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   <[email protected]>
  12  *   Yaniv Kamay  <[email protected]>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "irq.h"
  20 #include "mmu.h"
  21 #include "cpuid.h"
  22 #include "lapic.h"
  23
  24 #include <linux/kvm_host.h>
  25 #include <linux/module.h>
  26 #include <linux/kernel.h>
  27 #include <linux/mm.h>
  28 #include <linux/highmem.h>
  29 #include <linux/sched.h>
  30 #include <linux/moduleparam.h>
  31 #include <linux/mod_devicetable.h>
  32 #include <linux/trace_events.h>
  33 #include <linux/slab.h>
  34 #include <linux/tboot.h>
  35 #include <linux/hrtimer.h>
  36 #include <linux/frame.h>
  37 #include "kvm_cache_regs.h"
  38 #include "x86.h"
  39
  40 #include <asm/cpu.h>
  41 #include <asm/io.h>
  42 #include <asm/desc.h>
  43 #include <asm/vmx.h>
  44 #include <asm/virtext.h>
  45 #include <asm/mce.h>
  46 #include <asm/fpu/internal.h>
  47 #include <asm/perf_event.h>
  48 #include <asm/debugreg.h>
  49 #include <asm/kexec.h>
  50 #include <asm/apic.h>
  51 #include <asm/irq_remapping.h>
  52 #include <asm/mmu_context.h>
  53
  54 #include "trace.h"
  55 #include "pmu.h"
  56
  57 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  58 #define __ex_clear(x, reg) \
  59         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
  60
  61 MODULE_AUTHOR("Qumranet");
  62 MODULE_LICENSE("GPL");
  63
  64 static const struct x86_cpu_id vmx_cpu_id[] = {
  65         X86_FEATURE_MATCH(X86_FEATURE_VMX),
  66         {}
  67 };
  68 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
  69
  70 static bool __read_mostly enable_vpid = 1;
  71 module_param_named(vpid, enable_vpid, bool, 0444);
  72
  73 static bool __read_mostly enable_vnmi = 1;
  74 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
  75
  76 static bool __read_mostly flexpriority_enabled = 1;
  77 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
  78
  79 static bool __read_mostly enable_ept = 1;
  80 module_param_named(ept, enable_ept, bool, S_IRUGO);
  81
  82 static bool __read_mostly enable_unrestricted_guest = 1;
  83 module_param_named(unrestricted_guest,
  84                         enable_unrestricted_guest, bool, S_IRUGO);
  85
  86 static bool __read_mostly enable_ept_ad_bits = 1;
  87 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
  88
  89 static bool __read_mostly emulate_invalid_guest_state = true;
  90 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
  91
  92 static bool __read_mostly fasteoi = 1;
  93 module_param(fasteoi, bool, S_IRUGO);
  94
  95 static bool __read_mostly enable_apicv = 1;
  96 module_param(enable_apicv, bool, S_IRUGO);
  97
  98 static bool __read_mostly enable_shadow_vmcs = 1;
  99 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
 100 /*
 101  * If nested=1, nested virtualization is supported, i.e., guests may use
 102  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
 103  * use VMX instructions.
 104  */
 105 static bool __read_mostly nested = 0;
 106 module_param(nested, bool, S_IRUGO);
 107
 108 static u64 __read_mostly host_xss;
 109
 110 static bool __read_mostly enable_pml = 1;
 111 module_param_named(pml, enable_pml, bool, S_IRUGO);
 112
 113 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
 114
 115 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
 116 static int __read_mostly cpu_preemption_timer_multi;
 117 static bool __read_mostly enable_preemption_timer = 1;
 118 #ifdef CONFIG_X86_64
 119 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
 120 #endif
 121
 122 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
 123 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
 124 #define KVM_VM_CR0_ALWAYS_ON                                            \
 125         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
 126 #define KVM_CR4_GUEST_OWNED_BITS                                      \
 127         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
 128          | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
 129
 130 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
 131 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
 132
 133 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
 134
 135 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
 136
 137 /*
 138  * Hyper-V requires all of these, so mark them as supported even though
 139  * they are just treated the same as all-context.
 140  */
 141 #define VMX_VPID_EXTENT_SUPPORTED_MASK          \
 142         (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |  \
 143         VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |    \
 144         VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |    \
 145         VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
 146
 147 /*
 148  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 149  * ple_gap:    upper bound on the amount of time between two successive
 150  *             executions of PAUSE in a loop. Also indicate if ple enabled.
 151  *             According to test, this time is usually smaller than 128 cycles.
 152  * ple_window: upper bound on the amount of time a guest is allowed to execute
 153  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
 154  *             less than 2^12 cycles
 155  * Time is measured based on a counter that runs at the same rate as the TSC,
 156  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 157  */
 158 #define KVM_VMX_DEFAULT_PLE_GAP           128
 159 #define KVM_VMX_DEFAULT_PLE_WINDOW        4096
 160 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW   2
 161 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
 162 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX    \
 163                 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
 164
 165 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 166 module_param(ple_gap, int, S_IRUGO);
 167
 168 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 169 module_param(ple_window, int, S_IRUGO);
 170
 171 /* Default doubles per-vcpu window every exit. */
 172 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
 173 module_param(ple_window_grow, int, S_IRUGO);
 174
 175 /* Default resets per-vcpu window every exit to ple_window. */
 176 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
 177 module_param(ple_window_shrink, int, S_IRUGO);
 178
 179 /* Default is to compute the maximum so we can never overflow. */
 180 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
 181 static int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
 182 module_param(ple_window_max, int, S_IRUGO);
 183
 184 extern const ulong vmx_return;
 185
 186 #define NR_AUTOLOAD_MSRS 8
 187 #define VMCS02_POOL_SIZE 1
 188
 189 struct vmcs {
 190         u32 revision_id;
 191         u32 abort;
 192         char data[0];
 193 };
 194
 195 /*
 196  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
 197  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
 198  * loaded on this CPU (so we can clear them if the CPU goes down).
 199  */
 200 struct loaded_vmcs {
 201         struct vmcs *vmcs;
 202         struct vmcs *shadow_vmcs;
 203         int cpu;
 204         bool launched;
 205         bool nmi_known_unmasked;
 206         unsigned long vmcs_host_cr3;    /* May not match real cr3 */
 207         unsigned long vmcs_host_cr4;    /* May not match real cr4 */
 208         /* Support for vnmi-less CPUs */
 209         int soft_vnmi_blocked;
 210         ktime_t entry_time;
 211         s64 vnmi_blocked_time;
 212         struct list_head loaded_vmcss_on_cpu_link;
 213 };
 214
 215 struct shared_msr_entry {
 216         unsigned index;
 217         u64 data;
 218         u64 mask;
 219 };
 220
 221 /*
 222  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
 223  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
 224  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
 225  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
 226  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
 227  * More than one of these structures may exist, if L1 runs multiple L2 guests.
 228  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
 229  * underlying hardware which will be used to run L2.
 230  * This structure is packed to ensure that its layout is identical across
 231  * machines (necessary for live migration).
 232  * If there are changes in this struct, VMCS12_REVISION must be changed.
 233  */
 234 typedef u64 natural_width;
 235 struct __packed vmcs12 {
 236         /* According to the Intel spec, a VMCS region must start with the
 237          * following two fields. Then follow implementation-specific data.
 238          */
 239         u32 revision_id;
 240         u32 abort;
 241
 242         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
 243         u32 padding[7]; /* room for future expansion */
 244
 245         u64 io_bitmap_a;
 246         u64 io_bitmap_b;
 247         u64 msr_bitmap;
 248         u64 vm_exit_msr_store_addr;
 249         u64 vm_exit_msr_load_addr;
 250         u64 vm_entry_msr_load_addr;
 251         u64 tsc_offset;
 252         u64 virtual_apic_page_addr;
 253         u64 apic_access_addr;
 254         u64 posted_intr_desc_addr;
 255         u64 vm_function_control;
 256         u64 ept_pointer;
 257         u64 eoi_exit_bitmap0;
 258         u64 eoi_exit_bitmap1;
 259         u64 eoi_exit_bitmap2;
 260         u64 eoi_exit_bitmap3;
 261         u64 eptp_list_address;
 262         u64 xss_exit_bitmap;
 263         u64 guest_physical_address;
 264         u64 vmcs_link_pointer;
 265         u64 pml_address;
 266         u64 guest_ia32_debugctl;
 267         u64 guest_ia32_pat;
 268         u64 guest_ia32_efer;
 269         u64 guest_ia32_perf_global_ctrl;
 270         u64 guest_pdptr0;
 271         u64 guest_pdptr1;
 272         u64 guest_pdptr2;
 273         u64 guest_pdptr3;
 274         u64 guest_bndcfgs;
 275         u64 host_ia32_pat;
 276         u64 host_ia32_efer;
 277         u64 host_ia32_perf_global_ctrl;
 278         u64 padding64[8]; /* room for future expansion */
 279         /*
 280          * To allow migration of L1 (complete with its L2 guests) between
 281          * machines of different natural widths (32 or 64 bit), we cannot have
 282          * unsigned long fields with no explict size. We use u64 (aliased
 283          * natural_width) instead. Luckily, x86 is little-endian.
 284          */
 285         natural_width cr0_guest_host_mask;
 286         natural_width cr4_guest_host_mask;
 287         natural_width cr0_read_shadow;
 288         natural_width cr4_read_shadow;
 289         natural_width cr3_target_value0;
 290         natural_width cr3_target_value1;
 291         natural_width cr3_target_value2;
 292         natural_width cr3_target_value3;
 293         natural_width exit_qualification;
 294         natural_width guest_linear_address;
 295         natural_width guest_cr0;
 296         natural_width guest_cr3;
 297         natural_width guest_cr4;
 298         natural_width guest_es_base;
 299         natural_width guest_cs_base;
 300         natural_width guest_ss_base;
 301         natural_width guest_ds_base;
 302         natural_width guest_fs_base;
 303         natural_width guest_gs_base;
 304         natural_width guest_ldtr_base;
 305         natural_width guest_tr_base;
 306         natural_width guest_gdtr_base;
 307         natural_width guest_idtr_base;
 308         natural_width guest_dr7;
 309         natural_width guest_rsp;
 310         natural_width guest_rip;
 311         natural_width guest_rflags;
 312         natural_width guest_pending_dbg_exceptions;
 313         natural_width guest_sysenter_esp;
 314         natural_width guest_sysenter_eip;
 315         natural_width host_cr0;
 316         natural_width host_cr3;
 317         natural_width host_cr4;
 318         natural_width host_fs_base;
 319         natural_width host_gs_base;
 320         natural_width host_tr_base;
 321         natural_width host_gdtr_base;
 322         natural_width host_idtr_base;
 323         natural_width host_ia32_sysenter_esp;
 324         natural_width host_ia32_sysenter_eip;
 325         natural_width host_rsp;
 326         natural_width host_rip;
 327         natural_width paddingl[8]; /* room for future expansion */
 328         u32 pin_based_vm_exec_control;
 329         u32 cpu_based_vm_exec_control;
 330         u32 exception_bitmap;
 331         u32 page_fault_error_code_mask;
 332         u32 page_fault_error_code_match;
 333         u32 cr3_target_count;
 334         u32 vm_exit_controls;
 335         u32 vm_exit_msr_store_count;
 336         u32 vm_exit_msr_load_count;
 337         u32 vm_entry_controls;
 338         u32 vm_entry_msr_load_count;
 339         u32 vm_entry_intr_info_field;
 340         u32 vm_entry_exception_error_code;
 341         u32 vm_entry_instruction_len;
 342         u32 tpr_threshold;
 343         u32 secondary_vm_exec_control;
 344         u32 vm_instruction_error;
 345         u32 vm_exit_reason;
 346         u32 vm_exit_intr_info;
 347         u32 vm_exit_intr_error_code;
 348         u32 idt_vectoring_info_field;
 349         u32 idt_vectoring_error_code;
 350         u32 vm_exit_instruction_len;
 351         u32 vmx_instruction_info;
 352         u32 guest_es_limit;
 353         u32 guest_cs_limit;
 354         u32 guest_ss_limit;
 355         u32 guest_ds_limit;
 356         u32 guest_fs_limit;
 357         u32 guest_gs_limit;
 358         u32 guest_ldtr_limit;
 359         u32 guest_tr_limit;
 360         u32 guest_gdtr_limit;
 361         u32 guest_idtr_limit;
 362         u32 guest_es_ar_bytes;
 363         u32 guest_cs_ar_bytes;
 364         u32 guest_ss_ar_bytes;
 365         u32 guest_ds_ar_bytes;
 366         u32 guest_fs_ar_bytes;
 367         u32 guest_gs_ar_bytes;
 368         u32 guest_ldtr_ar_bytes;
 369         u32 guest_tr_ar_bytes;
 370         u32 guest_interruptibility_info;
 371         u32 guest_activity_state;
 372         u32 guest_sysenter_cs;
 373         u32 host_ia32_sysenter_cs;
 374         u32 vmx_preemption_timer_value;
 375         u32 padding32[7]; /* room for future expansion */
 376         u16 virtual_processor_id;
 377         u16 posted_intr_nv;
 378         u16 guest_es_selector;
 379         u16 guest_cs_selector;
 380         u16 guest_ss_selector;
 381         u16 guest_ds_selector;
 382         u16 guest_fs_selector;
 383         u16 guest_gs_selector;
 384         u16 guest_ldtr_selector;
 385         u16 guest_tr_selector;
 386         u16 guest_intr_status;
 387         u16 guest_pml_index;
 388         u16 host_es_selector;
 389         u16 host_cs_selector;
 390         u16 host_ss_selector;
 391         u16 host_ds_selector;
 392         u16 host_fs_selector;
 393         u16 host_gs_selector;
 394         u16 host_tr_selector;
 395 };
 396
 397 /*
 398  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
 399  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
 400  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
 401  */
 402 #define VMCS12_REVISION 0x11e57ed0
 403
 404 /*
 405  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
 406  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
 407  * current implementation, 4K are reserved to avoid future complications.
 408  */
 409 #define VMCS12_SIZE 0x1000
 410
 411 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
 412 struct vmcs02_list {
 413         struct list_head list;
 414         gpa_t vmptr;
 415         struct loaded_vmcs vmcs02;
 416 };
 417
 418 /*
 419  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 420  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 421  */
 422 struct nested_vmx {
 423         /* Has the level1 guest done vmxon? */
 424         bool vmxon;
 425         gpa_t vmxon_ptr;
 426         bool pml_full;
 427
 428         /* The guest-physical address of the current VMCS L1 keeps for L2 */
 429         gpa_t current_vmptr;
 430         /*
 431          * Cache of the guest's VMCS, existing outside of guest memory.
 432          * Loaded from guest memory during VMPTRLD. Flushed to guest
 433          * memory during VMCLEAR and VMPTRLD.
 434          */
 435         struct vmcs12 *cached_vmcs12;
 436         /*
 437          * Indicates if the shadow vmcs must be updated with the
 438          * data hold by vmcs12
 439          */
 440         bool sync_shadow_vmcs;
 441
 442         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
 443         struct list_head vmcs02_pool;
 444         int vmcs02_num;
 445         bool change_vmcs01_virtual_x2apic_mode;
 446         /* L2 must run next, and mustn't decide to exit to L1. */
 447         bool nested_run_pending;
 448         /*
 449          * Guest pages referred to in vmcs02 with host-physical pointers, so
 450          * we must keep them pinned while L2 runs.
 451          */
 452         struct page *apic_access_page;
 453         struct page *virtual_apic_page;
 454         struct page *pi_desc_page;
 455         struct pi_desc *pi_desc;
 456         bool pi_pending;
 457         u16 posted_intr_nv;
 458
 459         unsigned long *msr_bitmap;
 460
 461         struct hrtimer preemption_timer;
 462         bool preemption_timer_expired;
 463
 464         /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
 465         u64 vmcs01_debugctl;
 466
 467         u16 vpid02;
 468         u16 last_vpid;
 469
 470         /*
 471          * We only store the "true" versions of the VMX capability MSRs. We
 472          * generate the "non-true" versions by setting the must-be-1 bits
 473          * according to the SDM.
 474          */
 475         u32 nested_vmx_procbased_ctls_low;
 476         u32 nested_vmx_procbased_ctls_high;
 477         u32 nested_vmx_secondary_ctls_low;
 478         u32 nested_vmx_secondary_ctls_high;
 479         u32 nested_vmx_pinbased_ctls_low;
 480         u32 nested_vmx_pinbased_ctls_high;
 481         u32 nested_vmx_exit_ctls_low;
 482         u32 nested_vmx_exit_ctls_high;
 483         u32 nested_vmx_entry_ctls_low;
 484         u32 nested_vmx_entry_ctls_high;
 485         u32 nested_vmx_misc_low;
 486         u32 nested_vmx_misc_high;
 487         u32 nested_vmx_ept_caps;
 488         u32 nested_vmx_vpid_caps;
 489         u64 nested_vmx_basic;
 490         u64 nested_vmx_cr0_fixed0;
 491         u64 nested_vmx_cr0_fixed1;
 492         u64 nested_vmx_cr4_fixed0;
 493         u64 nested_vmx_cr4_fixed1;
 494         u64 nested_vmx_vmcs_enum;
 495         u64 nested_vmx_vmfunc_controls;
 496
 497         /* SMM related state */
 498         struct {
 499                 /* in VMX operation on SMM entry? */
 500                 bool vmxon;
 501                 /* in guest mode on SMM entry? */
 502                 bool guest_mode;
 503         } smm;
 504 };
 505
 506 #define POSTED_INTR_ON  0
 507 #define POSTED_INTR_SN  1
 508
 509 /* Posted-Interrupt Descriptor */
 510 struct pi_desc {
 511         u32 pir[8];     /* Posted interrupt requested */
 512         union {
 513                 struct {
 514                                 /* bit 256 - Outstanding Notification */
 515                         u16     on      : 1,
 516                                 /* bit 257 - Suppress Notification */
 517                                 sn      : 1,
 518                                 /* bit 271:258 - Reserved */
 519                                 rsvd_1  : 14;
 520                                 /* bit 279:272 - Notification Vector */
 521                         u8      nv;
 522                                 /* bit 287:280 - Reserved */
 523                         u8      rsvd_2;
 524                                 /* bit 319:288 - Notification Destination */
 525                         u32     ndst;
 526                 };
 527                 u64 control;
 528         };
 529         u32 rsvd[6];
 530 } __aligned(64);
 531
 532 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
 533 {
 534         return test_and_set_bit(POSTED_INTR_ON,
 535                         (unsigned long *)&pi_desc->control);
 536 }
 537
 538 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
 539 {
 540         return test_and_clear_bit(POSTED_INTR_ON,
 541                         (unsigned long *)&pi_desc->control);
 542 }
 543
 544 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
 545 {
 546         return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
 547 }
 548
 549 static inline void pi_clear_sn(struct pi_desc *pi_desc)
 550 {
 551         return clear_bit(POSTED_INTR_SN,
 552                         (unsigned long *)&pi_desc->control);
 553 }
 554
 555 static inline void pi_set_sn(struct pi_desc *pi_desc)
 556 {
 557         return set_bit(POSTED_INTR_SN,
 558                         (unsigned long *)&pi_desc->control);
 559 }
 560
 561 static inline void pi_clear_on(struct pi_desc *pi_desc)
 562 {
 563         clear_bit(POSTED_INTR_ON,
 564                   (unsigned long *)&pi_desc->control);
 565 }
 566
 567 static inline int pi_test_on(struct pi_desc *pi_desc)
 568 {
 569         return test_bit(POSTED_INTR_ON,
 570                         (unsigned long *)&pi_desc->control);
 571 }
 572
 573 static inline int pi_test_sn(struct pi_desc *pi_desc)
 574 {
 575         return test_bit(POSTED_INTR_SN,
 576                         (unsigned long *)&pi_desc->control);
 577 }
 578
 579 struct vcpu_vmx {
 580         struct kvm_vcpu       vcpu;
 581         unsigned long         host_rsp;
 582         u8                    fail;
 583         u32                   exit_intr_info;
 584         u32                   idt_vectoring_info;
 585         ulong                 rflags;
 586         struct shared_msr_entry *guest_msrs;
 587         int                   nmsrs;
 588         int                   save_nmsrs;
 589         unsigned long         host_idt_base;
 590 #ifdef CONFIG_X86_64
 591         u64                   msr_host_kernel_gs_base;
 592         u64                   msr_guest_kernel_gs_base;
 593 #endif
 594         u32 vm_entry_controls_shadow;
 595         u32 vm_exit_controls_shadow;
 596         u32 secondary_exec_control;
 597
 598         /*
 599          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 600          * non-nested (L1) guest, it always points to vmcs01. For a nested
 601          * guest (L2), it points to a different VMCS.
 602          */
 603         struct loaded_vmcs    vmcs01;
 604         struct loaded_vmcs   *loaded_vmcs;
 605         bool                  __launched; /* temporary, used in vmx_vcpu_run */
 606         struct msr_autoload {
 607                 unsigned nr;
 608                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 609                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 610         } msr_autoload;
 611         struct {
 612                 int           loaded;
 613                 u16           fs_sel, gs_sel, ldt_sel;
 614 #ifdef CONFIG_X86_64
 615                 u16           ds_sel, es_sel;
 616 #endif
 617                 int           gs_ldt_reload_needed;
 618                 int           fs_reload_needed;
 619                 u64           msr_host_bndcfgs;
 620         } host_state;
 621         struct {
 622                 int vm86_active;
 623                 ulong save_rflags;
 624                 struct kvm_segment segs[8];
 625         } rmode;
 626         struct {
 627                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 628                 struct kvm_save_segment {
 629                         u16 selector;
 630                         unsigned long base;
 631                         u32 limit;
 632                         u32 ar;
 633                 } seg[8];
 634         } segment_cache;
 635         int vpid;
 636         bool emulation_required;
 637
 638         u32 exit_reason;
 639
 640         /* Posted interrupt descriptor */
 641         struct pi_desc pi_desc;
 642
 643         /* Support for a guest hypervisor (nested VMX) */
 644         struct nested_vmx nested;
 645
 646         /* Dynamic PLE window. */
 647         int ple_window;
 648         bool ple_window_dirty;
 649
 650         /* Support for PML */
 651 #define PML_ENTITY_NUM          512
 652         struct page *pml_pg;
 653
 654         /* apic deadline value in host tsc */
 655         u64 hv_deadline_tsc;
 656
 657         u64 current_tsc_ratio;
 658
 659         u32 host_pkru;
 660
 661         /*
 662          * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
 663          * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
 664          * in msr_ia32_feature_control_valid_bits.
 665          */
 666         u64 msr_ia32_feature_control;
 667         u64 msr_ia32_feature_control_valid_bits;
 668 };
 669
 670 enum segment_cache_field {
 671         SEG_FIELD_SEL = 0,
 672         SEG_FIELD_BASE = 1,
 673         SEG_FIELD_LIMIT = 2,
 674         SEG_FIELD_AR = 3,
 675
 676         SEG_FIELD_NR = 4
 677 };
 678
 679 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 680 {
 681         return container_of(vcpu, struct vcpu_vmx, vcpu);
 682 }
 683
 684 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
 685 {
 686         return &(to_vmx(vcpu)->pi_desc);
 687 }
 688
 689 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
 690 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
 691 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
 692                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
 693
 694
 695 static unsigned long shadow_read_only_fields[] = {
 696         /*
 697          * We do NOT shadow fields that are modified when L0
 698          * traps and emulates any vmx instruction (e.g. VMPTRLD,
 699          * VMXON...) executed by L1.
 700          * For example, VM_INSTRUCTION_ERROR is read
 701          * by L1 if a vmx instruction fails (part of the error path).
 702          * Note the code assumes this logic. If for some reason
 703          * we start shadowing these fields then we need to
 704          * force a shadow sync when L0 emulates vmx instructions
 705          * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
 706          * by nested_vmx_failValid)
 707          */
 708         VM_EXIT_REASON,
 709         VM_EXIT_INTR_INFO,
 710         VM_EXIT_INSTRUCTION_LEN,
 711         IDT_VECTORING_INFO_FIELD,
 712         IDT_VECTORING_ERROR_CODE,
 713         VM_EXIT_INTR_ERROR_CODE,
 714         EXIT_QUALIFICATION,
 715         GUEST_LINEAR_ADDRESS,
 716         GUEST_PHYSICAL_ADDRESS
 717 };
 718 static int max_shadow_read_only_fields =
 719         ARRAY_SIZE(shadow_read_only_fields);
 720
 721 static unsigned long shadow_read_write_fields[] = {
 722         TPR_THRESHOLD,
 723         GUEST_RIP,
 724         GUEST_RSP,
 725         GUEST_CR0,
 726         GUEST_CR3,
 727         GUEST_CR4,
 728         GUEST_INTERRUPTIBILITY_INFO,
 729         GUEST_RFLAGS,
 730         GUEST_CS_SELECTOR,
 731         GUEST_CS_AR_BYTES,
 732         GUEST_CS_LIMIT,
 733         GUEST_CS_BASE,
 734         GUEST_ES_BASE,
 735         GUEST_BNDCFGS,
 736         CR0_GUEST_HOST_MASK,
 737         CR0_READ_SHADOW,
 738         CR4_READ_SHADOW,
 739         TSC_OFFSET,
 740         EXCEPTION_BITMAP,
 741         CPU_BASED_VM_EXEC_CONTROL,
 742         VM_ENTRY_EXCEPTION_ERROR_CODE,
 743         VM_ENTRY_INTR_INFO_FIELD,
 744         VM_ENTRY_INSTRUCTION_LEN,
 745         VM_ENTRY_EXCEPTION_ERROR_CODE,
 746         HOST_FS_BASE,
 747         HOST_GS_BASE,
 748         HOST_FS_SELECTOR,
 749         HOST_GS_SELECTOR
 750 };
 751 static int max_shadow_read_write_fields =
 752         ARRAY_SIZE(shadow_read_write_fields);
 753
 754 static const unsigned short vmcs_field_to_offset_table[] = {
 755         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 756         FIELD(POSTED_INTR_NV, posted_intr_nv),
 757         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
 758         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
 759         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
 760         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
 761         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
 762         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
 763         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
 764         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
 765         FIELD(GUEST_INTR_STATUS, guest_intr_status),
 766         FIELD(GUEST_PML_INDEX, guest_pml_index),
 767         FIELD(HOST_ES_SELECTOR, host_es_selector),
 768         FIELD(HOST_CS_SELECTOR, host_cs_selector),
 769         FIELD(HOST_SS_SELECTOR, host_ss_selector),
 770         FIELD(HOST_DS_SELECTOR, host_ds_selector),
 771         FIELD(HOST_FS_SELECTOR, host_fs_selector),
 772         FIELD(HOST_GS_SELECTOR, host_gs_selector),
 773         FIELD(HOST_TR_SELECTOR, host_tr_selector),
 774         FIELD64(IO_BITMAP_A, io_bitmap_a),
 775         FIELD64(IO_BITMAP_B, io_bitmap_b),
 776         FIELD64(MSR_BITMAP, msr_bitmap),
 777         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
 778         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
 779         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
 780         FIELD64(TSC_OFFSET, tsc_offset),
 781         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
 782         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
 783         FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
 784         FIELD64(VM_FUNCTION_CONTROL, vm_function_control),
 785         FIELD64(EPT_POINTER, ept_pointer),
 786         FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
 787         FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
 788         FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
 789         FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
 790         FIELD64(EPTP_LIST_ADDRESS, eptp_list_address),
 791         FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
 792         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
 793         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
 794         FIELD64(PML_ADDRESS, pml_address),
 795         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
 796         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
 797         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
 798         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
 799         FIELD64(GUEST_PDPTR0, guest_pdptr0),
 800         FIELD64(GUEST_PDPTR1, guest_pdptr1),
 801         FIELD64(GUEST_PDPTR2, guest_pdptr2),
 802         FIELD64(GUEST_PDPTR3, guest_pdptr3),
 803         FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
 804         FIELD64(HOST_IA32_PAT, host_ia32_pat),
 805         FIELD64(HOST_IA32_EFER, host_ia32_efer),
 806         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
 807         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
 808         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
 809         FIELD(EXCEPTION_BITMAP, exception_bitmap),
 810         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
 811         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
 812         FIELD(CR3_TARGET_COUNT, cr3_target_count),
 813         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
 814         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
 815         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
 816         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
 817         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
 818         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
 819         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
 820         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
 821         FIELD(TPR_THRESHOLD, tpr_threshold),
 822         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
 823         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
 824         FIELD(VM_EXIT_REASON, vm_exit_reason),
 825         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
 826         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
 827         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
 828         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
 829         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
 830         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
 831         FIELD(GUEST_ES_LIMIT, guest_es_limit),
 832         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
 833         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
 834         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
 835         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
 836         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
 837         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
 838         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
 839         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
 840         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
 841         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
 842         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
 843         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
 844         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
 845         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
 846         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
 847         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
 848         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
 849         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
 850         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
 851         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
 852         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
 853         FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
 854         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
 855         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
 856         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
 857         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
 858         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
 859         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
 860         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
 861         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
 862         FIELD(EXIT_QUALIFICATION, exit_qualification),
 863         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
 864         FIELD(GUEST_CR0, guest_cr0),
 865         FIELD(GUEST_CR3, guest_cr3),
 866         FIELD(GUEST_CR4, guest_cr4),
 867         FIELD(GUEST_ES_BASE, guest_es_base),
 868         FIELD(GUEST_CS_BASE, guest_cs_base),
 869         FIELD(GUEST_SS_BASE, guest_ss_base),
 870         FIELD(GUEST_DS_BASE, guest_ds_base),
 871         FIELD(GUEST_FS_BASE, guest_fs_base),
 872         FIELD(GUEST_GS_BASE, guest_gs_base),
 873         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
 874         FIELD(GUEST_TR_BASE, guest_tr_base),
 875         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
 876         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
 877         FIELD(GUEST_DR7, guest_dr7),
 878         FIELD(GUEST_RSP, guest_rsp),
 879         FIELD(GUEST_RIP, guest_rip),
 880         FIELD(GUEST_RFLAGS, guest_rflags),
 881         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
 882         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
 883         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
 884         FIELD(HOST_CR0, host_cr0),
 885         FIELD(HOST_CR3, host_cr3),
 886         FIELD(HOST_CR4, host_cr4),
 887         FIELD(HOST_FS_BASE, host_fs_base),
 888         FIELD(HOST_GS_BASE, host_gs_base),
 889         FIELD(HOST_TR_BASE, host_tr_base),
 890         FIELD(HOST_GDTR_BASE, host_gdtr_base),
 891         FIELD(HOST_IDTR_BASE, host_idtr_base),
 892         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
 893         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
 894         FIELD(HOST_RSP, host_rsp),
 895         FIELD(HOST_RIP, host_rip),
 896 };
 897
 898 static inline short vmcs_field_to_offset(unsigned long field)
 899 {
 900         BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
 901
 902         if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
 903             vmcs_field_to_offset_table[field] == 0)
 904                 return -ENOENT;
 905
 906         return vmcs_field_to_offset_table[field];
 907 }
 908
 909 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
 910 {
 911         return to_vmx(vcpu)->nested.cached_vmcs12;
 912 }
 913
 914 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
 915 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
 916 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
 917 static bool vmx_xsaves_supported(void);
 918 static void vmx_set_segment(struct kvm_vcpu *vcpu,
 919                             struct kvm_segment *var, int seg);
 920 static void vmx_get_segment(struct kvm_vcpu *vcpu,
 921                             struct kvm_segment *var, int seg);
 922 static bool guest_state_valid(struct kvm_vcpu *vcpu);
 923 static u32 vmx_segment_access_rights(struct kvm_segment *var);
 924 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
 925 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
 926 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
 927 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
 928                                             u16 error_code);
 929
 930 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 931 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 932 /*
 933  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
 934  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
 935  */
 936 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
 937
 938 /*
 939  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
 940  * can find which vCPU should be waken up.
 941  */
 942 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
 943 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
 944
 945 enum {
 946         VMX_IO_BITMAP_A,
 947         VMX_IO_BITMAP_B,
 948         VMX_MSR_BITMAP_LEGACY,
 949         VMX_MSR_BITMAP_LONGMODE,
 950         VMX_MSR_BITMAP_LEGACY_X2APIC_APICV,
 951         VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV,
 952         VMX_MSR_BITMAP_LEGACY_X2APIC,
 953         VMX_MSR_BITMAP_LONGMODE_X2APIC,
 954         VMX_VMREAD_BITMAP,
 955         VMX_VMWRITE_BITMAP,
 956         VMX_BITMAP_NR
 957 };
 958
 959 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
 960
 961 #define vmx_io_bitmap_a                      (vmx_bitmap[VMX_IO_BITMAP_A])
 962 #define vmx_io_bitmap_b                      (vmx_bitmap[VMX_IO_BITMAP_B])
 963 #define vmx_msr_bitmap_legacy                (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
 964 #define vmx_msr_bitmap_longmode              (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
 965 #define vmx_msr_bitmap_legacy_x2apic_apicv   (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
 966 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
 967 #define vmx_msr_bitmap_legacy_x2apic         (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
 968 #define vmx_msr_bitmap_longmode_x2apic       (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
 969 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
 970 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
 971
 972 static bool cpu_has_load_ia32_efer;
 973 static bool cpu_has_load_perf_global_ctrl;
 974
 975 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 976 static DEFINE_SPINLOCK(vmx_vpid_lock);
 977
 978 static struct vmcs_config {
 979         int size;
 980         int order;
 981         u32 basic_cap;
 982         u32 revision_id;
 983         u32 pin_based_exec_ctrl;
 984         u32 cpu_based_exec_ctrl;
 985         u32 cpu_based_2nd_exec_ctrl;
 986         u32 vmexit_ctrl;
 987         u32 vmentry_ctrl;
 988 } vmcs_config;
 989
 990 static struct vmx_capability {
 991         u32 ept;
 992         u32 vpid;
 993 } vmx_capability;
 994
 995 #define VMX_SEGMENT_FIELD(seg)                                  \
 996         [VCPU_SREG_##seg] = {                                   \
 997                 .selector = GUEST_##seg##_SELECTOR,             \
 998                 .base = GUEST_##seg##_BASE,                     \
 999                 .limit = GUEST_##seg##_LIMIT,                   \
1000                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
1001         }
1002
1003 static const struct kvm_vmx_segment_field {
1004         unsigned selector;
1005         unsigned base;
1006         unsigned limit;
1007         unsigned ar_bytes;
1008 } kvm_vmx_segment_fields[] = {
1009         VMX_SEGMENT_FIELD(CS),
1010         VMX_SEGMENT_FIELD(DS),
1011         VMX_SEGMENT_FIELD(ES),
1012         VMX_SEGMENT_FIELD(FS),
1013         VMX_SEGMENT_FIELD(GS),
1014         VMX_SEGMENT_FIELD(SS),
1015         VMX_SEGMENT_FIELD(TR),
1016         VMX_SEGMENT_FIELD(LDTR),
1017 };
1018
1019 static u64 host_efer;
1020
1021 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1022
1023 /*
1024  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1025  * away by decrementing the array size.
1026  */
1027 static const u32 vmx_msr_index[] = {
1028 #ifdef CONFIG_X86_64
1029         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1030 #endif
1031         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1032 };
1033
1034 static inline bool is_exception_n(u32 intr_info, u8 vector)
1035 {
1036         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1037                              INTR_INFO_VALID_MASK)) ==
1038                 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1039 }
1040
1041 static inline bool is_debug(u32 intr_info)
1042 {
1043         return is_exception_n(intr_info, DB_VECTOR);
1044 }
1045
1046 static inline bool is_breakpoint(u32 intr_info)
1047 {
1048         return is_exception_n(intr_info, BP_VECTOR);
1049 }
1050
1051 static inline bool is_page_fault(u32 intr_info)
1052 {
1053         return is_exception_n(intr_info, PF_VECTOR);
1054 }
1055
1056 static inline bool is_no_device(u32 intr_info)
1057 {
1058         return is_exception_n(intr_info, NM_VECTOR);
1059 }
1060
1061 static inline bool is_invalid_opcode(u32 intr_info)
1062 {
1063         return is_exception_n(intr_info, UD_VECTOR);
1064 }
1065
1066 static inline bool is_external_interrupt(u32 intr_info)
1067 {
1068         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1069                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1070 }
1071
1072 static inline bool is_machine_check(u32 intr_info)
1073 {
1074         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1075                              INTR_INFO_VALID_MASK)) ==
1076                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1077 }
1078
1079 static inline bool cpu_has_vmx_msr_bitmap(void)
1080 {
1081         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1082 }
1083
1084 static inline bool cpu_has_vmx_tpr_shadow(void)
1085 {
1086         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1087 }
1088
1089 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1090 {
1091         return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1092 }
1093
1094 static inline bool cpu_has_secondary_exec_ctrls(void)
1095 {
1096         return vmcs_config.cpu_based_exec_ctrl &
1097                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1098 }
1099
1100 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1101 {
1102         return vmcs_config.cpu_based_2nd_exec_ctrl &
1103                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1104 }
1105
1106 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1107 {
1108         return vmcs_config.cpu_based_2nd_exec_ctrl &
1109                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1110 }
1111
1112 static inline bool cpu_has_vmx_apic_register_virt(void)
1113 {
1114         return vmcs_config.cpu_based_2nd_exec_ctrl &
1115                 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1116 }
1117
1118 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1119 {
1120         return vmcs_config.cpu_based_2nd_exec_ctrl &
1121                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1122 }
1123
1124 /*
1125  * Comment's format: document - errata name - stepping - processor name.
1126  * Refer from
1127  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1128  */
1129 static u32 vmx_preemption_cpu_tfms[] = {
1130 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
1131 0x000206E6,
1132 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
1133 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1134 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1135 0x00020652,
1136 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1137 0x00020655,
1138 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
1139 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
1140 /*
1141  * 320767.pdf - AAP86  - B1 -
1142  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1143  */
1144 0x000106E5,
1145 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1146 0x000106A0,
1147 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1148 0x000106A1,
1149 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1150 0x000106A4,
1151  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1152  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1153  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1154 0x000106A5,
1155 };
1156
1157 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1158 {
1159         u32 eax = cpuid_eax(0x00000001), i;
1160
1161         /* Clear the reserved bits */
1162         eax &= ~(0x3U << 14 | 0xfU << 28);
1163         for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1164                 if (eax == vmx_preemption_cpu_tfms[i])
1165                         return true;
1166
1167         return false;
1168 }
1169
1170 static inline bool cpu_has_vmx_preemption_timer(void)
1171 {
1172         return vmcs_config.pin_based_exec_ctrl &
1173                 PIN_BASED_VMX_PREEMPTION_TIMER;
1174 }
1175
1176 static inline bool cpu_has_vmx_posted_intr(void)
1177 {
1178         return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1179                 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1180 }
1181
1182 static inline bool cpu_has_vmx_apicv(void)
1183 {
1184         return cpu_has_vmx_apic_register_virt() &&
1185                 cpu_has_vmx_virtual_intr_delivery() &&
1186                 cpu_has_vmx_posted_intr();
1187 }
1188
1189 static inline bool cpu_has_vmx_flexpriority(void)
1190 {
1191         return cpu_has_vmx_tpr_shadow() &&
1192                 cpu_has_vmx_virtualize_apic_accesses();
1193 }
1194
1195 static inline bool cpu_has_vmx_ept_execute_only(void)
1196 {
1197         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1198 }
1199
1200 static inline bool cpu_has_vmx_ept_2m_page(void)
1201 {
1202         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1203 }
1204
1205 static inline bool cpu_has_vmx_ept_1g_page(void)
1206 {
1207         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1208 }
1209
1210 static inline bool cpu_has_vmx_ept_4levels(void)
1211 {
1212         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1213 }
1214
1215 static inline bool cpu_has_vmx_ept_mt_wb(void)
1216 {
1217         return vmx_capability.ept & VMX_EPTP_WB_BIT;
1218 }
1219
1220 static inline bool cpu_has_vmx_ept_5levels(void)
1221 {
1222         return vmx_capability.ept & VMX_EPT_PAGE_WALK_5_BIT;
1223 }
1224
1225 static inline bool cpu_has_vmx_ept_ad_bits(void)
1226 {
1227         return vmx_capability.ept & VMX_EPT_AD_BIT;
1228 }
1229
1230 static inline bool cpu_has_vmx_invept_context(void)
1231 {
1232         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1233 }
1234
1235 static inline bool cpu_has_vmx_invept_global(void)
1236 {
1237         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1238 }
1239
1240 static inline bool cpu_has_vmx_invvpid_single(void)
1241 {
1242         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1243 }
1244
1245 static inline bool cpu_has_vmx_invvpid_global(void)
1246 {
1247         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1248 }
1249
1250 static inline bool cpu_has_vmx_invvpid(void)
1251 {
1252         return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1253 }
1254
1255 static inline bool cpu_has_vmx_ept(void)
1256 {
1257         return vmcs_config.cpu_based_2nd_exec_ctrl &
1258                 SECONDARY_EXEC_ENABLE_EPT;
1259 }
1260
1261 static inline bool cpu_has_vmx_unrestricted_guest(void)
1262 {
1263         return vmcs_config.cpu_based_2nd_exec_ctrl &
1264                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1265 }
1266
1267 static inline bool cpu_has_vmx_ple(void)
1268 {
1269         return vmcs_config.cpu_based_2nd_exec_ctrl &
1270                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1271 }
1272
1273 static inline bool cpu_has_vmx_basic_inout(void)
1274 {
1275         return  (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1276 }
1277
1278 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1279 {
1280         return flexpriority_enabled && lapic_in_kernel(vcpu);
1281 }
1282
1283 static inline bool cpu_has_vmx_vpid(void)
1284 {
1285         return vmcs_config.cpu_based_2nd_exec_ctrl &
1286                 SECONDARY_EXEC_ENABLE_VPID;
1287 }
1288
1289 static inline bool cpu_has_vmx_rdtscp(void)
1290 {
1291         return vmcs_config.cpu_based_2nd_exec_ctrl &
1292                 SECONDARY_EXEC_RDTSCP;
1293 }
1294
1295 static inline bool cpu_has_vmx_invpcid(void)
1296 {
1297         return vmcs_config.cpu_based_2nd_exec_ctrl &
1298                 SECONDARY_EXEC_ENABLE_INVPCID;
1299 }
1300
1301 static inline bool cpu_has_virtual_nmis(void)
1302 {
1303         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1304 }
1305
1306 static inline bool cpu_has_vmx_wbinvd_exit(void)
1307 {
1308         return vmcs_config.cpu_based_2nd_exec_ctrl &
1309                 SECONDARY_EXEC_WBINVD_EXITING;
1310 }
1311
1312 static inline bool cpu_has_vmx_shadow_vmcs(void)
1313 {
1314         u64 vmx_msr;
1315         rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1316         /* check if the cpu supports writing r/o exit information fields */
1317         if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1318                 return false;
1319
1320         return vmcs_config.cpu_based_2nd_exec_ctrl &
1321                 SECONDARY_EXEC_SHADOW_VMCS;
1322 }
1323
1324 static inline bool cpu_has_vmx_pml(void)
1325 {
1326         return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1327 }
1328
1329 static inline bool cpu_has_vmx_tsc_scaling(void)
1330 {
1331         return vmcs_config.cpu_based_2nd_exec_ctrl &
1332                 SECONDARY_EXEC_TSC_SCALING;
1333 }
1334
1335 static inline bool cpu_has_vmx_vmfunc(void)
1336 {
1337         return vmcs_config.cpu_based_2nd_exec_ctrl &
1338                 SECONDARY_EXEC_ENABLE_VMFUNC;
1339 }
1340
1341 static inline bool report_flexpriority(void)
1342 {
1343         return flexpriority_enabled;
1344 }
1345
1346 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1347 {
1348         return vmx_misc_cr3_count(to_vmx(vcpu)->nested.nested_vmx_misc_low);
1349 }
1350
1351 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1352 {
1353         return vmcs12->cpu_based_vm_exec_control & bit;
1354 }
1355
1356 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1357 {
1358         return (vmcs12->cpu_based_vm_exec_control &
1359                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1360                 (vmcs12->secondary_vm_exec_control & bit);
1361 }
1362
1363 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1364 {
1365         return vmcs12->pin_based_vm_exec_control &
1366                 PIN_BASED_VMX_PREEMPTION_TIMER;
1367 }
1368
1369 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1370 {
1371         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1372 }
1373
1374 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1375 {
1376         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
1377 }
1378
1379 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1380 {
1381         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1382 }
1383
1384 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1385 {
1386         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1387 }
1388
1389 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1390 {
1391         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1392 }
1393
1394 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1395 {
1396         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1397 }
1398
1399 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1400 {
1401         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1402 }
1403
1404 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1405 {
1406         return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1407 }
1408
1409 static inline bool nested_cpu_has_vmfunc(struct vmcs12 *vmcs12)
1410 {
1411         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VMFUNC);
1412 }
1413
1414 static inline bool nested_cpu_has_eptp_switching(struct vmcs12 *vmcs12)
1415 {
1416         return nested_cpu_has_vmfunc(vmcs12) &&
1417                 (vmcs12->vm_function_control &
1418                  VMX_VMFUNC_EPTP_SWITCHING);
1419 }
1420
1421 static inline bool is_nmi(u32 intr_info)
1422 {
1423         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1424                 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1425 }
1426
1427 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1428                               u32 exit_intr_info,
1429                               unsigned long exit_qualification);
1430 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1431                         struct vmcs12 *vmcs12,
1432                         u32 reason, unsigned long qualification);
1433
1434 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1435 {
1436         int i;
1437
1438         for (i = 0; i < vmx->nmsrs; ++i)
1439                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1440                         return i;
1441         return -1;
1442 }
1443
1444 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1445 {
1446     struct {
1447         u64 vpid : 16;
1448         u64 rsvd : 48;
1449         u64 gva;
1450     } operand = { vpid, 0, gva };
1451
1452     asm volatile (__ex(ASM_VMX_INVVPID)
1453                   /* CF==1 or ZF==1 --> rc = -1 */
1454                   "; ja 1f ; ud2 ; 1:"
1455                   : : "a"(&operand), "c"(ext) : "cc", "memory");
1456 }
1457
1458 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1459 {
1460         struct {
1461                 u64 eptp, gpa;
1462         } operand = {eptp, gpa};
1463
1464         asm volatile (__ex(ASM_VMX_INVEPT)
1465                         /* CF==1 or ZF==1 --> rc = -1 */
1466                         "; ja 1f ; ud2 ; 1:\n"
1467                         : : "a" (&operand), "c" (ext) : "cc", "memory");
1468 }
1469
1470 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1471 {
1472         int i;
1473
1474         i = __find_msr_index(vmx, msr);
1475         if (i >= 0)
1476                 return &vmx->guest_msrs[i];
1477         return NULL;
1478 }
1479
1480 static void vmcs_clear(struct vmcs *vmcs)
1481 {
1482         u64 phys_addr = __pa(vmcs);
1483         u8 error;
1484
1485         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1486                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1487                       : "cc", "memory");
1488         if (error)
1489                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1490                        vmcs, phys_addr);
1491 }
1492
1493 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1494 {
1495         vmcs_clear(loaded_vmcs->vmcs);
1496         if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1497                 vmcs_clear(loaded_vmcs->shadow_vmcs);
1498         loaded_vmcs->cpu = -1;
1499         loaded_vmcs->launched = 0;
1500 }
1501
1502 static void vmcs_load(struct vmcs *vmcs)
1503 {
1504         u64 phys_addr = __pa(vmcs);
1505         u8 error;
1506
1507         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1508                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1509                         : "cc", "memory");
1510         if (error)
1511                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1512                        vmcs, phys_addr);
1513 }
1514
1515 #ifdef CONFIG_KEXEC_CORE
1516 /*
1517  * This bitmap is used to indicate whether the vmclear
1518  * operation is enabled on all cpus. All disabled by
1519  * default.
1520  */
1521 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1522
1523 static inline void crash_enable_local_vmclear(int cpu)
1524 {
1525         cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1526 }
1527
1528 static inline void crash_disable_local_vmclear(int cpu)
1529 {
1530         cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1531 }
1532
1533 static inline int crash_local_vmclear_enabled(int cpu)
1534 {
1535         return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1536 }
1537
1538 static void crash_vmclear_local_loaded_vmcss(void)
1539 {
1540         int cpu = raw_smp_processor_id();
1541         struct loaded_vmcs *v;
1542
1543         if (!crash_local_vmclear_enabled(cpu))
1544                 return;
1545
1546         list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1547                             loaded_vmcss_on_cpu_link)
1548                 vmcs_clear(v->vmcs);
1549 }
1550 #else
1551 static inline void crash_enable_local_vmclear(int cpu) { }
1552 static inline void crash_disable_local_vmclear(int cpu) { }
1553 #endif /* CONFIG_KEXEC_CORE */
1554
1555 static void __loaded_vmcs_clear(void *arg)
1556 {
1557         struct loaded_vmcs *loaded_vmcs = arg;
1558         int cpu = raw_smp_processor_id();
1559
1560         if (loaded_vmcs->cpu != cpu)
1561                 return; /* vcpu migration can race with cpu offline */
1562         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1563                 per_cpu(current_vmcs, cpu) = NULL;
1564         crash_disable_local_vmclear(cpu);
1565         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1566
1567         /*
1568          * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1569          * is before setting loaded_vmcs->vcpu to -1 which is done in
1570          * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1571          * then adds the vmcs into percpu list before it is deleted.
1572          */
1573         smp_wmb();
1574
1575         loaded_vmcs_init(loaded_vmcs);
1576         crash_enable_local_vmclear(cpu);
1577 }
1578
1579 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1580 {
1581         int cpu = loaded_vmcs->cpu;
1582
1583         if (cpu != -1)
1584                 smp_call_function_single(cpu,
1585                          __loaded_vmcs_clear, loaded_vmcs, 1);
1586 }
1587
1588 static inline void vpid_sync_vcpu_single(int vpid)
1589 {
1590         if (vpid == 0)
1591                 return;
1592
1593         if (cpu_has_vmx_invvpid_single())
1594                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1595 }
1596
1597 static inline void vpid_sync_vcpu_global(void)
1598 {
1599         if (cpu_has_vmx_invvpid_global())
1600                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1601 }
1602
1603 static inline void vpid_sync_context(int vpid)
1604 {
1605         if (cpu_has_vmx_invvpid_single())
1606                 vpid_sync_vcpu_single(vpid);
1607         else
1608                 vpid_sync_vcpu_global();
1609 }
1610
1611 static inline void ept_sync_global(void)
1612 {
1613         __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1614 }
1615
1616 static inline void ept_sync_context(u64 eptp)
1617 {
1618         if (cpu_has_vmx_invept_context())
1619                 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1620         else
1621                 ept_sync_global();
1622 }
1623
1624 static __always_inline void vmcs_check16(unsigned long field)
1625 {
1626         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1627                          "16-bit accessor invalid for 64-bit field");
1628         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1629                          "16-bit accessor invalid for 64-bit high field");
1630         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1631                          "16-bit accessor invalid for 32-bit high field");
1632         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1633                          "16-bit accessor invalid for natural width field");
1634 }
1635
1636 static __always_inline void vmcs_check32(unsigned long field)
1637 {
1638         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1639                          "32-bit accessor invalid for 16-bit field");
1640         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1641                          "32-bit accessor invalid for natural width field");
1642 }
1643
1644 static __always_inline void vmcs_check64(unsigned long field)
1645 {
1646         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1647                          "64-bit accessor invalid for 16-bit field");
1648         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1649                          "64-bit accessor invalid for 64-bit high field");
1650         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1651                          "64-bit accessor invalid for 32-bit field");
1652         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1653                          "64-bit accessor invalid for natural width field");
1654 }
1655
1656 static __always_inline void vmcs_checkl(unsigned long field)
1657 {
1658         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1659                          "Natural width accessor invalid for 16-bit field");
1660         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1661                          "Natural width accessor invalid for 64-bit field");
1662         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1663                          "Natural width accessor invalid for 64-bit high field");
1664         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1665                          "Natural width accessor invalid for 32-bit field");
1666 }
1667
1668 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1669 {
1670         unsigned long value;
1671
1672         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1673                       : "=a"(value) : "d"(field) : "cc");
1674         return value;
1675 }
1676
1677 static __always_inline u16 vmcs_read16(unsigned long field)
1678 {
1679         vmcs_check16(field);
1680         return __vmcs_readl(field);
1681 }
1682
1683 static __always_inline u32 vmcs_read32(unsigned long field)
1684 {
1685         vmcs_check32(field);
1686         return __vmcs_readl(field);
1687 }
1688
1689 static __always_inline u64 vmcs_read64(unsigned long field)
1690 {
1691         vmcs_check64(field);
1692 #ifdef CONFIG_X86_64
1693         return __vmcs_readl(field);
1694 #else
1695         return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1696 #endif
1697 }
1698
1699 static __always_inline unsigned long vmcs_readl(unsigned long field)
1700 {
1701         vmcs_checkl(field);
1702         return __vmcs_readl(field);
1703 }
1704
1705 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1706 {
1707         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1708                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1709         dump_stack();
1710 }
1711
1712 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1713 {
1714         u8 error;
1715
1716         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1717                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1718         if (unlikely(error))
1719                 vmwrite_error(field, value);
1720 }
1721
1722 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1723 {
1724         vmcs_check16(field);
1725         __vmcs_writel(field, value);
1726 }
1727
1728 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1729 {
1730         vmcs_check32(field);
1731         __vmcs_writel(field, value);
1732 }
1733
1734 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1735 {
1736         vmcs_check64(field);
1737         __vmcs_writel(field, value);
1738 #ifndef CONFIG_X86_64
1739         asm volatile ("");
1740         __vmcs_writel(field+1, value >> 32);
1741 #endif
1742 }
1743
1744 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1745 {
1746         vmcs_checkl(field);
1747         __vmcs_writel(field, value);
1748 }
1749
1750 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1751 {
1752         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1753                          "vmcs_clear_bits does not support 64-bit fields");
1754         __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1755 }
1756
1757 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1758 {
1759         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1760                          "vmcs_set_bits does not support 64-bit fields");
1761         __vmcs_writel(field, __vmcs_readl(field) | mask);
1762 }
1763
1764 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1765 {
1766         vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1767 }
1768
1769 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1770 {
1771         vmcs_write32(VM_ENTRY_CONTROLS, val);
1772         vmx->vm_entry_controls_shadow = val;
1773 }
1774
1775 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1776 {
1777         if (vmx->vm_entry_controls_shadow != val)
1778                 vm_entry_controls_init(vmx, val);
1779 }
1780
1781 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1782 {
1783         return vmx->vm_entry_controls_shadow;
1784 }
1785
1786
1787 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1788 {
1789         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1790 }
1791
1792 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1793 {
1794         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1795 }
1796
1797 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1798 {
1799         vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1800 }
1801
1802 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1803 {
1804         vmcs_write32(VM_EXIT_CONTROLS, val);
1805         vmx->vm_exit_controls_shadow = val;
1806 }
1807
1808 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1809 {
1810         if (vmx->vm_exit_controls_shadow != val)
1811                 vm_exit_controls_init(vmx, val);
1812 }
1813
1814 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1815 {
1816         return vmx->vm_exit_controls_shadow;
1817 }
1818
1819
1820 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1821 {
1822         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1823 }
1824
1825 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1826 {
1827         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1828 }
1829
1830 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1831 {
1832         vmx->segment_cache.bitmask = 0;
1833 }
1834
1835 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1836                                        unsigned field)
1837 {
1838         bool ret;
1839         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1840
1841         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1842                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1843                 vmx->segment_cache.bitmask = 0;
1844         }
1845         ret = vmx->segment_cache.bitmask & mask;
1846         vmx->segment_cache.bitmask |= mask;
1847         return ret;
1848 }
1849
1850 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1851 {
1852         u16 *p = &vmx->segment_cache.seg[seg].selector;
1853
1854         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1855                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1856         return *p;
1857 }
1858
1859 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1860 {
1861         ulong *p = &vmx->segment_cache.seg[seg].base;
1862
1863         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1864                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1865         return *p;
1866 }
1867
1868 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1869 {
1870         u32 *p = &vmx->segment_cache.seg[seg].limit;
1871
1872         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1873                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1874         return *p;
1875 }
1876
1877 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1878 {
1879         u32 *p = &vmx->segment_cache.seg[seg].ar;
1880
1881         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1882                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1883         return *p;
1884 }
1885
1886 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1887 {
1888         u32 eb;
1889
1890         eb = (1u << PF_VECTOR) | (1u << MC_VECTOR) |
1891              (1u << DB_VECTOR) | (1u << AC_VECTOR);
1892         if ((vcpu->guest_debug &
1893              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1894             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1895                 eb |= 1u << BP_VECTOR;
1896         if (to_vmx(vcpu)->rmode.vm86_active)
1897                 eb = ~0;
1898         if (enable_ept)
1899                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1900
1901         /* When we are running a nested L2 guest and L1 specified for it a
1902          * certain exception bitmap, we must trap the same exceptions and pass
1903          * them to L1. When running L2, we will only handle the exceptions
1904          * specified above if L1 did not want them.
1905          */
1906         if (is_guest_mode(vcpu))
1907                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1908         else
1909                 eb |= 1u << UD_VECTOR;
1910
1911         vmcs_write32(EXCEPTION_BITMAP, eb);
1912 }
1913
1914 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1915                 unsigned long entry, unsigned long exit)
1916 {
1917         vm_entry_controls_clearbit(vmx, entry);
1918         vm_exit_controls_clearbit(vmx, exit);
1919 }
1920
1921 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1922 {
1923         unsigned i;
1924         struct msr_autoload *m = &vmx->msr_autoload;
1925
1926         switch (msr) {
1927         case MSR_EFER:
1928                 if (cpu_has_load_ia32_efer) {
1929                         clear_atomic_switch_msr_special(vmx,
1930                                         VM_ENTRY_LOAD_IA32_EFER,
1931                                         VM_EXIT_LOAD_IA32_EFER);
1932                         return;
1933                 }
1934                 break;
1935         case MSR_CORE_PERF_GLOBAL_CTRL:
1936                 if (cpu_has_load_perf_global_ctrl) {
1937                         clear_atomic_switch_msr_special(vmx,
1938                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1939                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1940                         return;
1941                 }
1942                 break;
1943         }
1944
1945         for (i = 0; i < m->nr; ++i)
1946                 if (m->guest[i].index == msr)
1947                         break;
1948
1949         if (i == m->nr)
1950                 return;
1951         --m->nr;
1952         m->guest[i] = m->guest[m->nr];
1953         m->host[i] = m->host[m->nr];
1954         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1955         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1956 }
1957
1958 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1959                 unsigned long entry, unsigned long exit,
1960                 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1961                 u64 guest_val, u64 host_val)
1962 {
1963         vmcs_write64(guest_val_vmcs, guest_val);
1964         vmcs_write64(host_val_vmcs, host_val);
1965         vm_entry_controls_setbit(vmx, entry);
1966         vm_exit_controls_setbit(vmx, exit);
1967 }
1968
1969 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1970                                   u64 guest_val, u64 host_val)
1971 {
1972         unsigned i;
1973         struct msr_autoload *m = &vmx->msr_autoload;
1974
1975         switch (msr) {
1976         case MSR_EFER:
1977                 if (cpu_has_load_ia32_efer) {
1978                         add_atomic_switch_msr_special(vmx,
1979                                         VM_ENTRY_LOAD_IA32_EFER,
1980                                         VM_EXIT_LOAD_IA32_EFER,
1981                                         GUEST_IA32_EFER,
1982                                         HOST_IA32_EFER,
1983                                         guest_val, host_val);
1984                         return;
1985                 }
1986                 break;
1987         case MSR_CORE_PERF_GLOBAL_CTRL:
1988                 if (cpu_has_load_perf_global_ctrl) {
1989                         add_atomic_switch_msr_special(vmx,
1990                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1991                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1992                                         GUEST_IA32_PERF_GLOBAL_CTRL,
1993                                         HOST_IA32_PERF_GLOBAL_CTRL,
1994                                         guest_val, host_val);
1995                         return;
1996                 }
1997                 break;
1998         case MSR_IA32_PEBS_ENABLE:
1999                 /* PEBS needs a quiescent period after being disabled (to write
2000                  * a record).  Disabling PEBS through VMX MSR swapping doesn't
2001                  * provide that period, so a CPU could write host's record into
2002                  * guest's memory.
2003                  */
2004                 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
2005         }
2006
2007         for (i = 0; i < m->nr; ++i)
2008                 if (m->guest[i].index == msr)
2009                         break;
2010
2011         if (i == NR_AUTOLOAD_MSRS) {
2012                 printk_once(KERN_WARNING "Not enough msr switch entries. "
2013                                 "Can't add msr %x\n", msr);
2014                 return;
2015         } else if (i == m->nr) {
2016                 ++m->nr;
2017                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2018                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2019         }
2020
2021         m->guest[i].index = msr;
2022         m->guest[i].value = guest_val;
2023         m->host[i].index = msr;
2024         m->host[i].value = host_val;
2025 }
2026
2027 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2028 {
2029         u64 guest_efer = vmx->vcpu.arch.efer;
2030         u64 ignore_bits = 0;
2031
2032         if (!enable_ept) {
2033                 /*
2034                  * NX is needed to handle CR0.WP=1, CR4.SMEP=1.  Testing
2035                  * host CPUID is more efficient than testing guest CPUID
2036                  * or CR4.  Host SMEP is anyway a requirement for guest SMEP.
2037                  */
2038                 if (boot_cpu_has(X86_FEATURE_SMEP))
2039                         guest_efer |= EFER_NX;
2040                 else if (!(guest_efer & EFER_NX))
2041                         ignore_bits |= EFER_NX;
2042         }
2043
2044         /*
2045          * LMA and LME handled by hardware; SCE meaningless outside long mode.
2046          */
2047         ignore_bits |= EFER_SCE;
2048 #ifdef CONFIG_X86_64
2049         ignore_bits |= EFER_LMA | EFER_LME;
2050         /* SCE is meaningful only in long mode on Intel */
2051         if (guest_efer & EFER_LMA)
2052                 ignore_bits &= ~(u64)EFER_SCE;
2053 #endif
2054
2055         clear_atomic_switch_msr(vmx, MSR_EFER);
2056
2057         /*
2058          * On EPT, we can't emulate NX, so we must switch EFER atomically.
2059          * On CPUs that support "load IA32_EFER", always switch EFER
2060          * atomically, since it's faster than switching it manually.
2061          */
2062         if (cpu_has_load_ia32_efer ||
2063             (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2064                 if (!(guest_efer & EFER_LMA))
2065                         guest_efer &= ~EFER_LME;
2066                 if (guest_efer != host_efer)
2067                         add_atomic_switch_msr(vmx, MSR_EFER,
2068                                               guest_efer, host_efer);
2069                 return false;
2070         } else {
2071                 guest_efer &= ~ignore_bits;
2072                 guest_efer |= host_efer & ignore_bits;
2073
2074                 vmx->guest_msrs[efer_offset].data = guest_efer;
2075                 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2076
2077                 return true;
2078         }
2079 }
2080
2081 #ifdef CONFIG_X86_32
2082 /*
2083  * On 32-bit kernels, VM exits still load the FS and GS bases from the
2084  * VMCS rather than the segment table.  KVM uses this helper to figure
2085  * out the current bases to poke them into the VMCS before entry.
2086  */
2087 static unsigned long segment_base(u16 selector)
2088 {
2089         struct desc_struct *table;
2090         unsigned long v;
2091
2092         if (!(selector & ~SEGMENT_RPL_MASK))
2093                 return 0;
2094
2095         table = get_current_gdt_ro();
2096
2097         if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2098                 u16 ldt_selector = kvm_read_ldt();
2099
2100                 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2101                         return 0;
2102
2103                 table = (struct desc_struct *)segment_base(ldt_selector);
2104         }
2105         v = get_desc_base(&table[selector >> 3]);
2106         return v;
2107 }
2108 #endif
2109
2110 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2111 {
2112         struct vcpu_vmx *vmx = to_vmx(vcpu);
2113         int i;
2114
2115         if (vmx->host_state.loaded)
2116                 return;
2117
2118         vmx->host_state.loaded = 1;
2119         /*
2120          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
2121          * allow segment selectors with cpl > 0 or ti == 1.
2122          */
2123         vmx->host_state.ldt_sel = kvm_read_ldt();
2124         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2125         savesegment(fs, vmx->host_state.fs_sel);
2126         if (!(vmx->host_state.fs_sel & 7)) {
2127                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2128                 vmx->host_state.fs_reload_needed = 0;
2129         } else {
2130                 vmcs_write16(HOST_FS_SELECTOR, 0);
2131                 vmx->host_state.fs_reload_needed = 1;
2132         }
2133         savesegment(gs, vmx->host_state.gs_sel);
2134         if (!(vmx->host_state.gs_sel & 7))
2135                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2136         else {
2137                 vmcs_write16(HOST_GS_SELECTOR, 0);
2138                 vmx->host_state.gs_ldt_reload_needed = 1;
2139         }
2140
2141 #ifdef CONFIG_X86_64
2142         savesegment(ds, vmx->host_state.ds_sel);
2143         savesegment(es, vmx->host_state.es_sel);
2144 #endif
2145
2146 #ifdef CONFIG_X86_64
2147         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2148         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2149 #else
2150         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2151         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2152 #endif
2153
2154 #ifdef CONFIG_X86_64
2155         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2156         if (is_long_mode(&vmx->vcpu))
2157                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2158 #endif
2159         if (boot_cpu_has(X86_FEATURE_MPX))
2160                 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2161         for (i = 0; i < vmx->save_nmsrs; ++i)
2162                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2163                                    vmx->guest_msrs[i].data,
2164                                    vmx->guest_msrs[i].mask);
2165 }
2166
2167 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2168 {
2169         if (!vmx->host_state.loaded)
2170                 return;
2171
2172         ++vmx->vcpu.stat.host_state_reload;
2173         vmx->host_state.loaded = 0;
2174 #ifdef CONFIG_X86_64
2175         if (is_long_mode(&vmx->vcpu))
2176                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2177 #endif
2178         if (vmx->host_state.gs_ldt_reload_needed) {
2179                 kvm_load_ldt(vmx->host_state.ldt_sel);
2180 #ifdef CONFIG_X86_64
2181                 load_gs_index(vmx->host_state.gs_sel);
2182 #else
2183                 loadsegment(gs, vmx->host_state.gs_sel);
2184 #endif
2185         }
2186         if (vmx->host_state.fs_reload_needed)
2187                 loadsegment(fs, vmx->host_state.fs_sel);
2188 #ifdef CONFIG_X86_64
2189         if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2190                 loadsegment(ds, vmx->host_state.ds_sel);
2191                 loadsegment(es, vmx->host_state.es_sel);
2192         }
2193 #endif
2194         invalidate_tss_limit();
2195 #ifdef CONFIG_X86_64
2196         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2197 #endif
2198         if (vmx->host_state.msr_host_bndcfgs)
2199                 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2200         load_fixmap_gdt(raw_smp_processor_id());
2201 }
2202
2203 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2204 {
2205         preempt_disable();
2206         __vmx_load_host_state(vmx);
2207         preempt_enable();
2208 }
2209
2210 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2211 {
2212         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2213         struct pi_desc old, new;
2214         unsigned int dest;
2215
2216         /*
2217          * In case of hot-plug or hot-unplug, we may have to undo
2218          * vmx_vcpu_pi_put even if there is no assigned device.  And we
2219          * always keep PI.NDST up to date for simplicity: it makes the
2220          * code easier, and CPU migration is not a fast path.
2221          */
2222         if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
2223                 return;
2224
2225         /*
2226          * First handle the simple case where no cmpxchg is necessary; just
2227          * allow posting non-urgent interrupts.
2228          *
2229          * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
2230          * PI.NDST: pi_post_block will do it for us and the wakeup_handler
2231          * expects the VCPU to be on the blocked_vcpu_list that matches
2232          * PI.NDST.
2233          */
2234         if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR ||
2235             vcpu->cpu == cpu) {
2236                 pi_clear_sn(pi_desc);
2237                 return;
2238         }
2239
2240         /* The full case.  */
2241         do {
2242                 old.control = new.control = pi_desc->control;
2243
2244                 dest = cpu_physical_id(cpu);
2245
2246                 if (x2apic_enabled())
2247                         new.ndst = dest;
2248                 else
2249                         new.ndst = (dest << 8) & 0xFF00;
2250
2251                 new.sn = 0;
2252         } while (cmpxchg64(&pi_desc->control, old.control,
2253                            new.control) != old.control);
2254 }
2255
2256 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2257 {
2258         vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2259         vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2260 }
2261
2262 /*
2263  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2264  * vcpu mutex is already taken.
2265  */
2266 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2267 {
2268         struct vcpu_vmx *vmx = to_vmx(vcpu);
2269         bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2270
2271         if (!already_loaded) {
2272                 loaded_vmcs_clear(vmx->loaded_vmcs);
2273                 local_irq_disable();
2274                 crash_disable_local_vmclear(cpu);
2275
2276                 /*
2277                  * Read loaded_vmcs->cpu should be before fetching
2278                  * loaded_vmcs->loaded_vmcss_on_cpu_link.
2279                  * See the comments in __loaded_vmcs_clear().
2280                  */
2281                 smp_rmb();
2282
2283                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2284                          &per_cpu(loaded_vmcss_on_cpu, cpu));
2285                 crash_enable_local_vmclear(cpu);
2286                 local_irq_enable();
2287         }
2288
2289         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2290                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2291                 vmcs_load(vmx->loaded_vmcs->vmcs);
2292         }
2293
2294         if (!already_loaded) {
2295                 void *gdt = get_current_gdt_ro();
2296                 unsigned long sysenter_esp;
2297
2298                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2299
2300                 /*
2301                  * Linux uses per-cpu TSS and GDT, so set these when switching
2302                  * processors.  See 22.2.4.
2303                  */
2304                 vmcs_writel(HOST_TR_BASE,
2305                             (unsigned long)this_cpu_ptr(&cpu_tss));
2306                 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
2307
2308                 /*
2309                  * VM exits change the host TR limit to 0x67 after a VM
2310                  * exit.  This is okay, since 0x67 covers everything except
2311                  * the IO bitmap and have have code to handle the IO bitmap
2312                  * being lost after a VM exit.
2313                  */
2314                 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2315
2316                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2317                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2318
2319                 vmx->loaded_vmcs->cpu = cpu;
2320         }
2321
2322         /* Setup TSC multiplier */
2323         if (kvm_has_tsc_control &&
2324             vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2325                 decache_tsc_multiplier(vmx);
2326
2327         vmx_vcpu_pi_load(vcpu, cpu);
2328         vmx->host_pkru = read_pkru();
2329 }
2330
2331 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2332 {
2333         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2334
2335         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2336                 !irq_remapping_cap(IRQ_POSTING_CAP)  ||
2337                 !kvm_vcpu_apicv_active(vcpu))
2338                 return;
2339
2340         /* Set SN when the vCPU is preempted */
2341         if (vcpu->preempted)
2342                 pi_set_sn(pi_desc);
2343 }
2344
2345 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2346 {
2347         vmx_vcpu_pi_put(vcpu);
2348
2349         __vmx_load_host_state(to_vmx(vcpu));
2350 }
2351
2352 static bool emulation_required(struct kvm_vcpu *vcpu)
2353 {
2354         return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2355 }
2356
2357 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2358
2359 /*
2360  * Return the cr0 value that a nested guest would read. This is a combination
2361  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2362  * its hypervisor (cr0_read_shadow).
2363  */
2364 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2365 {
2366         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2367                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2368 }
2369 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2370 {
2371         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2372                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2373 }
2374
2375 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2376 {
2377         unsigned long rflags, save_rflags;
2378
2379         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2380                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2381                 rflags = vmcs_readl(GUEST_RFLAGS);
2382                 if (to_vmx(vcpu)->rmode.vm86_active) {
2383                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2384                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2385                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2386                 }
2387                 to_vmx(vcpu)->rflags = rflags;
2388         }
2389         return to_vmx(vcpu)->rflags;
2390 }
2391
2392 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2393 {
2394         unsigned long old_rflags = vmx_get_rflags(vcpu);
2395
2396         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2397         to_vmx(vcpu)->rflags = rflags;
2398         if (to_vmx(vcpu)->rmode.vm86_active) {
2399                 to_vmx(vcpu)->rmode.save_rflags = rflags;
2400                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2401         }
2402         vmcs_writel(GUEST_RFLAGS, rflags);
2403
2404         if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2405                 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2406 }
2407
2408 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2409 {
2410         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2411         int ret = 0;
2412
2413         if (interruptibility & GUEST_INTR_STATE_STI)
2414                 ret |= KVM_X86_SHADOW_INT_STI;
2415         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2416                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2417
2418         return ret;
2419 }
2420
2421 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2422 {
2423         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2424         u32 interruptibility = interruptibility_old;
2425
2426         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2427
2428         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2429                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2430         else if (mask & KVM_X86_SHADOW_INT_STI)
2431                 interruptibility |= GUEST_INTR_STATE_STI;
2432
2433         if ((interruptibility != interruptibility_old))
2434                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2435 }
2436
2437 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2438 {
2439         unsigned long rip;
2440
2441         rip = kvm_rip_read(vcpu);
2442         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2443         kvm_rip_write(vcpu, rip);
2444
2445         /* skipping an emulated instruction also counts */
2446         vmx_set_interrupt_shadow(vcpu, 0);
2447 }
2448
2449 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2450                                                unsigned long exit_qual)
2451 {
2452         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2453         unsigned int nr = vcpu->arch.exception.nr;
2454         u32 intr_info = nr | INTR_INFO_VALID_MASK;
2455
2456         if (vcpu->arch.exception.has_error_code) {
2457                 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2458                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2459         }
2460
2461         if (kvm_exception_is_soft(nr))
2462                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2463         else
2464                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2465
2466         if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2467             vmx_get_nmi_mask(vcpu))
2468                 intr_info |= INTR_INFO_UNBLOCK_NMI;
2469
2470         nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2471 }
2472
2473 /*
2474  * KVM wants to inject page-faults which it got to the guest. This function
2475  * checks whether in a nested guest, we need to inject them to L1 or L2.
2476  */
2477 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
2478 {
2479         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2480         unsigned int nr = vcpu->arch.exception.nr;
2481
2482         if (nr == PF_VECTOR) {
2483                 if (vcpu->arch.exception.nested_apf) {
2484                         *exit_qual = vcpu->arch.apf.nested_apf_token;
2485                         return 1;
2486                 }
2487                 /*
2488                  * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2489                  * The fix is to add the ancillary datum (CR2 or DR6) to structs
2490                  * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2491                  * can be written only when inject_pending_event runs.  This should be
2492                  * conditional on a new capability---if the capability is disabled,
2493                  * kvm_multiple_exception would write the ancillary information to
2494                  * CR2 or DR6, for backwards ABI-compatibility.
2495                  */
2496                 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2497                                                     vcpu->arch.exception.error_code)) {
2498                         *exit_qual = vcpu->arch.cr2;
2499                         return 1;
2500                 }
2501         } else {
2502                 if (vmcs12->exception_bitmap & (1u << nr)) {
2503                         if (nr == DB_VECTOR)
2504                                 *exit_qual = vcpu->arch.dr6;
2505                         else
2506                                 *exit_qual = 0;
2507                         return 1;
2508                 }
2509         }
2510
2511         return 0;
2512 }
2513
2514 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
2515 {
2516         struct vcpu_vmx *vmx = to_vmx(vcpu);
2517         unsigned nr = vcpu->arch.exception.nr;
2518         bool has_error_code = vcpu->arch.exception.has_error_code;
2519         u32 error_code = vcpu->arch.exception.error_code;
2520         u32 intr_info = nr | INTR_INFO_VALID_MASK;
2521
2522         if (has_error_code) {
2523                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2524                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2525         }
2526
2527         if (vmx->rmode.vm86_active) {
2528                 int inc_eip = 0;
2529                 if (kvm_exception_is_soft(nr))
2530                         inc_eip = vcpu->arch.event_exit_inst_len;
2531                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2532                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2533                 return;
2534         }
2535
2536         if (kvm_exception_is_soft(nr)) {
2537                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2538                              vmx->vcpu.arch.event_exit_inst_len);
2539                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2540         } else
2541                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2542
2543         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2544 }
2545
2546 static bool vmx_rdtscp_supported(void)
2547 {
2548         return cpu_has_vmx_rdtscp();
2549 }
2550
2551 static bool vmx_invpcid_supported(void)
2552 {
2553         return cpu_has_vmx_invpcid() && enable_ept;
2554 }
2555
2556 /*
2557  * Swap MSR entry in host/guest MSR entry array.
2558  */
2559 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2560 {
2561         struct shared_msr_entry tmp;
2562
2563         tmp = vmx->guest_msrs[to];
2564         vmx->guest_msrs[to] = vmx->guest_msrs[from];
2565         vmx->guest_msrs[from] = tmp;
2566 }
2567
2568 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2569 {
2570         unsigned long *msr_bitmap;
2571
2572         if (is_guest_mode(vcpu))
2573                 msr_bitmap = to_vmx(vcpu)->nested.msr_bitmap;
2574         else if (cpu_has_secondary_exec_ctrls() &&
2575                  (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2576                   SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2577                 if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) {
2578                         if (is_long_mode(vcpu))
2579                                 msr_bitmap = vmx_msr_bitmap_longmode_x2apic_apicv;
2580                         else
2581                                 msr_bitmap = vmx_msr_bitmap_legacy_x2apic_apicv;
2582                 } else {
2583                         if (is_long_mode(vcpu))
2584                                 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2585                         else
2586                                 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2587                 }
2588         } else {
2589                 if (is_long_mode(vcpu))
2590                         msr_bitmap = vmx_msr_bitmap_longmode;
2591                 else
2592                         msr_bitmap = vmx_msr_bitmap_legacy;
2593         }
2594
2595         vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2596 }
2597
2598 /*
2599  * Set up the vmcs to automatically save and restore system
2600  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
2601  * mode, as fiddling with msrs is very expensive.
2602  */
2603 static void setup_msrs(struct vcpu_vmx *vmx)
2604 {
2605         int save_nmsrs, index;
2606
2607         save_nmsrs = 0;
2608 #ifdef CONFIG_X86_64
2609         if (is_long_mode(&vmx->vcpu)) {
2610                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2611                 if (index >= 0)
2612                         move_msr_up(vmx, index, save_nmsrs++);
2613                 index = __find_msr_index(vmx, MSR_LSTAR);
2614                 if (index >= 0)
2615                         move_msr_up(vmx, index, save_nmsrs++);
2616                 index = __find_msr_index(vmx, MSR_CSTAR);
2617                 if (index >= 0)
2618                         move_msr_up(vmx, index, save_nmsrs++);
2619                 index = __find_msr_index(vmx, MSR_TSC_AUX);
2620                 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
2621                         move_msr_up(vmx, index, save_nmsrs++);
2622                 /*
2623                  * MSR_STAR is only needed on long mode guests, and only
2624                  * if efer.sce is enabled.
2625                  */
2626                 index = __find_msr_index(vmx, MSR_STAR);
2627                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2628                         move_msr_up(vmx, index, save_nmsrs++);
2629         }
2630 #endif
2631         index = __find_msr_index(vmx, MSR_EFER);
2632         if (index >= 0 && update_transition_efer(vmx, index))
2633                 move_msr_up(vmx, index, save_nmsrs++);
2634
2635         vmx->save_nmsrs = save_nmsrs;
2636
2637         if (cpu_has_vmx_msr_bitmap())
2638                 vmx_set_msr_bitmap(&vmx->vcpu);
2639 }
2640
2641 /*
2642  * reads and returns guest's timestamp counter "register"
2643  * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2644  * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2645  */
2646 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2647 {
2648         u64 host_tsc, tsc_offset;
2649
2650         host_tsc = rdtsc();
2651         tsc_offset = vmcs_read64(TSC_OFFSET);
2652         return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2653 }
2654
2655 /*
2656  * writes 'offset' into guest's timestamp counter offset register
2657  */
2658 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2659 {
2660         if (is_guest_mode(vcpu)) {
2661                 /*
2662                  * We're here if L1 chose not to trap WRMSR to TSC. According
2663                  * to the spec, this should set L1's TSC; The offset that L1
2664                  * set for L2 remains unchanged, and still needs to be added
2665                  * to the newly set TSC to get L2's TSC.
2666                  */
2667                 struct vmcs12 *vmcs12;
2668                 /* recalculate vmcs02.TSC_OFFSET: */
2669                 vmcs12 = get_vmcs12(vcpu);
2670                 vmcs_write64(TSC_OFFSET, offset +
2671                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2672                          vmcs12->tsc_offset : 0));
2673         } else {
2674                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2675                                            vmcs_read64(TSC_OFFSET), offset);
2676                 vmcs_write64(TSC_OFFSET, offset);
2677         }
2678 }
2679
2680 /*
2681  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2682  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2683  * all guests if the "nested" module option is off, and can also be disabled
2684  * for a single guest by disabling its VMX cpuid bit.
2685  */
2686 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2687 {
2688         return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
2689 }
2690
2691 /*
2692  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2693  * returned for the various VMX controls MSRs when nested VMX is enabled.
2694  * The same values should also be used to verify that vmcs12 control fields are
2695  * valid during nested entry from L1 to L2.
2696  * Each of these control msrs has a low and high 32-bit half: A low bit is on
2697  * if the corresponding bit in the (32-bit) control field *must* be on, and a
2698  * bit in the high half is on if the corresponding bit in the control field
2699  * may be on. See also vmx_control_verify().
2700  */
2701 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2702 {
2703         /*
2704          * Note that as a general rule, the high half of the MSRs (bits in
2705          * the control fields which may be 1) should be initialized by the
2706          * intersection of the underlying hardware's MSR (i.e., features which
2707          * can be supported) and the list of features we want to expose -
2708          * because they are known to be properly supported in our code.
2709          * Also, usually, the low half of the MSRs (bits which must be 1) can
2710          * be set to 0, meaning that L1 may turn off any of these bits. The
2711          * reason is that if one of these bits is necessary, it will appear
2712          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2713          * fields of vmcs01 and vmcs02, will turn these bits off - and
2714          * nested_vmx_exit_reflected() will not pass related exits to L1.
2715          * These rules have exceptions below.
2716          */
2717
2718         /* pin-based controls */
2719         rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2720                 vmx->nested.nested_vmx_pinbased_ctls_low,
2721                 vmx->nested.nested_vmx_pinbased_ctls_high);
2722         vmx->nested.nested_vmx_pinbased_ctls_low |=
2723                 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2724         vmx->nested.nested_vmx_pinbased_ctls_high &=
2725                 PIN_BASED_EXT_INTR_MASK |
2726                 PIN_BASED_NMI_EXITING |
2727                 PIN_BASED_VIRTUAL_NMIS;
2728         vmx->nested.nested_vmx_pinbased_ctls_high |=
2729                 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2730                 PIN_BASED_VMX_PREEMPTION_TIMER;
2731         if (kvm_vcpu_apicv_active(&vmx->vcpu))
2732                 vmx->nested.nested_vmx_pinbased_ctls_high |=
2733                         PIN_BASED_POSTED_INTR;
2734
2735         /* exit controls */
2736         rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2737                 vmx->nested.nested_vmx_exit_ctls_low,
2738                 vmx->nested.nested_vmx_exit_ctls_high);
2739         vmx->nested.nested_vmx_exit_ctls_low =
2740                 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2741
2742         vmx->nested.nested_vmx_exit_ctls_high &=
2743 #ifdef CONFIG_X86_64
2744                 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2745 #endif
2746                 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2747         vmx->nested.nested_vmx_exit_ctls_high |=
2748                 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2749                 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2750                 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2751
2752         if (kvm_mpx_supported())
2753                 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2754
2755         /* We support free control of debug control saving. */
2756         vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2757
2758         /* entry controls */
2759         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2760                 vmx->nested.nested_vmx_entry_ctls_low,
2761                 vmx->nested.nested_vmx_entry_ctls_high);
2762         vmx->nested.nested_vmx_entry_ctls_low =
2763                 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2764         vmx->nested.nested_vmx_entry_ctls_high &=
2765 #ifdef CONFIG_X86_64
2766                 VM_ENTRY_IA32E_MODE |
2767 #endif
2768                 VM_ENTRY_LOAD_IA32_PAT;
2769         vmx->nested.nested_vmx_entry_ctls_high |=
2770                 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2771         if (kvm_mpx_supported())
2772                 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2773
2774         /* We support free control of debug control loading. */
2775         vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2776
2777         /* cpu-based controls */
2778         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2779                 vmx->nested.nested_vmx_procbased_ctls_low,
2780                 vmx->nested.nested_vmx_procbased_ctls_high);
2781         vmx->nested.nested_vmx_procbased_ctls_low =
2782                 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2783         vmx->nested.nested_vmx_procbased_ctls_high &=
2784                 CPU_BASED_VIRTUAL_INTR_PENDING |
2785                 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2786                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2787                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2788                 CPU_BASED_CR3_STORE_EXITING |
2789 #ifdef CONFIG_X86_64
2790                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2791 #endif
2792                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2793                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2794                 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2795                 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2796                 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2797         /*
2798          * We can allow some features even when not supported by the
2799          * hardware. For example, L1 can specify an MSR bitmap - and we
2800          * can use it to avoid exits to L1 - even when L0 runs L2
2801          * without MSR bitmaps.
2802          */
2803         vmx->nested.nested_vmx_procbased_ctls_high |=
2804                 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2805                 CPU_BASED_USE_MSR_BITMAPS;
2806
2807         /* We support free control of CR3 access interception. */
2808         vmx->nested.nested_vmx_procbased_ctls_low &=
2809                 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2810
2811         /*
2812          * secondary cpu-based controls.  Do not include those that
2813          * depend on CPUID bits, they are added later by vmx_cpuid_update.
2814          */
2815         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2816                 vmx->nested.nested_vmx_secondary_ctls_low,
2817                 vmx->nested.nested_vmx_secondary_ctls_high);
2818         vmx->nested.nested_vmx_secondary_ctls_low = 0;
2819         vmx->nested.nested_vmx_secondary_ctls_high &=
2820                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2821                 SECONDARY_EXEC_DESC |
2822                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2823                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2824                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2825                 SECONDARY_EXEC_WBINVD_EXITING;
2826
2827         if (enable_ept) {
2828                 /* nested EPT: emulate EPT also to L1 */
2829                 vmx->nested.nested_vmx_secondary_ctls_high |=
2830                         SECONDARY_EXEC_ENABLE_EPT;
2831                 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2832                          VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2833                 if (cpu_has_vmx_ept_execute_only())
2834                         vmx->nested.nested_vmx_ept_caps |=
2835                                 VMX_EPT_EXECUTE_ONLY_BIT;
2836                 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2837                 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2838                         VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2839                         VMX_EPT_1GB_PAGE_BIT;
2840                 if (enable_ept_ad_bits) {
2841                         vmx->nested.nested_vmx_secondary_ctls_high |=
2842                                 SECONDARY_EXEC_ENABLE_PML;
2843                         vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2844                 }
2845         }
2846
2847         if (cpu_has_vmx_vmfunc()) {
2848                 vmx->nested.nested_vmx_secondary_ctls_high |=
2849                         SECONDARY_EXEC_ENABLE_VMFUNC;
2850                 /*
2851                  * Advertise EPTP switching unconditionally
2852                  * since we emulate it
2853                  */
2854                 if (enable_ept)
2855                         vmx->nested.nested_vmx_vmfunc_controls =
2856                                 VMX_VMFUNC_EPTP_SWITCHING;
2857         }
2858
2859         /*
2860          * Old versions of KVM use the single-context version without
2861          * checking for support, so declare that it is supported even
2862          * though it is treated as global context.  The alternative is
2863          * not failing the single-context invvpid, and it is worse.
2864          */
2865         if (enable_vpid) {
2866                 vmx->nested.nested_vmx_secondary_ctls_high |=
2867                         SECONDARY_EXEC_ENABLE_VPID;
2868                 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2869                         VMX_VPID_EXTENT_SUPPORTED_MASK;
2870         }
2871
2872         if (enable_unrestricted_guest)
2873                 vmx->nested.nested_vmx_secondary_ctls_high |=
2874                         SECONDARY_EXEC_UNRESTRICTED_GUEST;
2875
2876         /* miscellaneous data */
2877         rdmsr(MSR_IA32_VMX_MISC,
2878                 vmx->nested.nested_vmx_misc_low,
2879                 vmx->nested.nested_vmx_misc_high);
2880         vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2881         vmx->nested.nested_vmx_misc_low |=
2882                 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2883                 VMX_MISC_ACTIVITY_HLT;
2884         vmx->nested.nested_vmx_misc_high = 0;
2885
2886         /*
2887          * This MSR reports some information about VMX support. We
2888          * should return information about the VMX we emulate for the
2889          * guest, and the VMCS structure we give it - not about the
2890          * VMX support of the underlying hardware.
2891          */
2892         vmx->nested.nested_vmx_basic =
2893                 VMCS12_REVISION |
2894                 VMX_BASIC_TRUE_CTLS |
2895                 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2896                 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2897
2898         if (cpu_has_vmx_basic_inout())
2899                 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2900
2901         /*
2902          * These MSRs specify bits which the guest must keep fixed on
2903          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2904          * We picked the standard core2 setting.
2905          */
2906 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2907 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
2908         vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2909         vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2910
2911         /* These MSRs specify bits which the guest must keep fixed off. */
2912         rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2913         rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2914
2915         /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2916         vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2917 }
2918
2919 /*
2920  * if fixed0[i] == 1: val[i] must be 1
2921  * if fixed1[i] == 0: val[i] must be 0
2922  */
2923 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2924 {
2925         return ((val & fixed1) | fixed0) == val;
2926 }
2927
2928 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2929 {
2930         return fixed_bits_valid(control, low, high);
2931 }
2932
2933 static inline u64 vmx_control_msr(u32 low, u32 high)
2934 {
2935         return low | ((u64)high << 32);
2936 }
2937
2938 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2939 {
2940         superset &= mask;
2941         subset &= mask;
2942
2943         return (superset | subset) == superset;
2944 }
2945
2946 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2947 {
2948         const u64 feature_and_reserved =
2949                 /* feature (except bit 48; see below) */
2950                 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2951                 /* reserved */
2952                 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2953         u64 vmx_basic = vmx->nested.nested_vmx_basic;
2954
2955         if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2956                 return -EINVAL;
2957
2958         /*
2959          * KVM does not emulate a version of VMX that constrains physical
2960          * addresses of VMX structures (e.g. VMCS) to 32-bits.
2961          */
2962         if (data & BIT_ULL(48))
2963                 return -EINVAL;
2964
2965         if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2966             vmx_basic_vmcs_revision_id(data))
2967                 return -EINVAL;
2968
2969         if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2970                 return -EINVAL;
2971
2972         vmx->nested.nested_vmx_basic = data;
2973         return 0;
2974 }
2975
2976 static int
2977 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2978 {
2979         u64 supported;
2980         u32 *lowp, *highp;
2981
2982         switch (msr_index) {
2983         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2984                 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2985                 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2986                 break;
2987         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2988                 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2989                 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2990                 break;
2991         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2992                 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2993                 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2994                 break;
2995         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2996                 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2997                 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2998                 break;
2999         case MSR_IA32_VMX_PROCBASED_CTLS2:
3000                 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
3001                 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
3002                 break;
3003         default:
3004                 BUG();
3005         }
3006
3007         supported = vmx_control_msr(*lowp, *highp);
3008
3009         /* Check must-be-1 bits are still 1. */
3010         if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
3011                 return -EINVAL;
3012
3013         /* Check must-be-0 bits are still 0. */
3014         if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3015                 return -EINVAL;
3016
3017         *lowp = data;
3018         *highp = data >> 32;
3019         return 0;
3020 }
3021
3022 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3023 {
3024         const u64 feature_and_reserved_bits =
3025                 /* feature */
3026                 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3027                 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3028                 /* reserved */
3029                 GENMASK_ULL(13, 9) | BIT_ULL(31);
3030         u64 vmx_misc;
3031
3032         vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
3033                                    vmx->nested.nested_vmx_misc_high);
3034
3035         if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3036                 return -EINVAL;
3037
3038         if ((vmx->nested.nested_vmx_pinbased_ctls_high &
3039              PIN_BASED_VMX_PREEMPTION_TIMER) &&
3040             vmx_misc_preemption_timer_rate(data) !=
3041             vmx_misc_preemption_timer_rate(vmx_misc))
3042                 return -EINVAL;
3043
3044         if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3045                 return -EINVAL;
3046
3047         if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3048                 return -EINVAL;
3049
3050         if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3051                 return -EINVAL;
3052
3053         vmx->nested.nested_vmx_misc_low = data;
3054         vmx->nested.nested_vmx_misc_high = data >> 32;
3055         return 0;
3056 }
3057
3058 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3059 {
3060         u64 vmx_ept_vpid_cap;
3061
3062         vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
3063                                            vmx->nested.nested_vmx_vpid_caps);
3064
3065         /* Every bit is either reserved or a feature bit. */
3066         if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3067                 return -EINVAL;
3068
3069         vmx->nested.nested_vmx_ept_caps = data;
3070         vmx->nested.nested_vmx_vpid_caps = data >> 32;
3071         return 0;
3072 }
3073
3074 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3075 {
3076         u64 *msr;
3077
3078         switch (msr_index) {
3079         case MSR_IA32_VMX_CR0_FIXED0:
3080                 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3081                 break;
3082         case MSR_IA32_VMX_CR4_FIXED0:
3083                 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3084                 break;
3085         default:
3086                 BUG();
3087         }
3088
3089         /*
3090          * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3091          * must be 1 in the restored value.
3092          */
3093         if (!is_bitwise_subset(data, *msr, -1ULL))
3094                 return -EINVAL;
3095
3096         *msr = data;
3097         return 0;
3098 }
3099
3100 /*
3101  * Called when userspace is restoring VMX MSRs.
3102  *
3103  * Returns 0 on success, non-0 otherwise.
3104  */
3105 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3106 {
3107         struct vcpu_vmx *vmx = to_vmx(vcpu);
3108
3109         switch (msr_index) {
3110         case MSR_IA32_VMX_BASIC:
3111                 return vmx_restore_vmx_basic(vmx, data);
3112         case MSR_IA32_VMX_PINBASED_CTLS:
3113         case MSR_IA32_VMX_PROCBASED_CTLS:
3114         case MSR_IA32_VMX_EXIT_CTLS:
3115         case MSR_IA32_VMX_ENTRY_CTLS:
3116                 /*
3117                  * The "non-true" VMX capability MSRs are generated from the
3118                  * "true" MSRs, so we do not support restoring them directly.
3119                  *
3120                  * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3121                  * should restore the "true" MSRs with the must-be-1 bits
3122                  * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3123                  * DEFAULT SETTINGS".
3124                  */
3125                 return -EINVAL;
3126         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3127         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3128         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3129         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3130         case MSR_IA32_VMX_PROCBASED_CTLS2:
3131                 return vmx_restore_control_msr(vmx, msr_index, data);
3132         case MSR_IA32_VMX_MISC:
3133                 return vmx_restore_vmx_misc(vmx, data);
3134         case MSR_IA32_VMX_CR0_FIXED0:
3135         case MSR_IA32_VMX_CR4_FIXED0:
3136                 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3137         case MSR_IA32_VMX_CR0_FIXED1:
3138         case MSR_IA32_VMX_CR4_FIXED1:
3139                 /*
3140                  * These MSRs are generated based on the vCPU's CPUID, so we
3141                  * do not support restoring them directly.
3142                  */
3143                 return -EINVAL;
3144         case MSR_IA32_VMX_EPT_VPID_CAP:
3145                 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3146         case MSR_IA32_VMX_VMCS_ENUM:
3147                 vmx->nested.nested_vmx_vmcs_enum = data;
3148                 return 0;
3149         default:
3150                 /*
3151                  * The rest of the VMX capability MSRs do not support restore.
3152                  */
3153                 return -EINVAL;
3154         }
3155 }
3156
3157 /* Returns 0 on success, non-0 otherwise. */
3158 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3159 {
3160         struct vcpu_vmx *vmx = to_vmx(vcpu);
3161
3162         switch (msr_index) {
3163         case MSR_IA32_VMX_BASIC:
3164                 *pdata = vmx->nested.nested_vmx_basic;
3165                 break;
3166         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3167         case MSR_IA32_VMX_PINBASED_CTLS:
3168                 *pdata = vmx_control_msr(
3169                         vmx->nested.nested_vmx_pinbased_ctls_low,
3170                         vmx->nested.nested_vmx_pinbased_ctls_high);
3171                 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3172                         *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3173                 break;
3174         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3175         case MSR_IA32_VMX_PROCBASED_CTLS:
3176                 *pdata = vmx_control_msr(
3177                         vmx->nested.nested_vmx_procbased_ctls_low,
3178                         vmx->nested.nested_vmx_procbased_ctls_high);
3179                 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3180                         *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3181                 break;
3182         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3183         case MSR_IA32_VMX_EXIT_CTLS:
3184                 *pdata = vmx_control_msr(
3185                         vmx->nested.nested_vmx_exit_ctls_low,
3186                         vmx->nested.nested_vmx_exit_ctls_high);
3187                 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3188                         *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3189                 break;
3190         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3191         case MSR_IA32_VMX_ENTRY_CTLS:
3192                 *pdata = vmx_control_msr(
3193                         vmx->nested.nested_vmx_entry_ctls_low,
3194                         vmx->nested.nested_vmx_entry_ctls_high);
3195                 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3196                         *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3197                 break;
3198         case MSR_IA32_VMX_MISC:
3199                 *pdata = vmx_control_msr(
3200                         vmx->nested.nested_vmx_misc_low,
3201                         vmx->nested.nested_vmx_misc_high);
3202                 break;
3203         case MSR_IA32_VMX_CR0_FIXED0:
3204                 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3205                 break;
3206         case MSR_IA32_VMX_CR0_FIXED1:
3207                 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3208                 break;
3209         case MSR_IA32_VMX_CR4_FIXED0:
3210                 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3211                 break;
3212         case MSR_IA32_VMX_CR4_FIXED1:
3213                 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3214                 break;
3215         case MSR_IA32_VMX_VMCS_ENUM:
3216                 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3217                 break;
3218         case MSR_IA32_VMX_PROCBASED_CTLS2:
3219                 *pdata = vmx_control_msr(
3220                         vmx->nested.nested_vmx_secondary_ctls_low,
3221                         vmx->nested.nested_vmx_secondary_ctls_high);
3222                 break;
3223         case MSR_IA32_VMX_EPT_VPID_CAP:
3224                 *pdata = vmx->nested.nested_vmx_ept_caps |
3225                         ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3226                 break;
3227         case MSR_IA32_VMX_VMFUNC:
3228                 *pdata = vmx->nested.nested_vmx_vmfunc_controls;
3229                 break;
3230         default:
3231                 return 1;
3232         }
3233
3234         return 0;
3235 }
3236
3237 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3238                                                  uint64_t val)
3239 {
3240         uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3241
3242         return !(val & ~valid_bits);
3243 }
3244
3245 /*
3246  * Reads an msr value (of 'msr_index') into 'pdata'.
3247  * Returns 0 on success, non-0 otherwise.
3248  * Assumes vcpu_load() was already called.
3249  */
3250 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3251 {
3252         struct shared_msr_entry *msr;
3253
3254         switch (msr_info->index) {
3255 #ifdef CONFIG_X86_64
3256         case MSR_FS_BASE:
3257                 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3258                 break;
3259         case MSR_GS_BASE:
3260                 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3261                 break;
3262         case MSR_KERNEL_GS_BASE:
3263                 vmx_load_host_state(to_vmx(vcpu));
3264                 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3265                 break;
3266 #endif
3267         case MSR_EFER:
3268                 return kvm_get_msr_common(vcpu, msr_info);
3269         case MSR_IA32_TSC:
3270                 msr_info->data = guest_read_tsc(vcpu);
3271                 break;
3272         case MSR_IA32_SYSENTER_CS:
3273                 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3274                 break;
3275         case MSR_IA32_SYSENTER_EIP:
3276                 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3277                 break;
3278         case MSR_IA32_SYSENTER_ESP:
3279                 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3280                 break;
3281         case MSR_IA32_BNDCFGS:
3282                 if (!kvm_mpx_supported() ||
3283                     (!msr_info->host_initiated &&
3284                      !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3285                         return 1;
3286                 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3287                 break;
3288         case MSR_IA32_MCG_EXT_CTL:
3289                 if (!msr_info->host_initiated &&
3290                     !(to_vmx(vcpu)->msr_ia32_feature_control &
3291                       FEATURE_CONTROL_LMCE))
3292                         return 1;
3293                 msr_info->data = vcpu->arch.mcg_ext_ctl;
3294                 break;
3295         case MSR_IA32_FEATURE_CONTROL:
3296                 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3297                 break;
3298         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3299                 if (!nested_vmx_allowed(vcpu))
3300                         return 1;
3301                 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3302         case MSR_IA32_XSS:
3303                 if (!vmx_xsaves_supported())
3304                         return 1;
3305                 msr_info->data = vcpu->arch.ia32_xss;
3306                 break;
3307         case MSR_TSC_AUX:
3308                 if (!msr_info->host_initiated &&
3309                     !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3310                         return 1;
3311                 /* Otherwise falls through */
3312         default:
3313                 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3314                 if (msr) {
3315                         msr_info->data = msr->data;
3316                         break;
3317                 }
3318                 return kvm_get_msr_common(vcpu, msr_info);
3319         }
3320
3321         return 0;
3322 }
3323
3324 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3325
3326 /*
3327  * Writes msr value into into the appropriate "register".
3328  * Returns 0 on success, non-0 otherwise.
3329  * Assumes vcpu_load() was already called.
3330  */
3331 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3332 {
3333         struct vcpu_vmx *vmx = to_vmx(vcpu);
3334         struct shared_msr_entry *msr;
3335         int ret = 0;
3336         u32 msr_index = msr_info->index;
3337         u64 data = msr_info->data;
3338
3339         switch (msr_index) {
3340         case MSR_EFER:
3341                 ret = kvm_set_msr_common(vcpu, msr_info);
3342                 break;
3343 #ifdef CONFIG_X86_64
3344         case MSR_FS_BASE:
3345                 vmx_segment_cache_clear(vmx);
3346                 vmcs_writel(GUEST_FS_BASE, data);
3347                 break;
3348         case MSR_GS_BASE:
3349                 vmx_segment_cache_clear(vmx);
3350                 vmcs_writel(GUEST_GS_BASE, data);
3351                 break;
3352         case MSR_KERNEL_GS_BASE:
3353                 vmx_load_host_state(vmx);
3354                 vmx->msr_guest_kernel_gs_base = data;
3355                 break;
3356 #endif
3357         case MSR_IA32_SYSENTER_CS:
3358                 vmcs_write32(GUEST_SYSENTER_CS, data);
3359                 break;
3360         case MSR_IA32_SYSENTER_EIP:
3361                 vmcs_writel(GUEST_SYSENTER_EIP, data);
3362                 break;
3363         case MSR_IA32_SYSENTER_ESP:
3364                 vmcs_writel(GUEST_SYSENTER_ESP, data);
3365                 break;
3366         case MSR_IA32_BNDCFGS:
3367                 if (!kvm_mpx_supported() ||
3368                     (!msr_info->host_initiated &&
3369                      !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3370                         return 1;
3371                 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
3372                     (data & MSR_IA32_BNDCFGS_RSVD))
3373                         return 1;
3374                 vmcs_write64(GUEST_BNDCFGS, data);
3375                 break;
3376         case MSR_IA32_TSC:
3377                 kvm_write_tsc(vcpu, msr_info);
3378                 break;
3379         case MSR_IA32_CR_PAT:
3380                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3381                         if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3382                                 return 1;
3383                         vmcs_write64(GUEST_IA32_PAT, data);
3384                         vcpu->arch.pat = data;
3385                         break;
3386                 }
3387                 ret = kvm_set_msr_common(vcpu, msr_info);
3388                 break;
3389         case MSR_IA32_TSC_ADJUST:
3390                 ret = kvm_set_msr_common(vcpu, msr_info);
3391                 break;
3392         case MSR_IA32_MCG_EXT_CTL:
3393                 if ((!msr_info->host_initiated &&
3394                      !(to_vmx(vcpu)->msr_ia32_feature_control &
3395                        FEATURE_CONTROL_LMCE)) ||
3396                     (data & ~MCG_EXT_CTL_LMCE_EN))
3397                         return 1;
3398                 vcpu->arch.mcg_ext_ctl = data;
3399                 break;
3400         case MSR_IA32_FEATURE_CONTROL:
3401                 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3402                     (to_vmx(vcpu)->msr_ia32_feature_control &
3403                      FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3404                         return 1;
3405                 vmx->msr_ia32_feature_control = data;
3406                 if (msr_info->host_initiated && data == 0)
3407                         vmx_leave_nested(vcpu);
3408                 break;
3409         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3410                 if (!msr_info->host_initiated)
3411                         return 1; /* they are read-only */
3412                 if (!nested_vmx_allowed(vcpu))
3413                         return 1;
3414                 return vmx_set_vmx_msr(vcpu, msr_index, data);
3415         case MSR_IA32_XSS:
3416                 if (!vmx_xsaves_supported())
3417                         return 1;
3418                 /*
3419                  * The only supported bit as of Skylake is bit 8, but
3420                  * it is not supported on KVM.
3421                  */
3422                 if (data != 0)
3423                         return 1;
3424                 vcpu->arch.ia32_xss = data;
3425                 if (vcpu->arch.ia32_xss != host_xss)
3426                         add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3427                                 vcpu->arch.ia32_xss, host_xss);
3428                 else
3429                         clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3430                 break;
3431         case MSR_TSC_AUX:
3432                 if (!msr_info->host_initiated &&
3433                     !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3434                         return 1;
3435                 /* Check reserved bit, higher 32 bits should be zero */
3436                 if ((data >> 32) != 0)
3437                         return 1;
3438                 /* Otherwise falls through */
3439         default:
3440                 msr = find_msr_entry(vmx, msr_index);
3441                 if (msr) {
3442                         u64 old_msr_data = msr->data;
3443                         msr->data = data;
3444                         if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3445                                 preempt_disable();
3446                                 ret = kvm_set_shared_msr(msr->index, msr->data,
3447                                                          msr->mask);
3448                                 preempt_enable();
3449                                 if (ret)
3450                                         msr->data = old_msr_data;
3451                         }
3452                         break;
3453                 }
3454                 ret = kvm_set_msr_common(vcpu, msr_info);
3455         }
3456
3457         return ret;
3458 }
3459
3460 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3461 {
3462         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3463         switch (reg) {
3464         case VCPU_REGS_RSP:
3465                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3466                 break;
3467         case VCPU_REGS_RIP:
3468                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3469                 break;
3470         case VCPU_EXREG_PDPTR:
3471                 if (enable_ept)
3472                         ept_save_pdptrs(vcpu);
3473                 break;
3474         default:
3475                 break;
3476         }
3477 }
3478
3479 static __init int cpu_has_kvm_support(void)
3480 {
3481         return cpu_has_vmx();
3482 }
3483
3484 static __init int vmx_disabled_by_bios(void)
3485 {
3486         u64 msr;
3487
3488         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3489         if (msr & FEATURE_CONTROL_LOCKED) {
3490                 /* launched w/ TXT and VMX disabled */
3491                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3492                         && tboot_enabled())
3493                         return 1;
3494                 /* launched w/o TXT and VMX only enabled w/ TXT */
3495                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3496                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3497                         && !tboot_enabled()) {
3498                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3499                                 "activate TXT before enabling KVM\n");
3500                         return 1;
3501                 }
3502                 /* launched w/o TXT and VMX disabled */
3503                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3504                         && !tboot_enabled())
3505                         return 1;
3506         }
3507
3508         return 0;
3509 }
3510
3511 static void kvm_cpu_vmxon(u64 addr)
3512 {
3513         cr4_set_bits(X86_CR4_VMXE);
3514         intel_pt_handle_vmx(1);
3515
3516         asm volatile (ASM_VMX_VMXON_RAX
3517                         : : "a"(&addr), "m"(addr)
3518                         : "memory", "cc");
3519 }
3520
3521 static int hardware_enable(void)
3522 {
3523         int cpu = raw_smp_processor_id();
3524         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3525         u64 old, test_bits;
3526
3527         if (cr4_read_shadow() & X86_CR4_VMXE)
3528                 return -EBUSY;
3529
3530         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3531         INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3532         spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3533
3534         /*
3535          * Now we can enable the vmclear operation in kdump
3536          * since the loaded_vmcss_on_cpu list on this cpu
3537          * has been initialized.
3538          *
3539          * Though the cpu is not in VMX operation now, there
3540          * is no problem to enable the vmclear operation
3541          * for the loaded_vmcss_on_cpu list is empty!
3542          */
3543         crash_enable_local_vmclear(cpu);
3544
3545         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3546
3547         test_bits = FEATURE_CONTROL_LOCKED;
3548         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3549         if (tboot_enabled())
3550                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3551
3552         if ((old & test_bits) != test_bits) {
3553                 /* enable and lock */
3554                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3555         }
3556         kvm_cpu_vmxon(phys_addr);
3557         if (enable_ept)
3558                 ept_sync_global();
3559
3560         return 0;
3561 }
3562
3563 static void vmclear_local_loaded_vmcss(void)
3564 {
3565         int cpu = raw_smp_processor_id();
3566         struct loaded_vmcs *v, *n;
3567
3568         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3569                                  loaded_vmcss_on_cpu_link)
3570                 __loaded_vmcs_clear(v);
3571 }
3572
3573
3574 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3575  * tricks.
3576  */
3577 static void kvm_cpu_vmxoff(void)
3578 {
3579         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3580
3581         intel_pt_handle_vmx(0);
3582         cr4_clear_bits(X86_CR4_VMXE);
3583 }
3584
3585 static void hardware_disable(void)
3586 {
3587         vmclear_local_loaded_vmcss();
3588         kvm_cpu_vmxoff();
3589 }
3590
3591 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3592                                       u32 msr, u32 *result)
3593 {
3594         u32 vmx_msr_low, vmx_msr_high;
3595         u32 ctl = ctl_min | ctl_opt;
3596
3597         rdmsr(msr, vmx_msr_low, vmx_msr_high);
3598
3599         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3600         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
3601
3602         /* Ensure minimum (required) set of control bits are supported. */
3603         if (ctl_min & ~ctl)
3604                 return -EIO;
3605
3606         *result = ctl;
3607         return 0;
3608 }
3609
3610 static __init bool allow_1_setting(u32 msr, u32 ctl)
3611 {
3612         u32 vmx_msr_low, vmx_msr_high;
3613
3614         rdmsr(msr, vmx_msr_low, vmx_msr_high);
3615         return vmx_msr_high & ctl;
3616 }
3617
3618 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3619 {
3620         u32 vmx_msr_low, vmx_msr_high;
3621         u32 min, opt, min2, opt2;
3622         u32 _pin_based_exec_control = 0;
3623         u32 _cpu_based_exec_control = 0;
3624         u32 _cpu_based_2nd_exec_control = 0;
3625         u32 _vmexit_control = 0;
3626         u32 _vmentry_control = 0;
3627
3628         min = CPU_BASED_HLT_EXITING |
3629 #ifdef CONFIG_X86_64
3630               CPU_BASED_CR8_LOAD_EXITING |
3631               CPU_BASED_CR8_STORE_EXITING |
3632 #endif
3633               CPU_BASED_CR3_LOAD_EXITING |
3634               CPU_BASED_CR3_STORE_EXITING |
3635               CPU_BASED_USE_IO_BITMAPS |
3636               CPU_BASED_MOV_DR_EXITING |
3637               CPU_BASED_USE_TSC_OFFSETING |
3638               CPU_BASED_INVLPG_EXITING |
3639               CPU_BASED_RDPMC_EXITING;
3640
3641         if (!kvm_mwait_in_guest())
3642                 min |= CPU_BASED_MWAIT_EXITING |
3643                         CPU_BASED_MONITOR_EXITING;
3644
3645         opt = CPU_BASED_TPR_SHADOW |
3646               CPU_BASED_USE_MSR_BITMAPS |
3647               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3648         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3649                                 &_cpu_based_exec_control) < 0)
3650                 return -EIO;
3651 #ifdef CONFIG_X86_64
3652         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3653                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3654                                            ~CPU_BASED_CR8_STORE_EXITING;
3655 #endif
3656         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3657                 min2 = 0;
3658                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3659                         SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3660                         SECONDARY_EXEC_WBINVD_EXITING |
3661                         SECONDARY_EXEC_ENABLE_VPID |
3662                         SECONDARY_EXEC_ENABLE_EPT |
3663                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
3664                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3665                         SECONDARY_EXEC_RDTSCP |
3666                         SECONDARY_EXEC_ENABLE_INVPCID |
3667                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
3668                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3669                         SECONDARY_EXEC_SHADOW_VMCS |
3670                         SECONDARY_EXEC_XSAVES |
3671                         SECONDARY_EXEC_RDSEED_EXITING |
3672                         SECONDARY_EXEC_RDRAND_EXITING |
3673                         SECONDARY_EXEC_ENABLE_PML |
3674                         SECONDARY_EXEC_TSC_SCALING |
3675                         SECONDARY_EXEC_ENABLE_VMFUNC;
3676                 if (adjust_vmx_controls(min2, opt2,
3677                                         MSR_IA32_VMX_PROCBASED_CTLS2,
3678                                         &_cpu_based_2nd_exec_control) < 0)
3679                         return -EIO;
3680         }
3681 #ifndef CONFIG_X86_64
3682         if (!(_cpu_based_2nd_exec_control &
3683                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3684                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3685 #endif
3686
3687         if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3688                 _cpu_based_2nd_exec_control &= ~(
3689                                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3690                                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3691                                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3692
3693         rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
3694                 &vmx_capability.ept, &vmx_capability.vpid);
3695
3696         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3697                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3698                    enabled */
3699                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3700                                              CPU_BASED_CR3_STORE_EXITING |
3701                                              CPU_BASED_INVLPG_EXITING);
3702         } else if (vmx_capability.ept) {
3703                 vmx_capability.ept = 0;
3704                 pr_warn_once("EPT CAP should not exist if not support "
3705                                 "1-setting enable EPT VM-execution control\n");
3706         }
3707         if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
3708                 vmx_capability.vpid) {
3709                 vmx_capability.vpid = 0;
3710                 pr_warn_once("VPID CAP should not exist if not support "
3711                                 "1-setting enable VPID VM-execution control\n");
3712         }
3713
3714         min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3715 #ifdef CONFIG_X86_64
3716         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3717 #endif
3718         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3719                 VM_EXIT_CLEAR_BNDCFGS;
3720         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3721                                 &_vmexit_control) < 0)
3722                 return -EIO;
3723
3724         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3725         opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
3726                  PIN_BASED_VMX_PREEMPTION_TIMER;
3727         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3728                                 &_pin_based_exec_control) < 0)
3729                 return -EIO;
3730
3731         if (cpu_has_broken_vmx_preemption_timer())
3732                 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3733         if (!(_cpu_based_2nd_exec_control &
3734                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3735                 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3736
3737         min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3738         opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3739         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3740                                 &_vmentry_control) < 0)
3741                 return -EIO;
3742
3743         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3744
3745         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3746         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3747                 return -EIO;
3748
3749 #ifdef CONFIG_X86_64
3750         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3751         if (vmx_msr_high & (1u<<16))
3752                 return -EIO;
3753 #endif
3754
3755         /* Require Write-Back (WB) memory type for VMCS accesses. */
3756         if (((vmx_msr_high >> 18) & 15) != 6)
3757                 return -EIO;
3758
3759         vmcs_conf->size = vmx_msr_high & 0x1fff;
3760         vmcs_conf->order = get_order(vmcs_conf->size);
3761         vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3762         vmcs_conf->revision_id = vmx_msr_low;
3763
3764         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3765         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3766         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3767         vmcs_conf->vmexit_ctrl         = _vmexit_control;
3768         vmcs_conf->vmentry_ctrl        = _vmentry_control;
3769
3770         cpu_has_load_ia32_efer =
3771                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3772                                 VM_ENTRY_LOAD_IA32_EFER)
3773                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3774                                    VM_EXIT_LOAD_IA32_EFER);
3775
3776         cpu_has_load_perf_global_ctrl =
3777                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3778                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3779                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3780                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3781
3782         /*
3783          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3784          * but due to errata below it can't be used. Workaround is to use
3785          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3786          *
3787          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3788          *
3789          * AAK155             (model 26)
3790          * AAP115             (model 30)
3791          * AAT100             (model 37)
3792          * BC86,AAY89,BD102   (model 44)
3793          * BA97               (model 46)
3794          *
3795          */
3796         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3797                 switch (boot_cpu_data.x86_model) {
3798                 case 26:
3799                 case 30:
3800                 case 37:
3801                 case 44:
3802                 case 46:
3803                         cpu_has_load_perf_global_ctrl = false;
3804                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3805                                         "does not work properly. Using workaround\n");
3806                         break;
3807                 default:
3808                         break;
3809                 }
3810         }
3811
3812         if (boot_cpu_has(X86_FEATURE_XSAVES))
3813                 rdmsrl(MSR_IA32_XSS, host_xss);
3814
3815         return 0;
3816 }
3817
3818 static struct vmcs *alloc_vmcs_cpu(int cpu)
3819 {
3820         int node = cpu_to_node(cpu);
3821         struct page *pages;
3822         struct vmcs *vmcs;
3823
3824         pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3825         if (!pages)
3826                 return NULL;
3827         vmcs = page_address(pages);
3828         memset(vmcs, 0, vmcs_config.size);
3829         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3830         return vmcs;
3831 }
3832
3833 static struct vmcs *alloc_vmcs(void)
3834 {
3835         return alloc_vmcs_cpu(raw_smp_processor_id());
3836 }
3837
3838 static void free_vmcs(struct vmcs *vmcs)
3839 {
3840         free_pages((unsigned long)vmcs, vmcs_config.order);
3841 }
3842
3843 /*
3844  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3845  */
3846 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3847 {
3848         if (!loaded_vmcs->vmcs)
3849                 return;
3850         loaded_vmcs_clear(loaded_vmcs);
3851         free_vmcs(loaded_vmcs->vmcs);
3852         loaded_vmcs->vmcs = NULL;
3853         WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3854 }
3855
3856 static void free_kvm_area(void)
3857 {
3858         int cpu;
3859
3860         for_each_possible_cpu(cpu) {
3861                 free_vmcs(per_cpu(vmxarea, cpu));
3862                 per_cpu(vmxarea, cpu) = NULL;
3863         }
3864 }
3865
3866 enum vmcs_field_type {
3867         VMCS_FIELD_TYPE_U16 = 0,
3868         VMCS_FIELD_TYPE_U64 = 1,
3869         VMCS_FIELD_TYPE_U32 = 2,
3870         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
3871 };
3872
3873 static inline int vmcs_field_type(unsigned long field)
3874 {
3875         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
3876                 return VMCS_FIELD_TYPE_U32;
3877         return (field >> 13) & 0x3 ;
3878 }
3879
3880 static inline int vmcs_field_readonly(unsigned long field)
3881 {
3882         return (((field >> 10) & 0x3) == 1);
3883 }
3884
3885 static void init_vmcs_shadow_fields(void)
3886 {
3887         int i, j;
3888
3889         /* No checks for read only fields yet */
3890
3891         for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3892                 switch (shadow_read_write_fields[i]) {
3893                 case GUEST_BNDCFGS:
3894                         if (!kvm_mpx_supported())
3895                                 continue;
3896                         break;
3897                 default:
3898                         break;
3899                 }
3900
3901                 if (j < i)
3902                         shadow_read_write_fields[j] =
3903                                 shadow_read_write_fields[i];
3904                 j++;
3905         }
3906         max_shadow_read_write_fields = j;
3907
3908         /* shadowed fields guest access without vmexit */
3909         for (i = 0; i < max_shadow_read_write_fields; i++) {
3910                 unsigned long field = shadow_read_write_fields[i];
3911
3912                 clear_bit(field, vmx_vmwrite_bitmap);
3913                 clear_bit(field, vmx_vmread_bitmap);
3914                 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64) {
3915                         clear_bit(field + 1, vmx_vmwrite_bitmap);
3916                         clear_bit(field + 1, vmx_vmread_bitmap);
3917                 }
3918         }
3919         for (i = 0; i < max_shadow_read_only_fields; i++) {
3920                 unsigned long field = shadow_read_only_fields[i];
3921
3922                 clear_bit(field, vmx_vmread_bitmap);
3923                 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64)
3924                         clear_bit(field + 1, vmx_vmread_bitmap);
3925         }
3926 }
3927
3928 static __init int alloc_kvm_area(void)
3929 {
3930         int cpu;
3931
3932         for_each_possible_cpu(cpu) {
3933                 struct vmcs *vmcs;
3934
3935                 vmcs = alloc_vmcs_cpu(cpu);
3936                 if (!vmcs) {
3937                         free_kvm_area();
3938                         return -ENOMEM;
3939                 }
3940
3941                 per_cpu(vmxarea, cpu) = vmcs;
3942         }
3943         return 0;
3944 }
3945
3946 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3947                 struct kvm_segment *save)
3948 {
3949         if (!emulate_invalid_guest_state) {
3950                 /*
3951                  * CS and SS RPL should be equal during guest entry according
3952                  * to VMX spec, but in reality it is not always so. Since vcpu
3953                  * is in the middle of the transition from real mode to
3954                  * protected mode it is safe to assume that RPL 0 is a good
3955                  * default value.
3956                  */
3957                 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3958                         save->selector &= ~SEGMENT_RPL_MASK;
3959                 save->dpl = save->selector & SEGMENT_RPL_MASK;
3960                 save->s = 1;
3961         }
3962         vmx_set_segment(vcpu, save, seg);
3963 }
3964
3965 static void enter_pmode(struct kvm_vcpu *vcpu)
3966 {
3967         unsigned long flags;
3968         struct vcpu_vmx *vmx = to_vmx(vcpu);
3969
3970         /*
3971          * Update real mode segment cache. It may be not up-to-date if sement
3972          * register was written while vcpu was in a guest mode.
3973          */
3974         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3975         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3976         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3977         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3978         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3979         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3980
3981         vmx->rmode.vm86_active = 0;
3982
3983         vmx_segment_cache_clear(vmx);
3984
3985         vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3986
3987         flags = vmcs_readl(GUEST_RFLAGS);
3988         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3989         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3990         vmcs_writel(GUEST_RFLAGS, flags);
3991
3992         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3993                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3994
3995         update_exception_bitmap(vcpu);
3996
3997         fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3998         fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3999         fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4000         fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4001         fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4002         fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4003 }
4004
4005 static void fix_rmode_seg(int seg, struct kvm_segment *save)
4006 {
4007         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4008         struct kvm_segment var = *save;
4009
4010         var.dpl = 0x3;
4011         if (seg == VCPU_SREG_CS)
4012                 var.type = 0x3;
4013
4014         if (!emulate_invalid_guest_state) {
4015                 var.selector = var.base >> 4;
4016                 var.base = var.base & 0xffff0;
4017                 var.limit = 0xffff;
4018                 var.g = 0;
4019                 var.db = 0;
4020                 var.present = 1;
4021                 var.s = 1;
4022                 var.l = 0;
4023                 var.unusable = 0;
4024                 var.type = 0x3;
4025                 var.avl = 0;
4026                 if (save->base & 0xf)
4027                         printk_once(KERN_WARNING "kvm: segment base is not "
4028                                         "paragraph aligned when entering "
4029                                         "protected mode (seg=%d)", seg);
4030         }
4031
4032         vmcs_write16(sf->selector, var.selector);
4033         vmcs_writel(sf->base, var.base);
4034         vmcs_write32(sf->limit, var.limit);
4035         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4036 }
4037
4038 static void enter_rmode(struct kvm_vcpu *vcpu)
4039 {
4040         unsigned long flags;
4041         struct vcpu_vmx *vmx = to_vmx(vcpu);
4042
4043         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4044         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4045         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4046         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4047         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4048         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4049         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4050
4051         vmx->rmode.vm86_active = 1;
4052
4053         /*
4054          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4055          * vcpu. Warn the user that an update is overdue.
4056          */
4057         if (!vcpu->kvm->arch.tss_addr)
4058                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4059                              "called before entering vcpu\n");
4060
4061         vmx_segment_cache_clear(vmx);
4062
4063         vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
4064         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4065         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4066
4067         flags = vmcs_readl(GUEST_RFLAGS);
4068         vmx->rmode.save_rflags = flags;
4069
4070         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4071
4072         vmcs_writel(GUEST_RFLAGS, flags);
4073         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4074         update_exception_bitmap(vcpu);
4075
4076         fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4077         fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4078         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4079         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4080         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4081         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4082
4083         kvm_mmu_reset_context(vcpu);
4084 }
4085
4086 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4087 {
4088         struct vcpu_vmx *vmx = to_vmx(vcpu);
4089         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4090
4091         if (!msr)
4092                 return;
4093
4094         /*
4095          * Force kernel_gs_base reloading before EFER changes, as control
4096          * of this msr depends on is_long_mode().
4097          */
4098         vmx_load_host_state(to_vmx(vcpu));
4099         vcpu->arch.efer = efer;
4100         if (efer & EFER_LMA) {
4101                 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4102                 msr->data = efer;
4103         } else {
4104                 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4105
4106                 msr->data = efer & ~EFER_LME;
4107         }
4108         setup_msrs(vmx);
4109 }
4110
4111 #ifdef CONFIG_X86_64
4112
4113 static void enter_lmode(struct kvm_vcpu *vcpu)
4114 {
4115         u32 guest_tr_ar;
4116
4117         vmx_segment_cache_clear(to_vmx(vcpu));
4118
4119         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4120         if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4121                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4122                                      __func__);
4123                 vmcs_write32(GUEST_TR_AR_BYTES,
4124                              (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4125                              | VMX_AR_TYPE_BUSY_64_TSS);
4126         }
4127         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4128 }
4129
4130 static void exit_lmode(struct kvm_vcpu *vcpu)
4131 {
4132         vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4133         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4134 }
4135
4136 #endif
4137
4138 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4139 {
4140         if (enable_ept) {
4141                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4142                         return;
4143                 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4144         } else {
4145                 vpid_sync_context(vpid);
4146         }
4147 }
4148
4149 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4150 {
4151         __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4152 }
4153
4154 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4155 {
4156         if (enable_ept)
4157                 vmx_flush_tlb(vcpu);
4158 }
4159
4160 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4161 {
4162         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4163
4164         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4165         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4166 }
4167
4168 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4169 {
4170         if (enable_ept && is_paging(vcpu))
4171                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4172         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4173 }
4174
4175 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4176 {
4177         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4178
4179         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4180         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4181 }
4182
4183 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4184 {
4185         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4186
4187         if (!test_bit(VCPU_EXREG_PDPTR,
4188                       (unsigned long *)&vcpu->arch.regs_dirty))
4189                 return;
4190
4191         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4192                 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4193                 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4194                 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4195                 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4196         }
4197 }
4198
4199 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4200 {
4201         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4202
4203         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4204                 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4205                 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4206                 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4207                 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4208         }
4209
4210         __set_bit(VCPU_EXREG_PDPTR,
4211                   (unsigned long *)&vcpu->arch.regs_avail);
4212         __set_bit(VCPU_EXREG_PDPTR,
4213                   (unsigned long *)&vcpu->arch.regs_dirty);
4214 }
4215
4216 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4217 {
4218         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4219         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4220         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4221
4222         if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4223                 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4224             nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4225                 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4226
4227         return fixed_bits_valid(val, fixed0, fixed1);
4228 }
4229
4230 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4231 {
4232         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4233         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4234
4235         return fixed_bits_valid(val, fixed0, fixed1);
4236 }
4237
4238 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4239 {
4240         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4241         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4242
4243         return fixed_bits_valid(val, fixed0, fixed1);
4244 }
4245
4246 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4247 #define nested_guest_cr4_valid  nested_cr4_valid
4248 #define nested_host_cr4_valid   nested_cr4_valid
4249
4250 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4251
4252 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4253                                         unsigned long cr0,
4254                                         struct kvm_vcpu *vcpu)
4255 {
4256         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4257                 vmx_decache_cr3(vcpu);
4258         if (!(cr0 & X86_CR0_PG)) {
4259                 /* From paging/starting to nonpaging */
4260                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4261                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4262                              (CPU_BASED_CR3_LOAD_EXITING |
4263                               CPU_BASED_CR3_STORE_EXITING));
4264                 vcpu->arch.cr0 = cr0;
4265                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4266         } else if (!is_paging(vcpu)) {
4267                 /* From nonpaging to paging */
4268                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4269                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4270                              ~(CPU_BASED_CR3_LOAD_EXITING |
4271                                CPU_BASED_CR3_STORE_EXITING));
4272                 vcpu->arch.cr0 = cr0;
4273                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4274         }
4275
4276         if (!(cr0 & X86_CR0_WP))
4277                 *hw_cr0 &= ~X86_CR0_WP;
4278 }
4279
4280 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4281 {
4282         struct vcpu_vmx *vmx = to_vmx(vcpu);
4283         unsigned long hw_cr0;
4284
4285         hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4286         if (enable_unrestricted_guest)
4287                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4288         else {
4289                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4290
4291                 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4292                         enter_pmode(vcpu);
4293
4294                 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4295                         enter_rmode(vcpu);
4296         }
4297
4298 #ifdef CONFIG_X86_64
4299         if (vcpu->arch.efer & EFER_LME) {
4300                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4301                         enter_lmode(vcpu);
4302                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4303                         exit_lmode(vcpu);
4304         }
4305 #endif
4306
4307         if (enable_ept)
4308                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4309
4310         vmcs_writel(CR0_READ_SHADOW, cr0);
4311         vmcs_writel(GUEST_CR0, hw_cr0);
4312         vcpu->arch.cr0 = cr0;
4313
4314         /* depends on vcpu->arch.cr0 to be set to a new value */
4315         vmx->emulation_required = emulation_required(vcpu);
4316 }
4317
4318 static int get_ept_level(struct kvm_vcpu *vcpu)
4319 {
4320         if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
4321                 return 5;
4322         return 4;
4323 }
4324
4325 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4326 {
4327         u64 eptp = VMX_EPTP_MT_WB;
4328
4329         eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
4330
4331         if (enable_ept_ad_bits &&
4332             (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4333                 eptp |= VMX_EPTP_AD_ENABLE_BIT;
4334         eptp |= (root_hpa & PAGE_MASK);
4335
4336         return eptp;
4337 }
4338
4339 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4340 {
4341         unsigned long guest_cr3;
4342         u64 eptp;
4343
4344         guest_cr3 = cr3;
4345         if (enable_ept) {
4346                 eptp = construct_eptp(vcpu, cr3);
4347                 vmcs_write64(EPT_POINTER, eptp);
4348                 if (is_paging(vcpu) || is_guest_mode(vcpu))
4349                         guest_cr3 = kvm_read_cr3(vcpu);
4350                 else
4351                         guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4352                 ept_load_pdptrs(vcpu);
4353         }
4354
4355         vmx_flush_tlb(vcpu);
4356         vmcs_writel(GUEST_CR3, guest_cr3);
4357 }
4358
4359 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4360 {
4361         /*
4362          * Pass through host's Machine Check Enable value to hw_cr4, which
4363          * is in force while we are in guest mode.  Do not let guests control
4364          * this bit, even if host CR4.MCE == 0.
4365          */
4366         unsigned long hw_cr4 =
4367                 (cr4_read_shadow() & X86_CR4_MCE) |
4368                 (cr4 & ~X86_CR4_MCE) |
4369                 (to_vmx(vcpu)->rmode.vm86_active ?
4370                  KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4371
4372         if (cr4 & X86_CR4_VMXE) {
4373                 /*
4374                  * To use VMXON (and later other VMX instructions), a guest
4375                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
4376                  * So basically the check on whether to allow nested VMX
4377                  * is here.
4378                  */
4379                 if (!nested_vmx_allowed(vcpu))
4380                         return 1;
4381         }
4382
4383         if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4384                 return 1;
4385
4386         vcpu->arch.cr4 = cr4;
4387         if (enable_ept) {
4388                 if (!is_paging(vcpu)) {
4389                         hw_cr4 &= ~X86_CR4_PAE;
4390                         hw_cr4 |= X86_CR4_PSE;
4391                 } else if (!(cr4 & X86_CR4_PAE)) {
4392                         hw_cr4 &= ~X86_CR4_PAE;
4393                 }
4394         }
4395
4396         if (!enable_unrestricted_guest && !is_paging(vcpu))
4397                 /*
4398                  * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4399                  * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
4400                  * to be manually disabled when guest switches to non-paging
4401                  * mode.
4402                  *
4403                  * If !enable_unrestricted_guest, the CPU is always running
4404                  * with CR0.PG=1 and CR4 needs to be modified.
4405                  * If enable_unrestricted_guest, the CPU automatically
4406                  * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4407                  */
4408                 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4409
4410         vmcs_writel(CR4_READ_SHADOW, cr4);
4411         vmcs_writel(GUEST_CR4, hw_cr4);
4412         return 0;
4413 }
4414
4415 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4416                             struct kvm_segment *var, int seg)
4417 {
4418         struct vcpu_vmx *vmx = to_vmx(vcpu);
4419         u32 ar;
4420
4421         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4422                 *var = vmx->rmode.segs[seg];
4423                 if (seg == VCPU_SREG_TR
4424                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4425                         return;
4426                 var->base = vmx_read_guest_seg_base(vmx, seg);
4427                 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4428                 return;
4429         }
4430         var->base = vmx_read_guest_seg_base(vmx, seg);
4431         var->limit = vmx_read_guest_seg_limit(vmx, seg);
4432         var->selector = vmx_read_guest_seg_selector(vmx, seg);
4433         ar = vmx_read_guest_seg_ar(vmx, seg);
4434         var->unusable = (ar >> 16) & 1;
4435         var->type = ar & 15;
4436         var->s = (ar >> 4) & 1;
4437         var->dpl = (ar >> 5) & 3;
4438         /*
4439          * Some userspaces do not preserve unusable property. Since usable
4440          * segment has to be present according to VMX spec we can use present
4441          * property to amend userspace bug by making unusable segment always
4442          * nonpresent. vmx_segment_access_rights() already marks nonpresent
4443          * segment as unusable.
4444          */
4445         var->present = !var->unusable;
4446         var->avl = (ar >> 12) & 1;
4447         var->l = (ar >> 13) & 1;
4448         var->db = (ar >> 14) & 1;
4449         var->g = (ar >> 15) & 1;
4450 }
4451
4452 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4453 {
4454         struct kvm_segment s;
4455
4456         if (to_vmx(vcpu)->rmode.vm86_active) {
4457                 vmx_get_segment(vcpu, &s, seg);
4458                 return s.base;
4459         }
4460         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4461 }
4462
4463 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4464 {
4465         struct vcpu_vmx *vmx = to_vmx(vcpu);
4466
4467         if (unlikely(vmx->rmode.vm86_active))
4468                 return 0;
4469         else {
4470                 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4471                 return VMX_AR_DPL(ar);
4472         }
4473 }
4474
4475 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4476 {
4477         u32 ar;
4478
4479         if (var->unusable || !var->present)
4480                 ar = 1 << 16;
4481         else {
4482                 ar = var->type & 15;
4483                 ar |= (var->s & 1) << 4;
4484                 ar |= (var->dpl & 3) << 5;
4485                 ar |= (var->present & 1) << 7;
4486                 ar |= (var->avl & 1) << 12;
4487                 ar |= (var->l & 1) << 13;
4488                 ar |= (var->db & 1) << 14;
4489                 ar |= (var->g & 1) << 15;
4490         }
4491
4492         return ar;
4493 }
4494
4495 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4496                             struct kvm_segment *var, int seg)
4497 {
4498         struct vcpu_vmx *vmx = to_vmx(vcpu);
4499         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4500
4501         vmx_segment_cache_clear(vmx);
4502
4503         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4504                 vmx->rmode.segs[seg] = *var;
4505                 if (seg == VCPU_SREG_TR)
4506                         vmcs_write16(sf->selector, var->selector);
4507                 else if (var->s)
4508                         fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4509                 goto out;
4510         }
4511
4512         vmcs_writel(sf->base, var->base);
4513         vmcs_write32(sf->limit, var->limit);
4514         vmcs_write16(sf->selector, var->selector);
4515
4516         /*
4517          *   Fix the "Accessed" bit in AR field of segment registers for older
4518          * qemu binaries.
4519          *   IA32 arch specifies that at the time of processor reset the
4520          * "Accessed" bit in the AR field of segment registers is 1. And qemu
4521          * is setting it to 0 in the userland code. This causes invalid guest
4522          * state vmexit when "unrestricted guest" mode is turned on.
4523          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
4524          * tree. Newer qemu binaries with that qemu fix would not need this
4525          * kvm hack.
4526          */
4527         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4528                 var->type |= 0x1; /* Accessed */
4529
4530         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4531
4532 out:
4533         vmx->emulation_required = emulation_required(vcpu);
4534 }
4535
4536 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4537 {
4538         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4539
4540         *db = (ar >> 14) & 1;
4541         *l = (ar >> 13) & 1;
4542 }
4543
4544 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4545 {
4546         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4547         dt->address = vmcs_readl(GUEST_IDTR_BASE);
4548 }
4549
4550 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4551 {
4552         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4553         vmcs_writel(GUEST_IDTR_BASE, dt->address);
4554 }
4555
4556 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4557 {
4558         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4559         dt->address = vmcs_readl(GUEST_GDTR_BASE);
4560 }
4561
4562 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4563 {
4564         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4565         vmcs_writel(GUEST_GDTR_BASE, dt->address);
4566 }
4567
4568 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4569 {
4570         struct kvm_segment var;
4571         u32 ar;
4572
4573         vmx_get_segment(vcpu, &var, seg);
4574         var.dpl = 0x3;
4575         if (seg == VCPU_SREG_CS)
4576                 var.type = 0x3;
4577         ar = vmx_segment_access_rights(&var);
4578
4579         if (var.base != (var.selector << 4))
4580                 return false;
4581         if (var.limit != 0xffff)
4582                 return false;
4583         if (ar != 0xf3)
4584                 return false;
4585
4586         return true;
4587 }
4588
4589 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4590 {
4591         struct kvm_segment cs;
4592         unsigned int cs_rpl;
4593
4594         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4595         cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4596
4597         if (cs.unusable)
4598                 return false;
4599         if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4600                 return false;
4601         if (!cs.s)
4602                 return false;
4603         if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4604                 if (cs.dpl > cs_rpl)
4605                         return false;
4606         } else {
4607                 if (cs.dpl != cs_rpl)
4608                         return false;
4609         }
4610         if (!cs.present)
4611                 return false;
4612
4613         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4614         return true;
4615 }
4616
4617 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4618 {
4619         struct kvm_segment ss;
4620         unsigned int ss_rpl;
4621
4622         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4623         ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4624
4625         if (ss.unusable)
4626                 return true;
4627         if (ss.type != 3 && ss.type != 7)
4628                 return false;
4629         if (!ss.s)
4630                 return false;
4631         if (ss.dpl != ss_rpl) /* DPL != RPL */
4632                 return false;
4633         if (!ss.present)
4634                 return false;
4635
4636         return true;
4637 }
4638
4639 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4640 {
4641         struct kvm_segment var;
4642         unsigned int rpl;
4643
4644         vmx_get_segment(vcpu, &var, seg);
4645         rpl = var.selector & SEGMENT_RPL_MASK;
4646
4647         if (var.unusable)
4648                 return true;
4649         if (!var.s)
4650                 return false;
4651         if (!var.present)
4652                 return false;
4653         if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4654                 if (var.dpl < rpl) /* DPL < RPL */
4655                         return false;
4656         }
4657
4658         /* TODO: Add other members to kvm_segment_field to allow checking for other access
4659          * rights flags
4660          */
4661         return true;
4662 }
4663
4664 static bool tr_valid(struct kvm_vcpu *vcpu)
4665 {
4666         struct kvm_segment tr;
4667
4668         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4669
4670         if (tr.unusable)
4671                 return false;
4672         if (tr.selector & SEGMENT_TI_MASK)      /* TI = 1 */
4673                 return false;
4674         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4675                 return false;
4676         if (!tr.present)
4677                 return false;
4678
4679         return true;
4680 }
4681
4682 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4683 {
4684         struct kvm_segment ldtr;
4685
4686         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4687
4688         if (ldtr.unusable)
4689                 return true;
4690         if (ldtr.selector & SEGMENT_TI_MASK)    /* TI = 1 */
4691                 return false;
4692         if (ldtr.type != 2)
4693                 return false;
4694         if (!ldtr.present)
4695                 return false;
4696
4697         return true;
4698 }
4699
4700 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4701 {
4702         struct kvm_segment cs, ss;
4703
4704         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4705         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4706
4707         return ((cs.selector & SEGMENT_RPL_MASK) ==
4708                  (ss.selector & SEGMENT_RPL_MASK));
4709 }
4710
4711 /*
4712  * Check if guest state is valid. Returns true if valid, false if
4713  * not.
4714  * We assume that registers are always usable
4715  */
4716 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4717 {
4718         if (enable_unrestricted_guest)
4719                 return true;
4720
4721         /* real mode guest state checks */
4722         if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4723                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4724                         return false;
4725                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4726                         return false;
4727                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4728                         return false;
4729                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4730                         return false;
4731                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4732                         return false;
4733                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4734                         return false;
4735         } else {
4736         /* protected mode guest state checks */
4737                 if (!cs_ss_rpl_check(vcpu))
4738                         return false;
4739                 if (!code_segment_valid(vcpu))
4740                         return false;
4741                 if (!stack_segment_valid(vcpu))
4742                         return false;
4743                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4744                         return false;
4745                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4746                         return false;
4747                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4748                         return false;
4749                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4750                         return false;
4751                 if (!tr_valid(vcpu))
4752                         return false;
4753                 if (!ldtr_valid(vcpu))
4754                         return false;
4755         }
4756         /* TODO:
4757          * - Add checks on RIP
4758          * - Add checks on RFLAGS
4759          */
4760
4761         return true;
4762 }
4763
4764 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
4765 {
4766         return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
4767 }
4768
4769 static int init_rmode_tss(struct kvm *kvm)
4770 {
4771         gfn_t fn;
4772         u16 data = 0;
4773         int idx, r;
4774
4775         idx = srcu_read_lock(&kvm->srcu);
4776         fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4777         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4778         if (r < 0)
4779                 goto out;
4780         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4781         r = kvm_write_guest_page(kvm, fn++, &data,
4782                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
4783         if (r < 0)
4784                 goto out;
4785         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4786         if (r < 0)
4787                 goto out;
4788         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4789         if (r < 0)
4790                 goto out;
4791         data = ~0;
4792         r = kvm_write_guest_page(kvm, fn, &data,
4793                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4794                                  sizeof(u8));
4795 out:
4796         srcu_read_unlock(&kvm->srcu, idx);
4797         return r;
4798 }
4799
4800 static int init_rmode_identity_map(struct kvm *kvm)
4801 {
4802         int i, idx, r = 0;
4803         kvm_pfn_t identity_map_pfn;
4804         u32 tmp;
4805
4806         /* Protect kvm->arch.ept_identity_pagetable_done. */
4807         mutex_lock(&kvm->slots_lock);
4808
4809         if (likely(kvm->arch.ept_identity_pagetable_done))
4810                 goto out2;
4811
4812         if (!kvm->arch.ept_identity_map_addr)
4813                 kvm->arch.ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4814         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4815
4816         r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4817                                     kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4818         if (r < 0)
4819                 goto out2;
4820
4821         idx = srcu_read_lock(&kvm->srcu);
4822         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4823         if (r < 0)
4824                 goto out;
4825         /* Set up identity-mapping pagetable for EPT in real mode */
4826         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4827                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4828                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4829                 r = kvm_write_guest_page(kvm, identity_map_pfn,
4830                                 &tmp, i * sizeof(tmp), sizeof(tmp));
4831                 if (r < 0)
4832                         goto out;
4833         }
4834         kvm->arch.ept_identity_pagetable_done = true;
4835
4836 out:
4837         srcu_read_unlock(&kvm->srcu, idx);
4838
4839 out2:
4840         mutex_unlock(&kvm->slots_lock);
4841         return r;
4842 }
4843
4844 static void seg_setup(int seg)
4845 {
4846         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4847         unsigned int ar;
4848
4849         vmcs_write16(sf->selector, 0);
4850         vmcs_writel(sf->base, 0);
4851         vmcs_write32(sf->limit, 0xffff);
4852         ar = 0x93;
4853         if (seg == VCPU_SREG_CS)
4854                 ar |= 0x08; /* code segment */
4855
4856         vmcs_write32(sf->ar_bytes, ar);
4857 }
4858
4859 static int alloc_apic_access_page(struct kvm *kvm)
4860 {
4861         struct page *page;
4862         int r = 0;
4863
4864         mutex_lock(&kvm->slots_lock);
4865         if (kvm->arch.apic_access_page_done)
4866                 goto out;
4867         r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4868                                     APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4869         if (r)
4870                 goto out;
4871
4872         page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4873         if (is_error_page(page)) {
4874                 r = -EFAULT;
4875                 goto out;
4876         }
4877
4878         /*
4879          * Do not pin the page in memory, so that memory hot-unplug
4880          * is able to migrate it.
4881          */
4882         put_page(page);
4883         kvm->arch.apic_access_page_done = true;
4884 out:
4885         mutex_unlock(&kvm->slots_lock);
4886         return r;
4887 }
4888
4889 static int allocate_vpid(void)
4890 {
4891         int vpid;
4892
4893         if (!enable_vpid)
4894                 return 0;
4895         spin_lock(&vmx_vpid_lock);
4896         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4897         if (vpid < VMX_NR_VPIDS)
4898                 __set_bit(vpid, vmx_vpid_bitmap);
4899         else
4900                 vpid = 0;
4901         spin_unlock(&vmx_vpid_lock);
4902         return vpid;
4903 }
4904
4905 static void free_vpid(int vpid)
4906 {
4907         if (!enable_vpid || vpid == 0)
4908                 return;
4909         spin_lock(&vmx_vpid_lock);
4910         __clear_bit(vpid, vmx_vpid_bitmap);
4911         spin_unlock(&vmx_vpid_lock);
4912 }
4913
4914 #define MSR_TYPE_R      1
4915 #define MSR_TYPE_W      2
4916 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4917                                                 u32 msr, int type)
4918 {
4919         int f = sizeof(unsigned long);
4920
4921         if (!cpu_has_vmx_msr_bitmap())
4922                 return;
4923
4924         /*
4925          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4926          * have the write-low and read-high bitmap offsets the wrong way round.
4927          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4928          */
4929         if (msr <= 0x1fff) {
4930                 if (type & MSR_TYPE_R)
4931                         /* read-low */
4932                         __clear_bit(msr, msr_bitmap + 0x000 / f);
4933
4934                 if (type & MSR_TYPE_W)
4935                         /* write-low */
4936                         __clear_bit(msr, msr_bitmap + 0x800 / f);
4937
4938         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4939                 msr &= 0x1fff;
4940                 if (type & MSR_TYPE_R)
4941                         /* read-high */
4942                         __clear_bit(msr, msr_bitmap + 0x400 / f);
4943
4944                 if (type & MSR_TYPE_W)
4945                         /* write-high */
4946                         __clear_bit(msr, msr_bitmap + 0xc00 / f);
4947
4948         }
4949 }
4950
4951 /*
4952  * If a msr is allowed by L0, we should check whether it is allowed by L1.
4953  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4954  */
4955 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4956                                                unsigned long *msr_bitmap_nested,
4957                                                u32 msr, int type)
4958 {
4959         int f = sizeof(unsigned long);
4960
4961         if (!cpu_has_vmx_msr_bitmap()) {
4962                 WARN_ON(1);
4963                 return;
4964         }
4965
4966         /*
4967          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4968          * have the write-low and read-high bitmap offsets the wrong way round.
4969          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4970          */
4971         if (msr <= 0x1fff) {
4972                 if (type & MSR_TYPE_R &&
4973                    !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4974                         /* read-low */
4975                         __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4976
4977                 if (type & MSR_TYPE_W &&
4978                    !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4979                         /* write-low */
4980                         __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4981
4982         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4983                 msr &= 0x1fff;
4984                 if (type & MSR_TYPE_R &&
4985                    !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4986                         /* read-high */
4987                         __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4988
4989                 if (type & MSR_TYPE_W &&
4990                    !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4991                         /* write-high */
4992                         __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4993
4994         }
4995 }
4996
4997 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4998 {
4999         if (!longmode_only)
5000                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
5001                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
5002         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
5003                                                 msr, MSR_TYPE_R | MSR_TYPE_W);
5004 }
5005
5006 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
5007 {
5008         if (apicv_active) {
5009                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
5010                                 msr, type);
5011                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
5012                                 msr, type);
5013         } else {
5014                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
5015                                 msr, type);
5016                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
5017                                 msr, type);
5018         }
5019 }
5020
5021 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
5022 {
5023         return enable_apicv;
5024 }
5025
5026 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
5027 {
5028         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5029         gfn_t gfn;
5030
5031         /*
5032          * Don't need to mark the APIC access page dirty; it is never
5033          * written to by the CPU during APIC virtualization.
5034          */
5035
5036         if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
5037                 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
5038                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5039         }
5040
5041         if (nested_cpu_has_posted_intr(vmcs12)) {
5042                 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
5043                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5044         }
5045 }
5046
5047
5048 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5049 {
5050         struct vcpu_vmx *vmx = to_vmx(vcpu);
5051         int max_irr;
5052         void *vapic_page;
5053         u16 status;
5054
5055         if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
5056                 return;
5057
5058         vmx->nested.pi_pending = false;
5059         if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5060                 return;
5061
5062         max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
5063         if (max_irr != 256) {
5064                 vapic_page = kmap(vmx->nested.virtual_apic_page);
5065                 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
5066                 kunmap(vmx->nested.virtual_apic_page);
5067
5068                 status = vmcs_read16(GUEST_INTR_STATUS);
5069                 if ((u8)max_irr > ((u8)status & 0xff)) {
5070                         status &= ~0xff;
5071                         status |= (u8)max_irr;
5072                         vmcs_write16(GUEST_INTR_STATUS, status);
5073                 }
5074         }
5075
5076         nested_mark_vmcs12_pages_dirty(vcpu);
5077 }
5078
5079 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5080                                                      bool nested)
5081 {
5082 #ifdef CONFIG_SMP
5083         int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5084
5085         if (vcpu->mode == IN_GUEST_MODE) {
5086                 /*
5087                  * The vector of interrupt to be delivered to vcpu had
5088                  * been set in PIR before this function.
5089                  *
5090                  * Following cases will be reached in this block, and
5091                  * we always send a notification event in all cases as
5092                  * explained below.
5093                  *
5094                  * Case 1: vcpu keeps in non-root mode. Sending a
5095                  * notification event posts the interrupt to vcpu.
5096                  *
5097                  * Case 2: vcpu exits to root mode and is still
5098                  * runnable. PIR will be synced to vIRR before the
5099                  * next vcpu entry. Sending a notification event in
5100                  * this case has no effect, as vcpu is not in root
5101                  * mode.
5102                  *
5103                  * Case 3: vcpu exits to root mode and is blocked.
5104                  * vcpu_block() has already synced PIR to vIRR and
5105                  * never blocks vcpu if vIRR is not cleared. Therefore,
5106                  * a blocked vcpu here does not wait for any requested
5107                  * interrupts in PIR, and sending a notification event
5108                  * which has no effect is safe here.
5109                  */
5110
5111                 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5112                 return true;
5113         }
5114 #endif
5115         return false;
5116 }
5117
5118 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5119                                                 int vector)
5120 {
5121         struct vcpu_vmx *vmx = to_vmx(vcpu);
5122
5123         if (is_guest_mode(vcpu) &&
5124             vector == vmx->nested.posted_intr_nv) {
5125                 /* the PIR and ON have been set by L1. */
5126                 kvm_vcpu_trigger_posted_interrupt(vcpu, true);
5127                 /*
5128                  * If a posted intr is not recognized by hardware,
5129                  * we will accomplish it in the next vmentry.
5130                  */
5131                 vmx->nested.pi_pending = true;
5132                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5133                 return 0;
5134         }
5135         return -1;
5136 }
5137 /*
5138  * Send interrupt to vcpu via posted interrupt way.
5139  * 1. If target vcpu is running(non-root mode), send posted interrupt
5140  * notification to vcpu and hardware will sync PIR to vIRR atomically.
5141  * 2. If target vcpu isn't running(root mode), kick it to pick up the
5142  * interrupt from PIR in next vmentry.
5143  */
5144 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5145 {
5146         struct vcpu_vmx *vmx = to_vmx(vcpu);
5147         int r;
5148
5149         r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5150         if (!r)
5151                 return;
5152
5153         if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5154                 return;
5155
5156         /* If a previous notification has sent the IPI, nothing to do.  */
5157         if (pi_test_and_set_on(&vmx->pi_desc))
5158                 return;
5159
5160         if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5161                 kvm_vcpu_kick(vcpu);
5162 }
5163
5164 /*
5165  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5166  * will not change in the lifetime of the guest.
5167  * Note that host-state that does change is set elsewhere. E.g., host-state
5168  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5169  */
5170 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5171 {
5172         u32 low32, high32;
5173         unsigned long tmpl;
5174         struct desc_ptr dt;
5175         unsigned long cr0, cr3, cr4;
5176
5177         cr0 = read_cr0();
5178         WARN_ON(cr0 & X86_CR0_TS);
5179         vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
5180
5181         /*
5182          * Save the most likely value for this task's CR3 in the VMCS.
5183          * We can't use __get_current_cr3_fast() because we're not atomic.
5184          */
5185         cr3 = __read_cr3();
5186         vmcs_writel(HOST_CR3, cr3);             /* 22.2.3  FIXME: shadow tables */
5187         vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
5188
5189         /* Save the most likely value for this task's CR4 in the VMCS. */
5190         cr4 = cr4_read_shadow();
5191         vmcs_writel(HOST_CR4, cr4);                     /* 22.2.3, 22.2.5 */
5192         vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
5193
5194         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
5195 #ifdef CONFIG_X86_64
5196         /*
5197          * Load null selectors, so we can avoid reloading them in
5198          * __vmx_load_host_state(), in case userspace uses the null selectors
5199          * too (the expected case).
5200          */
5201         vmcs_write16(HOST_DS_SELECTOR, 0);
5202         vmcs_write16(HOST_ES_SELECTOR, 0);
5203 #else
5204         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5205         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5206 #endif
5207         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5208         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
5209
5210         store_idt(&dt);
5211         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
5212         vmx->host_idt_base = dt.address;
5213
5214         vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5215
5216         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5217         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5218         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5219         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
5220
5221         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5222                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5223                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5224         }
5225 }
5226
5227 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5228 {
5229         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5230         if (enable_ept)
5231                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5232         if (is_guest_mode(&vmx->vcpu))
5233                 vmx->vcpu.arch.cr4_guest_owned_bits &=
5234                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5235         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5236 }
5237
5238 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5239 {
5240         u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5241
5242         if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5243                 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5244
5245         if (!enable_vnmi)
5246                 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
5247
5248         /* Enable the preemption timer dynamically */
5249         pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5250         return pin_based_exec_ctrl;
5251 }
5252
5253 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5254 {
5255         struct vcpu_vmx *vmx = to_vmx(vcpu);
5256
5257         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5258         if (cpu_has_secondary_exec_ctrls()) {
5259                 if (kvm_vcpu_apicv_active(vcpu))
5260                         vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5261                                       SECONDARY_EXEC_APIC_REGISTER_VIRT |
5262                                       SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5263                 else
5264                         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5265                                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
5266                                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5267         }
5268
5269         if (cpu_has_vmx_msr_bitmap())
5270                 vmx_set_msr_bitmap(vcpu);
5271 }
5272
5273 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5274 {
5275         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5276
5277         if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5278                 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5279
5280         if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5281                 exec_control &= ~CPU_BASED_TPR_SHADOW;
5282 #ifdef CONFIG_X86_64
5283                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5284                                 CPU_BASED_CR8_LOAD_EXITING;
5285 #endif
5286         }
5287         if (!enable_ept)
5288                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5289                                 CPU_BASED_CR3_LOAD_EXITING  |
5290                                 CPU_BASED_INVLPG_EXITING;
5291         return exec_control;
5292 }
5293
5294 static bool vmx_rdrand_supported(void)
5295 {
5296         return vmcs_config.cpu_based_2nd_exec_ctrl &
5297                 SECONDARY_EXEC_RDRAND_EXITING;
5298 }
5299
5300 static bool vmx_rdseed_supported(void)
5301 {
5302         return vmcs_config.cpu_based_2nd_exec_ctrl &
5303                 SECONDARY_EXEC_RDSEED_EXITING;
5304 }
5305
5306 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
5307 {
5308         struct kvm_vcpu *vcpu = &vmx->vcpu;
5309
5310         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5311         if (!cpu_need_virtualize_apic_accesses(vcpu))
5312                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5313         if (vmx->vpid == 0)
5314                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5315         if (!enable_ept) {
5316                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5317                 enable_unrestricted_guest = 0;
5318                 /* Enable INVPCID for non-ept guests may cause performance regression. */
5319                 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5320         }
5321         if (!enable_unrestricted_guest)
5322                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5323         if (!ple_gap)
5324                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5325         if (!kvm_vcpu_apicv_active(vcpu))
5326                 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5327                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5328         exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5329         /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5330            (handle_vmptrld).
5331            We can NOT enable shadow_vmcs here because we don't have yet
5332            a current VMCS12
5333         */
5334         exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5335
5336         if (!enable_pml)
5337                 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5338
5339         if (vmx_xsaves_supported()) {
5340                 /* Exposing XSAVES only when XSAVE is exposed */
5341                 bool xsaves_enabled =
5342                         guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
5343                         guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
5344
5345                 if (!xsaves_enabled)
5346                         exec_control &= ~SECONDARY_EXEC_XSAVES;
5347
5348                 if (nested) {
5349                         if (xsaves_enabled)
5350                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5351                                         SECONDARY_EXEC_XSAVES;
5352                         else
5353                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5354                                         ~SECONDARY_EXEC_XSAVES;
5355                 }
5356         }
5357
5358         if (vmx_rdtscp_supported()) {
5359                 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
5360                 if (!rdtscp_enabled)
5361                         exec_control &= ~SECONDARY_EXEC_RDTSCP;
5362
5363                 if (nested) {
5364                         if (rdtscp_enabled)
5365                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5366                                         SECONDARY_EXEC_RDTSCP;
5367                         else
5368                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5369                                         ~SECONDARY_EXEC_RDTSCP;
5370                 }
5371         }
5372
5373         if (vmx_invpcid_supported()) {
5374                 /* Exposing INVPCID only when PCID is exposed */
5375                 bool invpcid_enabled =
5376                         guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
5377                         guest_cpuid_has(vcpu, X86_FEATURE_PCID);
5378
5379                 if (!invpcid_enabled) {
5380                         exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5381                         guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
5382                 }
5383
5384                 if (nested) {
5385                         if (invpcid_enabled)
5386                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5387                                         SECONDARY_EXEC_ENABLE_INVPCID;
5388                         else
5389                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5390                                         ~SECONDARY_EXEC_ENABLE_INVPCID;
5391                 }
5392         }
5393
5394         if (vmx_rdrand_supported()) {
5395                 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
5396                 if (rdrand_enabled)
5397                         exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
5398
5399                 if (nested) {
5400                         if (rdrand_enabled)
5401                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5402                                         SECONDARY_EXEC_RDRAND_EXITING;
5403                         else
5404                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5405                                         ~SECONDARY_EXEC_RDRAND_EXITING;
5406                 }
5407         }
5408
5409         if (vmx_rdseed_supported()) {
5410                 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
5411                 if (rdseed_enabled)
5412                         exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
5413
5414                 if (nested) {
5415                         if (rdseed_enabled)
5416                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5417                                         SECONDARY_EXEC_RDSEED_EXITING;
5418                         else
5419                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5420                                         ~SECONDARY_EXEC_RDSEED_EXITING;
5421                 }
5422         }
5423
5424         vmx->secondary_exec_control = exec_control;
5425 }
5426
5427 static void ept_set_mmio_spte_mask(void)
5428 {
5429         /*
5430          * EPT Misconfigurations can be generated if the value of bits 2:0
5431          * of an EPT paging-structure entry is 110b (write/execute).
5432          */
5433         kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
5434                                    VMX_EPT_MISCONFIG_WX_VALUE);
5435 }
5436
5437 #define VMX_XSS_EXIT_BITMAP 0
5438 /*
5439  * Sets up the vmcs for emulated real mode.
5440  */
5441 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
5442 {
5443 #ifdef CONFIG_X86_64
5444         unsigned long a;
5445 #endif
5446         int i;
5447
5448         /* I/O */
5449         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5450         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5451
5452         if (enable_shadow_vmcs) {
5453                 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5454                 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5455         }
5456         if (cpu_has_vmx_msr_bitmap())
5457                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5458
5459         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5460
5461         /* Control */
5462         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5463         vmx->hv_deadline_tsc = -1;
5464
5465         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5466
5467         if (cpu_has_secondary_exec_ctrls()) {
5468                 vmx_compute_secondary_exec_control(vmx);
5469                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5470                              vmx->secondary_exec_control);
5471         }
5472
5473         if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5474                 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5475                 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5476                 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5477                 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5478
5479                 vmcs_write16(GUEST_INTR_STATUS, 0);
5480
5481                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5482                 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5483         }
5484
5485         if (ple_gap) {
5486                 vmcs_write32(PLE_GAP, ple_gap);
5487                 vmx->ple_window = ple_window;
5488                 vmx->ple_window_dirty = true;
5489         }
5490
5491         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5492         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5493         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
5494
5495         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
5496         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
5497         vmx_set_constant_host_state(vmx);
5498 #ifdef CONFIG_X86_64
5499         rdmsrl(MSR_FS_BASE, a);
5500         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5501         rdmsrl(MSR_GS_BASE, a);
5502         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5503 #else
5504         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5505         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5506 #endif
5507
5508         if (cpu_has_vmx_vmfunc())
5509                 vmcs_write64(VM_FUNCTION_CONTROL, 0);
5510
5511         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5512         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5513         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5514         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5515         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5516
5517         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5518                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5519
5520         for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5521                 u32 index = vmx_msr_index[i];
5522                 u32 data_low, data_high;
5523                 int j = vmx->nmsrs;
5524
5525                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5526                         continue;
5527                 if (wrmsr_safe(index, data_low, data_high) < 0)
5528                         continue;
5529                 vmx->guest_msrs[j].index = i;
5530                 vmx->guest_msrs[j].data = 0;
5531                 vmx->guest_msrs[j].mask = -1ull;
5532                 ++vmx->nmsrs;
5533         }
5534
5535
5536         vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5537
5538         /* 22.2.1, 20.8.1 */
5539         vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5540
5541         vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5542         vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5543
5544         set_cr4_guest_host_mask(vmx);
5545
5546         if (vmx_xsaves_supported())
5547                 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5548
5549         if (enable_pml) {
5550                 ASSERT(vmx->pml_pg);
5551                 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5552                 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5553         }
5554 }
5555
5556 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5557 {
5558         struct vcpu_vmx *vmx = to_vmx(vcpu);
5559         struct msr_data apic_base_msr;
5560         u64 cr0;
5561
5562         vmx->rmode.vm86_active = 0;
5563
5564         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5565         kvm_set_cr8(vcpu, 0);
5566
5567         if (!init_event) {
5568                 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5569                                      MSR_IA32_APICBASE_ENABLE;
5570                 if (kvm_vcpu_is_reset_bsp(vcpu))
5571                         apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5572                 apic_base_msr.host_initiated = true;
5573                 kvm_set_apic_base(vcpu, &apic_base_msr);
5574         }
5575
5576         vmx_segment_cache_clear(vmx);
5577
5578         seg_setup(VCPU_SREG_CS);
5579         vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5580         vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5581
5582         seg_setup(VCPU_SREG_DS);
5583         seg_setup(VCPU_SREG_ES);
5584         seg_setup(VCPU_SREG_FS);
5585         seg_setup(VCPU_SREG_GS);
5586         seg_setup(VCPU_SREG_SS);
5587
5588         vmcs_write16(GUEST_TR_SELECTOR, 0);
5589         vmcs_writel(GUEST_TR_BASE, 0);
5590         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5591         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5592
5593         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5594         vmcs_writel(GUEST_LDTR_BASE, 0);
5595         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5596         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5597
5598         if (!init_event) {
5599                 vmcs_write32(GUEST_SYSENTER_CS, 0);
5600                 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5601                 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5602                 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5603         }
5604
5605         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
5606         kvm_rip_write(vcpu, 0xfff0);
5607
5608         vmcs_writel(GUEST_GDTR_BASE, 0);
5609         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5610
5611         vmcs_writel(GUEST_IDTR_BASE, 0);
5612         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5613
5614         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5615         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5616         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5617         if (kvm_mpx_supported())
5618                 vmcs_write64(GUEST_BNDCFGS, 0);
5619
5620         setup_msrs(vmx);
5621
5622         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
5623
5624         if (cpu_has_vmx_tpr_shadow() && !init_event) {
5625                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5626                 if (cpu_need_tpr_shadow(vcpu))
5627                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5628                                      __pa(vcpu->arch.apic->regs));
5629                 vmcs_write32(TPR_THRESHOLD, 0);
5630         }
5631
5632         kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5633
5634         if (vmx->vpid != 0)
5635                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5636
5637         cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5638         vmx->vcpu.arch.cr0 = cr0;
5639         vmx_set_cr0(vcpu, cr0); /* enter rmode */
5640         vmx_set_cr4(vcpu, 0);
5641         vmx_set_efer(vcpu, 0);
5642
5643         update_exception_bitmap(vcpu);
5644
5645         vpid_sync_context(vmx->vpid);
5646 }
5647
5648 /*
5649  * In nested virtualization, check if L1 asked to exit on external interrupts.
5650  * For most existing hypervisors, this will always return true.
5651  */
5652 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5653 {
5654         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5655                 PIN_BASED_EXT_INTR_MASK;
5656 }
5657
5658 /*
5659  * In nested virtualization, check if L1 has set
5660  * VM_EXIT_ACK_INTR_ON_EXIT
5661  */
5662 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5663 {
5664         return get_vmcs12(vcpu)->vm_exit_controls &
5665                 VM_EXIT_ACK_INTR_ON_EXIT;
5666 }
5667
5668 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5669 {
5670         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5671                 PIN_BASED_NMI_EXITING;
5672 }
5673
5674 static void enable_irq_window(struct kvm_vcpu *vcpu)
5675 {
5676         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5677                       CPU_BASED_VIRTUAL_INTR_PENDING);
5678 }
5679
5680 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5681 {
5682         if (!enable_vnmi ||
5683             vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5684                 enable_irq_window(vcpu);
5685                 return;
5686         }
5687
5688         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5689                       CPU_BASED_VIRTUAL_NMI_PENDING);
5690 }
5691
5692 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5693 {
5694         struct vcpu_vmx *vmx = to_vmx(vcpu);
5695         uint32_t intr;
5696         int irq = vcpu->arch.interrupt.nr;
5697
5698         trace_kvm_inj_virq(irq);
5699
5700         ++vcpu->stat.irq_injections;
5701         if (vmx->rmode.vm86_active) {
5702                 int inc_eip = 0;
5703                 if (vcpu->arch.interrupt.soft)
5704                         inc_eip = vcpu->arch.event_exit_inst_len;
5705                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5706                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5707                 return;
5708         }
5709         intr = irq | INTR_INFO_VALID_MASK;
5710         if (vcpu->arch.interrupt.soft) {
5711                 intr |= INTR_TYPE_SOFT_INTR;
5712                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5713                              vmx->vcpu.arch.event_exit_inst_len);
5714         } else
5715                 intr |= INTR_TYPE_EXT_INTR;
5716         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5717 }
5718
5719 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5720 {
5721         struct vcpu_vmx *vmx = to_vmx(vcpu);
5722
5723         if (!enable_vnmi) {
5724                 /*
5725                  * Tracking the NMI-blocked state in software is built upon
5726                  * finding the next open IRQ window. This, in turn, depends on
5727                  * well-behaving guests: They have to keep IRQs disabled at
5728                  * least as long as the NMI handler runs. Otherwise we may
5729                  * cause NMI nesting, maybe breaking the guest. But as this is
5730                  * highly unlikely, we can live with the residual risk.
5731                  */
5732                 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
5733                 vmx->loaded_vmcs->vnmi_blocked_time = 0;
5734         }
5735
5736         ++vcpu->stat.nmi_injections;
5737         vmx->loaded_vmcs->nmi_known_unmasked = false;
5738
5739         if (vmx->rmode.vm86_active) {
5740                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5741                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5742                 return;
5743         }
5744
5745         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5746                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5747 }
5748
5749 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5750 {
5751         struct vcpu_vmx *vmx = to_vmx(vcpu);
5752         bool masked;
5753
5754         if (!enable_vnmi)
5755                 return vmx->loaded_vmcs->soft_vnmi_blocked;
5756         if (vmx->loaded_vmcs->nmi_known_unmasked)
5757                 return false;
5758         masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5759         vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5760         return masked;
5761 }
5762
5763 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5764 {
5765         struct vcpu_vmx *vmx = to_vmx(vcpu);
5766
5767         if (!enable_vnmi) {
5768                 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
5769                         vmx->loaded_vmcs->soft_vnmi_blocked = masked;
5770                         vmx->loaded_vmcs->vnmi_blocked_time = 0;
5771                 }
5772         } else {
5773                 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5774                 if (masked)
5775                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5776                                       GUEST_INTR_STATE_NMI);
5777                 else
5778                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5779                                         GUEST_INTR_STATE_NMI);
5780         }
5781 }
5782
5783 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5784 {
5785         if (to_vmx(vcpu)->nested.nested_run_pending)
5786                 return 0;
5787
5788         if (!enable_vnmi &&
5789             to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
5790                 return 0;
5791
5792         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5793                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5794                    | GUEST_INTR_STATE_NMI));
5795 }
5796
5797 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5798 {
5799         return (!to_vmx(vcpu)->nested.nested_run_pending &&
5800                 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5801                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5802                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5803 }
5804
5805 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5806 {
5807         int ret;
5808
5809         ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5810                                     PAGE_SIZE * 3);
5811         if (ret)
5812                 return ret;
5813         kvm->arch.tss_addr = addr;
5814         return init_rmode_tss(kvm);
5815 }
5816
5817 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5818 {
5819         switch (vec) {
5820         case BP_VECTOR:
5821                 /*
5822                  * Update instruction length as we may reinject the exception
5823                  * from user space while in guest debugging mode.
5824                  */
5825                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5826                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5827                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5828                         return false;
5829                 /* fall through */
5830         case DB_VECTOR:
5831                 if (vcpu->guest_debug &
5832                         (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5833                         return false;
5834                 /* fall through */
5835         case DE_VECTOR:
5836         case OF_VECTOR:
5837         case BR_VECTOR:
5838         case UD_VECTOR:
5839         case DF_VECTOR:
5840         case SS_VECTOR:
5841         case GP_VECTOR:
5842         case MF_VECTOR:
5843                 return true;
5844         break;
5845         }
5846         return false;
5847 }
5848
5849 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5850                                   int vec, u32 err_code)
5851 {
5852         /*
5853          * Instruction with address size override prefix opcode 0x67
5854          * Cause the #SS fault with 0 error code in VM86 mode.
5855          */
5856         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5857                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5858                         if (vcpu->arch.halt_request) {
5859                                 vcpu->arch.halt_request = 0;
5860                                 return kvm_vcpu_halt(vcpu);
5861                         }
5862                         return 1;
5863                 }
5864                 return 0;
5865         }
5866
5867         /*
5868          * Forward all other exceptions that are valid in real mode.
5869          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5870          *        the required debugging infrastructure rework.
5871          */
5872         kvm_queue_exception(vcpu, vec);
5873         return 1;
5874 }
5875
5876 /*
5877  * Trigger machine check on the host. We assume all the MSRs are already set up
5878  * by the CPU and that we still run on the same CPU as the MCE occurred on.
5879  * We pass a fake environment to the machine check handler because we want
5880  * the guest to be always treated like user space, no matter what context
5881  * it used internally.
5882  */
5883 static void kvm_machine_check(void)
5884 {
5885 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5886         struct pt_regs regs = {
5887                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5888                 .flags = X86_EFLAGS_IF,
5889         };
5890
5891         do_machine_check(&regs, 0);
5892 #endif
5893 }
5894
5895 static int handle_machine_check(struct kvm_vcpu *vcpu)
5896 {
5897         /* already handled by vcpu_run */
5898         return 1;
5899 }
5900
5901 static int handle_exception(struct kvm_vcpu *vcpu)
5902 {
5903         struct vcpu_vmx *vmx = to_vmx(vcpu);
5904         struct kvm_run *kvm_run = vcpu->run;
5905         u32 intr_info, ex_no, error_code;
5906         unsigned long cr2, rip, dr6;
5907         u32 vect_info;
5908         enum emulation_result er;
5909
5910         vect_info = vmx->idt_vectoring_info;
5911         intr_info = vmx->exit_intr_info;
5912
5913         if (is_machine_check(intr_info))
5914                 return handle_machine_check(vcpu);
5915
5916         if (is_nmi(intr_info))
5917                 return 1;  /* already handled by vmx_vcpu_run() */
5918
5919         if (is_invalid_opcode(intr_info)) {
5920                 WARN_ON_ONCE(is_guest_mode(vcpu));
5921                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5922                 if (er == EMULATE_USER_EXIT)
5923                         return 0;
5924                 if (er != EMULATE_DONE)
5925                         kvm_queue_exception(vcpu, UD_VECTOR);
5926                 return 1;
5927         }
5928
5929         error_code = 0;
5930         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5931                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5932
5933         /*
5934          * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5935          * MMIO, it is better to report an internal error.
5936          * See the comments in vmx_handle_exit.
5937          */
5938         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5939             !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5940                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5941                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5942                 vcpu->run->internal.ndata = 3;
5943                 vcpu->run->internal.data[0] = vect_info;
5944                 vcpu->run->internal.data[1] = intr_info;
5945                 vcpu->run->internal.data[2] = error_code;
5946                 return 0;
5947         }
5948
5949         if (is_page_fault(intr_info)) {
5950                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5951                 /* EPT won't cause page fault directly */
5952                 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
5953                 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
5954         }
5955
5956         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5957
5958         if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5959                 return handle_rmode_exception(vcpu, ex_no, error_code);
5960
5961         switch (ex_no) {
5962         case AC_VECTOR:
5963                 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5964                 return 1;
5965         case DB_VECTOR:
5966                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5967                 if (!(vcpu->guest_debug &
5968                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5969                         vcpu->arch.dr6 &= ~15;
5970                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
5971                         if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5972                                 skip_emulated_instruction(vcpu);
5973
5974                         kvm_queue_exception(vcpu, DB_VECTOR);
5975                         return 1;
5976                 }
5977                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5978                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5979                 /* fall through */
5980         case BP_VECTOR:
5981                 /*
5982                  * Update instruction length as we may reinject #BP from
5983                  * user space while in guest debugging mode. Reading it for
5984                  * #DB as well causes no harm, it is not used in that case.
5985                  */
5986                 vmx->vcpu.arch.event_exit_inst_len =
5987                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5988                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5989                 rip = kvm_rip_read(vcpu);
5990                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5991                 kvm_run->debug.arch.exception = ex_no;
5992                 break;
5993         default:
5994                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5995                 kvm_run->ex.exception = ex_no;
5996                 kvm_run->ex.error_code = error_code;
5997                 break;
5998         }
5999         return 0;
6000 }
6001
6002 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
6003 {
6004         ++vcpu->stat.irq_exits;
6005         return 1;
6006 }
6007
6008 static int handle_triple_fault(struct kvm_vcpu *vcpu)
6009 {
6010         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6011         vcpu->mmio_needed = 0;
6012         return 0;
6013 }
6014
6015 static int handle_io(struct kvm_vcpu *vcpu)
6016 {
6017         unsigned long exit_qualification;
6018         int size, in, string, ret;
6019         unsigned port;
6020
6021         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6022         string = (exit_qualification & 16) != 0;
6023         in = (exit_qualification & 8) != 0;
6024
6025         ++vcpu->stat.io_exits;
6026
6027         if (string || in)
6028                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6029
6030         port = exit_qualification >> 16;
6031         size = (exit_qualification & 7) + 1;
6032
6033         ret = kvm_skip_emulated_instruction(vcpu);
6034
6035         /*
6036          * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6037          * KVM_EXIT_DEBUG here.
6038          */
6039         return kvm_fast_pio_out(vcpu, size, port) && ret;
6040 }
6041
6042 static void
6043 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
6044 {
6045         /*
6046          * Patch in the VMCALL instruction:
6047          */
6048         hypercall[0] = 0x0f;
6049         hypercall[1] = 0x01;
6050         hypercall[2] = 0xc1;
6051 }
6052
6053 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
6054 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
6055 {
6056         if (is_guest_mode(vcpu)) {
6057                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6058                 unsigned long orig_val = val;
6059
6060                 /*
6061                  * We get here when L2 changed cr0 in a way that did not change
6062                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
6063                  * but did change L0 shadowed bits. So we first calculate the
6064                  * effective cr0 value that L1 would like to write into the
6065                  * hardware. It consists of the L2-owned bits from the new
6066                  * value combined with the L1-owned bits from L1's guest_cr0.
6067                  */
6068                 val = (val & ~vmcs12->cr0_guest_host_mask) |
6069                         (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
6070
6071                 if (!nested_guest_cr0_valid(vcpu, val))
6072                         return 1;
6073
6074                 if (kvm_set_cr0(vcpu, val))
6075                         return 1;
6076                 vmcs_writel(CR0_READ_SHADOW, orig_val);
6077                 return 0;
6078         } else {
6079                 if (to_vmx(vcpu)->nested.vmxon &&
6080                     !nested_host_cr0_valid(vcpu, val))
6081                         return 1;
6082
6083                 return kvm_set_cr0(vcpu, val);
6084         }
6085 }
6086
6087 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
6088 {
6089         if (is_guest_mode(vcpu)) {
6090                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6091                 unsigned long orig_val = val;
6092
6093                 /* analogously to handle_set_cr0 */
6094                 val = (val & ~vmcs12->cr4_guest_host_mask) |
6095                         (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
6096                 if (kvm_set_cr4(vcpu, val))
6097                         return 1;
6098                 vmcs_writel(CR4_READ_SHADOW, orig_val);
6099                 return 0;
6100         } else
6101                 return kvm_set_cr4(vcpu, val);
6102 }
6103
6104 static int handle_cr(struct kvm_vcpu *vcpu)
6105 {
6106         unsigned long exit_qualification, val;
6107         int cr;
6108         int reg;
6109         int err;
6110         int ret;
6111
6112         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6113         cr = exit_qualification & 15;
6114         reg = (exit_qualification >> 8) & 15;
6115         switch ((exit_qualification >> 4) & 3) {
6116         case 0: /* mov to cr */
6117                 val = kvm_register_readl(vcpu, reg);
6118                 trace_kvm_cr_write(cr, val);
6119                 switch (cr) {
6120                 case 0:
6121                         err = handle_set_cr0(vcpu, val);
6122                         return kvm_complete_insn_gp(vcpu, err);
6123                 case 3:
6124                         err = kvm_set_cr3(vcpu, val);
6125                         return kvm_complete_insn_gp(vcpu, err);
6126                 case 4:
6127                         err = handle_set_cr4(vcpu, val);
6128                         return kvm_complete_insn_gp(vcpu, err);
6129                 case 8: {
6130                                 u8 cr8_prev = kvm_get_cr8(vcpu);
6131                                 u8 cr8 = (u8)val;
6132                                 err = kvm_set_cr8(vcpu, cr8);
6133                                 ret = kvm_complete_insn_gp(vcpu, err);
6134                                 if (lapic_in_kernel(vcpu))
6135                                         return ret;
6136                                 if (cr8_prev <= cr8)
6137                                         return ret;
6138                                 /*
6139                                  * TODO: we might be squashing a
6140                                  * KVM_GUESTDBG_SINGLESTEP-triggered
6141                                  * KVM_EXIT_DEBUG here.
6142                                  */
6143                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
6144                                 return 0;
6145                         }
6146                 }
6147                 break;
6148         case 2: /* clts */
6149                 WARN_ONCE(1, "Guest should always own CR0.TS");
6150                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
6151                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
6152                 return kvm_skip_emulated_instruction(vcpu);
6153         case 1: /*mov from cr*/
6154                 switch (cr) {
6155                 case 3:
6156                         val = kvm_read_cr3(vcpu);
6157                         kvm_register_write(vcpu, reg, val);
6158                         trace_kvm_cr_read(cr, val);
6159                         return kvm_skip_emulated_instruction(vcpu);
6160                 case 8:
6161                         val = kvm_get_cr8(vcpu);
6162                         kvm_register_write(vcpu, reg, val);
6163                         trace_kvm_cr_read(cr, val);
6164                         return kvm_skip_emulated_instruction(vcpu);
6165                 }
6166                 break;
6167         case 3: /* lmsw */
6168                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6169                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
6170                 kvm_lmsw(vcpu, val);
6171
6172                 return kvm_skip_emulated_instruction(vcpu);
6173         default:
6174                 break;
6175         }
6176         vcpu->run->exit_reason = 0;
6177         vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6178                (int)(exit_qualification >> 4) & 3, cr);
6179         return 0;
6180 }
6181
6182 static int handle_dr(struct kvm_vcpu *vcpu)
6183 {
6184         unsigned long exit_qualification;
6185         int dr, dr7, reg;
6186
6187         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6188         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6189
6190         /* First, if DR does not exist, trigger UD */
6191         if (!kvm_require_dr(vcpu, dr))
6192                 return 1;
6193
6194         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6195         if (!kvm_require_cpl(vcpu, 0))
6196                 return 1;
6197         dr7 = vmcs_readl(GUEST_DR7);
6198         if (dr7 & DR7_GD) {
6199                 /*
6200                  * As the vm-exit takes precedence over the debug trap, we
6201                  * need to emulate the latter, either for the host or the
6202                  * guest debugging itself.
6203                  */
6204                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6205                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6206                         vcpu->run->debug.arch.dr7 = dr7;
6207                         vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6208                         vcpu->run->debug.arch.exception = DB_VECTOR;
6209                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6210                         return 0;
6211                 } else {
6212                         vcpu->arch.dr6 &= ~15;
6213                         vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6214                         kvm_queue_exception(vcpu, DB_VECTOR);
6215                         return 1;
6216                 }
6217         }
6218
6219         if (vcpu->guest_debug == 0) {
6220                 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6221                                 CPU_BASED_MOV_DR_EXITING);
6222
6223                 /*
6224                  * No more DR vmexits; force a reload of the debug registers
6225                  * and reenter on this instruction.  The next vmexit will
6226                  * retrieve the full state of the debug registers.
6227                  */
6228                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
6229                 return 1;
6230         }
6231
6232         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
6233         if (exit_qualification & TYPE_MOV_FROM_DR) {
6234                 unsigned long val;
6235
6236                 if (kvm_get_dr(vcpu, dr, &val))
6237                         return 1;
6238                 kvm_register_write(vcpu, reg, val);
6239         } else
6240                 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
6241                         return 1;
6242
6243         return kvm_skip_emulated_instruction(vcpu);
6244 }
6245
6246 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
6247 {
6248         return vcpu->arch.dr6;
6249 }
6250
6251 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
6252 {
6253 }
6254
6255 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
6256 {
6257         get_debugreg(vcpu->arch.db[0], 0);
6258         get_debugreg(vcpu->arch.db[1], 1);
6259         get_debugreg(vcpu->arch.db[2], 2);
6260         get_debugreg(vcpu->arch.db[3], 3);
6261         get_debugreg(vcpu->arch.dr6, 6);
6262         vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6263
6264         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6265         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6266 }
6267
6268 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6269 {
6270         vmcs_writel(GUEST_DR7, val);
6271 }
6272
6273 static int handle_cpuid(struct kvm_vcpu *vcpu)
6274 {
6275         return kvm_emulate_cpuid(vcpu);
6276 }
6277
6278 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6279 {
6280         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6281         struct msr_data msr_info;
6282
6283         msr_info.index = ecx;
6284         msr_info.host_initiated = false;
6285         if (vmx_get_msr(vcpu, &msr_info)) {
6286                 trace_kvm_msr_read_ex(ecx);
6287                 kvm_inject_gp(vcpu, 0);
6288                 return 1;
6289         }
6290
6291         trace_kvm_msr_read(ecx, msr_info.data);
6292
6293         /* FIXME: handling of bits 32:63 of rax, rdx */
6294         vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6295         vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6296         return kvm_skip_emulated_instruction(vcpu);
6297 }
6298
6299 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6300 {
6301         struct msr_data msr;
6302         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6303         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6304                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6305
6306         msr.data = data;
6307         msr.index = ecx;
6308         msr.host_initiated = false;
6309         if (kvm_set_msr(vcpu, &msr) != 0) {
6310                 trace_kvm_msr_write_ex(ecx, data);
6311                 kvm_inject_gp(vcpu, 0);
6312                 return 1;
6313         }
6314
6315         trace_kvm_msr_write(ecx, data);
6316         return kvm_skip_emulated_instruction(vcpu);
6317 }
6318
6319 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6320 {
6321         kvm_apic_update_ppr(vcpu);
6322         return 1;
6323 }
6324
6325 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6326 {
6327         vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6328                         CPU_BASED_VIRTUAL_INTR_PENDING);
6329
6330         kvm_make_request(KVM_REQ_EVENT, vcpu);
6331
6332         ++vcpu->stat.irq_window_exits;
6333         return 1;
6334 }
6335
6336 static int handle_halt(struct kvm_vcpu *vcpu)
6337 {
6338         return kvm_emulate_halt(vcpu);
6339 }
6340
6341 static int handle_vmcall(struct kvm_vcpu *vcpu)
6342 {
6343         return kvm_emulate_hypercall(vcpu);
6344 }
6345
6346 static int handle_invd(struct kvm_vcpu *vcpu)
6347 {
6348         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6349 }
6350
6351 static int handle_invlpg(struct kvm_vcpu *vcpu)
6352 {
6353         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6354
6355         kvm_mmu_invlpg(vcpu, exit_qualification);
6356         return kvm_skip_emulated_instruction(vcpu);
6357 }
6358
6359 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6360 {
6361         int err;
6362
6363         err = kvm_rdpmc(vcpu);
6364         return kvm_complete_insn_gp(vcpu, err);
6365 }
6366
6367 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6368 {
6369         return kvm_emulate_wbinvd(vcpu);
6370 }
6371
6372 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6373 {
6374         u64 new_bv = kvm_read_edx_eax(vcpu);
6375         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6376
6377         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6378                 return kvm_skip_emulated_instruction(vcpu);
6379         return 1;
6380 }
6381
6382 static int handle_xsaves(struct kvm_vcpu *vcpu)
6383 {
6384         kvm_skip_emulated_instruction(vcpu);
6385         WARN(1, "this should never happen\n");
6386         return 1;
6387 }
6388
6389 static int handle_xrstors(struct kvm_vcpu *vcpu)
6390 {
6391         kvm_skip_emulated_instruction(vcpu);
6392         WARN(1, "this should never happen\n");
6393         return 1;
6394 }
6395
6396 static int handle_apic_access(struct kvm_vcpu *vcpu)
6397 {
6398         if (likely(fasteoi)) {
6399                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6400                 int access_type, offset;
6401
6402                 access_type = exit_qualification & APIC_ACCESS_TYPE;
6403                 offset = exit_qualification & APIC_ACCESS_OFFSET;
6404                 /*
6405                  * Sane guest uses MOV to write EOI, with written value
6406                  * not cared. So make a short-circuit here by avoiding
6407                  * heavy instruction emulation.
6408                  */
6409                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6410                     (offset == APIC_EOI)) {
6411                         kvm_lapic_set_eoi(vcpu);
6412                         return kvm_skip_emulated_instruction(vcpu);
6413                 }
6414         }
6415         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6416 }
6417
6418 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6419 {
6420         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6421         int vector = exit_qualification & 0xff;
6422
6423         /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6424         kvm_apic_set_eoi_accelerated(vcpu, vector);
6425         return 1;
6426 }
6427
6428 static int handle_apic_write(struct kvm_vcpu *vcpu)
6429 {
6430         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6431         u32 offset = exit_qualification & 0xfff;
6432
6433         /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6434         kvm_apic_write_nodecode(vcpu, offset);
6435         return 1;
6436 }
6437
6438 static int handle_task_switch(struct kvm_vcpu *vcpu)
6439 {
6440         struct vcpu_vmx *vmx = to_vmx(vcpu);
6441         unsigned long exit_qualification;
6442         bool has_error_code = false;
6443         u32 error_code = 0;
6444         u16 tss_selector;
6445         int reason, type, idt_v, idt_index;
6446
6447         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6448         idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6449         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6450
6451         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6452
6453         reason = (u32)exit_qualification >> 30;
6454         if (reason == TASK_SWITCH_GATE && idt_v) {
6455                 switch (type) {
6456                 case INTR_TYPE_NMI_INTR:
6457                         vcpu->arch.nmi_injected = false;
6458                         vmx_set_nmi_mask(vcpu, true);
6459                         break;
6460                 case INTR_TYPE_EXT_INTR:
6461                 case INTR_TYPE_SOFT_INTR:
6462                         kvm_clear_interrupt_queue(vcpu);
6463                         break;
6464                 case INTR_TYPE_HARD_EXCEPTION:
6465                         if (vmx->idt_vectoring_info &
6466                             VECTORING_INFO_DELIVER_CODE_MASK) {
6467                                 has_error_code = true;
6468                                 error_code =
6469                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
6470                         }
6471                         /* fall through */
6472                 case INTR_TYPE_SOFT_EXCEPTION:
6473                         kvm_clear_exception_queue(vcpu);
6474                         break;
6475                 default:
6476                         break;
6477                 }
6478         }
6479         tss_selector = exit_qualification;
6480
6481         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6482                        type != INTR_TYPE_EXT_INTR &&
6483                        type != INTR_TYPE_NMI_INTR))
6484                 skip_emulated_instruction(vcpu);
6485
6486         if (kvm_task_switch(vcpu, tss_selector,
6487                             type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6488                             has_error_code, error_code) == EMULATE_FAIL) {
6489                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6490                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6491                 vcpu->run->internal.ndata = 0;
6492                 return 0;
6493         }
6494
6495         /*
6496          * TODO: What about debug traps on tss switch?
6497          *       Are we supposed to inject them and update dr6?
6498          */
6499
6500         return 1;
6501 }
6502
6503 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6504 {
6505         unsigned long exit_qualification;
6506         gpa_t gpa;
6507         u64 error_code;
6508
6509         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6510
6511         /*
6512          * EPT violation happened while executing iret from NMI,
6513          * "blocked by NMI" bit has to be set before next VM entry.
6514          * There are errata that may cause this bit to not be set:
6515          * AAK134, BY25.
6516          */
6517         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6518                         enable_vnmi &&
6519                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6520                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6521
6522         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6523         trace_kvm_page_fault(gpa, exit_qualification);
6524
6525         /* Is it a read fault? */
6526         error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6527                      ? PFERR_USER_MASK : 0;
6528         /* Is it a write fault? */
6529         error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6530                       ? PFERR_WRITE_MASK : 0;
6531         /* Is it a fetch fault? */
6532         error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6533                       ? PFERR_FETCH_MASK : 0;
6534         /* ept page table entry is present? */
6535         error_code |= (exit_qualification &
6536                        (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6537                         EPT_VIOLATION_EXECUTABLE))
6538                       ? PFERR_PRESENT_MASK : 0;
6539
6540         error_code |= (exit_qualification & 0x100) != 0 ?
6541                PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
6542
6543         vcpu->arch.exit_qualification = exit_qualification;
6544         return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6545 }
6546
6547 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6548 {
6549         int ret;
6550         gpa_t gpa;
6551
6552         /*
6553          * A nested guest cannot optimize MMIO vmexits, because we have an
6554          * nGPA here instead of the required GPA.
6555          */
6556         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6557         if (!is_guest_mode(vcpu) &&
6558             !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6559                 trace_kvm_fast_mmio(gpa);
6560                 return kvm_skip_emulated_instruction(vcpu);
6561         }
6562
6563         ret = kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
6564         if (ret >= 0)
6565                 return ret;
6566
6567         /* It is the real ept misconfig */
6568         WARN_ON(1);
6569
6570         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6571         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6572
6573         return 0;
6574 }
6575
6576 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6577 {
6578         WARN_ON_ONCE(!enable_vnmi);
6579         vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6580                         CPU_BASED_VIRTUAL_NMI_PENDING);
6581         ++vcpu->stat.nmi_window_exits;
6582         kvm_make_request(KVM_REQ_EVENT, vcpu);
6583
6584         return 1;
6585 }
6586
6587 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6588 {
6589         struct vcpu_vmx *vmx = to_vmx(vcpu);
6590         enum emulation_result err = EMULATE_DONE;
6591         int ret = 1;
6592         u32 cpu_exec_ctrl;
6593         bool intr_window_requested;
6594         unsigned count = 130;
6595
6596         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6597         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6598
6599         while (vmx->emulation_required && count-- != 0) {
6600                 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6601                         return handle_interrupt_window(&vmx->vcpu);
6602
6603                 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
6604                         return 1;
6605
6606                 err = emulate_instruction(vcpu, 0);
6607
6608                 if (err == EMULATE_USER_EXIT) {
6609                         ++vcpu->stat.mmio_exits;
6610                         ret = 0;
6611                         goto out;
6612                 }
6613
6614                 if (err != EMULATE_DONE) {
6615                         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6616                         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6617                         vcpu->run->internal.ndata = 0;
6618                         return 0;
6619                 }
6620
6621                 if (vcpu->arch.halt_request) {
6622                         vcpu->arch.halt_request = 0;
6623                         ret = kvm_vcpu_halt(vcpu);
6624                         goto out;
6625                 }
6626
6627                 if (signal_pending(current))
6628                         goto out;
6629                 if (need_resched())
6630                         schedule();
6631         }
6632
6633 out:
6634         return ret;
6635 }
6636
6637 static int __grow_ple_window(int val)
6638 {
6639         if (ple_window_grow < 1)
6640                 return ple_window;
6641
6642         val = min(val, ple_window_actual_max);
6643
6644         if (ple_window_grow < ple_window)
6645                 val *= ple_window_grow;
6646         else
6647                 val += ple_window_grow;
6648
6649         return val;
6650 }
6651
6652 static int __shrink_ple_window(int val, int modifier, int minimum)
6653 {
6654         if (modifier < 1)
6655                 return ple_window;
6656
6657         if (modifier < ple_window)
6658                 val /= modifier;
6659         else
6660                 val -= modifier;
6661
6662         return max(val, minimum);
6663 }
6664
6665 static void grow_ple_window(struct kvm_vcpu *vcpu)
6666 {
6667         struct vcpu_vmx *vmx = to_vmx(vcpu);
6668         int old = vmx->ple_window;
6669
6670         vmx->ple_window = __grow_ple_window(old);
6671
6672         if (vmx->ple_window != old)
6673                 vmx->ple_window_dirty = true;
6674
6675         trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6676 }
6677
6678 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6679 {
6680         struct vcpu_vmx *vmx = to_vmx(vcpu);
6681         int old = vmx->ple_window;
6682
6683         vmx->ple_window = __shrink_ple_window(old,
6684                                               ple_window_shrink, ple_window);
6685
6686         if (vmx->ple_window != old)
6687                 vmx->ple_window_dirty = true;
6688
6689         trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6690 }
6691
6692 /*
6693  * ple_window_actual_max is computed to be one grow_ple_window() below
6694  * ple_window_max. (See __grow_ple_window for the reason.)
6695  * This prevents overflows, because ple_window_max is int.
6696  * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6697  * this process.
6698  * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6699  */
6700 static void update_ple_window_actual_max(void)
6701 {
6702         ple_window_actual_max =
6703                         __shrink_ple_window(max(ple_window_max, ple_window),
6704                                             ple_window_grow, INT_MIN);
6705 }
6706
6707 /*
6708  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6709  */
6710 static void wakeup_handler(void)
6711 {
6712         struct kvm_vcpu *vcpu;
6713         int cpu = smp_processor_id();
6714
6715         spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6716         list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6717                         blocked_vcpu_list) {
6718                 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6719
6720                 if (pi_test_on(pi_desc) == 1)
6721                         kvm_vcpu_kick(vcpu);
6722         }
6723         spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6724 }
6725
6726 void vmx_enable_tdp(void)
6727 {
6728         kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6729                 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6730                 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6731                 0ull, VMX_EPT_EXECUTABLE_MASK,
6732                 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6733                 VMX_EPT_RWX_MASK, 0ull);
6734
6735         ept_set_mmio_spte_mask();
6736         kvm_enable_tdp();
6737 }
6738
6739 static __init int hardware_setup(void)
6740 {
6741         int r = -ENOMEM, i, msr;
6742
6743         rdmsrl_safe(MSR_EFER, &host_efer);
6744
6745         for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6746                 kvm_define_shared_msr(i, vmx_msr_index[i]);
6747
6748         for (i = 0; i < VMX_BITMAP_NR; i++) {
6749                 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6750                 if (!vmx_bitmap[i])
6751                         goto out;
6752         }
6753
6754         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6755         memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6756         memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6757
6758         /*
6759          * Allow direct access to the PC debug port (it is often used for I/O
6760          * delays, but the vmexits simply slow things down).
6761          */
6762         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6763         clear_bit(0x80, vmx_io_bitmap_a);
6764
6765         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6766
6767         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6768         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6769
6770         if (setup_vmcs_config(&vmcs_config) < 0) {
6771                 r = -EIO;
6772                 goto out;
6773         }
6774
6775         if (boot_cpu_has(X86_FEATURE_NX))
6776                 kvm_enable_efer_bits(EFER_NX);
6777
6778         if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6779                 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6780                 enable_vpid = 0;
6781
6782         if (!cpu_has_vmx_shadow_vmcs())
6783                 enable_shadow_vmcs = 0;
6784         if (enable_shadow_vmcs)
6785                 init_vmcs_shadow_fields();
6786
6787         if (!cpu_has_vmx_ept() ||
6788             !cpu_has_vmx_ept_4levels() ||
6789             !cpu_has_vmx_ept_mt_wb() ||
6790             !cpu_has_vmx_invept_global())
6791                 enable_ept = 0;
6792
6793         if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
6794                 enable_ept_ad_bits = 0;
6795
6796         if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
6797                 enable_unrestricted_guest = 0;
6798
6799         if (!cpu_has_vmx_flexpriority())
6800                 flexpriority_enabled = 0;
6801
6802         if (!cpu_has_virtual_nmis())
6803                 enable_vnmi = 0;
6804
6805         /*
6806          * set_apic_access_page_addr() is used to reload apic access
6807          * page upon invalidation.  No need to do anything if not
6808          * using the APIC_ACCESS_ADDR VMCS field.
6809          */
6810         if (!flexpriority_enabled)
6811                 kvm_x86_ops->set_apic_access_page_addr = NULL;
6812
6813         if (!cpu_has_vmx_tpr_shadow())
6814                 kvm_x86_ops->update_cr8_intercept = NULL;
6815
6816         if (enable_ept && !cpu_has_vmx_ept_2m_page())
6817                 kvm_disable_largepages();
6818
6819         if (!cpu_has_vmx_ple()) {
6820                 ple_gap = 0;
6821                 ple_window = 0;
6822                 ple_window_grow = 0;
6823                 ple_window_max = 0;
6824                 ple_window_shrink = 0;
6825         }
6826
6827         if (!cpu_has_vmx_apicv()) {
6828                 enable_apicv = 0;
6829                 kvm_x86_ops->sync_pir_to_irr = NULL;
6830         }
6831
6832         if (cpu_has_vmx_tsc_scaling()) {
6833                 kvm_has_tsc_control = true;
6834                 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6835                 kvm_tsc_scaling_ratio_frac_bits = 48;
6836         }
6837
6838         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6839         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6840         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6841         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6842         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6843         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6844
6845         memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6846                         vmx_msr_bitmap_legacy, PAGE_SIZE);
6847         memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6848                         vmx_msr_bitmap_longmode, PAGE_SIZE);
6849         memcpy(vmx_msr_bitmap_legacy_x2apic,
6850                         vmx_msr_bitmap_legacy, PAGE_SIZE);
6851         memcpy(vmx_msr_bitmap_longmode_x2apic,
6852                         vmx_msr_bitmap_longmode, PAGE_SIZE);
6853
6854         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6855
6856         for (msr = 0x800; msr <= 0x8ff; msr++) {
6857                 if (msr == 0x839 /* TMCCT */)
6858                         continue;
6859                 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6860         }
6861
6862         /*
6863          * TPR reads and writes can be virtualized even if virtual interrupt
6864          * delivery is not in use.
6865          */
6866         vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6867         vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6868
6869         /* EOI */
6870         vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6871         /* SELF-IPI */
6872         vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6873
6874         if (enable_ept)
6875                 vmx_enable_tdp();
6876         else
6877                 kvm_disable_tdp();
6878
6879         update_ple_window_actual_max();
6880
6881         /*
6882          * Only enable PML when hardware supports PML feature, and both EPT
6883          * and EPT A/D bit features are enabled -- PML depends on them to work.
6884          */
6885         if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6886                 enable_pml = 0;
6887
6888         if (!enable_pml) {
6889                 kvm_x86_ops->slot_enable_log_dirty = NULL;
6890                 kvm_x86_ops->slot_disable_log_dirty = NULL;
6891                 kvm_x86_ops->flush_log_dirty = NULL;
6892                 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6893         }
6894
6895         if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6896                 u64 vmx_msr;
6897
6898                 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6899                 cpu_preemption_timer_multi =
6900                          vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6901         } else {
6902                 kvm_x86_ops->set_hv_timer = NULL;
6903                 kvm_x86_ops->cancel_hv_timer = NULL;
6904         }
6905
6906         kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6907
6908         kvm_mce_cap_supported |= MCG_LMCE_P;
6909
6910         return alloc_kvm_area();
6911
6912 out:
6913         for (i = 0; i < VMX_BITMAP_NR; i++)
6914                 free_page((unsigned long)vmx_bitmap[i]);
6915
6916     return r;
6917 }
6918
6919 static __exit void hardware_unsetup(void)
6920 {
6921         int i;
6922
6923         for (i = 0; i < VMX_BITMAP_NR; i++)
6924                 free_page((unsigned long)vmx_bitmap[i]);
6925
6926         free_kvm_area();
6927 }
6928
6929 /*
6930  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6931  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6932  */
6933 static int handle_pause(struct kvm_vcpu *vcpu)
6934 {
6935         if (ple_gap)
6936                 grow_ple_window(vcpu);
6937
6938         /*
6939          * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
6940          * VM-execution control is ignored if CPL > 0. OTOH, KVM
6941          * never set PAUSE_EXITING and just set PLE if supported,
6942          * so the vcpu must be CPL=0 if it gets a PAUSE exit.
6943          */
6944         kvm_vcpu_on_spin(vcpu, true);
6945         return kvm_skip_emulated_instruction(vcpu);
6946 }
6947
6948 static int handle_nop(struct kvm_vcpu *vcpu)
6949 {
6950         return kvm_skip_emulated_instruction(vcpu);
6951 }
6952
6953 static int handle_mwait(struct kvm_vcpu *vcpu)
6954 {
6955         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6956         return handle_nop(vcpu);
6957 }
6958
6959 static int handle_invalid_op(struct kvm_vcpu *vcpu)
6960 {
6961         kvm_queue_exception(vcpu, UD_VECTOR);
6962         return 1;
6963 }
6964
6965 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6966 {
6967         return 1;
6968 }
6969
6970 static int handle_monitor(struct kvm_vcpu *vcpu)
6971 {
6972         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6973         return handle_nop(vcpu);
6974 }
6975
6976 /*
6977  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6978  * We could reuse a single VMCS for all the L2 guests, but we also want the
6979  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6980  * allows keeping them loaded on the processor, and in the future will allow
6981  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6982  * every entry if they never change.
6983  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6984  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6985  *
6986  * The following functions allocate and free a vmcs02 in this pool.
6987  */
6988
6989 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6990 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6991 {
6992         struct vmcs02_list *item;
6993         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6994                 if (item->vmptr == vmx->nested.current_vmptr) {
6995                         list_move(&item->list, &vmx->nested.vmcs02_pool);
6996                         return &item->vmcs02;
6997                 }
6998
6999         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
7000                 /* Recycle the least recently used VMCS. */
7001                 item = list_last_entry(&vmx->nested.vmcs02_pool,
7002                                        struct vmcs02_list, list);
7003                 item->vmptr = vmx->nested.current_vmptr;
7004                 list_move(&item->list, &vmx->nested.vmcs02_pool);
7005                 return &item->vmcs02;
7006         }
7007
7008         /* Create a new VMCS */
7009         item = kzalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
7010         if (!item)
7011                 return NULL;
7012         item->vmcs02.vmcs = alloc_vmcs();
7013         item->vmcs02.shadow_vmcs = NULL;
7014         if (!item->vmcs02.vmcs) {
7015                 kfree(item);
7016                 return NULL;
7017         }
7018         loaded_vmcs_init(&item->vmcs02);
7019         item->vmptr = vmx->nested.current_vmptr;
7020         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
7021         vmx->nested.vmcs02_num++;
7022         return &item->vmcs02;
7023 }
7024
7025 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
7026 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
7027 {
7028         struct vmcs02_list *item;
7029         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
7030                 if (item->vmptr == vmptr) {
7031                         free_loaded_vmcs(&item->vmcs02);
7032                         list_del(&item->list);
7033                         kfree(item);
7034                         vmx->nested.vmcs02_num--;
7035                         return;
7036                 }
7037 }
7038
7039 /*
7040  * Free all VMCSs saved for this vcpu, except the one pointed by
7041  * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
7042  * must be &vmx->vmcs01.
7043  */
7044 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
7045 {
7046         struct vmcs02_list *item, *n;
7047
7048         WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
7049         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
7050                 /*
7051                  * Something will leak if the above WARN triggers.  Better than
7052                  * a use-after-free.
7053                  */
7054                 if (vmx->loaded_vmcs == &item->vmcs02)
7055                         continue;
7056
7057                 free_loaded_vmcs(&item->vmcs02);
7058                 list_del(&item->list);
7059                 kfree(item);
7060                 vmx->nested.vmcs02_num--;
7061         }
7062 }
7063
7064 /*
7065  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
7066  * set the success or error code of an emulated VMX instruction, as specified
7067  * by Vol 2B, VMX Instruction Reference, "Conventions".
7068  */
7069 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
7070 {
7071         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
7072                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7073                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
7074 }
7075
7076 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
7077 {
7078         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7079                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
7080                             X86_EFLAGS_SF | X86_EFLAGS_OF))
7081                         | X86_EFLAGS_CF);
7082 }
7083
7084 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
7085                                         u32 vm_instruction_error)
7086 {
7087         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
7088                 /*
7089                  * failValid writes the error number to the current VMCS, which
7090                  * can't be done there isn't a current VMCS.
7091                  */
7092                 nested_vmx_failInvalid(vcpu);
7093                 return;
7094         }
7095         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7096                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7097                             X86_EFLAGS_SF | X86_EFLAGS_OF))
7098                         | X86_EFLAGS_ZF);
7099         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
7100         /*
7101          * We don't need to force a shadow sync because
7102          * VM_INSTRUCTION_ERROR is not shadowed
7103          */
7104 }
7105
7106 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
7107 {
7108         /* TODO: not to reset guest simply here. */
7109         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7110         pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
7111 }
7112
7113 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
7114 {
7115         struct vcpu_vmx *vmx =
7116                 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
7117
7118         vmx->nested.preemption_timer_expired = true;
7119         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
7120         kvm_vcpu_kick(&vmx->vcpu);
7121
7122         return HRTIMER_NORESTART;
7123 }
7124
7125 /*
7126  * Decode the memory-address operand of a vmx instruction, as recorded on an
7127  * exit caused by such an instruction (run by a guest hypervisor).
7128  * On success, returns 0. When the operand is invalid, returns 1 and throws
7129  * #UD or #GP.
7130  */
7131 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
7132                                  unsigned long exit_qualification,
7133                                  u32 vmx_instruction_info, bool wr, gva_t *ret)
7134 {
7135         gva_t off;
7136         bool exn;
7137         struct kvm_segment s;
7138
7139         /*
7140          * According to Vol. 3B, "Information for VM Exits Due to Instruction
7141          * Execution", on an exit, vmx_instruction_info holds most of the
7142          * addressing components of the operand. Only the displacement part
7143          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
7144          * For how an actual address is calculated from all these components,
7145          * refer to Vol. 1, "Operand Addressing".
7146          */
7147         int  scaling = vmx_instruction_info & 3;
7148         int  addr_size = (vmx_instruction_info >> 7) & 7;
7149         bool is_reg = vmx_instruction_info & (1u << 10);
7150         int  seg_reg = (vmx_instruction_info >> 15) & 7;
7151         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
7152         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
7153         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
7154         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
7155
7156         if (is_reg) {
7157                 kvm_queue_exception(vcpu, UD_VECTOR);
7158                 return 1;
7159         }
7160
7161         /* Addr = segment_base + offset */
7162         /* offset = base + [index * scale] + displacement */
7163         off = exit_qualification; /* holds the displacement */
7164         if (base_is_valid)
7165                 off += kvm_register_read(vcpu, base_reg);
7166         if (index_is_valid)
7167                 off += kvm_register_read(vcpu, index_reg)<<scaling;
7168         vmx_get_segment(vcpu, &s, seg_reg);
7169         *ret = s.base + off;
7170
7171         if (addr_size == 1) /* 32 bit */
7172                 *ret &= 0xffffffff;
7173
7174         /* Checks for #GP/#SS exceptions. */
7175         exn = false;
7176         if (is_long_mode(vcpu)) {
7177                 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
7178                  * non-canonical form. This is the only check on the memory
7179                  * destination for long mode!
7180                  */
7181                 exn = is_noncanonical_address(*ret, vcpu);
7182         } else if (is_protmode(vcpu)) {
7183                 /* Protected mode: apply checks for segment validity in the
7184                  * following order:
7185                  * - segment type check (#GP(0) may be thrown)
7186                  * - usability check (#GP(0)/#SS(0))
7187                  * - limit check (#GP(0)/#SS(0))
7188                  */
7189                 if (wr)
7190                         /* #GP(0) if the destination operand is located in a
7191                          * read-only data segment or any code segment.
7192                          */
7193                         exn = ((s.type & 0xa) == 0 || (s.type & 8));
7194                 else
7195                         /* #GP(0) if the source operand is located in an
7196                          * execute-only code segment
7197                          */
7198                         exn = ((s.type & 0xa) == 8);
7199                 if (exn) {
7200                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7201                         return 1;
7202                 }
7203                 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
7204                  */
7205                 exn = (s.unusable != 0);
7206                 /* Protected mode: #GP(0)/#SS(0) if the memory
7207                  * operand is outside the segment limit.
7208                  */
7209                 exn = exn || (off + sizeof(u64) > s.limit);
7210         }
7211         if (exn) {
7212                 kvm_queue_exception_e(vcpu,
7213                                       seg_reg == VCPU_SREG_SS ?
7214                                                 SS_VECTOR : GP_VECTOR,
7215                                       0);
7216                 return 1;
7217         }
7218
7219         return 0;
7220 }
7221
7222 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
7223 {
7224         gva_t gva;
7225         struct x86_exception e;
7226
7227         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7228                         vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
7229                 return 1;
7230
7231         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, vmpointer,
7232                                 sizeof(*vmpointer), &e)) {
7233                 kvm_inject_page_fault(vcpu, &e);
7234                 return 1;
7235         }
7236
7237         return 0;
7238 }
7239
7240 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7241 {
7242         struct vcpu_vmx *vmx = to_vmx(vcpu);
7243         struct vmcs *shadow_vmcs;
7244
7245         if (cpu_has_vmx_msr_bitmap()) {
7246                 vmx->nested.msr_bitmap =
7247                                 (unsigned long *)__get_free_page(GFP_KERNEL);
7248                 if (!vmx->nested.msr_bitmap)
7249                         goto out_msr_bitmap;
7250         }
7251
7252         vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7253         if (!vmx->nested.cached_vmcs12)
7254                 goto out_cached_vmcs12;
7255
7256         if (enable_shadow_vmcs) {
7257                 shadow_vmcs = alloc_vmcs();
7258                 if (!shadow_vmcs)
7259                         goto out_shadow_vmcs;
7260                 /* mark vmcs as shadow */
7261                 shadow_vmcs->revision_id |= (1u << 31);
7262                 /* init shadow vmcs */
7263                 vmcs_clear(shadow_vmcs);
7264                 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7265         }
7266
7267         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7268         vmx->nested.vmcs02_num = 0;
7269
7270         hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7271                      HRTIMER_MODE_REL_PINNED);
7272         vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7273
7274         vmx->nested.vmxon = true;
7275         return 0;
7276
7277 out_shadow_vmcs:
7278         kfree(vmx->nested.cached_vmcs12);
7279
7280 out_cached_vmcs12:
7281         free_page((unsigned long)vmx->nested.msr_bitmap);
7282
7283 out_msr_bitmap:
7284         return -ENOMEM;
7285 }
7286
7287 /*
7288  * Emulate the VMXON instruction.
7289  * Currently, we just remember that VMX is active, and do not save or even
7290  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7291  * do not currently need to store anything in that guest-allocated memory
7292  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7293  * argument is different from the VMXON pointer (which the spec says they do).
7294  */
7295 static int handle_vmon(struct kvm_vcpu *vcpu)
7296 {
7297         int ret;
7298         gpa_t vmptr;
7299         struct page *page;
7300         struct vcpu_vmx *vmx = to_vmx(vcpu);
7301         const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7302                 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7303
7304         /*
7305          * The Intel VMX Instruction Reference lists a bunch of bits that are
7306          * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7307          * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7308          * Otherwise, we should fail with #UD.  But most faulting conditions
7309          * have already been checked by hardware, prior to the VM-exit for
7310          * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
7311          * that bit set to 1 in non-root mode.
7312          */
7313         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7314                 kvm_queue_exception(vcpu, UD_VECTOR);
7315                 return 1;
7316         }
7317
7318         if (vmx->nested.vmxon) {
7319                 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7320                 return kvm_skip_emulated_instruction(vcpu);
7321         }
7322
7323         if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7324                         != VMXON_NEEDED_FEATURES) {
7325                 kvm_inject_gp(vcpu, 0);
7326                 return 1;
7327         }
7328
7329         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7330                 return 1;
7331
7332         /*
7333          * SDM 3: 24.11.5
7334          * The first 4 bytes of VMXON region contain the supported
7335          * VMCS revision identifier
7336          *
7337          * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7338          * which replaces physical address width with 32
7339          */
7340         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7341                 nested_vmx_failInvalid(vcpu);
7342                 return kvm_skip_emulated_instruction(vcpu);
7343         }
7344
7345         page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7346         if (is_error_page(page)) {
7347                 nested_vmx_failInvalid(vcpu);
7348                 return kvm_skip_emulated_instruction(vcpu);
7349         }
7350         if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7351                 kunmap(page);
7352                 kvm_release_page_clean(page);
7353                 nested_vmx_failInvalid(vcpu);
7354                 return kvm_skip_emulated_instruction(vcpu);
7355         }
7356         kunmap(page);
7357         kvm_release_page_clean(page);
7358
7359         vmx->nested.vmxon_ptr = vmptr;
7360         ret = enter_vmx_operation(vcpu);
7361         if (ret)
7362                 return ret;
7363
7364         nested_vmx_succeed(vcpu);
7365         return kvm_skip_emulated_instruction(vcpu);
7366 }
7367
7368 /*
7369  * Intel's VMX Instruction Reference specifies a common set of prerequisites
7370  * for running VMX instructions (except VMXON, whose prerequisites are
7371  * slightly different). It also specifies what exception to inject otherwise.
7372  * Note that many of these exceptions have priority over VM exits, so they
7373  * don't have to be checked again here.
7374  */
7375 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7376 {
7377         if (!to_vmx(vcpu)->nested.vmxon) {
7378                 kvm_queue_exception(vcpu, UD_VECTOR);
7379                 return 0;
7380         }
7381         return 1;
7382 }
7383
7384 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
7385 {
7386         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
7387         vmcs_write64(VMCS_LINK_POINTER, -1ull);
7388 }
7389
7390 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7391 {
7392         if (vmx->nested.current_vmptr == -1ull)
7393                 return;
7394
7395         if (enable_shadow_vmcs) {
7396                 /* copy to memory all shadowed fields in case
7397                    they were modified */
7398                 copy_shadow_to_vmcs12(vmx);
7399                 vmx->nested.sync_shadow_vmcs = false;
7400                 vmx_disable_shadow_vmcs(vmx);
7401         }
7402         vmx->nested.posted_intr_nv = -1;
7403
7404         /* Flush VMCS12 to guest memory */
7405         kvm_vcpu_write_guest_page(&vmx->vcpu,
7406                                   vmx->nested.current_vmptr >> PAGE_SHIFT,
7407                                   vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
7408
7409         vmx->nested.current_vmptr = -1ull;
7410 }
7411
7412 /*
7413  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7414  * just stops using VMX.
7415  */
7416 static void free_nested(struct vcpu_vmx *vmx)
7417 {
7418         if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
7419                 return;
7420
7421         vmx->nested.vmxon = false;
7422         vmx->nested.smm.vmxon = false;
7423         free_vpid(vmx->nested.vpid02);
7424         vmx->nested.posted_intr_nv = -1;
7425         vmx->nested.current_vmptr = -1ull;
7426         if (vmx->nested.msr_bitmap) {
7427                 free_page((unsigned long)vmx->nested.msr_bitmap);
7428                 vmx->nested.msr_bitmap = NULL;
7429         }
7430         if (enable_shadow_vmcs) {
7431                 vmx_disable_shadow_vmcs(vmx);
7432                 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7433                 free_vmcs(vmx->vmcs01.shadow_vmcs);
7434                 vmx->vmcs01.shadow_vmcs = NULL;
7435         }
7436         kfree(vmx->nested.cached_vmcs12);
7437         /* Unpin physical memory we referred to in current vmcs02 */
7438         if (vmx->nested.apic_access_page) {
7439                 kvm_release_page_dirty(vmx->nested.apic_access_page);
7440                 vmx->nested.apic_access_page = NULL;
7441         }
7442         if (vmx->nested.virtual_apic_page) {
7443                 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
7444                 vmx->nested.virtual_apic_page = NULL;
7445         }
7446         if (vmx->nested.pi_desc_page) {
7447                 kunmap(vmx->nested.pi_desc_page);
7448                 kvm_release_page_dirty(vmx->nested.pi_desc_page);
7449                 vmx->nested.pi_desc_page = NULL;
7450                 vmx->nested.pi_desc = NULL;
7451         }
7452
7453         nested_free_all_saved_vmcss(vmx);
7454 }
7455
7456 /* Emulate the VMXOFF instruction */
7457 static int handle_vmoff(struct kvm_vcpu *vcpu)
7458 {
7459         if (!nested_vmx_check_permission(vcpu))
7460                 return 1;
7461         free_nested(to_vmx(vcpu));
7462         nested_vmx_succeed(vcpu);
7463         return kvm_skip_emulated_instruction(vcpu);
7464 }
7465
7466 /* Emulate the VMCLEAR instruction */
7467 static int handle_vmclear(struct kvm_vcpu *vcpu)
7468 {
7469         struct vcpu_vmx *vmx = to_vmx(vcpu);
7470         u32 zero = 0;
7471         gpa_t vmptr;
7472
7473         if (!nested_vmx_check_permission(vcpu))
7474                 return 1;
7475
7476         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7477                 return 1;
7478
7479         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7480                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
7481                 return kvm_skip_emulated_instruction(vcpu);
7482         }
7483
7484         if (vmptr == vmx->nested.vmxon_ptr) {
7485                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
7486                 return kvm_skip_emulated_instruction(vcpu);
7487         }
7488
7489         if (vmptr == vmx->nested.current_vmptr)
7490                 nested_release_vmcs12(vmx);
7491
7492         kvm_vcpu_write_guest(vcpu,
7493                         vmptr + offsetof(struct vmcs12, launch_state),
7494                         &zero, sizeof(zero));
7495
7496         nested_free_vmcs02(vmx, vmptr);
7497
7498         nested_vmx_succeed(vcpu);
7499         return kvm_skip_emulated_instruction(vcpu);
7500 }
7501
7502 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7503
7504 /* Emulate the VMLAUNCH instruction */
7505 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7506 {
7507         return nested_vmx_run(vcpu, true);
7508 }
7509
7510 /* Emulate the VMRESUME instruction */
7511 static int handle_vmresume(struct kvm_vcpu *vcpu)
7512 {
7513
7514         return nested_vmx_run(vcpu, false);
7515 }
7516
7517 /*
7518  * Read a vmcs12 field. Since these can have varying lengths and we return
7519  * one type, we chose the biggest type (u64) and zero-extend the return value
7520  * to that size. Note that the caller, handle_vmread, might need to use only
7521  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7522  * 64-bit fields are to be returned).
7523  */
7524 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7525                                   unsigned long field, u64 *ret)
7526 {
7527         short offset = vmcs_field_to_offset(field);
7528         char *p;
7529
7530         if (offset < 0)
7531                 return offset;
7532
7533         p = ((char *)(get_vmcs12(vcpu))) + offset;
7534
7535         switch (vmcs_field_type(field)) {
7536         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7537                 *ret = *((natural_width *)p);
7538                 return 0;
7539         case VMCS_FIELD_TYPE_U16:
7540                 *ret = *((u16 *)p);
7541                 return 0;
7542         case VMCS_FIELD_TYPE_U32:
7543                 *ret = *((u32 *)p);
7544                 return 0;
7545         case VMCS_FIELD_TYPE_U64:
7546                 *ret = *((u64 *)p);
7547                 return 0;
7548         default:
7549                 WARN_ON(1);
7550                 return -ENOENT;
7551         }
7552 }
7553
7554
7555 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7556                                    unsigned long field, u64 field_value){
7557         short offset = vmcs_field_to_offset(field);
7558         char *p = ((char *) get_vmcs12(vcpu)) + offset;
7559         if (offset < 0)
7560                 return offset;
7561
7562         switch (vmcs_field_type(field)) {
7563         case VMCS_FIELD_TYPE_U16:
7564                 *(u16 *)p = field_value;
7565                 return 0;
7566         case VMCS_FIELD_TYPE_U32:
7567                 *(u32 *)p = field_value;
7568                 return 0;
7569         case VMCS_FIELD_TYPE_U64:
7570                 *(u64 *)p = field_value;
7571                 return 0;
7572         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7573                 *(natural_width *)p = field_value;
7574                 return 0;
7575         default:
7576                 WARN_ON(1);
7577                 return -ENOENT;
7578         }
7579
7580 }
7581
7582 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7583 {
7584         int i;
7585         unsigned long field;
7586         u64 field_value;
7587         struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7588         const unsigned long *fields = shadow_read_write_fields;
7589         const int num_fields = max_shadow_read_write_fields;
7590
7591         preempt_disable();
7592
7593         vmcs_load(shadow_vmcs);
7594
7595         for (i = 0; i < num_fields; i++) {
7596                 field = fields[i];
7597                 switch (vmcs_field_type(field)) {
7598                 case VMCS_FIELD_TYPE_U16:
7599                         field_value = vmcs_read16(field);
7600                         break;
7601                 case VMCS_FIELD_TYPE_U32:
7602                         field_value = vmcs_read32(field);
7603                         break;
7604                 case VMCS_FIELD_TYPE_U64:
7605                         field_value = vmcs_read64(field);
7606                         break;
7607                 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7608                         field_value = vmcs_readl(field);
7609                         break;
7610                 default:
7611                         WARN_ON(1);
7612                         continue;
7613                 }
7614                 vmcs12_write_any(&vmx->vcpu, field, field_value);
7615         }
7616
7617         vmcs_clear(shadow_vmcs);
7618         vmcs_load(vmx->loaded_vmcs->vmcs);
7619
7620         preempt_enable();
7621 }
7622
7623 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7624 {
7625         const unsigned long *fields[] = {
7626                 shadow_read_write_fields,
7627                 shadow_read_only_fields
7628         };
7629         const int max_fields[] = {
7630                 max_shadow_read_write_fields,
7631                 max_shadow_read_only_fields
7632         };
7633         int i, q;
7634         unsigned long field;
7635         u64 field_value = 0;
7636         struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7637
7638         vmcs_load(shadow_vmcs);
7639
7640         for (q = 0; q < ARRAY_SIZE(fields); q++) {
7641                 for (i = 0; i < max_fields[q]; i++) {
7642                         field = fields[q][i];
7643                         vmcs12_read_any(&vmx->vcpu, field, &field_value);
7644
7645                         switch (vmcs_field_type(field)) {
7646                         case VMCS_FIELD_TYPE_U16:
7647                                 vmcs_write16(field, (u16)field_value);
7648                                 break;
7649                         case VMCS_FIELD_TYPE_U32:
7650                                 vmcs_write32(field, (u32)field_value);
7651                                 break;
7652                         case VMCS_FIELD_TYPE_U64:
7653                                 vmcs_write64(field, (u64)field_value);
7654                                 break;
7655                         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7656                                 vmcs_writel(field, (long)field_value);
7657                                 break;
7658                         default:
7659                                 WARN_ON(1);
7660                                 break;
7661                         }
7662                 }
7663         }
7664
7665         vmcs_clear(shadow_vmcs);
7666         vmcs_load(vmx->loaded_vmcs->vmcs);
7667 }
7668
7669 /*
7670  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7671  * used before) all generate the same failure when it is missing.
7672  */
7673 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7674 {
7675         struct vcpu_vmx *vmx = to_vmx(vcpu);
7676         if (vmx->nested.current_vmptr == -1ull) {
7677                 nested_vmx_failInvalid(vcpu);
7678                 return 0;
7679         }
7680         return 1;
7681 }
7682
7683 static int handle_vmread(struct kvm_vcpu *vcpu)
7684 {
7685         unsigned long field;
7686         u64 field_value;
7687         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7688         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7689         gva_t gva = 0;
7690
7691         if (!nested_vmx_check_permission(vcpu))
7692                 return 1;
7693
7694         if (!nested_vmx_check_vmcs12(vcpu))
7695                 return kvm_skip_emulated_instruction(vcpu);
7696
7697         /* Decode instruction info and find the field to read */
7698         field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7699         /* Read the field, zero-extended to a u64 field_value */
7700         if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7701                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7702                 return kvm_skip_emulated_instruction(vcpu);
7703         }
7704         /*
7705          * Now copy part of this value to register or memory, as requested.
7706          * Note that the number of bits actually copied is 32 or 64 depending
7707          * on the guest's mode (32 or 64 bit), not on the given field's length.
7708          */
7709         if (vmx_instruction_info & (1u << 10)) {
7710                 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7711                         field_value);
7712         } else {
7713                 if (get_vmx_mem_address(vcpu, exit_qualification,
7714                                 vmx_instruction_info, true, &gva))
7715                         return 1;
7716                 /* _system ok, as hardware has verified cpl=0 */
7717                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7718                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7719         }
7720
7721         nested_vmx_succeed(vcpu);
7722         return kvm_skip_emulated_instruction(vcpu);
7723 }
7724
7725
7726 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7727 {
7728         unsigned long field;
7729         gva_t gva;
7730         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7731         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7732         /* The value to write might be 32 or 64 bits, depending on L1's long
7733          * mode, and eventually we need to write that into a field of several
7734          * possible lengths. The code below first zero-extends the value to 64
7735          * bit (field_value), and then copies only the appropriate number of
7736          * bits into the vmcs12 field.
7737          */
7738         u64 field_value = 0;
7739         struct x86_exception e;
7740
7741         if (!nested_vmx_check_permission(vcpu))
7742                 return 1;
7743
7744         if (!nested_vmx_check_vmcs12(vcpu))
7745                 return kvm_skip_emulated_instruction(vcpu);
7746
7747         if (vmx_instruction_info & (1u << 10))
7748                 field_value = kvm_register_readl(vcpu,
7749                         (((vmx_instruction_info) >> 3) & 0xf));
7750         else {
7751                 if (get_vmx_mem_address(vcpu, exit_qualification,
7752                                 vmx_instruction_info, false, &gva))
7753                         return 1;
7754                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7755                            &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7756                         kvm_inject_page_fault(vcpu, &e);
7757                         return 1;
7758                 }
7759         }
7760
7761
7762         field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7763         if (vmcs_field_readonly(field)) {
7764                 nested_vmx_failValid(vcpu,
7765                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7766                 return kvm_skip_emulated_instruction(vcpu);
7767         }
7768
7769         if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7770                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7771                 return kvm_skip_emulated_instruction(vcpu);
7772         }
7773
7774         nested_vmx_succeed(vcpu);
7775         return kvm_skip_emulated_instruction(vcpu);
7776 }
7777
7778 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7779 {
7780         vmx->nested.current_vmptr = vmptr;
7781         if (enable_shadow_vmcs) {
7782                 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7783                               SECONDARY_EXEC_SHADOW_VMCS);
7784                 vmcs_write64(VMCS_LINK_POINTER,
7785                              __pa(vmx->vmcs01.shadow_vmcs));
7786                 vmx->nested.sync_shadow_vmcs = true;
7787         }
7788 }
7789
7790 /* Emulate the VMPTRLD instruction */
7791 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7792 {
7793         struct vcpu_vmx *vmx = to_vmx(vcpu);
7794         gpa_t vmptr;
7795
7796         if (!nested_vmx_check_permission(vcpu))
7797                 return 1;
7798
7799         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7800                 return 1;
7801
7802         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7803                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
7804                 return kvm_skip_emulated_instruction(vcpu);
7805         }
7806
7807         if (vmptr == vmx->nested.vmxon_ptr) {
7808                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
7809                 return kvm_skip_emulated_instruction(vcpu);
7810         }
7811
7812         if (vmx->nested.current_vmptr != vmptr) {
7813                 struct vmcs12 *new_vmcs12;
7814                 struct page *page;
7815                 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7816                 if (is_error_page(page)) {
7817                         nested_vmx_failInvalid(vcpu);
7818                         return kvm_skip_emulated_instruction(vcpu);
7819                 }
7820                 new_vmcs12 = kmap(page);
7821                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7822                         kunmap(page);
7823                         kvm_release_page_clean(page);
7824                         nested_vmx_failValid(vcpu,
7825                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7826                         return kvm_skip_emulated_instruction(vcpu);
7827                 }
7828
7829                 nested_release_vmcs12(vmx);
7830                 /*
7831                  * Load VMCS12 from guest memory since it is not already
7832                  * cached.
7833                  */
7834                 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
7835                 kunmap(page);
7836                 kvm_release_page_clean(page);
7837
7838                 set_current_vmptr(vmx, vmptr);
7839         }
7840
7841         nested_vmx_succeed(vcpu);
7842         return kvm_skip_emulated_instruction(vcpu);
7843 }
7844
7845 /* Emulate the VMPTRST instruction */
7846 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7847 {
7848         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7849         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7850         gva_t vmcs_gva;
7851         struct x86_exception e;
7852
7853         if (!nested_vmx_check_permission(vcpu))
7854                 return 1;
7855
7856         if (get_vmx_mem_address(vcpu, exit_qualification,
7857                         vmx_instruction_info, true, &vmcs_gva))
7858                 return 1;
7859         /* ok to use *_system, as hardware has verified cpl=0 */
7860         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7861                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
7862                                  sizeof(u64), &e)) {
7863                 kvm_inject_page_fault(vcpu, &e);
7864                 return 1;
7865         }
7866         nested_vmx_succeed(vcpu);
7867         return kvm_skip_emulated_instruction(vcpu);
7868 }
7869
7870 /* Emulate the INVEPT instruction */
7871 static int handle_invept(struct kvm_vcpu *vcpu)
7872 {
7873         struct vcpu_vmx *vmx = to_vmx(vcpu);
7874         u32 vmx_instruction_info, types;
7875         unsigned long type;
7876         gva_t gva;
7877         struct x86_exception e;
7878         struct {
7879                 u64 eptp, gpa;
7880         } operand;
7881
7882         if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7883               SECONDARY_EXEC_ENABLE_EPT) ||
7884             !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7885                 kvm_queue_exception(vcpu, UD_VECTOR);
7886                 return 1;
7887         }
7888
7889         if (!nested_vmx_check_permission(vcpu))
7890                 return 1;
7891
7892         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7893         type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7894
7895         types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7896
7897         if (type >= 32 || !(types & (1 << type))) {
7898                 nested_vmx_failValid(vcpu,
7899                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7900                 return kvm_skip_emulated_instruction(vcpu);
7901         }
7902
7903         /* According to the Intel VMX instruction reference, the memory
7904          * operand is read even if it isn't needed (e.g., for type==global)
7905          */
7906         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7907                         vmx_instruction_info, false, &gva))
7908                 return 1;
7909         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7910                                 sizeof(operand), &e)) {
7911                 kvm_inject_page_fault(vcpu, &e);
7912                 return 1;
7913         }
7914
7915         switch (type) {
7916         case VMX_EPT_EXTENT_GLOBAL:
7917         /*
7918          * TODO: track mappings and invalidate
7919          * single context requests appropriately
7920          */
7921         case VMX_EPT_EXTENT_CONTEXT:
7922                 kvm_mmu_sync_roots(vcpu);
7923                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7924                 nested_vmx_succeed(vcpu);
7925                 break;
7926         default:
7927                 BUG_ON(1);
7928                 break;
7929         }
7930
7931         return kvm_skip_emulated_instruction(vcpu);
7932 }
7933
7934 static int handle_invvpid(struct kvm_vcpu *vcpu)
7935 {
7936         struct vcpu_vmx *vmx = to_vmx(vcpu);
7937         u32 vmx_instruction_info;
7938         unsigned long type, types;
7939         gva_t gva;
7940         struct x86_exception e;
7941         struct {
7942                 u64 vpid;
7943                 u64 gla;
7944         } operand;
7945
7946         if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7947               SECONDARY_EXEC_ENABLE_VPID) ||
7948                         !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7949                 kvm_queue_exception(vcpu, UD_VECTOR);
7950                 return 1;
7951         }
7952
7953         if (!nested_vmx_check_permission(vcpu))
7954                 return 1;
7955
7956         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7957         type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7958
7959         types = (vmx->nested.nested_vmx_vpid_caps &
7960                         VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7961
7962         if (type >= 32 || !(types & (1 << type))) {
7963                 nested_vmx_failValid(vcpu,
7964                         VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7965                 return kvm_skip_emulated_instruction(vcpu);
7966         }
7967
7968         /* according to the intel vmx instruction reference, the memory
7969          * operand is read even if it isn't needed (e.g., for type==global)
7970          */
7971         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7972                         vmx_instruction_info, false, &gva))
7973                 return 1;
7974         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7975                                 sizeof(operand), &e)) {
7976                 kvm_inject_page_fault(vcpu, &e);
7977                 return 1;
7978         }
7979         if (operand.vpid >> 16) {
7980                 nested_vmx_failValid(vcpu,
7981                         VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7982                 return kvm_skip_emulated_instruction(vcpu);
7983         }
7984
7985         switch (type) {
7986         case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7987                 if (is_noncanonical_address(operand.gla, vcpu)) {
7988                         nested_vmx_failValid(vcpu,
7989                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7990                         return kvm_skip_emulated_instruction(vcpu);
7991                 }
7992                 /* fall through */
7993         case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7994         case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7995                 if (!operand.vpid) {
7996                         nested_vmx_failValid(vcpu,
7997                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7998                         return kvm_skip_emulated_instruction(vcpu);
7999                 }
8000                 break;
8001         case VMX_VPID_EXTENT_ALL_CONTEXT:
8002                 break;
8003         default:
8004                 WARN_ON_ONCE(1);
8005                 return kvm_skip_emulated_instruction(vcpu);
8006         }
8007
8008         __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
8009         nested_vmx_succeed(vcpu);
8010
8011         return kvm_skip_emulated_instruction(vcpu);
8012 }
8013
8014 static int handle_pml_full(struct kvm_vcpu *vcpu)
8015 {
8016         unsigned long exit_qualification;
8017
8018         trace_kvm_pml_full(vcpu->vcpu_id);
8019
8020         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8021
8022         /*
8023          * PML buffer FULL happened while executing iret from NMI,
8024          * "blocked by NMI" bit has to be set before next VM entry.
8025          */
8026         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
8027                         enable_vnmi &&
8028                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
8029                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8030                                 GUEST_INTR_STATE_NMI);
8031
8032         /*
8033          * PML buffer already flushed at beginning of VMEXIT. Nothing to do
8034          * here.., and there's no userspace involvement needed for PML.
8035          */
8036         return 1;
8037 }
8038
8039 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
8040 {
8041         kvm_lapic_expired_hv_timer(vcpu);
8042         return 1;
8043 }
8044
8045 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
8046 {
8047         struct vcpu_vmx *vmx = to_vmx(vcpu);
8048         int maxphyaddr = cpuid_maxphyaddr(vcpu);
8049
8050         /* Check for memory type validity */
8051         switch (address & VMX_EPTP_MT_MASK) {
8052         case VMX_EPTP_MT_UC:
8053                 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_UC_BIT))
8054                         return false;
8055                 break;
8056         case VMX_EPTP_MT_WB:
8057                 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_WB_BIT))
8058                         return false;
8059                 break;
8060         default:
8061                 return false;
8062         }
8063
8064         /* only 4 levels page-walk length are valid */
8065         if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
8066                 return false;
8067
8068         /* Reserved bits should not be set */
8069         if (address >> maxphyaddr || ((address >> 7) & 0x1f))
8070                 return false;
8071
8072         /* AD, if set, should be supported */
8073         if (address & VMX_EPTP_AD_ENABLE_BIT) {
8074                 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPT_AD_BIT))
8075                         return false;
8076         }
8077
8078         return true;
8079 }
8080
8081 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
8082                                      struct vmcs12 *vmcs12)
8083 {
8084         u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
8085         u64 address;
8086         bool accessed_dirty;
8087         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
8088
8089         if (!nested_cpu_has_eptp_switching(vmcs12) ||
8090             !nested_cpu_has_ept(vmcs12))
8091                 return 1;
8092
8093         if (index >= VMFUNC_EPTP_ENTRIES)
8094                 return 1;
8095
8096
8097         if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
8098                                      &address, index * 8, 8))
8099                 return 1;
8100
8101         accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
8102
8103         /*
8104          * If the (L2) guest does a vmfunc to the currently
8105          * active ept pointer, we don't have to do anything else
8106          */
8107         if (vmcs12->ept_pointer != address) {
8108                 if (!valid_ept_address(vcpu, address))
8109                         return 1;
8110
8111                 kvm_mmu_unload(vcpu);
8112                 mmu->ept_ad = accessed_dirty;
8113                 mmu->base_role.ad_disabled = !accessed_dirty;
8114                 vmcs12->ept_pointer = address;
8115                 /*
8116                  * TODO: Check what's the correct approach in case
8117                  * mmu reload fails. Currently, we just let the next
8118                  * reload potentially fail
8119                  */
8120                 kvm_mmu_reload(vcpu);
8121         }
8122
8123         return 0;
8124 }
8125
8126 static int handle_vmfunc(struct kvm_vcpu *vcpu)
8127 {
8128         struct vcpu_vmx *vmx = to_vmx(vcpu);
8129         struct vmcs12 *vmcs12;
8130         u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
8131
8132         /*
8133          * VMFUNC is only supported for nested guests, but we always enable the
8134          * secondary control for simplicity; for non-nested mode, fake that we
8135          * didn't by injecting #UD.
8136          */
8137         if (!is_guest_mode(vcpu)) {
8138                 kvm_queue_exception(vcpu, UD_VECTOR);
8139                 return 1;
8140         }
8141
8142         vmcs12 = get_vmcs12(vcpu);
8143         if ((vmcs12->vm_function_control & (1 << function)) == 0)
8144                 goto fail;
8145
8146         switch (function) {
8147         case 0:
8148                 if (nested_vmx_eptp_switching(vcpu, vmcs12))
8149                         goto fail;
8150                 break;
8151         default:
8152                 goto fail;
8153         }
8154         return kvm_skip_emulated_instruction(vcpu);
8155
8156 fail:
8157         nested_vmx_vmexit(vcpu, vmx->exit_reason,
8158                           vmcs_read32(VM_EXIT_INTR_INFO),
8159                           vmcs_readl(EXIT_QUALIFICATION));
8160         return 1;
8161 }
8162
8163 /*
8164  * The exit handlers return 1 if the exit was handled fully and guest execution
8165  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
8166  * to be done to userspace and return 0.
8167  */
8168 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
8169         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
8170         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
8171         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
8172         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
8173         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
8174         [EXIT_REASON_CR_ACCESS]               = handle_cr,
8175         [EXIT_REASON_DR_ACCESS]               = handle_dr,
8176         [EXIT_REASON_CPUID]                   = handle_cpuid,
8177         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
8178         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
8179         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
8180         [EXIT_REASON_HLT]                     = handle_halt,
8181         [EXIT_REASON_INVD]                    = handle_invd,
8182         [EXIT_REASON_INVLPG]                  = handle_invlpg,
8183         [EXIT_REASON_RDPMC]                   = handle_rdpmc,
8184         [EXIT_REASON_VMCALL]                  = handle_vmcall,
8185         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
8186         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
8187         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
8188         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
8189         [EXIT_REASON_VMREAD]                  = handle_vmread,
8190         [EXIT_REASON_VMRESUME]                = handle_vmresume,
8191         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
8192         [EXIT_REASON_VMOFF]                   = handle_vmoff,
8193         [EXIT_REASON_VMON]                    = handle_vmon,
8194         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
8195         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
8196         [EXIT_REASON_APIC_WRITE]              = handle_apic_write,
8197         [EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
8198         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
8199         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
8200         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
8201         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
8202         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
8203         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
8204         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
8205         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_mwait,
8206         [EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
8207         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
8208         [EXIT_REASON_INVEPT]                  = handle_invept,
8209         [EXIT_REASON_INVVPID]                 = handle_invvpid,
8210         [EXIT_REASON_RDRAND]                  = handle_invalid_op,
8211         [EXIT_REASON_RDSEED]                  = handle_invalid_op,
8212         [EXIT_REASON_XSAVES]                  = handle_xsaves,
8213         [EXIT_REASON_XRSTORS]                 = handle_xrstors,
8214         [EXIT_REASON_PML_FULL]                = handle_pml_full,
8215         [EXIT_REASON_VMFUNC]                  = handle_vmfunc,
8216         [EXIT_REASON_PREEMPTION_TIMER]        = handle_preemption_timer,
8217 };
8218
8219 static const int kvm_vmx_max_exit_handlers =
8220         ARRAY_SIZE(kvm_vmx_exit_handlers);
8221
8222 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
8223                                        struct vmcs12 *vmcs12)
8224 {
8225         unsigned long exit_qualification;
8226         gpa_t bitmap, last_bitmap;
8227         unsigned int port;
8228         int size;
8229         u8 b;
8230
8231         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
8232                 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
8233
8234         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8235
8236         port = exit_qualification >> 16;
8237         size = (exit_qualification & 7) + 1;
8238
8239         last_bitmap = (gpa_t)-1;
8240         b = -1;
8241
8242         while (size > 0) {
8243                 if (port < 0x8000)
8244                         bitmap = vmcs12->io_bitmap_a;
8245                 else if (port < 0x10000)
8246                         bitmap = vmcs12->io_bitmap_b;
8247                 else
8248                         return true;
8249                 bitmap += (port & 0x7fff) / 8;
8250
8251                 if (last_bitmap != bitmap)
8252                         if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
8253                                 return true;
8254                 if (b & (1 << (port & 7)))
8255                         return true;
8256
8257                 port++;
8258                 size--;
8259                 last_bitmap = bitmap;
8260         }
8261
8262         return false;
8263 }
8264
8265 /*
8266  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
8267  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
8268  * disinterest in the current event (read or write a specific MSR) by using an
8269  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
8270  */
8271 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
8272         struct vmcs12 *vmcs12, u32 exit_reason)
8273 {
8274         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
8275         gpa_t bitmap;
8276
8277         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
8278                 return true;
8279
8280         /*
8281          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
8282          * for the four combinations of read/write and low/high MSR numbers.
8283          * First we need to figure out which of the four to use:
8284          */
8285         bitmap = vmcs12->msr_bitmap;
8286         if (exit_reason == EXIT_REASON_MSR_WRITE)
8287                 bitmap += 2048;
8288         if (msr_index >= 0xc0000000) {
8289                 msr_index -= 0xc0000000;
8290                 bitmap += 1024;
8291         }
8292
8293         /* Then read the msr_index'th bit from this bitmap: */
8294         if (msr_index < 1024*8) {
8295                 unsigned char b;
8296                 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
8297                         return true;
8298                 return 1 & (b >> (msr_index & 7));
8299         } else
8300                 return true; /* let L1 handle the wrong parameter */
8301 }
8302
8303 /*
8304  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
8305  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
8306  * intercept (via guest_host_mask etc.) the current event.
8307  */
8308 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
8309         struct vmcs12 *vmcs12)
8310 {
8311         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8312         int cr = exit_qualification & 15;
8313         int reg;
8314         unsigned long val;
8315
8316         switch ((exit_qualification >> 4) & 3) {
8317         case 0: /* mov to cr */
8318                 reg = (exit_qualification >> 8) & 15;
8319                 val = kvm_register_readl(vcpu, reg);
8320                 switch (cr) {
8321                 case 0:
8322                         if (vmcs12->cr0_guest_host_mask &
8323                             (val ^ vmcs12->cr0_read_shadow))
8324                                 return true;
8325                         break;
8326                 case 3:
8327                         if ((vmcs12->cr3_target_count >= 1 &&
8328                                         vmcs12->cr3_target_value0 == val) ||
8329                                 (vmcs12->cr3_target_count >= 2 &&
8330                                         vmcs12->cr3_target_value1 == val) ||
8331                                 (vmcs12->cr3_target_count >= 3 &&
8332                                         vmcs12->cr3_target_value2 == val) ||
8333                                 (vmcs12->cr3_target_count >= 4 &&
8334                                         vmcs12->cr3_target_value3 == val))
8335                                 return false;
8336                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8337                                 return true;
8338                         break;
8339                 case 4:
8340                         if (vmcs12->cr4_guest_host_mask &
8341                             (vmcs12->cr4_read_shadow ^ val))
8342                                 return true;
8343                         break;
8344                 case 8:
8345                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8346                                 return true;
8347                         break;
8348                 }
8349                 break;
8350         case 2: /* clts */
8351                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8352                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
8353                         return true;
8354                 break;
8355         case 1: /* mov from cr */
8356                 switch (cr) {
8357                 case 3:
8358                         if (vmcs12->cpu_based_vm_exec_control &
8359                             CPU_BASED_CR3_STORE_EXITING)
8360                                 return true;
8361                         break;
8362                 case 8:
8363                         if (vmcs12->cpu_based_vm_exec_control &
8364                             CPU_BASED_CR8_STORE_EXITING)
8365                                 return true;
8366                         break;
8367                 }
8368                 break;
8369         case 3: /* lmsw */
8370                 /*
8371                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8372                  * cr0. Other attempted changes are ignored, with no exit.
8373                  */
8374                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
8375                 if (vmcs12->cr0_guest_host_mask & 0xe &
8376                     (val ^ vmcs12->cr0_read_shadow))
8377                         return true;
8378                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8379                     !(vmcs12->cr0_read_shadow & 0x1) &&
8380                     (val & 0x1))
8381                         return true;
8382                 break;
8383         }
8384         return false;
8385 }
8386
8387 /*
8388  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8389  * should handle it ourselves in L0 (and then continue L2). Only call this
8390  * when in is_guest_mode (L2).
8391  */
8392 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
8393 {
8394         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8395         struct vcpu_vmx *vmx = to_vmx(vcpu);
8396         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8397
8398         if (vmx->nested.nested_run_pending)
8399                 return false;
8400
8401         if (unlikely(vmx->fail)) {
8402                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8403                                     vmcs_read32(VM_INSTRUCTION_ERROR));
8404                 return true;
8405         }
8406
8407         /*
8408          * The host physical addresses of some pages of guest memory
8409          * are loaded into VMCS02 (e.g. L1's Virtual APIC Page). The CPU
8410          * may write to these pages via their host physical address while
8411          * L2 is running, bypassing any address-translation-based dirty
8412          * tracking (e.g. EPT write protection).
8413          *
8414          * Mark them dirty on every exit from L2 to prevent them from
8415          * getting out of sync with dirty tracking.
8416          */
8417         nested_mark_vmcs12_pages_dirty(vcpu);
8418
8419         trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8420                                 vmcs_readl(EXIT_QUALIFICATION),
8421                                 vmx->idt_vectoring_info,
8422                                 intr_info,
8423                                 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8424                                 KVM_ISA_VMX);
8425
8426         switch (exit_reason) {
8427         case EXIT_REASON_EXCEPTION_NMI:
8428                 if (is_nmi(intr_info))
8429                         return false;
8430                 else if (is_page_fault(intr_info))
8431                         return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
8432                 else if (is_no_device(intr_info) &&
8433                          !(vmcs12->guest_cr0 & X86_CR0_TS))
8434                         return false;
8435                 else if (is_debug(intr_info) &&
8436                          vcpu->guest_debug &
8437                          (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8438                         return false;
8439                 else if (is_breakpoint(intr_info) &&
8440                          vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8441                         return false;
8442                 return vmcs12->exception_bitmap &
8443                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8444         case EXIT_REASON_EXTERNAL_INTERRUPT:
8445                 return false;
8446         case EXIT_REASON_TRIPLE_FAULT:
8447                 return true;
8448         case EXIT_REASON_PENDING_INTERRUPT:
8449                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8450         case EXIT_REASON_NMI_WINDOW:
8451                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8452         case EXIT_REASON_TASK_SWITCH:
8453                 return true;
8454         case EXIT_REASON_CPUID:
8455                 return true;
8456         case EXIT_REASON_HLT:
8457                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8458         case EXIT_REASON_INVD:
8459                 return true;
8460         case EXIT_REASON_INVLPG:
8461                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8462         case EXIT_REASON_RDPMC:
8463                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8464         case EXIT_REASON_RDRAND:
8465                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
8466         case EXIT_REASON_RDSEED:
8467                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
8468         case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8469                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8470         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8471         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8472         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8473         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8474         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8475         case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8476                 /*
8477                  * VMX instructions trap unconditionally. This allows L1 to
8478                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
8479                  */
8480                 return true;
8481         case EXIT_REASON_CR_ACCESS:
8482                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8483         case EXIT_REASON_DR_ACCESS:
8484                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8485         case EXIT_REASON_IO_INSTRUCTION:
8486                 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8487         case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8488                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8489         case EXIT_REASON_MSR_READ:
8490         case EXIT_REASON_MSR_WRITE:
8491                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8492         case EXIT_REASON_INVALID_STATE:
8493                 return true;
8494         case EXIT_REASON_MWAIT_INSTRUCTION:
8495                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8496         case EXIT_REASON_MONITOR_TRAP_FLAG:
8497                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8498         case EXIT_REASON_MONITOR_INSTRUCTION:
8499                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8500         case EXIT_REASON_PAUSE_INSTRUCTION:
8501                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8502                         nested_cpu_has2(vmcs12,
8503                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8504         case EXIT_REASON_MCE_DURING_VMENTRY:
8505                 return false;
8506         case EXIT_REASON_TPR_BELOW_THRESHOLD:
8507                 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8508         case EXIT_REASON_APIC_ACCESS:
8509                 return nested_cpu_has2(vmcs12,
8510                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8511         case EXIT_REASON_APIC_WRITE:
8512         case EXIT_REASON_EOI_INDUCED:
8513                 /* apic_write and eoi_induced should exit unconditionally. */
8514                 return true;
8515         case EXIT_REASON_EPT_VIOLATION:
8516                 /*
8517                  * L0 always deals with the EPT violation. If nested EPT is
8518                  * used, and the nested mmu code discovers that the address is
8519                  * missing in the guest EPT table (EPT12), the EPT violation
8520                  * will be injected with nested_ept_inject_page_fault()
8521                  */
8522                 return false;
8523         case EXIT_REASON_EPT_MISCONFIG:
8524                 /*
8525                  * L2 never uses directly L1's EPT, but rather L0's own EPT
8526                  * table (shadow on EPT) or a merged EPT table that L0 built
8527                  * (EPT on EPT). So any problems with the structure of the
8528                  * table is L0's fault.
8529                  */
8530                 return false;
8531         case EXIT_REASON_INVPCID:
8532                 return
8533                         nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
8534                         nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8535         case EXIT_REASON_WBINVD:
8536                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8537         case EXIT_REASON_XSETBV:
8538                 return true;
8539         case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8540                 /*
8541                  * This should never happen, since it is not possible to
8542                  * set XSS to a non-zero value---neither in L1 nor in L2.
8543                  * If if it were, XSS would have to be checked against
8544                  * the XSS exit bitmap in vmcs12.
8545                  */
8546                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8547         case EXIT_REASON_PREEMPTION_TIMER:
8548                 return false;
8549         case EXIT_REASON_PML_FULL:
8550                 /* We emulate PML support to L1. */
8551                 return false;
8552         case EXIT_REASON_VMFUNC:
8553                 /* VM functions are emulated through L2->L0 vmexits. */
8554                 return false;
8555         default:
8556                 return true;
8557         }
8558 }
8559
8560 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
8561 {
8562         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8563
8564         /*
8565          * At this point, the exit interruption info in exit_intr_info
8566          * is only valid for EXCEPTION_NMI exits.  For EXTERNAL_INTERRUPT
8567          * we need to query the in-kernel LAPIC.
8568          */
8569         WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
8570         if ((exit_intr_info &
8571              (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8572             (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
8573                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8574                 vmcs12->vm_exit_intr_error_code =
8575                         vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8576         }
8577
8578         nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
8579                           vmcs_readl(EXIT_QUALIFICATION));
8580         return 1;
8581 }
8582
8583 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8584 {
8585         *info1 = vmcs_readl(EXIT_QUALIFICATION);
8586         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8587 }
8588
8589 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8590 {
8591         if (vmx->pml_pg) {
8592                 __free_page(vmx->pml_pg);
8593                 vmx->pml_pg = NULL;
8594         }
8595 }
8596
8597 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8598 {
8599         struct vcpu_vmx *vmx = to_vmx(vcpu);
8600         u64 *pml_buf;
8601         u16 pml_idx;
8602
8603         pml_idx = vmcs_read16(GUEST_PML_INDEX);
8604
8605         /* Do nothing if PML buffer is empty */
8606         if (pml_idx == (PML_ENTITY_NUM - 1))
8607                 return;
8608
8609         /* PML index always points to next available PML buffer entity */
8610         if (pml_idx >= PML_ENTITY_NUM)
8611                 pml_idx = 0;
8612         else
8613                 pml_idx++;
8614
8615         pml_buf = page_address(vmx->pml_pg);
8616         for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8617                 u64 gpa;
8618
8619                 gpa = pml_buf[pml_idx];
8620                 WARN_ON(gpa & (PAGE_SIZE - 1));
8621                 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8622         }
8623
8624         /* reset PML index */
8625         vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8626 }
8627
8628 /*
8629  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8630  * Called before reporting dirty_bitmap to userspace.
8631  */
8632 static void kvm_flush_pml_buffers(struct kvm *kvm)
8633 {
8634         int i;
8635         struct kvm_vcpu *vcpu;
8636         /*
8637          * We only need to kick vcpu out of guest mode here, as PML buffer
8638          * is flushed at beginning of all VMEXITs, and it's obvious that only
8639          * vcpus running in guest are possible to have unflushed GPAs in PML
8640          * buffer.
8641          */
8642         kvm_for_each_vcpu(i, vcpu, kvm)
8643                 kvm_vcpu_kick(vcpu);
8644 }
8645
8646 static void vmx_dump_sel(char *name, uint32_t sel)
8647 {
8648         pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8649                name, vmcs_read16(sel),
8650                vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8651                vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8652                vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8653 }
8654
8655 static void vmx_dump_dtsel(char *name, uint32_t limit)
8656 {
8657         pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
8658                name, vmcs_read32(limit),
8659                vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8660 }
8661
8662 static void dump_vmcs(void)
8663 {
8664         u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8665         u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8666         u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8667         u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8668         u32 secondary_exec_control = 0;
8669         unsigned long cr4 = vmcs_readl(GUEST_CR4);
8670         u64 efer = vmcs_read64(GUEST_IA32_EFER);
8671         int i, n;
8672
8673         if (cpu_has_secondary_exec_ctrls())
8674                 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8675
8676         pr_err("*** Guest State ***\n");
8677         pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8678                vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8679                vmcs_readl(CR0_GUEST_HOST_MASK));
8680         pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8681                cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8682         pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8683         if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8684             (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8685         {
8686                 pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
8687                        vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8688                 pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
8689                        vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8690         }
8691         pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
8692                vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8693         pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
8694                vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8695         pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8696                vmcs_readl(GUEST_SYSENTER_ESP),
8697                vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8698         vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
8699         vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
8700         vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
8701         vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
8702         vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
8703         vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
8704         vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8705         vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8706         vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8707         vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
8708         if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8709             (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8710                 pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
8711                        efer, vmcs_read64(GUEST_IA32_PAT));
8712         pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
8713                vmcs_read64(GUEST_IA32_DEBUGCTL),
8714                vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8715         if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8716                 pr_err("PerfGlobCtl = 0x%016llx\n",
8717                        vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8718         if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8719                 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8720         pr_err("Interruptibility = %08x  ActivityState = %08x\n",
8721                vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8722                vmcs_read32(GUEST_ACTIVITY_STATE));
8723         if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8724                 pr_err("InterruptStatus = %04x\n",
8725                        vmcs_read16(GUEST_INTR_STATUS));
8726
8727         pr_err("*** Host State ***\n");
8728         pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
8729                vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8730         pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8731                vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8732                vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8733                vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8734                vmcs_read16(HOST_TR_SELECTOR));
8735         pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8736                vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8737                vmcs_readl(HOST_TR_BASE));
8738         pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8739                vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8740         pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8741                vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8742                vmcs_readl(HOST_CR4));
8743         pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8744                vmcs_readl(HOST_IA32_SYSENTER_ESP),
8745                vmcs_read32(HOST_IA32_SYSENTER_CS),
8746                vmcs_readl(HOST_IA32_SYSENTER_EIP));
8747         if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8748                 pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
8749                        vmcs_read64(HOST_IA32_EFER),
8750                        vmcs_read64(HOST_IA32_PAT));
8751         if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8752                 pr_err("PerfGlobCtl = 0x%016llx\n",
8753                        vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8754
8755         pr_err("*** Control State ***\n");
8756         pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8757                pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8758         pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8759         pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8760                vmcs_read32(EXCEPTION_BITMAP),
8761                vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8762                vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8763         pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8764                vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8765                vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8766                vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8767         pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8768                vmcs_read32(VM_EXIT_INTR_INFO),
8769                vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8770                vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8771         pr_err("        reason=%08x qualification=%016lx\n",
8772                vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8773         pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8774                vmcs_read32(IDT_VECTORING_INFO_FIELD),
8775                vmcs_read32(IDT_VECTORING_ERROR_CODE));
8776         pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8777         if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8778                 pr_err("TSC Multiplier = 0x%016llx\n",
8779                        vmcs_read64(TSC_MULTIPLIER));
8780         if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8781                 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8782         if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8783                 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8784         if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8785                 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8786         n = vmcs_read32(CR3_TARGET_COUNT);
8787         for (i = 0; i + 1 < n; i += 4)
8788                 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8789                        i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8790                        i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8791         if (i < n)
8792                 pr_err("CR3 target%u=%016lx\n",
8793                        i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8794         if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8795                 pr_err("PLE Gap=%08x Window=%08x\n",
8796                        vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8797         if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8798                 pr_err("Virtual processor ID = 0x%04x\n",
8799                        vmcs_read16(VIRTUAL_PROCESSOR_ID));
8800 }
8801
8802 /*
8803  * The guest has exited.  See if we can fix it or if we need userspace
8804  * assistance.
8805  */
8806 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8807 {
8808         struct vcpu_vmx *vmx = to_vmx(vcpu);
8809         u32 exit_reason = vmx->exit_reason;
8810         u32 vectoring_info = vmx->idt_vectoring_info;
8811
8812         trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8813
8814         /*
8815          * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8816          * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8817          * querying dirty_bitmap, we only need to kick all vcpus out of guest
8818          * mode as if vcpus is in root mode, the PML buffer must has been
8819          * flushed already.
8820          */
8821         if (enable_pml)
8822                 vmx_flush_pml_buffer(vcpu);
8823
8824         /* If guest state is invalid, start emulating */
8825         if (vmx->emulation_required)
8826                 return handle_invalid_guest_state(vcpu);
8827
8828         if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
8829                 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
8830
8831         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8832                 dump_vmcs();
8833                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8834                 vcpu->run->fail_entry.hardware_entry_failure_reason
8835                         = exit_reason;
8836                 return 0;
8837         }
8838
8839         if (unlikely(vmx->fail)) {
8840                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8841                 vcpu->run->fail_entry.hardware_entry_failure_reason
8842                         = vmcs_read32(VM_INSTRUCTION_ERROR);
8843                 return 0;
8844         }
8845
8846         /*
8847          * Note:
8848          * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8849          * delivery event since it indicates guest is accessing MMIO.
8850          * The vm-exit can be triggered again after return to guest that
8851          * will cause infinite loop.
8852          */
8853         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8854                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8855                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
8856                         exit_reason != EXIT_REASON_PML_FULL &&
8857                         exit_reason != EXIT_REASON_TASK_SWITCH)) {
8858                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8859                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8860                 vcpu->run->internal.ndata = 3;
8861                 vcpu->run->internal.data[0] = vectoring_info;
8862                 vcpu->run->internal.data[1] = exit_reason;
8863                 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
8864                 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
8865                         vcpu->run->internal.ndata++;
8866                         vcpu->run->internal.data[3] =
8867                                 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
8868                 }
8869                 return 0;
8870         }
8871
8872         if (unlikely(!enable_vnmi &&
8873                      vmx->loaded_vmcs->soft_vnmi_blocked)) {
8874                 if (vmx_interrupt_allowed(vcpu)) {
8875                         vmx->loaded_vmcs->soft_vnmi_blocked = 0;
8876                 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
8877                            vcpu->arch.nmi_pending) {
8878                         /*
8879                          * This CPU don't support us in finding the end of an
8880                          * NMI-blocked window if the guest runs with IRQs
8881                          * disabled. So we pull the trigger after 1 s of
8882                          * futile waiting, but inform the user about this.
8883                          */
8884                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8885                                "state on VCPU %d after 1 s timeout\n",
8886                                __func__, vcpu->vcpu_id);
8887                         vmx->loaded_vmcs->soft_vnmi_blocked = 0;
8888                 }
8889         }
8890
8891         if (exit_reason < kvm_vmx_max_exit_handlers
8892             && kvm_vmx_exit_handlers[exit_reason])
8893                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8894         else {
8895                 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8896                                 exit_reason);
8897                 kvm_queue_exception(vcpu, UD_VECTOR);
8898                 return 1;
8899         }
8900 }
8901
8902 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8903 {
8904         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8905
8906         if (is_guest_mode(vcpu) &&
8907                 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8908                 return;
8909
8910         if (irr == -1 || tpr < irr) {
8911                 vmcs_write32(TPR_THRESHOLD, 0);
8912                 return;
8913         }
8914
8915         vmcs_write32(TPR_THRESHOLD, irr);
8916 }
8917
8918 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8919 {
8920         u32 sec_exec_control;
8921
8922         /* Postpone execution until vmcs01 is the current VMCS. */
8923         if (is_guest_mode(vcpu)) {
8924                 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8925                 return;
8926         }
8927
8928         if (!cpu_has_vmx_virtualize_x2apic_mode())
8929                 return;
8930
8931         if (!cpu_need_tpr_shadow(vcpu))
8932                 return;
8933
8934         sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8935
8936         if (set) {
8937                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8938                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8939         } else {
8940                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8941                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8942                 vmx_flush_tlb_ept_only(vcpu);
8943         }
8944         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8945
8946         vmx_set_msr_bitmap(vcpu);
8947 }
8948
8949 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8950 {
8951         struct vcpu_vmx *vmx = to_vmx(vcpu);
8952
8953         /*
8954          * Currently we do not handle the nested case where L2 has an
8955          * APIC access page of its own; that page is still pinned.
8956          * Hence, we skip the case where the VCPU is in guest mode _and_
8957          * L1 prepared an APIC access page for L2.
8958          *
8959          * For the case where L1 and L2 share the same APIC access page
8960          * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8961          * in the vmcs12), this function will only update either the vmcs01
8962          * or the vmcs02.  If the former, the vmcs02 will be updated by
8963          * prepare_vmcs02.  If the latter, the vmcs01 will be updated in
8964          * the next L2->L1 exit.
8965          */
8966         if (!is_guest_mode(vcpu) ||
8967             !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8968                              SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8969                 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8970                 vmx_flush_tlb_ept_only(vcpu);
8971         }
8972 }
8973
8974 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8975 {
8976         u16 status;
8977         u8 old;
8978
8979         if (max_isr == -1)
8980                 max_isr = 0;
8981
8982         status = vmcs_read16(GUEST_INTR_STATUS);
8983         old = status >> 8;
8984         if (max_isr != old) {
8985                 status &= 0xff;
8986                 status |= max_isr << 8;
8987                 vmcs_write16(GUEST_INTR_STATUS, status);
8988         }
8989 }
8990
8991 static void vmx_set_rvi(int vector)
8992 {
8993         u16 status;
8994         u8 old;
8995
8996         if (vector == -1)
8997                 vector = 0;
8998
8999         status = vmcs_read16(GUEST_INTR_STATUS);
9000         old = (u8)status & 0xff;
9001         if ((u8)vector != old) {
9002                 status &= ~0xff;
9003                 status |= (u8)vector;
9004                 vmcs_write16(GUEST_INTR_STATUS, status);
9005         }
9006 }
9007
9008 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
9009 {
9010         if (!is_guest_mode(vcpu)) {
9011                 vmx_set_rvi(max_irr);
9012                 return;
9013         }
9014
9015         if (max_irr == -1)
9016                 return;
9017
9018         /*
9019          * In guest mode.  If a vmexit is needed, vmx_check_nested_events
9020          * handles it.
9021          */
9022         if (nested_exit_on_intr(vcpu))
9023                 return;
9024
9025         /*
9026          * Else, fall back to pre-APICv interrupt injection since L2
9027          * is run without virtual interrupt delivery.
9028          */
9029         if (!kvm_event_needs_reinjection(vcpu) &&
9030             vmx_interrupt_allowed(vcpu)) {
9031                 kvm_queue_interrupt(vcpu, max_irr, false);
9032                 vmx_inject_irq(vcpu);
9033         }
9034 }
9035
9036 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
9037 {
9038         struct vcpu_vmx *vmx = to_vmx(vcpu);
9039         int max_irr;
9040
9041         WARN_ON(!vcpu->arch.apicv_active);
9042         if (pi_test_on(&vmx->pi_desc)) {
9043                 pi_clear_on(&vmx->pi_desc);
9044                 /*
9045                  * IOMMU can write to PIR.ON, so the barrier matters even on UP.
9046                  * But on x86 this is just a compiler barrier anyway.
9047                  */
9048                 smp_mb__after_atomic();
9049                 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
9050         } else {
9051                 max_irr = kvm_lapic_find_highest_irr(vcpu);
9052         }
9053         vmx_hwapic_irr_update(vcpu, max_irr);
9054         return max_irr;
9055 }
9056
9057 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
9058 {
9059         if (!kvm_vcpu_apicv_active(vcpu))
9060                 return;
9061
9062         vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
9063         vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
9064         vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
9065         vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
9066 }
9067
9068 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
9069 {
9070         struct vcpu_vmx *vmx = to_vmx(vcpu);
9071
9072         pi_clear_on(&vmx->pi_desc);
9073         memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
9074 }
9075
9076 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
9077 {
9078         u32 exit_intr_info = 0;
9079         u16 basic_exit_reason = (u16)vmx->exit_reason;
9080
9081         if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
9082               || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
9083                 return;
9084
9085         if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9086                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9087         vmx->exit_intr_info = exit_intr_info;
9088
9089         /* if exit due to PF check for async PF */
9090         if (is_page_fault(exit_intr_info))
9091                 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
9092
9093         /* Handle machine checks before interrupts are enabled */
9094         if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
9095             is_machine_check(exit_intr_info))
9096                 kvm_machine_check();
9097
9098         /* We need to handle NMIs before interrupts are enabled */
9099         if (is_nmi(exit_intr_info)) {
9100                 kvm_before_handle_nmi(&vmx->vcpu);
9101                 asm("int $2");
9102                 kvm_after_handle_nmi(&vmx->vcpu);
9103         }
9104 }
9105
9106 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
9107 {
9108         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9109
9110         if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
9111                         == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
9112                 unsigned int vector;
9113                 unsigned long entry;
9114                 gate_desc *desc;
9115                 struct vcpu_vmx *vmx = to_vmx(vcpu);
9116 #ifdef CONFIG_X86_64
9117                 unsigned long tmp;
9118 #endif
9119
9120                 vector =  exit_intr_info & INTR_INFO_VECTOR_MASK;
9121                 desc = (gate_desc *)vmx->host_idt_base + vector;
9122                 entry = gate_offset(desc);
9123                 asm volatile(
9124 #ifdef CONFIG_X86_64
9125                         "mov %%" _ASM_SP ", %[sp]\n\t"
9126                         "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
9127                         "push $%c[ss]\n\t"
9128                         "push %[sp]\n\t"
9129 #endif
9130                         "pushf\n\t"
9131                         __ASM_SIZE(push) " $%c[cs]\n\t"
9132                         "call *%[entry]\n\t"
9133                         :
9134 #ifdef CONFIG_X86_64
9135                         [sp]"=&r"(tmp),
9136 #endif
9137                         ASM_CALL_CONSTRAINT
9138                         :
9139                         [entry]"r"(entry),
9140                         [ss]"i"(__KERNEL_DS),
9141                         [cs]"i"(__KERNEL_CS)
9142                         );
9143         }
9144 }
9145 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
9146
9147 static bool vmx_has_high_real_mode_segbase(void)
9148 {
9149         return enable_unrestricted_guest || emulate_invalid_guest_state;
9150 }
9151
9152 static bool vmx_mpx_supported(void)
9153 {
9154         return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
9155                 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
9156 }
9157
9158 static bool vmx_xsaves_supported(void)
9159 {
9160         return vmcs_config.cpu_based_2nd_exec_ctrl &
9161                 SECONDARY_EXEC_XSAVES;
9162 }
9163
9164 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
9165 {
9166         u32 exit_intr_info;
9167         bool unblock_nmi;
9168         u8 vector;
9169         bool idtv_info_valid;
9170
9171         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9172
9173         if (enable_vnmi) {
9174                 if (vmx->loaded_vmcs->nmi_known_unmasked)
9175                         return;
9176                 /*
9177                  * Can't use vmx->exit_intr_info since we're not sure what
9178                  * the exit reason is.
9179                  */
9180                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9181                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
9182                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9183                 /*
9184                  * SDM 3: 27.7.1.2 (September 2008)
9185                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
9186                  * a guest IRET fault.
9187                  * SDM 3: 23.2.2 (September 2008)
9188                  * Bit 12 is undefined in any of the following cases:
9189                  *  If the VM exit sets the valid bit in the IDT-vectoring
9190                  *   information field.
9191                  *  If the VM exit is due to a double fault.
9192                  */
9193                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
9194                     vector != DF_VECTOR && !idtv_info_valid)
9195                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
9196                                       GUEST_INTR_STATE_NMI);
9197                 else
9198                         vmx->loaded_vmcs->nmi_known_unmasked =
9199                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
9200                                   & GUEST_INTR_STATE_NMI);
9201         } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
9202                 vmx->loaded_vmcs->vnmi_blocked_time +=
9203                         ktime_to_ns(ktime_sub(ktime_get(),
9204                                               vmx->loaded_vmcs->entry_time));
9205 }
9206
9207 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
9208                                       u32 idt_vectoring_info,
9209                                       int instr_len_field,
9210                                       int error_code_field)
9211 {
9212         u8 vector;
9213         int type;
9214         bool idtv_info_valid;
9215
9216         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9217
9218         vcpu->arch.nmi_injected = false;
9219         kvm_clear_exception_queue(vcpu);
9220         kvm_clear_interrupt_queue(vcpu);
9221
9222         if (!idtv_info_valid)
9223                 return;
9224
9225         kvm_make_request(KVM_REQ_EVENT, vcpu);
9226
9227         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
9228         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
9229
9230         switch (type) {
9231         case INTR_TYPE_NMI_INTR:
9232                 vcpu->arch.nmi_injected = true;
9233                 /*
9234                  * SDM 3: 27.7.1.2 (September 2008)
9235                  * Clear bit "block by NMI" before VM entry if a NMI
9236                  * delivery faulted.
9237                  */
9238                 vmx_set_nmi_mask(vcpu, false);
9239                 break;
9240         case INTR_TYPE_SOFT_EXCEPTION:
9241                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9242                 /* fall through */
9243         case INTR_TYPE_HARD_EXCEPTION:
9244                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
9245                         u32 err = vmcs_read32(error_code_field);
9246                         kvm_requeue_exception_e(vcpu, vector, err);
9247                 } else
9248                         kvm_requeue_exception(vcpu, vector);
9249                 break;
9250         case INTR_TYPE_SOFT_INTR:
9251                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9252                 /* fall through */
9253         case INTR_TYPE_EXT_INTR:
9254                 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
9255                 break;
9256         default:
9257                 break;
9258         }
9259 }
9260
9261 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
9262 {
9263         __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
9264                                   VM_EXIT_INSTRUCTION_LEN,
9265                                   IDT_VECTORING_ERROR_CODE);
9266 }
9267
9268 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
9269 {
9270         __vmx_complete_interrupts(vcpu,
9271                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
9272                                   VM_ENTRY_INSTRUCTION_LEN,
9273                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
9274
9275         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9276 }
9277
9278 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
9279 {
9280         int i, nr_msrs;
9281         struct perf_guest_switch_msr *msrs;
9282
9283         msrs = perf_guest_get_msrs(&nr_msrs);
9284
9285         if (!msrs)
9286                 return;
9287
9288         for (i = 0; i < nr_msrs; i++)
9289                 if (msrs[i].host == msrs[i].guest)
9290                         clear_atomic_switch_msr(vmx, msrs[i].msr);
9291                 else
9292                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
9293                                         msrs[i].host);
9294 }
9295
9296 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
9297 {
9298         struct vcpu_vmx *vmx = to_vmx(vcpu);
9299         u64 tscl;
9300         u32 delta_tsc;
9301
9302         if (vmx->hv_deadline_tsc == -1)
9303                 return;
9304
9305         tscl = rdtsc();
9306         if (vmx->hv_deadline_tsc > tscl)
9307                 /* sure to be 32 bit only because checked on set_hv_timer */
9308                 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
9309                         cpu_preemption_timer_multi);
9310         else
9311                 delta_tsc = 0;
9312
9313         vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
9314 }
9315
9316 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
9317 {
9318         struct vcpu_vmx *vmx = to_vmx(vcpu);
9319         unsigned long debugctlmsr, cr3, cr4;
9320
9321         /* Record the guest's net vcpu time for enforced NMI injections. */
9322         if (unlikely(!enable_vnmi &&
9323                      vmx->loaded_vmcs->soft_vnmi_blocked))
9324                 vmx->loaded_vmcs->entry_time = ktime_get();
9325
9326         /* Don't enter VMX if guest state is invalid, let the exit handler
9327            start emulation until we arrive back to a valid state */
9328         if (vmx->emulation_required)
9329                 return;
9330
9331         if (vmx->ple_window_dirty) {
9332                 vmx->ple_window_dirty = false;
9333                 vmcs_write32(PLE_WINDOW, vmx->ple_window);
9334         }
9335
9336         if (vmx->nested.sync_shadow_vmcs) {
9337                 copy_vmcs12_to_shadow(vmx);
9338                 vmx->nested.sync_shadow_vmcs = false;
9339         }
9340
9341         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
9342                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
9343         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
9344                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
9345
9346         cr3 = __get_current_cr3_fast();
9347         if (unlikely(cr3 != vmx->loaded_vmcs->vmcs_host_cr3)) {
9348                 vmcs_writel(HOST_CR3, cr3);
9349                 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
9350         }
9351
9352         cr4 = cr4_read_shadow();
9353         if (unlikely(cr4 != vmx->loaded_vmcs->vmcs_host_cr4)) {
9354                 vmcs_writel(HOST_CR4, cr4);
9355                 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
9356         }
9357
9358         /* When single-stepping over STI and MOV SS, we must clear the
9359          * corresponding interruptibility bits in the guest state. Otherwise
9360          * vmentry fails as it then expects bit 14 (BS) in pending debug
9361          * exceptions being set, but that's not correct for the guest debugging
9362          * case. */
9363         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9364                 vmx_set_interrupt_shadow(vcpu, 0);
9365
9366         if (static_cpu_has(X86_FEATURE_PKU) &&
9367             kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
9368             vcpu->arch.pkru != vmx->host_pkru)
9369                 __write_pkru(vcpu->arch.pkru);
9370
9371         atomic_switch_perf_msrs(vmx);
9372         debugctlmsr = get_debugctlmsr();
9373
9374         vmx_arm_hv_timer(vcpu);
9375
9376         vmx->__launched = vmx->loaded_vmcs->launched;
9377         asm(
9378                 /* Store host registers */
9379                 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
9380                 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
9381                 "push %%" _ASM_CX " \n\t"
9382                 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9383                 "je 1f \n\t"
9384                 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9385                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
9386                 "1: \n\t"
9387                 /* Reload cr2 if changed */
9388                 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9389                 "mov %%cr2, %%" _ASM_DX " \n\t"
9390                 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9391                 "je 2f \n\t"
9392                 "mov %%" _ASM_AX", %%cr2 \n\t"
9393                 "2: \n\t"
9394                 /* Check if vmlaunch of vmresume is needed */
9395                 "cmpl $0, %c[launched](%0) \n\t"
9396                 /* Load guest registers.  Don't clobber flags. */
9397                 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9398                 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9399                 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9400                 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9401                 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9402                 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9403 #ifdef CONFIG_X86_64
9404                 "mov %c[r8](%0),  %%r8  \n\t"
9405                 "mov %c[r9](%0),  %%r9  \n\t"
9406                 "mov %c[r10](%0), %%r10 \n\t"
9407                 "mov %c[r11](%0), %%r11 \n\t"
9408                 "mov %c[r12](%0), %%r12 \n\t"
9409                 "mov %c[r13](%0), %%r13 \n\t"
9410                 "mov %c[r14](%0), %%r14 \n\t"
9411                 "mov %c[r15](%0), %%r15 \n\t"
9412 #endif
9413                 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9414
9415                 /* Enter guest mode */
9416                 "jne 1f \n\t"
9417                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9418                 "jmp 2f \n\t"
9419                 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9420                 "2: "
9421                 /* Save guest registers, load host registers, keep flags */
9422                 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9423                 "pop %0 \n\t"
9424                 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9425                 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9426                 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9427                 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9428                 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9429                 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9430                 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9431 #ifdef CONFIG_X86_64
9432                 "mov %%r8,  %c[r8](%0) \n\t"
9433                 "mov %%r9,  %c[r9](%0) \n\t"
9434                 "mov %%r10, %c[r10](%0) \n\t"
9435                 "mov %%r11, %c[r11](%0) \n\t"
9436                 "mov %%r12, %c[r12](%0) \n\t"
9437                 "mov %%r13, %c[r13](%0) \n\t"
9438                 "mov %%r14, %c[r14](%0) \n\t"
9439                 "mov %%r15, %c[r15](%0) \n\t"
9440 #endif
9441                 "mov %%cr2, %%" _ASM_AX "   \n\t"
9442                 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9443
9444                 "pop  %%" _ASM_BP "; pop  %%" _ASM_DX " \n\t"
9445                 "setbe %c[fail](%0) \n\t"
9446                 ".pushsection .rodata \n\t"
9447                 ".global vmx_return \n\t"
9448                 "vmx_return: " _ASM_PTR " 2b \n\t"
9449                 ".popsection"
9450               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9451                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9452                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9453                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9454                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9455                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9456                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9457                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9458                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9459                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9460                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9461 #ifdef CONFIG_X86_64
9462                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9463                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9464                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9465                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9466                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9467                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9468                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9469                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9470 #endif
9471                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9472                 [wordsize]"i"(sizeof(ulong))
9473               : "cc", "memory"
9474 #ifdef CONFIG_X86_64
9475                 , "rax", "rbx", "rdi", "rsi"
9476                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9477 #else
9478                 , "eax", "ebx", "edi", "esi"
9479 #endif
9480               );
9481
9482         /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9483         if (debugctlmsr)
9484                 update_debugctlmsr(debugctlmsr);
9485
9486 #ifndef CONFIG_X86_64
9487         /*
9488          * The sysexit path does not restore ds/es, so we must set them to
9489          * a reasonable value ourselves.
9490          *
9491          * We can't defer this to vmx_load_host_state() since that function
9492          * may be executed in interrupt context, which saves and restore segments
9493          * around it, nullifying its effect.
9494          */
9495         loadsegment(ds, __USER_DS);
9496         loadsegment(es, __USER_DS);
9497 #endif
9498
9499         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9500                                   | (1 << VCPU_EXREG_RFLAGS)
9501                                   | (1 << VCPU_EXREG_PDPTR)
9502                                   | (1 << VCPU_EXREG_SEGMENTS)
9503                                   | (1 << VCPU_EXREG_CR3));
9504         vcpu->arch.regs_dirty = 0;
9505
9506         /*
9507          * eager fpu is enabled if PKEY is supported and CR4 is switched
9508          * back on host, so it is safe to read guest PKRU from current
9509          * XSAVE.
9510          */
9511         if (static_cpu_has(X86_FEATURE_PKU) &&
9512             kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
9513                 vcpu->arch.pkru = __read_pkru();
9514                 if (vcpu->arch.pkru != vmx->host_pkru)
9515                         __write_pkru(vmx->host_pkru);
9516         }
9517
9518         /*
9519          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9520          * we did not inject a still-pending event to L1 now because of
9521          * nested_run_pending, we need to re-enable this bit.
9522          */
9523         if (vmx->nested.nested_run_pending)
9524                 kvm_make_request(KVM_REQ_EVENT, vcpu);
9525
9526         vmx->nested.nested_run_pending = 0;
9527         vmx->idt_vectoring_info = 0;
9528
9529         vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
9530         if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9531                 return;
9532
9533         vmx->loaded_vmcs->launched = 1;
9534         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9535
9536         vmx_complete_atomic_exit(vmx);
9537         vmx_recover_nmi_blocking(vmx);
9538         vmx_complete_interrupts(vmx);
9539 }
9540 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
9541
9542 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9543 {
9544         struct vcpu_vmx *vmx = to_vmx(vcpu);
9545         int cpu;
9546
9547         if (vmx->loaded_vmcs == vmcs)
9548                 return;
9549
9550         cpu = get_cpu();
9551         vmx->loaded_vmcs = vmcs;
9552         vmx_vcpu_put(vcpu);
9553         vmx_vcpu_load(vcpu, cpu);
9554         put_cpu();
9555 }
9556
9557 /*
9558  * Ensure that the current vmcs of the logical processor is the
9559  * vmcs01 of the vcpu before calling free_nested().
9560  */
9561 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9562 {
9563        struct vcpu_vmx *vmx = to_vmx(vcpu);
9564        int r;
9565
9566        r = vcpu_load(vcpu);
9567        BUG_ON(r);
9568        vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9569        free_nested(vmx);
9570        vcpu_put(vcpu);
9571 }
9572
9573 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9574 {
9575         struct vcpu_vmx *vmx = to_vmx(vcpu);
9576
9577         if (enable_pml)
9578                 vmx_destroy_pml_buffer(vmx);
9579         free_vpid(vmx->vpid);
9580         leave_guest_mode(vcpu);
9581         vmx_free_vcpu_nested(vcpu);
9582         free_loaded_vmcs(vmx->loaded_vmcs);
9583         kfree(vmx->guest_msrs);
9584         kvm_vcpu_uninit(vcpu);
9585         kmem_cache_free(kvm_vcpu_cache, vmx);
9586 }
9587
9588 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9589 {
9590         int err;
9591         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9592         int cpu;
9593
9594         if (!vmx)
9595                 return ERR_PTR(-ENOMEM);
9596
9597         vmx->vpid = allocate_vpid();
9598
9599         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9600         if (err)
9601                 goto free_vcpu;
9602
9603         err = -ENOMEM;
9604
9605         /*
9606          * If PML is turned on, failure on enabling PML just results in failure
9607          * of creating the vcpu, therefore we can simplify PML logic (by
9608          * avoiding dealing with cases, such as enabling PML partially on vcpus
9609          * for the guest, etc.
9610          */
9611         if (enable_pml) {
9612                 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9613                 if (!vmx->pml_pg)
9614                         goto uninit_vcpu;
9615         }
9616
9617         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9618         BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9619                      > PAGE_SIZE);
9620
9621         if (!vmx->guest_msrs)
9622                 goto free_pml;
9623
9624         vmx->loaded_vmcs = &vmx->vmcs01;
9625         vmx->loaded_vmcs->vmcs = alloc_vmcs();
9626         vmx->loaded_vmcs->shadow_vmcs = NULL;
9627         if (!vmx->loaded_vmcs->vmcs)
9628                 goto free_msrs;
9629         loaded_vmcs_init(vmx->loaded_vmcs);
9630
9631         cpu = get_cpu();
9632         vmx_vcpu_load(&vmx->vcpu, cpu);
9633         vmx->vcpu.cpu = cpu;
9634         vmx_vcpu_setup(vmx);
9635         vmx_vcpu_put(&vmx->vcpu);
9636         put_cpu();
9637         if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9638                 err = alloc_apic_access_page(kvm);
9639                 if (err)
9640                         goto free_vmcs;
9641         }
9642
9643         if (enable_ept) {
9644                 err = init_rmode_identity_map(kvm);
9645                 if (err)
9646                         goto free_vmcs;
9647         }
9648
9649         if (nested) {
9650                 nested_vmx_setup_ctls_msrs(vmx);
9651                 vmx->nested.vpid02 = allocate_vpid();
9652         }
9653
9654         vmx->nested.posted_intr_nv = -1;
9655         vmx->nested.current_vmptr = -1ull;
9656
9657         vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9658
9659         /*
9660          * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
9661          * or POSTED_INTR_WAKEUP_VECTOR.
9662          */
9663         vmx->pi_desc.nv = POSTED_INTR_VECTOR;
9664         vmx->pi_desc.sn = 1;
9665
9666         return &vmx->vcpu;
9667
9668 free_vmcs:
9669         free_vpid(vmx->nested.vpid02);
9670         free_loaded_vmcs(vmx->loaded_vmcs);
9671 free_msrs:
9672         kfree(vmx->guest_msrs);
9673 free_pml:
9674         vmx_destroy_pml_buffer(vmx);
9675 uninit_vcpu:
9676         kvm_vcpu_uninit(&vmx->vcpu);
9677 free_vcpu:
9678         free_vpid(vmx->vpid);
9679         kmem_cache_free(kvm_vcpu_cache, vmx);
9680         return ERR_PTR(err);
9681 }
9682
9683 static void __init vmx_check_processor_compat(void *rtn)
9684 {
9685         struct vmcs_config vmcs_conf;
9686
9687         *(int *)rtn = 0;
9688         if (setup_vmcs_config(&vmcs_conf) < 0)
9689                 *(int *)rtn = -EIO;
9690         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9691                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9692                                 smp_processor_id());
9693                 *(int *)rtn = -EIO;
9694         }
9695 }
9696
9697 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9698 {
9699         u8 cache;
9700         u64 ipat = 0;
9701
9702         /* For VT-d and EPT combination
9703          * 1. MMIO: always map as UC
9704          * 2. EPT with VT-d:
9705          *   a. VT-d without snooping control feature: can't guarantee the
9706          *      result, try to trust guest.
9707          *   b. VT-d with snooping control feature: snooping control feature of
9708          *      VT-d engine can guarantee the cache correctness. Just set it
9709          *      to WB to keep consistent with host. So the same as item 3.
9710          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9711          *    consistent with host MTRR
9712          */
9713         if (is_mmio) {
9714                 cache = MTRR_TYPE_UNCACHABLE;
9715                 goto exit;
9716         }
9717
9718         if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9719                 ipat = VMX_EPT_IPAT_BIT;
9720                 cache = MTRR_TYPE_WRBACK;
9721                 goto exit;
9722         }
9723
9724         if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9725                 ipat = VMX_EPT_IPAT_BIT;
9726                 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9727                         cache = MTRR_TYPE_WRBACK;
9728                 else
9729                         cache = MTRR_TYPE_UNCACHABLE;
9730                 goto exit;
9731         }
9732
9733         cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9734
9735 exit:
9736         return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9737 }
9738
9739 static int vmx_get_lpage_level(void)
9740 {
9741         if (enable_ept && !cpu_has_vmx_ept_1g_page())
9742                 return PT_DIRECTORY_LEVEL;
9743         else
9744                 /* For shadow and EPT supported 1GB page */
9745                 return PT_PDPE_LEVEL;
9746 }
9747
9748 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9749 {
9750         /*
9751          * These bits in the secondary execution controls field
9752          * are dynamic, the others are mostly based on the hypervisor
9753          * architecture and the guest's CPUID.  Do not touch the
9754          * dynamic bits.
9755          */
9756         u32 mask =
9757                 SECONDARY_EXEC_SHADOW_VMCS |
9758                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9759                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9760
9761         u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9762
9763         vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9764                      (new_ctl & ~mask) | (cur_ctl & mask));
9765 }
9766
9767 /*
9768  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9769  * (indicating "allowed-1") if they are supported in the guest's CPUID.
9770  */
9771 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9772 {
9773         struct vcpu_vmx *vmx = to_vmx(vcpu);
9774         struct kvm_cpuid_entry2 *entry;
9775
9776         vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9777         vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9778
9779 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {            \
9780         if (entry && (entry->_reg & (_cpuid_mask)))                     \
9781                 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask);       \
9782 } while (0)
9783
9784         entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9785         cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
9786         cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
9787         cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
9788         cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
9789         cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
9790         cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
9791         cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
9792         cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
9793         cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
9794         cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9795         cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
9796         cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
9797         cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
9798         cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
9799
9800         entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9801         cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
9802         cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
9803         cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
9804         cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
9805         cr4_fixed1_update(X86_CR4_UMIP,       ecx, bit(X86_FEATURE_UMIP));
9806
9807 #undef cr4_fixed1_update
9808 }
9809
9810 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9811 {
9812         struct vcpu_vmx *vmx = to_vmx(vcpu);
9813
9814         if (cpu_has_secondary_exec_ctrls()) {
9815                 vmx_compute_secondary_exec_control(vmx);
9816                 vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
9817         }
9818
9819         if (nested_vmx_allowed(vcpu))
9820                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9821                         FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9822         else
9823                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9824                         ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9825
9826         if (nested_vmx_allowed(vcpu))
9827                 nested_vmx_cr_fixed1_bits_update(vcpu);
9828 }
9829
9830 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9831 {
9832         if (func == 1 && nested)
9833                 entry->ecx |= bit(X86_FEATURE_VMX);
9834 }
9835
9836 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9837                 struct x86_exception *fault)
9838 {
9839         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9840         struct vcpu_vmx *vmx = to_vmx(vcpu);
9841         u32 exit_reason;
9842         unsigned long exit_qualification = vcpu->arch.exit_qualification;
9843
9844         if (vmx->nested.pml_full) {
9845                 exit_reason = EXIT_REASON_PML_FULL;
9846                 vmx->nested.pml_full = false;
9847                 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
9848         } else if (fault->error_code & PFERR_RSVD_MASK)
9849                 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9850         else
9851                 exit_reason = EXIT_REASON_EPT_VIOLATION;
9852
9853         nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
9854         vmcs12->guest_physical_address = fault->address;
9855 }
9856
9857 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
9858 {
9859         return nested_ept_get_cr3(vcpu) & VMX_EPTP_AD_ENABLE_BIT;
9860 }
9861
9862 /* Callbacks for nested_ept_init_mmu_context: */
9863
9864 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9865 {
9866         /* return the page table to be shadowed - in our case, EPT12 */
9867         return get_vmcs12(vcpu)->ept_pointer;
9868 }
9869
9870 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9871 {
9872         WARN_ON(mmu_is_nested(vcpu));
9873         if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
9874                 return 1;
9875
9876         kvm_mmu_unload(vcpu);
9877         kvm_init_shadow_ept_mmu(vcpu,
9878                         to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9879                         VMX_EPT_EXECUTE_ONLY_BIT,
9880                         nested_ept_ad_enabled(vcpu));
9881         vcpu->arch.mmu.set_cr3           = vmx_set_cr3;
9882         vcpu->arch.mmu.get_cr3           = nested_ept_get_cr3;
9883         vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9884
9885         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
9886         return 0;
9887 }
9888
9889 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9890 {
9891         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9892 }
9893
9894 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9895                                             u16 error_code)
9896 {
9897         bool inequality, bit;
9898
9899         bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9900         inequality =
9901                 (error_code & vmcs12->page_fault_error_code_mask) !=
9902                  vmcs12->page_fault_error_code_match;
9903         return inequality ^ bit;
9904 }
9905
9906 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9907                 struct x86_exception *fault)
9908 {
9909         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9910
9911         WARN_ON(!is_guest_mode(vcpu));
9912
9913         if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
9914                 !to_vmx(vcpu)->nested.nested_run_pending) {
9915                 vmcs12->vm_exit_intr_error_code = fault->error_code;
9916                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
9917                                   PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
9918                                   INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
9919                                   fault->address);
9920         } else {
9921                 kvm_inject_page_fault(vcpu, fault);
9922         }
9923 }
9924
9925 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9926                                                struct vmcs12 *vmcs12);
9927
9928 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9929                                         struct vmcs12 *vmcs12)
9930 {
9931         struct vcpu_vmx *vmx = to_vmx(vcpu);
9932         struct page *page;
9933         u64 hpa;
9934
9935         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9936                 /*
9937                  * Translate L1 physical address to host physical
9938                  * address for vmcs02. Keep the page pinned, so this
9939                  * physical address remains valid. We keep a reference
9940                  * to it so we can release it later.
9941                  */
9942                 if (vmx->nested.apic_access_page) { /* shouldn't happen */
9943                         kvm_release_page_dirty(vmx->nested.apic_access_page);
9944                         vmx->nested.apic_access_page = NULL;
9945                 }
9946                 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
9947                 /*
9948                  * If translation failed, no matter: This feature asks
9949                  * to exit when accessing the given address, and if it
9950                  * can never be accessed, this feature won't do
9951                  * anything anyway.
9952                  */
9953                 if (!is_error_page(page)) {
9954                         vmx->nested.apic_access_page = page;
9955                         hpa = page_to_phys(vmx->nested.apic_access_page);
9956                         vmcs_write64(APIC_ACCESS_ADDR, hpa);
9957                 } else {
9958                         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9959                                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9960                 }
9961         } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9962                    cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9963                 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9964                               SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9965                 kvm_vcpu_reload_apic_access_page(vcpu);
9966         }
9967
9968         if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9969                 if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
9970                         kvm_release_page_dirty(vmx->nested.virtual_apic_page);
9971                         vmx->nested.virtual_apic_page = NULL;
9972                 }
9973                 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
9974
9975                 /*
9976                  * If translation failed, VM entry will fail because
9977                  * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9978                  * Failing the vm entry is _not_ what the processor
9979                  * does but it's basically the only possibility we
9980                  * have.  We could still enter the guest if CR8 load
9981                  * exits are enabled, CR8 store exits are enabled, and
9982                  * virtualize APIC access is disabled; in this case
9983                  * the processor would never use the TPR shadow and we
9984                  * could simply clear the bit from the execution
9985                  * control.  But such a configuration is useless, so
9986                  * let's keep the code simple.
9987                  */
9988                 if (!is_error_page(page)) {
9989                         vmx->nested.virtual_apic_page = page;
9990                         hpa = page_to_phys(vmx->nested.virtual_apic_page);
9991                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9992                 }
9993         }
9994
9995         if (nested_cpu_has_posted_intr(vmcs12)) {
9996                 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9997                         kunmap(vmx->nested.pi_desc_page);
9998                         kvm_release_page_dirty(vmx->nested.pi_desc_page);
9999                         vmx->nested.pi_desc_page = NULL;
10000                 }
10001                 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
10002                 if (is_error_page(page))
10003                         return;
10004                 vmx->nested.pi_desc_page = page;
10005                 vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
10006                 vmx->nested.pi_desc =
10007                         (struct pi_desc *)((void *)vmx->nested.pi_desc +
10008                         (unsigned long)(vmcs12->posted_intr_desc_addr &
10009                         (PAGE_SIZE - 1)));
10010                 vmcs_write64(POSTED_INTR_DESC_ADDR,
10011                         page_to_phys(vmx->nested.pi_desc_page) +
10012                         (unsigned long)(vmcs12->posted_intr_desc_addr &
10013                         (PAGE_SIZE - 1)));
10014         }
10015         if (cpu_has_vmx_msr_bitmap() &&
10016             nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
10017             nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
10018                 ;
10019         else
10020                 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
10021                                 CPU_BASED_USE_MSR_BITMAPS);
10022 }
10023
10024 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
10025 {
10026         u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
10027         struct vcpu_vmx *vmx = to_vmx(vcpu);
10028
10029         if (vcpu->arch.virtual_tsc_khz == 0)
10030                 return;
10031
10032         /* Make sure short timeouts reliably trigger an immediate vmexit.
10033          * hrtimer_start does not guarantee this. */
10034         if (preemption_timeout <= 1) {
10035                 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
10036                 return;
10037         }
10038
10039         preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10040         preemption_timeout *= 1000000;
10041         do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
10042         hrtimer_start(&vmx->nested.preemption_timer,
10043                       ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
10044 }
10045
10046 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
10047                                                struct vmcs12 *vmcs12)
10048 {
10049         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
10050                 return 0;
10051
10052         if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
10053             !page_address_valid(vcpu, vmcs12->io_bitmap_b))
10054                 return -EINVAL;
10055
10056         return 0;
10057 }
10058
10059 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
10060                                                 struct vmcs12 *vmcs12)
10061 {
10062         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
10063                 return 0;
10064
10065         if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
10066                 return -EINVAL;
10067
10068         return 0;
10069 }
10070
10071 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
10072                                                 struct vmcs12 *vmcs12)
10073 {
10074         if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10075                 return 0;
10076
10077         if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
10078                 return -EINVAL;
10079
10080         return 0;
10081 }
10082
10083 /*
10084  * Merge L0's and L1's MSR bitmap, return false to indicate that
10085  * we do not use the hardware.
10086  */
10087 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
10088                                                struct vmcs12 *vmcs12)
10089 {
10090         int msr;
10091         struct page *page;
10092         unsigned long *msr_bitmap_l1;
10093         unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
10094
10095         /* This shortcut is ok because we support only x2APIC MSRs so far. */
10096         if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
10097                 return false;
10098
10099         page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
10100         if (is_error_page(page))
10101                 return false;
10102         msr_bitmap_l1 = (unsigned long *)kmap(page);
10103
10104         memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
10105
10106         if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
10107                 if (nested_cpu_has_apic_reg_virt(vmcs12))
10108                         for (msr = 0x800; msr <= 0x8ff; msr++)
10109                                 nested_vmx_disable_intercept_for_msr(
10110                                         msr_bitmap_l1, msr_bitmap_l0,
10111                                         msr, MSR_TYPE_R);
10112
10113                 nested_vmx_disable_intercept_for_msr(
10114                                 msr_bitmap_l1, msr_bitmap_l0,
10115                                 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
10116                                 MSR_TYPE_R | MSR_TYPE_W);
10117
10118                 if (nested_cpu_has_vid(vmcs12)) {
10119                         nested_vmx_disable_intercept_for_msr(
10120                                 msr_bitmap_l1, msr_bitmap_l0,
10121                                 APIC_BASE_MSR + (APIC_EOI >> 4),
10122                                 MSR_TYPE_W);
10123                         nested_vmx_disable_intercept_for_msr(
10124                                 msr_bitmap_l1, msr_bitmap_l0,
10125                                 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
10126                                 MSR_TYPE_W);
10127                 }
10128         }
10129         kunmap(page);
10130         kvm_release_page_clean(page);
10131
10132         return true;
10133 }
10134
10135 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
10136                                            struct vmcs12 *vmcs12)
10137 {
10138         if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10139             !nested_cpu_has_apic_reg_virt(vmcs12) &&
10140             !nested_cpu_has_vid(vmcs12) &&
10141             !nested_cpu_has_posted_intr(vmcs12))
10142                 return 0;
10143
10144         /*
10145          * If virtualize x2apic mode is enabled,
10146          * virtualize apic access must be disabled.
10147          */
10148         if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10149             nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
10150                 return -EINVAL;
10151
10152         /*
10153          * If virtual interrupt delivery is enabled,
10154          * we must exit on external interrupts.
10155          */
10156         if (nested_cpu_has_vid(vmcs12) &&
10157            !nested_exit_on_intr(vcpu))
10158                 return -EINVAL;
10159
10160         /*
10161          * bits 15:8 should be zero in posted_intr_nv,
10162          * the descriptor address has been already checked
10163          * in nested_get_vmcs12_pages.
10164          */
10165         if (nested_cpu_has_posted_intr(vmcs12) &&
10166            (!nested_cpu_has_vid(vmcs12) ||
10167             !nested_exit_intr_ack_set(vcpu) ||
10168             vmcs12->posted_intr_nv & 0xff00))
10169                 return -EINVAL;
10170
10171         /* tpr shadow is needed by all apicv features. */
10172         if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10173                 return -EINVAL;
10174
10175         return 0;
10176 }
10177
10178 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
10179                                        unsigned long count_field,
10180                                        unsigned long addr_field)
10181 {
10182         int maxphyaddr;
10183         u64 count, addr;
10184
10185         if (vmcs12_read_any(vcpu, count_field, &count) ||
10186             vmcs12_read_any(vcpu, addr_field, &addr)) {
10187                 WARN_ON(1);
10188                 return -EINVAL;
10189         }
10190         if (count == 0)
10191                 return 0;
10192         maxphyaddr = cpuid_maxphyaddr(vcpu);
10193         if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
10194             (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
10195                 pr_debug_ratelimited(
10196                         "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
10197                         addr_field, maxphyaddr, count, addr);
10198                 return -EINVAL;
10199         }
10200         return 0;
10201 }
10202
10203 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
10204                                                 struct vmcs12 *vmcs12)
10205 {
10206         if (vmcs12->vm_exit_msr_load_count == 0 &&
10207             vmcs12->vm_exit_msr_store_count == 0 &&
10208             vmcs12->vm_entry_msr_load_count == 0)
10209                 return 0; /* Fast path */
10210         if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
10211                                         VM_EXIT_MSR_LOAD_ADDR) ||
10212             nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
10213                                         VM_EXIT_MSR_STORE_ADDR) ||
10214             nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
10215                                         VM_ENTRY_MSR_LOAD_ADDR))
10216                 return -EINVAL;
10217         return 0;
10218 }
10219
10220 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
10221                                          struct vmcs12 *vmcs12)
10222 {
10223         u64 address = vmcs12->pml_address;
10224         int maxphyaddr = cpuid_maxphyaddr(vcpu);
10225
10226         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
10227                 if (!nested_cpu_has_ept(vmcs12) ||
10228                     !IS_ALIGNED(address, 4096)  ||
10229                     address >> maxphyaddr)
10230                         return -EINVAL;
10231         }
10232
10233         return 0;
10234 }
10235
10236 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
10237                                        struct vmx_msr_entry *e)
10238 {
10239         /* x2APIC MSR accesses are not allowed */
10240         if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
10241                 return -EINVAL;
10242         if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
10243             e->index == MSR_IA32_UCODE_REV)
10244                 return -EINVAL;
10245         if (e->reserved != 0)
10246                 return -EINVAL;
10247         return 0;
10248 }
10249
10250 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
10251                                      struct vmx_msr_entry *e)
10252 {
10253         if (e->index == MSR_FS_BASE ||
10254             e->index == MSR_GS_BASE ||
10255             e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
10256             nested_vmx_msr_check_common(vcpu, e))
10257                 return -EINVAL;
10258         return 0;
10259 }
10260
10261 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
10262                                       struct vmx_msr_entry *e)
10263 {
10264         if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
10265             nested_vmx_msr_check_common(vcpu, e))
10266                 return -EINVAL;
10267         return 0;
10268 }
10269
10270 /*
10271  * Load guest's/host's msr at nested entry/exit.
10272  * return 0 for success, entry index for failure.
10273  */
10274 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10275 {
10276         u32 i;
10277         struct vmx_msr_entry e;
10278         struct msr_data msr;
10279
10280         msr.host_initiated = false;
10281         for (i = 0; i < count; i++) {
10282                 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
10283                                         &e, sizeof(e))) {
10284                         pr_debug_ratelimited(
10285                                 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10286                                 __func__, i, gpa + i * sizeof(e));
10287                         goto fail;
10288                 }
10289                 if (nested_vmx_load_msr_check(vcpu, &e)) {
10290                         pr_debug_ratelimited(
10291                                 "%s check failed (%u, 0x%x, 0x%x)\n",
10292                                 __func__, i, e.index, e.reserved);
10293                         goto fail;
10294                 }
10295                 msr.index = e.index;
10296                 msr.data = e.value;
10297                 if (kvm_set_msr(vcpu, &msr)) {
10298                         pr_debug_ratelimited(
10299                                 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10300                                 __func__, i, e.index, e.value);
10301                         goto fail;
10302                 }
10303         }
10304         return 0;
10305 fail:
10306         return i + 1;
10307 }
10308
10309 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10310 {
10311         u32 i;
10312         struct vmx_msr_entry e;
10313
10314         for (i = 0; i < count; i++) {
10315                 struct msr_data msr_info;
10316                 if (kvm_vcpu_read_guest(vcpu,
10317                                         gpa + i * sizeof(e),
10318                                         &e, 2 * sizeof(u32))) {
10319                         pr_debug_ratelimited(
10320                                 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10321                                 __func__, i, gpa + i * sizeof(e));
10322                         return -EINVAL;
10323                 }
10324                 if (nested_vmx_store_msr_check(vcpu, &e)) {
10325                         pr_debug_ratelimited(
10326                                 "%s check failed (%u, 0x%x, 0x%x)\n",
10327                                 __func__, i, e.index, e.reserved);
10328                         return -EINVAL;
10329                 }
10330                 msr_info.host_initiated = false;
10331                 msr_info.index = e.index;
10332                 if (kvm_get_msr(vcpu, &msr_info)) {
10333                         pr_debug_ratelimited(
10334                                 "%s cannot read MSR (%u, 0x%x)\n",
10335                                 __func__, i, e.index);
10336                         return -EINVAL;
10337                 }
10338                 if (kvm_vcpu_write_guest(vcpu,
10339                                          gpa + i * sizeof(e) +
10340                                              offsetof(struct vmx_msr_entry, value),
10341                                          &msr_info.data, sizeof(msr_info.data))) {
10342                         pr_debug_ratelimited(
10343                                 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10344                                 __func__, i, e.index, msr_info.data);
10345                         return -EINVAL;
10346                 }
10347         }
10348         return 0;
10349 }
10350
10351 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
10352 {
10353         unsigned long invalid_mask;
10354
10355         invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
10356         return (val & invalid_mask) == 0;
10357 }
10358
10359 /*
10360  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
10361  * emulating VM entry into a guest with EPT enabled.
10362  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10363  * is assigned to entry_failure_code on failure.
10364  */
10365 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
10366                                u32 *entry_failure_code)
10367 {
10368         if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
10369                 if (!nested_cr3_valid(vcpu, cr3)) {
10370                         *entry_failure_code = ENTRY_FAIL_DEFAULT;
10371                         return 1;
10372                 }
10373
10374                 /*
10375                  * If PAE paging and EPT are both on, CR3 is not used by the CPU and
10376                  * must not be dereferenced.
10377                  */
10378                 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
10379                     !nested_ept) {
10380                         if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
10381                                 *entry_failure_code = ENTRY_FAIL_PDPTE;
10382                                 return 1;
10383                         }
10384                 }
10385
10386                 vcpu->arch.cr3 = cr3;
10387                 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
10388         }
10389
10390         kvm_mmu_reset_context(vcpu);
10391         return 0;
10392 }
10393
10394 /*
10395  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
10396  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
10397  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
10398  * guest in a way that will both be appropriate to L1's requests, and our
10399  * needs. In addition to modifying the active vmcs (which is vmcs02), this
10400  * function also has additional necessary side-effects, like setting various
10401  * vcpu->arch fields.
10402  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10403  * is assigned to entry_failure_code on failure.
10404  */
10405 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10406                           bool from_vmentry, u32 *entry_failure_code)
10407 {
10408         struct vcpu_vmx *vmx = to_vmx(vcpu);
10409         u32 exec_control, vmcs12_exec_ctrl;
10410
10411         vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10412         vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10413         vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10414         vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10415         vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10416         vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10417         vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10418         vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10419         vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10420         vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10421         vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10422         vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10423         vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10424         vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10425         vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10426         vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10427         vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10428         vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10429         vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10430         vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10431         vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10432         vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10433         vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10434         vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10435         vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10436         vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10437         vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10438         vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10439         vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10440         vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10441         vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10442         vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10443         vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10444         vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10445         vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10446         vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10447
10448         if (from_vmentry &&
10449             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10450                 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10451                 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10452         } else {
10453                 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10454                 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10455         }
10456         if (from_vmentry) {
10457                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10458                              vmcs12->vm_entry_intr_info_field);
10459                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10460                              vmcs12->vm_entry_exception_error_code);
10461                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10462                              vmcs12->vm_entry_instruction_len);
10463                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10464                              vmcs12->guest_interruptibility_info);
10465                 vmx->loaded_vmcs->nmi_known_unmasked =
10466                         !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
10467         } else {
10468                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10469         }
10470         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10471         vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10472         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10473                 vmcs12->guest_pending_dbg_exceptions);
10474         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10475         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10476
10477         if (nested_cpu_has_xsaves(vmcs12))
10478                 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10479         vmcs_write64(VMCS_LINK_POINTER, -1ull);
10480
10481         exec_control = vmcs12->pin_based_vm_exec_control;
10482
10483         /* Preemption timer setting is only taken from vmcs01.  */
10484         exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10485         exec_control |= vmcs_config.pin_based_exec_ctrl;
10486         if (vmx->hv_deadline_tsc == -1)
10487                 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10488
10489         /* Posted interrupts setting is only taken from vmcs12.  */
10490         if (nested_cpu_has_posted_intr(vmcs12)) {
10491                 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10492                 vmx->nested.pi_pending = false;
10493                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
10494         } else {
10495                 exec_control &= ~PIN_BASED_POSTED_INTR;
10496         }
10497
10498         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10499
10500         vmx->nested.preemption_timer_expired = false;
10501         if (nested_cpu_has_preemption_timer(vmcs12))
10502                 vmx_start_preemption_timer(vcpu);
10503
10504         /*
10505          * Whether page-faults are trapped is determined by a combination of
10506          * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10507          * If enable_ept, L0 doesn't care about page faults and we should
10508          * set all of these to L1's desires. However, if !enable_ept, L0 does
10509          * care about (at least some) page faults, and because it is not easy
10510          * (if at all possible?) to merge L0 and L1's desires, we simply ask
10511          * to exit on each and every L2 page fault. This is done by setting
10512          * MASK=MATCH=0 and (see below) EB.PF=1.
10513          * Note that below we don't need special code to set EB.PF beyond the
10514          * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10515          * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10516          * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10517          */
10518         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10519                 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10520         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10521                 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10522
10523         if (cpu_has_secondary_exec_ctrls()) {
10524                 exec_control = vmx->secondary_exec_control;
10525
10526                 /* Take the following fields only from vmcs12 */
10527                 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10528                                   SECONDARY_EXEC_ENABLE_INVPCID |
10529                                   SECONDARY_EXEC_RDTSCP |
10530                                   SECONDARY_EXEC_XSAVES |
10531                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10532                                   SECONDARY_EXEC_APIC_REGISTER_VIRT |
10533                                   SECONDARY_EXEC_ENABLE_VMFUNC);
10534                 if (nested_cpu_has(vmcs12,
10535                                    CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
10536                         vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
10537                                 ~SECONDARY_EXEC_ENABLE_PML;
10538                         exec_control |= vmcs12_exec_ctrl;
10539                 }
10540
10541                 /* All VMFUNCs are currently emulated through L0 vmexits.  */
10542                 if (exec_control & SECONDARY_EXEC_ENABLE_VMFUNC)
10543                         vmcs_write64(VM_FUNCTION_CONTROL, 0);
10544
10545                 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10546                         vmcs_write64(EOI_EXIT_BITMAP0,
10547                                 vmcs12->eoi_exit_bitmap0);
10548                         vmcs_write64(EOI_EXIT_BITMAP1,
10549                                 vmcs12->eoi_exit_bitmap1);
10550                         vmcs_write64(EOI_EXIT_BITMAP2,
10551                                 vmcs12->eoi_exit_bitmap2);
10552                         vmcs_write64(EOI_EXIT_BITMAP3,
10553                                 vmcs12->eoi_exit_bitmap3);
10554                         vmcs_write16(GUEST_INTR_STATUS,
10555                                 vmcs12->guest_intr_status);
10556                 }
10557
10558                 /*
10559                  * Write an illegal value to APIC_ACCESS_ADDR. Later,
10560                  * nested_get_vmcs12_pages will either fix it up or
10561                  * remove the VM execution control.
10562                  */
10563                 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10564                         vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10565
10566                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10567         }
10568
10569
10570         /*
10571          * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10572          * Some constant fields are set here by vmx_set_constant_host_state().
10573          * Other fields are different per CPU, and will be set later when
10574          * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10575          */
10576         vmx_set_constant_host_state(vmx);
10577
10578         /*
10579          * Set the MSR load/store lists to match L0's settings.
10580          */
10581         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10582         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10583         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10584         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10585         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10586
10587         /*
10588          * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10589          * entry, but only if the current (host) sp changed from the value
10590          * we wrote last (vmx->host_rsp). This cache is no longer relevant
10591          * if we switch vmcs, and rather than hold a separate cache per vmcs,
10592          * here we just force the write to happen on entry.
10593          */
10594         vmx->host_rsp = 0;
10595
10596         exec_control = vmx_exec_control(vmx); /* L0's desires */
10597         exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10598         exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10599         exec_control &= ~CPU_BASED_TPR_SHADOW;
10600         exec_control |= vmcs12->cpu_based_vm_exec_control;
10601
10602         /*
10603          * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10604          * nested_get_vmcs12_pages can't fix it up, the illegal value
10605          * will result in a VM entry failure.
10606          */
10607         if (exec_control & CPU_BASED_TPR_SHADOW) {
10608                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10609                 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10610         } else {
10611 #ifdef CONFIG_X86_64
10612                 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
10613                                 CPU_BASED_CR8_STORE_EXITING;
10614 #endif
10615         }
10616
10617         /*
10618          * Merging of IO bitmap not currently supported.
10619          * Rather, exit every time.
10620          */
10621         exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10622         exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10623
10624         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10625
10626         /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10627          * bitwise-or of what L1 wants to trap for L2, and what we want to
10628          * trap. Note that CR0.TS also needs updating - we do this later.
10629          */
10630         update_exception_bitmap(vcpu);
10631         vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10632         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10633
10634         /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10635          * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10636          * bits are further modified by vmx_set_efer() below.
10637          */
10638         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10639
10640         /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10641          * emulated by vmx_set_efer(), below.
10642          */
10643         vm_entry_controls_init(vmx,
10644                 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10645                         ~VM_ENTRY_IA32E_MODE) |
10646                 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10647
10648         if (from_vmentry &&
10649             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10650                 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10651                 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10652         } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10653                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10654         }
10655
10656         set_cr4_guest_host_mask(vmx);
10657
10658         if (from_vmentry &&
10659             vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10660                 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10661
10662         if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10663                 vmcs_write64(TSC_OFFSET,
10664                         vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10665         else
10666                 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10667         if (kvm_has_tsc_control)
10668                 decache_tsc_multiplier(vmx);
10669
10670         if (enable_vpid) {
10671                 /*
10672                  * There is no direct mapping between vpid02 and vpid12, the
10673                  * vpid02 is per-vCPU for L0 and reused while the value of
10674                  * vpid12 is changed w/ one invvpid during nested vmentry.
10675                  * The vpid12 is allocated by L1 for L2, so it will not
10676                  * influence global bitmap(for vpid01 and vpid02 allocation)
10677                  * even if spawn a lot of nested vCPUs.
10678                  */
10679                 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10680                         vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10681                         if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10682                                 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10683                                 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10684                         }
10685                 } else {
10686                         vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10687                         vmx_flush_tlb(vcpu);
10688                 }
10689
10690         }
10691
10692         if (enable_pml) {
10693                 /*
10694                  * Conceptually we want to copy the PML address and index from
10695                  * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10696                  * since we always flush the log on each vmexit, this happens
10697                  * to be equivalent to simply resetting the fields in vmcs02.
10698                  */
10699                 ASSERT(vmx->pml_pg);
10700                 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
10701                 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
10702         }
10703
10704         if (nested_cpu_has_ept(vmcs12)) {
10705                 if (nested_ept_init_mmu_context(vcpu)) {
10706                         *entry_failure_code = ENTRY_FAIL_DEFAULT;
10707                         return 1;
10708                 }
10709         } else if (nested_cpu_has2(vmcs12,
10710                                    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10711                 vmx_flush_tlb_ept_only(vcpu);
10712         }
10713
10714         /*
10715          * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10716          * bits which we consider mandatory enabled.
10717          * The CR0_READ_SHADOW is what L2 should have expected to read given
10718          * the specifications by L1; It's not enough to take
10719          * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10720          * have more bits than L1 expected.
10721          */
10722         vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10723         vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10724
10725         vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10726         vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10727
10728         if (from_vmentry &&
10729             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10730                 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10731         else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10732                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10733         else
10734                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10735         /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10736         vmx_set_efer(vcpu, vcpu->arch.efer);
10737
10738         /* Shadow page tables on either EPT or shadow page tables. */
10739         if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10740                                 entry_failure_code))
10741                 return 1;
10742
10743         if (!enable_ept)
10744                 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10745
10746         /*
10747          * L1 may access the L2's PDPTR, so save them to construct vmcs12
10748          */
10749         if (enable_ept) {
10750                 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10751                 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10752                 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10753                 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10754         }
10755
10756         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10757         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10758         return 0;
10759 }
10760
10761 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10762 {
10763         struct vcpu_vmx *vmx = to_vmx(vcpu);
10764
10765         if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10766             vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10767                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10768
10769         if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
10770                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10771
10772         if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10773                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10774
10775         if (nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12))
10776                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10777
10778         if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10779                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10780
10781         if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10782                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10783
10784         if (nested_vmx_check_pml_controls(vcpu, vmcs12))
10785                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10786
10787         if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10788                                 vmx->nested.nested_vmx_procbased_ctls_low,
10789                                 vmx->nested.nested_vmx_procbased_ctls_high) ||
10790             (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
10791              !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10792                                  vmx->nested.nested_vmx_secondary_ctls_low,
10793                                  vmx->nested.nested_vmx_secondary_ctls_high)) ||
10794             !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10795                                 vmx->nested.nested_vmx_pinbased_ctls_low,
10796                                 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10797             !vmx_control_verify(vmcs12->vm_exit_controls,
10798                                 vmx->nested.nested_vmx_exit_ctls_low,
10799                                 vmx->nested.nested_vmx_exit_ctls_high) ||
10800             !vmx_control_verify(vmcs12->vm_entry_controls,
10801                                 vmx->nested.nested_vmx_entry_ctls_low,
10802                                 vmx->nested.nested_vmx_entry_ctls_high))
10803                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10804
10805         if (nested_cpu_has_vmfunc(vmcs12)) {
10806                 if (vmcs12->vm_function_control &
10807                     ~vmx->nested.nested_vmx_vmfunc_controls)
10808                         return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10809
10810                 if (nested_cpu_has_eptp_switching(vmcs12)) {
10811                         if (!nested_cpu_has_ept(vmcs12) ||
10812                             !page_address_valid(vcpu, vmcs12->eptp_list_address))
10813                                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10814                 }
10815         }
10816
10817         if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
10818                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10819
10820         if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10821             !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10822             !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10823                 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10824
10825         return 0;
10826 }
10827
10828 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10829                                   u32 *exit_qual)
10830 {
10831         bool ia32e;
10832
10833         *exit_qual = ENTRY_FAIL_DEFAULT;
10834
10835         if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10836             !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10837                 return 1;
10838
10839         if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10840             vmcs12->vmcs_link_pointer != -1ull) {
10841                 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10842                 return 1;
10843         }
10844
10845         /*
10846          * If the load IA32_EFER VM-entry control is 1, the following checks
10847          * are performed on the field for the IA32_EFER MSR:
10848          * - Bits reserved in the IA32_EFER MSR must be 0.
10849          * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10850          *   the IA-32e mode guest VM-exit control. It must also be identical
10851          *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10852          *   CR0.PG) is 1.
10853          */
10854         if (to_vmx(vcpu)->nested.nested_run_pending &&
10855             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10856                 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10857                 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10858                     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10859                     ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10860                      ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10861                         return 1;
10862         }
10863
10864         /*
10865          * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10866          * IA32_EFER MSR must be 0 in the field for that register. In addition,
10867          * the values of the LMA and LME bits in the field must each be that of
10868          * the host address-space size VM-exit control.
10869          */
10870         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10871                 ia32e = (vmcs12->vm_exit_controls &
10872                          VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10873                 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10874                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10875                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10876                         return 1;
10877         }
10878
10879         if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
10880                 (is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu) ||
10881                 (vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD)))
10882                         return 1;
10883
10884         return 0;
10885 }
10886
10887 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10888 {
10889         struct vcpu_vmx *vmx = to_vmx(vcpu);
10890         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10891         struct loaded_vmcs *vmcs02;
10892         u32 msr_entry_idx;
10893         u32 exit_qual;
10894
10895         vmcs02 = nested_get_current_vmcs02(vmx);
10896         if (!vmcs02)
10897                 return -ENOMEM;
10898
10899         enter_guest_mode(vcpu);
10900
10901         if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10902                 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10903
10904         vmx_switch_vmcs(vcpu, vmcs02);
10905         vmx_segment_cache_clear(vmx);
10906
10907         if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10908                 leave_guest_mode(vcpu);
10909                 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10910                 nested_vmx_entry_failure(vcpu, vmcs12,
10911                                          EXIT_REASON_INVALID_STATE, exit_qual);
10912                 return 1;
10913         }
10914
10915         nested_get_vmcs12_pages(vcpu, vmcs12);
10916
10917         msr_entry_idx = nested_vmx_load_msr(vcpu,
10918                                             vmcs12->vm_entry_msr_load_addr,
10919                                             vmcs12->vm_entry_msr_load_count);
10920         if (msr_entry_idx) {
10921                 leave_guest_mode(vcpu);
10922                 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10923                 nested_vmx_entry_failure(vcpu, vmcs12,
10924                                 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10925                 return 1;
10926         }
10927
10928         /*
10929          * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10930          * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10931          * returned as far as L1 is concerned. It will only return (and set
10932          * the success flag) when L2 exits (see nested_vmx_vmexit()).
10933          */
10934         return 0;
10935 }
10936
10937 /*
10938  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10939  * for running an L2 nested guest.
10940  */
10941 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10942 {
10943         struct vmcs12 *vmcs12;
10944         struct vcpu_vmx *vmx = to_vmx(vcpu);
10945         u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
10946         u32 exit_qual;
10947         int ret;
10948
10949         if (!nested_vmx_check_permission(vcpu))
10950                 return 1;
10951
10952         if (!nested_vmx_check_vmcs12(vcpu))
10953                 goto out;
10954
10955         vmcs12 = get_vmcs12(vcpu);
10956
10957         if (enable_shadow_vmcs)
10958                 copy_shadow_to_vmcs12(vmx);
10959
10960         /*
10961          * The nested entry process starts with enforcing various prerequisites
10962          * on vmcs12 as required by the Intel SDM, and act appropriately when
10963          * they fail: As the SDM explains, some conditions should cause the
10964          * instruction to fail, while others will cause the instruction to seem
10965          * to succeed, but return an EXIT_REASON_INVALID_STATE.
10966          * To speed up the normal (success) code path, we should avoid checking
10967          * for misconfigurations which will anyway be caught by the processor
10968          * when using the merged vmcs02.
10969          */
10970         if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
10971                 nested_vmx_failValid(vcpu,
10972                                      VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
10973                 goto out;
10974         }
10975
10976         if (vmcs12->launch_state == launch) {
10977                 nested_vmx_failValid(vcpu,
10978                         launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10979                                : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10980                 goto out;
10981         }
10982
10983         ret = check_vmentry_prereqs(vcpu, vmcs12);
10984         if (ret) {
10985                 nested_vmx_failValid(vcpu, ret);
10986                 goto out;
10987         }
10988
10989         /*
10990          * After this point, the trap flag no longer triggers a singlestep trap
10991          * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10992          * This is not 100% correct; for performance reasons, we delegate most
10993          * of the checks on host state to the processor.  If those fail,
10994          * the singlestep trap is missed.
10995          */
10996         skip_emulated_instruction(vcpu);
10997
10998         ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10999         if (ret) {
11000                 nested_vmx_entry_failure(vcpu, vmcs12,
11001                                          EXIT_REASON_INVALID_STATE, exit_qual);
11002                 return 1;
11003         }
11004
11005         /*
11006          * We're finally done with prerequisite checking, and can start with
11007          * the nested entry.
11008          */
11009
11010         ret = enter_vmx_non_root_mode(vcpu, true);
11011         if (ret)
11012                 return ret;
11013
11014         if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
11015                 return kvm_vcpu_halt(vcpu);
11016
11017         vmx->nested.nested_run_pending = 1;
11018
11019         return 1;
11020
11021 out:
11022         return kvm_skip_emulated_instruction(vcpu);
11023 }
11024
11025 /*
11026  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
11027  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
11028  * This function returns the new value we should put in vmcs12.guest_cr0.
11029  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
11030  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
11031  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
11032  *     didn't trap the bit, because if L1 did, so would L0).
11033  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
11034  *     been modified by L2, and L1 knows it. So just leave the old value of
11035  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
11036  *     isn't relevant, because if L0 traps this bit it can set it to anything.
11037  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
11038  *     changed these bits, and therefore they need to be updated, but L0
11039  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
11040  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
11041  */
11042 static inline unsigned long
11043 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11044 {
11045         return
11046         /*1*/   (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
11047         /*2*/   (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
11048         /*3*/   (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
11049                         vcpu->arch.cr0_guest_owned_bits));
11050 }
11051
11052 static inline unsigned long
11053 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11054 {
11055         return
11056         /*1*/   (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
11057         /*2*/   (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
11058         /*3*/   (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
11059                         vcpu->arch.cr4_guest_owned_bits));
11060 }
11061
11062 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
11063                                        struct vmcs12 *vmcs12)
11064 {
11065         u32 idt_vectoring;
11066         unsigned int nr;
11067
11068         if (vcpu->arch.exception.injected) {
11069                 nr = vcpu->arch.exception.nr;
11070                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11071
11072                 if (kvm_exception_is_soft(nr)) {
11073                         vmcs12->vm_exit_instruction_len =
11074                                 vcpu->arch.event_exit_inst_len;
11075                         idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
11076                 } else
11077                         idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
11078
11079                 if (vcpu->arch.exception.has_error_code) {
11080                         idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
11081                         vmcs12->idt_vectoring_error_code =
11082                                 vcpu->arch.exception.error_code;
11083                 }
11084
11085                 vmcs12->idt_vectoring_info_field = idt_vectoring;
11086         } else if (vcpu->arch.nmi_injected) {
11087                 vmcs12->idt_vectoring_info_field =
11088                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
11089         } else if (vcpu->arch.interrupt.pending) {
11090                 nr = vcpu->arch.interrupt.nr;
11091                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11092
11093                 if (vcpu->arch.interrupt.soft) {
11094                         idt_vectoring |= INTR_TYPE_SOFT_INTR;
11095                         vmcs12->vm_entry_instruction_len =
11096                                 vcpu->arch.event_exit_inst_len;
11097                 } else
11098                         idt_vectoring |= INTR_TYPE_EXT_INTR;
11099
11100                 vmcs12->idt_vectoring_info_field = idt_vectoring;
11101         }
11102 }
11103
11104 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
11105 {
11106         struct vcpu_vmx *vmx = to_vmx(vcpu);
11107         unsigned long exit_qual;
11108         bool block_nested_events =
11109             vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
11110
11111         if (vcpu->arch.exception.pending &&
11112                 nested_vmx_check_exception(vcpu, &exit_qual)) {
11113                 if (block_nested_events)
11114                         return -EBUSY;
11115                 nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
11116                 vcpu->arch.exception.pending = false;
11117                 return 0;
11118         }
11119
11120         if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
11121             vmx->nested.preemption_timer_expired) {
11122                 if (block_nested_events)
11123                         return -EBUSY;
11124                 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
11125                 return 0;
11126         }
11127
11128         if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
11129                 if (block_nested_events)
11130                         return -EBUSY;
11131                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
11132                                   NMI_VECTOR | INTR_TYPE_NMI_INTR |
11133                                   INTR_INFO_VALID_MASK, 0);
11134                 /*
11135                  * The NMI-triggered VM exit counts as injection:
11136                  * clear this one and block further NMIs.
11137                  */
11138                 vcpu->arch.nmi_pending = 0;
11139                 vmx_set_nmi_mask(vcpu, true);
11140                 return 0;
11141         }
11142
11143         if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
11144             nested_exit_on_intr(vcpu)) {
11145                 if (block_nested_events)
11146                         return -EBUSY;
11147                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
11148                 return 0;
11149         }
11150
11151         vmx_complete_nested_posted_interrupt(vcpu);
11152         return 0;
11153 }
11154
11155 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
11156 {
11157         ktime_t remaining =
11158                 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
11159         u64 value;
11160
11161         if (ktime_to_ns(remaining) <= 0)
11162                 return 0;
11163
11164         value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
11165         do_div(value, 1000000);
11166         return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
11167 }
11168
11169 /*
11170  * Update the guest state fields of vmcs12 to reflect changes that
11171  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
11172  * VM-entry controls is also updated, since this is really a guest
11173  * state bit.)
11174  */
11175 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11176 {
11177         vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
11178         vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
11179
11180         vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
11181         vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
11182         vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
11183
11184         vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
11185         vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
11186         vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
11187         vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
11188         vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
11189         vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
11190         vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
11191         vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
11192         vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
11193         vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
11194         vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
11195         vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
11196         vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
11197         vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
11198         vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
11199         vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
11200         vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
11201         vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
11202         vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
11203         vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
11204         vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
11205         vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
11206         vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
11207         vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
11208         vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
11209         vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
11210         vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
11211         vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
11212         vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
11213         vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
11214         vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
11215         vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
11216         vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
11217         vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
11218         vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
11219         vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
11220
11221         vmcs12->guest_interruptibility_info =
11222                 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
11223         vmcs12->guest_pending_dbg_exceptions =
11224                 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
11225         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
11226                 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
11227         else
11228                 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
11229
11230         if (nested_cpu_has_preemption_timer(vmcs12)) {
11231                 if (vmcs12->vm_exit_controls &
11232                     VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
11233                         vmcs12->vmx_preemption_timer_value =
11234                                 vmx_get_preemption_timer_value(vcpu);
11235                 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
11236         }
11237
11238         /*
11239          * In some cases (usually, nested EPT), L2 is allowed to change its
11240          * own CR3 without exiting. If it has changed it, we must keep it.
11241          * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
11242          * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
11243          *
11244          * Additionally, restore L2's PDPTR to vmcs12.
11245          */
11246         if (enable_ept) {
11247                 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
11248                 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
11249                 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
11250                 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
11251                 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
11252         }
11253
11254         vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
11255
11256         if (nested_cpu_has_vid(vmcs12))
11257                 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
11258
11259         vmcs12->vm_entry_controls =
11260                 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
11261                 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
11262
11263         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
11264                 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
11265                 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11266         }
11267
11268         /* TODO: These cannot have changed unless we have MSR bitmaps and
11269          * the relevant bit asks not to trap the change */
11270         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
11271                 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
11272         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
11273                 vmcs12->guest_ia32_efer = vcpu->arch.efer;
11274         vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
11275         vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
11276         vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
11277         if (kvm_mpx_supported())
11278                 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
11279 }
11280
11281 /*
11282  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
11283  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
11284  * and this function updates it to reflect the changes to the guest state while
11285  * L2 was running (and perhaps made some exits which were handled directly by L0
11286  * without going back to L1), and to reflect the exit reason.
11287  * Note that we do not have to copy here all VMCS fields, just those that
11288  * could have changed by the L2 guest or the exit - i.e., the guest-state and
11289  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
11290  * which already writes to vmcs12 directly.
11291  */
11292 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11293                            u32 exit_reason, u32 exit_intr_info,
11294                            unsigned long exit_qualification)
11295 {
11296         /* update guest state fields: */
11297         sync_vmcs12(vcpu, vmcs12);
11298
11299         /* update exit information fields: */
11300
11301         vmcs12->vm_exit_reason = exit_reason;
11302         vmcs12->exit_qualification = exit_qualification;
11303         vmcs12->vm_exit_intr_info = exit_intr_info;
11304
11305         vmcs12->idt_vectoring_info_field = 0;
11306         vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
11307         vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
11308
11309         if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
11310                 vmcs12->launch_state = 1;
11311
11312                 /* vm_entry_intr_info_field is cleared on exit. Emulate this
11313                  * instead of reading the real value. */
11314                 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
11315
11316                 /*
11317                  * Transfer the event that L0 or L1 may wanted to inject into
11318                  * L2 to IDT_VECTORING_INFO_FIELD.
11319                  */
11320                 vmcs12_save_pending_event(vcpu, vmcs12);
11321         }
11322
11323         /*
11324          * Drop what we picked up for L2 via vmx_complete_interrupts. It is
11325          * preserved above and would only end up incorrectly in L1.
11326          */
11327         vcpu->arch.nmi_injected = false;
11328         kvm_clear_exception_queue(vcpu);
11329         kvm_clear_interrupt_queue(vcpu);
11330 }
11331
11332 static void load_vmcs12_mmu_host_state(struct kvm_vcpu *vcpu,
11333                         struct vmcs12 *vmcs12)
11334 {
11335         u32 entry_failure_code;
11336
11337         nested_ept_uninit_mmu_context(vcpu);
11338
11339         /*
11340          * Only PDPTE load can fail as the value of cr3 was checked on entry and
11341          * couldn't have changed.
11342          */
11343         if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
11344                 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
11345
11346         if (!enable_ept)
11347                 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
11348 }
11349
11350 /*
11351  * A part of what we need to when the nested L2 guest exits and we want to
11352  * run its L1 parent, is to reset L1's guest state to the host state specified
11353  * in vmcs12.
11354  * This function is to be called not only on normal nested exit, but also on
11355  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
11356  * Failures During or After Loading Guest State").
11357  * This function should be called when the active VMCS is L1's (vmcs01).
11358  */
11359 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
11360                                    struct vmcs12 *vmcs12)
11361 {
11362         struct kvm_segment seg;
11363
11364         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
11365                 vcpu->arch.efer = vmcs12->host_ia32_efer;
11366         else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11367                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
11368         else
11369                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
11370         vmx_set_efer(vcpu, vcpu->arch.efer);
11371
11372         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
11373         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
11374         vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
11375         /*
11376          * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
11377          * actually changed, because vmx_set_cr0 refers to efer set above.
11378          *
11379          * CR0_GUEST_HOST_MASK is already set in the original vmcs01
11380          * (KVM doesn't change it);
11381          */
11382         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
11383         vmx_set_cr0(vcpu, vmcs12->host_cr0);
11384
11385         /* Same as above - no reason to call set_cr4_guest_host_mask().  */
11386         vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
11387         vmx_set_cr4(vcpu, vmcs12->host_cr4);
11388
11389         load_vmcs12_mmu_host_state(vcpu, vmcs12);
11390
11391         if (enable_vpid) {
11392                 /*
11393                  * Trivially support vpid by letting L2s share their parent
11394                  * L1's vpid. TODO: move to a more elaborate solution, giving
11395                  * each L2 its own vpid and exposing the vpid feature to L1.
11396                  */
11397                 vmx_flush_tlb(vcpu);
11398         }
11399         /* Restore posted intr vector. */
11400         if (nested_cpu_has_posted_intr(vmcs12))
11401                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
11402
11403         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
11404         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
11405         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
11406         vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
11407         vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
11408         vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
11409         vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
11410
11411         /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
11412         if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
11413                 vmcs_write64(GUEST_BNDCFGS, 0);
11414
11415         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
11416                 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
11417                 vcpu->arch.pat = vmcs12->host_ia32_pat;
11418         }
11419         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
11420                 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
11421                         vmcs12->host_ia32_perf_global_ctrl);
11422
11423         /* Set L1 segment info according to Intel SDM
11424             27.5.2 Loading Host Segment and Descriptor-Table Registers */
11425         seg = (struct kvm_segment) {
11426                 .base = 0,
11427                 .limit = 0xFFFFFFFF,
11428                 .selector = vmcs12->host_cs_selector,
11429                 .type = 11,
11430                 .present = 1,
11431                 .s = 1,
11432                 .g = 1
11433         };
11434         if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11435                 seg.l = 1;
11436         else
11437                 seg.db = 1;
11438         vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
11439         seg = (struct kvm_segment) {
11440                 .base = 0,
11441                 .limit = 0xFFFFFFFF,
11442                 .type = 3,
11443                 .present = 1,
11444                 .s = 1,
11445                 .db = 1,
11446                 .g = 1
11447         };
11448         seg.selector = vmcs12->host_ds_selector;
11449         vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
11450         seg.selector = vmcs12->host_es_selector;
11451         vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
11452         seg.selector = vmcs12->host_ss_selector;
11453         vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
11454         seg.selector = vmcs12->host_fs_selector;
11455         seg.base = vmcs12->host_fs_base;
11456         vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
11457         seg.selector = vmcs12->host_gs_selector;
11458         seg.base = vmcs12->host_gs_base;
11459         vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
11460         seg = (struct kvm_segment) {
11461                 .base = vmcs12->host_tr_base,
11462                 .limit = 0x67,
11463                 .selector = vmcs12->host_tr_selector,
11464                 .type = 11,
11465                 .present = 1
11466         };
11467         vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
11468
11469         kvm_set_dr(vcpu, 7, 0x400);
11470         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11471
11472         if (cpu_has_vmx_msr_bitmap())
11473                 vmx_set_msr_bitmap(vcpu);
11474
11475         if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11476                                 vmcs12->vm_exit_msr_load_count))
11477                 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11478 }
11479
11480 /*
11481  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11482  * and modify vmcs12 to make it see what it would expect to see there if
11483  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11484  */
11485 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11486                               u32 exit_intr_info,
11487                               unsigned long exit_qualification)
11488 {
11489         struct vcpu_vmx *vmx = to_vmx(vcpu);
11490         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11491
11492         /* trying to cancel vmlaunch/vmresume is a bug */
11493         WARN_ON_ONCE(vmx->nested.nested_run_pending);
11494
11495         /*
11496          * The only expected VM-instruction error is "VM entry with
11497          * invalid control field(s)." Anything else indicates a
11498          * problem with L0.
11499          */
11500         WARN_ON_ONCE(vmx->fail && (vmcs_read32(VM_INSTRUCTION_ERROR) !=
11501                                    VMXERR_ENTRY_INVALID_CONTROL_FIELD));
11502
11503         leave_guest_mode(vcpu);
11504
11505         if (likely(!vmx->fail)) {
11506                 if (exit_reason == -1)
11507                         sync_vmcs12(vcpu, vmcs12);
11508                 else
11509                         prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11510                                        exit_qualification);
11511
11512                 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11513                                          vmcs12->vm_exit_msr_store_count))
11514                         nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11515         }
11516
11517         vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11518         vm_entry_controls_reset_shadow(vmx);
11519         vm_exit_controls_reset_shadow(vmx);
11520         vmx_segment_cache_clear(vmx);
11521
11522         /* if no vmcs02 cache requested, remove the one we used */
11523         if (VMCS02_POOL_SIZE == 0)
11524                 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
11525
11526         /* Update any VMCS fields that might have changed while L2 ran */
11527         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11528         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11529         vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11530         if (vmx->hv_deadline_tsc == -1)
11531                 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11532                                 PIN_BASED_VMX_PREEMPTION_TIMER);
11533         else
11534                 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11535                               PIN_BASED_VMX_PREEMPTION_TIMER);
11536         if (kvm_has_tsc_control)
11537                 decache_tsc_multiplier(vmx);
11538
11539         if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11540                 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11541                 vmx_set_virtual_x2apic_mode(vcpu,
11542                                 vcpu->arch.apic_base & X2APIC_ENABLE);
11543         } else if (!nested_cpu_has_ept(vmcs12) &&
11544                    nested_cpu_has2(vmcs12,
11545                                    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11546                 vmx_flush_tlb_ept_only(vcpu);
11547         }
11548
11549         /* This is needed for same reason as it was needed in prepare_vmcs02 */
11550         vmx->host_rsp = 0;
11551
11552         /* Unpin physical memory we referred to in vmcs02 */
11553         if (vmx->nested.apic_access_page) {
11554                 kvm_release_page_dirty(vmx->nested.apic_access_page);
11555                 vmx->nested.apic_access_page = NULL;
11556         }
11557         if (vmx->nested.virtual_apic_page) {
11558                 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
11559                 vmx->nested.virtual_apic_page = NULL;
11560         }
11561         if (vmx->nested.pi_desc_page) {
11562                 kunmap(vmx->nested.pi_desc_page);
11563                 kvm_release_page_dirty(vmx->nested.pi_desc_page);
11564                 vmx->nested.pi_desc_page = NULL;
11565                 vmx->nested.pi_desc = NULL;
11566         }
11567
11568         /*
11569          * We are now running in L2, mmu_notifier will force to reload the
11570          * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11571          */
11572         kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11573
11574         if (enable_shadow_vmcs && exit_reason != -1)
11575                 vmx->nested.sync_shadow_vmcs = true;
11576
11577         /* in case we halted in L2 */
11578         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11579
11580         if (likely(!vmx->fail)) {
11581                 /*
11582                  * TODO: SDM says that with acknowledge interrupt on
11583                  * exit, bit 31 of the VM-exit interrupt information
11584                  * (valid interrupt) is always set to 1 on
11585                  * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
11586                  * need kvm_cpu_has_interrupt().  See the commit
11587                  * message for details.
11588                  */
11589                 if (nested_exit_intr_ack_set(vcpu) &&
11590                     exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
11591                     kvm_cpu_has_interrupt(vcpu)) {
11592                         int irq = kvm_cpu_get_interrupt(vcpu);
11593                         WARN_ON(irq < 0);
11594                         vmcs12->vm_exit_intr_info = irq |
11595                                 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11596                 }
11597
11598                 if (exit_reason != -1)
11599                         trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11600                                                        vmcs12->exit_qualification,
11601                                                        vmcs12->idt_vectoring_info_field,
11602                                                        vmcs12->vm_exit_intr_info,
11603                                                        vmcs12->vm_exit_intr_error_code,
11604                                                        KVM_ISA_VMX);
11605
11606                 load_vmcs12_host_state(vcpu, vmcs12);
11607
11608                 return;
11609         }
11610
11611         /*
11612          * After an early L2 VM-entry failure, we're now back
11613          * in L1 which thinks it just finished a VMLAUNCH or
11614          * VMRESUME instruction, so we need to set the failure
11615          * flag and the VM-instruction error field of the VMCS
11616          * accordingly.
11617          */
11618         nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
11619
11620         load_vmcs12_mmu_host_state(vcpu, vmcs12);
11621
11622         /*
11623          * The emulated instruction was already skipped in
11624          * nested_vmx_run, but the updated RIP was never
11625          * written back to the vmcs01.
11626          */
11627         skip_emulated_instruction(vcpu);
11628         vmx->fail = 0;
11629 }
11630
11631 /*
11632  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11633  */
11634 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11635 {
11636         if (is_guest_mode(vcpu)) {
11637                 to_vmx(vcpu)->nested.nested_run_pending = 0;
11638                 nested_vmx_vmexit(vcpu, -1, 0, 0);
11639         }
11640         free_nested(to_vmx(vcpu));
11641 }
11642
11643 /*
11644  * L1's failure to enter L2 is a subset of a normal exit, as explained in
11645  * 23.7 "VM-entry failures during or after loading guest state" (this also
11646  * lists the acceptable exit-reason and exit-qualification parameters).
11647  * It should only be called before L2 actually succeeded to run, and when
11648  * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11649  */
11650 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11651                         struct vmcs12 *vmcs12,
11652                         u32 reason, unsigned long qualification)
11653 {
11654         load_vmcs12_host_state(vcpu, vmcs12);
11655         vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11656         vmcs12->exit_qualification = qualification;
11657         nested_vmx_succeed(vcpu);
11658         if (enable_shadow_vmcs)
11659                 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11660 }
11661
11662 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11663                                struct x86_instruction_info *info,
11664                                enum x86_intercept_stage stage)
11665 {
11666         return X86EMUL_CONTINUE;
11667 }
11668
11669 #ifdef CONFIG_X86_64
11670 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11671 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11672                                   u64 divisor, u64 *result)
11673 {
11674         u64 low = a << shift, high = a >> (64 - shift);
11675
11676         /* To avoid the overflow on divq */
11677         if (high >= divisor)
11678                 return 1;
11679
11680         /* Low hold the result, high hold rem which is discarded */
11681         asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11682             "rm" (divisor), "0" (low), "1" (high));
11683         *result = low;
11684
11685         return 0;
11686 }
11687
11688 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11689 {
11690         struct vcpu_vmx *vmx = to_vmx(vcpu);
11691         u64 tscl = rdtsc();
11692         u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11693         u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11694
11695         /* Convert to host delta tsc if tsc scaling is enabled */
11696         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11697                         u64_shl_div_u64(delta_tsc,
11698                                 kvm_tsc_scaling_ratio_frac_bits,
11699                                 vcpu->arch.tsc_scaling_ratio,
11700                                 &delta_tsc))
11701                 return -ERANGE;
11702
11703         /*
11704          * If the delta tsc can't fit in the 32 bit after the multi shift,
11705          * we can't use the preemption timer.
11706          * It's possible that it fits on later vmentries, but checking
11707          * on every vmentry is costly so we just use an hrtimer.
11708          */
11709         if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11710                 return -ERANGE;
11711
11712         vmx->hv_deadline_tsc = tscl + delta_tsc;
11713         vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11714                         PIN_BASED_VMX_PREEMPTION_TIMER);
11715
11716         return delta_tsc == 0;
11717 }
11718
11719 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11720 {
11721         struct vcpu_vmx *vmx = to_vmx(vcpu);
11722         vmx->hv_deadline_tsc = -1;
11723         vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11724                         PIN_BASED_VMX_PREEMPTION_TIMER);
11725 }
11726 #endif
11727
11728 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11729 {
11730         if (ple_gap)
11731                 shrink_ple_window(vcpu);
11732 }
11733
11734 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11735                                      struct kvm_memory_slot *slot)
11736 {
11737         kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11738         kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11739 }
11740
11741 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11742                                        struct kvm_memory_slot *slot)
11743 {
11744         kvm_mmu_slot_set_dirty(kvm, slot);
11745 }
11746
11747 static void vmx_flush_log_dirty(struct kvm *kvm)
11748 {
11749         kvm_flush_pml_buffers(kvm);
11750 }
11751
11752 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
11753 {
11754         struct vmcs12 *vmcs12;
11755         struct vcpu_vmx *vmx = to_vmx(vcpu);
11756         gpa_t gpa;
11757         struct page *page = NULL;
11758         u64 *pml_address;
11759
11760         if (is_guest_mode(vcpu)) {
11761                 WARN_ON_ONCE(vmx->nested.pml_full);
11762
11763                 /*
11764                  * Check if PML is enabled for the nested guest.
11765                  * Whether eptp bit 6 is set is already checked
11766                  * as part of A/D emulation.
11767                  */
11768                 vmcs12 = get_vmcs12(vcpu);
11769                 if (!nested_cpu_has_pml(vmcs12))
11770                         return 0;
11771
11772                 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
11773                         vmx->nested.pml_full = true;
11774                         return 1;
11775                 }
11776
11777                 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
11778
11779                 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
11780                 if (is_error_page(page))
11781                         return 0;
11782
11783                 pml_address = kmap(page);
11784                 pml_address[vmcs12->guest_pml_index--] = gpa;
11785                 kunmap(page);
11786                 kvm_release_page_clean(page);
11787         }
11788
11789         return 0;
11790 }
11791
11792 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11793                                            struct kvm_memory_slot *memslot,
11794                                            gfn_t offset, unsigned long mask)
11795 {
11796         kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11797 }
11798
11799 static void __pi_post_block(struct kvm_vcpu *vcpu)
11800 {
11801         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11802         struct pi_desc old, new;
11803         unsigned int dest;
11804
11805         do {
11806                 old.control = new.control = pi_desc->control;
11807                 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
11808                      "Wakeup handler not enabled while the VCPU is blocked\n");
11809
11810                 dest = cpu_physical_id(vcpu->cpu);
11811
11812                 if (x2apic_enabled())
11813                         new.ndst = dest;
11814                 else
11815                         new.ndst = (dest << 8) & 0xFF00;
11816
11817                 /* set 'NV' to 'notification vector' */
11818                 new.nv = POSTED_INTR_VECTOR;
11819         } while (cmpxchg64(&pi_desc->control, old.control,
11820                            new.control) != old.control);
11821
11822         if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
11823                 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11824                 list_del(&vcpu->blocked_vcpu_list);
11825                 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11826                 vcpu->pre_pcpu = -1;
11827         }
11828 }
11829
11830 /*
11831  * This routine does the following things for vCPU which is going
11832  * to be blocked if VT-d PI is enabled.
11833  * - Store the vCPU to the wakeup list, so when interrupts happen
11834  *   we can find the right vCPU to wake up.
11835  * - Change the Posted-interrupt descriptor as below:
11836  *      'NDST' <-- vcpu->pre_pcpu
11837  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11838  * - If 'ON' is set during this process, which means at least one
11839  *   interrupt is posted for this vCPU, we cannot block it, in
11840  *   this case, return 1, otherwise, return 0.
11841  *
11842  */
11843 static int pi_pre_block(struct kvm_vcpu *vcpu)
11844 {
11845         unsigned int dest;
11846         struct pi_desc old, new;
11847         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11848
11849         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11850                 !irq_remapping_cap(IRQ_POSTING_CAP)  ||
11851                 !kvm_vcpu_apicv_active(vcpu))
11852                 return 0;
11853
11854         WARN_ON(irqs_disabled());
11855         local_irq_disable();
11856         if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
11857                 vcpu->pre_pcpu = vcpu->cpu;
11858                 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11859                 list_add_tail(&vcpu->blocked_vcpu_list,
11860                               &per_cpu(blocked_vcpu_on_cpu,
11861                                        vcpu->pre_pcpu));
11862                 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11863         }
11864
11865         do {
11866                 old.control = new.control = pi_desc->control;
11867
11868                 WARN((pi_desc->sn == 1),
11869                      "Warning: SN field of posted-interrupts "
11870                      "is set before blocking\n");
11871
11872                 /*
11873                  * Since vCPU can be preempted during this process,
11874                  * vcpu->cpu could be different with pre_pcpu, we
11875                  * need to set pre_pcpu as the destination of wakeup
11876                  * notification event, then we can find the right vCPU
11877                  * to wakeup in wakeup handler if interrupts happen
11878                  * when the vCPU is in blocked state.
11879                  */
11880                 dest = cpu_physical_id(vcpu->pre_pcpu);
11881
11882                 if (x2apic_enabled())
11883                         new.ndst = dest;
11884                 else
11885                         new.ndst = (dest << 8) & 0xFF00;
11886
11887                 /* set 'NV' to 'wakeup vector' */
11888                 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11889         } while (cmpxchg64(&pi_desc->control, old.control,
11890                            new.control) != old.control);
11891
11892         /* We should not block the vCPU if an interrupt is posted for it.  */
11893         if (pi_test_on(pi_desc) == 1)
11894                 __pi_post_block(vcpu);
11895
11896         local_irq_enable();
11897         return (vcpu->pre_pcpu == -1);
11898 }
11899
11900 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11901 {
11902         if (pi_pre_block(vcpu))
11903                 return 1;
11904
11905         if (kvm_lapic_hv_timer_in_use(vcpu))
11906                 kvm_lapic_switch_to_sw_timer(vcpu);
11907
11908         return 0;
11909 }
11910
11911 static void pi_post_block(struct kvm_vcpu *vcpu)
11912 {
11913         if (vcpu->pre_pcpu == -1)
11914                 return;
11915
11916         WARN_ON(irqs_disabled());
11917         local_irq_disable();
11918         __pi_post_block(vcpu);
11919         local_irq_enable();
11920 }
11921
11922 static void vmx_post_block(struct kvm_vcpu *vcpu)
11923 {
11924         if (kvm_x86_ops->set_hv_timer)
11925                 kvm_lapic_switch_to_hv_timer(vcpu);
11926
11927         pi_post_block(vcpu);
11928 }
11929
11930 /*
11931  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11932  *
11933  * @kvm: kvm
11934  * @host_irq: host irq of the interrupt
11935  * @guest_irq: gsi of the interrupt
11936  * @set: set or unset PI
11937  * returns 0 on success, < 0 on failure
11938  */
11939 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11940                               uint32_t guest_irq, bool set)
11941 {
11942         struct kvm_kernel_irq_routing_entry *e;
11943         struct kvm_irq_routing_table *irq_rt;
11944         struct kvm_lapic_irq irq;
11945         struct kvm_vcpu *vcpu;
11946         struct vcpu_data vcpu_info;
11947         int idx, ret = 0;
11948
11949         if (!kvm_arch_has_assigned_device(kvm) ||
11950                 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11951                 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11952                 return 0;
11953
11954         idx = srcu_read_lock(&kvm->irq_srcu);
11955         irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11956         if (guest_irq >= irq_rt->nr_rt_entries ||
11957             hlist_empty(&irq_rt->map[guest_irq])) {
11958                 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
11959                              guest_irq, irq_rt->nr_rt_entries);
11960                 goto out;
11961         }
11962
11963         hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11964                 if (e->type != KVM_IRQ_ROUTING_MSI)
11965                         continue;
11966                 /*
11967                  * VT-d PI cannot support posting multicast/broadcast
11968                  * interrupts to a vCPU, we still use interrupt remapping
11969                  * for these kind of interrupts.
11970                  *
11971                  * For lowest-priority interrupts, we only support
11972                  * those with single CPU as the destination, e.g. user
11973                  * configures the interrupts via /proc/irq or uses
11974                  * irqbalance to make the interrupts single-CPU.
11975                  *
11976                  * We will support full lowest-priority interrupt later.
11977                  */
11978
11979                 kvm_set_msi_irq(kvm, e, &irq);
11980                 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11981                         /*
11982                          * Make sure the IRTE is in remapped mode if
11983                          * we don't handle it in posted mode.
11984                          */
11985                         ret = irq_set_vcpu_affinity(host_irq, NULL);
11986                         if (ret < 0) {
11987                                 printk(KERN_INFO
11988                                    "failed to back to remapped mode, irq: %u\n",
11989                                    host_irq);
11990                                 goto out;
11991                         }
11992
11993                         continue;
11994                 }
11995
11996                 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11997                 vcpu_info.vector = irq.vector;
11998
11999                 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
12000                                 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
12001
12002                 if (set)
12003                         ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
12004                 else
12005                         ret = irq_set_vcpu_affinity(host_irq, NULL);
12006
12007                 if (ret < 0) {
12008                         printk(KERN_INFO "%s: failed to update PI IRTE\n",
12009                                         __func__);
12010                         goto out;
12011                 }
12012         }
12013
12014         ret = 0;
12015 out:
12016         srcu_read_unlock(&kvm->irq_srcu, idx);
12017         return ret;
12018 }
12019
12020 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
12021 {
12022         if (vcpu->arch.mcg_cap & MCG_LMCE_P)
12023                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
12024                         FEATURE_CONTROL_LMCE;
12025         else
12026                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
12027                         ~FEATURE_CONTROL_LMCE;
12028 }
12029
12030 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
12031 {
12032         /* we need a nested vmexit to enter SMM, postpone if run is pending */
12033         if (to_vmx(vcpu)->nested.nested_run_pending)
12034                 return 0;
12035         return 1;
12036 }
12037
12038 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
12039 {
12040         struct vcpu_vmx *vmx = to_vmx(vcpu);
12041
12042         vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
12043         if (vmx->nested.smm.guest_mode)
12044                 nested_vmx_vmexit(vcpu, -1, 0, 0);
12045
12046         vmx->nested.smm.vmxon = vmx->nested.vmxon;
12047         vmx->nested.vmxon = false;
12048         return 0;
12049 }
12050
12051 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
12052 {
12053         struct vcpu_vmx *vmx = to_vmx(vcpu);
12054         int ret;
12055
12056         if (vmx->nested.smm.vmxon) {
12057                 vmx->nested.vmxon = true;
12058                 vmx->nested.smm.vmxon = false;
12059         }
12060
12061         if (vmx->nested.smm.guest_mode) {
12062                 vcpu->arch.hflags &= ~HF_SMM_MASK;
12063                 ret = enter_vmx_non_root_mode(vcpu, false);
12064                 vcpu->arch.hflags |= HF_SMM_MASK;
12065                 if (ret)
12066                         return ret;
12067
12068                 vmx->nested.smm.guest_mode = false;
12069         }
12070         return 0;
12071 }
12072
12073 static int enable_smi_window(struct kvm_vcpu *vcpu)
12074 {
12075         return 0;
12076 }
12077
12078 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
12079         .cpu_has_kvm_support = cpu_has_kvm_support,
12080         .disabled_by_bios = vmx_disabled_by_bios,
12081         .hardware_setup = hardware_setup,
12082         .hardware_unsetup = hardware_unsetup,
12083         .check_processor_compatibility = vmx_check_processor_compat,
12084         .hardware_enable = hardware_enable,
12085         .hardware_disable = hardware_disable,
12086         .cpu_has_accelerated_tpr = report_flexpriority,
12087         .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
12088
12089         .vcpu_create = vmx_create_vcpu,
12090         .vcpu_free = vmx_free_vcpu,
12091         .vcpu_reset = vmx_vcpu_reset,
12092
12093         .prepare_guest_switch = vmx_save_host_state,
12094         .vcpu_load = vmx_vcpu_load,
12095         .vcpu_put = vmx_vcpu_put,
12096
12097         .update_bp_intercept = update_exception_bitmap,
12098         .get_msr = vmx_get_msr,
12099         .set_msr = vmx_set_msr,
12100         .get_segment_base = vmx_get_segment_base,
12101         .get_segment = vmx_get_segment,
12102         .set_segment = vmx_set_segment,
12103         .get_cpl = vmx_get_cpl,
12104         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
12105         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
12106         .decache_cr3 = vmx_decache_cr3,
12107         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
12108         .set_cr0 = vmx_set_cr0,
12109         .set_cr3 = vmx_set_cr3,
12110         .set_cr4 = vmx_set_cr4,
12111         .set_efer = vmx_set_efer,
12112         .get_idt = vmx_get_idt,
12113         .set_idt = vmx_set_idt,
12114         .get_gdt = vmx_get_gdt,
12115         .set_gdt = vmx_set_gdt,
12116         .get_dr6 = vmx_get_dr6,
12117         .set_dr6 = vmx_set_dr6,
12118         .set_dr7 = vmx_set_dr7,
12119         .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
12120         .cache_reg = vmx_cache_reg,
12121         .get_rflags = vmx_get_rflags,
12122         .set_rflags = vmx_set_rflags,
12123
12124         .tlb_flush = vmx_flush_tlb,
12125
12126         .run = vmx_vcpu_run,
12127         .handle_exit = vmx_handle_exit,
12128         .skip_emulated_instruction = skip_emulated_instruction,
12129         .set_interrupt_shadow = vmx_set_interrupt_shadow,
12130         .get_interrupt_shadow = vmx_get_interrupt_shadow,
12131         .patch_hypercall = vmx_patch_hypercall,
12132         .set_irq = vmx_inject_irq,
12133         .set_nmi = vmx_inject_nmi,
12134         .queue_exception = vmx_queue_exception,
12135         .cancel_injection = vmx_cancel_injection,
12136         .interrupt_allowed = vmx_interrupt_allowed,
12137         .nmi_allowed = vmx_nmi_allowed,
12138         .get_nmi_mask = vmx_get_nmi_mask,
12139         .set_nmi_mask = vmx_set_nmi_mask,
12140         .enable_nmi_window = enable_nmi_window,
12141         .enable_irq_window = enable_irq_window,
12142         .update_cr8_intercept = update_cr8_intercept,
12143         .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
12144         .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
12145         .get_enable_apicv = vmx_get_enable_apicv,
12146         .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
12147         .load_eoi_exitmap = vmx_load_eoi_exitmap,
12148         .apicv_post_state_restore = vmx_apicv_post_state_restore,
12149         .hwapic_irr_update = vmx_hwapic_irr_update,
12150         .hwapic_isr_update = vmx_hwapic_isr_update,
12151         .sync_pir_to_irr = vmx_sync_pir_to_irr,
12152         .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
12153
12154         .set_tss_addr = vmx_set_tss_addr,
12155         .get_tdp_level = get_ept_level,
12156         .get_mt_mask = vmx_get_mt_mask,
12157
12158         .get_exit_info = vmx_get_exit_info,
12159
12160         .get_lpage_level = vmx_get_lpage_level,
12161
12162         .cpuid_update = vmx_cpuid_update,
12163
12164         .rdtscp_supported = vmx_rdtscp_supported,
12165         .invpcid_supported = vmx_invpcid_supported,
12166
12167         .set_supported_cpuid = vmx_set_supported_cpuid,
12168
12169         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
12170
12171         .write_tsc_offset = vmx_write_tsc_offset,
12172
12173         .set_tdp_cr3 = vmx_set_cr3,
12174
12175         .check_intercept = vmx_check_intercept,
12176         .handle_external_intr = vmx_handle_external_intr,
12177         .mpx_supported = vmx_mpx_supported,
12178         .xsaves_supported = vmx_xsaves_supported,
12179
12180         .check_nested_events = vmx_check_nested_events,
12181
12182         .sched_in = vmx_sched_in,
12183
12184         .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
12185         .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
12186         .flush_log_dirty = vmx_flush_log_dirty,
12187         .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
12188         .write_log_dirty = vmx_write_pml_buffer,
12189
12190         .pre_block = vmx_pre_block,
12191         .post_block = vmx_post_block,
12192
12193         .pmu_ops = &intel_pmu_ops,
12194
12195         .update_pi_irte = vmx_update_pi_irte,
12196
12197 #ifdef CONFIG_X86_64
12198         .set_hv_timer = vmx_set_hv_timer,
12199         .cancel_hv_timer = vmx_cancel_hv_timer,
12200 #endif
12201
12202         .setup_mce = vmx_setup_mce,
12203
12204         .smi_allowed = vmx_smi_allowed,
12205         .pre_enter_smm = vmx_pre_enter_smm,
12206         .pre_leave_smm = vmx_pre_leave_smm,
12207         .enable_smi_window = enable_smi_window,
12208 };
12209
12210 static int __init vmx_init(void)
12211 {
12212         int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
12213                      __alignof__(struct vcpu_vmx), THIS_MODULE);
12214         if (r)
12215                 return r;
12216
12217 #ifdef CONFIG_KEXEC_CORE
12218         rcu_assign_pointer(crash_vmclear_loaded_vmcss,
12219                            crash_vmclear_local_loaded_vmcss);
12220 #endif
12221
12222         return 0;
12223 }
12224
12225 static void __exit vmx_exit(void)
12226 {
12227 #ifdef CONFIG_KEXEC_CORE
12228         RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
12229         synchronize_rcu();
12230 #endif
12231
12232         kvm_exit();
12233 }
12234
12235 module_init(vmx_init)
12236 module_exit(vmx_exit)