Linux Audio

Check our new training course

Loading...
   1/*
   2 * Kernel-based Virtual Machine driver for Linux
   3 *
   4 * This module enables machines with Intel VT-x extensions to run virtual
   5 * machines without emulation or binary translation.
   6 *
   7 * Copyright (C) 2006 Qumranet, Inc.
   8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9 *
  10 * Authors:
  11 *   Avi Kivity   <avi@qumranet.com>
  12 *   Yaniv Kamay  <yaniv@qumranet.com>
  13 *
  14 * This work is licensed under the terms of the GNU GPL, version 2.  See
  15 * the COPYING file in the top-level directory.
  16 *
  17 */
  18
  19#include "irq.h"
  20#include "mmu.h"
  21#include "cpuid.h"
  22
  23#include <linux/kvm_host.h>
  24#include <linux/module.h>
  25#include <linux/kernel.h>
  26#include <linux/mm.h>
  27#include <linux/highmem.h>
  28#include <linux/sched.h>
  29#include <linux/moduleparam.h>
  30#include <linux/mod_devicetable.h>
  31#include <linux/ftrace_event.h>
  32#include <linux/slab.h>
  33#include <linux/tboot.h>
  34#include "kvm_cache_regs.h"
  35#include "x86.h"
  36
  37#include <asm/io.h>
  38#include <asm/desc.h>
  39#include <asm/vmx.h>
  40#include <asm/virtext.h>
  41#include <asm/mce.h>
  42#include <asm/i387.h>
  43#include <asm/xcr.h>
  44#include <asm/perf_event.h>
  45
  46#include "trace.h"
  47
  48#define __ex(x) __kvm_handle_fault_on_reboot(x)
  49#define __ex_clear(x, reg) \
  50	____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
  51
  52MODULE_AUTHOR("Qumranet");
  53MODULE_LICENSE("GPL");
  54
  55static const struct x86_cpu_id vmx_cpu_id[] = {
  56	X86_FEATURE_MATCH(X86_FEATURE_VMX),
  57	{}
  58};
  59MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
  60
  61static bool __read_mostly enable_vpid = 1;
  62module_param_named(vpid, enable_vpid, bool, 0444);
  63
  64static bool __read_mostly flexpriority_enabled = 1;
  65module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  66
  67static bool __read_mostly enable_ept = 1;
  68module_param_named(ept, enable_ept, bool, S_IRUGO);
  69
  70static bool __read_mostly enable_unrestricted_guest = 1;
  71module_param_named(unrestricted_guest,
  72			enable_unrestricted_guest, bool, S_IRUGO);
  73
  74static bool __read_mostly emulate_invalid_guest_state = 0;
  75module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  76
  77static bool __read_mostly vmm_exclusive = 1;
  78module_param(vmm_exclusive, bool, S_IRUGO);
  79
  80static bool __read_mostly fasteoi = 1;
  81module_param(fasteoi, bool, S_IRUGO);
  82
  83/*
  84 * If nested=1, nested virtualization is supported, i.e., guests may use
  85 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
  86 * use VMX instructions.
  87 */
  88static bool __read_mostly nested = 0;
  89module_param(nested, bool, S_IRUGO);
  90
  91#define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST				\
  92	(X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
  93#define KVM_GUEST_CR0_MASK						\
  94	(KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  95#define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST				\
  96	(X86_CR0_WP | X86_CR0_NE)
  97#define KVM_VM_CR0_ALWAYS_ON						\
  98	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
  99#define KVM_CR4_GUEST_OWNED_BITS				      \
 100	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
 101	 | X86_CR4_OSXMMEXCPT)
 102
 103#define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
 104#define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
 105
 106#define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
 107
 108/*
 109 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 110 * ple_gap:    upper bound on the amount of time between two successive
 111 *             executions of PAUSE in a loop. Also indicate if ple enabled.
 112 *             According to test, this time is usually smaller than 128 cycles.
 113 * ple_window: upper bound on the amount of time a guest is allowed to execute
 114 *             in a PAUSE loop. Tests indicate that most spinlocks are held for
 115 *             less than 2^12 cycles
 116 * Time is measured based on a counter that runs at the same rate as the TSC,
 117 * refer SDM volume 3b section 21.6.13 & 22.1.3.
 118 */
 119#define KVM_VMX_DEFAULT_PLE_GAP    128
 120#define KVM_VMX_DEFAULT_PLE_WINDOW 4096
 121static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 122module_param(ple_gap, int, S_IRUGO);
 123
 124static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 125module_param(ple_window, int, S_IRUGO);
 126
 127#define NR_AUTOLOAD_MSRS 8
 128#define VMCS02_POOL_SIZE 1
 129
 130struct vmcs {
 131	u32 revision_id;
 132	u32 abort;
 133	char data[0];
 134};
 135
 136/*
 137 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
 138 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
 139 * loaded on this CPU (so we can clear them if the CPU goes down).
 140 */
 141struct loaded_vmcs {
 142	struct vmcs *vmcs;
 143	int cpu;
 144	int launched;
 145	struct list_head loaded_vmcss_on_cpu_link;
 146};
 147
 148struct shared_msr_entry {
 149	unsigned index;
 150	u64 data;
 151	u64 mask;
 152};
 153
 154/*
 155 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
 156 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
 157 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
 158 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
 159 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
 160 * More than one of these structures may exist, if L1 runs multiple L2 guests.
 161 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
 162 * underlying hardware which will be used to run L2.
 163 * This structure is packed to ensure that its layout is identical across
 164 * machines (necessary for live migration).
 165 * If there are changes in this struct, VMCS12_REVISION must be changed.
 166 */
 167typedef u64 natural_width;
 168struct __packed vmcs12 {
 169	/* According to the Intel spec, a VMCS region must start with the
 170	 * following two fields. Then follow implementation-specific data.
 171	 */
 172	u32 revision_id;
 173	u32 abort;
 174
 175	u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
 176	u32 padding[7]; /* room for future expansion */
 177
 178	u64 io_bitmap_a;
 179	u64 io_bitmap_b;
 180	u64 msr_bitmap;
 181	u64 vm_exit_msr_store_addr;
 182	u64 vm_exit_msr_load_addr;
 183	u64 vm_entry_msr_load_addr;
 184	u64 tsc_offset;
 185	u64 virtual_apic_page_addr;
 186	u64 apic_access_addr;
 187	u64 ept_pointer;
 188	u64 guest_physical_address;
 189	u64 vmcs_link_pointer;
 190	u64 guest_ia32_debugctl;
 191	u64 guest_ia32_pat;
 192	u64 guest_ia32_efer;
 193	u64 guest_ia32_perf_global_ctrl;
 194	u64 guest_pdptr0;
 195	u64 guest_pdptr1;
 196	u64 guest_pdptr2;
 197	u64 guest_pdptr3;
 198	u64 host_ia32_pat;
 199	u64 host_ia32_efer;
 200	u64 host_ia32_perf_global_ctrl;
 201	u64 padding64[8]; /* room for future expansion */
 202	/*
 203	 * To allow migration of L1 (complete with its L2 guests) between
 204	 * machines of different natural widths (32 or 64 bit), we cannot have
 205	 * unsigned long fields with no explict size. We use u64 (aliased
 206	 * natural_width) instead. Luckily, x86 is little-endian.
 207	 */
 208	natural_width cr0_guest_host_mask;
 209	natural_width cr4_guest_host_mask;
 210	natural_width cr0_read_shadow;
 211	natural_width cr4_read_shadow;
 212	natural_width cr3_target_value0;
 213	natural_width cr3_target_value1;
 214	natural_width cr3_target_value2;
 215	natural_width cr3_target_value3;
 216	natural_width exit_qualification;
 217	natural_width guest_linear_address;
 218	natural_width guest_cr0;
 219	natural_width guest_cr3;
 220	natural_width guest_cr4;
 221	natural_width guest_es_base;
 222	natural_width guest_cs_base;
 223	natural_width guest_ss_base;
 224	natural_width guest_ds_base;
 225	natural_width guest_fs_base;
 226	natural_width guest_gs_base;
 227	natural_width guest_ldtr_base;
 228	natural_width guest_tr_base;
 229	natural_width guest_gdtr_base;
 230	natural_width guest_idtr_base;
 231	natural_width guest_dr7;
 232	natural_width guest_rsp;
 233	natural_width guest_rip;
 234	natural_width guest_rflags;
 235	natural_width guest_pending_dbg_exceptions;
 236	natural_width guest_sysenter_esp;
 237	natural_width guest_sysenter_eip;
 238	natural_width host_cr0;
 239	natural_width host_cr3;
 240	natural_width host_cr4;
 241	natural_width host_fs_base;
 242	natural_width host_gs_base;
 243	natural_width host_tr_base;
 244	natural_width host_gdtr_base;
 245	natural_width host_idtr_base;
 246	natural_width host_ia32_sysenter_esp;
 247	natural_width host_ia32_sysenter_eip;
 248	natural_width host_rsp;
 249	natural_width host_rip;
 250	natural_width paddingl[8]; /* room for future expansion */
 251	u32 pin_based_vm_exec_control;
 252	u32 cpu_based_vm_exec_control;
 253	u32 exception_bitmap;
 254	u32 page_fault_error_code_mask;
 255	u32 page_fault_error_code_match;
 256	u32 cr3_target_count;
 257	u32 vm_exit_controls;
 258	u32 vm_exit_msr_store_count;
 259	u32 vm_exit_msr_load_count;
 260	u32 vm_entry_controls;
 261	u32 vm_entry_msr_load_count;
 262	u32 vm_entry_intr_info_field;
 263	u32 vm_entry_exception_error_code;
 264	u32 vm_entry_instruction_len;
 265	u32 tpr_threshold;
 266	u32 secondary_vm_exec_control;
 267	u32 vm_instruction_error;
 268	u32 vm_exit_reason;
 269	u32 vm_exit_intr_info;
 270	u32 vm_exit_intr_error_code;
 271	u32 idt_vectoring_info_field;
 272	u32 idt_vectoring_error_code;
 273	u32 vm_exit_instruction_len;
 274	u32 vmx_instruction_info;
 275	u32 guest_es_limit;
 276	u32 guest_cs_limit;
 277	u32 guest_ss_limit;
 278	u32 guest_ds_limit;
 279	u32 guest_fs_limit;
 280	u32 guest_gs_limit;
 281	u32 guest_ldtr_limit;
 282	u32 guest_tr_limit;
 283	u32 guest_gdtr_limit;
 284	u32 guest_idtr_limit;
 285	u32 guest_es_ar_bytes;
 286	u32 guest_cs_ar_bytes;
 287	u32 guest_ss_ar_bytes;
 288	u32 guest_ds_ar_bytes;
 289	u32 guest_fs_ar_bytes;
 290	u32 guest_gs_ar_bytes;
 291	u32 guest_ldtr_ar_bytes;
 292	u32 guest_tr_ar_bytes;
 293	u32 guest_interruptibility_info;
 294	u32 guest_activity_state;
 295	u32 guest_sysenter_cs;
 296	u32 host_ia32_sysenter_cs;
 297	u32 padding32[8]; /* room for future expansion */
 298	u16 virtual_processor_id;
 299	u16 guest_es_selector;
 300	u16 guest_cs_selector;
 301	u16 guest_ss_selector;
 302	u16 guest_ds_selector;
 303	u16 guest_fs_selector;
 304	u16 guest_gs_selector;
 305	u16 guest_ldtr_selector;
 306	u16 guest_tr_selector;
 307	u16 host_es_selector;
 308	u16 host_cs_selector;
 309	u16 host_ss_selector;
 310	u16 host_ds_selector;
 311	u16 host_fs_selector;
 312	u16 host_gs_selector;
 313	u16 host_tr_selector;
 314};
 315
 316/*
 317 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
 318 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
 319 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
 320 */
 321#define VMCS12_REVISION 0x11e57ed0
 322
 323/*
 324 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
 325 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
 326 * current implementation, 4K are reserved to avoid future complications.
 327 */
 328#define VMCS12_SIZE 0x1000
 329
 330/* Used to remember the last vmcs02 used for some recently used vmcs12s */
 331struct vmcs02_list {
 332	struct list_head list;
 333	gpa_t vmptr;
 334	struct loaded_vmcs vmcs02;
 335};
 336
 337/*
 338 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 339 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 340 */
 341struct nested_vmx {
 342	/* Has the level1 guest done vmxon? */
 343	bool vmxon;
 344
 345	/* The guest-physical address of the current VMCS L1 keeps for L2 */
 346	gpa_t current_vmptr;
 347	/* The host-usable pointer to the above */
 348	struct page *current_vmcs12_page;
 349	struct vmcs12 *current_vmcs12;
 350
 351	/* vmcs02_list cache of VMCSs recently used to run L2 guests */
 352	struct list_head vmcs02_pool;
 353	int vmcs02_num;
 354	u64 vmcs01_tsc_offset;
 355	/* L2 must run next, and mustn't decide to exit to L1. */
 356	bool nested_run_pending;
 357	/*
 358	 * Guest pages referred to in vmcs02 with host-physical pointers, so
 359	 * we must keep them pinned while L2 runs.
 360	 */
 361	struct page *apic_access_page;
 362};
 363
 364struct vcpu_vmx {
 365	struct kvm_vcpu       vcpu;
 366	unsigned long         host_rsp;
 367	u8                    fail;
 368	u8                    cpl;
 369	bool                  nmi_known_unmasked;
 370	u32                   exit_intr_info;
 371	u32                   idt_vectoring_info;
 372	ulong                 rflags;
 373	struct shared_msr_entry *guest_msrs;
 374	int                   nmsrs;
 375	int                   save_nmsrs;
 376#ifdef CONFIG_X86_64
 377	u64 		      msr_host_kernel_gs_base;
 378	u64 		      msr_guest_kernel_gs_base;
 379#endif
 380	/*
 381	 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 382	 * non-nested (L1) guest, it always points to vmcs01. For a nested
 383	 * guest (L2), it points to a different VMCS.
 384	 */
 385	struct loaded_vmcs    vmcs01;
 386	struct loaded_vmcs   *loaded_vmcs;
 387	bool                  __launched; /* temporary, used in vmx_vcpu_run */
 388	struct msr_autoload {
 389		unsigned nr;
 390		struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 391		struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 392	} msr_autoload;
 393	struct {
 394		int           loaded;
 395		u16           fs_sel, gs_sel, ldt_sel;
 396#ifdef CONFIG_X86_64
 397		u16           ds_sel, es_sel;
 398#endif
 399		int           gs_ldt_reload_needed;
 400		int           fs_reload_needed;
 401	} host_state;
 402	struct {
 403		int vm86_active;
 404		ulong save_rflags;
 405		struct kvm_save_segment {
 406			u16 selector;
 407			unsigned long base;
 408			u32 limit;
 409			u32 ar;
 410		} tr, es, ds, fs, gs;
 411	} rmode;
 412	struct {
 413		u32 bitmask; /* 4 bits per segment (1 bit per field) */
 414		struct kvm_save_segment seg[8];
 415	} segment_cache;
 416	int vpid;
 417	bool emulation_required;
 418
 419	/* Support for vnmi-less CPUs */
 420	int soft_vnmi_blocked;
 421	ktime_t entry_time;
 422	s64 vnmi_blocked_time;
 423	u32 exit_reason;
 424
 425	bool rdtscp_enabled;
 426
 427	/* Support for a guest hypervisor (nested VMX) */
 428	struct nested_vmx nested;
 429};
 430
 431enum segment_cache_field {
 432	SEG_FIELD_SEL = 0,
 433	SEG_FIELD_BASE = 1,
 434	SEG_FIELD_LIMIT = 2,
 435	SEG_FIELD_AR = 3,
 436
 437	SEG_FIELD_NR = 4
 438};
 439
 440static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 441{
 442	return container_of(vcpu, struct vcpu_vmx, vcpu);
 443}
 444
 445#define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
 446#define FIELD(number, name)	[number] = VMCS12_OFFSET(name)
 447#define FIELD64(number, name)	[number] = VMCS12_OFFSET(name), \
 448				[number##_HIGH] = VMCS12_OFFSET(name)+4
 449
 450static unsigned short vmcs_field_to_offset_table[] = {
 451	FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 452	FIELD(GUEST_ES_SELECTOR, guest_es_selector),
 453	FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
 454	FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
 455	FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
 456	FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
 457	FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
 458	FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
 459	FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
 460	FIELD(HOST_ES_SELECTOR, host_es_selector),
 461	FIELD(HOST_CS_SELECTOR, host_cs_selector),
 462	FIELD(HOST_SS_SELECTOR, host_ss_selector),
 463	FIELD(HOST_DS_SELECTOR, host_ds_selector),
 464	FIELD(HOST_FS_SELECTOR, host_fs_selector),
 465	FIELD(HOST_GS_SELECTOR, host_gs_selector),
 466	FIELD(HOST_TR_SELECTOR, host_tr_selector),
 467	FIELD64(IO_BITMAP_A, io_bitmap_a),
 468	FIELD64(IO_BITMAP_B, io_bitmap_b),
 469	FIELD64(MSR_BITMAP, msr_bitmap),
 470	FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
 471	FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
 472	FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
 473	FIELD64(TSC_OFFSET, tsc_offset),
 474	FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
 475	FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
 476	FIELD64(EPT_POINTER, ept_pointer),
 477	FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
 478	FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
 479	FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
 480	FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
 481	FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
 482	FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
 483	FIELD64(GUEST_PDPTR0, guest_pdptr0),
 484	FIELD64(GUEST_PDPTR1, guest_pdptr1),
 485	FIELD64(GUEST_PDPTR2, guest_pdptr2),
 486	FIELD64(GUEST_PDPTR3, guest_pdptr3),
 487	FIELD64(HOST_IA32_PAT, host_ia32_pat),
 488	FIELD64(HOST_IA32_EFER, host_ia32_efer),
 489	FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
 490	FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
 491	FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
 492	FIELD(EXCEPTION_BITMAP, exception_bitmap),
 493	FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
 494	FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
 495	FIELD(CR3_TARGET_COUNT, cr3_target_count),
 496	FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
 497	FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
 498	FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
 499	FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
 500	FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
 501	FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
 502	FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
 503	FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
 504	FIELD(TPR_THRESHOLD, tpr_threshold),
 505	FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
 506	FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
 507	FIELD(VM_EXIT_REASON, vm_exit_reason),
 508	FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
 509	FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
 510	FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
 511	FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
 512	FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
 513	FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
 514	FIELD(GUEST_ES_LIMIT, guest_es_limit),
 515	FIELD(GUEST_CS_LIMIT, guest_cs_limit),
 516	FIELD(GUEST_SS_LIMIT, guest_ss_limit),
 517	FIELD(GUEST_DS_LIMIT, guest_ds_limit),
 518	FIELD(GUEST_FS_LIMIT, guest_fs_limit),
 519	FIELD(GUEST_GS_LIMIT, guest_gs_limit),
 520	FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
 521	FIELD(GUEST_TR_LIMIT, guest_tr_limit),
 522	FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
 523	FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
 524	FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
 525	FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
 526	FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
 527	FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
 528	FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
 529	FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
 530	FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
 531	FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
 532	FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
 533	FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
 534	FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
 535	FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
 536	FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
 537	FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
 538	FIELD(CR0_READ_SHADOW, cr0_read_shadow),
 539	FIELD(CR4_READ_SHADOW, cr4_read_shadow),
 540	FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
 541	FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
 542	FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
 543	FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
 544	FIELD(EXIT_QUALIFICATION, exit_qualification),
 545	FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
 546	FIELD(GUEST_CR0, guest_cr0),
 547	FIELD(GUEST_CR3, guest_cr3),
 548	FIELD(GUEST_CR4, guest_cr4),
 549	FIELD(GUEST_ES_BASE, guest_es_base),
 550	FIELD(GUEST_CS_BASE, guest_cs_base),
 551	FIELD(GUEST_SS_BASE, guest_ss_base),
 552	FIELD(GUEST_DS_BASE, guest_ds_base),
 553	FIELD(GUEST_FS_BASE, guest_fs_base),
 554	FIELD(GUEST_GS_BASE, guest_gs_base),
 555	FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
 556	FIELD(GUEST_TR_BASE, guest_tr_base),
 557	FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
 558	FIELD(GUEST_IDTR_BASE, guest_idtr_base),
 559	FIELD(GUEST_DR7, guest_dr7),
 560	FIELD(GUEST_RSP, guest_rsp),
 561	FIELD(GUEST_RIP, guest_rip),
 562	FIELD(GUEST_RFLAGS, guest_rflags),
 563	FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
 564	FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
 565	FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
 566	FIELD(HOST_CR0, host_cr0),
 567	FIELD(HOST_CR3, host_cr3),
 568	FIELD(HOST_CR4, host_cr4),
 569	FIELD(HOST_FS_BASE, host_fs_base),
 570	FIELD(HOST_GS_BASE, host_gs_base),
 571	FIELD(HOST_TR_BASE, host_tr_base),
 572	FIELD(HOST_GDTR_BASE, host_gdtr_base),
 573	FIELD(HOST_IDTR_BASE, host_idtr_base),
 574	FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
 575	FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
 576	FIELD(HOST_RSP, host_rsp),
 577	FIELD(HOST_RIP, host_rip),
 578};
 579static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
 580
 581static inline short vmcs_field_to_offset(unsigned long field)
 582{
 583	if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
 584		return -1;
 585	return vmcs_field_to_offset_table[field];
 586}
 587
 588static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
 589{
 590	return to_vmx(vcpu)->nested.current_vmcs12;
 591}
 592
 593static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
 594{
 595	struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
 596	if (is_error_page(page)) {
 597		kvm_release_page_clean(page);
 598		return NULL;
 599	}
 600	return page;
 601}
 602
 603static void nested_release_page(struct page *page)
 604{
 605	kvm_release_page_dirty(page);
 606}
 607
 608static void nested_release_page_clean(struct page *page)
 609{
 610	kvm_release_page_clean(page);
 611}
 612
 613static u64 construct_eptp(unsigned long root_hpa);
 614static void kvm_cpu_vmxon(u64 addr);
 615static void kvm_cpu_vmxoff(void);
 616static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
 617static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 618static void vmx_set_segment(struct kvm_vcpu *vcpu,
 619			    struct kvm_segment *var, int seg);
 620static void vmx_get_segment(struct kvm_vcpu *vcpu,
 621			    struct kvm_segment *var, int seg);
 622
 623static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 624static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 625/*
 626 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
 627 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
 628 */
 629static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
 630static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 631
 632static unsigned long *vmx_io_bitmap_a;
 633static unsigned long *vmx_io_bitmap_b;
 634static unsigned long *vmx_msr_bitmap_legacy;
 635static unsigned long *vmx_msr_bitmap_longmode;
 636
 637static bool cpu_has_load_ia32_efer;
 638static bool cpu_has_load_perf_global_ctrl;
 639
 640static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 641static DEFINE_SPINLOCK(vmx_vpid_lock);
 642
 643static struct vmcs_config {
 644	int size;
 645	int order;
 646	u32 revision_id;
 647	u32 pin_based_exec_ctrl;
 648	u32 cpu_based_exec_ctrl;
 649	u32 cpu_based_2nd_exec_ctrl;
 650	u32 vmexit_ctrl;
 651	u32 vmentry_ctrl;
 652} vmcs_config;
 653
 654static struct vmx_capability {
 655	u32 ept;
 656	u32 vpid;
 657} vmx_capability;
 658
 659#define VMX_SEGMENT_FIELD(seg)					\
 660	[VCPU_SREG_##seg] = {                                   \
 661		.selector = GUEST_##seg##_SELECTOR,		\
 662		.base = GUEST_##seg##_BASE,		   	\
 663		.limit = GUEST_##seg##_LIMIT,		   	\
 664		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
 665	}
 666
 667static struct kvm_vmx_segment_field {
 668	unsigned selector;
 669	unsigned base;
 670	unsigned limit;
 671	unsigned ar_bytes;
 672} kvm_vmx_segment_fields[] = {
 673	VMX_SEGMENT_FIELD(CS),
 674	VMX_SEGMENT_FIELD(DS),
 675	VMX_SEGMENT_FIELD(ES),
 676	VMX_SEGMENT_FIELD(FS),
 677	VMX_SEGMENT_FIELD(GS),
 678	VMX_SEGMENT_FIELD(SS),
 679	VMX_SEGMENT_FIELD(TR),
 680	VMX_SEGMENT_FIELD(LDTR),
 681};
 682
 683static u64 host_efer;
 684
 685static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 686
 687/*
 688 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
 689 * away by decrementing the array size.
 690 */
 691static const u32 vmx_msr_index[] = {
 692#ifdef CONFIG_X86_64
 693	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 694#endif
 695	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
 696};
 697#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
 698
 699static inline bool is_page_fault(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 | PF_VECTOR | INTR_INFO_VALID_MASK);
 704}
 705
 706static inline bool is_no_device(u32 intr_info)
 707{
 708	return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 709			     INTR_INFO_VALID_MASK)) ==
 710		(INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
 711}
 712
 713static inline bool is_invalid_opcode(u32 intr_info)
 714{
 715	return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 716			     INTR_INFO_VALID_MASK)) ==
 717		(INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
 718}
 719
 720static inline bool is_external_interrupt(u32 intr_info)
 721{
 722	return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 723		== (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
 724}
 725
 726static inline bool is_machine_check(u32 intr_info)
 727{
 728	return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 729			     INTR_INFO_VALID_MASK)) ==
 730		(INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
 731}
 732
 733static inline bool cpu_has_vmx_msr_bitmap(void)
 734{
 735	return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
 736}
 737
 738static inline bool cpu_has_vmx_tpr_shadow(void)
 739{
 740	return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
 741}
 742
 743static inline bool vm_need_tpr_shadow(struct kvm *kvm)
 744{
 745	return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
 746}
 747
 748static inline bool cpu_has_secondary_exec_ctrls(void)
 749{
 750	return vmcs_config.cpu_based_exec_ctrl &
 751		CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
 752}
 753
 754static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
 755{
 756	return vmcs_config.cpu_based_2nd_exec_ctrl &
 757		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
 758}
 759
 760static inline bool cpu_has_vmx_flexpriority(void)
 761{
 762	return cpu_has_vmx_tpr_shadow() &&
 763		cpu_has_vmx_virtualize_apic_accesses();
 764}
 765
 766static inline bool cpu_has_vmx_ept_execute_only(void)
 767{
 768	return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
 769}
 770
 771static inline bool cpu_has_vmx_eptp_uncacheable(void)
 772{
 773	return vmx_capability.ept & VMX_EPTP_UC_BIT;
 774}
 775
 776static inline bool cpu_has_vmx_eptp_writeback(void)
 777{
 778	return vmx_capability.ept & VMX_EPTP_WB_BIT;
 779}
 780
 781static inline bool cpu_has_vmx_ept_2m_page(void)
 782{
 783	return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
 784}
 785
 786static inline bool cpu_has_vmx_ept_1g_page(void)
 787{
 788	return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
 789}
 790
 791static inline bool cpu_has_vmx_ept_4levels(void)
 792{
 793	return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
 794}
 795
 796static inline bool cpu_has_vmx_invept_individual_addr(void)
 797{
 798	return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
 799}
 800
 801static inline bool cpu_has_vmx_invept_context(void)
 802{
 803	return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
 804}
 805
 806static inline bool cpu_has_vmx_invept_global(void)
 807{
 808	return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
 809}
 810
 811static inline bool cpu_has_vmx_invvpid_single(void)
 812{
 813	return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
 814}
 815
 816static inline bool cpu_has_vmx_invvpid_global(void)
 817{
 818	return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
 819}
 820
 821static inline bool cpu_has_vmx_ept(void)
 822{
 823	return vmcs_config.cpu_based_2nd_exec_ctrl &
 824		SECONDARY_EXEC_ENABLE_EPT;
 825}
 826
 827static inline bool cpu_has_vmx_unrestricted_guest(void)
 828{
 829	return vmcs_config.cpu_based_2nd_exec_ctrl &
 830		SECONDARY_EXEC_UNRESTRICTED_GUEST;
 831}
 832
 833static inline bool cpu_has_vmx_ple(void)
 834{
 835	return vmcs_config.cpu_based_2nd_exec_ctrl &
 836		SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 837}
 838
 839static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
 840{
 841	return flexpriority_enabled && irqchip_in_kernel(kvm);
 842}
 843
 844static inline bool cpu_has_vmx_vpid(void)
 845{
 846	return vmcs_config.cpu_based_2nd_exec_ctrl &
 847		SECONDARY_EXEC_ENABLE_VPID;
 848}
 849
 850static inline bool cpu_has_vmx_rdtscp(void)
 851{
 852	return vmcs_config.cpu_based_2nd_exec_ctrl &
 853		SECONDARY_EXEC_RDTSCP;
 854}
 855
 856static inline bool cpu_has_virtual_nmis(void)
 857{
 858	return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
 859}
 860
 861static inline bool cpu_has_vmx_wbinvd_exit(void)
 862{
 863	return vmcs_config.cpu_based_2nd_exec_ctrl &
 864		SECONDARY_EXEC_WBINVD_EXITING;
 865}
 866
 867static inline bool report_flexpriority(void)
 868{
 869	return flexpriority_enabled;
 870}
 871
 872static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
 873{
 874	return vmcs12->cpu_based_vm_exec_control & bit;
 875}
 876
 877static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
 878{
 879	return (vmcs12->cpu_based_vm_exec_control &
 880			CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
 881		(vmcs12->secondary_vm_exec_control & bit);
 882}
 883
 884static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
 885	struct kvm_vcpu *vcpu)
 886{
 887	return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
 888}
 889
 890static inline bool is_exception(u32 intr_info)
 891{
 892	return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
 893		== (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
 894}
 895
 896static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
 897static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
 898			struct vmcs12 *vmcs12,
 899			u32 reason, unsigned long qualification);
 900
 901static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
 902{
 903	int i;
 904
 905	for (i = 0; i < vmx->nmsrs; ++i)
 906		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
 907			return i;
 908	return -1;
 909}
 910
 911static inline void __invvpid(int ext, u16 vpid, gva_t gva)
 912{
 913    struct {
 914	u64 vpid : 16;
 915	u64 rsvd : 48;
 916	u64 gva;
 917    } operand = { vpid, 0, gva };
 918
 919    asm volatile (__ex(ASM_VMX_INVVPID)
 920		  /* CF==1 or ZF==1 --> rc = -1 */
 921		  "; ja 1f ; ud2 ; 1:"
 922		  : : "a"(&operand), "c"(ext) : "cc", "memory");
 923}
 924
 925static inline void __invept(int ext, u64 eptp, gpa_t gpa)
 926{
 927	struct {
 928		u64 eptp, gpa;
 929	} operand = {eptp, gpa};
 930
 931	asm volatile (__ex(ASM_VMX_INVEPT)
 932			/* CF==1 or ZF==1 --> rc = -1 */
 933			"; ja 1f ; ud2 ; 1:\n"
 934			: : "a" (&operand), "c" (ext) : "cc", "memory");
 935}
 936
 937static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
 938{
 939	int i;
 940
 941	i = __find_msr_index(vmx, msr);
 942	if (i >= 0)
 943		return &vmx->guest_msrs[i];
 944	return NULL;
 945}
 946
 947static void vmcs_clear(struct vmcs *vmcs)
 948{
 949	u64 phys_addr = __pa(vmcs);
 950	u8 error;
 951
 952	asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
 953		      : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
 954		      : "cc", "memory");
 955	if (error)
 956		printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
 957		       vmcs, phys_addr);
 958}
 959
 960static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
 961{
 962	vmcs_clear(loaded_vmcs->vmcs);
 963	loaded_vmcs->cpu = -1;
 964	loaded_vmcs->launched = 0;
 965}
 966
 967static void vmcs_load(struct vmcs *vmcs)
 968{
 969	u64 phys_addr = __pa(vmcs);
 970	u8 error;
 971
 972	asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
 973			: "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
 974			: "cc", "memory");
 975	if (error)
 976		printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
 977		       vmcs, phys_addr);
 978}
 979
 980static void __loaded_vmcs_clear(void *arg)
 981{
 982	struct loaded_vmcs *loaded_vmcs = arg;
 983	int cpu = raw_smp_processor_id();
 984
 985	if (loaded_vmcs->cpu != cpu)
 986		return; /* vcpu migration can race with cpu offline */
 987	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
 988		per_cpu(current_vmcs, cpu) = NULL;
 989	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
 990	loaded_vmcs_init(loaded_vmcs);
 991}
 992
 993static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
 994{
 995	if (loaded_vmcs->cpu != -1)
 996		smp_call_function_single(
 997			loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
 998}
 999
1000static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1001{
1002	if (vmx->vpid == 0)
1003		return;
1004
1005	if (cpu_has_vmx_invvpid_single())
1006		__invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1007}
1008
1009static inline void vpid_sync_vcpu_global(void)
1010{
1011	if (cpu_has_vmx_invvpid_global())
1012		__invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1013}
1014
1015static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1016{
1017	if (cpu_has_vmx_invvpid_single())
1018		vpid_sync_vcpu_single(vmx);
1019	else
1020		vpid_sync_vcpu_global();
1021}
1022
1023static inline void ept_sync_global(void)
1024{
1025	if (cpu_has_vmx_invept_global())
1026		__invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1027}
1028
1029static inline void ept_sync_context(u64 eptp)
1030{
1031	if (enable_ept) {
1032		if (cpu_has_vmx_invept_context())
1033			__invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1034		else
1035			ept_sync_global();
1036	}
1037}
1038
1039static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
1040{
1041	if (enable_ept) {
1042		if (cpu_has_vmx_invept_individual_addr())
1043			__invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
1044					eptp, gpa);
1045		else
1046			ept_sync_context(eptp);
1047	}
1048}
1049
1050static __always_inline unsigned long vmcs_readl(unsigned long field)
1051{
1052	unsigned long value;
1053
1054	asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1055		      : "=a"(value) : "d"(field) : "cc");
1056	return value;
1057}
1058
1059static __always_inline u16 vmcs_read16(unsigned long field)
1060{
1061	return vmcs_readl(field);
1062}
1063
1064static __always_inline u32 vmcs_read32(unsigned long field)
1065{
1066	return vmcs_readl(field);
1067}
1068
1069static __always_inline u64 vmcs_read64(unsigned long field)
1070{
1071#ifdef CONFIG_X86_64
1072	return vmcs_readl(field);
1073#else
1074	return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1075#endif
1076}
1077
1078static noinline void vmwrite_error(unsigned long field, unsigned long value)
1079{
1080	printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1081	       field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1082	dump_stack();
1083}
1084
1085static void vmcs_writel(unsigned long field, unsigned long value)
1086{
1087	u8 error;
1088
1089	asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1090		       : "=q"(error) : "a"(value), "d"(field) : "cc");
1091	if (unlikely(error))
1092		vmwrite_error(field, value);
1093}
1094
1095static void vmcs_write16(unsigned long field, u16 value)
1096{
1097	vmcs_writel(field, value);
1098}
1099
1100static void vmcs_write32(unsigned long field, u32 value)
1101{
1102	vmcs_writel(field, value);
1103}
1104
1105static void vmcs_write64(unsigned long field, u64 value)
1106{
1107	vmcs_writel(field, value);
1108#ifndef CONFIG_X86_64
1109	asm volatile ("");
1110	vmcs_writel(field+1, value >> 32);
1111#endif
1112}
1113
1114static void vmcs_clear_bits(unsigned long field, u32 mask)
1115{
1116	vmcs_writel(field, vmcs_readl(field) & ~mask);
1117}
1118
1119static void vmcs_set_bits(unsigned long field, u32 mask)
1120{
1121	vmcs_writel(field, vmcs_readl(field) | mask);
1122}
1123
1124static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1125{
1126	vmx->segment_cache.bitmask = 0;
1127}
1128
1129static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1130				       unsigned field)
1131{
1132	bool ret;
1133	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1134
1135	if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1136		vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1137		vmx->segment_cache.bitmask = 0;
1138	}
1139	ret = vmx->segment_cache.bitmask & mask;
1140	vmx->segment_cache.bitmask |= mask;
1141	return ret;
1142}
1143
1144static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1145{
1146	u16 *p = &vmx->segment_cache.seg[seg].selector;
1147
1148	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1149		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1150	return *p;
1151}
1152
1153static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1154{
1155	ulong *p = &vmx->segment_cache.seg[seg].base;
1156
1157	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1158		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1159	return *p;
1160}
1161
1162static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1163{
1164	u32 *p = &vmx->segment_cache.seg[seg].limit;
1165
1166	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1167		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1168	return *p;
1169}
1170
1171static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1172{
1173	u32 *p = &vmx->segment_cache.seg[seg].ar;
1174
1175	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1176		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1177	return *p;
1178}
1179
1180static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1181{
1182	u32 eb;
1183
1184	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1185	     (1u << NM_VECTOR) | (1u << DB_VECTOR);
1186	if ((vcpu->guest_debug &
1187	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1188	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1189		eb |= 1u << BP_VECTOR;
1190	if (to_vmx(vcpu)->rmode.vm86_active)
1191		eb = ~0;
1192	if (enable_ept)
1193		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1194	if (vcpu->fpu_active)
1195		eb &= ~(1u << NM_VECTOR);
1196
1197	/* When we are running a nested L2 guest and L1 specified for it a
1198	 * certain exception bitmap, we must trap the same exceptions and pass
1199	 * them to L1. When running L2, we will only handle the exceptions
1200	 * specified above if L1 did not want them.
1201	 */
1202	if (is_guest_mode(vcpu))
1203		eb |= get_vmcs12(vcpu)->exception_bitmap;
1204
1205	vmcs_write32(EXCEPTION_BITMAP, eb);
1206}
1207
1208static void clear_atomic_switch_msr_special(unsigned long entry,
1209		unsigned long exit)
1210{
1211	vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1212	vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1213}
1214
1215static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1216{
1217	unsigned i;
1218	struct msr_autoload *m = &vmx->msr_autoload;
1219
1220	switch (msr) {
1221	case MSR_EFER:
1222		if (cpu_has_load_ia32_efer) {
1223			clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1224					VM_EXIT_LOAD_IA32_EFER);
1225			return;
1226		}
1227		break;
1228	case MSR_CORE_PERF_GLOBAL_CTRL:
1229		if (cpu_has_load_perf_global_ctrl) {
1230			clear_atomic_switch_msr_special(
1231					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1232					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1233			return;
1234		}
1235		break;
1236	}
1237
1238	for (i = 0; i < m->nr; ++i)
1239		if (m->guest[i].index == msr)
1240			break;
1241
1242	if (i == m->nr)
1243		return;
1244	--m->nr;
1245	m->guest[i] = m->guest[m->nr];
1246	m->host[i] = m->host[m->nr];
1247	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1248	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1249}
1250
1251static void add_atomic_switch_msr_special(unsigned long entry,
1252		unsigned long exit, unsigned long guest_val_vmcs,
1253		unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1254{
1255	vmcs_write64(guest_val_vmcs, guest_val);
1256	vmcs_write64(host_val_vmcs, host_val);
1257	vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1258	vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1259}
1260
1261static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1262				  u64 guest_val, u64 host_val)
1263{
1264	unsigned i;
1265	struct msr_autoload *m = &vmx->msr_autoload;
1266
1267	switch (msr) {
1268	case MSR_EFER:
1269		if (cpu_has_load_ia32_efer) {
1270			add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1271					VM_EXIT_LOAD_IA32_EFER,
1272					GUEST_IA32_EFER,
1273					HOST_IA32_EFER,
1274					guest_val, host_val);
1275			return;
1276		}
1277		break;
1278	case MSR_CORE_PERF_GLOBAL_CTRL:
1279		if (cpu_has_load_perf_global_ctrl) {
1280			add_atomic_switch_msr_special(
1281					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1282					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1283					GUEST_IA32_PERF_GLOBAL_CTRL,
1284					HOST_IA32_PERF_GLOBAL_CTRL,
1285					guest_val, host_val);
1286			return;
1287		}
1288		break;
1289	}
1290
1291	for (i = 0; i < m->nr; ++i)
1292		if (m->guest[i].index == msr)
1293			break;
1294
1295	if (i == NR_AUTOLOAD_MSRS) {
1296		printk_once(KERN_WARNING"Not enough mst switch entries. "
1297				"Can't add msr %x\n", msr);
1298		return;
1299	} else if (i == m->nr) {
1300		++m->nr;
1301		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1302		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1303	}
1304
1305	m->guest[i].index = msr;
1306	m->guest[i].value = guest_val;
1307	m->host[i].index = msr;
1308	m->host[i].value = host_val;
1309}
1310
1311static void reload_tss(void)
1312{
1313	/*
1314	 * VT restores TR but not its size.  Useless.
1315	 */
1316	struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1317	struct desc_struct *descs;
1318
1319	descs = (void *)gdt->address;
1320	descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1321	load_TR_desc();
1322}
1323
1324static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1325{
1326	u64 guest_efer;
1327	u64 ignore_bits;
1328
1329	guest_efer = vmx->vcpu.arch.efer;
1330
1331	/*
1332	 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
1333	 * outside long mode
1334	 */
1335	ignore_bits = EFER_NX | EFER_SCE;
1336#ifdef CONFIG_X86_64
1337	ignore_bits |= EFER_LMA | EFER_LME;
1338	/* SCE is meaningful only in long mode on Intel */
1339	if (guest_efer & EFER_LMA)
1340		ignore_bits &= ~(u64)EFER_SCE;
1341#endif
1342	guest_efer &= ~ignore_bits;
1343	guest_efer |= host_efer & ignore_bits;
1344	vmx->guest_msrs[efer_offset].data = guest_efer;
1345	vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1346
1347	clear_atomic_switch_msr(vmx, MSR_EFER);
1348	/* On ept, can't emulate nx, and must switch nx atomically */
1349	if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1350		guest_efer = vmx->vcpu.arch.efer;
1351		if (!(guest_efer & EFER_LMA))
1352			guest_efer &= ~EFER_LME;
1353		add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1354		return false;
1355	}
1356
1357	return true;
1358}
1359
1360static unsigned long segment_base(u16 selector)
1361{
1362	struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1363	struct desc_struct *d;
1364	unsigned long table_base;
1365	unsigned long v;
1366
1367	if (!(selector & ~3))
1368		return 0;
1369
1370	table_base = gdt->address;
1371
1372	if (selector & 4) {           /* from ldt */
1373		u16 ldt_selector = kvm_read_ldt();
1374
1375		if (!(ldt_selector & ~3))
1376			return 0;
1377
1378		table_base = segment_base(ldt_selector);
1379	}
1380	d = (struct desc_struct *)(table_base + (selector & ~7));
1381	v = get_desc_base(d);
1382#ifdef CONFIG_X86_64
1383       if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1384               v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1385#endif
1386	return v;
1387}
1388
1389static inline unsigned long kvm_read_tr_base(void)
1390{
1391	u16 tr;
1392	asm("str %0" : "=g"(tr));
1393	return segment_base(tr);
1394}
1395
1396static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1397{
1398	struct vcpu_vmx *vmx = to_vmx(vcpu);
1399	int i;
1400
1401	if (vmx->host_state.loaded)
1402		return;
1403
1404	vmx->host_state.loaded = 1;
1405	/*
1406	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1407	 * allow segment selectors with cpl > 0 or ti == 1.
1408	 */
1409	vmx->host_state.ldt_sel = kvm_read_ldt();
1410	vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1411	savesegment(fs, vmx->host_state.fs_sel);
1412	if (!(vmx->host_state.fs_sel & 7)) {
1413		vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1414		vmx->host_state.fs_reload_needed = 0;
1415	} else {
1416		vmcs_write16(HOST_FS_SELECTOR, 0);
1417		vmx->host_state.fs_reload_needed = 1;
1418	}
1419	savesegment(gs, vmx->host_state.gs_sel);
1420	if (!(vmx->host_state.gs_sel & 7))
1421		vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1422	else {
1423		vmcs_write16(HOST_GS_SELECTOR, 0);
1424		vmx->host_state.gs_ldt_reload_needed = 1;
1425	}
1426
1427#ifdef CONFIG_X86_64
1428	savesegment(ds, vmx->host_state.ds_sel);
1429	savesegment(es, vmx->host_state.es_sel);
1430#endif
1431
1432#ifdef CONFIG_X86_64
1433	vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1434	vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1435#else
1436	vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1437	vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1438#endif
1439
1440#ifdef CONFIG_X86_64
1441	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1442	if (is_long_mode(&vmx->vcpu))
1443		wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1444#endif
1445	for (i = 0; i < vmx->save_nmsrs; ++i)
1446		kvm_set_shared_msr(vmx->guest_msrs[i].index,
1447				   vmx->guest_msrs[i].data,
1448				   vmx->guest_msrs[i].mask);
1449}
1450
1451static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1452{
1453	if (!vmx->host_state.loaded)
1454		return;
1455
1456	++vmx->vcpu.stat.host_state_reload;
1457	vmx->host_state.loaded = 0;
1458#ifdef CONFIG_X86_64
1459	if (is_long_mode(&vmx->vcpu))
1460		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1461#endif
1462	if (vmx->host_state.gs_ldt_reload_needed) {
1463		kvm_load_ldt(vmx->host_state.ldt_sel);
1464#ifdef CONFIG_X86_64
1465		load_gs_index(vmx->host_state.gs_sel);
1466#else
1467		loadsegment(gs, vmx->host_state.gs_sel);
1468#endif
1469	}
1470	if (vmx->host_state.fs_reload_needed)
1471		loadsegment(fs, vmx->host_state.fs_sel);
1472#ifdef CONFIG_X86_64
1473	if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1474		loadsegment(ds, vmx->host_state.ds_sel);
1475		loadsegment(es, vmx->host_state.es_sel);
1476	}
1477#endif
1478	reload_tss();
1479#ifdef CONFIG_X86_64
1480	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1481#endif
1482	if (user_has_fpu())
1483		clts();
1484	load_gdt(&__get_cpu_var(host_gdt));
1485}
1486
1487static void vmx_load_host_state(struct vcpu_vmx *vmx)
1488{
1489	preempt_disable();
1490	__vmx_load_host_state(vmx);
1491	preempt_enable();
1492}
1493
1494/*
1495 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1496 * vcpu mutex is already taken.
1497 */
1498static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1499{
1500	struct vcpu_vmx *vmx = to_vmx(vcpu);
1501	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1502
1503	if (!vmm_exclusive)
1504		kvm_cpu_vmxon(phys_addr);
1505	else if (vmx->loaded_vmcs->cpu != cpu)
1506		loaded_vmcs_clear(vmx->loaded_vmcs);
1507
1508	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1509		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1510		vmcs_load(vmx->loaded_vmcs->vmcs);
1511	}
1512
1513	if (vmx->loaded_vmcs->cpu != cpu) {
1514		struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1515		unsigned long sysenter_esp;
1516
1517		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1518		local_irq_disable();
1519		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1520			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1521		local_irq_enable();
1522
1523		/*
1524		 * Linux uses per-cpu TSS and GDT, so set these when switching
1525		 * processors.
1526		 */
1527		vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1528		vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
1529
1530		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1531		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1532		vmx->loaded_vmcs->cpu = cpu;
1533	}
1534}
1535
1536static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1537{
1538	__vmx_load_host_state(to_vmx(vcpu));
1539	if (!vmm_exclusive) {
1540		__loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1541		vcpu->cpu = -1;
1542		kvm_cpu_vmxoff();
1543	}
1544}
1545
1546static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1547{
1548	ulong cr0;
1549
1550	if (vcpu->fpu_active)
1551		return;
1552	vcpu->fpu_active = 1;
1553	cr0 = vmcs_readl(GUEST_CR0);
1554	cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1555	cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1556	vmcs_writel(GUEST_CR0, cr0);
1557	update_exception_bitmap(vcpu);
1558	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1559	if (is_guest_mode(vcpu))
1560		vcpu->arch.cr0_guest_owned_bits &=
1561			~get_vmcs12(vcpu)->cr0_guest_host_mask;
1562	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1563}
1564
1565static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1566
1567/*
1568 * Return the cr0 value that a nested guest would read. This is a combination
1569 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1570 * its hypervisor (cr0_read_shadow).
1571 */
1572static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1573{
1574	return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1575		(fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1576}
1577static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1578{
1579	return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1580		(fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1581}
1582
1583static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1584{
1585	/* Note that there is no vcpu->fpu_active = 0 here. The caller must
1586	 * set this *before* calling this function.
1587	 */
1588	vmx_decache_cr0_guest_bits(vcpu);
1589	vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1590	update_exception_bitmap(vcpu);
1591	vcpu->arch.cr0_guest_owned_bits = 0;
1592	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1593	if (is_guest_mode(vcpu)) {
1594		/*
1595		 * L1's specified read shadow might not contain the TS bit,
1596		 * so now that we turned on shadowing of this bit, we need to
1597		 * set this bit of the shadow. Like in nested_vmx_run we need
1598		 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1599		 * up-to-date here because we just decached cr0.TS (and we'll
1600		 * only update vmcs12->guest_cr0 on nested exit).
1601		 */
1602		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1603		vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1604			(vcpu->arch.cr0 & X86_CR0_TS);
1605		vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1606	} else
1607		vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1608}
1609
1610static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1611{
1612	unsigned long rflags, save_rflags;
1613
1614	if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1615		__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1616		rflags = vmcs_readl(GUEST_RFLAGS);
1617		if (to_vmx(vcpu)->rmode.vm86_active) {
1618			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1619			save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1620			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1621		}
1622		to_vmx(vcpu)->rflags = rflags;
1623	}
1624	return to_vmx(vcpu)->rflags;
1625}
1626
1627static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1628{
1629	__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1630	__clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1631	to_vmx(vcpu)->rflags = rflags;
1632	if (to_vmx(vcpu)->rmode.vm86_active) {
1633		to_vmx(vcpu)->rmode.save_rflags = rflags;
1634		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1635	}
1636	vmcs_writel(GUEST_RFLAGS, rflags);
1637}
1638
1639static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1640{
1641	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1642	int ret = 0;
1643
1644	if (interruptibility & GUEST_INTR_STATE_STI)
1645		ret |= KVM_X86_SHADOW_INT_STI;
1646	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1647		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1648
1649	return ret & mask;
1650}
1651
1652static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1653{
1654	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1655	u32 interruptibility = interruptibility_old;
1656
1657	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1658
1659	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1660		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1661	else if (mask & KVM_X86_SHADOW_INT_STI)
1662		interruptibility |= GUEST_INTR_STATE_STI;
1663
1664	if ((interruptibility != interruptibility_old))
1665		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1666}
1667
1668static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1669{
1670	unsigned long rip;
1671
1672	rip = kvm_rip_read(vcpu);
1673	rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1674	kvm_rip_write(vcpu, rip);
1675
1676	/* skipping an emulated instruction also counts */
1677	vmx_set_interrupt_shadow(vcpu, 0);
1678}
1679
1680/*
1681 * KVM wants to inject page-faults which it got to the guest. This function
1682 * checks whether in a nested guest, we need to inject them to L1 or L2.
1683 * This function assumes it is called with the exit reason in vmcs02 being
1684 * a #PF exception (this is the only case in which KVM injects a #PF when L2
1685 * is running).
1686 */
1687static int nested_pf_handled(struct kvm_vcpu *vcpu)
1688{
1689	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1690
1691	/* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1692	if (!(vmcs12->exception_bitmap & (1u << PF_VECTOR)))
1693		return 0;
1694
1695	nested_vmx_vmexit(vcpu);
1696	return 1;
1697}
1698
1699static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1700				bool has_error_code, u32 error_code,
1701				bool reinject)
1702{
1703	struct vcpu_vmx *vmx = to_vmx(vcpu);
1704	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1705
1706	if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1707		nested_pf_handled(vcpu))
1708		return;
1709
1710	if (has_error_code) {
1711		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1712		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1713	}
1714
1715	if (vmx->rmode.vm86_active) {
1716		int inc_eip = 0;
1717		if (kvm_exception_is_soft(nr))
1718			inc_eip = vcpu->arch.event_exit_inst_len;
1719		if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1720			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1721		return;
1722	}
1723
1724	if (kvm_exception_is_soft(nr)) {
1725		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1726			     vmx->vcpu.arch.event_exit_inst_len);
1727		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1728	} else
1729		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1730
1731	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1732}
1733
1734static bool vmx_rdtscp_supported(void)
1735{
1736	return cpu_has_vmx_rdtscp();
1737}
1738
1739/*
1740 * Swap MSR entry in host/guest MSR entry array.
1741 */
1742static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1743{
1744	struct shared_msr_entry tmp;
1745
1746	tmp = vmx->guest_msrs[to];
1747	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1748	vmx->guest_msrs[from] = tmp;
1749}
1750
1751/*
1752 * Set up the vmcs to automatically save and restore system
1753 * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1754 * mode, as fiddling with msrs is very expensive.
1755 */
1756static void setup_msrs(struct vcpu_vmx *vmx)
1757{
1758	int save_nmsrs, index;
1759	unsigned long *msr_bitmap;
1760
1761	save_nmsrs = 0;
1762#ifdef CONFIG_X86_64
1763	if (is_long_mode(&vmx->vcpu)) {
1764		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1765		if (index >= 0)
1766			move_msr_up(vmx, index, save_nmsrs++);
1767		index = __find_msr_index(vmx, MSR_LSTAR);
1768		if (index >= 0)
1769			move_msr_up(vmx, index, save_nmsrs++);
1770		index = __find_msr_index(vmx, MSR_CSTAR);
1771		if (index >= 0)
1772			move_msr_up(vmx, index, save_nmsrs++);
1773		index = __find_msr_index(vmx, MSR_TSC_AUX);
1774		if (index >= 0 && vmx->rdtscp_enabled)
1775			move_msr_up(vmx, index, save_nmsrs++);
1776		/*
1777		 * MSR_STAR is only needed on long mode guests, and only
1778		 * if efer.sce is enabled.
1779		 */
1780		index = __find_msr_index(vmx, MSR_STAR);
1781		if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1782			move_msr_up(vmx, index, save_nmsrs++);
1783	}
1784#endif
1785	index = __find_msr_index(vmx, MSR_EFER);
1786	if (index >= 0 && update_transition_efer(vmx, index))
1787		move_msr_up(vmx, index, save_nmsrs++);
1788
1789	vmx->save_nmsrs = save_nmsrs;
1790
1791	if (cpu_has_vmx_msr_bitmap()) {
1792		if (is_long_mode(&vmx->vcpu))
1793			msr_bitmap = vmx_msr_bitmap_longmode;
1794		else
1795			msr_bitmap = vmx_msr_bitmap_legacy;
1796
1797		vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1798	}
1799}
1800
1801/*
1802 * reads and returns guest's timestamp counter "register"
1803 * guest_tsc = host_tsc + tsc_offset    -- 21.3
1804 */
1805static u64 guest_read_tsc(void)
1806{
1807	u64 host_tsc, tsc_offset;
1808
1809	rdtscll(host_tsc);
1810	tsc_offset = vmcs_read64(TSC_OFFSET);
1811	return host_tsc + tsc_offset;
1812}
1813
1814/*
1815 * Like guest_read_tsc, but always returns L1's notion of the timestamp
1816 * counter, even if a nested guest (L2) is currently running.
1817 */
1818u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu)
1819{
1820	u64 host_tsc, tsc_offset;
1821
1822	rdtscll(host_tsc);
1823	tsc_offset = is_guest_mode(vcpu) ?
1824		to_vmx(vcpu)->nested.vmcs01_tsc_offset :
1825		vmcs_read64(TSC_OFFSET);
1826	return host_tsc + tsc_offset;
1827}
1828
1829/*
1830 * Engage any workarounds for mis-matched TSC rates.  Currently limited to
1831 * software catchup for faster rates on slower CPUs.
1832 */
1833static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1834{
1835	if (!scale)
1836		return;
1837
1838	if (user_tsc_khz > tsc_khz) {
1839		vcpu->arch.tsc_catchup = 1;
1840		vcpu->arch.tsc_always_catchup = 1;
1841	} else
1842		WARN(1, "user requested TSC rate below hardware speed\n");
1843}
1844
1845/*
1846 * writes 'offset' into guest's timestamp counter offset register
1847 */
1848static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1849{
1850	if (is_guest_mode(vcpu)) {
1851		/*
1852		 * We're here if L1 chose not to trap WRMSR to TSC. According
1853		 * to the spec, this should set L1's TSC; The offset that L1
1854		 * set for L2 remains unchanged, and still needs to be added
1855		 * to the newly set TSC to get L2's TSC.
1856		 */
1857		struct vmcs12 *vmcs12;
1858		to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
1859		/* recalculate vmcs02.TSC_OFFSET: */
1860		vmcs12 = get_vmcs12(vcpu);
1861		vmcs_write64(TSC_OFFSET, offset +
1862			(nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
1863			 vmcs12->tsc_offset : 0));
1864	} else {
1865		vmcs_write64(TSC_OFFSET, offset);
1866	}
1867}
1868
1869static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1870{
1871	u64 offset = vmcs_read64(TSC_OFFSET);
1872	vmcs_write64(TSC_OFFSET, offset + adjustment);
1873	if (is_guest_mode(vcpu)) {
1874		/* Even when running L2, the adjustment needs to apply to L1 */
1875		to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
1876	}
1877}
1878
1879static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1880{
1881	return target_tsc - native_read_tsc();
1882}
1883
1884static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1885{
1886	struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1887	return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1888}
1889
1890/*
1891 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1892 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1893 * all guests if the "nested" module option is off, and can also be disabled
1894 * for a single guest by disabling its VMX cpuid bit.
1895 */
1896static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1897{
1898	return nested && guest_cpuid_has_vmx(vcpu);
1899}
1900
1901/*
1902 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1903 * returned for the various VMX controls MSRs when nested VMX is enabled.
1904 * The same values should also be used to verify that vmcs12 control fields are
1905 * valid during nested entry from L1 to L2.
1906 * Each of these control msrs has a low and high 32-bit half: A low bit is on
1907 * if the corresponding bit in the (32-bit) control field *must* be on, and a
1908 * bit in the high half is on if the corresponding bit in the control field
1909 * may be on. See also vmx_control_verify().
1910 * TODO: allow these variables to be modified (downgraded) by module options
1911 * or other means.
1912 */
1913static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1914static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1915static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1916static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1917static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1918static __init void nested_vmx_setup_ctls_msrs(void)
1919{
1920	/*
1921	 * Note that as a general rule, the high half of the MSRs (bits in
1922	 * the control fields which may be 1) should be initialized by the
1923	 * intersection of the underlying hardware's MSR (i.e., features which
1924	 * can be supported) and the list of features we want to expose -
1925	 * because they are known to be properly supported in our code.
1926	 * Also, usually, the low half of the MSRs (bits which must be 1) can
1927	 * be set to 0, meaning that L1 may turn off any of these bits. The
1928	 * reason is that if one of these bits is necessary, it will appear
1929	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1930	 * fields of vmcs01 and vmcs02, will turn these bits off - and
1931	 * nested_vmx_exit_handled() will not pass related exits to L1.
1932	 * These rules have exceptions below.
1933	 */
1934
1935	/* pin-based controls */
1936	/*
1937	 * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1938	 * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1939	 */
1940	nested_vmx_pinbased_ctls_low = 0x16 ;
1941	nested_vmx_pinbased_ctls_high = 0x16 |
1942		PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1943		PIN_BASED_VIRTUAL_NMIS;
1944
1945	/* exit controls */
1946	nested_vmx_exit_ctls_low = 0;
1947	/* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
1948#ifdef CONFIG_X86_64
1949	nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1950#else
1951	nested_vmx_exit_ctls_high = 0;
1952#endif
1953
1954	/* entry controls */
1955	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1956		nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1957	nested_vmx_entry_ctls_low = 0;
1958	nested_vmx_entry_ctls_high &=
1959		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1960
1961	/* cpu-based controls */
1962	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1963		nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1964	nested_vmx_procbased_ctls_low = 0;
1965	nested_vmx_procbased_ctls_high &=
1966		CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1967		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1968		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1969		CPU_BASED_CR3_STORE_EXITING |
1970#ifdef CONFIG_X86_64
1971		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1972#endif
1973		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1974		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1975		CPU_BASED_RDPMC_EXITING |
1976		CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1977	/*
1978	 * We can allow some features even when not supported by the
1979	 * hardware. For example, L1 can specify an MSR bitmap - and we
1980	 * can use it to avoid exits to L1 - even when L0 runs L2
1981	 * without MSR bitmaps.
1982	 */
1983	nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
1984
1985	/* secondary cpu-based controls */
1986	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
1987		nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
1988	nested_vmx_secondary_ctls_low = 0;
1989	nested_vmx_secondary_ctls_high &=
1990		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1991}
1992
1993static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
1994{
1995	/*
1996	 * Bits 0 in high must be 0, and bits 1 in low must be 1.
1997	 */
1998	return ((control & high) | low) == control;
1999}
2000
2001static inline u64 vmx_control_msr(u32 low, u32 high)
2002{
2003	return low | ((u64)high << 32);
2004}
2005
2006/*
2007 * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
2008 * also let it use VMX-specific MSRs.
2009 * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
2010 * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
2011 * like all other MSRs).
2012 */
2013static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2014{
2015	if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2016		     msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2017		/*
2018		 * According to the spec, processors which do not support VMX
2019		 * should throw a #GP(0) when VMX capability MSRs are read.
2020		 */
2021		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2022		return 1;
2023	}
2024
2025	switch (msr_index) {
2026	case MSR_IA32_FEATURE_CONTROL:
2027		*pdata = 0;
2028		break;
2029	case MSR_IA32_VMX_BASIC:
2030		/*
2031		 * This MSR reports some information about VMX support. We
2032		 * should return information about the VMX we emulate for the
2033		 * guest, and the VMCS structure we give it - not about the
2034		 * VMX support of the underlying hardware.
2035		 */
2036		*pdata = VMCS12_REVISION |
2037			   ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2038			   (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2039		break;
2040	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2041	case MSR_IA32_VMX_PINBASED_CTLS:
2042		*pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2043					nested_vmx_pinbased_ctls_high);
2044		break;
2045	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2046	case MSR_IA32_VMX_PROCBASED_CTLS:
2047		*pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2048					nested_vmx_procbased_ctls_high);
2049		break;
2050	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2051	case MSR_IA32_VMX_EXIT_CTLS:
2052		*pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2053					nested_vmx_exit_ctls_high);
2054		break;
2055	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2056	case MSR_IA32_VMX_ENTRY_CTLS:
2057		*pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2058					nested_vmx_entry_ctls_high);
2059		break;
2060	case MSR_IA32_VMX_MISC:
2061		*pdata = 0;
2062		break;
2063	/*
2064	 * These MSRs specify bits which the guest must keep fixed (on or off)
2065	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2066	 * We picked the standard core2 setting.
2067	 */
2068#define VMXON_CR0_ALWAYSON	(X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2069#define VMXON_CR4_ALWAYSON	X86_CR4_VMXE
2070	case MSR_IA32_VMX_CR0_FIXED0:
2071		*pdata = VMXON_CR0_ALWAYSON;
2072		break;
2073	case MSR_IA32_VMX_CR0_FIXED1:
2074		*pdata = -1ULL;
2075		break;
2076	case MSR_IA32_VMX_CR4_FIXED0:
2077		*pdata = VMXON_CR4_ALWAYSON;
2078		break;
2079	case MSR_IA32_VMX_CR4_FIXED1:
2080		*pdata = -1ULL;
2081		break;
2082	case MSR_IA32_VMX_VMCS_ENUM:
2083		*pdata = 0x1f;
2084		break;
2085	case MSR_IA32_VMX_PROCBASED_CTLS2:
2086		*pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2087					nested_vmx_secondary_ctls_high);
2088		break;
2089	case MSR_IA32_VMX_EPT_VPID_CAP:
2090		/* Currently, no nested ept or nested vpid */
2091		*pdata = 0;
2092		break;
2093	default:
2094		return 0;
2095	}
2096
2097	return 1;
2098}
2099
2100static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2101{
2102	if (!nested_vmx_allowed(vcpu))
2103		return 0;
2104
2105	if (msr_index == MSR_IA32_FEATURE_CONTROL)
2106		/* TODO: the right thing. */
2107		return 1;
2108	/*
2109	 * No need to treat VMX capability MSRs specially: If we don't handle
2110	 * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2111	 */
2112	return 0;
2113}
2114
2115/*
2116 * Reads an msr value (of 'msr_index') into 'pdata'.
2117 * Returns 0 on success, non-0 otherwise.
2118 * Assumes vcpu_load() was already called.
2119 */
2120static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2121{
2122	u64 data;
2123	struct shared_msr_entry *msr;
2124
2125	if (!pdata) {
2126		printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2127		return -EINVAL;
2128	}
2129
2130	switch (msr_index) {
2131#ifdef CONFIG_X86_64
2132	case MSR_FS_BASE:
2133		data = vmcs_readl(GUEST_FS_BASE);
2134		break;
2135	case MSR_GS_BASE:
2136		data = vmcs_readl(GUEST_GS_BASE);
2137		break;
2138	case MSR_KERNEL_GS_BASE:
2139		vmx_load_host_state(to_vmx(vcpu));
2140		data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2141		break;
2142#endif
2143	case MSR_EFER:
2144		return kvm_get_msr_common(vcpu, msr_index, pdata);
2145	case MSR_IA32_TSC:
2146		data = guest_read_tsc();
2147		break;
2148	case MSR_IA32_SYSENTER_CS:
2149		data = vmcs_read32(GUEST_SYSENTER_CS);
2150		break;
2151	case MSR_IA32_SYSENTER_EIP:
2152		data = vmcs_readl(GUEST_SYSENTER_EIP);
2153		break;
2154	case MSR_IA32_SYSENTER_ESP:
2155		data = vmcs_readl(GUEST_SYSENTER_ESP);
2156		break;
2157	case MSR_TSC_AUX:
2158		if (!to_vmx(vcpu)->rdtscp_enabled)
2159			return 1;
2160		/* Otherwise falls through */
2161	default:
2162		if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2163			return 0;
2164		msr = find_msr_entry(to_vmx(vcpu), msr_index);
2165		if (msr) {
2166			data = msr->data;
2167			break;
2168		}
2169		return kvm_get_msr_common(vcpu, msr_index, pdata);
2170	}
2171
2172	*pdata = data;
2173	return 0;
2174}
2175
2176/*
2177 * Writes msr value into into the appropriate "register".
2178 * Returns 0 on success, non-0 otherwise.
2179 * Assumes vcpu_load() was already called.
2180 */
2181static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2182{
2183	struct vcpu_vmx *vmx = to_vmx(vcpu);
2184	struct shared_msr_entry *msr;
2185	int ret = 0;
2186
2187	switch (msr_index) {
2188	case MSR_EFER:
2189		ret = kvm_set_msr_common(vcpu, msr_index, data);
2190		break;
2191#ifdef CONFIG_X86_64
2192	case MSR_FS_BASE:
2193		vmx_segment_cache_clear(vmx);
2194		vmcs_writel(GUEST_FS_BASE, data);
2195		break;
2196	case MSR_GS_BASE:
2197		vmx_segment_cache_clear(vmx);
2198		vmcs_writel(GUEST_GS_BASE, data);
2199		break;
2200	case MSR_KERNEL_GS_BASE:
2201		vmx_load_host_state(vmx);
2202		vmx->msr_guest_kernel_gs_base = data;
2203		break;
2204#endif
2205	case MSR_IA32_SYSENTER_CS:
2206		vmcs_write32(GUEST_SYSENTER_CS, data);
2207		break;
2208	case MSR_IA32_SYSENTER_EIP:
2209		vmcs_writel(GUEST_SYSENTER_EIP, data);
2210		break;
2211	case MSR_IA32_SYSENTER_ESP:
2212		vmcs_writel(GUEST_SYSENTER_ESP, data);
2213		break;
2214	case MSR_IA32_TSC:
2215		kvm_write_tsc(vcpu, data);
2216		break;
2217	case MSR_IA32_CR_PAT:
2218		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2219			vmcs_write64(GUEST_IA32_PAT, data);
2220			vcpu->arch.pat = data;
2221			break;
2222		}
2223		ret = kvm_set_msr_common(vcpu, msr_index, data);
2224		break;
2225	case MSR_TSC_AUX:
2226		if (!vmx->rdtscp_enabled)
2227			return 1;
2228		/* Check reserved bit, higher 32 bits should be zero */
2229		if ((data >> 32) != 0)
2230			return 1;
2231		/* Otherwise falls through */
2232	default:
2233		if (vmx_set_vmx_msr(vcpu, msr_index, data))
2234			break;
2235		msr = find_msr_entry(vmx, msr_index);
2236		if (msr) {
2237			msr->data = data;
2238			if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2239				preempt_disable();
2240				kvm_set_shared_msr(msr->index, msr->data,
2241						   msr->mask);
2242				preempt_enable();
2243			}
2244			break;
2245		}
2246		ret = kvm_set_msr_common(vcpu, msr_index, data);
2247	}
2248
2249	return ret;
2250}
2251
2252static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2253{
2254	__set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2255	switch (reg) {
2256	case VCPU_REGS_RSP:
2257		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2258		break;
2259	case VCPU_REGS_RIP:
2260		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2261		break;
2262	case VCPU_EXREG_PDPTR:
2263		if (enable_ept)
2264			ept_save_pdptrs(vcpu);
2265		break;
2266	default:
2267		break;
2268	}
2269}
2270
2271static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
2272{
2273	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
2274		vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
2275	else
2276		vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2277
2278	update_exception_bitmap(vcpu);
2279}
2280
2281static __init int cpu_has_kvm_support(void)
2282{
2283	return cpu_has_vmx();
2284}
2285
2286static __init int vmx_disabled_by_bios(void)
2287{
2288	u64 msr;
2289
2290	rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2291	if (msr & FEATURE_CONTROL_LOCKED) {
2292		/* launched w/ TXT and VMX disabled */
2293		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2294			&& tboot_enabled())
2295			return 1;
2296		/* launched w/o TXT and VMX only enabled w/ TXT */
2297		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2298			&& (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2299			&& !tboot_enabled()) {
2300			printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2301				"activate TXT before enabling KVM\n");
2302			return 1;
2303		}
2304		/* launched w/o TXT and VMX disabled */
2305		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2306			&& !tboot_enabled())
2307			return 1;
2308	}
2309
2310	return 0;
2311}
2312
2313static void kvm_cpu_vmxon(u64 addr)
2314{
2315	asm volatile (ASM_VMX_VMXON_RAX
2316			: : "a"(&addr), "m"(addr)
2317			: "memory", "cc");
2318}
2319
2320static int hardware_enable(void *garbage)
2321{
2322	int cpu = raw_smp_processor_id();
2323	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2324	u64 old, test_bits;
2325
2326	if (read_cr4() & X86_CR4_VMXE)
2327		return -EBUSY;
2328
2329	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2330	rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2331
2332	test_bits = FEATURE_CONTROL_LOCKED;
2333	test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2334	if (tboot_enabled())
2335		test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2336
2337	if ((old & test_bits) != test_bits) {
2338		/* enable and lock */
2339		wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2340	}
2341	write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2342
2343	if (vmm_exclusive) {
2344		kvm_cpu_vmxon(phys_addr);
2345		ept_sync_global();
2346	}
2347
2348	store_gdt(&__get_cpu_var(host_gdt));
2349
2350	return 0;
2351}
2352
2353static void vmclear_local_loaded_vmcss(void)
2354{
2355	int cpu = raw_smp_processor_id();
2356	struct loaded_vmcs *v, *n;
2357
2358	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2359				 loaded_vmcss_on_cpu_link)
2360		__loaded_vmcs_clear(v);
2361}
2362
2363
2364/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2365 * tricks.
2366 */
2367static void kvm_cpu_vmxoff(void)
2368{
2369	asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2370}
2371
2372static void hardware_disable(void *garbage)
2373{
2374	if (vmm_exclusive) {
2375		vmclear_local_loaded_vmcss();
2376		kvm_cpu_vmxoff();
2377	}
2378	write_cr4(read_cr4() & ~X86_CR4_VMXE);
2379}
2380
2381static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2382				      u32 msr, u32 *result)
2383{
2384	u32 vmx_msr_low, vmx_msr_high;
2385	u32 ctl = ctl_min | ctl_opt;
2386
2387	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2388
2389	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2390	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2391
2392	/* Ensure minimum (required) set of control bits are supported. */
2393	if (ctl_min & ~ctl)
2394		return -EIO;
2395
2396	*result = ctl;
2397	return 0;
2398}
2399
2400static __init bool allow_1_setting(u32 msr, u32 ctl)
2401{
2402	u32 vmx_msr_low, vmx_msr_high;
2403
2404	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2405	return vmx_msr_high & ctl;
2406}
2407
2408static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2409{
2410	u32 vmx_msr_low, vmx_msr_high;
2411	u32 min, opt, min2, opt2;
2412	u32 _pin_based_exec_control = 0;
2413	u32 _cpu_based_exec_control = 0;
2414	u32 _cpu_based_2nd_exec_control = 0;
2415	u32 _vmexit_control = 0;
2416	u32 _vmentry_control = 0;
2417
2418	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2419	opt = PIN_BASED_VIRTUAL_NMIS;
2420	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2421				&_pin_based_exec_control) < 0)
2422		return -EIO;
2423
2424	min = CPU_BASED_HLT_EXITING |
2425#ifdef CONFIG_X86_64
2426	      CPU_BASED_CR8_LOAD_EXITING |
2427	      CPU_BASED_CR8_STORE_EXITING |
2428#endif
2429	      CPU_BASED_CR3_LOAD_EXITING |
2430	      CPU_BASED_CR3_STORE_EXITING |
2431	      CPU_BASED_USE_IO_BITMAPS |
2432	      CPU_BASED_MOV_DR_EXITING |
2433	      CPU_BASED_USE_TSC_OFFSETING |
2434	      CPU_BASED_MWAIT_EXITING |
2435	      CPU_BASED_MONITOR_EXITING |
2436	      CPU_BASED_INVLPG_EXITING |
2437	      CPU_BASED_RDPMC_EXITING;
2438
2439	opt = CPU_BASED_TPR_SHADOW |
2440	      CPU_BASED_USE_MSR_BITMAPS |
2441	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2442	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2443				&_cpu_based_exec_control) < 0)
2444		return -EIO;
2445#ifdef CONFIG_X86_64
2446	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2447		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2448					   ~CPU_BASED_CR8_STORE_EXITING;
2449#endif
2450	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2451		min2 = 0;
2452		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2453			SECONDARY_EXEC_WBINVD_EXITING |
2454			SECONDARY_EXEC_ENABLE_VPID |
2455			SECONDARY_EXEC_ENABLE_EPT |
2456			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2457			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2458			SECONDARY_EXEC_RDTSCP;
2459		if (adjust_vmx_controls(min2, opt2,
2460					MSR_IA32_VMX_PROCBASED_CTLS2,
2461					&_cpu_based_2nd_exec_control) < 0)
2462			return -EIO;
2463	}
2464#ifndef CONFIG_X86_64
2465	if (!(_cpu_based_2nd_exec_control &
2466				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2467		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2468#endif
2469	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2470		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2471		   enabled */
2472		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2473					     CPU_BASED_CR3_STORE_EXITING |
2474					     CPU_BASED_INVLPG_EXITING);
2475		rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2476		      vmx_capability.ept, vmx_capability.vpid);
2477	}
2478
2479	min = 0;
2480#ifdef CONFIG_X86_64
2481	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2482#endif
2483	opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2484	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2485				&_vmexit_control) < 0)
2486		return -EIO;
2487
2488	min = 0;
2489	opt = VM_ENTRY_LOAD_IA32_PAT;
2490	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2491				&_vmentry_control) < 0)
2492		return -EIO;
2493
2494	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2495
2496	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2497	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2498		return -EIO;
2499
2500#ifdef CONFIG_X86_64
2501	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2502	if (vmx_msr_high & (1u<<16))
2503		return -EIO;
2504#endif
2505
2506	/* Require Write-Back (WB) memory type for VMCS accesses. */
2507	if (((vmx_msr_high >> 18) & 15) != 6)
2508		return -EIO;
2509
2510	vmcs_conf->size = vmx_msr_high & 0x1fff;
2511	vmcs_conf->order = get_order(vmcs_config.size);
2512	vmcs_conf->revision_id = vmx_msr_low;
2513
2514	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2515	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2516	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2517	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2518	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2519
2520	cpu_has_load_ia32_efer =
2521		allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2522				VM_ENTRY_LOAD_IA32_EFER)
2523		&& allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2524				   VM_EXIT_LOAD_IA32_EFER);
2525
2526	cpu_has_load_perf_global_ctrl =
2527		allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2528				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2529		&& allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2530				   VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2531
2532	/*
2533	 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2534	 * but due to arrata below it can't be used. Workaround is to use
2535	 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2536	 *
2537	 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2538	 *
2539	 * AAK155             (model 26)
2540	 * AAP115             (model 30)
2541	 * AAT100             (model 37)
2542	 * BC86,AAY89,BD102   (model 44)
2543	 * BA97               (model 46)
2544	 *
2545	 */
2546	if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2547		switch (boot_cpu_data.x86_model) {
2548		case 26:
2549		case 30:
2550		case 37:
2551		case 44:
2552		case 46:
2553			cpu_has_load_perf_global_ctrl = false;
2554			printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2555					"does not work properly. Using workaround\n");
2556			break;
2557		default:
2558			break;
2559		}
2560	}
2561
2562	return 0;
2563}
2564
2565static struct vmcs *alloc_vmcs_cpu(int cpu)
2566{
2567	int node = cpu_to_node(cpu);
2568	struct page *pages;
2569	struct vmcs *vmcs;
2570
2571	pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2572	if (!pages)
2573		return NULL;
2574	vmcs = page_address(pages);
2575	memset(vmcs, 0, vmcs_config.size);
2576	vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2577	return vmcs;
2578}
2579
2580static struct vmcs *alloc_vmcs(void)
2581{
2582	return alloc_vmcs_cpu(raw_smp_processor_id());
2583}
2584
2585static void free_vmcs(struct vmcs *vmcs)
2586{
2587	free_pages((unsigned long)vmcs, vmcs_config.order);
2588}
2589
2590/*
2591 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2592 */
2593static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2594{
2595	if (!loaded_vmcs->vmcs)
2596		return;
2597	loaded_vmcs_clear(loaded_vmcs);
2598	free_vmcs(loaded_vmcs->vmcs);
2599	loaded_vmcs->vmcs = NULL;
2600}
2601
2602static void free_kvm_area(void)
2603{
2604	int cpu;
2605
2606	for_each_possible_cpu(cpu) {
2607		free_vmcs(per_cpu(vmxarea, cpu));
2608		per_cpu(vmxarea, cpu) = NULL;
2609	}
2610}
2611
2612static __init int alloc_kvm_area(void)
2613{
2614	int cpu;
2615
2616	for_each_possible_cpu(cpu) {
2617		struct vmcs *vmcs;
2618
2619		vmcs = alloc_vmcs_cpu(cpu);
2620		if (!vmcs) {
2621			free_kvm_area();
2622			return -ENOMEM;
2623		}
2624
2625		per_cpu(vmxarea, cpu) = vmcs;
2626	}
2627	return 0;
2628}
2629
2630static __init int hardware_setup(void)
2631{
2632	if (setup_vmcs_config(&vmcs_config) < 0)
2633		return -EIO;
2634
2635	if (boot_cpu_has(X86_FEATURE_NX))
2636		kvm_enable_efer_bits(EFER_NX);
2637
2638	if (!cpu_has_vmx_vpid())
2639		enable_vpid = 0;
2640
2641	if (!cpu_has_vmx_ept() ||
2642	    !cpu_has_vmx_ept_4levels()) {
2643		enable_ept = 0;
2644		enable_unrestricted_guest = 0;
2645	}
2646
2647	if (!cpu_has_vmx_unrestricted_guest())
2648		enable_unrestricted_guest = 0;
2649
2650	if (!cpu_has_vmx_flexpriority())
2651		flexpriority_enabled = 0;
2652
2653	if (!cpu_has_vmx_tpr_shadow())
2654		kvm_x86_ops->update_cr8_intercept = NULL;
2655
2656	if (enable_ept && !cpu_has_vmx_ept_2m_page())
2657		kvm_disable_largepages();
2658
2659	if (!cpu_has_vmx_ple())
2660		ple_gap = 0;
2661
2662	if (nested)
2663		nested_vmx_setup_ctls_msrs();
2664
2665	return alloc_kvm_area();
2666}
2667
2668static __exit void hardware_unsetup(void)
2669{
2670	free_kvm_area();
2671}
2672
2673static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
2674{
2675	struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2676
2677	if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
2678		vmcs_write16(sf->selector, save->selector);
2679		vmcs_writel(sf->base, save->base);
2680		vmcs_write32(sf->limit, save->limit);
2681		vmcs_write32(sf->ar_bytes, save->ar);
2682	} else {
2683		u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
2684			<< AR_DPL_SHIFT;
2685		vmcs_write32(sf->ar_bytes, 0x93 | dpl);
2686	}
2687}
2688
2689static void enter_pmode(struct kvm_vcpu *vcpu)
2690{
2691	unsigned long flags;
2692	struct vcpu_vmx *vmx = to_vmx(vcpu);
2693
2694	vmx->emulation_required = 1;
2695	vmx->rmode.vm86_active = 0;
2696
2697	vmx_segment_cache_clear(vmx);
2698
2699	vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
2700	vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
2701	vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
2702	vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
2703
2704	flags = vmcs_readl(GUEST_RFLAGS);
2705	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2706	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2707	vmcs_writel(GUEST_RFLAGS, flags);
2708
2709	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2710			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2711
2712	update_exception_bitmap(vcpu);
2713
2714	if (emulate_invalid_guest_state)
2715		return;
2716
2717	fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
2718	fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
2719	fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
2720	fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
2721
2722	vmx_segment_cache_clear(vmx);
2723
2724	vmcs_write16(GUEST_SS_SELECTOR, 0);
2725	vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2726
2727	vmcs_write16(GUEST_CS_SELECTOR,
2728		     vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2729	vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2730}
2731
2732static gva_t rmode_tss_base(struct kvm *kvm)
2733{
2734	if (!kvm->arch.tss_addr) {
2735		struct kvm_memslots *slots;
2736		struct kvm_memory_slot *slot;
2737		gfn_t base_gfn;
2738
2739		slots = kvm_memslots(kvm);
2740		slot = id_to_memslot(slots, 0);
2741		base_gfn = slot->base_gfn + slot->npages - 3;
2742
2743		return base_gfn << PAGE_SHIFT;
2744	}
2745	return kvm->arch.tss_addr;
2746}
2747
2748static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
2749{
2750	struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2751
2752	save->selector = vmcs_read16(sf->selector);
2753	save->base = vmcs_readl(sf->base);
2754	save->limit = vmcs_read32(sf->limit);
2755	save->ar = vmcs_read32(sf->ar_bytes);
2756	vmcs_write16(sf->selector, save->base >> 4);
2757	vmcs_write32(sf->base, save->base & 0xffff0);
2758	vmcs_write32(sf->limit, 0xffff);
2759	vmcs_write32(sf->ar_bytes, 0xf3);
2760	if (save->base & 0xf)
2761		printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2762			    " aligned when entering protected mode (seg=%d)",
2763			    seg);
2764}
2765
2766static void enter_rmode(struct kvm_vcpu *vcpu)
2767{
2768	unsigned long flags;
2769	struct vcpu_vmx *vmx = to_vmx(vcpu);
2770	struct kvm_segment var;
2771
2772	if (enable_unrestricted_guest)
2773		return;
2774
2775	vmx->emulation_required = 1;
2776	vmx->rmode.vm86_active = 1;
2777
2778	/*
2779	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2780	 * vcpu. Call it here with phys address pointing 16M below 4G.
2781	 */
2782	if (!vcpu->kvm->arch.tss_addr) {
2783		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2784			     "called before entering vcpu\n");
2785		srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2786		vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2787		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2788	}
2789
2790	vmx_segment_cache_clear(vmx);
2791
2792	vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
2793	vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
2794	vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
2795
2796	vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
2797	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2798
2799	vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
2800	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2801
2802	flags = vmcs_readl(GUEST_RFLAGS);
2803	vmx->rmode.save_rflags = flags;
2804
2805	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2806
2807	vmcs_writel(GUEST_RFLAGS, flags);
2808	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2809	update_exception_bitmap(vcpu);
2810
2811	if (emulate_invalid_guest_state)
2812		goto continue_rmode;
2813
2814	vmx_get_segment(vcpu, &var, VCPU_SREG_SS);
2815	vmx_set_segment(vcpu, &var, VCPU_SREG_SS);
2816
2817	vmx_get_segment(vcpu, &var, VCPU_SREG_CS);
2818	vmx_set_segment(vcpu, &var, VCPU_SREG_CS);
2819
2820	vmx_get_segment(vcpu, &var, VCPU_SREG_ES);
2821	vmx_set_segment(vcpu, &var, VCPU_SREG_ES);
2822
2823	vmx_get_segment(vcpu, &var, VCPU_SREG_DS);
2824	vmx_set_segment(vcpu, &var, VCPU_SREG_DS);
2825
2826	vmx_get_segment(vcpu, &var, VCPU_SREG_GS);
2827	vmx_set_segment(vcpu, &var, VCPU_SREG_GS);
2828
2829	vmx_get_segment(vcpu, &var, VCPU_SREG_FS);
2830	vmx_set_segment(vcpu, &var, VCPU_SREG_FS);
2831
2832continue_rmode:
2833	kvm_mmu_reset_context(vcpu);
2834}
2835
2836static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2837{
2838	struct vcpu_vmx *vmx = to_vmx(vcpu);
2839	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2840
2841	if (!msr)
2842		return;
2843
2844	/*
2845	 * Force kernel_gs_base reloading before EFER changes, as control
2846	 * of this msr depends on is_long_mode().
2847	 */
2848	vmx_load_host_state(to_vmx(vcpu));
2849	vcpu->arch.efer = efer;
2850	if (efer & EFER_LMA) {
2851		vmcs_write32(VM_ENTRY_CONTROLS,
2852			     vmcs_read32(VM_ENTRY_CONTROLS) |
2853			     VM_ENTRY_IA32E_MODE);
2854		msr->data = efer;
2855	} else {
2856		vmcs_write32(VM_ENTRY_CONTROLS,
2857			     vmcs_read32(VM_ENTRY_CONTROLS) &
2858			     ~VM_ENTRY_IA32E_MODE);
2859
2860		msr->data = efer & ~EFER_LME;
2861	}
2862	setup_msrs(vmx);
2863}
2864
2865#ifdef CONFIG_X86_64
2866
2867static void enter_lmode(struct kvm_vcpu *vcpu)
2868{
2869	u32 guest_tr_ar;
2870
2871	vmx_segment_cache_clear(to_vmx(vcpu));
2872
2873	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2874	if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
2875		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2876				     __func__);
2877		vmcs_write32(GUEST_TR_AR_BYTES,
2878			     (guest_tr_ar & ~AR_TYPE_MASK)
2879			     | AR_TYPE_BUSY_64_TSS);
2880	}
2881	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2882}
2883
2884static void exit_lmode(struct kvm_vcpu *vcpu)
2885{
2886	vmcs_write32(VM_ENTRY_CONTROLS,
2887		     vmcs_read32(VM_ENTRY_CONTROLS)
2888		     & ~VM_ENTRY_IA32E_MODE);
2889	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2890}
2891
2892#endif
2893
2894static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2895{
2896	vpid_sync_context(to_vmx(vcpu));
2897	if (enable_ept) {
2898		if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2899			return;
2900		ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2901	}
2902}
2903
2904static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2905{
2906	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2907
2908	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2909	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2910}
2911
2912static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2913{
2914	if (enable_ept && is_paging(vcpu))
2915		vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2916	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2917}
2918
2919static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2920{
2921	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2922
2923	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2924	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2925}
2926
2927static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2928{
2929	if (!test_bit(VCPU_EXREG_PDPTR,
2930		      (unsigned long *)&vcpu->arch.regs_dirty))
2931		return;
2932
2933	if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2934		vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2935		vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2936		vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2937		vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2938	}
2939}
2940
2941static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2942{
2943	if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2944		vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2945		vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2946		vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2947		vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2948	}
2949
2950	__set_bit(VCPU_EXREG_PDPTR,
2951		  (unsigned long *)&vcpu->arch.regs_avail);
2952	__set_bit(VCPU_EXREG_PDPTR,
2953		  (unsigned long *)&vcpu->arch.regs_dirty);
2954}
2955
2956static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2957
2958static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2959					unsigned long cr0,
2960					struct kvm_vcpu *vcpu)
2961{
2962	if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2963		vmx_decache_cr3(vcpu);
2964	if (!(cr0 & X86_CR0_PG)) {
2965		/* From paging/starting to nonpaging */
2966		vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2967			     vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2968			     (CPU_BASED_CR3_LOAD_EXITING |
2969			      CPU_BASED_CR3_STORE_EXITING));
2970		vcpu->arch.cr0 = cr0;
2971		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2972	} else if (!is_paging(vcpu)) {
2973		/* From nonpaging to paging */
2974		vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2975			     vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2976			     ~(CPU_BASED_CR3_LOAD_EXITING |
2977			       CPU_BASED_CR3_STORE_EXITING));
2978		vcpu->arch.cr0 = cr0;
2979		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2980	}
2981
2982	if (!(cr0 & X86_CR0_WP))
2983		*hw_cr0 &= ~X86_CR0_WP;
2984}
2985
2986static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2987{
2988	struct vcpu_vmx *vmx = to_vmx(vcpu);
2989	unsigned long hw_cr0;
2990
2991	if (enable_unrestricted_guest)
2992		hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
2993			| KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2994	else
2995		hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
2996
2997	if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2998		enter_pmode(vcpu);
2999
3000	if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3001		enter_rmode(vcpu);
3002
3003#ifdef CONFIG_X86_64
3004	if (vcpu->arch.efer & EFER_LME) {
3005		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3006			enter_lmode(vcpu);
3007		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3008			exit_lmode(vcpu);
3009	}
3010#endif
3011
3012	if (enable_ept)
3013		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3014
3015	if (!vcpu->fpu_active)
3016		hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3017
3018	vmcs_writel(CR0_READ_SHADOW, cr0);
3019	vmcs_writel(GUEST_CR0, hw_cr0);
3020	vcpu->arch.cr0 = cr0;
3021	__clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3022}
3023
3024static u64 construct_eptp(unsigned long root_hpa)
3025{
3026	u64 eptp;
3027
3028	/* TODO write the value reading from MSR */
3029	eptp = VMX_EPT_DEFAULT_MT |
3030		VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3031	eptp |= (root_hpa & PAGE_MASK);
3032
3033	return eptp;
3034}
3035
3036static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3037{
3038	unsigned long guest_cr3;
3039	u64 eptp;
3040
3041	guest_cr3 = cr3;
3042	if (enable_ept) {
3043		eptp = construct_eptp(cr3);
3044		vmcs_write64(EPT_POINTER, eptp);
3045		guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3046			vcpu->kvm->arch.ept_identity_map_addr;
3047		ept_load_pdptrs(vcpu);
3048	}
3049
3050	vmx_flush_tlb(vcpu);
3051	vmcs_writel(GUEST_CR3, guest_cr3);
3052}
3053
3054static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3055{
3056	unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3057		    KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3058
3059	if (cr4 & X86_CR4_VMXE) {
3060		/*
3061		 * To use VMXON (and later other VMX instructions), a guest
3062		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3063		 * So basically the check on whether to allow nested VMX
3064		 * is here.
3065		 */
3066		if (!nested_vmx_allowed(vcpu))
3067			return 1;
3068	} else if (to_vmx(vcpu)->nested.vmxon)
3069		return 1;
3070
3071	vcpu->arch.cr4 = cr4;
3072	if (enable_ept) {
3073		if (!is_paging(vcpu)) {
3074			hw_cr4 &= ~X86_CR4_PAE;
3075			hw_cr4 |= X86_CR4_PSE;
3076		} else if (!(cr4 & X86_CR4_PAE)) {
3077			hw_cr4 &= ~X86_CR4_PAE;
3078		}
3079	}
3080
3081	vmcs_writel(CR4_READ_SHADOW, cr4);
3082	vmcs_writel(GUEST_CR4, hw_cr4);
3083	return 0;
3084}
3085
3086static void vmx_get_segment(struct kvm_vcpu *vcpu,
3087			    struct kvm_segment *var, int seg)
3088{
3089	struct vcpu_vmx *vmx = to_vmx(vcpu);
3090	struct kvm_save_segment *save;
3091	u32 ar;
3092
3093	if (vmx->rmode.vm86_active
3094	    && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
3095		|| seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
3096		|| seg == VCPU_SREG_GS)
3097	    && !emulate_invalid_guest_state) {
3098		switch (seg) {
3099		case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
3100		case VCPU_SREG_ES: save = &vmx->rmode.es; break;
3101		case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
3102		case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
3103		case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
3104		default: BUG();
3105		}
3106		var->selector = save->selector;
3107		var->base = save->base;
3108		var->limit = save->limit;
3109		ar = save->ar;
3110		if (seg == VCPU_SREG_TR
3111		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3112			goto use_saved_rmode_seg;
3113	}
3114	var->base = vmx_read_guest_seg_base(vmx, seg);
3115	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3116	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3117	ar = vmx_read_guest_seg_ar(vmx, seg);
3118use_saved_rmode_seg:
3119	if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
3120		ar = 0;
3121	var->type = ar & 15;
3122	var->s = (ar >> 4) & 1;
3123	var->dpl = (ar >> 5) & 3;
3124	var->present = (ar >> 7) & 1;
3125	var->avl = (ar >> 12) & 1;
3126	var->l = (ar >> 13) & 1;
3127	var->db = (ar >> 14) & 1;
3128	var->g = (ar >> 15) & 1;
3129	var->unusable = (ar >> 16) & 1;
3130}
3131
3132static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3133{
3134	struct kvm_segment s;
3135
3136	if (to_vmx(vcpu)->rmode.vm86_active) {
3137		vmx_get_segment(vcpu, &s, seg);
3138		return s.base;
3139	}
3140	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3141}
3142
3143static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
3144{
3145	if (!is_protmode(vcpu))
3146		return 0;
3147
3148	if (!is_long_mode(vcpu)
3149	    && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3150		return 3;
3151
3152	return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
3153}
3154
3155static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3156{
3157	if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3158		__set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3159		to_vmx(vcpu)->cpl = __vmx_get_cpl(vcpu);
3160	}
3161	return to_vmx(vcpu)->cpl;
3162}
3163
3164
3165static u32 vmx_segment_access_rights(struct kvm_segment *var)
3166{
3167	u32 ar;
3168
3169	if (var->unusable)
3170		ar = 1 << 16;
3171	else {
3172		ar = var->type & 15;
3173		ar |= (var->s & 1) << 4;
3174		ar |= (var->dpl & 3) << 5;
3175		ar |= (var->present & 1) << 7;
3176		ar |= (var->avl & 1) << 12;
3177		ar |= (var->l & 1) << 13;
3178		ar |= (var->db & 1) << 14;
3179		ar |= (var->g & 1) << 15;
3180	}
3181	if (ar == 0) /* a 0 value means unusable */
3182		ar = AR_UNUSABLE_MASK;
3183
3184	return ar;
3185}
3186
3187static void vmx_set_segment(struct kvm_vcpu *vcpu,
3188			    struct kvm_segment *var, int seg)
3189{
3190	struct vcpu_vmx *vmx = to_vmx(vcpu);
3191	struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3192	u32 ar;
3193
3194	vmx_segment_cache_clear(vmx);
3195
3196	if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
3197		vmcs_write16(sf->selector, var->selector);
3198		vmx->rmode.tr.selector = var->selector;
3199		vmx->rmode.tr.base = var->base;
3200		vmx->rmode.tr.limit = var->limit;
3201		vmx->rmode.tr.ar = vmx_segment_access_rights(var);
3202		return;
3203	}
3204	vmcs_writel(sf->base, var->base);
3205	vmcs_write32(sf->limit, var->limit);
3206	vmcs_write16(sf->selector, var->selector);
3207	if (vmx->rmode.vm86_active && var->s) {
3208		/*
3209		 * Hack real-mode segments into vm86 compatibility.
3210		 */
3211		if (var->base == 0xffff0000 && var->selector == 0xf000)
3212			vmcs_writel(sf->base, 0xf0000);
3213		ar = 0xf3;
3214	} else
3215		ar = vmx_segment_access_rights(var);
3216
3217	/*
3218	 *   Fix the "Accessed" bit in AR field of segment registers for older
3219	 * qemu binaries.
3220	 *   IA32 arch specifies that at the time of processor reset the
3221	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3222	 * is setting it to 0 in the usedland code. This causes invalid guest
3223	 * state vmexit when "unrestricted guest" mode is turned on.
3224	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3225	 * tree. Newer qemu binaries with that qemu fix would not need this
3226	 * kvm hack.
3227	 */
3228	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3229		ar |= 0x1; /* Accessed */
3230
3231	vmcs_write32(sf->ar_bytes, ar);
3232	__clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3233
3234	/*
3235	 * Fix segments for real mode guest in hosts that don't have
3236	 * "unrestricted_mode" or it was disabled.
3237	 * This is done to allow migration of the guests from hosts with
3238	 * unrestricted guest like Westmere to older host that don't have
3239	 * unrestricted guest like Nehelem.
3240	 */
3241	if (!enable_unrestricted_guest && vmx->rmode.vm86_active) {
3242		switch (seg) {
3243		case VCPU_SREG_CS:
3244			vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
3245			vmcs_write32(GUEST_CS_LIMIT, 0xffff);
3246			if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
3247				vmcs_writel(GUEST_CS_BASE, 0xf0000);
3248			vmcs_write16(GUEST_CS_SELECTOR,
3249				     vmcs_readl(GUEST_CS_BASE) >> 4);
3250			break;
3251		case VCPU_SREG_ES:
3252			fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
3253			break;
3254		case VCPU_SREG_DS:
3255			fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
3256			break;
3257		case VCPU_SREG_GS:
3258			fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
3259			break;
3260		case VCPU_SREG_FS:
3261			fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
3262			break;
3263		case VCPU_SREG_SS:
3264			vmcs_write16(GUEST_SS_SELECTOR,
3265				     vmcs_readl(GUEST_SS_BASE) >> 4);
3266			vmcs_write32(GUEST_SS_LIMIT, 0xffff);
3267			vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
3268			break;
3269		}
3270	}
3271}
3272
3273static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3274{
3275	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3276
3277	*db = (ar >> 14) & 1;
3278	*l = (ar >> 13) & 1;
3279}
3280
3281static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3282{
3283	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3284	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3285}
3286
3287static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3288{
3289	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3290	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3291}
3292
3293static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3294{
3295	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3296	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3297}
3298
3299static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3300{
3301	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3302	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3303}
3304
3305static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3306{
3307	struct kvm_segment var;
3308	u32 ar;
3309
3310	vmx_get_segment(vcpu, &var, seg);
3311	ar = vmx_segment_access_rights(&var);
3312
3313	if (var.base != (var.selector << 4))
3314		return false;
3315	if (var.limit != 0xffff)
3316		return false;
3317	if (ar != 0xf3)
3318		return false;
3319
3320	return true;
3321}
3322
3323static bool code_segment_valid(struct kvm_vcpu *vcpu)
3324{
3325	struct kvm_segment cs;
3326	unsigned int cs_rpl;
3327
3328	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3329	cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3330
3331	if (cs.unusable)
3332		return false;
3333	if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3334		return false;
3335	if (!cs.s)
3336		return false;
3337	if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3338		if (cs.dpl > cs_rpl)
3339			return false;
3340	} else {
3341		if (cs.dpl != cs_rpl)
3342			return false;
3343	}
3344	if (!cs.present)
3345		return false;
3346
3347	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3348	return true;
3349}
3350
3351static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3352{
3353	struct kvm_segment ss;
3354	unsigned int ss_rpl;
3355
3356	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3357	ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3358
3359	if (ss.unusable)
3360		return true;
3361	if (ss.type != 3 && ss.type != 7)
3362		return false;
3363	if (!ss.s)
3364		return false;
3365	if (ss.dpl != ss_rpl) /* DPL != RPL */
3366		return false;
3367	if (!ss.present)
3368		return false;
3369
3370	return true;
3371}
3372
3373static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3374{
3375	struct kvm_segment var;
3376	unsigned int rpl;
3377
3378	vmx_get_segment(vcpu, &var, seg);
3379	rpl = var.selector & SELECTOR_RPL_MASK;
3380
3381	if (var.unusable)
3382		return true;
3383	if (!var.s)
3384		return false;
3385	if (!var.present)
3386		return false;
3387	if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3388		if (var.dpl < rpl) /* DPL < RPL */
3389			return false;
3390	}
3391
3392	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3393	 * rights flags
3394	 */
3395	return true;
3396}
3397
3398static bool tr_valid(struct kvm_vcpu *vcpu)
3399{
3400	struct kvm_segment tr;
3401
3402	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3403
3404	if (tr.unusable)
3405		return false;
3406	if (tr.selector & SELECTOR_TI_MASK)	/* TI = 1 */
3407		return false;
3408	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3409		return false;
3410	if (!tr.present)
3411		return false;
3412
3413	return true;
3414}
3415
3416static bool ldtr_valid(struct kvm_vcpu *vcpu)
3417{
3418	struct kvm_segment ldtr;
3419
3420	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3421
3422	if (ldtr.unusable)
3423		return true;
3424	if (ldtr.selector & SELECTOR_TI_MASK)	/* TI = 1 */
3425		return false;
3426	if (ldtr.type != 2)
3427		return false;
3428	if (!ldtr.present)
3429		return false;
3430
3431	return true;
3432}
3433
3434static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3435{
3436	struct kvm_segment cs, ss;
3437
3438	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3439	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3440
3441	return ((cs.selector & SELECTOR_RPL_MASK) ==
3442		 (ss.selector & SELECTOR_RPL_MASK));
3443}
3444
3445/*
3446 * Check if guest state is valid. Returns true if valid, false if
3447 * not.
3448 * We assume that registers are always usable
3449 */
3450static bool guest_state_valid(struct kvm_vcpu *vcpu)
3451{
3452	/* real mode guest state checks */
3453	if (!is_protmode(vcpu)) {
3454		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3455			return false;
3456		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3457			return false;
3458		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3459			return false;
3460		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3461			return false;
3462		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3463			return false;
3464		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3465			return false;
3466	} else {
3467	/* protected mode guest state checks */
3468		if (!cs_ss_rpl_check(vcpu))
3469			return false;
3470		if (!code_segment_valid(vcpu))
3471			return false;
3472		if (!stack_segment_valid(vcpu))
3473			return false;
3474		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3475			return false;
3476		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3477			return false;
3478		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3479			return false;
3480		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3481			return false;
3482		if (!tr_valid(vcpu))
3483			return false;
3484		if (!ldtr_valid(vcpu))
3485			return false;
3486	}
3487	/* TODO:
3488	 * - Add checks on RIP
3489	 * - Add checks on RFLAGS
3490	 */
3491
3492	return true;
3493}
3494
3495static int init_rmode_tss(struct kvm *kvm)
3496{
3497	gfn_t fn;
3498	u16 data = 0;
3499	int r, idx, ret = 0;
3500
3501	idx = srcu_read_lock(&kvm->srcu);
3502	fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
3503	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3504	if (r < 0)
3505		goto out;
3506	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3507	r = kvm_write_guest_page(kvm, fn++, &data,
3508			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3509	if (r < 0)
3510		goto out;
3511	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3512	if (r < 0)
3513		goto out;
3514	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3515	if (r < 0)
3516		goto out;
3517	data = ~0;
3518	r = kvm_write_guest_page(kvm, fn, &data,
3519				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3520				 sizeof(u8));
3521	if (r < 0)
3522		goto out;
3523
3524	ret = 1;
3525out:
3526	srcu_read_unlock(&kvm->srcu, idx);
3527	return ret;
3528}
3529
3530static int init_rmode_identity_map(struct kvm *kvm)
3531{
3532	int i, idx, r, ret;
3533	pfn_t identity_map_pfn;
3534	u32 tmp;
3535
3536	if (!enable_ept)
3537		return 1;
3538	if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3539		printk(KERN_ERR "EPT: identity-mapping pagetable "
3540			"haven't been allocated!\n");
3541		return 0;
3542	}
3543	if (likely(kvm->arch.ept_identity_pagetable_done))
3544		return 1;
3545	ret = 0;
3546	identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3547	idx = srcu_read_lock(&kvm->srcu);
3548	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3549	if (r < 0)
3550		goto out;
3551	/* Set up identity-mapping pagetable for EPT in real mode */
3552	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3553		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3554			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3555		r = kvm_write_guest_page(kvm, identity_map_pfn,
3556				&tmp, i * sizeof(tmp), sizeof(tmp));
3557		if (r < 0)
3558			goto out;
3559	}
3560	kvm->arch.ept_identity_pagetable_done = true;
3561	ret = 1;
3562out:
3563	srcu_read_unlock(&kvm->srcu, idx);
3564	return ret;
3565}
3566
3567static void seg_setup(int seg)
3568{
3569	struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3570	unsigned int ar;
3571
3572	vmcs_write16(sf->selector, 0);
3573	vmcs_writel(sf->base, 0);
3574	vmcs_write32(sf->limit, 0xffff);
3575	if (enable_unrestricted_guest) {
3576		ar = 0x93;
3577		if (seg == VCPU_SREG_CS)
3578			ar |= 0x08; /* code segment */
3579	} else
3580		ar = 0xf3;
3581
3582	vmcs_write32(sf->ar_bytes, ar);
3583}
3584
3585static int alloc_apic_access_page(struct kvm *kvm)
3586{
3587	struct kvm_userspace_memory_region kvm_userspace_mem;
3588	int r = 0;
3589
3590	mutex_lock(&kvm->slots_lock);
3591	if (kvm->arch.apic_access_page)
3592		goto out;
3593	kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3594	kvm_userspace_mem.flags = 0;
3595	kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3596	kvm_userspace_mem.memory_size = PAGE_SIZE;
3597	r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3598	if (r)
3599		goto out;
3600
3601	kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
3602out:
3603	mutex_unlock(&kvm->slots_lock);
3604	return r;
3605}
3606
3607static int alloc_identity_pagetable(struct kvm *kvm)
3608{
3609	struct kvm_userspace_memory_region kvm_userspace_mem;
3610	int r = 0;
3611
3612	mutex_lock(&kvm->slots_lock);
3613	if (kvm->arch.ept_identity_pagetable)
3614		goto out;
3615	kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3616	kvm_userspace_mem.flags = 0;
3617	kvm_userspace_mem.guest_phys_addr =
3618		kvm->arch.ept_identity_map_addr;
3619	kvm_userspace_mem.memory_size = PAGE_SIZE;
3620	r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3621	if (r)
3622		goto out;
3623
3624	kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
3625			kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3626out:
3627	mutex_unlock(&kvm->slots_lock);
3628	return r;
3629}
3630
3631static void allocate_vpid(struct vcpu_vmx *vmx)
3632{
3633	int vpid;
3634
3635	vmx->vpid = 0;
3636	if (!enable_vpid)
3637		return;
3638	spin_lock(&vmx_vpid_lock);
3639	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3640	if (vpid < VMX_NR_VPIDS) {
3641		vmx->vpid = vpid;
3642		__set_bit(vpid, vmx_vpid_bitmap);
3643	}
3644	spin_unlock(&vmx_vpid_lock);
3645}
3646
3647static void free_vpid(struct vcpu_vmx *vmx)
3648{
3649	if (!enable_vpid)
3650		return;
3651	spin_lock(&vmx_vpid_lock);
3652	if (vmx->vpid != 0)
3653		__clear_bit(vmx->vpid, vmx_vpid_bitmap);
3654	spin_unlock(&vmx_vpid_lock);
3655}
3656
3657static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
3658{
3659	int f = sizeof(unsigned long);
3660
3661	if (!cpu_has_vmx_msr_bitmap())
3662		return;
3663
3664	/*
3665	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3666	 * have the write-low and read-high bitmap offsets the wrong way round.
3667	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3668	 */
3669	if (msr <= 0x1fff) {
3670		__clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3671		__clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
3672	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3673		msr &= 0x1fff;
3674		__clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3675		__clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
3676	}
3677}
3678
3679static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3680{
3681	if (!longmode_only)
3682		__vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3683	__vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3684}
3685
3686/*
3687 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3688 * will not change in the lifetime of the guest.
3689 * Note that host-state that does change is set elsewhere. E.g., host-state
3690 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3691 */
3692static void vmx_set_constant_host_state(void)
3693{
3694	u32 low32, high32;
3695	unsigned long tmpl;
3696	struct desc_ptr dt;
3697
3698	vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS);  /* 22.2.3 */
3699	vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
3700	vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
3701
3702	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3703#ifdef CONFIG_X86_64
3704	/*
3705	 * Load null selectors, so we can avoid reloading them in
3706	 * __vmx_load_host_state(), in case userspace uses the null selectors
3707	 * too (the expected case).
3708	 */
3709	vmcs_write16(HOST_DS_SELECTOR, 0);
3710	vmcs_write16(HOST_ES_SELECTOR, 0);
3711#else
3712	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3713	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3714#endif
3715	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3716	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3717
3718	native_store_idt(&dt);
3719	vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
3720
3721	asm("mov $.Lkvm_vmx_return, %0" : "=r"(tmpl));
3722	vmcs_writel(HOST_RIP, tmpl); /* 22.2.5 */
3723
3724	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3725	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3726	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3727	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3728
3729	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3730		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3731		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3732	}
3733}
3734
3735static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3736{
3737	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3738	if (enable_ept)
3739		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3740	if (is_guest_mode(&vmx->vcpu))
3741		vmx->vcpu.arch.cr4_guest_owned_bits &=
3742			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3743	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3744}
3745
3746static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3747{
3748	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3749	if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3750		exec_control &= ~CPU_BASED_TPR_SHADOW;
3751#ifdef CONFIG_X86_64
3752		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3753				CPU_BASED_CR8_LOAD_EXITING;
3754#endif
3755	}
3756	if (!enable_ept)
3757		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3758				CPU_BASED_CR3_LOAD_EXITING  |
3759				CPU_BASED_INVLPG_EXITING;
3760	return exec_control;
3761}
3762
3763static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3764{
3765	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3766	if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3767		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3768	if (vmx->vpid == 0)
3769		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3770	if (!enable_ept) {
3771		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3772		enable_unrestricted_guest = 0;
3773	}
3774	if (!enable_unrestricted_guest)
3775		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3776	if (!ple_gap)
3777		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3778	return exec_control;
3779}
3780
3781static void ept_set_mmio_spte_mask(void)
3782{
3783	/*
3784	 * EPT Misconfigurations can be generated if the value of bits 2:0
3785	 * of an EPT paging-structure entry is 110b (write/execute).
3786	 * Also, magic bits (0xffull << 49) is set to quickly identify mmio
3787	 * spte.
3788	 */
3789	kvm_mmu_set_mmio_spte_mask(0xffull << 49 | 0x6ull);
3790}
3791
3792/*
3793 * Sets up the vmcs for emulated real mode.
3794 */
3795static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3796{
3797#ifdef CONFIG_X86_64
3798	unsigned long a;
3799#endif
3800	int i;
3801
3802	/* I/O */
3803	vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3804	vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
3805
3806	if (cpu_has_vmx_msr_bitmap())
3807		vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
3808
3809	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3810
3811	/* Control */
3812	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3813		vmcs_config.pin_based_exec_ctrl);
3814
3815	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
3816
3817	if (cpu_has_secondary_exec_ctrls()) {
3818		vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
3819				vmx_secondary_exec_control(vmx));
3820	}
3821
3822	if (ple_gap) {
3823		vmcs_write32(PLE_GAP, ple_gap);
3824		vmcs_write32(PLE_WINDOW, ple_window);
3825	}
3826
3827	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
3828	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
3829	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
3830
3831	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
3832	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
3833	vmx_set_constant_host_state();
3834#ifdef CONFIG_X86_64
3835	rdmsrl(MSR_FS_BASE, a);
3836	vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3837	rdmsrl(MSR_GS_BASE, a);
3838	vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3839#else
3840	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3841	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3842#endif
3843
3844	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3845	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3846	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
3847	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3848	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
3849
3850	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3851		u32 msr_low, msr_high;
3852		u64 host_pat;
3853		rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3854		host_pat = msr_low | ((u64) msr_high << 32);
3855		/* Write the default value follow host pat */
3856		vmcs_write64(GUEST_IA32_PAT, host_pat);
3857		/* Keep arch.pat sync with GUEST_IA32_PAT */
3858		vmx->vcpu.arch.pat = host_pat;
3859	}
3860
3861	for (i = 0; i < NR_VMX_MSR; ++i) {
3862		u32 index = vmx_msr_index[i];
3863		u32 data_low, data_high;
3864		int j = vmx->nmsrs;
3865
3866		if (rdmsr_safe(index, &data_low, &data_high) < 0)
3867			continue;
3868		if (wrmsr_safe(index, data_low, data_high) < 0)
3869			continue;
3870		vmx->guest_msrs[j].index = i;
3871		vmx->guest_msrs[j].data = 0;
3872		vmx->guest_msrs[j].mask = -1ull;
3873		++vmx->nmsrs;
3874	}
3875
3876	vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
3877
3878	/* 22.2.1, 20.8.1 */
3879	vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3880
3881	vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
3882	set_cr4_guest_host_mask(vmx);
3883
3884	kvm_write_tsc(&vmx->vcpu, 0);
3885
3886	return 0;
3887}
3888
3889static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3890{
3891	struct vcpu_vmx *vmx = to_vmx(vcpu);
3892	u64 msr;
3893	int ret;
3894
3895	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3896
3897	vmx->rmode.vm86_active = 0;
3898
3899	vmx->soft_vnmi_blocked = 0;
3900
3901	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
3902	kvm_set_cr8(&vmx->vcpu, 0);
3903	msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
3904	if (kvm_vcpu_is_bsp(&vmx->vcpu))
3905		msr |= MSR_IA32_APICBASE_BSP;
3906	kvm_set_apic_base(&vmx->vcpu, msr);
3907
3908	ret = fx_init(&vmx->vcpu);
3909	if (ret != 0)
3910		goto out;
3911
3912	vmx_segment_cache_clear(vmx);
3913
3914	seg_setup(VCPU_SREG_CS);
3915	/*
3916	 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3917	 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
3918	 */
3919	if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
3920		vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3921		vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3922	} else {
3923		vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3924		vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
3925	}
3926
3927	seg_setup(VCPU_SREG_DS);
3928	seg_setup(VCPU_SREG_ES);
3929	seg_setup(VCPU_SREG_FS);
3930	seg_setup(VCPU_SREG_GS);
3931	seg_setup(VCPU_SREG_SS);
3932
3933	vmcs_write16(GUEST_TR_SELECTOR, 0);
3934	vmcs_writel(GUEST_TR_BASE, 0);
3935	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3936	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3937
3938	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3939	vmcs_writel(GUEST_LDTR_BASE, 0);
3940	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3941	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3942
3943	vmcs_write32(GUEST_SYSENTER_CS, 0);
3944	vmcs_writel(GUEST_SYSENTER_ESP, 0);
3945	vmcs_writel(GUEST_SYSENTER_EIP, 0);
3946
3947	vmcs_writel(GUEST_RFLAGS, 0x02);
3948	if (kvm_vcpu_is_bsp(&vmx->vcpu))
3949		kvm_rip_write(vcpu, 0xfff0);
3950	else
3951		kvm_rip_write(vcpu, 0);
3952	kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
3953
3954	vmcs_writel(GUEST_DR7, 0x400);
3955
3956	vmcs_writel(GUEST_GDTR_BASE, 0);
3957	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3958
3959	vmcs_writel(GUEST_IDTR_BASE, 0);
3960	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3961
3962	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
3963	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3964	vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3965
3966	/* Special registers */
3967	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3968
3969	setup_msrs(vmx);
3970
3971	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
3972
3973	if (cpu_has_vmx_tpr_shadow()) {
3974		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3975		if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3976			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
3977				     __pa(vmx->vcpu.arch.apic->regs));
3978		vmcs_write32(TPR_THRESHOLD, 0);
3979	}
3980
3981	if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3982		vmcs_write64(APIC_ACCESS_ADDR,
3983			     page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
3984
3985	if (vmx->vpid != 0)
3986		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
3987
3988	vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
3989	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3990	vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
3991	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3992	vmx_set_cr4(&vmx->vcpu, 0);
3993	vmx_set_efer(&vmx->vcpu, 0);
3994	vmx_fpu_activate(&vmx->vcpu);
3995	update_exception_bitmap(&vmx->vcpu);
3996
3997	vpid_sync_context(vmx);
3998
3999	ret = 0;
4000
4001	/* HACK: Don't enable emulation on guest boot/reset */
4002	vmx->emulation_required = 0;
4003
4004out:
4005	return ret;
4006}
4007
4008/*
4009 * In nested virtualization, check if L1 asked to exit on external interrupts.
4010 * For most existing hypervisors, this will always return true.
4011 */
4012static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4013{
4014	return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4015		PIN_BASED_EXT_INTR_MASK;
4016}
4017
4018static void enable_irq_window(struct kvm_vcpu *vcpu)
4019{
4020	u32 cpu_based_vm_exec_control;
4021	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4022		/*
4023		 * We get here if vmx_interrupt_allowed() said we can't
4024		 * inject to L1 now because L2 must run. Ask L2 to exit
4025		 * right after entry, so we can inject to L1 more promptly.
4026		 */
4027		kvm_make_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
4028		return;
4029	}
4030
4031	cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4032	cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4033	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4034}
4035
4036static void enable_nmi_window(struct kvm_vcpu *vcpu)
4037{
4038	u32 cpu_based_vm_exec_control;
4039
4040	if (!cpu_has_virtual_nmis()) {
4041		enable_irq_window(vcpu);
4042		return;
4043	}
4044
4045	if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4046		enable_irq_window(vcpu);
4047		return;
4048	}
4049	cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4050	cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4051	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4052}
4053
4054static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4055{
4056	struct vcpu_vmx *vmx = to_vmx(vcpu);
4057	uint32_t intr;
4058	int irq = vcpu->arch.interrupt.nr;
4059
4060	trace_kvm_inj_virq(irq);
4061
4062	++vcpu->stat.irq_injections;
4063	if (vmx->rmode.vm86_active) {
4064		int inc_eip = 0;
4065		if (vcpu->arch.interrupt.soft)
4066			inc_eip = vcpu->arch.event_exit_inst_len;
4067		if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4068			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4069		return;
4070	}
4071	intr = irq | INTR_INFO_VALID_MASK;
4072	if (vcpu->arch.interrupt.soft) {
4073		intr |= INTR_TYPE_SOFT_INTR;
4074		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4075			     vmx->vcpu.arch.event_exit_inst_len);
4076	} else
4077		intr |= INTR_TYPE_EXT_INTR;
4078	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4079}
4080
4081static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4082{
4083	struct vcpu_vmx *vmx = to_vmx(vcpu);
4084
4085	if (is_guest_mode(vcpu))
4086		return;
4087
4088	if (!cpu_has_virtual_nmis()) {
4089		/*
4090		 * Tracking the NMI-blocked state in software is built upon
4091		 * finding the next open IRQ window. This, in turn, depends on
4092		 * well-behaving guests: They have to keep IRQs disabled at
4093		 * least as long as the NMI handler runs. Otherwise we may
4094		 * cause NMI nesting, maybe breaking the guest. But as this is
4095		 * highly unlikely, we can live with the residual risk.
4096		 */
4097		vmx->soft_vnmi_blocked = 1;
4098		vmx->vnmi_blocked_time = 0;
4099	}
4100
4101	++vcpu->stat.nmi_injections;
4102	vmx->nmi_known_unmasked = false;
4103	if (vmx->rmode.vm86_active) {
4104		if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4105			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4106		return;
4107	}
4108	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4109			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4110}
4111
4112static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4113{
4114	if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4115		return 0;
4116
4117	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4118		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4119		   | GUEST_INTR_STATE_NMI));
4120}
4121
4122static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4123{
4124	if (!cpu_has_virtual_nmis())
4125		return to_vmx(vcpu)->soft_vnmi_blocked;
4126	if (to_vmx(vcpu)->nmi_known_unmasked)
4127		return false;
4128	return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)	& GUEST_INTR_STATE_NMI;
4129}
4130
4131static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4132{
4133	struct vcpu_vmx *vmx = to_vmx(vcpu);
4134
4135	if (!cpu_has_virtual_nmis()) {
4136		if (vmx->soft_vnmi_blocked != masked) {
4137			vmx->soft_vnmi_blocked = masked;
4138			vmx->vnmi_blocked_time = 0;
4139		}
4140	} else {
4141		vmx->nmi_known_unmasked = !masked;
4142		if (masked)
4143			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4144				      GUEST_INTR_STATE_NMI);
4145		else
4146			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4147					GUEST_INTR_STATE_NMI);
4148	}
4149}
4150
4151static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4152{
4153	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4154		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4155		if (to_vmx(vcpu)->nested.nested_run_pending ||
4156		    (vmcs12->idt_vectoring_info_field &
4157		     VECTORING_INFO_VALID_MASK))
4158			return 0;
4159		nested_vmx_vmexit(vcpu);
4160		vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
4161		vmcs12->vm_exit_intr_info = 0;
4162		/* fall through to normal code, but now in L1, not L2 */
4163	}
4164
4165	return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4166		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4167			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4168}
4169
4170static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4171{
4172	int ret;
4173	struct kvm_userspace_memory_region tss_mem = {
4174		.slot = TSS_PRIVATE_MEMSLOT,
4175		.guest_phys_addr = addr,
4176		.memory_size = PAGE_SIZE * 3,
4177		.flags = 0,
4178	};
4179
4180	ret = kvm_set_memory_region(kvm, &tss_mem, 0);
4181	if (ret)
4182		return ret;
4183	kvm->arch.tss_addr = addr;
4184	if (!init_rmode_tss(kvm))
4185		return  -ENOMEM;
4186
4187	return 0;
4188}
4189
4190static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4191				  int vec, u32 err_code)
4192{
4193	/*
4194	 * Instruction with address size override prefix opcode 0x67
4195	 * Cause the #SS fault with 0 error code in VM86 mode.
4196	 */
4197	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
4198		if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
4199			return 1;
4200	/*
4201	 * Forward all other exceptions that are valid in real mode.
4202	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4203	 *        the required debugging infrastructure rework.
4204	 */
4205	switch (vec) {
4206	case DB_VECTOR:
4207		if (vcpu->guest_debug &
4208		    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4209			return 0;
4210		kvm_queue_exception(vcpu, vec);
4211		return 1;
4212	case BP_VECTOR:
4213		/*
4214		 * Update instruction length as we may reinject the exception
4215		 * from user space while in guest debugging mode.
4216		 */
4217		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4218			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4219		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4220			return 0;
4221		/* fall through */
4222	case DE_VECTOR:
4223	case OF_VECTOR:
4224	case BR_VECTOR:
4225	case UD_VECTOR:
4226	case DF_VECTOR:
4227	case SS_VECTOR:
4228	case GP_VECTOR:
4229	case MF_VECTOR:
4230		kvm_queue_exception(vcpu, vec);
4231		return 1;
4232	}
4233	return 0;
4234}
4235
4236/*
4237 * Trigger machine check on the host. We assume all the MSRs are already set up
4238 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4239 * We pass a fake environment to the machine check handler because we want
4240 * the guest to be always treated like user space, no matter what context
4241 * it used internally.
4242 */
4243static void kvm_machine_check(void)
4244{
4245#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4246	struct pt_regs regs = {
4247		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4248		.flags = X86_EFLAGS_IF,
4249	};
4250
4251	do_machine_check(&regs, 0);
4252#endif
4253}
4254
4255static int handle_machine_check(struct kvm_vcpu *vcpu)
4256{
4257	/* already handled by vcpu_run */
4258	return 1;
4259}
4260
4261static int handle_exception(struct kvm_vcpu *vcpu)
4262{
4263	struct vcpu_vmx *vmx = to_vmx(vcpu);
4264	struct kvm_run *kvm_run = vcpu->run;
4265	u32 intr_info, ex_no, error_code;
4266	unsigned long cr2, rip, dr6;
4267	u32 vect_info;
4268	enum emulation_result er;
4269
4270	vect_info = vmx->idt_vectoring_info;
4271	intr_info = vmx->exit_intr_info;
4272
4273	if (is_machine_check(intr_info))
4274		return handle_machine_check(vcpu);
4275
4276	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4277	    !is_page_fault(intr_info)) {
4278		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4279		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4280		vcpu->run->internal.ndata = 2;
4281		vcpu->run->internal.data[0] = vect_info;
4282		vcpu->run->internal.data[1] = intr_info;
4283		return 0;
4284	}
4285
4286	if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4287		return 1;  /* already handled by vmx_vcpu_run() */
4288
4289	if (is_no_device(intr_info)) {
4290		vmx_fpu_activate(vcpu);
4291		return 1;
4292	}
4293
4294	if (is_invalid_opcode(intr_info)) {
4295		er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4296		if (er != EMULATE_DONE)
4297			kvm_queue_exception(vcpu, UD_VECTOR);
4298		return 1;
4299	}
4300
4301	error_code = 0;
4302	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4303		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4304	if (is_page_fault(intr_info)) {
4305		/* EPT won't cause page fault directly */
4306		BUG_ON(enable_ept);
4307		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4308		trace_kvm_page_fault(cr2, error_code);
4309
4310		if (kvm_event_needs_reinjection(vcpu))
4311			kvm_mmu_unprotect_page_virt(vcpu, cr2);
4312		return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4313	}
4314
4315	if (vmx->rmode.vm86_active &&
4316	    handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
4317								error_code)) {
4318		if (vcpu->arch.halt_request) {
4319			vcpu->arch.halt_request = 0;
4320			return kvm_emulate_halt(vcpu);
4321		}
4322		return 1;
4323	}
4324
4325	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4326	switch (ex_no) {
4327	case DB_VECTOR:
4328		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4329		if (!(vcpu->guest_debug &
4330		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4331			vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4332			kvm_queue_exception(vcpu, DB_VECTOR);
4333			return 1;
4334		}
4335		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4336		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4337		/* fall through */
4338	case BP_VECTOR:
4339		/*
4340		 * Update instruction length as we may reinject #BP from
4341		 * user space while in guest debugging mode. Reading it for
4342		 * #DB as well causes no harm, it is not used in that case.
4343		 */
4344		vmx->vcpu.arch.event_exit_inst_len =
4345			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4346		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4347		rip = kvm_rip_read(vcpu);
4348		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4349		kvm_run->debug.arch.exception = ex_no;
4350		break;
4351	default:
4352		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4353		kvm_run->ex.exception = ex_no;
4354		kvm_run->ex.error_code = error_code;
4355		break;
4356	}
4357	return 0;
4358}
4359
4360static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4361{
4362	++vcpu->stat.irq_exits;
4363	return 1;
4364}
4365
4366static int handle_triple_fault(struct kvm_vcpu *vcpu)
4367{
4368	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4369	return 0;
4370}
4371
4372static int handle_io(struct kvm_vcpu *vcpu)
4373{
4374	unsigned long exit_qualification;
4375	int size, in, string;
4376	unsigned port;
4377
4378	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4379	string = (exit_qualification & 16) != 0;
4380	in = (exit_qualification & 8) != 0;
4381
4382	++vcpu->stat.io_exits;
4383
4384	if (string || in)
4385		return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4386
4387	port = exit_qualification >> 16;
4388	size = (exit_qualification & 7) + 1;
4389	skip_emulated_instruction(vcpu);
4390
4391	return kvm_fast_pio_out(vcpu, size, port);
4392}
4393
4394static void
4395vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4396{
4397	/*
4398	 * Patch in the VMCALL instruction:
4399	 */
4400	hypercall[0] = 0x0f;
4401	hypercall[1] = 0x01;
4402	hypercall[2] = 0xc1;
4403}
4404
4405/* called to set cr0 as approriate for a mov-to-cr0 exit. */
4406static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4407{
4408	if (to_vmx(vcpu)->nested.vmxon &&
4409	    ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4410		return 1;
4411
4412	if (is_guest_mode(vcpu)) {
4413		/*
4414		 * We get here when L2 changed cr0 in a way that did not change
4415		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4416		 * but did change L0 shadowed bits. This can currently happen
4417		 * with the TS bit: L0 may want to leave TS on (for lazy fpu
4418		 * loading) while pretending to allow the guest to change it.
4419		 */
4420		if (kvm_set_cr0(vcpu, (val & vcpu->arch.cr0_guest_owned_bits) |
4421			 (vcpu->arch.cr0 & ~vcpu->arch.cr0_guest_owned_bits)))
4422			return 1;
4423		vmcs_writel(CR0_READ_SHADOW, val);
4424		return 0;
4425	} else
4426		return kvm_set_cr0(vcpu, val);
4427}
4428
4429static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4430{
4431	if (is_guest_mode(vcpu)) {
4432		if (kvm_set_cr4(vcpu, (val & vcpu->arch.cr4_guest_owned_bits) |
4433			 (vcpu->arch.cr4 & ~vcpu->arch.cr4_guest_owned_bits)))
4434			return 1;
4435		vmcs_writel(CR4_READ_SHADOW, val);
4436		return 0;
4437	} else
4438		return kvm_set_cr4(vcpu, val);
4439}
4440
4441/* called to set cr0 as approriate for clts instruction exit. */
4442static void handle_clts(struct kvm_vcpu *vcpu)
4443{
4444	if (is_guest_mode(vcpu)) {
4445		/*
4446		 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4447		 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4448		 * just pretend it's off (also in arch.cr0 for fpu_activate).
4449		 */
4450		vmcs_writel(CR0_READ_SHADOW,
4451			vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4452		vcpu->arch.cr0 &= ~X86_CR0_TS;
4453	} else
4454		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4455}
4456
4457static int handle_cr(struct kvm_vcpu *vcpu)
4458{
4459	unsigned long exit_qualification, val;
4460	int cr;
4461	int reg;
4462	int err;
4463
4464	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4465	cr = exit_qualification & 15;
4466	reg = (exit_qualification >> 8) & 15;
4467	switch ((exit_qualification >> 4) & 3) {
4468	case 0: /* mov to cr */
4469		val = kvm_register_read(vcpu, reg);
4470		trace_kvm_cr_write(cr, val);
4471		switch (cr) {
4472		case 0:
4473			err = handle_set_cr0(vcpu, val);
4474			kvm_complete_insn_gp(vcpu, err);
4475			return 1;
4476		case 3:
4477			err = kvm_set_cr3(vcpu, val);
4478			kvm_complete_insn_gp(vcpu, err);
4479			return 1;
4480		case 4:
4481			err = handle_set_cr4(vcpu, val);
4482			kvm_complete_insn_gp(vcpu, err);
4483			return 1;
4484		case 8: {
4485				u8 cr8_prev = kvm_get_cr8(vcpu);
4486				u8 cr8 = kvm_register_read(vcpu, reg);
4487				err = kvm_set_cr8(vcpu, cr8);
4488				kvm_complete_insn_gp(vcpu, err);
4489				if (irqchip_in_kernel(vcpu->kvm))
4490					return 1;
4491				if (cr8_prev <= cr8)
4492					return 1;
4493				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4494				return 0;
4495			}
4496		};
4497		break;
4498	case 2: /* clts */
4499		handle_clts(vcpu);
4500		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4501		skip_emulated_instruction(vcpu);
4502		vmx_fpu_activate(vcpu);
4503		return 1;
4504	case 1: /*mov from cr*/
4505		switch (cr) {
4506		case 3:
4507			val = kvm_read_cr3(vcpu);
4508			kvm_register_write(vcpu, reg, val);
4509			trace_kvm_cr_read(cr, val);
4510			skip_emulated_instruction(vcpu);
4511			return 1;
4512		case 8:
4513			val = kvm_get_cr8(vcpu);
4514			kvm_register_write(vcpu, reg, val);
4515			trace_kvm_cr_read(cr, val);
4516			skip_emulated_instruction(vcpu);
4517			return 1;
4518		}
4519		break;
4520	case 3: /* lmsw */
4521		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4522		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4523		kvm_lmsw(vcpu, val);
4524
4525		skip_emulated_instruction(vcpu);
4526		return 1;
4527	default:
4528		break;
4529	}
4530	vcpu->run->exit_reason = 0;
4531	pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4532	       (int)(exit_qualification >> 4) & 3, cr);
4533	return 0;
4534}
4535
4536static int handle_dr(struct kvm_vcpu *vcpu)
4537{
4538	unsigned long exit_qualification;
4539	int dr, reg;
4540
4541	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4542	if (!kvm_require_cpl(vcpu, 0))
4543		return 1;
4544	dr = vmcs_readl(GUEST_DR7);
4545	if (dr & DR7_GD) {
4546		/*
4547		 * As the vm-exit takes precedence over the debug trap, we
4548		 * need to emulate the latter, either for the host or the
4549		 * guest debugging itself.
4550		 */
4551		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4552			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4553			vcpu->run->debug.arch.dr7 = dr;
4554			vcpu->run->debug.arch.pc =
4555				vmcs_readl(GUEST_CS_BASE) +
4556				vmcs_readl(GUEST_RIP);
4557			vcpu->run->debug.arch.exception = DB_VECTOR;
4558			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4559			return 0;
4560		} else {
4561			vcpu->arch.dr7 &= ~DR7_GD;
4562			vcpu->arch.dr6 |= DR6_BD;
4563			vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4564			kvm_queue_exception(vcpu, DB_VECTOR);
4565			return 1;
4566		}
4567	}
4568
4569	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4570	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4571	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4572	if (exit_qualification & TYPE_MOV_FROM_DR) {
4573		unsigned long val;
4574		if (!kvm_get_dr(vcpu, dr, &val))
4575			kvm_register_write(vcpu, reg, val);
4576	} else
4577		kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4578	skip_emulated_instruction(vcpu);
4579	return 1;
4580}
4581
4582static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4583{
4584	vmcs_writel(GUEST_DR7, val);
4585}
4586
4587static int handle_cpuid(struct kvm_vcpu *vcpu)
4588{
4589	kvm_emulate_cpuid(vcpu);
4590	return 1;
4591}
4592
4593static int handle_rdmsr(struct kvm_vcpu *vcpu)
4594{
4595	u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4596	u64 data;
4597
4598	if (vmx_get_msr(vcpu, ecx, &data)) {
4599		trace_kvm_msr_read_ex(ecx);
4600		kvm_inject_gp(vcpu, 0);
4601		return 1;
4602	}
4603
4604	trace_kvm_msr_read(ecx, data);
4605
4606	/* FIXME: handling of bits 32:63 of rax, rdx */
4607	vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4608	vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4609	skip_emulated_instruction(vcpu);
4610	return 1;
4611}
4612
4613static int handle_wrmsr(struct kvm_vcpu *vcpu)
4614{
4615	u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4616	u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4617		| ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4618
4619	if (vmx_set_msr(vcpu, ecx, data) != 0) {
4620		trace_kvm_msr_write_ex(ecx, data);
4621		kvm_inject_gp(vcpu, 0);
4622		return 1;
4623	}
4624
4625	trace_kvm_msr_write(ecx, data);
4626	skip_emulated_instruction(vcpu);
4627	return 1;
4628}
4629
4630static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4631{
4632	kvm_make_request(KVM_REQ_EVENT, vcpu);
4633	return 1;
4634}
4635
4636static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4637{
4638	u32 cpu_based_vm_exec_control;
4639
4640	/* clear pending irq */
4641	cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4642	cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4643	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4644
4645	kvm_make_request(KVM_REQ_EVENT, vcpu);
4646
4647	++vcpu->stat.irq_window_exits;
4648
4649	/*
4650	 * If the user space waits to inject interrupts, exit as soon as
4651	 * possible
4652	 */
4653	if (!irqchip_in_kernel(vcpu->kvm) &&
4654	    vcpu->run->request_interrupt_window &&
4655	    !kvm_cpu_has_interrupt(vcpu)) {
4656		vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4657		return 0;
4658	}
4659	return 1;
4660}
4661
4662static int handle_halt(struct kvm_vcpu *vcpu)
4663{
4664	skip_emulated_instruction(vcpu);
4665	return kvm_emulate_halt(vcpu);
4666}
4667
4668static int handle_vmcall(struct kvm_vcpu *vcpu)
4669{
4670	skip_emulated_instruction(vcpu);
4671	kvm_emulate_hypercall(vcpu);
4672	return 1;
4673}
4674
4675static int handle_invd(struct kvm_vcpu *vcpu)
4676{
4677	return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4678}
4679
4680static int handle_invlpg(struct kvm_vcpu *vcpu)
4681{
4682	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4683
4684	kvm_mmu_invlpg(vcpu, exit_qualification);
4685	skip_emulated_instruction(vcpu);
4686	return 1;
4687}
4688
4689static int handle_rdpmc(struct kvm_vcpu *vcpu)
4690{
4691	int err;
4692
4693	err = kvm_rdpmc(vcpu);
4694	kvm_complete_insn_gp(vcpu, err);
4695
4696	return 1;
4697}
4698
4699static int handle_wbinvd(struct kvm_vcpu *vcpu)
4700{
4701	skip_emulated_instruction(vcpu);
4702	kvm_emulate_wbinvd(vcpu);
4703	return 1;
4704}
4705
4706static int handle_xsetbv(struct kvm_vcpu *vcpu)
4707{
4708	u64 new_bv = kvm_read_edx_eax(vcpu);
4709	u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4710
4711	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4712		skip_emulated_instruction(vcpu);
4713	return 1;
4714}
4715
4716static int handle_apic_access(struct kvm_vcpu *vcpu)
4717{
4718	if (likely(fasteoi)) {
4719		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4720		int access_type, offset;
4721
4722		access_type = exit_qualification & APIC_ACCESS_TYPE;
4723		offset = exit_qualification & APIC_ACCESS_OFFSET;
4724		/*
4725		 * Sane guest uses MOV to write EOI, with written value
4726		 * not cared. So make a short-circuit here by avoiding
4727		 * heavy instruction emulation.
4728		 */
4729		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4730		    (offset == APIC_EOI)) {
4731			kvm_lapic_set_eoi(vcpu);
4732			skip_emulated_instruction(vcpu);
4733			return 1;
4734		}
4735	}
4736	return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4737}
4738
4739static int handle_task_switch(struct kvm_vcpu *vcpu)
4740{
4741	struct vcpu_vmx *vmx = to_vmx(vcpu);
4742	unsigned long exit_qualification;
4743	bool has_error_code = false;
4744	u32 error_code = 0;
4745	u16 tss_selector;
4746	int reason, type, idt_v, idt_index;
4747
4748	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4749	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
4750	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4751
4752	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4753
4754	reason = (u32)exit_qualification >> 30;
4755	if (reason == TASK_SWITCH_GATE && idt_v) {
4756		switch (type) {
4757		case INTR_TYPE_NMI_INTR:
4758			vcpu->arch.nmi_injected = false;
4759			vmx_set_nmi_mask(vcpu, true);
4760			break;
4761		case INTR_TYPE_EXT_INTR:
4762		case INTR_TYPE_SOFT_INTR:
4763			kvm_clear_interrupt_queue(vcpu);
4764			break;
4765		case INTR_TYPE_HARD_EXCEPTION:
4766			if (vmx->idt_vectoring_info &
4767			    VECTORING_INFO_DELIVER_CODE_MASK) {
4768				has_error_code = true;
4769				error_code =
4770					vmcs_read32(IDT_VECTORING_ERROR_CODE);
4771			}
4772			/* fall through */
4773		case INTR_TYPE_SOFT_EXCEPTION:
4774			kvm_clear_exception_queue(vcpu);
4775			break;
4776		default:
4777			break;
4778		}
4779	}
4780	tss_selector = exit_qualification;
4781
4782	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4783		       type != INTR_TYPE_EXT_INTR &&
4784		       type != INTR_TYPE_NMI_INTR))
4785		skip_emulated_instruction(vcpu);
4786
4787	if (kvm_task_switch(vcpu, tss_selector,
4788			    type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
4789			    has_error_code, error_code) == EMULATE_FAIL) {
4790		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4791		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4792		vcpu->run->internal.ndata = 0;
4793		return 0;
4794	}
4795
4796	/* clear all local breakpoint enable flags */
4797	vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4798
4799	/*
4800	 * TODO: What about debug traps on tss switch?
4801	 *       Are we supposed to inject them and update dr6?
4802	 */
4803
4804	return 1;
4805}
4806
4807static int handle_ept_violation(struct kvm_vcpu *vcpu)
4808{
4809	unsigned long exit_qualification;
4810	gpa_t gpa;
4811	int gla_validity;
4812
4813	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4814
4815	if (exit_qualification & (1 << 6)) {
4816		printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
4817		return -EINVAL;
4818	}
4819
4820	gla_validity = (exit_qualification >> 7) & 0x3;
4821	if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4822		printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4823		printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4824			(long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
4825			vmcs_readl(GUEST_LINEAR_ADDRESS));
4826		printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4827			(long unsigned int)exit_qualification);
4828		vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4829		vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
4830		return 0;
4831	}
4832
4833	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4834	trace_kvm_page_fault(gpa, exit_qualification);
4835	return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
4836}
4837
4838static u64 ept_rsvd_mask(u64 spte, int level)
4839{
4840	int i;
4841	u64 mask = 0;
4842
4843	for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4844		mask |= (1ULL << i);
4845
4846	if (level > 2)
4847		/* bits 7:3 reserved */
4848		mask |= 0xf8;
4849	else if (level == 2) {
4850		if (spte & (1ULL << 7))
4851			/* 2MB ref, bits 20:12 reserved */
4852			mask |= 0x1ff000;
4853		else
4854			/* bits 6:3 reserved */
4855			mask |= 0x78;
4856	}
4857
4858	return mask;
4859}
4860
4861static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4862				       int level)
4863{
4864	printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4865
4866	/* 010b (write-only) */
4867	WARN_ON((spte & 0x7) == 0x2);
4868
4869	/* 110b (write/execute) */
4870	WARN_ON((spte & 0x7) == 0x6);
4871
4872	/* 100b (execute-only) and value not supported by logical processor */
4873	if (!cpu_has_vmx_ept_execute_only())
4874		WARN_ON((spte & 0x7) == 0x4);
4875
4876	/* not 000b */
4877	if ((spte & 0x7)) {
4878		u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4879
4880		if (rsvd_bits != 0) {
4881			printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4882					 __func__, rsvd_bits);
4883			WARN_ON(1);
4884		}
4885
4886		if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4887			u64 ept_mem_type = (spte & 0x38) >> 3;
4888
4889			if (ept_mem_type == 2 || ept_mem_type == 3 ||
4890			    ept_mem_type == 7) {
4891				printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4892						__func__, ept_mem_type);
4893				WARN_ON(1);
4894			}
4895		}
4896	}
4897}
4898
4899static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
4900{
4901	u64 sptes[4];
4902	int nr_sptes, i, ret;
4903	gpa_t gpa;
4904
4905	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4906
4907	ret = handle_mmio_page_fault_common(vcpu, gpa, true);
4908	if (likely(ret == 1))
4909		return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
4910					      EMULATE_DONE;
4911	if (unlikely(!ret))
4912		return 1;
4913
4914	/* It is the real ept misconfig */
4915	printk(KERN_ERR "EPT: Misconfiguration.\n");
4916	printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4917
4918	nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4919
4920	for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4921		ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4922
4923	vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4924	vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
4925
4926	return 0;
4927}
4928
4929static int handle_nmi_window(struct kvm_vcpu *vcpu)
4930{
4931	u32 cpu_based_vm_exec_control;
4932
4933	/* clear pending NMI */
4934	cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4935	cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4936	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4937	++vcpu->stat.nmi_window_exits;
4938	kvm_make_request(KVM_REQ_EVENT, vcpu);
4939
4940	return 1;
4941}
4942
4943static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
4944{
4945	struct vcpu_vmx *vmx = to_vmx(vcpu);
4946	enum emulation_result err = EMULATE_DONE;
4947	int ret = 1;
4948	u32 cpu_exec_ctrl;
4949	bool intr_window_requested;
4950
4951	cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4952	intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
4953
4954	while (!guest_state_valid(vcpu)) {
4955		if (intr_window_requested
4956		    && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
4957			return handle_interrupt_window(&vmx->vcpu);
4958
4959		err = emulate_instruction(vcpu, 0);
4960
4961		if (err == EMULATE_DO_MMIO) {
4962			ret = 0;
4963			goto out;
4964		}
4965
4966		if (err != EMULATE_DONE)
4967			return 0;
4968
4969		if (signal_pending(current))
4970			goto out;
4971		if (need_resched())
4972			schedule();
4973	}
4974
4975	vmx->emulation_required = 0;
4976out:
4977	return ret;
4978}
4979
4980/*
4981 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
4982 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
4983 */
4984static int handle_pause(struct kvm_vcpu *vcpu)
4985{
4986	skip_emulated_instruction(vcpu);
4987	kvm_vcpu_on_spin(vcpu);
4988
4989	return 1;
4990}
4991
4992static int handle_invalid_op(struct kvm_vcpu *vcpu)
4993{
4994	kvm_queue_exception(vcpu, UD_VECTOR);
4995	return 1;
4996}
4997
4998/*
4999 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5000 * We could reuse a single VMCS for all the L2 guests, but we also want the
5001 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5002 * allows keeping them loaded on the processor, and in the future will allow
5003 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5004 * every entry if they never change.
5005 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5006 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5007 *
5008 * The following functions allocate and free a vmcs02 in this pool.
5009 */
5010
5011/* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5012static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5013{
5014	struct vmcs02_list *item;
5015	list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5016		if (item->vmptr == vmx->nested.current_vmptr) {
5017			list_move(&item->list, &vmx->nested.vmcs02_pool);
5018			return &item->vmcs02;
5019		}
5020
5021	if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5022		/* Recycle the least recently used VMCS. */
5023		item = list_entry(vmx->nested.vmcs02_pool.prev,
5024			struct vmcs02_list, list);
5025		item->vmptr = vmx->nested.current_vmptr;
5026		list_move(&item->list, &vmx->nested.vmcs02_pool);
5027		return &item->vmcs02;
5028	}
5029
5030	/* Create a new VMCS */
5031	item = (struct vmcs02_list *)
5032		kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5033	if (!item)
5034		return NULL;
5035	item->vmcs02.vmcs = alloc_vmcs();
5036	if (!item->vmcs02.vmcs) {
5037		kfree(item);
5038		return NULL;
5039	}
5040	loaded_vmcs_init(&item->vmcs02);
5041	item->vmptr = vmx->nested.current_vmptr;
5042	list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5043	vmx->nested.vmcs02_num++;
5044	return &item->vmcs02;
5045}
5046
5047/* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5048static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5049{
5050	struct vmcs02_list *item;
5051	list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5052		if (item->vmptr == vmptr) {
5053			free_loaded_vmcs(&item->vmcs02);
5054			list_del(&item->list);
5055			kfree(item);
5056			vmx->nested.vmcs02_num--;
5057			return;
5058		}
5059}
5060
5061/*
5062 * Free all VMCSs saved for this vcpu, except the one pointed by
5063 * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5064 * currently used, if running L2), and vmcs01 when running L2.
5065 */
5066static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5067{
5068	struct vmcs02_list *item, *n;
5069	list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5070		if (vmx->loaded_vmcs != &item->vmcs02)
5071			free_loaded_vmcs(&item->vmcs02);
5072		list_del(&item->list);
5073		kfree(item);
5074	}
5075	vmx->nested.vmcs02_num = 0;
5076
5077	if (vmx->loaded_vmcs != &vmx->vmcs01)
5078		free_loaded_vmcs(&vmx->vmcs01);
5079}
5080
5081/*
5082 * Emulate the VMXON instruction.
5083 * Currently, we just remember that VMX is active, and do not save or even
5084 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5085 * do not currently need to store anything in that guest-allocated memory
5086 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5087 * argument is different from the VMXON pointer (which the spec says they do).
5088 */
5089static int handle_vmon(struct kvm_vcpu *vcpu)
5090{
5091	struct kvm_segment cs;
5092	struct vcpu_vmx *vmx = to_vmx(vcpu);
5093
5094	/* The Intel VMX Instruction Reference lists a bunch of bits that
5095	 * are prerequisite to running VMXON, most notably cr4.VMXE must be
5096	 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5097	 * Otherwise, we should fail with #UD. We test these now:
5098	 */
5099	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5100	    !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5101	    (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5102		kvm_queue_exception(vcpu, UD_VECTOR);
5103		return 1;
5104	}
5105
5106	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5107	if (is_long_mode(vcpu) && !cs.l) {
5108		kvm_queue_exception(vcpu, UD_VECTOR);
5109		return 1;
5110	}
5111
5112	if (vmx_get_cpl(vcpu)) {
5113		kvm_inject_gp(vcpu, 0);
5114		return 1;
5115	}
5116
5117	INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5118	vmx->nested.vmcs02_num = 0;
5119
5120	vmx->nested.vmxon = true;
5121
5122	skip_emulated_instruction(vcpu);
5123	return 1;
5124}
5125
5126/*
5127 * Intel's VMX Instruction Reference specifies a common set of prerequisites
5128 * for running VMX instructions (except VMXON, whose prerequisites are
5129 * slightly different). It also specifies what exception to inject otherwise.
5130 */
5131static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5132{
5133	struct kvm_segment cs;
5134	struct vcpu_vmx *vmx = to_vmx(vcpu);
5135
5136	if (!vmx->nested.vmxon) {
5137		kvm_queue_exception(vcpu, UD_VECTOR);
5138		return 0;
5139	}
5140
5141	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5142	if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5143	    (is_long_mode(vcpu) && !cs.l)) {
5144		kvm_queue_exception(vcpu, UD_VECTOR);
5145		return 0;
5146	}
5147
5148	if (vmx_get_cpl(vcpu)) {
5149		kvm_inject_gp(vcpu, 0);
5150		return 0;
5151	}
5152
5153	return 1;
5154}
5155
5156/*
5157 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5158 * just stops using VMX.
5159 */
5160static void free_nested(struct vcpu_vmx *vmx)
5161{
5162	if (!vmx->nested.vmxon)
5163		return;
5164	vmx->nested.vmxon = false;
5165	if (vmx->nested.current_vmptr != -1ull) {
5166		kunmap(vmx->nested.current_vmcs12_page);
5167		nested_release_page(vmx->nested.current_vmcs12_page);
5168		vmx->nested.current_vmptr = -1ull;
5169		vmx->nested.current_vmcs12 = NULL;
5170	}
5171	/* Unpin physical memory we referred to in current vmcs02 */
5172	if (vmx->nested.apic_access_page) {
5173		nested_release_page(vmx->nested.apic_access_page);
5174		vmx->nested.apic_access_page = 0;
5175	}
5176
5177	nested_free_all_saved_vmcss(vmx);
5178}
5179
5180/* Emulate the VMXOFF instruction */
5181static int handle_vmoff(struct kvm_vcpu *vcpu)
5182{
5183	if (!nested_vmx_check_permission(vcpu))
5184		return 1;
5185	free_nested(to_vmx(vcpu));
5186	skip_emulated_instruction(vcpu);
5187	return 1;
5188}
5189
5190/*
5191 * Decode the memory-address operand of a vmx instruction, as recorded on an
5192 * exit caused by such an instruction (run by a guest hypervisor).
5193 * On success, returns 0. When the operand is invalid, returns 1 and throws
5194 * #UD or #GP.
5195 */
5196static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5197				 unsigned long exit_qualification,
5198				 u32 vmx_instruction_info, gva_t *ret)
5199{
5200	/*
5201	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5202	 * Execution", on an exit, vmx_instruction_info holds most of the
5203	 * addressing components of the operand. Only the displacement part
5204	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5205	 * For how an actual address is calculated from all these components,
5206	 * refer to Vol. 1, "Operand Addressing".
5207	 */
5208	int  scaling = vmx_instruction_info & 3;
5209	int  addr_size = (vmx_instruction_info >> 7) & 7;
5210	bool is_reg = vmx_instruction_info & (1u << 10);
5211	int  seg_reg = (vmx_instruction_info >> 15) & 7;
5212	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
5213	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5214	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
5215	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
5216
5217	if (is_reg) {
5218		kvm_queue_exception(vcpu, UD_VECTOR);
5219		return 1;
5220	}
5221
5222	/* Addr = segment_base + offset */
5223	/* offset = base + [index * scale] + displacement */
5224	*ret = vmx_get_segment_base(vcpu, seg_reg);
5225	if (base_is_valid)
5226		*ret += kvm_register_read(vcpu, base_reg);
5227	if (index_is_valid)
5228		*ret += kvm_register_read(vcpu, index_reg)<<scaling;
5229	*ret += exit_qualification; /* holds the displacement */
5230
5231	if (addr_size == 1) /* 32 bit */
5232		*ret &= 0xffffffff;
5233
5234	/*
5235	 * TODO: throw #GP (and return 1) in various cases that the VM*
5236	 * instructions require it - e.g., offset beyond segment limit,
5237	 * unusable or unreadable/unwritable segment, non-canonical 64-bit
5238	 * address, and so on. Currently these are not checked.
5239	 */
5240	return 0;
5241}
5242
5243/*
5244 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5245 * set the success or error code of an emulated VMX instruction, as specified
5246 * by Vol 2B, VMX Instruction Reference, "Conventions".
5247 */
5248static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5249{
5250	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5251			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5252			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5253}
5254
5255static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5256{
5257	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5258			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5259			    X86_EFLAGS_SF | X86_EFLAGS_OF))
5260			| X86_EFLAGS_CF);
5261}
5262
5263static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5264					u32 vm_instruction_error)
5265{
5266	if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5267		/*
5268		 * failValid writes the error number to the current VMCS, which
5269		 * can't be done there isn't a current VMCS.
5270		 */
5271		nested_vmx_failInvalid(vcpu);
5272		return;
5273	}
5274	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5275			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5276			    X86_EFLAGS_SF | X86_EFLAGS_OF))
5277			| X86_EFLAGS_ZF);
5278	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5279}
5280
5281/* Emulate the VMCLEAR instruction */
5282static int handle_vmclear(struct kvm_vcpu *vcpu)
5283{
5284	struct vcpu_vmx *vmx = to_vmx(vcpu);
5285	gva_t gva;
5286	gpa_t vmptr;
5287	struct vmcs12 *vmcs12;
5288	struct page *page;
5289	struct x86_exception e;
5290
5291	if (!nested_vmx_check_permission(vcpu))
5292		return 1;
5293
5294	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5295			vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5296		return 1;
5297
5298	if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5299				sizeof(vmptr), &e)) {
5300		kvm_inject_page_fault(vcpu, &e);
5301		return 1;
5302	}
5303
5304	if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5305		nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5306		skip_emulated_instruction(vcpu);
5307		return 1;
5308	}
5309
5310	if (vmptr == vmx->nested.current_vmptr) {
5311		kunmap(vmx->nested.current_vmcs12_page);
5312		nested_release_page(vmx->nested.current_vmcs12_page);
5313		vmx->nested.current_vmptr = -1ull;
5314		vmx->nested.current_vmcs12 = NULL;
5315	}
5316
5317	page = nested_get_page(vcpu, vmptr);
5318	if (page == NULL) {
5319		/*
5320		 * For accurate processor emulation, VMCLEAR beyond available
5321		 * physical memory should do nothing at all. However, it is
5322		 * possible that a nested vmx bug, not a guest hypervisor bug,
5323		 * resulted in this case, so let's shut down before doing any
5324		 * more damage:
5325		 */
5326		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5327		return 1;
5328	}
5329	vmcs12 = kmap(page);
5330	vmcs12->launch_state = 0;
5331	kunmap(page);
5332	nested_release_page(page);
5333
5334	nested_free_vmcs02(vmx, vmptr);
5335
5336	skip_emulated_instruction(vcpu);
5337	nested_vmx_succeed(vcpu);
5338	return 1;
5339}
5340
5341static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5342
5343/* Emulate the VMLAUNCH instruction */
5344static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5345{
5346	return nested_vmx_run(vcpu, true);
5347}
5348
5349/* Emulate the VMRESUME instruction */
5350static int handle_vmresume(struct kvm_vcpu *vcpu)
5351{
5352
5353	return nested_vmx_run(vcpu, false);
5354}
5355
5356enum vmcs_field_type {
5357	VMCS_FIELD_TYPE_U16 = 0,
5358	VMCS_FIELD_TYPE_U64 = 1,
5359	VMCS_FIELD_TYPE_U32 = 2,
5360	VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5361};
5362
5363static inline int vmcs_field_type(unsigned long field)
5364{
5365	if (0x1 & field)	/* the *_HIGH fields are all 32 bit */
5366		return VMCS_FIELD_TYPE_U32;
5367	return (field >> 13) & 0x3 ;
5368}
5369
5370static inline int vmcs_field_readonly(unsigned long field)
5371{
5372	return (((field >> 10) & 0x3) == 1);
5373}
5374
5375/*
5376 * Read a vmcs12 field. Since these can have varying lengths and we return
5377 * one type, we chose the biggest type (u64) and zero-extend the return value
5378 * to that size. Note that the caller, handle_vmread, might need to use only
5379 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5380 * 64-bit fields are to be returned).
5381 */
5382static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5383					unsigned long field, u64 *ret)
5384{
5385	short offset = vmcs_field_to_offset(field);
5386	char *p;
5387
5388	if (offset < 0)
5389		return 0;
5390
5391	p = ((char *)(get_vmcs12(vcpu))) + offset;
5392
5393	switch (vmcs_field_type(field)) {
5394	case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5395		*ret = *((natural_width *)p);
5396		return 1;
5397	case VMCS_FIELD_TYPE_U16:
5398		*ret = *((u16 *)p);
5399		return 1;
5400	case VMCS_FIELD_TYPE_U32:
5401		*ret = *((u32 *)p);
5402		return 1;
5403	case VMCS_FIELD_TYPE_U64:
5404		*ret = *((u64 *)p);
5405		return 1;
5406	default:
5407		return 0; /* can never happen. */
5408	}
5409}
5410
5411/*
5412 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5413 * used before) all generate the same failure when it is missing.
5414 */
5415static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5416{
5417	struct vcpu_vmx *vmx = to_vmx(vcpu);
5418	if (vmx->nested.current_vmptr == -1ull) {
5419		nested_vmx_failInvalid(vcpu);
5420		skip_emulated_instruction(vcpu);
5421		return 0;
5422	}
5423	return 1;
5424}
5425
5426static int handle_vmread(struct kvm_vcpu *vcpu)
5427{
5428	unsigned long field;
5429	u64 field_value;
5430	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5431	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5432	gva_t gva = 0;
5433
5434	if (!nested_vmx_check_permission(vcpu) ||
5435	    !nested_vmx_check_vmcs12(vcpu))
5436		return 1;
5437
5438	/* Decode instruction info and find the field to read */
5439	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5440	/* Read the field, zero-extended to a u64 field_value */
5441	if (!vmcs12_read_any(vcpu, field, &field_value)) {
5442		nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5443		skip_emulated_instruction(vcpu);
5444		return 1;
5445	}
5446	/*
5447	 * Now copy part of this value to register or memory, as requested.
5448	 * Note that the number of bits actually copied is 32 or 64 depending
5449	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5450	 */
5451	if (vmx_instruction_info & (1u << 10)) {
5452		kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5453			field_value);
5454	} else {
5455		if (get_vmx_mem_address(vcpu, exit_qualification,
5456				vmx_instruction_info, &gva))
5457			return 1;
5458		/* _system ok, as nested_vmx_check_permission verified cpl=0 */
5459		kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5460			     &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5461	}
5462
5463	nested_vmx_succeed(vcpu);
5464	skip_emulated_instruction(vcpu);
5465	return 1;
5466}
5467
5468
5469static int handle_vmwrite(struct kvm_vcpu *vcpu)
5470{
5471	unsigned long field;
5472	gva_t gva;
5473	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5474	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5475	char *p;
5476	short offset;
5477	/* The value to write might be 32 or 64 bits, depending on L1's long
5478	 * mode, and eventually we need to write that into a field of several
5479	 * possible lengths. The code below first zero-extends the value to 64
5480	 * bit (field_value), and then copies only the approriate number of
5481	 * bits into the vmcs12 field.
5482	 */
5483	u64 field_value = 0;
5484	struct x86_exception e;
5485
5486	if (!nested_vmx_check_permission(vcpu) ||
5487	    !nested_vmx_check_vmcs12(vcpu))
5488		return 1;
5489
5490	if (vmx_instruction_info & (1u << 10))
5491		field_value = kvm_register_read(vcpu,
5492			(((vmx_instruction_info) >> 3) & 0xf));
5493	else {
5494		if (get_vmx_mem_address(vcpu, exit_qualification,
5495				vmx_instruction_info, &gva))
5496			return 1;
5497		if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5498			   &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5499			kvm_inject_page_fault(vcpu, &e);
5500			return 1;
5501		}
5502	}
5503
5504
5505	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5506	if (vmcs_field_readonly(field)) {
5507		nested_vmx_failValid(vcpu,
5508			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5509		skip_emulated_instruction(vcpu);
5510		return 1;
5511	}
5512
5513	offset = vmcs_field_to_offset(field);
5514	if (offset < 0) {
5515		nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5516		skip_emulated_instruction(vcpu);
5517		return 1;
5518	}
5519	p = ((char *) get_vmcs12(vcpu)) + offset;
5520
5521	switch (vmcs_field_type(field)) {
5522	case VMCS_FIELD_TYPE_U16:
5523		*(u16 *)p = field_value;
5524		break;
5525	case VMCS_FIELD_TYPE_U32:
5526		*(u32 *)p = field_value;
5527		break;
5528	case VMCS_FIELD_TYPE_U64:
5529		*(u64 *)p = field_value;
5530		break;
5531	case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5532		*(natural_width *)p = field_value;
5533		break;
5534	default:
5535		nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5536		skip_emulated_instruction(vcpu);
5537		return 1;
5538	}
5539
5540	nested_vmx_succeed(vcpu);
5541	skip_emulated_instruction(vcpu);
5542	return 1;
5543}
5544
5545/* Emulate the VMPTRLD instruction */
5546static int handle_vmptrld(struct kvm_vcpu *vcpu)
5547{
5548	struct vcpu_vmx *vmx = to_vmx(vcpu);
5549	gva_t gva;
5550	gpa_t vmptr;
5551	struct x86_exception e;
5552
5553	if (!nested_vmx_check_permission(vcpu))
5554		return 1;
5555
5556	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5557			vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5558		return 1;
5559
5560	if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5561				sizeof(vmptr), &e)) {
5562		kvm_inject_page_fault(vcpu, &e);
5563		return 1;
5564	}
5565
5566	if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5567		nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5568		skip_emulated_instruction(vcpu);
5569		return 1;
5570	}
5571
5572	if (vmx->nested.current_vmptr != vmptr) {
5573		struct vmcs12 *new_vmcs12;
5574		struct page *page;
5575		page = nested_get_page(vcpu, vmptr);
5576		if (page == NULL) {
5577			nested_vmx_failInvalid(vcpu);
5578			skip_emulated_instruction(vcpu);
5579			return 1;
5580		}
5581		new_vmcs12 = kmap(page);
5582		if (new_vmcs12->revision_id != VMCS12_REVISION) {
5583			kunmap(page);
5584			nested_release_page_clean(page);
5585			nested_vmx_failValid(vcpu,
5586				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5587			skip_emulated_instruction(vcpu);
5588			return 1;
5589		}
5590		if (vmx->nested.current_vmptr != -1ull) {
5591			kunmap(vmx->nested.current_vmcs12_page);
5592			nested_release_page(vmx->nested.current_vmcs12_page);
5593		}
5594
5595		vmx->nested.current_vmptr = vmptr;
5596		vmx->nested.current_vmcs12 = new_vmcs12;
5597		vmx->nested.current_vmcs12_page = page;
5598	}
5599
5600	nested_vmx_succeed(vcpu);
5601	skip_emulated_instruction(vcpu);
5602	return 1;
5603}
5604
5605/* Emulate the VMPTRST instruction */
5606static int handle_vmptrst(struct kvm_vcpu *vcpu)
5607{
5608	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5609	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5610	gva_t vmcs_gva;
5611	struct x86_exception e;
5612
5613	if (!nested_vmx_check_permission(vcpu))
5614		return 1;
5615
5616	if (get_vmx_mem_address(vcpu, exit_qualification,
5617			vmx_instruction_info, &vmcs_gva))
5618		return 1;
5619	/* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5620	if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5621				 (void *)&to_vmx(vcpu)->nested.current_vmptr,
5622				 sizeof(u64), &e)) {
5623		kvm_inject_page_fault(vcpu, &e);
5624		return 1;
5625	}
5626	nested_vmx_succeed(vcpu);
5627	skip_emulated_instruction(vcpu);
5628	return 1;
5629}
5630
5631/*
5632 * The exit handlers return 1 if the exit was handled fully and guest execution
5633 * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5634 * to be done to userspace and return 0.
5635 */
5636static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5637	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
5638	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5639	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5640	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5641	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5642	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5643	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5644	[EXIT_REASON_CPUID]                   = handle_cpuid,
5645	[EXIT_REASON_MSR_READ]                = handle_rdmsr,
5646	[EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5647	[EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5648	[EXIT_REASON_HLT]                     = handle_halt,
5649	[EXIT_REASON_INVD]		      = handle_invd,
5650	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5651	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5652	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5653	[EXIT_REASON_VMCLEAR]	              = handle_vmclear,
5654	[EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
5655	[EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
5656	[EXIT_REASON_VMPTRST]                 = handle_vmptrst,
5657	[EXIT_REASON_VMREAD]                  = handle_vmread,
5658	[EXIT_REASON_VMRESUME]                = handle_vmresume,
5659	[EXIT_REASON_VMWRITE]                 = handle_vmwrite,
5660	[EXIT_REASON_VMOFF]                   = handle_vmoff,
5661	[EXIT_REASON_VMON]                    = handle_vmon,
5662	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5663	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5664	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5665	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5666	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5667	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5668	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5669	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5670	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5671	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_invalid_op,
5672	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
5673};
5674
5675static const int kvm_vmx_max_exit_handlers =
5676	ARRAY_SIZE(kvm_vmx_exit_handlers);
5677
5678/*
5679 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5680 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5681 * disinterest in the current event (read or write a specific MSR) by using an
5682 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5683 */
5684static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5685	struct vmcs12 *vmcs12, u32 exit_reason)
5686{
5687	u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
5688	gpa_t bitmap;
5689
5690	if (!nested_cpu_has(get_vmcs12(vcpu), CPU_BASED_USE_MSR_BITMAPS))
5691		return 1;
5692
5693	/*
5694	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5695	 * for the four combinations of read/write and low/high MSR numbers.
5696	 * First we need to figure out which of the four to use:
5697	 */
5698	bitmap = vmcs12->msr_bitmap;
5699	if (exit_reason == EXIT_REASON_MSR_WRITE)
5700		bitmap += 2048;
5701	if (msr_index >= 0xc0000000) {
5702		msr_index -= 0xc0000000;
5703		bitmap += 1024;
5704	}
5705
5706	/* Then read the msr_index'th bit from this bitmap: */
5707	if (msr_index < 1024*8) {
5708		unsigned char b;
5709		kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1);
5710		return 1 & (b >> (msr_index & 7));
5711	} else
5712		return 1; /* let L1 handle the wrong parameter */
5713}
5714
5715/*
5716 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5717 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5718 * intercept (via guest_host_mask etc.) the current event.
5719 */
5720static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5721	struct vmcs12 *vmcs12)
5722{
5723	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5724	int cr = exit_qualification & 15;
5725	int reg = (exit_qualification >> 8) & 15;
5726	unsigned long val = kvm_register_read(vcpu, reg);
5727
5728	switch ((exit_qualification >> 4) & 3) {
5729	case 0: /* mov to cr */
5730		switch (cr) {
5731		case 0:
5732			if (vmcs12->cr0_guest_host_mask &
5733			    (val ^ vmcs12->cr0_read_shadow))
5734				return 1;
5735			break;
5736		case 3:
5737			if ((vmcs12->cr3_target_count >= 1 &&
5738					vmcs12->cr3_target_value0 == val) ||
5739				(vmcs12->cr3_target_count >= 2 &&
5740					vmcs12->cr3_target_value1 == val) ||
5741				(vmcs12->cr3_target_count >= 3 &&
5742					vmcs12->cr3_target_value2 == val) ||
5743				(vmcs12->cr3_target_count >= 4 &&
5744					vmcs12->cr3_target_value3 == val))
5745				return 0;
5746			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5747				return 1;
5748			break;
5749		case 4:
5750			if (vmcs12->cr4_guest_host_mask &
5751			    (vmcs12->cr4_read_shadow ^ val))
5752				return 1;
5753			break;
5754		case 8:
5755			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5756				return 1;
5757			break;
5758		}
5759		break;
5760	case 2: /* clts */
5761		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5762		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5763			return 1;
5764		break;
5765	case 1: /* mov from cr */
5766		switch (cr) {
5767		case 3:
5768			if (vmcs12->cpu_based_vm_exec_control &
5769			    CPU_BASED_CR3_STORE_EXITING)
5770				return 1;
5771			break;
5772		case 8:
5773			if (vmcs12->cpu_based_vm_exec_control &
5774			    CPU_BASED_CR8_STORE_EXITING)
5775				return 1;
5776			break;
5777		}
5778		break;
5779	case 3: /* lmsw */
5780		/*
5781		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5782		 * cr0. Other attempted changes are ignored, with no exit.
5783		 */
5784		if (vmcs12->cr0_guest_host_mask & 0xe &
5785		    (val ^ vmcs12->cr0_read_shadow))
5786			return 1;
5787		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5788		    !(vmcs12->cr0_read_shadow & 0x1) &&
5789		    (val & 0x1))
5790			return 1;
5791		break;
5792	}
5793	return 0;
5794}
5795
5796/*
5797 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5798 * should handle it ourselves in L0 (and then continue L2). Only call this
5799 * when in is_guest_mode (L2).
5800 */
5801static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
5802{
5803	u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
5804	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5805	struct vcpu_vmx *vmx = to_vmx(vcpu);
5806	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5807
5808	if (vmx->nested.nested_run_pending)
5809		return 0;
5810
5811	if (unlikely(vmx->fail)) {
5812		pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5813				    vmcs_read32(VM_INSTRUCTION_ERROR));
5814		return 1;
5815	}
5816
5817	switch (exit_reason) {
5818	case EXIT_REASON_EXCEPTION_NMI:
5819		if (!is_exception(intr_info))
5820			return 0;
5821		else if (is_page_fault(intr_info))
5822			return enable_ept;
5823		return vmcs12->exception_bitmap &
5824				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5825	case EXIT_REASON_EXTERNAL_INTERRUPT:
5826		return 0;
5827	case EXIT_REASON_TRIPLE_FAULT:
5828		return 1;
5829	case EXIT_REASON_PENDING_INTERRUPT:
5830	case EXIT_REASON_NMI_WINDOW:
5831		/*
5832		 * prepare_vmcs02() set the CPU_BASED_VIRTUAL_INTR_PENDING bit
5833		 * (aka Interrupt Window Exiting) only when L1 turned it on,
5834		 * so if we got a PENDING_INTERRUPT exit, this must be for L1.
5835		 * Same for NMI Window Exiting.
5836		 */
5837		return 1;
5838	case EXIT_REASON_TASK_SWITCH:
5839		return 1;
5840	case EXIT_REASON_CPUID:
5841		return 1;
5842	case EXIT_REASON_HLT:
5843		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5844	case EXIT_REASON_INVD:
5845		return 1;
5846	case EXIT_REASON_INVLPG:
5847		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5848	case EXIT_REASON_RDPMC:
5849		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5850	case EXIT_REASON_RDTSC:
5851		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5852	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5853	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5854	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
5855	case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
5856	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5857		/*
5858		 * VMX instructions trap unconditionally. This allows L1 to
5859		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5860		 */
5861		return 1;
5862	case EXIT_REASON_CR_ACCESS:
5863		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5864	case EXIT_REASON_DR_ACCESS:
5865		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5866	case EXIT_REASON_IO_INSTRUCTION:
5867		/* TODO: support IO bitmaps */
5868		return 1;
5869	case EXIT_REASON_MSR_READ:
5870	case EXIT_REASON_MSR_WRITE:
5871		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5872	case EXIT_REASON_INVALID_STATE:
5873		return 1;
5874	case EXIT_REASON_MWAIT_INSTRUCTION:
5875		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5876	case EXIT_REASON_MONITOR_INSTRUCTION:
5877		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5878	case EXIT_REASON_PAUSE_INSTRUCTION:
5879		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5880			nested_cpu_has2(vmcs12,
5881				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5882	case EXIT_REASON_MCE_DURING_VMENTRY:
5883		return 0;
5884	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5885		return 1;
5886	case EXIT_REASON_APIC_ACCESS:
5887		return nested_cpu_has2(vmcs12,
5888			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
5889	case EXIT_REASON_EPT_VIOLATION:
5890	case EXIT_REASON_EPT_MISCONFIG:
5891		return 0;
5892	case EXIT_REASON_WBINVD:
5893		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5894	case EXIT_REASON_XSETBV:
5895		return 1;
5896	default:
5897		return 1;
5898	}
5899}
5900
5901static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5902{
5903	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5904	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5905}
5906
5907/*
5908 * The guest has exited.  See if we can fix it or if we need userspace
5909 * assistance.
5910 */
5911static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5912{
5913	struct vcpu_vmx *vmx = to_vmx(vcpu);
5914	u32 exit_reason = vmx->exit_reason;
5915	u32 vectoring_info = vmx->idt_vectoring_info;
5916
5917	/* If guest state is invalid, start emulating */
5918	if (vmx->emulation_required && emulate_invalid_guest_state)
5919		return handle_invalid_guest_state(vcpu);
5920
5921	/*
5922	 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
5923	 * we did not inject a still-pending event to L1 now because of
5924	 * nested_run_pending, we need to re-enable this bit.
5925	 */
5926	if (vmx->nested.nested_run_pending)
5927		kvm_make_request(KVM_REQ_EVENT, vcpu);
5928
5929	if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
5930	    exit_reason == EXIT_REASON_VMRESUME))
5931		vmx->nested.nested_run_pending = 1;
5932	else
5933		vmx->nested.nested_run_pending = 0;
5934
5935	if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
5936		nested_vmx_vmexit(vcpu);
5937		return 1;
5938	}
5939
5940	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5941		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5942		vcpu->run->fail_entry.hardware_entry_failure_reason
5943			= exit_reason;
5944		return 0;
5945	}
5946
5947	if (unlikely(vmx->fail)) {
5948		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5949		vcpu->run->fail_entry.hardware_entry_failure_reason
5950			= vmcs_read32(VM_INSTRUCTION_ERROR);
5951		return 0;
5952	}
5953
5954	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5955			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5956			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5957			exit_reason != EXIT_REASON_TASK_SWITCH))
5958		printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5959		       "(0x%x) and exit reason is 0x%x\n",
5960		       __func__, vectoring_info, exit_reason);
5961
5962	if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
5963	    !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
5964	                                get_vmcs12(vcpu), vcpu)))) {
5965		if (vmx_interrupt_allowed(vcpu)) {
5966			vmx->soft_vnmi_blocked = 0;
5967		} else if (vmx->vnmi_blocked_time > 1000000000LL &&
5968			   vcpu->arch.nmi_pending) {
5969			/*
5970			 * This CPU don't support us in finding the end of an
5971			 * NMI-blocked window if the guest runs with IRQs
5972			 * disabled. So we pull the trigger after 1 s of
5973			 * futile waiting, but inform the user about this.
5974			 */
5975			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5976			       "state on VCPU %d after 1 s timeout\n",
5977			       __func__, vcpu->vcpu_id);
5978			vmx->soft_vnmi_blocked = 0;
5979		}
5980	}
5981
5982	if (exit_reason < kvm_vmx_max_exit_handlers
5983	    && kvm_vmx_exit_handlers[exit_reason])
5984		return kvm_vmx_exit_handlers[exit_reason](vcpu);
5985	else {
5986		vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5987		vcpu->run->hw.hardware_exit_reason = exit_reason;
5988	}
5989	return 0;
5990}
5991
5992static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5993{
5994	if (irr == -1 || tpr < irr) {
5995		vmcs_write32(TPR_THRESHOLD, 0);
5996		return;
5997	}
5998
5999	vmcs_write32(TPR_THRESHOLD, irr);
6000}
6001
6002static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6003{
6004	u32 exit_intr_info;
6005
6006	if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6007	      || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
6008		return;
6009
6010	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6011	exit_intr_info = vmx->exit_intr_info;
6012
6013	/* Handle machine checks before interrupts are enabled */
6014	if (is_machine_check(exit_intr_info))
6015		kvm_machine_check();
6016
6017	/* We need to handle NMIs before interrupts are enabled */
6018	if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
6019	    (exit_intr_info & INTR_INFO_VALID_MASK)) {
6020		kvm_before_handle_nmi(&vmx->vcpu);
6021		asm("int $2");
6022		kvm_after_handle_nmi(&vmx->vcpu);
6023	}
6024}
6025
6026static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6027{
6028	u32 exit_intr_info;
6029	bool unblock_nmi;
6030	u8 vector;
6031	bool idtv_info_valid;
6032
6033	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6034
6035	if (cpu_has_virtual_nmis()) {
6036		if (vmx->nmi_known_unmasked)
6037			return;
6038		/*
6039		 * Can't use vmx->exit_intr_info since we're not sure what
6040		 * the exit reason is.
6041		 */
6042		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6043		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6044		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6045		/*
6046		 * SDM 3: 27.7.1.2 (September 2008)
6047		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6048		 * a guest IRET fault.
6049		 * SDM 3: 23.2.2 (September 2008)
6050		 * Bit 12 is undefined in any of the following cases:
6051		 *  If the VM exit sets the valid bit in the IDT-vectoring
6052		 *   information field.
6053		 *  If the VM exit is due to a double fault.
6054		 */
6055		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6056		    vector != DF_VECTOR && !idtv_info_valid)
6057			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6058				      GUEST_INTR_STATE_NMI);
6059		else
6060			vmx->nmi_known_unmasked =
6061				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6062				  & GUEST_INTR_STATE_NMI);
6063	} else if (unlikely(vmx->soft_vnmi_blocked))
6064		vmx->vnmi_blocked_time +=
6065			ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
6066}
6067
6068static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
6069				      u32 idt_vectoring_info,
6070				      int instr_len_field,
6071				      int error_code_field)
6072{
6073	u8 vector;
6074	int type;
6075	bool idtv_info_valid;
6076
6077	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6078
6079	vmx->vcpu.arch.nmi_injected = false;
6080	kvm_clear_exception_queue(&vmx->vcpu);
6081	kvm_clear_interrupt_queue(&vmx->vcpu);
6082
6083	if (!idtv_info_valid)
6084		return;
6085
6086	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6087
6088	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6089	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6090
6091	switch (type) {
6092	case INTR_TYPE_NMI_INTR:
6093		vmx->vcpu.arch.nmi_injected = true;
6094		/*
6095		 * SDM 3: 27.7.1.2 (September 2008)
6096		 * Clear bit "block by NMI" before VM entry if a NMI
6097		 * delivery faulted.
6098		 */
6099		vmx_set_nmi_mask(&vmx->vcpu, false);
6100		break;
6101	case INTR_TYPE_SOFT_EXCEPTION:
6102		vmx->vcpu.arch.event_exit_inst_len =
6103			vmcs_read32(instr_len_field);
6104		/* fall through */
6105	case INTR_TYPE_HARD_EXCEPTION:
6106		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6107			u32 err = vmcs_read32(error_code_field);
6108			kvm_queue_exception_e(&vmx->vcpu, vector, err);
6109		} else
6110			kvm_queue_exception(&vmx->vcpu, vector);
6111		break;
6112	case INTR_TYPE_SOFT_INTR:
6113		vmx->vcpu.arch.event_exit_inst_len =
6114			vmcs_read32(instr_len_field);
6115		/* fall through */
6116	case INTR_TYPE_EXT_INTR:
6117		kvm_queue_interrupt(&vmx->vcpu, vector,
6118			type == INTR_TYPE_SOFT_INTR);
6119		break;
6120	default:
6121		break;
6122	}
6123}
6124
6125static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6126{
6127	if (is_guest_mode(&vmx->vcpu))
6128		return;
6129	__vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
6130				  VM_EXIT_INSTRUCTION_LEN,
6131				  IDT_VECTORING_ERROR_CODE);
6132}
6133
6134static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6135{
6136	if (is_guest_mode(vcpu))
6137		return;
6138	__vmx_complete_interrupts(to_vmx(vcpu),
6139				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6140				  VM_ENTRY_INSTRUCTION_LEN,
6141				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6142
6143	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6144}
6145
6146static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6147{
6148	int i, nr_msrs;
6149	struct perf_guest_switch_msr *msrs;
6150
6151	msrs = perf_guest_get_msrs(&nr_msrs);
6152
6153	if (!msrs)
6154		return;
6155
6156	for (i = 0; i < nr_msrs; i++)
6157		if (msrs[i].host == msrs[i].guest)
6158			clear_atomic_switch_msr(vmx, msrs[i].msr);
6159		else
6160			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6161					msrs[i].host);
6162}
6163
6164#ifdef CONFIG_X86_64
6165#define R "r"
6166#define Q "q"
6167#else
6168#define R "e"
6169#define Q "l"
6170#endif
6171
6172static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6173{
6174	struct vcpu_vmx *vmx = to_vmx(vcpu);
6175
6176	if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
6177		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6178		if (vmcs12->idt_vectoring_info_field &
6179				VECTORING_INFO_VALID_MASK) {
6180			vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6181				vmcs12->idt_vectoring_info_field);
6182			vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6183				vmcs12->vm_exit_instruction_len);
6184			if (vmcs12->idt_vectoring_info_field &
6185					VECTORING_INFO_DELIVER_CODE_MASK)
6186				vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6187					vmcs12->idt_vectoring_error_code);
6188		}
6189	}
6190
6191	/* Record the guest's net vcpu time for enforced NMI injections. */
6192	if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
6193		vmx->entry_time = ktime_get();
6194
6195	/* Don't enter VMX if guest state is invalid, let the exit handler
6196	   start emulation until we arrive back to a valid state */
6197	if (vmx->emulation_required && emulate_invalid_guest_state)
6198		return;
6199
6200	if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6201		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6202	if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6203		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6204
6205	/* When single-stepping over STI and MOV SS, we must clear the
6206	 * corresponding interruptibility bits in the guest state. Otherwise
6207	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6208	 * exceptions being set, but that's not correct for the guest debugging
6209	 * case. */
6210	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6211		vmx_set_interrupt_shadow(vcpu, 0);
6212
6213	atomic_switch_perf_msrs(vmx);
6214
6215	vmx->__launched = vmx->loaded_vmcs->launched;
6216	asm(
6217		/* Store host registers */
6218		"push %%"R"dx; push %%"R"bp;"
6219		"push %%"R"cx \n\t" /* placeholder for guest rcx */
6220		"push %%"R"cx \n\t"
6221		"cmp %%"R"sp, %c[host_rsp](%0) \n\t"
6222		"je 1f \n\t"
6223		"mov %%"R"sp, %c[host_rsp](%0) \n\t"
6224		__ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
6225		"1: \n\t"
6226		/* Reload cr2 if changed */
6227		"mov %c[cr2](%0), %%"R"ax \n\t"
6228		"mov %%cr2, %%"R"dx \n\t"
6229		"cmp %%"R"ax, %%"R"dx \n\t"
6230		"je 2f \n\t"
6231		"mov %%"R"ax, %%cr2 \n\t"
6232		"2: \n\t"
6233		/* Check if vmlaunch of vmresume is needed */
6234		"cmpl $0, %c[launched](%0) \n\t"
6235		/* Load guest registers.  Don't clobber flags. */
6236		"mov %c[rax](%0), %%"R"ax \n\t"
6237		"mov %c[rbx](%0), %%"R"bx \n\t"
6238		"mov %c[rdx](%0), %%"R"dx \n\t"
6239		"mov %c[rsi](%0), %%"R"si \n\t"
6240		"mov %c[rdi](%0), %%"R"di \n\t"
6241		"mov %c[rbp](%0), %%"R"bp \n\t"
6242#ifdef CONFIG_X86_64
6243		"mov %c[r8](%0),  %%r8  \n\t"
6244		"mov %c[r9](%0),  %%r9  \n\t"
6245		"mov %c[r10](%0), %%r10 \n\t"
6246		"mov %c[r11](%0), %%r11 \n\t"
6247		"mov %c[r12](%0), %%r12 \n\t"
6248		"mov %c[r13](%0), %%r13 \n\t"
6249		"mov %c[r14](%0), %%r14 \n\t"
6250		"mov %c[r15](%0), %%r15 \n\t"
6251#endif
6252		"mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
6253
6254		/* Enter guest mode */
6255		"jne .Llaunched \n\t"
6256		__ex(ASM_VMX_VMLAUNCH) "\n\t"
6257		"jmp .Lkvm_vmx_return \n\t"
6258		".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
6259		".Lkvm_vmx_return: "
6260		/* Save guest registers, load host registers, keep flags */
6261		"mov %0, %c[wordsize](%%"R"sp) \n\t"
6262		"pop %0 \n\t"
6263		"mov %%"R"ax, %c[rax](%0) \n\t"
6264		"mov %%"R"bx, %c[rbx](%0) \n\t"
6265		"pop"Q" %c[rcx](%0) \n\t"
6266		"mov %%"R"dx, %c[rdx](%0) \n\t"
6267		"mov %%"R"si, %c[rsi](%0) \n\t"
6268		"mov %%"R"di, %c[rdi](%0) \n\t"
6269		"mov %%"R"bp, %c[rbp](%0) \n\t"
6270#ifdef CONFIG_X86_64
6271		"mov %%r8,  %c[r8](%0) \n\t"
6272		"mov %%r9,  %c[r9](%0) \n\t"
6273		"mov %%r10, %c[r10](%0) \n\t"
6274		"mov %%r11, %c[r11](%0) \n\t"
6275		"mov %%r12, %c[r12](%0) \n\t"
6276		"mov %%r13, %c[r13](%0) \n\t"
6277		"mov %%r14, %c[r14](%0) \n\t"
6278		"mov %%r15, %c[r15](%0) \n\t"
6279#endif
6280		"mov %%cr2, %%"R"ax   \n\t"
6281		"mov %%"R"ax, %c[cr2](%0) \n\t"
6282
6283		"pop  %%"R"bp; pop  %%"R"dx \n\t"
6284		"setbe %c[fail](%0) \n\t"
6285	      : : "c"(vmx), "d"((unsigned long)HOST_RSP),
6286		[launched]"i"(offsetof(struct vcpu_vmx, __launched)),
6287		[fail]"i"(offsetof(struct vcpu_vmx, fail)),
6288		[host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
6289		[rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
6290		[rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
6291		[rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
6292		[rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
6293		[rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
6294		[rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
6295		[rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
6296#ifdef CONFIG_X86_64
6297		[r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
6298		[r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
6299		[r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
6300		[r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
6301		[r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
6302		[r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
6303		[r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
6304		[r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6305#endif
6306		[cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
6307		[wordsize]"i"(sizeof(ulong))
6308	      : "cc", "memory"
6309		, R"ax", R"bx", R"di", R"si"
6310#ifdef CONFIG_X86_64
6311		, "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
6312#endif
6313	      );
6314
6315#ifndef CONFIG_X86_64
6316	/*
6317	 * The sysexit path does not restore ds/es, so we must set them to
6318	 * a reasonable value ourselves.
6319	 *
6320	 * We can't defer this to vmx_load_host_state() since that function
6321	 * may be executed in interrupt context, which saves and restore segments
6322	 * around it, nullifying its effect.
6323	 */
6324	loadsegment(ds, __USER_DS);
6325	loadsegment(es, __USER_DS);
6326#endif
6327
6328	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6329				  | (1 << VCPU_EXREG_RFLAGS)
6330				  | (1 << VCPU_EXREG_CPL)
6331				  | (1 << VCPU_EXREG_PDPTR)
6332				  | (1 << VCPU_EXREG_SEGMENTS)
6333				  | (1 << VCPU_EXREG_CR3));
6334	vcpu->arch.regs_dirty = 0;
6335
6336	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6337
6338	if (is_guest_mode(vcpu)) {
6339		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6340		vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
6341		if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
6342			vmcs12->idt_vectoring_error_code =
6343				vmcs_read32(IDT_VECTORING_ERROR_CODE);
6344			vmcs12->vm_exit_instruction_len =
6345				vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6346		}
6347	}
6348
6349	vmx->loaded_vmcs->launched = 1;
6350
6351	vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6352	trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
6353
6354	vmx_complete_atomic_exit(vmx);
6355	vmx_recover_nmi_blocking(vmx);
6356	vmx_complete_interrupts(vmx);
6357}
6358
6359#undef R
6360#undef Q
6361
6362static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6363{
6364	struct vcpu_vmx *vmx = to_vmx(vcpu);
6365
6366	free_vpid(vmx);
6367	free_nested(vmx);
6368	free_loaded_vmcs(vmx->loaded_vmcs);
6369	kfree(vmx->guest_msrs);
6370	kvm_vcpu_uninit(vcpu);
6371	kmem_cache_free(kvm_vcpu_cache, vmx);
6372}
6373
6374static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6375{
6376	int err;
6377	struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
6378	int cpu;
6379
6380	if (!vmx)
6381		return ERR_PTR(-ENOMEM);
6382
6383	allocate_vpid(vmx);
6384
6385	err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6386	if (err)
6387		goto free_vcpu;
6388
6389	vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
6390	err = -ENOMEM;
6391	if (!vmx->guest_msrs) {
6392		goto uninit_vcpu;
6393	}
6394
6395	vmx->loaded_vmcs = &vmx->vmcs01;
6396	vmx->loaded_vmcs->vmcs = alloc_vmcs();
6397	if (!vmx->loaded_vmcs->vmcs)
6398		goto free_msrs;
6399	if (!vmm_exclusive)
6400		kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
6401	loaded_vmcs_init(vmx->loaded_vmcs);
6402	if (!vmm_exclusive)
6403		kvm_cpu_vmxoff();
6404
6405	cpu = get_cpu();
6406	vmx_vcpu_load(&vmx->vcpu, cpu);
6407	vmx->vcpu.cpu = cpu;
6408	err = vmx_vcpu_setup(vmx);
6409	vmx_vcpu_put(&vmx->vcpu);
6410	put_cpu();
6411	if (err)
6412		goto free_vmcs;
6413	if (vm_need_virtualize_apic_accesses(kvm))
6414		err = alloc_apic_access_page(kvm);
6415		if (err)
6416			goto free_vmcs;
6417
6418	if (enable_ept) {
6419		if (!kvm->arch.ept_identity_map_addr)
6420			kvm->arch.ept_identity_map_addr =
6421				VMX_EPT_IDENTITY_PAGETABLE_ADDR;
6422		err = -ENOMEM;
6423		if (alloc_identity_pagetable(kvm) != 0)
6424			goto free_vmcs;
6425		if (!init_rmode_identity_map(kvm))
6426			goto free_vmcs;
6427	}
6428
6429	vmx->nested.current_vmptr = -1ull;
6430	vmx->nested.current_vmcs12 = NULL;
6431
6432	return &vmx->vcpu;
6433
6434free_vmcs:
6435	free_loaded_vmcs(vmx->loaded_vmcs);
6436free_msrs:
6437	kfree(vmx->guest_msrs);
6438uninit_vcpu:
6439	kvm_vcpu_uninit(&vmx->vcpu);
6440free_vcpu:
6441	free_vpid(vmx);
6442	kmem_cache_free(kvm_vcpu_cache, vmx);
6443	return ERR_PTR(err);
6444}
6445
6446static void __init vmx_check_processor_compat(void *rtn)
6447{
6448	struct vmcs_config vmcs_conf;
6449
6450	*(int *)rtn = 0;
6451	if (setup_vmcs_config(&vmcs_conf) < 0)
6452		*(int *)rtn = -EIO;
6453	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6454		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6455				smp_processor_id());
6456		*(int *)rtn = -EIO;
6457	}
6458}
6459
6460static int get_ept_level(void)
6461{
6462	return VMX_EPT_DEFAULT_GAW + 1;
6463}
6464
6465static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6466{
6467	u64 ret;
6468
6469	/* For VT-d and EPT combination
6470	 * 1. MMIO: always map as UC
6471	 * 2. EPT with VT-d:
6472	 *   a. VT-d without snooping control feature: can't guarantee the
6473	 *	result, try to trust guest.
6474	 *   b. VT-d with snooping control feature: snooping control feature of
6475	 *	VT-d engine can guarantee the cache correctness. Just set it
6476	 *	to WB to keep consistent with host. So the same as item 3.
6477	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6478	 *    consistent with host MTRR
6479	 */
6480	if (is_mmio)
6481		ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
6482	else if (vcpu->kvm->arch.iommu_domain &&
6483		!(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
6484		ret = kvm_get_guest_memory_type(vcpu, gfn) <<
6485		      VMX_EPT_MT_EPTE_SHIFT;
6486	else
6487		ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
6488			| VMX_EPT_IPAT_BIT;
6489
6490	return ret;
6491}
6492
6493static int vmx_get_lpage_level(void)
6494{
6495	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6496		return PT_DIRECTORY_LEVEL;
6497	else
6498		/* For shadow and EPT supported 1GB page */
6499		return PT_PDPE_LEVEL;
6500}
6501
6502static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
6503{
6504	struct kvm_cpuid_entry2 *best;
6505	struct vcpu_vmx *vmx = to_vmx(vcpu);
6506	u32 exec_control;
6507
6508	vmx->rdtscp_enabled = false;
6509	if (vmx_rdtscp_supported()) {
6510		exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6511		if (exec_control & SECONDARY_EXEC_RDTSCP) {
6512			best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
6513			if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
6514				vmx->rdtscp_enabled = true;
6515			else {
6516				exec_control &= ~SECONDARY_EXEC_RDTSCP;
6517				vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6518						exec_control);
6519			}
6520		}
6521	}
6522}
6523
6524static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
6525{
6526	if (func == 1 && nested)
6527		entry->ecx |= bit(X86_FEATURE_VMX);
6528}
6529
6530/*
6531 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
6532 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
6533 * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
6534 * guest in a way that will both be appropriate to L1's requests, and our
6535 * needs. In addition to modifying the active vmcs (which is vmcs02), this
6536 * function also has additional necessary side-effects, like setting various
6537 * vcpu->arch fields.
6538 */
6539static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6540{
6541	struct vcpu_vmx *vmx = to_vmx(vcpu);
6542	u32 exec_control;
6543
6544	vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
6545	vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
6546	vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
6547	vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
6548	vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
6549	vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
6550	vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
6551	vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
6552	vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
6553	vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
6554	vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
6555	vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
6556	vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
6557	vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
6558	vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
6559	vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
6560	vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
6561	vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
6562	vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
6563	vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
6564	vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
6565	vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
6566	vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
6567	vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
6568	vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
6569	vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
6570	vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
6571	vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
6572	vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
6573	vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
6574	vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
6575	vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
6576	vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
6577	vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
6578	vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
6579	vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
6580
6581	vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
6582	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6583		vmcs12->vm_entry_intr_info_field);
6584	vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6585		vmcs12->vm_entry_exception_error_code);
6586	vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6587		vmcs12->vm_entry_instruction_len);
6588	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
6589		vmcs12->guest_interruptibility_info);
6590	vmcs_write32(GUEST_ACTIVITY_STATE, vmcs12->guest_activity_state);
6591	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
6592	vmcs_writel(GUEST_DR7, vmcs12->guest_dr7);
6593	vmcs_writel(GUEST_RFLAGS, vmcs12->guest_rflags);
6594	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
6595		vmcs12->guest_pending_dbg_exceptions);
6596	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
6597	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
6598
6599	vmcs_write64(VMCS_LINK_POINTER, -1ull);
6600
6601	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
6602		(vmcs_config.pin_based_exec_ctrl |
6603		 vmcs12->pin_based_vm_exec_control));
6604
6605	/*
6606	 * Whether page-faults are trapped is determined by a combination of
6607	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
6608	 * If enable_ept, L0 doesn't care about page faults and we should
6609	 * set all of these to L1's desires. However, if !enable_ept, L0 does
6610	 * care about (at least some) page faults, and because it is not easy
6611	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
6612	 * to exit on each and every L2 page fault. This is done by setting
6613	 * MASK=MATCH=0 and (see below) EB.PF=1.
6614	 * Note that below we don't need special code to set EB.PF beyond the
6615	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
6616	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
6617	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
6618	 *
6619	 * A problem with this approach (when !enable_ept) is that L1 may be
6620	 * injected with more page faults than it asked for. This could have
6621	 * caused problems, but in practice existing hypervisors don't care.
6622	 * To fix this, we will need to emulate the PFEC checking (on the L1
6623	 * page tables), using walk_addr(), when injecting PFs to L1.
6624	 */
6625	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
6626		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
6627	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
6628		enable_ept ? vmcs12->page_fault_error_code_match : 0);
6629
6630	if (cpu_has_secondary_exec_ctrls()) {
6631		u32 exec_control = vmx_secondary_exec_control(vmx);
6632		if (!vmx->rdtscp_enabled)
6633			exec_control &= ~SECONDARY_EXEC_RDTSCP;
6634		/* Take the following fields only from vmcs12 */
6635		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6636		if (nested_cpu_has(vmcs12,
6637				CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
6638			exec_control |= vmcs12->secondary_vm_exec_control;
6639
6640		if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
6641			/*
6642			 * Translate L1 physical address to host physical
6643			 * address for vmcs02. Keep the page pinned, so this
6644			 * physical address remains valid. We keep a reference
6645			 * to it so we can release it later.
6646			 */
6647			if (vmx->nested.apic_access_page) /* shouldn't happen */
6648				nested_release_page(vmx->nested.apic_access_page);
6649			vmx->nested.apic_access_page =
6650				nested_get_page(vcpu, vmcs12->apic_access_addr);
6651			/*
6652			 * If translation failed, no matter: This feature asks
6653			 * to exit when accessing the given address, and if it
6654			 * can never be accessed, this feature won't do
6655			 * anything anyway.
6656			 */
6657			if (!vmx->nested.apic_access_page)
6658				exec_control &=
6659				  ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6660			else
6661				vmcs_write64(APIC_ACCESS_ADDR,
6662				  page_to_phys(vmx->nested.apic_access_page));
6663		}
6664
6665		vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6666	}
6667
6668
6669	/*
6670	 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
6671	 * Some constant fields are set here by vmx_set_constant_host_state().
6672	 * Other fields are different per CPU, and will be set later when
6673	 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
6674	 */
6675	vmx_set_constant_host_state();
6676
6677	/*
6678	 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
6679	 * entry, but only if the current (host) sp changed from the value
6680	 * we wrote last (vmx->host_rsp). This cache is no longer relevant
6681	 * if we switch vmcs, and rather than hold a separate cache per vmcs,
6682	 * here we just force the write to happen on entry.
6683	 */
6684	vmx->host_rsp = 0;
6685
6686	exec_control = vmx_exec_control(vmx); /* L0's desires */
6687	exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
6688	exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6689	exec_control &= ~CPU_BASED_TPR_SHADOW;
6690	exec_control |= vmcs12->cpu_based_vm_exec_control;
6691	/*
6692	 * Merging of IO and MSR bitmaps not currently supported.
6693	 * Rather, exit every time.
6694	 */
6695	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
6696	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
6697	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
6698
6699	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6700
6701	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
6702	 * bitwise-or of what L1 wants to trap for L2, and what we want to
6703	 * trap. Note that CR0.TS also needs updating - we do this later.
6704	 */
6705	update_exception_bitmap(vcpu);
6706	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
6707	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
6708
6709	/* Note: IA32_MODE, LOAD_IA32_EFER are modified by vmx_set_efer below */
6710	vmcs_write32(VM_EXIT_CONTROLS,
6711		vmcs12->vm_exit_controls | vmcs_config.vmexit_ctrl);
6712	vmcs_write32(VM_ENTRY_CONTROLS, vmcs12->vm_entry_controls |
6713		(vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
6714
6715	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)
6716		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
6717	else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
6718		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
6719
6720
6721	set_cr4_guest_host_mask(vmx);
6722
6723	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
6724		vmcs_write64(TSC_OFFSET,
6725			vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
6726	else
6727		vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
6728
6729	if (enable_vpid) {
6730		/*
6731		 * Trivially support vpid by letting L2s share their parent
6732		 * L1's vpid. TODO: move to a more elaborate solution, giving
6733		 * each L2 its own vpid and exposing the vpid feature to L1.
6734		 */
6735		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
6736		vmx_flush_tlb(vcpu);
6737	}
6738
6739	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
6740		vcpu->arch.efer = vmcs12->guest_ia32_efer;
6741	if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
6742		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
6743	else
6744		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
6745	/* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
6746	vmx_set_efer(vcpu, vcpu->arch.efer);
6747
6748	/*
6749	 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
6750	 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
6751	 * The CR0_READ_SHADOW is what L2 should have expected to read given
6752	 * the specifications by L1; It's not enough to take
6753	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
6754	 * have more bits than L1 expected.
6755	 */
6756	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
6757	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
6758
6759	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
6760	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
6761
6762	/* shadow page tables on either EPT or shadow page tables */
6763	kvm_set_cr3(vcpu, vmcs12->guest_cr3);
6764	kvm_mmu_reset_context(vcpu);
6765
6766	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
6767	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
6768}
6769
6770/*
6771 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
6772 * for running an L2 nested guest.
6773 */
6774static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
6775{
6776	struct vmcs12 *vmcs12;
6777	struct vcpu_vmx *vmx = to_vmx(vcpu);
6778	int cpu;
6779	struct loaded_vmcs *vmcs02;
6780
6781	if (!nested_vmx_check_permission(vcpu) ||
6782	    !nested_vmx_check_vmcs12(vcpu))
6783		return 1;
6784
6785	skip_emulated_instruction(vcpu);
6786	vmcs12 = get_vmcs12(vcpu);
6787
6788	/*
6789	 * The nested entry process starts with enforcing various prerequisites
6790	 * on vmcs12 as required by the Intel SDM, and act appropriately when
6791	 * they fail: As the SDM explains, some conditions should cause the
6792	 * instruction to fail, while others will cause the instruction to seem
6793	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
6794	 * To speed up the normal (success) code path, we should avoid checking
6795	 * for misconfigurations which will anyway be caught by the processor
6796	 * when using the merged vmcs02.
6797	 */
6798	if (vmcs12->launch_state == launch) {
6799		nested_vmx_failValid(vcpu,
6800			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
6801			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
6802		return 1;
6803	}
6804
6805	if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
6806			!IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
6807		/*TODO: Also verify bits beyond physical address width are 0*/
6808		nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6809		return 1;
6810	}
6811
6812	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
6813			!IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
6814		/*TODO: Also verify bits beyond physical address width are 0*/
6815		nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6816		return 1;
6817	}
6818
6819	if (vmcs12->vm_entry_msr_load_count > 0 ||
6820	    vmcs12->vm_exit_msr_load_count > 0 ||
6821	    vmcs12->vm_exit_msr_store_count > 0) {
6822		pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
6823				    __func__);
6824		nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6825		return 1;
6826	}
6827
6828	if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
6829	      nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
6830	    !vmx_control_verify(vmcs12->secondary_vm_exec_control,
6831	      nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
6832	    !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
6833	      nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
6834	    !vmx_control_verify(vmcs12->vm_exit_controls,
6835	      nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
6836	    !vmx_control_verify(vmcs12->vm_entry_controls,
6837	      nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
6838	{
6839		nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6840		return 1;
6841	}
6842
6843	if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6844	    ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6845		nested_vmx_failValid(vcpu,
6846			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6847		return 1;
6848	}
6849
6850	if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6851	    ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6852		nested_vmx_entry_failure(vcpu, vmcs12,
6853			EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
6854		return 1;
6855	}
6856	if (vmcs12->vmcs_link_pointer != -1ull) {
6857		nested_vmx_entry_failure(vcpu, vmcs12,
6858			EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
6859		return 1;
6860	}
6861
6862	/*
6863	 * We're finally done with prerequisite checking, and can start with
6864	 * the nested entry.
6865	 */
6866
6867	vmcs02 = nested_get_current_vmcs02(vmx);
6868	if (!vmcs02)
6869		return -ENOMEM;
6870
6871	enter_guest_mode(vcpu);
6872
6873	vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
6874
6875	cpu = get_cpu();
6876	vmx->loaded_vmcs = vmcs02;
6877	vmx_vcpu_put(vcpu);
6878	vmx_vcpu_load(vcpu, cpu);
6879	vcpu->cpu = cpu;
6880	put_cpu();
6881
6882	vmcs12->launch_state = 1;
6883
6884	prepare_vmcs02(vcpu, vmcs12);
6885
6886	/*
6887	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
6888	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
6889	 * returned as far as L1 is concerned. It will only return (and set
6890	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
6891	 */
6892	return 1;
6893}
6894
6895/*
6896 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
6897 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
6898 * This function returns the new value we should put in vmcs12.guest_cr0.
6899 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
6900 *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
6901 *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
6902 *     didn't trap the bit, because if L1 did, so would L0).
6903 *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
6904 *     been modified by L2, and L1 knows it. So just leave the old value of
6905 *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
6906 *     isn't relevant, because if L0 traps this bit it can set it to anything.
6907 *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
6908 *     changed these bits, and therefore they need to be updated, but L0
6909 *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
6910 *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
6911 */
6912static inline unsigned long
6913vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6914{
6915	return
6916	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
6917	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
6918	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
6919			vcpu->arch.cr0_guest_owned_bits));
6920}
6921
6922static inline unsigned long
6923vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6924{
6925	return
6926	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
6927	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
6928	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
6929			vcpu->arch.cr4_guest_owned_bits));
6930}
6931
6932/*
6933 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
6934 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
6935 * and this function updates it to reflect the changes to the guest state while
6936 * L2 was running (and perhaps made some exits which were handled directly by L0
6937 * without going back to L1), and to reflect the exit reason.
6938 * Note that we do not have to copy here all VMCS fields, just those that
6939 * could have changed by the L2 guest or the exit - i.e., the guest-state and
6940 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
6941 * which already writes to vmcs12 directly.
6942 */
6943void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6944{
6945	/* update guest state fields: */
6946	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
6947	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
6948
6949	kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
6950	vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6951	vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
6952	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
6953
6954	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
6955	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
6956	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
6957	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
6958	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
6959	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
6960	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
6961	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
6962	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
6963	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
6964	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
6965	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
6966	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
6967	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
6968	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
6969	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
6970	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
6971	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
6972	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
6973	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
6974	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
6975	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
6976	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
6977	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
6978	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
6979	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
6980	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
6981	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
6982	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
6983	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
6984	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
6985	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
6986	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
6987	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
6988	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
6989	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
6990
6991	vmcs12->guest_activity_state = vmcs_read32(GUEST_ACTIVITY_STATE);
6992	vmcs12->guest_interruptibility_info =
6993		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
6994	vmcs12->guest_pending_dbg_exceptions =
6995		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
6996
6997	/* TODO: These cannot have changed unless we have MSR bitmaps and
6998	 * the relevant bit asks not to trap the change */
6999	vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
7000	if (vmcs12->vm_entry_controls & VM_EXIT_SAVE_IA32_PAT)
7001		vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
7002	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
7003	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
7004	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
7005
7006	/* update exit information fields: */
7007
7008	vmcs12->vm_exit_reason  = vmcs_read32(VM_EXIT_REASON);
7009	vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7010
7011	vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7012	vmcs12->vm_exit_intr_error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
7013	vmcs12->idt_vectoring_info_field =
7014		vmcs_read32(IDT_VECTORING_INFO_FIELD);
7015	vmcs12->idt_vectoring_error_code =
7016		vmcs_read32(IDT_VECTORING_ERROR_CODE);
7017	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
7018	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7019
7020	/* clear vm-entry fields which are to be cleared on exit */
7021	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
7022		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
7023}
7024
7025/*
7026 * A part of what we need to when the nested L2 guest exits and we want to
7027 * run its L1 parent, is to reset L1's guest state to the host state specified
7028 * in vmcs12.
7029 * This function is to be called not only on normal nested exit, but also on
7030 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
7031 * Failures During or After Loading Guest State").
7032 * This function should be called when the active VMCS is L1's (vmcs01).
7033 */
7034void load_vmcs12_host_state(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7035{
7036	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
7037		vcpu->arch.efer = vmcs12->host_ia32_efer;
7038	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
7039		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
7040	else
7041		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
7042	vmx_set_efer(vcpu, vcpu->arch.efer);
7043
7044	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
7045	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
7046	/*
7047	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
7048	 * actually changed, because it depends on the current state of
7049	 * fpu_active (which may have changed).
7050	 * Note that vmx_set_cr0 refers to efer set above.
7051	 */
7052	kvm_set_cr0(vcpu, vmcs12->host_cr0);
7053	/*
7054	 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
7055	 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
7056	 * but we also need to update cr0_guest_host_mask and exception_bitmap.
7057	 */
7058	update_exception_bitmap(vcpu);
7059	vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
7060	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7061
7062	/*
7063	 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
7064	 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
7065	 */
7066	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
7067	kvm_set_cr4(vcpu, vmcs12->host_cr4);
7068
7069	/* shadow page tables on either EPT or shadow page tables */
7070	kvm_set_cr3(vcpu, vmcs12->host_cr3);
7071	kvm_mmu_reset_context(vcpu);
7072
7073	if (enable_vpid) {
7074		/*
7075		 * Trivially support vpid by letting L2s share their parent
7076		 * L1's vpid. TODO: move to a more elaborate solution, giving
7077		 * each L2 its own vpid and exposing the vpid feature to L1.
7078		 */
7079		vmx_flush_tlb(vcpu);
7080	}
7081
7082
7083	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
7084	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
7085	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
7086	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
7087	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
7088	vmcs_writel(GUEST_TR_BASE, vmcs12->host_tr_base);
7089	vmcs_writel(GUEST_GS_BASE, vmcs12->host_gs_base);
7090	vmcs_writel(GUEST_FS_BASE, vmcs12->host_fs_base);
7091	vmcs_write16(GUEST_ES_SELECTOR, vmcs12->host_es_selector);
7092	vmcs_write16(GUEST_CS_SELECTOR, vmcs12->host_cs_selector);
7093	vmcs_write16(GUEST_SS_SELECTOR, vmcs12->host_ss_selector);
7094	vmcs_write16(GUEST_DS_SELECTOR, vmcs12->host_ds_selector);
7095	vmcs_write16(GUEST_FS_SELECTOR, vmcs12->host_fs_selector);
7096	vmcs_write16(GUEST_GS_SELECTOR, vmcs12->host_gs_selector);
7097	vmcs_write16(GUEST_TR_SELECTOR, vmcs12->host_tr_selector);
7098
7099	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT)
7100		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
7101	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
7102		vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
7103			vmcs12->host_ia32_perf_global_ctrl);
7104}
7105
7106/*
7107 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
7108 * and modify vmcs12 to make it see what it would expect to see there if
7109 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
7110 */
7111static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
7112{
7113	struct vcpu_vmx *vmx = to_vmx(vcpu);
7114	int cpu;
7115	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7116
7117	leave_guest_mode(vcpu);
7118	prepare_vmcs12(vcpu, vmcs12);
7119
7120	cpu = get_cpu();
7121	vmx->loaded_vmcs = &vmx->vmcs01;
7122	vmx_vcpu_put(vcpu);
7123	vmx_vcpu_load(vcpu, cpu);
7124	vcpu->cpu = cpu;
7125	put_cpu();
7126
7127	/* if no vmcs02 cache requested, remove the one we used */
7128	if (VMCS02_POOL_SIZE == 0)
7129		nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
7130
7131	load_vmcs12_host_state(vcpu, vmcs12);
7132
7133	/* Update TSC_OFFSET if TSC was changed while L2 ran */
7134	vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
7135
7136	/* This is needed for same reason as it was needed in prepare_vmcs02 */
7137	vmx->host_rsp = 0;
7138
7139	/* Unpin physical memory we referred to in vmcs02 */
7140	if (vmx->nested.apic_access_page) {
7141		nested_release_page(vmx->nested.apic_access_page);
7142		vmx->nested.apic_access_page = 0;
7143	}
7144
7145	/*
7146	 * Exiting from L2 to L1, we're now back to L1 which thinks it just
7147	 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
7148	 * success or failure flag accordingly.
7149	 */
7150	if (unlikely(vmx->fail)) {
7151		vmx->fail = 0;
7152		nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
7153	} else
7154		nested_vmx_succeed(vcpu);
7155}
7156
7157/*
7158 * L1's failure to enter L2 is a subset of a normal exit, as explained in
7159 * 23.7 "VM-entry failures during or after loading guest state" (this also
7160 * lists the acceptable exit-reason and exit-qualification parameters).
7161 * It should only be called before L2 actually succeeded to run, and when
7162 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
7163 */
7164static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
7165			struct vmcs12 *vmcs12,
7166			u32 reason, unsigned long qualification)
7167{
7168	load_vmcs12_host_state(vcpu, vmcs12);
7169	vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
7170	vmcs12->exit_qualification = qualification;
7171	nested_vmx_succeed(vcpu);
7172}
7173
7174static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7175			       struct x86_instruction_info *info,
7176			       enum x86_intercept_stage stage)
7177{
7178	return X86EMUL_CONTINUE;
7179}
7180
7181static struct kvm_x86_ops vmx_x86_ops = {
7182	.cpu_has_kvm_support = cpu_has_kvm_support,
7183	.disabled_by_bios = vmx_disabled_by_bios,
7184	.hardware_setup = hardware_setup,
7185	.hardware_unsetup = hardware_unsetup,
7186	.check_processor_compatibility = vmx_check_processor_compat,
7187	.hardware_enable = hardware_enable,
7188	.hardware_disable = hardware_disable,
7189	.cpu_has_accelerated_tpr = report_flexpriority,
7190
7191	.vcpu_create = vmx_create_vcpu,
7192	.vcpu_free = vmx_free_vcpu,
7193	.vcpu_reset = vmx_vcpu_reset,
7194
7195	.prepare_guest_switch = vmx_save_host_state,
7196	.vcpu_load = vmx_vcpu_load,
7197	.vcpu_put = vmx_vcpu_put,
7198
7199	.set_guest_debug = set_guest_debug,
7200	.get_msr = vmx_get_msr,
7201	.set_msr = vmx_set_msr,
7202	.get_segment_base = vmx_get_segment_base,
7203	.get_segment = vmx_get_segment,
7204	.set_segment = vmx_set_segment,
7205	.get_cpl = vmx_get_cpl,
7206	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7207	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7208	.decache_cr3 = vmx_decache_cr3,
7209	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7210	.set_cr0 = vmx_set_cr0,
7211	.set_cr3 = vmx_set_cr3,
7212	.set_cr4 = vmx_set_cr4,
7213	.set_efer = vmx_set_efer,
7214	.get_idt = vmx_get_idt,
7215	.set_idt = vmx_set_idt,
7216	.get_gdt = vmx_get_gdt,
7217	.set_gdt = vmx_set_gdt,
7218	.set_dr7 = vmx_set_dr7,
7219	.cache_reg = vmx_cache_reg,
7220	.get_rflags = vmx_get_rflags,
7221	.set_rflags = vmx_set_rflags,
7222	.fpu_activate = vmx_fpu_activate,
7223	.fpu_deactivate = vmx_fpu_deactivate,
7224
7225	.tlb_flush = vmx_flush_tlb,
7226
7227	.run = vmx_vcpu_run,
7228	.handle_exit = vmx_handle_exit,
7229	.skip_emulated_instruction = skip_emulated_instruction,
7230	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7231	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7232	.patch_hypercall = vmx_patch_hypercall,
7233	.set_irq = vmx_inject_irq,
7234	.set_nmi = vmx_inject_nmi,
7235	.queue_exception = vmx_queue_exception,
7236	.cancel_injection = vmx_cancel_injection,
7237	.interrupt_allowed = vmx_interrupt_allowed,
7238	.nmi_allowed = vmx_nmi_allowed,
7239	.get_nmi_mask = vmx_get_nmi_mask,
7240	.set_nmi_mask = vmx_set_nmi_mask,
7241	.enable_nmi_window = enable_nmi_window,
7242	.enable_irq_window = enable_irq_window,
7243	.update_cr8_intercept = update_cr8_intercept,
7244
7245	.set_tss_addr = vmx_set_tss_addr,
7246	.get_tdp_level = get_ept_level,
7247	.get_mt_mask = vmx_get_mt_mask,
7248
7249	.get_exit_info = vmx_get_exit_info,
7250
7251	.get_lpage_level = vmx_get_lpage_level,
7252
7253	.cpuid_update = vmx_cpuid_update,
7254
7255	.rdtscp_supported = vmx_rdtscp_supported,
7256
7257	.set_supported_cpuid = vmx_set_supported_cpuid,
7258
7259	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7260
7261	.set_tsc_khz = vmx_set_tsc_khz,
7262	.write_tsc_offset = vmx_write_tsc_offset,
7263	.adjust_tsc_offset = vmx_adjust_tsc_offset,
7264	.compute_tsc_offset = vmx_compute_tsc_offset,
7265	.read_l1_tsc = vmx_read_l1_tsc,
7266
7267	.set_tdp_cr3 = vmx_set_cr3,
7268
7269	.check_intercept = vmx_check_intercept,
7270};
7271
7272static int __init vmx_init(void)
7273{
7274	int r, i;
7275
7276	rdmsrl_safe(MSR_EFER, &host_efer);
7277
7278	for (i = 0; i < NR_VMX_MSR; ++i)
7279		kvm_define_shared_msr(i, vmx_msr_index[i]);
7280
7281	vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
7282	if (!vmx_io_bitmap_a)
7283		return -ENOMEM;
7284
7285	vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
7286	if (!vmx_io_bitmap_b) {
7287		r = -ENOMEM;
7288		goto out;
7289	}
7290
7291	vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
7292	if (!vmx_msr_bitmap_legacy) {
7293		r = -ENOMEM;
7294		goto out1;
7295	}
7296
7297	vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
7298	if (!vmx_msr_bitmap_longmode) {
7299		r = -ENOMEM;
7300		goto out2;
7301	}
7302
7303	/*
7304	 * Allow direct access to the PC debug port (it is often used for I/O
7305	 * delays, but the vmexits simply slow things down).
7306	 */
7307	memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
7308	clear_bit(0x80, vmx_io_bitmap_a);
7309
7310	memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
7311
7312	memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
7313	memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
7314
7315	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7316
7317	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7318		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7319	if (r)
7320		goto out3;
7321
7322	vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
7323	vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
7324	vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
7325	vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
7326	vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
7327	vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
7328
7329	if (enable_ept) {
7330		kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
7331				VMX_EPT_EXECUTABLE_MASK);
7332		ept_set_mmio_spte_mask();
7333		kvm_enable_tdp();
7334	} else
7335		kvm_disable_tdp();
7336
7337	return 0;
7338
7339out3:
7340	free_page((unsigned long)vmx_msr_bitmap_longmode);
7341out2:
7342	free_page((unsigned long)vmx_msr_bitmap_legacy);
7343out1:
7344	free_page((unsigned long)vmx_io_bitmap_b);
7345out:
7346	free_page((unsigned long)vmx_io_bitmap_a);
7347	return r;
7348}
7349
7350static void __exit vmx_exit(void)
7351{
7352	free_page((unsigned long)vmx_msr_bitmap_legacy);
7353	free_page((unsigned long)vmx_msr_bitmap_longmode);
7354	free_page((unsigned long)vmx_io_bitmap_b);
7355	free_page((unsigned long)vmx_io_bitmap_a);
7356
7357	kvm_exit();
7358}
7359
7360module_init(vmx_init)
7361module_exit(vmx_exit)