arch/x86/kvm/vmx.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "irq.h"
  20 #include "mmu.h"
  21 #include "cpuid.h"
  22 #include "lapic.h"
  23
  24 #include <linux/kvm_host.h>
  25 #include <linux/module.h>
  26 #include <linux/kernel.h>
  27 #include <linux/mm.h>
  28 #include <linux/highmem.h>
  29 #include <linux/sched.h>
  30 #include <linux/moduleparam.h>
  31 #include <linux/mod_devicetable.h>
  32 #include <linux/trace_events.h>
  33 #include <linux/slab.h>
  34 #include <linux/tboot.h>
  35 #include <linux/hrtimer.h>
  36 #include <linux/frame.h>
  37 #include "kvm_cache_regs.h"
  38 #include "x86.h"
  39
  40 #include <asm/cpu.h>
  41 #include <asm/io.h>
  42 #include <asm/desc.h>
  43 #include <asm/vmx.h>
  44 #include <asm/virtext.h>
  45 #include <asm/mce.h>
  46 #include <asm/fpu/internal.h>
  47 #include <asm/perf_event.h>
  48 #include <asm/debugreg.h>
  49 #include <asm/kexec.h>
  50 #include <asm/apic.h>
  51 #include <asm/irq_remapping.h>
  52 #include <asm/mmu_context.h>
  53
  54 #include "trace.h"
  55 #include "pmu.h"
  56
  57 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  58 #define __ex_clear(x, reg) \
  59         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
  60
  61 MODULE_AUTHOR("Qumranet");
  62 MODULE_LICENSE("GPL");
  63
  64 static const struct x86_cpu_id vmx_cpu_id[] = {
  65         X86_FEATURE_MATCH(X86_FEATURE_VMX),
  66         {}
  67 };
  68 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
  69
  70 static bool __read_mostly enable_vpid = 1;
  71 module_param_named(vpid, enable_vpid, bool, 0444);
  72
  73 static bool __read_mostly flexpriority_enabled = 1;
  74 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  75
  76 static bool __read_mostly enable_ept = 1;
  77 module_param_named(ept, enable_ept, bool, S_IRUGO);
  78
  79 static bool __read_mostly enable_unrestricted_guest = 1;
  80 module_param_named(unrestricted_guest,
  81                         enable_unrestricted_guest, bool, S_IRUGO);
  82
  83 static bool __read_mostly enable_ept_ad_bits = 1;
  84 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
  85
  86 static bool __read_mostly emulate_invalid_guest_state = true;
  87 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  88
  89 static bool __read_mostly fasteoi = 1;
  90 module_param(fasteoi, bool, S_IRUGO);
  91
  92 static bool __read_mostly enable_apicv = 1;
  93 module_param(enable_apicv, bool, S_IRUGO);
  94
  95 static bool __read_mostly enable_shadow_vmcs = 1;
  96 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
  97 /*
  98  * If nested=1, nested virtualization is supported, i.e., guests may use
  99  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
 100  * use VMX instructions.
 101  */
 102 static bool __read_mostly nested = 0;
 103 module_param(nested, bool, S_IRUGO);
 104
 105 static u64 __read_mostly host_xss;
 106
 107 static bool __read_mostly enable_pml = 1;
 108 module_param_named(pml, enable_pml, bool, S_IRUGO);
 109
 110 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
 111
 112 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
 113 static int __read_mostly cpu_preemption_timer_multi;
 114 static bool __read_mostly enable_preemption_timer = 1;
 115 #ifdef CONFIG_X86_64
 116 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
 117 #endif
 118
 119 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
 120 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
 121 #define KVM_VM_CR0_ALWAYS_ON                                            \
 122         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
 123 #define KVM_CR4_GUEST_OWNED_BITS                                      \
 124         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
 125          | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
 126
 127 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
 128 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
 129
 130 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
 131
 132 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
 133
 134 /*
 135  * Hyper-V requires all of these, so mark them as supported even though
 136  * they are just treated the same as all-context.
 137  */
 138 #define VMX_VPID_EXTENT_SUPPORTED_MASK          \
 139         (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |  \
 140         VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |    \
 141         VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |    \
 142         VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
 143
 144 /*
 145  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 146  * ple_gap:    upper bound on the amount of time between two successive
 147  *             executions of PAUSE in a loop. Also indicate if ple enabled.
 148  *             According to test, this time is usually smaller than 128 cycles.
 149  * ple_window: upper bound on the amount of time a guest is allowed to execute
 150  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
 151  *             less than 2^12 cycles
 152  * Time is measured based on a counter that runs at the same rate as the TSC,
 153  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 154  */
 155 #define KVM_VMX_DEFAULT_PLE_GAP           128
 156 #define KVM_VMX_DEFAULT_PLE_WINDOW        4096
 157 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW   2
 158 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
 159 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX    \
 160                 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
 161
 162 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 163 module_param(ple_gap, int, S_IRUGO);
 164
 165 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 166 module_param(ple_window, int, S_IRUGO);
 167
 168 /* Default doubles per-vcpu window every exit. */
 169 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
 170 module_param(ple_window_grow, int, S_IRUGO);
 171
 172 /* Default resets per-vcpu window every exit to ple_window. */
 173 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
 174 module_param(ple_window_shrink, int, S_IRUGO);
 175
 176 /* Default is to compute the maximum so we can never overflow. */
 177 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
 178 static int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
 179 module_param(ple_window_max, int, S_IRUGO);
 180
 181 extern const ulong vmx_return;
 182
 183 #define NR_AUTOLOAD_MSRS 8
 184 #define VMCS02_POOL_SIZE 1
 185
 186 struct vmcs {
 187         u32 revision_id;
 188         u32 abort;
 189         char data[0];
 190 };
 191
 192 /*
 193  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
 194  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
 195  * loaded on this CPU (so we can clear them if the CPU goes down).
 196  */
 197 struct loaded_vmcs {
 198         struct vmcs *vmcs;
 199         struct vmcs *shadow_vmcs;
 200         int cpu;
 201         bool launched;
 202         bool nmi_known_unmasked;
 203         struct list_head loaded_vmcss_on_cpu_link;
 204 };
 205
 206 struct shared_msr_entry {
 207         unsigned index;
 208         u64 data;
 209         u64 mask;
 210 };
 211
 212 /*
 213  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
 214  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
 215  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
 216  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
 217  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
 218  * More than one of these structures may exist, if L1 runs multiple L2 guests.
 219  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
 220  * underlying hardware which will be used to run L2.
 221  * This structure is packed to ensure that its layout is identical across
 222  * machines (necessary for live migration).
 223  * If there are changes in this struct, VMCS12_REVISION must be changed.
 224  */
 225 typedef u64 natural_width;
 226 struct __packed vmcs12 {
 227         /* According to the Intel spec, a VMCS region must start with the
 228          * following two fields. Then follow implementation-specific data.
 229          */
 230         u32 revision_id;
 231         u32 abort;
 232
 233         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
 234         u32 padding[7]; /* room for future expansion */
 235
 236         u64 io_bitmap_a;
 237         u64 io_bitmap_b;
 238         u64 msr_bitmap;
 239         u64 vm_exit_msr_store_addr;
 240         u64 vm_exit_msr_load_addr;
 241         u64 vm_entry_msr_load_addr;
 242         u64 tsc_offset;
 243         u64 virtual_apic_page_addr;
 244         u64 apic_access_addr;
 245         u64 posted_intr_desc_addr;
 246         u64 ept_pointer;
 247         u64 eoi_exit_bitmap0;
 248         u64 eoi_exit_bitmap1;
 249         u64 eoi_exit_bitmap2;
 250         u64 eoi_exit_bitmap3;
 251         u64 xss_exit_bitmap;
 252         u64 guest_physical_address;
 253         u64 vmcs_link_pointer;
 254         u64 pml_address;
 255         u64 guest_ia32_debugctl;
 256         u64 guest_ia32_pat;
 257         u64 guest_ia32_efer;
 258         u64 guest_ia32_perf_global_ctrl;
 259         u64 guest_pdptr0;
 260         u64 guest_pdptr1;
 261         u64 guest_pdptr2;
 262         u64 guest_pdptr3;
 263         u64 guest_bndcfgs;
 264         u64 host_ia32_pat;
 265         u64 host_ia32_efer;
 266         u64 host_ia32_perf_global_ctrl;
 267         u64 padding64[8]; /* room for future expansion */
 268         /*
 269          * To allow migration of L1 (complete with its L2 guests) between
 270          * machines of different natural widths (32 or 64 bit), we cannot have
 271          * unsigned long fields with no explict size. We use u64 (aliased
 272          * natural_width) instead. Luckily, x86 is little-endian.
 273          */
 274         natural_width cr0_guest_host_mask;
 275         natural_width cr4_guest_host_mask;
 276         natural_width cr0_read_shadow;
 277         natural_width cr4_read_shadow;
 278         natural_width cr3_target_value0;
 279         natural_width cr3_target_value1;
 280         natural_width cr3_target_value2;
 281         natural_width cr3_target_value3;
 282         natural_width exit_qualification;
 283         natural_width guest_linear_address;
 284         natural_width guest_cr0;
 285         natural_width guest_cr3;
 286         natural_width guest_cr4;
 287         natural_width guest_es_base;
 288         natural_width guest_cs_base;
 289         natural_width guest_ss_base;
 290         natural_width guest_ds_base;
 291         natural_width guest_fs_base;
 292         natural_width guest_gs_base;
 293         natural_width guest_ldtr_base;
 294         natural_width guest_tr_base;
 295         natural_width guest_gdtr_base;
 296         natural_width guest_idtr_base;
 297         natural_width guest_dr7;
 298         natural_width guest_rsp;
 299         natural_width guest_rip;
 300         natural_width guest_rflags;
 301         natural_width guest_pending_dbg_exceptions;
 302         natural_width guest_sysenter_esp;
 303         natural_width guest_sysenter_eip;
 304         natural_width host_cr0;
 305         natural_width host_cr3;
 306         natural_width host_cr4;
 307         natural_width host_fs_base;
 308         natural_width host_gs_base;
 309         natural_width host_tr_base;
 310         natural_width host_gdtr_base;
 311         natural_width host_idtr_base;
 312         natural_width host_ia32_sysenter_esp;
 313         natural_width host_ia32_sysenter_eip;
 314         natural_width host_rsp;
 315         natural_width host_rip;
 316         natural_width paddingl[8]; /* room for future expansion */
 317         u32 pin_based_vm_exec_control;
 318         u32 cpu_based_vm_exec_control;
 319         u32 exception_bitmap;
 320         u32 page_fault_error_code_mask;
 321         u32 page_fault_error_code_match;
 322         u32 cr3_target_count;
 323         u32 vm_exit_controls;
 324         u32 vm_exit_msr_store_count;
 325         u32 vm_exit_msr_load_count;
 326         u32 vm_entry_controls;
 327         u32 vm_entry_msr_load_count;
 328         u32 vm_entry_intr_info_field;
 329         u32 vm_entry_exception_error_code;
 330         u32 vm_entry_instruction_len;
 331         u32 tpr_threshold;
 332         u32 secondary_vm_exec_control;
 333         u32 vm_instruction_error;
 334         u32 vm_exit_reason;
 335         u32 vm_exit_intr_info;
 336         u32 vm_exit_intr_error_code;
 337         u32 idt_vectoring_info_field;
 338         u32 idt_vectoring_error_code;
 339         u32 vm_exit_instruction_len;
 340         u32 vmx_instruction_info;
 341         u32 guest_es_limit;
 342         u32 guest_cs_limit;
 343         u32 guest_ss_limit;
 344         u32 guest_ds_limit;
 345         u32 guest_fs_limit;
 346         u32 guest_gs_limit;
 347         u32 guest_ldtr_limit;
 348         u32 guest_tr_limit;
 349         u32 guest_gdtr_limit;
 350         u32 guest_idtr_limit;
 351         u32 guest_es_ar_bytes;
 352         u32 guest_cs_ar_bytes;
 353         u32 guest_ss_ar_bytes;
 354         u32 guest_ds_ar_bytes;
 355         u32 guest_fs_ar_bytes;
 356         u32 guest_gs_ar_bytes;
 357         u32 guest_ldtr_ar_bytes;
 358         u32 guest_tr_ar_bytes;
 359         u32 guest_interruptibility_info;
 360         u32 guest_activity_state;
 361         u32 guest_sysenter_cs;
 362         u32 host_ia32_sysenter_cs;
 363         u32 vmx_preemption_timer_value;
 364         u32 padding32[7]; /* room for future expansion */
 365         u16 virtual_processor_id;
 366         u16 posted_intr_nv;
 367         u16 guest_es_selector;
 368         u16 guest_cs_selector;
 369         u16 guest_ss_selector;
 370         u16 guest_ds_selector;
 371         u16 guest_fs_selector;
 372         u16 guest_gs_selector;
 373         u16 guest_ldtr_selector;
 374         u16 guest_tr_selector;
 375         u16 guest_intr_status;
 376         u16 guest_pml_index;
 377         u16 host_es_selector;
 378         u16 host_cs_selector;
 379         u16 host_ss_selector;
 380         u16 host_ds_selector;
 381         u16 host_fs_selector;
 382         u16 host_gs_selector;
 383         u16 host_tr_selector;
 384 };
 385
 386 /*
 387  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
 388  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
 389  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
 390  */
 391 #define VMCS12_REVISION 0x11e57ed0
 392
 393 /*
 394  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
 395  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
 396  * current implementation, 4K are reserved to avoid future complications.
 397  */
 398 #define VMCS12_SIZE 0x1000
 399
 400 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
 401 struct vmcs02_list {
 402         struct list_head list;
 403         gpa_t vmptr;
 404         struct loaded_vmcs vmcs02;
 405 };
 406
 407 /*
 408  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 409  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 410  */
 411 struct nested_vmx {
 412         /* Has the level1 guest done vmxon? */
 413         bool vmxon;
 414         gpa_t vmxon_ptr;
 415         bool pml_full;
 416
 417         /* The guest-physical address of the current VMCS L1 keeps for L2 */
 418         gpa_t current_vmptr;
 419         /*
 420          * Cache of the guest's VMCS, existing outside of guest memory.
 421          * Loaded from guest memory during VMPTRLD. Flushed to guest
 422          * memory during VMCLEAR and VMPTRLD.
 423          */
 424         struct vmcs12 *cached_vmcs12;
 425         /*
 426          * Indicates if the shadow vmcs must be updated with the
 427          * data hold by vmcs12
 428          */
 429         bool sync_shadow_vmcs;
 430
 431         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
 432         struct list_head vmcs02_pool;
 433         int vmcs02_num;
 434         bool change_vmcs01_virtual_x2apic_mode;
 435         /* L2 must run next, and mustn't decide to exit to L1. */
 436         bool nested_run_pending;
 437         /*
 438          * Guest pages referred to in vmcs02 with host-physical pointers, so
 439          * we must keep them pinned while L2 runs.
 440          */
 441         struct page *apic_access_page;
 442         struct page *virtual_apic_page;
 443         struct page *pi_desc_page;
 444         struct pi_desc *pi_desc;
 445         bool pi_pending;
 446         u16 posted_intr_nv;
 447
 448         unsigned long *msr_bitmap;
 449
 450         struct hrtimer preemption_timer;
 451         bool preemption_timer_expired;
 452
 453         /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
 454         u64 vmcs01_debugctl;
 455
 456         u16 vpid02;
 457         u16 last_vpid;
 458
 459         /*
 460          * We only store the "true" versions of the VMX capability MSRs. We
 461          * generate the "non-true" versions by setting the must-be-1 bits
 462          * according to the SDM.
 463          */
 464         u32 nested_vmx_procbased_ctls_low;
 465         u32 nested_vmx_procbased_ctls_high;
 466         u32 nested_vmx_secondary_ctls_low;
 467         u32 nested_vmx_secondary_ctls_high;
 468         u32 nested_vmx_pinbased_ctls_low;
 469         u32 nested_vmx_pinbased_ctls_high;
 470         u32 nested_vmx_exit_ctls_low;
 471         u32 nested_vmx_exit_ctls_high;
 472         u32 nested_vmx_entry_ctls_low;
 473         u32 nested_vmx_entry_ctls_high;
 474         u32 nested_vmx_misc_low;
 475         u32 nested_vmx_misc_high;
 476         u32 nested_vmx_ept_caps;
 477         u32 nested_vmx_vpid_caps;
 478         u64 nested_vmx_basic;
 479         u64 nested_vmx_cr0_fixed0;
 480         u64 nested_vmx_cr0_fixed1;
 481         u64 nested_vmx_cr4_fixed0;
 482         u64 nested_vmx_cr4_fixed1;
 483         u64 nested_vmx_vmcs_enum;
 484 };
 485
 486 #define POSTED_INTR_ON  0
 487 #define POSTED_INTR_SN  1
 488
 489 /* Posted-Interrupt Descriptor */
 490 struct pi_desc {
 491         u32 pir[8];     /* Posted interrupt requested */
 492         union {
 493                 struct {
 494                                 /* bit 256 - Outstanding Notification */
 495                         u16     on      : 1,
 496                                 /* bit 257 - Suppress Notification */
 497                                 sn      : 1,
 498                                 /* bit 271:258 - Reserved */
 499                                 rsvd_1  : 14;
 500                                 /* bit 279:272 - Notification Vector */
 501                         u8      nv;
 502                                 /* bit 287:280 - Reserved */
 503                         u8      rsvd_2;
 504                                 /* bit 319:288 - Notification Destination */
 505                         u32     ndst;
 506                 };
 507                 u64 control;
 508         };
 509         u32 rsvd[6];
 510 } __aligned(64);
 511
 512 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
 513 {
 514         return test_and_set_bit(POSTED_INTR_ON,
 515                         (unsigned long *)&pi_desc->control);
 516 }
 517
 518 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
 519 {
 520         return test_and_clear_bit(POSTED_INTR_ON,
 521                         (unsigned long *)&pi_desc->control);
 522 }
 523
 524 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
 525 {
 526         return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
 527 }
 528
 529 static inline void pi_clear_sn(struct pi_desc *pi_desc)
 530 {
 531         return clear_bit(POSTED_INTR_SN,
 532                         (unsigned long *)&pi_desc->control);
 533 }
 534
 535 static inline void pi_set_sn(struct pi_desc *pi_desc)
 536 {
 537         return set_bit(POSTED_INTR_SN,
 538                         (unsigned long *)&pi_desc->control);
 539 }
 540
 541 static inline void pi_clear_on(struct pi_desc *pi_desc)
 542 {
 543         clear_bit(POSTED_INTR_ON,
 544                   (unsigned long *)&pi_desc->control);
 545 }
 546
 547 static inline int pi_test_on(struct pi_desc *pi_desc)
 548 {
 549         return test_bit(POSTED_INTR_ON,
 550                         (unsigned long *)&pi_desc->control);
 551 }
 552
 553 static inline int pi_test_sn(struct pi_desc *pi_desc)
 554 {
 555         return test_bit(POSTED_INTR_SN,
 556                         (unsigned long *)&pi_desc->control);
 557 }
 558
 559 struct vcpu_vmx {
 560         struct kvm_vcpu       vcpu;
 561         unsigned long         host_rsp;
 562         u8                    fail;
 563         u32                   exit_intr_info;
 564         u32                   idt_vectoring_info;
 565         ulong                 rflags;
 566         struct shared_msr_entry *guest_msrs;
 567         int                   nmsrs;
 568         int                   save_nmsrs;
 569         unsigned long         host_idt_base;
 570 #ifdef CONFIG_X86_64
 571         u64                   msr_host_kernel_gs_base;
 572         u64                   msr_guest_kernel_gs_base;
 573 #endif
 574         u32 vm_entry_controls_shadow;
 575         u32 vm_exit_controls_shadow;
 576         /*
 577          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 578          * non-nested (L1) guest, it always points to vmcs01. For a nested
 579          * guest (L2), it points to a different VMCS.
 580          */
 581         struct loaded_vmcs    vmcs01;
 582         struct loaded_vmcs   *loaded_vmcs;
 583         bool                  __launched; /* temporary, used in vmx_vcpu_run */
 584         struct msr_autoload {
 585                 unsigned nr;
 586                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 587                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 588         } msr_autoload;
 589         struct {
 590                 int           loaded;
 591                 u16           fs_sel, gs_sel, ldt_sel;
 592 #ifdef CONFIG_X86_64
 593                 u16           ds_sel, es_sel;
 594 #endif
 595                 int           gs_ldt_reload_needed;
 596                 int           fs_reload_needed;
 597                 u64           msr_host_bndcfgs;
 598                 unsigned long vmcs_host_cr3;    /* May not match real cr3 */
 599                 unsigned long vmcs_host_cr4;    /* May not match real cr4 */
 600         } host_state;
 601         struct {
 602                 int vm86_active;
 603                 ulong save_rflags;
 604                 struct kvm_segment segs[8];
 605         } rmode;
 606         struct {
 607                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 608                 struct kvm_save_segment {
 609                         u16 selector;
 610                         unsigned long base;
 611                         u32 limit;
 612                         u32 ar;
 613                 } seg[8];
 614         } segment_cache;
 615         int vpid;
 616         bool emulation_required;
 617
 618         u32 exit_reason;
 619
 620         /* Posted interrupt descriptor */
 621         struct pi_desc pi_desc;
 622
 623         /* Support for a guest hypervisor (nested VMX) */
 624         struct nested_vmx nested;
 625
 626         /* Dynamic PLE window. */
 627         int ple_window;
 628         bool ple_window_dirty;
 629
 630         /* Support for PML */
 631 #define PML_ENTITY_NUM          512
 632         struct page *pml_pg;
 633
 634         /* apic deadline value in host tsc */
 635         u64 hv_deadline_tsc;
 636
 637         u64 current_tsc_ratio;
 638
 639         bool guest_pkru_valid;
 640         u32 guest_pkru;
 641         u32 host_pkru;
 642
 643         /*
 644          * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
 645          * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
 646          * in msr_ia32_feature_control_valid_bits.
 647          */
 648         u64 msr_ia32_feature_control;
 649         u64 msr_ia32_feature_control_valid_bits;
 650 };
 651
 652 enum segment_cache_field {
 653         SEG_FIELD_SEL = 0,
 654         SEG_FIELD_BASE = 1,
 655         SEG_FIELD_LIMIT = 2,
 656         SEG_FIELD_AR = 3,
 657
 658         SEG_FIELD_NR = 4
 659 };
 660
 661 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 662 {
 663         return container_of(vcpu, struct vcpu_vmx, vcpu);
 664 }
 665
 666 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
 667 {
 668         return &(to_vmx(vcpu)->pi_desc);
 669 }
 670
 671 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
 672 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
 673 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
 674                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
 675
 676
 677 static unsigned long shadow_read_only_fields[] = {
 678         /*
 679          * We do NOT shadow fields that are modified when L0
 680          * traps and emulates any vmx instruction (e.g. VMPTRLD,
 681          * VMXON...) executed by L1.
 682          * For example, VM_INSTRUCTION_ERROR is read
 683          * by L1 if a vmx instruction fails (part of the error path).
 684          * Note the code assumes this logic. If for some reason
 685          * we start shadowing these fields then we need to
 686          * force a shadow sync when L0 emulates vmx instructions
 687          * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
 688          * by nested_vmx_failValid)
 689          */
 690         VM_EXIT_REASON,
 691         VM_EXIT_INTR_INFO,
 692         VM_EXIT_INSTRUCTION_LEN,
 693         IDT_VECTORING_INFO_FIELD,
 694         IDT_VECTORING_ERROR_CODE,
 695         VM_EXIT_INTR_ERROR_CODE,
 696         EXIT_QUALIFICATION,
 697         GUEST_LINEAR_ADDRESS,
 698         GUEST_PHYSICAL_ADDRESS
 699 };
 700 static int max_shadow_read_only_fields =
 701         ARRAY_SIZE(shadow_read_only_fields);
 702
 703 static unsigned long shadow_read_write_fields[] = {
 704         TPR_THRESHOLD,
 705         GUEST_RIP,
 706         GUEST_RSP,
 707         GUEST_CR0,
 708         GUEST_CR3,
 709         GUEST_CR4,
 710         GUEST_INTERRUPTIBILITY_INFO,
 711         GUEST_RFLAGS,
 712         GUEST_CS_SELECTOR,
 713         GUEST_CS_AR_BYTES,
 714         GUEST_CS_LIMIT,
 715         GUEST_CS_BASE,
 716         GUEST_ES_BASE,
 717         GUEST_BNDCFGS,
 718         CR0_GUEST_HOST_MASK,
 719         CR0_READ_SHADOW,
 720         CR4_READ_SHADOW,
 721         TSC_OFFSET,
 722         EXCEPTION_BITMAP,
 723         CPU_BASED_VM_EXEC_CONTROL,
 724         VM_ENTRY_EXCEPTION_ERROR_CODE,
 725         VM_ENTRY_INTR_INFO_FIELD,
 726         VM_ENTRY_INSTRUCTION_LEN,
 727         VM_ENTRY_EXCEPTION_ERROR_CODE,
 728         HOST_FS_BASE,
 729         HOST_GS_BASE,
 730         HOST_FS_SELECTOR,
 731         HOST_GS_SELECTOR
 732 };
 733 static int max_shadow_read_write_fields =
 734         ARRAY_SIZE(shadow_read_write_fields);
 735
 736 static const unsigned short vmcs_field_to_offset_table[] = {
 737         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 738         FIELD(POSTED_INTR_NV, posted_intr_nv),
 739         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
 740         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
 741         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
 742         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
 743         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
 744         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
 745         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
 746         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
 747         FIELD(GUEST_INTR_STATUS, guest_intr_status),
 748         FIELD(GUEST_PML_INDEX, guest_pml_index),
 749         FIELD(HOST_ES_SELECTOR, host_es_selector),
 750         FIELD(HOST_CS_SELECTOR, host_cs_selector),
 751         FIELD(HOST_SS_SELECTOR, host_ss_selector),
 752         FIELD(HOST_DS_SELECTOR, host_ds_selector),
 753         FIELD(HOST_FS_SELECTOR, host_fs_selector),
 754         FIELD(HOST_GS_SELECTOR, host_gs_selector),
 755         FIELD(HOST_TR_SELECTOR, host_tr_selector),
 756         FIELD64(IO_BITMAP_A, io_bitmap_a),
 757         FIELD64(IO_BITMAP_B, io_bitmap_b),
 758         FIELD64(MSR_BITMAP, msr_bitmap),
 759         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
 760         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
 761         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
 762         FIELD64(TSC_OFFSET, tsc_offset),
 763         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
 764         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
 765         FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
 766         FIELD64(EPT_POINTER, ept_pointer),
 767         FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
 768         FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
 769         FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
 770         FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
 771         FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
 772         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
 773         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
 774         FIELD64(PML_ADDRESS, pml_address),
 775         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
 776         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
 777         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
 778         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
 779         FIELD64(GUEST_PDPTR0, guest_pdptr0),
 780         FIELD64(GUEST_PDPTR1, guest_pdptr1),
 781         FIELD64(GUEST_PDPTR2, guest_pdptr2),
 782         FIELD64(GUEST_PDPTR3, guest_pdptr3),
 783         FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
 784         FIELD64(HOST_IA32_PAT, host_ia32_pat),
 785         FIELD64(HOST_IA32_EFER, host_ia32_efer),
 786         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
 787         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
 788         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
 789         FIELD(EXCEPTION_BITMAP, exception_bitmap),
 790         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
 791         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
 792         FIELD(CR3_TARGET_COUNT, cr3_target_count),
 793         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
 794         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
 795         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
 796         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
 797         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
 798         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
 799         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
 800         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
 801         FIELD(TPR_THRESHOLD, tpr_threshold),
 802         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
 803         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
 804         FIELD(VM_EXIT_REASON, vm_exit_reason),
 805         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
 806         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
 807         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
 808         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
 809         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
 810         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
 811         FIELD(GUEST_ES_LIMIT, guest_es_limit),
 812         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
 813         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
 814         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
 815         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
 816         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
 817         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
 818         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
 819         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
 820         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
 821         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
 822         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
 823         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
 824         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
 825         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
 826         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
 827         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
 828         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
 829         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
 830         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
 831         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
 832         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
 833         FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
 834         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
 835         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
 836         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
 837         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
 838         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
 839         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
 840         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
 841         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
 842         FIELD(EXIT_QUALIFICATION, exit_qualification),
 843         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
 844         FIELD(GUEST_CR0, guest_cr0),
 845         FIELD(GUEST_CR3, guest_cr3),
 846         FIELD(GUEST_CR4, guest_cr4),
 847         FIELD(GUEST_ES_BASE, guest_es_base),
 848         FIELD(GUEST_CS_BASE, guest_cs_base),
 849         FIELD(GUEST_SS_BASE, guest_ss_base),
 850         FIELD(GUEST_DS_BASE, guest_ds_base),
 851         FIELD(GUEST_FS_BASE, guest_fs_base),
 852         FIELD(GUEST_GS_BASE, guest_gs_base),
 853         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
 854         FIELD(GUEST_TR_BASE, guest_tr_base),
 855         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
 856         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
 857         FIELD(GUEST_DR7, guest_dr7),
 858         FIELD(GUEST_RSP, guest_rsp),
 859         FIELD(GUEST_RIP, guest_rip),
 860         FIELD(GUEST_RFLAGS, guest_rflags),
 861         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
 862         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
 863         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
 864         FIELD(HOST_CR0, host_cr0),
 865         FIELD(HOST_CR3, host_cr3),
 866         FIELD(HOST_CR4, host_cr4),
 867         FIELD(HOST_FS_BASE, host_fs_base),
 868         FIELD(HOST_GS_BASE, host_gs_base),
 869         FIELD(HOST_TR_BASE, host_tr_base),
 870         FIELD(HOST_GDTR_BASE, host_gdtr_base),
 871         FIELD(HOST_IDTR_BASE, host_idtr_base),
 872         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
 873         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
 874         FIELD(HOST_RSP, host_rsp),
 875         FIELD(HOST_RIP, host_rip),
 876 };
 877
 878 static inline short vmcs_field_to_offset(unsigned long field)
 879 {
 880         BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
 881
 882         if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
 883             vmcs_field_to_offset_table[field] == 0)
 884                 return -ENOENT;
 885
 886         return vmcs_field_to_offset_table[field];
 887 }
 888
 889 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
 890 {
 891         return to_vmx(vcpu)->nested.cached_vmcs12;
 892 }
 893
 894 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
 895 {
 896         struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
 897         if (is_error_page(page))
 898                 return NULL;
 899
 900         return page;
 901 }
 902
 903 static void nested_release_page(struct page *page)
 904 {
 905         kvm_release_page_dirty(page);
 906 }
 907
 908 static void nested_release_page_clean(struct page *page)
 909 {
 910         kvm_release_page_clean(page);
 911 }
 912
 913 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
 914 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
 915 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
 916 static bool vmx_xsaves_supported(void);
 917 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 918 static void vmx_set_segment(struct kvm_vcpu *vcpu,
 919                             struct kvm_segment *var, int seg);
 920 static void vmx_get_segment(struct kvm_vcpu *vcpu,
 921                             struct kvm_segment *var, int seg);
 922 static bool guest_state_valid(struct kvm_vcpu *vcpu);
 923 static u32 vmx_segment_access_rights(struct kvm_segment *var);
 924 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
 925 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
 926 static int alloc_identity_pagetable(struct kvm *kvm);
 927 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
 928 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
 929 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
 930                                             u16 error_code);
 931
 932 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 933 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 934 /*
 935  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
 936  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
 937  */
 938 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
 939
 940 /*
 941  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
 942  * can find which vCPU should be waken up.
 943  */
 944 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
 945 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
 946
 947 enum {
 948         VMX_IO_BITMAP_A,
 949         VMX_IO_BITMAP_B,
 950         VMX_MSR_BITMAP_LEGACY,
 951         VMX_MSR_BITMAP_LONGMODE,
 952         VMX_MSR_BITMAP_LEGACY_X2APIC_APICV,
 953         VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV,
 954         VMX_MSR_BITMAP_LEGACY_X2APIC,
 955         VMX_MSR_BITMAP_LONGMODE_X2APIC,
 956         VMX_VMREAD_BITMAP,
 957         VMX_VMWRITE_BITMAP,
 958         VMX_BITMAP_NR
 959 };
 960
 961 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
 962
 963 #define vmx_io_bitmap_a                      (vmx_bitmap[VMX_IO_BITMAP_A])
 964 #define vmx_io_bitmap_b                      (vmx_bitmap[VMX_IO_BITMAP_B])
 965 #define vmx_msr_bitmap_legacy                (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
 966 #define vmx_msr_bitmap_longmode              (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
 967 #define vmx_msr_bitmap_legacy_x2apic_apicv   (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
 968 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
 969 #define vmx_msr_bitmap_legacy_x2apic         (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
 970 #define vmx_msr_bitmap_longmode_x2apic       (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
 971 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
 972 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
 973
 974 static bool cpu_has_load_ia32_efer;
 975 static bool cpu_has_load_perf_global_ctrl;
 976
 977 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 978 static DEFINE_SPINLOCK(vmx_vpid_lock);
 979
 980 static struct vmcs_config {
 981         int size;
 982         int order;
 983         u32 basic_cap;
 984         u32 revision_id;
 985         u32 pin_based_exec_ctrl;
 986         u32 cpu_based_exec_ctrl;
 987         u32 cpu_based_2nd_exec_ctrl;
 988         u32 vmexit_ctrl;
 989         u32 vmentry_ctrl;
 990 } vmcs_config;
 991
 992 static struct vmx_capability {
 993         u32 ept;
 994         u32 vpid;
 995 } vmx_capability;
 996
 997 #define VMX_SEGMENT_FIELD(seg)                                  \
 998         [VCPU_SREG_##seg] = {                                   \
 999                 .selector = GUEST_##seg##_SELECTOR,             \
1000                 .base = GUEST_##seg##_BASE,                     \
1001                 .limit = GUEST_##seg##_LIMIT,                   \
1002                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
1003         }
1004
1005 static const struct kvm_vmx_segment_field {
1006         unsigned selector;
1007         unsigned base;
1008         unsigned limit;
1009         unsigned ar_bytes;
1010 } kvm_vmx_segment_fields[] = {
1011         VMX_SEGMENT_FIELD(CS),
1012         VMX_SEGMENT_FIELD(DS),
1013         VMX_SEGMENT_FIELD(ES),
1014         VMX_SEGMENT_FIELD(FS),
1015         VMX_SEGMENT_FIELD(GS),
1016         VMX_SEGMENT_FIELD(SS),
1017         VMX_SEGMENT_FIELD(TR),
1018         VMX_SEGMENT_FIELD(LDTR),
1019 };
1020
1021 static u64 host_efer;
1022
1023 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1024
1025 /*
1026  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1027  * away by decrementing the array size.
1028  */
1029 static const u32 vmx_msr_index[] = {
1030 #ifdef CONFIG_X86_64
1031         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1032 #endif
1033         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1034 };
1035
1036 static inline bool is_exception_n(u32 intr_info, u8 vector)
1037 {
1038         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1039                              INTR_INFO_VALID_MASK)) ==
1040                 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1041 }
1042
1043 static inline bool is_debug(u32 intr_info)
1044 {
1045         return is_exception_n(intr_info, DB_VECTOR);
1046 }
1047
1048 static inline bool is_breakpoint(u32 intr_info)
1049 {
1050         return is_exception_n(intr_info, BP_VECTOR);
1051 }
1052
1053 static inline bool is_page_fault(u32 intr_info)
1054 {
1055         return is_exception_n(intr_info, PF_VECTOR);
1056 }
1057
1058 static inline bool is_no_device(u32 intr_info)
1059 {
1060         return is_exception_n(intr_info, NM_VECTOR);
1061 }
1062
1063 static inline bool is_invalid_opcode(u32 intr_info)
1064 {
1065         return is_exception_n(intr_info, UD_VECTOR);
1066 }
1067
1068 static inline bool is_external_interrupt(u32 intr_info)
1069 {
1070         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1071                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1072 }
1073
1074 static inline bool is_machine_check(u32 intr_info)
1075 {
1076         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1077                              INTR_INFO_VALID_MASK)) ==
1078                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1079 }
1080
1081 static inline bool cpu_has_vmx_msr_bitmap(void)
1082 {
1083         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1084 }
1085
1086 static inline bool cpu_has_vmx_tpr_shadow(void)
1087 {
1088         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1089 }
1090
1091 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1092 {
1093         return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1094 }
1095
1096 static inline bool cpu_has_secondary_exec_ctrls(void)
1097 {
1098         return vmcs_config.cpu_based_exec_ctrl &
1099                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1100 }
1101
1102 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1103 {
1104         return vmcs_config.cpu_based_2nd_exec_ctrl &
1105                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1106 }
1107
1108 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1109 {
1110         return vmcs_config.cpu_based_2nd_exec_ctrl &
1111                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1112 }
1113
1114 static inline bool cpu_has_vmx_apic_register_virt(void)
1115 {
1116         return vmcs_config.cpu_based_2nd_exec_ctrl &
1117                 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1118 }
1119
1120 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1121 {
1122         return vmcs_config.cpu_based_2nd_exec_ctrl &
1123                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1124 }
1125
1126 /*
1127  * Comment's format: document - errata name - stepping - processor name.
1128  * Refer from
1129  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1130  */
1131 static u32 vmx_preemption_cpu_tfms[] = {
1132 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
1133 0x000206E6,
1134 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
1135 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1136 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1137 0x00020652,
1138 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1139 0x00020655,
1140 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
1141 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
1142 /*
1143  * 320767.pdf - AAP86  - B1 -
1144  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1145  */
1146 0x000106E5,
1147 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1148 0x000106A0,
1149 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1150 0x000106A1,
1151 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1152 0x000106A4,
1153  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1154  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1155  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1156 0x000106A5,
1157 };
1158
1159 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1160 {
1161         u32 eax = cpuid_eax(0x00000001), i;
1162
1163         /* Clear the reserved bits */
1164         eax &= ~(0x3U << 14 | 0xfU << 28);
1165         for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1166                 if (eax == vmx_preemption_cpu_tfms[i])
1167                         return true;
1168
1169         return false;
1170 }
1171
1172 static inline bool cpu_has_vmx_preemption_timer(void)
1173 {
1174         return vmcs_config.pin_based_exec_ctrl &
1175                 PIN_BASED_VMX_PREEMPTION_TIMER;
1176 }
1177
1178 static inline bool cpu_has_vmx_posted_intr(void)
1179 {
1180         return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1181                 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1182 }
1183
1184 static inline bool cpu_has_vmx_apicv(void)
1185 {
1186         return cpu_has_vmx_apic_register_virt() &&
1187                 cpu_has_vmx_virtual_intr_delivery() &&
1188                 cpu_has_vmx_posted_intr();
1189 }
1190
1191 static inline bool cpu_has_vmx_flexpriority(void)
1192 {
1193         return cpu_has_vmx_tpr_shadow() &&
1194                 cpu_has_vmx_virtualize_apic_accesses();
1195 }
1196
1197 static inline bool cpu_has_vmx_ept_execute_only(void)
1198 {
1199         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1200 }
1201
1202 static inline bool cpu_has_vmx_ept_2m_page(void)
1203 {
1204         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1205 }
1206
1207 static inline bool cpu_has_vmx_ept_1g_page(void)
1208 {
1209         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1210 }
1211
1212 static inline bool cpu_has_vmx_ept_4levels(void)
1213 {
1214         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1215 }
1216
1217 static inline bool cpu_has_vmx_ept_ad_bits(void)
1218 {
1219         return vmx_capability.ept & VMX_EPT_AD_BIT;
1220 }
1221
1222 static inline bool cpu_has_vmx_invept_context(void)
1223 {
1224         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1225 }
1226
1227 static inline bool cpu_has_vmx_invept_global(void)
1228 {
1229         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1230 }
1231
1232 static inline bool cpu_has_vmx_invvpid_single(void)
1233 {
1234         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1235 }
1236
1237 static inline bool cpu_has_vmx_invvpid_global(void)
1238 {
1239         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1240 }
1241
1242 static inline bool cpu_has_vmx_invvpid(void)
1243 {
1244         return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1245 }
1246
1247 static inline bool cpu_has_vmx_ept(void)
1248 {
1249         return vmcs_config.cpu_based_2nd_exec_ctrl &
1250                 SECONDARY_EXEC_ENABLE_EPT;
1251 }
1252
1253 static inline bool cpu_has_vmx_unrestricted_guest(void)
1254 {
1255         return vmcs_config.cpu_based_2nd_exec_ctrl &
1256                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1257 }
1258
1259 static inline bool cpu_has_vmx_ple(void)
1260 {
1261         return vmcs_config.cpu_based_2nd_exec_ctrl &
1262                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1263 }
1264
1265 static inline bool cpu_has_vmx_basic_inout(void)
1266 {
1267         return  (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1268 }
1269
1270 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1271 {
1272         return flexpriority_enabled && lapic_in_kernel(vcpu);
1273 }
1274
1275 static inline bool cpu_has_vmx_vpid(void)
1276 {
1277         return vmcs_config.cpu_based_2nd_exec_ctrl &
1278                 SECONDARY_EXEC_ENABLE_VPID;
1279 }
1280
1281 static inline bool cpu_has_vmx_rdtscp(void)
1282 {
1283         return vmcs_config.cpu_based_2nd_exec_ctrl &
1284                 SECONDARY_EXEC_RDTSCP;
1285 }
1286
1287 static inline bool cpu_has_vmx_invpcid(void)
1288 {
1289         return vmcs_config.cpu_based_2nd_exec_ctrl &
1290                 SECONDARY_EXEC_ENABLE_INVPCID;
1291 }
1292
1293 static inline bool cpu_has_vmx_wbinvd_exit(void)
1294 {
1295         return vmcs_config.cpu_based_2nd_exec_ctrl &
1296                 SECONDARY_EXEC_WBINVD_EXITING;
1297 }
1298
1299 static inline bool cpu_has_vmx_shadow_vmcs(void)
1300 {
1301         u64 vmx_msr;
1302         rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1303         /* check if the cpu supports writing r/o exit information fields */
1304         if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1305                 return false;
1306
1307         return vmcs_config.cpu_based_2nd_exec_ctrl &
1308                 SECONDARY_EXEC_SHADOW_VMCS;
1309 }
1310
1311 static inline bool cpu_has_vmx_pml(void)
1312 {
1313         return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1314 }
1315
1316 static inline bool cpu_has_vmx_tsc_scaling(void)
1317 {
1318         return vmcs_config.cpu_based_2nd_exec_ctrl &
1319                 SECONDARY_EXEC_TSC_SCALING;
1320 }
1321
1322 static inline bool report_flexpriority(void)
1323 {
1324         return flexpriority_enabled;
1325 }
1326
1327 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1328 {
1329         return vmx_misc_cr3_count(to_vmx(vcpu)->nested.nested_vmx_misc_low);
1330 }
1331
1332 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1333 {
1334         return vmcs12->cpu_based_vm_exec_control & bit;
1335 }
1336
1337 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1338 {
1339         return (vmcs12->cpu_based_vm_exec_control &
1340                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1341                 (vmcs12->secondary_vm_exec_control & bit);
1342 }
1343
1344 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1345 {
1346         return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1347 }
1348
1349 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1350 {
1351         return vmcs12->pin_based_vm_exec_control &
1352                 PIN_BASED_VMX_PREEMPTION_TIMER;
1353 }
1354
1355 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1356 {
1357         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1358 }
1359
1360 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1361 {
1362         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1363                 vmx_xsaves_supported();
1364 }
1365
1366 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1367 {
1368         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1369 }
1370
1371 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1372 {
1373         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1374 }
1375
1376 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1377 {
1378         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1379 }
1380
1381 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1382 {
1383         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1384 }
1385
1386 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1387 {
1388         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1389 }
1390
1391 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1392 {
1393         return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1394 }
1395
1396 static inline bool is_nmi(u32 intr_info)
1397 {
1398         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1399                 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1400 }
1401
1402 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1403                               u32 exit_intr_info,
1404                               unsigned long exit_qualification);
1405 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1406                         struct vmcs12 *vmcs12,
1407                         u32 reason, unsigned long qualification);
1408
1409 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1410 {
1411         int i;
1412
1413         for (i = 0; i < vmx->nmsrs; ++i)
1414                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1415                         return i;
1416         return -1;
1417 }
1418
1419 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1420 {
1421     struct {
1422         u64 vpid : 16;
1423         u64 rsvd : 48;
1424         u64 gva;
1425     } operand = { vpid, 0, gva };
1426
1427     asm volatile (__ex(ASM_VMX_INVVPID)
1428                   /* CF==1 or ZF==1 --> rc = -1 */
1429                   "; ja 1f ; ud2 ; 1:"
1430                   : : "a"(&operand), "c"(ext) : "cc", "memory");
1431 }
1432
1433 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1434 {
1435         struct {
1436                 u64 eptp, gpa;
1437         } operand = {eptp, gpa};
1438
1439         asm volatile (__ex(ASM_VMX_INVEPT)
1440                         /* CF==1 or ZF==1 --> rc = -1 */
1441                         "; ja 1f ; ud2 ; 1:\n"
1442                         : : "a" (&operand), "c" (ext) : "cc", "memory");
1443 }
1444
1445 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1446 {
1447         int i;
1448
1449         i = __find_msr_index(vmx, msr);
1450         if (i >= 0)
1451                 return &vmx->guest_msrs[i];
1452         return NULL;
1453 }
1454
1455 static void vmcs_clear(struct vmcs *vmcs)
1456 {
1457         u64 phys_addr = __pa(vmcs);
1458         u8 error;
1459
1460         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1461                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1462                       : "cc", "memory");
1463         if (error)
1464                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1465                        vmcs, phys_addr);
1466 }
1467
1468 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1469 {
1470         vmcs_clear(loaded_vmcs->vmcs);
1471         if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1472                 vmcs_clear(loaded_vmcs->shadow_vmcs);
1473         loaded_vmcs->cpu = -1;
1474         loaded_vmcs->launched = 0;
1475 }
1476
1477 static void vmcs_load(struct vmcs *vmcs)
1478 {
1479         u64 phys_addr = __pa(vmcs);
1480         u8 error;
1481
1482         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1483                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1484                         : "cc", "memory");
1485         if (error)
1486                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1487                        vmcs, phys_addr);
1488 }
1489
1490 #ifdef CONFIG_KEXEC_CORE
1491 /*
1492  * This bitmap is used to indicate whether the vmclear
1493  * operation is enabled on all cpus. All disabled by
1494  * default.
1495  */
1496 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1497
1498 static inline void crash_enable_local_vmclear(int cpu)
1499 {
1500         cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1501 }
1502
1503 static inline void crash_disable_local_vmclear(int cpu)
1504 {
1505         cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1506 }
1507
1508 static inline int crash_local_vmclear_enabled(int cpu)
1509 {
1510         return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1511 }
1512
1513 static void crash_vmclear_local_loaded_vmcss(void)
1514 {
1515         int cpu = raw_smp_processor_id();
1516         struct loaded_vmcs *v;
1517
1518         if (!crash_local_vmclear_enabled(cpu))
1519                 return;
1520
1521         list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1522                             loaded_vmcss_on_cpu_link)
1523                 vmcs_clear(v->vmcs);
1524 }
1525 #else
1526 static inline void crash_enable_local_vmclear(int cpu) { }
1527 static inline void crash_disable_local_vmclear(int cpu) { }
1528 #endif /* CONFIG_KEXEC_CORE */
1529
1530 static void __loaded_vmcs_clear(void *arg)
1531 {
1532         struct loaded_vmcs *loaded_vmcs = arg;
1533         int cpu = raw_smp_processor_id();
1534
1535         if (loaded_vmcs->cpu != cpu)
1536                 return; /* vcpu migration can race with cpu offline */
1537         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1538                 per_cpu(current_vmcs, cpu) = NULL;
1539         crash_disable_local_vmclear(cpu);
1540         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1541
1542         /*
1543          * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1544          * is before setting loaded_vmcs->vcpu to -1 which is done in
1545          * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1546          * then adds the vmcs into percpu list before it is deleted.
1547          */
1548         smp_wmb();
1549
1550         loaded_vmcs_init(loaded_vmcs);
1551         crash_enable_local_vmclear(cpu);
1552 }
1553
1554 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1555 {
1556         int cpu = loaded_vmcs->cpu;
1557
1558         if (cpu != -1)
1559                 smp_call_function_single(cpu,
1560                          __loaded_vmcs_clear, loaded_vmcs, 1);
1561 }
1562
1563 static inline void vpid_sync_vcpu_single(int vpid)
1564 {
1565         if (vpid == 0)
1566                 return;
1567
1568         if (cpu_has_vmx_invvpid_single())
1569                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1570 }
1571
1572 static inline void vpid_sync_vcpu_global(void)
1573 {
1574         if (cpu_has_vmx_invvpid_global())
1575                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1576 }
1577
1578 static inline void vpid_sync_context(int vpid)
1579 {
1580         if (cpu_has_vmx_invvpid_single())
1581                 vpid_sync_vcpu_single(vpid);
1582         else
1583                 vpid_sync_vcpu_global();
1584 }
1585
1586 static inline void ept_sync_global(void)
1587 {
1588         if (cpu_has_vmx_invept_global())
1589                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1590 }
1591
1592 static inline void ept_sync_context(u64 eptp)
1593 {
1594         if (enable_ept) {
1595                 if (cpu_has_vmx_invept_context())
1596                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1597                 else
1598                         ept_sync_global();
1599         }
1600 }
1601
1602 static __always_inline void vmcs_check16(unsigned long field)
1603 {
1604         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1605                          "16-bit accessor invalid for 64-bit field");
1606         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1607                          "16-bit accessor invalid for 64-bit high field");
1608         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1609                          "16-bit accessor invalid for 32-bit high field");
1610         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1611                          "16-bit accessor invalid for natural width field");
1612 }
1613
1614 static __always_inline void vmcs_check32(unsigned long field)
1615 {
1616         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1617                          "32-bit accessor invalid for 16-bit field");
1618         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1619                          "32-bit accessor invalid for natural width field");
1620 }
1621
1622 static __always_inline void vmcs_check64(unsigned long field)
1623 {
1624         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1625                          "64-bit accessor invalid for 16-bit field");
1626         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1627                          "64-bit accessor invalid for 64-bit high field");
1628         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1629                          "64-bit accessor invalid for 32-bit field");
1630         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1631                          "64-bit accessor invalid for natural width field");
1632 }
1633
1634 static __always_inline void vmcs_checkl(unsigned long field)
1635 {
1636         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1637                          "Natural width accessor invalid for 16-bit field");
1638         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1639                          "Natural width accessor invalid for 64-bit field");
1640         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1641                          "Natural width accessor invalid for 64-bit high field");
1642         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1643                          "Natural width accessor invalid for 32-bit field");
1644 }
1645
1646 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1647 {
1648         unsigned long value;
1649
1650         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1651                       : "=a"(value) : "d"(field) : "cc");
1652         return value;
1653 }
1654
1655 static __always_inline u16 vmcs_read16(unsigned long field)
1656 {
1657         vmcs_check16(field);
1658         return __vmcs_readl(field);
1659 }
1660
1661 static __always_inline u32 vmcs_read32(unsigned long field)
1662 {
1663         vmcs_check32(field);
1664         return __vmcs_readl(field);
1665 }
1666
1667 static __always_inline u64 vmcs_read64(unsigned long field)
1668 {
1669         vmcs_check64(field);
1670 #ifdef CONFIG_X86_64
1671         return __vmcs_readl(field);
1672 #else
1673         return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1674 #endif
1675 }
1676
1677 static __always_inline unsigned long vmcs_readl(unsigned long field)
1678 {
1679         vmcs_checkl(field);
1680         return __vmcs_readl(field);
1681 }
1682
1683 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1684 {
1685         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1686                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1687         dump_stack();
1688 }
1689
1690 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1691 {
1692         u8 error;
1693
1694         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1695                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1696         if (unlikely(error))
1697                 vmwrite_error(field, value);
1698 }
1699
1700 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1701 {
1702         vmcs_check16(field);
1703         __vmcs_writel(field, value);
1704 }
1705
1706 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1707 {
1708         vmcs_check32(field);
1709         __vmcs_writel(field, value);
1710 }
1711
1712 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1713 {
1714         vmcs_check64(field);
1715         __vmcs_writel(field, value);
1716 #ifndef CONFIG_X86_64
1717         asm volatile ("");
1718         __vmcs_writel(field+1, value >> 32);
1719 #endif
1720 }
1721
1722 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1723 {
1724         vmcs_checkl(field);
1725         __vmcs_writel(field, value);
1726 }
1727
1728 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1729 {
1730         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1731                          "vmcs_clear_bits does not support 64-bit fields");
1732         __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1733 }
1734
1735 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1736 {
1737         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1738                          "vmcs_set_bits does not support 64-bit fields");
1739         __vmcs_writel(field, __vmcs_readl(field) | mask);
1740 }
1741
1742 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1743 {
1744         vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1745 }
1746
1747 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1748 {
1749         vmcs_write32(VM_ENTRY_CONTROLS, val);
1750         vmx->vm_entry_controls_shadow = val;
1751 }
1752
1753 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1754 {
1755         if (vmx->vm_entry_controls_shadow != val)
1756                 vm_entry_controls_init(vmx, val);
1757 }
1758
1759 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1760 {
1761         return vmx->vm_entry_controls_shadow;
1762 }
1763
1764
1765 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1766 {
1767         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1768 }
1769
1770 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1771 {
1772         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1773 }
1774
1775 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1776 {
1777         vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1778 }
1779
1780 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1781 {
1782         vmcs_write32(VM_EXIT_CONTROLS, val);
1783         vmx->vm_exit_controls_shadow = val;
1784 }
1785
1786 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1787 {
1788         if (vmx->vm_exit_controls_shadow != val)
1789                 vm_exit_controls_init(vmx, val);
1790 }
1791
1792 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1793 {
1794         return vmx->vm_exit_controls_shadow;
1795 }
1796
1797
1798 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1799 {
1800         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1801 }
1802
1803 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1804 {
1805         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1806 }
1807
1808 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1809 {
1810         vmx->segment_cache.bitmask = 0;
1811 }
1812
1813 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1814                                        unsigned field)
1815 {
1816         bool ret;
1817         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1818
1819         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1820                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1821                 vmx->segment_cache.bitmask = 0;
1822         }
1823         ret = vmx->segment_cache.bitmask & mask;
1824         vmx->segment_cache.bitmask |= mask;
1825         return ret;
1826 }
1827
1828 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1829 {
1830         u16 *p = &vmx->segment_cache.seg[seg].selector;
1831
1832         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1833                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1834         return *p;
1835 }
1836
1837 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1838 {
1839         ulong *p = &vmx->segment_cache.seg[seg].base;
1840
1841         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1842                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1843         return *p;
1844 }
1845
1846 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1847 {
1848         u32 *p = &vmx->segment_cache.seg[seg].limit;
1849
1850         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1851                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1852         return *p;
1853 }
1854
1855 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1856 {
1857         u32 *p = &vmx->segment_cache.seg[seg].ar;
1858
1859         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1860                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1861         return *p;
1862 }
1863
1864 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1865 {
1866         u32 eb;
1867
1868         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1869              (1u << DB_VECTOR) | (1u << AC_VECTOR);
1870         if ((vcpu->guest_debug &
1871              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1872             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1873                 eb |= 1u << BP_VECTOR;
1874         if (to_vmx(vcpu)->rmode.vm86_active)
1875                 eb = ~0;
1876         if (enable_ept)
1877                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1878
1879         /* When we are running a nested L2 guest and L1 specified for it a
1880          * certain exception bitmap, we must trap the same exceptions and pass
1881          * them to L1. When running L2, we will only handle the exceptions
1882          * specified above if L1 did not want them.
1883          */
1884         if (is_guest_mode(vcpu))
1885                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1886
1887         vmcs_write32(EXCEPTION_BITMAP, eb);
1888 }
1889
1890 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1891                 unsigned long entry, unsigned long exit)
1892 {
1893         vm_entry_controls_clearbit(vmx, entry);
1894         vm_exit_controls_clearbit(vmx, exit);
1895 }
1896
1897 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1898 {
1899         unsigned i;
1900         struct msr_autoload *m = &vmx->msr_autoload;
1901
1902         switch (msr) {
1903         case MSR_EFER:
1904                 if (cpu_has_load_ia32_efer) {
1905                         clear_atomic_switch_msr_special(vmx,
1906                                         VM_ENTRY_LOAD_IA32_EFER,
1907                                         VM_EXIT_LOAD_IA32_EFER);
1908                         return;
1909                 }
1910                 break;
1911         case MSR_CORE_PERF_GLOBAL_CTRL:
1912                 if (cpu_has_load_perf_global_ctrl) {
1913                         clear_atomic_switch_msr_special(vmx,
1914                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1915                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1916                         return;
1917                 }
1918                 break;
1919         }
1920
1921         for (i = 0; i < m->nr; ++i)
1922                 if (m->guest[i].index == msr)
1923                         break;
1924
1925         if (i == m->nr)
1926                 return;
1927         --m->nr;
1928         m->guest[i] = m->guest[m->nr];
1929         m->host[i] = m->host[m->nr];
1930         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1931         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1932 }
1933
1934 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1935                 unsigned long entry, unsigned long exit,
1936                 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1937                 u64 guest_val, u64 host_val)
1938 {
1939         vmcs_write64(guest_val_vmcs, guest_val);
1940         vmcs_write64(host_val_vmcs, host_val);
1941         vm_entry_controls_setbit(vmx, entry);
1942         vm_exit_controls_setbit(vmx, exit);
1943 }
1944
1945 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1946                                   u64 guest_val, u64 host_val)
1947 {
1948         unsigned i;
1949         struct msr_autoload *m = &vmx->msr_autoload;
1950
1951         switch (msr) {
1952         case MSR_EFER:
1953                 if (cpu_has_load_ia32_efer) {
1954                         add_atomic_switch_msr_special(vmx,
1955                                         VM_ENTRY_LOAD_IA32_EFER,
1956                                         VM_EXIT_LOAD_IA32_EFER,
1957                                         GUEST_IA32_EFER,
1958                                         HOST_IA32_EFER,
1959                                         guest_val, host_val);
1960                         return;
1961                 }
1962                 break;
1963         case MSR_CORE_PERF_GLOBAL_CTRL:
1964                 if (cpu_has_load_perf_global_ctrl) {
1965                         add_atomic_switch_msr_special(vmx,
1966                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1967                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1968                                         GUEST_IA32_PERF_GLOBAL_CTRL,
1969                                         HOST_IA32_PERF_GLOBAL_CTRL,
1970                                         guest_val, host_val);
1971                         return;
1972                 }
1973                 break;
1974         case MSR_IA32_PEBS_ENABLE:
1975                 /* PEBS needs a quiescent period after being disabled (to write
1976                  * a record).  Disabling PEBS through VMX MSR swapping doesn't
1977                  * provide that period, so a CPU could write host's record into
1978                  * guest's memory.
1979                  */
1980                 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1981         }
1982
1983         for (i = 0; i < m->nr; ++i)
1984                 if (m->guest[i].index == msr)
1985                         break;
1986
1987         if (i == NR_AUTOLOAD_MSRS) {
1988                 printk_once(KERN_WARNING "Not enough msr switch entries. "
1989                                 "Can't add msr %x\n", msr);
1990                 return;
1991         } else if (i == m->nr) {
1992                 ++m->nr;
1993                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1994                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1995         }
1996
1997         m->guest[i].index = msr;
1998         m->guest[i].value = guest_val;
1999         m->host[i].index = msr;
2000         m->host[i].value = host_val;
2001 }
2002
2003 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2004 {
2005         u64 guest_efer = vmx->vcpu.arch.efer;
2006         u64 ignore_bits = 0;
2007
2008         if (!enable_ept) {
2009                 /*
2010                  * NX is needed to handle CR0.WP=1, CR4.SMEP=1.  Testing
2011                  * host CPUID is more efficient than testing guest CPUID
2012                  * or CR4.  Host SMEP is anyway a requirement for guest SMEP.
2013                  */
2014                 if (boot_cpu_has(X86_FEATURE_SMEP))
2015                         guest_efer |= EFER_NX;
2016                 else if (!(guest_efer & EFER_NX))
2017                         ignore_bits |= EFER_NX;
2018         }
2019
2020         /*
2021          * LMA and LME handled by hardware; SCE meaningless outside long mode.
2022          */
2023         ignore_bits |= EFER_SCE;
2024 #ifdef CONFIG_X86_64
2025         ignore_bits |= EFER_LMA | EFER_LME;
2026         /* SCE is meaningful only in long mode on Intel */
2027         if (guest_efer & EFER_LMA)
2028                 ignore_bits &= ~(u64)EFER_SCE;
2029 #endif
2030
2031         clear_atomic_switch_msr(vmx, MSR_EFER);
2032
2033         /*
2034          * On EPT, we can't emulate NX, so we must switch EFER atomically.
2035          * On CPUs that support "load IA32_EFER", always switch EFER
2036          * atomically, since it's faster than switching it manually.
2037          */
2038         if (cpu_has_load_ia32_efer ||
2039             (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2040                 if (!(guest_efer & EFER_LMA))
2041                         guest_efer &= ~EFER_LME;
2042                 if (guest_efer != host_efer)
2043                         add_atomic_switch_msr(vmx, MSR_EFER,
2044                                               guest_efer, host_efer);
2045                 return false;
2046         } else {
2047                 guest_efer &= ~ignore_bits;
2048                 guest_efer |= host_efer & ignore_bits;
2049
2050                 vmx->guest_msrs[efer_offset].data = guest_efer;
2051                 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2052
2053                 return true;
2054         }
2055 }
2056
2057 #ifdef CONFIG_X86_32
2058 /*
2059  * On 32-bit kernels, VM exits still load the FS and GS bases from the
2060  * VMCS rather than the segment table.  KVM uses this helper to figure
2061  * out the current bases to poke them into the VMCS before entry.
2062  */
2063 static unsigned long segment_base(u16 selector)
2064 {
2065         struct desc_struct *table;
2066         unsigned long v;
2067
2068         if (!(selector & ~SEGMENT_RPL_MASK))
2069                 return 0;
2070
2071         table = get_current_gdt_ro();
2072
2073         if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2074                 u16 ldt_selector = kvm_read_ldt();
2075
2076                 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2077                         return 0;
2078
2079                 table = (struct desc_struct *)segment_base(ldt_selector);
2080         }
2081         v = get_desc_base(&table[selector >> 3]);
2082         return v;
2083 }
2084 #endif
2085
2086 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2087 {
2088         struct vcpu_vmx *vmx = to_vmx(vcpu);
2089         int i;
2090
2091         if (vmx->host_state.loaded)
2092                 return;
2093
2094         vmx->host_state.loaded = 1;
2095         /*
2096          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
2097          * allow segment selectors with cpl > 0 or ti == 1.
2098          */
2099         vmx->host_state.ldt_sel = kvm_read_ldt();
2100         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2101         savesegment(fs, vmx->host_state.fs_sel);
2102         if (!(vmx->host_state.fs_sel & 7)) {
2103                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2104                 vmx->host_state.fs_reload_needed = 0;
2105         } else {
2106                 vmcs_write16(HOST_FS_SELECTOR, 0);
2107                 vmx->host_state.fs_reload_needed = 1;
2108         }
2109         savesegment(gs, vmx->host_state.gs_sel);
2110         if (!(vmx->host_state.gs_sel & 7))
2111                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2112         else {
2113                 vmcs_write16(HOST_GS_SELECTOR, 0);
2114                 vmx->host_state.gs_ldt_reload_needed = 1;
2115         }
2116
2117 #ifdef CONFIG_X86_64
2118         savesegment(ds, vmx->host_state.ds_sel);
2119         savesegment(es, vmx->host_state.es_sel);
2120 #endif
2121
2122 #ifdef CONFIG_X86_64
2123         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2124         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2125 #else
2126         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2127         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2128 #endif
2129
2130 #ifdef CONFIG_X86_64
2131         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2132         if (is_long_mode(&vmx->vcpu))
2133                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2134 #endif
2135         if (boot_cpu_has(X86_FEATURE_MPX))
2136                 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2137         for (i = 0; i < vmx->save_nmsrs; ++i)
2138                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2139                                    vmx->guest_msrs[i].data,
2140                                    vmx->guest_msrs[i].mask);
2141 }
2142
2143 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2144 {
2145         if (!vmx->host_state.loaded)
2146                 return;
2147
2148         ++vmx->vcpu.stat.host_state_reload;
2149         vmx->host_state.loaded = 0;
2150 #ifdef CONFIG_X86_64
2151         if (is_long_mode(&vmx->vcpu))
2152                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2153 #endif
2154         if (vmx->host_state.gs_ldt_reload_needed) {
2155                 kvm_load_ldt(vmx->host_state.ldt_sel);
2156 #ifdef CONFIG_X86_64
2157                 load_gs_index(vmx->host_state.gs_sel);
2158 #else
2159                 loadsegment(gs, vmx->host_state.gs_sel);
2160 #endif
2161         }
2162         if (vmx->host_state.fs_reload_needed)
2163                 loadsegment(fs, vmx->host_state.fs_sel);
2164 #ifdef CONFIG_X86_64
2165         if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2166                 loadsegment(ds, vmx->host_state.ds_sel);
2167                 loadsegment(es, vmx->host_state.es_sel);
2168         }
2169 #endif
2170         invalidate_tss_limit();
2171 #ifdef CONFIG_X86_64
2172         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2173 #endif
2174         if (vmx->host_state.msr_host_bndcfgs)
2175                 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2176         load_fixmap_gdt(raw_smp_processor_id());
2177 }
2178
2179 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2180 {
2181         preempt_disable();
2182         __vmx_load_host_state(vmx);
2183         preempt_enable();
2184 }
2185
2186 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2187 {
2188         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2189         struct pi_desc old, new;
2190         unsigned int dest;
2191
2192         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2193                 !irq_remapping_cap(IRQ_POSTING_CAP)  ||
2194                 !kvm_vcpu_apicv_active(vcpu))
2195                 return;
2196
2197         do {
2198                 old.control = new.control = pi_desc->control;
2199
2200                 /*
2201                  * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2202                  * are two possible cases:
2203                  * 1. After running 'pre_block', context switch
2204                  *    happened. For this case, 'sn' was set in
2205                  *    vmx_vcpu_put(), so we need to clear it here.
2206                  * 2. After running 'pre_block', we were blocked,
2207                  *    and woken up by some other guy. For this case,
2208                  *    we don't need to do anything, 'pi_post_block'
2209                  *    will do everything for us. However, we cannot
2210                  *    check whether it is case #1 or case #2 here
2211                  *    (maybe, not needed), so we also clear sn here,
2212                  *    I think it is not a big deal.
2213                  */
2214                 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2215                         if (vcpu->cpu != cpu) {
2216                                 dest = cpu_physical_id(cpu);
2217
2218                                 if (x2apic_enabled())
2219                                         new.ndst = dest;
2220                                 else
2221                                         new.ndst = (dest << 8) & 0xFF00;
2222                         }
2223
2224                         /* set 'NV' to 'notification vector' */
2225                         new.nv = POSTED_INTR_VECTOR;
2226                 }
2227
2228                 /* Allow posting non-urgent interrupts */
2229                 new.sn = 0;
2230         } while (cmpxchg(&pi_desc->control, old.control,
2231                         new.control) != old.control);
2232 }
2233
2234 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2235 {
2236         vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2237         vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2238 }
2239
2240 /*
2241  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2242  * vcpu mutex is already taken.
2243  */
2244 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2245 {
2246         struct vcpu_vmx *vmx = to_vmx(vcpu);
2247         bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2248
2249         if (!already_loaded) {
2250                 loaded_vmcs_clear(vmx->loaded_vmcs);
2251                 local_irq_disable();
2252                 crash_disable_local_vmclear(cpu);
2253
2254                 /*
2255                  * Read loaded_vmcs->cpu should be before fetching
2256                  * loaded_vmcs->loaded_vmcss_on_cpu_link.
2257                  * See the comments in __loaded_vmcs_clear().
2258                  */
2259                 smp_rmb();
2260
2261                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2262                          &per_cpu(loaded_vmcss_on_cpu, cpu));
2263                 crash_enable_local_vmclear(cpu);
2264                 local_irq_enable();
2265         }
2266
2267         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2268                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2269                 vmcs_load(vmx->loaded_vmcs->vmcs);
2270         }
2271
2272         if (!already_loaded) {
2273                 void *gdt = get_current_gdt_ro();
2274                 unsigned long sysenter_esp;
2275
2276                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2277
2278                 /*
2279                  * Linux uses per-cpu TSS and GDT, so set these when switching
2280                  * processors.  See 22.2.4.
2281                  */
2282                 vmcs_writel(HOST_TR_BASE,
2283                             (unsigned long)this_cpu_ptr(&cpu_tss));
2284                 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
2285
2286                 /*
2287                  * VM exits change the host TR limit to 0x67 after a VM
2288                  * exit.  This is okay, since 0x67 covers everything except
2289                  * the IO bitmap and have have code to handle the IO bitmap
2290                  * being lost after a VM exit.
2291                  */
2292                 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2293
2294                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2295                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2296
2297                 vmx->loaded_vmcs->cpu = cpu;
2298         }
2299
2300         /* Setup TSC multiplier */
2301         if (kvm_has_tsc_control &&
2302             vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2303                 decache_tsc_multiplier(vmx);
2304
2305         vmx_vcpu_pi_load(vcpu, cpu);
2306         vmx->host_pkru = read_pkru();
2307 }
2308
2309 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2310 {
2311         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2312
2313         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2314                 !irq_remapping_cap(IRQ_POSTING_CAP)  ||
2315                 !kvm_vcpu_apicv_active(vcpu))
2316                 return;
2317
2318         /* Set SN when the vCPU is preempted */
2319         if (vcpu->preempted)
2320                 pi_set_sn(pi_desc);
2321 }
2322
2323 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2324 {
2325         vmx_vcpu_pi_put(vcpu);
2326
2327         __vmx_load_host_state(to_vmx(vcpu));
2328 }
2329
2330 static bool emulation_required(struct kvm_vcpu *vcpu)
2331 {
2332         return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2333 }
2334
2335 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2336
2337 /*
2338  * Return the cr0 value that a nested guest would read. This is a combination
2339  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2340  * its hypervisor (cr0_read_shadow).
2341  */
2342 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2343 {
2344         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2345                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2346 }
2347 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2348 {
2349         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2350                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2351 }
2352
2353 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2354 {
2355         unsigned long rflags, save_rflags;
2356
2357         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2358                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2359                 rflags = vmcs_readl(GUEST_RFLAGS);
2360                 if (to_vmx(vcpu)->rmode.vm86_active) {
2361                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2362                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2363                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2364                 }
2365                 to_vmx(vcpu)->rflags = rflags;
2366         }
2367         return to_vmx(vcpu)->rflags;
2368 }
2369
2370 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2371 {
2372         unsigned long old_rflags = vmx_get_rflags(vcpu);
2373
2374         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2375         to_vmx(vcpu)->rflags = rflags;
2376         if (to_vmx(vcpu)->rmode.vm86_active) {
2377                 to_vmx(vcpu)->rmode.save_rflags = rflags;
2378                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2379         }
2380         vmcs_writel(GUEST_RFLAGS, rflags);
2381
2382         if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2383                 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2384 }
2385
2386 static u32 vmx_get_pkru(struct kvm_vcpu *vcpu)
2387 {
2388         return to_vmx(vcpu)->guest_pkru;
2389 }
2390
2391 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2392 {
2393         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2394         int ret = 0;
2395
2396         if (interruptibility & GUEST_INTR_STATE_STI)
2397                 ret |= KVM_X86_SHADOW_INT_STI;
2398         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2399                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2400
2401         return ret;
2402 }
2403
2404 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2405 {
2406         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2407         u32 interruptibility = interruptibility_old;
2408
2409         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2410
2411         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2412                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2413         else if (mask & KVM_X86_SHADOW_INT_STI)
2414                 interruptibility |= GUEST_INTR_STATE_STI;
2415
2416         if ((interruptibility != interruptibility_old))
2417                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2418 }
2419
2420 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2421 {
2422         unsigned long rip;
2423
2424         rip = kvm_rip_read(vcpu);
2425         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2426         kvm_rip_write(vcpu, rip);
2427
2428         /* skipping an emulated instruction also counts */
2429         vmx_set_interrupt_shadow(vcpu, 0);
2430 }
2431
2432 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2433                                                unsigned long exit_qual)
2434 {
2435         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2436         unsigned int nr = vcpu->arch.exception.nr;
2437         u32 intr_info = nr | INTR_INFO_VALID_MASK;
2438
2439         if (vcpu->arch.exception.has_error_code) {
2440                 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2441                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2442         }
2443
2444         if (kvm_exception_is_soft(nr))
2445                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2446         else
2447                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2448
2449         if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2450             vmx_get_nmi_mask(vcpu))
2451                 intr_info |= INTR_INFO_UNBLOCK_NMI;
2452
2453         nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2454 }
2455
2456 /*
2457  * KVM wants to inject page-faults which it got to the guest. This function
2458  * checks whether in a nested guest, we need to inject them to L1 or L2.
2459  */
2460 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu)
2461 {
2462         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2463         unsigned int nr = vcpu->arch.exception.nr;
2464
2465         if (nr == PF_VECTOR) {
2466                 if (vcpu->arch.exception.nested_apf) {
2467                         nested_vmx_inject_exception_vmexit(vcpu,
2468                                                            vcpu->arch.apf.nested_apf_token);
2469                         return 1;
2470                 }
2471                 /*
2472                  * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2473                  * The fix is to add the ancillary datum (CR2 or DR6) to structs
2474                  * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2475                  * can be written only when inject_pending_event runs.  This should be
2476                  * conditional on a new capability---if the capability is disabled,
2477                  * kvm_multiple_exception would write the ancillary information to
2478                  * CR2 or DR6, for backwards ABI-compatibility.
2479                  */
2480                 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2481                                                     vcpu->arch.exception.error_code)) {
2482                         nested_vmx_inject_exception_vmexit(vcpu, vcpu->arch.cr2);
2483                         return 1;
2484                 }
2485         } else {
2486                 unsigned long exit_qual = 0;
2487                 if (nr == DB_VECTOR)
2488                         exit_qual = vcpu->arch.dr6;
2489
2490                 if (vmcs12->exception_bitmap & (1u << nr)) {
2491                         nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
2492                         return 1;
2493                 }
2494         }
2495
2496         return 0;
2497 }
2498
2499 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
2500 {
2501         struct vcpu_vmx *vmx = to_vmx(vcpu);
2502         unsigned nr = vcpu->arch.exception.nr;
2503         bool has_error_code = vcpu->arch.exception.has_error_code;
2504         bool reinject = vcpu->arch.exception.reinject;
2505         u32 error_code = vcpu->arch.exception.error_code;
2506         u32 intr_info = nr | INTR_INFO_VALID_MASK;
2507
2508         if (!reinject && is_guest_mode(vcpu) &&
2509             nested_vmx_check_exception(vcpu))
2510                 return;
2511
2512         if (has_error_code) {
2513                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2514                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2515         }
2516
2517         if (vmx->rmode.vm86_active) {
2518                 int inc_eip = 0;
2519                 if (kvm_exception_is_soft(nr))
2520                         inc_eip = vcpu->arch.event_exit_inst_len;
2521                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2522                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2523                 return;
2524         }
2525
2526         if (kvm_exception_is_soft(nr)) {
2527                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2528                              vmx->vcpu.arch.event_exit_inst_len);
2529                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2530         } else
2531                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2532
2533         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2534 }
2535
2536 static bool vmx_rdtscp_supported(void)
2537 {
2538         return cpu_has_vmx_rdtscp();
2539 }
2540
2541 static bool vmx_invpcid_supported(void)
2542 {
2543         return cpu_has_vmx_invpcid() && enable_ept;
2544 }
2545
2546 /*
2547  * Swap MSR entry in host/guest MSR entry array.
2548  */
2549 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2550 {
2551         struct shared_msr_entry tmp;
2552
2553         tmp = vmx->guest_msrs[to];
2554         vmx->guest_msrs[to] = vmx->guest_msrs[from];
2555         vmx->guest_msrs[from] = tmp;
2556 }
2557
2558 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2559 {
2560         unsigned long *msr_bitmap;
2561
2562         if (is_guest_mode(vcpu))
2563                 msr_bitmap = to_vmx(vcpu)->nested.msr_bitmap;
2564         else if (cpu_has_secondary_exec_ctrls() &&
2565                  (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2566                   SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2567                 if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) {
2568                         if (is_long_mode(vcpu))
2569                                 msr_bitmap = vmx_msr_bitmap_longmode_x2apic_apicv;
2570                         else
2571                                 msr_bitmap = vmx_msr_bitmap_legacy_x2apic_apicv;
2572                 } else {
2573                         if (is_long_mode(vcpu))
2574                                 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2575                         else
2576                                 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2577                 }
2578         } else {
2579                 if (is_long_mode(vcpu))
2580                         msr_bitmap = vmx_msr_bitmap_longmode;
2581                 else
2582                         msr_bitmap = vmx_msr_bitmap_legacy;
2583         }
2584
2585         vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2586 }
2587
2588 /*
2589  * Set up the vmcs to automatically save and restore system
2590  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
2591  * mode, as fiddling with msrs is very expensive.
2592  */
2593 static void setup_msrs(struct vcpu_vmx *vmx)
2594 {
2595         int save_nmsrs, index;
2596
2597         save_nmsrs = 0;
2598 #ifdef CONFIG_X86_64
2599         if (is_long_mode(&vmx->vcpu)) {
2600                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2601                 if (index >= 0)
2602                         move_msr_up(vmx, index, save_nmsrs++);
2603                 index = __find_msr_index(vmx, MSR_LSTAR);
2604                 if (index >= 0)
2605                         move_msr_up(vmx, index, save_nmsrs++);
2606                 index = __find_msr_index(vmx, MSR_CSTAR);
2607                 if (index >= 0)
2608                         move_msr_up(vmx, index, save_nmsrs++);
2609                 index = __find_msr_index(vmx, MSR_TSC_AUX);
2610                 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2611                         move_msr_up(vmx, index, save_nmsrs++);
2612                 /*
2613                  * MSR_STAR is only needed on long mode guests, and only
2614                  * if efer.sce is enabled.
2615                  */
2616                 index = __find_msr_index(vmx, MSR_STAR);
2617                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2618                         move_msr_up(vmx, index, save_nmsrs++);
2619         }
2620 #endif
2621         index = __find_msr_index(vmx, MSR_EFER);
2622         if (index >= 0 && update_transition_efer(vmx, index))
2623                 move_msr_up(vmx, index, save_nmsrs++);
2624
2625         vmx->save_nmsrs = save_nmsrs;
2626
2627         if (cpu_has_vmx_msr_bitmap())
2628                 vmx_set_msr_bitmap(&vmx->vcpu);
2629 }
2630
2631 /*
2632  * reads and returns guest's timestamp counter "register"
2633  * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2634  * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2635  */
2636 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2637 {
2638         u64 host_tsc, tsc_offset;
2639
2640         host_tsc = rdtsc();
2641         tsc_offset = vmcs_read64(TSC_OFFSET);
2642         return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2643 }
2644
2645 /*
2646  * writes 'offset' into guest's timestamp counter offset register
2647  */
2648 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2649 {
2650         if (is_guest_mode(vcpu)) {
2651                 /*
2652                  * We're here if L1 chose not to trap WRMSR to TSC. According
2653                  * to the spec, this should set L1's TSC; The offset that L1
2654                  * set for L2 remains unchanged, and still needs to be added
2655                  * to the newly set TSC to get L2's TSC.
2656                  */
2657                 struct vmcs12 *vmcs12;
2658                 /* recalculate vmcs02.TSC_OFFSET: */
2659                 vmcs12 = get_vmcs12(vcpu);
2660                 vmcs_write64(TSC_OFFSET, offset +
2661                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2662                          vmcs12->tsc_offset : 0));
2663         } else {
2664                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2665                                            vmcs_read64(TSC_OFFSET), offset);
2666                 vmcs_write64(TSC_OFFSET, offset);
2667         }
2668 }
2669
2670 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2671 {
2672         struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2673         return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2674 }
2675
2676 /*
2677  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2678  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2679  * all guests if the "nested" module option is off, and can also be disabled
2680  * for a single guest by disabling its VMX cpuid bit.
2681  */
2682 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2683 {
2684         return nested && guest_cpuid_has_vmx(vcpu);
2685 }
2686
2687 /*
2688  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2689  * returned for the various VMX controls MSRs when nested VMX is enabled.
2690  * The same values should also be used to verify that vmcs12 control fields are
2691  * valid during nested entry from L1 to L2.
2692  * Each of these control msrs has a low and high 32-bit half: A low bit is on
2693  * if the corresponding bit in the (32-bit) control field *must* be on, and a
2694  * bit in the high half is on if the corresponding bit in the control field
2695  * may be on. See also vmx_control_verify().
2696  */
2697 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2698 {
2699         /*
2700          * Note that as a general rule, the high half of the MSRs (bits in
2701          * the control fields which may be 1) should be initialized by the
2702          * intersection of the underlying hardware's MSR (i.e., features which
2703          * can be supported) and the list of features we want to expose -
2704          * because they are known to be properly supported in our code.
2705          * Also, usually, the low half of the MSRs (bits which must be 1) can
2706          * be set to 0, meaning that L1 may turn off any of these bits. The
2707          * reason is that if one of these bits is necessary, it will appear
2708          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2709          * fields of vmcs01 and vmcs02, will turn these bits off - and
2710          * nested_vmx_exit_reflected() will not pass related exits to L1.
2711          * These rules have exceptions below.
2712          */
2713
2714         /* pin-based controls */
2715         rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2716                 vmx->nested.nested_vmx_pinbased_ctls_low,
2717                 vmx->nested.nested_vmx_pinbased_ctls_high);
2718         vmx->nested.nested_vmx_pinbased_ctls_low |=
2719                 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2720         vmx->nested.nested_vmx_pinbased_ctls_high &=
2721                 PIN_BASED_EXT_INTR_MASK |
2722                 PIN_BASED_NMI_EXITING |
2723                 PIN_BASED_VIRTUAL_NMIS;
2724         vmx->nested.nested_vmx_pinbased_ctls_high |=
2725                 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2726                 PIN_BASED_VMX_PREEMPTION_TIMER;
2727         if (kvm_vcpu_apicv_active(&vmx->vcpu))
2728                 vmx->nested.nested_vmx_pinbased_ctls_high |=
2729                         PIN_BASED_POSTED_INTR;
2730
2731         /* exit controls */
2732         rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2733                 vmx->nested.nested_vmx_exit_ctls_low,
2734                 vmx->nested.nested_vmx_exit_ctls_high);
2735         vmx->nested.nested_vmx_exit_ctls_low =
2736                 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2737
2738         vmx->nested.nested_vmx_exit_ctls_high &=
2739 #ifdef CONFIG_X86_64
2740                 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2741 #endif
2742                 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2743         vmx->nested.nested_vmx_exit_ctls_high |=
2744                 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2745                 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2746                 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2747
2748         if (kvm_mpx_supported())
2749                 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2750
2751         /* We support free control of debug control saving. */
2752         vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2753
2754         /* entry controls */
2755         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2756                 vmx->nested.nested_vmx_entry_ctls_low,
2757                 vmx->nested.nested_vmx_entry_ctls_high);
2758         vmx->nested.nested_vmx_entry_ctls_low =
2759                 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2760         vmx->nested.nested_vmx_entry_ctls_high &=
2761 #ifdef CONFIG_X86_64
2762                 VM_ENTRY_IA32E_MODE |
2763 #endif
2764                 VM_ENTRY_LOAD_IA32_PAT;
2765         vmx->nested.nested_vmx_entry_ctls_high |=
2766                 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2767         if (kvm_mpx_supported())
2768                 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2769
2770         /* We support free control of debug control loading. */
2771         vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2772
2773         /* cpu-based controls */
2774         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2775                 vmx->nested.nested_vmx_procbased_ctls_low,
2776                 vmx->nested.nested_vmx_procbased_ctls_high);
2777         vmx->nested.nested_vmx_procbased_ctls_low =
2778                 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2779         vmx->nested.nested_vmx_procbased_ctls_high &=
2780                 CPU_BASED_VIRTUAL_INTR_PENDING |
2781                 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2782                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2783                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2784                 CPU_BASED_CR3_STORE_EXITING |
2785 #ifdef CONFIG_X86_64
2786                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2787 #endif
2788                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2789                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2790                 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2791                 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2792                 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2793         /*
2794          * We can allow some features even when not supported by the
2795          * hardware. For example, L1 can specify an MSR bitmap - and we
2796          * can use it to avoid exits to L1 - even when L0 runs L2
2797          * without MSR bitmaps.
2798          */
2799         vmx->nested.nested_vmx_procbased_ctls_high |=
2800                 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2801                 CPU_BASED_USE_MSR_BITMAPS;
2802
2803         /* We support free control of CR3 access interception. */
2804         vmx->nested.nested_vmx_procbased_ctls_low &=
2805                 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2806
2807         /* secondary cpu-based controls */
2808         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2809                 vmx->nested.nested_vmx_secondary_ctls_low,
2810                 vmx->nested.nested_vmx_secondary_ctls_high);
2811         vmx->nested.nested_vmx_secondary_ctls_low = 0;
2812         vmx->nested.nested_vmx_secondary_ctls_high &=
2813                 SECONDARY_EXEC_RDRAND | SECONDARY_EXEC_RDSEED |
2814                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2815                 SECONDARY_EXEC_RDTSCP |
2816                 SECONDARY_EXEC_DESC |
2817                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2818                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2819                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2820                 SECONDARY_EXEC_WBINVD_EXITING |
2821                 SECONDARY_EXEC_XSAVES;
2822
2823         if (enable_ept) {
2824                 /* nested EPT: emulate EPT also to L1 */
2825                 vmx->nested.nested_vmx_secondary_ctls_high |=
2826                         SECONDARY_EXEC_ENABLE_EPT;
2827                 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2828                          VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2829                 if (cpu_has_vmx_ept_execute_only())
2830                         vmx->nested.nested_vmx_ept_caps |=
2831                                 VMX_EPT_EXECUTE_ONLY_BIT;
2832                 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2833                 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2834                         VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2835                         VMX_EPT_1GB_PAGE_BIT;
2836                 if (enable_ept_ad_bits) {
2837                         vmx->nested.nested_vmx_secondary_ctls_high |=
2838                                 SECONDARY_EXEC_ENABLE_PML;
2839                         vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2840                 }
2841         } else
2842                 vmx->nested.nested_vmx_ept_caps = 0;
2843
2844         /*
2845          * Old versions of KVM use the single-context version without
2846          * checking for support, so declare that it is supported even
2847          * though it is treated as global context.  The alternative is
2848          * not failing the single-context invvpid, and it is worse.
2849          */
2850         if (enable_vpid) {
2851                 vmx->nested.nested_vmx_secondary_ctls_high |=
2852                         SECONDARY_EXEC_ENABLE_VPID;
2853                 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2854                         VMX_VPID_EXTENT_SUPPORTED_MASK;
2855         } else
2856                 vmx->nested.nested_vmx_vpid_caps = 0;
2857
2858         if (enable_unrestricted_guest)
2859                 vmx->nested.nested_vmx_secondary_ctls_high |=
2860                         SECONDARY_EXEC_UNRESTRICTED_GUEST;
2861
2862         /* miscellaneous data */
2863         rdmsr(MSR_IA32_VMX_MISC,
2864                 vmx->nested.nested_vmx_misc_low,
2865                 vmx->nested.nested_vmx_misc_high);
2866         vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2867         vmx->nested.nested_vmx_misc_low |=
2868                 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2869                 VMX_MISC_ACTIVITY_HLT;
2870         vmx->nested.nested_vmx_misc_high = 0;
2871
2872         /*
2873          * This MSR reports some information about VMX support. We
2874          * should return information about the VMX we emulate for the
2875          * guest, and the VMCS structure we give it - not about the
2876          * VMX support of the underlying hardware.
2877          */
2878         vmx->nested.nested_vmx_basic =
2879                 VMCS12_REVISION |
2880                 VMX_BASIC_TRUE_CTLS |
2881                 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2882                 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2883
2884         if (cpu_has_vmx_basic_inout())
2885                 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2886
2887         /*
2888          * These MSRs specify bits which the guest must keep fixed on
2889          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2890          * We picked the standard core2 setting.
2891          */
2892 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2893 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
2894         vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2895         vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2896
2897         /* These MSRs specify bits which the guest must keep fixed off. */
2898         rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2899         rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2900
2901         /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2902         vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2903 }
2904
2905 /*
2906  * if fixed0[i] == 1: val[i] must be 1
2907  * if fixed1[i] == 0: val[i] must be 0
2908  */
2909 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2910 {
2911         return ((val & fixed1) | fixed0) == val;
2912 }
2913
2914 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2915 {
2916         return fixed_bits_valid(control, low, high);
2917 }
2918
2919 static inline u64 vmx_control_msr(u32 low, u32 high)
2920 {
2921         return low | ((u64)high << 32);
2922 }
2923
2924 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2925 {
2926         superset &= mask;
2927         subset &= mask;
2928
2929         return (superset | subset) == superset;
2930 }
2931
2932 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2933 {
2934         const u64 feature_and_reserved =
2935                 /* feature (except bit 48; see below) */
2936                 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2937                 /* reserved */
2938                 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2939         u64 vmx_basic = vmx->nested.nested_vmx_basic;
2940
2941         if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2942                 return -EINVAL;
2943
2944         /*
2945          * KVM does not emulate a version of VMX that constrains physical
2946          * addresses of VMX structures (e.g. VMCS) to 32-bits.
2947          */
2948         if (data & BIT_ULL(48))
2949                 return -EINVAL;
2950
2951         if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2952             vmx_basic_vmcs_revision_id(data))
2953                 return -EINVAL;
2954
2955         if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2956                 return -EINVAL;
2957
2958         vmx->nested.nested_vmx_basic = data;
2959         return 0;
2960 }
2961
2962 static int
2963 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2964 {
2965         u64 supported;
2966         u32 *lowp, *highp;
2967
2968         switch (msr_index) {
2969         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2970                 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2971                 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2972                 break;
2973         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2974                 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2975                 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2976                 break;
2977         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2978                 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2979                 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2980                 break;
2981         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2982                 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2983                 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2984                 break;
2985         case MSR_IA32_VMX_PROCBASED_CTLS2:
2986                 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2987                 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2988                 break;
2989         default:
2990                 BUG();
2991         }
2992
2993         supported = vmx_control_msr(*lowp, *highp);
2994
2995         /* Check must-be-1 bits are still 1. */
2996         if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
2997                 return -EINVAL;
2998
2999         /* Check must-be-0 bits are still 0. */
3000         if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3001                 return -EINVAL;
3002
3003         *lowp = data;
3004         *highp = data >> 32;
3005         return 0;
3006 }
3007
3008 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3009 {
3010         const u64 feature_and_reserved_bits =
3011                 /* feature */
3012                 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3013                 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3014                 /* reserved */
3015                 GENMASK_ULL(13, 9) | BIT_ULL(31);
3016         u64 vmx_misc;
3017
3018         vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
3019                                    vmx->nested.nested_vmx_misc_high);
3020
3021         if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3022                 return -EINVAL;
3023
3024         if ((vmx->nested.nested_vmx_pinbased_ctls_high &
3025              PIN_BASED_VMX_PREEMPTION_TIMER) &&
3026             vmx_misc_preemption_timer_rate(data) !=
3027             vmx_misc_preemption_timer_rate(vmx_misc))
3028                 return -EINVAL;
3029
3030         if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3031                 return -EINVAL;
3032
3033         if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3034                 return -EINVAL;
3035
3036         if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3037                 return -EINVAL;
3038
3039         vmx->nested.nested_vmx_misc_low = data;
3040         vmx->nested.nested_vmx_misc_high = data >> 32;
3041         return 0;
3042 }
3043
3044 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3045 {
3046         u64 vmx_ept_vpid_cap;
3047
3048         vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
3049                                            vmx->nested.nested_vmx_vpid_caps);
3050
3051         /* Every bit is either reserved or a feature bit. */
3052         if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3053                 return -EINVAL;
3054
3055         vmx->nested.nested_vmx_ept_caps = data;
3056         vmx->nested.nested_vmx_vpid_caps = data >> 32;
3057         return 0;
3058 }
3059
3060 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3061 {
3062         u64 *msr;
3063
3064         switch (msr_index) {
3065         case MSR_IA32_VMX_CR0_FIXED0:
3066                 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3067                 break;
3068         case MSR_IA32_VMX_CR4_FIXED0:
3069                 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3070                 break;
3071         default:
3072                 BUG();
3073         }
3074
3075         /*
3076          * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3077          * must be 1 in the restored value.
3078          */
3079         if (!is_bitwise_subset(data, *msr, -1ULL))
3080                 return -EINVAL;
3081
3082         *msr = data;
3083         return 0;
3084 }
3085
3086 /*
3087  * Called when userspace is restoring VMX MSRs.
3088  *
3089  * Returns 0 on success, non-0 otherwise.
3090  */
3091 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3092 {
3093         struct vcpu_vmx *vmx = to_vmx(vcpu);
3094
3095         switch (msr_index) {
3096         case MSR_IA32_VMX_BASIC:
3097                 return vmx_restore_vmx_basic(vmx, data);
3098         case MSR_IA32_VMX_PINBASED_CTLS:
3099         case MSR_IA32_VMX_PROCBASED_CTLS:
3100         case MSR_IA32_VMX_EXIT_CTLS:
3101         case MSR_IA32_VMX_ENTRY_CTLS:
3102                 /*
3103                  * The "non-true" VMX capability MSRs are generated from the
3104                  * "true" MSRs, so we do not support restoring them directly.
3105                  *
3106                  * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3107                  * should restore the "true" MSRs with the must-be-1 bits
3108                  * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3109                  * DEFAULT SETTINGS".
3110                  */
3111                 return -EINVAL;
3112         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3113         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3114         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3115         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3116         case MSR_IA32_VMX_PROCBASED_CTLS2:
3117                 return vmx_restore_control_msr(vmx, msr_index, data);
3118         case MSR_IA32_VMX_MISC:
3119                 return vmx_restore_vmx_misc(vmx, data);
3120         case MSR_IA32_VMX_CR0_FIXED0:
3121         case MSR_IA32_VMX_CR4_FIXED0:
3122                 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3123         case MSR_IA32_VMX_CR0_FIXED1:
3124         case MSR_IA32_VMX_CR4_FIXED1:
3125                 /*
3126                  * These MSRs are generated based on the vCPU's CPUID, so we
3127                  * do not support restoring them directly.
3128                  */
3129                 return -EINVAL;
3130         case MSR_IA32_VMX_EPT_VPID_CAP:
3131                 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3132         case MSR_IA32_VMX_VMCS_ENUM:
3133                 vmx->nested.nested_vmx_vmcs_enum = data;
3134                 return 0;
3135         default:
3136                 /*
3137                  * The rest of the VMX capability MSRs do not support restore.
3138                  */
3139                 return -EINVAL;
3140         }
3141 }
3142
3143 /* Returns 0 on success, non-0 otherwise. */
3144 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3145 {
3146         struct vcpu_vmx *vmx = to_vmx(vcpu);
3147
3148         switch (msr_index) {
3149         case MSR_IA32_VMX_BASIC:
3150                 *pdata = vmx->nested.nested_vmx_basic;
3151                 break;
3152         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3153         case MSR_IA32_VMX_PINBASED_CTLS:
3154                 *pdata = vmx_control_msr(
3155                         vmx->nested.nested_vmx_pinbased_ctls_low,
3156                         vmx->nested.nested_vmx_pinbased_ctls_high);
3157                 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3158                         *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3159                 break;
3160         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3161         case MSR_IA32_VMX_PROCBASED_CTLS:
3162                 *pdata = vmx_control_msr(
3163                         vmx->nested.nested_vmx_procbased_ctls_low,
3164                         vmx->nested.nested_vmx_procbased_ctls_high);
3165                 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3166                         *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3167                 break;
3168         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3169         case MSR_IA32_VMX_EXIT_CTLS:
3170                 *pdata = vmx_control_msr(
3171                         vmx->nested.nested_vmx_exit_ctls_low,
3172                         vmx->nested.nested_vmx_exit_ctls_high);
3173                 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3174                         *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3175                 break;
3176         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3177         case MSR_IA32_VMX_ENTRY_CTLS:
3178                 *pdata = vmx_control_msr(
3179                         vmx->nested.nested_vmx_entry_ctls_low,
3180                         vmx->nested.nested_vmx_entry_ctls_high);
3181                 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3182                         *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3183                 break;
3184         case MSR_IA32_VMX_MISC:
3185                 *pdata = vmx_control_msr(
3186                         vmx->nested.nested_vmx_misc_low,
3187                         vmx->nested.nested_vmx_misc_high);
3188                 break;
3189         case MSR_IA32_VMX_CR0_FIXED0:
3190                 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3191                 break;
3192         case MSR_IA32_VMX_CR0_FIXED1:
3193                 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3194                 break;
3195         case MSR_IA32_VMX_CR4_FIXED0:
3196                 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3197                 break;
3198         case MSR_IA32_VMX_CR4_FIXED1:
3199                 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3200                 break;
3201         case MSR_IA32_VMX_VMCS_ENUM:
3202                 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3203                 break;
3204         case MSR_IA32_VMX_PROCBASED_CTLS2:
3205                 *pdata = vmx_control_msr(
3206                         vmx->nested.nested_vmx_secondary_ctls_low,
3207                         vmx->nested.nested_vmx_secondary_ctls_high);
3208                 break;
3209         case MSR_IA32_VMX_EPT_VPID_CAP:
3210                 *pdata = vmx->nested.nested_vmx_ept_caps |
3211                         ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3212                 break;
3213         default:
3214                 return 1;
3215         }
3216
3217         return 0;
3218 }
3219
3220 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3221                                                  uint64_t val)
3222 {
3223         uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3224
3225         return !(val & ~valid_bits);
3226 }
3227
3228 /*
3229  * Reads an msr value (of 'msr_index') into 'pdata'.
3230  * Returns 0 on success, non-0 otherwise.
3231  * Assumes vcpu_load() was already called.
3232  */
3233 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3234 {
3235         struct shared_msr_entry *msr;
3236
3237         switch (msr_info->index) {
3238 #ifdef CONFIG_X86_64
3239         case MSR_FS_BASE:
3240                 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3241                 break;
3242         case MSR_GS_BASE:
3243                 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3244                 break;
3245         case MSR_KERNEL_GS_BASE:
3246                 vmx_load_host_state(to_vmx(vcpu));
3247                 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3248                 break;
3249 #endif
3250         case MSR_EFER:
3251                 return kvm_get_msr_common(vcpu, msr_info);
3252         case MSR_IA32_TSC:
3253                 msr_info->data = guest_read_tsc(vcpu);
3254                 break;
3255         case MSR_IA32_SYSENTER_CS:
3256                 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3257                 break;
3258         case MSR_IA32_SYSENTER_EIP:
3259                 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3260                 break;
3261         case MSR_IA32_SYSENTER_ESP:
3262                 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3263                 break;
3264         case MSR_IA32_BNDCFGS:
3265                 if (!kvm_mpx_supported() ||
3266                     (!msr_info->host_initiated && !guest_cpuid_has_mpx(vcpu)))
3267                         return 1;
3268                 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3269                 break;
3270         case MSR_IA32_MCG_EXT_CTL:
3271                 if (!msr_info->host_initiated &&
3272                     !(to_vmx(vcpu)->msr_ia32_feature_control &
3273                       FEATURE_CONTROL_LMCE))
3274                         return 1;
3275                 msr_info->data = vcpu->arch.mcg_ext_ctl;
3276                 break;
3277         case MSR_IA32_FEATURE_CONTROL:
3278                 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3279                 break;
3280         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3281                 if (!nested_vmx_allowed(vcpu))
3282                         return 1;
3283                 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3284         case MSR_IA32_XSS:
3285                 if (!vmx_xsaves_supported())
3286                         return 1;
3287                 msr_info->data = vcpu->arch.ia32_xss;
3288                 break;
3289         case MSR_TSC_AUX:
3290                 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3291                         return 1;
3292                 /* Otherwise falls through */
3293         default:
3294                 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3295                 if (msr) {
3296                         msr_info->data = msr->data;
3297                         break;
3298                 }
3299                 return kvm_get_msr_common(vcpu, msr_info);
3300         }
3301
3302         return 0;
3303 }
3304
3305 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3306
3307 /*
3308  * Writes msr value into into the appropriate "register".
3309  * Returns 0 on success, non-0 otherwise.
3310  * Assumes vcpu_load() was already called.
3311  */
3312 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3313 {
3314         struct vcpu_vmx *vmx = to_vmx(vcpu);
3315         struct shared_msr_entry *msr;
3316         int ret = 0;
3317         u32 msr_index = msr_info->index;
3318         u64 data = msr_info->data;
3319
3320         switch (msr_index) {
3321         case MSR_EFER:
3322                 ret = kvm_set_msr_common(vcpu, msr_info);
3323                 break;
3324 #ifdef CONFIG_X86_64
3325         case MSR_FS_BASE:
3326                 vmx_segment_cache_clear(vmx);
3327                 vmcs_writel(GUEST_FS_BASE, data);
3328                 break;
3329         case MSR_GS_BASE:
3330                 vmx_segment_cache_clear(vmx);
3331                 vmcs_writel(GUEST_GS_BASE, data);
3332                 break;
3333         case MSR_KERNEL_GS_BASE:
3334                 vmx_load_host_state(vmx);
3335                 vmx->msr_guest_kernel_gs_base = data;
3336                 break;
3337 #endif
3338         case MSR_IA32_SYSENTER_CS:
3339                 vmcs_write32(GUEST_SYSENTER_CS, data);
3340                 break;
3341         case MSR_IA32_SYSENTER_EIP:
3342                 vmcs_writel(GUEST_SYSENTER_EIP, data);
3343                 break;
3344         case MSR_IA32_SYSENTER_ESP:
3345                 vmcs_writel(GUEST_SYSENTER_ESP, data);
3346                 break;
3347         case MSR_IA32_BNDCFGS:
3348                 if (!kvm_mpx_supported() ||
3349                     (!msr_info->host_initiated && !guest_cpuid_has_mpx(vcpu)))
3350                         return 1;
3351                 if (is_noncanonical_address(data & PAGE_MASK) ||
3352                     (data & MSR_IA32_BNDCFGS_RSVD))
3353                         return 1;
3354                 vmcs_write64(GUEST_BNDCFGS, data);
3355                 break;
3356         case MSR_IA32_TSC:
3357                 kvm_write_tsc(vcpu, msr_info);
3358                 break;
3359         case MSR_IA32_CR_PAT:
3360                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3361                         if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3362                                 return 1;
3363                         vmcs_write64(GUEST_IA32_PAT, data);
3364                         vcpu->arch.pat = data;
3365                         break;
3366                 }
3367                 ret = kvm_set_msr_common(vcpu, msr_info);
3368                 break;
3369         case MSR_IA32_TSC_ADJUST:
3370                 ret = kvm_set_msr_common(vcpu, msr_info);
3371                 break;
3372         case MSR_IA32_MCG_EXT_CTL:
3373                 if ((!msr_info->host_initiated &&
3374                      !(to_vmx(vcpu)->msr_ia32_feature_control &
3375                        FEATURE_CONTROL_LMCE)) ||
3376                     (data & ~MCG_EXT_CTL_LMCE_EN))
3377                         return 1;
3378                 vcpu->arch.mcg_ext_ctl = data;
3379                 break;
3380         case MSR_IA32_FEATURE_CONTROL:
3381                 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3382                     (to_vmx(vcpu)->msr_ia32_feature_control &
3383                      FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3384                         return 1;
3385                 vmx->msr_ia32_feature_control = data;
3386                 if (msr_info->host_initiated && data == 0)
3387                         vmx_leave_nested(vcpu);
3388                 break;
3389         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3390                 if (!msr_info->host_initiated)
3391                         return 1; /* they are read-only */
3392                 if (!nested_vmx_allowed(vcpu))
3393                         return 1;
3394                 return vmx_set_vmx_msr(vcpu, msr_index, data);
3395         case MSR_IA32_XSS:
3396                 if (!vmx_xsaves_supported())
3397                         return 1;
3398                 /*
3399                  * The only supported bit as of Skylake is bit 8, but
3400                  * it is not supported on KVM.
3401                  */
3402                 if (data != 0)
3403                         return 1;
3404                 vcpu->arch.ia32_xss = data;
3405                 if (vcpu->arch.ia32_xss != host_xss)
3406                         add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3407                                 vcpu->arch.ia32_xss, host_xss);
3408                 else
3409                         clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3410                 break;
3411         case MSR_TSC_AUX:
3412                 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3413                         return 1;
3414                 /* Check reserved bit, higher 32 bits should be zero */
3415                 if ((data >> 32) != 0)
3416                         return 1;
3417                 /* Otherwise falls through */
3418         default:
3419                 msr = find_msr_entry(vmx, msr_index);
3420                 if (msr) {
3421                         u64 old_msr_data = msr->data;
3422                         msr->data = data;
3423                         if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3424                                 preempt_disable();
3425                                 ret = kvm_set_shared_msr(msr->index, msr->data,
3426                                                          msr->mask);
3427                                 preempt_enable();
3428                                 if (ret)
3429                                         msr->data = old_msr_data;
3430                         }
3431                         break;
3432                 }
3433                 ret = kvm_set_msr_common(vcpu, msr_info);
3434         }
3435
3436         return ret;
3437 }
3438
3439 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3440 {
3441         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3442         switch (reg) {
3443         case VCPU_REGS_RSP:
3444                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3445                 break;
3446         case VCPU_REGS_RIP:
3447                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3448                 break;
3449         case VCPU_EXREG_PDPTR:
3450                 if (enable_ept)
3451                         ept_save_pdptrs(vcpu);
3452                 break;
3453         default:
3454                 break;
3455         }
3456 }
3457
3458 static __init int cpu_has_kvm_support(void)
3459 {
3460         return cpu_has_vmx();
3461 }
3462
3463 static __init int vmx_disabled_by_bios(void)
3464 {
3465         u64 msr;
3466
3467         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3468         if (msr & FEATURE_CONTROL_LOCKED) {
3469                 /* launched w/ TXT and VMX disabled */
3470                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3471                         && tboot_enabled())
3472                         return 1;
3473                 /* launched w/o TXT and VMX only enabled w/ TXT */
3474                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3475                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3476                         && !tboot_enabled()) {
3477                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3478                                 "activate TXT before enabling KVM\n");
3479                         return 1;
3480                 }
3481                 /* launched w/o TXT and VMX disabled */
3482                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3483                         && !tboot_enabled())
3484                         return 1;
3485         }
3486
3487         return 0;
3488 }
3489
3490 static void kvm_cpu_vmxon(u64 addr)
3491 {
3492         cr4_set_bits(X86_CR4_VMXE);
3493         intel_pt_handle_vmx(1);
3494
3495         asm volatile (ASM_VMX_VMXON_RAX
3496                         : : "a"(&addr), "m"(addr)
3497                         : "memory", "cc");
3498 }
3499
3500 static int hardware_enable(void)
3501 {
3502         int cpu = raw_smp_processor_id();
3503         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3504         u64 old, test_bits;
3505
3506         if (cr4_read_shadow() & X86_CR4_VMXE)
3507                 return -EBUSY;
3508
3509         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3510         INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3511         spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3512
3513         /*
3514          * Now we can enable the vmclear operation in kdump
3515          * since the loaded_vmcss_on_cpu list on this cpu
3516          * has been initialized.
3517          *
3518          * Though the cpu is not in VMX operation now, there
3519          * is no problem to enable the vmclear operation
3520          * for the loaded_vmcss_on_cpu list is empty!
3521          */
3522         crash_enable_local_vmclear(cpu);
3523
3524         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3525
3526         test_bits = FEATURE_CONTROL_LOCKED;
3527         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3528         if (tboot_enabled())
3529                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3530
3531         if ((old & test_bits) != test_bits) {
3532                 /* enable and lock */
3533                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3534         }
3535         kvm_cpu_vmxon(phys_addr);
3536         ept_sync_global();
3537
3538         return 0;
3539 }
3540
3541 static void vmclear_local_loaded_vmcss(void)
3542 {
3543         int cpu = raw_smp_processor_id();
3544         struct loaded_vmcs *v, *n;
3545
3546         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3547                                  loaded_vmcss_on_cpu_link)
3548                 __loaded_vmcs_clear(v);
3549 }
3550
3551
3552 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3553  * tricks.
3554  */
3555 static void kvm_cpu_vmxoff(void)
3556 {
3557         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3558
3559         intel_pt_handle_vmx(0);
3560         cr4_clear_bits(X86_CR4_VMXE);
3561 }
3562
3563 static void hardware_disable(void)
3564 {
3565         vmclear_local_loaded_vmcss();
3566         kvm_cpu_vmxoff();
3567 }
3568
3569 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3570                                       u32 msr, u32 *result)
3571 {
3572         u32 vmx_msr_low, vmx_msr_high;
3573         u32 ctl = ctl_min | ctl_opt;
3574
3575         rdmsr(msr, vmx_msr_low, vmx_msr_high);
3576
3577         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3578         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
3579
3580         /* Ensure minimum (required) set of control bits are supported. */
3581         if (ctl_min & ~ctl)
3582                 return -EIO;
3583
3584         *result = ctl;
3585         return 0;
3586 }
3587
3588 static __init bool allow_1_setting(u32 msr, u32 ctl)
3589 {
3590         u32 vmx_msr_low, vmx_msr_high;
3591
3592         rdmsr(msr, vmx_msr_low, vmx_msr_high);
3593         return vmx_msr_high & ctl;
3594 }
3595
3596 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3597 {
3598         u32 vmx_msr_low, vmx_msr_high;
3599         u32 min, opt, min2, opt2;
3600         u32 _pin_based_exec_control = 0;
3601         u32 _cpu_based_exec_control = 0;
3602         u32 _cpu_based_2nd_exec_control = 0;
3603         u32 _vmexit_control = 0;
3604         u32 _vmentry_control = 0;
3605
3606         min = CPU_BASED_HLT_EXITING |
3607 #ifdef CONFIG_X86_64
3608               CPU_BASED_CR8_LOAD_EXITING |
3609               CPU_BASED_CR8_STORE_EXITING |
3610 #endif
3611               CPU_BASED_CR3_LOAD_EXITING |
3612               CPU_BASED_CR3_STORE_EXITING |
3613               CPU_BASED_USE_IO_BITMAPS |
3614               CPU_BASED_MOV_DR_EXITING |
3615               CPU_BASED_USE_TSC_OFFSETING |
3616               CPU_BASED_INVLPG_EXITING |
3617               CPU_BASED_RDPMC_EXITING;
3618
3619         if (!kvm_mwait_in_guest())
3620                 min |= CPU_BASED_MWAIT_EXITING |
3621                         CPU_BASED_MONITOR_EXITING;
3622
3623         opt = CPU_BASED_TPR_SHADOW |
3624               CPU_BASED_USE_MSR_BITMAPS |
3625               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3626         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3627                                 &_cpu_based_exec_control) < 0)
3628                 return -EIO;
3629 #ifdef CONFIG_X86_64
3630         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3631                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3632                                            ~CPU_BASED_CR8_STORE_EXITING;
3633 #endif
3634         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3635                 min2 = 0;
3636                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3637                         SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3638                         SECONDARY_EXEC_WBINVD_EXITING |
3639                         SECONDARY_EXEC_ENABLE_VPID |
3640                         SECONDARY_EXEC_ENABLE_EPT |
3641                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
3642                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3643                         SECONDARY_EXEC_RDTSCP |
3644                         SECONDARY_EXEC_ENABLE_INVPCID |
3645                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
3646                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3647                         SECONDARY_EXEC_SHADOW_VMCS |
3648                         SECONDARY_EXEC_XSAVES |
3649                         SECONDARY_EXEC_ENABLE_PML |
3650                         SECONDARY_EXEC_TSC_SCALING;
3651                 if (adjust_vmx_controls(min2, opt2,
3652                                         MSR_IA32_VMX_PROCBASED_CTLS2,
3653                                         &_cpu_based_2nd_exec_control) < 0)
3654                         return -EIO;
3655         }
3656 #ifndef CONFIG_X86_64
3657         if (!(_cpu_based_2nd_exec_control &
3658                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3659                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3660 #endif
3661
3662         if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3663                 _cpu_based_2nd_exec_control &= ~(
3664                                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3665                                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3666                                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3667
3668         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3669                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3670                    enabled */
3671                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3672                                              CPU_BASED_CR3_STORE_EXITING |
3673                                              CPU_BASED_INVLPG_EXITING);
3674                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3675                       vmx_capability.ept, vmx_capability.vpid);
3676         }
3677
3678         min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3679 #ifdef CONFIG_X86_64
3680         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3681 #endif
3682         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3683                 VM_EXIT_CLEAR_BNDCFGS;
3684         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3685                                 &_vmexit_control) < 0)
3686                 return -EIO;
3687
3688         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
3689                 PIN_BASED_VIRTUAL_NMIS;
3690         opt = PIN_BASED_POSTED_INTR | PIN_BASED_VMX_PREEMPTION_TIMER;
3691         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3692                                 &_pin_based_exec_control) < 0)
3693                 return -EIO;
3694
3695         if (cpu_has_broken_vmx_preemption_timer())
3696                 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3697         if (!(_cpu_based_2nd_exec_control &
3698                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3699                 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3700
3701         min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3702         opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3703         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3704                                 &_vmentry_control) < 0)
3705                 return -EIO;
3706
3707         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3708
3709         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3710         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3711                 return -EIO;
3712
3713 #ifdef CONFIG_X86_64
3714         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3715         if (vmx_msr_high & (1u<<16))
3716                 return -EIO;
3717 #endif
3718
3719         /* Require Write-Back (WB) memory type for VMCS accesses. */
3720         if (((vmx_msr_high >> 18) & 15) != 6)
3721                 return -EIO;
3722
3723         vmcs_conf->size = vmx_msr_high & 0x1fff;
3724         vmcs_conf->order = get_order(vmcs_conf->size);
3725         vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3726         vmcs_conf->revision_id = vmx_msr_low;
3727
3728         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3729         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3730         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3731         vmcs_conf->vmexit_ctrl         = _vmexit_control;
3732         vmcs_conf->vmentry_ctrl        = _vmentry_control;
3733
3734         cpu_has_load_ia32_efer =
3735                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3736                                 VM_ENTRY_LOAD_IA32_EFER)
3737                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3738                                    VM_EXIT_LOAD_IA32_EFER);
3739
3740         cpu_has_load_perf_global_ctrl =
3741                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3742                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3743                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3744                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3745
3746         /*
3747          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3748          * but due to errata below it can't be used. Workaround is to use
3749          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3750          *
3751          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3752          *
3753          * AAK155             (model 26)
3754          * AAP115             (model 30)
3755          * AAT100             (model 37)
3756          * BC86,AAY89,BD102   (model 44)
3757          * BA97               (model 46)
3758          *
3759          */
3760         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3761                 switch (boot_cpu_data.x86_model) {
3762                 case 26:
3763                 case 30:
3764                 case 37:
3765                 case 44:
3766                 case 46:
3767                         cpu_has_load_perf_global_ctrl = false;
3768                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3769                                         "does not work properly. Using workaround\n");
3770                         break;
3771                 default:
3772                         break;
3773                 }
3774         }
3775
3776         if (boot_cpu_has(X86_FEATURE_XSAVES))
3777                 rdmsrl(MSR_IA32_XSS, host_xss);
3778
3779         return 0;
3780 }
3781
3782 static struct vmcs *alloc_vmcs_cpu(int cpu)
3783 {
3784         int node = cpu_to_node(cpu);
3785         struct page *pages;
3786         struct vmcs *vmcs;
3787
3788         pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3789         if (!pages)
3790                 return NULL;
3791         vmcs = page_address(pages);
3792         memset(vmcs, 0, vmcs_config.size);
3793         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3794         return vmcs;
3795 }
3796
3797 static struct vmcs *alloc_vmcs(void)
3798 {
3799         return alloc_vmcs_cpu(raw_smp_processor_id());
3800 }
3801
3802 static void free_vmcs(struct vmcs *vmcs)
3803 {
3804         free_pages((unsigned long)vmcs, vmcs_config.order);
3805 }
3806
3807 /*
3808  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3809  */
3810 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3811 {
3812         if (!loaded_vmcs->vmcs)
3813                 return;
3814         loaded_vmcs_clear(loaded_vmcs);
3815         free_vmcs(loaded_vmcs->vmcs);
3816         loaded_vmcs->vmcs = NULL;
3817         WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3818 }
3819
3820 static void free_kvm_area(void)
3821 {
3822         int cpu;
3823
3824         for_each_possible_cpu(cpu) {
3825                 free_vmcs(per_cpu(vmxarea, cpu));
3826                 per_cpu(vmxarea, cpu) = NULL;
3827         }
3828 }
3829
3830 enum vmcs_field_type {
3831         VMCS_FIELD_TYPE_U16 = 0,
3832         VMCS_FIELD_TYPE_U64 = 1,
3833         VMCS_FIELD_TYPE_U32 = 2,
3834         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
3835 };
3836
3837 static inline int vmcs_field_type(unsigned long field)
3838 {
3839         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
3840                 return VMCS_FIELD_TYPE_U32;
3841         return (field >> 13) & 0x3 ;
3842 }
3843
3844 static inline int vmcs_field_readonly(unsigned long field)
3845 {
3846         return (((field >> 10) & 0x3) == 1);
3847 }
3848
3849 static void init_vmcs_shadow_fields(void)
3850 {
3851         int i, j;
3852
3853         /* No checks for read only fields yet */
3854
3855         for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3856                 switch (shadow_read_write_fields[i]) {
3857                 case GUEST_BNDCFGS:
3858                         if (!kvm_mpx_supported())
3859                                 continue;
3860                         break;
3861                 default:
3862                         break;
3863                 }
3864
3865                 if (j < i)
3866                         shadow_read_write_fields[j] =
3867                                 shadow_read_write_fields[i];
3868                 j++;
3869         }
3870         max_shadow_read_write_fields = j;
3871
3872         /* shadowed fields guest access without vmexit */
3873         for (i = 0; i < max_shadow_read_write_fields; i++) {
3874                 unsigned long field = shadow_read_write_fields[i];
3875
3876                 clear_bit(field, vmx_vmwrite_bitmap);
3877                 clear_bit(field, vmx_vmread_bitmap);
3878                 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64) {
3879                         clear_bit(field + 1, vmx_vmwrite_bitmap);
3880                         clear_bit(field + 1, vmx_vmread_bitmap);
3881                 }
3882         }
3883         for (i = 0; i < max_shadow_read_only_fields; i++) {
3884                 unsigned long field = shadow_read_only_fields[i];
3885
3886                 clear_bit(field, vmx_vmread_bitmap);
3887                 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64)
3888                         clear_bit(field + 1, vmx_vmread_bitmap);
3889         }
3890 }
3891
3892 static __init int alloc_kvm_area(void)
3893 {
3894         int cpu;
3895
3896         for_each_possible_cpu(cpu) {
3897                 struct vmcs *vmcs;
3898
3899                 vmcs = alloc_vmcs_cpu(cpu);
3900                 if (!vmcs) {
3901                         free_kvm_area();
3902                         return -ENOMEM;
3903                 }
3904
3905                 per_cpu(vmxarea, cpu) = vmcs;
3906         }
3907         return 0;
3908 }
3909
3910 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3911                 struct kvm_segment *save)
3912 {
3913         if (!emulate_invalid_guest_state) {
3914                 /*
3915                  * CS and SS RPL should be equal during guest entry according
3916                  * to VMX spec, but in reality it is not always so. Since vcpu
3917                  * is in the middle of the transition from real mode to
3918                  * protected mode it is safe to assume that RPL 0 is a good
3919                  * default value.
3920                  */
3921                 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3922                         save->selector &= ~SEGMENT_RPL_MASK;
3923                 save->dpl = save->selector & SEGMENT_RPL_MASK;
3924                 save->s = 1;
3925         }
3926         vmx_set_segment(vcpu, save, seg);
3927 }
3928
3929 static void enter_pmode(struct kvm_vcpu *vcpu)
3930 {
3931         unsigned long flags;
3932         struct vcpu_vmx *vmx = to_vmx(vcpu);
3933
3934         /*
3935          * Update real mode segment cache. It may be not up-to-date if sement
3936          * register was written while vcpu was in a guest mode.
3937          */
3938         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3939         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3940         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3941         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3942         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3943         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3944
3945         vmx->rmode.vm86_active = 0;
3946
3947         vmx_segment_cache_clear(vmx);
3948
3949         vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3950
3951         flags = vmcs_readl(GUEST_RFLAGS);
3952         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3953         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3954         vmcs_writel(GUEST_RFLAGS, flags);
3955
3956         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3957                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3958
3959         update_exception_bitmap(vcpu);
3960
3961         fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3962         fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3963         fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3964         fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3965         fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3966         fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3967 }
3968
3969 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3970 {
3971         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3972         struct kvm_segment var = *save;
3973
3974         var.dpl = 0x3;
3975         if (seg == VCPU_SREG_CS)
3976                 var.type = 0x3;
3977
3978         if (!emulate_invalid_guest_state) {
3979                 var.selector = var.base >> 4;
3980                 var.base = var.base & 0xffff0;
3981                 var.limit = 0xffff;
3982                 var.g = 0;
3983                 var.db = 0;
3984                 var.present = 1;
3985                 var.s = 1;
3986                 var.l = 0;
3987                 var.unusable = 0;
3988                 var.type = 0x3;
3989                 var.avl = 0;
3990                 if (save->base & 0xf)
3991                         printk_once(KERN_WARNING "kvm: segment base is not "
3992                                         "paragraph aligned when entering "
3993                                         "protected mode (seg=%d)", seg);
3994         }
3995
3996         vmcs_write16(sf->selector, var.selector);
3997         vmcs_writel(sf->base, var.base);
3998         vmcs_write32(sf->limit, var.limit);
3999         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4000 }
4001
4002 static void enter_rmode(struct kvm_vcpu *vcpu)
4003 {
4004         unsigned long flags;
4005         struct vcpu_vmx *vmx = to_vmx(vcpu);
4006
4007         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4008         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4009         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4010         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4011         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4012         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4013         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4014
4015         vmx->rmode.vm86_active = 1;
4016
4017         /*
4018          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4019          * vcpu. Warn the user that an update is overdue.
4020          */
4021         if (!vcpu->kvm->arch.tss_addr)
4022                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4023                              "called before entering vcpu\n");
4024
4025         vmx_segment_cache_clear(vmx);
4026
4027         vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
4028         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4029         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4030
4031         flags = vmcs_readl(GUEST_RFLAGS);
4032         vmx->rmode.save_rflags = flags;
4033
4034         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4035
4036         vmcs_writel(GUEST_RFLAGS, flags);
4037         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4038         update_exception_bitmap(vcpu);
4039
4040         fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4041         fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4042         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4043         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4044         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4045         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4046
4047         kvm_mmu_reset_context(vcpu);
4048 }
4049
4050 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4051 {
4052         struct vcpu_vmx *vmx = to_vmx(vcpu);
4053         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4054
4055         if (!msr)
4056                 return;
4057
4058         /*
4059          * Force kernel_gs_base reloading before EFER changes, as control
4060          * of this msr depends on is_long_mode().
4061          */
4062         vmx_load_host_state(to_vmx(vcpu));
4063         vcpu->arch.efer = efer;
4064         if (efer & EFER_LMA) {
4065                 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4066                 msr->data = efer;
4067         } else {
4068                 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4069
4070                 msr->data = efer & ~EFER_LME;
4071         }
4072         setup_msrs(vmx);
4073 }
4074
4075 #ifdef CONFIG_X86_64
4076
4077 static void enter_lmode(struct kvm_vcpu *vcpu)
4078 {
4079         u32 guest_tr_ar;
4080
4081         vmx_segment_cache_clear(to_vmx(vcpu));
4082
4083         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4084         if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4085                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4086                                      __func__);
4087                 vmcs_write32(GUEST_TR_AR_BYTES,
4088                              (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4089                              | VMX_AR_TYPE_BUSY_64_TSS);
4090         }
4091         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4092 }
4093
4094 static void exit_lmode(struct kvm_vcpu *vcpu)
4095 {
4096         vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4097         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4098 }
4099
4100 #endif
4101
4102 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4103 {
4104         if (enable_ept) {
4105                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4106                         return;
4107                 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4108         } else {
4109                 vpid_sync_context(vpid);
4110         }
4111 }
4112
4113 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4114 {
4115         __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4116 }
4117
4118 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4119 {
4120         if (enable_ept)
4121                 vmx_flush_tlb(vcpu);
4122 }
4123
4124 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4125 {
4126         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4127
4128         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4129         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4130 }
4131
4132 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4133 {
4134         if (enable_ept && is_paging(vcpu))
4135                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4136         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4137 }
4138
4139 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4140 {
4141         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4142
4143         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4144         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4145 }
4146
4147 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4148 {
4149         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4150
4151         if (!test_bit(VCPU_EXREG_PDPTR,
4152                       (unsigned long *)&vcpu->arch.regs_dirty))
4153                 return;
4154
4155         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4156                 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4157                 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4158                 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4159                 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4160         }
4161 }
4162
4163 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4164 {
4165         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4166
4167         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4168                 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4169                 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4170                 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4171                 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4172         }
4173
4174         __set_bit(VCPU_EXREG_PDPTR,
4175                   (unsigned long *)&vcpu->arch.regs_avail);
4176         __set_bit(VCPU_EXREG_PDPTR,
4177                   (unsigned long *)&vcpu->arch.regs_dirty);
4178 }
4179
4180 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4181 {
4182         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4183         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4184         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4185
4186         if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4187                 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4188             nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4189                 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4190
4191         return fixed_bits_valid(val, fixed0, fixed1);
4192 }
4193
4194 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4195 {
4196         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4197         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4198
4199         return fixed_bits_valid(val, fixed0, fixed1);
4200 }
4201
4202 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4203 {
4204         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4205         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4206
4207         return fixed_bits_valid(val, fixed0, fixed1);
4208 }
4209
4210 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4211 #define nested_guest_cr4_valid  nested_cr4_valid
4212 #define nested_host_cr4_valid   nested_cr4_valid
4213
4214 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4215
4216 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4217                                         unsigned long cr0,
4218                                         struct kvm_vcpu *vcpu)
4219 {
4220         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4221                 vmx_decache_cr3(vcpu);
4222         if (!(cr0 & X86_CR0_PG)) {
4223                 /* From paging/starting to nonpaging */
4224                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4225                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4226                              (CPU_BASED_CR3_LOAD_EXITING |
4227                               CPU_BASED_CR3_STORE_EXITING));
4228                 vcpu->arch.cr0 = cr0;
4229                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4230         } else if (!is_paging(vcpu)) {
4231                 /* From nonpaging to paging */
4232                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4233                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4234                              ~(CPU_BASED_CR3_LOAD_EXITING |
4235                                CPU_BASED_CR3_STORE_EXITING));
4236                 vcpu->arch.cr0 = cr0;
4237                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4238         }
4239
4240         if (!(cr0 & X86_CR0_WP))
4241                 *hw_cr0 &= ~X86_CR0_WP;
4242 }
4243
4244 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4245 {
4246         struct vcpu_vmx *vmx = to_vmx(vcpu);
4247         unsigned long hw_cr0;
4248
4249         hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4250         if (enable_unrestricted_guest)
4251                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4252         else {
4253                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4254
4255                 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4256                         enter_pmode(vcpu);
4257
4258                 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4259                         enter_rmode(vcpu);
4260         }
4261
4262 #ifdef CONFIG_X86_64
4263         if (vcpu->arch.efer & EFER_LME) {
4264                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4265                         enter_lmode(vcpu);
4266                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4267                         exit_lmode(vcpu);
4268         }
4269 #endif
4270
4271         if (enable_ept)
4272                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4273
4274         vmcs_writel(CR0_READ_SHADOW, cr0);
4275         vmcs_writel(GUEST_CR0, hw_cr0);
4276         vcpu->arch.cr0 = cr0;
4277
4278         /* depends on vcpu->arch.cr0 to be set to a new value */
4279         vmx->emulation_required = emulation_required(vcpu);
4280 }
4281
4282 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4283 {
4284         u64 eptp;
4285
4286         /* TODO write the value reading from MSR */
4287         eptp = VMX_EPT_DEFAULT_MT |
4288                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
4289         if (enable_ept_ad_bits &&
4290             (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4291                 eptp |= VMX_EPT_AD_ENABLE_BIT;
4292         eptp |= (root_hpa & PAGE_MASK);
4293
4294         return eptp;
4295 }
4296
4297 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4298 {
4299         unsigned long guest_cr3;
4300         u64 eptp;
4301
4302         guest_cr3 = cr3;
4303         if (enable_ept) {
4304                 eptp = construct_eptp(vcpu, cr3);
4305                 vmcs_write64(EPT_POINTER, eptp);
4306                 if (is_paging(vcpu) || is_guest_mode(vcpu))
4307                         guest_cr3 = kvm_read_cr3(vcpu);
4308                 else
4309                         guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4310                 ept_load_pdptrs(vcpu);
4311         }
4312
4313         vmx_flush_tlb(vcpu);
4314         vmcs_writel(GUEST_CR3, guest_cr3);
4315 }
4316
4317 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4318 {
4319         /*
4320          * Pass through host's Machine Check Enable value to hw_cr4, which
4321          * is in force while we are in guest mode.  Do not let guests control
4322          * this bit, even if host CR4.MCE == 0.
4323          */
4324         unsigned long hw_cr4 =
4325                 (cr4_read_shadow() & X86_CR4_MCE) |
4326                 (cr4 & ~X86_CR4_MCE) |
4327                 (to_vmx(vcpu)->rmode.vm86_active ?
4328                  KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4329
4330         if (cr4 & X86_CR4_VMXE) {
4331                 /*
4332                  * To use VMXON (and later other VMX instructions), a guest
4333                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
4334                  * So basically the check on whether to allow nested VMX
4335                  * is here.
4336                  */
4337                 if (!nested_vmx_allowed(vcpu))
4338                         return 1;
4339         }
4340
4341         if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4342                 return 1;
4343
4344         vcpu->arch.cr4 = cr4;
4345         if (enable_ept) {
4346                 if (!is_paging(vcpu)) {
4347                         hw_cr4 &= ~X86_CR4_PAE;
4348                         hw_cr4 |= X86_CR4_PSE;
4349                 } else if (!(cr4 & X86_CR4_PAE)) {
4350                         hw_cr4 &= ~X86_CR4_PAE;
4351                 }
4352         }
4353
4354         if (!enable_unrestricted_guest && !is_paging(vcpu))
4355                 /*
4356                  * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4357                  * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
4358                  * to be manually disabled when guest switches to non-paging
4359                  * mode.
4360                  *
4361                  * If !enable_unrestricted_guest, the CPU is always running
4362                  * with CR0.PG=1 and CR4 needs to be modified.
4363                  * If enable_unrestricted_guest, the CPU automatically
4364                  * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4365                  */
4366                 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4367
4368         vmcs_writel(CR4_READ_SHADOW, cr4);
4369         vmcs_writel(GUEST_CR4, hw_cr4);
4370         return 0;
4371 }
4372
4373 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4374                             struct kvm_segment *var, int seg)
4375 {
4376         struct vcpu_vmx *vmx = to_vmx(vcpu);
4377         u32 ar;
4378
4379         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4380                 *var = vmx->rmode.segs[seg];
4381                 if (seg == VCPU_SREG_TR
4382                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4383                         return;
4384                 var->base = vmx_read_guest_seg_base(vmx, seg);
4385                 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4386                 return;
4387         }
4388         var->base = vmx_read_guest_seg_base(vmx, seg);
4389         var->limit = vmx_read_guest_seg_limit(vmx, seg);
4390         var->selector = vmx_read_guest_seg_selector(vmx, seg);
4391         ar = vmx_read_guest_seg_ar(vmx, seg);
4392         var->unusable = (ar >> 16) & 1;
4393         var->type = ar & 15;
4394         var->s = (ar >> 4) & 1;
4395         var->dpl = (ar >> 5) & 3;
4396         /*
4397          * Some userspaces do not preserve unusable property. Since usable
4398          * segment has to be present according to VMX spec we can use present
4399          * property to amend userspace bug by making unusable segment always
4400          * nonpresent. vmx_segment_access_rights() already marks nonpresent
4401          * segment as unusable.
4402          */
4403         var->present = !var->unusable;
4404         var->avl = (ar >> 12) & 1;
4405         var->l = (ar >> 13) & 1;
4406         var->db = (ar >> 14) & 1;
4407         var->g = (ar >> 15) & 1;
4408 }
4409
4410 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4411 {
4412         struct kvm_segment s;
4413
4414         if (to_vmx(vcpu)->rmode.vm86_active) {
4415                 vmx_get_segment(vcpu, &s, seg);
4416                 return s.base;
4417         }
4418         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4419 }
4420
4421 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4422 {
4423         struct vcpu_vmx *vmx = to_vmx(vcpu);
4424
4425         if (unlikely(vmx->rmode.vm86_active))
4426                 return 0;
4427         else {
4428                 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4429                 return VMX_AR_DPL(ar);
4430         }
4431 }
4432
4433 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4434 {
4435         u32 ar;
4436
4437         if (var->unusable || !var->present)
4438                 ar = 1 << 16;
4439         else {
4440                 ar = var->type & 15;
4441                 ar |= (var->s & 1) << 4;
4442                 ar |= (var->dpl & 3) << 5;
4443                 ar |= (var->present & 1) << 7;
4444                 ar |= (var->avl & 1) << 12;
4445                 ar |= (var->l & 1) << 13;
4446                 ar |= (var->db & 1) << 14;
4447                 ar |= (var->g & 1) << 15;
4448         }
4449
4450         return ar;
4451 }
4452
4453 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4454                             struct kvm_segment *var, int seg)
4455 {
4456         struct vcpu_vmx *vmx = to_vmx(vcpu);
4457         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4458
4459         vmx_segment_cache_clear(vmx);
4460
4461         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4462                 vmx->rmode.segs[seg] = *var;
4463                 if (seg == VCPU_SREG_TR)
4464                         vmcs_write16(sf->selector, var->selector);
4465                 else if (var->s)
4466                         fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4467                 goto out;
4468         }
4469
4470         vmcs_writel(sf->base, var->base);
4471         vmcs_write32(sf->limit, var->limit);
4472         vmcs_write16(sf->selector, var->selector);
4473
4474         /*
4475          *   Fix the "Accessed" bit in AR field of segment registers for older
4476          * qemu binaries.
4477          *   IA32 arch specifies that at the time of processor reset the
4478          * "Accessed" bit in the AR field of segment registers is 1. And qemu
4479          * is setting it to 0 in the userland code. This causes invalid guest
4480          * state vmexit when "unrestricted guest" mode is turned on.
4481          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
4482          * tree. Newer qemu binaries with that qemu fix would not need this
4483          * kvm hack.
4484          */
4485         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4486                 var->type |= 0x1; /* Accessed */
4487
4488         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4489
4490 out:
4491         vmx->emulation_required = emulation_required(vcpu);
4492 }
4493
4494 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4495 {
4496         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4497
4498         *db = (ar >> 14) & 1;
4499         *l = (ar >> 13) & 1;
4500 }
4501
4502 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4503 {
4504         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4505         dt->address = vmcs_readl(GUEST_IDTR_BASE);
4506 }
4507
4508 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4509 {
4510         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4511         vmcs_writel(GUEST_IDTR_BASE, dt->address);
4512 }
4513
4514 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4515 {
4516         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4517         dt->address = vmcs_readl(GUEST_GDTR_BASE);
4518 }
4519
4520 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4521 {
4522         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4523         vmcs_writel(GUEST_GDTR_BASE, dt->address);
4524 }
4525
4526 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4527 {
4528         struct kvm_segment var;
4529         u32 ar;
4530
4531         vmx_get_segment(vcpu, &var, seg);
4532         var.dpl = 0x3;
4533         if (seg == VCPU_SREG_CS)
4534                 var.type = 0x3;
4535         ar = vmx_segment_access_rights(&var);
4536
4537         if (var.base != (var.selector << 4))
4538                 return false;
4539         if (var.limit != 0xffff)
4540                 return false;
4541         if (ar != 0xf3)
4542                 return false;
4543
4544         return true;
4545 }
4546
4547 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4548 {
4549         struct kvm_segment cs;
4550         unsigned int cs_rpl;
4551
4552         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4553         cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4554
4555         if (cs.unusable)
4556                 return false;
4557         if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4558                 return false;
4559         if (!cs.s)
4560                 return false;
4561         if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4562                 if (cs.dpl > cs_rpl)
4563                         return false;
4564         } else {
4565                 if (cs.dpl != cs_rpl)
4566                         return false;
4567         }
4568         if (!cs.present)
4569                 return false;
4570
4571         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4572         return true;
4573 }
4574
4575 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4576 {
4577         struct kvm_segment ss;
4578         unsigned int ss_rpl;
4579
4580         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4581         ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4582
4583         if (ss.unusable)
4584                 return true;
4585         if (ss.type != 3 && ss.type != 7)
4586                 return false;
4587         if (!ss.s)
4588                 return false;
4589         if (ss.dpl != ss_rpl) /* DPL != RPL */
4590                 return false;
4591         if (!ss.present)
4592                 return false;
4593
4594         return true;
4595 }
4596
4597 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4598 {
4599         struct kvm_segment var;
4600         unsigned int rpl;
4601
4602         vmx_get_segment(vcpu, &var, seg);
4603         rpl = var.selector & SEGMENT_RPL_MASK;
4604
4605         if (var.unusable)
4606                 return true;
4607         if (!var.s)
4608                 return false;
4609         if (!var.present)
4610                 return false;
4611         if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4612                 if (var.dpl < rpl) /* DPL < RPL */
4613                         return false;
4614         }
4615
4616         /* TODO: Add other members to kvm_segment_field to allow checking for other access
4617          * rights flags
4618          */
4619         return true;
4620 }
4621
4622 static bool tr_valid(struct kvm_vcpu *vcpu)
4623 {
4624         struct kvm_segment tr;
4625
4626         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4627
4628         if (tr.unusable)
4629                 return false;
4630         if (tr.selector & SEGMENT_TI_MASK)      /* TI = 1 */
4631                 return false;
4632         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4633                 return false;
4634         if (!tr.present)
4635                 return false;
4636
4637         return true;
4638 }
4639
4640 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4641 {
4642         struct kvm_segment ldtr;
4643
4644         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4645
4646         if (ldtr.unusable)
4647                 return true;
4648         if (ldtr.selector & SEGMENT_TI_MASK)    /* TI = 1 */
4649                 return false;
4650         if (ldtr.type != 2)
4651                 return false;
4652         if (!ldtr.present)
4653                 return false;
4654
4655         return true;
4656 }
4657
4658 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4659 {
4660         struct kvm_segment cs, ss;
4661
4662         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4663         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4664
4665         return ((cs.selector & SEGMENT_RPL_MASK) ==
4666                  (ss.selector & SEGMENT_RPL_MASK));
4667 }
4668
4669 /*
4670  * Check if guest state is valid. Returns true if valid, false if
4671  * not.
4672  * We assume that registers are always usable
4673  */
4674 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4675 {
4676         if (enable_unrestricted_guest)
4677                 return true;
4678
4679         /* real mode guest state checks */
4680         if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4681                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4682                         return false;
4683                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4684                         return false;
4685                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4686                         return false;
4687                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4688                         return false;
4689                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4690                         return false;
4691                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4692                         return false;
4693         } else {
4694         /* protected mode guest state checks */
4695                 if (!cs_ss_rpl_check(vcpu))
4696                         return false;
4697                 if (!code_segment_valid(vcpu))
4698                         return false;
4699                 if (!stack_segment_valid(vcpu))
4700                         return false;
4701                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4702                         return false;
4703                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4704                         return false;
4705                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4706                         return false;
4707                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4708                         return false;
4709                 if (!tr_valid(vcpu))
4710                         return false;
4711                 if (!ldtr_valid(vcpu))
4712                         return false;
4713         }
4714         /* TODO:
4715          * - Add checks on RIP
4716          * - Add checks on RFLAGS
4717          */
4718
4719         return true;
4720 }
4721
4722 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
4723 {
4724         return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
4725 }
4726
4727 static int init_rmode_tss(struct kvm *kvm)
4728 {
4729         gfn_t fn;
4730         u16 data = 0;
4731         int idx, r;
4732
4733         idx = srcu_read_lock(&kvm->srcu);
4734         fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4735         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4736         if (r < 0)
4737                 goto out;
4738         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4739         r = kvm_write_guest_page(kvm, fn++, &data,
4740                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
4741         if (r < 0)
4742                 goto out;
4743         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4744         if (r < 0)
4745                 goto out;
4746         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4747         if (r < 0)
4748                 goto out;
4749         data = ~0;
4750         r = kvm_write_guest_page(kvm, fn, &data,
4751                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4752                                  sizeof(u8));
4753 out:
4754         srcu_read_unlock(&kvm->srcu, idx);
4755         return r;
4756 }
4757
4758 static int init_rmode_identity_map(struct kvm *kvm)
4759 {
4760         int i, idx, r = 0;
4761         kvm_pfn_t identity_map_pfn;
4762         u32 tmp;
4763
4764         if (!enable_ept)
4765                 return 0;
4766
4767         /* Protect kvm->arch.ept_identity_pagetable_done. */
4768         mutex_lock(&kvm->slots_lock);
4769
4770         if (likely(kvm->arch.ept_identity_pagetable_done))
4771                 goto out2;
4772
4773         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4774
4775         r = alloc_identity_pagetable(kvm);
4776         if (r < 0)
4777                 goto out2;
4778
4779         idx = srcu_read_lock(&kvm->srcu);
4780         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4781         if (r < 0)
4782                 goto out;
4783         /* Set up identity-mapping pagetable for EPT in real mode */
4784         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4785                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4786                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4787                 r = kvm_write_guest_page(kvm, identity_map_pfn,
4788                                 &tmp, i * sizeof(tmp), sizeof(tmp));
4789                 if (r < 0)
4790                         goto out;
4791         }
4792         kvm->arch.ept_identity_pagetable_done = true;
4793
4794 out:
4795         srcu_read_unlock(&kvm->srcu, idx);
4796
4797 out2:
4798         mutex_unlock(&kvm->slots_lock);
4799         return r;
4800 }
4801
4802 static void seg_setup(int seg)
4803 {
4804         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4805         unsigned int ar;
4806
4807         vmcs_write16(sf->selector, 0);
4808         vmcs_writel(sf->base, 0);
4809         vmcs_write32(sf->limit, 0xffff);
4810         ar = 0x93;
4811         if (seg == VCPU_SREG_CS)
4812                 ar |= 0x08; /* code segment */
4813
4814         vmcs_write32(sf->ar_bytes, ar);
4815 }
4816
4817 static int alloc_apic_access_page(struct kvm *kvm)
4818 {
4819         struct page *page;
4820         int r = 0;
4821
4822         mutex_lock(&kvm->slots_lock);
4823         if (kvm->arch.apic_access_page_done)
4824                 goto out;
4825         r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4826                                     APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4827         if (r)
4828                 goto out;
4829
4830         page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4831         if (is_error_page(page)) {
4832                 r = -EFAULT;
4833                 goto out;
4834         }
4835
4836         /*
4837          * Do not pin the page in memory, so that memory hot-unplug
4838          * is able to migrate it.
4839          */
4840         put_page(page);
4841         kvm->arch.apic_access_page_done = true;
4842 out:
4843         mutex_unlock(&kvm->slots_lock);
4844         return r;
4845 }
4846
4847 static int alloc_identity_pagetable(struct kvm *kvm)
4848 {
4849         /* Called with kvm->slots_lock held. */
4850
4851         int r = 0;
4852
4853         BUG_ON(kvm->arch.ept_identity_pagetable_done);
4854
4855         r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4856                                     kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4857
4858         return r;
4859 }
4860
4861 static int allocate_vpid(void)
4862 {
4863         int vpid;
4864
4865         if (!enable_vpid)
4866                 return 0;
4867         spin_lock(&vmx_vpid_lock);
4868         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4869         if (vpid < VMX_NR_VPIDS)
4870                 __set_bit(vpid, vmx_vpid_bitmap);
4871         else
4872                 vpid = 0;
4873         spin_unlock(&vmx_vpid_lock);
4874         return vpid;
4875 }
4876
4877 static void free_vpid(int vpid)
4878 {
4879         if (!enable_vpid || vpid == 0)
4880                 return;
4881         spin_lock(&vmx_vpid_lock);
4882         __clear_bit(vpid, vmx_vpid_bitmap);
4883         spin_unlock(&vmx_vpid_lock);
4884 }
4885
4886 #define MSR_TYPE_R      1
4887 #define MSR_TYPE_W      2
4888 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4889                                                 u32 msr, int type)
4890 {
4891         int f = sizeof(unsigned long);
4892
4893         if (!cpu_has_vmx_msr_bitmap())
4894                 return;
4895
4896         /*
4897          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4898          * have the write-low and read-high bitmap offsets the wrong way round.
4899          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4900          */
4901         if (msr <= 0x1fff) {
4902                 if (type & MSR_TYPE_R)
4903                         /* read-low */
4904                         __clear_bit(msr, msr_bitmap + 0x000 / f);
4905
4906                 if (type & MSR_TYPE_W)
4907                         /* write-low */
4908                         __clear_bit(msr, msr_bitmap + 0x800 / f);
4909
4910         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4911                 msr &= 0x1fff;
4912                 if (type & MSR_TYPE_R)
4913                         /* read-high */
4914                         __clear_bit(msr, msr_bitmap + 0x400 / f);
4915
4916                 if (type & MSR_TYPE_W)
4917                         /* write-high */
4918                         __clear_bit(msr, msr_bitmap + 0xc00 / f);
4919
4920         }
4921 }
4922
4923 /*
4924  * If a msr is allowed by L0, we should check whether it is allowed by L1.
4925  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4926  */
4927 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4928                                                unsigned long *msr_bitmap_nested,
4929                                                u32 msr, int type)
4930 {
4931         int f = sizeof(unsigned long);
4932
4933         if (!cpu_has_vmx_msr_bitmap()) {
4934                 WARN_ON(1);
4935                 return;
4936         }
4937
4938         /*
4939          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4940          * have the write-low and read-high bitmap offsets the wrong way round.
4941          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4942          */
4943         if (msr <= 0x1fff) {
4944                 if (type & MSR_TYPE_R &&
4945                    !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4946                         /* read-low */
4947                         __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4948
4949                 if (type & MSR_TYPE_W &&
4950                    !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4951                         /* write-low */
4952                         __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4953
4954         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4955                 msr &= 0x1fff;
4956                 if (type & MSR_TYPE_R &&
4957                    !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4958                         /* read-high */
4959                         __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4960
4961                 if (type & MSR_TYPE_W &&
4962                    !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4963                         /* write-high */
4964                         __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4965
4966         }
4967 }
4968
4969 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4970 {
4971         if (!longmode_only)
4972                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4973                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
4974         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4975                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
4976 }
4977
4978 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
4979 {
4980         if (apicv_active) {
4981                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
4982                                 msr, type);
4983                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
4984                                 msr, type);
4985         } else {
4986                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4987                                 msr, type);
4988                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4989                                 msr, type);
4990         }
4991 }
4992
4993 static bool vmx_get_enable_apicv(void)
4994 {
4995         return enable_apicv;
4996 }
4997
4998 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
4999 {
5000         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5001         gfn_t gfn;
5002
5003         /*
5004          * Don't need to mark the APIC access page dirty; it is never
5005          * written to by the CPU during APIC virtualization.
5006          */
5007
5008         if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
5009                 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
5010                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5011         }
5012
5013         if (nested_cpu_has_posted_intr(vmcs12)) {
5014                 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
5015                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5016         }
5017 }
5018
5019
5020 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5021 {
5022         struct vcpu_vmx *vmx = to_vmx(vcpu);
5023         int max_irr;
5024         void *vapic_page;
5025         u16 status;
5026
5027         if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
5028                 return;
5029
5030         vmx->nested.pi_pending = false;
5031         if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5032                 return;
5033
5034         max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
5035         if (max_irr != 256) {
5036                 vapic_page = kmap(vmx->nested.virtual_apic_page);
5037                 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
5038                 kunmap(vmx->nested.virtual_apic_page);
5039
5040                 status = vmcs_read16(GUEST_INTR_STATUS);
5041                 if ((u8)max_irr > ((u8)status & 0xff)) {
5042                         status &= ~0xff;
5043                         status |= (u8)max_irr;
5044                         vmcs_write16(GUEST_INTR_STATUS, status);
5045                 }
5046         }
5047
5048         nested_mark_vmcs12_pages_dirty(vcpu);
5049 }
5050
5051 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5052                                                      bool nested)
5053 {
5054 #ifdef CONFIG_SMP
5055         int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5056
5057         if (vcpu->mode == IN_GUEST_MODE) {
5058                 struct vcpu_vmx *vmx = to_vmx(vcpu);
5059
5060                 /*
5061                  * Currently, we don't support urgent interrupt,
5062                  * all interrupts are recognized as non-urgent
5063                  * interrupt, so we cannot post interrupts when
5064                  * 'SN' is set.
5065                  *
5066                  * If the vcpu is in guest mode, it means it is
5067                  * running instead of being scheduled out and
5068                  * waiting in the run queue, and that's the only
5069                  * case when 'SN' is set currently, warning if
5070                  * 'SN' is set.
5071                  */
5072                 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
5073
5074                 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5075                 return true;
5076         }
5077 #endif
5078         return false;
5079 }
5080
5081 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5082                                                 int vector)
5083 {
5084         struct vcpu_vmx *vmx = to_vmx(vcpu);
5085
5086         if (is_guest_mode(vcpu) &&
5087             vector == vmx->nested.posted_intr_nv) {
5088                 /* the PIR and ON have been set by L1. */
5089                 kvm_vcpu_trigger_posted_interrupt(vcpu, true);
5090                 /*
5091                  * If a posted intr is not recognized by hardware,
5092                  * we will accomplish it in the next vmentry.
5093                  */
5094                 vmx->nested.pi_pending = true;
5095                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5096                 return 0;
5097         }
5098         return -1;
5099 }
5100 /*
5101  * Send interrupt to vcpu via posted interrupt way.
5102  * 1. If target vcpu is running(non-root mode), send posted interrupt
5103  * notification to vcpu and hardware will sync PIR to vIRR atomically.
5104  * 2. If target vcpu isn't running(root mode), kick it to pick up the
5105  * interrupt from PIR in next vmentry.
5106  */
5107 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5108 {
5109         struct vcpu_vmx *vmx = to_vmx(vcpu);
5110         int r;
5111
5112         r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5113         if (!r)
5114                 return;
5115
5116         if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5117                 return;
5118
5119         /* If a previous notification has sent the IPI, nothing to do.  */
5120         if (pi_test_and_set_on(&vmx->pi_desc))
5121                 return;
5122
5123         if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5124                 kvm_vcpu_kick(vcpu);
5125 }
5126
5127 /*
5128  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5129  * will not change in the lifetime of the guest.
5130  * Note that host-state that does change is set elsewhere. E.g., host-state
5131  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5132  */
5133 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5134 {
5135         u32 low32, high32;
5136         unsigned long tmpl;
5137         struct desc_ptr dt;
5138         unsigned long cr0, cr3, cr4;
5139
5140         cr0 = read_cr0();
5141         WARN_ON(cr0 & X86_CR0_TS);
5142         vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
5143
5144         /*
5145          * Save the most likely value for this task's CR3 in the VMCS.
5146          * We can't use __get_current_cr3_fast() because we're not atomic.
5147          */
5148         cr3 = __read_cr3();
5149         vmcs_writel(HOST_CR3, cr3);             /* 22.2.3  FIXME: shadow tables */
5150         vmx->host_state.vmcs_host_cr3 = cr3;
5151
5152         /* Save the most likely value for this task's CR4 in the VMCS. */
5153         cr4 = cr4_read_shadow();
5154         vmcs_writel(HOST_CR4, cr4);                     /* 22.2.3, 22.2.5 */
5155         vmx->host_state.vmcs_host_cr4 = cr4;
5156
5157         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
5158 #ifdef CONFIG_X86_64
5159         /*
5160          * Load null selectors, so we can avoid reloading them in
5161          * __vmx_load_host_state(), in case userspace uses the null selectors
5162          * too (the expected case).
5163          */
5164         vmcs_write16(HOST_DS_SELECTOR, 0);
5165         vmcs_write16(HOST_ES_SELECTOR, 0);
5166 #else
5167         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5168         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5169 #endif
5170         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5171         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
5172
5173         native_store_idt(&dt);
5174         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
5175         vmx->host_idt_base = dt.address;
5176
5177         vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5178
5179         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5180         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5181         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5182         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
5183
5184         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5185                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5186                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5187         }
5188 }
5189
5190 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5191 {
5192         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5193         if (enable_ept)
5194                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5195         if (is_guest_mode(&vmx->vcpu))
5196                 vmx->vcpu.arch.cr4_guest_owned_bits &=
5197                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5198         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5199 }
5200
5201 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5202 {
5203         u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5204
5205         if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5206                 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5207         /* Enable the preemption timer dynamically */
5208         pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5209         return pin_based_exec_ctrl;
5210 }
5211
5212 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5213 {
5214         struct vcpu_vmx *vmx = to_vmx(vcpu);
5215
5216         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5217         if (cpu_has_secondary_exec_ctrls()) {
5218                 if (kvm_vcpu_apicv_active(vcpu))
5219                         vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5220                                       SECONDARY_EXEC_APIC_REGISTER_VIRT |
5221                                       SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5222                 else
5223                         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5224                                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
5225                                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5226         }
5227
5228         if (cpu_has_vmx_msr_bitmap())
5229                 vmx_set_msr_bitmap(vcpu);
5230 }
5231
5232 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5233 {
5234         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5235
5236         if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5237                 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5238
5239         if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5240                 exec_control &= ~CPU_BASED_TPR_SHADOW;
5241 #ifdef CONFIG_X86_64
5242                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5243                                 CPU_BASED_CR8_LOAD_EXITING;
5244 #endif
5245         }
5246         if (!enable_ept)
5247                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5248                                 CPU_BASED_CR3_LOAD_EXITING  |
5249                                 CPU_BASED_INVLPG_EXITING;
5250         return exec_control;
5251 }
5252
5253 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
5254 {
5255         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5256         if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
5257                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5258         if (vmx->vpid == 0)
5259                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5260         if (!enable_ept) {
5261                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5262                 enable_unrestricted_guest = 0;
5263                 /* Enable INVPCID for non-ept guests may cause performance regression. */
5264                 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5265         }
5266         if (!enable_unrestricted_guest)
5267                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5268         if (!ple_gap)
5269                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5270         if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5271                 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5272                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5273         exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5274         /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5275            (handle_vmptrld).
5276            We can NOT enable shadow_vmcs here because we don't have yet
5277            a current VMCS12
5278         */
5279         exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5280
5281         if (!enable_pml)
5282                 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5283
5284         return exec_control;
5285 }
5286
5287 static void ept_set_mmio_spte_mask(void)
5288 {
5289         /*
5290          * EPT Misconfigurations can be generated if the value of bits 2:0
5291          * of an EPT paging-structure entry is 110b (write/execute).
5292          */
5293         kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
5294                                    VMX_EPT_MISCONFIG_WX_VALUE);
5295 }
5296
5297 #define VMX_XSS_EXIT_BITMAP 0
5298 /*
5299  * Sets up the vmcs for emulated real mode.
5300  */
5301 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5302 {
5303 #ifdef CONFIG_X86_64
5304         unsigned long a;
5305 #endif
5306         int i;
5307
5308         /* I/O */
5309         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5310         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5311
5312         if (enable_shadow_vmcs) {
5313                 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5314                 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5315         }
5316         if (cpu_has_vmx_msr_bitmap())
5317                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5318
5319         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5320
5321         /* Control */
5322         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5323         vmx->hv_deadline_tsc = -1;
5324
5325         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5326
5327         if (cpu_has_secondary_exec_ctrls()) {
5328                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5329                                 vmx_secondary_exec_control(vmx));
5330         }
5331
5332         if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5333                 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5334                 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5335                 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5336                 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5337
5338                 vmcs_write16(GUEST_INTR_STATUS, 0);
5339
5340                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5341                 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5342         }
5343
5344         if (ple_gap) {
5345                 vmcs_write32(PLE_GAP, ple_gap);
5346                 vmx->ple_window = ple_window;
5347                 vmx->ple_window_dirty = true;
5348         }
5349
5350         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5351         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5352         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
5353
5354         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
5355         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
5356         vmx_set_constant_host_state(vmx);
5357 #ifdef CONFIG_X86_64
5358         rdmsrl(MSR_FS_BASE, a);
5359         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5360         rdmsrl(MSR_GS_BASE, a);
5361         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5362 #else
5363         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5364         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5365 #endif
5366
5367         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5368         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5369         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5370         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5371         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5372
5373         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5374                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5375
5376         for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5377                 u32 index = vmx_msr_index[i];
5378                 u32 data_low, data_high;
5379                 int j = vmx->nmsrs;
5380
5381                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5382                         continue;
5383                 if (wrmsr_safe(index, data_low, data_high) < 0)
5384                         continue;
5385                 vmx->guest_msrs[j].index = i;
5386                 vmx->guest_msrs[j].data = 0;
5387                 vmx->guest_msrs[j].mask = -1ull;
5388                 ++vmx->nmsrs;
5389         }
5390
5391
5392         vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5393
5394         /* 22.2.1, 20.8.1 */
5395         vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5396
5397         vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5398         vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5399
5400         set_cr4_guest_host_mask(vmx);
5401
5402         if (vmx_xsaves_supported())
5403                 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5404
5405         if (enable_pml) {
5406                 ASSERT(vmx->pml_pg);
5407                 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5408                 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5409         }
5410
5411         return 0;
5412 }
5413
5414 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5415 {
5416         struct vcpu_vmx *vmx = to_vmx(vcpu);
5417         struct msr_data apic_base_msr;
5418         u64 cr0;
5419
5420         vmx->rmode.vm86_active = 0;
5421
5422         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5423         kvm_set_cr8(vcpu, 0);
5424
5425         if (!init_event) {
5426                 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5427                                      MSR_IA32_APICBASE_ENABLE;
5428                 if (kvm_vcpu_is_reset_bsp(vcpu))
5429                         apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5430                 apic_base_msr.host_initiated = true;
5431                 kvm_set_apic_base(vcpu, &apic_base_msr);
5432         }
5433
5434         vmx_segment_cache_clear(vmx);
5435
5436         seg_setup(VCPU_SREG_CS);
5437         vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5438         vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5439
5440         seg_setup(VCPU_SREG_DS);
5441         seg_setup(VCPU_SREG_ES);
5442         seg_setup(VCPU_SREG_FS);
5443         seg_setup(VCPU_SREG_GS);
5444         seg_setup(VCPU_SREG_SS);
5445
5446         vmcs_write16(GUEST_TR_SELECTOR, 0);
5447         vmcs_writel(GUEST_TR_BASE, 0);
5448         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5449         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5450
5451         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5452         vmcs_writel(GUEST_LDTR_BASE, 0);
5453         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5454         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5455
5456         if (!init_event) {
5457                 vmcs_write32(GUEST_SYSENTER_CS, 0);
5458                 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5459                 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5460                 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5461         }
5462
5463         vmcs_writel(GUEST_RFLAGS, 0x02);
5464         kvm_rip_write(vcpu, 0xfff0);
5465
5466         vmcs_writel(GUEST_GDTR_BASE, 0);
5467         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5468
5469         vmcs_writel(GUEST_IDTR_BASE, 0);
5470         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5471
5472         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5473         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5474         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5475
5476         setup_msrs(vmx);
5477
5478         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
5479
5480         if (cpu_has_vmx_tpr_shadow() && !init_event) {
5481                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5482                 if (cpu_need_tpr_shadow(vcpu))
5483                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5484                                      __pa(vcpu->arch.apic->regs));
5485                 vmcs_write32(TPR_THRESHOLD, 0);
5486         }
5487
5488         kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5489
5490         if (kvm_vcpu_apicv_active(vcpu))
5491                 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5492
5493         if (vmx->vpid != 0)
5494                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5495
5496         cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5497         vmx->vcpu.arch.cr0 = cr0;
5498         vmx_set_cr0(vcpu, cr0); /* enter rmode */
5499         vmx_set_cr4(vcpu, 0);
5500         vmx_set_efer(vcpu, 0);
5501
5502         update_exception_bitmap(vcpu);
5503
5504         vpid_sync_context(vmx->vpid);
5505 }
5506
5507 /*
5508  * In nested virtualization, check if L1 asked to exit on external interrupts.
5509  * For most existing hypervisors, this will always return true.
5510  */
5511 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5512 {
5513         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5514                 PIN_BASED_EXT_INTR_MASK;
5515 }
5516
5517 /*
5518  * In nested virtualization, check if L1 has set
5519  * VM_EXIT_ACK_INTR_ON_EXIT
5520  */
5521 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5522 {
5523         return get_vmcs12(vcpu)->vm_exit_controls &
5524                 VM_EXIT_ACK_INTR_ON_EXIT;
5525 }
5526
5527 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5528 {
5529         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5530                 PIN_BASED_NMI_EXITING;
5531 }
5532
5533 static void enable_irq_window(struct kvm_vcpu *vcpu)
5534 {
5535         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5536                       CPU_BASED_VIRTUAL_INTR_PENDING);
5537 }
5538
5539 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5540 {
5541         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5542                 enable_irq_window(vcpu);
5543                 return;
5544         }
5545
5546         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5547                       CPU_BASED_VIRTUAL_NMI_PENDING);
5548 }
5549
5550 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5551 {
5552         struct vcpu_vmx *vmx = to_vmx(vcpu);
5553         uint32_t intr;
5554         int irq = vcpu->arch.interrupt.nr;
5555
5556         trace_kvm_inj_virq(irq);
5557
5558         ++vcpu->stat.irq_injections;
5559         if (vmx->rmode.vm86_active) {
5560                 int inc_eip = 0;
5561                 if (vcpu->arch.interrupt.soft)
5562                         inc_eip = vcpu->arch.event_exit_inst_len;
5563                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5564                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5565                 return;
5566         }
5567         intr = irq | INTR_INFO_VALID_MASK;
5568         if (vcpu->arch.interrupt.soft) {
5569                 intr |= INTR_TYPE_SOFT_INTR;
5570                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5571                              vmx->vcpu.arch.event_exit_inst_len);
5572         } else
5573                 intr |= INTR_TYPE_EXT_INTR;
5574         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5575 }
5576
5577 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5578 {
5579         struct vcpu_vmx *vmx = to_vmx(vcpu);
5580
5581         ++vcpu->stat.nmi_injections;
5582         vmx->loaded_vmcs->nmi_known_unmasked = false;
5583
5584         if (vmx->rmode.vm86_active) {
5585                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5586                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5587                 return;
5588         }
5589
5590         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5591                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5592 }
5593
5594 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5595 {
5596         struct vcpu_vmx *vmx = to_vmx(vcpu);
5597         bool masked;
5598
5599         if (vmx->loaded_vmcs->nmi_known_unmasked)
5600                 return false;
5601         masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5602         vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5603         return masked;
5604 }
5605
5606 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5607 {
5608         struct vcpu_vmx *vmx = to_vmx(vcpu);
5609
5610         vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5611         if (masked)
5612                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5613                               GUEST_INTR_STATE_NMI);
5614         else
5615                 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5616                                 GUEST_INTR_STATE_NMI);
5617 }
5618
5619 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5620 {
5621         if (to_vmx(vcpu)->nested.nested_run_pending)
5622                 return 0;
5623
5624         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5625                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5626                    | GUEST_INTR_STATE_NMI));
5627 }
5628
5629 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5630 {
5631         return (!to_vmx(vcpu)->nested.nested_run_pending &&
5632                 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5633                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5634                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5635 }
5636
5637 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5638 {
5639         int ret;
5640
5641         ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5642                                     PAGE_SIZE * 3);
5643         if (ret)
5644                 return ret;
5645         kvm->arch.tss_addr = addr;
5646         return init_rmode_tss(kvm);
5647 }
5648
5649 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5650 {
5651         switch (vec) {
5652         case BP_VECTOR:
5653                 /*
5654                  * Update instruction length as we may reinject the exception
5655                  * from user space while in guest debugging mode.
5656                  */
5657                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5658                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5659                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5660                         return false;
5661                 /* fall through */
5662         case DB_VECTOR:
5663                 if (vcpu->guest_debug &
5664                         (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5665                         return false;
5666                 /* fall through */
5667         case DE_VECTOR:
5668         case OF_VECTOR:
5669         case BR_VECTOR:
5670         case UD_VECTOR:
5671         case DF_VECTOR:
5672         case SS_VECTOR:
5673         case GP_VECTOR:
5674         case MF_VECTOR:
5675                 return true;
5676         break;
5677         }
5678         return false;
5679 }
5680
5681 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5682                                   int vec, u32 err_code)
5683 {
5684         /*
5685          * Instruction with address size override prefix opcode 0x67
5686          * Cause the #SS fault with 0 error code in VM86 mode.
5687          */
5688         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5689                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5690                         if (vcpu->arch.halt_request) {
5691                                 vcpu->arch.halt_request = 0;
5692                                 return kvm_vcpu_halt(vcpu);
5693                         }
5694                         return 1;
5695                 }
5696                 return 0;
5697         }
5698
5699         /*
5700          * Forward all other exceptions that are valid in real mode.
5701          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5702          *        the required debugging infrastructure rework.
5703          */
5704         kvm_queue_exception(vcpu, vec);
5705         return 1;
5706 }
5707
5708 /*
5709  * Trigger machine check on the host. We assume all the MSRs are already set up
5710  * by the CPU and that we still run on the same CPU as the MCE occurred on.
5711  * We pass a fake environment to the machine check handler because we want
5712  * the guest to be always treated like user space, no matter what context
5713  * it used internally.
5714  */
5715 static void kvm_machine_check(void)
5716 {
5717 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5718         struct pt_regs regs = {
5719                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5720                 .flags = X86_EFLAGS_IF,
5721         };
5722
5723         do_machine_check(&regs, 0);
5724 #endif
5725 }
5726
5727 static int handle_machine_check(struct kvm_vcpu *vcpu)
5728 {
5729         /* already handled by vcpu_run */
5730         return 1;
5731 }
5732
5733 static int handle_exception(struct kvm_vcpu *vcpu)
5734 {
5735         struct vcpu_vmx *vmx = to_vmx(vcpu);
5736         struct kvm_run *kvm_run = vcpu->run;
5737         u32 intr_info, ex_no, error_code;
5738         unsigned long cr2, rip, dr6;
5739         u32 vect_info;
5740         enum emulation_result er;
5741
5742         vect_info = vmx->idt_vectoring_info;
5743         intr_info = vmx->exit_intr_info;
5744
5745         if (is_machine_check(intr_info))
5746                 return handle_machine_check(vcpu);
5747
5748         if (is_nmi(intr_info))
5749                 return 1;  /* already handled by vmx_vcpu_run() */
5750
5751         if (is_invalid_opcode(intr_info)) {
5752                 if (is_guest_mode(vcpu)) {
5753                         kvm_queue_exception(vcpu, UD_VECTOR);
5754                         return 1;
5755                 }
5756                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5757                 if (er != EMULATE_DONE)
5758                         kvm_queue_exception(vcpu, UD_VECTOR);
5759                 return 1;
5760         }
5761
5762         error_code = 0;
5763         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5764                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5765
5766         /*
5767          * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5768          * MMIO, it is better to report an internal error.
5769          * See the comments in vmx_handle_exit.
5770          */
5771         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5772             !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5773                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5774                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5775                 vcpu->run->internal.ndata = 3;
5776                 vcpu->run->internal.data[0] = vect_info;
5777                 vcpu->run->internal.data[1] = intr_info;
5778                 vcpu->run->internal.data[2] = error_code;
5779                 return 0;
5780         }
5781
5782         if (is_page_fault(intr_info)) {
5783                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5784                 /* EPT won't cause page fault directly */
5785                 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
5786                 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0,
5787                                 true);
5788         }
5789
5790         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5791
5792         if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5793                 return handle_rmode_exception(vcpu, ex_no, error_code);
5794
5795         switch (ex_no) {
5796         case AC_VECTOR:
5797                 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5798                 return 1;
5799         case DB_VECTOR:
5800                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5801                 if (!(vcpu->guest_debug &
5802                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5803                         vcpu->arch.dr6 &= ~15;
5804                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5805                         if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5806                                 skip_emulated_instruction(vcpu);
5807
5808                         kvm_queue_exception(vcpu, DB_VECTOR);
5809                         return 1;
5810                 }
5811                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5812                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5813                 /* fall through */
5814         case BP_VECTOR:
5815                 /*
5816                  * Update instruction length as we may reinject #BP from
5817                  * user space while in guest debugging mode. Reading it for
5818                  * #DB as well causes no harm, it is not used in that case.
5819                  */
5820                 vmx->vcpu.arch.event_exit_inst_len =
5821                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5822                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5823                 rip = kvm_rip_read(vcpu);
5824                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5825                 kvm_run->debug.arch.exception = ex_no;
5826                 break;
5827         default:
5828                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5829                 kvm_run->ex.exception = ex_no;
5830                 kvm_run->ex.error_code = error_code;
5831                 break;
5832         }
5833         return 0;
5834 }
5835
5836 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5837 {
5838         ++vcpu->stat.irq_exits;
5839         return 1;
5840 }
5841
5842 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5843 {
5844         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5845         return 0;
5846 }
5847
5848 static int handle_io(struct kvm_vcpu *vcpu)
5849 {
5850         unsigned long exit_qualification;
5851         int size, in, string, ret;
5852         unsigned port;
5853
5854         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5855         string = (exit_qualification & 16) != 0;
5856         in = (exit_qualification & 8) != 0;
5857
5858         ++vcpu->stat.io_exits;
5859
5860         if (string || in)
5861                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5862
5863         port = exit_qualification >> 16;
5864         size = (exit_qualification & 7) + 1;
5865
5866         ret = kvm_skip_emulated_instruction(vcpu);
5867
5868         /*
5869          * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
5870          * KVM_EXIT_DEBUG here.
5871          */
5872         return kvm_fast_pio_out(vcpu, size, port) && ret;
5873 }
5874
5875 static void
5876 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5877 {
5878         /*
5879          * Patch in the VMCALL instruction:
5880          */
5881         hypercall[0] = 0x0f;
5882         hypercall[1] = 0x01;
5883         hypercall[2] = 0xc1;
5884 }
5885
5886 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5887 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5888 {
5889         if (is_guest_mode(vcpu)) {
5890                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5891                 unsigned long orig_val = val;
5892
5893                 /*
5894                  * We get here when L2 changed cr0 in a way that did not change
5895                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5896                  * but did change L0 shadowed bits. So we first calculate the
5897                  * effective cr0 value that L1 would like to write into the
5898                  * hardware. It consists of the L2-owned bits from the new
5899                  * value combined with the L1-owned bits from L1's guest_cr0.
5900                  */
5901                 val = (val & ~vmcs12->cr0_guest_host_mask) |
5902                         (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5903
5904                 if (!nested_guest_cr0_valid(vcpu, val))
5905                         return 1;
5906
5907                 if (kvm_set_cr0(vcpu, val))
5908                         return 1;
5909                 vmcs_writel(CR0_READ_SHADOW, orig_val);
5910                 return 0;
5911         } else {
5912                 if (to_vmx(vcpu)->nested.vmxon &&
5913                     !nested_host_cr0_valid(vcpu, val))
5914                         return 1;
5915
5916                 return kvm_set_cr0(vcpu, val);
5917         }
5918 }
5919
5920 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5921 {
5922         if (is_guest_mode(vcpu)) {
5923                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5924                 unsigned long orig_val = val;
5925
5926                 /* analogously to handle_set_cr0 */
5927                 val = (val & ~vmcs12->cr4_guest_host_mask) |
5928                         (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5929                 if (kvm_set_cr4(vcpu, val))
5930                         return 1;
5931                 vmcs_writel(CR4_READ_SHADOW, orig_val);
5932                 return 0;
5933         } else
5934                 return kvm_set_cr4(vcpu, val);
5935 }
5936
5937 static int handle_cr(struct kvm_vcpu *vcpu)
5938 {
5939         unsigned long exit_qualification, val;
5940         int cr;
5941         int reg;
5942         int err;
5943         int ret;
5944
5945         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5946         cr = exit_qualification & 15;
5947         reg = (exit_qualification >> 8) & 15;
5948         switch ((exit_qualification >> 4) & 3) {
5949         case 0: /* mov to cr */
5950                 val = kvm_register_readl(vcpu, reg);
5951                 trace_kvm_cr_write(cr, val);
5952                 switch (cr) {
5953                 case 0:
5954                         err = handle_set_cr0(vcpu, val);
5955                         return kvm_complete_insn_gp(vcpu, err);
5956                 case 3:
5957                         err = kvm_set_cr3(vcpu, val);
5958                         return kvm_complete_insn_gp(vcpu, err);
5959                 case 4:
5960                         err = handle_set_cr4(vcpu, val);
5961                         return kvm_complete_insn_gp(vcpu, err);
5962                 case 8: {
5963                                 u8 cr8_prev = kvm_get_cr8(vcpu);
5964                                 u8 cr8 = (u8)val;
5965                                 err = kvm_set_cr8(vcpu, cr8);
5966                                 ret = kvm_complete_insn_gp(vcpu, err);
5967                                 if (lapic_in_kernel(vcpu))
5968                                         return ret;
5969                                 if (cr8_prev <= cr8)
5970                                         return ret;
5971                                 /*
5972                                  * TODO: we might be squashing a
5973                                  * KVM_GUESTDBG_SINGLESTEP-triggered
5974                                  * KVM_EXIT_DEBUG here.
5975                                  */
5976                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5977                                 return 0;
5978                         }
5979                 }
5980                 break;
5981         case 2: /* clts */
5982                 WARN_ONCE(1, "Guest should always own CR0.TS");
5983                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5984                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5985                 return kvm_skip_emulated_instruction(vcpu);
5986         case 1: /*mov from cr*/
5987                 switch (cr) {
5988                 case 3:
5989                         val = kvm_read_cr3(vcpu);
5990                         kvm_register_write(vcpu, reg, val);
5991                         trace_kvm_cr_read(cr, val);
5992                         return kvm_skip_emulated_instruction(vcpu);
5993                 case 8:
5994                         val = kvm_get_cr8(vcpu);
5995                         kvm_register_write(vcpu, reg, val);
5996                         trace_kvm_cr_read(cr, val);
5997                         return kvm_skip_emulated_instruction(vcpu);
5998                 }
5999                 break;
6000         case 3: /* lmsw */
6001                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6002                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
6003                 kvm_lmsw(vcpu, val);
6004
6005                 return kvm_skip_emulated_instruction(vcpu);
6006         default:
6007                 break;
6008         }
6009         vcpu->run->exit_reason = 0;
6010         vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6011                (int)(exit_qualification >> 4) & 3, cr);
6012         return 0;
6013 }
6014
6015 static int handle_dr(struct kvm_vcpu *vcpu)
6016 {
6017         unsigned long exit_qualification;
6018         int dr, dr7, reg;
6019
6020         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6021         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6022
6023         /* First, if DR does not exist, trigger UD */
6024         if (!kvm_require_dr(vcpu, dr))
6025                 return 1;
6026
6027         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6028         if (!kvm_require_cpl(vcpu, 0))
6029                 return 1;
6030         dr7 = vmcs_readl(GUEST_DR7);
6031         if (dr7 & DR7_GD) {
6032                 /*
6033                  * As the vm-exit takes precedence over the debug trap, we
6034                  * need to emulate the latter, either for the host or the
6035                  * guest debugging itself.
6036                  */
6037                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6038                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6039                         vcpu->run->debug.arch.dr7 = dr7;
6040                         vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6041                         vcpu->run->debug.arch.exception = DB_VECTOR;
6042                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6043                         return 0;
6044                 } else {
6045                         vcpu->arch.dr6 &= ~15;
6046                         vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6047                         kvm_queue_exception(vcpu, DB_VECTOR);
6048                         return 1;
6049                 }
6050         }
6051
6052         if (vcpu->guest_debug == 0) {
6053                 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6054                                 CPU_BASED_MOV_DR_EXITING);
6055
6056                 /*
6057                  * No more DR vmexits; force a reload of the debug registers
6058                  * and reenter on this instruction.  The next vmexit will
6059                  * retrieve the full state of the debug registers.
6060                  */
6061                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
6062                 return 1;
6063         }
6064
6065         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
6066         if (exit_qualification & TYPE_MOV_FROM_DR) {
6067                 unsigned long val;
6068
6069                 if (kvm_get_dr(vcpu, dr, &val))
6070                         return 1;
6071                 kvm_register_write(vcpu, reg, val);
6072         } else
6073                 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
6074                         return 1;
6075
6076         return kvm_skip_emulated_instruction(vcpu);
6077 }
6078
6079 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
6080 {
6081         return vcpu->arch.dr6;
6082 }
6083
6084 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
6085 {
6086 }
6087
6088 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
6089 {
6090         get_debugreg(vcpu->arch.db[0], 0);
6091         get_debugreg(vcpu->arch.db[1], 1);
6092         get_debugreg(vcpu->arch.db[2], 2);
6093         get_debugreg(vcpu->arch.db[3], 3);
6094         get_debugreg(vcpu->arch.dr6, 6);
6095         vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6096
6097         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6098         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6099 }
6100
6101 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6102 {
6103         vmcs_writel(GUEST_DR7, val);
6104 }
6105
6106 static int handle_cpuid(struct kvm_vcpu *vcpu)
6107 {
6108         return kvm_emulate_cpuid(vcpu);
6109 }
6110
6111 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6112 {
6113         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6114         struct msr_data msr_info;
6115
6116         msr_info.index = ecx;
6117         msr_info.host_initiated = false;
6118         if (vmx_get_msr(vcpu, &msr_info)) {
6119                 trace_kvm_msr_read_ex(ecx);
6120                 kvm_inject_gp(vcpu, 0);
6121                 return 1;
6122         }
6123
6124         trace_kvm_msr_read(ecx, msr_info.data);
6125
6126         /* FIXME: handling of bits 32:63 of rax, rdx */
6127         vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6128         vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6129         return kvm_skip_emulated_instruction(vcpu);
6130 }
6131
6132 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6133 {
6134         struct msr_data msr;
6135         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6136         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6137                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6138
6139         msr.data = data;
6140         msr.index = ecx;
6141         msr.host_initiated = false;
6142         if (kvm_set_msr(vcpu, &msr) != 0) {
6143                 trace_kvm_msr_write_ex(ecx, data);
6144                 kvm_inject_gp(vcpu, 0);
6145                 return 1;
6146         }
6147
6148         trace_kvm_msr_write(ecx, data);
6149         return kvm_skip_emulated_instruction(vcpu);
6150 }
6151
6152 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6153 {
6154         kvm_apic_update_ppr(vcpu);
6155         return 1;
6156 }
6157
6158 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6159 {
6160         vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6161                         CPU_BASED_VIRTUAL_INTR_PENDING);
6162
6163         kvm_make_request(KVM_REQ_EVENT, vcpu);
6164
6165         ++vcpu->stat.irq_window_exits;
6166         return 1;
6167 }
6168
6169 static int handle_halt(struct kvm_vcpu *vcpu)
6170 {
6171         return kvm_emulate_halt(vcpu);
6172 }
6173
6174 static int handle_vmcall(struct kvm_vcpu *vcpu)
6175 {
6176         return kvm_emulate_hypercall(vcpu);
6177 }
6178
6179 static int handle_invd(struct kvm_vcpu *vcpu)
6180 {
6181         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6182 }
6183
6184 static int handle_invlpg(struct kvm_vcpu *vcpu)
6185 {
6186         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6187
6188         kvm_mmu_invlpg(vcpu, exit_qualification);
6189         return kvm_skip_emulated_instruction(vcpu);
6190 }
6191
6192 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6193 {
6194         int err;
6195
6196         err = kvm_rdpmc(vcpu);
6197         return kvm_complete_insn_gp(vcpu, err);
6198 }
6199
6200 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6201 {
6202         return kvm_emulate_wbinvd(vcpu);
6203 }
6204
6205 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6206 {
6207         u64 new_bv = kvm_read_edx_eax(vcpu);
6208         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6209
6210         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6211                 return kvm_skip_emulated_instruction(vcpu);
6212         return 1;
6213 }
6214
6215 static int handle_xsaves(struct kvm_vcpu *vcpu)
6216 {
6217         kvm_skip_emulated_instruction(vcpu);
6218         WARN(1, "this should never happen\n");
6219         return 1;
6220 }
6221
6222 static int handle_xrstors(struct kvm_vcpu *vcpu)
6223 {
6224         kvm_skip_emulated_instruction(vcpu);
6225         WARN(1, "this should never happen\n");
6226         return 1;
6227 }
6228
6229 static int handle_apic_access(struct kvm_vcpu *vcpu)
6230 {
6231         if (likely(fasteoi)) {
6232                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6233                 int access_type, offset;
6234
6235                 access_type = exit_qualification & APIC_ACCESS_TYPE;
6236                 offset = exit_qualification & APIC_ACCESS_OFFSET;
6237                 /*
6238                  * Sane guest uses MOV to write EOI, with written value
6239                  * not cared. So make a short-circuit here by avoiding
6240                  * heavy instruction emulation.
6241                  */
6242                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6243                     (offset == APIC_EOI)) {
6244                         kvm_lapic_set_eoi(vcpu);
6245                         return kvm_skip_emulated_instruction(vcpu);
6246                 }
6247         }
6248         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6249 }
6250
6251 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6252 {
6253         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6254         int vector = exit_qualification & 0xff;
6255
6256         /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6257         kvm_apic_set_eoi_accelerated(vcpu, vector);
6258         return 1;
6259 }
6260
6261 static int handle_apic_write(struct kvm_vcpu *vcpu)
6262 {
6263         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6264         u32 offset = exit_qualification & 0xfff;
6265
6266         /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6267         kvm_apic_write_nodecode(vcpu, offset);
6268         return 1;
6269 }
6270
6271 static int handle_task_switch(struct kvm_vcpu *vcpu)
6272 {
6273         struct vcpu_vmx *vmx = to_vmx(vcpu);
6274         unsigned long exit_qualification;
6275         bool has_error_code = false;
6276         u32 error_code = 0;
6277         u16 tss_selector;
6278         int reason, type, idt_v, idt_index;
6279
6280         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6281         idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6282         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6283
6284         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6285
6286         reason = (u32)exit_qualification >> 30;
6287         if (reason == TASK_SWITCH_GATE && idt_v) {
6288                 switch (type) {
6289                 case INTR_TYPE_NMI_INTR:
6290                         vcpu->arch.nmi_injected = false;
6291                         vmx_set_nmi_mask(vcpu, true);
6292                         break;
6293                 case INTR_TYPE_EXT_INTR:
6294                 case INTR_TYPE_SOFT_INTR:
6295                         kvm_clear_interrupt_queue(vcpu);
6296                         break;
6297                 case INTR_TYPE_HARD_EXCEPTION:
6298                         if (vmx->idt_vectoring_info &
6299                             VECTORING_INFO_DELIVER_CODE_MASK) {
6300                                 has_error_code = true;
6301                                 error_code =
6302                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
6303                         }
6304                         /* fall through */
6305                 case INTR_TYPE_SOFT_EXCEPTION:
6306                         kvm_clear_exception_queue(vcpu);
6307                         break;
6308                 default:
6309                         break;
6310                 }
6311         }
6312         tss_selector = exit_qualification;
6313
6314         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6315                        type != INTR_TYPE_EXT_INTR &&
6316                        type != INTR_TYPE_NMI_INTR))
6317                 skip_emulated_instruction(vcpu);
6318
6319         if (kvm_task_switch(vcpu, tss_selector,
6320                             type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6321                             has_error_code, error_code) == EMULATE_FAIL) {
6322                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6323                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6324                 vcpu->run->internal.ndata = 0;
6325                 return 0;
6326         }
6327
6328         /*
6329          * TODO: What about debug traps on tss switch?
6330          *       Are we supposed to inject them and update dr6?
6331          */
6332
6333         return 1;
6334 }
6335
6336 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6337 {
6338         unsigned long exit_qualification;
6339         gpa_t gpa;
6340         u32 error_code;
6341
6342         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6343
6344         /*
6345          * EPT violation happened while executing iret from NMI,
6346          * "blocked by NMI" bit has to be set before next VM entry.
6347          * There are errata that may cause this bit to not be set:
6348          * AAK134, BY25.
6349          */
6350         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6351                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6352                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6353
6354         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6355         trace_kvm_page_fault(gpa, exit_qualification);
6356
6357         /* Is it a read fault? */
6358         error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6359                      ? PFERR_USER_MASK : 0;
6360         /* Is it a write fault? */
6361         error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6362                       ? PFERR_WRITE_MASK : 0;
6363         /* Is it a fetch fault? */
6364         error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6365                       ? PFERR_FETCH_MASK : 0;
6366         /* ept page table entry is present? */
6367         error_code |= (exit_qualification &
6368                        (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6369                         EPT_VIOLATION_EXECUTABLE))
6370                       ? PFERR_PRESENT_MASK : 0;
6371
6372         vcpu->arch.gpa_available = true;
6373         vcpu->arch.exit_qualification = exit_qualification;
6374
6375         return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6376 }
6377
6378 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6379 {
6380         int ret;
6381         gpa_t gpa;
6382
6383         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6384         if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6385                 trace_kvm_fast_mmio(gpa);
6386                 return kvm_skip_emulated_instruction(vcpu);
6387         }
6388
6389         ret = handle_mmio_page_fault(vcpu, gpa, true);
6390         vcpu->arch.gpa_available = true;
6391         if (likely(ret == RET_MMIO_PF_EMULATE))
6392                 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6393                                               EMULATE_DONE;
6394
6395         if (unlikely(ret == RET_MMIO_PF_INVALID))
6396                 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6397
6398         if (unlikely(ret == RET_MMIO_PF_RETRY))
6399                 return 1;
6400
6401         /* It is the real ept misconfig */
6402         WARN_ON(1);
6403
6404         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6405         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6406
6407         return 0;
6408 }
6409
6410 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6411 {
6412         vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6413                         CPU_BASED_VIRTUAL_NMI_PENDING);
6414         ++vcpu->stat.nmi_window_exits;
6415         kvm_make_request(KVM_REQ_EVENT, vcpu);
6416
6417         return 1;
6418 }
6419
6420 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6421 {
6422         struct vcpu_vmx *vmx = to_vmx(vcpu);
6423         enum emulation_result err = EMULATE_DONE;
6424         int ret = 1;
6425         u32 cpu_exec_ctrl;
6426         bool intr_window_requested;
6427         unsigned count = 130;
6428
6429         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6430         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6431
6432         while (vmx->emulation_required && count-- != 0) {
6433                 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6434                         return handle_interrupt_window(&vmx->vcpu);
6435
6436                 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
6437                         return 1;
6438
6439                 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6440
6441                 if (err == EMULATE_USER_EXIT) {
6442                         ++vcpu->stat.mmio_exits;
6443                         ret = 0;
6444                         goto out;
6445                 }
6446
6447                 if (err != EMULATE_DONE) {
6448                         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6449                         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6450                         vcpu->run->internal.ndata = 0;
6451                         return 0;
6452                 }
6453
6454                 if (vcpu->arch.halt_request) {
6455                         vcpu->arch.halt_request = 0;
6456                         ret = kvm_vcpu_halt(vcpu);
6457                         goto out;
6458                 }
6459
6460                 if (signal_pending(current))
6461                         goto out;
6462                 if (need_resched())
6463                         schedule();
6464         }
6465
6466 out:
6467         return ret;
6468 }
6469
6470 static int __grow_ple_window(int val)
6471 {
6472         if (ple_window_grow < 1)
6473                 return ple_window;
6474
6475         val = min(val, ple_window_actual_max);
6476
6477         if (ple_window_grow < ple_window)
6478                 val *= ple_window_grow;
6479         else
6480                 val += ple_window_grow;
6481
6482         return val;
6483 }
6484
6485 static int __shrink_ple_window(int val, int modifier, int minimum)
6486 {
6487         if (modifier < 1)
6488                 return ple_window;
6489
6490         if (modifier < ple_window)
6491                 val /= modifier;
6492         else
6493                 val -= modifier;
6494
6495         return max(val, minimum);
6496 }
6497
6498 static void grow_ple_window(struct kvm_vcpu *vcpu)
6499 {
6500         struct vcpu_vmx *vmx = to_vmx(vcpu);
6501         int old = vmx->ple_window;
6502
6503         vmx->ple_window = __grow_ple_window(old);
6504
6505         if (vmx->ple_window != old)
6506                 vmx->ple_window_dirty = true;
6507
6508         trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6509 }
6510
6511 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6512 {
6513         struct vcpu_vmx *vmx = to_vmx(vcpu);
6514         int old = vmx->ple_window;
6515
6516         vmx->ple_window = __shrink_ple_window(old,
6517                                               ple_window_shrink, ple_window);
6518
6519         if (vmx->ple_window != old)
6520                 vmx->ple_window_dirty = true;
6521
6522         trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6523 }
6524
6525 /*
6526  * ple_window_actual_max is computed to be one grow_ple_window() below
6527  * ple_window_max. (See __grow_ple_window for the reason.)
6528  * This prevents overflows, because ple_window_max is int.
6529  * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6530  * this process.
6531  * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6532  */
6533 static void update_ple_window_actual_max(void)
6534 {
6535         ple_window_actual_max =
6536                         __shrink_ple_window(max(ple_window_max, ple_window),
6537                                             ple_window_grow, INT_MIN);
6538 }
6539
6540 /*
6541  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6542  */
6543 static void wakeup_handler(void)
6544 {
6545         struct kvm_vcpu *vcpu;
6546         int cpu = smp_processor_id();
6547
6548         spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6549         list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6550                         blocked_vcpu_list) {
6551                 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6552
6553                 if (pi_test_on(pi_desc) == 1)
6554                         kvm_vcpu_kick(vcpu);
6555         }
6556         spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6557 }
6558
6559 void vmx_enable_tdp(void)
6560 {
6561         kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6562                 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6563                 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6564                 0ull, VMX_EPT_EXECUTABLE_MASK,
6565                 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6566                 VMX_EPT_RWX_MASK);
6567
6568         ept_set_mmio_spte_mask();
6569         kvm_enable_tdp();
6570 }
6571
6572 static __init int hardware_setup(void)
6573 {
6574         int r = -ENOMEM, i, msr;
6575
6576         rdmsrl_safe(MSR_EFER, &host_efer);
6577
6578         for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6579                 kvm_define_shared_msr(i, vmx_msr_index[i]);
6580
6581         for (i = 0; i < VMX_BITMAP_NR; i++) {
6582                 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6583                 if (!vmx_bitmap[i])
6584                         goto out;
6585         }
6586
6587         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6588         memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6589         memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6590
6591         /*
6592          * Allow direct access to the PC debug port (it is often used for I/O
6593          * delays, but the vmexits simply slow things down).
6594          */
6595         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6596         clear_bit(0x80, vmx_io_bitmap_a);
6597
6598         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6599
6600         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6601         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6602
6603         if (setup_vmcs_config(&vmcs_config) < 0) {
6604                 r = -EIO;
6605                 goto out;
6606         }
6607
6608         if (boot_cpu_has(X86_FEATURE_NX))
6609                 kvm_enable_efer_bits(EFER_NX);
6610
6611         if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6612                 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6613                 enable_vpid = 0;
6614
6615         if (!cpu_has_vmx_shadow_vmcs())
6616                 enable_shadow_vmcs = 0;
6617         if (enable_shadow_vmcs)
6618                 init_vmcs_shadow_fields();
6619
6620         if (!cpu_has_vmx_ept() ||
6621             !cpu_has_vmx_ept_4levels()) {
6622                 enable_ept = 0;
6623                 enable_unrestricted_guest = 0;
6624                 enable_ept_ad_bits = 0;
6625         }
6626
6627         if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
6628                 enable_ept_ad_bits = 0;
6629
6630         if (!cpu_has_vmx_unrestricted_guest())
6631                 enable_unrestricted_guest = 0;
6632
6633         if (!cpu_has_vmx_flexpriority())
6634                 flexpriority_enabled = 0;
6635
6636         /*
6637          * set_apic_access_page_addr() is used to reload apic access
6638          * page upon invalidation.  No need to do anything if not
6639          * using the APIC_ACCESS_ADDR VMCS field.
6640          */
6641         if (!flexpriority_enabled)
6642                 kvm_x86_ops->set_apic_access_page_addr = NULL;
6643
6644         if (!cpu_has_vmx_tpr_shadow())
6645                 kvm_x86_ops->update_cr8_intercept = NULL;
6646
6647         if (enable_ept && !cpu_has_vmx_ept_2m_page())
6648                 kvm_disable_largepages();
6649
6650         if (!cpu_has_vmx_ple())
6651                 ple_gap = 0;
6652
6653         if (!cpu_has_vmx_apicv()) {
6654                 enable_apicv = 0;
6655                 kvm_x86_ops->sync_pir_to_irr = NULL;
6656         }
6657
6658         if (cpu_has_vmx_tsc_scaling()) {
6659                 kvm_has_tsc_control = true;
6660                 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6661                 kvm_tsc_scaling_ratio_frac_bits = 48;
6662         }
6663
6664         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6665         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6666         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6667         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6668         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6669         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6670
6671         memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6672                         vmx_msr_bitmap_legacy, PAGE_SIZE);
6673         memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6674                         vmx_msr_bitmap_longmode, PAGE_SIZE);
6675         memcpy(vmx_msr_bitmap_legacy_x2apic,
6676                         vmx_msr_bitmap_legacy, PAGE_SIZE);
6677         memcpy(vmx_msr_bitmap_longmode_x2apic,
6678                         vmx_msr_bitmap_longmode, PAGE_SIZE);
6679
6680         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6681
6682         for (msr = 0x800; msr <= 0x8ff; msr++) {
6683                 if (msr == 0x839 /* TMCCT */)
6684                         continue;
6685                 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6686         }
6687
6688         /*
6689          * TPR reads and writes can be virtualized even if virtual interrupt
6690          * delivery is not in use.
6691          */
6692         vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6693         vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6694
6695         /* EOI */
6696         vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6697         /* SELF-IPI */
6698         vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6699
6700         if (enable_ept)
6701                 vmx_enable_tdp();
6702         else
6703                 kvm_disable_tdp();
6704
6705         update_ple_window_actual_max();
6706
6707         /*
6708          * Only enable PML when hardware supports PML feature, and both EPT
6709          * and EPT A/D bit features are enabled -- PML depends on them to work.
6710          */
6711         if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6712                 enable_pml = 0;
6713
6714         if (!enable_pml) {
6715                 kvm_x86_ops->slot_enable_log_dirty = NULL;
6716                 kvm_x86_ops->slot_disable_log_dirty = NULL;
6717                 kvm_x86_ops->flush_log_dirty = NULL;
6718                 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6719         }
6720
6721         if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6722                 u64 vmx_msr;
6723
6724                 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6725                 cpu_preemption_timer_multi =
6726                          vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6727         } else {
6728                 kvm_x86_ops->set_hv_timer = NULL;
6729                 kvm_x86_ops->cancel_hv_timer = NULL;
6730         }
6731
6732         kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6733
6734         kvm_mce_cap_supported |= MCG_LMCE_P;
6735
6736         return alloc_kvm_area();
6737
6738 out:
6739         for (i = 0; i < VMX_BITMAP_NR; i++)
6740                 free_page((unsigned long)vmx_bitmap[i]);
6741
6742     return r;
6743 }
6744
6745 static __exit void hardware_unsetup(void)
6746 {
6747         int i;
6748
6749         for (i = 0; i < VMX_BITMAP_NR; i++)
6750                 free_page((unsigned long)vmx_bitmap[i]);
6751
6752         free_kvm_area();
6753 }
6754
6755 /*
6756  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6757  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6758  */
6759 static int handle_pause(struct kvm_vcpu *vcpu)
6760 {
6761         if (ple_gap)
6762                 grow_ple_window(vcpu);
6763
6764         kvm_vcpu_on_spin(vcpu);
6765         return kvm_skip_emulated_instruction(vcpu);
6766 }
6767
6768 static int handle_nop(struct kvm_vcpu *vcpu)
6769 {
6770         return kvm_skip_emulated_instruction(vcpu);
6771 }
6772
6773 static int handle_mwait(struct kvm_vcpu *vcpu)
6774 {
6775         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6776         return handle_nop(vcpu);
6777 }
6778
6779 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6780 {
6781         return 1;
6782 }
6783
6784 static int handle_monitor(struct kvm_vcpu *vcpu)
6785 {
6786         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6787         return handle_nop(vcpu);
6788 }
6789
6790 /*
6791  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6792  * We could reuse a single VMCS for all the L2 guests, but we also want the
6793  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6794  * allows keeping them loaded on the processor, and in the future will allow
6795  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6796  * every entry if they never change.
6797  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6798  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6799  *
6800  * The following functions allocate and free a vmcs02 in this pool.
6801  */
6802
6803 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6804 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6805 {
6806         struct vmcs02_list *item;
6807         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6808                 if (item->vmptr == vmx->nested.current_vmptr) {
6809                         list_move(&item->list, &vmx->nested.vmcs02_pool);
6810                         return &item->vmcs02;
6811                 }
6812
6813         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6814                 /* Recycle the least recently used VMCS. */
6815                 item = list_last_entry(&vmx->nested.vmcs02_pool,
6816                                        struct vmcs02_list, list);
6817                 item->vmptr = vmx->nested.current_vmptr;
6818                 list_move(&item->list, &vmx->nested.vmcs02_pool);
6819                 return &item->vmcs02;
6820         }
6821
6822         /* Create a new VMCS */
6823         item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6824         if (!item)
6825                 return NULL;
6826         item->vmcs02.vmcs = alloc_vmcs();
6827         item->vmcs02.shadow_vmcs = NULL;
6828         if (!item->vmcs02.vmcs) {
6829                 kfree(item);
6830                 return NULL;
6831         }
6832         loaded_vmcs_init(&item->vmcs02);
6833         item->vmptr = vmx->nested.current_vmptr;
6834         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6835         vmx->nested.vmcs02_num++;
6836         return &item->vmcs02;
6837 }
6838
6839 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6840 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6841 {
6842         struct vmcs02_list *item;
6843         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6844                 if (item->vmptr == vmptr) {
6845                         free_loaded_vmcs(&item->vmcs02);
6846                         list_del(&item->list);
6847                         kfree(item);
6848                         vmx->nested.vmcs02_num--;
6849                         return;
6850                 }
6851 }
6852
6853 /*
6854  * Free all VMCSs saved for this vcpu, except the one pointed by
6855  * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6856  * must be &vmx->vmcs01.
6857  */
6858 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6859 {
6860         struct vmcs02_list *item, *n;
6861
6862         WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6863         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6864                 /*
6865                  * Something will leak if the above WARN triggers.  Better than
6866                  * a use-after-free.
6867                  */
6868                 if (vmx->loaded_vmcs == &item->vmcs02)
6869                         continue;
6870
6871                 free_loaded_vmcs(&item->vmcs02);
6872                 list_del(&item->list);
6873                 kfree(item);
6874                 vmx->nested.vmcs02_num--;
6875         }
6876 }
6877
6878 /*
6879  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6880  * set the success or error code of an emulated VMX instruction, as specified
6881  * by Vol 2B, VMX Instruction Reference, "Conventions".
6882  */
6883 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6884 {
6885         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6886                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6887                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6888 }
6889
6890 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6891 {
6892         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6893                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6894                             X86_EFLAGS_SF | X86_EFLAGS_OF))
6895                         | X86_EFLAGS_CF);
6896 }
6897
6898 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6899                                         u32 vm_instruction_error)
6900 {
6901         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6902                 /*
6903                  * failValid writes the error number to the current VMCS, which
6904                  * can't be done there isn't a current VMCS.
6905                  */
6906                 nested_vmx_failInvalid(vcpu);
6907                 return;
6908         }
6909         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6910                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6911                             X86_EFLAGS_SF | X86_EFLAGS_OF))
6912                         | X86_EFLAGS_ZF);
6913         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6914         /*
6915          * We don't need to force a shadow sync because
6916          * VM_INSTRUCTION_ERROR is not shadowed
6917          */
6918 }
6919
6920 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6921 {
6922         /* TODO: not to reset guest simply here. */
6923         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6924         pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
6925 }
6926
6927 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6928 {
6929         struct vcpu_vmx *vmx =
6930                 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6931
6932         vmx->nested.preemption_timer_expired = true;
6933         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6934         kvm_vcpu_kick(&vmx->vcpu);
6935
6936         return HRTIMER_NORESTART;
6937 }
6938
6939 /*
6940  * Decode the memory-address operand of a vmx instruction, as recorded on an
6941  * exit caused by such an instruction (run by a guest hypervisor).
6942  * On success, returns 0. When the operand is invalid, returns 1 and throws
6943  * #UD or #GP.
6944  */
6945 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6946                                  unsigned long exit_qualification,
6947                                  u32 vmx_instruction_info, bool wr, gva_t *ret)
6948 {
6949         gva_t off;
6950         bool exn;
6951         struct kvm_segment s;
6952
6953         /*
6954          * According to Vol. 3B, "Information for VM Exits Due to Instruction
6955          * Execution", on an exit, vmx_instruction_info holds most of the
6956          * addressing components of the operand. Only the displacement part
6957          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6958          * For how an actual address is calculated from all these components,
6959          * refer to Vol. 1, "Operand Addressing".
6960          */
6961         int  scaling = vmx_instruction_info & 3;
6962         int  addr_size = (vmx_instruction_info >> 7) & 7;
6963         bool is_reg = vmx_instruction_info & (1u << 10);
6964         int  seg_reg = (vmx_instruction_info >> 15) & 7;
6965         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
6966         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6967         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
6968         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
6969
6970         if (is_reg) {
6971                 kvm_queue_exception(vcpu, UD_VECTOR);
6972                 return 1;
6973         }
6974
6975         /* Addr = segment_base + offset */
6976         /* offset = base + [index * scale] + displacement */
6977         off = exit_qualification; /* holds the displacement */
6978         if (base_is_valid)
6979                 off += kvm_register_read(vcpu, base_reg);
6980         if (index_is_valid)
6981                 off += kvm_register_read(vcpu, index_reg)<<scaling;
6982         vmx_get_segment(vcpu, &s, seg_reg);
6983         *ret = s.base + off;
6984
6985         if (addr_size == 1) /* 32 bit */
6986                 *ret &= 0xffffffff;
6987
6988         /* Checks for #GP/#SS exceptions. */
6989         exn = false;
6990         if (is_long_mode(vcpu)) {
6991                 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6992                  * non-canonical form. This is the only check on the memory
6993                  * destination for long mode!
6994                  */
6995                 exn = is_noncanonical_address(*ret);
6996         } else if (is_protmode(vcpu)) {
6997                 /* Protected mode: apply checks for segment validity in the
6998                  * following order:
6999                  * - segment type check (#GP(0) may be thrown)
7000                  * - usability check (#GP(0)/#SS(0))
7001                  * - limit check (#GP(0)/#SS(0))
7002                  */
7003                 if (wr)
7004                         /* #GP(0) if the destination operand is located in a
7005                          * read-only data segment or any code segment.
7006                          */
7007                         exn = ((s.type & 0xa) == 0 || (s.type & 8));
7008                 else
7009                         /* #GP(0) if the source operand is located in an
7010                          * execute-only code segment
7011                          */
7012                         exn = ((s.type & 0xa) == 8);
7013                 if (exn) {
7014                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7015                         return 1;
7016                 }
7017                 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
7018                  */
7019                 exn = (s.unusable != 0);
7020                 /* Protected mode: #GP(0)/#SS(0) if the memory
7021                  * operand is outside the segment limit.
7022                  */
7023                 exn = exn || (off + sizeof(u64) > s.limit);
7024         }
7025         if (exn) {
7026                 kvm_queue_exception_e(vcpu,
7027                                       seg_reg == VCPU_SREG_SS ?
7028                                                 SS_VECTOR : GP_VECTOR,
7029                                       0);
7030                 return 1;
7031         }
7032
7033         return 0;
7034 }
7035
7036 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
7037 {
7038         gva_t gva;
7039         struct x86_exception e;
7040
7041         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7042                         vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
7043                 return 1;
7044
7045         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, vmpointer,
7046                                 sizeof(*vmpointer), &e)) {
7047                 kvm_inject_page_fault(vcpu, &e);
7048                 return 1;
7049         }
7050
7051         return 0;
7052 }
7053
7054 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7055 {
7056         struct vcpu_vmx *vmx = to_vmx(vcpu);
7057         struct vmcs *shadow_vmcs;
7058
7059         if (cpu_has_vmx_msr_bitmap()) {
7060                 vmx->nested.msr_bitmap =
7061                                 (unsigned long *)__get_free_page(GFP_KERNEL);
7062                 if (!vmx->nested.msr_bitmap)
7063                         goto out_msr_bitmap;
7064         }
7065
7066         vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7067         if (!vmx->nested.cached_vmcs12)
7068                 goto out_cached_vmcs12;
7069
7070         if (enable_shadow_vmcs) {
7071                 shadow_vmcs = alloc_vmcs();
7072                 if (!shadow_vmcs)
7073                         goto out_shadow_vmcs;
7074                 /* mark vmcs as shadow */
7075                 shadow_vmcs->revision_id |= (1u << 31);
7076                 /* init shadow vmcs */
7077                 vmcs_clear(shadow_vmcs);
7078                 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7079         }
7080
7081         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7082         vmx->nested.vmcs02_num = 0;
7083
7084         hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7085                      HRTIMER_MODE_REL_PINNED);
7086         vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7087
7088         vmx->nested.vmxon = true;
7089         return 0;
7090
7091 out_shadow_vmcs:
7092         kfree(vmx->nested.cached_vmcs12);
7093
7094 out_cached_vmcs12:
7095         free_page((unsigned long)vmx->nested.msr_bitmap);
7096
7097 out_msr_bitmap:
7098         return -ENOMEM;
7099 }
7100
7101 /*
7102  * Emulate the VMXON instruction.
7103  * Currently, we just remember that VMX is active, and do not save or even
7104  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7105  * do not currently need to store anything in that guest-allocated memory
7106  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7107  * argument is different from the VMXON pointer (which the spec says they do).
7108  */
7109 static int handle_vmon(struct kvm_vcpu *vcpu)
7110 {
7111         int ret;
7112         gpa_t vmptr;
7113         struct page *page;
7114         struct vcpu_vmx *vmx = to_vmx(vcpu);
7115         const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7116                 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7117
7118         /*
7119          * The Intel VMX Instruction Reference lists a bunch of bits that are
7120          * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7121          * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7122          * Otherwise, we should fail with #UD.  But most faulting conditions
7123          * have already been checked by hardware, prior to the VM-exit for
7124          * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
7125          * that bit set to 1 in non-root mode.
7126          */
7127         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7128                 kvm_queue_exception(vcpu, UD_VECTOR);
7129                 return 1;
7130         }
7131
7132         if (vmx->nested.vmxon) {
7133                 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7134                 return kvm_skip_emulated_instruction(vcpu);
7135         }
7136
7137         if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7138                         != VMXON_NEEDED_FEATURES) {
7139                 kvm_inject_gp(vcpu, 0);
7140                 return 1;
7141         }
7142
7143         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7144                 return 1;
7145
7146         /*
7147          * SDM 3: 24.11.5
7148          * The first 4 bytes of VMXON region contain the supported
7149          * VMCS revision identifier
7150          *
7151          * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7152          * which replaces physical address width with 32
7153          */
7154         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7155                 nested_vmx_failInvalid(vcpu);
7156                 return kvm_skip_emulated_instruction(vcpu);
7157         }
7158
7159         page = nested_get_page(vcpu, vmptr);
7160         if (page == NULL) {
7161                 nested_vmx_failInvalid(vcpu);
7162                 return kvm_skip_emulated_instruction(vcpu);
7163         }
7164         if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7165                 kunmap(page);
7166                 nested_release_page_clean(page);
7167                 nested_vmx_failInvalid(vcpu);
7168                 return kvm_skip_emulated_instruction(vcpu);
7169         }
7170         kunmap(page);
7171         nested_release_page_clean(page);
7172
7173         vmx->nested.vmxon_ptr = vmptr;
7174         ret = enter_vmx_operation(vcpu);
7175         if (ret)
7176                 return ret;
7177
7178         nested_vmx_succeed(vcpu);
7179         return kvm_skip_emulated_instruction(vcpu);
7180 }
7181
7182 /*
7183  * Intel's VMX Instruction Reference specifies a common set of prerequisites
7184  * for running VMX instructions (except VMXON, whose prerequisites are
7185  * slightly different). It also specifies what exception to inject otherwise.
7186  * Note that many of these exceptions have priority over VM exits, so they
7187  * don't have to be checked again here.
7188  */
7189 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7190 {
7191         if (!to_vmx(vcpu)->nested.vmxon) {
7192                 kvm_queue_exception(vcpu, UD_VECTOR);
7193                 return 0;
7194         }
7195         return 1;
7196 }
7197
7198 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
7199 {
7200         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
7201         vmcs_write64(VMCS_LINK_POINTER, -1ull);
7202 }
7203
7204 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7205 {
7206         if (vmx->nested.current_vmptr == -1ull)
7207                 return;
7208
7209         if (enable_shadow_vmcs) {
7210                 /* copy to memory all shadowed fields in case
7211                    they were modified */
7212                 copy_shadow_to_vmcs12(vmx);
7213                 vmx->nested.sync_shadow_vmcs = false;
7214                 vmx_disable_shadow_vmcs(vmx);
7215         }
7216         vmx->nested.posted_intr_nv = -1;
7217
7218         /* Flush VMCS12 to guest memory */
7219         kvm_vcpu_write_guest_page(&vmx->vcpu,
7220                                   vmx->nested.current_vmptr >> PAGE_SHIFT,
7221                                   vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
7222
7223         vmx->nested.current_vmptr = -1ull;
7224 }
7225
7226 /*
7227  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7228  * just stops using VMX.
7229  */
7230 static void free_nested(struct vcpu_vmx *vmx)
7231 {
7232         if (!vmx->nested.vmxon)
7233                 return;
7234
7235         vmx->nested.vmxon = false;
7236         free_vpid(vmx->nested.vpid02);
7237         vmx->nested.posted_intr_nv = -1;
7238         vmx->nested.current_vmptr = -1ull;
7239         if (vmx->nested.msr_bitmap) {
7240                 free_page((unsigned long)vmx->nested.msr_bitmap);
7241                 vmx->nested.msr_bitmap = NULL;
7242         }
7243         if (enable_shadow_vmcs) {
7244                 vmx_disable_shadow_vmcs(vmx);
7245                 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7246                 free_vmcs(vmx->vmcs01.shadow_vmcs);
7247                 vmx->vmcs01.shadow_vmcs = NULL;
7248         }
7249         kfree(vmx->nested.cached_vmcs12);
7250         /* Unpin physical memory we referred to in current vmcs02 */
7251         if (vmx->nested.apic_access_page) {
7252                 nested_release_page(vmx->nested.apic_access_page);
7253                 vmx->nested.apic_access_page = NULL;
7254         }
7255         if (vmx->nested.virtual_apic_page) {
7256                 nested_release_page(vmx->nested.virtual_apic_page);
7257                 vmx->nested.virtual_apic_page = NULL;
7258         }
7259         if (vmx->nested.pi_desc_page) {
7260                 kunmap(vmx->nested.pi_desc_page);
7261                 nested_release_page(vmx->nested.pi_desc_page);
7262                 vmx->nested.pi_desc_page = NULL;
7263                 vmx->nested.pi_desc = NULL;
7264         }
7265
7266         nested_free_all_saved_vmcss(vmx);
7267 }
7268
7269 /* Emulate the VMXOFF instruction */
7270 static int handle_vmoff(struct kvm_vcpu *vcpu)
7271 {
7272         if (!nested_vmx_check_permission(vcpu))
7273                 return 1;
7274         free_nested(to_vmx(vcpu));
7275         nested_vmx_succeed(vcpu);
7276         return kvm_skip_emulated_instruction(vcpu);
7277 }
7278
7279 /* Emulate the VMCLEAR instruction */
7280 static int handle_vmclear(struct kvm_vcpu *vcpu)
7281 {
7282         struct vcpu_vmx *vmx = to_vmx(vcpu);
7283         u32 zero = 0;
7284         gpa_t vmptr;
7285
7286         if (!nested_vmx_check_permission(vcpu))
7287                 return 1;
7288
7289         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7290                 return 1;
7291
7292         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7293                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
7294                 return kvm_skip_emulated_instruction(vcpu);
7295         }
7296
7297         if (vmptr == vmx->nested.vmxon_ptr) {
7298                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
7299                 return kvm_skip_emulated_instruction(vcpu);
7300         }
7301
7302         if (vmptr == vmx->nested.current_vmptr)
7303                 nested_release_vmcs12(vmx);
7304
7305         kvm_vcpu_write_guest(vcpu,
7306                         vmptr + offsetof(struct vmcs12, launch_state),
7307                         &zero, sizeof(zero));
7308
7309         nested_free_vmcs02(vmx, vmptr);
7310
7311         nested_vmx_succeed(vcpu);
7312         return kvm_skip_emulated_instruction(vcpu);
7313 }
7314
7315 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7316
7317 /* Emulate the VMLAUNCH instruction */
7318 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7319 {
7320         return nested_vmx_run(vcpu, true);
7321 }
7322
7323 /* Emulate the VMRESUME instruction */
7324 static int handle_vmresume(struct kvm_vcpu *vcpu)
7325 {
7326
7327         return nested_vmx_run(vcpu, false);
7328 }
7329
7330 /*
7331  * Read a vmcs12 field. Since these can have varying lengths and we return
7332  * one type, we chose the biggest type (u64) and zero-extend the return value
7333  * to that size. Note that the caller, handle_vmread, might need to use only
7334  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7335  * 64-bit fields are to be returned).
7336  */
7337 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7338                                   unsigned long field, u64 *ret)
7339 {
7340         short offset = vmcs_field_to_offset(field);
7341         char *p;
7342
7343         if (offset < 0)
7344                 return offset;
7345
7346         p = ((char *)(get_vmcs12(vcpu))) + offset;
7347
7348         switch (vmcs_field_type(field)) {
7349         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7350                 *ret = *((natural_width *)p);
7351                 return 0;
7352         case VMCS_FIELD_TYPE_U16:
7353                 *ret = *((u16 *)p);
7354                 return 0;
7355         case VMCS_FIELD_TYPE_U32:
7356                 *ret = *((u32 *)p);
7357                 return 0;
7358         case VMCS_FIELD_TYPE_U64:
7359                 *ret = *((u64 *)p);
7360                 return 0;
7361         default:
7362                 WARN_ON(1);
7363                 return -ENOENT;
7364         }
7365 }
7366
7367
7368 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7369                                    unsigned long field, u64 field_value){
7370         short offset = vmcs_field_to_offset(field);
7371         char *p = ((char *) get_vmcs12(vcpu)) + offset;
7372         if (offset < 0)
7373                 return offset;
7374
7375         switch (vmcs_field_type(field)) {
7376         case VMCS_FIELD_TYPE_U16:
7377                 *(u16 *)p = field_value;
7378                 return 0;
7379         case VMCS_FIELD_TYPE_U32:
7380                 *(u32 *)p = field_value;
7381                 return 0;
7382         case VMCS_FIELD_TYPE_U64:
7383                 *(u64 *)p = field_value;
7384                 return 0;
7385         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7386                 *(natural_width *)p = field_value;
7387                 return 0;
7388         default:
7389                 WARN_ON(1);
7390                 return -ENOENT;
7391         }
7392
7393 }
7394
7395 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7396 {
7397         int i;
7398         unsigned long field;
7399         u64 field_value;
7400         struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7401         const unsigned long *fields = shadow_read_write_fields;
7402         const int num_fields = max_shadow_read_write_fields;
7403
7404         preempt_disable();
7405
7406         vmcs_load(shadow_vmcs);
7407
7408         for (i = 0; i < num_fields; i++) {
7409                 field = fields[i];
7410                 switch (vmcs_field_type(field)) {
7411                 case VMCS_FIELD_TYPE_U16:
7412                         field_value = vmcs_read16(field);
7413                         break;
7414                 case VMCS_FIELD_TYPE_U32:
7415                         field_value = vmcs_read32(field);
7416                         break;
7417                 case VMCS_FIELD_TYPE_U64:
7418                         field_value = vmcs_read64(field);
7419                         break;
7420                 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7421                         field_value = vmcs_readl(field);
7422                         break;
7423                 default:
7424                         WARN_ON(1);
7425                         continue;
7426                 }
7427                 vmcs12_write_any(&vmx->vcpu, field, field_value);
7428         }
7429
7430         vmcs_clear(shadow_vmcs);
7431         vmcs_load(vmx->loaded_vmcs->vmcs);
7432
7433         preempt_enable();
7434 }
7435
7436 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7437 {
7438         const unsigned long *fields[] = {
7439                 shadow_read_write_fields,
7440                 shadow_read_only_fields
7441         };
7442         const int max_fields[] = {
7443                 max_shadow_read_write_fields,
7444                 max_shadow_read_only_fields
7445         };
7446         int i, q;
7447         unsigned long field;
7448         u64 field_value = 0;
7449         struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7450
7451         vmcs_load(shadow_vmcs);
7452
7453         for (q = 0; q < ARRAY_SIZE(fields); q++) {
7454                 for (i = 0; i < max_fields[q]; i++) {
7455                         field = fields[q][i];
7456                         vmcs12_read_any(&vmx->vcpu, field, &field_value);
7457
7458                         switch (vmcs_field_type(field)) {
7459                         case VMCS_FIELD_TYPE_U16:
7460                                 vmcs_write16(field, (u16)field_value);
7461                                 break;
7462                         case VMCS_FIELD_TYPE_U32:
7463                                 vmcs_write32(field, (u32)field_value);
7464                                 break;
7465                         case VMCS_FIELD_TYPE_U64:
7466                                 vmcs_write64(field, (u64)field_value);
7467                                 break;
7468                         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7469                                 vmcs_writel(field, (long)field_value);
7470                                 break;
7471                         default:
7472                                 WARN_ON(1);
7473                                 break;
7474                         }
7475                 }
7476         }
7477
7478         vmcs_clear(shadow_vmcs);
7479         vmcs_load(vmx->loaded_vmcs->vmcs);
7480 }
7481
7482 /*
7483  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7484  * used before) all generate the same failure when it is missing.
7485  */
7486 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7487 {
7488         struct vcpu_vmx *vmx = to_vmx(vcpu);
7489         if (vmx->nested.current_vmptr == -1ull) {
7490                 nested_vmx_failInvalid(vcpu);
7491                 return 0;
7492         }
7493         return 1;
7494 }
7495
7496 static int handle_vmread(struct kvm_vcpu *vcpu)
7497 {
7498         unsigned long field;
7499         u64 field_value;
7500         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7501         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7502         gva_t gva = 0;
7503
7504         if (!nested_vmx_check_permission(vcpu))
7505                 return 1;
7506
7507         if (!nested_vmx_check_vmcs12(vcpu))
7508                 return kvm_skip_emulated_instruction(vcpu);
7509
7510         /* Decode instruction info and find the field to read */
7511         field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7512         /* Read the field, zero-extended to a u64 field_value */
7513         if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7514                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7515                 return kvm_skip_emulated_instruction(vcpu);
7516         }
7517         /*
7518          * Now copy part of this value to register or memory, as requested.
7519          * Note that the number of bits actually copied is 32 or 64 depending
7520          * on the guest's mode (32 or 64 bit), not on the given field's length.
7521          */
7522         if (vmx_instruction_info & (1u << 10)) {
7523                 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7524                         field_value);
7525         } else {
7526                 if (get_vmx_mem_address(vcpu, exit_qualification,
7527                                 vmx_instruction_info, true, &gva))
7528                         return 1;
7529                 /* _system ok, as hardware has verified cpl=0 */
7530                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7531                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7532         }
7533
7534         nested_vmx_succeed(vcpu);
7535         return kvm_skip_emulated_instruction(vcpu);
7536 }
7537
7538
7539 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7540 {
7541         unsigned long field;
7542         gva_t gva;
7543         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7544         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7545         /* The value to write might be 32 or 64 bits, depending on L1's long
7546          * mode, and eventually we need to write that into a field of several
7547          * possible lengths. The code below first zero-extends the value to 64
7548          * bit (field_value), and then copies only the appropriate number of
7549          * bits into the vmcs12 field.
7550          */
7551         u64 field_value = 0;
7552         struct x86_exception e;
7553
7554         if (!nested_vmx_check_permission(vcpu))
7555                 return 1;
7556
7557         if (!nested_vmx_check_vmcs12(vcpu))
7558                 return kvm_skip_emulated_instruction(vcpu);
7559
7560         if (vmx_instruction_info & (1u << 10))
7561                 field_value = kvm_register_readl(vcpu,
7562                         (((vmx_instruction_info) >> 3) & 0xf));
7563         else {
7564                 if (get_vmx_mem_address(vcpu, exit_qualification,
7565                                 vmx_instruction_info, false, &gva))
7566                         return 1;
7567                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7568                            &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7569                         kvm_inject_page_fault(vcpu, &e);
7570                         return 1;
7571                 }
7572         }
7573
7574
7575         field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7576         if (vmcs_field_readonly(field)) {
7577                 nested_vmx_failValid(vcpu,
7578                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7579                 return kvm_skip_emulated_instruction(vcpu);
7580         }
7581
7582         if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7583                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7584                 return kvm_skip_emulated_instruction(vcpu);
7585         }
7586
7587         nested_vmx_succeed(vcpu);
7588         return kvm_skip_emulated_instruction(vcpu);
7589 }
7590
7591 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7592 {
7593         vmx->nested.current_vmptr = vmptr;
7594         if (enable_shadow_vmcs) {
7595                 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7596                               SECONDARY_EXEC_SHADOW_VMCS);
7597                 vmcs_write64(VMCS_LINK_POINTER,
7598                              __pa(vmx->vmcs01.shadow_vmcs));
7599                 vmx->nested.sync_shadow_vmcs = true;
7600         }
7601 }
7602
7603 /* Emulate the VMPTRLD instruction */
7604 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7605 {
7606         struct vcpu_vmx *vmx = to_vmx(vcpu);
7607         gpa_t vmptr;
7608
7609         if (!nested_vmx_check_permission(vcpu))
7610                 return 1;
7611
7612         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7613                 return 1;
7614
7615         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7616                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
7617                 return kvm_skip_emulated_instruction(vcpu);
7618         }
7619
7620         if (vmptr == vmx->nested.vmxon_ptr) {
7621                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
7622                 return kvm_skip_emulated_instruction(vcpu);
7623         }
7624
7625         if (vmx->nested.current_vmptr != vmptr) {
7626                 struct vmcs12 *new_vmcs12;
7627                 struct page *page;
7628                 page = nested_get_page(vcpu, vmptr);
7629                 if (page == NULL) {
7630                         nested_vmx_failInvalid(vcpu);
7631                         return kvm_skip_emulated_instruction(vcpu);
7632                 }
7633                 new_vmcs12 = kmap(page);
7634                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7635                         kunmap(page);
7636                         nested_release_page_clean(page);
7637                         nested_vmx_failValid(vcpu,
7638                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7639                         return kvm_skip_emulated_instruction(vcpu);
7640                 }
7641
7642                 nested_release_vmcs12(vmx);
7643                 /*
7644                  * Load VMCS12 from guest memory since it is not already
7645                  * cached.
7646                  */
7647                 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
7648                 kunmap(page);
7649                 nested_release_page_clean(page);
7650
7651                 set_current_vmptr(vmx, vmptr);
7652         }
7653
7654         nested_vmx_succeed(vcpu);
7655         return kvm_skip_emulated_instruction(vcpu);
7656 }
7657
7658 /* Emulate the VMPTRST instruction */
7659 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7660 {
7661         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7662         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7663         gva_t vmcs_gva;
7664         struct x86_exception e;
7665
7666         if (!nested_vmx_check_permission(vcpu))
7667                 return 1;
7668
7669         if (get_vmx_mem_address(vcpu, exit_qualification,
7670                         vmx_instruction_info, true, &vmcs_gva))
7671                 return 1;
7672         /* ok to use *_system, as hardware has verified cpl=0 */
7673         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7674                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
7675                                  sizeof(u64), &e)) {
7676                 kvm_inject_page_fault(vcpu, &e);
7677                 return 1;
7678         }
7679         nested_vmx_succeed(vcpu);
7680         return kvm_skip_emulated_instruction(vcpu);
7681 }
7682
7683 /* Emulate the INVEPT instruction */
7684 static int handle_invept(struct kvm_vcpu *vcpu)
7685 {
7686         struct vcpu_vmx *vmx = to_vmx(vcpu);
7687         u32 vmx_instruction_info, types;
7688         unsigned long type;
7689         gva_t gva;
7690         struct x86_exception e;
7691         struct {
7692                 u64 eptp, gpa;
7693         } operand;
7694
7695         if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7696               SECONDARY_EXEC_ENABLE_EPT) ||
7697             !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7698                 kvm_queue_exception(vcpu, UD_VECTOR);
7699                 return 1;
7700         }
7701
7702         if (!nested_vmx_check_permission(vcpu))
7703                 return 1;
7704
7705         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7706         type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7707
7708         types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7709
7710         if (type >= 32 || !(types & (1 << type))) {
7711                 nested_vmx_failValid(vcpu,
7712                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7713                 return kvm_skip_emulated_instruction(vcpu);
7714         }
7715
7716         /* According to the Intel VMX instruction reference, the memory
7717          * operand is read even if it isn't needed (e.g., for type==global)
7718          */
7719         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7720                         vmx_instruction_info, false, &gva))
7721                 return 1;
7722         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7723                                 sizeof(operand), &e)) {
7724                 kvm_inject_page_fault(vcpu, &e);
7725                 return 1;
7726         }
7727
7728         switch (type) {
7729         case VMX_EPT_EXTENT_GLOBAL:
7730         /*
7731          * TODO: track mappings and invalidate
7732          * single context requests appropriately
7733          */
7734         case VMX_EPT_EXTENT_CONTEXT:
7735                 kvm_mmu_sync_roots(vcpu);
7736                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7737                 nested_vmx_succeed(vcpu);
7738                 break;
7739         default:
7740                 BUG_ON(1);
7741                 break;
7742         }
7743
7744         return kvm_skip_emulated_instruction(vcpu);
7745 }
7746
7747 static int handle_invvpid(struct kvm_vcpu *vcpu)
7748 {
7749         struct vcpu_vmx *vmx = to_vmx(vcpu);
7750         u32 vmx_instruction_info;
7751         unsigned long type, types;
7752         gva_t gva;
7753         struct x86_exception e;
7754         struct {
7755                 u64 vpid;
7756                 u64 gla;
7757         } operand;
7758
7759         if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7760               SECONDARY_EXEC_ENABLE_VPID) ||
7761                         !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7762                 kvm_queue_exception(vcpu, UD_VECTOR);
7763                 return 1;
7764         }
7765
7766         if (!nested_vmx_check_permission(vcpu))
7767                 return 1;
7768
7769         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7770         type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7771
7772         types = (vmx->nested.nested_vmx_vpid_caps &
7773                         VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7774
7775         if (type >= 32 || !(types & (1 << type))) {
7776                 nested_vmx_failValid(vcpu,
7777                         VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7778                 return kvm_skip_emulated_instruction(vcpu);
7779         }
7780
7781         /* according to the intel vmx instruction reference, the memory
7782          * operand is read even if it isn't needed (e.g., for type==global)
7783          */
7784         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7785                         vmx_instruction_info, false, &gva))
7786                 return 1;
7787         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7788                                 sizeof(operand), &e)) {
7789                 kvm_inject_page_fault(vcpu, &e);
7790                 return 1;
7791         }
7792         if (operand.vpid >> 16) {
7793                 nested_vmx_failValid(vcpu,
7794                         VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7795                 return kvm_skip_emulated_instruction(vcpu);
7796         }
7797
7798         switch (type) {
7799         case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7800                 if (is_noncanonical_address(operand.gla)) {
7801                         nested_vmx_failValid(vcpu,
7802                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7803                         return kvm_skip_emulated_instruction(vcpu);
7804                 }
7805                 /* fall through */
7806         case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7807         case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7808                 if (!operand.vpid) {
7809                         nested_vmx_failValid(vcpu,
7810                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7811                         return kvm_skip_emulated_instruction(vcpu);
7812                 }
7813                 break;
7814         case VMX_VPID_EXTENT_ALL_CONTEXT:
7815                 break;
7816         default:
7817                 WARN_ON_ONCE(1);
7818                 return kvm_skip_emulated_instruction(vcpu);
7819         }
7820
7821         __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
7822         nested_vmx_succeed(vcpu);
7823
7824         return kvm_skip_emulated_instruction(vcpu);
7825 }
7826
7827 static int handle_pml_full(struct kvm_vcpu *vcpu)
7828 {
7829         unsigned long exit_qualification;
7830
7831         trace_kvm_pml_full(vcpu->vcpu_id);
7832
7833         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7834
7835         /*
7836          * PML buffer FULL happened while executing iret from NMI,
7837          * "blocked by NMI" bit has to be set before next VM entry.
7838          */
7839         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7840                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7841                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7842                                 GUEST_INTR_STATE_NMI);
7843
7844         /*
7845          * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7846          * here.., and there's no userspace involvement needed for PML.
7847          */
7848         return 1;
7849 }
7850
7851 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7852 {
7853         kvm_lapic_expired_hv_timer(vcpu);
7854         return 1;
7855 }
7856
7857 /*
7858  * The exit handlers return 1 if the exit was handled fully and guest execution
7859  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
7860  * to be done to userspace and return 0.
7861  */
7862 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7863         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
7864         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
7865         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
7866         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
7867         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
7868         [EXIT_REASON_CR_ACCESS]               = handle_cr,
7869         [EXIT_REASON_DR_ACCESS]               = handle_dr,
7870         [EXIT_REASON_CPUID]                   = handle_cpuid,
7871         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
7872         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
7873         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
7874         [EXIT_REASON_HLT]                     = handle_halt,
7875         [EXIT_REASON_INVD]                    = handle_invd,
7876         [EXIT_REASON_INVLPG]                  = handle_invlpg,
7877         [EXIT_REASON_RDPMC]                   = handle_rdpmc,
7878         [EXIT_REASON_VMCALL]                  = handle_vmcall,
7879         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
7880         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
7881         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
7882         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
7883         [EXIT_REASON_VMREAD]                  = handle_vmread,
7884         [EXIT_REASON_VMRESUME]                = handle_vmresume,
7885         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
7886         [EXIT_REASON_VMOFF]                   = handle_vmoff,
7887         [EXIT_REASON_VMON]                    = handle_vmon,
7888         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
7889         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
7890         [EXIT_REASON_APIC_WRITE]              = handle_apic_write,
7891         [EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
7892         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
7893         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
7894         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
7895         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
7896         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
7897         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
7898         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
7899         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_mwait,
7900         [EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
7901         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
7902         [EXIT_REASON_INVEPT]                  = handle_invept,
7903         [EXIT_REASON_INVVPID]                 = handle_invvpid,
7904         [EXIT_REASON_XSAVES]                  = handle_xsaves,
7905         [EXIT_REASON_XRSTORS]                 = handle_xrstors,
7906         [EXIT_REASON_PML_FULL]                = handle_pml_full,
7907         [EXIT_REASON_PREEMPTION_TIMER]        = handle_preemption_timer,
7908 };
7909
7910 static const int kvm_vmx_max_exit_handlers =
7911         ARRAY_SIZE(kvm_vmx_exit_handlers);
7912
7913 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7914                                        struct vmcs12 *vmcs12)
7915 {
7916         unsigned long exit_qualification;
7917         gpa_t bitmap, last_bitmap;
7918         unsigned int port;
7919         int size;
7920         u8 b;
7921
7922         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7923                 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7924
7925         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7926
7927         port = exit_qualification >> 16;
7928         size = (exit_qualification & 7) + 1;
7929
7930         last_bitmap = (gpa_t)-1;
7931         b = -1;
7932
7933         while (size > 0) {
7934                 if (port < 0x8000)
7935                         bitmap = vmcs12->io_bitmap_a;
7936                 else if (port < 0x10000)
7937                         bitmap = vmcs12->io_bitmap_b;
7938                 else
7939                         return true;
7940                 bitmap += (port & 0x7fff) / 8;
7941
7942                 if (last_bitmap != bitmap)
7943                         if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7944                                 return true;
7945                 if (b & (1 << (port & 7)))
7946                         return true;
7947
7948                 port++;
7949                 size--;
7950                 last_bitmap = bitmap;
7951         }
7952
7953         return false;
7954 }
7955
7956 /*
7957  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7958  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7959  * disinterest in the current event (read or write a specific MSR) by using an
7960  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7961  */
7962 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7963         struct vmcs12 *vmcs12, u32 exit_reason)
7964 {
7965         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7966         gpa_t bitmap;
7967
7968         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7969                 return true;
7970
7971         /*
7972          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7973          * for the four combinations of read/write and low/high MSR numbers.
7974          * First we need to figure out which of the four to use:
7975          */
7976         bitmap = vmcs12->msr_bitmap;
7977         if (exit_reason == EXIT_REASON_MSR_WRITE)
7978                 bitmap += 2048;
7979         if (msr_index >= 0xc0000000) {
7980                 msr_index -= 0xc0000000;
7981                 bitmap += 1024;
7982         }
7983
7984         /* Then read the msr_index'th bit from this bitmap: */
7985         if (msr_index < 1024*8) {
7986                 unsigned char b;
7987                 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7988                         return true;
7989                 return 1 & (b >> (msr_index & 7));
7990         } else
7991                 return true; /* let L1 handle the wrong parameter */
7992 }
7993
7994 /*
7995  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7996  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7997  * intercept (via guest_host_mask etc.) the current event.
7998  */
7999 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
8000         struct vmcs12 *vmcs12)
8001 {
8002         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8003         int cr = exit_qualification & 15;
8004         int reg;
8005         unsigned long val;
8006
8007         switch ((exit_qualification >> 4) & 3) {
8008         case 0: /* mov to cr */
8009                 reg = (exit_qualification >> 8) & 15;
8010                 val = kvm_register_readl(vcpu, reg);
8011                 switch (cr) {
8012                 case 0:
8013                         if (vmcs12->cr0_guest_host_mask &
8014                             (val ^ vmcs12->cr0_read_shadow))
8015                                 return true;
8016                         break;
8017                 case 3:
8018                         if ((vmcs12->cr3_target_count >= 1 &&
8019                                         vmcs12->cr3_target_value0 == val) ||
8020                                 (vmcs12->cr3_target_count >= 2 &&
8021                                         vmcs12->cr3_target_value1 == val) ||
8022                                 (vmcs12->cr3_target_count >= 3 &&
8023                                         vmcs12->cr3_target_value2 == val) ||
8024                                 (vmcs12->cr3_target_count >= 4 &&
8025                                         vmcs12->cr3_target_value3 == val))
8026                                 return false;
8027                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8028                                 return true;
8029                         break;
8030                 case 4:
8031                         if (vmcs12->cr4_guest_host_mask &
8032                             (vmcs12->cr4_read_shadow ^ val))
8033                                 return true;
8034                         break;
8035                 case 8:
8036                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8037                                 return true;
8038                         break;
8039                 }
8040                 break;
8041         case 2: /* clts */
8042                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8043                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
8044                         return true;
8045                 break;
8046         case 1: /* mov from cr */
8047                 switch (cr) {
8048                 case 3:
8049                         if (vmcs12->cpu_based_vm_exec_control &
8050                             CPU_BASED_CR3_STORE_EXITING)
8051                                 return true;
8052                         break;
8053                 case 8:
8054                         if (vmcs12->cpu_based_vm_exec_control &
8055                             CPU_BASED_CR8_STORE_EXITING)
8056                                 return true;
8057                         break;
8058                 }
8059                 break;
8060         case 3: /* lmsw */
8061                 /*
8062                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8063                  * cr0. Other attempted changes are ignored, with no exit.
8064                  */
8065                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
8066                 if (vmcs12->cr0_guest_host_mask & 0xe &
8067                     (val ^ vmcs12->cr0_read_shadow))
8068                         return true;
8069                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8070                     !(vmcs12->cr0_read_shadow & 0x1) &&
8071                     (val & 0x1))
8072                         return true;
8073                 break;
8074         }
8075         return false;
8076 }
8077
8078 /*
8079  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8080  * should handle it ourselves in L0 (and then continue L2). Only call this
8081  * when in is_guest_mode (L2).
8082  */
8083 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
8084 {
8085         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8086         struct vcpu_vmx *vmx = to_vmx(vcpu);
8087         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8088
8089         trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8090                                 vmcs_readl(EXIT_QUALIFICATION),
8091                                 vmx->idt_vectoring_info,
8092                                 intr_info,
8093                                 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8094                                 KVM_ISA_VMX);
8095
8096         /*
8097          * The host physical addresses of some pages of guest memory
8098          * are loaded into VMCS02 (e.g. L1's Virtual APIC Page). The CPU
8099          * may write to these pages via their host physical address while
8100          * L2 is running, bypassing any address-translation-based dirty
8101          * tracking (e.g. EPT write protection).
8102          *
8103          * Mark them dirty on every exit from L2 to prevent them from
8104          * getting out of sync with dirty tracking.
8105          */
8106         nested_mark_vmcs12_pages_dirty(vcpu);
8107
8108         if (vmx->nested.nested_run_pending)
8109                 return false;
8110
8111         if (unlikely(vmx->fail)) {
8112                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8113                                     vmcs_read32(VM_INSTRUCTION_ERROR));
8114                 return true;
8115         }
8116
8117         switch (exit_reason) {
8118         case EXIT_REASON_EXCEPTION_NMI:
8119                 if (is_nmi(intr_info))
8120                         return false;
8121                 else if (is_page_fault(intr_info))
8122                         return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
8123                 else if (is_no_device(intr_info) &&
8124                          !(vmcs12->guest_cr0 & X86_CR0_TS))
8125                         return false;
8126                 else if (is_debug(intr_info) &&
8127                          vcpu->guest_debug &
8128                          (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8129                         return false;
8130                 else if (is_breakpoint(intr_info) &&
8131                          vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8132                         return false;
8133                 return vmcs12->exception_bitmap &
8134                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8135         case EXIT_REASON_EXTERNAL_INTERRUPT:
8136                 return false;
8137         case EXIT_REASON_TRIPLE_FAULT:
8138                 return true;
8139         case EXIT_REASON_PENDING_INTERRUPT:
8140                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8141         case EXIT_REASON_NMI_WINDOW:
8142                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8143         case EXIT_REASON_TASK_SWITCH:
8144                 return true;
8145         case EXIT_REASON_CPUID:
8146                 return true;
8147         case EXIT_REASON_HLT:
8148                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8149         case EXIT_REASON_INVD:
8150                 return true;
8151         case EXIT_REASON_INVLPG:
8152                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8153         case EXIT_REASON_RDPMC:
8154                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8155         case EXIT_REASON_RDRAND:
8156                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND);
8157         case EXIT_REASON_RDSEED:
8158                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED);
8159         case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8160                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8161         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8162         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8163         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8164         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8165         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8166         case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8167                 /*
8168                  * VMX instructions trap unconditionally. This allows L1 to
8169                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
8170                  */
8171                 return true;
8172         case EXIT_REASON_CR_ACCESS:
8173                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8174         case EXIT_REASON_DR_ACCESS:
8175                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8176         case EXIT_REASON_IO_INSTRUCTION:
8177                 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8178         case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8179                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8180         case EXIT_REASON_MSR_READ:
8181         case EXIT_REASON_MSR_WRITE:
8182                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8183         case EXIT_REASON_INVALID_STATE:
8184                 return true;
8185         case EXIT_REASON_MWAIT_INSTRUCTION:
8186                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8187         case EXIT_REASON_MONITOR_TRAP_FLAG:
8188                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8189         case EXIT_REASON_MONITOR_INSTRUCTION:
8190                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8191         case EXIT_REASON_PAUSE_INSTRUCTION:
8192                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8193                         nested_cpu_has2(vmcs12,
8194                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8195         case EXIT_REASON_MCE_DURING_VMENTRY:
8196                 return false;
8197         case EXIT_REASON_TPR_BELOW_THRESHOLD:
8198                 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8199         case EXIT_REASON_APIC_ACCESS:
8200                 return nested_cpu_has2(vmcs12,
8201                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8202         case EXIT_REASON_APIC_WRITE:
8203         case EXIT_REASON_EOI_INDUCED:
8204                 /* apic_write and eoi_induced should exit unconditionally. */
8205                 return true;
8206         case EXIT_REASON_EPT_VIOLATION:
8207                 /*
8208                  * L0 always deals with the EPT violation. If nested EPT is
8209                  * used, and the nested mmu code discovers that the address is
8210                  * missing in the guest EPT table (EPT12), the EPT violation
8211                  * will be injected with nested_ept_inject_page_fault()
8212                  */
8213                 return false;
8214         case EXIT_REASON_EPT_MISCONFIG:
8215                 /*
8216                  * L2 never uses directly L1's EPT, but rather L0's own EPT
8217                  * table (shadow on EPT) or a merged EPT table that L0 built
8218                  * (EPT on EPT). So any problems with the structure of the
8219                  * table is L0's fault.
8220                  */
8221                 return false;
8222         case EXIT_REASON_WBINVD:
8223                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8224         case EXIT_REASON_XSETBV:
8225                 return true;
8226         case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8227                 /*
8228                  * This should never happen, since it is not possible to
8229                  * set XSS to a non-zero value---neither in L1 nor in L2.
8230                  * If if it were, XSS would have to be checked against
8231                  * the XSS exit bitmap in vmcs12.
8232                  */
8233                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8234         case EXIT_REASON_PREEMPTION_TIMER:
8235                 return false;
8236         case EXIT_REASON_PML_FULL:
8237                 /* We emulate PML support to L1. */
8238                 return false;
8239         default:
8240                 return true;
8241         }
8242 }
8243
8244 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
8245 {
8246         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8247
8248         /*
8249          * At this point, the exit interruption info in exit_intr_info
8250          * is only valid for EXCEPTION_NMI exits.  For EXTERNAL_INTERRUPT
8251          * we need to query the in-kernel LAPIC.
8252          */
8253         WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
8254         if ((exit_intr_info &
8255              (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8256             (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
8257                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8258                 vmcs12->vm_exit_intr_error_code =
8259                         vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8260         }
8261
8262         nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
8263                           vmcs_readl(EXIT_QUALIFICATION));
8264         return 1;
8265 }
8266
8267 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8268 {
8269         *info1 = vmcs_readl(EXIT_QUALIFICATION);
8270         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8271 }
8272
8273 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8274 {
8275         if (vmx->pml_pg) {
8276                 __free_page(vmx->pml_pg);
8277                 vmx->pml_pg = NULL;
8278         }
8279 }
8280
8281 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8282 {
8283         struct vcpu_vmx *vmx = to_vmx(vcpu);
8284         u64 *pml_buf;
8285         u16 pml_idx;
8286
8287         pml_idx = vmcs_read16(GUEST_PML_INDEX);
8288
8289         /* Do nothing if PML buffer is empty */
8290         if (pml_idx == (PML_ENTITY_NUM - 1))
8291                 return;
8292
8293         /* PML index always points to next available PML buffer entity */
8294         if (pml_idx >= PML_ENTITY_NUM)
8295                 pml_idx = 0;
8296         else
8297                 pml_idx++;
8298
8299         pml_buf = page_address(vmx->pml_pg);
8300         for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8301                 u64 gpa;
8302
8303                 gpa = pml_buf[pml_idx];
8304                 WARN_ON(gpa & (PAGE_SIZE - 1));
8305                 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8306         }
8307
8308         /* reset PML index */
8309         vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8310 }
8311
8312 /*
8313  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8314  * Called before reporting dirty_bitmap to userspace.
8315  */
8316 static void kvm_flush_pml_buffers(struct kvm *kvm)
8317 {
8318         int i;
8319         struct kvm_vcpu *vcpu;
8320         /*
8321          * We only need to kick vcpu out of guest mode here, as PML buffer
8322          * is flushed at beginning of all VMEXITs, and it's obvious that only
8323          * vcpus running in guest are possible to have unflushed GPAs in PML
8324          * buffer.
8325          */
8326         kvm_for_each_vcpu(i, vcpu, kvm)
8327                 kvm_vcpu_kick(vcpu);
8328 }
8329
8330 static void vmx_dump_sel(char *name, uint32_t sel)
8331 {
8332         pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8333                name, vmcs_read16(sel),
8334                vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8335                vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8336                vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8337 }
8338
8339 static void vmx_dump_dtsel(char *name, uint32_t limit)
8340 {
8341         pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
8342                name, vmcs_read32(limit),
8343                vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8344 }
8345
8346 static void dump_vmcs(void)
8347 {
8348         u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8349         u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8350         u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8351         u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8352         u32 secondary_exec_control = 0;
8353         unsigned long cr4 = vmcs_readl(GUEST_CR4);
8354         u64 efer = vmcs_read64(GUEST_IA32_EFER);
8355         int i, n;
8356
8357         if (cpu_has_secondary_exec_ctrls())
8358                 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8359
8360         pr_err("*** Guest State ***\n");
8361         pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8362                vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8363                vmcs_readl(CR0_GUEST_HOST_MASK));
8364         pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8365                cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8366         pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8367         if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8368             (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8369         {
8370                 pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
8371                        vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8372                 pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
8373                        vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8374         }
8375         pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
8376                vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8377         pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
8378                vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8379         pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8380                vmcs_readl(GUEST_SYSENTER_ESP),
8381                vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8382         vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
8383         vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
8384         vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
8385         vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
8386         vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
8387         vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
8388         vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8389         vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8390         vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8391         vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
8392         if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8393             (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8394                 pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
8395                        efer, vmcs_read64(GUEST_IA32_PAT));
8396         pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
8397                vmcs_read64(GUEST_IA32_DEBUGCTL),
8398                vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8399         if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8400                 pr_err("PerfGlobCtl = 0x%016llx\n",
8401                        vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8402         if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8403                 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8404         pr_err("Interruptibility = %08x  ActivityState = %08x\n",
8405                vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8406                vmcs_read32(GUEST_ACTIVITY_STATE));
8407         if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8408                 pr_err("InterruptStatus = %04x\n",
8409                        vmcs_read16(GUEST_INTR_STATUS));
8410
8411         pr_err("*** Host State ***\n");
8412         pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
8413                vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8414         pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8415                vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8416                vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8417                vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8418                vmcs_read16(HOST_TR_SELECTOR));
8419         pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8420                vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8421                vmcs_readl(HOST_TR_BASE));
8422         pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8423                vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8424         pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8425                vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8426                vmcs_readl(HOST_CR4));
8427         pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8428                vmcs_readl(HOST_IA32_SYSENTER_ESP),
8429                vmcs_read32(HOST_IA32_SYSENTER_CS),
8430                vmcs_readl(HOST_IA32_SYSENTER_EIP));
8431         if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8432                 pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
8433                        vmcs_read64(HOST_IA32_EFER),
8434                        vmcs_read64(HOST_IA32_PAT));
8435         if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8436                 pr_err("PerfGlobCtl = 0x%016llx\n",
8437                        vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8438
8439         pr_err("*** Control State ***\n");
8440         pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8441                pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8442         pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8443         pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8444                vmcs_read32(EXCEPTION_BITMAP),
8445                vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8446                vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8447         pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8448                vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8449                vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8450                vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8451         pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8452                vmcs_read32(VM_EXIT_INTR_INFO),
8453                vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8454                vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8455         pr_err("        reason=%08x qualification=%016lx\n",
8456                vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8457         pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8458                vmcs_read32(IDT_VECTORING_INFO_FIELD),
8459                vmcs_read32(IDT_VECTORING_ERROR_CODE));
8460         pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8461         if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8462                 pr_err("TSC Multiplier = 0x%016llx\n",
8463                        vmcs_read64(TSC_MULTIPLIER));
8464         if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8465                 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8466         if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8467                 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8468         if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8469                 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8470         n = vmcs_read32(CR3_TARGET_COUNT);
8471         for (i = 0; i + 1 < n; i += 4)
8472                 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8473                        i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8474                        i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8475         if (i < n)
8476                 pr_err("CR3 target%u=%016lx\n",
8477                        i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8478         if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8479                 pr_err("PLE Gap=%08x Window=%08x\n",
8480                        vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8481         if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8482                 pr_err("Virtual processor ID = 0x%04x\n",
8483                        vmcs_read16(VIRTUAL_PROCESSOR_ID));
8484 }
8485
8486 /*
8487  * The guest has exited.  See if we can fix it or if we need userspace
8488  * assistance.
8489  */
8490 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8491 {
8492         struct vcpu_vmx *vmx = to_vmx(vcpu);
8493         u32 exit_reason = vmx->exit_reason;
8494         u32 vectoring_info = vmx->idt_vectoring_info;
8495
8496         trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8497         vcpu->arch.gpa_available = false;
8498
8499         /*
8500          * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8501          * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8502          * querying dirty_bitmap, we only need to kick all vcpus out of guest
8503          * mode as if vcpus is in root mode, the PML buffer must has been
8504          * flushed already.
8505          */
8506         if (enable_pml)
8507                 vmx_flush_pml_buffer(vcpu);
8508
8509         /* If guest state is invalid, start emulating */
8510         if (vmx->emulation_required)
8511                 return handle_invalid_guest_state(vcpu);
8512
8513         if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
8514                 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
8515
8516         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8517                 dump_vmcs();
8518                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8519                 vcpu->run->fail_entry.hardware_entry_failure_reason
8520                         = exit_reason;
8521                 return 0;
8522         }
8523
8524         if (unlikely(vmx->fail)) {
8525                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8526                 vcpu->run->fail_entry.hardware_entry_failure_reason
8527                         = vmcs_read32(VM_INSTRUCTION_ERROR);
8528                 return 0;
8529         }
8530
8531         /*
8532          * Note:
8533          * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8534          * delivery event since it indicates guest is accessing MMIO.
8535          * The vm-exit can be triggered again after return to guest that
8536          * will cause infinite loop.
8537          */
8538         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8539                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8540                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
8541                         exit_reason != EXIT_REASON_PML_FULL &&
8542                         exit_reason != EXIT_REASON_TASK_SWITCH)) {
8543                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8544                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8545                 vcpu->run->internal.ndata = 3;
8546                 vcpu->run->internal.data[0] = vectoring_info;
8547                 vcpu->run->internal.data[1] = exit_reason;
8548                 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
8549                 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
8550                         vcpu->run->internal.ndata++;
8551                         vcpu->run->internal.data[3] =
8552                                 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
8553                 }
8554                 return 0;
8555         }
8556
8557         if (exit_reason < kvm_vmx_max_exit_handlers
8558             && kvm_vmx_exit_handlers[exit_reason])
8559                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8560         else {
8561                 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8562                                 exit_reason);
8563                 kvm_queue_exception(vcpu, UD_VECTOR);
8564                 return 1;
8565         }
8566 }
8567
8568 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8569 {
8570         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8571
8572         if (is_guest_mode(vcpu) &&
8573                 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8574                 return;
8575
8576         if (irr == -1 || tpr < irr) {
8577                 vmcs_write32(TPR_THRESHOLD, 0);
8578                 return;
8579         }
8580
8581         vmcs_write32(TPR_THRESHOLD, irr);
8582 }
8583
8584 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8585 {
8586         u32 sec_exec_control;
8587
8588         /* Postpone execution until vmcs01 is the current VMCS. */
8589         if (is_guest_mode(vcpu)) {
8590                 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8591                 return;
8592         }
8593
8594         if (!cpu_has_vmx_virtualize_x2apic_mode())
8595                 return;
8596
8597         if (!cpu_need_tpr_shadow(vcpu))
8598                 return;
8599
8600         sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8601
8602         if (set) {
8603                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8604                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8605         } else {
8606                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8607                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8608                 vmx_flush_tlb_ept_only(vcpu);
8609         }
8610         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8611
8612         vmx_set_msr_bitmap(vcpu);
8613 }
8614
8615 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8616 {
8617         struct vcpu_vmx *vmx = to_vmx(vcpu);
8618
8619         /*
8620          * Currently we do not handle the nested case where L2 has an
8621          * APIC access page of its own; that page is still pinned.
8622          * Hence, we skip the case where the VCPU is in guest mode _and_
8623          * L1 prepared an APIC access page for L2.
8624          *
8625          * For the case where L1 and L2 share the same APIC access page
8626          * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8627          * in the vmcs12), this function will only update either the vmcs01
8628          * or the vmcs02.  If the former, the vmcs02 will be updated by
8629          * prepare_vmcs02.  If the latter, the vmcs01 will be updated in
8630          * the next L2->L1 exit.
8631          */
8632         if (!is_guest_mode(vcpu) ||
8633             !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8634                              SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8635                 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8636                 vmx_flush_tlb_ept_only(vcpu);
8637         }
8638 }
8639
8640 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8641 {
8642         u16 status;
8643         u8 old;
8644
8645         if (max_isr == -1)
8646                 max_isr = 0;
8647
8648         status = vmcs_read16(GUEST_INTR_STATUS);
8649         old = status >> 8;
8650         if (max_isr != old) {
8651                 status &= 0xff;
8652                 status |= max_isr << 8;
8653                 vmcs_write16(GUEST_INTR_STATUS, status);
8654         }
8655 }
8656
8657 static void vmx_set_rvi(int vector)
8658 {
8659         u16 status;
8660         u8 old;
8661
8662         if (vector == -1)
8663                 vector = 0;
8664
8665         status = vmcs_read16(GUEST_INTR_STATUS);
8666         old = (u8)status & 0xff;
8667         if ((u8)vector != old) {
8668                 status &= ~0xff;
8669                 status |= (u8)vector;
8670                 vmcs_write16(GUEST_INTR_STATUS, status);
8671         }
8672 }
8673
8674 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8675 {
8676         if (!is_guest_mode(vcpu)) {
8677                 vmx_set_rvi(max_irr);
8678                 return;
8679         }
8680
8681         if (max_irr == -1)
8682                 return;
8683
8684         /*
8685          * In guest mode.  If a vmexit is needed, vmx_check_nested_events
8686          * handles it.
8687          */
8688         if (nested_exit_on_intr(vcpu))
8689                 return;
8690
8691         /*
8692          * Else, fall back to pre-APICv interrupt injection since L2
8693          * is run without virtual interrupt delivery.
8694          */
8695         if (!kvm_event_needs_reinjection(vcpu) &&
8696             vmx_interrupt_allowed(vcpu)) {
8697                 kvm_queue_interrupt(vcpu, max_irr, false);
8698                 vmx_inject_irq(vcpu);
8699         }
8700 }
8701
8702 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
8703 {
8704         struct vcpu_vmx *vmx = to_vmx(vcpu);
8705         int max_irr;
8706
8707         WARN_ON(!vcpu->arch.apicv_active);
8708         if (pi_test_on(&vmx->pi_desc)) {
8709                 pi_clear_on(&vmx->pi_desc);
8710                 /*
8711                  * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8712                  * But on x86 this is just a compiler barrier anyway.
8713                  */
8714                 smp_mb__after_atomic();
8715                 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
8716         } else {
8717                 max_irr = kvm_lapic_find_highest_irr(vcpu);
8718         }
8719         vmx_hwapic_irr_update(vcpu, max_irr);
8720         return max_irr;
8721 }
8722
8723 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8724 {
8725         if (!kvm_vcpu_apicv_active(vcpu))
8726                 return;
8727
8728         vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8729         vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8730         vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8731         vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8732 }
8733
8734 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
8735 {
8736         struct vcpu_vmx *vmx = to_vmx(vcpu);
8737
8738         pi_clear_on(&vmx->pi_desc);
8739         memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
8740 }
8741
8742 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8743 {
8744         u32 exit_intr_info = 0;
8745         u16 basic_exit_reason = (u16)vmx->exit_reason;
8746
8747         if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8748               || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
8749                 return;
8750
8751         if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
8752                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8753         vmx->exit_intr_info = exit_intr_info;
8754
8755         /* if exit due to PF check for async PF */
8756         if (is_page_fault(exit_intr_info))
8757                 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
8758
8759         /* Handle machine checks before interrupts are enabled */
8760         if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
8761             is_machine_check(exit_intr_info))
8762                 kvm_machine_check();
8763
8764         /* We need to handle NMIs before interrupts are enabled */
8765         if (is_nmi(exit_intr_info)) {
8766                 kvm_before_handle_nmi(&vmx->vcpu);
8767                 asm("int $2");
8768                 kvm_after_handle_nmi(&vmx->vcpu);
8769         }
8770 }
8771
8772 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8773 {
8774         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8775         register void *__sp asm(_ASM_SP);
8776
8777         if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8778                         == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8779                 unsigned int vector;
8780                 unsigned long entry;
8781                 gate_desc *desc;
8782                 struct vcpu_vmx *vmx = to_vmx(vcpu);
8783 #ifdef CONFIG_X86_64
8784                 unsigned long tmp;
8785 #endif
8786
8787                 vector =  exit_intr_info & INTR_INFO_VECTOR_MASK;
8788                 desc = (gate_desc *)vmx->host_idt_base + vector;
8789                 entry = gate_offset(*desc);
8790                 asm volatile(
8791 #ifdef CONFIG_X86_64
8792                         "mov %%" _ASM_SP ", %[sp]\n\t"
8793                         "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8794                         "push $%c[ss]\n\t"
8795                         "push %[sp]\n\t"
8796 #endif
8797                         "pushf\n\t"
8798                         __ASM_SIZE(push) " $%c[cs]\n\t"
8799                         "call *%[entry]\n\t"
8800                         :
8801 #ifdef CONFIG_X86_64
8802                         [sp]"=&r"(tmp),
8803 #endif
8804                         "+r"(__sp)
8805                         :
8806                         [entry]"r"(entry),
8807                         [ss]"i"(__KERNEL_DS),
8808                         [cs]"i"(__KERNEL_CS)
8809                         );
8810         }
8811 }
8812 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
8813
8814 static bool vmx_has_high_real_mode_segbase(void)
8815 {
8816         return enable_unrestricted_guest || emulate_invalid_guest_state;
8817 }
8818
8819 static bool vmx_mpx_supported(void)
8820 {
8821         return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8822                 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8823 }
8824
8825 static bool vmx_xsaves_supported(void)
8826 {
8827         return vmcs_config.cpu_based_2nd_exec_ctrl &
8828                 SECONDARY_EXEC_XSAVES;
8829 }
8830
8831 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8832 {
8833         u32 exit_intr_info;
8834         bool unblock_nmi;
8835         u8 vector;
8836         bool idtv_info_valid;
8837
8838         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8839
8840         if (vmx->loaded_vmcs->nmi_known_unmasked)
8841                 return;
8842         /*
8843          * Can't use vmx->exit_intr_info since we're not sure what
8844          * the exit reason is.
8845          */
8846         exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8847         unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8848         vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8849         /*
8850          * SDM 3: 27.7.1.2 (September 2008)
8851          * Re-set bit "block by NMI" before VM entry if vmexit caused by
8852          * a guest IRET fault.
8853          * SDM 3: 23.2.2 (September 2008)
8854          * Bit 12 is undefined in any of the following cases:
8855          *  If the VM exit sets the valid bit in the IDT-vectoring
8856          *   information field.
8857          *  If the VM exit is due to a double fault.
8858          */
8859         if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8860             vector != DF_VECTOR && !idtv_info_valid)
8861                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8862                               GUEST_INTR_STATE_NMI);
8863         else
8864                 vmx->loaded_vmcs->nmi_known_unmasked =
8865                         !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8866                           & GUEST_INTR_STATE_NMI);
8867 }
8868
8869 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8870                                       u32 idt_vectoring_info,
8871                                       int instr_len_field,
8872                                       int error_code_field)
8873 {
8874         u8 vector;
8875         int type;
8876         bool idtv_info_valid;
8877
8878         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8879
8880         vcpu->arch.nmi_injected = false;
8881         kvm_clear_exception_queue(vcpu);
8882         kvm_clear_interrupt_queue(vcpu);
8883
8884         if (!idtv_info_valid)
8885                 return;
8886
8887         kvm_make_request(KVM_REQ_EVENT, vcpu);
8888
8889         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8890         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8891
8892         switch (type) {
8893         case INTR_TYPE_NMI_INTR:
8894                 vcpu->arch.nmi_injected = true;
8895                 /*
8896                  * SDM 3: 27.7.1.2 (September 2008)
8897                  * Clear bit "block by NMI" before VM entry if a NMI
8898                  * delivery faulted.
8899                  */
8900                 vmx_set_nmi_mask(vcpu, false);
8901                 break;
8902         case INTR_TYPE_SOFT_EXCEPTION:
8903                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8904                 /* fall through */
8905         case INTR_TYPE_HARD_EXCEPTION:
8906                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8907                         u32 err = vmcs_read32(error_code_field);
8908                         kvm_requeue_exception_e(vcpu, vector, err);
8909                 } else
8910                         kvm_requeue_exception(vcpu, vector);
8911                 break;
8912         case INTR_TYPE_SOFT_INTR:
8913                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8914                 /* fall through */
8915         case INTR_TYPE_EXT_INTR:
8916                 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8917                 break;
8918         default:
8919                 break;
8920         }
8921 }
8922
8923 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8924 {
8925         __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8926                                   VM_EXIT_INSTRUCTION_LEN,
8927                                   IDT_VECTORING_ERROR_CODE);
8928 }
8929
8930 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8931 {
8932         __vmx_complete_interrupts(vcpu,
8933                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8934                                   VM_ENTRY_INSTRUCTION_LEN,
8935                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
8936
8937         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8938 }
8939
8940 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8941 {
8942         int i, nr_msrs;
8943         struct perf_guest_switch_msr *msrs;
8944
8945         msrs = perf_guest_get_msrs(&nr_msrs);
8946
8947         if (!msrs)
8948                 return;
8949
8950         for (i = 0; i < nr_msrs; i++)
8951                 if (msrs[i].host == msrs[i].guest)
8952                         clear_atomic_switch_msr(vmx, msrs[i].msr);
8953                 else
8954                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8955                                         msrs[i].host);
8956 }
8957
8958 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8959 {
8960         struct vcpu_vmx *vmx = to_vmx(vcpu);
8961         u64 tscl;
8962         u32 delta_tsc;
8963
8964         if (vmx->hv_deadline_tsc == -1)
8965                 return;
8966
8967         tscl = rdtsc();
8968         if (vmx->hv_deadline_tsc > tscl)
8969                 /* sure to be 32 bit only because checked on set_hv_timer */
8970                 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8971                         cpu_preemption_timer_multi);
8972         else
8973                 delta_tsc = 0;
8974
8975         vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8976 }
8977
8978 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8979 {
8980         struct vcpu_vmx *vmx = to_vmx(vcpu);
8981         unsigned long debugctlmsr, cr3, cr4;
8982
8983         /* Don't enter VMX if guest state is invalid, let the exit handler
8984            start emulation until we arrive back to a valid state */
8985         if (vmx->emulation_required)
8986                 return;
8987
8988         if (vmx->ple_window_dirty) {
8989                 vmx->ple_window_dirty = false;
8990                 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8991         }
8992
8993         if (vmx->nested.sync_shadow_vmcs) {
8994                 copy_vmcs12_to_shadow(vmx);
8995                 vmx->nested.sync_shadow_vmcs = false;
8996         }
8997
8998         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8999                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
9000         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
9001                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
9002
9003         cr3 = __get_current_cr3_fast();
9004         if (unlikely(cr3 != vmx->host_state.vmcs_host_cr3)) {
9005                 vmcs_writel(HOST_CR3, cr3);
9006                 vmx->host_state.vmcs_host_cr3 = cr3;
9007         }
9008
9009         cr4 = cr4_read_shadow();
9010         if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
9011                 vmcs_writel(HOST_CR4, cr4);
9012                 vmx->host_state.vmcs_host_cr4 = cr4;
9013         }
9014
9015         /* When single-stepping over STI and MOV SS, we must clear the
9016          * corresponding interruptibility bits in the guest state. Otherwise
9017          * vmentry fails as it then expects bit 14 (BS) in pending debug
9018          * exceptions being set, but that's not correct for the guest debugging
9019          * case. */
9020         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9021                 vmx_set_interrupt_shadow(vcpu, 0);
9022
9023         if (vmx->guest_pkru_valid)
9024                 __write_pkru(vmx->guest_pkru);
9025
9026         atomic_switch_perf_msrs(vmx);
9027         debugctlmsr = get_debugctlmsr();
9028
9029         vmx_arm_hv_timer(vcpu);
9030
9031         vmx->__launched = vmx->loaded_vmcs->launched;
9032         asm(
9033                 /* Store host registers */
9034                 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
9035                 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
9036                 "push %%" _ASM_CX " \n\t"
9037                 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9038                 "je 1f \n\t"
9039                 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9040                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
9041                 "1: \n\t"
9042                 /* Reload cr2 if changed */
9043                 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9044                 "mov %%cr2, %%" _ASM_DX " \n\t"
9045                 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9046                 "je 2f \n\t"
9047                 "mov %%" _ASM_AX", %%cr2 \n\t"
9048                 "2: \n\t"
9049                 /* Check if vmlaunch of vmresume is needed */
9050                 "cmpl $0, %c[launched](%0) \n\t"
9051                 /* Load guest registers.  Don't clobber flags. */
9052                 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9053                 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9054                 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9055                 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9056                 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9057                 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9058 #ifdef CONFIG_X86_64
9059                 "mov %c[r8](%0),  %%r8  \n\t"
9060                 "mov %c[r9](%0),  %%r9  \n\t"
9061                 "mov %c[r10](%0), %%r10 \n\t"
9062                 "mov %c[r11](%0), %%r11 \n\t"
9063                 "mov %c[r12](%0), %%r12 \n\t"
9064                 "mov %c[r13](%0), %%r13 \n\t"
9065                 "mov %c[r14](%0), %%r14 \n\t"
9066                 "mov %c[r15](%0), %%r15 \n\t"
9067 #endif
9068                 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9069
9070                 /* Enter guest mode */
9071                 "jne 1f \n\t"
9072                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9073                 "jmp 2f \n\t"
9074                 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9075                 "2: "
9076                 /* Save guest registers, load host registers, keep flags */
9077                 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9078                 "pop %0 \n\t"
9079                 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9080                 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9081                 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9082                 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9083                 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9084                 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9085                 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9086 #ifdef CONFIG_X86_64
9087                 "mov %%r8,  %c[r8](%0) \n\t"
9088                 "mov %%r9,  %c[r9](%0) \n\t"
9089                 "mov %%r10, %c[r10](%0) \n\t"
9090                 "mov %%r11, %c[r11](%0) \n\t"
9091                 "mov %%r12, %c[r12](%0) \n\t"
9092                 "mov %%r13, %c[r13](%0) \n\t"
9093                 "mov %%r14, %c[r14](%0) \n\t"
9094                 "mov %%r15, %c[r15](%0) \n\t"
9095 #endif
9096                 "mov %%cr2, %%" _ASM_AX "   \n\t"
9097                 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9098
9099                 "pop  %%" _ASM_BP "; pop  %%" _ASM_DX " \n\t"
9100                 "setbe %c[fail](%0) \n\t"
9101                 ".pushsection .rodata \n\t"
9102                 ".global vmx_return \n\t"
9103                 "vmx_return: " _ASM_PTR " 2b \n\t"
9104                 ".popsection"
9105               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9106                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9107                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9108                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9109                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9110                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9111                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9112                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9113                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9114                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9115                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9116 #ifdef CONFIG_X86_64
9117                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9118                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9119                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9120                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9121                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9122                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9123                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9124                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9125 #endif
9126                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9127                 [wordsize]"i"(sizeof(ulong))
9128               : "cc", "memory"
9129 #ifdef CONFIG_X86_64
9130                 , "rax", "rbx", "rdi", "rsi"
9131                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9132 #else
9133                 , "eax", "ebx", "edi", "esi"
9134 #endif
9135               );
9136
9137         /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9138         if (debugctlmsr)
9139                 update_debugctlmsr(debugctlmsr);
9140
9141 #ifndef CONFIG_X86_64
9142         /*
9143          * The sysexit path does not restore ds/es, so we must set them to
9144          * a reasonable value ourselves.
9145          *
9146          * We can't defer this to vmx_load_host_state() since that function
9147          * may be executed in interrupt context, which saves and restore segments
9148          * around it, nullifying its effect.
9149          */
9150         loadsegment(ds, __USER_DS);
9151         loadsegment(es, __USER_DS);
9152 #endif
9153
9154         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9155                                   | (1 << VCPU_EXREG_RFLAGS)
9156                                   | (1 << VCPU_EXREG_PDPTR)
9157                                   | (1 << VCPU_EXREG_SEGMENTS)
9158                                   | (1 << VCPU_EXREG_CR3));
9159         vcpu->arch.regs_dirty = 0;
9160
9161         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9162
9163         vmx->loaded_vmcs->launched = 1;
9164
9165         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
9166
9167         /*
9168          * eager fpu is enabled if PKEY is supported and CR4 is switched
9169          * back on host, so it is safe to read guest PKRU from current
9170          * XSAVE.
9171          */
9172         if (boot_cpu_has(X86_FEATURE_OSPKE)) {
9173                 vmx->guest_pkru = __read_pkru();
9174                 if (vmx->guest_pkru != vmx->host_pkru) {
9175                         vmx->guest_pkru_valid = true;
9176                         __write_pkru(vmx->host_pkru);
9177                 } else
9178                         vmx->guest_pkru_valid = false;
9179         }
9180
9181         /*
9182          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9183          * we did not inject a still-pending event to L1 now because of
9184          * nested_run_pending, we need to re-enable this bit.
9185          */
9186         if (vmx->nested.nested_run_pending)
9187                 kvm_make_request(KVM_REQ_EVENT, vcpu);
9188
9189         vmx->nested.nested_run_pending = 0;
9190
9191         vmx_complete_atomic_exit(vmx);
9192         vmx_recover_nmi_blocking(vmx);
9193         vmx_complete_interrupts(vmx);
9194 }
9195 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
9196
9197 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9198 {
9199         struct vcpu_vmx *vmx = to_vmx(vcpu);
9200         int cpu;
9201
9202         if (vmx->loaded_vmcs == vmcs)
9203                 return;
9204
9205         cpu = get_cpu();
9206         vmx->loaded_vmcs = vmcs;
9207         vmx_vcpu_put(vcpu);
9208         vmx_vcpu_load(vcpu, cpu);
9209         vcpu->cpu = cpu;
9210         put_cpu();
9211 }
9212
9213 /*
9214  * Ensure that the current vmcs of the logical processor is the
9215  * vmcs01 of the vcpu before calling free_nested().
9216  */
9217 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9218 {
9219        struct vcpu_vmx *vmx = to_vmx(vcpu);
9220        int r;
9221
9222        r = vcpu_load(vcpu);
9223        BUG_ON(r);
9224        vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9225        free_nested(vmx);
9226        vcpu_put(vcpu);
9227 }
9228
9229 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9230 {
9231         struct vcpu_vmx *vmx = to_vmx(vcpu);
9232
9233         if (enable_pml)
9234                 vmx_destroy_pml_buffer(vmx);
9235         free_vpid(vmx->vpid);
9236         leave_guest_mode(vcpu);
9237         vmx_free_vcpu_nested(vcpu);
9238         free_loaded_vmcs(vmx->loaded_vmcs);
9239         kfree(vmx->guest_msrs);
9240         kvm_vcpu_uninit(vcpu);
9241         kmem_cache_free(kvm_vcpu_cache, vmx);
9242 }
9243
9244 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9245 {
9246         int err;
9247         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9248         int cpu;
9249
9250         if (!vmx)
9251                 return ERR_PTR(-ENOMEM);
9252
9253         vmx->vpid = allocate_vpid();
9254
9255         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9256         if (err)
9257                 goto free_vcpu;
9258
9259         err = -ENOMEM;
9260
9261         /*
9262          * If PML is turned on, failure on enabling PML just results in failure
9263          * of creating the vcpu, therefore we can simplify PML logic (by
9264          * avoiding dealing with cases, such as enabling PML partially on vcpus
9265          * for the guest, etc.
9266          */
9267         if (enable_pml) {
9268                 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9269                 if (!vmx->pml_pg)
9270                         goto uninit_vcpu;
9271         }
9272
9273         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9274         BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9275                      > PAGE_SIZE);
9276
9277         if (!vmx->guest_msrs)
9278                 goto free_pml;
9279
9280         vmx->loaded_vmcs = &vmx->vmcs01;
9281         vmx->loaded_vmcs->vmcs = alloc_vmcs();
9282         vmx->loaded_vmcs->shadow_vmcs = NULL;
9283         if (!vmx->loaded_vmcs->vmcs)
9284                 goto free_msrs;
9285         loaded_vmcs_init(vmx->loaded_vmcs);
9286
9287         cpu = get_cpu();
9288         vmx_vcpu_load(&vmx->vcpu, cpu);
9289         vmx->vcpu.cpu = cpu;
9290         err = vmx_vcpu_setup(vmx);
9291         vmx_vcpu_put(&vmx->vcpu);
9292         put_cpu();
9293         if (err)
9294                 goto free_vmcs;
9295         if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9296                 err = alloc_apic_access_page(kvm);
9297                 if (err)
9298                         goto free_vmcs;
9299         }
9300
9301         if (enable_ept) {
9302                 if (!kvm->arch.ept_identity_map_addr)
9303                         kvm->arch.ept_identity_map_addr =
9304                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9305                 err = init_rmode_identity_map(kvm);
9306                 if (err)
9307                         goto free_vmcs;
9308         }
9309
9310         if (nested) {
9311                 nested_vmx_setup_ctls_msrs(vmx);
9312                 vmx->nested.vpid02 = allocate_vpid();
9313         }
9314
9315         vmx->nested.posted_intr_nv = -1;
9316         vmx->nested.current_vmptr = -1ull;
9317
9318         vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9319
9320         return &vmx->vcpu;
9321
9322 free_vmcs:
9323         free_vpid(vmx->nested.vpid02);
9324         free_loaded_vmcs(vmx->loaded_vmcs);
9325 free_msrs:
9326         kfree(vmx->guest_msrs);
9327 free_pml:
9328         vmx_destroy_pml_buffer(vmx);
9329 uninit_vcpu:
9330         kvm_vcpu_uninit(&vmx->vcpu);
9331 free_vcpu:
9332         free_vpid(vmx->vpid);
9333         kmem_cache_free(kvm_vcpu_cache, vmx);
9334         return ERR_PTR(err);
9335 }
9336
9337 static void __init vmx_check_processor_compat(void *rtn)
9338 {
9339         struct vmcs_config vmcs_conf;
9340
9341         *(int *)rtn = 0;
9342         if (setup_vmcs_config(&vmcs_conf) < 0)
9343                 *(int *)rtn = -EIO;
9344         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9345                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9346                                 smp_processor_id());
9347                 *(int *)rtn = -EIO;
9348         }
9349 }
9350
9351 static int get_ept_level(void)
9352 {
9353         return VMX_EPT_DEFAULT_GAW + 1;
9354 }
9355
9356 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9357 {
9358         u8 cache;
9359         u64 ipat = 0;
9360
9361         /* For VT-d and EPT combination
9362          * 1. MMIO: always map as UC
9363          * 2. EPT with VT-d:
9364          *   a. VT-d without snooping control feature: can't guarantee the
9365          *      result, try to trust guest.
9366          *   b. VT-d with snooping control feature: snooping control feature of
9367          *      VT-d engine can guarantee the cache correctness. Just set it
9368          *      to WB to keep consistent with host. So the same as item 3.
9369          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9370          *    consistent with host MTRR
9371          */
9372         if (is_mmio) {
9373                 cache = MTRR_TYPE_UNCACHABLE;
9374                 goto exit;
9375         }
9376
9377         if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9378                 ipat = VMX_EPT_IPAT_BIT;
9379                 cache = MTRR_TYPE_WRBACK;
9380                 goto exit;
9381         }
9382
9383         if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9384                 ipat = VMX_EPT_IPAT_BIT;
9385                 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9386                         cache = MTRR_TYPE_WRBACK;
9387                 else
9388                         cache = MTRR_TYPE_UNCACHABLE;
9389                 goto exit;
9390         }
9391
9392         cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9393
9394 exit:
9395         return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9396 }
9397
9398 static int vmx_get_lpage_level(void)
9399 {
9400         if (enable_ept && !cpu_has_vmx_ept_1g_page())
9401                 return PT_DIRECTORY_LEVEL;
9402         else
9403                 /* For shadow and EPT supported 1GB page */
9404                 return PT_PDPE_LEVEL;
9405 }
9406
9407 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9408 {
9409         /*
9410          * These bits in the secondary execution controls field
9411          * are dynamic, the others are mostly based on the hypervisor
9412          * architecture and the guest's CPUID.  Do not touch the
9413          * dynamic bits.
9414          */
9415         u32 mask =
9416                 SECONDARY_EXEC_SHADOW_VMCS |
9417                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9418                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9419
9420         u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9421
9422         vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9423                      (new_ctl & ~mask) | (cur_ctl & mask));
9424 }
9425
9426 /*
9427  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9428  * (indicating "allowed-1") if they are supported in the guest's CPUID.
9429  */
9430 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9431 {
9432         struct vcpu_vmx *vmx = to_vmx(vcpu);
9433         struct kvm_cpuid_entry2 *entry;
9434
9435         vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9436         vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9437
9438 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {            \
9439         if (entry && (entry->_reg & (_cpuid_mask)))                     \
9440                 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask);       \
9441 } while (0)
9442
9443         entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9444         cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
9445         cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
9446         cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
9447         cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
9448         cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
9449         cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
9450         cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
9451         cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
9452         cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
9453         cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9454         cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
9455         cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
9456         cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
9457         cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
9458
9459         entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9460         cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
9461         cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
9462         cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
9463         cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
9464         /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9465         cr4_fixed1_update(bit(11),            ecx, bit(2));
9466
9467 #undef cr4_fixed1_update
9468 }
9469
9470 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9471 {
9472         struct kvm_cpuid_entry2 *best;
9473         struct vcpu_vmx *vmx = to_vmx(vcpu);
9474         u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9475
9476         if (vmx_rdtscp_supported()) {
9477                 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9478                 if (!rdtscp_enabled)
9479                         secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9480
9481                 if (nested) {
9482                         if (rdtscp_enabled)
9483                                 vmx->nested.nested_vmx_secondary_ctls_high |=
9484                                         SECONDARY_EXEC_RDTSCP;
9485                         else
9486                                 vmx->nested.nested_vmx_secondary_ctls_high &=
9487                                         ~SECONDARY_EXEC_RDTSCP;
9488                 }
9489         }
9490
9491         /* Exposing INVPCID only when PCID is exposed */
9492         best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9493         if (vmx_invpcid_supported() &&
9494             (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9495             !guest_cpuid_has_pcid(vcpu))) {
9496                 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9497
9498                 if (best)
9499                         best->ebx &= ~bit(X86_FEATURE_INVPCID);
9500         }
9501
9502         if (cpu_has_secondary_exec_ctrls())
9503                 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9504
9505         if (nested_vmx_allowed(vcpu))
9506                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9507                         FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9508         else
9509                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9510                         ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9511
9512         if (nested_vmx_allowed(vcpu))
9513                 nested_vmx_cr_fixed1_bits_update(vcpu);
9514 }
9515
9516 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9517 {
9518         if (func == 1 && nested)
9519                 entry->ecx |= bit(X86_FEATURE_VMX);
9520 }
9521
9522 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9523                 struct x86_exception *fault)
9524 {
9525         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9526         struct vcpu_vmx *vmx = to_vmx(vcpu);
9527         u32 exit_reason;
9528         unsigned long exit_qualification = vcpu->arch.exit_qualification;
9529
9530         if (vmx->nested.pml_full) {
9531                 exit_reason = EXIT_REASON_PML_FULL;
9532                 vmx->nested.pml_full = false;
9533                 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
9534         } else if (fault->error_code & PFERR_RSVD_MASK)
9535                 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9536         else
9537                 exit_reason = EXIT_REASON_EPT_VIOLATION;
9538
9539         nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
9540         vmcs12->guest_physical_address = fault->address;
9541 }
9542
9543 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
9544 {
9545         return nested_ept_get_cr3(vcpu) & VMX_EPT_AD_ENABLE_BIT;
9546 }
9547
9548 /* Callbacks for nested_ept_init_mmu_context: */
9549
9550 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9551 {
9552         /* return the page table to be shadowed - in our case, EPT12 */
9553         return get_vmcs12(vcpu)->ept_pointer;
9554 }
9555
9556 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9557 {
9558         bool wants_ad;
9559
9560         WARN_ON(mmu_is_nested(vcpu));
9561         wants_ad = nested_ept_ad_enabled(vcpu);
9562         if (wants_ad && !enable_ept_ad_bits)
9563                 return 1;
9564
9565         kvm_mmu_unload(vcpu);
9566         kvm_init_shadow_ept_mmu(vcpu,
9567                         to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9568                         VMX_EPT_EXECUTE_ONLY_BIT,
9569                         wants_ad);
9570         vcpu->arch.mmu.set_cr3           = vmx_set_cr3;
9571         vcpu->arch.mmu.get_cr3           = nested_ept_get_cr3;
9572         vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9573
9574         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
9575         return 0;
9576 }
9577
9578 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9579 {
9580         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9581 }
9582
9583 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9584                                             u16 error_code)
9585 {
9586         bool inequality, bit;
9587
9588         bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9589         inequality =
9590                 (error_code & vmcs12->page_fault_error_code_mask) !=
9591                  vmcs12->page_fault_error_code_match;
9592         return inequality ^ bit;
9593 }
9594
9595 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9596                 struct x86_exception *fault)
9597 {
9598         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9599
9600         WARN_ON(!is_guest_mode(vcpu));
9601
9602         if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code)) {
9603                 vmcs12->vm_exit_intr_error_code = fault->error_code;
9604                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
9605                                   PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
9606                                   INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
9607                                   fault->address);
9608         } else {
9609                 kvm_inject_page_fault(vcpu, fault);
9610         }
9611 }
9612
9613 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9614                                                struct vmcs12 *vmcs12);
9615
9616 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9617                                         struct vmcs12 *vmcs12)
9618 {
9619         struct vcpu_vmx *vmx = to_vmx(vcpu);
9620         u64 hpa;
9621
9622         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9623                 /*
9624                  * Translate L1 physical address to host physical
9625                  * address for vmcs02. Keep the page pinned, so this
9626                  * physical address remains valid. We keep a reference
9627                  * to it so we can release it later.
9628                  */
9629                 if (vmx->nested.apic_access_page) /* shouldn't happen */
9630                         nested_release_page(vmx->nested.apic_access_page);
9631                 vmx->nested.apic_access_page =
9632                         nested_get_page(vcpu, vmcs12->apic_access_addr);
9633                 /*
9634                  * If translation failed, no matter: This feature asks
9635                  * to exit when accessing the given address, and if it
9636                  * can never be accessed, this feature won't do
9637                  * anything anyway.
9638                  */
9639                 if (vmx->nested.apic_access_page) {
9640                         hpa = page_to_phys(vmx->nested.apic_access_page);
9641                         vmcs_write64(APIC_ACCESS_ADDR, hpa);
9642                 } else {
9643                         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9644                                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9645                 }
9646         } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9647                    cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9648                 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9649                               SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9650                 kvm_vcpu_reload_apic_access_page(vcpu);
9651         }
9652
9653         if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9654                 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9655                         nested_release_page(vmx->nested.virtual_apic_page);
9656                 vmx->nested.virtual_apic_page =
9657                         nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9658
9659                 /*
9660                  * If translation failed, VM entry will fail because
9661                  * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9662                  * Failing the vm entry is _not_ what the processor
9663                  * does but it's basically the only possibility we
9664                  * have.  We could still enter the guest if CR8 load
9665                  * exits are enabled, CR8 store exits are enabled, and
9666                  * virtualize APIC access is disabled; in this case
9667                  * the processor would never use the TPR shadow and we
9668                  * could simply clear the bit from the execution
9669                  * control.  But such a configuration is useless, so
9670                  * let's keep the code simple.
9671                  */
9672                 if (vmx->nested.virtual_apic_page) {
9673                         hpa = page_to_phys(vmx->nested.virtual_apic_page);
9674                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9675                 }
9676         }
9677
9678         if (nested_cpu_has_posted_intr(vmcs12)) {
9679                 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9680                         kunmap(vmx->nested.pi_desc_page);
9681                         nested_release_page(vmx->nested.pi_desc_page);
9682                 }
9683                 vmx->nested.pi_desc_page =
9684                         nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9685                 vmx->nested.pi_desc =
9686                         (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9687                 if (!vmx->nested.pi_desc) {
9688                         nested_release_page_clean(vmx->nested.pi_desc_page);
9689                         return;
9690                 }
9691                 vmx->nested.pi_desc =
9692                         (struct pi_desc *)((void *)vmx->nested.pi_desc +
9693                         (unsigned long)(vmcs12->posted_intr_desc_addr &
9694                         (PAGE_SIZE - 1)));
9695                 vmcs_write64(POSTED_INTR_DESC_ADDR,
9696                         page_to_phys(vmx->nested.pi_desc_page) +
9697                         (unsigned long)(vmcs12->posted_intr_desc_addr &
9698                         (PAGE_SIZE - 1)));
9699         }
9700         if (cpu_has_vmx_msr_bitmap() &&
9701             nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
9702             nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
9703                 ;
9704         else
9705                 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
9706                                 CPU_BASED_USE_MSR_BITMAPS);
9707 }
9708
9709 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9710 {
9711         u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9712         struct vcpu_vmx *vmx = to_vmx(vcpu);
9713
9714         if (vcpu->arch.virtual_tsc_khz == 0)
9715                 return;
9716
9717         /* Make sure short timeouts reliably trigger an immediate vmexit.
9718          * hrtimer_start does not guarantee this. */
9719         if (preemption_timeout <= 1) {
9720                 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9721                 return;
9722         }
9723
9724         preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9725         preemption_timeout *= 1000000;
9726         do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9727         hrtimer_start(&vmx->nested.preemption_timer,
9728                       ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9729 }
9730
9731 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
9732                                                struct vmcs12 *vmcs12)
9733 {
9734         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
9735                 return 0;
9736
9737         if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
9738             !page_address_valid(vcpu, vmcs12->io_bitmap_b))
9739                 return -EINVAL;
9740
9741         return 0;
9742 }
9743
9744 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9745                                                 struct vmcs12 *vmcs12)
9746 {
9747         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9748                 return 0;
9749
9750         if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
9751                 return -EINVAL;
9752
9753         return 0;
9754 }
9755
9756 /*
9757  * Merge L0's and L1's MSR bitmap, return false to indicate that
9758  * we do not use the hardware.
9759  */
9760 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9761                                                struct vmcs12 *vmcs12)
9762 {
9763         int msr;
9764         struct page *page;
9765         unsigned long *msr_bitmap_l1;
9766         unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
9767
9768         /* This shortcut is ok because we support only x2APIC MSRs so far. */
9769         if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9770                 return false;
9771
9772         page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9773         if (!page)
9774                 return false;
9775         msr_bitmap_l1 = (unsigned long *)kmap(page);
9776
9777         memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
9778
9779         if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9780                 if (nested_cpu_has_apic_reg_virt(vmcs12))
9781                         for (msr = 0x800; msr <= 0x8ff; msr++)
9782                                 nested_vmx_disable_intercept_for_msr(
9783                                         msr_bitmap_l1, msr_bitmap_l0,
9784                                         msr, MSR_TYPE_R);
9785
9786                 nested_vmx_disable_intercept_for_msr(
9787                                 msr_bitmap_l1, msr_bitmap_l0,
9788                                 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9789                                 MSR_TYPE_R | MSR_TYPE_W);
9790
9791                 if (nested_cpu_has_vid(vmcs12)) {
9792                         nested_vmx_disable_intercept_for_msr(
9793                                 msr_bitmap_l1, msr_bitmap_l0,
9794                                 APIC_BASE_MSR + (APIC_EOI >> 4),
9795                                 MSR_TYPE_W);
9796                         nested_vmx_disable_intercept_for_msr(
9797                                 msr_bitmap_l1, msr_bitmap_l0,
9798                                 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9799                                 MSR_TYPE_W);
9800                 }
9801         }
9802         kunmap(page);
9803         nested_release_page_clean(page);
9804
9805         return true;
9806 }
9807
9808 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9809                                            struct vmcs12 *vmcs12)
9810 {
9811         if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9812             !nested_cpu_has_apic_reg_virt(vmcs12) &&
9813             !nested_cpu_has_vid(vmcs12) &&
9814             !nested_cpu_has_posted_intr(vmcs12))
9815                 return 0;
9816
9817         /*
9818          * If virtualize x2apic mode is enabled,
9819          * virtualize apic access must be disabled.
9820          */
9821         if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9822             nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9823                 return -EINVAL;
9824
9825         /*
9826          * If virtual interrupt delivery is enabled,
9827          * we must exit on external interrupts.
9828          */
9829         if (nested_cpu_has_vid(vmcs12) &&
9830            !nested_exit_on_intr(vcpu))
9831                 return -EINVAL;
9832
9833         /*
9834          * bits 15:8 should be zero in posted_intr_nv,
9835          * the descriptor address has been already checked
9836          * in nested_get_vmcs12_pages.
9837          */
9838         if (nested_cpu_has_posted_intr(vmcs12) &&
9839            (!nested_cpu_has_vid(vmcs12) ||
9840             !nested_exit_intr_ack_set(vcpu) ||
9841             vmcs12->posted_intr_nv & 0xff00))
9842                 return -EINVAL;
9843
9844         /* tpr shadow is needed by all apicv features. */
9845         if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9846                 return -EINVAL;
9847
9848         return 0;
9849 }
9850
9851 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9852                                        unsigned long count_field,
9853                                        unsigned long addr_field)
9854 {
9855         int maxphyaddr;
9856         u64 count, addr;
9857
9858         if (vmcs12_read_any(vcpu, count_field, &count) ||
9859             vmcs12_read_any(vcpu, addr_field, &addr)) {
9860                 WARN_ON(1);
9861                 return -EINVAL;
9862         }
9863         if (count == 0)
9864                 return 0;
9865         maxphyaddr = cpuid_maxphyaddr(vcpu);
9866         if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9867             (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9868                 pr_debug_ratelimited(
9869                         "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9870                         addr_field, maxphyaddr, count, addr);
9871                 return -EINVAL;
9872         }
9873         return 0;
9874 }
9875
9876 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9877                                                 struct vmcs12 *vmcs12)
9878 {
9879         if (vmcs12->vm_exit_msr_load_count == 0 &&
9880             vmcs12->vm_exit_msr_store_count == 0 &&
9881             vmcs12->vm_entry_msr_load_count == 0)
9882                 return 0; /* Fast path */
9883         if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9884                                         VM_EXIT_MSR_LOAD_ADDR) ||
9885             nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9886                                         VM_EXIT_MSR_STORE_ADDR) ||
9887             nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9888                                         VM_ENTRY_MSR_LOAD_ADDR))
9889                 return -EINVAL;
9890         return 0;
9891 }
9892
9893 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
9894                                          struct vmcs12 *vmcs12)
9895 {
9896         u64 address = vmcs12->pml_address;
9897         int maxphyaddr = cpuid_maxphyaddr(vcpu);
9898
9899         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
9900                 if (!nested_cpu_has_ept(vmcs12) ||
9901                     !IS_ALIGNED(address, 4096)  ||
9902                     address >> maxphyaddr)
9903                         return -EINVAL;
9904         }
9905
9906         return 0;
9907 }
9908
9909 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9910                                        struct vmx_msr_entry *e)
9911 {
9912         /* x2APIC MSR accesses are not allowed */
9913         if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9914                 return -EINVAL;
9915         if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9916             e->index == MSR_IA32_UCODE_REV)
9917                 return -EINVAL;
9918         if (e->reserved != 0)
9919                 return -EINVAL;
9920         return 0;
9921 }
9922
9923 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9924                                      struct vmx_msr_entry *e)
9925 {
9926         if (e->index == MSR_FS_BASE ||
9927             e->index == MSR_GS_BASE ||
9928             e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9929             nested_vmx_msr_check_common(vcpu, e))
9930                 return -EINVAL;
9931         return 0;
9932 }
9933
9934 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9935                                       struct vmx_msr_entry *e)
9936 {
9937         if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9938             nested_vmx_msr_check_common(vcpu, e))
9939                 return -EINVAL;
9940         return 0;
9941 }
9942
9943 /*
9944  * Load guest's/host's msr at nested entry/exit.
9945  * return 0 for success, entry index for failure.
9946  */
9947 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9948 {
9949         u32 i;
9950         struct vmx_msr_entry e;
9951         struct msr_data msr;
9952
9953         msr.host_initiated = false;
9954         for (i = 0; i < count; i++) {
9955                 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9956                                         &e, sizeof(e))) {
9957                         pr_debug_ratelimited(
9958                                 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9959                                 __func__, i, gpa + i * sizeof(e));
9960                         goto fail;
9961                 }
9962                 if (nested_vmx_load_msr_check(vcpu, &e)) {
9963                         pr_debug_ratelimited(
9964                                 "%s check failed (%u, 0x%x, 0x%x)\n",
9965                                 __func__, i, e.index, e.reserved);
9966                         goto fail;
9967                 }
9968                 msr.index = e.index;
9969                 msr.data = e.value;
9970                 if (kvm_set_msr(vcpu, &msr)) {
9971                         pr_debug_ratelimited(
9972                                 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9973                                 __func__, i, e.index, e.value);
9974                         goto fail;
9975                 }
9976         }
9977         return 0;
9978 fail:
9979         return i + 1;
9980 }
9981
9982 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9983 {
9984         u32 i;
9985         struct vmx_msr_entry e;
9986
9987         for (i = 0; i < count; i++) {
9988                 struct msr_data msr_info;
9989                 if (kvm_vcpu_read_guest(vcpu,
9990                                         gpa + i * sizeof(e),
9991                                         &e, 2 * sizeof(u32))) {
9992                         pr_debug_ratelimited(
9993                                 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9994                                 __func__, i, gpa + i * sizeof(e));
9995                         return -EINVAL;
9996                 }
9997                 if (nested_vmx_store_msr_check(vcpu, &e)) {
9998                         pr_debug_ratelimited(
9999                                 "%s check failed (%u, 0x%x, 0x%x)\n",
10000                                 __func__, i, e.index, e.reserved);
10001                         return -EINVAL;
10002                 }
10003                 msr_info.host_initiated = false;
10004                 msr_info.index = e.index;
10005                 if (kvm_get_msr(vcpu, &msr_info)) {
10006                         pr_debug_ratelimited(
10007                                 "%s cannot read MSR (%u, 0x%x)\n",
10008                                 __func__, i, e.index);
10009                         return -EINVAL;
10010                 }
10011                 if (kvm_vcpu_write_guest(vcpu,
10012                                          gpa + i * sizeof(e) +
10013                                              offsetof(struct vmx_msr_entry, value),
10014                                          &msr_info.data, sizeof(msr_info.data))) {
10015                         pr_debug_ratelimited(
10016                                 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10017                                 __func__, i, e.index, msr_info.data);
10018                         return -EINVAL;
10019                 }
10020         }
10021         return 0;
10022 }
10023
10024 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
10025 {
10026         unsigned long invalid_mask;
10027
10028         invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
10029         return (val & invalid_mask) == 0;
10030 }
10031
10032 /*
10033  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
10034  * emulating VM entry into a guest with EPT enabled.
10035  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10036  * is assigned to entry_failure_code on failure.
10037  */
10038 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
10039                                u32 *entry_failure_code)
10040 {
10041         if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
10042                 if (!nested_cr3_valid(vcpu, cr3)) {
10043                         *entry_failure_code = ENTRY_FAIL_DEFAULT;
10044                         return 1;
10045                 }
10046
10047                 /*
10048                  * If PAE paging and EPT are both on, CR3 is not used by the CPU and
10049                  * must not be dereferenced.
10050                  */
10051                 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
10052                     !nested_ept) {
10053                         if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
10054                                 *entry_failure_code = ENTRY_FAIL_PDPTE;
10055                                 return 1;
10056                         }
10057                 }
10058
10059                 vcpu->arch.cr3 = cr3;
10060                 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
10061         }
10062
10063         kvm_mmu_reset_context(vcpu);
10064         return 0;
10065 }
10066
10067 /*
10068  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
10069  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
10070  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
10071  * guest in a way that will both be appropriate to L1's requests, and our
10072  * needs. In addition to modifying the active vmcs (which is vmcs02), this
10073  * function also has additional necessary side-effects, like setting various
10074  * vcpu->arch fields.
10075  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10076  * is assigned to entry_failure_code on failure.
10077  */
10078 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10079                           bool from_vmentry, u32 *entry_failure_code)
10080 {
10081         struct vcpu_vmx *vmx = to_vmx(vcpu);
10082         u32 exec_control, vmcs12_exec_ctrl;
10083
10084         vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10085         vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10086         vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10087         vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10088         vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10089         vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10090         vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10091         vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10092         vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10093         vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10094         vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10095         vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10096         vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10097         vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10098         vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10099         vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10100         vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10101         vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10102         vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10103         vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10104         vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10105         vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10106         vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10107         vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10108         vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10109         vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10110         vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10111         vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10112         vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10113         vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10114         vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10115         vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10116         vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10117         vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10118         vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10119         vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10120
10121         if (from_vmentry &&
10122             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10123                 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10124                 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10125         } else {
10126                 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10127                 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10128         }
10129         if (from_vmentry) {
10130                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10131                              vmcs12->vm_entry_intr_info_field);
10132                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10133                              vmcs12->vm_entry_exception_error_code);
10134                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10135                              vmcs12->vm_entry_instruction_len);
10136                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10137                              vmcs12->guest_interruptibility_info);
10138                 vmx->loaded_vmcs->nmi_known_unmasked =
10139                         !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
10140         } else {
10141                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10142         }
10143         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10144         vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10145         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10146                 vmcs12->guest_pending_dbg_exceptions);
10147         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10148         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10149
10150         if (nested_cpu_has_xsaves(vmcs12))
10151                 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10152         vmcs_write64(VMCS_LINK_POINTER, -1ull);
10153
10154         exec_control = vmcs12->pin_based_vm_exec_control;
10155
10156         /* Preemption timer setting is only taken from vmcs01.  */
10157         exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10158         exec_control |= vmcs_config.pin_based_exec_ctrl;
10159         if (vmx->hv_deadline_tsc == -1)
10160                 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10161
10162         /* Posted interrupts setting is only taken from vmcs12.  */
10163         if (nested_cpu_has_posted_intr(vmcs12)) {
10164                 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10165                 vmx->nested.pi_pending = false;
10166                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
10167         } else {
10168                 exec_control &= ~PIN_BASED_POSTED_INTR;
10169         }
10170
10171         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10172
10173         vmx->nested.preemption_timer_expired = false;
10174         if (nested_cpu_has_preemption_timer(vmcs12))
10175                 vmx_start_preemption_timer(vcpu);
10176
10177         /*
10178          * Whether page-faults are trapped is determined by a combination of
10179          * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10180          * If enable_ept, L0 doesn't care about page faults and we should
10181          * set all of these to L1's desires. However, if !enable_ept, L0 does
10182          * care about (at least some) page faults, and because it is not easy
10183          * (if at all possible?) to merge L0 and L1's desires, we simply ask
10184          * to exit on each and every L2 page fault. This is done by setting
10185          * MASK=MATCH=0 and (see below) EB.PF=1.
10186          * Note that below we don't need special code to set EB.PF beyond the
10187          * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10188          * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10189          * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10190          */
10191         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10192                 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10193         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10194                 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10195
10196         if (cpu_has_secondary_exec_ctrls()) {
10197                 exec_control = vmx_secondary_exec_control(vmx);
10198
10199                 /* Take the following fields only from vmcs12 */
10200                 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10201                                   SECONDARY_EXEC_RDTSCP |
10202                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10203                                   SECONDARY_EXEC_APIC_REGISTER_VIRT);
10204                 if (nested_cpu_has(vmcs12,
10205                                    CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
10206                         vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
10207                                 ~SECONDARY_EXEC_ENABLE_PML;
10208                         exec_control |= vmcs12_exec_ctrl;
10209                 }
10210
10211                 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10212                         vmcs_write64(EOI_EXIT_BITMAP0,
10213                                 vmcs12->eoi_exit_bitmap0);
10214                         vmcs_write64(EOI_EXIT_BITMAP1,
10215                                 vmcs12->eoi_exit_bitmap1);
10216                         vmcs_write64(EOI_EXIT_BITMAP2,
10217                                 vmcs12->eoi_exit_bitmap2);
10218                         vmcs_write64(EOI_EXIT_BITMAP3,
10219                                 vmcs12->eoi_exit_bitmap3);
10220                         vmcs_write16(GUEST_INTR_STATUS,
10221                                 vmcs12->guest_intr_status);
10222                 }
10223
10224                 /*
10225                  * Write an illegal value to APIC_ACCESS_ADDR. Later,
10226                  * nested_get_vmcs12_pages will either fix it up or
10227                  * remove the VM execution control.
10228                  */
10229                 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10230                         vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10231
10232                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10233         }
10234
10235
10236         /*
10237          * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10238          * Some constant fields are set here by vmx_set_constant_host_state().
10239          * Other fields are different per CPU, and will be set later when
10240          * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10241          */
10242         vmx_set_constant_host_state(vmx);
10243
10244         /*
10245          * Set the MSR load/store lists to match L0's settings.
10246          */
10247         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10248         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10249         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10250         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10251         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10252
10253         /*
10254          * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10255          * entry, but only if the current (host) sp changed from the value
10256          * we wrote last (vmx->host_rsp). This cache is no longer relevant
10257          * if we switch vmcs, and rather than hold a separate cache per vmcs,
10258          * here we just force the write to happen on entry.
10259          */
10260         vmx->host_rsp = 0;
10261
10262         exec_control = vmx_exec_control(vmx); /* L0's desires */
10263         exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10264         exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10265         exec_control &= ~CPU_BASED_TPR_SHADOW;
10266         exec_control |= vmcs12->cpu_based_vm_exec_control;
10267
10268         /*
10269          * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10270          * nested_get_vmcs12_pages can't fix it up, the illegal value
10271          * will result in a VM entry failure.
10272          */
10273         if (exec_control & CPU_BASED_TPR_SHADOW) {
10274                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10275                 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10276         }
10277
10278         /*
10279          * Merging of IO bitmap not currently supported.
10280          * Rather, exit every time.
10281          */
10282         exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10283         exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10284
10285         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10286
10287         /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10288          * bitwise-or of what L1 wants to trap for L2, and what we want to
10289          * trap. Note that CR0.TS also needs updating - we do this later.
10290          */
10291         update_exception_bitmap(vcpu);
10292         vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10293         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10294
10295         /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10296          * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10297          * bits are further modified by vmx_set_efer() below.
10298          */
10299         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10300
10301         /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10302          * emulated by vmx_set_efer(), below.
10303          */
10304         vm_entry_controls_init(vmx,
10305                 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10306                         ~VM_ENTRY_IA32E_MODE) |
10307                 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10308
10309         if (from_vmentry &&
10310             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10311                 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10312                 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10313         } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10314                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10315         }
10316
10317         set_cr4_guest_host_mask(vmx);
10318
10319         if (from_vmentry &&
10320             vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10321                 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10322
10323         if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10324                 vmcs_write64(TSC_OFFSET,
10325                         vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10326         else
10327                 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10328         if (kvm_has_tsc_control)
10329                 decache_tsc_multiplier(vmx);
10330
10331         if (enable_vpid) {
10332                 /*
10333                  * There is no direct mapping between vpid02 and vpid12, the
10334                  * vpid02 is per-vCPU for L0 and reused while the value of
10335                  * vpid12 is changed w/ one invvpid during nested vmentry.
10336                  * The vpid12 is allocated by L1 for L2, so it will not
10337                  * influence global bitmap(for vpid01 and vpid02 allocation)
10338                  * even if spawn a lot of nested vCPUs.
10339                  */
10340                 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10341                         vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10342                         if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10343                                 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10344                                 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10345                         }
10346                 } else {
10347                         vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10348                         vmx_flush_tlb(vcpu);
10349                 }
10350
10351         }
10352
10353         if (enable_pml) {
10354                 /*
10355                  * Conceptually we want to copy the PML address and index from
10356                  * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10357                  * since we always flush the log on each vmexit, this happens
10358                  * to be equivalent to simply resetting the fields in vmcs02.
10359                  */
10360                 ASSERT(vmx->pml_pg);
10361                 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
10362                 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
10363         }
10364
10365         if (nested_cpu_has_ept(vmcs12)) {
10366                 if (nested_ept_init_mmu_context(vcpu)) {
10367                         *entry_failure_code = ENTRY_FAIL_DEFAULT;
10368                         return 1;
10369                 }
10370         } else if (nested_cpu_has2(vmcs12,
10371                                    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10372                 vmx_flush_tlb_ept_only(vcpu);
10373         }
10374
10375         /*
10376          * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10377          * bits which we consider mandatory enabled.
10378          * The CR0_READ_SHADOW is what L2 should have expected to read given
10379          * the specifications by L1; It's not enough to take
10380          * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10381          * have more bits than L1 expected.
10382          */
10383         vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10384         vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10385
10386         vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10387         vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10388
10389         if (from_vmentry &&
10390             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10391                 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10392         else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10393                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10394         else
10395                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10396         /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10397         vmx_set_efer(vcpu, vcpu->arch.efer);
10398
10399         /* Shadow page tables on either EPT or shadow page tables. */
10400         if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10401                                 entry_failure_code))
10402                 return 1;
10403
10404         if (!enable_ept)
10405                 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10406
10407         /*
10408          * L1 may access the L2's PDPTR, so save them to construct vmcs12
10409          */
10410         if (enable_ept) {
10411                 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10412                 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10413                 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10414                 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10415         }
10416
10417         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10418         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10419         return 0;
10420 }
10421
10422 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10423 {
10424         struct vcpu_vmx *vmx = to_vmx(vcpu);
10425
10426         if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10427             vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10428                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10429
10430         if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
10431                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10432
10433         if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10434                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10435
10436         if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10437                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10438
10439         if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10440                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10441
10442         if (nested_vmx_check_pml_controls(vcpu, vmcs12))
10443                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10444
10445         if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10446                                 vmx->nested.nested_vmx_procbased_ctls_low,
10447                                 vmx->nested.nested_vmx_procbased_ctls_high) ||
10448             (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
10449              !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10450                                  vmx->nested.nested_vmx_secondary_ctls_low,
10451                                  vmx->nested.nested_vmx_secondary_ctls_high)) ||
10452             !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10453                                 vmx->nested.nested_vmx_pinbased_ctls_low,
10454                                 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10455             !vmx_control_verify(vmcs12->vm_exit_controls,
10456                                 vmx->nested.nested_vmx_exit_ctls_low,
10457                                 vmx->nested.nested_vmx_exit_ctls_high) ||
10458             !vmx_control_verify(vmcs12->vm_entry_controls,
10459                                 vmx->nested.nested_vmx_entry_ctls_low,
10460                                 vmx->nested.nested_vmx_entry_ctls_high))
10461                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10462
10463         if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
10464                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10465
10466         if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10467             !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10468             !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10469                 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10470
10471         return 0;
10472 }
10473
10474 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10475                                   u32 *exit_qual)
10476 {
10477         bool ia32e;
10478
10479         *exit_qual = ENTRY_FAIL_DEFAULT;
10480
10481         if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10482             !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10483                 return 1;
10484
10485         if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10486             vmcs12->vmcs_link_pointer != -1ull) {
10487                 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10488                 return 1;
10489         }
10490
10491         /*
10492          * If the load IA32_EFER VM-entry control is 1, the following checks
10493          * are performed on the field for the IA32_EFER MSR:
10494          * - Bits reserved in the IA32_EFER MSR must be 0.
10495          * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10496          *   the IA-32e mode guest VM-exit control. It must also be identical
10497          *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10498          *   CR0.PG) is 1.
10499          */
10500         if (to_vmx(vcpu)->nested.nested_run_pending &&
10501             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10502                 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10503                 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10504                     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10505                     ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10506                      ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10507                         return 1;
10508         }
10509
10510         /*
10511          * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10512          * IA32_EFER MSR must be 0 in the field for that register. In addition,
10513          * the values of the LMA and LME bits in the field must each be that of
10514          * the host address-space size VM-exit control.
10515          */
10516         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10517                 ia32e = (vmcs12->vm_exit_controls &
10518                          VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10519                 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10520                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10521                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10522                         return 1;
10523         }
10524
10525         return 0;
10526 }
10527
10528 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10529 {
10530         struct vcpu_vmx *vmx = to_vmx(vcpu);
10531         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10532         struct loaded_vmcs *vmcs02;
10533         u32 msr_entry_idx;
10534         u32 exit_qual;
10535
10536         vmcs02 = nested_get_current_vmcs02(vmx);
10537         if (!vmcs02)
10538                 return -ENOMEM;
10539
10540         enter_guest_mode(vcpu);
10541
10542         if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10543                 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10544
10545         vmx_switch_vmcs(vcpu, vmcs02);
10546         vmx_segment_cache_clear(vmx);
10547
10548         if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10549                 leave_guest_mode(vcpu);
10550                 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10551                 nested_vmx_entry_failure(vcpu, vmcs12,
10552                                          EXIT_REASON_INVALID_STATE, exit_qual);
10553                 return 1;
10554         }
10555
10556         nested_get_vmcs12_pages(vcpu, vmcs12);
10557
10558         msr_entry_idx = nested_vmx_load_msr(vcpu,
10559                                             vmcs12->vm_entry_msr_load_addr,
10560                                             vmcs12->vm_entry_msr_load_count);
10561         if (msr_entry_idx) {
10562                 leave_guest_mode(vcpu);
10563                 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10564                 nested_vmx_entry_failure(vcpu, vmcs12,
10565                                 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10566                 return 1;
10567         }
10568
10569         /*
10570          * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10571          * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10572          * returned as far as L1 is concerned. It will only return (and set
10573          * the success flag) when L2 exits (see nested_vmx_vmexit()).
10574          */
10575         return 0;
10576 }
10577
10578 /*
10579  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10580  * for running an L2 nested guest.
10581  */
10582 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10583 {
10584         struct vmcs12 *vmcs12;
10585         struct vcpu_vmx *vmx = to_vmx(vcpu);
10586         u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
10587         u32 exit_qual;
10588         int ret;
10589
10590         if (!nested_vmx_check_permission(vcpu))
10591                 return 1;
10592
10593         if (!nested_vmx_check_vmcs12(vcpu))
10594                 goto out;
10595
10596         vmcs12 = get_vmcs12(vcpu);
10597
10598         if (enable_shadow_vmcs)
10599                 copy_shadow_to_vmcs12(vmx);
10600
10601         /*
10602          * The nested entry process starts with enforcing various prerequisites
10603          * on vmcs12 as required by the Intel SDM, and act appropriately when
10604          * they fail: As the SDM explains, some conditions should cause the
10605          * instruction to fail, while others will cause the instruction to seem
10606          * to succeed, but return an EXIT_REASON_INVALID_STATE.
10607          * To speed up the normal (success) code path, we should avoid checking
10608          * for misconfigurations which will anyway be caught by the processor
10609          * when using the merged vmcs02.
10610          */
10611         if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
10612                 nested_vmx_failValid(vcpu,
10613                                      VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
10614                 goto out;
10615         }
10616
10617         if (vmcs12->launch_state == launch) {
10618                 nested_vmx_failValid(vcpu,
10619                         launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10620                                : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10621                 goto out;
10622         }
10623
10624         ret = check_vmentry_prereqs(vcpu, vmcs12);
10625         if (ret) {
10626                 nested_vmx_failValid(vcpu, ret);
10627                 goto out;
10628         }
10629
10630         /*
10631          * After this point, the trap flag no longer triggers a singlestep trap
10632          * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10633          * This is not 100% correct; for performance reasons, we delegate most
10634          * of the checks on host state to the processor.  If those fail,
10635          * the singlestep trap is missed.
10636          */
10637         skip_emulated_instruction(vcpu);
10638
10639         ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10640         if (ret) {
10641                 nested_vmx_entry_failure(vcpu, vmcs12,
10642                                          EXIT_REASON_INVALID_STATE, exit_qual);
10643                 return 1;
10644         }
10645
10646         /*
10647          * We're finally done with prerequisite checking, and can start with
10648          * the nested entry.
10649          */
10650
10651         ret = enter_vmx_non_root_mode(vcpu, true);
10652         if (ret)
10653                 return ret;
10654
10655         if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10656                 return kvm_vcpu_halt(vcpu);
10657
10658         vmx->nested.nested_run_pending = 1;
10659
10660         return 1;
10661
10662 out:
10663         return kvm_skip_emulated_instruction(vcpu);
10664 }
10665
10666 /*
10667  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10668  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10669  * This function returns the new value we should put in vmcs12.guest_cr0.
10670  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10671  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10672  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10673  *     didn't trap the bit, because if L1 did, so would L0).
10674  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10675  *     been modified by L2, and L1 knows it. So just leave the old value of
10676  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10677  *     isn't relevant, because if L0 traps this bit it can set it to anything.
10678  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10679  *     changed these bits, and therefore they need to be updated, but L0
10680  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10681  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10682  */
10683 static inline unsigned long
10684 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10685 {
10686         return
10687         /*1*/   (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10688         /*2*/   (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10689         /*3*/   (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10690                         vcpu->arch.cr0_guest_owned_bits));
10691 }
10692
10693 static inline unsigned long
10694 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10695 {
10696         return
10697         /*1*/   (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10698         /*2*/   (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10699         /*3*/   (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10700                         vcpu->arch.cr4_guest_owned_bits));
10701 }
10702
10703 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10704                                        struct vmcs12 *vmcs12)
10705 {
10706         u32 idt_vectoring;
10707         unsigned int nr;
10708
10709         if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10710                 nr = vcpu->arch.exception.nr;
10711                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10712
10713                 if (kvm_exception_is_soft(nr)) {
10714                         vmcs12->vm_exit_instruction_len =
10715                                 vcpu->arch.event_exit_inst_len;
10716                         idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10717                 } else
10718                         idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10719
10720                 if (vcpu->arch.exception.has_error_code) {
10721                         idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10722                         vmcs12->idt_vectoring_error_code =
10723                                 vcpu->arch.exception.error_code;
10724                 }
10725
10726                 vmcs12->idt_vectoring_info_field = idt_vectoring;
10727         } else if (vcpu->arch.nmi_injected) {
10728                 vmcs12->idt_vectoring_info_field =
10729                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10730         } else if (vcpu->arch.interrupt.pending) {
10731                 nr = vcpu->arch.interrupt.nr;
10732                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10733
10734                 if (vcpu->arch.interrupt.soft) {
10735                         idt_vectoring |= INTR_TYPE_SOFT_INTR;
10736                         vmcs12->vm_entry_instruction_len =
10737                                 vcpu->arch.event_exit_inst_len;
10738                 } else
10739                         idt_vectoring |= INTR_TYPE_EXT_INTR;
10740
10741                 vmcs12->idt_vectoring_info_field = idt_vectoring;
10742         }
10743 }
10744
10745 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10746 {
10747         struct vcpu_vmx *vmx = to_vmx(vcpu);
10748
10749         if (vcpu->arch.exception.pending ||
10750                 vcpu->arch.nmi_injected ||
10751                 vcpu->arch.interrupt.pending)
10752                 return -EBUSY;
10753
10754         if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10755             vmx->nested.preemption_timer_expired) {
10756                 if (vmx->nested.nested_run_pending)
10757                         return -EBUSY;
10758                 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10759                 return 0;
10760         }
10761
10762         if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10763                 if (vmx->nested.nested_run_pending)
10764                         return -EBUSY;
10765                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10766                                   NMI_VECTOR | INTR_TYPE_NMI_INTR |
10767                                   INTR_INFO_VALID_MASK, 0);
10768                 /*
10769                  * The NMI-triggered VM exit counts as injection:
10770                  * clear this one and block further NMIs.
10771                  */
10772                 vcpu->arch.nmi_pending = 0;
10773                 vmx_set_nmi_mask(vcpu, true);
10774                 return 0;
10775         }
10776
10777         if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10778             nested_exit_on_intr(vcpu)) {
10779                 if (vmx->nested.nested_run_pending)
10780                         return -EBUSY;
10781                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10782                 return 0;
10783         }
10784
10785         vmx_complete_nested_posted_interrupt(vcpu);
10786         return 0;
10787 }
10788
10789 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10790 {
10791         ktime_t remaining =
10792                 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10793         u64 value;
10794
10795         if (ktime_to_ns(remaining) <= 0)
10796                 return 0;
10797
10798         value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10799         do_div(value, 1000000);
10800         return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10801 }
10802
10803 /*
10804  * Update the guest state fields of vmcs12 to reflect changes that
10805  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
10806  * VM-entry controls is also updated, since this is really a guest
10807  * state bit.)
10808  */
10809 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10810 {
10811         vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10812         vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10813
10814         vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10815         vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10816         vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10817
10818         vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10819         vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10820         vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10821         vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10822         vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10823         vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10824         vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10825         vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10826         vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10827         vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10828         vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10829         vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10830         vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10831         vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10832         vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10833         vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10834         vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10835         vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10836         vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10837         vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10838         vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10839         vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10840         vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10841         vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10842         vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10843         vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10844         vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10845         vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10846         vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10847         vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10848         vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10849         vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10850         vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10851         vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10852         vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10853         vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10854
10855         vmcs12->guest_interruptibility_info =
10856                 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10857         vmcs12->guest_pending_dbg_exceptions =
10858                 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10859         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10860                 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10861         else
10862                 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10863
10864         if (nested_cpu_has_preemption_timer(vmcs12)) {
10865                 if (vmcs12->vm_exit_controls &
10866                     VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10867                         vmcs12->vmx_preemption_timer_value =
10868                                 vmx_get_preemption_timer_value(vcpu);
10869                 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10870         }
10871
10872         /*
10873          * In some cases (usually, nested EPT), L2 is allowed to change its
10874          * own CR3 without exiting. If it has changed it, we must keep it.
10875          * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10876          * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10877          *
10878          * Additionally, restore L2's PDPTR to vmcs12.
10879          */
10880         if (enable_ept) {
10881                 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10882                 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10883                 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10884                 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10885                 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10886         }
10887
10888         vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
10889
10890         if (nested_cpu_has_vid(vmcs12))
10891                 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10892
10893         vmcs12->vm_entry_controls =
10894                 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10895                 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10896
10897         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10898                 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10899                 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10900         }
10901
10902         /* TODO: These cannot have changed unless we have MSR bitmaps and
10903          * the relevant bit asks not to trap the change */
10904         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10905                 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10906         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10907                 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10908         vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10909         vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10910         vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10911         if (kvm_mpx_supported())
10912                 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10913 }
10914
10915 /*
10916  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10917  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10918  * and this function updates it to reflect the changes to the guest state while
10919  * L2 was running (and perhaps made some exits which were handled directly by L0
10920  * without going back to L1), and to reflect the exit reason.
10921  * Note that we do not have to copy here all VMCS fields, just those that
10922  * could have changed by the L2 guest or the exit - i.e., the guest-state and
10923  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10924  * which already writes to vmcs12 directly.
10925  */
10926 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10927                            u32 exit_reason, u32 exit_intr_info,
10928                            unsigned long exit_qualification)
10929 {
10930         /* update guest state fields: */
10931         sync_vmcs12(vcpu, vmcs12);
10932
10933         /* update exit information fields: */
10934
10935         vmcs12->vm_exit_reason = exit_reason;
10936         vmcs12->exit_qualification = exit_qualification;
10937         vmcs12->vm_exit_intr_info = exit_intr_info;
10938
10939         vmcs12->idt_vectoring_info_field = 0;
10940         vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10941         vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10942
10943         if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10944                 vmcs12->launch_state = 1;
10945
10946                 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10947                  * instead of reading the real value. */
10948                 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10949
10950                 /*
10951                  * Transfer the event that L0 or L1 may wanted to inject into
10952                  * L2 to IDT_VECTORING_INFO_FIELD.
10953                  */
10954                 vmcs12_save_pending_event(vcpu, vmcs12);
10955         }
10956
10957         /*
10958          * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10959          * preserved above and would only end up incorrectly in L1.
10960          */
10961         vcpu->arch.nmi_injected = false;
10962         kvm_clear_exception_queue(vcpu);
10963         kvm_clear_interrupt_queue(vcpu);
10964 }
10965
10966 /*
10967  * A part of what we need to when the nested L2 guest exits and we want to
10968  * run its L1 parent, is to reset L1's guest state to the host state specified
10969  * in vmcs12.
10970  * This function is to be called not only on normal nested exit, but also on
10971  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10972  * Failures During or After Loading Guest State").
10973  * This function should be called when the active VMCS is L1's (vmcs01).
10974  */
10975 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10976                                    struct vmcs12 *vmcs12)
10977 {
10978         struct kvm_segment seg;
10979         u32 entry_failure_code;
10980
10981         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10982                 vcpu->arch.efer = vmcs12->host_ia32_efer;
10983         else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10984                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10985         else
10986                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10987         vmx_set_efer(vcpu, vcpu->arch.efer);
10988
10989         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10990         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10991         vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10992         /*
10993          * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10994          * actually changed, because vmx_set_cr0 refers to efer set above.
10995          *
10996          * CR0_GUEST_HOST_MASK is already set in the original vmcs01
10997          * (KVM doesn't change it);
10998          */
10999         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
11000         vmx_set_cr0(vcpu, vmcs12->host_cr0);
11001
11002         /* Same as above - no reason to call set_cr4_guest_host_mask().  */
11003         vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
11004         kvm_set_cr4(vcpu, vmcs12->host_cr4);
11005
11006         nested_ept_uninit_mmu_context(vcpu);
11007
11008         /*
11009          * Only PDPTE load can fail as the value of cr3 was checked on entry and
11010          * couldn't have changed.
11011          */
11012         if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
11013                 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
11014
11015         if (!enable_ept)
11016                 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
11017
11018         if (enable_vpid) {
11019                 /*
11020                  * Trivially support vpid by letting L2s share their parent
11021                  * L1's vpid. TODO: move to a more elaborate solution, giving
11022                  * each L2 its own vpid and exposing the vpid feature to L1.
11023                  */
11024                 vmx_flush_tlb(vcpu);
11025         }
11026         /* Restore posted intr vector. */
11027         if (nested_cpu_has_posted_intr(vmcs12))
11028                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
11029
11030         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
11031         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
11032         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
11033         vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
11034         vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
11035
11036         /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
11037         if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
11038                 vmcs_write64(GUEST_BNDCFGS, 0);
11039
11040         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
11041                 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
11042                 vcpu->arch.pat = vmcs12->host_ia32_pat;
11043         }
11044         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
11045                 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
11046                         vmcs12->host_ia32_perf_global_ctrl);
11047
11048         /* Set L1 segment info according to Intel SDM
11049             27.5.2 Loading Host Segment and Descriptor-Table Registers */
11050         seg = (struct kvm_segment) {
11051                 .base = 0,
11052                 .limit = 0xFFFFFFFF,
11053                 .selector = vmcs12->host_cs_selector,
11054                 .type = 11,
11055                 .present = 1,
11056                 .s = 1,
11057                 .g = 1
11058         };
11059         if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11060                 seg.l = 1;
11061         else
11062                 seg.db = 1;
11063         vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
11064         seg = (struct kvm_segment) {
11065                 .base = 0,
11066                 .limit = 0xFFFFFFFF,
11067                 .type = 3,
11068                 .present = 1,
11069                 .s = 1,
11070                 .db = 1,
11071                 .g = 1
11072         };
11073         seg.selector = vmcs12->host_ds_selector;
11074         vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
11075         seg.selector = vmcs12->host_es_selector;
11076         vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
11077         seg.selector = vmcs12->host_ss_selector;
11078         vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
11079         seg.selector = vmcs12->host_fs_selector;
11080         seg.base = vmcs12->host_fs_base;
11081         vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
11082         seg.selector = vmcs12->host_gs_selector;
11083         seg.base = vmcs12->host_gs_base;
11084         vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
11085         seg = (struct kvm_segment) {
11086                 .base = vmcs12->host_tr_base,
11087                 .limit = 0x67,
11088                 .selector = vmcs12->host_tr_selector,
11089                 .type = 11,
11090                 .present = 1
11091         };
11092         vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
11093
11094         kvm_set_dr(vcpu, 7, 0x400);
11095         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11096
11097         if (cpu_has_vmx_msr_bitmap())
11098                 vmx_set_msr_bitmap(vcpu);
11099
11100         if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11101                                 vmcs12->vm_exit_msr_load_count))
11102                 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11103 }
11104
11105 /*
11106  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11107  * and modify vmcs12 to make it see what it would expect to see there if
11108  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11109  */
11110 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11111                               u32 exit_intr_info,
11112                               unsigned long exit_qualification)
11113 {
11114         struct vcpu_vmx *vmx = to_vmx(vcpu);
11115         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11116         u32 vm_inst_error = 0;
11117
11118         /* trying to cancel vmlaunch/vmresume is a bug */
11119         WARN_ON_ONCE(vmx->nested.nested_run_pending);
11120
11121         leave_guest_mode(vcpu);
11122         prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11123                        exit_qualification);
11124
11125         if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11126                                  vmcs12->vm_exit_msr_store_count))
11127                 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11128
11129         if (unlikely(vmx->fail))
11130                 vm_inst_error = vmcs_read32(VM_INSTRUCTION_ERROR);
11131
11132         vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11133
11134         /*
11135          * TODO: SDM says that with acknowledge interrupt on exit, bit 31 of
11136          * the VM-exit interrupt information (valid interrupt) is always set to
11137          * 1 on EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't need
11138          * kvm_cpu_has_interrupt().  See the commit message for details.
11139          */
11140         if (nested_exit_intr_ack_set(vcpu) &&
11141             exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
11142             kvm_cpu_has_interrupt(vcpu)) {
11143                 int irq = kvm_cpu_get_interrupt(vcpu);
11144                 WARN_ON(irq < 0);
11145                 vmcs12->vm_exit_intr_info = irq |
11146                         INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11147         }
11148
11149         trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11150                                        vmcs12->exit_qualification,
11151                                        vmcs12->idt_vectoring_info_field,
11152                                        vmcs12->vm_exit_intr_info,
11153                                        vmcs12->vm_exit_intr_error_code,
11154                                        KVM_ISA_VMX);
11155
11156         vm_entry_controls_reset_shadow(vmx);
11157         vm_exit_controls_reset_shadow(vmx);
11158         vmx_segment_cache_clear(vmx);
11159
11160         /* if no vmcs02 cache requested, remove the one we used */
11161         if (VMCS02_POOL_SIZE == 0)
11162                 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
11163
11164         load_vmcs12_host_state(vcpu, vmcs12);
11165
11166         /* Update any VMCS fields that might have changed while L2 ran */
11167         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11168         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11169         vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11170         if (vmx->hv_deadline_tsc == -1)
11171                 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11172                                 PIN_BASED_VMX_PREEMPTION_TIMER);
11173         else
11174                 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11175                               PIN_BASED_VMX_PREEMPTION_TIMER);
11176         if (kvm_has_tsc_control)
11177                 decache_tsc_multiplier(vmx);
11178
11179         if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11180                 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11181                 vmx_set_virtual_x2apic_mode(vcpu,
11182                                 vcpu->arch.apic_base & X2APIC_ENABLE);
11183         } else if (!nested_cpu_has_ept(vmcs12) &&
11184                    nested_cpu_has2(vmcs12,
11185                                    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11186                 vmx_flush_tlb_ept_only(vcpu);
11187         }
11188
11189         /* This is needed for same reason as it was needed in prepare_vmcs02 */
11190         vmx->host_rsp = 0;
11191
11192         /* Unpin physical memory we referred to in vmcs02 */
11193         if (vmx->nested.apic_access_page) {
11194                 nested_release_page(vmx->nested.apic_access_page);
11195                 vmx->nested.apic_access_page = NULL;
11196         }
11197         if (vmx->nested.virtual_apic_page) {
11198                 nested_release_page(vmx->nested.virtual_apic_page);
11199                 vmx->nested.virtual_apic_page = NULL;
11200         }
11201         if (vmx->nested.pi_desc_page) {
11202                 kunmap(vmx->nested.pi_desc_page);
11203                 nested_release_page(vmx->nested.pi_desc_page);
11204                 vmx->nested.pi_desc_page = NULL;
11205                 vmx->nested.pi_desc = NULL;
11206         }
11207
11208         /*
11209          * We are now running in L2, mmu_notifier will force to reload the
11210          * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11211          */
11212         kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11213
11214         /*
11215          * Exiting from L2 to L1, we're now back to L1 which thinks it just
11216          * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
11217          * success or failure flag accordingly.
11218          */
11219         if (unlikely(vmx->fail)) {
11220                 vmx->fail = 0;
11221                 nested_vmx_failValid(vcpu, vm_inst_error);
11222         } else
11223                 nested_vmx_succeed(vcpu);
11224         if (enable_shadow_vmcs)
11225                 vmx->nested.sync_shadow_vmcs = true;
11226
11227         /* in case we halted in L2 */
11228         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11229 }
11230
11231 /*
11232  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11233  */
11234 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11235 {
11236         if (is_guest_mode(vcpu)) {
11237                 to_vmx(vcpu)->nested.nested_run_pending = 0;
11238                 nested_vmx_vmexit(vcpu, -1, 0, 0);
11239         }
11240         free_nested(to_vmx(vcpu));
11241 }
11242
11243 /*
11244  * L1's failure to enter L2 is a subset of a normal exit, as explained in
11245  * 23.7 "VM-entry failures during or after loading guest state" (this also
11246  * lists the acceptable exit-reason and exit-qualification parameters).
11247  * It should only be called before L2 actually succeeded to run, and when
11248  * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11249  */
11250 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11251                         struct vmcs12 *vmcs12,
11252                         u32 reason, unsigned long qualification)
11253 {
11254         load_vmcs12_host_state(vcpu, vmcs12);
11255         vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11256         vmcs12->exit_qualification = qualification;
11257         nested_vmx_succeed(vcpu);
11258         if (enable_shadow_vmcs)
11259                 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11260 }
11261
11262 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11263                                struct x86_instruction_info *info,
11264                                enum x86_intercept_stage stage)
11265 {
11266         return X86EMUL_CONTINUE;
11267 }
11268
11269 #ifdef CONFIG_X86_64
11270 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11271 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11272                                   u64 divisor, u64 *result)
11273 {
11274         u64 low = a << shift, high = a >> (64 - shift);
11275
11276         /* To avoid the overflow on divq */
11277         if (high >= divisor)
11278                 return 1;
11279
11280         /* Low hold the result, high hold rem which is discarded */
11281         asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11282             "rm" (divisor), "0" (low), "1" (high));
11283         *result = low;
11284
11285         return 0;
11286 }
11287
11288 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11289 {
11290         struct vcpu_vmx *vmx = to_vmx(vcpu);
11291         u64 tscl = rdtsc();
11292         u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11293         u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11294
11295         /* Convert to host delta tsc if tsc scaling is enabled */
11296         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11297                         u64_shl_div_u64(delta_tsc,
11298                                 kvm_tsc_scaling_ratio_frac_bits,
11299                                 vcpu->arch.tsc_scaling_ratio,
11300                                 &delta_tsc))
11301                 return -ERANGE;
11302
11303         /*
11304          * If the delta tsc can't fit in the 32 bit after the multi shift,
11305          * we can't use the preemption timer.
11306          * It's possible that it fits on later vmentries, but checking
11307          * on every vmentry is costly so we just use an hrtimer.
11308          */
11309         if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11310                 return -ERANGE;
11311
11312         vmx->hv_deadline_tsc = tscl + delta_tsc;
11313         vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11314                         PIN_BASED_VMX_PREEMPTION_TIMER);
11315
11316         return delta_tsc == 0;
11317 }
11318
11319 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11320 {
11321         struct vcpu_vmx *vmx = to_vmx(vcpu);
11322         vmx->hv_deadline_tsc = -1;
11323         vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11324                         PIN_BASED_VMX_PREEMPTION_TIMER);
11325 }
11326 #endif
11327
11328 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11329 {
11330         if (ple_gap)
11331                 shrink_ple_window(vcpu);
11332 }
11333
11334 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11335                                      struct kvm_memory_slot *slot)
11336 {
11337         kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11338         kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11339 }
11340
11341 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11342                                        struct kvm_memory_slot *slot)
11343 {
11344         kvm_mmu_slot_set_dirty(kvm, slot);
11345 }
11346
11347 static void vmx_flush_log_dirty(struct kvm *kvm)
11348 {
11349         kvm_flush_pml_buffers(kvm);
11350 }
11351
11352 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
11353 {
11354         struct vmcs12 *vmcs12;
11355         struct vcpu_vmx *vmx = to_vmx(vcpu);
11356         gpa_t gpa;
11357         struct page *page = NULL;
11358         u64 *pml_address;
11359
11360         if (is_guest_mode(vcpu)) {
11361                 WARN_ON_ONCE(vmx->nested.pml_full);
11362
11363                 /*
11364                  * Check if PML is enabled for the nested guest.
11365                  * Whether eptp bit 6 is set is already checked
11366                  * as part of A/D emulation.
11367                  */
11368                 vmcs12 = get_vmcs12(vcpu);
11369                 if (!nested_cpu_has_pml(vmcs12))
11370                         return 0;
11371
11372                 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
11373                         vmx->nested.pml_full = true;
11374                         return 1;
11375                 }
11376
11377                 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
11378
11379                 page = nested_get_page(vcpu, vmcs12->pml_address);
11380                 if (!page)
11381                         return 0;
11382
11383                 pml_address = kmap(page);
11384                 pml_address[vmcs12->guest_pml_index--] = gpa;
11385                 kunmap(page);
11386                 nested_release_page_clean(page);
11387         }
11388
11389         return 0;
11390 }
11391
11392 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11393                                            struct kvm_memory_slot *memslot,
11394                                            gfn_t offset, unsigned long mask)
11395 {
11396         kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11397 }
11398
11399 /*
11400  * This routine does the following things for vCPU which is going
11401  * to be blocked if VT-d PI is enabled.
11402  * - Store the vCPU to the wakeup list, so when interrupts happen
11403  *   we can find the right vCPU to wake up.
11404  * - Change the Posted-interrupt descriptor as below:
11405  *      'NDST' <-- vcpu->pre_pcpu
11406  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11407  * - If 'ON' is set during this process, which means at least one
11408  *   interrupt is posted for this vCPU, we cannot block it, in
11409  *   this case, return 1, otherwise, return 0.
11410  *
11411  */
11412 static int pi_pre_block(struct kvm_vcpu *vcpu)
11413 {
11414         unsigned long flags;
11415         unsigned int dest;
11416         struct pi_desc old, new;
11417         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11418
11419         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11420                 !irq_remapping_cap(IRQ_POSTING_CAP)  ||
11421                 !kvm_vcpu_apicv_active(vcpu))
11422                 return 0;
11423
11424         vcpu->pre_pcpu = vcpu->cpu;
11425         spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11426                           vcpu->pre_pcpu), flags);
11427         list_add_tail(&vcpu->blocked_vcpu_list,
11428                       &per_cpu(blocked_vcpu_on_cpu,
11429                       vcpu->pre_pcpu));
11430         spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
11431                                vcpu->pre_pcpu), flags);
11432
11433         do {
11434                 old.control = new.control = pi_desc->control;
11435
11436                 /*
11437                  * We should not block the vCPU if
11438                  * an interrupt is posted for it.
11439                  */
11440                 if (pi_test_on(pi_desc) == 1) {
11441                         spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11442                                           vcpu->pre_pcpu), flags);
11443                         list_del(&vcpu->blocked_vcpu_list);
11444                         spin_unlock_irqrestore(
11445                                         &per_cpu(blocked_vcpu_on_cpu_lock,
11446                                         vcpu->pre_pcpu), flags);
11447                         vcpu->pre_pcpu = -1;
11448
11449                         return 1;
11450                 }
11451
11452                 WARN((pi_desc->sn == 1),
11453                      "Warning: SN field of posted-interrupts "
11454                      "is set before blocking\n");
11455
11456                 /*
11457                  * Since vCPU can be preempted during this process,
11458                  * vcpu->cpu could be different with pre_pcpu, we
11459                  * need to set pre_pcpu as the destination of wakeup
11460                  * notification event, then we can find the right vCPU
11461                  * to wakeup in wakeup handler if interrupts happen
11462                  * when the vCPU is in blocked state.
11463                  */
11464                 dest = cpu_physical_id(vcpu->pre_pcpu);
11465
11466                 if (x2apic_enabled())
11467                         new.ndst = dest;
11468                 else
11469                         new.ndst = (dest << 8) & 0xFF00;
11470
11471                 /* set 'NV' to 'wakeup vector' */
11472                 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11473         } while (cmpxchg(&pi_desc->control, old.control,
11474                         new.control) != old.control);
11475
11476         return 0;
11477 }
11478
11479 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11480 {
11481         if (pi_pre_block(vcpu))
11482                 return 1;
11483
11484         if (kvm_lapic_hv_timer_in_use(vcpu))
11485                 kvm_lapic_switch_to_sw_timer(vcpu);
11486
11487         return 0;
11488 }
11489
11490 static void pi_post_block(struct kvm_vcpu *vcpu)
11491 {
11492         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11493         struct pi_desc old, new;
11494         unsigned int dest;
11495         unsigned long flags;
11496
11497         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11498                 !irq_remapping_cap(IRQ_POSTING_CAP)  ||
11499                 !kvm_vcpu_apicv_active(vcpu))
11500                 return;
11501
11502         do {
11503                 old.control = new.control = pi_desc->control;
11504
11505                 dest = cpu_physical_id(vcpu->cpu);
11506
11507                 if (x2apic_enabled())
11508                         new.ndst = dest;
11509                 else
11510                         new.ndst = (dest << 8) & 0xFF00;
11511
11512                 /* Allow posting non-urgent interrupts */
11513                 new.sn = 0;
11514
11515                 /* set 'NV' to 'notification vector' */
11516                 new.nv = POSTED_INTR_VECTOR;
11517         } while (cmpxchg(&pi_desc->control, old.control,
11518                         new.control) != old.control);
11519
11520         if(vcpu->pre_pcpu != -1) {
11521                 spin_lock_irqsave(
11522                         &per_cpu(blocked_vcpu_on_cpu_lock,
11523                         vcpu->pre_pcpu), flags);
11524                 list_del(&vcpu->blocked_vcpu_list);
11525                 spin_unlock_irqrestore(
11526                         &per_cpu(blocked_vcpu_on_cpu_lock,
11527                         vcpu->pre_pcpu), flags);
11528                 vcpu->pre_pcpu = -1;
11529         }
11530 }
11531
11532 static void vmx_post_block(struct kvm_vcpu *vcpu)
11533 {
11534         if (kvm_x86_ops->set_hv_timer)
11535                 kvm_lapic_switch_to_hv_timer(vcpu);
11536
11537         pi_post_block(vcpu);
11538 }
11539
11540 /*
11541  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11542  *
11543  * @kvm: kvm
11544  * @host_irq: host irq of the interrupt
11545  * @guest_irq: gsi of the interrupt
11546  * @set: set or unset PI
11547  * returns 0 on success, < 0 on failure
11548  */
11549 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11550                               uint32_t guest_irq, bool set)
11551 {
11552         struct kvm_kernel_irq_routing_entry *e;
11553         struct kvm_irq_routing_table *irq_rt;
11554         struct kvm_lapic_irq irq;
11555         struct kvm_vcpu *vcpu;
11556         struct vcpu_data vcpu_info;
11557         int idx, ret = -EINVAL;
11558
11559         if (!kvm_arch_has_assigned_device(kvm) ||
11560                 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11561                 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11562                 return 0;
11563
11564         idx = srcu_read_lock(&kvm->irq_srcu);
11565         irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11566         BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11567
11568         hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11569                 if (e->type != KVM_IRQ_ROUTING_MSI)
11570                         continue;
11571                 /*
11572                  * VT-d PI cannot support posting multicast/broadcast
11573                  * interrupts to a vCPU, we still use interrupt remapping
11574                  * for these kind of interrupts.
11575                  *
11576                  * For lowest-priority interrupts, we only support
11577                  * those with single CPU as the destination, e.g. user
11578                  * configures the interrupts via /proc/irq or uses
11579                  * irqbalance to make the interrupts single-CPU.
11580                  *
11581                  * We will support full lowest-priority interrupt later.
11582                  */
11583
11584                 kvm_set_msi_irq(kvm, e, &irq);
11585                 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11586                         /*
11587                          * Make sure the IRTE is in remapped mode if
11588                          * we don't handle it in posted mode.
11589                          */
11590                         ret = irq_set_vcpu_affinity(host_irq, NULL);
11591                         if (ret < 0) {
11592                                 printk(KERN_INFO
11593                                    "failed to back to remapped mode, irq: %u\n",
11594                                    host_irq);
11595                                 goto out;
11596                         }
11597
11598                         continue;
11599                 }
11600
11601                 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11602                 vcpu_info.vector = irq.vector;
11603
11604                 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11605                                 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11606
11607                 if (set)
11608                         ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11609                 else {
11610                         /* suppress notification event before unposting */
11611                         pi_set_sn(vcpu_to_pi_desc(vcpu));
11612                         ret = irq_set_vcpu_affinity(host_irq, NULL);
11613                         pi_clear_sn(vcpu_to_pi_desc(vcpu));
11614                 }
11615
11616                 if (ret < 0) {
11617                         printk(KERN_INFO "%s: failed to update PI IRTE\n",
11618                                         __func__);
11619                         goto out;
11620                 }
11621         }
11622
11623         ret = 0;
11624 out:
11625         srcu_read_unlock(&kvm->irq_srcu, idx);
11626         return ret;
11627 }
11628
11629 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11630 {
11631         if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11632                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11633                         FEATURE_CONTROL_LMCE;
11634         else
11635                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11636                         ~FEATURE_CONTROL_LMCE;
11637 }
11638
11639 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
11640         .cpu_has_kvm_support = cpu_has_kvm_support,
11641         .disabled_by_bios = vmx_disabled_by_bios,
11642         .hardware_setup = hardware_setup,
11643         .hardware_unsetup = hardware_unsetup,
11644         .check_processor_compatibility = vmx_check_processor_compat,
11645         .hardware_enable = hardware_enable,
11646         .hardware_disable = hardware_disable,
11647         .cpu_has_accelerated_tpr = report_flexpriority,
11648         .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11649
11650         .vcpu_create = vmx_create_vcpu,
11651         .vcpu_free = vmx_free_vcpu,
11652         .vcpu_reset = vmx_vcpu_reset,
11653
11654         .prepare_guest_switch = vmx_save_host_state,
11655         .vcpu_load = vmx_vcpu_load,
11656         .vcpu_put = vmx_vcpu_put,
11657
11658         .update_bp_intercept = update_exception_bitmap,
11659         .get_msr = vmx_get_msr,
11660         .set_msr = vmx_set_msr,
11661         .get_segment_base = vmx_get_segment_base,
11662         .get_segment = vmx_get_segment,
11663         .set_segment = vmx_set_segment,
11664         .get_cpl = vmx_get_cpl,
11665         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11666         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11667         .decache_cr3 = vmx_decache_cr3,
11668         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11669         .set_cr0 = vmx_set_cr0,
11670         .set_cr3 = vmx_set_cr3,
11671         .set_cr4 = vmx_set_cr4,
11672         .set_efer = vmx_set_efer,
11673         .get_idt = vmx_get_idt,
11674         .set_idt = vmx_set_idt,
11675         .get_gdt = vmx_get_gdt,
11676         .set_gdt = vmx_set_gdt,
11677         .get_dr6 = vmx_get_dr6,
11678         .set_dr6 = vmx_set_dr6,
11679         .set_dr7 = vmx_set_dr7,
11680         .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11681         .cache_reg = vmx_cache_reg,
11682         .get_rflags = vmx_get_rflags,
11683         .set_rflags = vmx_set_rflags,
11684
11685         .get_pkru = vmx_get_pkru,
11686
11687         .tlb_flush = vmx_flush_tlb,
11688
11689         .run = vmx_vcpu_run,
11690         .handle_exit = vmx_handle_exit,
11691         .skip_emulated_instruction = skip_emulated_instruction,
11692         .set_interrupt_shadow = vmx_set_interrupt_shadow,
11693         .get_interrupt_shadow = vmx_get_interrupt_shadow,
11694         .patch_hypercall = vmx_patch_hypercall,
11695         .set_irq = vmx_inject_irq,
11696         .set_nmi = vmx_inject_nmi,
11697         .queue_exception = vmx_queue_exception,
11698         .cancel_injection = vmx_cancel_injection,
11699         .interrupt_allowed = vmx_interrupt_allowed,
11700         .nmi_allowed = vmx_nmi_allowed,
11701         .get_nmi_mask = vmx_get_nmi_mask,
11702         .set_nmi_mask = vmx_set_nmi_mask,
11703         .enable_nmi_window = enable_nmi_window,
11704         .enable_irq_window = enable_irq_window,
11705         .update_cr8_intercept = update_cr8_intercept,
11706         .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11707         .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11708         .get_enable_apicv = vmx_get_enable_apicv,
11709         .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11710         .load_eoi_exitmap = vmx_load_eoi_exitmap,
11711         .apicv_post_state_restore = vmx_apicv_post_state_restore,
11712         .hwapic_irr_update = vmx_hwapic_irr_update,
11713         .hwapic_isr_update = vmx_hwapic_isr_update,
11714         .sync_pir_to_irr = vmx_sync_pir_to_irr,
11715         .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11716
11717         .set_tss_addr = vmx_set_tss_addr,
11718         .get_tdp_level = get_ept_level,
11719         .get_mt_mask = vmx_get_mt_mask,
11720
11721         .get_exit_info = vmx_get_exit_info,
11722
11723         .get_lpage_level = vmx_get_lpage_level,
11724
11725         .cpuid_update = vmx_cpuid_update,
11726
11727         .rdtscp_supported = vmx_rdtscp_supported,
11728         .invpcid_supported = vmx_invpcid_supported,
11729
11730         .set_supported_cpuid = vmx_set_supported_cpuid,
11731
11732         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11733
11734         .write_tsc_offset = vmx_write_tsc_offset,
11735
11736         .set_tdp_cr3 = vmx_set_cr3,
11737
11738         .check_intercept = vmx_check_intercept,
11739         .handle_external_intr = vmx_handle_external_intr,
11740         .mpx_supported = vmx_mpx_supported,
11741         .xsaves_supported = vmx_xsaves_supported,
11742
11743         .check_nested_events = vmx_check_nested_events,
11744
11745         .sched_in = vmx_sched_in,
11746
11747         .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11748         .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11749         .flush_log_dirty = vmx_flush_log_dirty,
11750         .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11751         .write_log_dirty = vmx_write_pml_buffer,
11752
11753         .pre_block = vmx_pre_block,
11754         .post_block = vmx_post_block,
11755
11756         .pmu_ops = &intel_pmu_ops,
11757
11758         .update_pi_irte = vmx_update_pi_irte,
11759
11760 #ifdef CONFIG_X86_64
11761         .set_hv_timer = vmx_set_hv_timer,
11762         .cancel_hv_timer = vmx_cancel_hv_timer,
11763 #endif
11764
11765         .setup_mce = vmx_setup_mce,
11766 };
11767
11768 static int __init vmx_init(void)
11769 {
11770         int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11771                      __alignof__(struct vcpu_vmx), THIS_MODULE);
11772         if (r)
11773                 return r;
11774
11775 #ifdef CONFIG_KEXEC_CORE
11776         rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11777                            crash_vmclear_local_loaded_vmcss);
11778 #endif
11779
11780         return 0;
11781 }
11782
11783 static void __exit vmx_exit(void)
11784 {
11785 #ifdef CONFIG_KEXEC_CORE
11786         RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11787         synchronize_rcu();
11788 #endif
11789
11790         kvm_exit();
11791 }
11792
11793 module_init(vmx_init)
11794 module_exit(vmx_exit)