381
8382 /*
8383  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
8384  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
8385  * and this function updates it to reflect the changes to the guest state while
8386  * L2 was running (and perhaps made some exits which were handled directly by L0
8387  * without going back to L1), and to reflect the exit reason.
8388  * Note that we do not have to copy here all VMCS fields, just those that
8389  * could have changed by the L2 guest or the exit - i.e., the guest-state and
8390  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
8391  * which already writes to vmcs12 directly.
8392  */
8393 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
8394                            u32 exit_reason, u32 exit_intr_info,
8395                            unsigned long exit_qualification)
8396 {
8397         /* update guest state fields: */
8398         vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
8399         vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
8400
8401         kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
8402         vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8403         vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
8404         vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
8405
8406         vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
8407         vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
8408         vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
8409         vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
8410         vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
8411         vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
8412         vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
8413         vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
8414         vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
8415         vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
8416         vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
8417         vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
8418         vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
8419         vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
8420         vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
8421         vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
8422         vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
8423         vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
8424         vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
8425         vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
8426         vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
8427         vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
8428         vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
8429         vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
8430         vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
8431         vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
8432         vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
8433         vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
8434         vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
8435         vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
8436         vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
8437         vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
8438         vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
8439         vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
8440         vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
8441         vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
8442
8443         vmcs12->guest_interruptibility_info =
8444                 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
8445         vmcs12->guest_pending_dbg_exceptions =
8446                 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
8447         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
8448                 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
8449         else
8450                 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
8451
8452         if (nested_cpu_has_preemption_timer(vmcs12)) {
8453                 if (vmcs12->vm_exit_controls &
8454                     VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
8455                         vmcs12->vmx_preemption_timer_value =
8456                                 vmx_get_preemption_timer_value(vcpu);
8457                 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
8458         }
8459
8460         /*
8461          * In some cases (usually, nested EPT), L2 is allowed to change its
8462          * own CR3 without exiting. If it has changed it, we must keep it.
8463          * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
8464          * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
8465          *
8466          * Additionally, restore L2's PDPTR to vmcs12.
8467          */
8468         if (enable_ept) {
8469                 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
8470                 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
8471                 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
8472                 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
8473                 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
8474         }
8475
8476         vmcs12->vm_entry_controls =
8477                 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
8478                 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
8479
8480         /* TODO: These cannot have changed unless we have MSR bitmaps and
8481          * the relevant bit asks not to trap the change */
8482         vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8483         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
8484                 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
8485         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
8486                 vmcs12->guest_ia32_efer = vcpu->arch.efer;
8487         vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
8488         vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
8489         vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
8490         if (vmx_mpx_supported())
8491                 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
8492
8493         /* update exit information fields: */
8494
8495         vmcs12->vm_exit_reason = exit_reason;
8496         vmcs12->exit_qualification = exit_qualification;
8497
8498         vmcs12->vm_exit_intr_info = exit_intr_info;
8499         if ((vmcs12->vm_exit_intr_info &
8500              (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8501             (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
8502                 vmcs12->vm_exit_intr_error_code =
8503                         vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8504         vmcs12->idt_vectoring_info_field = 0;
8505         vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
8506         vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8507
8508         if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
8509                 /* vm_entry_intr_info_field is cleared on exit. Emulate this
8510                  * instead of reading the real value. */
8511                 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
8512
8513                 /*
8514                  * Transfer the event that L0 or L1 may wanted to inject into
8515                  * L2 to IDT_VECTORING_INFO_FIELD.
8516                  */
8517                 vmcs12_save_pending_event(vcpu, vmcs12);
8518         }
8519
8520         /*
8521          * Drop what we picked up for L2 via vmx_complete_interrupts. It is
8522          * preserved above and would only end up incorrectly in L1.
8523          */
8524         vcpu->arch.nmi_injected = false;
8525         kvm_clear_exception_queue(vcpu);
8526         kvm_clear_interrupt_queue(vcpu);
8527 }
8528
8529 /*
8530  * A part of what we need to when the nested L2 guest exits and we want to
8531  * run its L1 parent, is to reset L1's guest state to the host state specified
8532  * in vmcs12.
8533  * This function is to be called not only on normal nested exit, but also on
8534  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
8535  * Failures During or After Loading Guest State").
8536  * This function should be called when the active VMCS is L1's (vmcs01).
8537  */
8538 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
8539                                    struct vmcs12 *vmcs12)
8540 {
8541         struct kvm_segment seg;
8542
8543         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
8544                 vcpu->arch.efer = vmcs12->host_ia32_efer;
8545         else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8546                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8547         else
8548                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8549         vmx_set_efer(vcpu, vcpu->arch.efer);
8550
8551         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
8552         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
8553         vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
8554         /*
8555          * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
8556          * actually changed, because it depends on the current state of
8557          * fpu_active (which may have changed).
8558          * Note that vmx_set_cr0 refers to efer set above.
8559          */
8560         vmx_set_cr0(vcpu, vmcs12->host_cr0);
8561         /*
8562          * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
8563          * to apply the same changes to L1's vmcs. We just set cr0 correctly,
8564          * but we also need to update cr0_guest_host_mask and exception_bitmap.
8565          */
8566         update_exception_bitmap(vcpu);
8567         vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
8568         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8569
8570         /*
8571          * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
8572          * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
8573          */
8574         vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
8575         kvm_set_cr4(vcpu, vmcs12->host_cr4);
8576
8577         nested_ept_uninit_mmu_context(vcpu);
8578
8579         kvm_set_cr3(vcpu, vmcs12->host_cr3);
8580         kvm_mmu_reset_context(vcpu);
8581
8582         if (!enable_ept)
8583                 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
8584
8585         if (enable_vpid) {
8586                 /*
8587                  * Trivially support vpid by letting L2s share their parent
8588                  * L1's vpid. TODO: move to a more elaborate solution, giving
8589                  * each L2 its own vpid and exposing the vpid feature to L1.
8590                  */
8591                 vmx_flush_tlb(vcpu);
8592         }
8593
8594
8595         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
8596         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
8597         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
8598         vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
8599         vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
8600
8601         /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
8602         if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
8603                 vmcs_write64(GUEST_BNDCFGS, 0);
8604
8605         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
8606                 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
8607                 vcpu->arch.pat = vmcs12->host_ia32_pat;
8608         }
8609         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8610                 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
8611                         vmcs12->host_ia32_perf_global_ctrl);
8612
8613         /* Set L1 segment info according to Intel SDM
8614             27.5.2 Loading Host Segment and Descriptor-Table Registers */
8615         seg = (struct kvm_segment) {
8616                 .base = 0,
8617                 .limit = 0xFFFFFFFF,
8618                 .selector = vmcs12->host_cs_selector,
8619                 .type = 11,
8620                 .present = 1,
8621                 .s = 1,
8622                 .g = 1
8623         };
8624         if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8625                 seg.l = 1;
8626         else
8627                 seg.db = 1;
8628         vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
8629         seg = (struct kvm_segment) {
8630                 .base = 0,
8631                 .limit = 0xFFFFFFFF,
8632                 .type = 3,
8633                 .present = 1,
8634                 .s = 1,
8635                 .db = 1,
8636                 .g = 1
8637         };
8638         seg.selector = vmcs12->host_ds_selector;
8639         vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
8640         seg.selector = vmcs12->host_es_selector;
8641         vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
8642         seg.selector = vmcs12->host_ss_selector;
8643         vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
8644         seg.selector = vmcs12->host_fs_selector;
8645         seg.base = vmcs12->host_fs_base;
8646         vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
8647         seg.selector = vmcs12->host_gs_selector;
8648         seg.base = vmcs12->host_gs_base;
8649         vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
8650         seg = (struct kvm_segment) {
8651                 .base = vmcs12->host_tr_base,
8652                 .limit = 0x67,
8653                 .selector = vmcs12->host_tr_selector,
8654                 .type = 11,
8655                 .present = 1
8656         };
8657         vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
8658
8659         kvm_set_dr(vcpu, 7, 0x400);
8660         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
8661 }
8662
8663 /*
8664  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
8665  * and modify vmcs12 to make it see what it would expect to see there if
8666  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
8667  */
8668 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
8669                               u32 exit_intr_info,
8670                               unsigned long exit_qualification)
8671 {
8672         struct vcpu_vmx *vmx = to_vmx(vcpu);
8673         int cpu;
8674         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8675
8676         /* trying to cancel vmlaunch/vmresume is a bug */
8677         WARN_ON_ONCE(vmx->nested.nested_run_pending);
8678
8679         leave_guest_mode(vcpu);
8680         prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
8681                        exit_qualification);
8682
8683         if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
8684             && nested_exit_intr_ack_set(vcpu)) {
8685                 int irq = kvm_cpu_get_interrupt(vcpu);
8686                 WARN_ON(irq < 0);
8687                 vmcs12->vm_exit_intr_info = irq |
8688                         INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
8689         }
8690
8691         trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
8692                                        vmcs12->exit_qualification,
8693                                        vmcs12->idt_vectoring_info_field,
8694                                        vmcs12->vm_exit_intr_info,
8695                                        vmcs12->vm_exit_intr_error_code,
8696                                        KVM_ISA_VMX);
8697
8698         cpu = get_cpu();
8699         vmx->loaded_vmcs = &vmx->vmcs01;
8700         vmx_vcpu_put(vcpu);
8701         vmx_vcpu_load(vcpu, cpu);
8702         vcpu->cpu = cpu;
8703         put_cpu();
8704
8705         vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
8706         vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
8707         vmx_segment_cache_clear(vmx);
8708
8709         /* if no vmcs02 cache requested, remove the one we used */
8710         if (VMCS02_POOL_SIZE == 0)
8711                 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
8712
8713         load_vmcs12_host_state(vcpu, vmcs12);
8714
8715         /* Update TSC_OFFSET if TSC was changed while L2 ran */
8716         vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8717
8718         /* This is needed for same reason as it was needed in prepare_vmcs02 */
8719         vmx->host_rsp = 0;
8720
8721         /* Unpin physical memory we referred to in vmcs02 */
8722         if (vmx->nested.apic_access_page) {
8723                 nested_release_page(vmx->nested.apic_access_page);
8724                 vmx->nested.apic_access_page = 0;
8725         }
8726
8727         /*
8728          * Exiting from L2 to L1, we're now back to L1 which thinks it just
8729          * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
8730          * success or failure flag accordingly.
8731          */
8732         if (unlikely(vmx->fail)) {
8733                 vmx->fail = 0;
8734                 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
8735         } else
8736                 nested_vmx_succeed(vcpu);
8737         if (enable_shadow_vmcs)
8738                 vmx->nested.sync_shadow_vmcs = true;
8739
8740         /* in case we halted in L2 */
8741         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8742 }
8743
8744 /*
8745  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
8746  */
8747 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
8748 {
8749         if (is_guest_mode(vcpu))
8750                 nested_vmx_vmexit(vcpu, -1, 0, 0);
8751         free_nested(to_vmx(vcpu));
8752 }
8753
8754 /*
8755  * L1's failure to enter L2 is a subset of a normal exit, as explained in
8756  * 23.7 "VM-entry failures during or after loading guest state" (this also
8757  * lists the acceptable exit-reason and exit-qualification parameters).
8758  * It should only be called before L2 actually succeeded to run, and when
8759  * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
8760  */
8761 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
8762                         struct vmcs12 *vmcs12,
8763                         u32 reason, unsigned long qualification)
8764 {
8765         load_vmcs12_host_state(vcpu, vmcs12);
8766         vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
8767         vmcs12->exit_qualification = qualification;
8768         nested_vmx_succeed(vcpu);
8769         if (enable_shadow_vmcs)
8770                 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
8771 }
8772
8773 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
8774                                struct x86_instruction_info *info,
8775                                enum x86_intercept_stage stage)
8776 {
8777         return X86EMUL_CONTINUE;
8778 }
8779
8780 static struct kvm_x86_ops vmx_x86_ops = {
8781         .cpu_has_kvm_support = cpu_has_kvm_support,
8782         .disabled_by_bios = vmx_disabled_by_bios,
8783         .hardware_setup = hardware_setup,
8784         .hardware_unsetup = hardware_unsetup,
8785         .check_processor_compatibility = vmx_check_processor_compat,
8786         .hardware_enable = hardware_enable,
8787         .hardware_disable = hardware_disable,
8788         .cpu_has_accelerated_tpr = report_flexpriority,
8789
8790         .vcpu_create = vmx_create_vcpu,
8791         .vcpu_free = vmx_free_vcpu,
8792         .vcpu_reset = vmx_vcpu_reset,
8793
8794         .prepare_guest_switch = vmx_save_host_state,
8795         .vcpu_load = vmx_vcpu_load,
8796         .vcpu_put = vmx_vcpu_put,
8797
8798         .update_db_bp_intercept = update_exception_bitmap,
8799         .get_msr = vmx_get_msr,
8800         .set_msr = vmx_set_msr,
8801         .get_segment_base = vmx_get_segment_base,
8802         .get_segment = vmx_get_segment,
8803         .set_segment = vmx_set_segment,
8804         .get_cpl = vmx_get_cpl,
8805         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
8806         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
8807         .decache_cr3 = vmx_decache_cr3,
8808         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
8809         .set_cr0 = vmx_set_cr0,
8810         .set_cr3 = vmx_set_cr3,
8811         .set_cr4 = vmx_set_cr4,
8812         .set_efer = vmx_set_efer,
8813         .get_idt = vmx_get_idt,
8814         .set_idt = vmx_set_idt,
8815         .get_gdt = vmx_get_gdt,
8816         .set_gdt = vmx_set_gdt,
8817         .get_dr6 = vmx_get_dr6,
8818         .set_dr6 = vmx_set_dr6,
8819         .set_dr7 = vmx_set_dr7,
8820         .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
8821         .cache_reg = vmx_cache_reg,
8822         .get_rflags = vmx_get_rflags,
8823         .set_rflags = vmx_set_rflags,
8824         .fpu_activate = vmx_fpu_activate,
8825         .fpu_deactivate = vmx_fpu_deactivate,
8826
8827         .tlb_flush = vmx_flush_tlb,
8828
8829         .run = vmx_vcpu_run,
8830         .handle_exit = vmx_handle_exit,
8831         .skip_emulated_instruction = skip_emulated_instruction,
8832         .set_interrupt_shadow = vmx_set_interrupt_shadow,
8833         .get_interrupt_shadow = vmx_get_interrupt_shadow,
8834         .patch_hypercall = vmx_patch_hypercall,
8835         .set_irq = vmx_inject_irq,
8836         .set_nmi = vmx_inject_nmi,
8837         .queue_exception = vmx_queue_exception,
8838         .cancel_injection = vmx_cancel_injection,
8839         .interrupt_allowed = vmx_interrupt_allowed,
8840         .nmi_allowed = vmx_nmi_allowed,
8841         .get_nmi_mask = vmx_get_nmi_mask,
8842         .set_nmi_mask = vmx_set_nmi_mask,
8843         .enable_nmi_window = enable_nmi_window,
8844         .enable_irq_window = enable_irq_window,
8845         .update_cr8_intercept = update_cr8_intercept,
8846         .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
8847         .vm_has_apicv = vmx_vm_has_apicv,
8848         .load_eoi_exitmap = vmx_load_eoi_exitmap,
8849         .hwapic_irr_update = vmx_hwapic_irr_update,
8850         .hwapic_isr_update = vmx_hwapic_isr_update,
8851         .sync_pir_to_irr = vmx_sync_pir_to_irr,
8852         .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
8853
8854         .set_tss_addr = vmx_set_tss_addr,
8855         .get_tdp_level = get_ept_level,
8856         .get_mt_mask = vmx_get_mt_mask,
8857
8858         .get_exit_info = vmx_get_exit_info,
8859
8860         .get_lpage_level = vmx_get_lpage_level,
8861
8862         .cpuid_update = vmx_cpuid_update,
8863
8864         .rdtscp_supported = vmx_rdtscp_supported,
8865         .invpcid_supported = vmx_invpcid_supported,
8866
8867         .set_supported_cpuid = vmx_set_supported_cpuid,
8868
8869         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
8870
8871         .set_tsc_khz = vmx_set_tsc_khz,
8872         .read_tsc_offset = vmx_read_tsc_offset,
8873         .write_tsc_offset = vmx_write_tsc_offset,
8874         .adjust_tsc_offset = vmx_adjust_tsc_offset,
8875         .compute_tsc_offset = vmx_compute_tsc_offset,
8876         .read_l1_tsc = vmx_read_l1_tsc,
8877
8878         .set_tdp_cr3 = vmx_set_cr3,
8879
8880         .check_intercept = vmx_check_intercept,
8881         .handle_external_intr = vmx_handle_external_intr,
8882         .mpx_supported = vmx_mpx_supported,
8883
8884         .check_nested_events = vmx_check_nested_events,
8885 };
8886
8887 static int __init vmx_init(void)
8888 {
8889         int r, i, msr;
8890
8891         rdmsrl_safe(MSR_EFER, &host_efer);
8892
8893         for (i = 0; i < NR_VMX_MSR; ++i)
8894                 kvm_define_shared_msr(i, vmx_msr_index[i]);
8895
8896         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
8897         if (!vmx_io_bitmap_a)
8898                 return -ENOMEM;
8899
8900         r = -ENOMEM;
8901
8902         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
8903         if (!vmx_io_bitmap_b)
8904                 goto out;
8905
8906         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
8907         if (!vmx_msr_bitmap_legacy)
8908                 goto out1;
8909
8910         vmx_msr_bitmap_legacy_x2apic =
8911                                 (unsigned long *)__get_free_page(GFP_KERNEL);
8912         if (!vmx_msr_bitmap_legacy_x2apic)
8913                 goto out2;
8914
8915         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
8916         if (!vmx_msr_bitmap_longmode)
8917                 goto out3;
8918
8919         vmx_msr_bitmap_longmode_x2apic =
8920                                 (unsigned long *)__get_free_page(GFP_KERNEL);
8921         if (!vmx_msr_bitmap_longmode_x2apic)
8922                 goto out4;
8923         vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
8924         if (!vmx_vmread_bitmap)
8925                 goto out5;
8926
8927         vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
8928         if (!vmx_vmwrite_bitmap)
8929                 goto out6;
8930
8931         memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
8932         memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
8933
8934         /*
8935          * Allow direct access to the PC debug port (it is often used for I/O
8936          * delays, but the vmexits simply slow things down).
8937          */
8938         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
8939         clear_bit(0x80, vmx_io_bitmap_a);
8940
8941         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
8942
8943         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
8944         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
8945
8946         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8947
8948         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
8949                      __alignof__(struct vcpu_vmx), THIS_MODULE);
8950         if (r)
8951                 goto out7;
8952
8953 #ifdef CONFIG_KEXEC
8954         rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8955                            crash_vmclear_local_loaded_vmcss);
8956 #endif
8957
8958         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
8959         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
8960         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
8961         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
8962         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
8963         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
8964         vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
8965
8966         memcpy(vmx_msr_bitmap_legacy_x2apic,
8967                         vmx_msr_bitmap_legacy, PAGE_SIZE);
8968         memcpy(vmx_msr_bitmap_longmode_x2apic,
8969                         vmx_msr_bitmap_longmode, PAGE_SIZE);
8970
8971         if (enable_apicv) {
8972                 for (msr = 0x800; msr <= 0x8ff; msr++)
8973                         vmx_disable_intercept_msr_read_x2apic(msr);
8974
8975                 /* According SDM, in x2apic mode, the whole id reg is used.
8976                  * But in KVM, it only use the highest eight bits. Need to
8977                  * intercept it */
8978                 vmx_enable_intercept_msr_read_x2apic(0x802);
8979                 /* TMCCT */
8980                 vmx_enable_intercept_msr_read_x2apic(0x839);
8981                 /* TPR */
8982                 vmx_disable_intercept_msr_write_x2apic(0x808);
8983                 /* EOI */
8984                 vmx_disable_intercept_msr_write_x2apic(0x80b);
8985                 /* SELF-IPI */
8986                 vmx_disable_intercept_msr_write_x2apic(0x83f);
8987         }
8988
8989         if (enable_ept) {
8990                 kvm_mmu_set_mask_ptes(0ull,
8991                         (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
8992                         (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
8993                         0ull, VMX_EPT_EXECUTABLE_MASK);
8994                 ept_set_mmio_spte_mask();
8995                 kvm_enable_tdp();
8996         } else
8997                 kvm_disable_tdp();
8998
8999         return 0;
9000
9001 out7:
9002         free_page((unsigned long)vmx_vmwrite_bitmap);
9003 out6:
9004         free_page((unsigned long)vmx_vmread_bitmap);
9005 out5:
9006         free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
9007 out4:
9008         free_page((unsigned long)vmx_msr_bitmap_longmode);
9009 out3:
9010         free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
9011 out2:
9012         free_page((unsigned long)vmx_msr_bitmap_legacy);
9013 out1:
9014         free_page((unsigned long)vmx_io_bitmap_b);
9015 out:
9016         free_page((unsigned long)vmx_io_bitmap_a);
9017         return r;
9018 }
9019
9020 static void __exit vmx_exit(void)
9021 {
9022         free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
9023         free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
9024         free_page((unsigned long)vmx_msr_bitmap_legacy);
9025         free_page((unsigned long)vmx_msr_bitmap_longmode);
9026         free_page((unsigned long)vmx_io_bitmap_b);
9027         free_page((unsigned long)vmx_io_bitmap_a);
9028         free_page((unsigned long)vmx_vmwrite_bitmap);
9029         free_page((unsigned long)vmx_vmread_bitmap);
9030
9031 #ifdef CONFIG_KEXEC
9032         rcu_assign_pointer(crash_vmclear_loaded_vmcss, NULL);
9033         synchronize_rcu();
9034 #endif
9035
9036         kvm_exit();
9037 }
9038
9039 module_init(vmx_init)
9040 module_exit(vmx_exit)