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 "kvm_cache_regs.h"
  37 #include "x86.h"
  38
  39 #include <asm/cpu.h>
  40 #include <asm/io.h>
  41 #include <asm/desc.h>
  42 #include <asm/vmx.h>
  43 #include <asm/virtext.h>
  44 #include <asm/mce.h>
  45 #include <asm/fpu/internal.h>
  46 #include <asm/perf_event.h>
  47 #include <asm/debugreg.h>
  48 #include <asm/kexec.h>
  49 #include <asm/apic.h>
  50 #include <asm/irq_remapping.h>
  51
  52 #include "trace.h"
  53 #include "pmu.h"
  54
  55 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  56 #define __ex_clear(x, reg) \
  57         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
  58
  59 MODULE_AUTHOR("Qumranet");
  60 MODULE_LICENSE("GPL");
  61
  62 static const struct x86_cpu_id vmx_cpu_id[] = {
  63         X86_FEATURE_MATCH(X86_FEATURE_VMX),
  64         {}
  65 };
  66 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
  67
  68 static bool __read_mostly enable_vpid = 1;
  69 module_param_named(vpid, enable_vpid, bool, 0444);
  70
  71 static bool __read_mostly flexpriority_enabled = 1;
  72 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  73
  74 static bool __read_mostly enable_ept = 1;
  75 module_param_named(ept, enable_ept, bool, S_IRUGO);
  76
  77 static bool __read_mostly enable_unrestricted_guest = 1;
  78 module_param_named(unrestricted_guest,
  79                         enable_unrestricted_guest, bool, S_IRUGO);
  80
  81 static bool __read_mostly enable_ept_ad_bits = 1;
  82 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
  83
  84 static bool __read_mostly emulate_invalid_guest_state = true;
  85 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  86
  87 static bool __read_mostly vmm_exclusive = 1;
  88 module_param(vmm_exclusive, bool, S_IRUGO);
  89
  90 static bool __read_mostly fasteoi = 1;
  91 module_param(fasteoi, bool, S_IRUGO);
  92
  93 static bool __read_mostly enable_apicv = 1;
  94 module_param(enable_apicv, bool, S_IRUGO);
  95
  96 static bool __read_mostly enable_shadow_vmcs = 1;
  97 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
  98 /*
  99  * If nested=1, nested virtualization is supported, i.e., guests may use
 100  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
 101  * use VMX instructions.
 102  */
 103 static bool __read_mostly nested = 0;
 104 module_param(nested, bool, S_IRUGO);
 105
 106 static u64 __read_mostly host_xss;
 107
 108 static bool __read_mostly enable_pml = 1;
 109 module_param_named(pml, enable_pml, bool, S_IRUGO);
 110
 111 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
 112
 113 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
 114 static int __read_mostly cpu_preemption_timer_multi;
 115 static bool __read_mostly enable_preemption_timer = 1;
 116 #ifdef CONFIG_X86_64
 117 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
 118 #endif
 119
 120 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
 121 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
 122 #define KVM_VM_CR0_ALWAYS_ON                                            \
 123         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
 124 #define KVM_CR4_GUEST_OWNED_BITS                                      \
 125         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
 126          | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
 127
 128 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
 129 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
 130
 131 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
 132
 133 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
 134
 135 /*
 136  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 137  * ple_gap:    upper bound on the amount of time between two successive
 138  *             executions of PAUSE in a loop. Also indicate if ple enabled.
 139  *             According to test, this time is usually smaller than 128 cycles.
 140  * ple_window: upper bound on the amount of time a guest is allowed to execute
 141  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
 142  *             less than 2^12 cycles
 143  * Time is measured based on a counter that runs at the same rate as the TSC,
 144  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 145  */
 146 #define KVM_VMX_DEFAULT_PLE_GAP           128
 147 #define KVM_VMX_DEFAULT_PLE_WINDOW        4096
 148 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW   2
 149 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
 150 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX    \
 151                 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
 152
 153 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 154 module_param(ple_gap, int, S_IRUGO);
 155
 156 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 157 module_param(ple_window, int, S_IRUGO);
 158
 159 /* Default doubles per-vcpu window every exit. */
 160 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
 161 module_param(ple_window_grow, int, S_IRUGO);
 162
 163 /* Default resets per-vcpu window every exit to ple_window. */
 164 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
 165 module_param(ple_window_shrink, int, S_IRUGO);
 166
 167 /* Default is to compute the maximum so we can never overflow. */
 168 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
 169 static int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
 170 module_param(ple_window_max, int, S_IRUGO);
 171
 172 extern const ulong vmx_return;
 173
 174 #define NR_AUTOLOAD_MSRS 8
 175 #define VMCS02_POOL_SIZE 1
 176
 177 struct vmcs {
 178         u32 revision_id;
 179         u32 abort;
 180         char data[0];
 181 };
 182
 183 /*
 184  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
 185  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
 186  * loaded on this CPU (so we can clear them if the CPU goes down).
 187  */
 188 struct loaded_vmcs {
 189         struct vmcs *vmcs;
 190         int cpu;
 191         int launched;
 192         struct list_head loaded_vmcss_on_cpu_link;
 193 };
 194
 195 struct shared_msr_entry {
 196         unsigned index;
 197         u64 data;
 198         u64 mask;
 199 };
 200
 201 /*
 202  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
 203  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
 204  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
 205  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
 206  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
 207  * More than one of these structures may exist, if L1 runs multiple L2 guests.
 208  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
 209  * underlying hardware which will be used to run L2.
 210  * This structure is packed to ensure that its layout is identical across
 211  * machines (necessary for live migration).
 212  * If there are changes in this struct, VMCS12_REVISION must be changed.
 213  */
 214 typedef u64 natural_width;
 215 struct __packed vmcs12 {
 216         /* According to the Intel spec, a VMCS region must start with the
 217          * following two fields. Then follow implementation-specific data.
 218          */
 219         u32 revision_id;
 220         u32 abort;
 221
 222         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
 223         u32 padding[7]; /* room for future expansion */
 224
 225         u64 io_bitmap_a;
 226         u64 io_bitmap_b;
 227         u64 msr_bitmap;
 228         u64 vm_exit_msr_store_addr;
 229         u64 vm_exit_msr_load_addr;
 230         u64 vm_entry_msr_load_addr;
 231         u64 tsc_offset;
 232         u64 virtual_apic_page_addr;
 233         u64 apic_access_addr;
 234         u64 posted_intr_desc_addr;
 235         u64 ept_pointer;
 236         u64 eoi_exit_bitmap0;
 237         u64 eoi_exit_bitmap1;
 238         u64 eoi_exit_bitmap2;
 239         u64 eoi_exit_bitmap3;
 240         u64 xss_exit_bitmap;
 241         u64 guest_physical_address;
 242         u64 vmcs_link_pointer;
 243         u64 guest_ia32_debugctl;
 244         u64 guest_ia32_pat;
 245         u64 guest_ia32_efer;
 246         u64 guest_ia32_perf_global_ctrl;
 247         u64 guest_pdptr0;
 248         u64 guest_pdptr1;
 249         u64 guest_pdptr2;
 250         u64 guest_pdptr3;
 251         u64 guest_bndcfgs;
 252         u64 host_ia32_pat;
 253         u64 host_ia32_efer;
 254         u64 host_ia32_perf_global_ctrl;
 255         u64 padding64[8]; /* room for future expansion */
 256         /*
 257          * To allow migration of L1 (complete with its L2 guests) between
 258          * machines of different natural widths (32 or 64 bit), we cannot have
 259          * unsigned long fields with no explict size. We use u64 (aliased
 260          * natural_width) instead. Luckily, x86 is little-endian.
 261          */
 262         natural_width cr0_guest_host_mask;
 263         natural_width cr4_guest_host_mask;
 264         natural_width cr0_read_shadow;
 265         natural_width cr4_read_shadow;
 266         natural_width cr3_target_value0;
 267         natural_width cr3_target_value1;
 268         natural_width cr3_target_value2;
 269         natural_width cr3_target_value3;
 270         natural_width exit_qualification;
 271         natural_width guest_linear_address;
 272         natural_width guest_cr0;
 273         natural_width guest_cr3;
 274         natural_width guest_cr4;
 275         natural_width guest_es_base;
 276         natural_width guest_cs_base;
 277         natural_width guest_ss_base;
 278         natural_width guest_ds_base;
 279         natural_width guest_fs_base;
 280         natural_width guest_gs_base;
 281         natural_width guest_ldtr_base;
 282         natural_width guest_tr_base;
 283         natural_width guest_gdtr_base;
 284         natural_width guest_idtr_base;
 285         natural_width guest_dr7;
 286         natural_width guest_rsp;
 287         natural_width guest_rip;
 288         natural_width guest_rflags;
 289         natural_width guest_pending_dbg_exceptions;
 290         natural_width guest_sysenter_esp;
 291         natural_width guest_sysenter_eip;
 292         natural_width host_cr0;
 293         natural_width host_cr3;
 294         natural_width host_cr4;
 295         natural_width host_fs_base;
 296         natural_width host_gs_base;
 297         natural_width host_tr_base;
 298         natural_width host_gdtr_base;
 299         natural_width host_idtr_base;
 300         natural_width host_ia32_sysenter_esp;
 301         natural_width host_ia32_sysenter_eip;
 302         natural_width host_rsp;
 303         natural_width host_rip;
 304         natural_width paddingl[8]; /* room for future expansion */
 305         u32 pin_based_vm_exec_control;
 306         u32 cpu_based_vm_exec_control;
 307         u32 exception_bitmap;
 308         u32 page_fault_error_code_mask;
 309         u32 page_fault_error_code_match;
 310         u32 cr3_target_count;
 311         u32 vm_exit_controls;
 312         u32 vm_exit_msr_store_count;
 313         u32 vm_exit_msr_load_count;
 314         u32 vm_entry_controls;
 315         u32 vm_entry_msr_load_count;
 316         u32 vm_entry_intr_info_field;
 317         u32 vm_entry_exception_error_code;
 318         u32 vm_entry_instruction_len;
 319         u32 tpr_threshold;
 320         u32 secondary_vm_exec_control;
 321         u32 vm_instruction_error;
 322         u32 vm_exit_reason;
 323         u32 vm_exit_intr_info;
 324         u32 vm_exit_intr_error_code;
 325         u32 idt_vectoring_info_field;
 326         u32 idt_vectoring_error_code;
 327         u32 vm_exit_instruction_len;
 328         u32 vmx_instruction_info;
 329         u32 guest_es_limit;
 330         u32 guest_cs_limit;
 331         u32 guest_ss_limit;
 332         u32 guest_ds_limit;
 333         u32 guest_fs_limit;
 334         u32 guest_gs_limit;
 335         u32 guest_ldtr_limit;
 336         u32 guest_tr_limit;
 337         u32 guest_gdtr_limit;
 338         u32 guest_idtr_limit;
 339         u32 guest_es_ar_bytes;
 340         u32 guest_cs_ar_bytes;
 341         u32 guest_ss_ar_bytes;
 342         u32 guest_ds_ar_bytes;
 343         u32 guest_fs_ar_bytes;
 344         u32 guest_gs_ar_bytes;
 345         u32 guest_ldtr_ar_bytes;
 346         u32 guest_tr_ar_bytes;
 347         u32 guest_interruptibility_info;
 348         u32 guest_activity_state;
 349         u32 guest_sysenter_cs;
 350         u32 host_ia32_sysenter_cs;
 351         u32 vmx_preemption_timer_value;
 352         u32 padding32[7]; /* room for future expansion */
 353         u16 virtual_processor_id;
 354         u16 posted_intr_nv;
 355         u16 guest_es_selector;
 356         u16 guest_cs_selector;
 357         u16 guest_ss_selector;
 358         u16 guest_ds_selector;
 359         u16 guest_fs_selector;
 360         u16 guest_gs_selector;
 361         u16 guest_ldtr_selector;
 362         u16 guest_tr_selector;
 363         u16 guest_intr_status;
 364         u16 host_es_selector;
 365         u16 host_cs_selector;
 366         u16 host_ss_selector;
 367         u16 host_ds_selector;
 368         u16 host_fs_selector;
 369         u16 host_gs_selector;
 370         u16 host_tr_selector;
 371 };
 372
 373 /*
 374  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
 375  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
 376  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
 377  */
 378 #define VMCS12_REVISION 0x11e57ed0
 379
 380 /*
 381  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
 382  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
 383  * current implementation, 4K are reserved to avoid future complications.
 384  */
 385 #define VMCS12_SIZE 0x1000
 386
 387 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
 388 struct vmcs02_list {
 389         struct list_head list;
 390         gpa_t vmptr;
 391         struct loaded_vmcs vmcs02;
 392 };
 393
 394 /*
 395  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 396  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 397  */
 398 struct nested_vmx {
 399         /* Has the level1 guest done vmxon? */
 400         bool vmxon;
 401         gpa_t vmxon_ptr;
 402
 403         /* The guest-physical address of the current VMCS L1 keeps for L2 */
 404         gpa_t current_vmptr;
 405         /* The host-usable pointer to the above */
 406         struct page *current_vmcs12_page;
 407         struct vmcs12 *current_vmcs12;
 408         struct vmcs *current_shadow_vmcs;
 409         /*
 410          * Indicates if the shadow vmcs must be updated with the
 411          * data hold by vmcs12
 412          */
 413         bool sync_shadow_vmcs;
 414
 415         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
 416         struct list_head vmcs02_pool;
 417         int vmcs02_num;
 418         u64 vmcs01_tsc_offset;
 419         /* L2 must run next, and mustn't decide to exit to L1. */
 420         bool nested_run_pending;
 421         /*
 422          * Guest pages referred to in vmcs02 with host-physical pointers, so
 423          * we must keep them pinned while L2 runs.
 424          */
 425         struct page *apic_access_page;
 426         struct page *virtual_apic_page;
 427         struct page *pi_desc_page;
 428         struct pi_desc *pi_desc;
 429         bool pi_pending;
 430         u16 posted_intr_nv;
 431
 432         struct hrtimer preemption_timer;
 433         bool preemption_timer_expired;
 434
 435         /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
 436         u64 vmcs01_debugctl;
 437
 438         u16 vpid02;
 439         u16 last_vpid;
 440
 441         u32 nested_vmx_procbased_ctls_low;
 442         u32 nested_vmx_procbased_ctls_high;
 443         u32 nested_vmx_true_procbased_ctls_low;
 444         u32 nested_vmx_secondary_ctls_low;
 445         u32 nested_vmx_secondary_ctls_high;
 446         u32 nested_vmx_pinbased_ctls_low;
 447         u32 nested_vmx_pinbased_ctls_high;
 448         u32 nested_vmx_exit_ctls_low;
 449         u32 nested_vmx_exit_ctls_high;
 450         u32 nested_vmx_true_exit_ctls_low;
 451         u32 nested_vmx_entry_ctls_low;
 452         u32 nested_vmx_entry_ctls_high;
 453         u32 nested_vmx_true_entry_ctls_low;
 454         u32 nested_vmx_misc_low;
 455         u32 nested_vmx_misc_high;
 456         u32 nested_vmx_ept_caps;
 457         u32 nested_vmx_vpid_caps;
 458 };
 459
 460 #define POSTED_INTR_ON  0
 461 #define POSTED_INTR_SN  1
 462
 463 /* Posted-Interrupt Descriptor */
 464 struct pi_desc {
 465         u32 pir[8];     /* Posted interrupt requested */
 466         union {
 467                 struct {
 468                                 /* bit 256 - Outstanding Notification */
 469                         u16     on      : 1,
 470                                 /* bit 257 - Suppress Notification */
 471                                 sn      : 1,
 472                                 /* bit 271:258 - Reserved */
 473                                 rsvd_1  : 14;
 474                                 /* bit 279:272 - Notification Vector */
 475                         u8      nv;
 476                                 /* bit 287:280 - Reserved */
 477                         u8      rsvd_2;
 478                                 /* bit 319:288 - Notification Destination */
 479                         u32     ndst;
 480                 };
 481                 u64 control;
 482         };
 483         u32 rsvd[6];
 484 } __aligned(64);
 485
 486 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
 487 {
 488         return test_and_set_bit(POSTED_INTR_ON,
 489                         (unsigned long *)&pi_desc->control);
 490 }
 491
 492 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
 493 {
 494         return test_and_clear_bit(POSTED_INTR_ON,
 495                         (unsigned long *)&pi_desc->control);
 496 }
 497
 498 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
 499 {
 500         return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
 501 }
 502
 503 static inline void pi_clear_sn(struct pi_desc *pi_desc)
 504 {
 505         return clear_bit(POSTED_INTR_SN,
 506                         (unsigned long *)&pi_desc->control);
 507 }
 508
 509 static inline void pi_set_sn(struct pi_desc *pi_desc)
 510 {
 511         return set_bit(POSTED_INTR_SN,
 512                         (unsigned long *)&pi_desc->control);
 513 }
 514
 515 static inline int pi_test_on(struct pi_desc *pi_desc)
 516 {
 517         return test_bit(POSTED_INTR_ON,
 518                         (unsigned long *)&pi_desc->control);
 519 }
 520
 521 static inline int pi_test_sn(struct pi_desc *pi_desc)
 522 {
 523         return test_bit(POSTED_INTR_SN,
 524                         (unsigned long *)&pi_desc->control);
 525 }
 526
 527 struct vcpu_vmx {
 528         struct kvm_vcpu       vcpu;
 529         unsigned long         host_rsp;
 530         u8                    fail;
 531         bool                  nmi_known_unmasked;
 532         u32                   exit_intr_info;
 533         u32                   idt_vectoring_info;
 534         ulong                 rflags;
 535         struct shared_msr_entry *guest_msrs;
 536         int                   nmsrs;
 537         int                   save_nmsrs;
 538         unsigned long         host_idt_base;
 539 #ifdef CONFIG_X86_64
 540         u64                   msr_host_kernel_gs_base;
 541         u64                   msr_guest_kernel_gs_base;
 542 #endif
 543         u32 vm_entry_controls_shadow;
 544         u32 vm_exit_controls_shadow;
 545         /*
 546          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 547          * non-nested (L1) guest, it always points to vmcs01. For a nested
 548          * guest (L2), it points to a different VMCS.
 549          */
 550         struct loaded_vmcs    vmcs01;
 551         struct loaded_vmcs   *loaded_vmcs;
 552         bool                  __launched; /* temporary, used in vmx_vcpu_run */
 553         struct msr_autoload {
 554                 unsigned nr;
 555                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 556                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 557         } msr_autoload;
 558         struct {
 559                 int           loaded;
 560                 u16           fs_sel, gs_sel, ldt_sel;
 561 #ifdef CONFIG_X86_64
 562                 u16           ds_sel, es_sel;
 563 #endif
 564                 int           gs_ldt_reload_needed;
 565                 int           fs_reload_needed;
 566                 u64           msr_host_bndcfgs;
 567                 unsigned long vmcs_host_cr4;    /* May not match real cr4 */
 568         } host_state;
 569         struct {
 570                 int vm86_active;
 571                 ulong save_rflags;
 572                 struct kvm_segment segs[8];
 573         } rmode;
 574         struct {
 575                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 576                 struct kvm_save_segment {
 577                         u16 selector;
 578                         unsigned long base;
 579                         u32 limit;
 580                         u32 ar;
 581                 } seg[8];
 582         } segment_cache;
 583         int vpid;
 584         bool emulation_required;
 585
 586         /* Support for vnmi-less CPUs */
 587         int soft_vnmi_blocked;
 588         ktime_t entry_time;
 589         s64 vnmi_blocked_time;
 590         u32 exit_reason;
 591
 592         /* Posted interrupt descriptor */
 593         struct pi_desc pi_desc;
 594
 595         /* Support for a guest hypervisor (nested VMX) */
 596         struct nested_vmx nested;
 597
 598         /* Dynamic PLE window. */
 599         int ple_window;
 600         bool ple_window_dirty;
 601
 602         /* Support for PML */
 603 #define PML_ENTITY_NUM          512
 604         struct page *pml_pg;
 605
 606         /* apic deadline value in host tsc */
 607         u64 hv_deadline_tsc;
 608
 609         u64 current_tsc_ratio;
 610
 611         bool guest_pkru_valid;
 612         u32 guest_pkru;
 613         u32 host_pkru;
 614
 615         /*
 616          * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
 617          * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
 618          * in msr_ia32_feature_control_valid_bits.
 619          */
 620         u64 msr_ia32_feature_control;
 621         u64 msr_ia32_feature_control_valid_bits;
 622 };
 623
 624 enum segment_cache_field {
 625         SEG_FIELD_SEL = 0,
 626         SEG_FIELD_BASE = 1,
 627         SEG_FIELD_LIMIT = 2,
 628         SEG_FIELD_AR = 3,
 629
 630         SEG_FIELD_NR = 4
 631 };
 632
 633 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 634 {
 635         return container_of(vcpu, struct vcpu_vmx, vcpu);
 636 }
 637
 638 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
 639 {
 640         return &(to_vmx(vcpu)->pi_desc);
 641 }
 642
 643 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
 644 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
 645 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
 646                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
 647
 648
 649 static unsigned long shadow_read_only_fields[] = {
 650         /*
 651          * We do NOT shadow fields that are modified when L0
 652          * traps and emulates any vmx instruction (e.g. VMPTRLD,
 653          * VMXON...) executed by L1.
 654          * For example, VM_INSTRUCTION_ERROR is read
 655          * by L1 if a vmx instruction fails (part of the error path).
 656          * Note the code assumes this logic. If for some reason
 657          * we start shadowing these fields then we need to
 658          * force a shadow sync when L0 emulates vmx instructions
 659          * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
 660          * by nested_vmx_failValid)
 661          */
 662         VM_EXIT_REASON,
 663         VM_EXIT_INTR_INFO,
 664         VM_EXIT_INSTRUCTION_LEN,
 665         IDT_VECTORING_INFO_FIELD,
 666         IDT_VECTORING_ERROR_CODE,
 667         VM_EXIT_INTR_ERROR_CODE,
 668         EXIT_QUALIFICATION,
 669         GUEST_LINEAR_ADDRESS,
 670         GUEST_PHYSICAL_ADDRESS
 671 };
 672 static int max_shadow_read_only_fields =
 673         ARRAY_SIZE(shadow_read_only_fields);
 674
 675 static unsigned long shadow_read_write_fields[] = {
 676         TPR_THRESHOLD,
 677         GUEST_RIP,
 678         GUEST_RSP,
 679         GUEST_CR0,
 680         GUEST_CR3,
 681         GUEST_CR4,
 682         GUEST_INTERRUPTIBILITY_INFO,
 683         GUEST_RFLAGS,
 684         GUEST_CS_SELECTOR,
 685         GUEST_CS_AR_BYTES,
 686         GUEST_CS_LIMIT,
 687         GUEST_CS_BASE,
 688         GUEST_ES_BASE,
 689         GUEST_BNDCFGS,
 690         CR0_GUEST_HOST_MASK,
 691         CR0_READ_SHADOW,
 692         CR4_READ_SHADOW,
 693         TSC_OFFSET,
 694         EXCEPTION_BITMAP,
 695         CPU_BASED_VM_EXEC_CONTROL,
 696         VM_ENTRY_EXCEPTION_ERROR_CODE,
 697         VM_ENTRY_INTR_INFO_FIELD,
 698         VM_ENTRY_INSTRUCTION_LEN,
 699         VM_ENTRY_EXCEPTION_ERROR_CODE,
 700         HOST_FS_BASE,
 701         HOST_GS_BASE,
 702         HOST_FS_SELECTOR,
 703         HOST_GS_SELECTOR
 704 };
 705 static int max_shadow_read_write_fields =
 706         ARRAY_SIZE(shadow_read_write_fields);
 707
 708 static const unsigned short vmcs_field_to_offset_table[] = {
 709         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 710         FIELD(POSTED_INTR_NV, posted_intr_nv),
 711         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
 712         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
 713         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
 714         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
 715         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
 716         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
 717         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
 718         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
 719         FIELD(GUEST_INTR_STATUS, guest_intr_status),
 720         FIELD(HOST_ES_SELECTOR, host_es_selector),
 721         FIELD(HOST_CS_SELECTOR, host_cs_selector),
 722         FIELD(HOST_SS_SELECTOR, host_ss_selector),
 723         FIELD(HOST_DS_SELECTOR, host_ds_selector),
 724         FIELD(HOST_FS_SELECTOR, host_fs_selector),
 725         FIELD(HOST_GS_SELECTOR, host_gs_selector),
 726         FIELD(HOST_TR_SELECTOR, host_tr_selector),
 727         FIELD64(IO_BITMAP_A, io_bitmap_a),
 728         FIELD64(IO_BITMAP_B, io_bitmap_b),
 729         FIELD64(MSR_BITMAP, msr_bitmap),
 730         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
 731         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
 732         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
 733         FIELD64(TSC_OFFSET, tsc_offset),
 734         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
 735         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
 736         FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
 737         FIELD64(EPT_POINTER, ept_pointer),
 738         FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
 739         FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
 740         FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
 741         FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
 742         FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
 743         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
 744         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
 745         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
 746         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
 747         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
 748         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
 749         FIELD64(GUEST_PDPTR0, guest_pdptr0),
 750         FIELD64(GUEST_PDPTR1, guest_pdptr1),
 751         FIELD64(GUEST_PDPTR2, guest_pdptr2),
 752         FIELD64(GUEST_PDPTR3, guest_pdptr3),
 753         FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
 754         FIELD64(HOST_IA32_PAT, host_ia32_pat),
 755         FIELD64(HOST_IA32_EFER, host_ia32_efer),
 756         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
 757         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
 758         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
 759         FIELD(EXCEPTION_BITMAP, exception_bitmap),
 760         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
 761         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
 762         FIELD(CR3_TARGET_COUNT, cr3_target_count),
 763         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
 764         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
 765         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
 766         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
 767         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
 768         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
 769         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
 770         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
 771         FIELD(TPR_THRESHOLD, tpr_threshold),
 772         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
 773         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
 774         FIELD(VM_EXIT_REASON, vm_exit_reason),
 775         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
 776         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
 777         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
 778         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
 779         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
 780         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
 781         FIELD(GUEST_ES_LIMIT, guest_es_limit),
 782         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
 783         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
 784         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
 785         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
 786         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
 787         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
 788         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
 789         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
 790         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
 791         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
 792         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
 793         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
 794         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
 795         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
 796         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
 797         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
 798         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
 799         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
 800         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
 801         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
 802         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
 803         FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
 804         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
 805         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
 806         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
 807         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
 808         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
 809         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
 810         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
 811         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
 812         FIELD(EXIT_QUALIFICATION, exit_qualification),
 813         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
 814         FIELD(GUEST_CR0, guest_cr0),
 815         FIELD(GUEST_CR3, guest_cr3),
 816         FIELD(GUEST_CR4, guest_cr4),
 817         FIELD(GUEST_ES_BASE, guest_es_base),
 818         FIELD(GUEST_CS_BASE, guest_cs_base),
 819         FIELD(GUEST_SS_BASE, guest_ss_base),
 820         FIELD(GUEST_DS_BASE, guest_ds_base),
 821         FIELD(GUEST_FS_BASE, guest_fs_base),
 822         FIELD(GUEST_GS_BASE, guest_gs_base),
 823         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
 824         FIELD(GUEST_TR_BASE, guest_tr_base),
 825         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
 826         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
 827         FIELD(GUEST_DR7, guest_dr7),
 828         FIELD(GUEST_RSP, guest_rsp),
 829         FIELD(GUEST_RIP, guest_rip),
 830         FIELD(GUEST_RFLAGS, guest_rflags),
 831         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
 832         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
 833         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
 834         FIELD(HOST_CR0, host_cr0),
 835         FIELD(HOST_CR3, host_cr3),
 836         FIELD(HOST_CR4, host_cr4),
 837         FIELD(HOST_FS_BASE, host_fs_base),
 838         FIELD(HOST_GS_BASE, host_gs_base),
 839         FIELD(HOST_TR_BASE, host_tr_base),
 840         FIELD(HOST_GDTR_BASE, host_gdtr_base),
 841         FIELD(HOST_IDTR_BASE, host_idtr_base),
 842         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
 843         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
 844         FIELD(HOST_RSP, host_rsp),
 845         FIELD(HOST_RIP, host_rip),
 846 };
 847
 848 static inline short vmcs_field_to_offset(unsigned long field)
 849 {
 850         BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
 851
 852         if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
 853             vmcs_field_to_offset_table[field] == 0)
 854                 return -ENOENT;
 855
 856         return vmcs_field_to_offset_table[field];
 857 }
 858
 859 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
 860 {
 861         return to_vmx(vcpu)->nested.current_vmcs12;
 862 }
 863
 864 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
 865 {
 866         struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
 867         if (is_error_page(page))
 868                 return NULL;
 869
 870         return page;
 871 }
 872
 873 static void nested_release_page(struct page *page)
 874 {
 875         kvm_release_page_dirty(page);
 876 }
 877
 878 static void nested_release_page_clean(struct page *page)
 879 {
 880         kvm_release_page_clean(page);
 881 }
 882
 883 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
 884 static u64 construct_eptp(unsigned long root_hpa);
 885 static void kvm_cpu_vmxon(u64 addr);
 886 static void kvm_cpu_vmxoff(void);
 887 static bool vmx_xsaves_supported(void);
 888 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 889 static void vmx_set_segment(struct kvm_vcpu *vcpu,
 890                             struct kvm_segment *var, int seg);
 891 static void vmx_get_segment(struct kvm_vcpu *vcpu,
 892                             struct kvm_segment *var, int seg);
 893 static bool guest_state_valid(struct kvm_vcpu *vcpu);
 894 static u32 vmx_segment_access_rights(struct kvm_segment *var);
 895 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
 896 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
 897 static int alloc_identity_pagetable(struct kvm *kvm);
 898
 899 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 900 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 901 /*
 902  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
 903  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
 904  */
 905 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
 906 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 907
 908 /*
 909  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
 910  * can find which vCPU should be waken up.
 911  */
 912 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
 913 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
 914
 915 static unsigned long *vmx_io_bitmap_a;
 916 static unsigned long *vmx_io_bitmap_b;
 917 static unsigned long *vmx_msr_bitmap_legacy;
 918 static unsigned long *vmx_msr_bitmap_longmode;
 919 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
 920 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
 921 static unsigned long *vmx_msr_bitmap_nested;
 922 static unsigned long *vmx_vmread_bitmap;
 923 static unsigned long *vmx_vmwrite_bitmap;
 924
 925 static bool cpu_has_load_ia32_efer;
 926 static bool cpu_has_load_perf_global_ctrl;
 927
 928 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 929 static DEFINE_SPINLOCK(vmx_vpid_lock);
 930
 931 static struct vmcs_config {
 932         int size;
 933         int order;
 934         u32 revision_id;
 935         u32 pin_based_exec_ctrl;
 936         u32 cpu_based_exec_ctrl;
 937         u32 cpu_based_2nd_exec_ctrl;
 938         u32 vmexit_ctrl;
 939         u32 vmentry_ctrl;
 940 } vmcs_config;
 941
 942 static struct vmx_capability {
 943         u32 ept;
 944         u32 vpid;
 945 } vmx_capability;
 946
 947 #define VMX_SEGMENT_FIELD(seg)                                  \
 948         [VCPU_SREG_##seg] = {                                   \
 949                 .selector = GUEST_##seg##_SELECTOR,             \
 950                 .base = GUEST_##seg##_BASE,                     \
 951                 .limit = GUEST_##seg##_LIMIT,                   \
 952                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 953         }
 954
 955 static const struct kvm_vmx_segment_field {
 956         unsigned selector;
 957         unsigned base;
 958         unsigned limit;
 959         unsigned ar_bytes;
 960 } kvm_vmx_segment_fields[] = {
 961         VMX_SEGMENT_FIELD(CS),
 962         VMX_SEGMENT_FIELD(DS),
 963         VMX_SEGMENT_FIELD(ES),
 964         VMX_SEGMENT_FIELD(FS),
 965         VMX_SEGMENT_FIELD(GS),
 966         VMX_SEGMENT_FIELD(SS),
 967         VMX_SEGMENT_FIELD(TR),
 968         VMX_SEGMENT_FIELD(LDTR),
 969 };
 970
 971 static u64 host_efer;
 972
 973 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
 974
 975 /*
 976  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
 977  * away by decrementing the array size.
 978  */
 979 static const u32 vmx_msr_index[] = {
 980 #ifdef CONFIG_X86_64
 981         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
 982 #endif
 983         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
 984 };
 985
 986 static inline bool is_exception_n(u32 intr_info, u8 vector)
 987 {
 988         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
 989                              INTR_INFO_VALID_MASK)) ==
 990                 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
 991 }
 992
 993 static inline bool is_debug(u32 intr_info)
 994 {
 995         return is_exception_n(intr_info, DB_VECTOR);
 996 }
 997
 998 static inline bool is_breakpoint(u32 intr_info)
 999 {
1000         return is_exception_n(intr_info, BP_VECTOR);
1001 }
1002
1003 static inline bool is_page_fault(u32 intr_info)
1004 {
1005         return is_exception_n(intr_info, PF_VECTOR);
1006 }
1007
1008 static inline bool is_no_device(u32 intr_info)
1009 {
1010         return is_exception_n(intr_info, NM_VECTOR);
1011 }
1012
1013 static inline bool is_invalid_opcode(u32 intr_info)
1014 {
1015         return is_exception_n(intr_info, UD_VECTOR);
1016 }
1017
1018 static inline bool is_external_interrupt(u32 intr_info)
1019 {
1020         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1021                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1022 }
1023
1024 static inline bool is_machine_check(u32 intr_info)
1025 {
1026         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1027                              INTR_INFO_VALID_MASK)) ==
1028                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1029 }
1030
1031 static inline bool cpu_has_vmx_msr_bitmap(void)
1032 {
1033         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1034 }
1035
1036 static inline bool cpu_has_vmx_tpr_shadow(void)
1037 {
1038         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1039 }
1040
1041 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1042 {
1043         return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1044 }
1045
1046 static inline bool cpu_has_secondary_exec_ctrls(void)
1047 {
1048         return vmcs_config.cpu_based_exec_ctrl &
1049                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1050 }
1051
1052 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1053 {
1054         return vmcs_config.cpu_based_2nd_exec_ctrl &
1055                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1056 }
1057
1058 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1059 {
1060         return vmcs_config.cpu_based_2nd_exec_ctrl &
1061                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1062 }
1063
1064 static inline bool cpu_has_vmx_apic_register_virt(void)
1065 {
1066         return vmcs_config.cpu_based_2nd_exec_ctrl &
1067                 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1068 }
1069
1070 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1071 {
1072         return vmcs_config.cpu_based_2nd_exec_ctrl &
1073                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1074 }
1075
1076 /*
1077  * Comment's format: document - errata name - stepping - processor name.
1078  * Refer from
1079  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1080  */
1081 static u32 vmx_preemption_cpu_tfms[] = {
1082 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
1083 0x000206E6,
1084 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
1085 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1086 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1087 0x00020652,
1088 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1089 0x00020655,
1090 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
1091 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
1092 /*
1093  * 320767.pdf - AAP86  - B1 -
1094  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1095  */
1096 0x000106E5,
1097 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1098 0x000106A0,
1099 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1100 0x000106A1,
1101 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1102 0x000106A4,
1103  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1104  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1105  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1106 0x000106A5,
1107 };
1108
1109 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1110 {
1111         u32 eax = cpuid_eax(0x00000001), i;
1112
1113         /* Clear the reserved bits */
1114         eax &= ~(0x3U << 14 | 0xfU << 28);
1115         for (i = 0; i < sizeof(vmx_preemption_cpu_tfms)/sizeof(u32); i++)
1116                 if (eax == vmx_preemption_cpu_tfms[i])
1117                         return true;
1118
1119         return false;
1120 }
1121
1122 static inline bool cpu_has_vmx_preemption_timer(void)
1123 {
1124         if (cpu_has_broken_vmx_preemption_timer())
1125                 return false;
1126
1127         return vmcs_config.pin_based_exec_ctrl &
1128                 PIN_BASED_VMX_PREEMPTION_TIMER;
1129 }
1130
1131 static inline bool cpu_has_vmx_posted_intr(void)
1132 {
1133         return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1134                 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1135 }
1136
1137 static inline bool cpu_has_vmx_apicv(void)
1138 {
1139         return cpu_has_vmx_apic_register_virt() &&
1140                 cpu_has_vmx_virtual_intr_delivery() &&
1141                 cpu_has_vmx_posted_intr();
1142 }
1143
1144 static inline bool cpu_has_vmx_flexpriority(void)
1145 {
1146         return cpu_has_vmx_tpr_shadow() &&
1147                 cpu_has_vmx_virtualize_apic_accesses();
1148 }
1149
1150 static inline bool cpu_has_vmx_ept_execute_only(void)
1151 {
1152         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1153 }
1154
1155 static inline bool cpu_has_vmx_ept_2m_page(void)
1156 {
1157         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1158 }
1159
1160 static inline bool cpu_has_vmx_ept_1g_page(void)
1161 {
1162         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1163 }
1164
1165 static inline bool cpu_has_vmx_ept_4levels(void)
1166 {
1167         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1168 }
1169
1170 static inline bool cpu_has_vmx_ept_ad_bits(void)
1171 {
1172         return vmx_capability.ept & VMX_EPT_AD_BIT;
1173 }
1174
1175 static inline bool cpu_has_vmx_invept_context(void)
1176 {
1177         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1178 }
1179
1180 static inline bool cpu_has_vmx_invept_global(void)
1181 {
1182         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1183 }
1184
1185 static inline bool cpu_has_vmx_invvpid_single(void)
1186 {
1187         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1188 }
1189
1190 static inline bool cpu_has_vmx_invvpid_global(void)
1191 {
1192         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1193 }
1194
1195 static inline bool cpu_has_vmx_ept(void)
1196 {
1197         return vmcs_config.cpu_based_2nd_exec_ctrl &
1198                 SECONDARY_EXEC_ENABLE_EPT;
1199 }
1200
1201 static inline bool cpu_has_vmx_unrestricted_guest(void)
1202 {
1203         return vmcs_config.cpu_based_2nd_exec_ctrl &
1204                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1205 }
1206
1207 static inline bool cpu_has_vmx_ple(void)
1208 {
1209         return vmcs_config.cpu_based_2nd_exec_ctrl &
1210                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1211 }
1212
1213 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1214 {
1215         return flexpriority_enabled && lapic_in_kernel(vcpu);
1216 }
1217
1218 static inline bool cpu_has_vmx_vpid(void)
1219 {
1220         return vmcs_config.cpu_based_2nd_exec_ctrl &
1221                 SECONDARY_EXEC_ENABLE_VPID;
1222 }
1223
1224 static inline bool cpu_has_vmx_rdtscp(void)
1225 {
1226         return vmcs_config.cpu_based_2nd_exec_ctrl &
1227                 SECONDARY_EXEC_RDTSCP;
1228 }
1229
1230 static inline bool cpu_has_vmx_invpcid(void)
1231 {
1232         return vmcs_config.cpu_based_2nd_exec_ctrl &
1233                 SECONDARY_EXEC_ENABLE_INVPCID;
1234 }
1235
1236 static inline bool cpu_has_virtual_nmis(void)
1237 {
1238         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1239 }
1240
1241 static inline bool cpu_has_vmx_wbinvd_exit(void)
1242 {
1243         return vmcs_config.cpu_based_2nd_exec_ctrl &
1244                 SECONDARY_EXEC_WBINVD_EXITING;
1245 }
1246
1247 static inline bool cpu_has_vmx_shadow_vmcs(void)
1248 {
1249         u64 vmx_msr;
1250         rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1251         /* check if the cpu supports writing r/o exit information fields */
1252         if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1253                 return false;
1254
1255         return vmcs_config.cpu_based_2nd_exec_ctrl &
1256                 SECONDARY_EXEC_SHADOW_VMCS;
1257 }
1258
1259 static inline bool cpu_has_vmx_pml(void)
1260 {
1261         return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1262 }
1263
1264 static inline bool cpu_has_vmx_tsc_scaling(void)
1265 {
1266         return vmcs_config.cpu_based_2nd_exec_ctrl &
1267                 SECONDARY_EXEC_TSC_SCALING;
1268 }
1269
1270 static inline bool report_flexpriority(void)
1271 {
1272         return flexpriority_enabled;
1273 }
1274
1275 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1276 {
1277         return vmcs12->cpu_based_vm_exec_control & bit;
1278 }
1279
1280 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1281 {
1282         return (vmcs12->cpu_based_vm_exec_control &
1283                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1284                 (vmcs12->secondary_vm_exec_control & bit);
1285 }
1286
1287 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1288 {
1289         return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1290 }
1291
1292 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1293 {
1294         return vmcs12->pin_based_vm_exec_control &
1295                 PIN_BASED_VMX_PREEMPTION_TIMER;
1296 }
1297
1298 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1299 {
1300         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1301 }
1302
1303 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1304 {
1305         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1306                 vmx_xsaves_supported();
1307 }
1308
1309 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1310 {
1311         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1312 }
1313
1314 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1315 {
1316         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1317 }
1318
1319 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1320 {
1321         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1322 }
1323
1324 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1325 {
1326         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1327 }
1328
1329 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1330 {
1331         return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1332 }
1333
1334 static inline bool is_exception(u32 intr_info)
1335 {
1336         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1337                 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1338 }
1339
1340 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1341                               u32 exit_intr_info,
1342                               unsigned long exit_qualification);
1343 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1344                         struct vmcs12 *vmcs12,
1345                         u32 reason, unsigned long qualification);
1346
1347 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1348 {
1349         int i;
1350
1351         for (i = 0; i < vmx->nmsrs; ++i)
1352                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1353                         return i;
1354         return -1;
1355 }
1356
1357 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1358 {
1359     struct {
1360         u64 vpid : 16;
1361         u64 rsvd : 48;
1362         u64 gva;
1363     } operand = { vpid, 0, gva };
1364
1365     asm volatile (__ex(ASM_VMX_INVVPID)
1366                   /* CF==1 or ZF==1 --> rc = -1 */
1367                   "; ja 1f ; ud2 ; 1:"
1368                   : : "a"(&operand), "c"(ext) : "cc", "memory");
1369 }
1370
1371 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1372 {
1373         struct {
1374                 u64 eptp, gpa;
1375         } operand = {eptp, gpa};
1376
1377         asm volatile (__ex(ASM_VMX_INVEPT)
1378                         /* CF==1 or ZF==1 --> rc = -1 */
1379                         "; ja 1f ; ud2 ; 1:\n"
1380                         : : "a" (&operand), "c" (ext) : "cc", "memory");
1381 }
1382
1383 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1384 {
1385         int i;
1386
1387         i = __find_msr_index(vmx, msr);
1388         if (i >= 0)
1389                 return &vmx->guest_msrs[i];
1390         return NULL;
1391 }
1392
1393 static void vmcs_clear(struct vmcs *vmcs)
1394 {
1395         u64 phys_addr = __pa(vmcs);
1396         u8 error;
1397
1398         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1399                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1400                       : "cc", "memory");
1401         if (error)
1402                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1403                        vmcs, phys_addr);
1404 }
1405
1406 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1407 {
1408         vmcs_clear(loaded_vmcs->vmcs);
1409         loaded_vmcs->cpu = -1;
1410         loaded_vmcs->launched = 0;
1411 }
1412
1413 static void vmcs_load(struct vmcs *vmcs)
1414 {
1415         u64 phys_addr = __pa(vmcs);
1416         u8 error;
1417
1418         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1419                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1420                         : "cc", "memory");
1421         if (error)
1422                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1423                        vmcs, phys_addr);
1424 }
1425
1426 #ifdef CONFIG_KEXEC_CORE
1427 /*
1428  * This bitmap is used to indicate whether the vmclear
1429  * operation is enabled on all cpus. All disabled by
1430  * default.
1431  */
1432 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1433
1434 static inline void crash_enable_local_vmclear(int cpu)
1435 {
1436         cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1437 }
1438
1439 static inline void crash_disable_local_vmclear(int cpu)
1440 {
1441         cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1442 }
1443
1444 static inline int crash_local_vmclear_enabled(int cpu)
1445 {
1446         return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1447 }
1448
1449 static void crash_vmclear_local_loaded_vmcss(void)
1450 {
1451         int cpu = raw_smp_processor_id();
1452         struct loaded_vmcs *v;
1453
1454         if (!crash_local_vmclear_enabled(cpu))
1455                 return;
1456
1457         list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1458                             loaded_vmcss_on_cpu_link)
1459                 vmcs_clear(v->vmcs);
1460 }
1461 #else
1462 static inline void crash_enable_local_vmclear(int cpu) { }
1463 static inline void crash_disable_local_vmclear(int cpu) { }
1464 #endif /* CONFIG_KEXEC_CORE */
1465
1466 static void __loaded_vmcs_clear(void *arg)
1467 {
1468         struct loaded_vmcs *loaded_vmcs = arg;
1469         int cpu = raw_smp_processor_id();
1470
1471         if (loaded_vmcs->cpu != cpu)
1472                 return; /* vcpu migration can race with cpu offline */
1473         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1474                 per_cpu(current_vmcs, cpu) = NULL;
1475         crash_disable_local_vmclear(cpu);
1476         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1477
1478         /*
1479          * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1480          * is before setting loaded_vmcs->vcpu to -1 which is done in
1481          * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1482          * then adds the vmcs into percpu list before it is deleted.
1483          */
1484         smp_wmb();
1485
1486         loaded_vmcs_init(loaded_vmcs);
1487         crash_enable_local_vmclear(cpu);
1488 }
1489
1490 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1491 {
1492         int cpu = loaded_vmcs->cpu;
1493
1494         if (cpu != -1)
1495                 smp_call_function_single(cpu,
1496                          __loaded_vmcs_clear, loaded_vmcs, 1);
1497 }
1498
1499 static inline void vpid_sync_vcpu_single(int vpid)
1500 {
1501         if (vpid == 0)
1502                 return;
1503
1504         if (cpu_has_vmx_invvpid_single())
1505                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1506 }
1507
1508 static inline void vpid_sync_vcpu_global(void)
1509 {
1510         if (cpu_has_vmx_invvpid_global())
1511                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1512 }
1513
1514 static inline void vpid_sync_context(int vpid)
1515 {
1516         if (cpu_has_vmx_invvpid_single())
1517                 vpid_sync_vcpu_single(vpid);
1518         else
1519                 vpid_sync_vcpu_global();
1520 }
1521
1522 static inline void ept_sync_global(void)
1523 {
1524         if (cpu_has_vmx_invept_global())
1525                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1526 }
1527
1528 static inline void ept_sync_context(u64 eptp)
1529 {
1530         if (enable_ept) {
1531                 if (cpu_has_vmx_invept_context())
1532                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1533                 else
1534                         ept_sync_global();
1535         }
1536 }
1537
1538 static __always_inline void vmcs_check16(unsigned long field)
1539 {
1540         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1541                          "16-bit accessor invalid for 64-bit field");
1542         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1543                          "16-bit accessor invalid for 64-bit high field");
1544         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1545                          "16-bit accessor invalid for 32-bit high field");
1546         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1547                          "16-bit accessor invalid for natural width field");
1548 }
1549
1550 static __always_inline void vmcs_check32(unsigned long field)
1551 {
1552         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1553                          "32-bit accessor invalid for 16-bit field");
1554         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1555                          "32-bit accessor invalid for natural width field");
1556 }
1557
1558 static __always_inline void vmcs_check64(unsigned long field)
1559 {
1560         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1561                          "64-bit accessor invalid for 16-bit field");
1562         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1563                          "64-bit accessor invalid for 64-bit high field");
1564         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1565                          "64-bit accessor invalid for 32-bit field");
1566         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1567                          "64-bit accessor invalid for natural width field");
1568 }
1569
1570 static __always_inline void vmcs_checkl(unsigned long field)
1571 {
1572         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1573                          "Natural width accessor invalid for 16-bit field");
1574         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1575                          "Natural width accessor invalid for 64-bit field");
1576         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1577                          "Natural width accessor invalid for 64-bit high field");
1578         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1579                          "Natural width accessor invalid for 32-bit field");
1580 }
1581
1582 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1583 {
1584         unsigned long value;
1585
1586         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1587                       : "=a"(value) : "d"(field) : "cc");
1588         return value;
1589 }
1590
1591 static __always_inline u16 vmcs_read16(unsigned long field)
1592 {
1593         vmcs_check16(field);
1594         return __vmcs_readl(field);
1595 }
1596
1597 static __always_inline u32 vmcs_read32(unsigned long field)
1598 {
1599         vmcs_check32(field);
1600         return __vmcs_readl(field);
1601 }
1602
1603 static __always_inline u64 vmcs_read64(unsigned long field)
1604 {
1605         vmcs_check64(field);
1606 #ifdef CONFIG_X86_64
1607         return __vmcs_readl(field);
1608 #else
1609         return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1610 #endif
1611 }
1612
1613 static __always_inline unsigned long vmcs_readl(unsigned long field)
1614 {
1615         vmcs_checkl(field);
1616         return __vmcs_readl(field);
1617 }
1618
1619 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1620 {
1621         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1622                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1623         dump_stack();
1624 }
1625
1626 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1627 {
1628         u8 error;
1629
1630         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1631                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1632         if (unlikely(error))
1633                 vmwrite_error(field, value);
1634 }
1635
1636 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1637 {
1638         vmcs_check16(field);
1639         __vmcs_writel(field, value);
1640 }
1641
1642 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1643 {
1644         vmcs_check32(field);
1645         __vmcs_writel(field, value);
1646 }
1647
1648 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1649 {
1650         vmcs_check64(field);
1651         __vmcs_writel(field, value);
1652 #ifndef CONFIG_X86_64
1653         asm volatile ("");
1654         __vmcs_writel(field+1, value >> 32);
1655 #endif
1656 }
1657
1658 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1659 {
1660         vmcs_checkl(field);
1661         __vmcs_writel(field, value);
1662 }
1663
1664 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1665 {
1666         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1667                          "vmcs_clear_bits does not support 64-bit fields");
1668         __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1669 }
1670
1671 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1672 {
1673         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1674                          "vmcs_set_bits does not support 64-bit fields");
1675         __vmcs_writel(field, __vmcs_readl(field) | mask);
1676 }
1677
1678 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1679 {
1680         vmcs_write32(VM_ENTRY_CONTROLS, val);
1681         vmx->vm_entry_controls_shadow = val;
1682 }
1683
1684 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1685 {
1686         if (vmx->vm_entry_controls_shadow != val)
1687                 vm_entry_controls_init(vmx, val);
1688 }
1689
1690 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1691 {
1692         return vmx->vm_entry_controls_shadow;
1693 }
1694
1695
1696 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1697 {
1698         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1699 }
1700
1701 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1702 {
1703         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1704 }
1705
1706 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1707 {
1708         vmcs_write32(VM_EXIT_CONTROLS, val);
1709         vmx->vm_exit_controls_shadow = val;
1710 }
1711
1712 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1713 {
1714         if (vmx->vm_exit_controls_shadow != val)
1715                 vm_exit_controls_init(vmx, val);
1716 }
1717
1718 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1719 {
1720         return vmx->vm_exit_controls_shadow;
1721 }
1722
1723
1724 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1725 {
1726         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1727 }
1728
1729 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1730 {
1731         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1732 }
1733
1734 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1735 {
1736         vmx->segment_cache.bitmask = 0;
1737 }
1738
1739 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1740                                        unsigned field)
1741 {
1742         bool ret;
1743         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1744
1745         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1746                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1747                 vmx->segment_cache.bitmask = 0;
1748         }
1749         ret = vmx->segment_cache.bitmask & mask;
1750         vmx->segment_cache.bitmask |= mask;
1751         return ret;
1752 }
1753
1754 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1755 {
1756         u16 *p = &vmx->segment_cache.seg[seg].selector;
1757
1758         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1759                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1760         return *p;
1761 }
1762
1763 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1764 {
1765         ulong *p = &vmx->segment_cache.seg[seg].base;
1766
1767         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1768                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1769         return *p;
1770 }
1771
1772 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1773 {
1774         u32 *p = &vmx->segment_cache.seg[seg].limit;
1775
1776         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1777                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1778         return *p;
1779 }
1780
1781 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1782 {
1783         u32 *p = &vmx->segment_cache.seg[seg].ar;
1784
1785         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1786                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1787         return *p;
1788 }
1789
1790 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1791 {
1792         u32 eb;
1793
1794         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1795              (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR);
1796         if ((vcpu->guest_debug &
1797              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1798             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1799                 eb |= 1u << BP_VECTOR;
1800         if (to_vmx(vcpu)->rmode.vm86_active)
1801                 eb = ~0;
1802         if (enable_ept)
1803                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1804         if (vcpu->fpu_active)
1805                 eb &= ~(1u << NM_VECTOR);
1806
1807         /* When we are running a nested L2 guest and L1 specified for it a
1808          * certain exception bitmap, we must trap the same exceptions and pass
1809          * them to L1. When running L2, we will only handle the exceptions
1810          * specified above if L1 did not want them.
1811          */
1812         if (is_guest_mode(vcpu))
1813                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1814
1815         vmcs_write32(EXCEPTION_BITMAP, eb);
1816 }
1817
1818 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1819                 unsigned long entry, unsigned long exit)
1820 {
1821         vm_entry_controls_clearbit(vmx, entry);
1822         vm_exit_controls_clearbit(vmx, exit);
1823 }
1824
1825 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1826 {
1827         unsigned i;
1828         struct msr_autoload *m = &vmx->msr_autoload;
1829
1830         switch (msr) {
1831         case MSR_EFER:
1832                 if (cpu_has_load_ia32_efer) {
1833                         clear_atomic_switch_msr_special(vmx,
1834                                         VM_ENTRY_LOAD_IA32_EFER,
1835                                         VM_EXIT_LOAD_IA32_EFER);
1836                         return;
1837                 }
1838                 break;
1839         case MSR_CORE_PERF_GLOBAL_CTRL:
1840                 if (cpu_has_load_perf_global_ctrl) {
1841                         clear_atomic_switch_msr_special(vmx,
1842                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1843                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1844                         return;
1845                 }
1846                 break;
1847         }
1848
1849         for (i = 0; i < m->nr; ++i)
1850                 if (m->guest[i].index == msr)
1851                         break;
1852
1853         if (i == m->nr)
1854                 return;
1855         --m->nr;
1856         m->guest[i] = m->guest[m->nr];
1857         m->host[i] = m->host[m->nr];
1858         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1859         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1860 }
1861
1862 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1863                 unsigned long entry, unsigned long exit,
1864                 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1865                 u64 guest_val, u64 host_val)
1866 {
1867         vmcs_write64(guest_val_vmcs, guest_val);
1868         vmcs_write64(host_val_vmcs, host_val);
1869         vm_entry_controls_setbit(vmx, entry);
1870         vm_exit_controls_setbit(vmx, exit);
1871 }
1872
1873 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1874                                   u64 guest_val, u64 host_val)
1875 {
1876         unsigned i;
1877         struct msr_autoload *m = &vmx->msr_autoload;
1878
1879         switch (msr) {
1880         case MSR_EFER:
1881                 if (cpu_has_load_ia32_efer) {
1882                         add_atomic_switch_msr_special(vmx,
1883                                         VM_ENTRY_LOAD_IA32_EFER,
1884                                         VM_EXIT_LOAD_IA32_EFER,
1885                                         GUEST_IA32_EFER,
1886                                         HOST_IA32_EFER,
1887                                         guest_val, host_val);
1888                         return;
1889                 }
1890                 break;
1891         case MSR_CORE_PERF_GLOBAL_CTRL:
1892                 if (cpu_has_load_perf_global_ctrl) {
1893                         add_atomic_switch_msr_special(vmx,
1894                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1895                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1896                                         GUEST_IA32_PERF_GLOBAL_CTRL,
1897                                         HOST_IA32_PERF_GLOBAL_CTRL,
1898                                         guest_val, host_val);
1899                         return;
1900                 }
1901                 break;
1902         case MSR_IA32_PEBS_ENABLE:
1903                 /* PEBS needs a quiescent period after being disabled (to write
1904                  * a record).  Disabling PEBS through VMX MSR swapping doesn't
1905                  * provide that period, so a CPU could write host's record into
1906                  * guest's memory.
1907                  */
1908                 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1909         }
1910
1911         for (i = 0; i < m->nr; ++i)
1912                 if (m->guest[i].index == msr)
1913                         break;
1914
1915         if (i == NR_AUTOLOAD_MSRS) {
1916                 printk_once(KERN_WARNING "Not enough msr switch entries. "
1917                                 "Can't add msr %x\n", msr);
1918                 return;
1919         } else if (i == m->nr) {
1920                 ++m->nr;
1921                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1922                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1923         }
1924
1925         m->guest[i].index = msr;
1926         m->guest[i].value = guest_val;
1927         m->host[i].index = msr;
1928         m->host[i].value = host_val;
1929 }
1930
1931 static void reload_tss(void)
1932 {
1933         /*
1934          * VT restores TR but not its size.  Useless.
1935          */
1936         struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1937         struct desc_struct *descs;
1938
1939         descs = (void *)gdt->address;
1940         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1941         load_TR_desc();
1942 }
1943
1944 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1945 {
1946         u64 guest_efer = vmx->vcpu.arch.efer;
1947         u64 ignore_bits = 0;
1948
1949         if (!enable_ept) {
1950                 /*
1951                  * NX is needed to handle CR0.WP=1, CR4.SMEP=1.  Testing
1952                  * host CPUID is more efficient than testing guest CPUID
1953                  * or CR4.  Host SMEP is anyway a requirement for guest SMEP.
1954                  */
1955                 if (boot_cpu_has(X86_FEATURE_SMEP))
1956                         guest_efer |= EFER_NX;
1957                 else if (!(guest_efer & EFER_NX))
1958                         ignore_bits |= EFER_NX;
1959         }
1960
1961         /*
1962          * LMA and LME handled by hardware; SCE meaningless outside long mode.
1963          */
1964         ignore_bits |= EFER_SCE;
1965 #ifdef CONFIG_X86_64
1966         ignore_bits |= EFER_LMA | EFER_LME;
1967         /* SCE is meaningful only in long mode on Intel */
1968         if (guest_efer & EFER_LMA)
1969                 ignore_bits &= ~(u64)EFER_SCE;
1970 #endif
1971
1972         clear_atomic_switch_msr(vmx, MSR_EFER);
1973
1974         /*
1975          * On EPT, we can't emulate NX, so we must switch EFER atomically.
1976          * On CPUs that support "load IA32_EFER", always switch EFER
1977          * atomically, since it's faster than switching it manually.
1978          */
1979         if (cpu_has_load_ia32_efer ||
1980             (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1981                 if (!(guest_efer & EFER_LMA))
1982                         guest_efer &= ~EFER_LME;
1983                 if (guest_efer != host_efer)
1984                         add_atomic_switch_msr(vmx, MSR_EFER,
1985                                               guest_efer, host_efer);
1986                 return false;
1987         } else {
1988                 guest_efer &= ~ignore_bits;
1989                 guest_efer |= host_efer & ignore_bits;
1990
1991                 vmx->guest_msrs[efer_offset].data = guest_efer;
1992                 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1993
1994                 return true;
1995         }
1996 }
1997
1998 static unsigned long segment_base(u16 selector)
1999 {
2000         struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2001         struct desc_struct *d;
2002         unsigned long table_base;
2003         unsigned long v;
2004
2005         if (!(selector & ~3))
2006                 return 0;
2007
2008         table_base = gdt->address;
2009
2010         if (selector & 4) {           /* from ldt */
2011                 u16 ldt_selector = kvm_read_ldt();
2012
2013                 if (!(ldt_selector & ~3))
2014                         return 0;
2015
2016                 table_base = segment_base(ldt_selector);
2017         }
2018         d = (struct desc_struct *)(table_base + (selector & ~7));
2019         v = get_desc_base(d);
2020 #ifdef CONFIG_X86_64
2021        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
2022                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
2023 #endif
2024         return v;
2025 }
2026
2027 static inline unsigned long kvm_read_tr_base(void)
2028 {
2029         u16 tr;
2030         asm("str %0" : "=g"(tr));
2031         return segment_base(tr);
2032 }
2033
2034 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2035 {
2036         struct vcpu_vmx *vmx = to_vmx(vcpu);
2037         int i;
2038
2039         if (vmx->host_state.loaded)
2040                 return;
2041
2042         vmx->host_state.loaded = 1;
2043         /*
2044          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
2045          * allow segment selectors with cpl > 0 or ti == 1.
2046          */
2047         vmx->host_state.ldt_sel = kvm_read_ldt();
2048         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2049         savesegment(fs, vmx->host_state.fs_sel);
2050         if (!(vmx->host_state.fs_sel & 7)) {
2051                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2052                 vmx->host_state.fs_reload_needed = 0;
2053         } else {
2054                 vmcs_write16(HOST_FS_SELECTOR, 0);
2055                 vmx->host_state.fs_reload_needed = 1;
2056         }
2057         savesegment(gs, vmx->host_state.gs_sel);
2058         if (!(vmx->host_state.gs_sel & 7))
2059                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2060         else {
2061                 vmcs_write16(HOST_GS_SELECTOR, 0);
2062                 vmx->host_state.gs_ldt_reload_needed = 1;
2063         }
2064
2065 #ifdef CONFIG_X86_64
2066         savesegment(ds, vmx->host_state.ds_sel);
2067         savesegment(es, vmx->host_state.es_sel);
2068 #endif
2069
2070 #ifdef CONFIG_X86_64
2071         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2072         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2073 #else
2074         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2075         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2076 #endif
2077
2078 #ifdef CONFIG_X86_64
2079         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2080         if (is_long_mode(&vmx->vcpu))
2081                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2082 #endif
2083         if (boot_cpu_has(X86_FEATURE_MPX))
2084                 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2085         for (i = 0; i < vmx->save_nmsrs; ++i)
2086                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2087                                    vmx->guest_msrs[i].data,
2088                                    vmx->guest_msrs[i].mask);
2089 }
2090
2091 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2092 {
2093         if (!vmx->host_state.loaded)
2094                 return;
2095
2096         ++vmx->vcpu.stat.host_state_reload;
2097         vmx->host_state.loaded = 0;
2098 #ifdef CONFIG_X86_64
2099         if (is_long_mode(&vmx->vcpu))
2100                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2101 #endif
2102         if (vmx->host_state.gs_ldt_reload_needed) {
2103                 kvm_load_ldt(vmx->host_state.ldt_sel);
2104 #ifdef CONFIG_X86_64
2105                 load_gs_index(vmx->host_state.gs_sel);
2106 #else
2107                 loadsegment(gs, vmx->host_state.gs_sel);
2108 #endif
2109         }
2110         if (vmx->host_state.fs_reload_needed)
2111                 loadsegment(fs, vmx->host_state.fs_sel);
2112 #ifdef CONFIG_X86_64
2113         if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2114                 loadsegment(ds, vmx->host_state.ds_sel);
2115                 loadsegment(es, vmx->host_state.es_sel);
2116         }
2117 #endif
2118         reload_tss();
2119 #ifdef CONFIG_X86_64
2120         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2121 #endif
2122         if (vmx->host_state.msr_host_bndcfgs)
2123                 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2124         /*
2125          * If the FPU is not active (through the host task or
2126          * the guest vcpu), then restore the cr0.TS bit.
2127          */
2128         if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
2129                 stts();
2130         load_gdt(this_cpu_ptr(&host_gdt));
2131 }
2132
2133 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2134 {
2135         preempt_disable();
2136         __vmx_load_host_state(vmx);
2137         preempt_enable();
2138 }
2139
2140 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2141 {
2142         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2143         struct pi_desc old, new;
2144         unsigned int dest;
2145
2146         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2147                 !irq_remapping_cap(IRQ_POSTING_CAP))
2148                 return;
2149
2150         do {
2151                 old.control = new.control = pi_desc->control;
2152
2153                 /*
2154                  * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2155                  * are two possible cases:
2156                  * 1. After running 'pre_block', context switch
2157                  *    happened. For this case, 'sn' was set in
2158                  *    vmx_vcpu_put(), so we need to clear it here.
2159                  * 2. After running 'pre_block', we were blocked,
2160                  *    and woken up by some other guy. For this case,
2161                  *    we don't need to do anything, 'pi_post_block'
2162                  *    will do everything for us. However, we cannot
2163                  *    check whether it is case #1 or case #2 here
2164                  *    (maybe, not needed), so we also clear sn here,
2165                  *    I think it is not a big deal.
2166                  */
2167                 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2168                         if (vcpu->cpu != cpu) {
2169                                 dest = cpu_physical_id(cpu);
2170
2171                                 if (x2apic_enabled())
2172                                         new.ndst = dest;
2173                                 else
2174                                         new.ndst = (dest << 8) & 0xFF00;
2175                         }
2176
2177                         /* set 'NV' to 'notification vector' */
2178                         new.nv = POSTED_INTR_VECTOR;
2179                 }
2180
2181                 /* Allow posting non-urgent interrupts */
2182                 new.sn = 0;
2183         } while (cmpxchg(&pi_desc->control, old.control,
2184                         new.control) != old.control);
2185 }
2186
2187 /*
2188  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2189  * vcpu mutex is already taken.
2190  */
2191 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2192 {
2193         struct vcpu_vmx *vmx = to_vmx(vcpu);
2194         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2195
2196         if (!vmm_exclusive)
2197                 kvm_cpu_vmxon(phys_addr);
2198         else if (vmx->loaded_vmcs->cpu != cpu)
2199                 loaded_vmcs_clear(vmx->loaded_vmcs);
2200
2201         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2202                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2203                 vmcs_load(vmx->loaded_vmcs->vmcs);
2204         }
2205
2206         if (vmx->loaded_vmcs->cpu != cpu) {
2207                 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2208                 unsigned long sysenter_esp;
2209
2210                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2211                 local_irq_disable();
2212                 crash_disable_local_vmclear(cpu);
2213
2214                 /*
2215                  * Read loaded_vmcs->cpu should be before fetching
2216                  * loaded_vmcs->loaded_vmcss_on_cpu_link.
2217                  * See the comments in __loaded_vmcs_clear().
2218                  */
2219                 smp_rmb();
2220
2221                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2222                          &per_cpu(loaded_vmcss_on_cpu, cpu));
2223                 crash_enable_local_vmclear(cpu);
2224                 local_irq_enable();
2225
2226                 /*
2227                  * Linux uses per-cpu TSS and GDT, so set these when switching
2228                  * processors.
2229                  */
2230                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2231                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
2232
2233                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2234                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2235
2236                 vmx->loaded_vmcs->cpu = cpu;
2237         }
2238
2239         /* Setup TSC multiplier */
2240         if (kvm_has_tsc_control &&
2241             vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio) {
2242                 vmx->current_tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2243                 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2244         }
2245
2246         vmx_vcpu_pi_load(vcpu, cpu);
2247         vmx->host_pkru = read_pkru();
2248 }
2249
2250 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2251 {
2252         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2253
2254         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2255                 !irq_remapping_cap(IRQ_POSTING_CAP))
2256                 return;
2257
2258         /* Set SN when the vCPU is preempted */
2259         if (vcpu->preempted)
2260                 pi_set_sn(pi_desc);
2261 }
2262
2263 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2264 {
2265         vmx_vcpu_pi_put(vcpu);
2266
2267         __vmx_load_host_state(to_vmx(vcpu));
2268         if (!vmm_exclusive) {
2269                 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2270                 vcpu->cpu = -1;
2271                 kvm_cpu_vmxoff();
2272         }
2273 }
2274
2275 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
2276 {
2277         ulong cr0;
2278
2279         if (vcpu->fpu_active)
2280                 return;
2281         vcpu->fpu_active = 1;
2282         cr0 = vmcs_readl(GUEST_CR0);
2283         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
2284         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
2285         vmcs_writel(GUEST_CR0, cr0);
2286         update_exception_bitmap(vcpu);
2287         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
2288         if (is_guest_mode(vcpu))
2289                 vcpu->arch.cr0_guest_owned_bits &=
2290                         ~get_vmcs12(vcpu)->cr0_guest_host_mask;
2291         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2292 }
2293
2294 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2295
2296 /*
2297  * Return the cr0 value that a nested guest would read. This is a combination
2298  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2299  * its hypervisor (cr0_read_shadow).
2300  */
2301 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2302 {
2303         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2304                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2305 }
2306 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2307 {
2308         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2309                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2310 }
2311
2312 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
2313 {
2314         /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2315          * set this *before* calling this function.
2316          */
2317         vmx_decache_cr0_guest_bits(vcpu);
2318         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2319         update_exception_bitmap(vcpu);
2320         vcpu->arch.cr0_guest_owned_bits = 0;
2321         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2322         if (is_guest_mode(vcpu)) {
2323                 /*
2324                  * L1's specified read shadow might not contain the TS bit,
2325                  * so now that we turned on shadowing of this bit, we need to
2326                  * set this bit of the shadow. Like in nested_vmx_run we need
2327                  * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2328                  * up-to-date here because we just decached cr0.TS (and we'll
2329                  * only update vmcs12->guest_cr0 on nested exit).
2330                  */
2331                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2332                 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2333                         (vcpu->arch.cr0 & X86_CR0_TS);
2334                 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2335         } else
2336                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2337 }
2338
2339 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2340 {
2341         unsigned long rflags, save_rflags;
2342
2343         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2344                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2345                 rflags = vmcs_readl(GUEST_RFLAGS);
2346                 if (to_vmx(vcpu)->rmode.vm86_active) {
2347                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2348                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2349                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2350                 }
2351                 to_vmx(vcpu)->rflags = rflags;
2352         }
2353         return to_vmx(vcpu)->rflags;
2354 }
2355
2356 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2357 {
2358         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2359         to_vmx(vcpu)->rflags = rflags;
2360         if (to_vmx(vcpu)->rmode.vm86_active) {
2361                 to_vmx(vcpu)->rmode.save_rflags = rflags;
2362                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2363         }
2364         vmcs_writel(GUEST_RFLAGS, rflags);
2365 }
2366
2367 static u32 vmx_get_pkru(struct kvm_vcpu *vcpu)
2368 {
2369         return to_vmx(vcpu)->guest_pkru;
2370 }
2371
2372 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2373 {
2374         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2375         int ret = 0;
2376
2377         if (interruptibility & GUEST_INTR_STATE_STI)
2378                 ret |= KVM_X86_SHADOW_INT_STI;
2379         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2380                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2381
2382         return ret;
2383 }
2384
2385 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2386 {
2387         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2388         u32 interruptibility = interruptibility_old;
2389
2390         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2391
2392         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2393                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2394         else if (mask & KVM_X86_SHADOW_INT_STI)
2395                 interruptibility |= GUEST_INTR_STATE_STI;
2396
2397         if ((interruptibility != interruptibility_old))
2398                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2399 }
2400
2401 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2402 {
2403         unsigned long rip;
2404
2405         rip = kvm_rip_read(vcpu);
2406         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2407         kvm_rip_write(vcpu, rip);
2408
2409         /* skipping an emulated instruction also counts */
2410         vmx_set_interrupt_shadow(vcpu, 0);
2411 }
2412
2413 /*
2414  * KVM wants to inject page-faults which it got to the guest. This function
2415  * checks whether in a nested guest, we need to inject them to L1 or L2.
2416  */
2417 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2418 {
2419         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2420
2421         if (!(vmcs12->exception_bitmap & (1u << nr)))
2422                 return 0;
2423
2424         nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2425                           vmcs_read32(VM_EXIT_INTR_INFO),
2426                           vmcs_readl(EXIT_QUALIFICATION));
2427         return 1;
2428 }
2429
2430 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2431                                 bool has_error_code, u32 error_code,
2432                                 bool reinject)
2433 {
2434         struct vcpu_vmx *vmx = to_vmx(vcpu);
2435         u32 intr_info = nr | INTR_INFO_VALID_MASK;
2436
2437         if (!reinject && is_guest_mode(vcpu) &&
2438             nested_vmx_check_exception(vcpu, nr))
2439                 return;
2440
2441         if (has_error_code) {
2442                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2443                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2444         }
2445
2446         if (vmx->rmode.vm86_active) {
2447                 int inc_eip = 0;
2448                 if (kvm_exception_is_soft(nr))
2449                         inc_eip = vcpu->arch.event_exit_inst_len;
2450                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2451                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2452                 return;
2453         }
2454
2455         if (kvm_exception_is_soft(nr)) {
2456                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2457                              vmx->vcpu.arch.event_exit_inst_len);
2458                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2459         } else
2460                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2461
2462         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2463 }
2464
2465 static bool vmx_rdtscp_supported(void)
2466 {
2467         return cpu_has_vmx_rdtscp();
2468 }
2469
2470 static bool vmx_invpcid_supported(void)
2471 {
2472         return cpu_has_vmx_invpcid() && enable_ept;
2473 }
2474
2475 /*
2476  * Swap MSR entry in host/guest MSR entry array.
2477  */
2478 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2479 {
2480         struct shared_msr_entry tmp;
2481
2482         tmp = vmx->guest_msrs[to];
2483         vmx->guest_msrs[to] = vmx->guest_msrs[from];
2484         vmx->guest_msrs[from] = tmp;
2485 }
2486
2487 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2488 {
2489         unsigned long *msr_bitmap;
2490
2491         if (is_guest_mode(vcpu))
2492                 msr_bitmap = vmx_msr_bitmap_nested;
2493         else if (cpu_has_secondary_exec_ctrls() &&
2494                  (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2495                   SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2496                 if (is_long_mode(vcpu))
2497                         msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2498                 else
2499                         msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2500         } else {
2501                 if (is_long_mode(vcpu))
2502                         msr_bitmap = vmx_msr_bitmap_longmode;
2503                 else
2504                         msr_bitmap = vmx_msr_bitmap_legacy;
2505         }
2506
2507         vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2508 }
2509
2510 /*
2511  * Set up the vmcs to automatically save and restore system
2512  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
2513  * mode, as fiddling with msrs is very expensive.
2514  */
2515 static void setup_msrs(struct vcpu_vmx *vmx)
2516 {
2517         int save_nmsrs, index;
2518
2519         save_nmsrs = 0;
2520 #ifdef CONFIG_X86_64
2521         if (is_long_mode(&vmx->vcpu)) {
2522                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2523                 if (index >= 0)
2524                         move_msr_up(vmx, index, save_nmsrs++);
2525                 index = __find_msr_index(vmx, MSR_LSTAR);
2526                 if (index >= 0)
2527                         move_msr_up(vmx, index, save_nmsrs++);
2528                 index = __find_msr_index(vmx, MSR_CSTAR);
2529                 if (index >= 0)
2530                         move_msr_up(vmx, index, save_nmsrs++);
2531                 index = __find_msr_index(vmx, MSR_TSC_AUX);
2532                 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2533                         move_msr_up(vmx, index, save_nmsrs++);
2534                 /*
2535                  * MSR_STAR is only needed on long mode guests, and only
2536                  * if efer.sce is enabled.
2537                  */
2538                 index = __find_msr_index(vmx, MSR_STAR);
2539                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2540                         move_msr_up(vmx, index, save_nmsrs++);
2541         }
2542 #endif
2543         index = __find_msr_index(vmx, MSR_EFER);
2544         if (index >= 0 && update_transition_efer(vmx, index))
2545                 move_msr_up(vmx, index, save_nmsrs++);
2546
2547         vmx->save_nmsrs = save_nmsrs;
2548
2549         if (cpu_has_vmx_msr_bitmap())
2550                 vmx_set_msr_bitmap(&vmx->vcpu);
2551 }
2552
2553 /*
2554  * reads and returns guest's timestamp counter "register"
2555  * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2556  * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2557  */
2558 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2559 {
2560         u64 host_tsc, tsc_offset;
2561
2562         host_tsc = rdtsc();
2563         tsc_offset = vmcs_read64(TSC_OFFSET);
2564         return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2565 }
2566
2567 /*
2568  * Like guest_read_tsc, but always returns L1's notion of the timestamp
2569  * counter, even if a nested guest (L2) is currently running.
2570  */
2571 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2572 {
2573         u64 tsc_offset;
2574
2575         tsc_offset = is_guest_mode(vcpu) ?
2576                 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2577                 vmcs_read64(TSC_OFFSET);
2578         return host_tsc + tsc_offset;
2579 }
2580
2581 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2582 {
2583         return vmcs_read64(TSC_OFFSET);
2584 }
2585
2586 /*
2587  * writes 'offset' into guest's timestamp counter offset register
2588  */
2589 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2590 {
2591         if (is_guest_mode(vcpu)) {
2592                 /*
2593                  * We're here if L1 chose not to trap WRMSR to TSC. According
2594                  * to the spec, this should set L1's TSC; The offset that L1
2595                  * set for L2 remains unchanged, and still needs to be added
2596                  * to the newly set TSC to get L2's TSC.
2597                  */
2598                 struct vmcs12 *vmcs12;
2599                 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2600                 /* recalculate vmcs02.TSC_OFFSET: */
2601                 vmcs12 = get_vmcs12(vcpu);
2602                 vmcs_write64(TSC_OFFSET, offset +
2603                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2604                          vmcs12->tsc_offset : 0));
2605         } else {
2606                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2607                                            vmcs_read64(TSC_OFFSET), offset);
2608                 vmcs_write64(TSC_OFFSET, offset);
2609         }
2610 }
2611
2612 static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
2613 {
2614         u64 offset = vmcs_read64(TSC_OFFSET);
2615
2616         vmcs_write64(TSC_OFFSET, offset + adjustment);
2617         if (is_guest_mode(vcpu)) {
2618                 /* Even when running L2, the adjustment needs to apply to L1 */
2619                 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2620         } else
2621                 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2622                                            offset + adjustment);
2623 }
2624
2625 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2626 {
2627         struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2628         return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2629 }
2630
2631 /*
2632  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2633  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2634  * all guests if the "nested" module option is off, and can also be disabled
2635  * for a single guest by disabling its VMX cpuid bit.
2636  */
2637 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2638 {
2639         return nested && guest_cpuid_has_vmx(vcpu);
2640 }
2641
2642 /*
2643  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2644  * returned for the various VMX controls MSRs when nested VMX is enabled.
2645  * The same values should also be used to verify that vmcs12 control fields are
2646  * valid during nested entry from L1 to L2.
2647  * Each of these control msrs has a low and high 32-bit half: A low bit is on
2648  * if the corresponding bit in the (32-bit) control field *must* be on, and a
2649  * bit in the high half is on if the corresponding bit in the control field
2650  * may be on. See also vmx_control_verify().
2651  */
2652 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2653 {
2654         /*
2655          * Note that as a general rule, the high half of the MSRs (bits in
2656          * the control fields which may be 1) should be initialized by the
2657          * intersection of the underlying hardware's MSR (i.e., features which
2658          * can be supported) and the list of features we want to expose -
2659          * because they are known to be properly supported in our code.
2660          * Also, usually, the low half of the MSRs (bits which must be 1) can
2661          * be set to 0, meaning that L1 may turn off any of these bits. The
2662          * reason is that if one of these bits is necessary, it will appear
2663          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2664          * fields of vmcs01 and vmcs02, will turn these bits off - and
2665          * nested_vmx_exit_handled() will not pass related exits to L1.
2666          * These rules have exceptions below.
2667          */
2668
2669         /* pin-based controls */
2670         rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2671                 vmx->nested.nested_vmx_pinbased_ctls_low,
2672                 vmx->nested.nested_vmx_pinbased_ctls_high);
2673         vmx->nested.nested_vmx_pinbased_ctls_low |=
2674                 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2675         vmx->nested.nested_vmx_pinbased_ctls_high &=
2676                 PIN_BASED_EXT_INTR_MASK |
2677                 PIN_BASED_NMI_EXITING |
2678                 PIN_BASED_VIRTUAL_NMIS;
2679         vmx->nested.nested_vmx_pinbased_ctls_high |=
2680                 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2681                 PIN_BASED_VMX_PREEMPTION_TIMER;
2682         if (kvm_vcpu_apicv_active(&vmx->vcpu))
2683                 vmx->nested.nested_vmx_pinbased_ctls_high |=
2684                         PIN_BASED_POSTED_INTR;
2685
2686         /* exit controls */
2687         rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2688                 vmx->nested.nested_vmx_exit_ctls_low,
2689                 vmx->nested.nested_vmx_exit_ctls_high);
2690         vmx->nested.nested_vmx_exit_ctls_low =
2691                 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2692
2693         vmx->nested.nested_vmx_exit_ctls_high &=
2694 #ifdef CONFIG_X86_64
2695                 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2696 #endif
2697                 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2698         vmx->nested.nested_vmx_exit_ctls_high |=
2699                 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2700                 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2701                 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2702
2703         if (kvm_mpx_supported())
2704                 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2705
2706         /* We support free control of debug control saving. */
2707         vmx->nested.nested_vmx_true_exit_ctls_low =
2708                 vmx->nested.nested_vmx_exit_ctls_low &
2709                 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2710
2711         /* entry controls */
2712         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2713                 vmx->nested.nested_vmx_entry_ctls_low,
2714                 vmx->nested.nested_vmx_entry_ctls_high);
2715         vmx->nested.nested_vmx_entry_ctls_low =
2716                 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2717         vmx->nested.nested_vmx_entry_ctls_high &=
2718 #ifdef CONFIG_X86_64
2719                 VM_ENTRY_IA32E_MODE |
2720 #endif
2721                 VM_ENTRY_LOAD_IA32_PAT;
2722         vmx->nested.nested_vmx_entry_ctls_high |=
2723                 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2724         if (kvm_mpx_supported())
2725                 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2726
2727         /* We support free control of debug control loading. */
2728         vmx->nested.nested_vmx_true_entry_ctls_low =
2729                 vmx->nested.nested_vmx_entry_ctls_low &
2730                 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2731
2732         /* cpu-based controls */
2733         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2734                 vmx->nested.nested_vmx_procbased_ctls_low,
2735                 vmx->nested.nested_vmx_procbased_ctls_high);
2736         vmx->nested.nested_vmx_procbased_ctls_low =
2737                 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2738         vmx->nested.nested_vmx_procbased_ctls_high &=
2739                 CPU_BASED_VIRTUAL_INTR_PENDING |
2740                 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2741                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2742                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2743                 CPU_BASED_CR3_STORE_EXITING |
2744 #ifdef CONFIG_X86_64
2745                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2746 #endif
2747                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2748                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2749                 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2750                 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2751                 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2752         /*
2753          * We can allow some features even when not supported by the
2754          * hardware. For example, L1 can specify an MSR bitmap - and we
2755          * can use it to avoid exits to L1 - even when L0 runs L2
2756          * without MSR bitmaps.
2757          */
2758         vmx->nested.nested_vmx_procbased_ctls_high |=
2759                 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2760                 CPU_BASED_USE_MSR_BITMAPS;
2761
2762         /* We support free control of CR3 access interception. */
2763         vmx->nested.nested_vmx_true_procbased_ctls_low =
2764                 vmx->nested.nested_vmx_procbased_ctls_low &
2765                 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2766
2767         /* secondary cpu-based controls */
2768         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2769                 vmx->nested.nested_vmx_secondary_ctls_low,
2770                 vmx->nested.nested_vmx_secondary_ctls_high);
2771         vmx->nested.nested_vmx_secondary_ctls_low = 0;
2772         vmx->nested.nested_vmx_secondary_ctls_high &=
2773                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2774                 SECONDARY_EXEC_RDTSCP |
2775                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2776                 SECONDARY_EXEC_ENABLE_VPID |
2777                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2778                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2779                 SECONDARY_EXEC_WBINVD_EXITING |
2780                 SECONDARY_EXEC_XSAVES |
2781                 SECONDARY_EXEC_PCOMMIT;
2782
2783         if (enable_ept) {
2784                 /* nested EPT: emulate EPT also to L1 */
2785                 vmx->nested.nested_vmx_secondary_ctls_high |=
2786                         SECONDARY_EXEC_ENABLE_EPT;
2787                 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2788                          VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2789                          VMX_EPT_INVEPT_BIT;
2790                 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2791                 /*
2792                  * For nested guests, we don't do anything specific
2793                  * for single context invalidation. Hence, only advertise
2794                  * support for global context invalidation.
2795                  */
2796                 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2797         } else
2798                 vmx->nested.nested_vmx_ept_caps = 0;
2799
2800         /*
2801          * Old versions of KVM use the single-context version without
2802          * checking for support, so declare that it is supported even
2803          * though it is treated as global context.  The alternative is
2804          * not failing the single-context invvpid, and it is worse.
2805          */
2806         if (enable_vpid)
2807                 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2808                                 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |
2809                                 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
2810         else
2811                 vmx->nested.nested_vmx_vpid_caps = 0;
2812
2813         if (enable_unrestricted_guest)
2814                 vmx->nested.nested_vmx_secondary_ctls_high |=
2815                         SECONDARY_EXEC_UNRESTRICTED_GUEST;
2816
2817         /* miscellaneous data */
2818         rdmsr(MSR_IA32_VMX_MISC,
2819                 vmx->nested.nested_vmx_misc_low,
2820                 vmx->nested.nested_vmx_misc_high);
2821         vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2822         vmx->nested.nested_vmx_misc_low |=
2823                 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2824                 VMX_MISC_ACTIVITY_HLT;
2825         vmx->nested.nested_vmx_misc_high = 0;
2826 }
2827
2828 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2829 {
2830         /*
2831          * Bits 0 in high must be 0, and bits 1 in low must be 1.
2832          */
2833         return ((control & high) | low) == control;
2834 }
2835
2836 static inline u64 vmx_control_msr(u32 low, u32 high)
2837 {
2838         return low | ((u64)high << 32);
2839 }
2840
2841 /* Returns 0 on success, non-0 otherwise. */
2842 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2843 {
2844         struct vcpu_vmx *vmx = to_vmx(vcpu);
2845
2846         switch (msr_index) {
2847         case MSR_IA32_VMX_BASIC:
2848                 /*
2849                  * This MSR reports some information about VMX support. We
2850                  * should return information about the VMX we emulate for the
2851                  * guest, and the VMCS structure we give it - not about the
2852                  * VMX support of the underlying hardware.
2853                  */
2854                 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2855                            ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2856                            (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2857                 break;
2858         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2859         case MSR_IA32_VMX_PINBASED_CTLS:
2860                 *pdata = vmx_control_msr(
2861                         vmx->nested.nested_vmx_pinbased_ctls_low,
2862                         vmx->nested.nested_vmx_pinbased_ctls_high);
2863                 break;
2864         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2865                 *pdata = vmx_control_msr(
2866                         vmx->nested.nested_vmx_true_procbased_ctls_low,
2867                         vmx->nested.nested_vmx_procbased_ctls_high);
2868                 break;
2869         case MSR_IA32_VMX_PROCBASED_CTLS:
2870                 *pdata = vmx_control_msr(
2871                         vmx->nested.nested_vmx_procbased_ctls_low,
2872                         vmx->nested.nested_vmx_procbased_ctls_high);
2873                 break;
2874         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2875                 *pdata = vmx_control_msr(
2876                         vmx->nested.nested_vmx_true_exit_ctls_low,
2877                         vmx->nested.nested_vmx_exit_ctls_high);
2878                 break;
2879         case MSR_IA32_VMX_EXIT_CTLS:
2880                 *pdata = vmx_control_msr(
2881                         vmx->nested.nested_vmx_exit_ctls_low,
2882                         vmx->nested.nested_vmx_exit_ctls_high);
2883                 break;
2884         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2885                 *pdata = vmx_control_msr(
2886                         vmx->nested.nested_vmx_true_entry_ctls_low,
2887                         vmx->nested.nested_vmx_entry_ctls_high);
2888                 break;
2889         case MSR_IA32_VMX_ENTRY_CTLS:
2890                 *pdata = vmx_control_msr(
2891                         vmx->nested.nested_vmx_entry_ctls_low,
2892                         vmx->nested.nested_vmx_entry_ctls_high);
2893                 break;
2894         case MSR_IA32_VMX_MISC:
2895                 *pdata = vmx_control_msr(
2896                         vmx->nested.nested_vmx_misc_low,
2897                         vmx->nested.nested_vmx_misc_high);
2898                 break;
2899         /*
2900          * These MSRs specify bits which the guest must keep fixed (on or off)
2901          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2902          * We picked the standard core2 setting.
2903          */
2904 #define VMXON_CR0_ALWAYSON      (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2905 #define VMXON_CR4_ALWAYSON      X86_CR4_VMXE
2906         case MSR_IA32_VMX_CR0_FIXED0:
2907                 *pdata = VMXON_CR0_ALWAYSON;
2908                 break;
2909         case MSR_IA32_VMX_CR0_FIXED1:
2910                 *pdata = -1ULL;
2911                 break;
2912         case MSR_IA32_VMX_CR4_FIXED0:
2913                 *pdata = VMXON_CR4_ALWAYSON;
2914                 break;
2915         case MSR_IA32_VMX_CR4_FIXED1:
2916                 *pdata = -1ULL;
2917                 break;
2918         case MSR_IA32_VMX_VMCS_ENUM:
2919                 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2920                 break;
2921         case MSR_IA32_VMX_PROCBASED_CTLS2:
2922                 *pdata = vmx_control_msr(
2923                         vmx->nested.nested_vmx_secondary_ctls_low,
2924                         vmx->nested.nested_vmx_secondary_ctls_high);
2925                 break;
2926         case MSR_IA32_VMX_EPT_VPID_CAP:
2927                 /* Currently, no nested vpid support */
2928                 *pdata = vmx->nested.nested_vmx_ept_caps |
2929                         ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
2930                 break;
2931         default:
2932                 return 1;
2933         }
2934
2935         return 0;
2936 }
2937
2938 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
2939                                                  uint64_t val)
2940 {
2941         uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
2942
2943         return !(val & ~valid_bits);
2944 }
2945
2946 /*
2947  * Reads an msr value (of 'msr_index') into 'pdata'.
2948  * Returns 0 on success, non-0 otherwise.
2949  * Assumes vcpu_load() was already called.
2950  */
2951 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2952 {
2953         struct shared_msr_entry *msr;
2954
2955         switch (msr_info->index) {
2956 #ifdef CONFIG_X86_64
2957         case MSR_FS_BASE:
2958                 msr_info->data = vmcs_readl(GUEST_FS_BASE);
2959                 break;
2960         case MSR_GS_BASE:
2961                 msr_info->data = vmcs_readl(GUEST_GS_BASE);
2962                 break;
2963         case MSR_KERNEL_GS_BASE:
2964                 vmx_load_host_state(to_vmx(vcpu));
2965                 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2966                 break;
2967 #endif
2968         case MSR_EFER:
2969                 return kvm_get_msr_common(vcpu, msr_info);
2970         case MSR_IA32_TSC:
2971                 msr_info->data = guest_read_tsc(vcpu);
2972                 break;
2973         case MSR_IA32_SYSENTER_CS:
2974                 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2975                 break;
2976         case MSR_IA32_SYSENTER_EIP:
2977                 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2978                 break;
2979         case MSR_IA32_SYSENTER_ESP:
2980                 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2981                 break;
2982         case MSR_IA32_BNDCFGS:
2983                 if (!kvm_mpx_supported())
2984                         return 1;
2985                 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2986                 break;
2987         case MSR_IA32_MCG_EXT_CTL:
2988                 if (!msr_info->host_initiated &&
2989                     !(to_vmx(vcpu)->msr_ia32_feature_control &
2990                       FEATURE_CONTROL_LMCE))
2991                         return 1;
2992                 msr_info->data = vcpu->arch.mcg_ext_ctl;
2993                 break;
2994         case MSR_IA32_FEATURE_CONTROL:
2995                 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
2996                 break;
2997         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2998                 if (!nested_vmx_allowed(vcpu))
2999                         return 1;
3000                 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3001         case MSR_IA32_XSS:
3002                 if (!vmx_xsaves_supported())
3003                         return 1;
3004                 msr_info->data = vcpu->arch.ia32_xss;
3005                 break;
3006         case MSR_TSC_AUX:
3007                 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3008                         return 1;
3009                 /* Otherwise falls through */
3010         default:
3011                 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3012                 if (msr) {
3013                         msr_info->data = msr->data;
3014                         break;
3015                 }
3016                 return kvm_get_msr_common(vcpu, msr_info);
3017         }
3018
3019         return 0;
3020 }
3021
3022 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3023
3024 /*
3025  * Writes msr value into into the appropriate "register".
3026  * Returns 0 on success, non-0 otherwise.
3027  * Assumes vcpu_load() was already called.
3028  */
3029 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3030 {
3031         struct vcpu_vmx *vmx = to_vmx(vcpu);
3032         struct shared_msr_entry *msr;
3033         int ret = 0;
3034         u32 msr_index = msr_info->index;
3035         u64 data = msr_info->data;
3036
3037         switch (msr_index) {
3038         case MSR_EFER:
3039                 ret = kvm_set_msr_common(vcpu, msr_info);
3040                 break;
3041 #ifdef CONFIG_X86_64
3042         case MSR_FS_BASE:
3043                 vmx_segment_cache_clear(vmx);
3044                 vmcs_writel(GUEST_FS_BASE, data);
3045                 break;
3046         case MSR_GS_BASE:
3047                 vmx_segment_cache_clear(vmx);
3048                 vmcs_writel(GUEST_GS_BASE, data);
3049                 break;
3050         case MSR_KERNEL_GS_BASE:
3051                 vmx_load_host_state(vmx);
3052                 vmx->msr_guest_kernel_gs_base = data;
3053                 break;
3054 #endif
3055         case MSR_IA32_SYSENTER_CS:
3056                 vmcs_write32(GUEST_SYSENTER_CS, data);
3057                 break;
3058         case MSR_IA32_SYSENTER_EIP:
3059                 vmcs_writel(GUEST_SYSENTER_EIP, data);
3060                 break;
3061         case MSR_IA32_SYSENTER_ESP:
3062                 vmcs_writel(GUEST_SYSENTER_ESP, data);
3063                 break;
3064         case MSR_IA32_BNDCFGS:
3065                 if (!kvm_mpx_supported())
3066                         return 1;
3067                 vmcs_write64(GUEST_BNDCFGS, data);
3068                 break;
3069         case MSR_IA32_TSC:
3070                 kvm_write_tsc(vcpu, msr_info);
3071                 break;
3072         case MSR_IA32_CR_PAT:
3073                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3074                         if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3075                                 return 1;
3076                         vmcs_write64(GUEST_IA32_PAT, data);
3077                         vcpu->arch.pat = data;
3078                         break;
3079                 }
3080                 ret = kvm_set_msr_common(vcpu, msr_info);
3081                 break;
3082         case MSR_IA32_TSC_ADJUST:
3083                 ret = kvm_set_msr_common(vcpu, msr_info);
3084                 break;
3085         case MSR_IA32_MCG_EXT_CTL:
3086                 if ((!msr_info->host_initiated &&
3087                      !(to_vmx(vcpu)->msr_ia32_feature_control &
3088                        FEATURE_CONTROL_LMCE)) ||
3089                     (data & ~MCG_EXT_CTL_LMCE_EN))
3090                         return 1;
3091                 vcpu->arch.mcg_ext_ctl = data;
3092                 break;
3093         case MSR_IA32_FEATURE_CONTROL:
3094                 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3095                     (to_vmx(vcpu)->msr_ia32_feature_control &
3096                      FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3097                         return 1;
3098                 vmx->msr_ia32_feature_control = data;
3099                 if (msr_info->host_initiated && data == 0)
3100                         vmx_leave_nested(vcpu);
3101                 break;
3102         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3103                 return 1; /* they are read-only */
3104         case MSR_IA32_XSS:
3105                 if (!vmx_xsaves_supported())
3106                         return 1;
3107                 /*
3108                  * The only supported bit as of Skylake is bit 8, but
3109                  * it is not supported on KVM.
3110                  */
3111                 if (data != 0)
3112                         return 1;
3113                 vcpu->arch.ia32_xss = data;
3114                 if (vcpu->arch.ia32_xss != host_xss)
3115                         add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3116                                 vcpu->arch.ia32_xss, host_xss);
3117                 else
3118                         clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3119                 break;
3120         case MSR_TSC_AUX:
3121                 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3122                         return 1;
3123                 /* Check reserved bit, higher 32 bits should be zero */
3124                 if ((data >> 32) != 0)
3125                         return 1;
3126                 /* Otherwise falls through */
3127         default:
3128                 msr = find_msr_entry(vmx, msr_index);
3129                 if (msr) {
3130                         u64 old_msr_data = msr->data;
3131                         msr->data = data;
3132                         if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3133                                 preempt_disable();
3134                                 ret = kvm_set_shared_msr(msr->index, msr->data,
3135                                                          msr->mask);
3136                                 preempt_enable();
3137                                 if (ret)
3138                                         msr->data = old_msr_data;
3139                         }
3140                         break;
3141                 }
3142                 ret = kvm_set_msr_common(vcpu, msr_info);
3143         }
3144
3145         return ret;
3146 }
3147
3148 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3149 {
3150         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3151         switch (reg) {
3152         case VCPU_REGS_RSP:
3153                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3154                 break;
3155         case VCPU_REGS_RIP:
3156                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3157                 break;
3158         case VCPU_EXREG_PDPTR:
3159                 if (enable_ept)
3160                         ept_save_pdptrs(vcpu);
3161                 break;
3162         default:
3163                 break;
3164         }
3165 }
3166
3167 static __init int cpu_has_kvm_support(void)
3168 {
3169         return cpu_has_vmx();
3170 }
3171
3172 static __init int vmx_disabled_by_bios(void)
3173 {
3174         u64 msr;
3175
3176         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3177         if (msr & FEATURE_CONTROL_LOCKED) {
3178                 /* launched w/ TXT and VMX disabled */
3179                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3180                         && tboot_enabled())
3181                         return 1;
3182                 /* launched w/o TXT and VMX only enabled w/ TXT */
3183                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3184                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3185                         && !tboot_enabled()) {
3186                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3187                                 "activate TXT before enabling KVM\n");
3188                         return 1;
3189                 }
3190                 /* launched w/o TXT and VMX disabled */
3191                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3192                         && !tboot_enabled())
3193                         return 1;
3194         }
3195
3196         return 0;
3197 }
3198
3199 static void kvm_cpu_vmxon(u64 addr)
3200 {
3201         intel_pt_handle_vmx(1);
3202
3203         asm volatile (ASM_VMX_VMXON_RAX
3204                         : : "a"(&addr), "m"(addr)
3205                         : "memory", "cc");
3206 }
3207
3208 static int hardware_enable(void)
3209 {
3210         int cpu = raw_smp_processor_id();
3211         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3212         u64 old, test_bits;
3213
3214         if (cr4_read_shadow() & X86_CR4_VMXE)
3215                 return -EBUSY;
3216
3217         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3218         INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3219         spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3220
3221         /*
3222          * Now we can enable the vmclear operation in kdump
3223          * since the loaded_vmcss_on_cpu list on this cpu
3224          * has been initialized.
3225          *
3226          * Though the cpu is not in VMX operation now, there
3227          * is no problem to enable the vmclear operation
3228          * for the loaded_vmcss_on_cpu list is empty!
3229          */
3230         crash_enable_local_vmclear(cpu);
3231
3232         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3233
3234         test_bits = FEATURE_CONTROL_LOCKED;
3235         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3236         if (tboot_enabled())
3237                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3238
3239         if ((old & test_bits) != test_bits) {
3240                 /* enable and lock */
3241                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3242         }
3243         cr4_set_bits(X86_CR4_VMXE);
3244
3245         if (vmm_exclusive) {
3246                 kvm_cpu_vmxon(phys_addr);
3247                 ept_sync_global();
3248         }
3249
3250         native_store_gdt(this_cpu_ptr(&host_gdt));
3251
3252         return 0;
3253 }
3254
3255 static void vmclear_local_loaded_vmcss(void)
3256 {
3257         int cpu = raw_smp_processor_id();
3258         struct loaded_vmcs *v, *n;
3259
3260         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3261                                  loaded_vmcss_on_cpu_link)
3262                 __loaded_vmcs_clear(v);
3263 }
3264
3265
3266 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3267  * tricks.
3268  */
3269 static void kvm_cpu_vmxoff(void)
3270 {
3271         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3272
3273         intel_pt_handle_vmx(0);
3274 }
3275
3276 static void hardware_disable(void)
3277 {
3278         if (vmm_exclusive) {
3279                 vmclear_local_loaded_vmcss();
3280                 kvm_cpu_vmxoff();
3281         }
3282         cr4_clear_bits(X86_CR4_VMXE);
3283 }
3284
3285 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3286                                       u32 msr, u32 *result)
3287 {
3288         u32 vmx_msr_low, vmx_msr_high;
3289         u32 ctl = ctl_min | ctl_opt;
3290
3291         rdmsr(msr, vmx_msr_low, vmx_msr_high);
3292
3293         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3294         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
3295
3296         /* Ensure minimum (required) set of control bits are supported. */
3297         if (ctl_min & ~ctl)
3298                 return -EIO;
3299
3300         *result = ctl;
3301         return 0;
3302 }
3303
3304 static __init bool allow_1_setting(u32 msr, u32 ctl)
3305 {
3306         u32 vmx_msr_low, vmx_msr_high;
3307
3308         rdmsr(msr, vmx_msr_low, vmx_msr_high);
3309         return vmx_msr_high & ctl;
3310 }
3311
3312 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3313 {
3314         u32 vmx_msr_low, vmx_msr_high;
3315         u32 min, opt, min2, opt2;
3316         u32 _pin_based_exec_control = 0;
3317         u32 _cpu_based_exec_control = 0;
3318         u32 _cpu_based_2nd_exec_control = 0;
3319         u32 _vmexit_control = 0;
3320         u32 _vmentry_control = 0;
3321
3322         min = CPU_BASED_HLT_EXITING |
3323 #ifdef CONFIG_X86_64
3324               CPU_BASED_CR8_LOAD_EXITING |
3325               CPU_BASED_CR8_STORE_EXITING |
3326 #endif
3327               CPU_BASED_CR3_LOAD_EXITING |
3328               CPU_BASED_CR3_STORE_EXITING |
3329               CPU_BASED_USE_IO_BITMAPS |
3330               CPU_BASED_MOV_DR_EXITING |
3331               CPU_BASED_USE_TSC_OFFSETING |
3332               CPU_BASED_MWAIT_EXITING |
3333               CPU_BASED_MONITOR_EXITING |
3334               CPU_BASED_INVLPG_EXITING |
3335               CPU_BASED_RDPMC_EXITING;
3336
3337         opt = CPU_BASED_TPR_SHADOW |
3338               CPU_BASED_USE_MSR_BITMAPS |
3339               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3340         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3341                                 &_cpu_based_exec_control) < 0)
3342                 return -EIO;
3343 #ifdef CONFIG_X86_64
3344         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3345                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3346                                            ~CPU_BASED_CR8_STORE_EXITING;
3347 #endif
3348         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3349                 min2 = 0;
3350                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3351                         SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3352                         SECONDARY_EXEC_WBINVD_EXITING |
3353                         SECONDARY_EXEC_ENABLE_VPID |
3354                         SECONDARY_EXEC_ENABLE_EPT |
3355                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
3356                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3357                         SECONDARY_EXEC_RDTSCP |
3358                         SECONDARY_EXEC_ENABLE_INVPCID |
3359                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
3360                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3361                         SECONDARY_EXEC_SHADOW_VMCS |
3362                         SECONDARY_EXEC_XSAVES |
3363                         SECONDARY_EXEC_ENABLE_PML |
3364                         SECONDARY_EXEC_PCOMMIT |
3365                         SECONDARY_EXEC_TSC_SCALING;
3366                 if (adjust_vmx_controls(min2, opt2,
3367                                         MSR_IA32_VMX_PROCBASED_CTLS2,
3368                                         &_cpu_based_2nd_exec_control) < 0)
3369                         return -EIO;
3370         }
3371 #ifndef CONFIG_X86_64
3372         if (!(_cpu_based_2nd_exec_control &
3373                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3374                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3375 #endif
3376
3377         if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3378                 _cpu_based_2nd_exec_control &= ~(
3379                                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3380                                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3381                                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3382
3383         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3384                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3385                    enabled */
3386                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3387                                              CPU_BASED_CR3_STORE_EXITING |
3388                                              CPU_BASED_INVLPG_EXITING);
3389                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3390                       vmx_capability.ept, vmx_capability.vpid);
3391         }
3392
3393         min = VM_EXIT_SAVE_DEBUG_CONTROLS;
3394 #ifdef CONFIG_X86_64
3395         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3396 #endif
3397         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3398                 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
3399         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3400                                 &_vmexit_control) < 0)
3401                 return -EIO;
3402
3403         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3404         opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
3405                  PIN_BASED_VMX_PREEMPTION_TIMER;
3406         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3407                                 &_pin_based_exec_control) < 0)
3408                 return -EIO;
3409
3410         if (!(_cpu_based_2nd_exec_control &
3411                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
3412                 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
3413                 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3414
3415         min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3416         opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3417         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3418                                 &_vmentry_control) < 0)
3419                 return -EIO;
3420
3421         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3422
3423         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3424         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3425                 return -EIO;
3426
3427 #ifdef CONFIG_X86_64
3428         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3429         if (vmx_msr_high & (1u<<16))
3430                 return -EIO;
3431 #endif
3432
3433         /* Require Write-Back (WB) memory type for VMCS accesses. */
3434         if (((vmx_msr_high >> 18) & 15) != 6)
3435                 return -EIO;
3436
3437         vmcs_conf->size = vmx_msr_high & 0x1fff;
3438         vmcs_conf->order = get_order(vmcs_config.size);
3439         vmcs_conf->revision_id = vmx_msr_low;
3440
3441         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3442         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3443         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3444         vmcs_conf->vmexit_ctrl         = _vmexit_control;
3445         vmcs_conf->vmentry_ctrl        = _vmentry_control;
3446
3447         cpu_has_load_ia32_efer =
3448                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3449                                 VM_ENTRY_LOAD_IA32_EFER)
3450                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3451                                    VM_EXIT_LOAD_IA32_EFER);
3452
3453         cpu_has_load_perf_global_ctrl =
3454                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3455                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3456                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3457                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3458
3459         /*
3460          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3461          * but due to errata below it can't be used. Workaround is to use
3462          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3463          *
3464          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3465          *
3466          * AAK155             (model 26)
3467          * AAP115             (model 30)
3468          * AAT100             (model 37)
3469          * BC86,AAY89,BD102   (model 44)
3470          * BA97               (model 46)
3471          *
3472          */
3473         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3474                 switch (boot_cpu_data.x86_model) {
3475                 case 26:
3476                 case 30:
3477                 case 37:
3478                 case 44:
3479                 case 46:
3480                         cpu_has_load_perf_global_ctrl = false;
3481                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3482                                         "does not work properly. Using workaround\n");
3483                         break;
3484                 default:
3485                         break;
3486                 }
3487         }
3488
3489         if (boot_cpu_has(X86_FEATURE_XSAVES))
3490                 rdmsrl(MSR_IA32_XSS, host_xss);
3491
3492         return 0;
3493 }
3494
3495 static struct vmcs *alloc_vmcs_cpu(int cpu)
3496 {
3497         int node = cpu_to_node(cpu);
3498         struct page *pages;
3499         struct vmcs *vmcs;
3500
3501         pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3502         if (!pages)
3503                 return NULL;
3504         vmcs = page_address(pages);
3505         memset(vmcs, 0, vmcs_config.size);
3506         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3507         return vmcs;
3508 }
3509
3510 static struct vmcs *alloc_vmcs(void)
3511 {
3512         return alloc_vmcs_cpu(raw_smp_processor_id());
3513 }
3514
3515 static void free_vmcs(struct vmcs *vmcs)
3516 {
3517         free_pages((unsigned long)vmcs, vmcs_config.order);
3518 }
3519
3520 /*
3521  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3522  */
3523 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3524 {
3525         if (!loaded_vmcs->vmcs)
3526                 return;
3527         loaded_vmcs_clear(loaded_vmcs);
3528         free_vmcs(loaded_vmcs->vmcs);
3529         loaded_vmcs->vmcs = NULL;
3530 }
3531
3532 static void free_kvm_area(void)
3533 {
3534         int cpu;
3535
3536         for_each_possible_cpu(cpu) {
3537                 free_vmcs(per_cpu(vmxarea, cpu));
3538                 per_cpu(vmxarea, cpu) = NULL;
3539         }
3540 }
3541
3542 static void init_vmcs_shadow_fields(void)
3543 {
3544         int i, j;
3545
3546         /* No checks for read only fields yet */
3547
3548         for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3549                 switch (shadow_read_write_fields[i]) {
3550                 case GUEST_BNDCFGS:
3551                         if (!kvm_mpx_supported())
3552                                 continue;
3553                         break;
3554                 default:
3555                         break;
3556                 }
3557
3558                 if (j < i)
3559                         shadow_read_write_fields[j] =
3560                                 shadow_read_write_fields[i];
3561                 j++;
3562         }
3563         max_shadow_read_write_fields = j;
3564
3565         /* shadowed fields guest access without vmexit */
3566         for (i = 0; i < max_shadow_read_write_fields; i++) {
3567                 clear_bit(shadow_read_write_fields[i],
3568                           vmx_vmwrite_bitmap);
3569                 clear_bit(shadow_read_write_fields[i],
3570                           vmx_vmread_bitmap);
3571         }
3572         for (i = 0; i < max_shadow_read_only_fields; i++)
3573                 clear_bit(shadow_read_only_fields[i],
3574                           vmx_vmread_bitmap);
3575 }
3576
3577 static __init int alloc_kvm_area(void)
3578 {
3579         int cpu;
3580
3581         for_each_possible_cpu(cpu) {
3582                 struct vmcs *vmcs;
3583
3584                 vmcs = alloc_vmcs_cpu(cpu);
3585                 if (!vmcs) {
3586                         free_kvm_area();
3587                         return -ENOMEM;
3588                 }
3589
3590                 per_cpu(vmxarea, cpu) = vmcs;
3591         }
3592         return 0;
3593 }
3594
3595 static bool emulation_required(struct kvm_vcpu *vcpu)
3596 {
3597         return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3598 }
3599
3600 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3601                 struct kvm_segment *save)
3602 {
3603         if (!emulate_invalid_guest_state) {
3604                 /*
3605                  * CS and SS RPL should be equal during guest entry according
3606                  * to VMX spec, but in reality it is not always so. Since vcpu
3607                  * is in the middle of the transition from real mode to
3608                  * protected mode it is safe to assume that RPL 0 is a good
3609                  * default value.
3610                  */
3611                 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3612                         save->selector &= ~SEGMENT_RPL_MASK;
3613                 save->dpl = save->selector & SEGMENT_RPL_MASK;
3614                 save->s = 1;
3615         }
3616         vmx_set_segment(vcpu, save, seg);
3617 }
3618
3619 static void enter_pmode(struct kvm_vcpu *vcpu)
3620 {
3621         unsigned long flags;
3622         struct vcpu_vmx *vmx = to_vmx(vcpu);
3623
3624         /*
3625          * Update real mode segment cache. It may be not up-to-date if sement
3626          * register was written while vcpu was in a guest mode.
3627          */
3628         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3629         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3630         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3631         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3632         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3633         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3634
3635         vmx->rmode.vm86_active = 0;
3636
3637         vmx_segment_cache_clear(vmx);
3638
3639         vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3640
3641         flags = vmcs_readl(GUEST_RFLAGS);
3642         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3643         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3644         vmcs_writel(GUEST_RFLAGS, flags);
3645
3646         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3647                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3648
3649         update_exception_bitmap(vcpu);
3650
3651         fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3652         fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3653         fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3654         fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3655         fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3656         fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3657 }
3658
3659 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3660 {
3661         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3662         struct kvm_segment var = *save;
3663
3664         var.dpl = 0x3;
3665         if (seg == VCPU_SREG_CS)
3666                 var.type = 0x3;
3667
3668         if (!emulate_invalid_guest_state) {
3669                 var.selector = var.base >> 4;
3670                 var.base = var.base & 0xffff0;
3671                 var.limit = 0xffff;
3672                 var.g = 0;
3673                 var.db = 0;
3674                 var.present = 1;
3675                 var.s = 1;
3676                 var.l = 0;
3677                 var.unusable = 0;
3678                 var.type = 0x3;
3679                 var.avl = 0;
3680                 if (save->base & 0xf)
3681                         printk_once(KERN_WARNING "kvm: segment base is not "
3682                                         "paragraph aligned when entering "
3683                                         "protected mode (seg=%d)", seg);
3684         }
3685
3686         vmcs_write16(sf->selector, var.selector);
3687         vmcs_write32(sf->base, var.base);
3688         vmcs_write32(sf->limit, var.limit);
3689         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3690 }
3691
3692 static void enter_rmode(struct kvm_vcpu *vcpu)
3693 {
3694         unsigned long flags;
3695         struct vcpu_vmx *vmx = to_vmx(vcpu);
3696
3697         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3698         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3699         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3700         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3701         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3702         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3703         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3704
3705         vmx->rmode.vm86_active = 1;
3706
3707         /*
3708          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3709          * vcpu. Warn the user that an update is overdue.
3710          */
3711         if (!vcpu->kvm->arch.tss_addr)
3712                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3713                              "called before entering vcpu\n");
3714
3715         vmx_segment_cache_clear(vmx);
3716
3717         vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3718         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3719         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3720
3721         flags = vmcs_readl(GUEST_RFLAGS);
3722         vmx->rmode.save_rflags = flags;
3723
3724         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3725
3726         vmcs_writel(GUEST_RFLAGS, flags);
3727         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3728         update_exception_bitmap(vcpu);
3729
3730         fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3731         fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3732         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3733         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3734         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3735         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3736
3737         kvm_mmu_reset_context(vcpu);
3738 }
3739
3740 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3741 {
3742         struct vcpu_vmx *vmx = to_vmx(vcpu);
3743         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3744
3745         if (!msr)
3746                 return;
3747
3748         /*
3749          * Force kernel_gs_base reloading before EFER changes, as control
3750          * of this msr depends on is_long_mode().
3751          */
3752         vmx_load_host_state(to_vmx(vcpu));
3753         vcpu->arch.efer = efer;
3754         if (efer & EFER_LMA) {
3755                 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3756                 msr->data = efer;
3757         } else {
3758                 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3759
3760                 msr->data = efer & ~EFER_LME;
3761         }
3762         setup_msrs(vmx);
3763 }
3764
3765 #ifdef CONFIG_X86_64
3766
3767 static void enter_lmode(struct kvm_vcpu *vcpu)
3768 {
3769         u32 guest_tr_ar;
3770
3771         vmx_segment_cache_clear(to_vmx(vcpu));
3772
3773         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3774         if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3775                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3776                                      __func__);
3777                 vmcs_write32(GUEST_TR_AR_BYTES,
3778                              (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3779                              | VMX_AR_TYPE_BUSY_64_TSS);
3780         }
3781         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3782 }
3783
3784 static void exit_lmode(struct kvm_vcpu *vcpu)
3785 {
3786         vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3787         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3788 }
3789
3790 #endif
3791
3792 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3793 {
3794         vpid_sync_context(vpid);
3795         if (enable_ept) {
3796                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3797                         return;
3798                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3799         }
3800 }
3801
3802 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3803 {
3804         __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
3805 }
3806
3807 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3808 {
3809         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3810
3811         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3812         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3813 }
3814
3815 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3816 {
3817         if (enable_ept && is_paging(vcpu))
3818                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3819         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3820 }
3821
3822 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3823 {
3824         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3825
3826         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3827         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3828 }
3829
3830 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3831 {
3832         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3833
3834         if (!test_bit(VCPU_EXREG_PDPTR,
3835                       (unsigned long *)&vcpu->arch.regs_dirty))
3836                 return;
3837
3838         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3839                 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3840                 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3841                 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3842                 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3843         }
3844 }
3845
3846 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3847 {
3848         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3849
3850         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3851                 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3852                 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3853                 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3854                 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3855         }
3856
3857         __set_bit(VCPU_EXREG_PDPTR,
3858                   (unsigned long *)&vcpu->arch.regs_avail);
3859         __set_bit(VCPU_EXREG_PDPTR,
3860                   (unsigned long *)&vcpu->arch.regs_dirty);
3861 }
3862
3863 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3864
3865 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3866                                         unsigned long cr0,
3867                                         struct kvm_vcpu *vcpu)
3868 {
3869         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3870                 vmx_decache_cr3(vcpu);
3871         if (!(cr0 & X86_CR0_PG)) {
3872                 /* From paging/starting to nonpaging */
3873                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3874                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3875                              (CPU_BASED_CR3_LOAD_EXITING |
3876                               CPU_BASED_CR3_STORE_EXITING));
3877                 vcpu->arch.cr0 = cr0;
3878                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3879         } else if (!is_paging(vcpu)) {
3880                 /* From nonpaging to paging */
3881                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3882                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3883                              ~(CPU_BASED_CR3_LOAD_EXITING |
3884                                CPU_BASED_CR3_STORE_EXITING));
3885                 vcpu->arch.cr0 = cr0;
3886                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3887         }
3888
3889         if (!(cr0 & X86_CR0_WP))
3890                 *hw_cr0 &= ~X86_CR0_WP;
3891 }
3892
3893 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3894 {
3895         struct vcpu_vmx *vmx = to_vmx(vcpu);
3896         unsigned long hw_cr0;
3897
3898         hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3899         if (enable_unrestricted_guest)
3900                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3901         else {
3902                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3903
3904                 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3905                         enter_pmode(vcpu);
3906
3907                 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3908                         enter_rmode(vcpu);
3909         }
3910
3911 #ifdef CONFIG_X86_64
3912         if (vcpu->arch.efer & EFER_LME) {
3913                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3914                         enter_lmode(vcpu);
3915                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3916                         exit_lmode(vcpu);
3917         }
3918 #endif
3919
3920         if (enable_ept)
3921                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3922
3923         if (!vcpu->fpu_active)
3924                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3925
3926         vmcs_writel(CR0_READ_SHADOW, cr0);
3927         vmcs_writel(GUEST_CR0, hw_cr0);
3928         vcpu->arch.cr0 = cr0;
3929
3930         /* depends on vcpu->arch.cr0 to be set to a new value */
3931         vmx->emulation_required = emulation_required(vcpu);
3932 }
3933
3934 static u64 construct_eptp(unsigned long root_hpa)
3935 {
3936         u64 eptp;
3937
3938         /* TODO write the value reading from MSR */
3939         eptp = VMX_EPT_DEFAULT_MT |
3940                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3941         if (enable_ept_ad_bits)
3942                 eptp |= VMX_EPT_AD_ENABLE_BIT;
3943         eptp |= (root_hpa & PAGE_MASK);
3944
3945         return eptp;
3946 }
3947
3948 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3949 {
3950         unsigned long guest_cr3;
3951         u64 eptp;
3952
3953         guest_cr3 = cr3;
3954         if (enable_ept) {
3955                 eptp = construct_eptp(cr3);
3956                 vmcs_write64(EPT_POINTER, eptp);
3957                 if (is_paging(vcpu) || is_guest_mode(vcpu))
3958                         guest_cr3 = kvm_read_cr3(vcpu);
3959                 else
3960                         guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3961                 ept_load_pdptrs(vcpu);
3962         }
3963
3964         vmx_flush_tlb(vcpu);
3965         vmcs_writel(GUEST_CR3, guest_cr3);
3966 }
3967
3968 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3969 {
3970         /*
3971          * Pass through host's Machine Check Enable value to hw_cr4, which
3972          * is in force while we are in guest mode.  Do not let guests control
3973          * this bit, even if host CR4.MCE == 0.
3974          */
3975         unsigned long hw_cr4 =
3976                 (cr4_read_shadow() & X86_CR4_MCE) |
3977                 (cr4 & ~X86_CR4_MCE) |
3978                 (to_vmx(vcpu)->rmode.vm86_active ?
3979                  KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3980
3981         if (cr4 & X86_CR4_VMXE) {
3982                 /*
3983                  * To use VMXON (and later other VMX instructions), a guest
3984                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
3985                  * So basically the check on whether to allow nested VMX
3986                  * is here.
3987                  */
3988                 if (!nested_vmx_allowed(vcpu))
3989                         return 1;
3990         }
3991         if (to_vmx(vcpu)->nested.vmxon &&
3992             ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3993                 return 1;
3994
3995         vcpu->arch.cr4 = cr4;
3996         if (enable_ept) {
3997                 if (!is_paging(vcpu)) {
3998                         hw_cr4 &= ~X86_CR4_PAE;
3999                         hw_cr4 |= X86_CR4_PSE;
4000                 } else if (!(cr4 & X86_CR4_PAE)) {
4001                         hw_cr4 &= ~X86_CR4_PAE;
4002                 }
4003         }
4004
4005         if (!enable_unrestricted_guest && !is_paging(vcpu))
4006                 /*
4007                  * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4008                  * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
4009                  * to be manually disabled when guest switches to non-paging
4010                  * mode.
4011                  *
4012                  * If !enable_unrestricted_guest, the CPU is always running
4013                  * with CR0.PG=1 and CR4 needs to be modified.
4014                  * If enable_unrestricted_guest, the CPU automatically
4015                  * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4016                  */
4017                 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4018
4019         vmcs_writel(CR4_READ_SHADOW, cr4);
4020         vmcs_writel(GUEST_CR4, hw_cr4);
4021         return 0;
4022 }
4023
4024 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4025                             struct kvm_segment *var, int seg)
4026 {
4027         struct vcpu_vmx *vmx = to_vmx(vcpu);
4028         u32 ar;
4029
4030         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4031                 *var = vmx->rmode.segs[seg];
4032                 if (seg == VCPU_SREG_TR
4033                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4034                         return;
4035                 var->base = vmx_read_guest_seg_base(vmx, seg);
4036                 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4037                 return;
4038         }
4039         var->base = vmx_read_guest_seg_base(vmx, seg);
4040         var->limit = vmx_read_guest_seg_limit(vmx, seg);
4041         var->selector = vmx_read_guest_seg_selector(vmx, seg);
4042         ar = vmx_read_guest_seg_ar(vmx, seg);
4043         var->unusable = (ar >> 16) & 1;
4044         var->type = ar & 15;
4045         var->s = (ar >> 4) & 1;
4046         var->dpl = (ar >> 5) & 3;
4047         /*
4048          * Some userspaces do not preserve unusable property. Since usable
4049          * segment has to be present according to VMX spec we can use present
4050          * property to amend userspace bug by making unusable segment always
4051          * nonpresent. vmx_segment_access_rights() already marks nonpresent
4052          * segment as unusable.
4053          */
4054         var->present = !var->unusable;
4055         var->avl = (ar >> 12) & 1;
4056         var->l = (ar >> 13) & 1;
4057         var->db = (ar >> 14) & 1;
4058         var->g = (ar >> 15) & 1;
4059 }
4060
4061 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4062 {
4063         struct kvm_segment s;
4064
4065         if (to_vmx(vcpu)->rmode.vm86_active) {
4066                 vmx_get_segment(vcpu, &s, seg);
4067                 return s.base;
4068         }
4069         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4070 }
4071
4072 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4073 {
4074         struct vcpu_vmx *vmx = to_vmx(vcpu);
4075
4076         if (unlikely(vmx->rmode.vm86_active))
4077                 return 0;
4078         else {
4079                 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4080                 return VMX_AR_DPL(ar);
4081         }
4082 }
4083
4084 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4085 {
4086         u32 ar;
4087
4088         if (var->unusable || !var->present)
4089                 ar = 1 << 16;
4090         else {
4091                 ar = var->type & 15;
4092                 ar |= (var->s & 1) << 4;
4093                 ar |= (var->dpl & 3) << 5;
4094                 ar |= (var->present & 1) << 7;
4095                 ar |= (var->avl & 1) << 12;
4096                 ar |= (var->l & 1) << 13;
4097                 ar |= (var->db & 1) << 14;
4098                 ar |= (var->g & 1) << 15;
4099         }
4100
4101         return ar;
4102 }
4103
4104 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4105                             struct kvm_segment *var, int seg)
4106 {
4107         struct vcpu_vmx *vmx = to_vmx(vcpu);
4108         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4109
4110         vmx_segment_cache_clear(vmx);
4111
4112         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4113                 vmx->rmode.segs[seg] = *var;
4114                 if (seg == VCPU_SREG_TR)
4115                         vmcs_write16(sf->selector, var->selector);
4116                 else if (var->s)
4117                         fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4118                 goto out;
4119         }
4120
4121         vmcs_writel(sf->base, var->base);
4122         vmcs_write32(sf->limit, var->limit);
4123         vmcs_write16(sf->selector, var->selector);
4124
4125         /*
4126          *   Fix the "Accessed" bit in AR field of segment registers for older
4127          * qemu binaries.
4128          *   IA32 arch specifies that at the time of processor reset the
4129          * "Accessed" bit in the AR field of segment registers is 1. And qemu
4130          * is setting it to 0 in the userland code. This causes invalid guest
4131          * state vmexit when "unrestricted guest" mode is turned on.
4132          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
4133          * tree. Newer qemu binaries with that qemu fix would not need this
4134          * kvm hack.
4135          */
4136         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4137                 var->type |= 0x1; /* Accessed */
4138
4139         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4140
4141 out:
4142         vmx->emulation_required = emulation_required(vcpu);
4143 }
4144
4145 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4146 {
4147         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4148
4149         *db = (ar >> 14) & 1;
4150         *l = (ar >> 13) & 1;
4151 }
4152
4153 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4154 {
4155         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4156         dt->address = vmcs_readl(GUEST_IDTR_BASE);
4157 }
4158
4159 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4160 {
4161         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4162         vmcs_writel(GUEST_IDTR_BASE, dt->address);
4163 }
4164
4165 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4166 {
4167         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4168         dt->address = vmcs_readl(GUEST_GDTR_BASE);
4169 }
4170
4171 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4172 {
4173         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4174         vmcs_writel(GUEST_GDTR_BASE, dt->address);
4175 }
4176
4177 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4178 {
4179         struct kvm_segment var;
4180         u32 ar;
4181
4182         vmx_get_segment(vcpu, &var, seg);
4183         var.dpl = 0x3;
4184         if (seg == VCPU_SREG_CS)
4185                 var.type = 0x3;
4186         ar = vmx_segment_access_rights(&var);
4187
4188         if (var.base != (var.selector << 4))
4189                 return false;
4190         if (var.limit != 0xffff)
4191                 return false;
4192         if (ar != 0xf3)
4193                 return false;
4194
4195         return true;
4196 }
4197
4198 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4199 {
4200         struct kvm_segment cs;
4201         unsigned int cs_rpl;
4202
4203         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4204         cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4205
4206         if (cs.unusable)
4207                 return false;
4208         if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4209                 return false;
4210         if (!cs.s)
4211                 return false;
4212         if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4213                 if (cs.dpl > cs_rpl)
4214                         return false;
4215         } else {
4216                 if (cs.dpl != cs_rpl)
4217                         return false;
4218         }
4219         if (!cs.present)
4220                 return false;
4221
4222         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4223         return true;
4224 }
4225
4226 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4227 {
4228         struct kvm_segment ss;
4229         unsigned int ss_rpl;
4230
4231         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4232         ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4233
4234         if (ss.unusable)
4235                 return true;
4236         if (ss.type != 3 && ss.type != 7)
4237                 return false;
4238         if (!ss.s)
4239                 return false;
4240         if (ss.dpl != ss_rpl) /* DPL != RPL */
4241                 return false;
4242         if (!ss.present)
4243                 return false;
4244
4245         return true;
4246 }
4247
4248 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4249 {
4250         struct kvm_segment var;
4251         unsigned int rpl;
4252
4253         vmx_get_segment(vcpu, &var, seg);
4254         rpl = var.selector & SEGMENT_RPL_MASK;
4255
4256         if (var.unusable)
4257                 return true;
4258         if (!var.s)
4259                 return false;
4260         if (!var.present)
4261                 return false;
4262         if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4263                 if (var.dpl < rpl) /* DPL < RPL */
4264                         return false;
4265         }
4266
4267         /* TODO: Add other members to kvm_segment_field to allow checking for other access
4268          * rights flags
4269          */
4270         return true;
4271 }
4272
4273 static bool tr_valid(struct kvm_vcpu *vcpu)
4274 {
4275         struct kvm_segment tr;
4276
4277         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4278
4279         if (tr.unusable)
4280                 return false;
4281         if (tr.selector & SEGMENT_TI_MASK)      /* TI = 1 */
4282                 return false;
4283         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4284                 return false;
4285         if (!tr.present)
4286                 return false;
4287
4288         return true;
4289 }
4290
4291 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4292 {
4293         struct kvm_segment ldtr;
4294
4295         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4296
4297         if (ldtr.unusable)
4298                 return true;
4299         if (ldtr.selector & SEGMENT_TI_MASK)    /* TI = 1 */
4300                 return false;
4301         if (ldtr.type != 2)
4302                 return false;
4303         if (!ldtr.present)
4304                 return false;
4305
4306         return true;
4307 }
4308
4309 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4310 {
4311         struct kvm_segment cs, ss;
4312
4313         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4314         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4315
4316         return ((cs.selector & SEGMENT_RPL_MASK) ==
4317                  (ss.selector & SEGMENT_RPL_MASK));
4318 }
4319
4320 /*
4321  * Check if guest state is valid. Returns true if valid, false if
4322  * not.
4323  * We assume that registers are always usable
4324  */
4325 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4326 {
4327         if (enable_unrestricted_guest)
4328                 return true;
4329
4330         /* real mode guest state checks */
4331         if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4332                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4333                         return false;
4334                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4335                         return false;
4336                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4337                         return false;
4338                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4339                         return false;
4340                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4341                         return false;
4342                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4343                         return false;
4344         } else {
4345         /* protected mode guest state checks */
4346                 if (!cs_ss_rpl_check(vcpu))
4347                         return false;
4348                 if (!code_segment_valid(vcpu))
4349                         return false;
4350                 if (!stack_segment_valid(vcpu))
4351                         return false;
4352                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4353                         return false;
4354                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4355                         return false;
4356                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4357                         return false;
4358                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4359                         return false;
4360                 if (!tr_valid(vcpu))
4361                         return false;
4362                 if (!ldtr_valid(vcpu))
4363                         return false;
4364         }
4365         /* TODO:
4366          * - Add checks on RIP
4367          * - Add checks on RFLAGS
4368          */
4369
4370         return true;
4371 }
4372
4373 static int init_rmode_tss(struct kvm *kvm)
4374 {
4375         gfn_t fn;
4376         u16 data = 0;
4377         int idx, r;
4378
4379         idx = srcu_read_lock(&kvm->srcu);
4380         fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4381         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4382         if (r < 0)
4383                 goto out;
4384         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4385         r = kvm_write_guest_page(kvm, fn++, &data,
4386                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
4387         if (r < 0)
4388                 goto out;
4389         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4390         if (r < 0)
4391                 goto out;
4392         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4393         if (r < 0)
4394                 goto out;
4395         data = ~0;
4396         r = kvm_write_guest_page(kvm, fn, &data,
4397                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4398                                  sizeof(u8));
4399 out:
4400         srcu_read_unlock(&kvm->srcu, idx);
4401         return r;
4402 }
4403
4404 static int init_rmode_identity_map(struct kvm *kvm)
4405 {
4406         int i, idx, r = 0;
4407         kvm_pfn_t identity_map_pfn;
4408         u32 tmp;
4409
4410         if (!enable_ept)
4411                 return 0;
4412
4413         /* Protect kvm->arch.ept_identity_pagetable_done. */
4414         mutex_lock(&kvm->slots_lock);
4415
4416         if (likely(kvm->arch.ept_identity_pagetable_done))
4417                 goto out2;
4418
4419         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4420
4421         r = alloc_identity_pagetable(kvm);
4422         if (r < 0)
4423                 goto out2;
4424
4425         idx = srcu_read_lock(&kvm->srcu);
4426         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4427         if (r < 0)
4428                 goto out;
4429         /* Set up identity-mapping pagetable for EPT in real mode */
4430         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4431                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4432                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4433                 r = kvm_write_guest_page(kvm, identity_map_pfn,
4434                                 &tmp, i * sizeof(tmp), sizeof(tmp));
4435                 if (r < 0)
4436                         goto out;
4437         }
4438         kvm->arch.ept_identity_pagetable_done = true;
4439
4440 out:
4441         srcu_read_unlock(&kvm->srcu, idx);
4442
4443 out2:
4444         mutex_unlock(&kvm->slots_lock);
4445         return r;
4446 }
4447
4448 static void seg_setup(int seg)
4449 {
4450         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4451         unsigned int ar;
4452
4453         vmcs_write16(sf->selector, 0);
4454         vmcs_writel(sf->base, 0);
4455         vmcs_write32(sf->limit, 0xffff);
4456         ar = 0x93;
4457         if (seg == VCPU_SREG_CS)
4458                 ar |= 0x08; /* code segment */
4459
4460         vmcs_write32(sf->ar_bytes, ar);
4461 }
4462
4463 static int alloc_apic_access_page(struct kvm *kvm)
4464 {
4465         struct page *page;
4466         int r = 0;
4467
4468         mutex_lock(&kvm->slots_lock);
4469         if (kvm->arch.apic_access_page_done)
4470                 goto out;
4471         r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4472                                     APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4473         if (r)
4474                 goto out;
4475
4476         page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4477         if (is_error_page(page)) {
4478                 r = -EFAULT;
4479                 goto out;
4480         }
4481
4482         /*
4483          * Do not pin the page in memory, so that memory hot-unplug
4484          * is able to migrate it.
4485          */
4486         put_page(page);
4487         kvm->arch.apic_access_page_done = true;
4488 out:
4489         mutex_unlock(&kvm->slots_lock);
4490         return r;
4491 }
4492
4493 static int alloc_identity_pagetable(struct kvm *kvm)
4494 {
4495         /* Called with kvm->slots_lock held. */
4496
4497         int r = 0;
4498
4499         BUG_ON(kvm->arch.ept_identity_pagetable_done);
4500
4501         r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4502                                     kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4503
4504         return r;
4505 }
4506
4507 static int allocate_vpid(void)
4508 {
4509         int vpid;
4510
4511         if (!enable_vpid)
4512                 return 0;
4513         spin_lock(&vmx_vpid_lock);
4514         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4515         if (vpid < VMX_NR_VPIDS)
4516                 __set_bit(vpid, vmx_vpid_bitmap);
4517         else
4518                 vpid = 0;
4519         spin_unlock(&vmx_vpid_lock);
4520         return vpid;
4521 }
4522
4523 static void free_vpid(int vpid)
4524 {
4525         if (!enable_vpid || vpid == 0)
4526                 return;
4527         spin_lock(&vmx_vpid_lock);
4528         __clear_bit(vpid, vmx_vpid_bitmap);
4529         spin_unlock(&vmx_vpid_lock);
4530 }
4531
4532 #define MSR_TYPE_R      1
4533 #define MSR_TYPE_W      2
4534 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4535                                                 u32 msr, int type)
4536 {
4537         int f = sizeof(unsigned long);
4538
4539         if (!cpu_has_vmx_msr_bitmap())
4540                 return;
4541
4542         /*
4543          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4544          * have the write-low and read-high bitmap offsets the wrong way round.
4545          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4546          */
4547         if (msr <= 0x1fff) {
4548                 if (type & MSR_TYPE_R)
4549                         /* read-low */
4550                         __clear_bit(msr, msr_bitmap + 0x000 / f);
4551
4552                 if (type & MSR_TYPE_W)
4553                         /* write-low */
4554                         __clear_bit(msr, msr_bitmap + 0x800 / f);
4555
4556         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4557                 msr &= 0x1fff;
4558                 if (type & MSR_TYPE_R)
4559                         /* read-high */
4560                         __clear_bit(msr, msr_bitmap + 0x400 / f);
4561
4562                 if (type & MSR_TYPE_W)
4563                         /* write-high */
4564                         __clear_bit(msr, msr_bitmap + 0xc00 / f);
4565
4566         }
4567 }
4568
4569 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4570                                                 u32 msr, int type)
4571 {
4572         int f = sizeof(unsigned long);
4573
4574         if (!cpu_has_vmx_msr_bitmap())
4575                 return;
4576
4577         /*
4578          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4579          * have the write-low and read-high bitmap offsets the wrong way round.
4580          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4581          */
4582         if (msr <= 0x1fff) {
4583                 if (type & MSR_TYPE_R)
4584                         /* read-low */
4585                         __set_bit(msr, msr_bitmap + 0x000 / f);
4586
4587                 if (type & MSR_TYPE_W)
4588                         /* write-low */
4589                         __set_bit(msr, msr_bitmap + 0x800 / f);
4590
4591         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4592                 msr &= 0x1fff;
4593                 if (type & MSR_TYPE_R)
4594                         /* read-high */
4595                         __set_bit(msr, msr_bitmap + 0x400 / f);
4596
4597                 if (type & MSR_TYPE_W)
4598                         /* write-high */
4599                         __set_bit(msr, msr_bitmap + 0xc00 / f);
4600
4601         }
4602 }
4603
4604 /*
4605  * If a msr is allowed by L0, we should check whether it is allowed by L1.
4606  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4607  */
4608 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4609                                                unsigned long *msr_bitmap_nested,
4610                                                u32 msr, int type)
4611 {
4612         int f = sizeof(unsigned long);
4613
4614         if (!cpu_has_vmx_msr_bitmap()) {
4615                 WARN_ON(1);
4616                 return;
4617         }
4618
4619         /*
4620          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4621          * have the write-low and read-high bitmap offsets the wrong way round.
4622          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4623          */
4624         if (msr <= 0x1fff) {
4625                 if (type & MSR_TYPE_R &&
4626                    !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4627                         /* read-low */
4628                         __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4629
4630                 if (type & MSR_TYPE_W &&
4631                    !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4632                         /* write-low */
4633                         __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4634
4635         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4636                 msr &= 0x1fff;
4637                 if (type & MSR_TYPE_R &&
4638                    !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4639                         /* read-high */
4640                         __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4641
4642                 if (type & MSR_TYPE_W &&
4643                    !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4644                         /* write-high */
4645                         __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4646
4647         }
4648 }
4649
4650 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4651 {
4652         if (!longmode_only)
4653                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4654                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
4655         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4656                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
4657 }
4658
4659 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4660 {
4661         __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4662                         msr, MSR_TYPE_R);
4663         __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4664                         msr, MSR_TYPE_R);
4665 }
4666
4667 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4668 {
4669         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4670                         msr, MSR_TYPE_R);
4671         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4672                         msr, MSR_TYPE_R);
4673 }
4674
4675 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4676 {
4677         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4678                         msr, MSR_TYPE_W);
4679         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4680                         msr, MSR_TYPE_W);
4681 }
4682
4683 static bool vmx_get_enable_apicv(void)
4684 {
4685         return enable_apicv;
4686 }
4687
4688 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4689 {
4690         struct vcpu_vmx *vmx = to_vmx(vcpu);
4691         int max_irr;
4692         void *vapic_page;
4693         u16 status;
4694
4695         if (vmx->nested.pi_desc &&
4696             vmx->nested.pi_pending) {
4697                 vmx->nested.pi_pending = false;
4698                 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4699                         return 0;
4700
4701                 max_irr = find_last_bit(
4702                         (unsigned long *)vmx->nested.pi_desc->pir, 256);
4703
4704                 if (max_irr == 256)
4705                         return 0;
4706
4707                 vapic_page = kmap(vmx->nested.virtual_apic_page);
4708                 if (!vapic_page) {
4709                         WARN_ON(1);
4710                         return -ENOMEM;
4711                 }
4712                 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4713                 kunmap(vmx->nested.virtual_apic_page);
4714
4715                 status = vmcs_read16(GUEST_INTR_STATUS);
4716                 if ((u8)max_irr > ((u8)status & 0xff)) {
4717                         status &= ~0xff;
4718                         status |= (u8)max_irr;
4719                         vmcs_write16(GUEST_INTR_STATUS, status);
4720                 }
4721         }
4722         return 0;
4723 }
4724
4725 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4726 {
4727 #ifdef CONFIG_SMP
4728         if (vcpu->mode == IN_GUEST_MODE) {
4729                 struct vcpu_vmx *vmx = to_vmx(vcpu);
4730
4731                 /*
4732                  * Currently, we don't support urgent interrupt,
4733                  * all interrupts are recognized as non-urgent
4734                  * interrupt, so we cannot post interrupts when
4735                  * 'SN' is set.
4736                  *
4737                  * If the vcpu is in guest mode, it means it is
4738                  * running instead of being scheduled out and
4739                  * waiting in the run queue, and that's the only
4740                  * case when 'SN' is set currently, warning if
4741                  * 'SN' is set.
4742                  */
4743                 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4744
4745                 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4746                                 POSTED_INTR_VECTOR);
4747                 return true;
4748         }
4749 #endif
4750         return false;
4751 }
4752
4753 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4754                                                 int vector)
4755 {
4756         struct vcpu_vmx *vmx = to_vmx(vcpu);
4757
4758         if (is_guest_mode(vcpu) &&
4759             vector == vmx->nested.posted_intr_nv) {
4760                 /* the PIR and ON have been set by L1. */
4761                 kvm_vcpu_trigger_posted_interrupt(vcpu);
4762                 /*
4763                  * If a posted intr is not recognized by hardware,
4764                  * we will accomplish it in the next vmentry.
4765                  */
4766                 vmx->nested.pi_pending = true;
4767                 kvm_make_request(KVM_REQ_EVENT, vcpu);
4768                 return 0;
4769         }
4770         return -1;
4771 }
4772 /*
4773  * Send interrupt to vcpu via posted interrupt way.
4774  * 1. If target vcpu is running(non-root mode), send posted interrupt
4775  * notification to vcpu and hardware will sync PIR to vIRR atomically.
4776  * 2. If target vcpu isn't running(root mode), kick it to pick up the
4777  * interrupt from PIR in next vmentry.
4778  */
4779 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4780 {
4781         struct vcpu_vmx *vmx = to_vmx(vcpu);
4782         int r;
4783
4784         r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4785         if (!r)
4786                 return;
4787
4788         if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4789                 return;
4790
4791         r = pi_test_and_set_on(&vmx->pi_desc);
4792         kvm_make_request(KVM_REQ_EVENT, vcpu);
4793         if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4794                 kvm_vcpu_kick(vcpu);
4795 }
4796
4797 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4798 {
4799         struct vcpu_vmx *vmx = to_vmx(vcpu);
4800
4801         if (!pi_test_and_clear_on(&vmx->pi_desc))
4802                 return;
4803
4804         kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4805 }
4806
4807 /*
4808  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4809  * will not change in the lifetime of the guest.
4810  * Note that host-state that does change is set elsewhere. E.g., host-state
4811  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4812  */
4813 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4814 {
4815         u32 low32, high32;
4816         unsigned long tmpl;
4817         struct desc_ptr dt;
4818         unsigned long cr4;
4819
4820         vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS);  /* 22.2.3 */
4821         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
4822
4823         /* Save the most likely value for this task's CR4 in the VMCS. */
4824         cr4 = cr4_read_shadow();
4825         vmcs_writel(HOST_CR4, cr4);                     /* 22.2.3, 22.2.5 */
4826         vmx->host_state.vmcs_host_cr4 = cr4;
4827
4828         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
4829 #ifdef CONFIG_X86_64
4830         /*
4831          * Load null selectors, so we can avoid reloading them in
4832          * __vmx_load_host_state(), in case userspace uses the null selectors
4833          * too (the expected case).
4834          */
4835         vmcs_write16(HOST_DS_SELECTOR, 0);
4836         vmcs_write16(HOST_ES_SELECTOR, 0);
4837 #else
4838         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4839         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4840 #endif
4841         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4842         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
4843
4844         native_store_idt(&dt);
4845         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
4846         vmx->host_idt_base = dt.address;
4847
4848         vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4849
4850         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4851         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4852         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4853         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
4854
4855         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4856                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4857                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4858         }
4859 }
4860
4861 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4862 {
4863         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4864         if (enable_ept)
4865                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4866         if (is_guest_mode(&vmx->vcpu))
4867                 vmx->vcpu.arch.cr4_guest_owned_bits &=
4868                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4869         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4870 }
4871
4872 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4873 {
4874         u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4875
4876         if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4877                 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4878         /* Enable the preemption timer dynamically */
4879         pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4880         return pin_based_exec_ctrl;
4881 }
4882
4883 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4884 {
4885         struct vcpu_vmx *vmx = to_vmx(vcpu);
4886
4887         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4888         if (cpu_has_secondary_exec_ctrls()) {
4889                 if (kvm_vcpu_apicv_active(vcpu))
4890                         vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
4891                                       SECONDARY_EXEC_APIC_REGISTER_VIRT |
4892                                       SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4893                 else
4894                         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
4895                                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
4896                                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4897         }
4898
4899         if (cpu_has_vmx_msr_bitmap())
4900                 vmx_set_msr_bitmap(vcpu);
4901 }
4902
4903 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4904 {
4905         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4906
4907         if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4908                 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4909
4910         if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4911                 exec_control &= ~CPU_BASED_TPR_SHADOW;
4912 #ifdef CONFIG_X86_64
4913                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4914                                 CPU_BASED_CR8_LOAD_EXITING;
4915 #endif
4916         }
4917         if (!enable_ept)
4918                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4919                                 CPU_BASED_CR3_LOAD_EXITING  |
4920                                 CPU_BASED_INVLPG_EXITING;
4921         return exec_control;
4922 }
4923
4924 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4925 {
4926         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4927         if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
4928                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4929         if (vmx->vpid == 0)
4930                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4931         if (!enable_ept) {
4932                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4933                 enable_unrestricted_guest = 0;
4934                 /* Enable INVPCID for non-ept guests may cause performance regression. */
4935                 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4936         }
4937         if (!enable_unrestricted_guest)
4938                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4939         if (!ple_gap)
4940                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4941         if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4942                 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4943                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4944         exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4945         /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4946            (handle_vmptrld).
4947            We can NOT enable shadow_vmcs here because we don't have yet
4948            a current VMCS12
4949         */
4950         exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4951
4952         if (!enable_pml)
4953                 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4954
4955         /* Currently, we allow L1 guest to directly run pcommit instruction. */
4956         exec_control &= ~SECONDARY_EXEC_PCOMMIT;
4957
4958         return exec_control;
4959 }
4960
4961 static void ept_set_mmio_spte_mask(void)
4962 {
4963         /*
4964          * EPT Misconfigurations can be generated if the value of bits 2:0
4965          * of an EPT paging-structure entry is 110b (write/execute).
4966          * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4967          * spte.
4968          */
4969         kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4970 }
4971
4972 #define VMX_XSS_EXIT_BITMAP 0
4973 /*
4974  * Sets up the vmcs for emulated real mode.
4975  */
4976 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4977 {
4978 #ifdef CONFIG_X86_64
4979         unsigned long a;
4980 #endif
4981         int i;
4982
4983         /* I/O */
4984         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4985         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4986
4987         if (enable_shadow_vmcs) {
4988                 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4989                 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4990         }
4991         if (cpu_has_vmx_msr_bitmap())
4992                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4993
4994         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4995
4996         /* Control */
4997         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4998         vmx->hv_deadline_tsc = -1;
4999
5000         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5001
5002         if (cpu_has_secondary_exec_ctrls())
5003                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5004                                 vmx_secondary_exec_control(vmx));
5005
5006         if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5007                 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5008                 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5009                 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5010                 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5011
5012                 vmcs_write16(GUEST_INTR_STATUS, 0);
5013
5014                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5015                 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5016         }
5017
5018         if (ple_gap) {
5019                 vmcs_write32(PLE_GAP, ple_gap);
5020                 vmx->ple_window = ple_window;
5021                 vmx->ple_window_dirty = true;
5022         }
5023
5024         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5025         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5026         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
5027
5028         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
5029         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
5030         vmx_set_constant_host_state(vmx);
5031 #ifdef CONFIG_X86_64
5032         rdmsrl(MSR_FS_BASE, a);
5033         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5034         rdmsrl(MSR_GS_BASE, a);
5035         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5036 #else
5037         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5038         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5039 #endif
5040
5041         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5042         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5043         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5044         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5045         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5046
5047         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5048                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5049
5050         for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5051                 u32 index = vmx_msr_index[i];
5052                 u32 data_low, data_high;
5053                 int j = vmx->nmsrs;
5054
5055                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5056                         continue;
5057                 if (wrmsr_safe(index, data_low, data_high) < 0)
5058                         continue;
5059                 vmx->guest_msrs[j].index = i;
5060                 vmx->guest_msrs[j].data = 0;
5061                 vmx->guest_msrs[j].mask = -1ull;
5062                 ++vmx->nmsrs;
5063         }
5064
5065
5066         vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5067
5068         /* 22.2.1, 20.8.1 */
5069         vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5070
5071         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
5072         set_cr4_guest_host_mask(vmx);
5073
5074         if (vmx_xsaves_supported())
5075                 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5076
5077         return 0;
5078 }
5079
5080 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5081 {
5082         struct vcpu_vmx *vmx = to_vmx(vcpu);
5083         struct msr_data apic_base_msr;
5084         u64 cr0;
5085
5086         vmx->rmode.vm86_active = 0;
5087
5088         vmx->soft_vnmi_blocked = 0;
5089
5090         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5091         kvm_set_cr8(vcpu, 0);
5092
5093         if (!init_event) {
5094                 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5095                                      MSR_IA32_APICBASE_ENABLE;
5096                 if (kvm_vcpu_is_reset_bsp(vcpu))
5097                         apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5098                 apic_base_msr.host_initiated = true;
5099                 kvm_set_apic_base(vcpu, &apic_base_msr);
5100         }
5101
5102         vmx_segment_cache_clear(vmx);
5103
5104         seg_setup(VCPU_SREG_CS);
5105         vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5106         vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5107
5108         seg_setup(VCPU_SREG_DS);
5109         seg_setup(VCPU_SREG_ES);
5110         seg_setup(VCPU_SREG_FS);
5111         seg_setup(VCPU_SREG_GS);
5112         seg_setup(VCPU_SREG_SS);
5113
5114         vmcs_write16(GUEST_TR_SELECTOR, 0);
5115         vmcs_writel(GUEST_TR_BASE, 0);
5116         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5117         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5118
5119         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5120         vmcs_writel(GUEST_LDTR_BASE, 0);
5121         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5122         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5123
5124         if (!init_event) {
5125                 vmcs_write32(GUEST_SYSENTER_CS, 0);
5126                 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5127                 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5128                 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5129         }
5130
5131         vmcs_writel(GUEST_RFLAGS, 0x02);
5132         kvm_rip_write(vcpu, 0xfff0);
5133
5134         vmcs_writel(GUEST_GDTR_BASE, 0);
5135         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5136
5137         vmcs_writel(GUEST_IDTR_BASE, 0);
5138         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5139
5140         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5141         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5142         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5143
5144         setup_msrs(vmx);
5145
5146         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
5147
5148         if (cpu_has_vmx_tpr_shadow() && !init_event) {
5149                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5150                 if (cpu_need_tpr_shadow(vcpu))
5151                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5152                                      __pa(vcpu->arch.apic->regs));
5153                 vmcs_write32(TPR_THRESHOLD, 0);
5154         }
5155
5156         kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5157
5158         if (kvm_vcpu_apicv_active(vcpu))
5159                 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5160
5161         if (vmx->vpid != 0)
5162                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5163
5164         cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5165         vmx->vcpu.arch.cr0 = cr0;
5166         vmx_set_cr0(vcpu, cr0); /* enter rmode */
5167         vmx_set_cr4(vcpu, 0);
5168         vmx_set_efer(vcpu, 0);
5169         vmx_fpu_activate(vcpu);
5170         update_exception_bitmap(vcpu);
5171
5172         vpid_sync_context(vmx->vpid);
5173 }
5174
5175 /*
5176  * In nested virtualization, check if L1 asked to exit on external interrupts.
5177  * For most existing hypervisors, this will always return true.
5178  */
5179 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5180 {
5181         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5182                 PIN_BASED_EXT_INTR_MASK;
5183 }
5184
5185 /*
5186  * In nested virtualization, check if L1 has set
5187  * VM_EXIT_ACK_INTR_ON_EXIT
5188  */
5189 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5190 {
5191         return get_vmcs12(vcpu)->vm_exit_controls &
5192                 VM_EXIT_ACK_INTR_ON_EXIT;
5193 }
5194
5195 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5196 {
5197         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5198                 PIN_BASED_NMI_EXITING;
5199 }
5200
5201 static void enable_irq_window(struct kvm_vcpu *vcpu)
5202 {
5203         u32 cpu_based_vm_exec_control;
5204
5205         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5206         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
5207         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5208 }
5209
5210 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5211 {
5212         u32 cpu_based_vm_exec_control;
5213
5214         if (!cpu_has_virtual_nmis() ||
5215             vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5216                 enable_irq_window(vcpu);
5217                 return;
5218         }
5219
5220         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5221         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
5222         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5223 }
5224
5225 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5226 {
5227         struct vcpu_vmx *vmx = to_vmx(vcpu);
5228         uint32_t intr;
5229         int irq = vcpu->arch.interrupt.nr;
5230
5231         trace_kvm_inj_virq(irq);
5232
5233         ++vcpu->stat.irq_injections;
5234         if (vmx->rmode.vm86_active) {
5235                 int inc_eip = 0;
5236                 if (vcpu->arch.interrupt.soft)
5237                         inc_eip = vcpu->arch.event_exit_inst_len;
5238                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5239                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5240                 return;
5241         }
5242         intr = irq | INTR_INFO_VALID_MASK;
5243         if (vcpu->arch.interrupt.soft) {
5244                 intr |= INTR_TYPE_SOFT_INTR;
5245                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5246                              vmx->vcpu.arch.event_exit_inst_len);
5247         } else
5248                 intr |= INTR_TYPE_EXT_INTR;
5249         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5250 }
5251
5252 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5253 {
5254         struct vcpu_vmx *vmx = to_vmx(vcpu);
5255
5256         if (is_guest_mode(vcpu))
5257                 return;
5258
5259         if (!cpu_has_virtual_nmis()) {
5260                 /*
5261                  * Tracking the NMI-blocked state in software is built upon
5262                  * finding the next open IRQ window. This, in turn, depends on
5263                  * well-behaving guests: They have to keep IRQs disabled at
5264                  * least as long as the NMI handler runs. Otherwise we may
5265                  * cause NMI nesting, maybe breaking the guest. But as this is
5266                  * highly unlikely, we can live with the residual risk.
5267                  */
5268                 vmx->soft_vnmi_blocked = 1;
5269                 vmx->vnmi_blocked_time = 0;
5270         }
5271
5272         ++vcpu->stat.nmi_injections;
5273         vmx->nmi_known_unmasked = false;
5274         if (vmx->rmode.vm86_active) {
5275                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5276                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5277                 return;
5278         }
5279         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5280                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5281 }
5282
5283 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5284 {
5285         if (!cpu_has_virtual_nmis())
5286                 return to_vmx(vcpu)->soft_vnmi_blocked;
5287         if (to_vmx(vcpu)->nmi_known_unmasked)
5288                 return false;
5289         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5290 }
5291
5292 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5293 {
5294         struct vcpu_vmx *vmx = to_vmx(vcpu);
5295
5296         if (!cpu_has_virtual_nmis()) {
5297                 if (vmx->soft_vnmi_blocked != masked) {
5298                         vmx->soft_vnmi_blocked = masked;
5299                         vmx->vnmi_blocked_time = 0;
5300                 }
5301         } else {
5302                 vmx->nmi_known_unmasked = !masked;
5303                 if (masked)
5304                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5305                                       GUEST_INTR_STATE_NMI);
5306                 else
5307                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5308                                         GUEST_INTR_STATE_NMI);
5309         }
5310 }
5311
5312 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5313 {
5314         if (to_vmx(vcpu)->nested.nested_run_pending)
5315                 return 0;
5316
5317         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5318                 return 0;
5319
5320         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5321                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5322                    | GUEST_INTR_STATE_NMI));
5323 }
5324
5325 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5326 {
5327         return (!to_vmx(vcpu)->nested.nested_run_pending &&
5328                 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5329                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5330                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5331 }
5332
5333 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5334 {
5335         int ret;
5336
5337         ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5338                                     PAGE_SIZE * 3);
5339         if (ret)
5340                 return ret;
5341         kvm->arch.tss_addr = addr;
5342         return init_rmode_tss(kvm);
5343 }
5344
5345 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5346 {
5347         switch (vec) {
5348         case BP_VECTOR:
5349                 /*
5350                  * Update instruction length as we may reinject the exception
5351                  * from user space while in guest debugging mode.
5352                  */
5353                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5354                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5355                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5356                         return false;
5357                 /* fall through */
5358         case DB_VECTOR:
5359                 if (vcpu->guest_debug &
5360                         (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5361                         return false;
5362                 /* fall through */
5363         case DE_VECTOR:
5364         case OF_VECTOR:
5365         case BR_VECTOR:
5366         case UD_VECTOR:
5367         case DF_VECTOR:
5368         case SS_VECTOR:
5369         case GP_VECTOR:
5370         case MF_VECTOR:
5371                 return true;
5372         break;
5373         }
5374         return false;
5375 }
5376
5377 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5378                                   int vec, u32 err_code)
5379 {
5380         /*
5381          * Instruction with address size override prefix opcode 0x67
5382          * Cause the #SS fault with 0 error code in VM86 mode.
5383          */
5384         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5385                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5386                         if (vcpu->arch.halt_request) {
5387                                 vcpu->arch.halt_request = 0;
5388                                 return kvm_vcpu_halt(vcpu);
5389                         }
5390                         return 1;
5391                 }
5392                 return 0;
5393         }
5394
5395         /*
5396          * Forward all other exceptions that are valid in real mode.
5397          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5398          *        the required debugging infrastructure rework.
5399          */
5400         kvm_queue_exception(vcpu, vec);
5401         return 1;
5402 }
5403
5404 /*
5405  * Trigger machine check on the host. We assume all the MSRs are already set up
5406  * by the CPU and that we still run on the same CPU as the MCE occurred on.
5407  * We pass a fake environment to the machine check handler because we want
5408  * the guest to be always treated like user space, no matter what context
5409  * it used internally.
5410  */
5411 static void kvm_machine_check(void)
5412 {
5413 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5414         struct pt_regs regs = {
5415                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5416                 .flags = X86_EFLAGS_IF,
5417         };
5418
5419         do_machine_check(&regs, 0);
5420 #endif
5421 }
5422
5423 static int handle_machine_check(struct kvm_vcpu *vcpu)
5424 {
5425         /* already handled by vcpu_run */
5426         return 1;
5427 }
5428
5429 static int handle_exception(struct kvm_vcpu *vcpu)
5430 {
5431         struct vcpu_vmx *vmx = to_vmx(vcpu);
5432         struct kvm_run *kvm_run = vcpu->run;
5433         u32 intr_info, ex_no, error_code;
5434         unsigned long cr2, rip, dr6;
5435         u32 vect_info;
5436         enum emulation_result er;
5437
5438         vect_info = vmx->idt_vectoring_info;
5439         intr_info = vmx->exit_intr_info;
5440
5441         if (is_machine_check(intr_info))
5442                 return handle_machine_check(vcpu);
5443
5444         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
5445                 return 1;  /* already handled by vmx_vcpu_run() */
5446
5447         if (is_no_device(intr_info)) {
5448                 vmx_fpu_activate(vcpu);
5449                 return 1;
5450         }
5451
5452         if (is_invalid_opcode(intr_info)) {
5453                 if (is_guest_mode(vcpu)) {
5454                         kvm_queue_exception(vcpu, UD_VECTOR);
5455                         return 1;
5456                 }
5457                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5458                 if (er != EMULATE_DONE)
5459                         kvm_queue_exception(vcpu, UD_VECTOR);
5460                 return 1;
5461         }
5462
5463         error_code = 0;
5464         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5465                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5466
5467         /*
5468          * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5469          * MMIO, it is better to report an internal error.
5470          * See the comments in vmx_handle_exit.
5471          */
5472         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5473             !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5474                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5475                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5476                 vcpu->run->internal.ndata = 3;
5477                 vcpu->run->internal.data[0] = vect_info;
5478                 vcpu->run->internal.data[1] = intr_info;
5479                 vcpu->run->internal.data[2] = error_code;
5480                 return 0;
5481         }
5482
5483         if (is_page_fault(intr_info)) {
5484                 /* EPT won't cause page fault directly */
5485                 BUG_ON(enable_ept);
5486                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5487                 trace_kvm_page_fault(cr2, error_code);
5488
5489                 if (kvm_event_needs_reinjection(vcpu))
5490                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
5491                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5492         }
5493
5494         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5495
5496         if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5497                 return handle_rmode_exception(vcpu, ex_no, error_code);
5498
5499         switch (ex_no) {
5500         case AC_VECTOR:
5501                 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5502                 return 1;
5503         case DB_VECTOR:
5504                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5505                 if (!(vcpu->guest_debug &
5506                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5507                         vcpu->arch.dr6 &= ~15;
5508                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5509                         if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5510                                 skip_emulated_instruction(vcpu);
5511
5512                         kvm_queue_exception(vcpu, DB_VECTOR);
5513                         return 1;
5514                 }
5515                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5516                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5517                 /* fall through */
5518         case BP_VECTOR:
5519                 /*
5520                  * Update instruction length as we may reinject #BP from
5521                  * user space while in guest debugging mode. Reading it for
5522                  * #DB as well causes no harm, it is not used in that case.
5523                  */
5524                 vmx->vcpu.arch.event_exit_inst_len =
5525                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5526                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5527                 rip = kvm_rip_read(vcpu);
5528                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5529                 kvm_run->debug.arch.exception = ex_no;
5530                 break;
5531         default:
5532                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5533                 kvm_run->ex.exception = ex_no;
5534                 kvm_run->ex.error_code = error_code;
5535                 break;
5536         }
5537         return 0;
5538 }
5539
5540 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5541 {
5542         ++vcpu->stat.irq_exits;
5543         return 1;
5544 }
5545
5546 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5547 {
5548         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5549         return 0;
5550 }
5551
5552 static int handle_io(struct kvm_vcpu *vcpu)
5553 {
5554         unsigned long exit_qualification;
5555         int size, in, string;
5556         unsigned port;
5557
5558         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5559         string = (exit_qualification & 16) != 0;
5560         in = (exit_qualification & 8) != 0;
5561
5562         ++vcpu->stat.io_exits;
5563
5564         if (string || in)
5565                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5566
5567         port = exit_qualification >> 16;
5568         size = (exit_qualification & 7) + 1;
5569         skip_emulated_instruction(vcpu);
5570
5571         return kvm_fast_pio_out(vcpu, size, port);
5572 }
5573
5574 static void
5575 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5576 {
5577         /*
5578          * Patch in the VMCALL instruction:
5579          */
5580         hypercall[0] = 0x0f;
5581         hypercall[1] = 0x01;
5582         hypercall[2] = 0xc1;
5583 }
5584
5585 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5586 {
5587         unsigned long always_on = VMXON_CR0_ALWAYSON;
5588         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5589
5590         if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5591                 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5592             nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5593                 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5594         return (val & always_on) == always_on;
5595 }
5596
5597 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5598 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5599 {
5600         if (is_guest_mode(vcpu)) {
5601                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5602                 unsigned long orig_val = val;
5603
5604                 /*
5605                  * We get here when L2 changed cr0 in a way that did not change
5606                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5607                  * but did change L0 shadowed bits. So we first calculate the
5608                  * effective cr0 value that L1 would like to write into the
5609                  * hardware. It consists of the L2-owned bits from the new
5610                  * value combined with the L1-owned bits from L1's guest_cr0.
5611                  */
5612                 val = (val & ~vmcs12->cr0_guest_host_mask) |
5613                         (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5614
5615                 if (!nested_cr0_valid(vcpu, val))
5616                         return 1;
5617
5618                 if (kvm_set_cr0(vcpu, val))
5619                         return 1;
5620                 vmcs_writel(CR0_READ_SHADOW, orig_val);
5621                 return 0;
5622         } else {
5623                 if (to_vmx(vcpu)->nested.vmxon &&
5624                     ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5625                         return 1;
5626                 return kvm_set_cr0(vcpu, val);
5627         }
5628 }
5629
5630 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5631 {
5632         if (is_guest_mode(vcpu)) {
5633                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5634                 unsigned long orig_val = val;
5635
5636                 /* analogously to handle_set_cr0 */
5637                 val = (val & ~vmcs12->cr4_guest_host_mask) |
5638                         (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5639                 if (kvm_set_cr4(vcpu, val))
5640                         return 1;
5641                 vmcs_writel(CR4_READ_SHADOW, orig_val);
5642                 return 0;
5643         } else
5644                 return kvm_set_cr4(vcpu, val);
5645 }
5646
5647 /* called to set cr0 as appropriate for clts instruction exit. */
5648 static void handle_clts(struct kvm_vcpu *vcpu)
5649 {
5650         if (is_guest_mode(vcpu)) {
5651                 /*
5652                  * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5653                  * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5654                  * just pretend it's off (also in arch.cr0 for fpu_activate).
5655                  */
5656                 vmcs_writel(CR0_READ_SHADOW,
5657                         vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5658                 vcpu->arch.cr0 &= ~X86_CR0_TS;
5659         } else
5660                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5661 }
5662
5663 static int handle_cr(struct kvm_vcpu *vcpu)
5664 {
5665         unsigned long exit_qualification, val;
5666         int cr;
5667         int reg;
5668         int err;
5669
5670         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5671         cr = exit_qualification & 15;
5672         reg = (exit_qualification >> 8) & 15;
5673         switch ((exit_qualification >> 4) & 3) {
5674         case 0: /* mov to cr */
5675                 val = kvm_register_readl(vcpu, reg);
5676                 trace_kvm_cr_write(cr, val);
5677                 switch (cr) {
5678                 case 0:
5679                         err = handle_set_cr0(vcpu, val);
5680                         kvm_complete_insn_gp(vcpu, err);
5681                         return 1;
5682                 case 3:
5683                         err = kvm_set_cr3(vcpu, val);
5684                         kvm_complete_insn_gp(vcpu, err);
5685                         return 1;
5686                 case 4:
5687                         err = handle_set_cr4(vcpu, val);
5688                         kvm_complete_insn_gp(vcpu, err);
5689                         return 1;
5690                 case 8: {
5691                                 u8 cr8_prev = kvm_get_cr8(vcpu);
5692                                 u8 cr8 = (u8)val;
5693                                 err = kvm_set_cr8(vcpu, cr8);
5694                                 kvm_complete_insn_gp(vcpu, err);
5695                                 if (lapic_in_kernel(vcpu))
5696                                         return 1;
5697                                 if (cr8_prev <= cr8)
5698                                         return 1;
5699                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5700                                 return 0;
5701                         }
5702                 }
5703                 break;
5704         case 2: /* clts */
5705                 handle_clts(vcpu);
5706                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5707                 skip_emulated_instruction(vcpu);
5708                 vmx_fpu_activate(vcpu);
5709                 return 1;
5710         case 1: /*mov from cr*/
5711                 switch (cr) {
5712                 case 3:
5713                         val = kvm_read_cr3(vcpu);
5714                         kvm_register_write(vcpu, reg, val);
5715                         trace_kvm_cr_read(cr, val);
5716                         skip_emulated_instruction(vcpu);
5717                         return 1;
5718                 case 8:
5719                         val = kvm_get_cr8(vcpu);
5720                         kvm_register_write(vcpu, reg, val);
5721                         trace_kvm_cr_read(cr, val);
5722                         skip_emulated_instruction(vcpu);
5723                         return 1;
5724                 }
5725                 break;
5726         case 3: /* lmsw */
5727                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5728                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5729                 kvm_lmsw(vcpu, val);
5730
5731                 skip_emulated_instruction(vcpu);
5732                 return 1;
5733         default:
5734                 break;
5735         }
5736         vcpu->run->exit_reason = 0;
5737         vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5738                (int)(exit_qualification >> 4) & 3, cr);
5739         return 0;
5740 }
5741
5742 static int handle_dr(struct kvm_vcpu *vcpu)
5743 {
5744         unsigned long exit_qualification;
5745         int dr, dr7, reg;
5746
5747         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5748         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5749
5750         /* First, if DR does not exist, trigger UD */
5751         if (!kvm_require_dr(vcpu, dr))
5752                 return 1;
5753
5754         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5755         if (!kvm_require_cpl(vcpu, 0))
5756                 return 1;
5757         dr7 = vmcs_readl(GUEST_DR7);
5758         if (dr7 & DR7_GD) {
5759                 /*
5760                  * As the vm-exit takes precedence over the debug trap, we
5761                  * need to emulate the latter, either for the host or the
5762                  * guest debugging itself.
5763                  */
5764                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5765                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5766                         vcpu->run->debug.arch.dr7 = dr7;
5767                         vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5768                         vcpu->run->debug.arch.exception = DB_VECTOR;
5769                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5770                         return 0;
5771                 } else {
5772                         vcpu->arch.dr6 &= ~15;
5773                         vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5774                         kvm_queue_exception(vcpu, DB_VECTOR);
5775                         return 1;
5776                 }
5777         }
5778
5779         if (vcpu->guest_debug == 0) {
5780                 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5781                                 CPU_BASED_MOV_DR_EXITING);
5782
5783                 /*
5784                  * No more DR vmexits; force a reload of the debug registers
5785                  * and reenter on this instruction.  The next vmexit will
5786                  * retrieve the full state of the debug registers.
5787                  */
5788                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5789                 return 1;
5790         }
5791
5792         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5793         if (exit_qualification & TYPE_MOV_FROM_DR) {
5794                 unsigned long val;
5795
5796                 if (kvm_get_dr(vcpu, dr, &val))
5797                         return 1;
5798                 kvm_register_write(vcpu, reg, val);
5799         } else
5800                 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5801                         return 1;
5802
5803         skip_emulated_instruction(vcpu);
5804         return 1;
5805 }
5806
5807 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5808 {
5809         return vcpu->arch.dr6;
5810 }
5811
5812 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5813 {
5814 }
5815
5816 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5817 {
5818         get_debugreg(vcpu->arch.db[0], 0);
5819         get_debugreg(vcpu->arch.db[1], 1);
5820         get_debugreg(vcpu->arch.db[2], 2);
5821         get_debugreg(vcpu->arch.db[3], 3);
5822         get_debugreg(vcpu->arch.dr6, 6);
5823         vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5824
5825         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5826         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
5827 }
5828
5829 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5830 {
5831         vmcs_writel(GUEST_DR7, val);
5832 }
5833
5834 static int handle_cpuid(struct kvm_vcpu *vcpu)
5835 {
5836         kvm_emulate_cpuid(vcpu);
5837         return 1;
5838 }
5839
5840 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5841 {
5842         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5843         struct msr_data msr_info;
5844
5845         msr_info.index = ecx;
5846         msr_info.host_initiated = false;
5847         if (vmx_get_msr(vcpu, &msr_info)) {
5848                 trace_kvm_msr_read_ex(ecx);
5849                 kvm_inject_gp(vcpu, 0);
5850                 return 1;
5851         }
5852
5853         trace_kvm_msr_read(ecx, msr_info.data);
5854
5855         /* FIXME: handling of bits 32:63 of rax, rdx */
5856         vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5857         vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5858         skip_emulated_instruction(vcpu);
5859         return 1;
5860 }
5861
5862 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5863 {
5864         struct msr_data msr;
5865         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5866         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5867                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5868
5869         msr.data = data;
5870         msr.index = ecx;
5871         msr.host_initiated = false;
5872         if (kvm_set_msr(vcpu, &msr) != 0) {
5873                 trace_kvm_msr_write_ex(ecx, data);
5874                 kvm_inject_gp(vcpu, 0);
5875                 return 1;
5876         }
5877
5878         trace_kvm_msr_write(ecx, data);
5879         skip_emulated_instruction(vcpu);
5880         return 1;
5881 }
5882
5883 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5884 {
5885         kvm_make_request(KVM_REQ_EVENT, vcpu);
5886         return 1;
5887 }
5888
5889 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5890 {
5891         u32 cpu_based_vm_exec_control;
5892
5893         /* clear pending irq */
5894         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5895         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5896         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5897
5898         kvm_make_request(KVM_REQ_EVENT, vcpu);
5899
5900         ++vcpu->stat.irq_window_exits;
5901         return 1;
5902 }
5903
5904 static int handle_halt(struct kvm_vcpu *vcpu)
5905 {
5906         return kvm_emulate_halt(vcpu);
5907 }
5908
5909 static int handle_vmcall(struct kvm_vcpu *vcpu)
5910 {
5911         return kvm_emulate_hypercall(vcpu);
5912 }
5913
5914 static int handle_invd(struct kvm_vcpu *vcpu)
5915 {
5916         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5917 }
5918
5919 static int handle_invlpg(struct kvm_vcpu *vcpu)
5920 {
5921         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5922
5923         kvm_mmu_invlpg(vcpu, exit_qualification);
5924         skip_emulated_instruction(vcpu);
5925         return 1;
5926 }
5927
5928 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5929 {
5930         int err;
5931
5932         err = kvm_rdpmc(vcpu);
5933         kvm_complete_insn_gp(vcpu, err);
5934
5935         return 1;
5936 }
5937
5938 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5939 {
5940         kvm_emulate_wbinvd(vcpu);
5941         return 1;
5942 }
5943
5944 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5945 {
5946         u64 new_bv = kvm_read_edx_eax(vcpu);
5947         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5948
5949         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5950                 skip_emulated_instruction(vcpu);
5951         return 1;
5952 }
5953
5954 static int handle_xsaves(struct kvm_vcpu *vcpu)
5955 {
5956         skip_emulated_instruction(vcpu);
5957         WARN(1, "this should never happen\n");
5958         return 1;
5959 }
5960
5961 static int handle_xrstors(struct kvm_vcpu *vcpu)
5962 {
5963         skip_emulated_instruction(vcpu);
5964         WARN(1, "this should never happen\n");
5965         return 1;
5966 }
5967
5968 static int handle_apic_access(struct kvm_vcpu *vcpu)
5969 {
5970         if (likely(fasteoi)) {
5971                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5972                 int access_type, offset;
5973
5974                 access_type = exit_qualification & APIC_ACCESS_TYPE;
5975                 offset = exit_qualification & APIC_ACCESS_OFFSET;
5976                 /*
5977                  * Sane guest uses MOV to write EOI, with written value
5978                  * not cared. So make a short-circuit here by avoiding
5979                  * heavy instruction emulation.
5980                  */
5981                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5982                     (offset == APIC_EOI)) {
5983                         kvm_lapic_set_eoi(vcpu);
5984                         skip_emulated_instruction(vcpu);
5985                         return 1;
5986                 }
5987         }
5988         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5989 }
5990
5991 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5992 {
5993         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5994         int vector = exit_qualification & 0xff;
5995
5996         /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5997         kvm_apic_set_eoi_accelerated(vcpu, vector);
5998         return 1;
5999 }
6000
6001 static int handle_apic_write(struct kvm_vcpu *vcpu)
6002 {
6003         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6004         u32 offset = exit_qualification & 0xfff;
6005
6006         /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6007         kvm_apic_write_nodecode(vcpu, offset);
6008         return 1;
6009 }
6010
6011 static int handle_task_switch(struct kvm_vcpu *vcpu)
6012 {
6013         struct vcpu_vmx *vmx = to_vmx(vcpu);
6014         unsigned long exit_qualification;
6015         bool has_error_code = false;
6016         u32 error_code = 0;
6017         u16 tss_selector;
6018         int reason, type, idt_v, idt_index;
6019
6020         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6021         idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6022         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6023
6024         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6025
6026         reason = (u32)exit_qualification >> 30;
6027         if (reason == TASK_SWITCH_GATE && idt_v) {
6028                 switch (type) {
6029                 case INTR_TYPE_NMI_INTR:
6030                         vcpu->arch.nmi_injected = false;
6031                         vmx_set_nmi_mask(vcpu, true);
6032                         break;
6033                 case INTR_TYPE_EXT_INTR:
6034                 case INTR_TYPE_SOFT_INTR:
6035                         kvm_clear_interrupt_queue(vcpu);
6036                         break;
6037                 case INTR_TYPE_HARD_EXCEPTION:
6038                         if (vmx->idt_vectoring_info &
6039                             VECTORING_INFO_DELIVER_CODE_MASK) {
6040                                 has_error_code = true;
6041                                 error_code =
6042                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
6043                         }
6044                         /* fall through */
6045                 case INTR_TYPE_SOFT_EXCEPTION:
6046                         kvm_clear_exception_queue(vcpu);
6047                         break;
6048                 default:
6049                         break;
6050                 }
6051         }
6052         tss_selector = exit_qualification;
6053
6054         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6055                        type != INTR_TYPE_EXT_INTR &&
6056                        type != INTR_TYPE_NMI_INTR))
6057                 skip_emulated_instruction(vcpu);
6058
6059         if (kvm_task_switch(vcpu, tss_selector,
6060                             type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6061                             has_error_code, error_code) == EMULATE_FAIL) {
6062                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6063                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6064                 vcpu->run->internal.ndata = 0;
6065                 return 0;
6066         }
6067
6068         /*
6069          * TODO: What about debug traps on tss switch?
6070          *       Are we supposed to inject them and update dr6?
6071          */
6072
6073         return 1;
6074 }
6075
6076 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6077 {
6078         unsigned long exit_qualification;
6079         gpa_t gpa;
6080         u32 error_code;
6081         int gla_validity;
6082
6083         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6084
6085         gla_validity = (exit_qualification >> 7) & 0x3;
6086         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
6087                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
6088                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
6089                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
6090                         vmcs_readl(GUEST_LINEAR_ADDRESS));
6091                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
6092                         (long unsigned int)exit_qualification);
6093                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6094                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
6095                 return 0;
6096         }
6097
6098         /*
6099          * EPT violation happened while executing iret from NMI,
6100          * "blocked by NMI" bit has to be set before next VM entry.
6101          * There are errata that may cause this bit to not be set:
6102          * AAK134, BY25.
6103          */
6104         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6105                         cpu_has_virtual_nmis() &&
6106                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6107                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6108
6109         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6110         trace_kvm_page_fault(gpa, exit_qualification);
6111
6112         /* It is a write fault? */
6113         error_code = exit_qualification & PFERR_WRITE_MASK;
6114         /* It is a fetch fault? */
6115         error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
6116         /* ept page table is present? */
6117         error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
6118
6119         vcpu->arch.exit_qualification = exit_qualification;
6120
6121         return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6122 }
6123
6124 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6125 {
6126         int ret;
6127         gpa_t gpa;
6128
6129         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6130         if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6131                 skip_emulated_instruction(vcpu);
6132                 trace_kvm_fast_mmio(gpa);
6133                 return 1;
6134         }
6135
6136         ret = handle_mmio_page_fault(vcpu, gpa, true);
6137         if (likely(ret == RET_MMIO_PF_EMULATE))
6138                 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6139                                               EMULATE_DONE;
6140
6141         if (unlikely(ret == RET_MMIO_PF_INVALID))
6142                 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6143
6144         if (unlikely(ret == RET_MMIO_PF_RETRY))
6145                 return 1;
6146
6147         /* It is the real ept misconfig */
6148         WARN_ON(1);
6149
6150         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6151         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6152
6153         return 0;
6154 }
6155
6156 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6157 {
6158         u32 cpu_based_vm_exec_control;
6159
6160         /* clear pending NMI */
6161         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6162         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6163         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
6164         ++vcpu->stat.nmi_window_exits;
6165         kvm_make_request(KVM_REQ_EVENT, vcpu);
6166
6167         return 1;
6168 }
6169
6170 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6171 {
6172         struct vcpu_vmx *vmx = to_vmx(vcpu);
6173         enum emulation_result err = EMULATE_DONE;
6174         int ret = 1;
6175         u32 cpu_exec_ctrl;
6176         bool intr_window_requested;
6177         unsigned count = 130;
6178
6179         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6180         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6181
6182         while (vmx->emulation_required && count-- != 0) {
6183                 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6184                         return handle_interrupt_window(&vmx->vcpu);
6185
6186                 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6187                         return 1;
6188
6189                 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6190
6191                 if (err == EMULATE_USER_EXIT) {
6192                         ++vcpu->stat.mmio_exits;
6193                         ret = 0;
6194                         goto out;
6195                 }
6196
6197                 if (err != EMULATE_DONE) {
6198                         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6199                         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6200                         vcpu->run->internal.ndata = 0;
6201                         return 0;
6202                 }
6203
6204                 if (vcpu->arch.halt_request) {
6205                         vcpu->arch.halt_request = 0;
6206                         ret = kvm_vcpu_halt(vcpu);
6207                         goto out;
6208                 }
6209
6210                 if (signal_pending(current))
6211                         goto out;
6212                 if (need_resched())
6213                         schedule();
6214         }
6215
6216 out:
6217         return ret;
6218 }
6219
6220 static int __grow_ple_window(int val)
6221 {
6222         if (ple_window_grow < 1)
6223                 return ple_window;
6224
6225         val = min(val, ple_window_actual_max);
6226
6227         if (ple_window_grow < ple_window)
6228                 val *= ple_window_grow;
6229         else
6230                 val += ple_window_grow;
6231
6232         return val;
6233 }
6234
6235 static int __shrink_ple_window(int val, int modifier, int minimum)
6236 {
6237         if (modifier < 1)
6238                 return ple_window;
6239
6240         if (modifier < ple_window)
6241                 val /= modifier;
6242         else
6243                 val -= modifier;
6244
6245         return max(val, minimum);
6246 }
6247
6248 static void grow_ple_window(struct kvm_vcpu *vcpu)
6249 {
6250         struct vcpu_vmx *vmx = to_vmx(vcpu);
6251         int old = vmx->ple_window;
6252
6253         vmx->ple_window = __grow_ple_window(old);
6254
6255         if (vmx->ple_window != old)
6256                 vmx->ple_window_dirty = true;
6257
6258         trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6259 }
6260
6261 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6262 {
6263         struct vcpu_vmx *vmx = to_vmx(vcpu);
6264         int old = vmx->ple_window;
6265
6266         vmx->ple_window = __shrink_ple_window(old,
6267                                               ple_window_shrink, ple_window);
6268
6269         if (vmx->ple_window != old)
6270                 vmx->ple_window_dirty = true;
6271
6272         trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6273 }
6274
6275 /*
6276  * ple_window_actual_max is computed to be one grow_ple_window() below
6277  * ple_window_max. (See __grow_ple_window for the reason.)
6278  * This prevents overflows, because ple_window_max is int.
6279  * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6280  * this process.
6281  * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6282  */
6283 static void update_ple_window_actual_max(void)
6284 {
6285         ple_window_actual_max =
6286                         __shrink_ple_window(max(ple_window_max, ple_window),
6287                                             ple_window_grow, INT_MIN);
6288 }
6289
6290 /*
6291  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6292  */
6293 static void wakeup_handler(void)
6294 {
6295         struct kvm_vcpu *vcpu;
6296         int cpu = smp_processor_id();
6297
6298         spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6299         list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6300                         blocked_vcpu_list) {
6301                 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6302
6303                 if (pi_test_on(pi_desc) == 1)
6304                         kvm_vcpu_kick(vcpu);
6305         }
6306         spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6307 }
6308
6309 static __init int hardware_setup(void)
6310 {
6311         int r = -ENOMEM, i, msr;
6312
6313         rdmsrl_safe(MSR_EFER, &host_efer);
6314
6315         for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6316                 kvm_define_shared_msr(i, vmx_msr_index[i]);
6317
6318         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6319         if (!vmx_io_bitmap_a)
6320                 return r;
6321
6322         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6323         if (!vmx_io_bitmap_b)
6324                 goto out;
6325
6326         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6327         if (!vmx_msr_bitmap_legacy)
6328                 goto out1;
6329
6330         vmx_msr_bitmap_legacy_x2apic =
6331                                 (unsigned long *)__get_free_page(GFP_KERNEL);
6332         if (!vmx_msr_bitmap_legacy_x2apic)
6333                 goto out2;
6334
6335         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6336         if (!vmx_msr_bitmap_longmode)
6337                 goto out3;
6338
6339         vmx_msr_bitmap_longmode_x2apic =
6340                                 (unsigned long *)__get_free_page(GFP_KERNEL);
6341         if (!vmx_msr_bitmap_longmode_x2apic)
6342                 goto out4;
6343
6344         if (nested) {
6345                 vmx_msr_bitmap_nested =
6346                         (unsigned long *)__get_free_page(GFP_KERNEL);
6347                 if (!vmx_msr_bitmap_nested)
6348                         goto out5;
6349         }
6350
6351         vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6352         if (!vmx_vmread_bitmap)
6353                 goto out6;
6354
6355         vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6356         if (!vmx_vmwrite_bitmap)
6357                 goto out7;
6358
6359         memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6360         memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6361
6362         /*
6363          * Allow direct access to the PC debug port (it is often used for I/O
6364          * delays, but the vmexits simply slow things down).
6365          */
6366         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6367         clear_bit(0x80, vmx_io_bitmap_a);
6368
6369         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6370
6371         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6372         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6373         if (nested)
6374                 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6375
6376         if (setup_vmcs_config(&vmcs_config) < 0) {
6377                 r = -EIO;
6378                 goto out8;
6379         }
6380
6381         if (boot_cpu_has(X86_FEATURE_NX))
6382                 kvm_enable_efer_bits(EFER_NX);
6383
6384         if (!cpu_has_vmx_vpid())
6385                 enable_vpid = 0;
6386         if (!cpu_has_vmx_shadow_vmcs())
6387                 enable_shadow_vmcs = 0;
6388         if (enable_shadow_vmcs)
6389                 init_vmcs_shadow_fields();
6390
6391         if (!cpu_has_vmx_ept() ||
6392             !cpu_has_vmx_ept_4levels()) {
6393                 enable_ept = 0;
6394                 enable_unrestricted_guest = 0;
6395                 enable_ept_ad_bits = 0;
6396         }
6397
6398         if (!cpu_has_vmx_ept_ad_bits())
6399                 enable_ept_ad_bits = 0;
6400
6401         if (!cpu_has_vmx_unrestricted_guest())
6402                 enable_unrestricted_guest = 0;
6403
6404         if (!cpu_has_vmx_flexpriority())
6405                 flexpriority_enabled = 0;
6406
6407         /*
6408          * set_apic_access_page_addr() is used to reload apic access
6409          * page upon invalidation.  No need to do anything if not
6410          * using the APIC_ACCESS_ADDR VMCS field.
6411          */
6412         if (!flexpriority_enabled)
6413                 kvm_x86_ops->set_apic_access_page_addr = NULL;
6414
6415         if (!cpu_has_vmx_tpr_shadow())
6416                 kvm_x86_ops->update_cr8_intercept = NULL;
6417
6418         if (enable_ept && !cpu_has_vmx_ept_2m_page())
6419                 kvm_disable_largepages();
6420
6421         if (!cpu_has_vmx_ple())
6422                 ple_gap = 0;
6423
6424         if (!cpu_has_vmx_apicv())
6425                 enable_apicv = 0;
6426
6427         if (cpu_has_vmx_tsc_scaling()) {
6428                 kvm_has_tsc_control = true;
6429                 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6430                 kvm_tsc_scaling_ratio_frac_bits = 48;
6431         }
6432
6433         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6434         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6435         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6436         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6437         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6438         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6439         vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6440
6441         memcpy(vmx_msr_bitmap_legacy_x2apic,
6442                         vmx_msr_bitmap_legacy, PAGE_SIZE);
6443         memcpy(vmx_msr_bitmap_longmode_x2apic,
6444                         vmx_msr_bitmap_longmode, PAGE_SIZE);
6445
6446         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6447
6448         for (msr = 0x800; msr <= 0x8ff; msr++)
6449                 vmx_disable_intercept_msr_read_x2apic(msr);
6450
6451         /* According SDM, in x2apic mode, the whole id reg is used.  But in
6452          * KVM, it only use the highest eight bits. Need to intercept it */
6453         vmx_enable_intercept_msr_read_x2apic(0x802);
6454         /* TMCCT */
6455         vmx_enable_intercept_msr_read_x2apic(0x839);
6456         /* TPR */
6457         vmx_disable_intercept_msr_write_x2apic(0x808);
6458         /* EOI */
6459         vmx_disable_intercept_msr_write_x2apic(0x80b);
6460         /* SELF-IPI */
6461         vmx_disable_intercept_msr_write_x2apic(0x83f);
6462
6463         if (enable_ept) {
6464                 kvm_mmu_set_mask_ptes(0ull,
6465                         (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6466                         (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6467                         0ull, VMX_EPT_EXECUTABLE_MASK);
6468                 ept_set_mmio_spte_mask();
6469                 kvm_enable_tdp();
6470         } else
6471                 kvm_disable_tdp();
6472
6473         update_ple_window_actual_max();
6474
6475         /*
6476          * Only enable PML when hardware supports PML feature, and both EPT
6477          * and EPT A/D bit features are enabled -- PML depends on them to work.
6478          */
6479         if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6480                 enable_pml = 0;
6481
6482         if (!enable_pml) {
6483                 kvm_x86_ops->slot_enable_log_dirty = NULL;
6484                 kvm_x86_ops->slot_disable_log_dirty = NULL;
6485                 kvm_x86_ops->flush_log_dirty = NULL;
6486                 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6487         }
6488
6489         if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6490                 u64 vmx_msr;
6491
6492                 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6493                 cpu_preemption_timer_multi =
6494                          vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6495         } else {
6496                 kvm_x86_ops->set_hv_timer = NULL;
6497                 kvm_x86_ops->cancel_hv_timer = NULL;
6498         }
6499
6500         kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6501
6502         kvm_mce_cap_supported |= MCG_LMCE_P;
6503
6504         return alloc_kvm_area();
6505
6506 out8:
6507         free_page((unsigned long)vmx_vmwrite_bitmap);
6508 out7:
6509         free_page((unsigned long)vmx_vmread_bitmap);
6510 out6:
6511         if (nested)
6512                 free_page((unsigned long)vmx_msr_bitmap_nested);
6513 out5:
6514         free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6515 out4:
6516         free_page((unsigned long)vmx_msr_bitmap_longmode);
6517 out3:
6518         free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6519 out2:
6520         free_page((unsigned long)vmx_msr_bitmap_legacy);
6521 out1:
6522         free_page((unsigned long)vmx_io_bitmap_b);
6523 out:
6524         free_page((unsigned long)vmx_io_bitmap_a);
6525
6526     return r;
6527 }
6528
6529 static __exit void hardware_unsetup(void)
6530 {
6531         free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6532         free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6533         free_page((unsigned long)vmx_msr_bitmap_legacy);
6534         free_page((unsigned long)vmx_msr_bitmap_longmode);
6535         free_page((unsigned long)vmx_io_bitmap_b);
6536         free_page((unsigned long)vmx_io_bitmap_a);
6537         free_page((unsigned long)vmx_vmwrite_bitmap);
6538         free_page((unsigned long)vmx_vmread_bitmap);
6539         if (nested)
6540                 free_page((unsigned long)vmx_msr_bitmap_nested);
6541
6542         free_kvm_area();
6543 }
6544
6545 /*
6546  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6547  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6548  */
6549 static int handle_pause(struct kvm_vcpu *vcpu)
6550 {
6551         if (ple_gap)
6552                 grow_ple_window(vcpu);
6553
6554         skip_emulated_instruction(vcpu);
6555         kvm_vcpu_on_spin(vcpu);
6556
6557         return 1;
6558 }
6559
6560 static int handle_nop(struct kvm_vcpu *vcpu)
6561 {
6562         skip_emulated_instruction(vcpu);
6563         return 1;
6564 }
6565
6566 static int handle_mwait(struct kvm_vcpu *vcpu)
6567 {
6568         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6569         return handle_nop(vcpu);
6570 }
6571
6572 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6573 {
6574         return 1;
6575 }
6576
6577 static int handle_monitor(struct kvm_vcpu *vcpu)
6578 {
6579         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6580         return handle_nop(vcpu);
6581 }
6582
6583 /*
6584  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6585  * We could reuse a single VMCS for all the L2 guests, but we also want the
6586  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6587  * allows keeping them loaded on the processor, and in the future will allow
6588  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6589  * every entry if they never change.
6590  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6591  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6592  *
6593  * The following functions allocate and free a vmcs02 in this pool.
6594  */
6595
6596 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6597 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6598 {
6599         struct vmcs02_list *item;
6600         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6601                 if (item->vmptr == vmx->nested.current_vmptr) {
6602                         list_move(&item->list, &vmx->nested.vmcs02_pool);
6603                         return &item->vmcs02;
6604                 }
6605
6606         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6607                 /* Recycle the least recently used VMCS. */
6608                 item = list_last_entry(&vmx->nested.vmcs02_pool,
6609                                        struct vmcs02_list, list);
6610                 item->vmptr = vmx->nested.current_vmptr;
6611                 list_move(&item->list, &vmx->nested.vmcs02_pool);
6612                 return &item->vmcs02;
6613         }
6614
6615         /* Create a new VMCS */
6616         item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6617         if (!item)
6618                 return NULL;
6619         item->vmcs02.vmcs = alloc_vmcs();
6620         if (!item->vmcs02.vmcs) {
6621                 kfree(item);
6622                 return NULL;
6623         }
6624         loaded_vmcs_init(&item->vmcs02);
6625         item->vmptr = vmx->nested.current_vmptr;
6626         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6627         vmx->nested.vmcs02_num++;
6628         return &item->vmcs02;
6629 }
6630
6631 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6632 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6633 {
6634         struct vmcs02_list *item;
6635         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6636                 if (item->vmptr == vmptr) {
6637                         free_loaded_vmcs(&item->vmcs02);
6638                         list_del(&item->list);
6639                         kfree(item);
6640                         vmx->nested.vmcs02_num--;
6641                         return;
6642                 }
6643 }
6644
6645 /*
6646  * Free all VMCSs saved for this vcpu, except the one pointed by
6647  * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6648  * must be &vmx->vmcs01.
6649  */
6650 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6651 {
6652         struct vmcs02_list *item, *n;
6653
6654         WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6655         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6656                 /*
6657                  * Something will leak if the above WARN triggers.  Better than
6658                  * a use-after-free.
6659                  */
6660                 if (vmx->loaded_vmcs == &item->vmcs02)
6661                         continue;
6662
6663                 free_loaded_vmcs(&item->vmcs02);
6664                 list_del(&item->list);
6665                 kfree(item);
6666                 vmx->nested.vmcs02_num--;
6667         }
6668 }
6669
6670 /*
6671  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6672  * set the success or error code of an emulated VMX instruction, as specified
6673  * by Vol 2B, VMX Instruction Reference, "Conventions".
6674  */
6675 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6676 {
6677         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6678                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6679                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6680 }
6681
6682 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6683 {
6684         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6685                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6686                             X86_EFLAGS_SF | X86_EFLAGS_OF))
6687                         | X86_EFLAGS_CF);
6688 }
6689
6690 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6691                                         u32 vm_instruction_error)
6692 {
6693         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6694                 /*
6695                  * failValid writes the error number to the current VMCS, which
6696                  * can't be done there isn't a current VMCS.
6697                  */
6698                 nested_vmx_failInvalid(vcpu);
6699                 return;
6700         }
6701         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6702                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6703                             X86_EFLAGS_SF | X86_EFLAGS_OF))
6704                         | X86_EFLAGS_ZF);
6705         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6706         /*
6707          * We don't need to force a shadow sync because
6708          * VM_INSTRUCTION_ERROR is not shadowed
6709          */
6710 }
6711
6712 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6713 {
6714         /* TODO: not to reset guest simply here. */
6715         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6716         pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6717 }
6718
6719 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6720 {
6721         struct vcpu_vmx *vmx =
6722                 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6723
6724         vmx->nested.preemption_timer_expired = true;
6725         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6726         kvm_vcpu_kick(&vmx->vcpu);
6727
6728         return HRTIMER_NORESTART;
6729 }
6730
6731 /*
6732  * Decode the memory-address operand of a vmx instruction, as recorded on an
6733  * exit caused by such an instruction (run by a guest hypervisor).
6734  * On success, returns 0. When the operand is invalid, returns 1 and throws
6735  * #UD or #GP.
6736  */
6737 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6738                                  unsigned long exit_qualification,
6739                                  u32 vmx_instruction_info, bool wr, gva_t *ret)
6740 {
6741         gva_t off;
6742         bool exn;
6743         struct kvm_segment s;
6744
6745         /*
6746          * According to Vol. 3B, "Information for VM Exits Due to Instruction
6747          * Execution", on an exit, vmx_instruction_info holds most of the
6748          * addressing components of the operand. Only the displacement part
6749          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6750          * For how an actual address is calculated from all these components,
6751          * refer to Vol. 1, "Operand Addressing".
6752          */
6753         int  scaling = vmx_instruction_info & 3;
6754         int  addr_size = (vmx_instruction_info >> 7) & 7;
6755         bool is_reg = vmx_instruction_info & (1u << 10);
6756         int  seg_reg = (vmx_instruction_info >> 15) & 7;
6757         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
6758         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6759         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
6760         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
6761
6762         if (is_reg) {
6763                 kvm_queue_exception(vcpu, UD_VECTOR);
6764                 return 1;
6765         }
6766
6767         /* Addr = segment_base + offset */
6768         /* offset = base + [index * scale] + displacement */
6769         off = exit_qualification; /* holds the displacement */
6770         if (base_is_valid)
6771                 off += kvm_register_read(vcpu, base_reg);
6772         if (index_is_valid)
6773                 off += kvm_register_read(vcpu, index_reg)<<scaling;
6774         vmx_get_segment(vcpu, &s, seg_reg);
6775         *ret = s.base + off;
6776
6777         if (addr_size == 1) /* 32 bit */
6778                 *ret &= 0xffffffff;
6779
6780         /* Checks for #GP/#SS exceptions. */
6781         exn = false;
6782         if (is_protmode(vcpu)) {
6783                 /* Protected mode: apply checks for segment validity in the
6784                  * following order:
6785                  * - segment type check (#GP(0) may be thrown)
6786                  * - usability check (#GP(0)/#SS(0))
6787                  * - limit check (#GP(0)/#SS(0))
6788                  */
6789                 if (wr)
6790                         /* #GP(0) if the destination operand is located in a
6791                          * read-only data segment or any code segment.
6792                          */
6793                         exn = ((s.type & 0xa) == 0 || (s.type & 8));
6794                 else
6795                         /* #GP(0) if the source operand is located in an
6796                          * execute-only code segment
6797                          */
6798                         exn = ((s.type & 0xa) == 8);
6799         }
6800         if (exn) {
6801                 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6802                 return 1;
6803         }
6804         if (is_long_mode(vcpu)) {
6805                 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6806                  * non-canonical form. This is an only check for long mode.
6807                  */
6808                 exn = is_noncanonical_address(*ret);
6809         } else if (is_protmode(vcpu)) {
6810                 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6811                  */
6812                 exn = (s.unusable != 0);
6813                 /* Protected mode: #GP(0)/#SS(0) if the memory
6814                  * operand is outside the segment limit.
6815                  */
6816                 exn = exn || (off + sizeof(u64) > s.limit);
6817         }
6818         if (exn) {
6819                 kvm_queue_exception_e(vcpu,
6820                                       seg_reg == VCPU_SREG_SS ?
6821                                                 SS_VECTOR : GP_VECTOR,
6822                                       0);
6823                 return 1;
6824         }
6825
6826         return 0;
6827 }
6828
6829 /*
6830  * This function performs the various checks including
6831  * - if it's 4KB aligned
6832  * - No bits beyond the physical address width are set
6833  * - Returns 0 on success or else 1
6834  * (Intel SDM Section 30.3)
6835  */
6836 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6837                                   gpa_t *vmpointer)
6838 {
6839         gva_t gva;
6840         gpa_t vmptr;
6841         struct x86_exception e;
6842         struct page *page;
6843         struct vcpu_vmx *vmx = to_vmx(vcpu);
6844         int maxphyaddr = cpuid_maxphyaddr(vcpu);
6845
6846         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6847                         vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6848                 return 1;
6849
6850         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6851                                 sizeof(vmptr), &e)) {
6852                 kvm_inject_page_fault(vcpu, &e);
6853                 return 1;
6854         }
6855
6856         switch (exit_reason) {
6857         case EXIT_REASON_VMON:
6858                 /*
6859                  * SDM 3: 24.11.5
6860                  * The first 4 bytes of VMXON region contain the supported
6861                  * VMCS revision identifier
6862                  *
6863                  * Note - IA32_VMX_BASIC[48] will never be 1
6864                  * for the nested case;
6865                  * which replaces physical address width with 32
6866                  *
6867                  */
6868                 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6869                         nested_vmx_failInvalid(vcpu);
6870                         skip_emulated_instruction(vcpu);
6871                         return 1;
6872                 }
6873
6874                 page = nested_get_page(vcpu, vmptr);
6875                 if (page == NULL ||
6876                     *(u32 *)kmap(page) != VMCS12_REVISION) {
6877                         nested_vmx_failInvalid(vcpu);
6878                         kunmap(page);
6879                         skip_emulated_instruction(vcpu);
6880                         return 1;
6881                 }
6882                 kunmap(page);
6883                 vmx->nested.vmxon_ptr = vmptr;
6884                 break;
6885         case EXIT_REASON_VMCLEAR:
6886                 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6887                         nested_vmx_failValid(vcpu,
6888                                              VMXERR_VMCLEAR_INVALID_ADDRESS);
6889                         skip_emulated_instruction(vcpu);
6890                         return 1;
6891                 }
6892
6893                 if (vmptr == vmx->nested.vmxon_ptr) {
6894                         nested_vmx_failValid(vcpu,
6895                                              VMXERR_VMCLEAR_VMXON_POINTER);
6896                         skip_emulated_instruction(vcpu);
6897                         return 1;
6898                 }
6899                 break;
6900         case EXIT_REASON_VMPTRLD:
6901                 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6902                         nested_vmx_failValid(vcpu,
6903                                              VMXERR_VMPTRLD_INVALID_ADDRESS);
6904                         skip_emulated_instruction(vcpu);
6905                         return 1;
6906                 }
6907
6908                 if (vmptr == vmx->nested.vmxon_ptr) {
6909                         nested_vmx_failValid(vcpu,
6910                                              VMXERR_VMCLEAR_VMXON_POINTER);
6911                         skip_emulated_instruction(vcpu);
6912                         return 1;
6913                 }
6914                 break;
6915         default:
6916                 return 1; /* shouldn't happen */
6917         }
6918
6919         if (vmpointer)
6920                 *vmpointer = vmptr;
6921         return 0;
6922 }
6923
6924 /*
6925  * Emulate the VMXON instruction.
6926  * Currently, we just remember that VMX is active, and do not save or even
6927  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6928  * do not currently need to store anything in that guest-allocated memory
6929  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6930  * argument is different from the VMXON pointer (which the spec says they do).
6931  */
6932 static int handle_vmon(struct kvm_vcpu *vcpu)
6933 {
6934         struct kvm_segment cs;
6935         struct vcpu_vmx *vmx = to_vmx(vcpu);
6936         struct vmcs *shadow_vmcs;
6937         const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6938                 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6939
6940         /* The Intel VMX Instruction Reference lists a bunch of bits that
6941          * are prerequisite to running VMXON, most notably cr4.VMXE must be
6942          * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6943          * Otherwise, we should fail with #UD. We test these now:
6944          */
6945         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6946             !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6947             (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6948                 kvm_queue_exception(vcpu, UD_VECTOR);
6949                 return 1;
6950         }
6951
6952         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6953         if (is_long_mode(vcpu) && !cs.l) {
6954                 kvm_queue_exception(vcpu, UD_VECTOR);
6955                 return 1;
6956         }
6957
6958         if (vmx_get_cpl(vcpu)) {
6959                 kvm_inject_gp(vcpu, 0);
6960                 return 1;
6961         }
6962
6963         if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6964                 return 1;
6965
6966         if (vmx->nested.vmxon) {
6967                 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6968                 skip_emulated_instruction(vcpu);
6969                 return 1;
6970         }
6971
6972         if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6973                         != VMXON_NEEDED_FEATURES) {
6974                 kvm_inject_gp(vcpu, 0);
6975                 return 1;
6976         }
6977
6978         if (enable_shadow_vmcs) {
6979                 shadow_vmcs = alloc_vmcs();
6980                 if (!shadow_vmcs)
6981                         return -ENOMEM;
6982                 /* mark vmcs as shadow */
6983                 shadow_vmcs->revision_id |= (1u << 31);
6984                 /* init shadow vmcs */
6985                 vmcs_clear(shadow_vmcs);
6986                 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6987         }
6988
6989         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6990         vmx->nested.vmcs02_num = 0;
6991
6992         hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6993                      HRTIMER_MODE_REL);
6994         vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6995
6996         vmx->nested.vmxon = true;
6997
6998         skip_emulated_instruction(vcpu);
6999         nested_vmx_succeed(vcpu);
7000         return 1;
7001 }
7002
7003 /*
7004  * Intel's VMX Instruction Reference specifies a common set of prerequisites
7005  * for running VMX instructions (except VMXON, whose prerequisites are
7006  * slightly different). It also specifies what exception to inject otherwise.
7007  */
7008 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7009 {
7010         struct kvm_segment cs;
7011         struct vcpu_vmx *vmx = to_vmx(vcpu);
7012
7013         if (!vmx->nested.vmxon) {
7014                 kvm_queue_exception(vcpu, UD_VECTOR);
7015                 return 0;
7016         }
7017
7018         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7019         if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
7020             (is_long_mode(vcpu) && !cs.l)) {
7021                 kvm_queue_exception(vcpu, UD_VECTOR);
7022                 return 0;
7023         }
7024
7025         if (vmx_get_cpl(vcpu)) {
7026                 kvm_inject_gp(vcpu, 0);
7027                 return 0;
7028         }
7029
7030         return 1;
7031 }
7032
7033 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7034 {
7035         if (vmx->nested.current_vmptr == -1ull)
7036                 return;
7037
7038         /* current_vmptr and current_vmcs12 are always set/reset together */
7039         if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
7040                 return;
7041
7042         if (enable_shadow_vmcs) {
7043                 /* copy to memory all shadowed fields in case
7044                    they were modified */
7045                 copy_shadow_to_vmcs12(vmx);
7046                 vmx->nested.sync_shadow_vmcs = false;
7047                 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
7048                                 SECONDARY_EXEC_SHADOW_VMCS);
7049                 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7050         }
7051         vmx->nested.posted_intr_nv = -1;
7052         kunmap(vmx->nested.current_vmcs12_page);
7053         nested_release_page(vmx->nested.current_vmcs12_page);
7054         vmx->nested.current_vmptr = -1ull;
7055         vmx->nested.current_vmcs12 = NULL;
7056 }
7057
7058 /*
7059  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7060  * just stops using VMX.
7061  */
7062 static void free_nested(struct vcpu_vmx *vmx)
7063 {
7064         if (!vmx->nested.vmxon)
7065                 return;
7066
7067         vmx->nested.vmxon = false;
7068         free_vpid(vmx->nested.vpid02);
7069         nested_release_vmcs12(vmx);
7070         if (enable_shadow_vmcs)
7071                 free_vmcs(vmx->nested.current_shadow_vmcs);
7072         /* Unpin physical memory we referred to in current vmcs02 */
7073         if (vmx->nested.apic_access_page) {
7074                 nested_release_page(vmx->nested.apic_access_page);
7075                 vmx->nested.apic_access_page = NULL;
7076         }
7077         if (vmx->nested.virtual_apic_page) {
7078                 nested_release_page(vmx->nested.virtual_apic_page);
7079                 vmx->nested.virtual_apic_page = NULL;
7080         }
7081         if (vmx->nested.pi_desc_page) {
7082                 kunmap(vmx->nested.pi_desc_page);
7083                 nested_release_page(vmx->nested.pi_desc_page);
7084                 vmx->nested.pi_desc_page = NULL;
7085                 vmx->nested.pi_desc = NULL;
7086         }
7087
7088         nested_free_all_saved_vmcss(vmx);
7089 }
7090
7091 /* Emulate the VMXOFF instruction */
7092 static int handle_vmoff(struct kvm_vcpu *vcpu)
7093 {
7094         if (!nested_vmx_check_permission(vcpu))
7095                 return 1;
7096         free_nested(to_vmx(vcpu));
7097         skip_emulated_instruction(vcpu);
7098         nested_vmx_succeed(vcpu);
7099         return 1;
7100 }
7101
7102 /* Emulate the VMCLEAR instruction */
7103 static int handle_vmclear(struct kvm_vcpu *vcpu)
7104 {
7105         struct vcpu_vmx *vmx = to_vmx(vcpu);
7106         gpa_t vmptr;
7107         struct vmcs12 *vmcs12;
7108         struct page *page;
7109
7110         if (!nested_vmx_check_permission(vcpu))
7111                 return 1;
7112
7113         if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
7114                 return 1;
7115
7116         if (vmptr == vmx->nested.current_vmptr)
7117                 nested_release_vmcs12(vmx);
7118
7119         page = nested_get_page(vcpu, vmptr);
7120         if (page == NULL) {
7121                 /*
7122                  * For accurate processor emulation, VMCLEAR beyond available
7123                  * physical memory should do nothing at all. However, it is
7124                  * possible that a nested vmx bug, not a guest hypervisor bug,
7125                  * resulted in this case, so let's shut down before doing any
7126                  * more damage:
7127                  */
7128                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7129                 return 1;
7130         }
7131         vmcs12 = kmap(page);
7132         vmcs12->launch_state = 0;
7133         kunmap(page);
7134         nested_release_page(page);
7135
7136         nested_free_vmcs02(vmx, vmptr);
7137
7138         skip_emulated_instruction(vcpu);
7139         nested_vmx_succeed(vcpu);
7140         return 1;
7141 }
7142
7143 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7144
7145 /* Emulate the VMLAUNCH instruction */
7146 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7147 {
7148         return nested_vmx_run(vcpu, true);
7149 }
7150
7151 /* Emulate the VMRESUME instruction */
7152 static int handle_vmresume(struct kvm_vcpu *vcpu)
7153 {
7154
7155         return nested_vmx_run(vcpu, false);
7156 }
7157
7158 enum vmcs_field_type {
7159         VMCS_FIELD_TYPE_U16 = 0,
7160         VMCS_FIELD_TYPE_U64 = 1,
7161         VMCS_FIELD_TYPE_U32 = 2,
7162         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
7163 };
7164
7165 static inline int vmcs_field_type(unsigned long field)
7166 {
7167         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
7168                 return VMCS_FIELD_TYPE_U32;
7169         return (field >> 13) & 0x3 ;
7170 }
7171
7172 static inline int vmcs_field_readonly(unsigned long field)
7173 {
7174         return (((field >> 10) & 0x3) == 1);
7175 }
7176
7177 /*
7178  * Read a vmcs12 field. Since these can have varying lengths and we return
7179  * one type, we chose the biggest type (u64) and zero-extend the return value
7180  * to that size. Note that the caller, handle_vmread, might need to use only
7181  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7182  * 64-bit fields are to be returned).
7183  */
7184 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7185                                   unsigned long field, u64 *ret)
7186 {
7187         short offset = vmcs_field_to_offset(field);
7188         char *p;
7189
7190         if (offset < 0)
7191                 return offset;
7192
7193         p = ((char *)(get_vmcs12(vcpu))) + offset;
7194
7195         switch (vmcs_field_type(field)) {
7196         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7197                 *ret = *((natural_width *)p);
7198                 return 0;
7199         case VMCS_FIELD_TYPE_U16:
7200                 *ret = *((u16 *)p);
7201                 return 0;
7202         case VMCS_FIELD_TYPE_U32:
7203                 *ret = *((u32 *)p);
7204                 return 0;
7205         case VMCS_FIELD_TYPE_U64:
7206                 *ret = *((u64 *)p);
7207                 return 0;
7208         default:
7209                 WARN_ON(1);
7210                 return -ENOENT;
7211         }
7212 }
7213
7214
7215 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7216                                    unsigned long field, u64 field_value){
7217         short offset = vmcs_field_to_offset(field);
7218         char *p = ((char *) get_vmcs12(vcpu)) + offset;
7219         if (offset < 0)
7220                 return offset;
7221
7222         switch (vmcs_field_type(field)) {
7223         case VMCS_FIELD_TYPE_U16:
7224                 *(u16 *)p = field_value;
7225                 return 0;
7226         case VMCS_FIELD_TYPE_U32:
7227                 *(u32 *)p = field_value;
7228                 return 0;
7229         case VMCS_FIELD_TYPE_U64:
7230                 *(u64 *)p = field_value;
7231                 return 0;
7232         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7233                 *(natural_width *)p = field_value;
7234                 return 0;
7235         default:
7236                 WARN_ON(1);
7237                 return -ENOENT;
7238         }
7239
7240 }
7241
7242 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7243 {
7244         int i;
7245         unsigned long field;
7246         u64 field_value;
7247         struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7248         const unsigned long *fields = shadow_read_write_fields;
7249         const int num_fields = max_shadow_read_write_fields;
7250
7251         preempt_disable();
7252
7253         vmcs_load(shadow_vmcs);
7254
7255         for (i = 0; i < num_fields; i++) {
7256                 field = fields[i];
7257                 switch (vmcs_field_type(field)) {
7258                 case VMCS_FIELD_TYPE_U16:
7259                         field_value = vmcs_read16(field);
7260                         break;
7261                 case VMCS_FIELD_TYPE_U32:
7262                         field_value = vmcs_read32(field);
7263                         break;
7264                 case VMCS_FIELD_TYPE_U64:
7265                         field_value = vmcs_read64(field);
7266                         break;
7267                 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7268                         field_value = vmcs_readl(field);
7269                         break;
7270                 default:
7271                         WARN_ON(1);
7272                         continue;
7273                 }
7274                 vmcs12_write_any(&vmx->vcpu, field, field_value);
7275         }
7276
7277         vmcs_clear(shadow_vmcs);
7278         vmcs_load(vmx->loaded_vmcs->vmcs);
7279
7280         preempt_enable();
7281 }
7282
7283 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7284 {
7285         const unsigned long *fields[] = {
7286                 shadow_read_write_fields,
7287                 shadow_read_only_fields
7288         };
7289         const int max_fields[] = {
7290                 max_shadow_read_write_fields,
7291                 max_shadow_read_only_fields
7292         };
7293         int i, q;
7294         unsigned long field;
7295         u64 field_value = 0;
7296         struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7297
7298         vmcs_load(shadow_vmcs);
7299
7300         for (q = 0; q < ARRAY_SIZE(fields); q++) {
7301                 for (i = 0; i < max_fields[q]; i++) {
7302                         field = fields[q][i];
7303                         vmcs12_read_any(&vmx->vcpu, field, &field_value);
7304
7305                         switch (vmcs_field_type(field)) {
7306                         case VMCS_FIELD_TYPE_U16:
7307                                 vmcs_write16(field, (u16)field_value);
7308                                 break;
7309                         case VMCS_FIELD_TYPE_U32:
7310                                 vmcs_write32(field, (u32)field_value);
7311                                 break;
7312                         case VMCS_FIELD_TYPE_U64:
7313                                 vmcs_write64(field, (u64)field_value);
7314                                 break;
7315                         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7316                                 vmcs_writel(field, (long)field_value);
7317                                 break;
7318                         default:
7319                                 WARN_ON(1);
7320                                 break;
7321                         }
7322                 }
7323         }
7324
7325         vmcs_clear(shadow_vmcs);
7326         vmcs_load(vmx->loaded_vmcs->vmcs);
7327 }
7328
7329 /*
7330  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7331  * used before) all generate the same failure when it is missing.
7332  */
7333 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7334 {
7335         struct vcpu_vmx *vmx = to_vmx(vcpu);
7336         if (vmx->nested.current_vmptr == -1ull) {
7337                 nested_vmx_failInvalid(vcpu);
7338                 skip_emulated_instruction(vcpu);
7339                 return 0;
7340         }
7341         return 1;
7342 }
7343
7344 static int handle_vmread(struct kvm_vcpu *vcpu)
7345 {
7346         unsigned long field;
7347         u64 field_value;
7348         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7349         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7350         gva_t gva = 0;
7351
7352         if (!nested_vmx_check_permission(vcpu) ||
7353             !nested_vmx_check_vmcs12(vcpu))
7354                 return 1;
7355
7356         /* Decode instruction info and find the field to read */
7357         field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7358         /* Read the field, zero-extended to a u64 field_value */
7359         if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7360                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7361                 skip_emulated_instruction(vcpu);
7362                 return 1;
7363         }
7364         /*
7365          * Now copy part of this value to register or memory, as requested.
7366          * Note that the number of bits actually copied is 32 or 64 depending
7367          * on the guest's mode (32 or 64 bit), not on the given field's length.
7368          */
7369         if (vmx_instruction_info & (1u << 10)) {
7370                 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7371                         field_value);
7372         } else {
7373                 if (get_vmx_mem_address(vcpu, exit_qualification,
7374                                 vmx_instruction_info, true, &gva))
7375                         return 1;
7376                 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7377                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7378                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7379         }
7380
7381         nested_vmx_succeed(vcpu);
7382         skip_emulated_instruction(vcpu);
7383         return 1;
7384 }
7385
7386
7387 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7388 {
7389         unsigned long field;
7390         gva_t gva;
7391         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7392         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7393         /* The value to write might be 32 or 64 bits, depending on L1's long
7394          * mode, and eventually we need to write that into a field of several
7395          * possible lengths. The code below first zero-extends the value to 64
7396          * bit (field_value), and then copies only the appropriate number of
7397          * bits into the vmcs12 field.
7398          */
7399         u64 field_value = 0;
7400         struct x86_exception e;
7401
7402         if (!nested_vmx_check_permission(vcpu) ||
7403             !nested_vmx_check_vmcs12(vcpu))
7404                 return 1;
7405
7406         if (vmx_instruction_info & (1u << 10))
7407                 field_value = kvm_register_readl(vcpu,
7408                         (((vmx_instruction_info) >> 3) & 0xf));
7409         else {
7410                 if (get_vmx_mem_address(vcpu, exit_qualification,
7411                                 vmx_instruction_info, false, &gva))
7412                         return 1;
7413                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7414                            &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7415                         kvm_inject_page_fault(vcpu, &e);
7416                         return 1;
7417                 }
7418         }
7419
7420
7421         field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7422         if (vmcs_field_readonly(field)) {
7423                 nested_vmx_failValid(vcpu,
7424                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7425                 skip_emulated_instruction(vcpu);
7426                 return 1;
7427         }
7428
7429         if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7430                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7431                 skip_emulated_instruction(vcpu);
7432                 return 1;
7433         }
7434
7435         nested_vmx_succeed(vcpu);
7436         skip_emulated_instruction(vcpu);
7437         return 1;
7438 }
7439
7440 /* Emulate the VMPTRLD instruction */
7441 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7442 {
7443         struct vcpu_vmx *vmx = to_vmx(vcpu);
7444         gpa_t vmptr;
7445
7446         if (!nested_vmx_check_permission(vcpu))
7447                 return 1;
7448
7449         if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7450                 return 1;
7451
7452         if (vmx->nested.current_vmptr != vmptr) {
7453                 struct vmcs12 *new_vmcs12;
7454                 struct page *page;
7455                 page = nested_get_page(vcpu, vmptr);
7456                 if (page == NULL) {
7457                         nested_vmx_failInvalid(vcpu);
7458                         skip_emulated_instruction(vcpu);
7459                         return 1;
7460                 }
7461                 new_vmcs12 = kmap(page);
7462                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7463                         kunmap(page);
7464                         nested_release_page_clean(page);
7465                         nested_vmx_failValid(vcpu,
7466                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7467                         skip_emulated_instruction(vcpu);
7468                         return 1;
7469                 }
7470
7471                 nested_release_vmcs12(vmx);
7472                 vmx->nested.current_vmptr = vmptr;
7473                 vmx->nested.current_vmcs12 = new_vmcs12;
7474                 vmx->nested.current_vmcs12_page = page;
7475                 if (enable_shadow_vmcs) {
7476                         vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7477                                       SECONDARY_EXEC_SHADOW_VMCS);
7478                         vmcs_write64(VMCS_LINK_POINTER,
7479                                      __pa(vmx->nested.current_shadow_vmcs));
7480                         vmx->nested.sync_shadow_vmcs = true;
7481                 }
7482         }
7483
7484         nested_vmx_succeed(vcpu);
7485         skip_emulated_instruction(vcpu);
7486         return 1;
7487 }
7488
7489 /* Emulate the VMPTRST instruction */
7490 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7491 {
7492         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7493         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7494         gva_t vmcs_gva;
7495         struct x86_exception e;
7496
7497         if (!nested_vmx_check_permission(vcpu))
7498                 return 1;
7499
7500         if (get_vmx_mem_address(vcpu, exit_qualification,
7501                         vmx_instruction_info, true, &vmcs_gva))
7502                 return 1;
7503         /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7504         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7505                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
7506                                  sizeof(u64), &e)) {
7507                 kvm_inject_page_fault(vcpu, &e);
7508                 return 1;
7509         }
7510         nested_vmx_succeed(vcpu);
7511         skip_emulated_instruction(vcpu);
7512         return 1;
7513 }
7514
7515 /* Emulate the INVEPT instruction */
7516 static int handle_invept(struct kvm_vcpu *vcpu)
7517 {
7518         struct vcpu_vmx *vmx = to_vmx(vcpu);
7519         u32 vmx_instruction_info, types;
7520         unsigned long type;
7521         gva_t gva;
7522         struct x86_exception e;
7523         struct {
7524                 u64 eptp, gpa;
7525         } operand;
7526
7527         if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7528               SECONDARY_EXEC_ENABLE_EPT) ||
7529             !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7530                 kvm_queue_exception(vcpu, UD_VECTOR);
7531                 return 1;
7532         }
7533
7534         if (!nested_vmx_check_permission(vcpu))
7535                 return 1;
7536
7537         if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7538                 kvm_queue_exception(vcpu, UD_VECTOR);
7539                 return 1;
7540         }
7541
7542         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7543         type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7544
7545         types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7546
7547         if (!(types & (1UL << type))) {
7548                 nested_vmx_failValid(vcpu,
7549                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7550                 skip_emulated_instruction(vcpu);
7551                 return 1;
7552         }
7553
7554         /* According to the Intel VMX instruction reference, the memory
7555          * operand is read even if it isn't needed (e.g., for type==global)
7556          */
7557         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7558                         vmx_instruction_info, false, &gva))
7559                 return 1;
7560         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7561                                 sizeof(operand), &e)) {
7562                 kvm_inject_page_fault(vcpu, &e);
7563                 return 1;
7564         }
7565
7566         switch (type) {
7567         case VMX_EPT_EXTENT_GLOBAL:
7568                 kvm_mmu_sync_roots(vcpu);
7569                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7570                 nested_vmx_succeed(vcpu);
7571                 break;
7572         default:
7573                 /* Trap single context invalidation invept calls */
7574                 BUG_ON(1);
7575                 break;
7576         }
7577
7578         skip_emulated_instruction(vcpu);
7579         return 1;
7580 }
7581
7582 static int handle_invvpid(struct kvm_vcpu *vcpu)
7583 {
7584         struct vcpu_vmx *vmx = to_vmx(vcpu);
7585         u32 vmx_instruction_info;
7586         unsigned long type, types;
7587         gva_t gva;
7588         struct x86_exception e;
7589         int vpid;
7590
7591         if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7592               SECONDARY_EXEC_ENABLE_VPID) ||
7593                         !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7594                 kvm_queue_exception(vcpu, UD_VECTOR);
7595                 return 1;
7596         }
7597
7598         if (!nested_vmx_check_permission(vcpu))
7599                 return 1;
7600
7601         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7602         type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7603
7604         types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7605
7606         if (!(types & (1UL << type))) {
7607                 nested_vmx_failValid(vcpu,
7608                         VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7609                 skip_emulated_instruction(vcpu);
7610                 return 1;
7611         }
7612
7613         /* according to the intel vmx instruction reference, the memory
7614          * operand is read even if it isn't needed (e.g., for type==global)
7615          */
7616         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7617                         vmx_instruction_info, false, &gva))
7618                 return 1;
7619         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7620                                 sizeof(u32), &e)) {
7621                 kvm_inject_page_fault(vcpu, &e);
7622                 return 1;
7623         }
7624
7625         switch (type) {
7626         case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7627                 /*
7628                  * Old versions of KVM use the single-context version so we
7629                  * have to support it; just treat it the same as all-context.
7630                  */
7631         case VMX_VPID_EXTENT_ALL_CONTEXT:
7632                 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
7633                 nested_vmx_succeed(vcpu);
7634                 break;
7635         default:
7636                 /* Trap individual address invalidation invvpid calls */
7637                 BUG_ON(1);
7638                 break;
7639         }
7640
7641         skip_emulated_instruction(vcpu);
7642         return 1;
7643 }
7644
7645 static int handle_pml_full(struct kvm_vcpu *vcpu)
7646 {
7647         unsigned long exit_qualification;
7648
7649         trace_kvm_pml_full(vcpu->vcpu_id);
7650
7651         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7652
7653         /*
7654          * PML buffer FULL happened while executing iret from NMI,
7655          * "blocked by NMI" bit has to be set before next VM entry.
7656          */
7657         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7658                         cpu_has_virtual_nmis() &&
7659                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7660                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7661                                 GUEST_INTR_STATE_NMI);
7662
7663         /*
7664          * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7665          * here.., and there's no userspace involvement needed for PML.
7666          */
7667         return 1;
7668 }
7669
7670 static int handle_pcommit(struct kvm_vcpu *vcpu)
7671 {
7672         /* we never catch pcommit instruct for L1 guest. */
7673         WARN_ON(1);
7674         return 1;
7675 }
7676
7677 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7678 {
7679         kvm_lapic_expired_hv_timer(vcpu);
7680         return 1;
7681 }
7682
7683 /*
7684  * The exit handlers return 1 if the exit was handled fully and guest execution
7685  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
7686  * to be done to userspace and return 0.
7687  */
7688 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7689         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
7690         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
7691         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
7692         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
7693         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
7694         [EXIT_REASON_CR_ACCESS]               = handle_cr,
7695         [EXIT_REASON_DR_ACCESS]               = handle_dr,
7696         [EXIT_REASON_CPUID]                   = handle_cpuid,
7697         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
7698         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
7699         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
7700         [EXIT_REASON_HLT]                     = handle_halt,
7701         [EXIT_REASON_INVD]                    = handle_invd,
7702         [EXIT_REASON_INVLPG]                  = handle_invlpg,
7703         [EXIT_REASON_RDPMC]                   = handle_rdpmc,
7704         [EXIT_REASON_VMCALL]                  = handle_vmcall,
7705         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
7706         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
7707         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
7708         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
7709         [EXIT_REASON_VMREAD]                  = handle_vmread,
7710         [EXIT_REASON_VMRESUME]                = handle_vmresume,
7711         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
7712         [EXIT_REASON_VMOFF]                   = handle_vmoff,
7713         [EXIT_REASON_VMON]                    = handle_vmon,
7714         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
7715         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
7716         [EXIT_REASON_APIC_WRITE]              = handle_apic_write,
7717         [EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
7718         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
7719         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
7720         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
7721         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
7722         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
7723         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
7724         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
7725         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_mwait,
7726         [EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
7727         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
7728         [EXIT_REASON_INVEPT]                  = handle_invept,
7729         [EXIT_REASON_INVVPID]                 = handle_invvpid,
7730         [EXIT_REASON_XSAVES]                  = handle_xsaves,
7731         [EXIT_REASON_XRSTORS]                 = handle_xrstors,
7732         [EXIT_REASON_PML_FULL]                = handle_pml_full,
7733         [EXIT_REASON_PCOMMIT]                 = handle_pcommit,
7734         [EXIT_REASON_PREEMPTION_TIMER]        = handle_preemption_timer,
7735 };
7736
7737 static const int kvm_vmx_max_exit_handlers =
7738         ARRAY_SIZE(kvm_vmx_exit_handlers);
7739
7740 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7741                                        struct vmcs12 *vmcs12)
7742 {
7743         unsigned long exit_qualification;
7744         gpa_t bitmap, last_bitmap;
7745         unsigned int port;
7746         int size;
7747         u8 b;
7748
7749         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7750                 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7751
7752         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7753
7754         port = exit_qualification >> 16;
7755         size = (exit_qualification & 7) + 1;
7756
7757         last_bitmap = (gpa_t)-1;
7758         b = -1;
7759
7760         while (size > 0) {
7761                 if (port < 0x8000)
7762                         bitmap = vmcs12->io_bitmap_a;
7763                 else if (port < 0x10000)
7764                         bitmap = vmcs12->io_bitmap_b;
7765                 else
7766                         return true;
7767                 bitmap += (port & 0x7fff) / 8;
7768
7769                 if (last_bitmap != bitmap)
7770                         if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7771                                 return true;
7772                 if (b & (1 << (port & 7)))
7773                         return true;
7774
7775                 port++;
7776                 size--;
7777                 last_bitmap = bitmap;
7778         }
7779
7780         return false;
7781 }
7782
7783 /*
7784  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7785  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7786  * disinterest in the current event (read or write a specific MSR) by using an
7787  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7788  */
7789 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7790         struct vmcs12 *vmcs12, u32 exit_reason)
7791 {
7792         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7793         gpa_t bitmap;
7794
7795         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7796                 return true;
7797
7798         /*
7799          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7800          * for the four combinations of read/write and low/high MSR numbers.
7801          * First we need to figure out which of the four to use:
7802          */
7803         bitmap = vmcs12->msr_bitmap;
7804         if (exit_reason == EXIT_REASON_MSR_WRITE)
7805                 bitmap += 2048;
7806         if (msr_index >= 0xc0000000) {
7807                 msr_index -= 0xc0000000;
7808                 bitmap += 1024;
7809         }
7810
7811         /* Then read the msr_index'th bit from this bitmap: */
7812         if (msr_index < 1024*8) {
7813                 unsigned char b;
7814                 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7815                         return true;
7816                 return 1 & (b >> (msr_index & 7));
7817         } else
7818                 return true; /* let L1 handle the wrong parameter */
7819 }
7820
7821 /*
7822  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7823  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7824  * intercept (via guest_host_mask etc.) the current event.
7825  */
7826 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7827         struct vmcs12 *vmcs12)
7828 {
7829         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7830         int cr = exit_qualification & 15;
7831         int reg = (exit_qualification >> 8) & 15;
7832         unsigned long val = kvm_register_readl(vcpu, reg);
7833
7834         switch ((exit_qualification >> 4) & 3) {
7835         case 0: /* mov to cr */
7836                 switch (cr) {
7837                 case 0:
7838                         if (vmcs12->cr0_guest_host_mask &
7839                             (val ^ vmcs12->cr0_read_shadow))
7840                                 return true;
7841                         break;
7842                 case 3:
7843                         if ((vmcs12->cr3_target_count >= 1 &&
7844                                         vmcs12->cr3_target_value0 == val) ||
7845                                 (vmcs12->cr3_target_count >= 2 &&
7846                                         vmcs12->cr3_target_value1 == val) ||
7847                                 (vmcs12->cr3_target_count >= 3 &&
7848                                         vmcs12->cr3_target_value2 == val) ||
7849                                 (vmcs12->cr3_target_count >= 4 &&
7850                                         vmcs12->cr3_target_value3 == val))
7851                                 return false;
7852                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7853                                 return true;
7854                         break;
7855                 case 4:
7856                         if (vmcs12->cr4_guest_host_mask &
7857                             (vmcs12->cr4_read_shadow ^ val))
7858                                 return true;
7859                         break;
7860                 case 8:
7861                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7862                                 return true;
7863                         break;
7864                 }
7865                 break;
7866         case 2: /* clts */
7867                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7868                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
7869                         return true;
7870                 break;
7871         case 1: /* mov from cr */
7872                 switch (cr) {
7873                 case 3:
7874                         if (vmcs12->cpu_based_vm_exec_control &
7875                             CPU_BASED_CR3_STORE_EXITING)
7876                                 return true;
7877                         break;
7878                 case 8:
7879                         if (vmcs12->cpu_based_vm_exec_control &
7880                             CPU_BASED_CR8_STORE_EXITING)
7881                                 return true;
7882                         break;
7883                 }
7884                 break;
7885         case 3: /* lmsw */
7886                 /*
7887                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7888                  * cr0. Other attempted changes are ignored, with no exit.
7889                  */
7890                 if (vmcs12->cr0_guest_host_mask & 0xe &
7891                     (val ^ vmcs12->cr0_read_shadow))
7892                         return true;
7893                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7894                     !(vmcs12->cr0_read_shadow & 0x1) &&
7895                     (val & 0x1))
7896                         return true;
7897                 break;
7898         }
7899         return false;
7900 }
7901
7902 /*
7903  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7904  * should handle it ourselves in L0 (and then continue L2). Only call this
7905  * when in is_guest_mode (L2).
7906  */
7907 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7908 {
7909         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7910         struct vcpu_vmx *vmx = to_vmx(vcpu);
7911         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7912         u32 exit_reason = vmx->exit_reason;
7913
7914         trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7915                                 vmcs_readl(EXIT_QUALIFICATION),
7916                                 vmx->idt_vectoring_info,
7917                                 intr_info,
7918                                 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7919                                 KVM_ISA_VMX);
7920
7921         if (vmx->nested.nested_run_pending)
7922                 return false;
7923
7924         if (unlikely(vmx->fail)) {
7925                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7926                                     vmcs_read32(VM_INSTRUCTION_ERROR));
7927                 return true;
7928         }
7929
7930         switch (exit_reason) {
7931         case EXIT_REASON_EXCEPTION_NMI:
7932                 if (!is_exception(intr_info))
7933                         return false;
7934                 else if (is_page_fault(intr_info))
7935                         return enable_ept;
7936                 else if (is_no_device(intr_info) &&
7937                          !(vmcs12->guest_cr0 & X86_CR0_TS))
7938                         return false;
7939                 else if (is_debug(intr_info) &&
7940                          vcpu->guest_debug &
7941                          (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
7942                         return false;
7943                 else if (is_breakpoint(intr_info) &&
7944                          vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
7945                         return false;
7946                 return vmcs12->exception_bitmap &
7947                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7948         case EXIT_REASON_EXTERNAL_INTERRUPT:
7949                 return false;
7950         case EXIT_REASON_TRIPLE_FAULT:
7951                 return true;
7952         case EXIT_REASON_PENDING_INTERRUPT:
7953                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7954         case EXIT_REASON_NMI_WINDOW:
7955                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7956         case EXIT_REASON_TASK_SWITCH:
7957                 return true;
7958         case EXIT_REASON_CPUID:
7959                 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7960                         return false;
7961                 return true;
7962         case EXIT_REASON_HLT:
7963                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7964         case EXIT_REASON_INVD:
7965                 return true;
7966         case EXIT_REASON_INVLPG:
7967                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7968         case EXIT_REASON_RDPMC:
7969                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7970         case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7971                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7972         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7973         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7974         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7975         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7976         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7977         case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7978                 /*
7979                  * VMX instructions trap unconditionally. This allows L1 to
7980                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
7981                  */
7982                 return true;
7983         case EXIT_REASON_CR_ACCESS:
7984                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7985         case EXIT_REASON_DR_ACCESS:
7986                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7987         case EXIT_REASON_IO_INSTRUCTION:
7988                 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7989         case EXIT_REASON_MSR_READ:
7990         case EXIT_REASON_MSR_WRITE:
7991                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7992         case EXIT_REASON_INVALID_STATE:
7993                 return true;
7994         case EXIT_REASON_MWAIT_INSTRUCTION:
7995                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7996         case EXIT_REASON_MONITOR_TRAP_FLAG:
7997                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
7998         case EXIT_REASON_MONITOR_INSTRUCTION:
7999                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8000         case EXIT_REASON_PAUSE_INSTRUCTION:
8001                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8002                         nested_cpu_has2(vmcs12,
8003                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8004         case EXIT_REASON_MCE_DURING_VMENTRY:
8005                 return false;
8006         case EXIT_REASON_TPR_BELOW_THRESHOLD:
8007                 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8008         case EXIT_REASON_APIC_ACCESS:
8009                 return nested_cpu_has2(vmcs12,
8010                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8011         case EXIT_REASON_APIC_WRITE:
8012         case EXIT_REASON_EOI_INDUCED:
8013                 /* apic_write and eoi_induced should exit unconditionally. */
8014                 return true;
8015         case EXIT_REASON_EPT_VIOLATION:
8016                 /*
8017                  * L0 always deals with the EPT violation. If nested EPT is
8018                  * used, and the nested mmu code discovers that the address is
8019                  * missing in the guest EPT table (EPT12), the EPT violation
8020                  * will be injected with nested_ept_inject_page_fault()
8021                  */
8022                 return false;
8023         case EXIT_REASON_EPT_MISCONFIG:
8024                 /*
8025                  * L2 never uses directly L1's EPT, but rather L0's own EPT
8026                  * table (shadow on EPT) or a merged EPT table that L0 built
8027                  * (EPT on EPT). So any problems with the structure of the
8028                  * table is L0's fault.
8029                  */
8030                 return false;
8031         case EXIT_REASON_WBINVD:
8032                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8033         case EXIT_REASON_XSETBV:
8034                 return true;
8035         case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8036                 /*
8037                  * This should never happen, since it is not possible to
8038                  * set XSS to a non-zero value---neither in L1 nor in L2.
8039                  * If if it were, XSS would have to be checked against
8040                  * the XSS exit bitmap in vmcs12.
8041                  */
8042                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8043         case EXIT_REASON_PCOMMIT:
8044                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
8045         default:
8046                 return true;
8047         }
8048 }
8049
8050 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8051 {
8052         *info1 = vmcs_readl(EXIT_QUALIFICATION);
8053         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8054 }
8055
8056 static int vmx_create_pml_buffer(struct vcpu_vmx *vmx)
8057 {
8058         struct page *pml_pg;
8059
8060         pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
8061         if (!pml_pg)
8062                 return -ENOMEM;
8063
8064         vmx->pml_pg = pml_pg;
8065
8066         vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
8067         vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8068
8069         return 0;
8070 }
8071
8072 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8073 {
8074         if (vmx->pml_pg) {
8075                 __free_page(vmx->pml_pg);
8076                 vmx->pml_pg = NULL;
8077         }
8078 }
8079
8080 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8081 {
8082         struct vcpu_vmx *vmx = to_vmx(vcpu);
8083         u64 *pml_buf;
8084         u16 pml_idx;
8085
8086         pml_idx = vmcs_read16(GUEST_PML_INDEX);
8087
8088         /* Do nothing if PML buffer is empty */
8089         if (pml_idx == (PML_ENTITY_NUM - 1))
8090                 return;
8091
8092         /* PML index always points to next available PML buffer entity */
8093         if (pml_idx >= PML_ENTITY_NUM)
8094                 pml_idx = 0;
8095         else
8096                 pml_idx++;
8097
8098         pml_buf = page_address(vmx->pml_pg);
8099         for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8100                 u64 gpa;
8101
8102                 gpa = pml_buf[pml_idx];
8103                 WARN_ON(gpa & (PAGE_SIZE - 1));
8104                 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8105         }
8106
8107         /* reset PML index */
8108         vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8109 }
8110
8111 /*
8112  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8113  * Called before reporting dirty_bitmap to userspace.
8114  */
8115 static void kvm_flush_pml_buffers(struct kvm *kvm)
8116 {
8117         int i;
8118         struct kvm_vcpu *vcpu;
8119         /*
8120          * We only need to kick vcpu out of guest mode here, as PML buffer
8121          * is flushed at beginning of all VMEXITs, and it's obvious that only
8122          * vcpus running in guest are possible to have unflushed GPAs in PML
8123          * buffer.
8124          */
8125         kvm_for_each_vcpu(i, vcpu, kvm)
8126                 kvm_vcpu_kick(vcpu);
8127 }
8128
8129 static void vmx_dump_sel(char *name, uint32_t sel)
8130 {
8131         pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8132                name, vmcs_read32(sel),
8133                vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8134                vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8135                vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8136 }
8137
8138 static void vmx_dump_dtsel(char *name, uint32_t limit)
8139 {
8140         pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
8141                name, vmcs_read32(limit),
8142                vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8143 }
8144
8145 static void dump_vmcs(void)
8146 {
8147         u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8148         u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8149         u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8150         u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8151         u32 secondary_exec_control = 0;
8152         unsigned long cr4 = vmcs_readl(GUEST_CR4);
8153         u64 efer = vmcs_read64(GUEST_IA32_EFER);
8154         int i, n;
8155
8156         if (cpu_has_secondary_exec_ctrls())
8157                 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8158
8159         pr_err("*** Guest State ***\n");
8160         pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8161                vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8162                vmcs_readl(CR0_GUEST_HOST_MASK));
8163         pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8164                cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8165         pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8166         if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8167             (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8168         {
8169                 pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
8170                        vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8171                 pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
8172                        vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8173         }
8174         pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
8175                vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8176         pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
8177                vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8178         pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8179                vmcs_readl(GUEST_SYSENTER_ESP),
8180                vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8181         vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
8182         vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
8183         vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
8184         vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
8185         vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
8186         vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
8187         vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8188         vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8189         vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8190         vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
8191         if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8192             (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8193                 pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
8194                        efer, vmcs_read64(GUEST_IA32_PAT));
8195         pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
8196                vmcs_read64(GUEST_IA32_DEBUGCTL),
8197                vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8198         if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8199                 pr_err("PerfGlobCtl = 0x%016llx\n",
8200                        vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8201         if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8202                 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8203         pr_err("Interruptibility = %08x  ActivityState = %08x\n",
8204                vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8205                vmcs_read32(GUEST_ACTIVITY_STATE));
8206         if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8207                 pr_err("InterruptStatus = %04x\n",
8208                        vmcs_read16(GUEST_INTR_STATUS));
8209
8210         pr_err("*** Host State ***\n");
8211         pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
8212                vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8213         pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8214                vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8215                vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8216                vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8217                vmcs_read16(HOST_TR_SELECTOR));
8218         pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8219                vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8220                vmcs_readl(HOST_TR_BASE));
8221         pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8222                vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8223         pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8224                vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8225                vmcs_readl(HOST_CR4));
8226         pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8227                vmcs_readl(HOST_IA32_SYSENTER_ESP),
8228                vmcs_read32(HOST_IA32_SYSENTER_CS),
8229                vmcs_readl(HOST_IA32_SYSENTER_EIP));
8230         if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8231                 pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
8232                        vmcs_read64(HOST_IA32_EFER),
8233                        vmcs_read64(HOST_IA32_PAT));
8234         if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8235                 pr_err("PerfGlobCtl = 0x%016llx\n",
8236                        vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8237
8238         pr_err("*** Control State ***\n");
8239         pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8240                pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8241         pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8242         pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8243                vmcs_read32(EXCEPTION_BITMAP),
8244                vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8245                vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8246         pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8247                vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8248                vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8249                vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8250         pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8251                vmcs_read32(VM_EXIT_INTR_INFO),
8252                vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8253                vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8254         pr_err("        reason=%08x qualification=%016lx\n",
8255                vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8256         pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8257                vmcs_read32(IDT_VECTORING_INFO_FIELD),
8258                vmcs_read32(IDT_VECTORING_ERROR_CODE));
8259         pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8260         if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8261                 pr_err("TSC Multiplier = 0x%016llx\n",
8262                        vmcs_read64(TSC_MULTIPLIER));
8263         if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8264                 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8265         if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8266                 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8267         if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8268                 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8269         n = vmcs_read32(CR3_TARGET_COUNT);
8270         for (i = 0; i + 1 < n; i += 4)
8271                 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8272                        i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8273                        i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8274         if (i < n)
8275                 pr_err("CR3 target%u=%016lx\n",
8276                        i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8277         if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8278                 pr_err("PLE Gap=%08x Window=%08x\n",
8279                        vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8280         if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8281                 pr_err("Virtual processor ID = 0x%04x\n",
8282                        vmcs_read16(VIRTUAL_PROCESSOR_ID));
8283 }
8284
8285 /*
8286  * The guest has exited.  See if we can fix it or if we need userspace
8287  * assistance.
8288  */
8289 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8290 {
8291         struct vcpu_vmx *vmx = to_vmx(vcpu);
8292         u32 exit_reason = vmx->exit_reason;
8293         u32 vectoring_info = vmx->idt_vectoring_info;
8294
8295         trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8296
8297         /*
8298          * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8299          * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8300          * querying dirty_bitmap, we only need to kick all vcpus out of guest
8301          * mode as if vcpus is in root mode, the PML buffer must has been
8302          * flushed already.
8303          */
8304         if (enable_pml)
8305                 vmx_flush_pml_buffer(vcpu);
8306
8307         /* If guest state is invalid, start emulating */
8308         if (vmx->emulation_required)
8309                 return handle_invalid_guest_state(vcpu);
8310
8311         if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8312                 nested_vmx_vmexit(vcpu, exit_reason,
8313                                   vmcs_read32(VM_EXIT_INTR_INFO),
8314                                   vmcs_readl(EXIT_QUALIFICATION));
8315                 return 1;
8316         }
8317
8318         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8319                 dump_vmcs();
8320                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8321                 vcpu->run->fail_entry.hardware_entry_failure_reason
8322                         = exit_reason;
8323                 return 0;
8324         }
8325
8326         if (unlikely(vmx->fail)) {
8327                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8328                 vcpu->run->fail_entry.hardware_entry_failure_reason
8329                         = vmcs_read32(VM_INSTRUCTION_ERROR);
8330                 return 0;
8331         }
8332
8333         /*
8334          * Note:
8335          * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8336          * delivery event since it indicates guest is accessing MMIO.
8337          * The vm-exit can be triggered again after return to guest that
8338          * will cause infinite loop.
8339          */
8340         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8341                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8342                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
8343                         exit_reason != EXIT_REASON_TASK_SWITCH)) {
8344                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8345                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8346                 vcpu->run->internal.ndata = 2;
8347                 vcpu->run->internal.data[0] = vectoring_info;
8348                 vcpu->run->internal.data[1] = exit_reason;
8349                 return 0;
8350         }
8351
8352         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8353             !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8354                                         get_vmcs12(vcpu))))) {
8355                 if (vmx_interrupt_allowed(vcpu)) {
8356                         vmx->soft_vnmi_blocked = 0;
8357                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8358                            vcpu->arch.nmi_pending) {
8359                         /*
8360                          * This CPU don't support us in finding the end of an
8361                          * NMI-blocked window if the guest runs with IRQs
8362                          * disabled. So we pull the trigger after 1 s of
8363                          * futile waiting, but inform the user about this.
8364                          */
8365                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8366                                "state on VCPU %d after 1 s timeout\n",
8367                                __func__, vcpu->vcpu_id);
8368                         vmx->soft_vnmi_blocked = 0;
8369                 }
8370         }
8371
8372         if (exit_reason < kvm_vmx_max_exit_handlers
8373             && kvm_vmx_exit_handlers[exit_reason])
8374                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8375         else {
8376                 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8377                 kvm_queue_exception(vcpu, UD_VECTOR);
8378                 return 1;
8379         }
8380 }
8381
8382 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8383 {
8384         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8385
8386         if (is_guest_mode(vcpu) &&
8387                 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8388                 return;
8389
8390         if (irr == -1 || tpr < irr) {
8391                 vmcs_write32(TPR_THRESHOLD, 0);
8392                 return;
8393         }
8394
8395         vmcs_write32(TPR_THRESHOLD, irr);
8396 }
8397
8398 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8399 {
8400         u32 sec_exec_control;
8401
8402         /*
8403          * There is not point to enable virtualize x2apic without enable
8404          * apicv
8405          */
8406         if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8407                                 !kvm_vcpu_apicv_active(vcpu))
8408                 return;
8409
8410         if (!cpu_need_tpr_shadow(vcpu))
8411                 return;
8412
8413         sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8414
8415         if (set) {
8416                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8417                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8418         } else {
8419                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8420                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8421         }
8422         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8423
8424         vmx_set_msr_bitmap(vcpu);
8425 }
8426
8427 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8428 {
8429         struct vcpu_vmx *vmx = to_vmx(vcpu);
8430
8431         /*
8432          * Currently we do not handle the nested case where L2 has an
8433          * APIC access page of its own; that page is still pinned.
8434          * Hence, we skip the case where the VCPU is in guest mode _and_
8435          * L1 prepared an APIC access page for L2.
8436          *
8437          * For the case where L1 and L2 share the same APIC access page
8438          * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8439          * in the vmcs12), this function will only update either the vmcs01
8440          * or the vmcs02.  If the former, the vmcs02 will be updated by
8441          * prepare_vmcs02.  If the latter, the vmcs01 will be updated in
8442          * the next L2->L1 exit.
8443          */
8444         if (!is_guest_mode(vcpu) ||
8445             !nested_cpu_has2(vmx->nested.current_vmcs12,
8446                              SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8447                 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8448 }
8449
8450 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8451 {
8452         u16 status;
8453         u8 old;
8454
8455         if (max_isr == -1)
8456                 max_isr = 0;
8457
8458         status = vmcs_read16(GUEST_INTR_STATUS);
8459         old = status >> 8;
8460         if (max_isr != old) {
8461                 status &= 0xff;
8462                 status |= max_isr << 8;
8463                 vmcs_write16(GUEST_INTR_STATUS, status);
8464         }
8465 }
8466
8467 static void vmx_set_rvi(int vector)
8468 {
8469         u16 status;
8470         u8 old;
8471
8472         if (vector == -1)
8473                 vector = 0;
8474
8475         status = vmcs_read16(GUEST_INTR_STATUS);
8476         old = (u8)status & 0xff;
8477         if ((u8)vector != old) {
8478                 status &= ~0xff;
8479                 status |= (u8)vector;
8480                 vmcs_write16(GUEST_INTR_STATUS, status);
8481         }
8482 }
8483
8484 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8485 {
8486         if (!is_guest_mode(vcpu)) {
8487                 vmx_set_rvi(max_irr);
8488                 return;
8489         }
8490
8491         if (max_irr == -1)
8492                 return;
8493
8494         /*
8495          * In guest mode.  If a vmexit is needed, vmx_check_nested_events
8496          * handles it.
8497          */
8498         if (nested_exit_on_intr(vcpu))
8499                 return;
8500
8501         /*
8502          * Else, fall back to pre-APICv interrupt injection since L2
8503          * is run without virtual interrupt delivery.
8504          */
8505         if (!kvm_event_needs_reinjection(vcpu) &&
8506             vmx_interrupt_allowed(vcpu)) {
8507                 kvm_queue_interrupt(vcpu, max_irr, false);
8508                 vmx_inject_irq(vcpu);
8509         }
8510 }
8511
8512 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8513 {
8514         if (!kvm_vcpu_apicv_active(vcpu))
8515                 return;
8516
8517         vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8518         vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8519         vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8520         vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8521 }
8522
8523 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8524 {
8525         u32 exit_intr_info;
8526
8527         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8528               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8529                 return;
8530
8531         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8532         exit_intr_info = vmx->exit_intr_info;
8533
8534         /* Handle machine checks before interrupts are enabled */
8535         if (is_machine_check(exit_intr_info))
8536                 kvm_machine_check();
8537
8538         /* We need to handle NMIs before interrupts are enabled */
8539         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8540             (exit_intr_info & INTR_INFO_VALID_MASK)) {
8541                 kvm_before_handle_nmi(&vmx->vcpu);
8542                 asm("int $2");
8543                 kvm_after_handle_nmi(&vmx->vcpu);
8544         }
8545 }
8546
8547 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8548 {
8549         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8550         register void *__sp asm(_ASM_SP);
8551
8552         /*
8553          * If external interrupt exists, IF bit is set in rflags/eflags on the
8554          * interrupt stack frame, and interrupt will be enabled on a return
8555          * from interrupt handler.
8556          */
8557         if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8558                         == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8559                 unsigned int vector;
8560                 unsigned long entry;
8561                 gate_desc *desc;
8562                 struct vcpu_vmx *vmx = to_vmx(vcpu);
8563 #ifdef CONFIG_X86_64
8564                 unsigned long tmp;
8565 #endif
8566
8567                 vector =  exit_intr_info & INTR_INFO_VECTOR_MASK;
8568                 desc = (gate_desc *)vmx->host_idt_base + vector;
8569                 entry = gate_offset(*desc);
8570                 asm volatile(
8571 #ifdef CONFIG_X86_64
8572                         "mov %%" _ASM_SP ", %[sp]\n\t"
8573                         "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8574                         "push $%c[ss]\n\t"
8575                         "push %[sp]\n\t"
8576 #endif
8577                         "pushf\n\t"
8578                         "orl $0x200, (%%" _ASM_SP ")\n\t"
8579                         __ASM_SIZE(push) " $%c[cs]\n\t"
8580                         "call *%[entry]\n\t"
8581                         :
8582 #ifdef CONFIG_X86_64
8583                         [sp]"=&r"(tmp),
8584 #endif
8585                         "+r"(__sp)
8586                         :
8587                         [entry]"r"(entry),
8588                         [ss]"i"(__KERNEL_DS),
8589                         [cs]"i"(__KERNEL_CS)
8590                         );
8591         } else
8592                 local_irq_enable();
8593 }
8594
8595 static bool vmx_has_high_real_mode_segbase(void)
8596 {
8597         return enable_unrestricted_guest || emulate_invalid_guest_state;
8598 }
8599
8600 static bool vmx_mpx_supported(void)
8601 {
8602         return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8603                 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8604 }
8605
8606 static bool vmx_xsaves_supported(void)
8607 {
8608         return vmcs_config.cpu_based_2nd_exec_ctrl &
8609                 SECONDARY_EXEC_XSAVES;
8610 }
8611
8612 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8613 {
8614         u32 exit_intr_info;
8615         bool unblock_nmi;
8616         u8 vector;
8617         bool idtv_info_valid;
8618
8619         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8620
8621         if (cpu_has_virtual_nmis()) {
8622                 if (vmx->nmi_known_unmasked)
8623                         return;
8624                 /*
8625                  * Can't use vmx->exit_intr_info since we're not sure what
8626                  * the exit reason is.
8627                  */
8628                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8629                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8630                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8631                 /*
8632                  * SDM 3: 27.7.1.2 (September 2008)
8633                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
8634                  * a guest IRET fault.
8635                  * SDM 3: 23.2.2 (September 2008)
8636                  * Bit 12 is undefined in any of the following cases:
8637                  *  If the VM exit sets the valid bit in the IDT-vectoring
8638                  *   information field.
8639                  *  If the VM exit is due to a double fault.
8640                  */
8641                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8642                     vector != DF_VECTOR && !idtv_info_valid)
8643                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8644                                       GUEST_INTR_STATE_NMI);
8645                 else
8646                         vmx->nmi_known_unmasked =
8647                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8648                                   & GUEST_INTR_STATE_NMI);
8649         } else if (unlikely(vmx->soft_vnmi_blocked))
8650                 vmx->vnmi_blocked_time +=
8651                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8652 }
8653
8654 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8655                                       u32 idt_vectoring_info,
8656                                       int instr_len_field,
8657                                       int error_code_field)
8658 {
8659         u8 vector;
8660         int type;
8661         bool idtv_info_valid;
8662
8663         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8664
8665         vcpu->arch.nmi_injected = false;
8666         kvm_clear_exception_queue(vcpu);
8667         kvm_clear_interrupt_queue(vcpu);
8668
8669         if (!idtv_info_valid)
8670                 return;
8671
8672         kvm_make_request(KVM_REQ_EVENT, vcpu);
8673
8674         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8675         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8676
8677         switch (type) {
8678         case INTR_TYPE_NMI_INTR:
8679                 vcpu->arch.nmi_injected = true;
8680                 /*
8681                  * SDM 3: 27.7.1.2 (September 2008)
8682                  * Clear bit "block by NMI" before VM entry if a NMI
8683                  * delivery faulted.
8684                  */
8685                 vmx_set_nmi_mask(vcpu, false);
8686                 break;
8687         case INTR_TYPE_SOFT_EXCEPTION:
8688                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8689                 /* fall through */
8690         case INTR_TYPE_HARD_EXCEPTION:
8691                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8692                         u32 err = vmcs_read32(error_code_field);
8693                         kvm_requeue_exception_e(vcpu, vector, err);
8694                 } else
8695                         kvm_requeue_exception(vcpu, vector);
8696                 break;
8697         case INTR_TYPE_SOFT_INTR:
8698                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8699                 /* fall through */
8700         case INTR_TYPE_EXT_INTR:
8701                 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8702                 break;
8703         default:
8704                 break;
8705         }
8706 }
8707
8708 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8709 {
8710         __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8711                                   VM_EXIT_INSTRUCTION_LEN,
8712                                   IDT_VECTORING_ERROR_CODE);
8713 }
8714
8715 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8716 {
8717         __vmx_complete_interrupts(vcpu,
8718                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8719                                   VM_ENTRY_INSTRUCTION_LEN,
8720                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
8721
8722         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8723 }
8724
8725 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8726 {
8727         int i, nr_msrs;
8728         struct perf_guest_switch_msr *msrs;
8729
8730         msrs = perf_guest_get_msrs(&nr_msrs);
8731
8732         if (!msrs)
8733                 return;
8734
8735         for (i = 0; i < nr_msrs; i++)
8736                 if (msrs[i].host == msrs[i].guest)
8737                         clear_atomic_switch_msr(vmx, msrs[i].msr);
8738                 else
8739                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8740                                         msrs[i].host);
8741 }
8742
8743 void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8744 {
8745         struct vcpu_vmx *vmx = to_vmx(vcpu);
8746         u64 tscl;
8747         u32 delta_tsc;
8748
8749         if (vmx->hv_deadline_tsc == -1)
8750                 return;
8751
8752         tscl = rdtsc();
8753         if (vmx->hv_deadline_tsc > tscl)
8754                 /* sure to be 32 bit only because checked on set_hv_timer */
8755                 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8756                         cpu_preemption_timer_multi);
8757         else
8758                 delta_tsc = 0;
8759
8760         vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8761 }
8762
8763 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8764 {
8765         struct vcpu_vmx *vmx = to_vmx(vcpu);
8766         unsigned long debugctlmsr, cr4;
8767
8768         /* Record the guest's net vcpu time for enforced NMI injections. */
8769         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8770                 vmx->entry_time = ktime_get();
8771
8772         /* Don't enter VMX if guest state is invalid, let the exit handler
8773            start emulation until we arrive back to a valid state */
8774         if (vmx->emulation_required)
8775                 return;
8776
8777         if (vmx->ple_window_dirty) {
8778                 vmx->ple_window_dirty = false;
8779                 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8780         }
8781
8782         if (vmx->nested.sync_shadow_vmcs) {
8783                 copy_vmcs12_to_shadow(vmx);
8784                 vmx->nested.sync_shadow_vmcs = false;
8785         }
8786
8787         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8788                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8789         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8790                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8791
8792         cr4 = cr4_read_shadow();
8793         if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8794                 vmcs_writel(HOST_CR4, cr4);
8795                 vmx->host_state.vmcs_host_cr4 = cr4;
8796         }
8797
8798         /* When single-stepping over STI and MOV SS, we must clear the
8799          * corresponding interruptibility bits in the guest state. Otherwise
8800          * vmentry fails as it then expects bit 14 (BS) in pending debug
8801          * exceptions being set, but that's not correct for the guest debugging
8802          * case. */
8803         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8804                 vmx_set_interrupt_shadow(vcpu, 0);
8805
8806         if (vmx->guest_pkru_valid)
8807                 __write_pkru(vmx->guest_pkru);
8808
8809         atomic_switch_perf_msrs(vmx);
8810         debugctlmsr = get_debugctlmsr();
8811
8812         vmx_arm_hv_timer(vcpu);
8813
8814         vmx->__launched = vmx->loaded_vmcs->launched;
8815         asm(
8816                 /* Store host registers */
8817                 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8818                 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8819                 "push %%" _ASM_CX " \n\t"
8820                 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8821                 "je 1f \n\t"
8822                 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8823                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8824                 "1: \n\t"
8825                 /* Reload cr2 if changed */
8826                 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8827                 "mov %%cr2, %%" _ASM_DX " \n\t"
8828                 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8829                 "je 2f \n\t"
8830                 "mov %%" _ASM_AX", %%cr2 \n\t"
8831                 "2: \n\t"
8832                 /* Check if vmlaunch of vmresume is needed */
8833                 "cmpl $0, %c[launched](%0) \n\t"
8834                 /* Load guest registers.  Don't clobber flags. */
8835                 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8836                 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8837                 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8838                 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8839                 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8840                 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8841 #ifdef CONFIG_X86_64
8842                 "mov %c[r8](%0),  %%r8  \n\t"
8843                 "mov %c[r9](%0),  %%r9  \n\t"
8844                 "mov %c[r10](%0), %%r10 \n\t"
8845                 "mov %c[r11](%0), %%r11 \n\t"
8846                 "mov %c[r12](%0), %%r12 \n\t"
8847                 "mov %c[r13](%0), %%r13 \n\t"
8848                 "mov %c[r14](%0), %%r14 \n\t"
8849                 "mov %c[r15](%0), %%r15 \n\t"
8850 #endif
8851                 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8852
8853                 /* Enter guest mode */
8854                 "jne 1f \n\t"
8855                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8856                 "jmp 2f \n\t"
8857                 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8858                 "2: "
8859                 /* Save guest registers, load host registers, keep flags */
8860                 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8861                 "pop %0 \n\t"
8862                 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8863                 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8864                 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8865                 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8866                 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8867                 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8868                 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8869 #ifdef CONFIG_X86_64
8870                 "mov %%r8,  %c[r8](%0) \n\t"
8871                 "mov %%r9,  %c[r9](%0) \n\t"
8872                 "mov %%r10, %c[r10](%0) \n\t"
8873                 "mov %%r11, %c[r11](%0) \n\t"
8874                 "mov %%r12, %c[r12](%0) \n\t"
8875                 "mov %%r13, %c[r13](%0) \n\t"
8876                 "mov %%r14, %c[r14](%0) \n\t"
8877                 "mov %%r15, %c[r15](%0) \n\t"
8878 #endif
8879                 "mov %%cr2, %%" _ASM_AX "   \n\t"
8880                 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8881
8882                 "pop  %%" _ASM_BP "; pop  %%" _ASM_DX " \n\t"
8883                 "setbe %c[fail](%0) \n\t"
8884                 ".pushsection .rodata \n\t"
8885                 ".global vmx_return \n\t"
8886                 "vmx_return: " _ASM_PTR " 2b \n\t"
8887                 ".popsection"
8888               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8889                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8890                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8891                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8892                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8893                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8894                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8895                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8896                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8897                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8898                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8899 #ifdef CONFIG_X86_64
8900                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8901                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8902                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8903                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8904                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8905                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8906                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8907                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8908 #endif
8909                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8910                 [wordsize]"i"(sizeof(ulong))
8911               : "cc", "memory"
8912 #ifdef CONFIG_X86_64
8913                 , "rax", "rbx", "rdi", "rsi"
8914                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8915 #else
8916                 , "eax", "ebx", "edi", "esi"
8917 #endif
8918               );
8919
8920         /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8921         if (debugctlmsr)
8922                 update_debugctlmsr(debugctlmsr);
8923
8924 #ifndef CONFIG_X86_64
8925         /*
8926          * The sysexit path does not restore ds/es, so we must set them to
8927          * a reasonable value ourselves.
8928          *
8929          * We can't defer this to vmx_load_host_state() since that function
8930          * may be executed in interrupt context, which saves and restore segments
8931          * around it, nullifying its effect.
8932          */
8933         loadsegment(ds, __USER_DS);
8934         loadsegment(es, __USER_DS);
8935 #endif
8936
8937         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8938                                   | (1 << VCPU_EXREG_RFLAGS)
8939                                   | (1 << VCPU_EXREG_PDPTR)
8940                                   | (1 << VCPU_EXREG_SEGMENTS)
8941                                   | (1 << VCPU_EXREG_CR3));
8942         vcpu->arch.regs_dirty = 0;
8943
8944         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8945
8946         vmx->loaded_vmcs->launched = 1;
8947
8948         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8949
8950         /*
8951          * eager fpu is enabled if PKEY is supported and CR4 is switched
8952          * back on host, so it is safe to read guest PKRU from current
8953          * XSAVE.
8954          */
8955         if (boot_cpu_has(X86_FEATURE_OSPKE)) {
8956                 vmx->guest_pkru = __read_pkru();
8957                 if (vmx->guest_pkru != vmx->host_pkru) {
8958                         vmx->guest_pkru_valid = true;
8959                         __write_pkru(vmx->host_pkru);
8960                 } else
8961                         vmx->guest_pkru_valid = false;
8962         }
8963
8964         /*
8965          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8966          * we did not inject a still-pending event to L1 now because of
8967          * nested_run_pending, we need to re-enable this bit.
8968          */
8969         if (vmx->nested.nested_run_pending)
8970                 kvm_make_request(KVM_REQ_EVENT, vcpu);
8971
8972         vmx->nested.nested_run_pending = 0;
8973
8974         vmx_complete_atomic_exit(vmx);
8975         vmx_recover_nmi_blocking(vmx);
8976         vmx_complete_interrupts(vmx);
8977 }
8978
8979 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8980 {
8981         struct vcpu_vmx *vmx = to_vmx(vcpu);
8982         int cpu;
8983
8984         if (vmx->loaded_vmcs == &vmx->vmcs01)
8985                 return;
8986
8987         cpu = get_cpu();
8988         vmx->loaded_vmcs = &vmx->vmcs01;
8989         vmx_vcpu_put(vcpu);
8990         vmx_vcpu_load(vcpu, cpu);
8991         vcpu->cpu = cpu;
8992         put_cpu();
8993 }
8994
8995 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8996 {
8997         struct vcpu_vmx *vmx = to_vmx(vcpu);
8998
8999         if (enable_pml)
9000                 vmx_destroy_pml_buffer(vmx);
9001         free_vpid(vmx->vpid);
9002         leave_guest_mode(vcpu);
9003         vmx_load_vmcs01(vcpu);
9004         free_nested(vmx);
9005         free_loaded_vmcs(vmx->loaded_vmcs);
9006         kfree(vmx->guest_msrs);
9007         kvm_vcpu_uninit(vcpu);
9008         kmem_cache_free(kvm_vcpu_cache, vmx);
9009 }
9010
9011 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9012 {
9013         int err;
9014         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9015         int cpu;
9016
9017         if (!vmx)
9018                 return ERR_PTR(-ENOMEM);
9019
9020         vmx->vpid = allocate_vpid();
9021
9022         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9023         if (err)
9024                 goto free_vcpu;
9025
9026         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9027         BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9028                      > PAGE_SIZE);
9029
9030         err = -ENOMEM;
9031         if (!vmx->guest_msrs) {
9032                 goto uninit_vcpu;
9033         }
9034
9035         vmx->loaded_vmcs = &vmx->vmcs01;
9036         vmx->loaded_vmcs->vmcs = alloc_vmcs();
9037         if (!vmx->loaded_vmcs->vmcs)
9038                 goto free_msrs;
9039         if (!vmm_exclusive)
9040                 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
9041         loaded_vmcs_init(vmx->loaded_vmcs);
9042         if (!vmm_exclusive)
9043                 kvm_cpu_vmxoff();
9044
9045         cpu = get_cpu();
9046         vmx_vcpu_load(&vmx->vcpu, cpu);
9047         vmx->vcpu.cpu = cpu;
9048         err = vmx_vcpu_setup(vmx);
9049         vmx_vcpu_put(&vmx->vcpu);
9050         put_cpu();
9051         if (err)
9052                 goto free_vmcs;
9053         if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9054                 err = alloc_apic_access_page(kvm);
9055                 if (err)
9056                         goto free_vmcs;
9057         }
9058
9059         if (enable_ept) {
9060                 if (!kvm->arch.ept_identity_map_addr)
9061                         kvm->arch.ept_identity_map_addr =
9062                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9063                 err = init_rmode_identity_map(kvm);
9064                 if (err)
9065                         goto free_vmcs;
9066         }
9067
9068         if (nested) {
9069                 nested_vmx_setup_ctls_msrs(vmx);
9070                 vmx->nested.vpid02 = allocate_vpid();
9071         }
9072
9073         vmx->nested.posted_intr_nv = -1;
9074         vmx->nested.current_vmptr = -1ull;
9075         vmx->nested.current_vmcs12 = NULL;
9076
9077         /*
9078          * If PML is turned on, failure on enabling PML just results in failure
9079          * of creating the vcpu, therefore we can simplify PML logic (by
9080          * avoiding dealing with cases, such as enabling PML partially on vcpus
9081          * for the guest, etc.
9082          */
9083         if (enable_pml) {
9084                 err = vmx_create_pml_buffer(vmx);
9085                 if (err)
9086                         goto free_vmcs;
9087         }
9088
9089         vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9090
9091         return &vmx->vcpu;
9092
9093 free_vmcs:
9094         free_vpid(vmx->nested.vpid02);
9095         free_loaded_vmcs(vmx->loaded_vmcs);
9096 free_msrs:
9097         kfree(vmx->guest_msrs);
9098 uninit_vcpu:
9099         kvm_vcpu_uninit(&vmx->vcpu);
9100 free_vcpu:
9101         free_vpid(vmx->vpid);
9102         kmem_cache_free(kvm_vcpu_cache, vmx);
9103         return ERR_PTR(err);
9104 }
9105
9106 static void __init vmx_check_processor_compat(void *rtn)
9107 {
9108         struct vmcs_config vmcs_conf;
9109
9110         *(int *)rtn = 0;
9111         if (setup_vmcs_config(&vmcs_conf) < 0)
9112                 *(int *)rtn = -EIO;
9113         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9114                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9115                                 smp_processor_id());
9116                 *(int *)rtn = -EIO;
9117         }
9118 }
9119
9120 static int get_ept_level(void)
9121 {
9122         return VMX_EPT_DEFAULT_GAW + 1;
9123 }
9124
9125 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9126 {
9127         u8 cache;
9128         u64 ipat = 0;
9129
9130         /* For VT-d and EPT combination
9131          * 1. MMIO: always map as UC
9132          * 2. EPT with VT-d:
9133          *   a. VT-d without snooping control feature: can't guarantee the
9134          *      result, try to trust guest.
9135          *   b. VT-d with snooping control feature: snooping control feature of
9136          *      VT-d engine can guarantee the cache correctness. Just set it
9137          *      to WB to keep consistent with host. So the same as item 3.
9138          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9139          *    consistent with host MTRR
9140          */
9141         if (is_mmio) {
9142                 cache = MTRR_TYPE_UNCACHABLE;
9143                 goto exit;
9144         }
9145
9146         if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9147                 ipat = VMX_EPT_IPAT_BIT;
9148                 cache = MTRR_TYPE_WRBACK;
9149                 goto exit;
9150         }
9151
9152         if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9153                 ipat = VMX_EPT_IPAT_BIT;
9154                 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9155                         cache = MTRR_TYPE_WRBACK;
9156                 else
9157                         cache = MTRR_TYPE_UNCACHABLE;
9158                 goto exit;
9159         }
9160
9161         cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9162
9163 exit:
9164         return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9165 }
9166
9167 static int vmx_get_lpage_level(void)
9168 {
9169         if (enable_ept && !cpu_has_vmx_ept_1g_page())
9170                 return PT_DIRECTORY_LEVEL;
9171         else
9172                 /* For shadow and EPT supported 1GB page */
9173                 return PT_PDPE_LEVEL;
9174 }
9175
9176 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9177 {
9178         /*
9179          * These bits in the secondary execution controls field
9180          * are dynamic, the others are mostly based on the hypervisor
9181          * architecture and the guest's CPUID.  Do not touch the
9182          * dynamic bits.
9183          */
9184         u32 mask =
9185                 SECONDARY_EXEC_SHADOW_VMCS |
9186                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9187                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9188
9189         u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9190
9191         vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9192                      (new_ctl & ~mask) | (cur_ctl & mask));
9193 }
9194
9195 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9196 {
9197         struct kvm_cpuid_entry2 *best;
9198         struct vcpu_vmx *vmx = to_vmx(vcpu);
9199         u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9200
9201         if (vmx_rdtscp_supported()) {
9202                 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9203                 if (!rdtscp_enabled)
9204                         secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9205
9206                 if (nested) {
9207                         if (rdtscp_enabled)
9208                                 vmx->nested.nested_vmx_secondary_ctls_high |=
9209                                         SECONDARY_EXEC_RDTSCP;
9210                         else
9211                                 vmx->nested.nested_vmx_secondary_ctls_high &=
9212                                         ~SECONDARY_EXEC_RDTSCP;
9213                 }
9214         }
9215
9216         /* Exposing INVPCID only when PCID is exposed */
9217         best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9218         if (vmx_invpcid_supported() &&
9219             (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9220             !guest_cpuid_has_pcid(vcpu))) {
9221                 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9222
9223                 if (best)
9224                         best->ebx &= ~bit(X86_FEATURE_INVPCID);
9225         }
9226
9227         if (cpu_has_secondary_exec_ctrls())
9228                 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9229
9230         if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
9231                 if (guest_cpuid_has_pcommit(vcpu))
9232                         vmx->nested.nested_vmx_secondary_ctls_high |=
9233                                 SECONDARY_EXEC_PCOMMIT;
9234                 else
9235                         vmx->nested.nested_vmx_secondary_ctls_high &=
9236                                 ~SECONDARY_EXEC_PCOMMIT;
9237         }
9238
9239         if (nested_vmx_allowed(vcpu))
9240                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9241                         FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9242         else
9243                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9244                         ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9245 }
9246
9247 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9248 {
9249         if (func == 1 && nested)
9250                 entry->ecx |= bit(X86_FEATURE_VMX);
9251 }
9252
9253 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9254                 struct x86_exception *fault)
9255 {
9256         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9257         u32 exit_reason;
9258
9259         if (fault->error_code & PFERR_RSVD_MASK)
9260                 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9261         else
9262                 exit_reason = EXIT_REASON_EPT_VIOLATION;
9263         nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9264         vmcs12->guest_physical_address = fault->address;
9265 }
9266
9267 /* Callbacks for nested_ept_init_mmu_context: */
9268
9269 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9270 {
9271         /* return the page table to be shadowed - in our case, EPT12 */
9272         return get_vmcs12(vcpu)->ept_pointer;
9273 }
9274
9275 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9276 {
9277         WARN_ON(mmu_is_nested(vcpu));
9278         kvm_init_shadow_ept_mmu(vcpu,
9279                         to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9280                         VMX_EPT_EXECUTE_ONLY_BIT);
9281         vcpu->arch.mmu.set_cr3           = vmx_set_cr3;
9282         vcpu->arch.mmu.get_cr3           = nested_ept_get_cr3;
9283         vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9284
9285         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
9286 }
9287
9288 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9289 {
9290         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9291 }
9292
9293 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9294                                             u16 error_code)
9295 {
9296         bool inequality, bit;
9297
9298         bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9299         inequality =
9300                 (error_code & vmcs12->page_fault_error_code_mask) !=
9301                  vmcs12->page_fault_error_code_match;
9302         return inequality ^ bit;
9303 }
9304
9305 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9306                 struct x86_exception *fault)
9307 {
9308         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9309
9310         WARN_ON(!is_guest_mode(vcpu));
9311
9312         if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9313                 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9314                                   vmcs_read32(VM_EXIT_INTR_INFO),
9315                                   vmcs_readl(EXIT_QUALIFICATION));
9316         else
9317                 kvm_inject_page_fault(vcpu, fault);
9318 }
9319
9320 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9321                                         struct vmcs12 *vmcs12)
9322 {
9323         struct vcpu_vmx *vmx = to_vmx(vcpu);
9324         int maxphyaddr = cpuid_maxphyaddr(vcpu);
9325
9326         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9327                 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9328                     vmcs12->apic_access_addr >> maxphyaddr)
9329                         return false;
9330
9331                 /*
9332                  * Translate L1 physical address to host physical
9333                  * address for vmcs02. Keep the page pinned, so this
9334                  * physical address remains valid. We keep a reference
9335                  * to it so we can release it later.
9336                  */
9337                 if (vmx->nested.apic_access_page) /* shouldn't happen */
9338                         nested_release_page(vmx->nested.apic_access_page);
9339                 vmx->nested.apic_access_page =
9340                         nested_get_page(vcpu, vmcs12->apic_access_addr);
9341         }
9342
9343         if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9344                 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9345                     vmcs12->virtual_apic_page_addr >> maxphyaddr)
9346                         return false;
9347
9348                 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9349                         nested_release_page(vmx->nested.virtual_apic_page);
9350                 vmx->nested.virtual_apic_page =
9351                         nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9352
9353                 /*
9354                  * Failing the vm entry is _not_ what the processor does
9355                  * but it's basically the only possibility we have.
9356                  * We could still enter the guest if CR8 load exits are
9357                  * enabled, CR8 store exits are enabled, and virtualize APIC
9358                  * access is disabled; in this case the processor would never
9359                  * use the TPR shadow and we could simply clear the bit from
9360                  * the execution control.  But such a configuration is useless,
9361                  * so let's keep the code simple.
9362                  */
9363                 if (!vmx->nested.virtual_apic_page)
9364                         return false;
9365         }
9366
9367         if (nested_cpu_has_posted_intr(vmcs12)) {
9368                 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9369                     vmcs12->posted_intr_desc_addr >> maxphyaddr)
9370                         return false;
9371
9372                 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9373                         kunmap(vmx->nested.pi_desc_page);
9374                         nested_release_page(vmx->nested.pi_desc_page);
9375                 }
9376                 vmx->nested.pi_desc_page =
9377                         nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9378                 if (!vmx->nested.pi_desc_page)
9379                         return false;
9380
9381                 vmx->nested.pi_desc =
9382                         (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9383                 if (!vmx->nested.pi_desc) {
9384                         nested_release_page_clean(vmx->nested.pi_desc_page);
9385                         return false;
9386                 }
9387                 vmx->nested.pi_desc =
9388                         (struct pi_desc *)((void *)vmx->nested.pi_desc +
9389                         (unsigned long)(vmcs12->posted_intr_desc_addr &
9390                         (PAGE_SIZE - 1)));
9391         }
9392
9393         return true;
9394 }
9395
9396 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9397 {
9398         u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9399         struct vcpu_vmx *vmx = to_vmx(vcpu);
9400
9401         if (vcpu->arch.virtual_tsc_khz == 0)
9402                 return;
9403
9404         /* Make sure short timeouts reliably trigger an immediate vmexit.
9405          * hrtimer_start does not guarantee this. */
9406         if (preemption_timeout <= 1) {
9407                 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9408                 return;
9409         }
9410
9411         preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9412         preemption_timeout *= 1000000;
9413         do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9414         hrtimer_start(&vmx->nested.preemption_timer,
9415                       ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9416 }
9417
9418 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9419                                                 struct vmcs12 *vmcs12)
9420 {
9421         int maxphyaddr;
9422         u64 addr;
9423
9424         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9425                 return 0;
9426
9427         if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9428                 WARN_ON(1);
9429                 return -EINVAL;
9430         }
9431         maxphyaddr = cpuid_maxphyaddr(vcpu);
9432
9433         if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9434            ((addr + PAGE_SIZE) >> maxphyaddr))
9435                 return -EINVAL;
9436
9437         return 0;
9438 }
9439
9440 /*
9441  * Merge L0's and L1's MSR bitmap, return false to indicate that
9442  * we do not use the hardware.
9443  */
9444 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9445                                                struct vmcs12 *vmcs12)
9446 {
9447         int msr;
9448         struct page *page;
9449         unsigned long *msr_bitmap;
9450
9451         if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9452                 return false;
9453
9454         page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9455         if (!page) {
9456                 WARN_ON(1);
9457                 return false;
9458         }
9459         msr_bitmap = (unsigned long *)kmap(page);
9460         if (!msr_bitmap) {
9461                 nested_release_page_clean(page);
9462                 WARN_ON(1);
9463                 return false;
9464         }
9465
9466         if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9467                 if (nested_cpu_has_apic_reg_virt(vmcs12))
9468                         for (msr = 0x800; msr <= 0x8ff; msr++)
9469                                 nested_vmx_disable_intercept_for_msr(
9470                                         msr_bitmap,
9471                                         vmx_msr_bitmap_nested,
9472                                         msr, MSR_TYPE_R);
9473                 /* TPR is allowed */
9474                 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9475                                 vmx_msr_bitmap_nested,
9476                                 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9477                                 MSR_TYPE_R | MSR_TYPE_W);
9478                 if (nested_cpu_has_vid(vmcs12)) {
9479                         /* EOI and self-IPI are allowed */
9480                         nested_vmx_disable_intercept_for_msr(
9481                                 msr_bitmap,
9482                                 vmx_msr_bitmap_nested,
9483                                 APIC_BASE_MSR + (APIC_EOI >> 4),
9484                                 MSR_TYPE_W);
9485                         nested_vmx_disable_intercept_for_msr(
9486                                 msr_bitmap,
9487                                 vmx_msr_bitmap_nested,
9488                                 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9489                                 MSR_TYPE_W);
9490                 }
9491         } else {
9492                 /*
9493                  * Enable reading intercept of all the x2apic
9494                  * MSRs. We should not rely on vmcs12 to do any
9495                  * optimizations here, it may have been modified
9496                  * by L1.
9497                  */
9498                 for (msr = 0x800; msr <= 0x8ff; msr++)
9499                         __vmx_enable_intercept_for_msr(
9500                                 vmx_msr_bitmap_nested,
9501                                 msr,
9502                                 MSR_TYPE_R);
9503
9504                 __vmx_enable_intercept_for_msr(
9505                                 vmx_msr_bitmap_nested,
9506                                 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9507                                 MSR_TYPE_W);
9508                 __vmx_enable_intercept_for_msr(
9509                                 vmx_msr_bitmap_nested,
9510                                 APIC_BASE_MSR + (APIC_EOI >> 4),
9511                                 MSR_TYPE_W);
9512                 __vmx_enable_intercept_for_msr(
9513                                 vmx_msr_bitmap_nested,
9514                                 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9515                                 MSR_TYPE_W);
9516         }
9517         kunmap(page);
9518         nested_release_page_clean(page);
9519
9520         return true;
9521 }
9522
9523 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9524                                            struct vmcs12 *vmcs12)
9525 {
9526         if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9527             !nested_cpu_has_apic_reg_virt(vmcs12) &&
9528             !nested_cpu_has_vid(vmcs12) &&
9529             !nested_cpu_has_posted_intr(vmcs12))
9530                 return 0;
9531
9532         /*
9533          * If virtualize x2apic mode is enabled,
9534          * virtualize apic access must be disabled.
9535          */
9536         if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9537             nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9538                 return -EINVAL;
9539
9540         /*
9541          * If virtual interrupt delivery is enabled,
9542          * we must exit on external interrupts.
9543          */
9544         if (nested_cpu_has_vid(vmcs12) &&
9545            !nested_exit_on_intr(vcpu))
9546                 return -EINVAL;
9547
9548         /*
9549          * bits 15:8 should be zero in posted_intr_nv,
9550          * the descriptor address has been already checked
9551          * in nested_get_vmcs12_pages.
9552          */
9553         if (nested_cpu_has_posted_intr(vmcs12) &&
9554            (!nested_cpu_has_vid(vmcs12) ||
9555             !nested_exit_intr_ack_set(vcpu) ||
9556             vmcs12->posted_intr_nv & 0xff00))
9557                 return -EINVAL;
9558
9559         /* tpr shadow is needed by all apicv features. */
9560         if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9561                 return -EINVAL;
9562
9563         return 0;
9564 }
9565
9566 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9567                                        unsigned long count_field,
9568                                        unsigned long addr_field)
9569 {
9570         int maxphyaddr;
9571         u64 count, addr;
9572
9573         if (vmcs12_read_any(vcpu, count_field, &count) ||
9574             vmcs12_read_any(vcpu, addr_field, &addr)) {
9575                 WARN_ON(1);
9576                 return -EINVAL;
9577         }
9578         if (count == 0)
9579                 return 0;
9580         maxphyaddr = cpuid_maxphyaddr(vcpu);
9581         if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9582             (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9583                 pr_warn_ratelimited(
9584                         "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9585                         addr_field, maxphyaddr, count, addr);
9586                 return -EINVAL;
9587         }
9588         return 0;
9589 }
9590
9591 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9592                                                 struct vmcs12 *vmcs12)
9593 {
9594         if (vmcs12->vm_exit_msr_load_count == 0 &&
9595             vmcs12->vm_exit_msr_store_count == 0 &&
9596             vmcs12->vm_entry_msr_load_count == 0)
9597                 return 0; /* Fast path */
9598         if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9599                                         VM_EXIT_MSR_LOAD_ADDR) ||
9600             nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9601                                         VM_EXIT_MSR_STORE_ADDR) ||
9602             nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9603                                         VM_ENTRY_MSR_LOAD_ADDR))
9604                 return -EINVAL;
9605         return 0;
9606 }
9607
9608 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9609                                        struct vmx_msr_entry *e)
9610 {
9611         /* x2APIC MSR accesses are not allowed */
9612         if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9613                 return -EINVAL;
9614         if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9615             e->index == MSR_IA32_UCODE_REV)
9616                 return -EINVAL;
9617         if (e->reserved != 0)
9618                 return -EINVAL;
9619         return 0;
9620 }
9621
9622 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9623                                      struct vmx_msr_entry *e)
9624 {
9625         if (e->index == MSR_FS_BASE ||
9626             e->index == MSR_GS_BASE ||
9627             e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9628             nested_vmx_msr_check_common(vcpu, e))
9629                 return -EINVAL;
9630         return 0;
9631 }
9632
9633 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9634                                       struct vmx_msr_entry *e)
9635 {
9636         if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9637             nested_vmx_msr_check_common(vcpu, e))
9638                 return -EINVAL;
9639         return 0;
9640 }
9641
9642 /*
9643  * Load guest's/host's msr at nested entry/exit.
9644  * return 0 for success, entry index for failure.
9645  */
9646 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9647 {
9648         u32 i;
9649         struct vmx_msr_entry e;
9650         struct msr_data msr;
9651
9652         msr.host_initiated = false;
9653         for (i = 0; i < count; i++) {
9654                 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9655                                         &e, sizeof(e))) {
9656                         pr_warn_ratelimited(
9657                                 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9658                                 __func__, i, gpa + i * sizeof(e));
9659                         goto fail;
9660                 }
9661                 if (nested_vmx_load_msr_check(vcpu, &e)) {
9662                         pr_warn_ratelimited(
9663                                 "%s check failed (%u, 0x%x, 0x%x)\n",
9664                                 __func__, i, e.index, e.reserved);
9665                         goto fail;
9666                 }
9667                 msr.index = e.index;
9668                 msr.data = e.value;
9669                 if (kvm_set_msr(vcpu, &msr)) {
9670                         pr_warn_ratelimited(
9671                                 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9672                                 __func__, i, e.index, e.value);
9673                         goto fail;
9674                 }
9675         }
9676         return 0;
9677 fail:
9678         return i + 1;
9679 }
9680
9681 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9682 {
9683         u32 i;
9684         struct vmx_msr_entry e;
9685
9686         for (i = 0; i < count; i++) {
9687                 struct msr_data msr_info;
9688                 if (kvm_vcpu_read_guest(vcpu,
9689                                         gpa + i * sizeof(e),
9690                                         &e, 2 * sizeof(u32))) {
9691                         pr_warn_ratelimited(
9692                                 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9693                                 __func__, i, gpa + i * sizeof(e));
9694                         return -EINVAL;
9695                 }
9696                 if (nested_vmx_store_msr_check(vcpu, &e)) {
9697                         pr_warn_ratelimited(
9698                                 "%s check failed (%u, 0x%x, 0x%x)\n",
9699                                 __func__, i, e.index, e.reserved);
9700                         return -EINVAL;
9701                 }
9702                 msr_info.host_initiated = false;
9703                 msr_info.index = e.index;
9704                 if (kvm_get_msr(vcpu, &msr_info)) {
9705                         pr_warn_ratelimited(
9706                                 "%s cannot read MSR (%u, 0x%x)\n",
9707                                 __func__, i, e.index);
9708                         return -EINVAL;
9709                 }
9710                 if (kvm_vcpu_write_guest(vcpu,
9711                                          gpa + i * sizeof(e) +
9712                                              offsetof(struct vmx_msr_entry, value),
9713                                          &msr_info.data, sizeof(msr_info.data))) {
9714                         pr_warn_ratelimited(
9715                                 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9716                                 __func__, i, e.index, msr_info.data);
9717                         return -EINVAL;
9718                 }
9719         }
9720         return 0;
9721 }
9722
9723 /*
9724  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9725  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9726  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9727  * guest in a way that will both be appropriate to L1's requests, and our
9728  * needs. In addition to modifying the active vmcs (which is vmcs02), this
9729  * function also has additional necessary side-effects, like setting various
9730  * vcpu->arch fields.
9731  */
9732 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9733 {
9734         struct vcpu_vmx *vmx = to_vmx(vcpu);
9735         u32 exec_control;
9736
9737         vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9738         vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9739         vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9740         vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9741         vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9742         vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9743         vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9744         vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9745         vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9746         vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9747         vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9748         vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9749         vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9750         vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9751         vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9752         vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9753         vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9754         vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9755         vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9756         vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9757         vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9758         vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9759         vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9760         vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9761         vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9762         vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9763         vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9764         vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9765         vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9766         vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9767         vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9768         vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9769         vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9770         vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9771         vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9772         vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9773
9774         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9775                 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9776                 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9777         } else {
9778                 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9779                 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9780         }
9781         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9782                 vmcs12->vm_entry_intr_info_field);
9783         vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9784                 vmcs12->vm_entry_exception_error_code);
9785         vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9786                 vmcs12->vm_entry_instruction_len);
9787         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9788                 vmcs12->guest_interruptibility_info);
9789         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9790         vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9791         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9792                 vmcs12->guest_pending_dbg_exceptions);
9793         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9794         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9795
9796         if (nested_cpu_has_xsaves(vmcs12))
9797                 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9798         vmcs_write64(VMCS_LINK_POINTER, -1ull);
9799
9800         exec_control = vmcs12->pin_based_vm_exec_control;
9801         exec_control |= vmcs_config.pin_based_exec_ctrl;
9802         exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9803
9804         if (nested_cpu_has_posted_intr(vmcs12)) {
9805                 /*
9806                  * Note that we use L0's vector here and in
9807                  * vmx_deliver_nested_posted_interrupt.
9808                  */
9809                 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9810                 vmx->nested.pi_pending = false;
9811                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9812                 vmcs_write64(POSTED_INTR_DESC_ADDR,
9813                         page_to_phys(vmx->nested.pi_desc_page) +
9814                         (unsigned long)(vmcs12->posted_intr_desc_addr &
9815                         (PAGE_SIZE - 1)));
9816         } else
9817                 exec_control &= ~PIN_BASED_POSTED_INTR;
9818
9819         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9820
9821         vmx->nested.preemption_timer_expired = false;
9822         if (nested_cpu_has_preemption_timer(vmcs12))
9823                 vmx_start_preemption_timer(vcpu);
9824
9825         /*
9826          * Whether page-faults are trapped is determined by a combination of
9827          * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9828          * If enable_ept, L0 doesn't care about page faults and we should
9829          * set all of these to L1's desires. However, if !enable_ept, L0 does
9830          * care about (at least some) page faults, and because it is not easy
9831          * (if at all possible?) to merge L0 and L1's desires, we simply ask
9832          * to exit on each and every L2 page fault. This is done by setting
9833          * MASK=MATCH=0 and (see below) EB.PF=1.
9834          * Note that below we don't need special code to set EB.PF beyond the
9835          * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9836          * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9837          * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9838          *
9839          * A problem with this approach (when !enable_ept) is that L1 may be
9840          * injected with more page faults than it asked for. This could have
9841          * caused problems, but in practice existing hypervisors don't care.
9842          * To fix this, we will need to emulate the PFEC checking (on the L1
9843          * page tables), using walk_addr(), when injecting PFs to L1.
9844          */
9845         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9846                 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9847         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9848                 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9849
9850         if (cpu_has_secondary_exec_ctrls()) {
9851                 exec_control = vmx_secondary_exec_control(vmx);
9852
9853                 /* Take the following fields only from vmcs12 */
9854                 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9855                                   SECONDARY_EXEC_RDTSCP |
9856                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9857                                   SECONDARY_EXEC_APIC_REGISTER_VIRT |
9858                                   SECONDARY_EXEC_PCOMMIT);
9859                 if (nested_cpu_has(vmcs12,
9860                                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9861                         exec_control |= vmcs12->secondary_vm_exec_control;
9862
9863                 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9864                         /*
9865                          * If translation failed, no matter: This feature asks
9866                          * to exit when accessing the given address, and if it
9867                          * can never be accessed, this feature won't do
9868                          * anything anyway.
9869                          */
9870                         if (!vmx->nested.apic_access_page)
9871                                 exec_control &=
9872                                   ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9873                         else
9874                                 vmcs_write64(APIC_ACCESS_ADDR,
9875                                   page_to_phys(vmx->nested.apic_access_page));
9876                 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9877                             cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9878                         exec_control |=
9879                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9880                         kvm_vcpu_reload_apic_access_page(vcpu);
9881                 }
9882
9883                 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9884                         vmcs_write64(EOI_EXIT_BITMAP0,
9885                                 vmcs12->eoi_exit_bitmap0);
9886                         vmcs_write64(EOI_EXIT_BITMAP1,
9887                                 vmcs12->eoi_exit_bitmap1);
9888                         vmcs_write64(EOI_EXIT_BITMAP2,
9889                                 vmcs12->eoi_exit_bitmap2);
9890                         vmcs_write64(EOI_EXIT_BITMAP3,
9891                                 vmcs12->eoi_exit_bitmap3);
9892                         vmcs_write16(GUEST_INTR_STATUS,
9893                                 vmcs12->guest_intr_status);
9894                 }
9895
9896                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9897         }
9898
9899
9900         /*
9901          * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9902          * Some constant fields are set here by vmx_set_constant_host_state().
9903          * Other fields are different per CPU, and will be set later when
9904          * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9905          */
9906         vmx_set_constant_host_state(vmx);
9907
9908         /*
9909          * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9910          * entry, but only if the current (host) sp changed from the value
9911          * we wrote last (vmx->host_rsp). This cache is no longer relevant
9912          * if we switch vmcs, and rather than hold a separate cache per vmcs,
9913          * here we just force the write to happen on entry.
9914          */
9915         vmx->host_rsp = 0;
9916
9917         exec_control = vmx_exec_control(vmx); /* L0's desires */
9918         exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9919         exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9920         exec_control &= ~CPU_BASED_TPR_SHADOW;
9921         exec_control |= vmcs12->cpu_based_vm_exec_control;
9922
9923         if (exec_control & CPU_BASED_TPR_SHADOW) {
9924                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9925                                 page_to_phys(vmx->nested.virtual_apic_page));
9926                 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9927         }
9928
9929         if (cpu_has_vmx_msr_bitmap() &&
9930             exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9931                 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9932                 /* MSR_BITMAP will be set by following vmx_set_efer. */
9933         } else
9934                 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9935
9936         /*
9937          * Merging of IO bitmap not currently supported.
9938          * Rather, exit every time.
9939          */
9940         exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9941         exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9942
9943         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9944
9945         /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9946          * bitwise-or of what L1 wants to trap for L2, and what we want to
9947          * trap. Note that CR0.TS also needs updating - we do this later.
9948          */
9949         update_exception_bitmap(vcpu);
9950         vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9951         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9952
9953         /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9954          * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9955          * bits are further modified by vmx_set_efer() below.
9956          */
9957         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9958
9959         /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9960          * emulated by vmx_set_efer(), below.
9961          */
9962         vm_entry_controls_init(vmx,
9963                 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9964                         ~VM_ENTRY_IA32E_MODE) |
9965                 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9966
9967         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9968                 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9969                 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9970         } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9971                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9972
9973
9974         set_cr4_guest_host_mask(vmx);
9975
9976         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9977                 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9978
9979         if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9980                 vmcs_write64(TSC_OFFSET,
9981                         vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9982         else
9983                 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9984
9985         if (enable_vpid) {
9986                 /*
9987                  * There is no direct mapping between vpid02 and vpid12, the
9988                  * vpid02 is per-vCPU for L0 and reused while the value of
9989                  * vpid12 is changed w/ one invvpid during nested vmentry.
9990                  * The vpid12 is allocated by L1 for L2, so it will not
9991                  * influence global bitmap(for vpid01 and vpid02 allocation)
9992                  * even if spawn a lot of nested vCPUs.
9993                  */
9994                 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
9995                         vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
9996                         if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
9997                                 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
9998                                 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
9999                         }
10000                 } else {
10001                         vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10002                         vmx_flush_tlb(vcpu);
10003                 }
10004
10005         }
10006
10007         if (nested_cpu_has_ept(vmcs12)) {
10008                 kvm_mmu_unload(vcpu);
10009                 nested_ept_init_mmu_context(vcpu);
10010         }
10011
10012         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
10013                 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10014         else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10015                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10016         else
10017                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10018         /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10019         vmx_set_efer(vcpu, vcpu->arch.efer);
10020
10021         /*
10022          * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
10023          * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
10024          * The CR0_READ_SHADOW is what L2 should have expected to read given
10025          * the specifications by L1; It's not enough to take
10026          * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10027          * have more bits than L1 expected.
10028          */
10029         vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10030         vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10031
10032         vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10033         vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10034
10035         /* shadow page tables on either EPT or shadow page tables */
10036         kvm_set_cr3(vcpu, vmcs12->guest_cr3);
10037         kvm_mmu_reset_context(vcpu);
10038
10039         if (!enable_ept)
10040                 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10041
10042         /*
10043          * L1 may access the L2's PDPTR, so save them to construct vmcs12
10044          */
10045         if (enable_ept) {
10046                 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10047                 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10048                 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10049                 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10050         }
10051
10052         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10053         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10054 }
10055
10056 /*
10057  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10058  * for running an L2 nested guest.
10059  */
10060 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10061 {
10062         struct vmcs12 *vmcs12;
10063         struct vcpu_vmx *vmx = to_vmx(vcpu);
10064         int cpu;
10065         struct loaded_vmcs *vmcs02;
10066         bool ia32e;
10067         u32 msr_entry_idx;
10068
10069         if (!nested_vmx_check_permission(vcpu) ||
10070             !nested_vmx_check_vmcs12(vcpu))
10071                 return 1;
10072
10073         skip_emulated_instruction(vcpu);
10074         vmcs12 = get_vmcs12(vcpu);
10075
10076         if (enable_shadow_vmcs)
10077                 copy_shadow_to_vmcs12(vmx);
10078
10079         /*
10080          * The nested entry process starts with enforcing various prerequisites
10081          * on vmcs12 as required by the Intel SDM, and act appropriately when
10082          * they fail: As the SDM explains, some conditions should cause the
10083          * instruction to fail, while others will cause the instruction to seem
10084          * to succeed, but return an EXIT_REASON_INVALID_STATE.
10085          * To speed up the normal (success) code path, we should avoid checking
10086          * for misconfigurations which will anyway be caught by the processor
10087          * when using the merged vmcs02.
10088          */
10089         if (vmcs12->launch_state == launch) {
10090                 nested_vmx_failValid(vcpu,
10091                         launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10092                                : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10093                 return 1;
10094         }
10095
10096         if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10097             vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
10098                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10099                 return 1;
10100         }
10101
10102         if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
10103                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10104                 return 1;
10105         }
10106
10107         if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
10108                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10109                 return 1;
10110         }
10111
10112         if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
10113                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10114                 return 1;
10115         }
10116
10117         if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
10118                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10119                 return 1;
10120         }
10121
10122         if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10123                                 vmx->nested.nested_vmx_true_procbased_ctls_low,
10124                                 vmx->nested.nested_vmx_procbased_ctls_high) ||
10125             !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10126                                 vmx->nested.nested_vmx_secondary_ctls_low,
10127                                 vmx->nested.nested_vmx_secondary_ctls_high) ||
10128             !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10129                                 vmx->nested.nested_vmx_pinbased_ctls_low,
10130                                 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10131             !vmx_control_verify(vmcs12->vm_exit_controls,
10132                                 vmx->nested.nested_vmx_true_exit_ctls_low,
10133                                 vmx->nested.nested_vmx_exit_ctls_high) ||
10134             !vmx_control_verify(vmcs12->vm_entry_controls,
10135                                 vmx->nested.nested_vmx_true_entry_ctls_low,
10136                                 vmx->nested.nested_vmx_entry_ctls_high))
10137         {
10138                 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10139                 return 1;
10140         }
10141
10142         if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
10143             ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
10144                 nested_vmx_failValid(vcpu,
10145                         VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
10146                 return 1;
10147         }
10148
10149         if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10150             ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
10151                 nested_vmx_entry_failure(vcpu, vmcs12,
10152                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10153                 return 1;
10154         }
10155         if (vmcs12->vmcs_link_pointer != -1ull) {
10156                 nested_vmx_entry_failure(vcpu, vmcs12,
10157                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
10158                 return 1;
10159         }
10160
10161         /*
10162          * If the load IA32_EFER VM-entry control is 1, the following checks
10163          * are performed on the field for the IA32_EFER MSR:
10164          * - Bits reserved in the IA32_EFER MSR must be 0.
10165          * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10166          *   the IA-32e mode guest VM-exit control. It must also be identical
10167          *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10168          *   CR0.PG) is 1.
10169          */
10170         if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
10171                 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10172                 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10173                     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10174                     ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10175                      ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
10176                         nested_vmx_entry_failure(vcpu, vmcs12,
10177                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10178                         return 1;
10179                 }
10180         }
10181
10182         /*
10183          * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10184          * IA32_EFER MSR must be 0 in the field for that register. In addition,
10185          * the values of the LMA and LME bits in the field must each be that of
10186          * the host address-space size VM-exit control.
10187          */
10188         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10189                 ia32e = (vmcs12->vm_exit_controls &
10190                          VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10191                 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10192                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10193                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
10194                         nested_vmx_entry_failure(vcpu, vmcs12,
10195                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10196                         return 1;
10197                 }
10198         }
10199
10200         /*
10201          * We're finally done with prerequisite checking, and can start with
10202          * the nested entry.
10203          */
10204
10205         vmcs02 = nested_get_current_vmcs02(vmx);
10206         if (!vmcs02)
10207                 return -ENOMEM;
10208
10209         enter_guest_mode(vcpu);
10210
10211         vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
10212
10213         if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10214                 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10215
10216         cpu = get_cpu();
10217         vmx->loaded_vmcs = vmcs02;
10218         vmx_vcpu_put(vcpu);
10219         vmx_vcpu_load(vcpu, cpu);
10220         vcpu->cpu = cpu;
10221         put_cpu();
10222
10223         vmx_segment_cache_clear(vmx);
10224
10225         prepare_vmcs02(vcpu, vmcs12);
10226
10227         msr_entry_idx = nested_vmx_load_msr(vcpu,
10228                                             vmcs12->vm_entry_msr_load_addr,
10229                                             vmcs12->vm_entry_msr_load_count);
10230         if (msr_entry_idx) {
10231                 leave_guest_mode(vcpu);
10232                 vmx_load_vmcs01(vcpu);
10233                 nested_vmx_entry_failure(vcpu, vmcs12,
10234                                 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10235                 return 1;
10236         }
10237
10238         vmcs12->launch_state = 1;
10239
10240         if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10241                 return kvm_vcpu_halt(vcpu);
10242
10243         vmx->nested.nested_run_pending = 1;
10244
10245         /*
10246          * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10247          * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10248          * returned as far as L1 is concerned. It will only return (and set
10249          * the success flag) when L2 exits (see nested_vmx_vmexit()).
10250          */
10251         return 1;
10252 }
10253
10254 /*
10255  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10256  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10257  * This function returns the new value we should put in vmcs12.guest_cr0.
10258  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10259  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10260  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10261  *     didn't trap the bit, because if L1 did, so would L0).
10262  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10263  *     been modified by L2, and L1 knows it. So just leave the old value of
10264  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10265  *     isn't relevant, because if L0 traps this bit it can set it to anything.
10266  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10267  *     changed these bits, and therefore they need to be updated, but L0
10268  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10269  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10270  */
10271 static inline unsigned long
10272 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10273 {
10274         return
10275         /*1*/   (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10276         /*2*/   (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10277         /*3*/   (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10278                         vcpu->arch.cr0_guest_owned_bits));
10279 }
10280
10281 static inline unsigned long
10282 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10283 {
10284         return
10285         /*1*/   (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10286         /*2*/   (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10287         /*3*/   (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10288                         vcpu->arch.cr4_guest_owned_bits));
10289 }
10290
10291 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10292                                        struct vmcs12 *vmcs12)
10293 {
10294         u32 idt_vectoring;
10295         unsigned int nr;
10296
10297         if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10298                 nr = vcpu->arch.exception.nr;
10299                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10300
10301                 if (kvm_exception_is_soft(nr)) {
10302                         vmcs12->vm_exit_instruction_len =
10303                                 vcpu->arch.event_exit_inst_len;
10304                         idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10305                 } else
10306                         idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10307
10308                 if (vcpu->arch.exception.has_error_code) {
10309                         idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10310                         vmcs12->idt_vectoring_error_code =
10311                                 vcpu->arch.exception.error_code;
10312                 }
10313
10314                 vmcs12->idt_vectoring_info_field = idt_vectoring;
10315         } else if (vcpu->arch.nmi_injected) {
10316                 vmcs12->idt_vectoring_info_field =
10317                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10318         } else if (vcpu->arch.interrupt.pending) {
10319                 nr = vcpu->arch.interrupt.nr;
10320                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10321
10322                 if (vcpu->arch.interrupt.soft) {
10323                         idt_vectoring |= INTR_TYPE_SOFT_INTR;
10324                         vmcs12->vm_entry_instruction_len =
10325                                 vcpu->arch.event_exit_inst_len;
10326                 } else
10327                         idt_vectoring |= INTR_TYPE_EXT_INTR;
10328
10329                 vmcs12->idt_vectoring_info_field = idt_vectoring;
10330         }
10331 }
10332
10333 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10334 {
10335         struct vcpu_vmx *vmx = to_vmx(vcpu);
10336
10337         if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10338             vmx->nested.preemption_timer_expired) {
10339                 if (vmx->nested.nested_run_pending)
10340                         return -EBUSY;
10341                 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10342                 return 0;
10343         }
10344
10345         if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10346                 if (vmx->nested.nested_run_pending ||
10347                     vcpu->arch.interrupt.pending)
10348                         return -EBUSY;
10349                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10350                                   NMI_VECTOR | INTR_TYPE_NMI_INTR |
10351                                   INTR_INFO_VALID_MASK, 0);
10352                 /*
10353                  * The NMI-triggered VM exit counts as injection:
10354                  * clear this one and block further NMIs.
10355                  */
10356                 vcpu->arch.nmi_pending = 0;
10357                 vmx_set_nmi_mask(vcpu, true);
10358                 return 0;
10359         }
10360
10361         if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10362             nested_exit_on_intr(vcpu)) {
10363                 if (vmx->nested.nested_run_pending)
10364                         return -EBUSY;
10365                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10366                 return 0;
10367         }
10368
10369         return vmx_complete_nested_posted_interrupt(vcpu);
10370 }
10371
10372 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10373 {
10374         ktime_t remaining =
10375                 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10376         u64 value;
10377
10378         if (ktime_to_ns(remaining) <= 0)
10379                 return 0;
10380
10381         value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10382         do_div(value, 1000000);
10383         return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10384 }
10385
10386 /*
10387  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10388  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10389  * and this function updates it to reflect the changes to the guest state while
10390  * L2 was running (and perhaps made some exits which were handled directly by L0
10391  * without going back to L1), and to reflect the exit reason.
10392  * Note that we do not have to copy here all VMCS fields, just those that
10393  * could have changed by the L2 guest or the exit - i.e., the guest-state and
10394  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10395  * which already writes to vmcs12 directly.
10396  */
10397 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10398                            u32 exit_reason, u32 exit_intr_info,
10399                            unsigned long exit_qualification)
10400 {
10401         /* update guest state fields: */
10402         vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10403         vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10404
10405         vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10406         vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10407         vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10408
10409         vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10410         vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10411         vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10412         vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10413         vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10414         vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10415         vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10416         vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10417         vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10418         vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10419         vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10420         vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10421         vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10422         vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10423         vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10424         vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10425         vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10426         vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10427         vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10428         vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10429         vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10430         vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10431         vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10432         vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10433         vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10434         vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10435         vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10436         vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10437         vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10438         vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10439         vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10440         vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10441         vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10442         vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10443         vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10444         vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10445
10446         vmcs12->guest_interruptibility_info =
10447                 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10448         vmcs12->guest_pending_dbg_exceptions =
10449                 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10450         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10451                 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10452         else
10453                 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10454
10455         if (nested_cpu_has_preemption_timer(vmcs12)) {
10456                 if (vmcs12->vm_exit_controls &
10457                     VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10458                         vmcs12->vmx_preemption_timer_value =
10459                                 vmx_get_preemption_timer_value(vcpu);
10460                 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10461         }
10462
10463         /*
10464          * In some cases (usually, nested EPT), L2 is allowed to change its
10465          * own CR3 without exiting. If it has changed it, we must keep it.
10466          * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10467          * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10468          *
10469          * Additionally, restore L2's PDPTR to vmcs12.
10470          */
10471         if (enable_ept) {
10472                 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10473                 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10474                 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10475                 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10476                 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10477         }
10478
10479         if (nested_cpu_has_vid(vmcs12))
10480                 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10481
10482         vmcs12->vm_entry_controls =
10483                 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10484                 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10485
10486         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10487                 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10488                 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10489         }
10490
10491         /* TODO: These cannot have changed unless we have MSR bitmaps and
10492          * the relevant bit asks not to trap the change */
10493         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10494                 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10495         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10496                 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10497         vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10498         vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10499         vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10500         if (kvm_mpx_supported())
10501                 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10502         if (nested_cpu_has_xsaves(vmcs12))
10503                 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10504
10505         /* update exit information fields: */
10506
10507         vmcs12->vm_exit_reason = exit_reason;
10508         vmcs12->exit_qualification = exit_qualification;
10509
10510         vmcs12->vm_exit_intr_info = exit_intr_info;
10511         if ((vmcs12->vm_exit_intr_info &
10512              (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10513             (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10514                 vmcs12->vm_exit_intr_error_code =
10515                         vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10516         vmcs12->idt_vectoring_info_field = 0;
10517         vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10518         vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10519
10520         if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10521                 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10522                  * instead of reading the real value. */
10523                 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10524
10525                 /*
10526                  * Transfer the event that L0 or L1 may wanted to inject into
10527                  * L2 to IDT_VECTORING_INFO_FIELD.
10528                  */
10529                 vmcs12_save_pending_event(vcpu, vmcs12);
10530         }
10531
10532         /*
10533          * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10534          * preserved above and would only end up incorrectly in L1.
10535          */
10536         vcpu->arch.nmi_injected = false;
10537         kvm_clear_exception_queue(vcpu);
10538         kvm_clear_interrupt_queue(vcpu);
10539 }
10540
10541 /*
10542  * A part of what we need to when the nested L2 guest exits and we want to
10543  * run its L1 parent, is to reset L1's guest state to the host state specified
10544  * in vmcs12.
10545  * This function is to be called not only on normal nested exit, but also on
10546  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10547  * Failures During or After Loading Guest State").
10548  * This function should be called when the active VMCS is L1's (vmcs01).
10549  */
10550 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10551                                    struct vmcs12 *vmcs12)
10552 {
10553         struct kvm_segment seg;
10554
10555         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10556                 vcpu->arch.efer = vmcs12->host_ia32_efer;
10557         else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10558                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10559         else
10560                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10561         vmx_set_efer(vcpu, vcpu->arch.efer);
10562
10563         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10564         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10565         vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10566         /*
10567          * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10568          * actually changed, because it depends on the current state of
10569          * fpu_active (which may have changed).
10570          * Note that vmx_set_cr0 refers to efer set above.
10571          */
10572         vmx_set_cr0(vcpu, vmcs12->host_cr0);
10573         /*
10574          * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10575          * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10576          * but we also need to update cr0_guest_host_mask and exception_bitmap.
10577          */
10578         update_exception_bitmap(vcpu);
10579         vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10580         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10581
10582         /*
10583          * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10584          * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10585          */
10586         vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10587         kvm_set_cr4(vcpu, vmcs12->host_cr4);
10588
10589         nested_ept_uninit_mmu_context(vcpu);
10590
10591         kvm_set_cr3(vcpu, vmcs12->host_cr3);
10592         kvm_mmu_reset_context(vcpu);
10593
10594         if (!enable_ept)
10595                 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10596
10597         if (enable_vpid) {
10598                 /*
10599                  * Trivially support vpid by letting L2s share their parent
10600                  * L1's vpid. TODO: move to a more elaborate solution, giving
10601                  * each L2 its own vpid and exposing the vpid feature to L1.
10602                  */
10603                 vmx_flush_tlb(vcpu);
10604         }
10605
10606
10607         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10608         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10609         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10610         vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10611         vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10612
10613         /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
10614         if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10615                 vmcs_write64(GUEST_BNDCFGS, 0);
10616
10617         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10618                 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10619                 vcpu->arch.pat = vmcs12->host_ia32_pat;
10620         }
10621         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10622                 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10623                         vmcs12->host_ia32_perf_global_ctrl);
10624
10625         /* Set L1 segment info according to Intel SDM
10626             27.5.2 Loading Host Segment and Descriptor-Table Registers */
10627         seg = (struct kvm_segment) {
10628                 .base = 0,
10629                 .limit = 0xFFFFFFFF,
10630                 .selector = vmcs12->host_cs_selector,
10631                 .type = 11,
10632                 .present = 1,
10633                 .s = 1,
10634                 .g = 1
10635         };
10636         if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10637                 seg.l = 1;
10638         else
10639                 seg.db = 1;
10640         vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10641         seg = (struct kvm_segment) {
10642                 .base = 0,
10643                 .limit = 0xFFFFFFFF,
10644                 .type = 3,
10645                 .present = 1,
10646                 .s = 1,
10647                 .db = 1,
10648                 .g = 1
10649         };
10650         seg.selector = vmcs12->host_ds_selector;
10651         vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10652         seg.selector = vmcs12->host_es_selector;
10653         vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10654         seg.selector = vmcs12->host_ss_selector;
10655         vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10656         seg.selector = vmcs12->host_fs_selector;
10657         seg.base = vmcs12->host_fs_base;
10658         vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10659         seg.selector = vmcs12->host_gs_selector;
10660         seg.base = vmcs12->host_gs_base;
10661         vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10662         seg = (struct kvm_segment) {
10663                 .base = vmcs12->host_tr_base,
10664                 .limit = 0x67,
10665                 .selector = vmcs12->host_tr_selector,
10666                 .type = 11,
10667                 .present = 1
10668         };
10669         vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10670
10671         kvm_set_dr(vcpu, 7, 0x400);
10672         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10673
10674         if (cpu_has_vmx_msr_bitmap())
10675                 vmx_set_msr_bitmap(vcpu);
10676
10677         if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10678                                 vmcs12->vm_exit_msr_load_count))
10679                 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10680 }
10681
10682 /*
10683  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10684  * and modify vmcs12 to make it see what it would expect to see there if
10685  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10686  */
10687 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10688                               u32 exit_intr_info,
10689                               unsigned long exit_qualification)
10690 {
10691         struct vcpu_vmx *vmx = to_vmx(vcpu);
10692         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10693
10694         /* trying to cancel vmlaunch/vmresume is a bug */
10695         WARN_ON_ONCE(vmx->nested.nested_run_pending);
10696
10697         leave_guest_mode(vcpu);
10698         prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10699                        exit_qualification);
10700
10701         if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10702                                  vmcs12->vm_exit_msr_store_count))
10703                 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10704
10705         vmx_load_vmcs01(vcpu);
10706
10707         if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10708             && nested_exit_intr_ack_set(vcpu)) {
10709                 int irq = kvm_cpu_get_interrupt(vcpu);
10710                 WARN_ON(irq < 0);
10711                 vmcs12->vm_exit_intr_info = irq |
10712                         INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10713         }
10714
10715         trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10716                                        vmcs12->exit_qualification,
10717                                        vmcs12->idt_vectoring_info_field,
10718                                        vmcs12->vm_exit_intr_info,
10719                                        vmcs12->vm_exit_intr_error_code,
10720                                        KVM_ISA_VMX);
10721
10722         vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10723         vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10724         vmx_segment_cache_clear(vmx);
10725
10726         /* if no vmcs02 cache requested, remove the one we used */
10727         if (VMCS02_POOL_SIZE == 0)
10728                 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10729
10730         load_vmcs12_host_state(vcpu, vmcs12);
10731
10732         /* Update TSC_OFFSET if TSC was changed while L2 ran */
10733         vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10734
10735         /* This is needed for same reason as it was needed in prepare_vmcs02 */
10736         vmx->host_rsp = 0;
10737
10738         /* Unpin physical memory we referred to in vmcs02 */
10739         if (vmx->nested.apic_access_page) {
10740                 nested_release_page(vmx->nested.apic_access_page);
10741                 vmx->nested.apic_access_page = NULL;
10742         }
10743         if (vmx->nested.virtual_apic_page) {
10744                 nested_release_page(vmx->nested.virtual_apic_page);
10745                 vmx->nested.virtual_apic_page = NULL;
10746         }
10747         if (vmx->nested.pi_desc_page) {
10748                 kunmap(vmx->nested.pi_desc_page);
10749                 nested_release_page(vmx->nested.pi_desc_page);
10750                 vmx->nested.pi_desc_page = NULL;
10751                 vmx->nested.pi_desc = NULL;
10752         }
10753
10754         /*
10755          * We are now running in L2, mmu_notifier will force to reload the
10756          * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10757          */
10758         kvm_vcpu_reload_apic_access_page(vcpu);
10759
10760         /*
10761          * Exiting from L2 to L1, we're now back to L1 which thinks it just
10762          * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10763          * success or failure flag accordingly.
10764          */
10765         if (unlikely(vmx->fail)) {
10766                 vmx->fail = 0;
10767                 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10768         } else
10769                 nested_vmx_succeed(vcpu);
10770         if (enable_shadow_vmcs)
10771                 vmx->nested.sync_shadow_vmcs = true;
10772
10773         /* in case we halted in L2 */
10774         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10775 }
10776
10777 /*
10778  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10779  */
10780 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10781 {
10782         if (is_guest_mode(vcpu))
10783                 nested_vmx_vmexit(vcpu, -1, 0, 0);
10784         free_nested(to_vmx(vcpu));
10785 }
10786
10787 /*
10788  * L1's failure to enter L2 is a subset of a normal exit, as explained in
10789  * 23.7 "VM-entry failures during or after loading guest state" (this also
10790  * lists the acceptable exit-reason and exit-qualification parameters).
10791  * It should only be called before L2 actually succeeded to run, and when
10792  * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10793  */
10794 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10795                         struct vmcs12 *vmcs12,
10796                         u32 reason, unsigned long qualification)
10797 {
10798         load_vmcs12_host_state(vcpu, vmcs12);
10799         vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10800         vmcs12->exit_qualification = qualification;
10801         nested_vmx_succeed(vcpu);
10802         if (enable_shadow_vmcs)
10803                 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10804 }
10805
10806 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10807                                struct x86_instruction_info *info,
10808                                enum x86_intercept_stage stage)
10809 {
10810         return X86EMUL_CONTINUE;
10811 }
10812
10813 #ifdef CONFIG_X86_64
10814 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
10815 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
10816                                   u64 divisor, u64 *result)
10817 {
10818         u64 low = a << shift, high = a >> (64 - shift);
10819
10820         /* To avoid the overflow on divq */
10821         if (high >= divisor)
10822                 return 1;
10823
10824         /* Low hold the result, high hold rem which is discarded */
10825         asm("divq %2\n\t" : "=a" (low), "=d" (high) :
10826             "rm" (divisor), "0" (low), "1" (high));
10827         *result = low;
10828
10829         return 0;
10830 }
10831
10832 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
10833 {
10834         struct vcpu_vmx *vmx = to_vmx(vcpu);
10835         u64 tscl = rdtsc(), delta_tsc;
10836
10837         delta_tsc = guest_deadline_tsc - kvm_read_l1_tsc(vcpu, tscl);
10838
10839         /* Convert to host delta tsc if tsc scaling is enabled */
10840         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
10841                         u64_shl_div_u64(delta_tsc,
10842                                 kvm_tsc_scaling_ratio_frac_bits,
10843                                 vcpu->arch.tsc_scaling_ratio,
10844                                 &delta_tsc))
10845                 return -ERANGE;
10846
10847         /*
10848          * If the delta tsc can't fit in the 32 bit after the multi shift,
10849          * we can't use the preemption timer.
10850          * It's possible that it fits on later vmentries, but checking
10851          * on every vmentry is costly so we just use an hrtimer.
10852          */
10853         if (delta_tsc >> (cpu_preemption_timer_multi + 32))
10854                 return -ERANGE;
10855
10856         vmx->hv_deadline_tsc = tscl + delta_tsc;
10857         vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
10858                         PIN_BASED_VMX_PREEMPTION_TIMER);
10859         return 0;
10860 }
10861
10862 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
10863 {
10864         struct vcpu_vmx *vmx = to_vmx(vcpu);
10865         vmx->hv_deadline_tsc = -1;
10866         vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
10867                         PIN_BASED_VMX_PREEMPTION_TIMER);
10868 }
10869 #endif
10870
10871 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10872 {
10873         if (ple_gap)
10874                 shrink_ple_window(vcpu);
10875 }
10876
10877 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10878                                      struct kvm_memory_slot *slot)
10879 {
10880         kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10881         kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10882 }
10883
10884 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10885                                        struct kvm_memory_slot *slot)
10886 {
10887         kvm_mmu_slot_set_dirty(kvm, slot);
10888 }
10889
10890 static void vmx_flush_log_dirty(struct kvm *kvm)
10891 {
10892         kvm_flush_pml_buffers(kvm);
10893 }
10894
10895 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10896                                            struct kvm_memory_slot *memslot,
10897                                            gfn_t offset, unsigned long mask)
10898 {
10899         kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10900 }
10901
10902 /*
10903  * This routine does the following things for vCPU which is going
10904  * to be blocked if VT-d PI is enabled.
10905  * - Store the vCPU to the wakeup list, so when interrupts happen
10906  *   we can find the right vCPU to wake up.
10907  * - Change the Posted-interrupt descriptor as below:
10908  *      'NDST' <-- vcpu->pre_pcpu
10909  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10910  * - If 'ON' is set during this process, which means at least one
10911  *   interrupt is posted for this vCPU, we cannot block it, in
10912  *   this case, return 1, otherwise, return 0.
10913  *
10914  */
10915 static int pi_pre_block(struct kvm_vcpu *vcpu)
10916 {
10917         unsigned long flags;
10918         unsigned int dest;
10919         struct pi_desc old, new;
10920         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10921
10922         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10923                 !irq_remapping_cap(IRQ_POSTING_CAP))
10924                 return 0;
10925
10926         vcpu->pre_pcpu = vcpu->cpu;
10927         spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10928                           vcpu->pre_pcpu), flags);
10929         list_add_tail(&vcpu->blocked_vcpu_list,
10930                       &per_cpu(blocked_vcpu_on_cpu,
10931                       vcpu->pre_pcpu));
10932         spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10933                                vcpu->pre_pcpu), flags);
10934
10935         do {
10936                 old.control = new.control = pi_desc->control;
10937
10938                 /*
10939                  * We should not block the vCPU if
10940                  * an interrupt is posted for it.
10941                  */
10942                 if (pi_test_on(pi_desc) == 1) {
10943                         spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10944                                           vcpu->pre_pcpu), flags);
10945                         list_del(&vcpu->blocked_vcpu_list);
10946                         spin_unlock_irqrestore(
10947                                         &per_cpu(blocked_vcpu_on_cpu_lock,
10948                                         vcpu->pre_pcpu), flags);
10949                         vcpu->pre_pcpu = -1;
10950
10951                         return 1;
10952                 }
10953
10954                 WARN((pi_desc->sn == 1),
10955                      "Warning: SN field of posted-interrupts "
10956                      "is set before blocking\n");
10957
10958                 /*
10959                  * Since vCPU can be preempted during this process,
10960                  * vcpu->cpu could be different with pre_pcpu, we
10961                  * need to set pre_pcpu as the destination of wakeup
10962                  * notification event, then we can find the right vCPU
10963                  * to wakeup in wakeup handler if interrupts happen
10964                  * when the vCPU is in blocked state.
10965                  */
10966                 dest = cpu_physical_id(vcpu->pre_pcpu);
10967
10968                 if (x2apic_enabled())
10969                         new.ndst = dest;
10970                 else
10971                         new.ndst = (dest << 8) & 0xFF00;
10972
10973                 /* set 'NV' to 'wakeup vector' */
10974                 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10975         } while (cmpxchg(&pi_desc->control, old.control,
10976                         new.control) != old.control);
10977
10978         return 0;
10979 }
10980
10981 static int vmx_pre_block(struct kvm_vcpu *vcpu)
10982 {
10983         if (pi_pre_block(vcpu))
10984                 return 1;
10985
10986         if (kvm_lapic_hv_timer_in_use(vcpu))
10987                 kvm_lapic_switch_to_sw_timer(vcpu);
10988
10989         return 0;
10990 }
10991
10992 static void pi_post_block(struct kvm_vcpu *vcpu)
10993 {
10994         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10995         struct pi_desc old, new;
10996         unsigned int dest;
10997         unsigned long flags;
10998
10999         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11000                 !irq_remapping_cap(IRQ_POSTING_CAP))
11001                 return;
11002
11003         do {
11004                 old.control = new.control = pi_desc->control;
11005
11006                 dest = cpu_physical_id(vcpu->cpu);
11007
11008                 if (x2apic_enabled())
11009                         new.ndst = dest;
11010                 else
11011                         new.ndst = (dest << 8) & 0xFF00;
11012
11013                 /* Allow posting non-urgent interrupts */
11014                 new.sn = 0;
11015
11016                 /* set 'NV' to 'notification vector' */
11017                 new.nv = POSTED_INTR_VECTOR;
11018         } while (cmpxchg(&pi_desc->control, old.control,
11019                         new.control) != old.control);
11020
11021         if(vcpu->pre_pcpu != -1) {
11022                 spin_lock_irqsave(
11023                         &per_cpu(blocked_vcpu_on_cpu_lock,
11024                         vcpu->pre_pcpu), flags);
11025                 list_del(&vcpu->blocked_vcpu_list);
11026                 spin_unlock_irqrestore(
11027                         &per_cpu(blocked_vcpu_on_cpu_lock,
11028                         vcpu->pre_pcpu), flags);
11029                 vcpu->pre_pcpu = -1;
11030         }
11031 }
11032
11033 static void vmx_post_block(struct kvm_vcpu *vcpu)
11034 {
11035         if (kvm_x86_ops->set_hv_timer)
11036                 kvm_lapic_switch_to_hv_timer(vcpu);
11037
11038         pi_post_block(vcpu);
11039 }
11040
11041 /*
11042  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11043  *
11044  * @kvm: kvm
11045  * @host_irq: host irq of the interrupt
11046  * @guest_irq: gsi of the interrupt
11047  * @set: set or unset PI
11048  * returns 0 on success, < 0 on failure
11049  */
11050 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11051                               uint32_t guest_irq, bool set)
11052 {
11053         struct kvm_kernel_irq_routing_entry *e;
11054         struct kvm_irq_routing_table *irq_rt;
11055         struct kvm_lapic_irq irq;
11056         struct kvm_vcpu *vcpu;
11057         struct vcpu_data vcpu_info;
11058         int idx, ret = -EINVAL;
11059
11060         if (!kvm_arch_has_assigned_device(kvm) ||
11061                 !irq_remapping_cap(IRQ_POSTING_CAP))
11062                 return 0;
11063
11064         idx = srcu_read_lock(&kvm->irq_srcu);
11065         irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11066         BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11067
11068         hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11069                 if (e->type != KVM_IRQ_ROUTING_MSI)
11070                         continue;
11071                 /*
11072                  * VT-d PI cannot support posting multicast/broadcast
11073                  * interrupts to a vCPU, we still use interrupt remapping
11074                  * for these kind of interrupts.
11075                  *
11076                  * For lowest-priority interrupts, we only support
11077                  * those with single CPU as the destination, e.g. user
11078                  * configures the interrupts via /proc/irq or uses
11079                  * irqbalance to make the interrupts single-CPU.
11080                  *
11081                  * We will support full lowest-priority interrupt later.
11082                  */
11083
11084                 kvm_set_msi_irq(e, &irq);
11085                 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11086                         /*
11087                          * Make sure the IRTE is in remapped mode if
11088                          * we don't handle it in posted mode.
11089                          */
11090                         ret = irq_set_vcpu_affinity(host_irq, NULL);
11091                         if (ret < 0) {
11092                                 printk(KERN_INFO
11093                                    "failed to back to remapped mode, irq: %u\n",
11094                                    host_irq);
11095                                 goto out;
11096                         }
11097
11098                         continue;
11099                 }
11100
11101                 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11102                 vcpu_info.vector = irq.vector;
11103
11104                 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11105                                 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11106
11107                 if (set)
11108                         ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11109                 else {
11110                         /* suppress notification event before unposting */
11111                         pi_set_sn(vcpu_to_pi_desc(vcpu));
11112                         ret = irq_set_vcpu_affinity(host_irq, NULL);
11113                         pi_clear_sn(vcpu_to_pi_desc(vcpu));
11114                 }
11115
11116                 if (ret < 0) {
11117                         printk(KERN_INFO "%s: failed to update PI IRTE\n",
11118                                         __func__);
11119                         goto out;
11120                 }
11121         }
11122
11123         ret = 0;
11124 out:
11125         srcu_read_unlock(&kvm->irq_srcu, idx);
11126         return ret;
11127 }
11128
11129 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11130 {
11131         if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11132                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11133                         FEATURE_CONTROL_LMCE;
11134         else
11135                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11136                         ~FEATURE_CONTROL_LMCE;
11137 }
11138
11139 static struct kvm_x86_ops vmx_x86_ops = {
11140         .cpu_has_kvm_support = cpu_has_kvm_support,
11141         .disabled_by_bios = vmx_disabled_by_bios,
11142         .hardware_setup = hardware_setup,
11143         .hardware_unsetup = hardware_unsetup,
11144         .check_processor_compatibility = vmx_check_processor_compat,
11145         .hardware_enable = hardware_enable,
11146         .hardware_disable = hardware_disable,
11147         .cpu_has_accelerated_tpr = report_flexpriority,
11148         .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11149
11150         .vcpu_create = vmx_create_vcpu,
11151         .vcpu_free = vmx_free_vcpu,
11152         .vcpu_reset = vmx_vcpu_reset,
11153
11154         .prepare_guest_switch = vmx_save_host_state,
11155         .vcpu_load = vmx_vcpu_load,
11156         .vcpu_put = vmx_vcpu_put,
11157
11158         .update_bp_intercept = update_exception_bitmap,
11159         .get_msr = vmx_get_msr,
11160         .set_msr = vmx_set_msr,
11161         .get_segment_base = vmx_get_segment_base,
11162         .get_segment = vmx_get_segment,
11163         .set_segment = vmx_set_segment,
11164         .get_cpl = vmx_get_cpl,
11165         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11166         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11167         .decache_cr3 = vmx_decache_cr3,
11168         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11169         .set_cr0 = vmx_set_cr0,
11170         .set_cr3 = vmx_set_cr3,
11171         .set_cr4 = vmx_set_cr4,
11172         .set_efer = vmx_set_efer,
11173         .get_idt = vmx_get_idt,
11174         .set_idt = vmx_set_idt,
11175         .get_gdt = vmx_get_gdt,
11176         .set_gdt = vmx_set_gdt,
11177         .get_dr6 = vmx_get_dr6,
11178         .set_dr6 = vmx_set_dr6,
11179         .set_dr7 = vmx_set_dr7,
11180         .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11181         .cache_reg = vmx_cache_reg,
11182         .get_rflags = vmx_get_rflags,
11183         .set_rflags = vmx_set_rflags,
11184
11185         .get_pkru = vmx_get_pkru,
11186
11187         .fpu_activate = vmx_fpu_activate,
11188         .fpu_deactivate = vmx_fpu_deactivate,
11189
11190         .tlb_flush = vmx_flush_tlb,
11191
11192         .run = vmx_vcpu_run,
11193         .handle_exit = vmx_handle_exit,
11194         .skip_emulated_instruction = skip_emulated_instruction,
11195         .set_interrupt_shadow = vmx_set_interrupt_shadow,
11196         .get_interrupt_shadow = vmx_get_interrupt_shadow,
11197         .patch_hypercall = vmx_patch_hypercall,
11198         .set_irq = vmx_inject_irq,
11199         .set_nmi = vmx_inject_nmi,
11200         .queue_exception = vmx_queue_exception,
11201         .cancel_injection = vmx_cancel_injection,
11202         .interrupt_allowed = vmx_interrupt_allowed,
11203         .nmi_allowed = vmx_nmi_allowed,
11204         .get_nmi_mask = vmx_get_nmi_mask,
11205         .set_nmi_mask = vmx_set_nmi_mask,
11206         .enable_nmi_window = enable_nmi_window,
11207         .enable_irq_window = enable_irq_window,
11208         .update_cr8_intercept = update_cr8_intercept,
11209         .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11210         .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11211         .get_enable_apicv = vmx_get_enable_apicv,
11212         .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11213         .load_eoi_exitmap = vmx_load_eoi_exitmap,
11214         .hwapic_irr_update = vmx_hwapic_irr_update,
11215         .hwapic_isr_update = vmx_hwapic_isr_update,
11216         .sync_pir_to_irr = vmx_sync_pir_to_irr,
11217         .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11218
11219         .set_tss_addr = vmx_set_tss_addr,
11220         .get_tdp_level = get_ept_level,
11221         .get_mt_mask = vmx_get_mt_mask,
11222
11223         .get_exit_info = vmx_get_exit_info,
11224
11225         .get_lpage_level = vmx_get_lpage_level,
11226
11227         .cpuid_update = vmx_cpuid_update,
11228
11229         .rdtscp_supported = vmx_rdtscp_supported,
11230         .invpcid_supported = vmx_invpcid_supported,
11231
11232         .set_supported_cpuid = vmx_set_supported_cpuid,
11233
11234         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11235
11236         .read_tsc_offset = vmx_read_tsc_offset,
11237         .write_tsc_offset = vmx_write_tsc_offset,
11238         .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
11239         .read_l1_tsc = vmx_read_l1_tsc,
11240
11241         .set_tdp_cr3 = vmx_set_cr3,
11242
11243         .check_intercept = vmx_check_intercept,
11244         .handle_external_intr = vmx_handle_external_intr,
11245         .mpx_supported = vmx_mpx_supported,
11246         .xsaves_supported = vmx_xsaves_supported,
11247
11248         .check_nested_events = vmx_check_nested_events,
11249
11250         .sched_in = vmx_sched_in,
11251
11252         .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11253         .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11254         .flush_log_dirty = vmx_flush_log_dirty,
11255         .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11256
11257         .pre_block = vmx_pre_block,
11258         .post_block = vmx_post_block,
11259
11260         .pmu_ops = &intel_pmu_ops,
11261
11262         .update_pi_irte = vmx_update_pi_irte,
11263
11264 #ifdef CONFIG_X86_64
11265         .set_hv_timer = vmx_set_hv_timer,
11266         .cancel_hv_timer = vmx_cancel_hv_timer,
11267 #endif
11268
11269         .setup_mce = vmx_setup_mce,
11270 };
11271
11272 static int __init vmx_init(void)
11273 {
11274         int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11275                      __alignof__(struct vcpu_vmx), THIS_MODULE);
11276         if (r)
11277                 return r;
11278
11279 #ifdef CONFIG_KEXEC_CORE
11280         rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11281                            crash_vmclear_local_loaded_vmcss);
11282 #endif
11283
11284         return 0;
11285 }
11286
11287 static void __exit vmx_exit(void)
11288 {
11289 #ifdef CONFIG_KEXEC_CORE
11290         RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11291         synchronize_rcu();
11292 #endif
11293
11294         kvm_exit();
11295 }
11296
11297 module_init(vmx_init)
11298 module_exit(vmx_exit)