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
  22 #include <linux/kvm_host.h>
  23 #include <linux/module.h>
  24 #include <linux/kernel.h>
  25 #include <linux/mm.h>
  26 #include <linux/highmem.h>
  27 #include <linux/sched.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/ftrace_event.h>
  30 #include <linux/slab.h>
  31 #include <linux/tboot.h>
  32 #include "kvm_cache_regs.h"
  33 #include "x86.h"
  34
  35 #include <asm/io.h>
  36 #include <asm/desc.h>
  37 #include <asm/vmx.h>
  38 #include <asm/virtext.h>
  39 #include <asm/mce.h>
  40 #include <asm/i387.h>
  41 #include <asm/xcr.h>
  42
  43 #include "trace.h"
  44
  45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  46 #define __ex_clear(x, reg) \
  47         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
  48
  49 MODULE_AUTHOR("Qumranet");
  50 MODULE_LICENSE("GPL");
  51
  52 static int __read_mostly bypass_guest_pf = 1;
  53 module_param(bypass_guest_pf, bool, S_IRUGO);
  54
  55 static int __read_mostly enable_vpid = 1;
  56 module_param_named(vpid, enable_vpid, bool, 0444);
  57
  58 static int __read_mostly flexpriority_enabled = 1;
  59 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  60
  61 static int __read_mostly enable_ept = 1;
  62 module_param_named(ept, enable_ept, bool, S_IRUGO);
  63
  64 static int __read_mostly enable_unrestricted_guest = 1;
  65 module_param_named(unrestricted_guest,
  66                         enable_unrestricted_guest, bool, S_IRUGO);
  67
  68 static int __read_mostly emulate_invalid_guest_state = 0;
  69 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  70
  71 static int __read_mostly vmm_exclusive = 1;
  72 module_param(vmm_exclusive, bool, S_IRUGO);
  73
  74 static int __read_mostly yield_on_hlt = 1;
  75 module_param(yield_on_hlt, bool, S_IRUGO);
  76
  77 /*
  78  * If nested=1, nested virtualization is supported, i.e., guests may use
  79  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
  80  * use VMX instructions.
  81  */
  82 static int __read_mostly nested = 0;
  83 module_param(nested, bool, S_IRUGO);
  84
  85 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST                           \
  86         (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
  87 #define KVM_GUEST_CR0_MASK                                              \
  88         (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  89 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST                         \
  90         (X86_CR0_WP | X86_CR0_NE)
  91 #define KVM_VM_CR0_ALWAYS_ON                                            \
  92         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  93 #define KVM_CR4_GUEST_OWNED_BITS                                      \
  94         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
  95          | X86_CR4_OSXMMEXCPT)
  96
  97 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
  98 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
  99
 100 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
 101
 102 /*
 103  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 104  * ple_gap:    upper bound on the amount of time between two successive
 105  *             executions of PAUSE in a loop. Also indicate if ple enabled.
 106  *             According to test, this time is usually smaller than 128 cycles.
 107  * ple_window: upper bound on the amount of time a guest is allowed to execute
 108  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
 109  *             less than 2^12 cycles
 110  * Time is measured based on a counter that runs at the same rate as the TSC,
 111  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 112  */
 113 #define KVM_VMX_DEFAULT_PLE_GAP    128
 114 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
 115 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 116 module_param(ple_gap, int, S_IRUGO);
 117
 118 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 119 module_param(ple_window, int, S_IRUGO);
 120
 121 #define NR_AUTOLOAD_MSRS 1
 122 #define VMCS02_POOL_SIZE 1
 123
 124 struct vmcs {
 125         u32 revision_id;
 126         u32 abort;
 127         char data[0];
 128 };
 129
 130 /*
 131  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
 132  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
 133  * loaded on this CPU (so we can clear them if the CPU goes down).
 134  */
 135 struct loaded_vmcs {
 136         struct vmcs *vmcs;
 137         int cpu;
 138         int launched;
 139         struct list_head loaded_vmcss_on_cpu_link;
 140 };
 141
 142 struct shared_msr_entry {
 143         unsigned index;
 144         u64 data;
 145         u64 mask;
 146 };
 147
 148 /*
 149  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
 150  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
 151  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
 152  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
 153  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
 154  * More than one of these structures may exist, if L1 runs multiple L2 guests.
 155  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
 156  * underlying hardware which will be used to run L2.
 157  * This structure is packed to ensure that its layout is identical across
 158  * machines (necessary for live migration).
 159  * If there are changes in this struct, VMCS12_REVISION must be changed.
 160  */
 161 typedef u64 natural_width;
 162 struct __packed vmcs12 {
 163         /* According to the Intel spec, a VMCS region must start with the
 164          * following two fields. Then follow implementation-specific data.
 165          */
 166         u32 revision_id;
 167         u32 abort;
 168
 169         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
 170         u32 padding[7]; /* room for future expansion */
 171
 172         u64 io_bitmap_a;
 173         u64 io_bitmap_b;
 174         u64 msr_bitmap;
 175         u64 vm_exit_msr_store_addr;
 176         u64 vm_exit_msr_load_addr;
 177         u64 vm_entry_msr_load_addr;
 178         u64 tsc_offset;
 179         u64 virtual_apic_page_addr;
 180         u64 apic_access_addr;
 181         u64 ept_pointer;
 182         u64 guest_physical_address;
 183         u64 vmcs_link_pointer;
 184         u64 guest_ia32_debugctl;
 185         u64 guest_ia32_pat;
 186         u64 guest_ia32_efer;
 187         u64 guest_ia32_perf_global_ctrl;
 188         u64 guest_pdptr0;
 189         u64 guest_pdptr1;
 190         u64 guest_pdptr2;
 191         u64 guest_pdptr3;
 192         u64 host_ia32_pat;
 193         u64 host_ia32_efer;
 194         u64 host_ia32_perf_global_ctrl;
 195         u64 padding64[8]; /* room for future expansion */
 196         /*
 197          * To allow migration of L1 (complete with its L2 guests) between
 198          * machines of different natural widths (32 or 64 bit), we cannot have
 199          * unsigned long fields with no explict size. We use u64 (aliased
 200          * natural_width) instead. Luckily, x86 is little-endian.
 201          */
 202         natural_width cr0_guest_host_mask;
 203         natural_width cr4_guest_host_mask;
 204         natural_width cr0_read_shadow;
 205         natural_width cr4_read_shadow;
 206         natural_width cr3_target_value0;
 207         natural_width cr3_target_value1;
 208         natural_width cr3_target_value2;
 209         natural_width cr3_target_value3;
 210         natural_width exit_qualification;
 211         natural_width guest_linear_address;
 212         natural_width guest_cr0;
 213         natural_width guest_cr3;
 214         natural_width guest_cr4;
 215         natural_width guest_es_base;
 216         natural_width guest_cs_base;
 217         natural_width guest_ss_base;
 218         natural_width guest_ds_base;
 219         natural_width guest_fs_base;
 220         natural_width guest_gs_base;
 221         natural_width guest_ldtr_base;
 222         natural_width guest_tr_base;
 223         natural_width guest_gdtr_base;
 224         natural_width guest_idtr_base;
 225         natural_width guest_dr7;
 226         natural_width guest_rsp;
 227         natural_width guest_rip;
 228         natural_width guest_rflags;
 229         natural_width guest_pending_dbg_exceptions;
 230         natural_width guest_sysenter_esp;
 231         natural_width guest_sysenter_eip;
 232         natural_width host_cr0;
 233         natural_width host_cr3;
 234         natural_width host_cr4;
 235         natural_width host_fs_base;
 236         natural_width host_gs_base;
 237         natural_width host_tr_base;
 238         natural_width host_gdtr_base;
 239         natural_width host_idtr_base;
 240         natural_width host_ia32_sysenter_esp;
 241         natural_width host_ia32_sysenter_eip;
 242         natural_width host_rsp;
 243         natural_width host_rip;
 244         natural_width paddingl[8]; /* room for future expansion */
 245         u32 pin_based_vm_exec_control;
 246         u32 cpu_based_vm_exec_control;
 247         u32 exception_bitmap;
 248         u32 page_fault_error_code_mask;
 249         u32 page_fault_error_code_match;
 250         u32 cr3_target_count;
 251         u32 vm_exit_controls;
 252         u32 vm_exit_msr_store_count;
 253         u32 vm_exit_msr_load_count;
 254         u32 vm_entry_controls;
 255         u32 vm_entry_msr_load_count;
 256         u32 vm_entry_intr_info_field;
 257         u32 vm_entry_exception_error_code;
 258         u32 vm_entry_instruction_len;
 259         u32 tpr_threshold;
 260         u32 secondary_vm_exec_control;
 261         u32 vm_instruction_error;
 262         u32 vm_exit_reason;
 263         u32 vm_exit_intr_info;
 264         u32 vm_exit_intr_error_code;
 265         u32 idt_vectoring_info_field;
 266         u32 idt_vectoring_error_code;
 267         u32 vm_exit_instruction_len;
 268         u32 vmx_instruction_info;
 269         u32 guest_es_limit;
 270         u32 guest_cs_limit;
 271         u32 guest_ss_limit;
 272         u32 guest_ds_limit;
 273         u32 guest_fs_limit;
 274         u32 guest_gs_limit;
 275         u32 guest_ldtr_limit;
 276         u32 guest_tr_limit;
 277         u32 guest_gdtr_limit;
 278         u32 guest_idtr_limit;
 279         u32 guest_es_ar_bytes;
 280         u32 guest_cs_ar_bytes;
 281         u32 guest_ss_ar_bytes;
 282         u32 guest_ds_ar_bytes;
 283         u32 guest_fs_ar_bytes;
 284         u32 guest_gs_ar_bytes;
 285         u32 guest_ldtr_ar_bytes;
 286         u32 guest_tr_ar_bytes;
 287         u32 guest_interruptibility_info;
 288         u32 guest_activity_state;
 289         u32 guest_sysenter_cs;
 290         u32 host_ia32_sysenter_cs;
 291         u32 padding32[8]; /* room for future expansion */
 292         u16 virtual_processor_id;
 293         u16 guest_es_selector;
 294         u16 guest_cs_selector;
 295         u16 guest_ss_selector;
 296         u16 guest_ds_selector;
 297         u16 guest_fs_selector;
 298         u16 guest_gs_selector;
 299         u16 guest_ldtr_selector;
 300         u16 guest_tr_selector;
 301         u16 host_es_selector;
 302         u16 host_cs_selector;
 303         u16 host_ss_selector;
 304         u16 host_ds_selector;
 305         u16 host_fs_selector;
 306         u16 host_gs_selector;
 307         u16 host_tr_selector;
 308 };
 309
 310 /*
 311  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
 312  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
 313  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
 314  */
 315 #define VMCS12_REVISION 0x11e57ed0
 316
 317 /*
 318  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
 319  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
 320  * current implementation, 4K are reserved to avoid future complications.
 321  */
 322 #define VMCS12_SIZE 0x1000
 323
 324 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
 325 struct vmcs02_list {
 326         struct list_head list;
 327         gpa_t vmptr;
 328         struct loaded_vmcs vmcs02;
 329 };
 330
 331 /*
 332  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 333  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 334  */
 335 struct nested_vmx {
 336         /* Has the level1 guest done vmxon? */
 337         bool vmxon;
 338
 339         /* The guest-physical address of the current VMCS L1 keeps for L2 */
 340         gpa_t current_vmptr;
 341         /* The host-usable pointer to the above */
 342         struct page *current_vmcs12_page;
 343         struct vmcs12 *current_vmcs12;
 344
 345         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
 346         struct list_head vmcs02_pool;
 347         int vmcs02_num;
 348         u64 vmcs01_tsc_offset;
 349         /* L2 must run next, and mustn't decide to exit to L1. */
 350         bool nested_run_pending;
 351         /*
 352          * Guest pages referred to in vmcs02 with host-physical pointers, so
 353          * we must keep them pinned while L2 runs.
 354          */
 355         struct page *apic_access_page;
 356 };
 357
 358 struct vcpu_vmx {
 359         struct kvm_vcpu       vcpu;
 360         unsigned long         host_rsp;
 361         u8                    fail;
 362         u8                    cpl;
 363         bool                  nmi_known_unmasked;
 364         u32                   exit_intr_info;
 365         u32                   idt_vectoring_info;
 366         ulong                 rflags;
 367         struct shared_msr_entry *guest_msrs;
 368         int                   nmsrs;
 369         int                   save_nmsrs;
 370 #ifdef CONFIG_X86_64
 371         u64                   msr_host_kernel_gs_base;
 372         u64                   msr_guest_kernel_gs_base;
 373 #endif
 374         /*
 375          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 376          * non-nested (L1) guest, it always points to vmcs01. For a nested
 377          * guest (L2), it points to a different VMCS.
 378          */
 379         struct loaded_vmcs    vmcs01;
 380         struct loaded_vmcs   *loaded_vmcs;
 381         bool                  __launched; /* temporary, used in vmx_vcpu_run */
 382         struct msr_autoload {
 383                 unsigned nr;
 384                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 385                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 386         } msr_autoload;
 387         struct {
 388                 int           loaded;
 389                 u16           fs_sel, gs_sel, ldt_sel;
 390                 int           gs_ldt_reload_needed;
 391                 int           fs_reload_needed;
 392         } host_state;
 393         struct {
 394                 int vm86_active;
 395                 ulong save_rflags;
 396                 struct kvm_save_segment {
 397                         u16 selector;
 398                         unsigned long base;
 399                         u32 limit;
 400                         u32 ar;
 401                 } tr, es, ds, fs, gs;
 402         } rmode;
 403         struct {
 404                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 405                 struct kvm_save_segment seg[8];
 406         } segment_cache;
 407         int vpid;
 408         bool emulation_required;
 409
 410         /* Support for vnmi-less CPUs */
 411         int soft_vnmi_blocked;
 412         ktime_t entry_time;
 413         s64 vnmi_blocked_time;
 414         u32 exit_reason;
 415
 416         bool rdtscp_enabled;
 417
 418         /* Support for a guest hypervisor (nested VMX) */
 419         struct nested_vmx nested;
 420 };
 421
 422 enum segment_cache_field {
 423         SEG_FIELD_SEL = 0,
 424         SEG_FIELD_BASE = 1,
 425         SEG_FIELD_LIMIT = 2,
 426         SEG_FIELD_AR = 3,
 427
 428         SEG_FIELD_NR = 4
 429 };
 430
 431 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 432 {
 433         return container_of(vcpu, struct vcpu_vmx, vcpu);
 434 }
 435
 436 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
 437 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
 438 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
 439                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
 440
 441 static unsigned short vmcs_field_to_offset_table[] = {
 442         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 443         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
 444         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
 445         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
 446         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
 447         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
 448         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
 449         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
 450         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
 451         FIELD(HOST_ES_SELECTOR, host_es_selector),
 452         FIELD(HOST_CS_SELECTOR, host_cs_selector),
 453         FIELD(HOST_SS_SELECTOR, host_ss_selector),
 454         FIELD(HOST_DS_SELECTOR, host_ds_selector),
 455         FIELD(HOST_FS_SELECTOR, host_fs_selector),
 456         FIELD(HOST_GS_SELECTOR, host_gs_selector),
 457         FIELD(HOST_TR_SELECTOR, host_tr_selector),
 458         FIELD64(IO_BITMAP_A, io_bitmap_a),
 459         FIELD64(IO_BITMAP_B, io_bitmap_b),
 460         FIELD64(MSR_BITMAP, msr_bitmap),
 461         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
 462         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
 463         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
 464         FIELD64(TSC_OFFSET, tsc_offset),
 465         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
 466         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
 467         FIELD64(EPT_POINTER, ept_pointer),
 468         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
 469         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
 470         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
 471         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
 472         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
 473         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
 474         FIELD64(GUEST_PDPTR0, guest_pdptr0),
 475         FIELD64(GUEST_PDPTR1, guest_pdptr1),
 476         FIELD64(GUEST_PDPTR2, guest_pdptr2),
 477         FIELD64(GUEST_PDPTR3, guest_pdptr3),
 478         FIELD64(HOST_IA32_PAT, host_ia32_pat),
 479         FIELD64(HOST_IA32_EFER, host_ia32_efer),
 480         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
 481         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
 482         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
 483         FIELD(EXCEPTION_BITMAP, exception_bitmap),
 484         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
 485         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
 486         FIELD(CR3_TARGET_COUNT, cr3_target_count),
 487         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
 488         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
 489         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
 490         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
 491         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
 492         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
 493         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
 494         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
 495         FIELD(TPR_THRESHOLD, tpr_threshold),
 496         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
 497         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
 498         FIELD(VM_EXIT_REASON, vm_exit_reason),
 499         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
 500         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
 501         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
 502         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
 503         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
 504         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
 505         FIELD(GUEST_ES_LIMIT, guest_es_limit),
 506         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
 507         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
 508         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
 509         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
 510         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
 511         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
 512         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
 513         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
 514         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
 515         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
 516         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
 517         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
 518         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
 519         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
 520         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
 521         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
 522         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
 523         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
 524         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
 525         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
 526         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
 527         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
 528         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
 529         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
 530         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
 531         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
 532         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
 533         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
 534         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
 535         FIELD(EXIT_QUALIFICATION, exit_qualification),
 536         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
 537         FIELD(GUEST_CR0, guest_cr0),
 538         FIELD(GUEST_CR3, guest_cr3),
 539         FIELD(GUEST_CR4, guest_cr4),
 540         FIELD(GUEST_ES_BASE, guest_es_base),
 541         FIELD(GUEST_CS_BASE, guest_cs_base),
 542         FIELD(GUEST_SS_BASE, guest_ss_base),
 543         FIELD(GUEST_DS_BASE, guest_ds_base),
 544         FIELD(GUEST_FS_BASE, guest_fs_base),
 545         FIELD(GUEST_GS_BASE, guest_gs_base),
 546         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
 547         FIELD(GUEST_TR_BASE, guest_tr_base),
 548         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
 549         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
 550         FIELD(GUEST_DR7, guest_dr7),
 551         FIELD(GUEST_RSP, guest_rsp),
 552         FIELD(GUEST_RIP, guest_rip),
 553         FIELD(GUEST_RFLAGS, guest_rflags),
 554         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
 555         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
 556         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
 557         FIELD(HOST_CR0, host_cr0),
 558         FIELD(HOST_CR3, host_cr3),
 559         FIELD(HOST_CR4, host_cr4),
 560         FIELD(HOST_FS_BASE, host_fs_base),
 561         FIELD(HOST_GS_BASE, host_gs_base),
 562         FIELD(HOST_TR_BASE, host_tr_base),
 563         FIELD(HOST_GDTR_BASE, host_gdtr_base),
 564         FIELD(HOST_IDTR_BASE, host_idtr_base),
 565         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
 566         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
 567         FIELD(HOST_RSP, host_rsp),
 568         FIELD(HOST_RIP, host_rip),
 569 };
 570 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
 571
 572 static inline short vmcs_field_to_offset(unsigned long field)
 573 {
 574         if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
 575                 return -1;
 576         return vmcs_field_to_offset_table[field];
 577 }
 578
 579 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
 580 {
 581         return to_vmx(vcpu)->nested.current_vmcs12;
 582 }
 583
 584 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
 585 {
 586         struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
 587         if (is_error_page(page)) {
 588                 kvm_release_page_clean(page);
 589                 return NULL;
 590         }
 591         return page;
 592 }
 593
 594 static void nested_release_page(struct page *page)
 595 {
 596         kvm_release_page_dirty(page);
 597 }
 598
 599 static void nested_release_page_clean(struct page *page)
 600 {
 601         kvm_release_page_clean(page);
 602 }
 603
 604 static u64 construct_eptp(unsigned long root_hpa);
 605 static void kvm_cpu_vmxon(u64 addr);
 606 static void kvm_cpu_vmxoff(void);
 607 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
 608 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 609
 610 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 611 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 612 /*
 613  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
 614  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
 615  */
 616 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
 617 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 618
 619 static unsigned long *vmx_io_bitmap_a;
 620 static unsigned long *vmx_io_bitmap_b;
 621 static unsigned long *vmx_msr_bitmap_legacy;
 622 static unsigned long *vmx_msr_bitmap_longmode;
 623
 624 static bool cpu_has_load_ia32_efer;
 625
 626 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 627 static DEFINE_SPINLOCK(vmx_vpid_lock);
 628
 629 static struct vmcs_config {
 630         int size;
 631         int order;
 632         u32 revision_id;
 633         u32 pin_based_exec_ctrl;
 634         u32 cpu_based_exec_ctrl;
 635         u32 cpu_based_2nd_exec_ctrl;
 636         u32 vmexit_ctrl;
 637         u32 vmentry_ctrl;
 638 } vmcs_config;
 639
 640 static struct vmx_capability {
 641         u32 ept;
 642         u32 vpid;
 643 } vmx_capability;
 644
 645 #define VMX_SEGMENT_FIELD(seg)                                  \
 646         [VCPU_SREG_##seg] = {                                   \
 647                 .selector = GUEST_##seg##_SELECTOR,             \
 648                 .base = GUEST_##seg##_BASE,                     \
 649                 .limit = GUEST_##seg##_LIMIT,                   \
 650                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 651         }
 652
 653 static struct kvm_vmx_segment_field {
 654         unsigned selector;
 655         unsigned base;
 656         unsigned limit;
 657         unsigned ar_bytes;
 658 } kvm_vmx_segment_fields[] = {
 659         VMX_SEGMENT_FIELD(CS),
 660         VMX_SEGMENT_FIELD(DS),
 661         VMX_SEGMENT_FIELD(ES),
 662         VMX_SEGMENT_FIELD(FS),
 663         VMX_SEGMENT_FIELD(GS),
 664         VMX_SEGMENT_FIELD(SS),
 665         VMX_SEGMENT_FIELD(TR),
 666         VMX_SEGMENT_FIELD(LDTR),
 667 };
 668
 669 static u64 host_efer;
 670
 671 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 672
 673 /*
 674  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
 675  * away by decrementing the array size.
 676  */
 677 static const u32 vmx_msr_index[] = {
 678 #ifdef CONFIG_X86_64
 679         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 680 #endif
 681         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
 682 };
 683 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 684
 685 static inline bool is_page_fault(u32 intr_info)
 686 {
 687         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 688                              INTR_INFO_VALID_MASK)) ==
 689                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
 690 }
 691
 692 static inline bool is_no_device(u32 intr_info)
 693 {
 694         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 695                              INTR_INFO_VALID_MASK)) ==
 696                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 697 }
 698
 699 static inline bool is_invalid_opcode(u32 intr_info)
 700 {
 701         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 702                              INTR_INFO_VALID_MASK)) ==
 703                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 704 }
 705
 706 static inline bool is_external_interrupt(u32 intr_info)
 707 {
 708         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 709                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 710 }
 711
 712 static inline bool is_machine_check(u32 intr_info)
 713 {
 714         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 715                              INTR_INFO_VALID_MASK)) ==
 716                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
 717 }
 718
 719 static inline bool cpu_has_vmx_msr_bitmap(void)
 720 {
 721         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
 722 }
 723
 724 static inline bool cpu_has_vmx_tpr_shadow(void)
 725 {
 726         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
 727 }
 728
 729 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
 730 {
 731         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
 732 }
 733
 734 static inline bool cpu_has_secondary_exec_ctrls(void)
 735 {
 736         return vmcs_config.cpu_based_exec_ctrl &
 737                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
 738 }
 739
 740 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 741 {
 742         return vmcs_config.cpu_based_2nd_exec_ctrl &
 743                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
 744 }
 745
 746 static inline bool cpu_has_vmx_flexpriority(void)
 747 {
 748         return cpu_has_vmx_tpr_shadow() &&
 749                 cpu_has_vmx_virtualize_apic_accesses();
 750 }
 751
 752 static inline bool cpu_has_vmx_ept_execute_only(void)
 753 {
 754         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
 755 }
 756
 757 static inline bool cpu_has_vmx_eptp_uncacheable(void)
 758 {
 759         return vmx_capability.ept & VMX_EPTP_UC_BIT;
 760 }
 761
 762 static inline bool cpu_has_vmx_eptp_writeback(void)
 763 {
 764         return vmx_capability.ept & VMX_EPTP_WB_BIT;
 765 }
 766
 767 static inline bool cpu_has_vmx_ept_2m_page(void)
 768 {
 769         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
 770 }
 771
 772 static inline bool cpu_has_vmx_ept_1g_page(void)
 773 {
 774         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
 775 }
 776
 777 static inline bool cpu_has_vmx_ept_4levels(void)
 778 {
 779         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
 780 }
 781
 782 static inline bool cpu_has_vmx_invept_individual_addr(void)
 783 {
 784         return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
 785 }
 786
 787 static inline bool cpu_has_vmx_invept_context(void)
 788 {
 789         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
 790 }
 791
 792 static inline bool cpu_has_vmx_invept_global(void)
 793 {
 794         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
 795 }
 796
 797 static inline bool cpu_has_vmx_invvpid_single(void)
 798 {
 799         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
 800 }
 801
 802 static inline bool cpu_has_vmx_invvpid_global(void)
 803 {
 804         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
 805 }
 806
 807 static inline bool cpu_has_vmx_ept(void)
 808 {
 809         return vmcs_config.cpu_based_2nd_exec_ctrl &
 810                 SECONDARY_EXEC_ENABLE_EPT;
 811 }
 812
 813 static inline bool cpu_has_vmx_unrestricted_guest(void)
 814 {
 815         return vmcs_config.cpu_based_2nd_exec_ctrl &
 816                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
 817 }
 818
 819 static inline bool cpu_has_vmx_ple(void)
 820 {
 821         return vmcs_config.cpu_based_2nd_exec_ctrl &
 822                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 823 }
 824
 825 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
 826 {
 827         return flexpriority_enabled && irqchip_in_kernel(kvm);
 828 }
 829
 830 static inline bool cpu_has_vmx_vpid(void)
 831 {
 832         return vmcs_config.cpu_based_2nd_exec_ctrl &
 833                 SECONDARY_EXEC_ENABLE_VPID;
 834 }
 835
 836 static inline bool cpu_has_vmx_rdtscp(void)
 837 {
 838         return vmcs_config.cpu_based_2nd_exec_ctrl &
 839                 SECONDARY_EXEC_RDTSCP;
 840 }
 841
 842 static inline bool cpu_has_virtual_nmis(void)
 843 {
 844         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
 845 }
 846
 847 static inline bool cpu_has_vmx_wbinvd_exit(void)
 848 {
 849         return vmcs_config.cpu_based_2nd_exec_ctrl &
 850                 SECONDARY_EXEC_WBINVD_EXITING;
 851 }
 852
 853 static inline bool report_flexpriority(void)
 854 {
 855         return flexpriority_enabled;
 856 }
 857
 858 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
 859 {
 860         return vmcs12->cpu_based_vm_exec_control & bit;
 861 }
 862
 863 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
 864 {
 865         return (vmcs12->cpu_based_vm_exec_control &
 866                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
 867                 (vmcs12->secondary_vm_exec_control & bit);
 868 }
 869
 870 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
 871         struct kvm_vcpu *vcpu)
 872 {
 873         return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
 874 }
 875
 876 static inline bool is_exception(u32 intr_info)
 877 {
 878         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 879                 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
 880 }
 881
 882 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
 883 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
 884                         struct vmcs12 *vmcs12,
 885                         u32 reason, unsigned long qualification);
 886
 887 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
 888 {
 889         int i;
 890
 891         for (i = 0; i < vmx->nmsrs; ++i)
 892                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
 893                         return i;
 894         return -1;
 895 }
 896
 897 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
 898 {
 899     struct {
 900         u64 vpid : 16;
 901         u64 rsvd : 48;
 902         u64 gva;
 903     } operand = { vpid, 0, gva };
 904
 905     asm volatile (__ex(ASM_VMX_INVVPID)
 906                   /* CF==1 or ZF==1 --> rc = -1 */
 907                   "; ja 1f ; ud2 ; 1:"
 908                   : : "a"(&operand), "c"(ext) : "cc", "memory");
 909 }
 910
 911 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
 912 {
 913         struct {
 914                 u64 eptp, gpa;
 915         } operand = {eptp, gpa};
 916
 917         asm volatile (__ex(ASM_VMX_INVEPT)
 918                         /* CF==1 or ZF==1 --> rc = -1 */
 919                         "; ja 1f ; ud2 ; 1:\n"
 920                         : : "a" (&operand), "c" (ext) : "cc", "memory");
 921 }
 922
 923 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
 924 {
 925         int i;
 926
 927         i = __find_msr_index(vmx, msr);
 928         if (i >= 0)
 929                 return &vmx->guest_msrs[i];
 930         return NULL;
 931 }
 932
 933 static void vmcs_clear(struct vmcs *vmcs)
 934 {
 935         u64 phys_addr = __pa(vmcs);
 936         u8 error;
 937
 938         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
 939                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
 940                       : "cc", "memory");
 941         if (error)
 942                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 943                        vmcs, phys_addr);
 944 }
 945
 946 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
 947 {
 948         vmcs_clear(loaded_vmcs->vmcs);
 949         loaded_vmcs->cpu = -1;
 950         loaded_vmcs->launched = 0;
 951 }
 952
 953 static void vmcs_load(struct vmcs *vmcs)
 954 {
 955         u64 phys_addr = __pa(vmcs);
 956         u8 error;
 957
 958         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
 959                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
 960                         : "cc", "memory");
 961         if (error)
 962                 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
 963                        vmcs, phys_addr);
 964 }
 965
 966 static void __loaded_vmcs_clear(void *arg)
 967 {
 968         struct loaded_vmcs *loaded_vmcs = arg;
 969         int cpu = raw_smp_processor_id();
 970
 971         if (loaded_vmcs->cpu != cpu)
 972                 return; /* vcpu migration can race with cpu offline */
 973         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
 974                 per_cpu(current_vmcs, cpu) = NULL;
 975         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
 976         loaded_vmcs_init(loaded_vmcs);
 977 }
 978
 979 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
 980 {
 981         if (loaded_vmcs->cpu != -1)
 982                 smp_call_function_single(
 983                         loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
 984 }
 985
 986 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
 987 {
 988         if (vmx->vpid == 0)
 989                 return;
 990
 991         if (cpu_has_vmx_invvpid_single())
 992                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
 993 }
 994
 995 static inline void vpid_sync_vcpu_global(void)
 996 {
 997         if (cpu_has_vmx_invvpid_global())
 998                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
 999 }
1000
1001 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1002 {
1003         if (cpu_has_vmx_invvpid_single())
1004                 vpid_sync_vcpu_single(vmx);
1005         else
1006                 vpid_sync_vcpu_global();
1007 }
1008
1009 static inline void ept_sync_global(void)
1010 {
1011         if (cpu_has_vmx_invept_global())
1012                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1013 }
1014
1015 static inline void ept_sync_context(u64 eptp)
1016 {
1017         if (enable_ept) {
1018                 if (cpu_has_vmx_invept_context())
1019                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1020                 else
1021                         ept_sync_global();
1022         }
1023 }
1024
1025 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
1026 {
1027         if (enable_ept) {
1028                 if (cpu_has_vmx_invept_individual_addr())
1029                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
1030                                         eptp, gpa);
1031                 else
1032                         ept_sync_context(eptp);
1033         }
1034 }
1035
1036 static __always_inline unsigned long vmcs_readl(unsigned long field)
1037 {
1038         unsigned long value;
1039
1040         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1041                       : "=a"(value) : "d"(field) : "cc");
1042         return value;
1043 }
1044
1045 static __always_inline u16 vmcs_read16(unsigned long field)
1046 {
1047         return vmcs_readl(field);
1048 }
1049
1050 static __always_inline u32 vmcs_read32(unsigned long field)
1051 {
1052         return vmcs_readl(field);
1053 }
1054
1055 static __always_inline u64 vmcs_read64(unsigned long field)
1056 {
1057 #ifdef CONFIG_X86_64
1058         return vmcs_readl(field);
1059 #else
1060         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1061 #endif
1062 }
1063
1064 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1065 {
1066         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1067                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1068         dump_stack();
1069 }
1070
1071 static void vmcs_writel(unsigned long field, unsigned long value)
1072 {
1073         u8 error;
1074
1075         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1076                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1077         if (unlikely(error))
1078                 vmwrite_error(field, value);
1079 }
1080
1081 static void vmcs_write16(unsigned long field, u16 value)
1082 {
1083         vmcs_writel(field, value);
1084 }
1085
1086 static void vmcs_write32(unsigned long field, u32 value)
1087 {
1088         vmcs_writel(field, value);
1089 }
1090
1091 static void vmcs_write64(unsigned long field, u64 value)
1092 {
1093         vmcs_writel(field, value);
1094 #ifndef CONFIG_X86_64
1095         asm volatile ("");
1096         vmcs_writel(field+1, value >> 32);
1097 #endif
1098 }
1099
1100 static void vmcs_clear_bits(unsigned long field, u32 mask)
1101 {
1102         vmcs_writel(field, vmcs_readl(field) & ~mask);
1103 }
1104
1105 static void vmcs_set_bits(unsigned long field, u32 mask)
1106 {
1107         vmcs_writel(field, vmcs_readl(field) | mask);
1108 }
1109
1110 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1111 {
1112         vmx->segment_cache.bitmask = 0;
1113 }
1114
1115 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1116                                        unsigned field)
1117 {
1118         bool ret;
1119         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1120
1121         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1122                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1123                 vmx->segment_cache.bitmask = 0;
1124         }
1125         ret = vmx->segment_cache.bitmask & mask;
1126         vmx->segment_cache.bitmask |= mask;
1127         return ret;
1128 }
1129
1130 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1131 {
1132         u16 *p = &vmx->segment_cache.seg[seg].selector;
1133
1134         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1135                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1136         return *p;
1137 }
1138
1139 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1140 {
1141         ulong *p = &vmx->segment_cache.seg[seg].base;
1142
1143         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1144                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1145         return *p;
1146 }
1147
1148 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1149 {
1150         u32 *p = &vmx->segment_cache.seg[seg].limit;
1151
1152         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1153                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1154         return *p;
1155 }
1156
1157 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1158 {
1159         u32 *p = &vmx->segment_cache.seg[seg].ar;
1160
1161         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1162                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1163         return *p;
1164 }
1165
1166 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1167 {
1168         u32 eb;
1169
1170         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1171              (1u << NM_VECTOR) | (1u << DB_VECTOR);
1172         if ((vcpu->guest_debug &
1173              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1174             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1175                 eb |= 1u << BP_VECTOR;
1176         if (to_vmx(vcpu)->rmode.vm86_active)
1177                 eb = ~0;
1178         if (enable_ept)
1179                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1180         if (vcpu->fpu_active)
1181                 eb &= ~(1u << NM_VECTOR);
1182
1183         /* When we are running a nested L2 guest and L1 specified for it a
1184          * certain exception bitmap, we must trap the same exceptions and pass
1185          * them to L1. When running L2, we will only handle the exceptions
1186          * specified above if L1 did not want them.
1187          */
1188         if (is_guest_mode(vcpu))
1189                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1190
1191         vmcs_write32(EXCEPTION_BITMAP, eb);
1192 }
1193
1194 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1195 {
1196         unsigned i;
1197         struct msr_autoload *m = &vmx->msr_autoload;
1198
1199         if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
1200                 vmcs_clear_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
1201                 vmcs_clear_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
1202                 return;
1203         }
1204
1205         for (i = 0; i < m->nr; ++i)
1206                 if (m->guest[i].index == msr)
1207                         break;
1208
1209         if (i == m->nr)
1210                 return;
1211         --m->nr;
1212         m->guest[i] = m->guest[m->nr];
1213         m->host[i] = m->host[m->nr];
1214         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1215         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1216 }
1217
1218 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1219                                   u64 guest_val, u64 host_val)
1220 {
1221         unsigned i;
1222         struct msr_autoload *m = &vmx->msr_autoload;
1223
1224         if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
1225                 vmcs_write64(GUEST_IA32_EFER, guest_val);
1226                 vmcs_write64(HOST_IA32_EFER, host_val);
1227                 vmcs_set_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
1228                 vmcs_set_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
1229                 return;
1230         }
1231
1232         for (i = 0; i < m->nr; ++i)
1233                 if (m->guest[i].index == msr)
1234                         break;
1235
1236         if (i == m->nr) {
1237                 ++m->nr;
1238                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1239                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1240         }
1241
1242         m->guest[i].index = msr;
1243         m->guest[i].value = guest_val;
1244         m->host[i].index = msr;
1245         m->host[i].value = host_val;
1246 }
1247
1248 static void reload_tss(void)
1249 {
1250         /*
1251          * VT restores TR but not its size.  Useless.
1252          */
1253         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1254         struct desc_struct *descs;
1255
1256         descs = (void *)gdt->address;
1257         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1258         load_TR_desc();
1259 }
1260
1261 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1262 {
1263         u64 guest_efer;
1264         u64 ignore_bits;
1265
1266         guest_efer = vmx->vcpu.arch.efer;
1267
1268         /*
1269          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
1270          * outside long mode
1271          */
1272         ignore_bits = EFER_NX | EFER_SCE;
1273 #ifdef CONFIG_X86_64
1274         ignore_bits |= EFER_LMA | EFER_LME;
1275         /* SCE is meaningful only in long mode on Intel */
1276         if (guest_efer & EFER_LMA)
1277                 ignore_bits &= ~(u64)EFER_SCE;
1278 #endif
1279         guest_efer &= ~ignore_bits;
1280         guest_efer |= host_efer & ignore_bits;
1281         vmx->guest_msrs[efer_offset].data = guest_efer;
1282         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1283
1284         clear_atomic_switch_msr(vmx, MSR_EFER);
1285         /* On ept, can't emulate nx, and must switch nx atomically */
1286         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1287                 guest_efer = vmx->vcpu.arch.efer;
1288                 if (!(guest_efer & EFER_LMA))
1289                         guest_efer &= ~EFER_LME;
1290                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1291                 return false;
1292         }
1293
1294         return true;
1295 }
1296
1297 static unsigned long segment_base(u16 selector)
1298 {
1299         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1300         struct desc_struct *d;
1301         unsigned long table_base;
1302         unsigned long v;
1303
1304         if (!(selector & ~3))
1305                 return 0;
1306
1307         table_base = gdt->address;
1308
1309         if (selector & 4) {           /* from ldt */
1310                 u16 ldt_selector = kvm_read_ldt();
1311
1312                 if (!(ldt_selector & ~3))
1313                         return 0;
1314
1315                 table_base = segment_base(ldt_selector);
1316         }
1317         d = (struct desc_struct *)(table_base + (selector & ~7));
1318         v = get_desc_base(d);
1319 #ifdef CONFIG_X86_64
1320        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1321                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1322 #endif
1323         return v;
1324 }
1325
1326 static inline unsigned long kvm_read_tr_base(void)
1327 {
1328         u16 tr;
1329         asm("str %0" : "=g"(tr));
1330         return segment_base(tr);
1331 }
1332
1333 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1334 {
1335         struct vcpu_vmx *vmx = to_vmx(vcpu);
1336         int i;
1337
1338         if (vmx->host_state.loaded)
1339                 return;
1340
1341         vmx->host_state.loaded = 1;
1342         /*
1343          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1344          * allow segment selectors with cpl > 0 or ti == 1.
1345          */
1346         vmx->host_state.ldt_sel = kvm_read_ldt();
1347         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1348         savesegment(fs, vmx->host_state.fs_sel);
1349         if (!(vmx->host_state.fs_sel & 7)) {
1350                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1351                 vmx->host_state.fs_reload_needed = 0;
1352         } else {
1353                 vmcs_write16(HOST_FS_SELECTOR, 0);
1354                 vmx->host_state.fs_reload_needed = 1;
1355         }
1356         savesegment(gs, vmx->host_state.gs_sel);
1357         if (!(vmx->host_state.gs_sel & 7))
1358                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1359         else {
1360                 vmcs_write16(HOST_GS_SELECTOR, 0);
1361                 vmx->host_state.gs_ldt_reload_needed = 1;
1362         }
1363
1364 #ifdef CONFIG_X86_64
1365         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1366         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1367 #else
1368         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1369         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1370 #endif
1371
1372 #ifdef CONFIG_X86_64
1373         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1374         if (is_long_mode(&vmx->vcpu))
1375                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1376 #endif
1377         for (i = 0; i < vmx->save_nmsrs; ++i)
1378                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1379                                    vmx->guest_msrs[i].data,
1380                                    vmx->guest_msrs[i].mask);
1381 }
1382
1383 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1384 {
1385         if (!vmx->host_state.loaded)
1386                 return;
1387
1388         ++vmx->vcpu.stat.host_state_reload;
1389         vmx->host_state.loaded = 0;
1390 #ifdef CONFIG_X86_64
1391         if (is_long_mode(&vmx->vcpu))
1392                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1393 #endif
1394         if (vmx->host_state.gs_ldt_reload_needed) {
1395                 kvm_load_ldt(vmx->host_state.ldt_sel);
1396 #ifdef CONFIG_X86_64
1397                 load_gs_index(vmx->host_state.gs_sel);
1398 #else
1399                 loadsegment(gs, vmx->host_state.gs_sel);
1400 #endif
1401         }
1402         if (vmx->host_state.fs_reload_needed)
1403                 loadsegment(fs, vmx->host_state.fs_sel);
1404         reload_tss();
1405 #ifdef CONFIG_X86_64
1406         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1407 #endif
1408         if (current_thread_info()->status & TS_USEDFPU)
1409                 clts();
1410         load_gdt(&__get_cpu_var(host_gdt));
1411 }
1412
1413 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1414 {
1415         preempt_disable();
1416         __vmx_load_host_state(vmx);
1417         preempt_enable();
1418 }
1419
1420 /*
1421  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1422  * vcpu mutex is already taken.
1423  */
1424 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1425 {
1426         struct vcpu_vmx *vmx = to_vmx(vcpu);
1427         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1428
1429         if (!vmm_exclusive)
1430                 kvm_cpu_vmxon(phys_addr);
1431         else if (vmx->loaded_vmcs->cpu != cpu)
1432                 loaded_vmcs_clear(vmx->loaded_vmcs);
1433
1434         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1435                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1436                 vmcs_load(vmx->loaded_vmcs->vmcs);
1437         }
1438
1439         if (vmx->loaded_vmcs->cpu != cpu) {
1440                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1441                 unsigned long sysenter_esp;
1442
1443                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1444                 local_irq_disable();
1445                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1446                          &per_cpu(loaded_vmcss_on_cpu, cpu));
1447                 local_irq_enable();
1448
1449                 /*
1450                  * Linux uses per-cpu TSS and GDT, so set these when switching
1451                  * processors.
1452                  */
1453                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1454                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
1455
1456                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1457                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1458                 vmx->loaded_vmcs->cpu = cpu;
1459         }
1460 }
1461
1462 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1463 {
1464         __vmx_load_host_state(to_vmx(vcpu));
1465         if (!vmm_exclusive) {
1466                 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1467                 vcpu->cpu = -1;
1468                 kvm_cpu_vmxoff();
1469         }
1470 }
1471
1472 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1473 {
1474         ulong cr0;
1475
1476         if (vcpu->fpu_active)
1477                 return;
1478         vcpu->fpu_active = 1;
1479         cr0 = vmcs_readl(GUEST_CR0);
1480         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1481         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1482         vmcs_writel(GUEST_CR0, cr0);
1483         update_exception_bitmap(vcpu);
1484         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1485         if (is_guest_mode(vcpu))
1486                 vcpu->arch.cr0_guest_owned_bits &=
1487                         ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1488         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1489 }
1490
1491 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1492
1493 /*
1494  * Return the cr0 value that a nested guest would read. This is a combination
1495  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1496  * its hypervisor (cr0_read_shadow).
1497  */
1498 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1499 {
1500         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1501                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1502 }
1503 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1504 {
1505         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1506                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1507 }
1508
1509 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1510 {
1511         /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1512          * set this *before* calling this function.
1513          */
1514         vmx_decache_cr0_guest_bits(vcpu);
1515         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1516         update_exception_bitmap(vcpu);
1517         vcpu->arch.cr0_guest_owned_bits = 0;
1518         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1519         if (is_guest_mode(vcpu)) {
1520                 /*
1521                  * L1's specified read shadow might not contain the TS bit,
1522                  * so now that we turned on shadowing of this bit, we need to
1523                  * set this bit of the shadow. Like in nested_vmx_run we need
1524                  * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1525                  * up-to-date here because we just decached cr0.TS (and we'll
1526                  * only update vmcs12->guest_cr0 on nested exit).
1527                  */
1528                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1529                 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1530                         (vcpu->arch.cr0 & X86_CR0_TS);
1531                 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1532         } else
1533                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1534 }
1535
1536 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1537 {
1538         unsigned long rflags, save_rflags;
1539
1540         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1541                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1542                 rflags = vmcs_readl(GUEST_RFLAGS);
1543                 if (to_vmx(vcpu)->rmode.vm86_active) {
1544                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1545                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1546                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1547                 }
1548                 to_vmx(vcpu)->rflags = rflags;
1549         }
1550         return to_vmx(vcpu)->rflags;
1551 }
1552
1553 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1554 {
1555         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1556         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1557         to_vmx(vcpu)->rflags = rflags;
1558         if (to_vmx(vcpu)->rmode.vm86_active) {
1559                 to_vmx(vcpu)->rmode.save_rflags = rflags;
1560                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1561         }
1562         vmcs_writel(GUEST_RFLAGS, rflags);
1563 }
1564
1565 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1566 {
1567         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1568         int ret = 0;
1569
1570         if (interruptibility & GUEST_INTR_STATE_STI)
1571                 ret |= KVM_X86_SHADOW_INT_STI;
1572         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1573                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1574
1575         return ret & mask;
1576 }
1577
1578 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1579 {
1580         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1581         u32 interruptibility = interruptibility_old;
1582
1583         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1584
1585         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1586                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1587         else if (mask & KVM_X86_SHADOW_INT_STI)
1588                 interruptibility |= GUEST_INTR_STATE_STI;
1589
1590         if ((interruptibility != interruptibility_old))
1591                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1592 }
1593
1594 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1595 {
1596         unsigned long rip;
1597
1598         rip = kvm_rip_read(vcpu);
1599         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1600         kvm_rip_write(vcpu, rip);
1601
1602         /* skipping an emulated instruction also counts */
1603         vmx_set_interrupt_shadow(vcpu, 0);
1604 }
1605
1606 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1607 {
1608         /* Ensure that we clear the HLT state in the VMCS.  We don't need to
1609          * explicitly skip the instruction because if the HLT state is set, then
1610          * the instruction is already executing and RIP has already been
1611          * advanced. */
1612         if (!yield_on_hlt &&
1613             vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1614                 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1615 }
1616
1617 /*
1618  * KVM wants to inject page-faults which it got to the guest. This function
1619  * checks whether in a nested guest, we need to inject them to L1 or L2.
1620  * This function assumes it is called with the exit reason in vmcs02 being
1621  * a #PF exception (this is the only case in which KVM injects a #PF when L2
1622  * is running).
1623  */
1624 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1625 {
1626         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1627
1628         /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1629         if (!(vmcs12->exception_bitmap & PF_VECTOR))
1630                 return 0;
1631
1632         nested_vmx_vmexit(vcpu);
1633         return 1;
1634 }
1635
1636 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1637                                 bool has_error_code, u32 error_code,
1638                                 bool reinject)
1639 {
1640         struct vcpu_vmx *vmx = to_vmx(vcpu);
1641         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1642
1643         if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1644                 nested_pf_handled(vcpu))
1645                 return;
1646
1647         if (has_error_code) {
1648                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1649                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1650         }
1651
1652         if (vmx->rmode.vm86_active) {
1653                 int inc_eip = 0;
1654                 if (kvm_exception_is_soft(nr))
1655                         inc_eip = vcpu->arch.event_exit_inst_len;
1656                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1657                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1658                 return;
1659         }
1660
1661         if (kvm_exception_is_soft(nr)) {
1662                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1663                              vmx->vcpu.arch.event_exit_inst_len);
1664                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1665         } else
1666                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1667
1668         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1669         vmx_clear_hlt(vcpu);
1670 }
1671
1672 static bool vmx_rdtscp_supported(void)
1673 {
1674         return cpu_has_vmx_rdtscp();
1675 }
1676
1677 /*
1678  * Swap MSR entry in host/guest MSR entry array.
1679  */
1680 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1681 {
1682         struct shared_msr_entry tmp;
1683
1684         tmp = vmx->guest_msrs[to];
1685         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1686         vmx->guest_msrs[from] = tmp;
1687 }
1688
1689 /*
1690  * Set up the vmcs to automatically save and restore system
1691  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1692  * mode, as fiddling with msrs is very expensive.
1693  */
1694 static void setup_msrs(struct vcpu_vmx *vmx)
1695 {
1696         int save_nmsrs, index;
1697         unsigned long *msr_bitmap;
1698
1699         vmx_load_host_state(vmx);
1700         save_nmsrs = 0;
1701 #ifdef CONFIG_X86_64
1702         if (is_long_mode(&vmx->vcpu)) {
1703                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1704                 if (index >= 0)
1705                         move_msr_up(vmx, index, save_nmsrs++);
1706                 index = __find_msr_index(vmx, MSR_LSTAR);
1707                 if (index >= 0)
1708                         move_msr_up(vmx, index, save_nmsrs++);
1709                 index = __find_msr_index(vmx, MSR_CSTAR);
1710                 if (index >= 0)
1711                         move_msr_up(vmx, index, save_nmsrs++);
1712                 index = __find_msr_index(vmx, MSR_TSC_AUX);
1713                 if (index >= 0 && vmx->rdtscp_enabled)
1714                         move_msr_up(vmx, index, save_nmsrs++);
1715                 /*
1716                  * MSR_STAR is only needed on long mode guests, and only
1717                  * if efer.sce is enabled.
1718                  */
1719                 index = __find_msr_index(vmx, MSR_STAR);
1720                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1721                         move_msr_up(vmx, index, save_nmsrs++);
1722         }
1723 #endif
1724         index = __find_msr_index(vmx, MSR_EFER);
1725         if (index >= 0 && update_transition_efer(vmx, index))
1726                 move_msr_up(vmx, index, save_nmsrs++);
1727
1728         vmx->save_nmsrs = save_nmsrs;
1729
1730         if (cpu_has_vmx_msr_bitmap()) {
1731                 if (is_long_mode(&vmx->vcpu))
1732                         msr_bitmap = vmx_msr_bitmap_longmode;
1733                 else
1734                         msr_bitmap = vmx_msr_bitmap_legacy;
1735
1736                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1737         }
1738 }
1739
1740 /*
1741  * reads and returns guest's timestamp counter "register"
1742  * guest_tsc = host_tsc + tsc_offset    -- 21.3
1743  */
1744 static u64 guest_read_tsc(void)
1745 {
1746         u64 host_tsc, tsc_offset;
1747
1748         rdtscll(host_tsc);
1749         tsc_offset = vmcs_read64(TSC_OFFSET);
1750         return host_tsc + tsc_offset;
1751 }
1752
1753 /*
1754  * Empty call-back. Needs to be implemented when VMX enables the SET_TSC_KHZ
1755  * ioctl. In this case the call-back should update internal vmx state to make
1756  * the changes effective.
1757  */
1758 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1759 {
1760         /* Nothing to do here */
1761 }
1762
1763 /*
1764  * writes 'offset' into guest's timestamp counter offset register
1765  */
1766 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1767 {
1768         vmcs_write64(TSC_OFFSET, offset);
1769 }
1770
1771 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1772 {
1773         u64 offset = vmcs_read64(TSC_OFFSET);
1774         vmcs_write64(TSC_OFFSET, offset + adjustment);
1775 }
1776
1777 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1778 {
1779         return target_tsc - native_read_tsc();
1780 }
1781
1782 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1783 {
1784         struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1785         return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1786 }
1787
1788 /*
1789  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1790  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1791  * all guests if the "nested" module option is off, and can also be disabled
1792  * for a single guest by disabling its VMX cpuid bit.
1793  */
1794 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1795 {
1796         return nested && guest_cpuid_has_vmx(vcpu);
1797 }
1798
1799 /*
1800  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1801  * returned for the various VMX controls MSRs when nested VMX is enabled.
1802  * The same values should also be used to verify that vmcs12 control fields are
1803  * valid during nested entry from L1 to L2.
1804  * Each of these control msrs has a low and high 32-bit half: A low bit is on
1805  * if the corresponding bit in the (32-bit) control field *must* be on, and a
1806  * bit in the high half is on if the corresponding bit in the control field
1807  * may be on. See also vmx_control_verify().
1808  * TODO: allow these variables to be modified (downgraded) by module options
1809  * or other means.
1810  */
1811 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1812 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1813 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1814 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1815 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1816 static __init void nested_vmx_setup_ctls_msrs(void)
1817 {
1818         /*
1819          * Note that as a general rule, the high half of the MSRs (bits in
1820          * the control fields which may be 1) should be initialized by the
1821          * intersection of the underlying hardware's MSR (i.e., features which
1822          * can be supported) and the list of features we want to expose -
1823          * because they are known to be properly supported in our code.
1824          * Also, usually, the low half of the MSRs (bits which must be 1) can
1825          * be set to 0, meaning that L1 may turn off any of these bits. The
1826          * reason is that if one of these bits is necessary, it will appear
1827          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1828          * fields of vmcs01 and vmcs02, will turn these bits off - and
1829          * nested_vmx_exit_handled() will not pass related exits to L1.
1830          * These rules have exceptions below.
1831          */
1832
1833         /* pin-based controls */
1834         /*
1835          * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1836          * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1837          */
1838         nested_vmx_pinbased_ctls_low = 0x16 ;
1839         nested_vmx_pinbased_ctls_high = 0x16 |
1840                 PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1841                 PIN_BASED_VIRTUAL_NMIS;
1842
1843         /* exit controls */
1844         nested_vmx_exit_ctls_low = 0;
1845         /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
1846 #ifdef CONFIG_X86_64
1847         nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1848 #else
1849         nested_vmx_exit_ctls_high = 0;
1850 #endif
1851
1852         /* entry controls */
1853         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1854                 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1855         nested_vmx_entry_ctls_low = 0;
1856         nested_vmx_entry_ctls_high &=
1857                 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1858
1859         /* cpu-based controls */
1860         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1861                 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1862         nested_vmx_procbased_ctls_low = 0;
1863         nested_vmx_procbased_ctls_high &=
1864                 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1865                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1866                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1867                 CPU_BASED_CR3_STORE_EXITING |
1868 #ifdef CONFIG_X86_64
1869                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1870 #endif
1871                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1872                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1873                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1874         /*
1875          * We can allow some features even when not supported by the
1876          * hardware. For example, L1 can specify an MSR bitmap - and we
1877          * can use it to avoid exits to L1 - even when L0 runs L2
1878          * without MSR bitmaps.
1879          */
1880         nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
1881
1882         /* secondary cpu-based controls */
1883         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
1884                 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
1885         nested_vmx_secondary_ctls_low = 0;
1886         nested_vmx_secondary_ctls_high &=
1887                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1888 }
1889
1890 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
1891 {
1892         /*
1893          * Bits 0 in high must be 0, and bits 1 in low must be 1.
1894          */
1895         return ((control & high) | low) == control;
1896 }
1897
1898 static inline u64 vmx_control_msr(u32 low, u32 high)
1899 {
1900         return low | ((u64)high << 32);
1901 }
1902
1903 /*
1904  * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
1905  * also let it use VMX-specific MSRs.
1906  * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
1907  * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
1908  * like all other MSRs).
1909  */
1910 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1911 {
1912         if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
1913                      msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
1914                 /*
1915                  * According to the spec, processors which do not support VMX
1916                  * should throw a #GP(0) when VMX capability MSRs are read.
1917                  */
1918                 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
1919                 return 1;
1920         }
1921
1922         switch (msr_index) {
1923         case MSR_IA32_FEATURE_CONTROL:
1924                 *pdata = 0;
1925                 break;
1926         case MSR_IA32_VMX_BASIC:
1927                 /*
1928                  * This MSR reports some information about VMX support. We
1929                  * should return information about the VMX we emulate for the
1930                  * guest, and the VMCS structure we give it - not about the
1931                  * VMX support of the underlying hardware.
1932                  */
1933                 *pdata = VMCS12_REVISION |
1934                            ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
1935                            (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
1936                 break;
1937         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1938         case MSR_IA32_VMX_PINBASED_CTLS:
1939                 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
1940                                         nested_vmx_pinbased_ctls_high);
1941                 break;
1942         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1943         case MSR_IA32_VMX_PROCBASED_CTLS:
1944                 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
1945                                         nested_vmx_procbased_ctls_high);
1946                 break;
1947         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1948         case MSR_IA32_VMX_EXIT_CTLS:
1949                 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
1950                                         nested_vmx_exit_ctls_high);
1951                 break;
1952         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1953         case MSR_IA32_VMX_ENTRY_CTLS:
1954                 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
1955                                         nested_vmx_entry_ctls_high);
1956                 break;
1957         case MSR_IA32_VMX_MISC:
1958                 *pdata = 0;
1959                 break;
1960         /*
1961          * These MSRs specify bits which the guest must keep fixed (on or off)
1962          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
1963          * We picked the standard core2 setting.
1964          */
1965 #define VMXON_CR0_ALWAYSON      (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
1966 #define VMXON_CR4_ALWAYSON      X86_CR4_VMXE
1967         case MSR_IA32_VMX_CR0_FIXED0:
1968                 *pdata = VMXON_CR0_ALWAYSON;
1969                 break;
1970         case MSR_IA32_VMX_CR0_FIXED1:
1971                 *pdata = -1ULL;
1972                 break;
1973         case MSR_IA32_VMX_CR4_FIXED0:
1974                 *pdata = VMXON_CR4_ALWAYSON;
1975                 break;
1976         case MSR_IA32_VMX_CR4_FIXED1:
1977                 *pdata = -1ULL;
1978                 break;
1979         case MSR_IA32_VMX_VMCS_ENUM:
1980                 *pdata = 0x1f;
1981                 break;
1982         case MSR_IA32_VMX_PROCBASED_CTLS2:
1983                 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
1984                                         nested_vmx_secondary_ctls_high);
1985                 break;
1986         case MSR_IA32_VMX_EPT_VPID_CAP:
1987                 /* Currently, no nested ept or nested vpid */
1988                 *pdata = 0;
1989                 break;
1990         default:
1991                 return 0;
1992         }
1993
1994         return 1;
1995 }
1996
1997 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1998 {
1999         if (!nested_vmx_allowed(vcpu))
2000                 return 0;
2001
2002         if (msr_index == MSR_IA32_FEATURE_CONTROL)
2003                 /* TODO: the right thing. */
2004                 return 1;
2005         /*
2006          * No need to treat VMX capability MSRs specially: If we don't handle
2007          * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2008          */
2009         return 0;
2010 }
2011
2012 /*
2013  * Reads an msr value (of 'msr_index') into 'pdata'.
2014  * Returns 0 on success, non-0 otherwise.
2015  * Assumes vcpu_load() was already called.
2016  */
2017 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2018 {
2019         u64 data;
2020         struct shared_msr_entry *msr;
2021
2022         if (!pdata) {
2023                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2024                 return -EINVAL;
2025         }
2026
2027         switch (msr_index) {
2028 #ifdef CONFIG_X86_64
2029         case MSR_FS_BASE:
2030                 data = vmcs_readl(GUEST_FS_BASE);
2031                 break;
2032         case MSR_GS_BASE:
2033                 data = vmcs_readl(GUEST_GS_BASE);
2034                 break;
2035         case MSR_KERNEL_GS_BASE:
2036                 vmx_load_host_state(to_vmx(vcpu));
2037                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2038                 break;
2039 #endif
2040         case MSR_EFER:
2041                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2042         case MSR_IA32_TSC:
2043                 data = guest_read_tsc();
2044                 break;
2045         case MSR_IA32_SYSENTER_CS:
2046                 data = vmcs_read32(GUEST_SYSENTER_CS);
2047                 break;
2048         case MSR_IA32_SYSENTER_EIP:
2049                 data = vmcs_readl(GUEST_SYSENTER_EIP);
2050                 break;
2051         case MSR_IA32_SYSENTER_ESP:
2052                 data = vmcs_readl(GUEST_SYSENTER_ESP);
2053                 break;
2054         case MSR_TSC_AUX:
2055                 if (!to_vmx(vcpu)->rdtscp_enabled)
2056                         return 1;
2057                 /* Otherwise falls through */
2058         default:
2059                 vmx_load_host_state(to_vmx(vcpu));
2060                 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2061                         return 0;
2062                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2063                 if (msr) {
2064                         vmx_load_host_state(to_vmx(vcpu));
2065                         data = msr->data;
2066                         break;
2067                 }
2068                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2069         }
2070
2071         *pdata = data;
2072         return 0;
2073 }
2074
2075 /*
2076  * Writes msr value into into the appropriate "register".
2077  * Returns 0 on success, non-0 otherwise.
2078  * Assumes vcpu_load() was already called.
2079  */
2080 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2081 {
2082         struct vcpu_vmx *vmx = to_vmx(vcpu);
2083         struct shared_msr_entry *msr;
2084         int ret = 0;
2085
2086         switch (msr_index) {
2087         case MSR_EFER:
2088                 vmx_load_host_state(vmx);
2089                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2090                 break;
2091 #ifdef CONFIG_X86_64
2092         case MSR_FS_BASE:
2093                 vmx_segment_cache_clear(vmx);
2094                 vmcs_writel(GUEST_FS_BASE, data);
2095                 break;
2096         case MSR_GS_BASE:
2097                 vmx_segment_cache_clear(vmx);
2098                 vmcs_writel(GUEST_GS_BASE, data);
2099                 break;
2100         case MSR_KERNEL_GS_BASE:
2101                 vmx_load_host_state(vmx);
2102                 vmx->msr_guest_kernel_gs_base = data;
2103                 break;
2104 #endif
2105         case MSR_IA32_SYSENTER_CS:
2106                 vmcs_write32(GUEST_SYSENTER_CS, data);
2107                 break;
2108         case MSR_IA32_SYSENTER_EIP:
2109                 vmcs_writel(GUEST_SYSENTER_EIP, data);
2110                 break;
2111         case MSR_IA32_SYSENTER_ESP:
2112                 vmcs_writel(GUEST_SYSENTER_ESP, data);
2113                 break;
2114         case MSR_IA32_TSC:
2115                 kvm_write_tsc(vcpu, data);
2116                 break;
2117         case MSR_IA32_CR_PAT:
2118                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2119                         vmcs_write64(GUEST_IA32_PAT, data);
2120                         vcpu->arch.pat = data;
2121                         break;
2122                 }
2123                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2124                 break;
2125         case MSR_TSC_AUX:
2126                 if (!vmx->rdtscp_enabled)
2127                         return 1;
2128                 /* Check reserved bit, higher 32 bits should be zero */
2129                 if ((data >> 32) != 0)
2130                         return 1;
2131                 /* Otherwise falls through */
2132         default:
2133                 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2134                         break;
2135                 msr = find_msr_entry(vmx, msr_index);
2136                 if (msr) {
2137                         vmx_load_host_state(vmx);
2138                         msr->data = data;
2139                         break;
2140                 }
2141                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2142         }
2143
2144         return ret;
2145 }
2146
2147 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2148 {
2149         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2150         switch (reg) {
2151         case VCPU_REGS_RSP:
2152                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2153                 break;
2154         case VCPU_REGS_RIP:
2155                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2156                 break;
2157         case VCPU_EXREG_PDPTR:
2158                 if (enable_ept)
2159                         ept_save_pdptrs(vcpu);
2160                 break;
2161         default:
2162                 break;
2163         }
2164 }
2165
2166 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
2167 {
2168         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
2169                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
2170         else
2171                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2172
2173         update_exception_bitmap(vcpu);
2174 }
2175
2176 static __init int cpu_has_kvm_support(void)
2177 {
2178         return cpu_has_vmx();
2179 }
2180
2181 static __init int vmx_disabled_by_bios(void)
2182 {
2183         u64 msr;
2184
2185         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2186         if (msr & FEATURE_CONTROL_LOCKED) {
2187                 /* launched w/ TXT and VMX disabled */
2188                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2189                         && tboot_enabled())
2190                         return 1;
2191                 /* launched w/o TXT and VMX only enabled w/ TXT */
2192                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2193                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2194                         && !tboot_enabled()) {
2195                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2196                                 "activate TXT before enabling KVM\n");
2197                         return 1;
2198                 }
2199                 /* launched w/o TXT and VMX disabled */
2200                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2201                         && !tboot_enabled())
2202                         return 1;
2203         }
2204
2205         return 0;
2206 }
2207
2208 static void kvm_cpu_vmxon(u64 addr)
2209 {
2210         asm volatile (ASM_VMX_VMXON_RAX
2211                         : : "a"(&addr), "m"(addr)
2212                         : "memory", "cc");
2213 }
2214
2215 static int hardware_enable(void *garbage)
2216 {
2217         int cpu = raw_smp_processor_id();
2218         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2219         u64 old, test_bits;
2220
2221         if (read_cr4() & X86_CR4_VMXE)
2222                 return -EBUSY;
2223
2224         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2225         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2226
2227         test_bits = FEATURE_CONTROL_LOCKED;
2228         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2229         if (tboot_enabled())
2230                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2231
2232         if ((old & test_bits) != test_bits) {
2233                 /* enable and lock */
2234                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2235         }
2236         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2237
2238         if (vmm_exclusive) {
2239                 kvm_cpu_vmxon(phys_addr);
2240                 ept_sync_global();
2241         }
2242
2243         store_gdt(&__get_cpu_var(host_gdt));
2244
2245         return 0;
2246 }
2247
2248 static void vmclear_local_loaded_vmcss(void)
2249 {
2250         int cpu = raw_smp_processor_id();
2251         struct loaded_vmcs *v, *n;
2252
2253         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2254                                  loaded_vmcss_on_cpu_link)
2255                 __loaded_vmcs_clear(v);
2256 }
2257
2258
2259 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2260  * tricks.
2261  */
2262 static void kvm_cpu_vmxoff(void)
2263 {
2264         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2265 }
2266
2267 static void hardware_disable(void *garbage)
2268 {
2269         if (vmm_exclusive) {
2270                 vmclear_local_loaded_vmcss();
2271                 kvm_cpu_vmxoff();
2272         }
2273         write_cr4(read_cr4() & ~X86_CR4_VMXE);
2274 }
2275
2276 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2277                                       u32 msr, u32 *result)
2278 {
2279         u32 vmx_msr_low, vmx_msr_high;
2280         u32 ctl = ctl_min | ctl_opt;
2281
2282         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2283
2284         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2285         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2286
2287         /* Ensure minimum (required) set of control bits are supported. */
2288         if (ctl_min & ~ctl)
2289                 return -EIO;
2290
2291         *result = ctl;
2292         return 0;
2293 }
2294
2295 static __init bool allow_1_setting(u32 msr, u32 ctl)
2296 {
2297         u32 vmx_msr_low, vmx_msr_high;
2298
2299         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2300         return vmx_msr_high & ctl;
2301 }
2302
2303 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2304 {
2305         u32 vmx_msr_low, vmx_msr_high;
2306         u32 min, opt, min2, opt2;
2307         u32 _pin_based_exec_control = 0;
2308         u32 _cpu_based_exec_control = 0;
2309         u32 _cpu_based_2nd_exec_control = 0;
2310         u32 _vmexit_control = 0;
2311         u32 _vmentry_control = 0;
2312
2313         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2314         opt = PIN_BASED_VIRTUAL_NMIS;
2315         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2316                                 &_pin_based_exec_control) < 0)
2317                 return -EIO;
2318
2319         min =
2320 #ifdef CONFIG_X86_64
2321               CPU_BASED_CR8_LOAD_EXITING |
2322               CPU_BASED_CR8_STORE_EXITING |
2323 #endif
2324               CPU_BASED_CR3_LOAD_EXITING |
2325               CPU_BASED_CR3_STORE_EXITING |
2326               CPU_BASED_USE_IO_BITMAPS |
2327               CPU_BASED_MOV_DR_EXITING |
2328               CPU_BASED_USE_TSC_OFFSETING |
2329               CPU_BASED_MWAIT_EXITING |
2330               CPU_BASED_MONITOR_EXITING |
2331               CPU_BASED_INVLPG_EXITING;
2332
2333         if (yield_on_hlt)
2334                 min |= CPU_BASED_HLT_EXITING;
2335
2336         opt = CPU_BASED_TPR_SHADOW |
2337               CPU_BASED_USE_MSR_BITMAPS |
2338               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2339         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2340                                 &_cpu_based_exec_control) < 0)
2341                 return -EIO;
2342 #ifdef CONFIG_X86_64
2343         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2344                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2345                                            ~CPU_BASED_CR8_STORE_EXITING;
2346 #endif
2347         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2348                 min2 = 0;
2349                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2350                         SECONDARY_EXEC_WBINVD_EXITING |
2351                         SECONDARY_EXEC_ENABLE_VPID |
2352                         SECONDARY_EXEC_ENABLE_EPT |
2353                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
2354                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2355                         SECONDARY_EXEC_RDTSCP;
2356                 if (adjust_vmx_controls(min2, opt2,
2357                                         MSR_IA32_VMX_PROCBASED_CTLS2,
2358                                         &_cpu_based_2nd_exec_control) < 0)
2359                         return -EIO;
2360         }
2361 #ifndef CONFIG_X86_64
2362         if (!(_cpu_based_2nd_exec_control &
2363                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2364                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2365 #endif
2366         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2367                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2368                    enabled */
2369                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2370                                              CPU_BASED_CR3_STORE_EXITING |
2371                                              CPU_BASED_INVLPG_EXITING);
2372                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2373                       vmx_capability.ept, vmx_capability.vpid);
2374         }
2375
2376         min = 0;
2377 #ifdef CONFIG_X86_64
2378         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2379 #endif
2380         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2381         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2382                                 &_vmexit_control) < 0)
2383                 return -EIO;
2384
2385         min = 0;
2386         opt = VM_ENTRY_LOAD_IA32_PAT;
2387         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2388                                 &_vmentry_control) < 0)
2389                 return -EIO;
2390
2391         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2392
2393         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2394         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2395                 return -EIO;
2396
2397 #ifdef CONFIG_X86_64
2398         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2399         if (vmx_msr_high & (1u<<16))
2400                 return -EIO;
2401 #endif
2402
2403         /* Require Write-Back (WB) memory type for VMCS accesses. */
2404         if (((vmx_msr_high >> 18) & 15) != 6)
2405                 return -EIO;
2406
2407         vmcs_conf->size = vmx_msr_high & 0x1fff;
2408         vmcs_conf->order = get_order(vmcs_config.size);
2409         vmcs_conf->revision_id = vmx_msr_low;
2410
2411         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2412         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2413         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2414         vmcs_conf->vmexit_ctrl         = _vmexit_control;
2415         vmcs_conf->vmentry_ctrl        = _vmentry_control;
2416
2417         cpu_has_load_ia32_efer =
2418                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2419                                 VM_ENTRY_LOAD_IA32_EFER)
2420                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2421                                    VM_EXIT_LOAD_IA32_EFER);
2422
2423         return 0;
2424 }
2425
2426 static struct vmcs *alloc_vmcs_cpu(int cpu)
2427 {
2428         int node = cpu_to_node(cpu);
2429         struct page *pages;
2430         struct vmcs *vmcs;
2431
2432         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2433         if (!pages)
2434                 return NULL;
2435         vmcs = page_address(pages);
2436         memset(vmcs, 0, vmcs_config.size);
2437         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2438         return vmcs;
2439 }
2440
2441 static struct vmcs *alloc_vmcs(void)
2442 {
2443         return alloc_vmcs_cpu(raw_smp_processor_id());
2444 }
2445
2446 static void free_vmcs(struct vmcs *vmcs)
2447 {
2448         free_pages((unsigned long)vmcs, vmcs_config.order);
2449 }
2450
2451 /*
2452  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2453  */
2454 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2455 {
2456         if (!loaded_vmcs->vmcs)
2457                 return;
2458         loaded_vmcs_clear(loaded_vmcs);
2459         free_vmcs(loaded_vmcs->vmcs);
2460         loaded_vmcs->vmcs = NULL;
2461 }
2462
2463 static void free_kvm_area(void)
2464 {
2465         int cpu;
2466
2467         for_each_possible_cpu(cpu) {
2468                 free_vmcs(per_cpu(vmxarea, cpu));
2469                 per_cpu(vmxarea, cpu) = NULL;
2470         }
2471 }
2472
2473 static __init int alloc_kvm_area(void)
2474 {
2475         int cpu;
2476
2477         for_each_possible_cpu(cpu) {
2478                 struct vmcs *vmcs;
2479
2480                 vmcs = alloc_vmcs_cpu(cpu);
2481                 if (!vmcs) {
2482                         free_kvm_area();
2483                         return -ENOMEM;
2484                 }
2485
2486                 per_cpu(vmxarea, cpu) = vmcs;
2487         }
2488         return 0;
2489 }
2490
2491 static __init int hardware_setup(void)
2492 {
2493         if (setup_vmcs_config(&vmcs_config) < 0)
2494                 return -EIO;
2495
2496         if (boot_cpu_has(X86_FEATURE_NX))
2497                 kvm_enable_efer_bits(EFER_NX);
2498
2499         if (!cpu_has_vmx_vpid())
2500                 enable_vpid = 0;
2501
2502         if (!cpu_has_vmx_ept() ||
2503             !cpu_has_vmx_ept_4levels()) {
2504                 enable_ept = 0;
2505                 enable_unrestricted_guest = 0;
2506         }
2507
2508         if (!cpu_has_vmx_unrestricted_guest())
2509                 enable_unrestricted_guest = 0;
2510
2511         if (!cpu_has_vmx_flexpriority())
2512                 flexpriority_enabled = 0;
2513
2514         if (!cpu_has_vmx_tpr_shadow())
2515                 kvm_x86_ops->update_cr8_intercept = NULL;
2516
2517         if (enable_ept && !cpu_has_vmx_ept_2m_page())
2518                 kvm_disable_largepages();
2519
2520         if (!cpu_has_vmx_ple())
2521                 ple_gap = 0;
2522
2523         if (nested)
2524                 nested_vmx_setup_ctls_msrs();
2525
2526         return alloc_kvm_area();
2527 }
2528
2529 static __exit void hardware_unsetup(void)
2530 {
2531         free_kvm_area();
2532 }
2533
2534 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
2535 {
2536         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2537
2538         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
2539                 vmcs_write16(sf->selector, save->selector);
2540                 vmcs_writel(sf->base, save->base);
2541                 vmcs_write32(sf->limit, save->limit);
2542                 vmcs_write32(sf->ar_bytes, save->ar);
2543         } else {
2544                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
2545                         << AR_DPL_SHIFT;
2546                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
2547         }
2548 }
2549
2550 static void enter_pmode(struct kvm_vcpu *vcpu)
2551 {
2552         unsigned long flags;
2553         struct vcpu_vmx *vmx = to_vmx(vcpu);
2554
2555         vmx->emulation_required = 1;
2556         vmx->rmode.vm86_active = 0;
2557
2558         vmx_segment_cache_clear(vmx);
2559
2560         vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
2561         vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
2562         vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
2563         vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
2564
2565         flags = vmcs_readl(GUEST_RFLAGS);
2566         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2567         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2568         vmcs_writel(GUEST_RFLAGS, flags);
2569
2570         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2571                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2572
2573         update_exception_bitmap(vcpu);
2574
2575         if (emulate_invalid_guest_state)
2576                 return;
2577
2578         fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
2579         fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
2580         fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
2581         fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
2582
2583         vmx_segment_cache_clear(vmx);
2584
2585         vmcs_write16(GUEST_SS_SELECTOR, 0);
2586         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2587
2588         vmcs_write16(GUEST_CS_SELECTOR,
2589                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2590         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2591 }
2592
2593 static gva_t rmode_tss_base(struct kvm *kvm)
2594 {
2595         if (!kvm->arch.tss_addr) {
2596                 struct kvm_memslots *slots;
2597                 gfn_t base_gfn;
2598
2599                 slots = kvm_memslots(kvm);
2600                 base_gfn = slots->memslots[0].base_gfn +
2601                                  kvm->memslots->memslots[0].npages - 3;
2602                 return base_gfn << PAGE_SHIFT;
2603         }
2604         return kvm->arch.tss_addr;
2605 }
2606
2607 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
2608 {
2609         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2610
2611         save->selector = vmcs_read16(sf->selector);
2612         save->base = vmcs_readl(sf->base);
2613         save->limit = vmcs_read32(sf->limit);
2614         save->ar = vmcs_read32(sf->ar_bytes);
2615         vmcs_write16(sf->selector, save->base >> 4);
2616         vmcs_write32(sf->base, save->base & 0xffff0);
2617         vmcs_write32(sf->limit, 0xffff);
2618         vmcs_write32(sf->ar_bytes, 0xf3);
2619         if (save->base & 0xf)
2620                 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2621                             " aligned when entering protected mode (seg=%d)",
2622                             seg);
2623 }
2624
2625 static void enter_rmode(struct kvm_vcpu *vcpu)
2626 {
2627         unsigned long flags;
2628         struct vcpu_vmx *vmx = to_vmx(vcpu);
2629
2630         if (enable_unrestricted_guest)
2631                 return;
2632
2633         vmx->emulation_required = 1;
2634         vmx->rmode.vm86_active = 1;
2635
2636         /*
2637          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2638          * vcpu. Call it here with phys address pointing 16M below 4G.
2639          */
2640         if (!vcpu->kvm->arch.tss_addr) {
2641                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2642                              "called before entering vcpu\n");
2643                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2644                 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2645                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2646         }
2647
2648         vmx_segment_cache_clear(vmx);
2649
2650         vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
2651         vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
2652         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
2653
2654         vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
2655         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2656
2657         vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
2658         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2659
2660         flags = vmcs_readl(GUEST_RFLAGS);
2661         vmx->rmode.save_rflags = flags;
2662
2663         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2664
2665         vmcs_writel(GUEST_RFLAGS, flags);
2666         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2667         update_exception_bitmap(vcpu);
2668
2669         if (emulate_invalid_guest_state)
2670                 goto continue_rmode;
2671
2672         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
2673         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
2674         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
2675
2676         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
2677         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
2678         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
2679                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
2680         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
2681
2682         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
2683         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
2684         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
2685         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
2686
2687 continue_rmode:
2688         kvm_mmu_reset_context(vcpu);
2689 }
2690
2691 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2692 {
2693         struct vcpu_vmx *vmx = to_vmx(vcpu);
2694         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2695
2696         if (!msr)
2697                 return;
2698
2699         /*
2700          * Force kernel_gs_base reloading before EFER changes, as control
2701          * of this msr depends on is_long_mode().
2702          */
2703         vmx_load_host_state(to_vmx(vcpu));
2704         vcpu->arch.efer = efer;
2705         if (efer & EFER_LMA) {
2706                 vmcs_write32(VM_ENTRY_CONTROLS,
2707                              vmcs_read32(VM_ENTRY_CONTROLS) |
2708                              VM_ENTRY_IA32E_MODE);
2709                 msr->data = efer;
2710         } else {
2711                 vmcs_write32(VM_ENTRY_CONTROLS,
2712                              vmcs_read32(VM_ENTRY_CONTROLS) &
2713                              ~VM_ENTRY_IA32E_MODE);
2714
2715                 msr->data = efer & ~EFER_LME;
2716         }
2717         setup_msrs(vmx);
2718 }
2719
2720 #ifdef CONFIG_X86_64
2721
2722 static void enter_lmode(struct kvm_vcpu *vcpu)
2723 {
2724         u32 guest_tr_ar;
2725
2726         vmx_segment_cache_clear(to_vmx(vcpu));
2727
2728         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2729         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
2730                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
2731                        __func__);
2732                 vmcs_write32(GUEST_TR_AR_BYTES,
2733                              (guest_tr_ar & ~AR_TYPE_MASK)
2734                              | AR_TYPE_BUSY_64_TSS);
2735         }
2736         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2737 }
2738
2739 static void exit_lmode(struct kvm_vcpu *vcpu)
2740 {
2741         vmcs_write32(VM_ENTRY_CONTROLS,
2742                      vmcs_read32(VM_ENTRY_CONTROLS)
2743                      & ~VM_ENTRY_IA32E_MODE);
2744         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2745 }
2746
2747 #endif
2748
2749 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2750 {
2751         vpid_sync_context(to_vmx(vcpu));
2752         if (enable_ept) {
2753                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2754                         return;
2755                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2756         }
2757 }
2758
2759 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2760 {
2761         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2762
2763         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2764         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2765 }
2766
2767 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2768 {
2769         if (enable_ept && is_paging(vcpu))
2770                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2771         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2772 }
2773
2774 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2775 {
2776         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2777
2778         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2779         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2780 }
2781
2782 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2783 {
2784         if (!test_bit(VCPU_EXREG_PDPTR,
2785                       (unsigned long *)&vcpu->arch.regs_dirty))
2786                 return;
2787
2788         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2789                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2790                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2791                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2792                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2793         }
2794 }
2795
2796 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2797 {
2798         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2799                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2800                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2801                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2802                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2803         }
2804
2805         __set_bit(VCPU_EXREG_PDPTR,
2806                   (unsigned long *)&vcpu->arch.regs_avail);
2807         __set_bit(VCPU_EXREG_PDPTR,
2808                   (unsigned long *)&vcpu->arch.regs_dirty);
2809 }
2810
2811 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2812
2813 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2814                                         unsigned long cr0,
2815                                         struct kvm_vcpu *vcpu)
2816 {
2817         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2818                 vmx_decache_cr3(vcpu);
2819         if (!(cr0 & X86_CR0_PG)) {
2820                 /* From paging/starting to nonpaging */
2821                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2822                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2823                              (CPU_BASED_CR3_LOAD_EXITING |
2824                               CPU_BASED_CR3_STORE_EXITING));
2825                 vcpu->arch.cr0 = cr0;
2826                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2827         } else if (!is_paging(vcpu)) {
2828                 /* From nonpaging to paging */
2829                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2830                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2831                              ~(CPU_BASED_CR3_LOAD_EXITING |
2832                                CPU_BASED_CR3_STORE_EXITING));
2833                 vcpu->arch.cr0 = cr0;
2834                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2835         }
2836
2837         if (!(cr0 & X86_CR0_WP))
2838                 *hw_cr0 &= ~X86_CR0_WP;
2839 }
2840
2841 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2842 {
2843         struct vcpu_vmx *vmx = to_vmx(vcpu);
2844         unsigned long hw_cr0;
2845
2846         if (enable_unrestricted_guest)
2847                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
2848                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2849         else
2850                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
2851
2852         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2853                 enter_pmode(vcpu);
2854
2855         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2856                 enter_rmode(vcpu);
2857
2858 #ifdef CONFIG_X86_64
2859         if (vcpu->arch.efer & EFER_LME) {
2860                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2861                         enter_lmode(vcpu);
2862                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2863                         exit_lmode(vcpu);
2864         }
2865 #endif
2866
2867         if (enable_ept)
2868                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2869
2870         if (!vcpu->fpu_active)
2871                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
2872
2873         vmcs_writel(CR0_READ_SHADOW, cr0);
2874         vmcs_writel(GUEST_CR0, hw_cr0);
2875         vcpu->arch.cr0 = cr0;
2876         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2877 }
2878
2879 static u64 construct_eptp(unsigned long root_hpa)
2880 {
2881         u64 eptp;
2882
2883         /* TODO write the value reading from MSR */
2884         eptp = VMX_EPT_DEFAULT_MT |
2885                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
2886         eptp |= (root_hpa & PAGE_MASK);
2887
2888         return eptp;
2889 }
2890
2891 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2892 {
2893         unsigned long guest_cr3;
2894         u64 eptp;
2895
2896         guest_cr3 = cr3;
2897         if (enable_ept) {
2898                 eptp = construct_eptp(cr3);
2899                 vmcs_write64(EPT_POINTER, eptp);
2900                 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
2901                         vcpu->kvm->arch.ept_identity_map_addr;
2902                 ept_load_pdptrs(vcpu);
2903         }
2904
2905         vmx_flush_tlb(vcpu);
2906         vmcs_writel(GUEST_CR3, guest_cr3);
2907 }
2908
2909 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2910 {
2911         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
2912                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
2913
2914         if (cr4 & X86_CR4_VMXE) {
2915                 /*
2916                  * To use VMXON (and later other VMX instructions), a guest
2917                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
2918                  * So basically the check on whether to allow nested VMX
2919                  * is here.
2920                  */
2921                 if (!nested_vmx_allowed(vcpu))
2922                         return 1;
2923         } else if (to_vmx(vcpu)->nested.vmxon)
2924                 return 1;
2925
2926         vcpu->arch.cr4 = cr4;
2927         if (enable_ept) {
2928                 if (!is_paging(vcpu)) {
2929                         hw_cr4 &= ~X86_CR4_PAE;
2930                         hw_cr4 |= X86_CR4_PSE;
2931                 } else if (!(cr4 & X86_CR4_PAE)) {
2932                         hw_cr4 &= ~X86_CR4_PAE;
2933                 }
2934         }
2935
2936         vmcs_writel(CR4_READ_SHADOW, cr4);
2937         vmcs_writel(GUEST_CR4, hw_cr4);
2938         return 0;
2939 }
2940
2941 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2942                             struct kvm_segment *var, int seg)
2943 {
2944         struct vcpu_vmx *vmx = to_vmx(vcpu);
2945         struct kvm_save_segment *save;
2946         u32 ar;
2947
2948         if (vmx->rmode.vm86_active
2949             && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
2950                 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
2951                 || seg == VCPU_SREG_GS)
2952             && !emulate_invalid_guest_state) {
2953                 switch (seg) {
2954                 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
2955                 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
2956                 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
2957                 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
2958                 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
2959                 default: BUG();
2960                 }
2961                 var->selector = save->selector;
2962                 var->base = save->base;
2963                 var->limit = save->limit;
2964                 ar = save->ar;
2965                 if (seg == VCPU_SREG_TR
2966                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
2967                         goto use_saved_rmode_seg;
2968         }
2969         var->base = vmx_read_guest_seg_base(vmx, seg);
2970         var->limit = vmx_read_guest_seg_limit(vmx, seg);
2971         var->selector = vmx_read_guest_seg_selector(vmx, seg);
2972         ar = vmx_read_guest_seg_ar(vmx, seg);
2973 use_saved_rmode_seg:
2974         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2975                 ar = 0;
2976         var->type = ar & 15;
2977         var->s = (ar >> 4) & 1;
2978         var->dpl = (ar >> 5) & 3;
2979         var->present = (ar >> 7) & 1;
2980         var->avl = (ar >> 12) & 1;
2981         var->l = (ar >> 13) & 1;
2982         var->db = (ar >> 14) & 1;
2983         var->g = (ar >> 15) & 1;
2984         var->unusable = (ar >> 16) & 1;
2985 }
2986
2987 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2988 {
2989         struct kvm_segment s;
2990
2991         if (to_vmx(vcpu)->rmode.vm86_active) {
2992                 vmx_get_segment(vcpu, &s, seg);
2993                 return s.base;
2994         }
2995         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
2996 }
2997
2998 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
2999 {
3000         if (!is_protmode(vcpu))
3001                 return 0;
3002
3003         if (!is_long_mode(vcpu)
3004             && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3005                 return 3;
3006
3007         return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
3008 }
3009
3010 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3011 {
3012         if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3013                 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3014                 to_vmx(vcpu)->cpl = __vmx_get_cpl(vcpu);
3015         }
3016         return to_vmx(vcpu)->cpl;
3017 }
3018
3019
3020 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3021 {
3022         u32 ar;
3023
3024         if (var->unusable)
3025                 ar = 1 << 16;
3026         else {
3027                 ar = var->type & 15;
3028                 ar |= (var->s & 1) << 4;
3029                 ar |= (var->dpl & 3) << 5;
3030                 ar |= (var->present & 1) << 7;
3031                 ar |= (var->avl & 1) << 12;
3032                 ar |= (var->l & 1) << 13;
3033                 ar |= (var->db & 1) << 14;
3034                 ar |= (var->g & 1) << 15;
3035         }
3036         if (ar == 0) /* a 0 value means unusable */
3037                 ar = AR_UNUSABLE_MASK;
3038
3039         return ar;
3040 }
3041
3042 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3043                             struct kvm_segment *var, int seg)
3044 {
3045         struct vcpu_vmx *vmx = to_vmx(vcpu);
3046         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3047         u32 ar;
3048
3049         vmx_segment_cache_clear(vmx);
3050
3051         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
3052                 vmcs_write16(sf->selector, var->selector);
3053                 vmx->rmode.tr.selector = var->selector;
3054                 vmx->rmode.tr.base = var->base;
3055                 vmx->rmode.tr.limit = var->limit;
3056                 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
3057                 return;
3058         }
3059         vmcs_writel(sf->base, var->base);
3060         vmcs_write32(sf->limit, var->limit);
3061         vmcs_write16(sf->selector, var->selector);
3062         if (vmx->rmode.vm86_active && var->s) {
3063                 /*
3064                  * Hack real-mode segments into vm86 compatibility.
3065                  */
3066                 if (var->base == 0xffff0000 && var->selector == 0xf000)
3067                         vmcs_writel(sf->base, 0xf0000);
3068                 ar = 0xf3;
3069         } else
3070                 ar = vmx_segment_access_rights(var);
3071
3072         /*
3073          *   Fix the "Accessed" bit in AR field of segment registers for older
3074          * qemu binaries.
3075          *   IA32 arch specifies that at the time of processor reset the
3076          * "Accessed" bit in the AR field of segment registers is 1. And qemu
3077          * is setting it to 0 in the usedland code. This causes invalid guest
3078          * state vmexit when "unrestricted guest" mode is turned on.
3079          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3080          * tree. Newer qemu binaries with that qemu fix would not need this
3081          * kvm hack.
3082          */
3083         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3084                 ar |= 0x1; /* Accessed */
3085
3086         vmcs_write32(sf->ar_bytes, ar);
3087         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3088 }
3089
3090 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3091 {
3092         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3093
3094         *db = (ar >> 14) & 1;
3095         *l = (ar >> 13) & 1;
3096 }
3097
3098 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3099 {
3100         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3101         dt->address = vmcs_readl(GUEST_IDTR_BASE);
3102 }
3103
3104 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3105 {
3106         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3107         vmcs_writel(GUEST_IDTR_BASE, dt->address);
3108 }
3109
3110 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3111 {
3112         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3113         dt->address = vmcs_readl(GUEST_GDTR_BASE);
3114 }
3115
3116 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3117 {
3118         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3119         vmcs_writel(GUEST_GDTR_BASE, dt->address);
3120 }
3121
3122 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3123 {
3124         struct kvm_segment var;
3125         u32 ar;
3126
3127         vmx_get_segment(vcpu, &var, seg);
3128         ar = vmx_segment_access_rights(&var);
3129
3130         if (var.base != (var.selector << 4))
3131                 return false;
3132         if (var.limit != 0xffff)
3133                 return false;
3134         if (ar != 0xf3)
3135                 return false;
3136
3137         return true;
3138 }
3139
3140 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3141 {
3142         struct kvm_segment cs;
3143         unsigned int cs_rpl;
3144
3145         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3146         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3147
3148         if (cs.unusable)
3149                 return false;
3150         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3151                 return false;
3152         if (!cs.s)
3153                 return false;
3154         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3155                 if (cs.dpl > cs_rpl)
3156                         return false;
3157         } else {
3158                 if (cs.dpl != cs_rpl)
3159                         return false;
3160         }
3161         if (!cs.present)
3162                 return false;
3163
3164         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3165         return true;
3166 }
3167
3168 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3169 {
3170         struct kvm_segment ss;
3171         unsigned int ss_rpl;
3172
3173         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3174         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3175
3176         if (ss.unusable)
3177                 return true;
3178         if (ss.type != 3 && ss.type != 7)
3179                 return false;
3180         if (!ss.s)
3181                 return false;
3182         if (ss.dpl != ss_rpl) /* DPL != RPL */
3183                 return false;
3184         if (!ss.present)
3185                 return false;
3186
3187         return true;
3188 }
3189
3190 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3191 {
3192         struct kvm_segment var;
3193         unsigned int rpl;
3194
3195         vmx_get_segment(vcpu, &var, seg);
3196         rpl = var.selector & SELECTOR_RPL_MASK;
3197
3198         if (var.unusable)
3199                 return true;
3200         if (!var.s)
3201                 return false;
3202         if (!var.present)
3203                 return false;
3204         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3205                 if (var.dpl < rpl) /* DPL < RPL */
3206                         return false;
3207         }
3208
3209         /* TODO: Add other members to kvm_segment_field to allow checking for other access
3210          * rights flags
3211          */
3212         return true;
3213 }
3214
3215 static bool tr_valid(struct kvm_vcpu *vcpu)
3216 {
3217         struct kvm_segment tr;
3218
3219         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3220
3221         if (tr.unusable)
3222                 return false;
3223         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
3224                 return false;
3225         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3226                 return false;
3227         if (!tr.present)
3228                 return false;
3229
3230         return true;
3231 }
3232
3233 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3234 {
3235         struct kvm_segment ldtr;
3236
3237         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3238
3239         if (ldtr.unusable)
3240                 return true;
3241         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
3242                 return false;
3243         if (ldtr.type != 2)
3244                 return false;
3245         if (!ldtr.present)
3246                 return false;
3247
3248         return true;
3249 }
3250
3251 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3252 {
3253         struct kvm_segment cs, ss;
3254
3255         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3256         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3257
3258         return ((cs.selector & SELECTOR_RPL_MASK) ==
3259                  (ss.selector & SELECTOR_RPL_MASK));
3260 }
3261
3262 /*
3263  * Check if guest state is valid. Returns true if valid, false if
3264  * not.
3265  * We assume that registers are always usable
3266  */
3267 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3268 {
3269         /* real mode guest state checks */
3270         if (!is_protmode(vcpu)) {
3271                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3272                         return false;
3273                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3274                         return false;
3275                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3276                         return false;
3277                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3278                         return false;
3279                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3280                         return false;
3281                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3282                         return false;
3283         } else {
3284         /* protected mode guest state checks */
3285                 if (!cs_ss_rpl_check(vcpu))
3286                         return false;
3287                 if (!code_segment_valid(vcpu))
3288                         return false;
3289                 if (!stack_segment_valid(vcpu))
3290                         return false;
3291                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3292                         return false;
3293                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3294                         return false;
3295                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3296                         return false;
3297                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3298                         return false;
3299                 if (!tr_valid(vcpu))
3300                         return false;
3301                 if (!ldtr_valid(vcpu))
3302                         return false;
3303         }
3304         /* TODO:
3305          * - Add checks on RIP
3306          * - Add checks on RFLAGS
3307          */
3308
3309         return true;
3310 }
3311
3312 static int init_rmode_tss(struct kvm *kvm)
3313 {
3314         gfn_t fn;
3315         u16 data = 0;
3316         int r, idx, ret = 0;
3317
3318         idx = srcu_read_lock(&kvm->srcu);
3319         fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
3320         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3321         if (r < 0)
3322                 goto out;
3323         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3324         r = kvm_write_guest_page(kvm, fn++, &data,
3325                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
3326         if (r < 0)
3327                 goto out;
3328         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3329         if (r < 0)
3330                 goto out;
3331         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3332         if (r < 0)
3333                 goto out;
3334         data = ~0;
3335         r = kvm_write_guest_page(kvm, fn, &data,
3336                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3337                                  sizeof(u8));
3338         if (r < 0)
3339                 goto out;
3340
3341         ret = 1;
3342 out:
3343         srcu_read_unlock(&kvm->srcu, idx);
3344         return ret;
3345 }
3346
3347 static int init_rmode_identity_map(struct kvm *kvm)
3348 {
3349         int i, idx, r, ret;
3350         pfn_t identity_map_pfn;
3351         u32 tmp;
3352
3353         if (!enable_ept)
3354                 return 1;
3355         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3356                 printk(KERN_ERR "EPT: identity-mapping pagetable "
3357                         "haven't been allocated!\n");
3358                 return 0;
3359         }
3360         if (likely(kvm->arch.ept_identity_pagetable_done))
3361                 return 1;
3362         ret = 0;
3363         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3364         idx = srcu_read_lock(&kvm->srcu);
3365         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3366         if (r < 0)
3367                 goto out;
3368         /* Set up identity-mapping pagetable for EPT in real mode */
3369         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3370                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3371                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3372                 r = kvm_write_guest_page(kvm, identity_map_pfn,
3373                                 &tmp, i * sizeof(tmp), sizeof(tmp));
3374                 if (r < 0)
3375                         goto out;
3376         }
3377         kvm->arch.ept_identity_pagetable_done = true;
3378         ret = 1;
3379 out:
3380         srcu_read_unlock(&kvm->srcu, idx);
3381         return ret;
3382 }
3383
3384 static void seg_setup(int seg)
3385 {
3386         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3387         unsigned int ar;
3388
3389         vmcs_write16(sf->selector, 0);
3390         vmcs_writel(sf->base, 0);
3391         vmcs_write32(sf->limit, 0xffff);
3392         if (enable_unrestricted_guest) {
3393                 ar = 0x93;
3394                 if (seg == VCPU_SREG_CS)
3395                         ar |= 0x08; /* code segment */
3396         } else
3397                 ar = 0xf3;
3398
3399         vmcs_write32(sf->ar_bytes, ar);
3400 }
3401
3402 static int alloc_apic_access_page(struct kvm *kvm)
3403 {
3404         struct kvm_userspace_memory_region kvm_userspace_mem;
3405         int r = 0;
3406
3407         mutex_lock(&kvm->slots_lock);
3408         if (kvm->arch.apic_access_page)
3409                 goto out;
3410         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3411         kvm_userspace_mem.flags = 0;
3412         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3413         kvm_userspace_mem.memory_size = PAGE_SIZE;
3414         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3415         if (r)
3416                 goto out;
3417
3418         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
3419 out:
3420         mutex_unlock(&kvm->slots_lock);
3421         return r;
3422 }
3423
3424 static int alloc_identity_pagetable(struct kvm *kvm)
3425 {
3426         struct kvm_userspace_memory_region kvm_userspace_mem;
3427         int r = 0;
3428
3429         mutex_lock(&kvm->slots_lock);
3430         if (kvm->arch.ept_identity_pagetable)
3431                 goto out;
3432         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3433         kvm_userspace_mem.flags = 0;
3434         kvm_userspace_mem.guest_phys_addr =
3435                 kvm->arch.ept_identity_map_addr;
3436         kvm_userspace_mem.memory_size = PAGE_SIZE;
3437         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3438         if (r)
3439                 goto out;
3440
3441         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
3442                         kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3443 out:
3444         mutex_unlock(&kvm->slots_lock);
3445         return r;
3446 }
3447
3448 static void allocate_vpid(struct vcpu_vmx *vmx)
3449 {
3450         int vpid;
3451
3452         vmx->vpid = 0;
3453         if (!enable_vpid)
3454                 return;
3455         spin_lock(&vmx_vpid_lock);
3456         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3457         if (vpid < VMX_NR_VPIDS) {
3458                 vmx->vpid = vpid;
3459                 __set_bit(vpid, vmx_vpid_bitmap);
3460         }
3461         spin_unlock(&vmx_vpid_lock);
3462 }
3463
3464 static void free_vpid(struct vcpu_vmx *vmx)
3465 {
3466         if (!enable_vpid)
3467                 return;
3468         spin_lock(&vmx_vpid_lock);
3469         if (vmx->vpid != 0)
3470                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3471         spin_unlock(&vmx_vpid_lock);
3472 }
3473
3474 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
3475 {
3476         int f = sizeof(unsigned long);
3477
3478         if (!cpu_has_vmx_msr_bitmap())
3479                 return;
3480
3481         /*
3482          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3483          * have the write-low and read-high bitmap offsets the wrong way round.
3484          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3485          */
3486         if (msr <= 0x1fff) {
3487                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3488                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
3489         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3490                 msr &= 0x1fff;
3491                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3492                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
3493         }
3494 }
3495
3496 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3497 {
3498         if (!longmode_only)
3499                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3500         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3501 }
3502
3503 /*
3504  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3505  * will not change in the lifetime of the guest.
3506  * Note that host-state that does change is set elsewhere. E.g., host-state
3507  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3508  */
3509 static void vmx_set_constant_host_state(void)
3510 {
3511         u32 low32, high32;
3512         unsigned long tmpl;
3513         struct desc_ptr dt;
3514
3515         vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS);  /* 22.2.3 */
3516         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
3517         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
3518
3519         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3520         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3521         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3522         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3523         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3524
3525         native_store_idt(&dt);
3526         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
3527
3528         asm("mov $.Lkvm_vmx_return, %0" : "=r"(tmpl));
3529         vmcs_writel(HOST_RIP, tmpl); /* 22.2.5 */
3530
3531         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3532         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3533         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3534         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3535
3536         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3537                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3538                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3539         }
3540 }
3541
3542 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3543 {
3544         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3545         if (enable_ept)
3546                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3547         if (is_guest_mode(&vmx->vcpu))
3548                 vmx->vcpu.arch.cr4_guest_owned_bits &=
3549                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3550         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3551 }
3552
3553 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3554 {
3555         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3556         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3557                 exec_control &= ~CPU_BASED_TPR_SHADOW;
3558 #ifdef CONFIG_X86_64
3559                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3560                                 CPU_BASED_CR8_LOAD_EXITING;
3561 #endif
3562         }
3563         if (!enable_ept)
3564                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3565                                 CPU_BASED_CR3_LOAD_EXITING  |
3566                                 CPU_BASED_INVLPG_EXITING;
3567         return exec_control;
3568 }
3569
3570 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3571 {
3572         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3573         if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3574                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3575         if (vmx->vpid == 0)
3576                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3577         if (!enable_ept) {
3578                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3579                 enable_unrestricted_guest = 0;
3580         }
3581         if (!enable_unrestricted_guest)
3582                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3583         if (!ple_gap)
3584                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3585         return exec_control;
3586 }
3587
3588 /*
3589  * Sets up the vmcs for emulated real mode.
3590  */
3591 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3592 {
3593         unsigned long a;
3594         int i;
3595
3596         /* I/O */
3597         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3598         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
3599
3600         if (cpu_has_vmx_msr_bitmap())
3601                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
3602
3603         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3604
3605         /* Control */
3606         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3607                 vmcs_config.pin_based_exec_ctrl);
3608
3609         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
3610
3611         if (cpu_has_secondary_exec_ctrls()) {
3612                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
3613                                 vmx_secondary_exec_control(vmx));
3614         }
3615
3616         if (ple_gap) {
3617                 vmcs_write32(PLE_GAP, ple_gap);
3618                 vmcs_write32(PLE_WINDOW, ple_window);
3619         }
3620
3621         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
3622         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
3623         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
3624
3625         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
3626         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
3627         vmx_set_constant_host_state();
3628 #ifdef CONFIG_X86_64
3629         rdmsrl(MSR_FS_BASE, a);
3630         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3631         rdmsrl(MSR_GS_BASE, a);
3632         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3633 #else
3634         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3635         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3636 #endif
3637
3638         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3639         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3640         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
3641         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3642         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
3643
3644         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3645                 u32 msr_low, msr_high;
3646                 u64 host_pat;
3647                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3648                 host_pat = msr_low | ((u64) msr_high << 32);
3649                 /* Write the default value follow host pat */
3650                 vmcs_write64(GUEST_IA32_PAT, host_pat);
3651                 /* Keep arch.pat sync with GUEST_IA32_PAT */
3652                 vmx->vcpu.arch.pat = host_pat;
3653         }
3654
3655         for (i = 0; i < NR_VMX_MSR; ++i) {
3656                 u32 index = vmx_msr_index[i];
3657                 u32 data_low, data_high;
3658                 int j = vmx->nmsrs;
3659
3660                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
3661                         continue;
3662                 if (wrmsr_safe(index, data_low, data_high) < 0)
3663                         continue;
3664                 vmx->guest_msrs[j].index = i;
3665                 vmx->guest_msrs[j].data = 0;
3666                 vmx->guest_msrs[j].mask = -1ull;
3667                 ++vmx->nmsrs;
3668         }
3669
3670         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
3671
3672         /* 22.2.1, 20.8.1 */
3673         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3674
3675         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
3676         set_cr4_guest_host_mask(vmx);
3677
3678         kvm_write_tsc(&vmx->vcpu, 0);
3679
3680         return 0;
3681 }
3682
3683 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3684 {
3685         struct vcpu_vmx *vmx = to_vmx(vcpu);
3686         u64 msr;
3687         int ret;
3688
3689         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3690
3691         vmx->rmode.vm86_active = 0;
3692
3693         vmx->soft_vnmi_blocked = 0;
3694
3695         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
3696         kvm_set_cr8(&vmx->vcpu, 0);
3697         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
3698         if (kvm_vcpu_is_bsp(&vmx->vcpu))
3699                 msr |= MSR_IA32_APICBASE_BSP;
3700         kvm_set_apic_base(&vmx->vcpu, msr);
3701
3702         ret = fx_init(&vmx->vcpu);
3703         if (ret != 0)
3704                 goto out;
3705
3706         vmx_segment_cache_clear(vmx);
3707
3708         seg_setup(VCPU_SREG_CS);
3709         /*
3710          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3711          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
3712          */
3713         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
3714                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3715                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3716         } else {
3717                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3718                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
3719         }
3720
3721         seg_setup(VCPU_SREG_DS);
3722         seg_setup(VCPU_SREG_ES);
3723         seg_setup(VCPU_SREG_FS);
3724         seg_setup(VCPU_SREG_GS);
3725         seg_setup(VCPU_SREG_SS);
3726
3727         vmcs_write16(GUEST_TR_SELECTOR, 0);
3728         vmcs_writel(GUEST_TR_BASE, 0);
3729         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3730         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3731
3732         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3733         vmcs_writel(GUEST_LDTR_BASE, 0);
3734         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3735         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3736
3737         vmcs_write32(GUEST_SYSENTER_CS, 0);
3738         vmcs_writel(GUEST_SYSENTER_ESP, 0);
3739         vmcs_writel(GUEST_SYSENTER_EIP, 0);
3740
3741         vmcs_writel(GUEST_RFLAGS, 0x02);
3742         if (kvm_vcpu_is_bsp(&vmx->vcpu))
3743                 kvm_rip_write(vcpu, 0xfff0);
3744         else
3745                 kvm_rip_write(vcpu, 0);
3746         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
3747
3748         vmcs_writel(GUEST_DR7, 0x400);
3749
3750         vmcs_writel(GUEST_GDTR_BASE, 0);
3751         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3752
3753         vmcs_writel(GUEST_IDTR_BASE, 0);
3754         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3755
3756         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
3757         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3758         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3759
3760         /* Special registers */
3761         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3762
3763         setup_msrs(vmx);
3764
3765         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
3766
3767         if (cpu_has_vmx_tpr_shadow()) {
3768                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3769                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3770                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
3771                                      __pa(vmx->vcpu.arch.apic->regs));
3772                 vmcs_write32(TPR_THRESHOLD, 0);
3773         }
3774
3775         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3776                 vmcs_write64(APIC_ACCESS_ADDR,
3777                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
3778
3779         if (vmx->vpid != 0)
3780                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
3781
3782         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
3783         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
3784         vmx_set_cr4(&vmx->vcpu, 0);
3785         vmx_set_efer(&vmx->vcpu, 0);
3786         vmx_fpu_activate(&vmx->vcpu);
3787         update_exception_bitmap(&vmx->vcpu);
3788
3789         vpid_sync_context(vmx);
3790
3791         ret = 0;
3792
3793         /* HACK: Don't enable emulation on guest boot/reset */
3794         vmx->emulation_required = 0;
3795
3796 out:
3797         return ret;
3798 }
3799
3800 /*
3801  * In nested virtualization, check if L1 asked to exit on external interrupts.
3802  * For most existing hypervisors, this will always return true.
3803  */
3804 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
3805 {
3806         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
3807                 PIN_BASED_EXT_INTR_MASK;
3808 }
3809
3810 static void enable_irq_window(struct kvm_vcpu *vcpu)
3811 {
3812         u32 cpu_based_vm_exec_control;
3813         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
3814                 /* We can get here when nested_run_pending caused
3815                  * vmx_interrupt_allowed() to return false. In this case, do
3816                  * nothing - the interrupt will be injected later.
3817                  */
3818                 return;
3819
3820         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3821         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
3822         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3823 }
3824
3825 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3826 {
3827         u32 cpu_based_vm_exec_control;
3828
3829         if (!cpu_has_virtual_nmis()) {
3830                 enable_irq_window(vcpu);
3831                 return;
3832         }
3833
3834         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
3835                 enable_irq_window(vcpu);
3836                 return;
3837         }
3838         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3839         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
3840         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3841 }
3842
3843 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
3844 {
3845         struct vcpu_vmx *vmx = to_vmx(vcpu);
3846         uint32_t intr;
3847         int irq = vcpu->arch.interrupt.nr;
3848
3849         trace_kvm_inj_virq(irq);
3850
3851         ++vcpu->stat.irq_injections;
3852         if (vmx->rmode.vm86_active) {
3853                 int inc_eip = 0;
3854                 if (vcpu->arch.interrupt.soft)
3855                         inc_eip = vcpu->arch.event_exit_inst_len;
3856                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
3857                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3858                 return;
3859         }
3860         intr = irq | INTR_INFO_VALID_MASK;
3861         if (vcpu->arch.interrupt.soft) {
3862                 intr |= INTR_TYPE_SOFT_INTR;
3863                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
3864                              vmx->vcpu.arch.event_exit_inst_len);
3865         } else
3866                 intr |= INTR_TYPE_EXT_INTR;
3867         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
3868         vmx_clear_hlt(vcpu);
3869 }
3870
3871 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
3872 {
3873         struct vcpu_vmx *vmx = to_vmx(vcpu);
3874
3875         if (is_guest_mode(vcpu))
3876                 return;
3877
3878         if (!cpu_has_virtual_nmis()) {
3879                 /*
3880                  * Tracking the NMI-blocked state in software is built upon
3881                  * finding the next open IRQ window. This, in turn, depends on
3882                  * well-behaving guests: They have to keep IRQs disabled at
3883                  * least as long as the NMI handler runs. Otherwise we may
3884                  * cause NMI nesting, maybe breaking the guest. But as this is
3885                  * highly unlikely, we can live with the residual risk.
3886                  */
3887                 vmx->soft_vnmi_blocked = 1;
3888                 vmx->vnmi_blocked_time = 0;
3889         }
3890
3891         ++vcpu->stat.nmi_injections;
3892         vmx->nmi_known_unmasked = false;
3893         if (vmx->rmode.vm86_active) {
3894                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
3895                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3896                 return;
3897         }
3898         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
3899                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
3900         vmx_clear_hlt(vcpu);
3901 }
3902
3903 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
3904 {
3905         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
3906                 return 0;
3907
3908         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3909                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
3910                    | GUEST_INTR_STATE_NMI));
3911 }
3912
3913 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
3914 {
3915         if (!cpu_has_virtual_nmis())
3916                 return to_vmx(vcpu)->soft_vnmi_blocked;
3917         if (to_vmx(vcpu)->nmi_known_unmasked)
3918                 return false;
3919         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
3920 }
3921
3922 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3923 {
3924         struct vcpu_vmx *vmx = to_vmx(vcpu);
3925
3926         if (!cpu_has_virtual_nmis()) {
3927                 if (vmx->soft_vnmi_blocked != masked) {
3928                         vmx->soft_vnmi_blocked = masked;
3929                         vmx->vnmi_blocked_time = 0;
3930                 }
3931         } else {
3932                 vmx->nmi_known_unmasked = !masked;
3933                 if (masked)
3934                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3935                                       GUEST_INTR_STATE_NMI);
3936                 else
3937                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3938                                         GUEST_INTR_STATE_NMI);
3939         }
3940 }
3941
3942 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
3943 {
3944         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
3945                 struct vmcs12 *vmcs12;
3946                 if (to_vmx(vcpu)->nested.nested_run_pending)
3947                         return 0;
3948                 nested_vmx_vmexit(vcpu);
3949                 vmcs12 = get_vmcs12(vcpu);
3950                 vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
3951                 vmcs12->vm_exit_intr_info = 0;
3952                 /* fall through to normal code, but now in L1, not L2 */
3953         }
3954
3955         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
3956                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3957                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
3958 }
3959
3960 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
3961 {
3962         int ret;
3963         struct kvm_userspace_memory_region tss_mem = {
3964                 .slot = TSS_PRIVATE_MEMSLOT,
3965                 .guest_phys_addr = addr,
3966                 .memory_size = PAGE_SIZE * 3,
3967                 .flags = 0,
3968         };
3969
3970         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
3971         if (ret)
3972                 return ret;
3973         kvm->arch.tss_addr = addr;
3974         if (!init_rmode_tss(kvm))
3975                 return  -ENOMEM;
3976
3977         return 0;
3978 }
3979
3980 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
3981                                   int vec, u32 err_code)
3982 {
3983         /*
3984          * Instruction with address size override prefix opcode 0x67
3985          * Cause the #SS fault with 0 error code in VM86 mode.
3986          */
3987         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3988                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
3989                         return 1;
3990         /*
3991          * Forward all other exceptions that are valid in real mode.
3992          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
3993          *        the required debugging infrastructure rework.
3994          */
3995         switch (vec) {
3996         case DB_VECTOR:
3997                 if (vcpu->guest_debug &
3998                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
3999                         return 0;
4000                 kvm_queue_exception(vcpu, vec);
4001                 return 1;
4002         case BP_VECTOR:
4003                 /*
4004                  * Update instruction length as we may reinject the exception
4005                  * from user space while in guest debugging mode.
4006                  */
4007                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4008                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4009                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4010                         return 0;
4011                 /* fall through */
4012         case DE_VECTOR:
4013         case OF_VECTOR:
4014         case BR_VECTOR:
4015         case UD_VECTOR:
4016         case DF_VECTOR:
4017         case SS_VECTOR:
4018         case GP_VECTOR:
4019         case MF_VECTOR:
4020                 kvm_queue_exception(vcpu, vec);
4021                 return 1;
4022         }
4023         return 0;
4024 }
4025
4026 /*
4027  * Trigger machine check on the host. We assume all the MSRs are already set up
4028  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4029  * We pass a fake environment to the machine check handler because we want
4030  * the guest to be always treated like user space, no matter what context
4031  * it used internally.
4032  */
4033 static void kvm_machine_check(void)
4034 {
4035 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4036         struct pt_regs regs = {
4037                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4038                 .flags = X86_EFLAGS_IF,
4039         };
4040
4041         do_machine_check(&regs, 0);
4042 #endif
4043 }
4044
4045 static int handle_machine_check(struct kvm_vcpu *vcpu)
4046 {
4047         /* already handled by vcpu_run */
4048         return 1;
4049 }
4050
4051 static int handle_exception(struct kvm_vcpu *vcpu)
4052 {
4053         struct vcpu_vmx *vmx = to_vmx(vcpu);
4054         struct kvm_run *kvm_run = vcpu->run;
4055         u32 intr_info, ex_no, error_code;
4056         unsigned long cr2, rip, dr6;
4057         u32 vect_info;
4058         enum emulation_result er;
4059
4060         vect_info = vmx->idt_vectoring_info;
4061         intr_info = vmx->exit_intr_info;
4062
4063         if (is_machine_check(intr_info))
4064                 return handle_machine_check(vcpu);
4065
4066         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4067             !is_page_fault(intr_info)) {
4068                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4069                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4070                 vcpu->run->internal.ndata = 2;
4071                 vcpu->run->internal.data[0] = vect_info;
4072                 vcpu->run->internal.data[1] = intr_info;
4073                 return 0;
4074         }
4075
4076         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4077                 return 1;  /* already handled by vmx_vcpu_run() */
4078
4079         if (is_no_device(intr_info)) {
4080                 vmx_fpu_activate(vcpu);
4081                 return 1;
4082         }
4083
4084         if (is_invalid_opcode(intr_info)) {
4085                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4086                 if (er != EMULATE_DONE)
4087                         kvm_queue_exception(vcpu, UD_VECTOR);
4088                 return 1;
4089         }
4090
4091         error_code = 0;
4092         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4093                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4094         if (is_page_fault(intr_info)) {
4095                 /* EPT won't cause page fault directly */
4096                 if (enable_ept)
4097                         BUG();
4098                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4099                 trace_kvm_page_fault(cr2, error_code);
4100
4101                 if (kvm_event_needs_reinjection(vcpu))
4102                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
4103                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4104         }
4105
4106         if (vmx->rmode.vm86_active &&
4107             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
4108                                                                 error_code)) {
4109                 if (vcpu->arch.halt_request) {
4110                         vcpu->arch.halt_request = 0;
4111                         return kvm_emulate_halt(vcpu);
4112                 }
4113                 return 1;
4114         }
4115
4116         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4117         switch (ex_no) {
4118         case DB_VECTOR:
4119                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4120                 if (!(vcpu->guest_debug &
4121                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4122                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4123                         kvm_queue_exception(vcpu, DB_VECTOR);
4124                         return 1;
4125                 }
4126                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4127                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4128                 /* fall through */
4129         case BP_VECTOR:
4130                 /*
4131                  * Update instruction length as we may reinject #BP from
4132                  * user space while in guest debugging mode. Reading it for
4133                  * #DB as well causes no harm, it is not used in that case.
4134                  */
4135                 vmx->vcpu.arch.event_exit_inst_len =
4136                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4137                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4138                 rip = kvm_rip_read(vcpu);
4139                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4140                 kvm_run->debug.arch.exception = ex_no;
4141                 break;
4142         default:
4143                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4144                 kvm_run->ex.exception = ex_no;
4145                 kvm_run->ex.error_code = error_code;
4146                 break;
4147         }
4148         return 0;
4149 }
4150
4151 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4152 {
4153         ++vcpu->stat.irq_exits;
4154         return 1;
4155 }
4156
4157 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4158 {
4159         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4160         return 0;
4161 }
4162
4163 static int handle_io(struct kvm_vcpu *vcpu)
4164 {
4165         unsigned long exit_qualification;
4166         int size, in, string;
4167         unsigned port;
4168
4169         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4170         string = (exit_qualification & 16) != 0;
4171         in = (exit_qualification & 8) != 0;
4172
4173         ++vcpu->stat.io_exits;
4174
4175         if (string || in)
4176                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4177
4178         port = exit_qualification >> 16;
4179         size = (exit_qualification & 7) + 1;
4180         skip_emulated_instruction(vcpu);
4181
4182         return kvm_fast_pio_out(vcpu, size, port);
4183 }
4184
4185 static void
4186 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4187 {
4188         /*
4189          * Patch in the VMCALL instruction:
4190          */
4191         hypercall[0] = 0x0f;
4192         hypercall[1] = 0x01;
4193         hypercall[2] = 0xc1;
4194 }
4195
4196 /* called to set cr0 as approriate for a mov-to-cr0 exit. */
4197 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4198 {
4199         if (to_vmx(vcpu)->nested.vmxon &&
4200             ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4201                 return 1;
4202
4203         if (is_guest_mode(vcpu)) {
4204                 /*
4205                  * We get here when L2 changed cr0 in a way that did not change
4206                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4207                  * but did change L0 shadowed bits. This can currently happen
4208                  * with the TS bit: L0 may want to leave TS on (for lazy fpu
4209                  * loading) while pretending to allow the guest to change it.
4210                  */
4211                 if (kvm_set_cr0(vcpu, (val & vcpu->arch.cr0_guest_owned_bits) |
4212                          (vcpu->arch.cr0 & ~vcpu->arch.cr0_guest_owned_bits)))
4213                         return 1;
4214                 vmcs_writel(CR0_READ_SHADOW, val);
4215                 return 0;
4216         } else
4217                 return kvm_set_cr0(vcpu, val);
4218 }
4219
4220 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4221 {
4222         if (is_guest_mode(vcpu)) {
4223                 if (kvm_set_cr4(vcpu, (val & vcpu->arch.cr4_guest_owned_bits) |
4224                          (vcpu->arch.cr4 & ~vcpu->arch.cr4_guest_owned_bits)))
4225                         return 1;
4226                 vmcs_writel(CR4_READ_SHADOW, val);
4227                 return 0;
4228         } else
4229                 return kvm_set_cr4(vcpu, val);
4230 }
4231
4232 /* called to set cr0 as approriate for clts instruction exit. */
4233 static void handle_clts(struct kvm_vcpu *vcpu)
4234 {
4235         if (is_guest_mode(vcpu)) {
4236                 /*
4237                  * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4238                  * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4239                  * just pretend it's off (also in arch.cr0 for fpu_activate).
4240                  */
4241                 vmcs_writel(CR0_READ_SHADOW,
4242                         vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4243                 vcpu->arch.cr0 &= ~X86_CR0_TS;
4244         } else
4245                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4246 }
4247
4248 static int handle_cr(struct kvm_vcpu *vcpu)
4249 {
4250         unsigned long exit_qualification, val;
4251         int cr;
4252         int reg;
4253         int err;
4254
4255         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4256         cr = exit_qualification & 15;
4257         reg = (exit_qualification >> 8) & 15;
4258         switch ((exit_qualification >> 4) & 3) {
4259         case 0: /* mov to cr */
4260                 val = kvm_register_read(vcpu, reg);
4261                 trace_kvm_cr_write(cr, val);
4262                 switch (cr) {
4263                 case 0:
4264                         err = handle_set_cr0(vcpu, val);
4265                         kvm_complete_insn_gp(vcpu, err);
4266                         return 1;
4267                 case 3:
4268                         err = kvm_set_cr3(vcpu, val);
4269                         kvm_complete_insn_gp(vcpu, err);
4270                         return 1;
4271                 case 4:
4272                         err = handle_set_cr4(vcpu, val);
4273                         kvm_complete_insn_gp(vcpu, err);
4274                         return 1;
4275                 case 8: {
4276                                 u8 cr8_prev = kvm_get_cr8(vcpu);
4277                                 u8 cr8 = kvm_register_read(vcpu, reg);
4278                                 err = kvm_set_cr8(vcpu, cr8);
4279                                 kvm_complete_insn_gp(vcpu, err);
4280                                 if (irqchip_in_kernel(vcpu->kvm))
4281                                         return 1;
4282                                 if (cr8_prev <= cr8)
4283                                         return 1;
4284                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4285                                 return 0;
4286                         }
4287                 };
4288                 break;
4289         case 2: /* clts */
4290                 handle_clts(vcpu);
4291                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4292                 skip_emulated_instruction(vcpu);
4293                 vmx_fpu_activate(vcpu);
4294                 return 1;
4295         case 1: /*mov from cr*/
4296                 switch (cr) {
4297                 case 3:
4298                         val = kvm_read_cr3(vcpu);
4299                         kvm_register_write(vcpu, reg, val);
4300                         trace_kvm_cr_read(cr, val);
4301                         skip_emulated_instruction(vcpu);
4302                         return 1;
4303                 case 8:
4304                         val = kvm_get_cr8(vcpu);
4305                         kvm_register_write(vcpu, reg, val);
4306                         trace_kvm_cr_read(cr, val);
4307                         skip_emulated_instruction(vcpu);
4308                         return 1;
4309                 }
4310                 break;
4311         case 3: /* lmsw */
4312                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4313                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4314                 kvm_lmsw(vcpu, val);
4315
4316                 skip_emulated_instruction(vcpu);
4317                 return 1;
4318         default:
4319                 break;
4320         }
4321         vcpu->run->exit_reason = 0;
4322         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4323                (int)(exit_qualification >> 4) & 3, cr);
4324         return 0;
4325 }
4326
4327 static int handle_dr(struct kvm_vcpu *vcpu)
4328 {
4329         unsigned long exit_qualification;
4330         int dr, reg;
4331
4332         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4333         if (!kvm_require_cpl(vcpu, 0))
4334                 return 1;
4335         dr = vmcs_readl(GUEST_DR7);
4336         if (dr & DR7_GD) {
4337                 /*
4338                  * As the vm-exit takes precedence over the debug trap, we
4339                  * need to emulate the latter, either for the host or the
4340                  * guest debugging itself.
4341                  */
4342                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4343                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4344                         vcpu->run->debug.arch.dr7 = dr;
4345                         vcpu->run->debug.arch.pc =
4346                                 vmcs_readl(GUEST_CS_BASE) +
4347                                 vmcs_readl(GUEST_RIP);
4348                         vcpu->run->debug.arch.exception = DB_VECTOR;
4349                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4350                         return 0;
4351                 } else {
4352                         vcpu->arch.dr7 &= ~DR7_GD;
4353                         vcpu->arch.dr6 |= DR6_BD;
4354                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4355                         kvm_queue_exception(vcpu, DB_VECTOR);
4356                         return 1;
4357                 }
4358         }
4359
4360         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4361         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4362         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4363         if (exit_qualification & TYPE_MOV_FROM_DR) {
4364                 unsigned long val;
4365                 if (!kvm_get_dr(vcpu, dr, &val))
4366                         kvm_register_write(vcpu, reg, val);
4367         } else
4368                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4369         skip_emulated_instruction(vcpu);
4370         return 1;
4371 }
4372
4373 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4374 {
4375         vmcs_writel(GUEST_DR7, val);
4376 }
4377
4378 static int handle_cpuid(struct kvm_vcpu *vcpu)
4379 {
4380         kvm_emulate_cpuid(vcpu);
4381         return 1;
4382 }
4383
4384 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4385 {
4386         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4387         u64 data;
4388
4389         if (vmx_get_msr(vcpu, ecx, &data)) {
4390                 trace_kvm_msr_read_ex(ecx);
4391                 kvm_inject_gp(vcpu, 0);
4392                 return 1;
4393         }
4394
4395         trace_kvm_msr_read(ecx, data);
4396
4397         /* FIXME: handling of bits 32:63 of rax, rdx */
4398         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4399         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4400         skip_emulated_instruction(vcpu);
4401         return 1;
4402 }
4403
4404 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4405 {
4406         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4407         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4408                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4409
4410         if (vmx_set_msr(vcpu, ecx, data) != 0) {
4411                 trace_kvm_msr_write_ex(ecx, data);
4412                 kvm_inject_gp(vcpu, 0);
4413                 return 1;
4414         }
4415
4416         trace_kvm_msr_write(ecx, data);
4417         skip_emulated_instruction(vcpu);
4418         return 1;
4419 }
4420
4421 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4422 {
4423         kvm_make_request(KVM_REQ_EVENT, vcpu);
4424         return 1;
4425 }
4426
4427 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4428 {
4429         u32 cpu_based_vm_exec_control;
4430
4431         /* clear pending irq */
4432         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4433         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4434         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4435
4436         kvm_make_request(KVM_REQ_EVENT, vcpu);
4437
4438         ++vcpu->stat.irq_window_exits;
4439
4440         /*
4441          * If the user space waits to inject interrupts, exit as soon as
4442          * possible
4443          */
4444         if (!irqchip_in_kernel(vcpu->kvm) &&
4445             vcpu->run->request_interrupt_window &&
4446             !kvm_cpu_has_interrupt(vcpu)) {
4447                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4448                 return 0;
4449         }
4450         return 1;
4451 }
4452
4453 static int handle_halt(struct kvm_vcpu *vcpu)
4454 {
4455         skip_emulated_instruction(vcpu);
4456         return kvm_emulate_halt(vcpu);
4457 }
4458
4459 static int handle_vmcall(struct kvm_vcpu *vcpu)
4460 {
4461         skip_emulated_instruction(vcpu);
4462         kvm_emulate_hypercall(vcpu);
4463         return 1;
4464 }
4465
4466 static int handle_invd(struct kvm_vcpu *vcpu)
4467 {
4468         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4469 }
4470
4471 static int handle_invlpg(struct kvm_vcpu *vcpu)
4472 {
4473         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4474
4475         kvm_mmu_invlpg(vcpu, exit_qualification);
4476         skip_emulated_instruction(vcpu);
4477         return 1;
4478 }
4479
4480 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4481 {
4482         skip_emulated_instruction(vcpu);
4483         kvm_emulate_wbinvd(vcpu);
4484         return 1;
4485 }
4486
4487 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4488 {
4489         u64 new_bv = kvm_read_edx_eax(vcpu);
4490         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4491
4492         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4493                 skip_emulated_instruction(vcpu);
4494         return 1;
4495 }
4496
4497 static int handle_apic_access(struct kvm_vcpu *vcpu)
4498 {
4499         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4500 }
4501
4502 static int handle_task_switch(struct kvm_vcpu *vcpu)
4503 {
4504         struct vcpu_vmx *vmx = to_vmx(vcpu);
4505         unsigned long exit_qualification;
4506         bool has_error_code = false;
4507         u32 error_code = 0;
4508         u16 tss_selector;
4509         int reason, type, idt_v;
4510
4511         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4512         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4513
4514         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4515
4516         reason = (u32)exit_qualification >> 30;
4517         if (reason == TASK_SWITCH_GATE && idt_v) {
4518                 switch (type) {
4519                 case INTR_TYPE_NMI_INTR:
4520                         vcpu->arch.nmi_injected = false;
4521                         vmx_set_nmi_mask(vcpu, true);
4522                         break;
4523                 case INTR_TYPE_EXT_INTR:
4524                 case INTR_TYPE_SOFT_INTR:
4525                         kvm_clear_interrupt_queue(vcpu);
4526                         break;
4527                 case INTR_TYPE_HARD_EXCEPTION:
4528                         if (vmx->idt_vectoring_info &
4529                             VECTORING_INFO_DELIVER_CODE_MASK) {
4530                                 has_error_code = true;
4531                                 error_code =
4532                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
4533                         }
4534                         /* fall through */
4535                 case INTR_TYPE_SOFT_EXCEPTION:
4536                         kvm_clear_exception_queue(vcpu);
4537                         break;
4538                 default:
4539                         break;
4540                 }
4541         }
4542         tss_selector = exit_qualification;
4543
4544         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4545                        type != INTR_TYPE_EXT_INTR &&
4546                        type != INTR_TYPE_NMI_INTR))
4547                 skip_emulated_instruction(vcpu);
4548
4549         if (kvm_task_switch(vcpu, tss_selector, reason,
4550                                 has_error_code, error_code) == EMULATE_FAIL) {
4551                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4552                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4553                 vcpu->run->internal.ndata = 0;
4554                 return 0;
4555         }
4556
4557         /* clear all local breakpoint enable flags */
4558         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4559
4560         /*
4561          * TODO: What about debug traps on tss switch?
4562          *       Are we supposed to inject them and update dr6?
4563          */
4564
4565         return 1;
4566 }
4567
4568 static int handle_ept_violation(struct kvm_vcpu *vcpu)
4569 {
4570         unsigned long exit_qualification;
4571         gpa_t gpa;
4572         int gla_validity;
4573
4574         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4575
4576         if (exit_qualification & (1 << 6)) {
4577                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
4578                 return -EINVAL;
4579         }
4580
4581         gla_validity = (exit_qualification >> 7) & 0x3;
4582         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4583                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4584                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4585                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
4586                         vmcs_readl(GUEST_LINEAR_ADDRESS));
4587                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4588                         (long unsigned int)exit_qualification);
4589                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4590                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
4591                 return 0;
4592         }
4593
4594         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4595         trace_kvm_page_fault(gpa, exit_qualification);
4596         return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
4597 }
4598
4599 static u64 ept_rsvd_mask(u64 spte, int level)
4600 {
4601         int i;
4602         u64 mask = 0;
4603
4604         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4605                 mask |= (1ULL << i);
4606
4607         if (level > 2)
4608                 /* bits 7:3 reserved */
4609                 mask |= 0xf8;
4610         else if (level == 2) {
4611                 if (spte & (1ULL << 7))
4612                         /* 2MB ref, bits 20:12 reserved */
4613                         mask |= 0x1ff000;
4614                 else
4615                         /* bits 6:3 reserved */
4616                         mask |= 0x78;
4617         }
4618
4619         return mask;
4620 }
4621
4622 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4623                                        int level)
4624 {
4625         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4626
4627         /* 010b (write-only) */
4628         WARN_ON((spte & 0x7) == 0x2);
4629
4630         /* 110b (write/execute) */
4631         WARN_ON((spte & 0x7) == 0x6);
4632
4633         /* 100b (execute-only) and value not supported by logical processor */
4634         if (!cpu_has_vmx_ept_execute_only())
4635                 WARN_ON((spte & 0x7) == 0x4);
4636
4637         /* not 000b */
4638         if ((spte & 0x7)) {
4639                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4640
4641                 if (rsvd_bits != 0) {
4642                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4643                                          __func__, rsvd_bits);
4644                         WARN_ON(1);
4645                 }
4646
4647                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4648                         u64 ept_mem_type = (spte & 0x38) >> 3;
4649
4650                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
4651                             ept_mem_type == 7) {
4652                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4653                                                 __func__, ept_mem_type);
4654                                 WARN_ON(1);
4655                         }
4656                 }
4657         }
4658 }
4659
4660 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
4661 {
4662         u64 sptes[4];
4663         int nr_sptes, i;
4664         gpa_t gpa;
4665
4666         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4667
4668         printk(KERN_ERR "EPT: Misconfiguration.\n");
4669         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4670
4671         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4672
4673         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4674                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4675
4676         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4677         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
4678
4679         return 0;
4680 }
4681
4682 static int handle_nmi_window(struct kvm_vcpu *vcpu)
4683 {
4684         u32 cpu_based_vm_exec_control;
4685
4686         /* clear pending NMI */
4687         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4688         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4689         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4690         ++vcpu->stat.nmi_window_exits;
4691         kvm_make_request(KVM_REQ_EVENT, vcpu);
4692
4693         return 1;
4694 }
4695
4696 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
4697 {
4698         struct vcpu_vmx *vmx = to_vmx(vcpu);
4699         enum emulation_result err = EMULATE_DONE;
4700         int ret = 1;
4701         u32 cpu_exec_ctrl;
4702         bool intr_window_requested;
4703
4704         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4705         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
4706
4707         while (!guest_state_valid(vcpu)) {
4708                 if (intr_window_requested
4709                     && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
4710                         return handle_interrupt_window(&vmx->vcpu);
4711
4712                 err = emulate_instruction(vcpu, 0);
4713
4714                 if (err == EMULATE_DO_MMIO) {
4715                         ret = 0;
4716                         goto out;
4717                 }
4718
4719                 if (err != EMULATE_DONE)
4720                         return 0;
4721
4722                 if (signal_pending(current))
4723                         goto out;
4724                 if (need_resched())
4725                         schedule();
4726         }
4727
4728         vmx->emulation_required = 0;
4729 out:
4730         return ret;
4731 }
4732
4733 /*
4734  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
4735  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
4736  */
4737 static int handle_pause(struct kvm_vcpu *vcpu)
4738 {
4739         skip_emulated_instruction(vcpu);
4740         kvm_vcpu_on_spin(vcpu);
4741
4742         return 1;
4743 }
4744
4745 static int handle_invalid_op(struct kvm_vcpu *vcpu)
4746 {
4747         kvm_queue_exception(vcpu, UD_VECTOR);
4748         return 1;
4749 }
4750
4751 /*
4752  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
4753  * We could reuse a single VMCS for all the L2 guests, but we also want the
4754  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
4755  * allows keeping them loaded on the processor, and in the future will allow
4756  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
4757  * every entry if they never change.
4758  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
4759  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
4760  *
4761  * The following functions allocate and free a vmcs02 in this pool.
4762  */
4763
4764 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
4765 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
4766 {
4767         struct vmcs02_list *item;
4768         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4769                 if (item->vmptr == vmx->nested.current_vmptr) {
4770                         list_move(&item->list, &vmx->nested.vmcs02_pool);
4771                         return &item->vmcs02;
4772                 }
4773
4774         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
4775                 /* Recycle the least recently used VMCS. */
4776                 item = list_entry(vmx->nested.vmcs02_pool.prev,
4777                         struct vmcs02_list, list);
4778                 item->vmptr = vmx->nested.current_vmptr;
4779                 list_move(&item->list, &vmx->nested.vmcs02_pool);
4780                 return &item->vmcs02;
4781         }
4782
4783         /* Create a new VMCS */
4784         item = (struct vmcs02_list *)
4785                 kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
4786         if (!item)
4787                 return NULL;
4788         item->vmcs02.vmcs = alloc_vmcs();
4789         if (!item->vmcs02.vmcs) {
4790                 kfree(item);
4791                 return NULL;
4792         }
4793         loaded_vmcs_init(&item->vmcs02);
4794         item->vmptr = vmx->nested.current_vmptr;
4795         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
4796         vmx->nested.vmcs02_num++;
4797         return &item->vmcs02;
4798 }
4799
4800 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
4801 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
4802 {
4803         struct vmcs02_list *item;
4804         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4805                 if (item->vmptr == vmptr) {
4806                         free_loaded_vmcs(&item->vmcs02);
4807                         list_del(&item->list);
4808                         kfree(item);
4809                         vmx->nested.vmcs02_num--;
4810                         return;
4811                 }
4812 }
4813
4814 /*
4815  * Free all VMCSs saved for this vcpu, except the one pointed by
4816  * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
4817  * currently used, if running L2), and vmcs01 when running L2.
4818  */
4819 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
4820 {
4821         struct vmcs02_list *item, *n;
4822         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
4823                 if (vmx->loaded_vmcs != &item->vmcs02)
4824                         free_loaded_vmcs(&item->vmcs02);
4825                 list_del(&item->list);
4826                 kfree(item);
4827         }
4828         vmx->nested.vmcs02_num = 0;
4829
4830         if (vmx->loaded_vmcs != &vmx->vmcs01)
4831                 free_loaded_vmcs(&vmx->vmcs01);
4832 }
4833
4834 /*
4835  * Emulate the VMXON instruction.
4836  * Currently, we just remember that VMX is active, and do not save or even
4837  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4838  * do not currently need to store anything in that guest-allocated memory
4839  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4840  * argument is different from the VMXON pointer (which the spec says they do).
4841  */
4842 static int handle_vmon(struct kvm_vcpu *vcpu)
4843 {
4844         struct kvm_segment cs;
4845         struct vcpu_vmx *vmx = to_vmx(vcpu);
4846
4847         /* The Intel VMX Instruction Reference lists a bunch of bits that
4848          * are prerequisite to running VMXON, most notably cr4.VMXE must be
4849          * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
4850          * Otherwise, we should fail with #UD. We test these now:
4851          */
4852         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
4853             !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
4854             (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4855                 kvm_queue_exception(vcpu, UD_VECTOR);
4856                 return 1;
4857         }
4858
4859         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4860         if (is_long_mode(vcpu) && !cs.l) {
4861                 kvm_queue_exception(vcpu, UD_VECTOR);
4862                 return 1;
4863         }
4864
4865         if (vmx_get_cpl(vcpu)) {
4866                 kvm_inject_gp(vcpu, 0);
4867                 return 1;
4868         }
4869
4870         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
4871         vmx->nested.vmcs02_num = 0;
4872
4873         vmx->nested.vmxon = true;
4874
4875         skip_emulated_instruction(vcpu);
4876         return 1;
4877 }
4878
4879 /*
4880  * Intel's VMX Instruction Reference specifies a common set of prerequisites
4881  * for running VMX instructions (except VMXON, whose prerequisites are
4882  * slightly different). It also specifies what exception to inject otherwise.
4883  */
4884 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
4885 {
4886         struct kvm_segment cs;
4887         struct vcpu_vmx *vmx = to_vmx(vcpu);
4888
4889         if (!vmx->nested.vmxon) {
4890                 kvm_queue_exception(vcpu, UD_VECTOR);
4891                 return 0;
4892         }
4893
4894         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4895         if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
4896             (is_long_mode(vcpu) && !cs.l)) {
4897                 kvm_queue_exception(vcpu, UD_VECTOR);
4898                 return 0;
4899         }
4900
4901         if (vmx_get_cpl(vcpu)) {
4902                 kvm_inject_gp(vcpu, 0);
4903                 return 0;
4904         }
4905
4906         return 1;
4907 }
4908
4909 /*
4910  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
4911  * just stops using VMX.
4912  */
4913 static void free_nested(struct vcpu_vmx *vmx)
4914 {
4915         if (!vmx->nested.vmxon)
4916                 return;
4917         vmx->nested.vmxon = false;
4918         if (vmx->nested.current_vmptr != -1ull) {
4919                 kunmap(vmx->nested.current_vmcs12_page);
4920                 nested_release_page(vmx->nested.current_vmcs12_page);
4921                 vmx->nested.current_vmptr = -1ull;
4922                 vmx->nested.current_vmcs12 = NULL;
4923         }
4924         /* Unpin physical memory we referred to in current vmcs02 */
4925         if (vmx->nested.apic_access_page) {
4926                 nested_release_page(vmx->nested.apic_access_page);
4927                 vmx->nested.apic_access_page = 0;
4928         }
4929
4930         nested_free_all_saved_vmcss(vmx);
4931 }
4932
4933 /* Emulate the VMXOFF instruction */
4934 static int handle_vmoff(struct kvm_vcpu *vcpu)
4935 {
4936         if (!nested_vmx_check_permission(vcpu))
4937                 return 1;
4938         free_nested(to_vmx(vcpu));
4939         skip_emulated_instruction(vcpu);
4940         return 1;
4941 }
4942
4943 /*
4944  * Decode the memory-address operand of a vmx instruction, as recorded on an
4945  * exit caused by such an instruction (run by a guest hypervisor).
4946  * On success, returns 0. When the operand is invalid, returns 1 and throws
4947  * #UD or #GP.
4948  */
4949 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
4950                                  unsigned long exit_qualification,
4951                                  u32 vmx_instruction_info, gva_t *ret)
4952 {
4953         /*
4954          * According to Vol. 3B, "Information for VM Exits Due to Instruction
4955          * Execution", on an exit, vmx_instruction_info holds most of the
4956          * addressing components of the operand. Only the displacement part
4957          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4958          * For how an actual address is calculated from all these components,
4959          * refer to Vol. 1, "Operand Addressing".
4960          */
4961         int  scaling = vmx_instruction_info & 3;
4962         int  addr_size = (vmx_instruction_info >> 7) & 7;
4963         bool is_reg = vmx_instruction_info & (1u << 10);
4964         int  seg_reg = (vmx_instruction_info >> 15) & 7;
4965         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4966         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4967         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4968         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4969
4970         if (is_reg) {
4971                 kvm_queue_exception(vcpu, UD_VECTOR);
4972                 return 1;
4973         }
4974
4975         /* Addr = segment_base + offset */
4976         /* offset = base + [index * scale] + displacement */
4977         *ret = vmx_get_segment_base(vcpu, seg_reg);
4978         if (base_is_valid)
4979                 *ret += kvm_register_read(vcpu, base_reg);
4980         if (index_is_valid)
4981                 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
4982         *ret += exit_qualification; /* holds the displacement */
4983
4984         if (addr_size == 1) /* 32 bit */
4985                 *ret &= 0xffffffff;
4986
4987         /*
4988          * TODO: throw #GP (and return 1) in various cases that the VM*
4989          * instructions require it - e.g., offset beyond segment limit,
4990          * unusable or unreadable/unwritable segment, non-canonical 64-bit
4991          * address, and so on. Currently these are not checked.
4992          */
4993         return 0;
4994 }
4995
4996 /*
4997  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
4998  * set the success or error code of an emulated VMX instruction, as specified
4999  * by Vol 2B, VMX Instruction Reference, "Conventions".
5000  */
5001 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5002 {
5003         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5004                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5005                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5006 }
5007
5008 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5009 {
5010         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5011                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5012                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5013                         | X86_EFLAGS_CF);
5014 }
5015
5016 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5017                                         u32 vm_instruction_error)
5018 {
5019         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5020                 /*
5021                  * failValid writes the error number to the current VMCS, which
5022                  * can't be done there isn't a current VMCS.
5023                  */
5024                 nested_vmx_failInvalid(vcpu);
5025                 return;
5026         }
5027         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5028                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5029                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5030                         | X86_EFLAGS_ZF);
5031         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5032 }
5033
5034 /* Emulate the VMCLEAR instruction */
5035 static int handle_vmclear(struct kvm_vcpu *vcpu)
5036 {
5037         struct vcpu_vmx *vmx = to_vmx(vcpu);
5038         gva_t gva;
5039         gpa_t vmptr;
5040         struct vmcs12 *vmcs12;
5041         struct page *page;
5042         struct x86_exception e;
5043
5044         if (!nested_vmx_check_permission(vcpu))
5045                 return 1;
5046
5047         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5048                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5049                 return 1;
5050
5051         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5052                                 sizeof(vmptr), &e)) {
5053                 kvm_inject_page_fault(vcpu, &e);
5054                 return 1;
5055         }
5056
5057         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5058                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5059                 skip_emulated_instruction(vcpu);
5060                 return 1;
5061         }
5062
5063         if (vmptr == vmx->nested.current_vmptr) {
5064                 kunmap(vmx->nested.current_vmcs12_page);
5065                 nested_release_page(vmx->nested.current_vmcs12_page);
5066                 vmx->nested.current_vmptr = -1ull;
5067                 vmx->nested.current_vmcs12 = NULL;
5068         }
5069
5070         page = nested_get_page(vcpu, vmptr);
5071         if (page == NULL) {
5072                 /*
5073                  * For accurate processor emulation, VMCLEAR beyond available
5074                  * physical memory should do nothing at all. However, it is
5075                  * possible that a nested vmx bug, not a guest hypervisor bug,
5076                  * resulted in this case, so let's shut down before doing any
5077                  * more damage:
5078                  */
5079                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5080                 return 1;
5081         }
5082         vmcs12 = kmap(page);
5083         vmcs12->launch_state = 0;
5084         kunmap(page);
5085         nested_release_page(page);
5086
5087         nested_free_vmcs02(vmx, vmptr);
5088
5089         skip_emulated_instruction(vcpu);
5090         nested_vmx_succeed(vcpu);
5091         return 1;
5092 }
5093
5094 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5095
5096 /* Emulate the VMLAUNCH instruction */
5097 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5098 {
5099         return nested_vmx_run(vcpu, true);
5100 }
5101
5102 /* Emulate the VMRESUME instruction */
5103 static int handle_vmresume(struct kvm_vcpu *vcpu)
5104 {
5105
5106         return nested_vmx_run(vcpu, false);
5107 }
5108
5109 enum vmcs_field_type {
5110         VMCS_FIELD_TYPE_U16 = 0,
5111         VMCS_FIELD_TYPE_U64 = 1,
5112         VMCS_FIELD_TYPE_U32 = 2,
5113         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5114 };
5115
5116 static inline int vmcs_field_type(unsigned long field)
5117 {
5118         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
5119                 return VMCS_FIELD_TYPE_U32;
5120         return (field >> 13) & 0x3 ;
5121 }
5122
5123 static inline int vmcs_field_readonly(unsigned long field)
5124 {
5125         return (((field >> 10) & 0x3) == 1);
5126 }
5127
5128 /*
5129  * Read a vmcs12 field. Since these can have varying lengths and we return
5130  * one type, we chose the biggest type (u64) and zero-extend the return value
5131  * to that size. Note that the caller, handle_vmread, might need to use only
5132  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5133  * 64-bit fields are to be returned).
5134  */
5135 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5136                                         unsigned long field, u64 *ret)
5137 {
5138         short offset = vmcs_field_to_offset(field);
5139         char *p;
5140
5141         if (offset < 0)
5142                 return 0;
5143
5144         p = ((char *)(get_vmcs12(vcpu))) + offset;
5145
5146         switch (vmcs_field_type(field)) {
5147         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5148                 *ret = *((natural_width *)p);
5149                 return 1;
5150         case VMCS_FIELD_TYPE_U16:
5151                 *ret = *((u16 *)p);
5152                 return 1;
5153         case VMCS_FIELD_TYPE_U32:
5154                 *ret = *((u32 *)p);
5155                 return 1;
5156         case VMCS_FIELD_TYPE_U64:
5157                 *ret = *((u64 *)p);
5158                 return 1;
5159         default:
5160                 return 0; /* can never happen. */
5161         }
5162 }
5163
5164 /*
5165  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5166  * used before) all generate the same failure when it is missing.
5167  */
5168 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5169 {
5170         struct vcpu_vmx *vmx = to_vmx(vcpu);
5171         if (vmx->nested.current_vmptr == -1ull) {
5172                 nested_vmx_failInvalid(vcpu);
5173                 skip_emulated_instruction(vcpu);
5174                 return 0;
5175         }
5176         return 1;
5177 }
5178
5179 static int handle_vmread(struct kvm_vcpu *vcpu)
5180 {
5181         unsigned long field;
5182         u64 field_value;
5183         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5184         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5185         gva_t gva = 0;
5186
5187         if (!nested_vmx_check_permission(vcpu) ||
5188             !nested_vmx_check_vmcs12(vcpu))
5189                 return 1;
5190
5191         /* Decode instruction info and find the field to read */
5192         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5193         /* Read the field, zero-extended to a u64 field_value */
5194         if (!vmcs12_read_any(vcpu, field, &field_value)) {
5195                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5196                 skip_emulated_instruction(vcpu);
5197                 return 1;
5198         }
5199         /*
5200          * Now copy part of this value to register or memory, as requested.
5201          * Note that the number of bits actually copied is 32 or 64 depending
5202          * on the guest's mode (32 or 64 bit), not on the given field's length.
5203          */
5204         if (vmx_instruction_info & (1u << 10)) {
5205                 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5206                         field_value);
5207         } else {
5208                 if (get_vmx_mem_address(vcpu, exit_qualification,
5209                                 vmx_instruction_info, &gva))
5210                         return 1;
5211                 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
5212                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5213                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5214         }
5215
5216         nested_vmx_succeed(vcpu);
5217         skip_emulated_instruction(vcpu);
5218         return 1;
5219 }
5220
5221
5222 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5223 {
5224         unsigned long field;
5225         gva_t gva;
5226         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5227         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5228         char *p;
5229         short offset;
5230         /* The value to write might be 32 or 64 bits, depending on L1's long
5231          * mode, and eventually we need to write that into a field of several
5232          * possible lengths. The code below first zero-extends the value to 64
5233          * bit (field_value), and then copies only the approriate number of
5234          * bits into the vmcs12 field.
5235          */
5236         u64 field_value = 0;
5237         struct x86_exception e;
5238
5239         if (!nested_vmx_check_permission(vcpu) ||
5240             !nested_vmx_check_vmcs12(vcpu))
5241                 return 1;
5242
5243         if (vmx_instruction_info & (1u << 10))
5244                 field_value = kvm_register_read(vcpu,
5245                         (((vmx_instruction_info) >> 3) & 0xf));
5246         else {
5247                 if (get_vmx_mem_address(vcpu, exit_qualification,
5248                                 vmx_instruction_info, &gva))
5249                         return 1;
5250                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5251                            &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5252                         kvm_inject_page_fault(vcpu, &e);
5253                         return 1;
5254                 }
5255         }
5256
5257
5258         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5259         if (vmcs_field_readonly(field)) {
5260                 nested_vmx_failValid(vcpu,
5261                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5262                 skip_emulated_instruction(vcpu);
5263                 return 1;
5264         }
5265
5266         offset = vmcs_field_to_offset(field);
5267         if (offset < 0) {
5268                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5269                 skip_emulated_instruction(vcpu);
5270                 return 1;
5271         }
5272         p = ((char *) get_vmcs12(vcpu)) + offset;
5273
5274         switch (vmcs_field_type(field)) {
5275         case VMCS_FIELD_TYPE_U16:
5276                 *(u16 *)p = field_value;
5277                 break;
5278         case VMCS_FIELD_TYPE_U32:
5279                 *(u32 *)p = field_value;
5280                 break;
5281         case VMCS_FIELD_TYPE_U64:
5282                 *(u64 *)p = field_value;
5283                 break;
5284         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5285                 *(natural_width *)p = field_value;
5286                 break;
5287         default:
5288                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5289                 skip_emulated_instruction(vcpu);
5290                 return 1;
5291         }
5292
5293         nested_vmx_succeed(vcpu);
5294         skip_emulated_instruction(vcpu);
5295         return 1;
5296 }
5297
5298 /* Emulate the VMPTRLD instruction */
5299 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5300 {
5301         struct vcpu_vmx *vmx = to_vmx(vcpu);
5302         gva_t gva;
5303         gpa_t vmptr;
5304         struct x86_exception e;
5305
5306         if (!nested_vmx_check_permission(vcpu))
5307                 return 1;
5308
5309         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5310                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5311                 return 1;
5312
5313         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5314                                 sizeof(vmptr), &e)) {
5315                 kvm_inject_page_fault(vcpu, &e);
5316                 return 1;
5317         }
5318
5319         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5320                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5321                 skip_emulated_instruction(vcpu);
5322                 return 1;
5323         }
5324
5325         if (vmx->nested.current_vmptr != vmptr) {
5326                 struct vmcs12 *new_vmcs12;
5327                 struct page *page;
5328                 page = nested_get_page(vcpu, vmptr);
5329                 if (page == NULL) {
5330                         nested_vmx_failInvalid(vcpu);
5331                         skip_emulated_instruction(vcpu);
5332                         return 1;
5333                 }
5334                 new_vmcs12 = kmap(page);
5335                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5336                         kunmap(page);
5337                         nested_release_page_clean(page);
5338                         nested_vmx_failValid(vcpu,
5339                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5340                         skip_emulated_instruction(vcpu);
5341                         return 1;
5342                 }
5343                 if (vmx->nested.current_vmptr != -1ull) {
5344                         kunmap(vmx->nested.current_vmcs12_page);
5345                         nested_release_page(vmx->nested.current_vmcs12_page);
5346                 }
5347
5348                 vmx->nested.current_vmptr = vmptr;
5349                 vmx->nested.current_vmcs12 = new_vmcs12;
5350                 vmx->nested.current_vmcs12_page = page;
5351         }
5352
5353         nested_vmx_succeed(vcpu);
5354         skip_emulated_instruction(vcpu);
5355         return 1;
5356 }
5357
5358 /* Emulate the VMPTRST instruction */
5359 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5360 {
5361         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5362         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5363         gva_t vmcs_gva;
5364         struct x86_exception e;
5365
5366         if (!nested_vmx_check_permission(vcpu))
5367                 return 1;
5368
5369         if (get_vmx_mem_address(vcpu, exit_qualification,
5370                         vmx_instruction_info, &vmcs_gva))
5371                 return 1;
5372         /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5373         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5374                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
5375                                  sizeof(u64), &e)) {
5376                 kvm_inject_page_fault(vcpu, &e);
5377                 return 1;
5378         }
5379         nested_vmx_succeed(vcpu);
5380         skip_emulated_instruction(vcpu);
5381         return 1;
5382 }
5383
5384 /*
5385  * The exit handlers return 1 if the exit was handled fully and guest execution
5386  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5387  * to be done to userspace and return 0.
5388  */
5389 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5390         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
5391         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5392         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5393         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
5394         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5395         [EXIT_REASON_CR_ACCESS]               = handle_cr,
5396         [EXIT_REASON_DR_ACCESS]               = handle_dr,
5397         [EXIT_REASON_CPUID]                   = handle_cpuid,
5398         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
5399         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5400         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5401         [EXIT_REASON_HLT]                     = handle_halt,
5402         [EXIT_REASON_INVD]                    = handle_invd,
5403         [EXIT_REASON_INVLPG]                  = handle_invlpg,
5404         [EXIT_REASON_VMCALL]                  = handle_vmcall,
5405         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
5406         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
5407         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
5408         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
5409         [EXIT_REASON_VMREAD]                  = handle_vmread,
5410         [EXIT_REASON_VMRESUME]                = handle_vmresume,
5411         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
5412         [EXIT_REASON_VMOFF]                   = handle_vmoff,
5413         [EXIT_REASON_VMON]                    = handle_vmon,
5414         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5415         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5416         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
5417         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
5418         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5419         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5420         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
5421         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5422         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5423         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
5424         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
5425 };
5426
5427 static const int kvm_vmx_max_exit_handlers =
5428         ARRAY_SIZE(kvm_vmx_exit_handlers);
5429
5430 /*
5431  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5432  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5433  * disinterest in the current event (read or write a specific MSR) by using an
5434  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5435  */
5436 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5437         struct vmcs12 *vmcs12, u32 exit_reason)
5438 {
5439         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
5440         gpa_t bitmap;
5441
5442         if (!nested_cpu_has(get_vmcs12(vcpu), CPU_BASED_USE_MSR_BITMAPS))
5443                 return 1;
5444
5445         /*
5446          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5447          * for the four combinations of read/write and low/high MSR numbers.
5448          * First we need to figure out which of the four to use:
5449          */
5450         bitmap = vmcs12->msr_bitmap;
5451         if (exit_reason == EXIT_REASON_MSR_WRITE)
5452                 bitmap += 2048;
5453         if (msr_index >= 0xc0000000) {
5454                 msr_index -= 0xc0000000;
5455                 bitmap += 1024;
5456         }
5457
5458         /* Then read the msr_index'th bit from this bitmap: */
5459         if (msr_index < 1024*8) {
5460                 unsigned char b;
5461                 kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1);
5462                 return 1 & (b >> (msr_index & 7));
5463         } else
5464                 return 1; /* let L1 handle the wrong parameter */
5465 }
5466
5467 /*
5468  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5469  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5470  * intercept (via guest_host_mask etc.) the current event.
5471  */
5472 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5473         struct vmcs12 *vmcs12)
5474 {
5475         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5476         int cr = exit_qualification & 15;
5477         int reg = (exit_qualification >> 8) & 15;
5478         unsigned long val = kvm_register_read(vcpu, reg);
5479
5480         switch ((exit_qualification >> 4) & 3) {
5481         case 0: /* mov to cr */
5482                 switch (cr) {
5483                 case 0:
5484                         if (vmcs12->cr0_guest_host_mask &
5485                             (val ^ vmcs12->cr0_read_shadow))
5486                                 return 1;
5487                         break;
5488                 case 3:
5489                         if ((vmcs12->cr3_target_count >= 1 &&
5490                                         vmcs12->cr3_target_value0 == val) ||
5491                                 (vmcs12->cr3_target_count >= 2 &&
5492                                         vmcs12->cr3_target_value1 == val) ||
5493                                 (vmcs12->cr3_target_count >= 3 &&
5494                                         vmcs12->cr3_target_value2 == val) ||
5495                                 (vmcs12->cr3_target_count >= 4 &&
5496                                         vmcs12->cr3_target_value3 == val))
5497                                 return 0;
5498                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5499                                 return 1;
5500                         break;
5501                 case 4:
5502                         if (vmcs12->cr4_guest_host_mask &
5503                             (vmcs12->cr4_read_shadow ^ val))
5504                                 return 1;
5505                         break;
5506                 case 8:
5507                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5508                                 return 1;
5509                         break;
5510                 }
5511                 break;
5512         case 2: /* clts */
5513                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5514                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
5515                         return 1;
5516                 break;
5517         case 1: /* mov from cr */
5518                 switch (cr) {
5519                 case 3:
5520                         if (vmcs12->cpu_based_vm_exec_control &
5521                             CPU_BASED_CR3_STORE_EXITING)
5522                                 return 1;
5523                         break;
5524                 case 8:
5525                         if (vmcs12->cpu_based_vm_exec_control &
5526                             CPU_BASED_CR8_STORE_EXITING)
5527                                 return 1;
5528                         break;
5529                 }
5530                 break;
5531         case 3: /* lmsw */
5532                 /*
5533                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5534                  * cr0. Other attempted changes are ignored, with no exit.
5535                  */
5536                 if (vmcs12->cr0_guest_host_mask & 0xe &
5537                     (val ^ vmcs12->cr0_read_shadow))
5538                         return 1;
5539                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5540                     !(vmcs12->cr0_read_shadow & 0x1) &&
5541                     (val & 0x1))
5542                         return 1;
5543                 break;
5544         }
5545         return 0;
5546 }
5547
5548 /*
5549  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5550  * should handle it ourselves in L0 (and then continue L2). Only call this
5551  * when in is_guest_mode (L2).
5552  */
5553 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
5554 {
5555         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
5556         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5557         struct vcpu_vmx *vmx = to_vmx(vcpu);
5558         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5559
5560         if (vmx->nested.nested_run_pending)
5561                 return 0;
5562
5563         if (unlikely(vmx->fail)) {
5564                 printk(KERN_INFO "%s failed vm entry %x\n",
5565                        __func__, vmcs_read32(VM_INSTRUCTION_ERROR));
5566                 return 1;
5567         }
5568
5569         switch (exit_reason) {
5570         case EXIT_REASON_EXCEPTION_NMI:
5571                 if (!is_exception(intr_info))
5572                         return 0;
5573                 else if (is_page_fault(intr_info))
5574                         return enable_ept;
5575                 return vmcs12->exception_bitmap &
5576                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
5577         case EXIT_REASON_EXTERNAL_INTERRUPT:
5578                 return 0;
5579         case EXIT_REASON_TRIPLE_FAULT:
5580                 return 1;
5581         case EXIT_REASON_PENDING_INTERRUPT:
5582         case EXIT_REASON_NMI_WINDOW:
5583                 /*
5584                  * prepare_vmcs02() set the CPU_BASED_VIRTUAL_INTR_PENDING bit
5585                  * (aka Interrupt Window Exiting) only when L1 turned it on,
5586                  * so if we got a PENDING_INTERRUPT exit, this must be for L1.
5587                  * Same for NMI Window Exiting.
5588                  */
5589                 return 1;
5590         case EXIT_REASON_TASK_SWITCH:
5591                 return 1;
5592         case EXIT_REASON_CPUID:
5593                 return 1;
5594         case EXIT_REASON_HLT:
5595                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5596         case EXIT_REASON_INVD:
5597                 return 1;
5598         case EXIT_REASON_INVLPG:
5599                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5600         case EXIT_REASON_RDPMC:
5601                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5602         case EXIT_REASON_RDTSC:
5603                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5604         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5605         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5606         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
5607         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
5608         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5609                 /*
5610                  * VMX instructions trap unconditionally. This allows L1 to
5611                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
5612                  */
5613                 return 1;
5614         case EXIT_REASON_CR_ACCESS:
5615                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5616         case EXIT_REASON_DR_ACCESS:
5617                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5618         case EXIT_REASON_IO_INSTRUCTION:
5619                 /* TODO: support IO bitmaps */
5620                 return 1;
5621         case EXIT_REASON_MSR_READ:
5622         case EXIT_REASON_MSR_WRITE:
5623                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5624         case EXIT_REASON_INVALID_STATE:
5625                 return 1;
5626         case EXIT_REASON_MWAIT_INSTRUCTION:
5627                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5628         case EXIT_REASON_MONITOR_INSTRUCTION:
5629                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5630         case EXIT_REASON_PAUSE_INSTRUCTION:
5631                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5632                         nested_cpu_has2(vmcs12,
5633                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5634         case EXIT_REASON_MCE_DURING_VMENTRY:
5635                 return 0;
5636         case EXIT_REASON_TPR_BELOW_THRESHOLD:
5637                 return 1;
5638         case EXIT_REASON_APIC_ACCESS:
5639                 return nested_cpu_has2(vmcs12,
5640                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
5641         case EXIT_REASON_EPT_VIOLATION:
5642         case EXIT_REASON_EPT_MISCONFIG:
5643                 return 0;
5644         case EXIT_REASON_WBINVD:
5645                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5646         case EXIT_REASON_XSETBV:
5647                 return 1;
5648         default:
5649                 return 1;
5650         }
5651 }
5652
5653 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5654 {
5655         *info1 = vmcs_readl(EXIT_QUALIFICATION);
5656         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5657 }
5658
5659 /*
5660  * The guest has exited.  See if we can fix it or if we need userspace
5661  * assistance.
5662  */
5663 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5664 {
5665         struct vcpu_vmx *vmx = to_vmx(vcpu);
5666         u32 exit_reason = vmx->exit_reason;
5667         u32 vectoring_info = vmx->idt_vectoring_info;
5668
5669         trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5670
5671         /* If guest state is invalid, start emulating */
5672         if (vmx->emulation_required && emulate_invalid_guest_state)
5673                 return handle_invalid_guest_state(vcpu);
5674
5675         /*
5676          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
5677          * we did not inject a still-pending event to L1 now because of
5678          * nested_run_pending, we need to re-enable this bit.
5679          */
5680         if (vmx->nested.nested_run_pending)
5681                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5682
5683         if (exit_reason == EXIT_REASON_VMLAUNCH ||
5684             exit_reason == EXIT_REASON_VMRESUME)
5685                 vmx->nested.nested_run_pending = 1;
5686         else
5687                 vmx->nested.nested_run_pending = 0;
5688
5689         if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
5690                 nested_vmx_vmexit(vcpu);
5691                 return 1;
5692         }
5693
5694         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5695                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5696                 vcpu->run->fail_entry.hardware_entry_failure_reason
5697                         = exit_reason;
5698                 return 0;
5699         }
5700
5701         if (unlikely(vmx->fail)) {
5702                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5703                 vcpu->run->fail_entry.hardware_entry_failure_reason
5704                         = vmcs_read32(VM_INSTRUCTION_ERROR);
5705                 return 0;
5706         }
5707
5708         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5709                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5710                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
5711                         exit_reason != EXIT_REASON_TASK_SWITCH))
5712                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5713                        "(0x%x) and exit reason is 0x%x\n",
5714                        __func__, vectoring_info, exit_reason);
5715
5716         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
5717             !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
5718                                         get_vmcs12(vcpu), vcpu)))) {
5719                 if (vmx_interrupt_allowed(vcpu)) {
5720                         vmx->soft_vnmi_blocked = 0;
5721                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
5722                            vcpu->arch.nmi_pending) {
5723                         /*
5724                          * This CPU don't support us in finding the end of an
5725                          * NMI-blocked window if the guest runs with IRQs
5726                          * disabled. So we pull the trigger after 1 s of
5727                          * futile waiting, but inform the user about this.
5728                          */
5729                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5730                                "state on VCPU %d after 1 s timeout\n",
5731                                __func__, vcpu->vcpu_id);
5732                         vmx->soft_vnmi_blocked = 0;
5733                 }
5734         }
5735
5736         if (exit_reason < kvm_vmx_max_exit_handlers
5737             && kvm_vmx_exit_handlers[exit_reason])
5738                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5739         else {
5740                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5741                 vcpu->run->hw.hardware_exit_reason = exit_reason;
5742         }
5743         return 0;
5744 }
5745
5746 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5747 {
5748         if (irr == -1 || tpr < irr) {
5749                 vmcs_write32(TPR_THRESHOLD, 0);
5750                 return;
5751         }
5752
5753         vmcs_write32(TPR_THRESHOLD, irr);
5754 }
5755
5756 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
5757 {
5758         u32 exit_intr_info;
5759
5760         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
5761               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
5762                 return;
5763
5764         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5765         exit_intr_info = vmx->exit_intr_info;
5766
5767         /* Handle machine checks before interrupts are enabled */
5768         if (is_machine_check(exit_intr_info))
5769                 kvm_machine_check();
5770
5771         /* We need to handle NMIs before interrupts are enabled */
5772         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
5773             (exit_intr_info & INTR_INFO_VALID_MASK)) {
5774                 kvm_before_handle_nmi(&vmx->vcpu);
5775                 asm("int $2");
5776                 kvm_after_handle_nmi(&vmx->vcpu);
5777         }
5778 }
5779
5780 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
5781 {
5782         u32 exit_intr_info;
5783         bool unblock_nmi;
5784         u8 vector;
5785         bool idtv_info_valid;
5786
5787         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5788
5789         if (cpu_has_virtual_nmis()) {
5790                 if (vmx->nmi_known_unmasked)
5791                         return;
5792                 /*
5793                  * Can't use vmx->exit_intr_info since we're not sure what
5794                  * the exit reason is.
5795                  */
5796                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5797                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
5798                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
5799                 /*
5800                  * SDM 3: 27.7.1.2 (September 2008)
5801                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
5802                  * a guest IRET fault.
5803                  * SDM 3: 23.2.2 (September 2008)
5804                  * Bit 12 is undefined in any of the following cases:
5805                  *  If the VM exit sets the valid bit in the IDT-vectoring
5806                  *   information field.
5807                  *  If the VM exit is due to a double fault.
5808                  */
5809                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
5810                     vector != DF_VECTOR && !idtv_info_valid)
5811                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5812                                       GUEST_INTR_STATE_NMI);
5813                 else
5814                         vmx->nmi_known_unmasked =
5815                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
5816                                   & GUEST_INTR_STATE_NMI);
5817         } else if (unlikely(vmx->soft_vnmi_blocked))
5818                 vmx->vnmi_blocked_time +=
5819                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
5820 }
5821
5822 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
5823                                       u32 idt_vectoring_info,
5824                                       int instr_len_field,
5825                                       int error_code_field)
5826 {
5827         u8 vector;
5828         int type;
5829         bool idtv_info_valid;
5830
5831         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5832
5833         vmx->vcpu.arch.nmi_injected = false;
5834         kvm_clear_exception_queue(&vmx->vcpu);
5835         kvm_clear_interrupt_queue(&vmx->vcpu);
5836
5837         if (!idtv_info_valid)
5838                 return;
5839
5840         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
5841
5842         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
5843         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
5844
5845         switch (type) {
5846         case INTR_TYPE_NMI_INTR:
5847                 vmx->vcpu.arch.nmi_injected = true;
5848                 /*
5849                  * SDM 3: 27.7.1.2 (September 2008)
5850                  * Clear bit "block by NMI" before VM entry if a NMI
5851                  * delivery faulted.
5852                  */
5853                 vmx_set_nmi_mask(&vmx->vcpu, false);
5854                 break;
5855         case INTR_TYPE_SOFT_EXCEPTION:
5856                 vmx->vcpu.arch.event_exit_inst_len =
5857                         vmcs_read32(instr_len_field);
5858                 /* fall through */
5859         case INTR_TYPE_HARD_EXCEPTION:
5860                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
5861                         u32 err = vmcs_read32(error_code_field);
5862                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
5863                 } else
5864                         kvm_queue_exception(&vmx->vcpu, vector);
5865                 break;
5866         case INTR_TYPE_SOFT_INTR:
5867                 vmx->vcpu.arch.event_exit_inst_len =
5868                         vmcs_read32(instr_len_field);
5869                 /* fall through */
5870         case INTR_TYPE_EXT_INTR:
5871                 kvm_queue_interrupt(&vmx->vcpu, vector,
5872                         type == INTR_TYPE_SOFT_INTR);
5873                 break;
5874         default:
5875                 break;
5876         }
5877 }
5878
5879 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
5880 {
5881         if (is_guest_mode(&vmx->vcpu))
5882                 return;
5883         __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
5884                                   VM_EXIT_INSTRUCTION_LEN,
5885                                   IDT_VECTORING_ERROR_CODE);
5886 }
5887
5888 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
5889 {
5890         if (is_guest_mode(vcpu))
5891                 return;
5892         __vmx_complete_interrupts(to_vmx(vcpu),
5893                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5894                                   VM_ENTRY_INSTRUCTION_LEN,
5895                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
5896
5897         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
5898 }
5899
5900 #ifdef CONFIG_X86_64
5901 #define R "r"
5902 #define Q "q"
5903 #else
5904 #define R "e"
5905 #define Q "l"
5906 #endif
5907
5908 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
5909 {
5910         struct vcpu_vmx *vmx = to_vmx(vcpu);
5911
5912         if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
5913                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5914                 if (vmcs12->idt_vectoring_info_field &
5915                                 VECTORING_INFO_VALID_MASK) {
5916                         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5917                                 vmcs12->idt_vectoring_info_field);
5918                         vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5919                                 vmcs12->vm_exit_instruction_len);
5920                         if (vmcs12->idt_vectoring_info_field &
5921                                         VECTORING_INFO_DELIVER_CODE_MASK)
5922                                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
5923                                         vmcs12->idt_vectoring_error_code);
5924                 }
5925         }
5926
5927         /* Record the guest's net vcpu time for enforced NMI injections. */
5928         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
5929                 vmx->entry_time = ktime_get();
5930
5931         /* Don't enter VMX if guest state is invalid, let the exit handler
5932            start emulation until we arrive back to a valid state */
5933         if (vmx->emulation_required && emulate_invalid_guest_state)
5934                 return;
5935
5936         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
5937                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
5938         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
5939                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
5940
5941         /* When single-stepping over STI and MOV SS, we must clear the
5942          * corresponding interruptibility bits in the guest state. Otherwise
5943          * vmentry fails as it then expects bit 14 (BS) in pending debug
5944          * exceptions being set, but that's not correct for the guest debugging
5945          * case. */
5946         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5947                 vmx_set_interrupt_shadow(vcpu, 0);
5948
5949         vmx->__launched = vmx->loaded_vmcs->launched;
5950         asm(
5951                 /* Store host registers */
5952                 "push %%"R"dx; push %%"R"bp;"
5953                 "push %%"R"cx \n\t" /* placeholder for guest rcx */
5954                 "push %%"R"cx \n\t"
5955                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
5956                 "je 1f \n\t"
5957                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
5958                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
5959                 "1: \n\t"
5960                 /* Reload cr2 if changed */
5961                 "mov %c[cr2](%0), %%"R"ax \n\t"
5962                 "mov %%cr2, %%"R"dx \n\t"
5963                 "cmp %%"R"ax, %%"R"dx \n\t"
5964                 "je 2f \n\t"
5965                 "mov %%"R"ax, %%cr2 \n\t"
5966                 "2: \n\t"
5967                 /* Check if vmlaunch of vmresume is needed */
5968                 "cmpl $0, %c[launched](%0) \n\t"
5969                 /* Load guest registers.  Don't clobber flags. */
5970                 "mov %c[rax](%0), %%"R"ax \n\t"
5971                 "mov %c[rbx](%0), %%"R"bx \n\t"
5972                 "mov %c[rdx](%0), %%"R"dx \n\t"
5973                 "mov %c[rsi](%0), %%"R"si \n\t"
5974                 "mov %c[rdi](%0), %%"R"di \n\t"
5975                 "mov %c[rbp](%0), %%"R"bp \n\t"
5976 #ifdef CONFIG_X86_64
5977                 "mov %c[r8](%0),  %%r8  \n\t"
5978                 "mov %c[r9](%0),  %%r9  \n\t"
5979                 "mov %c[r10](%0), %%r10 \n\t"
5980                 "mov %c[r11](%0), %%r11 \n\t"
5981                 "mov %c[r12](%0), %%r12 \n\t"
5982                 "mov %c[r13](%0), %%r13 \n\t"
5983                 "mov %c[r14](%0), %%r14 \n\t"
5984                 "mov %c[r15](%0), %%r15 \n\t"
5985 #endif
5986                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
5987
5988                 /* Enter guest mode */
5989                 "jne .Llaunched \n\t"
5990                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
5991                 "jmp .Lkvm_vmx_return \n\t"
5992                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
5993                 ".Lkvm_vmx_return: "
5994                 /* Save guest registers, load host registers, keep flags */
5995                 "mov %0, %c[wordsize](%%"R"sp) \n\t"
5996                 "pop %0 \n\t"
5997                 "mov %%"R"ax, %c[rax](%0) \n\t"
5998                 "mov %%"R"bx, %c[rbx](%0) \n\t"
5999                 "pop"Q" %c[rcx](%0) \n\t"
6000                 "mov %%"R"dx, %c[rdx](%0) \n\t"
6001                 "mov %%"R"si, %c[rsi](%0) \n\t"
6002                 "mov %%"R"di, %c[rdi](%0) \n\t"
6003                 "mov %%"R"bp, %c[rbp](%0) \n\t"
6004 #ifdef CONFIG_X86_64
6005                 "mov %%r8,  %c[r8](%0) \n\t"
6006                 "mov %%r9,  %c[r9](%0) \n\t"
6007                 "mov %%r10, %c[r10](%0) \n\t"
6008                 "mov %%r11, %c[r11](%0) \n\t"
6009                 "mov %%r12, %c[r12](%0) \n\t"
6010                 "mov %%r13, %c[r13](%0) \n\t"
6011                 "mov %%r14, %c[r14](%0) \n\t"
6012                 "mov %%r15, %c[r15](%0) \n\t"
6013 #endif
6014                 "mov %%cr2, %%"R"ax   \n\t"
6015                 "mov %%"R"ax, %c[cr2](%0) \n\t"
6016
6017                 "pop  %%"R"bp; pop  %%"R"dx \n\t"
6018                 "setbe %c[fail](%0) \n\t"
6019               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
6020                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
6021                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
6022                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
6023                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
6024                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
6025                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
6026                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
6027                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
6028                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
6029                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
6030 #ifdef CONFIG_X86_64
6031                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
6032                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
6033                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
6034                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
6035                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
6036                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
6037                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
6038                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6039 #endif
6040                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
6041                 [wordsize]"i"(sizeof(ulong))
6042               : "cc", "memory"
6043                 , R"ax", R"bx", R"di", R"si"
6044 #ifdef CONFIG_X86_64
6045                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
6046 #endif
6047               );
6048
6049         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6050                                   | (1 << VCPU_EXREG_RFLAGS)
6051                                   | (1 << VCPU_EXREG_CPL)
6052                                   | (1 << VCPU_EXREG_PDPTR)
6053                                   | (1 << VCPU_EXREG_SEGMENTS)
6054                                   | (1 << VCPU_EXREG_CR3));
6055         vcpu->arch.regs_dirty = 0;
6056
6057         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6058
6059         if (is_guest_mode(vcpu)) {
6060                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6061                 vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
6062                 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
6063                         vmcs12->idt_vectoring_error_code =
6064                                 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6065                         vmcs12->vm_exit_instruction_len =
6066                                 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6067                 }
6068         }
6069
6070         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
6071         vmx->loaded_vmcs->launched = 1;
6072
6073         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6074
6075         vmx_complete_atomic_exit(vmx);
6076         vmx_recover_nmi_blocking(vmx);
6077         vmx_complete_interrupts(vmx);
6078 }
6079
6080 #undef R
6081 #undef Q
6082
6083 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6084 {
6085         struct vcpu_vmx *vmx = to_vmx(vcpu);
6086
6087         free_vpid(vmx);
6088         free_nested(vmx);
6089         free_loaded_vmcs(vmx->loaded_vmcs);
6090         kfree(vmx->guest_msrs);
6091         kvm_vcpu_uninit(vcpu);
6092         kmem_cache_free(kvm_vcpu_cache, vmx);
6093 }
6094
6095 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6096 {
6097         int err;
6098         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
6099         int cpu;
6100
6101         if (!vmx)
6102                 return ERR_PTR(-ENOMEM);
6103
6104         allocate_vpid(vmx);
6105
6106         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6107         if (err)
6108                 goto free_vcpu;
6109
6110         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
6111         err = -ENOMEM;
6112         if (!vmx->guest_msrs) {
6113                 goto uninit_vcpu;
6114         }
6115
6116         vmx->loaded_vmcs = &vmx->vmcs01;
6117         vmx->loaded_vmcs->vmcs = alloc_vmcs();
6118         if (!vmx->loaded_vmcs->vmcs)
6119                 goto free_msrs;
6120         if (!vmm_exclusive)
6121                 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
6122         loaded_vmcs_init(vmx->loaded_vmcs);
6123         if (!vmm_exclusive)
6124                 kvm_cpu_vmxoff();
6125
6126         cpu = get_cpu();
6127         vmx_vcpu_load(&vmx->vcpu, cpu);
6128         vmx->vcpu.cpu = cpu;
6129         err = vmx_vcpu_setup(vmx);
6130         vmx_vcpu_put(&vmx->vcpu);
6131         put_cpu();
6132         if (err)
6133                 goto free_vmcs;
6134         if (vm_need_virtualize_apic_accesses(kvm))
6135                 err = alloc_apic_access_page(kvm);
6136                 if (err)
6137                         goto free_vmcs;
6138
6139         if (enable_ept) {
6140                 if (!kvm->arch.ept_identity_map_addr)
6141                         kvm->arch.ept_identity_map_addr =
6142                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
6143                 err = -ENOMEM;
6144                 if (alloc_identity_pagetable(kvm) != 0)
6145                         goto free_vmcs;
6146                 if (!init_rmode_identity_map(kvm))
6147                         goto free_vmcs;
6148         }
6149
6150         vmx->nested.current_vmptr = -1ull;
6151         vmx->nested.current_vmcs12 = NULL;
6152
6153         return &vmx->vcpu;
6154
6155 free_vmcs:
6156         free_vmcs(vmx->loaded_vmcs->vmcs);
6157 free_msrs:
6158         kfree(vmx->guest_msrs);
6159 uninit_vcpu:
6160         kvm_vcpu_uninit(&vmx->vcpu);
6161 free_vcpu:
6162         free_vpid(vmx);
6163         kmem_cache_free(kvm_vcpu_cache, vmx);
6164         return ERR_PTR(err);
6165 }
6166
6167 static void __init vmx_check_processor_compat(void *rtn)
6168 {
6169         struct vmcs_config vmcs_conf;
6170
6171         *(int *)rtn = 0;
6172         if (setup_vmcs_config(&vmcs_conf) < 0)
6173                 *(int *)rtn = -EIO;
6174         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6175                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6176                                 smp_processor_id());
6177                 *(int *)rtn = -EIO;
6178         }
6179 }
6180
6181 static int get_ept_level(void)
6182 {
6183         return VMX_EPT_DEFAULT_GAW + 1;
6184 }
6185
6186 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6187 {
6188         u64 ret;
6189
6190         /* For VT-d and EPT combination
6191          * 1. MMIO: always map as UC
6192          * 2. EPT with VT-d:
6193          *   a. VT-d without snooping control feature: can't guarantee the
6194          *      result, try to trust guest.
6195          *   b. VT-d with snooping control feature: snooping control feature of
6196          *      VT-d engine can guarantee the cache correctness. Just set it
6197          *      to WB to keep consistent with host. So the same as item 3.
6198          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6199          *    consistent with host MTRR
6200          */
6201         if (is_mmio)
6202                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
6203         else if (vcpu->kvm->arch.iommu_domain &&
6204                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
6205                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
6206                       VMX_EPT_MT_EPTE_SHIFT;
6207         else
6208                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
6209                         | VMX_EPT_IPAT_BIT;
6210
6211         return ret;
6212 }
6213
6214 #define _ER(x) { EXIT_REASON_##x, #x }
6215
6216 static const struct trace_print_flags vmx_exit_reasons_str[] = {
6217         _ER(EXCEPTION_NMI),
6218         _ER(EXTERNAL_INTERRUPT),
6219         _ER(TRIPLE_FAULT),
6220         _ER(PENDING_INTERRUPT),
6221         _ER(NMI_WINDOW),
6222         _ER(TASK_SWITCH),
6223         _ER(CPUID),
6224         _ER(HLT),
6225         _ER(INVLPG),
6226         _ER(RDPMC),
6227         _ER(RDTSC),
6228         _ER(VMCALL),
6229         _ER(VMCLEAR),
6230         _ER(VMLAUNCH),
6231         _ER(VMPTRLD),
6232         _ER(VMPTRST),
6233         _ER(VMREAD),
6234         _ER(VMRESUME),
6235         _ER(VMWRITE),
6236         _ER(VMOFF),
6237         _ER(VMON),
6238         _ER(CR_ACCESS),
6239         _ER(DR_ACCESS),
6240         _ER(IO_INSTRUCTION),
6241         _ER(MSR_READ),
6242         _ER(MSR_WRITE),
6243         _ER(MWAIT_INSTRUCTION),
6244         _ER(MONITOR_INSTRUCTION),
6245         _ER(PAUSE_INSTRUCTION),
6246         _ER(MCE_DURING_VMENTRY),
6247         _ER(TPR_BELOW_THRESHOLD),
6248         _ER(APIC_ACCESS),
6249         _ER(EPT_VIOLATION),
6250         _ER(EPT_MISCONFIG),
6251         _ER(WBINVD),
6252         { -1, NULL }
6253 };
6254
6255 #undef _ER
6256
6257 static int vmx_get_lpage_level(void)
6258 {
6259         if (enable_ept && !cpu_has_vmx_ept_1g_page())
6260                 return PT_DIRECTORY_LEVEL;
6261         else
6262                 /* For shadow and EPT supported 1GB page */
6263                 return PT_PDPE_LEVEL;
6264 }
6265
6266 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
6267 {
6268         struct kvm_cpuid_entry2 *best;
6269         struct vcpu_vmx *vmx = to_vmx(vcpu);
6270         u32 exec_control;
6271
6272         vmx->rdtscp_enabled = false;
6273         if (vmx_rdtscp_supported()) {
6274                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6275                 if (exec_control & SECONDARY_EXEC_RDTSCP) {
6276                         best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
6277                         if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
6278                                 vmx->rdtscp_enabled = true;
6279                         else {
6280                                 exec_control &= ~SECONDARY_EXEC_RDTSCP;
6281                                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6282                                                 exec_control);
6283                         }
6284                 }
6285         }
6286 }
6287
6288 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
6289 {
6290 }
6291
6292 /*
6293  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
6294  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
6295  * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
6296  * guest in a way that will both be appropriate to L1's requests, and our
6297  * needs. In addition to modifying the active vmcs (which is vmcs02), this
6298  * function also has additional necessary side-effects, like setting various
6299  * vcpu->arch fields.
6300  */
6301 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6302 {
6303         struct vcpu_vmx *vmx = to_vmx(vcpu);
6304         u32 exec_control;
6305
6306         vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
6307         vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
6308         vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
6309         vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
6310         vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
6311         vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
6312         vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
6313         vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
6314         vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
6315         vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
6316         vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
6317         vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
6318         vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
6319         vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
6320         vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
6321         vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
6322         vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
6323         vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
6324         vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
6325         vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
6326         vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
6327         vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
6328         vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
6329         vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
6330         vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
6331         vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
6332         vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
6333         vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
6334         vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
6335         vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
6336         vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
6337         vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
6338         vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
6339         vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
6340         vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
6341         vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
6342
6343         vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
6344         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6345                 vmcs12->vm_entry_intr_info_field);
6346         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6347                 vmcs12->vm_entry_exception_error_code);
6348         vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6349                 vmcs12->vm_entry_instruction_len);
6350         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
6351                 vmcs12->guest_interruptibility_info);
6352         vmcs_write32(GUEST_ACTIVITY_STATE, vmcs12->guest_activity_state);
6353         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
6354         vmcs_writel(GUEST_DR7, vmcs12->guest_dr7);
6355         vmcs_writel(GUEST_RFLAGS, vmcs12->guest_rflags);
6356         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
6357                 vmcs12->guest_pending_dbg_exceptions);
6358         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
6359         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
6360
6361         vmcs_write64(VMCS_LINK_POINTER, -1ull);
6362
6363         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
6364                 (vmcs_config.pin_based_exec_ctrl |
6365                  vmcs12->pin_based_vm_exec_control));
6366
6367         /*
6368          * Whether page-faults are trapped is determined by a combination of
6369          * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
6370          * If enable_ept, L0 doesn't care about page faults and we should
6371          * set all of these to L1's desires. However, if !enable_ept, L0 does
6372          * care about (at least some) page faults, and because it is not easy
6373          * (if at all possible?) to merge L0 and L1's desires, we simply ask
6374          * to exit on each and every L2 page fault. This is done by setting
6375          * MASK=MATCH=0 and (see below) EB.PF=1.
6376          * Note that below we don't need special code to set EB.PF beyond the
6377          * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
6378          * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
6379          * !enable_ept, EB.PF is 1, so the "or" will always be 1.
6380          *
6381          * A problem with this approach (when !enable_ept) is that L1 may be
6382          * injected with more page faults than it asked for. This could have
6383          * caused problems, but in practice existing hypervisors don't care.
6384          * To fix this, we will need to emulate the PFEC checking (on the L1
6385          * page tables), using walk_addr(), when injecting PFs to L1.
6386          */
6387         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
6388                 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
6389         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
6390                 enable_ept ? vmcs12->page_fault_error_code_match : 0);
6391
6392         if (cpu_has_secondary_exec_ctrls()) {
6393                 u32 exec_control = vmx_secondary_exec_control(vmx);
6394                 if (!vmx->rdtscp_enabled)
6395                         exec_control &= ~SECONDARY_EXEC_RDTSCP;
6396                 /* Take the following fields only from vmcs12 */
6397                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6398                 if (nested_cpu_has(vmcs12,
6399                                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
6400                         exec_control |= vmcs12->secondary_vm_exec_control;
6401
6402                 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
6403                         /*
6404                          * Translate L1 physical address to host physical
6405                          * address for vmcs02. Keep the page pinned, so this
6406                          * physical address remains valid. We keep a reference
6407                          * to it so we can release it later.
6408                          */
6409                         if (vmx->nested.apic_access_page) /* shouldn't happen */
6410                                 nested_release_page(vmx->nested.apic_access_page);
6411                         vmx->nested.apic_access_page =
6412                                 nested_get_page(vcpu, vmcs12->apic_access_addr);
6413                         /*
6414                          * If translation failed, no matter: This feature asks
6415                          * to exit when accessing the given address, and if it
6416                          * can never be accessed, this feature won't do
6417                          * anything anyway.
6418                          */
6419                         if (!vmx->nested.apic_access_page)
6420                                 exec_control &=
6421                                   ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6422                         else
6423                                 vmcs_write64(APIC_ACCESS_ADDR,
6424                                   page_to_phys(vmx->nested.apic_access_page));
6425                 }
6426
6427                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6428         }
6429
6430
6431         /*
6432          * Set host-state according to L0's settings (vmcs12 is irrelevant here)
6433          * Some constant fields are set here by vmx_set_constant_host_state().
6434          * Other fields are different per CPU, and will be set later when
6435          * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
6436          */
6437         vmx_set_constant_host_state();
6438
6439         /*
6440          * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
6441          * entry, but only if the current (host) sp changed from the value
6442          * we wrote last (vmx->host_rsp). This cache is no longer relevant
6443          * if we switch vmcs, and rather than hold a separate cache per vmcs,
6444          * here we just force the write to happen on entry.
6445          */
6446         vmx->host_rsp = 0;
6447
6448         exec_control = vmx_exec_control(vmx); /* L0's desires */
6449         exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
6450         exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6451         exec_control &= ~CPU_BASED_TPR_SHADOW;
6452         exec_control |= vmcs12->cpu_based_vm_exec_control;
6453         /*
6454          * Merging of IO and MSR bitmaps not currently supported.
6455          * Rather, exit every time.
6456          */
6457         exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
6458         exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
6459         exec_control |= CPU_BASED_UNCOND_IO_EXITING;
6460
6461         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6462
6463         /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
6464          * bitwise-or of what L1 wants to trap for L2, and what we want to
6465          * trap. Note that CR0.TS also needs updating - we do this later.
6466          */
6467         update_exception_bitmap(vcpu);
6468         vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
6469         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
6470
6471         /* Note: IA32_MODE, LOAD_IA32_EFER are modified by vmx_set_efer below */
6472         vmcs_write32(VM_EXIT_CONTROLS,
6473                 vmcs12->vm_exit_controls | vmcs_config.vmexit_ctrl);
6474         vmcs_write32(VM_ENTRY_CONTROLS, vmcs12->vm_entry_controls |
6475                 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
6476
6477         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)
6478                 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
6479         else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
6480                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
6481
6482
6483         set_cr4_guest_host_mask(vmx);
6484
6485         vmcs_write64(TSC_OFFSET,
6486                 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
6487
6488         if (enable_vpid) {
6489                 /*
6490                  * Trivially support vpid by letting L2s share their parent
6491                  * L1's vpid. TODO: move to a more elaborate solution, giving
6492                  * each L2 its own vpid and exposing the vpid feature to L1.
6493                  */
6494                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
6495                 vmx_flush_tlb(vcpu);
6496         }
6497
6498         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
6499                 vcpu->arch.efer = vmcs12->guest_ia32_efer;
6500         if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
6501                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
6502         else
6503                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
6504         /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
6505         vmx_set_efer(vcpu, vcpu->arch.efer);
6506
6507         /*
6508          * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
6509          * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
6510          * The CR0_READ_SHADOW is what L2 should have expected to read given
6511          * the specifications by L1; It's not enough to take
6512          * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
6513          * have more bits than L1 expected.
6514          */
6515         vmx_set_cr0(vcpu, vmcs12->guest_cr0);
6516         vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
6517
6518         vmx_set_cr4(vcpu, vmcs12->guest_cr4);
6519         vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
6520
6521         /* shadow page tables on either EPT or shadow page tables */
6522         kvm_set_cr3(vcpu, vmcs12->guest_cr3);
6523         kvm_mmu_reset_context(vcpu);
6524
6525         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
6526         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
6527 }
6528
6529 /*
6530  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
6531  * for running an L2 nested guest.
6532  */
6533 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
6534 {
6535         struct vmcs12 *vmcs12;
6536         struct vcpu_vmx *vmx = to_vmx(vcpu);
6537         int cpu;
6538         struct loaded_vmcs *vmcs02;
6539
6540         if (!nested_vmx_check_permission(vcpu) ||
6541             !nested_vmx_check_vmcs12(vcpu))
6542                 return 1;
6543
6544         skip_emulated_instruction(vcpu);
6545         vmcs12 = get_vmcs12(vcpu);
6546
6547         /*
6548          * The nested entry process starts with enforcing various prerequisites
6549          * on vmcs12 as required by the Intel SDM, and act appropriately when
6550          * they fail: As the SDM explains, some conditions should cause the
6551          * instruction to fail, while others will cause the instruction to seem
6552          * to succeed, but return an EXIT_REASON_INVALID_STATE.
6553          * To speed up the normal (success) code path, we should avoid checking
6554          * for misconfigurations which will anyway be caught by the processor
6555          * when using the merged vmcs02.
6556          */
6557         if (vmcs12->launch_state == launch) {
6558                 nested_vmx_failValid(vcpu,
6559                         launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
6560                                : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
6561                 return 1;
6562         }
6563
6564         if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
6565                         !IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
6566                 /*TODO: Also verify bits beyond physical address width are 0*/
6567                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6568                 return 1;
6569         }
6570
6571         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
6572                         !IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
6573                 /*TODO: Also verify bits beyond physical address width are 0*/
6574                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6575                 return 1;
6576         }
6577
6578         if (vmcs12->vm_entry_msr_load_count > 0 ||
6579             vmcs12->vm_exit_msr_load_count > 0 ||
6580             vmcs12->vm_exit_msr_store_count > 0) {
6581                 if (printk_ratelimit())
6582                         printk(KERN_WARNING
6583                           "%s: VMCS MSR_{LOAD,STORE} unsupported\n", __func__);
6584                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6585                 return 1;
6586         }
6587
6588         if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
6589               nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
6590             !vmx_control_verify(vmcs12->secondary_vm_exec_control,
6591               nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
6592             !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
6593               nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
6594             !vmx_control_verify(vmcs12->vm_exit_controls,
6595               nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
6596             !vmx_control_verify(vmcs12->vm_entry_controls,
6597               nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
6598         {
6599                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6600                 return 1;
6601         }
6602
6603         if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6604             ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6605                 nested_vmx_failValid(vcpu,
6606                         VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6607                 return 1;
6608         }
6609
6610         if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6611             ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6612                 nested_vmx_entry_failure(vcpu, vmcs12,
6613                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
6614                 return 1;
6615         }
6616         if (vmcs12->vmcs_link_pointer != -1ull) {
6617                 nested_vmx_entry_failure(vcpu, vmcs12,
6618                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
6619                 return 1;
6620         }
6621
6622         /*
6623          * We're finally done with prerequisite checking, and can start with
6624          * the nested entry.
6625          */
6626
6627         vmcs02 = nested_get_current_vmcs02(vmx);
6628         if (!vmcs02)
6629                 return -ENOMEM;
6630
6631         enter_guest_mode(vcpu);
6632
6633         vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
6634
6635         cpu = get_cpu();
6636         vmx->loaded_vmcs = vmcs02;
6637         vmx_vcpu_put(vcpu);
6638         vmx_vcpu_load(vcpu, cpu);
6639         vcpu->cpu = cpu;
6640         put_cpu();
6641
6642         vmcs12->launch_state = 1;
6643
6644         prepare_vmcs02(vcpu, vmcs12);
6645
6646         /*
6647          * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
6648          * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
6649          * returned as far as L1 is concerned. It will only return (and set
6650          * the success flag) when L2 exits (see nested_vmx_vmexit()).
6651          */
6652         return 1;
6653 }
6654
6655 /*
6656  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
6657  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
6658  * This function returns the new value we should put in vmcs12.guest_cr0.
6659  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
6660  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
6661  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
6662  *     didn't trap the bit, because if L1 did, so would L0).
6663  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
6664  *     been modified by L2, and L1 knows it. So just leave the old value of
6665  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
6666  *     isn't relevant, because if L0 traps this bit it can set it to anything.
6667  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
6668  *     changed these bits, and therefore they need to be updated, but L0
6669  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
6670  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
6671  */
6672 static inline unsigned long
6673 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6674 {
6675         return
6676         /*1*/   (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
6677         /*2*/   (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
6678         /*3*/   (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
6679                         vcpu->arch.cr0_guest_owned_bits));
6680 }
6681
6682 static inline unsigned long
6683 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6684 {
6685         return
6686         /*1*/   (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
6687         /*2*/   (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
6688         /*3*/   (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
6689                         vcpu->arch.cr4_guest_owned_bits));
6690 }
6691
6692 /*
6693  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
6694  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
6695  * and this function updates it to reflect the changes to the guest state while
6696  * L2 was running (and perhaps made some exits which were handled directly by L0
6697  * without going back to L1), and to reflect the exit reason.
6698  * Note that we do not have to copy here all VMCS fields, just those that
6699  * could have changed by the L2 guest or the exit - i.e., the guest-state and
6700  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
6701  * which already writes to vmcs12 directly.
6702  */
6703 void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6704 {
6705         /* update guest state fields: */
6706         vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
6707         vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
6708
6709         kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
6710         vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6711         vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
6712         vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
6713
6714         vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
6715         vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
6716         vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
6717         vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
6718         vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
6719         vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
6720         vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
6721         vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
6722         vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
6723         vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
6724         vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
6725         vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
6726         vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
6727         vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
6728         vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
6729         vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
6730         vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
6731         vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
6732         vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
6733         vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
6734         vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
6735         vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
6736         vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
6737         vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
6738         vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
6739         vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
6740         vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
6741         vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
6742         vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
6743         vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
6744         vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
6745         vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
6746         vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
6747         vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
6748         vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
6749         vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
6750
6751         vmcs12->guest_activity_state = vmcs_read32(GUEST_ACTIVITY_STATE);
6752         vmcs12->guest_interruptibility_info =
6753                 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
6754         vmcs12->guest_pending_dbg_exceptions =
6755                 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
6756
6757         /* TODO: These cannot have changed unless we have MSR bitmaps and
6758          * the relevant bit asks not to trap the change */
6759         vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
6760         if (vmcs12->vm_entry_controls & VM_EXIT_SAVE_IA32_PAT)
6761                 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
6762         vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
6763         vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
6764         vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
6765
6766         /* update exit information fields: */
6767
6768         vmcs12->vm_exit_reason  = vmcs_read32(VM_EXIT_REASON);
6769         vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6770
6771         vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6772         vmcs12->vm_exit_intr_error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6773         vmcs12->idt_vectoring_info_field =
6774                 vmcs_read32(IDT_VECTORING_INFO_FIELD);
6775         vmcs12->idt_vectoring_error_code =
6776                 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6777         vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6778         vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6779
6780         /* clear vm-entry fields which are to be cleared on exit */
6781         if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6782                 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
6783 }
6784
6785 /*
6786  * A part of what we need to when the nested L2 guest exits and we want to
6787  * run its L1 parent, is to reset L1's guest state to the host state specified
6788  * in vmcs12.
6789  * This function is to be called not only on normal nested exit, but also on
6790  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
6791  * Failures During or After Loading Guest State").
6792  * This function should be called when the active VMCS is L1's (vmcs01).
6793  */
6794 void load_vmcs12_host_state(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6795 {
6796         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
6797                 vcpu->arch.efer = vmcs12->host_ia32_efer;
6798         if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
6799                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
6800         else
6801                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
6802         vmx_set_efer(vcpu, vcpu->arch.efer);
6803
6804         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
6805         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
6806         /*
6807          * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
6808          * actually changed, because it depends on the current state of
6809          * fpu_active (which may have changed).
6810          * Note that vmx_set_cr0 refers to efer set above.
6811          */
6812         kvm_set_cr0(vcpu, vmcs12->host_cr0);
6813         /*
6814          * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
6815          * to apply the same changes to L1's vmcs. We just set cr0 correctly,
6816          * but we also need to update cr0_guest_host_mask and exception_bitmap.
6817          */
6818         update_exception_bitmap(vcpu);
6819         vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
6820         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
6821
6822         /*
6823          * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
6824          * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
6825          */
6826         vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
6827         kvm_set_cr4(vcpu, vmcs12->host_cr4);
6828
6829         /* shadow page tables on either EPT or shadow page tables */
6830         kvm_set_cr3(vcpu, vmcs12->host_cr3);
6831         kvm_mmu_reset_context(vcpu);
6832
6833         if (enable_vpid) {
6834                 /*
6835                  * Trivially support vpid by letting L2s share their parent
6836                  * L1's vpid. TODO: move to a more elaborate solution, giving
6837                  * each L2 its own vpid and exposing the vpid feature to L1.
6838                  */
6839                 vmx_flush_tlb(vcpu);
6840         }
6841
6842
6843         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
6844         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
6845         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
6846         vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
6847         vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
6848         vmcs_writel(GUEST_TR_BASE, vmcs12->host_tr_base);
6849         vmcs_writel(GUEST_GS_BASE, vmcs12->host_gs_base);
6850         vmcs_writel(GUEST_FS_BASE, vmcs12->host_fs_base);
6851         vmcs_write16(GUEST_ES_SELECTOR, vmcs12->host_es_selector);
6852         vmcs_write16(GUEST_CS_SELECTOR, vmcs12->host_cs_selector);
6853         vmcs_write16(GUEST_SS_SELECTOR, vmcs12->host_ss_selector);
6854         vmcs_write16(GUEST_DS_SELECTOR, vmcs12->host_ds_selector);
6855         vmcs_write16(GUEST_FS_SELECTOR, vmcs12->host_fs_selector);
6856         vmcs_write16(GUEST_GS_SELECTOR, vmcs12->host_gs_selector);
6857         vmcs_write16(GUEST_TR_SELECTOR, vmcs12->host_tr_selector);
6858
6859         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT)
6860                 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
6861         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
6862                 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
6863                         vmcs12->host_ia32_perf_global_ctrl);
6864 }
6865
6866 /*
6867  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
6868  * and modify vmcs12 to make it see what it would expect to see there if
6869  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
6870  */
6871 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
6872 {
6873         struct vcpu_vmx *vmx = to_vmx(vcpu);
6874         int cpu;
6875         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6876
6877         leave_guest_mode(vcpu);
6878         prepare_vmcs12(vcpu, vmcs12);
6879
6880         cpu = get_cpu();
6881         vmx->loaded_vmcs = &vmx->vmcs01;
6882         vmx_vcpu_put(vcpu);
6883         vmx_vcpu_load(vcpu, cpu);
6884         vcpu->cpu = cpu;
6885         put_cpu();
6886
6887         /* if no vmcs02 cache requested, remove the one we used */
6888         if (VMCS02_POOL_SIZE == 0)
6889                 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
6890
6891         load_vmcs12_host_state(vcpu, vmcs12);
6892
6893         /* Update TSC_OFFSET if vmx_adjust_tsc_offset() was used while L2 ran */
6894         vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
6895
6896         /* This is needed for same reason as it was needed in prepare_vmcs02 */
6897         vmx->host_rsp = 0;
6898
6899         /* Unpin physical memory we referred to in vmcs02 */
6900         if (vmx->nested.apic_access_page) {
6901                 nested_release_page(vmx->nested.apic_access_page);
6902                 vmx->nested.apic_access_page = 0;
6903         }
6904
6905         /*
6906          * Exiting from L2 to L1, we're now back to L1 which thinks it just
6907          * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
6908          * success or failure flag accordingly.
6909          */
6910         if (unlikely(vmx->fail)) {
6911                 vmx->fail = 0;
6912                 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
6913         } else
6914                 nested_vmx_succeed(vcpu);
6915 }
6916
6917 /*
6918  * L1's failure to enter L2 is a subset of a normal exit, as explained in
6919  * 23.7 "VM-entry failures during or after loading guest state" (this also
6920  * lists the acceptable exit-reason and exit-qualification parameters).
6921  * It should only be called before L2 actually succeeded to run, and when
6922  * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
6923  */
6924 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
6925                         struct vmcs12 *vmcs12,
6926                         u32 reason, unsigned long qualification)
6927 {
6928         load_vmcs12_host_state(vcpu, vmcs12);
6929         vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
6930         vmcs12->exit_qualification = qualification;
6931         nested_vmx_succeed(vcpu);
6932 }
6933
6934 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
6935                                struct x86_instruction_info *info,
6936                                enum x86_intercept_stage stage)
6937 {
6938         return X86EMUL_CONTINUE;
6939 }
6940
6941 static struct kvm_x86_ops vmx_x86_ops = {
6942         .cpu_has_kvm_support = cpu_has_kvm_support,
6943         .disabled_by_bios = vmx_disabled_by_bios,
6944         .hardware_setup = hardware_setup,
6945         .hardware_unsetup = hardware_unsetup,
6946         .check_processor_compatibility = vmx_check_processor_compat,
6947         .hardware_enable = hardware_enable,
6948         .hardware_disable = hardware_disable,
6949         .cpu_has_accelerated_tpr = report_flexpriority,
6950
6951         .vcpu_create = vmx_create_vcpu,
6952         .vcpu_free = vmx_free_vcpu,
6953         .vcpu_reset = vmx_vcpu_reset,
6954
6955         .prepare_guest_switch = vmx_save_host_state,
6956         .vcpu_load = vmx_vcpu_load,
6957         .vcpu_put = vmx_vcpu_put,
6958
6959         .set_guest_debug = set_guest_debug,
6960         .get_msr = vmx_get_msr,
6961         .set_msr = vmx_set_msr,
6962         .get_segment_base = vmx_get_segment_base,
6963         .get_segment = vmx_get_segment,
6964         .set_segment = vmx_set_segment,
6965         .get_cpl = vmx_get_cpl,
6966         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
6967         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
6968         .decache_cr3 = vmx_decache_cr3,
6969         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
6970         .set_cr0 = vmx_set_cr0,
6971         .set_cr3 = vmx_set_cr3,
6972         .set_cr4 = vmx_set_cr4,
6973         .set_efer = vmx_set_efer,
6974         .get_idt = vmx_get_idt,
6975         .set_idt = vmx_set_idt,
6976         .get_gdt = vmx_get_gdt,
6977         .set_gdt = vmx_set_gdt,
6978         .set_dr7 = vmx_set_dr7,
6979         .cache_reg = vmx_cache_reg,
6980         .get_rflags = vmx_get_rflags,
6981         .set_rflags = vmx_set_rflags,
6982         .fpu_activate = vmx_fpu_activate,
6983         .fpu_deactivate = vmx_fpu_deactivate,
6984
6985         .tlb_flush = vmx_flush_tlb,
6986
6987         .run = vmx_vcpu_run,
6988         .handle_exit = vmx_handle_exit,
6989         .skip_emulated_instruction = skip_emulated_instruction,
6990         .set_interrupt_shadow = vmx_set_interrupt_shadow,
6991         .get_interrupt_shadow = vmx_get_interrupt_shadow,
6992         .patch_hypercall = vmx_patch_hypercall,
6993         .set_irq = vmx_inject_irq,
6994         .set_nmi = vmx_inject_nmi,
6995         .queue_exception = vmx_queue_exception,
6996         .cancel_injection = vmx_cancel_injection,
6997         .interrupt_allowed = vmx_interrupt_allowed,
6998         .nmi_allowed = vmx_nmi_allowed,
6999         .get_nmi_mask = vmx_get_nmi_mask,
7000         .set_nmi_mask = vmx_set_nmi_mask,
7001         .enable_nmi_window = enable_nmi_window,
7002         .enable_irq_window = enable_irq_window,
7003         .update_cr8_intercept = update_cr8_intercept,
7004
7005         .set_tss_addr = vmx_set_tss_addr,
7006         .get_tdp_level = get_ept_level,
7007         .get_mt_mask = vmx_get_mt_mask,
7008
7009         .get_exit_info = vmx_get_exit_info,
7010         .exit_reasons_str = vmx_exit_reasons_str,
7011
7012         .get_lpage_level = vmx_get_lpage_level,
7013
7014         .cpuid_update = vmx_cpuid_update,
7015
7016         .rdtscp_supported = vmx_rdtscp_supported,
7017
7018         .set_supported_cpuid = vmx_set_supported_cpuid,
7019
7020         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7021
7022         .set_tsc_khz = vmx_set_tsc_khz,
7023         .write_tsc_offset = vmx_write_tsc_offset,
7024         .adjust_tsc_offset = vmx_adjust_tsc_offset,
7025         .compute_tsc_offset = vmx_compute_tsc_offset,
7026
7027         .set_tdp_cr3 = vmx_set_cr3,
7028
7029         .check_intercept = vmx_check_intercept,
7030 };
7031
7032 static int __init vmx_init(void)
7033 {
7034         int r, i;
7035
7036         rdmsrl_safe(MSR_EFER, &host_efer);
7037
7038         for (i = 0; i < NR_VMX_MSR; ++i)
7039                 kvm_define_shared_msr(i, vmx_msr_index[i]);
7040
7041         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
7042         if (!vmx_io_bitmap_a)
7043                 return -ENOMEM;
7044
7045         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
7046         if (!vmx_io_bitmap_b) {
7047                 r = -ENOMEM;
7048                 goto out;
7049         }
7050
7051         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
7052         if (!vmx_msr_bitmap_legacy) {
7053                 r = -ENOMEM;
7054                 goto out1;
7055         }
7056
7057         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
7058         if (!vmx_msr_bitmap_longmode) {
7059                 r = -ENOMEM;
7060                 goto out2;
7061         }
7062
7063         /*
7064          * Allow direct access to the PC debug port (it is often used for I/O
7065          * delays, but the vmexits simply slow things down).
7066          */
7067         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
7068         clear_bit(0x80, vmx_io_bitmap_a);
7069
7070         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
7071
7072         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
7073         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
7074
7075         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7076
7077         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7078                      __alignof__(struct vcpu_vmx), THIS_MODULE);
7079         if (r)
7080                 goto out3;
7081
7082         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
7083         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
7084         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
7085         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
7086         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
7087         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
7088
7089         if (enable_ept) {
7090                 bypass_guest_pf = 0;
7091                 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
7092                                 VMX_EPT_EXECUTABLE_MASK);
7093                 kvm_enable_tdp();
7094         } else
7095                 kvm_disable_tdp();
7096
7097         if (bypass_guest_pf)
7098                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
7099
7100         return 0;
7101
7102 out3:
7103         free_page((unsigned long)vmx_msr_bitmap_longmode);
7104 out2:
7105         free_page((unsigned long)vmx_msr_bitmap_legacy);
7106 out1:
7107         free_page((unsigned long)vmx_io_bitmap_b);
7108 out:
7109         free_page((unsigned long)vmx_io_bitmap_a);
7110         return r;
7111 }
7112
7113 static void __exit vmx_exit(void)
7114 {
7115         free_page((unsigned long)vmx_msr_bitmap_legacy);
7116         free_page((unsigned long)vmx_msr_bitmap_longmode);
7117         free_page((unsigned long)vmx_io_bitmap_b);
7118         free_page((unsigned long)vmx_io_bitmap_a);
7119
7120         kvm_exit();
7121 }
7122
7123 module_init(vmx_init)
7124 module_exit(vmx_exit)