arch/x86/kvm/vmx.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "irq.h"
  20 #include "mmu.h"
  21 #include "cpuid.h"
  22 #include "lapic.h"
  23
  24 #include <linux/kvm_host.h>
  25 #include <linux/module.h>
  26 #include <linux/kernel.h>
  27 #include <linux/mm.h>
  28 #include <linux/highmem.h>
  29 #include <linux/sched.h>
  30 #include <linux/moduleparam.h>
  31 #include <linux/mod_devicetable.h>
  32 #include <linux/trace_events.h>
  33 #include <linux/slab.h>
  34 #include <linux/tboot.h>
  35 #include <linux/hrtimer.h>
  36 #include <linux/frame.h>
  37 #include <linux/nospec.h>
  38 #include "kvm_cache_regs.h"
  39 #include "x86.h"
  40
  41 #include <asm/cpu.h>
  42 #include <asm/io.h>
  43 #include <asm/desc.h>
  44 #include <asm/vmx.h>
  45 #include <asm/virtext.h>
  46 #include <asm/mce.h>
  47 #include <asm/fpu/internal.h>
  48 #include <asm/perf_event.h>
  49 #include <asm/debugreg.h>
  50 #include <asm/kexec.h>
  51 #include <asm/apic.h>
  52 #include <asm/irq_remapping.h>
  53 #include <asm/mmu_context.h>
  54 #include <asm/microcode.h>
  55 #include <asm/spec-ctrl.h>
  56
  57 #include "trace.h"
  58 #include "pmu.h"
  59
  60 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  61 #define __ex_clear(x, reg) \
  62         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 static const struct x86_cpu_id vmx_cpu_id[] = {
  68         X86_FEATURE_MATCH(X86_FEATURE_VMX),
  69         {}
  70 };
  71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
  72
  73 static bool __read_mostly enable_vpid = 1;
  74 module_param_named(vpid, enable_vpid, bool, 0444);
  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 MSR_TYPE_R      1
 114 #define MSR_TYPE_W      2
 115 #define MSR_TYPE_RW     3
 116
 117 #define MSR_BITMAP_MODE_X2APIC          1
 118 #define MSR_BITMAP_MODE_X2APIC_APICV    2
 119 #define MSR_BITMAP_MODE_LM              4
 120
 121 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
 122
 123 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
 124 static int __read_mostly cpu_preemption_timer_multi;
 125 static bool __read_mostly enable_preemption_timer = 1;
 126 #ifdef CONFIG_X86_64
 127 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
 128 #endif
 129
 130 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
 131 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
 132 #define KVM_VM_CR0_ALWAYS_ON                                            \
 133         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
 134 #define KVM_CR4_GUEST_OWNED_BITS                                      \
 135         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
 136          | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
 137
 138 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
 139 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
 140
 141 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
 142
 143 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
 144
 145 /*
 146  * Hyper-V requires all of these, so mark them as supported even though
 147  * they are just treated the same as all-context.
 148  */
 149 #define VMX_VPID_EXTENT_SUPPORTED_MASK          \
 150         (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |  \
 151         VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |    \
 152         VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |    \
 153         VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
 154
 155 /*
 156  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 157  * ple_gap:    upper bound on the amount of time between two successive
 158  *             executions of PAUSE in a loop. Also indicate if ple enabled.
 159  *             According to test, this time is usually smaller than 128 cycles.
 160  * ple_window: upper bound on the amount of time a guest is allowed to execute
 161  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
 162  *             less than 2^12 cycles
 163  * Time is measured based on a counter that runs at the same rate as the TSC,
 164  * refer SDM volume 3b section 21.6.13 & 22.1.3.
 165  */
 166 #define KVM_VMX_DEFAULT_PLE_GAP           128
 167 #define KVM_VMX_DEFAULT_PLE_WINDOW        4096
 168 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW   2
 169 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
 170 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX    \
 171                 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
 172
 173 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
 174 module_param(ple_gap, int, S_IRUGO);
 175
 176 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
 177 module_param(ple_window, int, S_IRUGO);
 178
 179 /* Default doubles per-vcpu window every exit. */
 180 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
 181 module_param(ple_window_grow, int, S_IRUGO);
 182
 183 /* Default resets per-vcpu window every exit to ple_window. */
 184 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
 185 module_param(ple_window_shrink, int, S_IRUGO);
 186
 187 /* Default is to compute the maximum so we can never overflow. */
 188 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
 189 static int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
 190 module_param(ple_window_max, int, S_IRUGO);
 191
 192 extern const ulong vmx_return;
 193
 194 #define NR_AUTOLOAD_MSRS 8
 195
 196 struct vmcs {
 197         u32 revision_id;
 198         u32 abort;
 199         char data[0];
 200 };
 201
 202 /*
 203  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
 204  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
 205  * loaded on this CPU (so we can clear them if the CPU goes down).
 206  */
 207 struct loaded_vmcs {
 208         struct vmcs *vmcs;
 209         struct vmcs *shadow_vmcs;
 210         int cpu;
 211         bool launched;
 212         bool nmi_known_unmasked;
 213         unsigned long vmcs_host_cr3;    /* May not match real cr3 */
 214         unsigned long vmcs_host_cr4;    /* May not match real cr4 */
 215         /* Support for vnmi-less CPUs */
 216         int soft_vnmi_blocked;
 217         ktime_t entry_time;
 218         s64 vnmi_blocked_time;
 219         unsigned long *msr_bitmap;
 220         struct list_head loaded_vmcss_on_cpu_link;
 221 };
 222
 223 struct shared_msr_entry {
 224         unsigned index;
 225         u64 data;
 226         u64 mask;
 227 };
 228
 229 /*
 230  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
 231  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
 232  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
 233  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
 234  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
 235  * More than one of these structures may exist, if L1 runs multiple L2 guests.
 236  * nested_vmx_run() will use the data here to build the vmcs02: a VMCS for the
 237  * underlying hardware which will be used to run L2.
 238  * This structure is packed to ensure that its layout is identical across
 239  * machines (necessary for live migration).
 240  * If there are changes in this struct, VMCS12_REVISION must be changed.
 241  */
 242 typedef u64 natural_width;
 243 struct __packed vmcs12 {
 244         /* According to the Intel spec, a VMCS region must start with the
 245          * following two fields. Then follow implementation-specific data.
 246          */
 247         u32 revision_id;
 248         u32 abort;
 249
 250         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
 251         u32 padding[7]; /* room for future expansion */
 252
 253         u64 io_bitmap_a;
 254         u64 io_bitmap_b;
 255         u64 msr_bitmap;
 256         u64 vm_exit_msr_store_addr;
 257         u64 vm_exit_msr_load_addr;
 258         u64 vm_entry_msr_load_addr;
 259         u64 tsc_offset;
 260         u64 virtual_apic_page_addr;
 261         u64 apic_access_addr;
 262         u64 posted_intr_desc_addr;
 263         u64 vm_function_control;
 264         u64 ept_pointer;
 265         u64 eoi_exit_bitmap0;
 266         u64 eoi_exit_bitmap1;
 267         u64 eoi_exit_bitmap2;
 268         u64 eoi_exit_bitmap3;
 269         u64 eptp_list_address;
 270         u64 xss_exit_bitmap;
 271         u64 guest_physical_address;
 272         u64 vmcs_link_pointer;
 273         u64 pml_address;
 274         u64 guest_ia32_debugctl;
 275         u64 guest_ia32_pat;
 276         u64 guest_ia32_efer;
 277         u64 guest_ia32_perf_global_ctrl;
 278         u64 guest_pdptr0;
 279         u64 guest_pdptr1;
 280         u64 guest_pdptr2;
 281         u64 guest_pdptr3;
 282         u64 guest_bndcfgs;
 283         u64 host_ia32_pat;
 284         u64 host_ia32_efer;
 285         u64 host_ia32_perf_global_ctrl;
 286         u64 padding64[8]; /* room for future expansion */
 287         /*
 288          * To allow migration of L1 (complete with its L2 guests) between
 289          * machines of different natural widths (32 or 64 bit), we cannot have
 290          * unsigned long fields with no explict size. We use u64 (aliased
 291          * natural_width) instead. Luckily, x86 is little-endian.
 292          */
 293         natural_width cr0_guest_host_mask;
 294         natural_width cr4_guest_host_mask;
 295         natural_width cr0_read_shadow;
 296         natural_width cr4_read_shadow;
 297         natural_width cr3_target_value0;
 298         natural_width cr3_target_value1;
 299         natural_width cr3_target_value2;
 300         natural_width cr3_target_value3;
 301         natural_width exit_qualification;
 302         natural_width guest_linear_address;
 303         natural_width guest_cr0;
 304         natural_width guest_cr3;
 305         natural_width guest_cr4;
 306         natural_width guest_es_base;
 307         natural_width guest_cs_base;
 308         natural_width guest_ss_base;
 309         natural_width guest_ds_base;
 310         natural_width guest_fs_base;
 311         natural_width guest_gs_base;
 312         natural_width guest_ldtr_base;
 313         natural_width guest_tr_base;
 314         natural_width guest_gdtr_base;
 315         natural_width guest_idtr_base;
 316         natural_width guest_dr7;
 317         natural_width guest_rsp;
 318         natural_width guest_rip;
 319         natural_width guest_rflags;
 320         natural_width guest_pending_dbg_exceptions;
 321         natural_width guest_sysenter_esp;
 322         natural_width guest_sysenter_eip;
 323         natural_width host_cr0;
 324         natural_width host_cr3;
 325         natural_width host_cr4;
 326         natural_width host_fs_base;
 327         natural_width host_gs_base;
 328         natural_width host_tr_base;
 329         natural_width host_gdtr_base;
 330         natural_width host_idtr_base;
 331         natural_width host_ia32_sysenter_esp;
 332         natural_width host_ia32_sysenter_eip;
 333         natural_width host_rsp;
 334         natural_width host_rip;
 335         natural_width paddingl[8]; /* room for future expansion */
 336         u32 pin_based_vm_exec_control;
 337         u32 cpu_based_vm_exec_control;
 338         u32 exception_bitmap;
 339         u32 page_fault_error_code_mask;
 340         u32 page_fault_error_code_match;
 341         u32 cr3_target_count;
 342         u32 vm_exit_controls;
 343         u32 vm_exit_msr_store_count;
 344         u32 vm_exit_msr_load_count;
 345         u32 vm_entry_controls;
 346         u32 vm_entry_msr_load_count;
 347         u32 vm_entry_intr_info_field;
 348         u32 vm_entry_exception_error_code;
 349         u32 vm_entry_instruction_len;
 350         u32 tpr_threshold;
 351         u32 secondary_vm_exec_control;
 352         u32 vm_instruction_error;
 353         u32 vm_exit_reason;
 354         u32 vm_exit_intr_info;
 355         u32 vm_exit_intr_error_code;
 356         u32 idt_vectoring_info_field;
 357         u32 idt_vectoring_error_code;
 358         u32 vm_exit_instruction_len;
 359         u32 vmx_instruction_info;
 360         u32 guest_es_limit;
 361         u32 guest_cs_limit;
 362         u32 guest_ss_limit;
 363         u32 guest_ds_limit;
 364         u32 guest_fs_limit;
 365         u32 guest_gs_limit;
 366         u32 guest_ldtr_limit;
 367         u32 guest_tr_limit;
 368         u32 guest_gdtr_limit;
 369         u32 guest_idtr_limit;
 370         u32 guest_es_ar_bytes;
 371         u32 guest_cs_ar_bytes;
 372         u32 guest_ss_ar_bytes;
 373         u32 guest_ds_ar_bytes;
 374         u32 guest_fs_ar_bytes;
 375         u32 guest_gs_ar_bytes;
 376         u32 guest_ldtr_ar_bytes;
 377         u32 guest_tr_ar_bytes;
 378         u32 guest_interruptibility_info;
 379         u32 guest_activity_state;
 380         u32 guest_sysenter_cs;
 381         u32 host_ia32_sysenter_cs;
 382         u32 vmx_preemption_timer_value;
 383         u32 padding32[7]; /* room for future expansion */
 384         u16 virtual_processor_id;
 385         u16 posted_intr_nv;
 386         u16 guest_es_selector;
 387         u16 guest_cs_selector;
 388         u16 guest_ss_selector;
 389         u16 guest_ds_selector;
 390         u16 guest_fs_selector;
 391         u16 guest_gs_selector;
 392         u16 guest_ldtr_selector;
 393         u16 guest_tr_selector;
 394         u16 guest_intr_status;
 395         u16 guest_pml_index;
 396         u16 host_es_selector;
 397         u16 host_cs_selector;
 398         u16 host_ss_selector;
 399         u16 host_ds_selector;
 400         u16 host_fs_selector;
 401         u16 host_gs_selector;
 402         u16 host_tr_selector;
 403 };
 404
 405 /*
 406  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
 407  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
 408  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
 409  */
 410 #define VMCS12_REVISION 0x11e57ed0
 411
 412 /*
 413  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
 414  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
 415  * current implementation, 4K are reserved to avoid future complications.
 416  */
 417 #define VMCS12_SIZE 0x1000
 418
 419 /*
 420  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
 421  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
 422  */
 423 struct nested_vmx {
 424         /* Has the level1 guest done vmxon? */
 425         bool vmxon;
 426         gpa_t vmxon_ptr;
 427         bool pml_full;
 428
 429         /* The guest-physical address of the current VMCS L1 keeps for L2 */
 430         gpa_t current_vmptr;
 431         /*
 432          * Cache of the guest's VMCS, existing outside of guest memory.
 433          * Loaded from guest memory during VMPTRLD. Flushed to guest
 434          * memory during VMCLEAR and VMPTRLD.
 435          */
 436         struct vmcs12 *cached_vmcs12;
 437         /*
 438          * Indicates if the shadow vmcs must be updated with the
 439          * data hold by vmcs12
 440          */
 441         bool sync_shadow_vmcs;
 442
 443         bool change_vmcs01_virtual_x2apic_mode;
 444         /* L2 must run next, and mustn't decide to exit to L1. */
 445         bool nested_run_pending;
 446
 447         struct loaded_vmcs vmcs02;
 448
 449         /*
 450          * Guest pages referred to in the vmcs02 with host-physical
 451          * pointers, so we must keep them pinned while L2 runs.
 452          */
 453         struct page *apic_access_page;
 454         struct page *virtual_apic_page;
 455         struct page *pi_desc_page;
 456         struct pi_desc *pi_desc;
 457         bool pi_pending;
 458         u16 posted_intr_nv;
 459
 460         struct hrtimer preemption_timer;
 461         bool preemption_timer_expired;
 462
 463         /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
 464         u64 vmcs01_debugctl;
 465
 466         u16 vpid02;
 467         u16 last_vpid;
 468
 469         /*
 470          * We only store the "true" versions of the VMX capability MSRs. We
 471          * generate the "non-true" versions by setting the must-be-1 bits
 472          * according to the SDM.
 473          */
 474         u32 nested_vmx_procbased_ctls_low;
 475         u32 nested_vmx_procbased_ctls_high;
 476         u32 nested_vmx_secondary_ctls_low;
 477         u32 nested_vmx_secondary_ctls_high;
 478         u32 nested_vmx_pinbased_ctls_low;
 479         u32 nested_vmx_pinbased_ctls_high;
 480         u32 nested_vmx_exit_ctls_low;
 481         u32 nested_vmx_exit_ctls_high;
 482         u32 nested_vmx_entry_ctls_low;
 483         u32 nested_vmx_entry_ctls_high;
 484         u32 nested_vmx_misc_low;
 485         u32 nested_vmx_misc_high;
 486         u32 nested_vmx_ept_caps;
 487         u32 nested_vmx_vpid_caps;
 488         u64 nested_vmx_basic;
 489         u64 nested_vmx_cr0_fixed0;
 490         u64 nested_vmx_cr0_fixed1;
 491         u64 nested_vmx_cr4_fixed0;
 492         u64 nested_vmx_cr4_fixed1;
 493         u64 nested_vmx_vmcs_enum;
 494         u64 nested_vmx_vmfunc_controls;
 495 };
 496
 497 #define POSTED_INTR_ON  0
 498 #define POSTED_INTR_SN  1
 499
 500 /* Posted-Interrupt Descriptor */
 501 struct pi_desc {
 502         u32 pir[8];     /* Posted interrupt requested */
 503         union {
 504                 struct {
 505                                 /* bit 256 - Outstanding Notification */
 506                         u16     on      : 1,
 507                                 /* bit 257 - Suppress Notification */
 508                                 sn      : 1,
 509                                 /* bit 271:258 - Reserved */
 510                                 rsvd_1  : 14;
 511                                 /* bit 279:272 - Notification Vector */
 512                         u8      nv;
 513                                 /* bit 287:280 - Reserved */
 514                         u8      rsvd_2;
 515                                 /* bit 319:288 - Notification Destination */
 516                         u32     ndst;
 517                 };
 518                 u64 control;
 519         };
 520         u32 rsvd[6];
 521 } __aligned(64);
 522
 523 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
 524 {
 525         return test_and_set_bit(POSTED_INTR_ON,
 526                         (unsigned long *)&pi_desc->control);
 527 }
 528
 529 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
 530 {
 531         return test_and_clear_bit(POSTED_INTR_ON,
 532                         (unsigned long *)&pi_desc->control);
 533 }
 534
 535 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
 536 {
 537         return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
 538 }
 539
 540 static inline void pi_clear_sn(struct pi_desc *pi_desc)
 541 {
 542         return clear_bit(POSTED_INTR_SN,
 543                         (unsigned long *)&pi_desc->control);
 544 }
 545
 546 static inline void pi_set_sn(struct pi_desc *pi_desc)
 547 {
 548         return set_bit(POSTED_INTR_SN,
 549                         (unsigned long *)&pi_desc->control);
 550 }
 551
 552 static inline void pi_clear_on(struct pi_desc *pi_desc)
 553 {
 554         clear_bit(POSTED_INTR_ON,
 555                   (unsigned long *)&pi_desc->control);
 556 }
 557
 558 static inline int pi_test_on(struct pi_desc *pi_desc)
 559 {
 560         return test_bit(POSTED_INTR_ON,
 561                         (unsigned long *)&pi_desc->control);
 562 }
 563
 564 static inline int pi_test_sn(struct pi_desc *pi_desc)
 565 {
 566         return test_bit(POSTED_INTR_SN,
 567                         (unsigned long *)&pi_desc->control);
 568 }
 569
 570 struct vcpu_vmx {
 571         struct kvm_vcpu       vcpu;
 572         unsigned long         host_rsp;
 573         u8                    fail;
 574         u8                    msr_bitmap_mode;
 575         u32                   exit_intr_info;
 576         u32                   idt_vectoring_info;
 577         ulong                 rflags;
 578         struct shared_msr_entry *guest_msrs;
 579         int                   nmsrs;
 580         int                   save_nmsrs;
 581         unsigned long         host_idt_base;
 582 #ifdef CONFIG_X86_64
 583         u64                   msr_host_kernel_gs_base;
 584         u64                   msr_guest_kernel_gs_base;
 585 #endif
 586
 587         u64                   arch_capabilities;
 588         u64                   spec_ctrl;
 589
 590         u32 vm_entry_controls_shadow;
 591         u32 vm_exit_controls_shadow;
 592         u32 secondary_exec_control;
 593
 594         /*
 595          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
 596          * non-nested (L1) guest, it always points to vmcs01. For a nested
 597          * guest (L2), it points to a different VMCS.
 598          */
 599         struct loaded_vmcs    vmcs01;
 600         struct loaded_vmcs   *loaded_vmcs;
 601         bool                  __launched; /* temporary, used in vmx_vcpu_run */
 602         struct msr_autoload {
 603                 unsigned nr;
 604                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
 605                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
 606         } msr_autoload;
 607         struct {
 608                 int           loaded;
 609                 u16           fs_sel, gs_sel, ldt_sel;
 610 #ifdef CONFIG_X86_64
 611                 u16           ds_sel, es_sel;
 612 #endif
 613                 int           gs_ldt_reload_needed;
 614                 int           fs_reload_needed;
 615                 u64           msr_host_bndcfgs;
 616         } host_state;
 617         struct {
 618                 int vm86_active;
 619                 ulong save_rflags;
 620                 struct kvm_segment segs[8];
 621         } rmode;
 622         struct {
 623                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
 624                 struct kvm_save_segment {
 625                         u16 selector;
 626                         unsigned long base;
 627                         u32 limit;
 628                         u32 ar;
 629                 } seg[8];
 630         } segment_cache;
 631         int vpid;
 632         bool emulation_required;
 633
 634         u32 exit_reason;
 635
 636         /* Posted interrupt descriptor */
 637         struct pi_desc pi_desc;
 638
 639         /* Support for a guest hypervisor (nested VMX) */
 640         struct nested_vmx nested;
 641
 642         /* Dynamic PLE window. */
 643         int ple_window;
 644         bool ple_window_dirty;
 645
 646         /* Support for PML */
 647 #define PML_ENTITY_NUM          512
 648         struct page *pml_pg;
 649
 650         /* apic deadline value in host tsc */
 651         u64 hv_deadline_tsc;
 652
 653         u64 current_tsc_ratio;
 654
 655         u32 host_pkru;
 656
 657         /*
 658          * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
 659          * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
 660          * in msr_ia32_feature_control_valid_bits.
 661          */
 662         u64 msr_ia32_feature_control;
 663         u64 msr_ia32_feature_control_valid_bits;
 664 };
 665
 666 enum segment_cache_field {
 667         SEG_FIELD_SEL = 0,
 668         SEG_FIELD_BASE = 1,
 669         SEG_FIELD_LIMIT = 2,
 670         SEG_FIELD_AR = 3,
 671
 672         SEG_FIELD_NR = 4
 673 };
 674
 675 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
 676 {
 677         return container_of(vcpu, struct vcpu_vmx, vcpu);
 678 }
 679
 680 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
 681 {
 682         return &(to_vmx(vcpu)->pi_desc);
 683 }
 684
 685 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
 686 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
 687 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
 688                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
 689
 690
 691 static unsigned long shadow_read_only_fields[] = {
 692         /*
 693          * We do NOT shadow fields that are modified when L0
 694          * traps and emulates any vmx instruction (e.g. VMPTRLD,
 695          * VMXON...) executed by L1.
 696          * For example, VM_INSTRUCTION_ERROR is read
 697          * by L1 if a vmx instruction fails (part of the error path).
 698          * Note the code assumes this logic. If for some reason
 699          * we start shadowing these fields then we need to
 700          * force a shadow sync when L0 emulates vmx instructions
 701          * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
 702          * by nested_vmx_failValid)
 703          */
 704         VM_EXIT_REASON,
 705         VM_EXIT_INTR_INFO,
 706         VM_EXIT_INSTRUCTION_LEN,
 707         IDT_VECTORING_INFO_FIELD,
 708         IDT_VECTORING_ERROR_CODE,
 709         VM_EXIT_INTR_ERROR_CODE,
 710         EXIT_QUALIFICATION,
 711         GUEST_LINEAR_ADDRESS,
 712         GUEST_PHYSICAL_ADDRESS
 713 };
 714 static int max_shadow_read_only_fields =
 715         ARRAY_SIZE(shadow_read_only_fields);
 716
 717 static unsigned long shadow_read_write_fields[] = {
 718         TPR_THRESHOLD,
 719         GUEST_RIP,
 720         GUEST_RSP,
 721         GUEST_CR0,
 722         GUEST_CR3,
 723         GUEST_CR4,
 724         GUEST_INTERRUPTIBILITY_INFO,
 725         GUEST_RFLAGS,
 726         GUEST_CS_SELECTOR,
 727         GUEST_CS_AR_BYTES,
 728         GUEST_CS_LIMIT,
 729         GUEST_CS_BASE,
 730         GUEST_ES_BASE,
 731         GUEST_BNDCFGS,
 732         CR0_GUEST_HOST_MASK,
 733         CR0_READ_SHADOW,
 734         CR4_READ_SHADOW,
 735         TSC_OFFSET,
 736         EXCEPTION_BITMAP,
 737         CPU_BASED_VM_EXEC_CONTROL,
 738         VM_ENTRY_EXCEPTION_ERROR_CODE,
 739         VM_ENTRY_INTR_INFO_FIELD,
 740         VM_ENTRY_INSTRUCTION_LEN,
 741         VM_ENTRY_EXCEPTION_ERROR_CODE,
 742         HOST_FS_BASE,
 743         HOST_GS_BASE,
 744         HOST_FS_SELECTOR,
 745         HOST_GS_SELECTOR
 746 };
 747 static int max_shadow_read_write_fields =
 748         ARRAY_SIZE(shadow_read_write_fields);
 749
 750 static const unsigned short vmcs_field_to_offset_table[] = {
 751         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 752         FIELD(POSTED_INTR_NV, posted_intr_nv),
 753         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
 754         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
 755         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
 756         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
 757         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
 758         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
 759         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
 760         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
 761         FIELD(GUEST_INTR_STATUS, guest_intr_status),
 762         FIELD(GUEST_PML_INDEX, guest_pml_index),
 763         FIELD(HOST_ES_SELECTOR, host_es_selector),
 764         FIELD(HOST_CS_SELECTOR, host_cs_selector),
 765         FIELD(HOST_SS_SELECTOR, host_ss_selector),
 766         FIELD(HOST_DS_SELECTOR, host_ds_selector),
 767         FIELD(HOST_FS_SELECTOR, host_fs_selector),
 768         FIELD(HOST_GS_SELECTOR, host_gs_selector),
 769         FIELD(HOST_TR_SELECTOR, host_tr_selector),
 770         FIELD64(IO_BITMAP_A, io_bitmap_a),
 771         FIELD64(IO_BITMAP_B, io_bitmap_b),
 772         FIELD64(MSR_BITMAP, msr_bitmap),
 773         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
 774         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
 775         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
 776         FIELD64(TSC_OFFSET, tsc_offset),
 777         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
 778         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
 779         FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
 780         FIELD64(VM_FUNCTION_CONTROL, vm_function_control),
 781         FIELD64(EPT_POINTER, ept_pointer),
 782         FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
 783         FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
 784         FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
 785         FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
 786         FIELD64(EPTP_LIST_ADDRESS, eptp_list_address),
 787         FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
 788         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
 789         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
 790         FIELD64(PML_ADDRESS, pml_address),
 791         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
 792         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
 793         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
 794         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
 795         FIELD64(GUEST_PDPTR0, guest_pdptr0),
 796         FIELD64(GUEST_PDPTR1, guest_pdptr1),
 797         FIELD64(GUEST_PDPTR2, guest_pdptr2),
 798         FIELD64(GUEST_PDPTR3, guest_pdptr3),
 799         FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
 800         FIELD64(HOST_IA32_PAT, host_ia32_pat),
 801         FIELD64(HOST_IA32_EFER, host_ia32_efer),
 802         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
 803         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
 804         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
 805         FIELD(EXCEPTION_BITMAP, exception_bitmap),
 806         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
 807         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
 808         FIELD(CR3_TARGET_COUNT, cr3_target_count),
 809         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
 810         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
 811         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
 812         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
 813         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
 814         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
 815         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
 816         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
 817         FIELD(TPR_THRESHOLD, tpr_threshold),
 818         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
 819         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
 820         FIELD(VM_EXIT_REASON, vm_exit_reason),
 821         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
 822         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
 823         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
 824         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
 825         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
 826         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
 827         FIELD(GUEST_ES_LIMIT, guest_es_limit),
 828         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
 829         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
 830         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
 831         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
 832         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
 833         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
 834         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
 835         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
 836         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
 837         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
 838         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
 839         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
 840         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
 841         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
 842         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
 843         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
 844         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
 845         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
 846         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
 847         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
 848         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
 849         FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
 850         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
 851         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
 852         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
 853         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
 854         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
 855         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
 856         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
 857         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
 858         FIELD(EXIT_QUALIFICATION, exit_qualification),
 859         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
 860         FIELD(GUEST_CR0, guest_cr0),
 861         FIELD(GUEST_CR3, guest_cr3),
 862         FIELD(GUEST_CR4, guest_cr4),
 863         FIELD(GUEST_ES_BASE, guest_es_base),
 864         FIELD(GUEST_CS_BASE, guest_cs_base),
 865         FIELD(GUEST_SS_BASE, guest_ss_base),
 866         FIELD(GUEST_DS_BASE, guest_ds_base),
 867         FIELD(GUEST_FS_BASE, guest_fs_base),
 868         FIELD(GUEST_GS_BASE, guest_gs_base),
 869         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
 870         FIELD(GUEST_TR_BASE, guest_tr_base),
 871         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
 872         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
 873         FIELD(GUEST_DR7, guest_dr7),
 874         FIELD(GUEST_RSP, guest_rsp),
 875         FIELD(GUEST_RIP, guest_rip),
 876         FIELD(GUEST_RFLAGS, guest_rflags),
 877         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
 878         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
 879         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
 880         FIELD(HOST_CR0, host_cr0),
 881         FIELD(HOST_CR3, host_cr3),
 882         FIELD(HOST_CR4, host_cr4),
 883         FIELD(HOST_FS_BASE, host_fs_base),
 884         FIELD(HOST_GS_BASE, host_gs_base),
 885         FIELD(HOST_TR_BASE, host_tr_base),
 886         FIELD(HOST_GDTR_BASE, host_gdtr_base),
 887         FIELD(HOST_IDTR_BASE, host_idtr_base),
 888         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
 889         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
 890         FIELD(HOST_RSP, host_rsp),
 891         FIELD(HOST_RIP, host_rip),
 892 };
 893
 894 static inline short vmcs_field_to_offset(unsigned long field)
 895 {
 896         const size_t size = ARRAY_SIZE(vmcs_field_to_offset_table);
 897         unsigned short offset;
 898
 899         BUILD_BUG_ON(size > SHRT_MAX);
 900         if (field >= size)
 901                 return -ENOENT;
 902
 903         field = array_index_nospec(field, size);
 904         offset = vmcs_field_to_offset_table[field];
 905         if (offset == 0)
 906                 return -ENOENT;
 907         return offset;
 908 }
 909
 910 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
 911 {
 912         return to_vmx(vcpu)->nested.cached_vmcs12;
 913 }
 914
 915 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
 916 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
 917 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
 918 static bool vmx_xsaves_supported(void);
 919 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
 920 static void vmx_set_segment(struct kvm_vcpu *vcpu,
 921                             struct kvm_segment *var, int seg);
 922 static void vmx_get_segment(struct kvm_vcpu *vcpu,
 923                             struct kvm_segment *var, int seg);
 924 static bool guest_state_valid(struct kvm_vcpu *vcpu);
 925 static u32 vmx_segment_access_rights(struct kvm_segment *var);
 926 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
 927 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
 928 static int alloc_identity_pagetable(struct kvm *kvm);
 929 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
 930 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
 931 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
 932                                             u16 error_code);
 933 static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
 934 static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
 935                                                           u32 msr, int type);
 936
 937 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
 938 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
 939 /*
 940  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
 941  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
 942  */
 943 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
 944
 945 /*
 946  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
 947  * can find which vCPU should be waken up.
 948  */
 949 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
 950 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
 951
 952 enum {
 953         VMX_IO_BITMAP_A,
 954         VMX_IO_BITMAP_B,
 955         VMX_VMREAD_BITMAP,
 956         VMX_VMWRITE_BITMAP,
 957         VMX_BITMAP_NR
 958 };
 959
 960 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
 961
 962 #define vmx_io_bitmap_a                      (vmx_bitmap[VMX_IO_BITMAP_A])
 963 #define vmx_io_bitmap_b                      (vmx_bitmap[VMX_IO_BITMAP_B])
 964 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
 965 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
 966
 967 static bool cpu_has_load_ia32_efer;
 968 static bool cpu_has_load_perf_global_ctrl;
 969
 970 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
 971 static DEFINE_SPINLOCK(vmx_vpid_lock);
 972
 973 static struct vmcs_config {
 974         int size;
 975         int order;
 976         u32 basic_cap;
 977         u32 revision_id;
 978         u32 pin_based_exec_ctrl;
 979         u32 cpu_based_exec_ctrl;
 980         u32 cpu_based_2nd_exec_ctrl;
 981         u32 vmexit_ctrl;
 982         u32 vmentry_ctrl;
 983 } vmcs_config;
 984
 985 static struct vmx_capability {
 986         u32 ept;
 987         u32 vpid;
 988 } vmx_capability;
 989
 990 #define VMX_SEGMENT_FIELD(seg)                                  \
 991         [VCPU_SREG_##seg] = {                                   \
 992                 .selector = GUEST_##seg##_SELECTOR,             \
 993                 .base = GUEST_##seg##_BASE,                     \
 994                 .limit = GUEST_##seg##_LIMIT,                   \
 995                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
 996         }
 997
 998 static const struct kvm_vmx_segment_field {
 999         unsigned selector;
1000         unsigned base;
1001         unsigned limit;
1002         unsigned ar_bytes;
1003 } kvm_vmx_segment_fields[] = {
1004         VMX_SEGMENT_FIELD(CS),
1005         VMX_SEGMENT_FIELD(DS),
1006         VMX_SEGMENT_FIELD(ES),
1007         VMX_SEGMENT_FIELD(FS),
1008         VMX_SEGMENT_FIELD(GS),
1009         VMX_SEGMENT_FIELD(SS),
1010         VMX_SEGMENT_FIELD(TR),
1011         VMX_SEGMENT_FIELD(LDTR),
1012 };
1013
1014 static u64 host_efer;
1015
1016 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1017
1018 /*
1019  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1020  * away by decrementing the array size.
1021  */
1022 static const u32 vmx_msr_index[] = {
1023 #ifdef CONFIG_X86_64
1024         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1025 #endif
1026         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1027 };
1028
1029 static inline bool is_exception_n(u32 intr_info, u8 vector)
1030 {
1031         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1032                              INTR_INFO_VALID_MASK)) ==
1033                 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1034 }
1035
1036 static inline bool is_debug(u32 intr_info)
1037 {
1038         return is_exception_n(intr_info, DB_VECTOR);
1039 }
1040
1041 static inline bool is_breakpoint(u32 intr_info)
1042 {
1043         return is_exception_n(intr_info, BP_VECTOR);
1044 }
1045
1046 static inline bool is_page_fault(u32 intr_info)
1047 {
1048         return is_exception_n(intr_info, PF_VECTOR);
1049 }
1050
1051 static inline bool is_no_device(u32 intr_info)
1052 {
1053         return is_exception_n(intr_info, NM_VECTOR);
1054 }
1055
1056 static inline bool is_invalid_opcode(u32 intr_info)
1057 {
1058         return is_exception_n(intr_info, UD_VECTOR);
1059 }
1060
1061 static inline bool is_external_interrupt(u32 intr_info)
1062 {
1063         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1064                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1065 }
1066
1067 static inline bool is_machine_check(u32 intr_info)
1068 {
1069         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1070                              INTR_INFO_VALID_MASK)) ==
1071                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1072 }
1073
1074 /* Undocumented: icebp/int1 */
1075 static inline bool is_icebp(u32 intr_info)
1076 {
1077         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1078                 == (INTR_TYPE_PRIV_SW_EXCEPTION | INTR_INFO_VALID_MASK);
1079 }
1080
1081 static inline bool cpu_has_vmx_msr_bitmap(void)
1082 {
1083         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1084 }
1085
1086 static inline bool cpu_has_vmx_tpr_shadow(void)
1087 {
1088         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1089 }
1090
1091 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1092 {
1093         return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1094 }
1095
1096 static inline bool cpu_has_secondary_exec_ctrls(void)
1097 {
1098         return vmcs_config.cpu_based_exec_ctrl &
1099                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1100 }
1101
1102 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1103 {
1104         return vmcs_config.cpu_based_2nd_exec_ctrl &
1105                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1106 }
1107
1108 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1109 {
1110         return vmcs_config.cpu_based_2nd_exec_ctrl &
1111                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1112 }
1113
1114 static inline bool cpu_has_vmx_apic_register_virt(void)
1115 {
1116         return vmcs_config.cpu_based_2nd_exec_ctrl &
1117                 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1118 }
1119
1120 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1121 {
1122         return vmcs_config.cpu_based_2nd_exec_ctrl &
1123                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1124 }
1125
1126 /*
1127  * Comment's format: document - errata name - stepping - processor name.
1128  * Refer from
1129  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1130  */
1131 static u32 vmx_preemption_cpu_tfms[] = {
1132 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
1133 0x000206E6,
1134 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
1135 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1136 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1137 0x00020652,
1138 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1139 0x00020655,
1140 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
1141 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
1142 /*
1143  * 320767.pdf - AAP86  - B1 -
1144  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1145  */
1146 0x000106E5,
1147 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1148 0x000106A0,
1149 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1150 0x000106A1,
1151 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1152 0x000106A4,
1153  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1154  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1155  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1156 0x000106A5,
1157 };
1158
1159 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1160 {
1161         u32 eax = cpuid_eax(0x00000001), i;
1162
1163         /* Clear the reserved bits */
1164         eax &= ~(0x3U << 14 | 0xfU << 28);
1165         for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1166                 if (eax == vmx_preemption_cpu_tfms[i])
1167                         return true;
1168
1169         return false;
1170 }
1171
1172 static inline bool cpu_has_vmx_preemption_timer(void)
1173 {
1174         return vmcs_config.pin_based_exec_ctrl &
1175                 PIN_BASED_VMX_PREEMPTION_TIMER;
1176 }
1177
1178 static inline bool cpu_has_vmx_posted_intr(void)
1179 {
1180         return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1181                 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1182 }
1183
1184 static inline bool cpu_has_vmx_apicv(void)
1185 {
1186         return cpu_has_vmx_apic_register_virt() &&
1187                 cpu_has_vmx_virtual_intr_delivery() &&
1188                 cpu_has_vmx_posted_intr();
1189 }
1190
1191 static inline bool cpu_has_vmx_flexpriority(void)
1192 {
1193         return cpu_has_vmx_tpr_shadow() &&
1194                 cpu_has_vmx_virtualize_apic_accesses();
1195 }
1196
1197 static inline bool cpu_has_vmx_ept_execute_only(void)
1198 {
1199         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1200 }
1201
1202 static inline bool cpu_has_vmx_ept_2m_page(void)
1203 {
1204         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1205 }
1206
1207 static inline bool cpu_has_vmx_ept_1g_page(void)
1208 {
1209         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1210 }
1211
1212 static inline bool cpu_has_vmx_ept_4levels(void)
1213 {
1214         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1215 }
1216
1217 static inline bool cpu_has_vmx_ept_mt_wb(void)
1218 {
1219         return vmx_capability.ept & VMX_EPTP_WB_BIT;
1220 }
1221
1222 static inline bool cpu_has_vmx_ept_5levels(void)
1223 {
1224         return vmx_capability.ept & VMX_EPT_PAGE_WALK_5_BIT;
1225 }
1226
1227 static inline bool cpu_has_vmx_ept_ad_bits(void)
1228 {
1229         return vmx_capability.ept & VMX_EPT_AD_BIT;
1230 }
1231
1232 static inline bool cpu_has_vmx_invept_context(void)
1233 {
1234         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1235 }
1236
1237 static inline bool cpu_has_vmx_invept_global(void)
1238 {
1239         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1240 }
1241
1242 static inline bool cpu_has_vmx_invvpid_single(void)
1243 {
1244         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1245 }
1246
1247 static inline bool cpu_has_vmx_invvpid_global(void)
1248 {
1249         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1250 }
1251
1252 static inline bool cpu_has_vmx_invvpid(void)
1253 {
1254         return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1255 }
1256
1257 static inline bool cpu_has_vmx_ept(void)
1258 {
1259         return vmcs_config.cpu_based_2nd_exec_ctrl &
1260                 SECONDARY_EXEC_ENABLE_EPT;
1261 }
1262
1263 static inline bool cpu_has_vmx_unrestricted_guest(void)
1264 {
1265         return vmcs_config.cpu_based_2nd_exec_ctrl &
1266                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1267 }
1268
1269 static inline bool cpu_has_vmx_ple(void)
1270 {
1271         return vmcs_config.cpu_based_2nd_exec_ctrl &
1272                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1273 }
1274
1275 static inline bool cpu_has_vmx_basic_inout(void)
1276 {
1277         return  (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1278 }
1279
1280 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1281 {
1282         return flexpriority_enabled && lapic_in_kernel(vcpu);
1283 }
1284
1285 static inline bool cpu_has_vmx_vpid(void)
1286 {
1287         return vmcs_config.cpu_based_2nd_exec_ctrl &
1288                 SECONDARY_EXEC_ENABLE_VPID;
1289 }
1290
1291 static inline bool cpu_has_vmx_rdtscp(void)
1292 {
1293         return vmcs_config.cpu_based_2nd_exec_ctrl &
1294                 SECONDARY_EXEC_RDTSCP;
1295 }
1296
1297 static inline bool cpu_has_vmx_invpcid(void)
1298 {
1299         return vmcs_config.cpu_based_2nd_exec_ctrl &
1300                 SECONDARY_EXEC_ENABLE_INVPCID;
1301 }
1302
1303 static inline bool cpu_has_virtual_nmis(void)
1304 {
1305         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1306 }
1307
1308 static inline bool cpu_has_vmx_wbinvd_exit(void)
1309 {
1310         return vmcs_config.cpu_based_2nd_exec_ctrl &
1311                 SECONDARY_EXEC_WBINVD_EXITING;
1312 }
1313
1314 static inline bool cpu_has_vmx_shadow_vmcs(void)
1315 {
1316         u64 vmx_msr;
1317         rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1318         /* check if the cpu supports writing r/o exit information fields */
1319         if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1320                 return false;
1321
1322         return vmcs_config.cpu_based_2nd_exec_ctrl &
1323                 SECONDARY_EXEC_SHADOW_VMCS;
1324 }
1325
1326 static inline bool cpu_has_vmx_pml(void)
1327 {
1328         return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1329 }
1330
1331 static inline bool cpu_has_vmx_tsc_scaling(void)
1332 {
1333         return vmcs_config.cpu_based_2nd_exec_ctrl &
1334                 SECONDARY_EXEC_TSC_SCALING;
1335 }
1336
1337 static inline bool cpu_has_vmx_vmfunc(void)
1338 {
1339         return vmcs_config.cpu_based_2nd_exec_ctrl &
1340                 SECONDARY_EXEC_ENABLE_VMFUNC;
1341 }
1342
1343 static inline bool report_flexpriority(void)
1344 {
1345         return flexpriority_enabled;
1346 }
1347
1348 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1349 {
1350         return vmx_misc_cr3_count(to_vmx(vcpu)->nested.nested_vmx_misc_low);
1351 }
1352
1353 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1354 {
1355         return vmcs12->cpu_based_vm_exec_control & bit;
1356 }
1357
1358 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1359 {
1360         return (vmcs12->cpu_based_vm_exec_control &
1361                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1362                 (vmcs12->secondary_vm_exec_control & bit);
1363 }
1364
1365 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1366 {
1367         return vmcs12->pin_based_vm_exec_control &
1368                 PIN_BASED_VMX_PREEMPTION_TIMER;
1369 }
1370
1371 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1372 {
1373         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1374 }
1375
1376 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1377 {
1378         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
1379 }
1380
1381 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1382 {
1383         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1384 }
1385
1386 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1387 {
1388         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1389 }
1390
1391 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1392 {
1393         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1394 }
1395
1396 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1397 {
1398         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1399 }
1400
1401 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1402 {
1403         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1404 }
1405
1406 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1407 {
1408         return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1409 }
1410
1411 static inline bool nested_cpu_has_vmfunc(struct vmcs12 *vmcs12)
1412 {
1413         return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VMFUNC);
1414 }
1415
1416 static inline bool nested_cpu_has_eptp_switching(struct vmcs12 *vmcs12)
1417 {
1418         return nested_cpu_has_vmfunc(vmcs12) &&
1419                 (vmcs12->vm_function_control &
1420                  VMX_VMFUNC_EPTP_SWITCHING);
1421 }
1422
1423 static inline bool is_nmi(u32 intr_info)
1424 {
1425         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1426                 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1427 }
1428
1429 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1430                               u32 exit_intr_info,
1431                               unsigned long exit_qualification);
1432 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1433                         struct vmcs12 *vmcs12,
1434                         u32 reason, unsigned long qualification);
1435
1436 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1437 {
1438         int i;
1439
1440         for (i = 0; i < vmx->nmsrs; ++i)
1441                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1442                         return i;
1443         return -1;
1444 }
1445
1446 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1447 {
1448     struct {
1449         u64 vpid : 16;
1450         u64 rsvd : 48;
1451         u64 gva;
1452     } operand = { vpid, 0, gva };
1453
1454     asm volatile (__ex(ASM_VMX_INVVPID)
1455                   /* CF==1 or ZF==1 --> rc = -1 */
1456                   "; ja 1f ; ud2 ; 1:"
1457                   : : "a"(&operand), "c"(ext) : "cc", "memory");
1458 }
1459
1460 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1461 {
1462         struct {
1463                 u64 eptp, gpa;
1464         } operand = {eptp, gpa};
1465
1466         asm volatile (__ex(ASM_VMX_INVEPT)
1467                         /* CF==1 or ZF==1 --> rc = -1 */
1468                         "; ja 1f ; ud2 ; 1:\n"
1469                         : : "a" (&operand), "c" (ext) : "cc", "memory");
1470 }
1471
1472 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1473 {
1474         int i;
1475
1476         i = __find_msr_index(vmx, msr);
1477         if (i >= 0)
1478                 return &vmx->guest_msrs[i];
1479         return NULL;
1480 }
1481
1482 static void vmcs_clear(struct vmcs *vmcs)
1483 {
1484         u64 phys_addr = __pa(vmcs);
1485         u8 error;
1486
1487         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1488                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1489                       : "cc", "memory");
1490         if (error)
1491                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1492                        vmcs, phys_addr);
1493 }
1494
1495 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1496 {
1497         vmcs_clear(loaded_vmcs->vmcs);
1498         if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1499                 vmcs_clear(loaded_vmcs->shadow_vmcs);
1500         loaded_vmcs->cpu = -1;
1501         loaded_vmcs->launched = 0;
1502 }
1503
1504 static void vmcs_load(struct vmcs *vmcs)
1505 {
1506         u64 phys_addr = __pa(vmcs);
1507         u8 error;
1508
1509         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1510                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1511                         : "cc", "memory");
1512         if (error)
1513                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1514                        vmcs, phys_addr);
1515 }
1516
1517 #ifdef CONFIG_KEXEC_CORE
1518 /*
1519  * This bitmap is used to indicate whether the vmclear
1520  * operation is enabled on all cpus. All disabled by
1521  * default.
1522  */
1523 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1524
1525 static inline void crash_enable_local_vmclear(int cpu)
1526 {
1527         cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1528 }
1529
1530 static inline void crash_disable_local_vmclear(int cpu)
1531 {
1532         cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1533 }
1534
1535 static inline int crash_local_vmclear_enabled(int cpu)
1536 {
1537         return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1538 }
1539
1540 static void crash_vmclear_local_loaded_vmcss(void)
1541 {
1542         int cpu = raw_smp_processor_id();
1543         struct loaded_vmcs *v;
1544
1545         if (!crash_local_vmclear_enabled(cpu))
1546                 return;
1547
1548         list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1549                             loaded_vmcss_on_cpu_link)
1550                 vmcs_clear(v->vmcs);
1551 }
1552 #else
1553 static inline void crash_enable_local_vmclear(int cpu) { }
1554 static inline void crash_disable_local_vmclear(int cpu) { }
1555 #endif /* CONFIG_KEXEC_CORE */
1556
1557 static void __loaded_vmcs_clear(void *arg)
1558 {
1559         struct loaded_vmcs *loaded_vmcs = arg;
1560         int cpu = raw_smp_processor_id();
1561
1562         if (loaded_vmcs->cpu != cpu)
1563                 return; /* vcpu migration can race with cpu offline */
1564         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1565                 per_cpu(current_vmcs, cpu) = NULL;
1566         crash_disable_local_vmclear(cpu);
1567         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1568
1569         /*
1570          * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1571          * is before setting loaded_vmcs->vcpu to -1 which is done in
1572          * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1573          * then adds the vmcs into percpu list before it is deleted.
1574          */
1575         smp_wmb();
1576
1577         loaded_vmcs_init(loaded_vmcs);
1578         crash_enable_local_vmclear(cpu);
1579 }
1580
1581 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1582 {
1583         int cpu = loaded_vmcs->cpu;
1584
1585         if (cpu != -1)
1586                 smp_call_function_single(cpu,
1587                          __loaded_vmcs_clear, loaded_vmcs, 1);
1588 }
1589
1590 static inline void vpid_sync_vcpu_single(int vpid)
1591 {
1592         if (vpid == 0)
1593                 return;
1594
1595         if (cpu_has_vmx_invvpid_single())
1596                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1597 }
1598
1599 static inline void vpid_sync_vcpu_global(void)
1600 {
1601         if (cpu_has_vmx_invvpid_global())
1602                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1603 }
1604
1605 static inline void vpid_sync_context(int vpid)
1606 {
1607         if (cpu_has_vmx_invvpid_single())
1608                 vpid_sync_vcpu_single(vpid);
1609         else
1610                 vpid_sync_vcpu_global();
1611 }
1612
1613 static inline void ept_sync_global(void)
1614 {
1615         if (cpu_has_vmx_invept_global())
1616                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1617 }
1618
1619 static inline void ept_sync_context(u64 eptp)
1620 {
1621         if (enable_ept) {
1622                 if (cpu_has_vmx_invept_context())
1623                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1624                 else
1625                         ept_sync_global();
1626         }
1627 }
1628
1629 static __always_inline void vmcs_check16(unsigned long field)
1630 {
1631         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1632                          "16-bit accessor invalid for 64-bit field");
1633         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1634                          "16-bit accessor invalid for 64-bit high field");
1635         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1636                          "16-bit accessor invalid for 32-bit high field");
1637         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1638                          "16-bit accessor invalid for natural width field");
1639 }
1640
1641 static __always_inline void vmcs_check32(unsigned long field)
1642 {
1643         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1644                          "32-bit accessor invalid for 16-bit field");
1645         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1646                          "32-bit accessor invalid for natural width field");
1647 }
1648
1649 static __always_inline void vmcs_check64(unsigned long field)
1650 {
1651         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1652                          "64-bit accessor invalid for 16-bit field");
1653         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1654                          "64-bit accessor invalid for 64-bit high field");
1655         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1656                          "64-bit accessor invalid for 32-bit field");
1657         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1658                          "64-bit accessor invalid for natural width field");
1659 }
1660
1661 static __always_inline void vmcs_checkl(unsigned long field)
1662 {
1663         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1664                          "Natural width accessor invalid for 16-bit field");
1665         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1666                          "Natural width accessor invalid for 64-bit field");
1667         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1668                          "Natural width accessor invalid for 64-bit high field");
1669         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1670                          "Natural width accessor invalid for 32-bit field");
1671 }
1672
1673 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1674 {
1675         unsigned long value;
1676
1677         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1678                       : "=a"(value) : "d"(field) : "cc");
1679         return value;
1680 }
1681
1682 static __always_inline u16 vmcs_read16(unsigned long field)
1683 {
1684         vmcs_check16(field);
1685         return __vmcs_readl(field);
1686 }
1687
1688 static __always_inline u32 vmcs_read32(unsigned long field)
1689 {
1690         vmcs_check32(field);
1691         return __vmcs_readl(field);
1692 }
1693
1694 static __always_inline u64 vmcs_read64(unsigned long field)
1695 {
1696         vmcs_check64(field);
1697 #ifdef CONFIG_X86_64
1698         return __vmcs_readl(field);
1699 #else
1700         return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1701 #endif
1702 }
1703
1704 static __always_inline unsigned long vmcs_readl(unsigned long field)
1705 {
1706         vmcs_checkl(field);
1707         return __vmcs_readl(field);
1708 }
1709
1710 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1711 {
1712         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1713                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1714         dump_stack();
1715 }
1716
1717 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1718 {
1719         u8 error;
1720
1721         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1722                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1723         if (unlikely(error))
1724                 vmwrite_error(field, value);
1725 }
1726
1727 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1728 {
1729         vmcs_check16(field);
1730         __vmcs_writel(field, value);
1731 }
1732
1733 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1734 {
1735         vmcs_check32(field);
1736         __vmcs_writel(field, value);
1737 }
1738
1739 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1740 {
1741         vmcs_check64(field);
1742         __vmcs_writel(field, value);
1743 #ifndef CONFIG_X86_64
1744         asm volatile ("");
1745         __vmcs_writel(field+1, value >> 32);
1746 #endif
1747 }
1748
1749 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1750 {
1751         vmcs_checkl(field);
1752         __vmcs_writel(field, value);
1753 }
1754
1755 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1756 {
1757         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1758                          "vmcs_clear_bits does not support 64-bit fields");
1759         __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1760 }
1761
1762 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1763 {
1764         BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1765                          "vmcs_set_bits does not support 64-bit fields");
1766         __vmcs_writel(field, __vmcs_readl(field) | mask);
1767 }
1768
1769 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1770 {
1771         vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1772 }
1773
1774 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1775 {
1776         vmcs_write32(VM_ENTRY_CONTROLS, val);
1777         vmx->vm_entry_controls_shadow = val;
1778 }
1779
1780 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1781 {
1782         if (vmx->vm_entry_controls_shadow != val)
1783                 vm_entry_controls_init(vmx, val);
1784 }
1785
1786 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1787 {
1788         return vmx->vm_entry_controls_shadow;
1789 }
1790
1791
1792 static inline void vm_entry_controls_setbit(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_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1798 {
1799         vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1800 }
1801
1802 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1803 {
1804         vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1805 }
1806
1807 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1808 {
1809         vmcs_write32(VM_EXIT_CONTROLS, val);
1810         vmx->vm_exit_controls_shadow = val;
1811 }
1812
1813 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1814 {
1815         if (vmx->vm_exit_controls_shadow != val)
1816                 vm_exit_controls_init(vmx, val);
1817 }
1818
1819 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1820 {
1821         return vmx->vm_exit_controls_shadow;
1822 }
1823
1824
1825 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1826 {
1827         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1828 }
1829
1830 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1831 {
1832         vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1833 }
1834
1835 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1836 {
1837         vmx->segment_cache.bitmask = 0;
1838 }
1839
1840 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1841                                        unsigned field)
1842 {
1843         bool ret;
1844         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1845
1846         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1847                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1848                 vmx->segment_cache.bitmask = 0;
1849         }
1850         ret = vmx->segment_cache.bitmask & mask;
1851         vmx->segment_cache.bitmask |= mask;
1852         return ret;
1853 }
1854
1855 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1856 {
1857         u16 *p = &vmx->segment_cache.seg[seg].selector;
1858
1859         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1860                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1861         return *p;
1862 }
1863
1864 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1865 {
1866         ulong *p = &vmx->segment_cache.seg[seg].base;
1867
1868         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1869                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1870         return *p;
1871 }
1872
1873 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1874 {
1875         u32 *p = &vmx->segment_cache.seg[seg].limit;
1876
1877         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1878                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1879         return *p;
1880 }
1881
1882 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1883 {
1884         u32 *p = &vmx->segment_cache.seg[seg].ar;
1885
1886         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1887                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1888         return *p;
1889 }
1890
1891 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1892 {
1893         u32 eb;
1894
1895         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1896              (1u << DB_VECTOR) | (1u << AC_VECTOR);
1897         if ((vcpu->guest_debug &
1898              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1899             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1900                 eb |= 1u << BP_VECTOR;
1901         if (to_vmx(vcpu)->rmode.vm86_active)
1902                 eb = ~0;
1903         if (enable_ept)
1904                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1905
1906         /* When we are running a nested L2 guest and L1 specified for it a
1907          * certain exception bitmap, we must trap the same exceptions and pass
1908          * them to L1. When running L2, we will only handle the exceptions
1909          * specified above if L1 did not want them.
1910          */
1911         if (is_guest_mode(vcpu))
1912                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1913
1914         vmcs_write32(EXCEPTION_BITMAP, eb);
1915 }
1916
1917 /*
1918  * Check if MSR is intercepted for currently loaded MSR bitmap.
1919  */
1920 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
1921 {
1922         unsigned long *msr_bitmap;
1923         int f = sizeof(unsigned long);
1924
1925         if (!cpu_has_vmx_msr_bitmap())
1926                 return true;
1927
1928         msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
1929
1930         if (msr <= 0x1fff) {
1931                 return !!test_bit(msr, msr_bitmap + 0x800 / f);
1932         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
1933                 msr &= 0x1fff;
1934                 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
1935         }
1936
1937         return true;
1938 }
1939
1940 /*
1941  * Check if MSR is intercepted for L01 MSR bitmap.
1942  */
1943 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
1944 {
1945         unsigned long *msr_bitmap;
1946         int f = sizeof(unsigned long);
1947
1948         if (!cpu_has_vmx_msr_bitmap())
1949                 return true;
1950
1951         msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
1952
1953         if (msr <= 0x1fff) {
1954                 return !!test_bit(msr, msr_bitmap + 0x800 / f);
1955         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
1956                 msr &= 0x1fff;
1957                 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
1958         }
1959
1960         return true;
1961 }
1962
1963 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1964                 unsigned long entry, unsigned long exit)
1965 {
1966         vm_entry_controls_clearbit(vmx, entry);
1967         vm_exit_controls_clearbit(vmx, exit);
1968 }
1969
1970 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
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                         clear_atomic_switch_msr_special(vmx,
1979                                         VM_ENTRY_LOAD_IA32_EFER,
1980                                         VM_EXIT_LOAD_IA32_EFER);
1981                         return;
1982                 }
1983                 break;
1984         case MSR_CORE_PERF_GLOBAL_CTRL:
1985                 if (cpu_has_load_perf_global_ctrl) {
1986                         clear_atomic_switch_msr_special(vmx,
1987                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1988                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1989                         return;
1990                 }
1991                 break;
1992         }
1993
1994         for (i = 0; i < m->nr; ++i)
1995                 if (m->guest[i].index == msr)
1996                         break;
1997
1998         if (i == m->nr)
1999                 return;
2000         --m->nr;
2001         m->guest[i] = m->guest[m->nr];
2002         m->host[i] = m->host[m->nr];
2003         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2004         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2005 }
2006
2007 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
2008                 unsigned long entry, unsigned long exit,
2009                 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
2010                 u64 guest_val, u64 host_val)
2011 {
2012         vmcs_write64(guest_val_vmcs, guest_val);
2013         vmcs_write64(host_val_vmcs, host_val);
2014         vm_entry_controls_setbit(vmx, entry);
2015         vm_exit_controls_setbit(vmx, exit);
2016 }
2017
2018 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
2019                                   u64 guest_val, u64 host_val)
2020 {
2021         unsigned i;
2022         struct msr_autoload *m = &vmx->msr_autoload;
2023
2024         switch (msr) {
2025         case MSR_EFER:
2026                 if (cpu_has_load_ia32_efer) {
2027                         add_atomic_switch_msr_special(vmx,
2028                                         VM_ENTRY_LOAD_IA32_EFER,
2029                                         VM_EXIT_LOAD_IA32_EFER,
2030                                         GUEST_IA32_EFER,
2031                                         HOST_IA32_EFER,
2032                                         guest_val, host_val);
2033                         return;
2034                 }
2035                 break;
2036         case MSR_CORE_PERF_GLOBAL_CTRL:
2037                 if (cpu_has_load_perf_global_ctrl) {
2038                         add_atomic_switch_msr_special(vmx,
2039                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
2040                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
2041                                         GUEST_IA32_PERF_GLOBAL_CTRL,
2042                                         HOST_IA32_PERF_GLOBAL_CTRL,
2043                                         guest_val, host_val);
2044                         return;
2045                 }
2046                 break;
2047         case MSR_IA32_PEBS_ENABLE:
2048                 /* PEBS needs a quiescent period after being disabled (to write
2049                  * a record).  Disabling PEBS through VMX MSR swapping doesn't
2050                  * provide that period, so a CPU could write host's record into
2051                  * guest's memory.
2052                  */
2053                 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
2054         }
2055
2056         for (i = 0; i < m->nr; ++i)
2057                 if (m->guest[i].index == msr)
2058                         break;
2059
2060         if (i == NR_AUTOLOAD_MSRS) {
2061                 printk_once(KERN_WARNING "Not enough msr switch entries. "
2062                                 "Can't add msr %x\n", msr);
2063                 return;
2064         } else if (i == m->nr) {
2065                 ++m->nr;
2066                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2067                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2068         }
2069
2070         m->guest[i].index = msr;
2071         m->guest[i].value = guest_val;
2072         m->host[i].index = msr;
2073         m->host[i].value = host_val;
2074 }
2075
2076 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2077 {
2078         u64 guest_efer = vmx->vcpu.arch.efer;
2079         u64 ignore_bits = 0;
2080
2081         if (!enable_ept) {
2082                 /*
2083                  * NX is needed to handle CR0.WP=1, CR4.SMEP=1.  Testing
2084                  * host CPUID is more efficient than testing guest CPUID
2085                  * or CR4.  Host SMEP is anyway a requirement for guest SMEP.
2086                  */
2087                 if (boot_cpu_has(X86_FEATURE_SMEP))
2088                         guest_efer |= EFER_NX;
2089                 else if (!(guest_efer & EFER_NX))
2090                         ignore_bits |= EFER_NX;
2091         }
2092
2093         /*
2094          * LMA and LME handled by hardware; SCE meaningless outside long mode.
2095          */
2096         ignore_bits |= EFER_SCE;
2097 #ifdef CONFIG_X86_64
2098         ignore_bits |= EFER_LMA | EFER_LME;
2099         /* SCE is meaningful only in long mode on Intel */
2100         if (guest_efer & EFER_LMA)
2101                 ignore_bits &= ~(u64)EFER_SCE;
2102 #endif
2103
2104         clear_atomic_switch_msr(vmx, MSR_EFER);
2105
2106         /*
2107          * On EPT, we can't emulate NX, so we must switch EFER atomically.
2108          * On CPUs that support "load IA32_EFER", always switch EFER
2109          * atomically, since it's faster than switching it manually.
2110          */
2111         if (cpu_has_load_ia32_efer ||
2112             (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2113                 if (!(guest_efer & EFER_LMA))
2114                         guest_efer &= ~EFER_LME;
2115                 if (guest_efer != host_efer)
2116                         add_atomic_switch_msr(vmx, MSR_EFER,
2117                                               guest_efer, host_efer);
2118                 return false;
2119         } else {
2120                 guest_efer &= ~ignore_bits;
2121                 guest_efer |= host_efer & ignore_bits;
2122
2123                 vmx->guest_msrs[efer_offset].data = guest_efer;
2124                 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2125
2126                 return true;
2127         }
2128 }
2129
2130 #ifdef CONFIG_X86_32
2131 /*
2132  * On 32-bit kernels, VM exits still load the FS and GS bases from the
2133  * VMCS rather than the segment table.  KVM uses this helper to figure
2134  * out the current bases to poke them into the VMCS before entry.
2135  */
2136 static unsigned long segment_base(u16 selector)
2137 {
2138         struct desc_struct *table;
2139         unsigned long v;
2140
2141         if (!(selector & ~SEGMENT_RPL_MASK))
2142                 return 0;
2143
2144         table = get_current_gdt_ro();
2145
2146         if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2147                 u16 ldt_selector = kvm_read_ldt();
2148
2149                 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2150                         return 0;
2151
2152                 table = (struct desc_struct *)segment_base(ldt_selector);
2153         }
2154         v = get_desc_base(&table[selector >> 3]);
2155         return v;
2156 }
2157 #endif
2158
2159 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2160 {
2161         struct vcpu_vmx *vmx = to_vmx(vcpu);
2162         int i;
2163
2164         if (vmx->host_state.loaded)
2165                 return;
2166
2167         vmx->host_state.loaded = 1;
2168         /*
2169          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
2170          * allow segment selectors with cpl > 0 or ti == 1.
2171          */
2172         vmx->host_state.ldt_sel = kvm_read_ldt();
2173         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2174         savesegment(fs, vmx->host_state.fs_sel);
2175         if (!(vmx->host_state.fs_sel & 7)) {
2176                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2177                 vmx->host_state.fs_reload_needed = 0;
2178         } else {
2179                 vmcs_write16(HOST_FS_SELECTOR, 0);
2180                 vmx->host_state.fs_reload_needed = 1;
2181         }
2182         savesegment(gs, vmx->host_state.gs_sel);
2183         if (!(vmx->host_state.gs_sel & 7))
2184                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2185         else {
2186                 vmcs_write16(HOST_GS_SELECTOR, 0);
2187                 vmx->host_state.gs_ldt_reload_needed = 1;
2188         }
2189
2190 #ifdef CONFIG_X86_64
2191         savesegment(ds, vmx->host_state.ds_sel);
2192         savesegment(es, vmx->host_state.es_sel);
2193 #endif
2194
2195 #ifdef CONFIG_X86_64
2196         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2197         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2198 #else
2199         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2200         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2201 #endif
2202
2203 #ifdef CONFIG_X86_64
2204         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2205         if (is_long_mode(&vmx->vcpu))
2206                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2207 #endif
2208         if (boot_cpu_has(X86_FEATURE_MPX))
2209                 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2210         for (i = 0; i < vmx->save_nmsrs; ++i)
2211                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2212                                    vmx->guest_msrs[i].data,
2213                                    vmx->guest_msrs[i].mask);
2214 }
2215
2216 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2217 {
2218         if (!vmx->host_state.loaded)
2219                 return;
2220
2221         ++vmx->vcpu.stat.host_state_reload;
2222         vmx->host_state.loaded = 0;
2223 #ifdef CONFIG_X86_64
2224         if (is_long_mode(&vmx->vcpu))
2225                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2226 #endif
2227         if (vmx->host_state.gs_ldt_reload_needed) {
2228                 kvm_load_ldt(vmx->host_state.ldt_sel);
2229 #ifdef CONFIG_X86_64
2230                 load_gs_index(vmx->host_state.gs_sel);
2231 #else
2232                 loadsegment(gs, vmx->host_state.gs_sel);
2233 #endif
2234         }
2235         if (vmx->host_state.fs_reload_needed)
2236                 loadsegment(fs, vmx->host_state.fs_sel);
2237 #ifdef CONFIG_X86_64
2238         if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2239                 loadsegment(ds, vmx->host_state.ds_sel);
2240                 loadsegment(es, vmx->host_state.es_sel);
2241         }
2242 #endif
2243         invalidate_tss_limit();
2244 #ifdef CONFIG_X86_64
2245         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2246 #endif
2247         if (vmx->host_state.msr_host_bndcfgs)
2248                 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2249         load_fixmap_gdt(raw_smp_processor_id());
2250 }
2251
2252 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2253 {
2254         preempt_disable();
2255         __vmx_load_host_state(vmx);
2256         preempt_enable();
2257 }
2258
2259 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2260 {
2261         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2262         struct pi_desc old, new;
2263         unsigned int dest;
2264
2265         /*
2266          * In case of hot-plug or hot-unplug, we may have to undo
2267          * vmx_vcpu_pi_put even if there is no assigned device.  And we
2268          * always keep PI.NDST up to date for simplicity: it makes the
2269          * code easier, and CPU migration is not a fast path.
2270          */
2271         if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
2272                 return;
2273
2274         /*
2275          * First handle the simple case where no cmpxchg is necessary; just
2276          * allow posting non-urgent interrupts.
2277          *
2278          * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
2279          * PI.NDST: pi_post_block will do it for us and the wakeup_handler
2280          * expects the VCPU to be on the blocked_vcpu_list that matches
2281          * PI.NDST.
2282          */
2283         if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR ||
2284             vcpu->cpu == cpu) {
2285                 pi_clear_sn(pi_desc);
2286                 return;
2287         }
2288
2289         /* The full case.  */
2290         do {
2291                 old.control = new.control = pi_desc->control;
2292
2293                 dest = cpu_physical_id(cpu);
2294
2295                 if (x2apic_enabled())
2296                         new.ndst = dest;
2297                 else
2298                         new.ndst = (dest << 8) & 0xFF00;
2299
2300                 new.sn = 0;
2301         } while (cmpxchg64(&pi_desc->control, old.control,
2302                            new.control) != old.control);
2303 }
2304
2305 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2306 {
2307         vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2308         vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2309 }
2310
2311 /*
2312  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2313  * vcpu mutex is already taken.
2314  */
2315 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2316 {
2317         struct vcpu_vmx *vmx = to_vmx(vcpu);
2318         bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2319
2320         if (!already_loaded) {
2321                 loaded_vmcs_clear(vmx->loaded_vmcs);
2322                 local_irq_disable();
2323                 crash_disable_local_vmclear(cpu);
2324
2325                 /*
2326                  * Read loaded_vmcs->cpu should be before fetching
2327                  * loaded_vmcs->loaded_vmcss_on_cpu_link.
2328                  * See the comments in __loaded_vmcs_clear().
2329                  */
2330                 smp_rmb();
2331
2332                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2333                          &per_cpu(loaded_vmcss_on_cpu, cpu));
2334                 crash_enable_local_vmclear(cpu);
2335                 local_irq_enable();
2336         }
2337
2338         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2339                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2340                 vmcs_load(vmx->loaded_vmcs->vmcs);
2341                 indirect_branch_prediction_barrier();
2342         }
2343
2344         if (!already_loaded) {
2345                 void *gdt = get_current_gdt_ro();
2346                 unsigned long sysenter_esp;
2347
2348                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2349
2350                 /*
2351                  * Linux uses per-cpu TSS and GDT, so set these when switching
2352                  * processors.  See 22.2.4.
2353                  */
2354                 vmcs_writel(HOST_TR_BASE,
2355                             (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
2356                 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
2357
2358                 /*
2359                  * VM exits change the host TR limit to 0x67 after a VM
2360                  * exit.  This is okay, since 0x67 covers everything except
2361                  * the IO bitmap and have have code to handle the IO bitmap
2362                  * being lost after a VM exit.
2363                  */
2364                 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2365
2366                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2367                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2368
2369                 vmx->loaded_vmcs->cpu = cpu;
2370         }
2371
2372         /* Setup TSC multiplier */
2373         if (kvm_has_tsc_control &&
2374             vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2375                 decache_tsc_multiplier(vmx);
2376
2377         vmx_vcpu_pi_load(vcpu, cpu);
2378         vmx->host_pkru = read_pkru();
2379 }
2380
2381 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2382 {
2383         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2384
2385         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2386                 !irq_remapping_cap(IRQ_POSTING_CAP)  ||
2387                 !kvm_vcpu_apicv_active(vcpu))
2388                 return;
2389
2390         /* Set SN when the vCPU is preempted */
2391         if (vcpu->preempted)
2392                 pi_set_sn(pi_desc);
2393 }
2394
2395 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2396 {
2397         vmx_vcpu_pi_put(vcpu);
2398
2399         __vmx_load_host_state(to_vmx(vcpu));
2400 }
2401
2402 static bool emulation_required(struct kvm_vcpu *vcpu)
2403 {
2404         return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2405 }
2406
2407 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2408
2409 /*
2410  * Return the cr0 value that a nested guest would read. This is a combination
2411  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2412  * its hypervisor (cr0_read_shadow).
2413  */
2414 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2415 {
2416         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2417                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2418 }
2419 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2420 {
2421         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2422                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2423 }
2424
2425 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2426 {
2427         unsigned long rflags, save_rflags;
2428
2429         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2430                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2431                 rflags = vmcs_readl(GUEST_RFLAGS);
2432                 if (to_vmx(vcpu)->rmode.vm86_active) {
2433                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2434                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2435                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2436                 }
2437                 to_vmx(vcpu)->rflags = rflags;
2438         }
2439         return to_vmx(vcpu)->rflags;
2440 }
2441
2442 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2443 {
2444         unsigned long old_rflags = vmx_get_rflags(vcpu);
2445
2446         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2447         to_vmx(vcpu)->rflags = rflags;
2448         if (to_vmx(vcpu)->rmode.vm86_active) {
2449                 to_vmx(vcpu)->rmode.save_rflags = rflags;
2450                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2451         }
2452         vmcs_writel(GUEST_RFLAGS, rflags);
2453
2454         if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2455                 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2456 }
2457
2458 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2459 {
2460         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2461         int ret = 0;
2462
2463         if (interruptibility & GUEST_INTR_STATE_STI)
2464                 ret |= KVM_X86_SHADOW_INT_STI;
2465         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2466                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2467
2468         return ret;
2469 }
2470
2471 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2472 {
2473         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2474         u32 interruptibility = interruptibility_old;
2475
2476         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2477
2478         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2479                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2480         else if (mask & KVM_X86_SHADOW_INT_STI)
2481                 interruptibility |= GUEST_INTR_STATE_STI;
2482
2483         if ((interruptibility != interruptibility_old))
2484                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2485 }
2486
2487 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2488 {
2489         unsigned long rip;
2490
2491         rip = kvm_rip_read(vcpu);
2492         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2493         kvm_rip_write(vcpu, rip);
2494
2495         /* skipping an emulated instruction also counts */
2496         vmx_set_interrupt_shadow(vcpu, 0);
2497 }
2498
2499 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2500                                                unsigned long exit_qual)
2501 {
2502         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2503         unsigned int nr = vcpu->arch.exception.nr;
2504         u32 intr_info = nr | INTR_INFO_VALID_MASK;
2505
2506         if (vcpu->arch.exception.has_error_code) {
2507                 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2508                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2509         }
2510
2511         if (kvm_exception_is_soft(nr))
2512                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2513         else
2514                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2515
2516         if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2517             vmx_get_nmi_mask(vcpu))
2518                 intr_info |= INTR_INFO_UNBLOCK_NMI;
2519
2520         nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2521 }
2522
2523 /*
2524  * KVM wants to inject page-faults which it got to the guest. This function
2525  * checks whether in a nested guest, we need to inject them to L1 or L2.
2526  */
2527 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
2528 {
2529         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2530         unsigned int nr = vcpu->arch.exception.nr;
2531
2532         if (nr == PF_VECTOR) {
2533                 if (vcpu->arch.exception.nested_apf) {
2534                         *exit_qual = vcpu->arch.apf.nested_apf_token;
2535                         return 1;
2536                 }
2537                 /*
2538                  * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2539                  * The fix is to add the ancillary datum (CR2 or DR6) to structs
2540                  * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2541                  * can be written only when inject_pending_event runs.  This should be
2542                  * conditional on a new capability---if the capability is disabled,
2543                  * kvm_multiple_exception would write the ancillary information to
2544                  * CR2 or DR6, for backwards ABI-compatibility.
2545                  */
2546                 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2547                                                     vcpu->arch.exception.error_code)) {
2548                         *exit_qual = vcpu->arch.cr2;
2549                         return 1;
2550                 }
2551         } else {
2552                 if (vmcs12->exception_bitmap & (1u << nr)) {
2553                         if (nr == DB_VECTOR)
2554                                 *exit_qual = vcpu->arch.dr6;
2555                         else
2556                                 *exit_qual = 0;
2557                         return 1;
2558                 }
2559         }
2560
2561         return 0;
2562 }
2563
2564 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
2565 {
2566         struct vcpu_vmx *vmx = to_vmx(vcpu);
2567         unsigned nr = vcpu->arch.exception.nr;
2568         bool has_error_code = vcpu->arch.exception.has_error_code;
2569         u32 error_code = vcpu->arch.exception.error_code;
2570         u32 intr_info = nr | INTR_INFO_VALID_MASK;
2571
2572         if (has_error_code) {
2573                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2574                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2575         }
2576
2577         if (vmx->rmode.vm86_active) {
2578                 int inc_eip = 0;
2579                 if (kvm_exception_is_soft(nr))
2580                         inc_eip = vcpu->arch.event_exit_inst_len;
2581                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2582                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2583                 return;
2584         }
2585
2586         if (kvm_exception_is_soft(nr)) {
2587                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2588                              vmx->vcpu.arch.event_exit_inst_len);
2589                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2590         } else
2591                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2592
2593         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2594 }
2595
2596 static bool vmx_rdtscp_supported(void)
2597 {
2598         return cpu_has_vmx_rdtscp();
2599 }
2600
2601 static bool vmx_invpcid_supported(void)
2602 {
2603         return cpu_has_vmx_invpcid() && enable_ept;
2604 }
2605
2606 /*
2607  * Swap MSR entry in host/guest MSR entry array.
2608  */
2609 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2610 {
2611         struct shared_msr_entry tmp;
2612
2613         tmp = vmx->guest_msrs[to];
2614         vmx->guest_msrs[to] = vmx->guest_msrs[from];
2615         vmx->guest_msrs[from] = tmp;
2616 }
2617
2618 /*
2619  * Set up the vmcs to automatically save and restore system
2620  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
2621  * mode, as fiddling with msrs is very expensive.
2622  */
2623 static void setup_msrs(struct vcpu_vmx *vmx)
2624 {
2625         int save_nmsrs, index;
2626
2627         save_nmsrs = 0;
2628 #ifdef CONFIG_X86_64
2629         if (is_long_mode(&vmx->vcpu)) {
2630                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2631                 if (index >= 0)
2632                         move_msr_up(vmx, index, save_nmsrs++);
2633                 index = __find_msr_index(vmx, MSR_LSTAR);
2634                 if (index >= 0)
2635                         move_msr_up(vmx, index, save_nmsrs++);
2636                 index = __find_msr_index(vmx, MSR_CSTAR);
2637                 if (index >= 0)
2638                         move_msr_up(vmx, index, save_nmsrs++);
2639                 index = __find_msr_index(vmx, MSR_TSC_AUX);
2640                 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
2641                         move_msr_up(vmx, index, save_nmsrs++);
2642                 /*
2643                  * MSR_STAR is only needed on long mode guests, and only
2644                  * if efer.sce is enabled.
2645                  */
2646                 index = __find_msr_index(vmx, MSR_STAR);
2647                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2648                         move_msr_up(vmx, index, save_nmsrs++);
2649         }
2650 #endif
2651         index = __find_msr_index(vmx, MSR_EFER);
2652         if (index >= 0 && update_transition_efer(vmx, index))
2653                 move_msr_up(vmx, index, save_nmsrs++);
2654
2655         vmx->save_nmsrs = save_nmsrs;
2656
2657         if (cpu_has_vmx_msr_bitmap())
2658                 vmx_update_msr_bitmap(&vmx->vcpu);
2659 }
2660
2661 /*
2662  * reads and returns guest's timestamp counter "register"
2663  * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2664  * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2665  */
2666 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2667 {
2668         u64 host_tsc, tsc_offset;
2669
2670         host_tsc = rdtsc();
2671         tsc_offset = vmcs_read64(TSC_OFFSET);
2672         return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2673 }
2674
2675 /*
2676  * writes 'offset' into guest's timestamp counter offset register
2677  */
2678 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2679 {
2680         if (is_guest_mode(vcpu)) {
2681                 /*
2682                  * We're here if L1 chose not to trap WRMSR to TSC. According
2683                  * to the spec, this should set L1's TSC; The offset that L1
2684                  * set for L2 remains unchanged, and still needs to be added
2685                  * to the newly set TSC to get L2's TSC.
2686                  */
2687                 struct vmcs12 *vmcs12;
2688                 /* recalculate vmcs02.TSC_OFFSET: */
2689                 vmcs12 = get_vmcs12(vcpu);
2690                 vmcs_write64(TSC_OFFSET, offset +
2691                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2692                          vmcs12->tsc_offset : 0));
2693         } else {
2694                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2695                                            vmcs_read64(TSC_OFFSET), offset);
2696                 vmcs_write64(TSC_OFFSET, offset);
2697         }
2698 }
2699
2700 /*
2701  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2702  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2703  * all guests if the "nested" module option is off, and can also be disabled
2704  * for a single guest by disabling its VMX cpuid bit.
2705  */
2706 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2707 {
2708         return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
2709 }
2710
2711 /*
2712  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2713  * returned for the various VMX controls MSRs when nested VMX is enabled.
2714  * The same values should also be used to verify that vmcs12 control fields are
2715  * valid during nested entry from L1 to L2.
2716  * Each of these control msrs has a low and high 32-bit half: A low bit is on
2717  * if the corresponding bit in the (32-bit) control field *must* be on, and a
2718  * bit in the high half is on if the corresponding bit in the control field
2719  * may be on. See also vmx_control_verify().
2720  */
2721 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2722 {
2723         /*
2724          * Note that as a general rule, the high half of the MSRs (bits in
2725          * the control fields which may be 1) should be initialized by the
2726          * intersection of the underlying hardware's MSR (i.e., features which
2727          * can be supported) and the list of features we want to expose -
2728          * because they are known to be properly supported in our code.
2729          * Also, usually, the low half of the MSRs (bits which must be 1) can
2730          * be set to 0, meaning that L1 may turn off any of these bits. The
2731          * reason is that if one of these bits is necessary, it will appear
2732          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2733          * fields of vmcs01 and vmcs02, will turn these bits off - and
2734          * nested_vmx_exit_reflected() will not pass related exits to L1.
2735          * These rules have exceptions below.
2736          */
2737
2738         /* pin-based controls */
2739         rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2740                 vmx->nested.nested_vmx_pinbased_ctls_low,
2741                 vmx->nested.nested_vmx_pinbased_ctls_high);
2742         vmx->nested.nested_vmx_pinbased_ctls_low |=
2743                 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2744         vmx->nested.nested_vmx_pinbased_ctls_high &=
2745                 PIN_BASED_EXT_INTR_MASK |
2746                 PIN_BASED_NMI_EXITING |
2747                 PIN_BASED_VIRTUAL_NMIS;
2748         vmx->nested.nested_vmx_pinbased_ctls_high |=
2749                 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2750                 PIN_BASED_VMX_PREEMPTION_TIMER;
2751         if (kvm_vcpu_apicv_active(&vmx->vcpu))
2752                 vmx->nested.nested_vmx_pinbased_ctls_high |=
2753                         PIN_BASED_POSTED_INTR;
2754
2755         /* exit controls */
2756         rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2757                 vmx->nested.nested_vmx_exit_ctls_low,
2758                 vmx->nested.nested_vmx_exit_ctls_high);
2759         vmx->nested.nested_vmx_exit_ctls_low =
2760                 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2761
2762         vmx->nested.nested_vmx_exit_ctls_high &=
2763 #ifdef CONFIG_X86_64
2764                 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2765 #endif
2766                 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2767         vmx->nested.nested_vmx_exit_ctls_high |=
2768                 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2769                 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2770                 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2771
2772         if (kvm_mpx_supported())
2773                 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2774
2775         /* We support free control of debug control saving. */
2776         vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2777
2778         /* entry controls */
2779         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2780                 vmx->nested.nested_vmx_entry_ctls_low,
2781                 vmx->nested.nested_vmx_entry_ctls_high);
2782         vmx->nested.nested_vmx_entry_ctls_low =
2783                 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2784         vmx->nested.nested_vmx_entry_ctls_high &=
2785 #ifdef CONFIG_X86_64
2786                 VM_ENTRY_IA32E_MODE |
2787 #endif
2788                 VM_ENTRY_LOAD_IA32_PAT;
2789         vmx->nested.nested_vmx_entry_ctls_high |=
2790                 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2791         if (kvm_mpx_supported())
2792                 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2793
2794         /* We support free control of debug control loading. */
2795         vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2796
2797         /* cpu-based controls */
2798         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2799                 vmx->nested.nested_vmx_procbased_ctls_low,
2800                 vmx->nested.nested_vmx_procbased_ctls_high);
2801         vmx->nested.nested_vmx_procbased_ctls_low =
2802                 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2803         vmx->nested.nested_vmx_procbased_ctls_high &=
2804                 CPU_BASED_VIRTUAL_INTR_PENDING |
2805                 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2806                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2807                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2808                 CPU_BASED_CR3_STORE_EXITING |
2809 #ifdef CONFIG_X86_64
2810                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2811 #endif
2812                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2813                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2814                 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2815                 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2816                 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2817         /*
2818          * We can allow some features even when not supported by the
2819          * hardware. For example, L1 can specify an MSR bitmap - and we
2820          * can use it to avoid exits to L1 - even when L0 runs L2
2821          * without MSR bitmaps.
2822          */
2823         vmx->nested.nested_vmx_procbased_ctls_high |=
2824                 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2825                 CPU_BASED_USE_MSR_BITMAPS;
2826
2827         /* We support free control of CR3 access interception. */
2828         vmx->nested.nested_vmx_procbased_ctls_low &=
2829                 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2830
2831         /*
2832          * secondary cpu-based controls.  Do not include those that
2833          * depend on CPUID bits, they are added later by vmx_cpuid_update.
2834          */
2835         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2836                 vmx->nested.nested_vmx_secondary_ctls_low,
2837                 vmx->nested.nested_vmx_secondary_ctls_high);
2838         vmx->nested.nested_vmx_secondary_ctls_low = 0;
2839         vmx->nested.nested_vmx_secondary_ctls_high &=
2840                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2841                 SECONDARY_EXEC_DESC |
2842                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2843                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2844                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2845                 SECONDARY_EXEC_WBINVD_EXITING;
2846
2847         if (enable_ept) {
2848                 /* nested EPT: emulate EPT also to L1 */
2849                 vmx->nested.nested_vmx_secondary_ctls_high |=
2850                         SECONDARY_EXEC_ENABLE_EPT;
2851                 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2852                          VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2853                 if (cpu_has_vmx_ept_execute_only())
2854                         vmx->nested.nested_vmx_ept_caps |=
2855                                 VMX_EPT_EXECUTE_ONLY_BIT;
2856                 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2857                 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2858                         VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2859                         VMX_EPT_1GB_PAGE_BIT;
2860                 if (enable_ept_ad_bits) {
2861                         vmx->nested.nested_vmx_secondary_ctls_high |=
2862                                 SECONDARY_EXEC_ENABLE_PML;
2863                         vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2864                 }
2865         } else
2866                 vmx->nested.nested_vmx_ept_caps = 0;
2867
2868         if (cpu_has_vmx_vmfunc()) {
2869                 vmx->nested.nested_vmx_secondary_ctls_high |=
2870                         SECONDARY_EXEC_ENABLE_VMFUNC;
2871                 /*
2872                  * Advertise EPTP switching unconditionally
2873                  * since we emulate it
2874                  */
2875                 if (enable_ept)
2876                         vmx->nested.nested_vmx_vmfunc_controls =
2877                                 VMX_VMFUNC_EPTP_SWITCHING;
2878         }
2879
2880         /*
2881          * Old versions of KVM use the single-context version without
2882          * checking for support, so declare that it is supported even
2883          * though it is treated as global context.  The alternative is
2884          * not failing the single-context invvpid, and it is worse.
2885          */
2886         if (enable_vpid) {
2887                 vmx->nested.nested_vmx_secondary_ctls_high |=
2888                         SECONDARY_EXEC_ENABLE_VPID;
2889                 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2890                         VMX_VPID_EXTENT_SUPPORTED_MASK;
2891         } else
2892                 vmx->nested.nested_vmx_vpid_caps = 0;
2893
2894         if (enable_unrestricted_guest)
2895                 vmx->nested.nested_vmx_secondary_ctls_high |=
2896                         SECONDARY_EXEC_UNRESTRICTED_GUEST;
2897
2898         /* miscellaneous data */
2899         rdmsr(MSR_IA32_VMX_MISC,
2900                 vmx->nested.nested_vmx_misc_low,
2901                 vmx->nested.nested_vmx_misc_high);
2902         vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2903         vmx->nested.nested_vmx_misc_low |=
2904                 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2905                 VMX_MISC_ACTIVITY_HLT;
2906         vmx->nested.nested_vmx_misc_high = 0;
2907
2908         /*
2909          * This MSR reports some information about VMX support. We
2910          * should return information about the VMX we emulate for the
2911          * guest, and the VMCS structure we give it - not about the
2912          * VMX support of the underlying hardware.
2913          */
2914         vmx->nested.nested_vmx_basic =
2915                 VMCS12_REVISION |
2916                 VMX_BASIC_TRUE_CTLS |
2917                 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2918                 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2919
2920         if (cpu_has_vmx_basic_inout())
2921                 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2922
2923         /*
2924          * These MSRs specify bits which the guest must keep fixed on
2925          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2926          * We picked the standard core2 setting.
2927          */
2928 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2929 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
2930         vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2931         vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2932
2933         /* These MSRs specify bits which the guest must keep fixed off. */
2934         rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2935         rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2936
2937         /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2938         vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2939 }
2940
2941 /*
2942  * if fixed0[i] == 1: val[i] must be 1
2943  * if fixed1[i] == 0: val[i] must be 0
2944  */
2945 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2946 {
2947         return ((val & fixed1) | fixed0) == val;
2948 }
2949
2950 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2951 {
2952         return fixed_bits_valid(control, low, high);
2953 }
2954
2955 static inline u64 vmx_control_msr(u32 low, u32 high)
2956 {
2957         return low | ((u64)high << 32);
2958 }
2959
2960 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2961 {
2962         superset &= mask;
2963         subset &= mask;
2964
2965         return (superset | subset) == superset;
2966 }
2967
2968 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2969 {
2970         const u64 feature_and_reserved =
2971                 /* feature (except bit 48; see below) */
2972                 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2973                 /* reserved */
2974                 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2975         u64 vmx_basic = vmx->nested.nested_vmx_basic;
2976
2977         if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2978                 return -EINVAL;
2979
2980         /*
2981          * KVM does not emulate a version of VMX that constrains physical
2982          * addresses of VMX structures (e.g. VMCS) to 32-bits.
2983          */
2984         if (data & BIT_ULL(48))
2985                 return -EINVAL;
2986
2987         if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2988             vmx_basic_vmcs_revision_id(data))
2989                 return -EINVAL;
2990
2991         if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2992                 return -EINVAL;
2993
2994         vmx->nested.nested_vmx_basic = data;
2995         return 0;
2996 }
2997
2998 static int
2999 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3000 {
3001         u64 supported;
3002         u32 *lowp, *highp;
3003
3004         switch (msr_index) {
3005         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3006                 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
3007                 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
3008                 break;
3009         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3010                 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
3011                 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
3012                 break;
3013         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3014                 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
3015                 highp = &vmx->nested.nested_vmx_exit_ctls_high;
3016                 break;
3017         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3018                 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
3019                 highp = &vmx->nested.nested_vmx_entry_ctls_high;
3020                 break;
3021         case MSR_IA32_VMX_PROCBASED_CTLS2:
3022                 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
3023                 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
3024                 break;
3025         default:
3026                 BUG();
3027         }
3028
3029         supported = vmx_control_msr(*lowp, *highp);
3030
3031         /* Check must-be-1 bits are still 1. */
3032         if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
3033                 return -EINVAL;
3034
3035         /* Check must-be-0 bits are still 0. */
3036         if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3037                 return -EINVAL;
3038
3039         *lowp = data;
3040         *highp = data >> 32;
3041         return 0;
3042 }
3043
3044 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3045 {
3046         const u64 feature_and_reserved_bits =
3047                 /* feature */
3048                 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3049                 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3050                 /* reserved */
3051                 GENMASK_ULL(13, 9) | BIT_ULL(31);
3052         u64 vmx_misc;
3053
3054         vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
3055                                    vmx->nested.nested_vmx_misc_high);
3056
3057         if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3058                 return -EINVAL;
3059
3060         if ((vmx->nested.nested_vmx_pinbased_ctls_high &
3061              PIN_BASED_VMX_PREEMPTION_TIMER) &&
3062             vmx_misc_preemption_timer_rate(data) !=
3063             vmx_misc_preemption_timer_rate(vmx_misc))
3064                 return -EINVAL;
3065
3066         if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3067                 return -EINVAL;
3068
3069         if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3070                 return -EINVAL;
3071
3072         if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3073                 return -EINVAL;
3074
3075         vmx->nested.nested_vmx_misc_low = data;
3076         vmx->nested.nested_vmx_misc_high = data >> 32;
3077         return 0;
3078 }
3079
3080 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3081 {
3082         u64 vmx_ept_vpid_cap;
3083
3084         vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
3085                                            vmx->nested.nested_vmx_vpid_caps);
3086
3087         /* Every bit is either reserved or a feature bit. */
3088         if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3089                 return -EINVAL;
3090
3091         vmx->nested.nested_vmx_ept_caps = data;
3092         vmx->nested.nested_vmx_vpid_caps = data >> 32;
3093         return 0;
3094 }
3095
3096 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3097 {
3098         u64 *msr;
3099
3100         switch (msr_index) {
3101         case MSR_IA32_VMX_CR0_FIXED0:
3102                 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3103                 break;
3104         case MSR_IA32_VMX_CR4_FIXED0:
3105                 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3106                 break;
3107         default:
3108                 BUG();
3109         }
3110
3111         /*
3112          * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3113          * must be 1 in the restored value.
3114          */
3115         if (!is_bitwise_subset(data, *msr, -1ULL))
3116                 return -EINVAL;
3117
3118         *msr = data;
3119         return 0;
3120 }
3121
3122 /*
3123  * Called when userspace is restoring VMX MSRs.
3124  *
3125  * Returns 0 on success, non-0 otherwise.
3126  */
3127 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3128 {
3129         struct vcpu_vmx *vmx = to_vmx(vcpu);
3130
3131         switch (msr_index) {
3132         case MSR_IA32_VMX_BASIC:
3133                 return vmx_restore_vmx_basic(vmx, data);
3134         case MSR_IA32_VMX_PINBASED_CTLS:
3135         case MSR_IA32_VMX_PROCBASED_CTLS:
3136         case MSR_IA32_VMX_EXIT_CTLS:
3137         case MSR_IA32_VMX_ENTRY_CTLS:
3138                 /*
3139                  * The "non-true" VMX capability MSRs are generated from the
3140                  * "true" MSRs, so we do not support restoring them directly.
3141                  *
3142                  * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3143                  * should restore the "true" MSRs with the must-be-1 bits
3144                  * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3145                  * DEFAULT SETTINGS".
3146                  */
3147                 return -EINVAL;
3148         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3149         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3150         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3151         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3152         case MSR_IA32_VMX_PROCBASED_CTLS2:
3153                 return vmx_restore_control_msr(vmx, msr_index, data);
3154         case MSR_IA32_VMX_MISC:
3155                 return vmx_restore_vmx_misc(vmx, data);
3156         case MSR_IA32_VMX_CR0_FIXED0:
3157         case MSR_IA32_VMX_CR4_FIXED0:
3158                 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3159         case MSR_IA32_VMX_CR0_FIXED1:
3160         case MSR_IA32_VMX_CR4_FIXED1:
3161                 /*
3162                  * These MSRs are generated based on the vCPU's CPUID, so we
3163                  * do not support restoring them directly.
3164                  */
3165                 return -EINVAL;
3166         case MSR_IA32_VMX_EPT_VPID_CAP:
3167                 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3168         case MSR_IA32_VMX_VMCS_ENUM:
3169                 vmx->nested.nested_vmx_vmcs_enum = data;
3170                 return 0;
3171         default:
3172                 /*
3173                  * The rest of the VMX capability MSRs do not support restore.
3174                  */
3175                 return -EINVAL;
3176         }
3177 }
3178
3179 /* Returns 0 on success, non-0 otherwise. */
3180 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3181 {
3182         struct vcpu_vmx *vmx = to_vmx(vcpu);
3183
3184         switch (msr_index) {
3185         case MSR_IA32_VMX_BASIC:
3186                 *pdata = vmx->nested.nested_vmx_basic;
3187                 break;
3188         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3189         case MSR_IA32_VMX_PINBASED_CTLS:
3190                 *pdata = vmx_control_msr(
3191                         vmx->nested.nested_vmx_pinbased_ctls_low,
3192                         vmx->nested.nested_vmx_pinbased_ctls_high);
3193                 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3194                         *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3195                 break;
3196         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3197         case MSR_IA32_VMX_PROCBASED_CTLS:
3198                 *pdata = vmx_control_msr(
3199                         vmx->nested.nested_vmx_procbased_ctls_low,
3200                         vmx->nested.nested_vmx_procbased_ctls_high);
3201                 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3202                         *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3203                 break;
3204         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3205         case MSR_IA32_VMX_EXIT_CTLS:
3206                 *pdata = vmx_control_msr(
3207                         vmx->nested.nested_vmx_exit_ctls_low,
3208                         vmx->nested.nested_vmx_exit_ctls_high);
3209                 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3210                         *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3211                 break;
3212         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3213         case MSR_IA32_VMX_ENTRY_CTLS:
3214                 *pdata = vmx_control_msr(
3215                         vmx->nested.nested_vmx_entry_ctls_low,
3216                         vmx->nested.nested_vmx_entry_ctls_high);
3217                 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3218                         *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3219                 break;
3220         case MSR_IA32_VMX_MISC:
3221                 *pdata = vmx_control_msr(
3222                         vmx->nested.nested_vmx_misc_low,
3223                         vmx->nested.nested_vmx_misc_high);
3224                 break;
3225         case MSR_IA32_VMX_CR0_FIXED0:
3226                 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3227                 break;
3228         case MSR_IA32_VMX_CR0_FIXED1:
3229                 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3230                 break;
3231         case MSR_IA32_VMX_CR4_FIXED0:
3232                 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3233                 break;
3234         case MSR_IA32_VMX_CR4_FIXED1:
3235                 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3236                 break;
3237         case MSR_IA32_VMX_VMCS_ENUM:
3238                 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3239                 break;
3240         case MSR_IA32_VMX_PROCBASED_CTLS2:
3241                 *pdata = vmx_control_msr(
3242                         vmx->nested.nested_vmx_secondary_ctls_low,
3243                         vmx->nested.nested_vmx_secondary_ctls_high);
3244                 break;
3245         case MSR_IA32_VMX_EPT_VPID_CAP:
3246                 *pdata = vmx->nested.nested_vmx_ept_caps |
3247                         ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3248                 break;
3249         case MSR_IA32_VMX_VMFUNC:
3250                 *pdata = vmx->nested.nested_vmx_vmfunc_controls;
3251                 break;
3252         default:
3253                 return 1;
3254         }
3255
3256         return 0;
3257 }
3258
3259 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3260                                                  uint64_t val)
3261 {
3262         uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3263
3264         return !(val & ~valid_bits);
3265 }
3266
3267 /*
3268  * Reads an msr value (of 'msr_index') into 'pdata'.
3269  * Returns 0 on success, non-0 otherwise.
3270  * Assumes vcpu_load() was already called.
3271  */
3272 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3273 {
3274         struct shared_msr_entry *msr;
3275
3276         switch (msr_info->index) {
3277 #ifdef CONFIG_X86_64
3278         case MSR_FS_BASE:
3279                 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3280                 break;
3281         case MSR_GS_BASE:
3282                 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3283                 break;
3284         case MSR_KERNEL_GS_BASE:
3285                 vmx_load_host_state(to_vmx(vcpu));
3286                 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3287                 break;
3288 #endif
3289         case MSR_EFER:
3290                 return kvm_get_msr_common(vcpu, msr_info);
3291         case MSR_IA32_TSC:
3292                 msr_info->data = guest_read_tsc(vcpu);
3293                 break;
3294         case MSR_IA32_SPEC_CTRL:
3295                 if (!msr_info->host_initiated &&
3296                     !guest_cpuid_has(vcpu, X86_FEATURE_IBRS) &&
3297                     !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
3298                     !guest_cpuid_has(vcpu, X86_FEATURE_RDS))
3299                         return 1;
3300
3301                 msr_info->data = to_vmx(vcpu)->spec_ctrl;
3302                 break;
3303         case MSR_IA32_ARCH_CAPABILITIES:
3304                 if (!msr_info->host_initiated &&
3305                     !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3306                         return 1;
3307                 msr_info->data = to_vmx(vcpu)->arch_capabilities;
3308                 break;
3309         case MSR_IA32_SYSENTER_CS:
3310                 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3311                 break;
3312         case MSR_IA32_SYSENTER_EIP:
3313                 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3314                 break;
3315         case MSR_IA32_SYSENTER_ESP:
3316                 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3317                 break;
3318         case MSR_IA32_BNDCFGS:
3319                 if (!kvm_mpx_supported() ||
3320                     (!msr_info->host_initiated &&
3321                      !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3322                         return 1;
3323                 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3324                 break;
3325         case MSR_IA32_MCG_EXT_CTL:
3326                 if (!msr_info->host_initiated &&
3327                     !(to_vmx(vcpu)->msr_ia32_feature_control &
3328                       FEATURE_CONTROL_LMCE))
3329                         return 1;
3330                 msr_info->data = vcpu->arch.mcg_ext_ctl;
3331                 break;
3332         case MSR_IA32_FEATURE_CONTROL:
3333                 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3334                 break;
3335         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3336                 if (!nested_vmx_allowed(vcpu))
3337                         return 1;
3338                 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3339         case MSR_IA32_XSS:
3340                 if (!vmx_xsaves_supported())
3341                         return 1;
3342                 msr_info->data = vcpu->arch.ia32_xss;
3343                 break;
3344         case MSR_TSC_AUX:
3345                 if (!msr_info->host_initiated &&
3346                     !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3347                         return 1;
3348                 /* Otherwise falls through */
3349         default:
3350                 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3351                 if (msr) {
3352                         msr_info->data = msr->data;
3353                         break;
3354                 }
3355                 return kvm_get_msr_common(vcpu, msr_info);
3356         }
3357
3358         return 0;
3359 }
3360
3361 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3362
3363 /*
3364  * Writes msr value into into the appropriate "register".
3365  * Returns 0 on success, non-0 otherwise.
3366  * Assumes vcpu_load() was already called.
3367  */
3368 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3369 {
3370         struct vcpu_vmx *vmx = to_vmx(vcpu);
3371         struct shared_msr_entry *msr;
3372         int ret = 0;
3373         u32 msr_index = msr_info->index;
3374         u64 data = msr_info->data;
3375
3376         switch (msr_index) {
3377         case MSR_EFER:
3378                 ret = kvm_set_msr_common(vcpu, msr_info);
3379                 break;
3380 #ifdef CONFIG_X86_64
3381         case MSR_FS_BASE:
3382                 vmx_segment_cache_clear(vmx);
3383                 vmcs_writel(GUEST_FS_BASE, data);
3384                 break;
3385         case MSR_GS_BASE:
3386                 vmx_segment_cache_clear(vmx);
3387                 vmcs_writel(GUEST_GS_BASE, data);
3388                 break;
3389         case MSR_KERNEL_GS_BASE:
3390                 vmx_load_host_state(vmx);
3391                 vmx->msr_guest_kernel_gs_base = data;
3392                 break;
3393 #endif
3394         case MSR_IA32_SYSENTER_CS:
3395                 vmcs_write32(GUEST_SYSENTER_CS, data);
3396                 break;
3397         case MSR_IA32_SYSENTER_EIP:
3398                 vmcs_writel(GUEST_SYSENTER_EIP, data);
3399                 break;
3400         case MSR_IA32_SYSENTER_ESP:
3401                 vmcs_writel(GUEST_SYSENTER_ESP, data);
3402                 break;
3403         case MSR_IA32_BNDCFGS:
3404                 if (!kvm_mpx_supported() ||
3405                     (!msr_info->host_initiated &&
3406                      !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3407                         return 1;
3408                 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
3409                     (data & MSR_IA32_BNDCFGS_RSVD))
3410                         return 1;
3411                 vmcs_write64(GUEST_BNDCFGS, data);
3412                 break;
3413         case MSR_IA32_TSC:
3414                 kvm_write_tsc(vcpu, msr_info);
3415                 break;
3416         case MSR_IA32_SPEC_CTRL:
3417                 if (!msr_info->host_initiated &&
3418                     !guest_cpuid_has(vcpu, X86_FEATURE_IBRS) &&
3419                     !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
3420                     !guest_cpuid_has(vcpu, X86_FEATURE_RDS))
3421                         return 1;
3422
3423                 /* The STIBP bit doesn't fault even if it's not advertised */
3424                 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_RDS))
3425                         return 1;
3426
3427                 vmx->spec_ctrl = data;
3428
3429                 if (!data)
3430                         break;
3431
3432                 /*
3433                  * For non-nested:
3434                  * When it's written (to non-zero) for the first time, pass
3435                  * it through.
3436                  *
3437                  * For nested:
3438                  * The handling of the MSR bitmap for L2 guests is done in
3439                  * nested_vmx_merge_msr_bitmap. We should not touch the
3440                  * vmcs02.msr_bitmap here since it gets completely overwritten
3441                  * in the merging. We update the vmcs01 here for L1 as well
3442                  * since it will end up touching the MSR anyway now.
3443                  */
3444                 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
3445                                               MSR_IA32_SPEC_CTRL,
3446                                               MSR_TYPE_RW);
3447                 break;
3448         case MSR_IA32_PRED_CMD:
3449                 if (!msr_info->host_initiated &&
3450                     !guest_cpuid_has(vcpu, X86_FEATURE_IBPB) &&
3451                     !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3452                         return 1;
3453
3454                 if (data & ~PRED_CMD_IBPB)
3455                         return 1;
3456
3457                 if (!data)
3458                         break;
3459
3460                 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
3461
3462                 /*
3463                  * For non-nested:
3464                  * When it's written (to non-zero) for the first time, pass
3465                  * it through.
3466                  *
3467                  * For nested:
3468                  * The handling of the MSR bitmap for L2 guests is done in
3469                  * nested_vmx_merge_msr_bitmap. We should not touch the
3470                  * vmcs02.msr_bitmap here since it gets completely overwritten
3471                  * in the merging.
3472                  */
3473                 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
3474                                               MSR_TYPE_W);
3475                 break;
3476         case MSR_IA32_ARCH_CAPABILITIES:
3477                 if (!msr_info->host_initiated)
3478                         return 1;
3479                 vmx->arch_capabilities = data;
3480                 break;
3481         case MSR_IA32_CR_PAT:
3482                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3483                         if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3484                                 return 1;
3485                         vmcs_write64(GUEST_IA32_PAT, data);
3486                         vcpu->arch.pat = data;
3487                         break;
3488                 }
3489                 ret = kvm_set_msr_common(vcpu, msr_info);
3490                 break;
3491         case MSR_IA32_TSC_ADJUST:
3492                 ret = kvm_set_msr_common(vcpu, msr_info);
3493                 break;
3494         case MSR_IA32_MCG_EXT_CTL:
3495                 if ((!msr_info->host_initiated &&
3496                      !(to_vmx(vcpu)->msr_ia32_feature_control &
3497                        FEATURE_CONTROL_LMCE)) ||
3498                     (data & ~MCG_EXT_CTL_LMCE_EN))
3499                         return 1;
3500                 vcpu->arch.mcg_ext_ctl = data;
3501                 break;
3502         case MSR_IA32_FEATURE_CONTROL:
3503                 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3504                     (to_vmx(vcpu)->msr_ia32_feature_control &
3505                      FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3506                         return 1;
3507                 vmx->msr_ia32_feature_control = data;
3508                 if (msr_info->host_initiated && data == 0)
3509                         vmx_leave_nested(vcpu);
3510                 break;
3511         case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3512                 if (!msr_info->host_initiated)
3513                         return 1; /* they are read-only */
3514                 if (!nested_vmx_allowed(vcpu))
3515                         return 1;
3516                 return vmx_set_vmx_msr(vcpu, msr_index, data);
3517         case MSR_IA32_XSS:
3518                 if (!vmx_xsaves_supported())
3519                         return 1;
3520                 /*
3521                  * The only supported bit as of Skylake is bit 8, but
3522                  * it is not supported on KVM.
3523                  */
3524                 if (data != 0)
3525                         return 1;
3526                 vcpu->arch.ia32_xss = data;
3527                 if (vcpu->arch.ia32_xss != host_xss)
3528                         add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3529                                 vcpu->arch.ia32_xss, host_xss);
3530                 else
3531                         clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3532                 break;
3533         case MSR_TSC_AUX:
3534                 if (!msr_info->host_initiated &&
3535                     !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3536                         return 1;
3537                 /* Check reserved bit, higher 32 bits should be zero */
3538                 if ((data >> 32) != 0)
3539                         return 1;
3540                 /* Otherwise falls through */
3541         default:
3542                 msr = find_msr_entry(vmx, msr_index);
3543                 if (msr) {
3544                         u64 old_msr_data = msr->data;
3545                         msr->data = data;
3546                         if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3547                                 preempt_disable();
3548                                 ret = kvm_set_shared_msr(msr->index, msr->data,
3549                                                          msr->mask);
3550                                 preempt_enable();
3551                                 if (ret)
3552                                         msr->data = old_msr_data;
3553                         }
3554                         break;
3555                 }
3556                 ret = kvm_set_msr_common(vcpu, msr_info);
3557         }
3558
3559         return ret;
3560 }
3561
3562 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3563 {
3564         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3565         switch (reg) {
3566         case VCPU_REGS_RSP:
3567                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3568                 break;
3569         case VCPU_REGS_RIP:
3570                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3571                 break;
3572         case VCPU_EXREG_PDPTR:
3573                 if (enable_ept)
3574                         ept_save_pdptrs(vcpu);
3575                 break;
3576         default:
3577                 break;
3578         }
3579 }
3580
3581 static __init int cpu_has_kvm_support(void)
3582 {
3583         return cpu_has_vmx();
3584 }
3585
3586 static __init int vmx_disabled_by_bios(void)
3587 {
3588         u64 msr;
3589
3590         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3591         if (msr & FEATURE_CONTROL_LOCKED) {
3592                 /* launched w/ TXT and VMX disabled */
3593                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3594                         && tboot_enabled())
3595                         return 1;
3596                 /* launched w/o TXT and VMX only enabled w/ TXT */
3597                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3598                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3599                         && !tboot_enabled()) {
3600                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3601                                 "activate TXT before enabling KVM\n");
3602                         return 1;
3603                 }
3604                 /* launched w/o TXT and VMX disabled */
3605                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3606                         && !tboot_enabled())
3607                         return 1;
3608         }
3609
3610         return 0;
3611 }
3612
3613 static void kvm_cpu_vmxon(u64 addr)
3614 {
3615         cr4_set_bits(X86_CR4_VMXE);
3616         intel_pt_handle_vmx(1);
3617
3618         asm volatile (ASM_VMX_VMXON_RAX
3619                         : : "a"(&addr), "m"(addr)
3620                         : "memory", "cc");
3621 }
3622
3623 static int hardware_enable(void)
3624 {
3625         int cpu = raw_smp_processor_id();
3626         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3627         u64 old, test_bits;
3628
3629         if (cr4_read_shadow() & X86_CR4_VMXE)
3630                 return -EBUSY;
3631
3632         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3633         INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3634         spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3635
3636         /*
3637          * Now we can enable the vmclear operation in kdump
3638          * since the loaded_vmcss_on_cpu list on this cpu
3639          * has been initialized.
3640          *
3641          * Though the cpu is not in VMX operation now, there
3642          * is no problem to enable the vmclear operation
3643          * for the loaded_vmcss_on_cpu list is empty!
3644          */
3645         crash_enable_local_vmclear(cpu);
3646
3647         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3648
3649         test_bits = FEATURE_CONTROL_LOCKED;
3650         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3651         if (tboot_enabled())
3652                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3653
3654         if ((old & test_bits) != test_bits) {
3655                 /* enable and lock */
3656                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3657         }
3658         kvm_cpu_vmxon(phys_addr);
3659         ept_sync_global();
3660
3661         return 0;
3662 }
3663
3664 static void vmclear_local_loaded_vmcss(void)
3665 {
3666         int cpu = raw_smp_processor_id();
3667         struct loaded_vmcs *v, *n;
3668
3669         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3670                                  loaded_vmcss_on_cpu_link)
3671                 __loaded_vmcs_clear(v);
3672 }
3673
3674
3675 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3676  * tricks.
3677  */
3678 static void kvm_cpu_vmxoff(void)
3679 {
3680         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3681
3682         intel_pt_handle_vmx(0);
3683         cr4_clear_bits(X86_CR4_VMXE);
3684 }
3685
3686 static void hardware_disable(void)
3687 {
3688         vmclear_local_loaded_vmcss();
3689         kvm_cpu_vmxoff();
3690 }
3691
3692 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3693                                       u32 msr, u32 *result)
3694 {
3695         u32 vmx_msr_low, vmx_msr_high;
3696         u32 ctl = ctl_min | ctl_opt;
3697
3698         rdmsr(msr, vmx_msr_low, vmx_msr_high);
3699
3700         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3701         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
3702
3703         /* Ensure minimum (required) set of control bits are supported. */
3704         if (ctl_min & ~ctl)
3705                 return -EIO;
3706
3707         *result = ctl;
3708         return 0;
3709 }
3710
3711 static __init bool allow_1_setting(u32 msr, u32 ctl)
3712 {
3713         u32 vmx_msr_low, vmx_msr_high;
3714
3715         rdmsr(msr, vmx_msr_low, vmx_msr_high);
3716         return vmx_msr_high & ctl;
3717 }
3718
3719 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3720 {
3721         u32 vmx_msr_low, vmx_msr_high;
3722         u32 min, opt, min2, opt2;
3723         u32 _pin_based_exec_control = 0;
3724         u32 _cpu_based_exec_control = 0;
3725         u32 _cpu_based_2nd_exec_control = 0;
3726         u32 _vmexit_control = 0;
3727         u32 _vmentry_control = 0;
3728
3729         min = CPU_BASED_HLT_EXITING |
3730 #ifdef CONFIG_X86_64
3731               CPU_BASED_CR8_LOAD_EXITING |
3732               CPU_BASED_CR8_STORE_EXITING |
3733 #endif
3734               CPU_BASED_CR3_LOAD_EXITING |
3735               CPU_BASED_CR3_STORE_EXITING |
3736               CPU_BASED_USE_IO_BITMAPS |
3737               CPU_BASED_MOV_DR_EXITING |
3738               CPU_BASED_USE_TSC_OFFSETING |
3739               CPU_BASED_INVLPG_EXITING |
3740               CPU_BASED_RDPMC_EXITING;
3741
3742         if (!kvm_mwait_in_guest())
3743                 min |= CPU_BASED_MWAIT_EXITING |
3744                         CPU_BASED_MONITOR_EXITING;
3745
3746         opt = CPU_BASED_TPR_SHADOW |
3747               CPU_BASED_USE_MSR_BITMAPS |
3748               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3749         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3750                                 &_cpu_based_exec_control) < 0)
3751                 return -EIO;
3752 #ifdef CONFIG_X86_64
3753         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3754                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3755                                            ~CPU_BASED_CR8_STORE_EXITING;
3756 #endif
3757         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3758                 min2 = 0;
3759                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3760                         SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3761                         SECONDARY_EXEC_WBINVD_EXITING |
3762                         SECONDARY_EXEC_ENABLE_VPID |
3763                         SECONDARY_EXEC_ENABLE_EPT |
3764                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
3765                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3766                         SECONDARY_EXEC_RDTSCP |
3767                         SECONDARY_EXEC_ENABLE_INVPCID |
3768                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
3769                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3770                         SECONDARY_EXEC_SHADOW_VMCS |
3771                         SECONDARY_EXEC_XSAVES |
3772                         SECONDARY_EXEC_RDSEED |
3773                         SECONDARY_EXEC_RDRAND |
3774                         SECONDARY_EXEC_ENABLE_PML |
3775                         SECONDARY_EXEC_TSC_SCALING |
3776                         SECONDARY_EXEC_ENABLE_VMFUNC;
3777                 if (adjust_vmx_controls(min2, opt2,
3778                                         MSR_IA32_VMX_PROCBASED_CTLS2,
3779                                         &_cpu_based_2nd_exec_control) < 0)
3780                         return -EIO;
3781         }
3782 #ifndef CONFIG_X86_64
3783         if (!(_cpu_based_2nd_exec_control &
3784                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3785                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3786 #endif
3787
3788         if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3789                 _cpu_based_2nd_exec_control &= ~(
3790                                 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3791                                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3792                                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3793
3794         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3795                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3796                    enabled */
3797                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3798                                              CPU_BASED_CR3_STORE_EXITING |
3799                                              CPU_BASED_INVLPG_EXITING);
3800                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3801                       vmx_capability.ept, vmx_capability.vpid);
3802         }
3803
3804         min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3805 #ifdef CONFIG_X86_64
3806         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3807 #endif
3808         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3809                 VM_EXIT_CLEAR_BNDCFGS;
3810         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3811                                 &_vmexit_control) < 0)
3812                 return -EIO;
3813
3814         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3815         opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
3816                  PIN_BASED_VMX_PREEMPTION_TIMER;
3817         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3818                                 &_pin_based_exec_control) < 0)
3819                 return -EIO;
3820
3821         if (cpu_has_broken_vmx_preemption_timer())
3822                 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3823         if (!(_cpu_based_2nd_exec_control &
3824                 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3825                 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3826
3827         min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3828         opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3829         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3830                                 &_vmentry_control) < 0)
3831                 return -EIO;
3832
3833         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3834
3835         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3836         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3837                 return -EIO;
3838
3839 #ifdef CONFIG_X86_64
3840         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3841         if (vmx_msr_high & (1u<<16))
3842                 return -EIO;
3843 #endif
3844
3845         /* Require Write-Back (WB) memory type for VMCS accesses. */
3846         if (((vmx_msr_high >> 18) & 15) != 6)
3847                 return -EIO;
3848
3849         vmcs_conf->size = vmx_msr_high & 0x1fff;
3850         vmcs_conf->order = get_order(vmcs_conf->size);
3851         vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3852         vmcs_conf->revision_id = vmx_msr_low;
3853
3854         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3855         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3856         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3857         vmcs_conf->vmexit_ctrl         = _vmexit_control;
3858         vmcs_conf->vmentry_ctrl        = _vmentry_control;
3859
3860         cpu_has_load_ia32_efer =
3861                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3862                                 VM_ENTRY_LOAD_IA32_EFER)
3863                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3864                                    VM_EXIT_LOAD_IA32_EFER);
3865
3866         cpu_has_load_perf_global_ctrl =
3867                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3868                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3869                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3870                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3871
3872         /*
3873          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3874          * but due to errata below it can't be used. Workaround is to use
3875          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3876          *
3877          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3878          *
3879          * AAK155             (model 26)
3880          * AAP115             (model 30)
3881          * AAT100             (model 37)
3882          * BC86,AAY89,BD102   (model 44)
3883          * BA97               (model 46)
3884          *
3885          */
3886         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3887                 switch (boot_cpu_data.x86_model) {
3888                 case 26:
3889                 case 30:
3890                 case 37:
3891                 case 44:
3892                 case 46:
3893                         cpu_has_load_perf_global_ctrl = false;
3894                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3895                                         "does not work properly. Using workaround\n");
3896                         break;
3897                 default:
3898                         break;
3899                 }
3900         }
3901
3902         if (boot_cpu_has(X86_FEATURE_XSAVES))
3903                 rdmsrl(MSR_IA32_XSS, host_xss);
3904
3905         return 0;
3906 }
3907
3908 static struct vmcs *alloc_vmcs_cpu(int cpu)
3909 {
3910         int node = cpu_to_node(cpu);
3911         struct page *pages;
3912         struct vmcs *vmcs;
3913
3914         pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3915         if (!pages)
3916                 return NULL;
3917         vmcs = page_address(pages);
3918         memset(vmcs, 0, vmcs_config.size);
3919         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3920         return vmcs;
3921 }
3922
3923 static void free_vmcs(struct vmcs *vmcs)
3924 {
3925         free_pages((unsigned long)vmcs, vmcs_config.order);
3926 }
3927
3928 /*
3929  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3930  */
3931 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3932 {
3933         if (!loaded_vmcs->vmcs)
3934                 return;
3935         loaded_vmcs_clear(loaded_vmcs);
3936         free_vmcs(loaded_vmcs->vmcs);
3937         loaded_vmcs->vmcs = NULL;
3938         if (loaded_vmcs->msr_bitmap)
3939                 free_page((unsigned long)loaded_vmcs->msr_bitmap);
3940         WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3941 }
3942
3943 static struct vmcs *alloc_vmcs(void)
3944 {
3945         return alloc_vmcs_cpu(raw_smp_processor_id());
3946 }
3947
3948 static int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3949 {
3950         loaded_vmcs->vmcs = alloc_vmcs();
3951         if (!loaded_vmcs->vmcs)
3952                 return -ENOMEM;
3953
3954         loaded_vmcs->shadow_vmcs = NULL;
3955         loaded_vmcs_init(loaded_vmcs);
3956
3957         if (cpu_has_vmx_msr_bitmap()) {
3958                 loaded_vmcs->msr_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
3959                 if (!loaded_vmcs->msr_bitmap)
3960                         goto out_vmcs;
3961                 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
3962         }
3963         return 0;
3964
3965 out_vmcs:
3966         free_loaded_vmcs(loaded_vmcs);
3967         return -ENOMEM;
3968 }
3969
3970 static void free_kvm_area(void)
3971 {
3972         int cpu;
3973
3974         for_each_possible_cpu(cpu) {
3975                 free_vmcs(per_cpu(vmxarea, cpu));
3976                 per_cpu(vmxarea, cpu) = NULL;
3977         }
3978 }
3979
3980 enum vmcs_field_type {
3981         VMCS_FIELD_TYPE_U16 = 0,
3982         VMCS_FIELD_TYPE_U64 = 1,
3983         VMCS_FIELD_TYPE_U32 = 2,
3984         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
3985 };
3986
3987 static inline int vmcs_field_type(unsigned long field)
3988 {
3989         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
3990                 return VMCS_FIELD_TYPE_U32;
3991         return (field >> 13) & 0x3 ;
3992 }
3993
3994 static inline int vmcs_field_readonly(unsigned long field)
3995 {
3996         return (((field >> 10) & 0x3) == 1);
3997 }
3998
3999 static void init_vmcs_shadow_fields(void)
4000 {
4001         int i, j;
4002
4003         /* No checks for read only fields yet */
4004
4005         for (i = j = 0; i < max_shadow_read_write_fields; i++) {
4006                 switch (shadow_read_write_fields[i]) {
4007                 case GUEST_BNDCFGS:
4008                         if (!kvm_mpx_supported())
4009                                 continue;
4010                         break;
4011                 default:
4012                         break;
4013                 }
4014
4015                 if (j < i)
4016                         shadow_read_write_fields[j] =
4017                                 shadow_read_write_fields[i];
4018                 j++;
4019         }
4020         max_shadow_read_write_fields = j;
4021
4022         /* shadowed fields guest access without vmexit */
4023         for (i = 0; i < max_shadow_read_write_fields; i++) {
4024                 unsigned long field = shadow_read_write_fields[i];
4025
4026                 clear_bit(field, vmx_vmwrite_bitmap);
4027                 clear_bit(field, vmx_vmread_bitmap);
4028                 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64) {
4029                         clear_bit(field + 1, vmx_vmwrite_bitmap);
4030                         clear_bit(field + 1, vmx_vmread_bitmap);
4031                 }
4032         }
4033         for (i = 0; i < max_shadow_read_only_fields; i++) {
4034                 unsigned long field = shadow_read_only_fields[i];
4035
4036                 clear_bit(field, vmx_vmread_bitmap);
4037                 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64)
4038                         clear_bit(field + 1, vmx_vmread_bitmap);
4039         }
4040 }
4041
4042 static __init int alloc_kvm_area(void)
4043 {
4044         int cpu;
4045
4046         for_each_possible_cpu(cpu) {
4047                 struct vmcs *vmcs;
4048
4049                 vmcs = alloc_vmcs_cpu(cpu);
4050                 if (!vmcs) {
4051                         free_kvm_area();
4052                         return -ENOMEM;
4053                 }
4054
4055                 per_cpu(vmxarea, cpu) = vmcs;
4056         }
4057         return 0;
4058 }
4059
4060 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
4061                 struct kvm_segment *save)
4062 {
4063         if (!emulate_invalid_guest_state) {
4064                 /*
4065                  * CS and SS RPL should be equal during guest entry according
4066                  * to VMX spec, but in reality it is not always so. Since vcpu
4067                  * is in the middle of the transition from real mode to
4068                  * protected mode it is safe to assume that RPL 0 is a good
4069                  * default value.
4070                  */
4071                 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4072                         save->selector &= ~SEGMENT_RPL_MASK;
4073                 save->dpl = save->selector & SEGMENT_RPL_MASK;
4074                 save->s = 1;
4075         }
4076         vmx_set_segment(vcpu, save, seg);
4077 }
4078
4079 static void enter_pmode(struct kvm_vcpu *vcpu)
4080 {
4081         unsigned long flags;
4082         struct vcpu_vmx *vmx = to_vmx(vcpu);
4083
4084         /*
4085          * Update real mode segment cache. It may be not up-to-date if sement
4086          * register was written while vcpu was in a guest mode.
4087          */
4088         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4089         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4090         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4091         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4092         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4093         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4094
4095         vmx->rmode.vm86_active = 0;
4096
4097         vmx_segment_cache_clear(vmx);
4098
4099         vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4100
4101         flags = vmcs_readl(GUEST_RFLAGS);
4102         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
4103         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
4104         vmcs_writel(GUEST_RFLAGS, flags);
4105
4106         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
4107                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
4108
4109         update_exception_bitmap(vcpu);
4110
4111         fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4112         fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4113         fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4114         fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4115         fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4116         fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4117 }
4118
4119 static void fix_rmode_seg(int seg, struct kvm_segment *save)
4120 {
4121         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4122         struct kvm_segment var = *save;
4123
4124         var.dpl = 0x3;
4125         if (seg == VCPU_SREG_CS)
4126                 var.type = 0x3;
4127
4128         if (!emulate_invalid_guest_state) {
4129                 var.selector = var.base >> 4;
4130                 var.base = var.base & 0xffff0;
4131                 var.limit = 0xffff;
4132                 var.g = 0;
4133                 var.db = 0;
4134                 var.present = 1;
4135                 var.s = 1;
4136                 var.l = 0;
4137                 var.unusable = 0;
4138                 var.type = 0x3;
4139                 var.avl = 0;
4140                 if (save->base & 0xf)
4141                         printk_once(KERN_WARNING "kvm: segment base is not "
4142                                         "paragraph aligned when entering "
4143                                         "protected mode (seg=%d)", seg);
4144         }
4145
4146         vmcs_write16(sf->selector, var.selector);
4147         vmcs_writel(sf->base, var.base);
4148         vmcs_write32(sf->limit, var.limit);
4149         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4150 }
4151
4152 static void enter_rmode(struct kvm_vcpu *vcpu)
4153 {
4154         unsigned long flags;
4155         struct vcpu_vmx *vmx = to_vmx(vcpu);
4156
4157         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4158         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4159         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4160         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4161         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4162         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4163         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4164
4165         vmx->rmode.vm86_active = 1;
4166
4167         /*
4168          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4169          * vcpu. Warn the user that an update is overdue.
4170          */
4171         if (!vcpu->kvm->arch.tss_addr)
4172                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4173                              "called before entering vcpu\n");
4174
4175         vmx_segment_cache_clear(vmx);
4176
4177         vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
4178         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4179         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4180
4181         flags = vmcs_readl(GUEST_RFLAGS);
4182         vmx->rmode.save_rflags = flags;
4183
4184         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4185
4186         vmcs_writel(GUEST_RFLAGS, flags);
4187         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4188         update_exception_bitmap(vcpu);
4189
4190         fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4191         fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4192         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4193         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4194         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4195         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4196
4197         kvm_mmu_reset_context(vcpu);
4198 }
4199
4200 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4201 {
4202         struct vcpu_vmx *vmx = to_vmx(vcpu);
4203         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4204
4205         if (!msr)
4206                 return;
4207
4208         /*
4209          * Force kernel_gs_base reloading before EFER changes, as control
4210          * of this msr depends on is_long_mode().
4211          */
4212         vmx_load_host_state(to_vmx(vcpu));
4213         vcpu->arch.efer = efer;
4214         if (efer & EFER_LMA) {
4215                 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4216                 msr->data = efer;
4217         } else {
4218                 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4219
4220                 msr->data = efer & ~EFER_LME;
4221         }
4222         setup_msrs(vmx);
4223 }
4224
4225 #ifdef CONFIG_X86_64
4226
4227 static void enter_lmode(struct kvm_vcpu *vcpu)
4228 {
4229         u32 guest_tr_ar;
4230
4231         vmx_segment_cache_clear(to_vmx(vcpu));
4232
4233         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4234         if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4235                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4236                                      __func__);
4237                 vmcs_write32(GUEST_TR_AR_BYTES,
4238                              (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4239                              | VMX_AR_TYPE_BUSY_64_TSS);
4240         }
4241         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4242 }
4243
4244 static void exit_lmode(struct kvm_vcpu *vcpu)
4245 {
4246         vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4247         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4248 }
4249
4250 #endif
4251
4252 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4253 {
4254         if (enable_ept) {
4255                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4256                         return;
4257                 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4258         } else {
4259                 vpid_sync_context(vpid);
4260         }
4261 }
4262
4263 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4264 {
4265         __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4266 }
4267
4268 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4269 {
4270         if (enable_ept)
4271                 vmx_flush_tlb(vcpu);
4272 }
4273
4274 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4275 {
4276         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4277
4278         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4279         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4280 }
4281
4282 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4283 {
4284         if (enable_ept && is_paging(vcpu))
4285                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4286         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4287 }
4288
4289 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4290 {
4291         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4292
4293         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4294         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4295 }
4296
4297 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4298 {
4299         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4300
4301         if (!test_bit(VCPU_EXREG_PDPTR,
4302                       (unsigned long *)&vcpu->arch.regs_dirty))
4303                 return;
4304
4305         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4306                 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4307                 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4308                 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4309                 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4310         }
4311 }
4312
4313 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4314 {
4315         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4316
4317         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4318                 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4319                 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4320                 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4321                 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4322         }
4323
4324         __set_bit(VCPU_EXREG_PDPTR,
4325                   (unsigned long *)&vcpu->arch.regs_avail);
4326         __set_bit(VCPU_EXREG_PDPTR,
4327                   (unsigned long *)&vcpu->arch.regs_dirty);
4328 }
4329
4330 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4331 {
4332         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4333         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4334         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4335
4336         if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4337                 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4338             nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4339                 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4340
4341         return fixed_bits_valid(val, fixed0, fixed1);
4342 }
4343
4344 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4345 {
4346         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4347         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4348
4349         return fixed_bits_valid(val, fixed0, fixed1);
4350 }
4351
4352 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4353 {
4354         u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4355         u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4356
4357         return fixed_bits_valid(val, fixed0, fixed1);
4358 }
4359
4360 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4361 #define nested_guest_cr4_valid  nested_cr4_valid
4362 #define nested_host_cr4_valid   nested_cr4_valid
4363
4364 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4365
4366 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4367                                         unsigned long cr0,
4368                                         struct kvm_vcpu *vcpu)
4369 {
4370         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4371                 vmx_decache_cr3(vcpu);
4372         if (!(cr0 & X86_CR0_PG)) {
4373                 /* From paging/starting to nonpaging */
4374                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4375                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4376                              (CPU_BASED_CR3_LOAD_EXITING |
4377                               CPU_BASED_CR3_STORE_EXITING));
4378                 vcpu->arch.cr0 = cr0;
4379                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4380         } else if (!is_paging(vcpu)) {
4381                 /* From nonpaging to paging */
4382                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4383                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4384                              ~(CPU_BASED_CR3_LOAD_EXITING |
4385                                CPU_BASED_CR3_STORE_EXITING));
4386                 vcpu->arch.cr0 = cr0;
4387                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4388         }
4389
4390         if (!(cr0 & X86_CR0_WP))
4391                 *hw_cr0 &= ~X86_CR0_WP;
4392 }
4393
4394 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4395 {
4396         struct vcpu_vmx *vmx = to_vmx(vcpu);
4397         unsigned long hw_cr0;
4398
4399         hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4400         if (enable_unrestricted_guest)
4401                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4402         else {
4403                 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4404
4405                 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4406                         enter_pmode(vcpu);
4407
4408                 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4409                         enter_rmode(vcpu);
4410         }
4411
4412 #ifdef CONFIG_X86_64
4413         if (vcpu->arch.efer & EFER_LME) {
4414                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4415                         enter_lmode(vcpu);
4416                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4417                         exit_lmode(vcpu);
4418         }
4419 #endif
4420
4421         if (enable_ept)
4422                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4423
4424         vmcs_writel(CR0_READ_SHADOW, cr0);
4425         vmcs_writel(GUEST_CR0, hw_cr0);
4426         vcpu->arch.cr0 = cr0;
4427
4428         /* depends on vcpu->arch.cr0 to be set to a new value */
4429         vmx->emulation_required = emulation_required(vcpu);
4430 }
4431
4432 static int get_ept_level(struct kvm_vcpu *vcpu)
4433 {
4434         if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
4435                 return 5;
4436         return 4;
4437 }
4438
4439 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4440 {
4441         u64 eptp = VMX_EPTP_MT_WB;
4442
4443         eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
4444
4445         if (enable_ept_ad_bits &&
4446             (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4447                 eptp |= VMX_EPTP_AD_ENABLE_BIT;
4448         eptp |= (root_hpa & PAGE_MASK);
4449
4450         return eptp;
4451 }
4452
4453 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4454 {
4455         unsigned long guest_cr3;
4456         u64 eptp;
4457
4458         guest_cr3 = cr3;
4459         if (enable_ept) {
4460                 eptp = construct_eptp(vcpu, cr3);
4461                 vmcs_write64(EPT_POINTER, eptp);
4462                 if (is_paging(vcpu) || is_guest_mode(vcpu))
4463                         guest_cr3 = kvm_read_cr3(vcpu);
4464                 else
4465                         guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4466                 ept_load_pdptrs(vcpu);
4467         }
4468
4469         vmx_flush_tlb(vcpu);
4470         vmcs_writel(GUEST_CR3, guest_cr3);
4471 }
4472
4473 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4474 {
4475         /*
4476          * Pass through host's Machine Check Enable value to hw_cr4, which
4477          * is in force while we are in guest mode.  Do not let guests control
4478          * this bit, even if host CR4.MCE == 0.
4479          */
4480         unsigned long hw_cr4 =
4481                 (cr4_read_shadow() & X86_CR4_MCE) |
4482                 (cr4 & ~X86_CR4_MCE) |
4483                 (to_vmx(vcpu)->rmode.vm86_active ?
4484                  KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4485
4486         if (cr4 & X86_CR4_VMXE) {
4487                 /*
4488                  * To use VMXON (and later other VMX instructions), a guest
4489                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
4490                  * So basically the check on whether to allow nested VMX
4491                  * is here.
4492                  */
4493                 if (!nested_vmx_allowed(vcpu))
4494                         return 1;
4495         }
4496
4497         if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4498                 return 1;
4499
4500         vcpu->arch.cr4 = cr4;
4501         if (enable_ept) {
4502                 if (!is_paging(vcpu)) {
4503                         hw_cr4 &= ~X86_CR4_PAE;
4504                         hw_cr4 |= X86_CR4_PSE;
4505                 } else if (!(cr4 & X86_CR4_PAE)) {
4506                         hw_cr4 &= ~X86_CR4_PAE;
4507                 }
4508         }
4509
4510         if (!enable_unrestricted_guest && !is_paging(vcpu))
4511                 /*
4512                  * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4513                  * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
4514                  * to be manually disabled when guest switches to non-paging
4515                  * mode.
4516                  *
4517                  * If !enable_unrestricted_guest, the CPU is always running
4518                  * with CR0.PG=1 and CR4 needs to be modified.
4519                  * If enable_unrestricted_guest, the CPU automatically
4520                  * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4521                  */
4522                 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4523
4524         vmcs_writel(CR4_READ_SHADOW, cr4);
4525         vmcs_writel(GUEST_CR4, hw_cr4);
4526         return 0;
4527 }
4528
4529 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4530                             struct kvm_segment *var, int seg)
4531 {
4532         struct vcpu_vmx *vmx = to_vmx(vcpu);
4533         u32 ar;
4534
4535         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4536                 *var = vmx->rmode.segs[seg];
4537                 if (seg == VCPU_SREG_TR
4538                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4539                         return;
4540                 var->base = vmx_read_guest_seg_base(vmx, seg);
4541                 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4542                 return;
4543         }
4544         var->base = vmx_read_guest_seg_base(vmx, seg);
4545         var->limit = vmx_read_guest_seg_limit(vmx, seg);
4546         var->selector = vmx_read_guest_seg_selector(vmx, seg);
4547         ar = vmx_read_guest_seg_ar(vmx, seg);
4548         var->unusable = (ar >> 16) & 1;
4549         var->type = ar & 15;
4550         var->s = (ar >> 4) & 1;
4551         var->dpl = (ar >> 5) & 3;
4552         /*
4553          * Some userspaces do not preserve unusable property. Since usable
4554          * segment has to be present according to VMX spec we can use present
4555          * property to amend userspace bug by making unusable segment always
4556          * nonpresent. vmx_segment_access_rights() already marks nonpresent
4557          * segment as unusable.
4558          */
4559         var->present = !var->unusable;
4560         var->avl = (ar >> 12) & 1;
4561         var->l = (ar >> 13) & 1;
4562         var->db = (ar >> 14) & 1;
4563         var->g = (ar >> 15) & 1;
4564 }
4565
4566 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4567 {
4568         struct kvm_segment s;
4569
4570         if (to_vmx(vcpu)->rmode.vm86_active) {
4571                 vmx_get_segment(vcpu, &s, seg);
4572                 return s.base;
4573         }
4574         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4575 }
4576
4577 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4578 {
4579         struct vcpu_vmx *vmx = to_vmx(vcpu);
4580
4581         if (unlikely(vmx->rmode.vm86_active))
4582                 return 0;
4583         else {
4584                 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4585                 return VMX_AR_DPL(ar);
4586         }
4587 }
4588
4589 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4590 {
4591         u32 ar;
4592
4593         if (var->unusable || !var->present)
4594                 ar = 1 << 16;
4595         else {
4596                 ar = var->type & 15;
4597                 ar |= (var->s & 1) << 4;
4598                 ar |= (var->dpl & 3) << 5;
4599                 ar |= (var->present & 1) << 7;
4600                 ar |= (var->avl & 1) << 12;
4601                 ar |= (var->l & 1) << 13;
4602                 ar |= (var->db & 1) << 14;
4603                 ar |= (var->g & 1) << 15;
4604         }
4605
4606         return ar;
4607 }
4608
4609 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4610                             struct kvm_segment *var, int seg)
4611 {
4612         struct vcpu_vmx *vmx = to_vmx(vcpu);
4613         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4614
4615         vmx_segment_cache_clear(vmx);
4616
4617         if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4618                 vmx->rmode.segs[seg] = *var;
4619                 if (seg == VCPU_SREG_TR)
4620                         vmcs_write16(sf->selector, var->selector);
4621                 else if (var->s)
4622                         fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4623                 goto out;
4624         }
4625
4626         vmcs_writel(sf->base, var->base);
4627         vmcs_write32(sf->limit, var->limit);
4628         vmcs_write16(sf->selector, var->selector);
4629
4630         /*
4631          *   Fix the "Accessed" bit in AR field of segment registers for older
4632          * qemu binaries.
4633          *   IA32 arch specifies that at the time of processor reset the
4634          * "Accessed" bit in the AR field of segment registers is 1. And qemu
4635          * is setting it to 0 in the userland code. This causes invalid guest
4636          * state vmexit when "unrestricted guest" mode is turned on.
4637          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
4638          * tree. Newer qemu binaries with that qemu fix would not need this
4639          * kvm hack.
4640          */
4641         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4642                 var->type |= 0x1; /* Accessed */
4643
4644         vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4645
4646 out:
4647         vmx->emulation_required = emulation_required(vcpu);
4648 }
4649
4650 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4651 {
4652         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4653
4654         *db = (ar >> 14) & 1;
4655         *l = (ar >> 13) & 1;
4656 }
4657
4658 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4659 {
4660         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4661         dt->address = vmcs_readl(GUEST_IDTR_BASE);
4662 }
4663
4664 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4665 {
4666         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4667         vmcs_writel(GUEST_IDTR_BASE, dt->address);
4668 }
4669
4670 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4671 {
4672         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4673         dt->address = vmcs_readl(GUEST_GDTR_BASE);
4674 }
4675
4676 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4677 {
4678         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4679         vmcs_writel(GUEST_GDTR_BASE, dt->address);
4680 }
4681
4682 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4683 {
4684         struct kvm_segment var;
4685         u32 ar;
4686
4687         vmx_get_segment(vcpu, &var, seg);
4688         var.dpl = 0x3;
4689         if (seg == VCPU_SREG_CS)
4690                 var.type = 0x3;
4691         ar = vmx_segment_access_rights(&var);
4692
4693         if (var.base != (var.selector << 4))
4694                 return false;
4695         if (var.limit != 0xffff)
4696                 return false;
4697         if (ar != 0xf3)
4698                 return false;
4699
4700         return true;
4701 }
4702
4703 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4704 {
4705         struct kvm_segment cs;
4706         unsigned int cs_rpl;
4707
4708         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4709         cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4710
4711         if (cs.unusable)
4712                 return false;
4713         if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4714                 return false;
4715         if (!cs.s)
4716                 return false;
4717         if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4718                 if (cs.dpl > cs_rpl)
4719                         return false;
4720         } else {
4721                 if (cs.dpl != cs_rpl)
4722                         return false;
4723         }
4724         if (!cs.present)
4725                 return false;
4726
4727         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4728         return true;
4729 }
4730
4731 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4732 {
4733         struct kvm_segment ss;
4734         unsigned int ss_rpl;
4735
4736         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4737         ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4738
4739         if (ss.unusable)
4740                 return true;
4741         if (ss.type != 3 && ss.type != 7)
4742                 return false;
4743         if (!ss.s)
4744                 return false;
4745         if (ss.dpl != ss_rpl) /* DPL != RPL */
4746                 return false;
4747         if (!ss.present)
4748                 return false;
4749
4750         return true;
4751 }
4752
4753 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4754 {
4755         struct kvm_segment var;
4756         unsigned int rpl;
4757
4758         vmx_get_segment(vcpu, &var, seg);
4759         rpl = var.selector & SEGMENT_RPL_MASK;
4760
4761         if (var.unusable)
4762                 return true;
4763         if (!var.s)
4764                 return false;
4765         if (!var.present)
4766                 return false;
4767         if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4768                 if (var.dpl < rpl) /* DPL < RPL */
4769                         return false;
4770         }
4771
4772         /* TODO: Add other members to kvm_segment_field to allow checking for other access
4773          * rights flags
4774          */
4775         return true;
4776 }
4777
4778 static bool tr_valid(struct kvm_vcpu *vcpu)
4779 {
4780         struct kvm_segment tr;
4781
4782         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4783
4784         if (tr.unusable)
4785                 return false;
4786         if (tr.selector & SEGMENT_TI_MASK)      /* TI = 1 */
4787                 return false;
4788         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4789                 return false;
4790         if (!tr.present)
4791                 return false;
4792
4793         return true;
4794 }
4795
4796 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4797 {
4798         struct kvm_segment ldtr;
4799
4800         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4801
4802         if (ldtr.unusable)
4803                 return true;
4804         if (ldtr.selector & SEGMENT_TI_MASK)    /* TI = 1 */
4805                 return false;
4806         if (ldtr.type != 2)
4807                 return false;
4808         if (!ldtr.present)
4809                 return false;
4810
4811         return true;
4812 }
4813
4814 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4815 {
4816         struct kvm_segment cs, ss;
4817
4818         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4819         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4820
4821         return ((cs.selector & SEGMENT_RPL_MASK) ==
4822                  (ss.selector & SEGMENT_RPL_MASK));
4823 }
4824
4825 /*
4826  * Check if guest state is valid. Returns true if valid, false if
4827  * not.
4828  * We assume that registers are always usable
4829  */
4830 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4831 {
4832         if (enable_unrestricted_guest)
4833                 return true;
4834
4835         /* real mode guest state checks */
4836         if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4837                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4838                         return false;
4839                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4840                         return false;
4841                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4842                         return false;
4843                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4844                         return false;
4845                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4846                         return false;
4847                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4848                         return false;
4849         } else {
4850         /* protected mode guest state checks */
4851                 if (!cs_ss_rpl_check(vcpu))
4852                         return false;
4853                 if (!code_segment_valid(vcpu))
4854                         return false;
4855                 if (!stack_segment_valid(vcpu))
4856                         return false;
4857                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4858                         return false;
4859                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4860                         return false;
4861                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4862                         return false;
4863                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4864                         return false;
4865                 if (!tr_valid(vcpu))
4866                         return false;
4867                 if (!ldtr_valid(vcpu))
4868                         return false;
4869         }
4870         /* TODO:
4871          * - Add checks on RIP
4872          * - Add checks on RFLAGS
4873          */
4874
4875         return true;
4876 }
4877
4878 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
4879 {
4880         return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
4881 }
4882
4883 static int init_rmode_tss(struct kvm *kvm)
4884 {
4885         gfn_t fn;
4886         u16 data = 0;
4887         int idx, r;
4888
4889         idx = srcu_read_lock(&kvm->srcu);
4890         fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4891         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4892         if (r < 0)
4893                 goto out;
4894         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4895         r = kvm_write_guest_page(kvm, fn++, &data,
4896                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
4897         if (r < 0)
4898                 goto out;
4899         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4900         if (r < 0)
4901                 goto out;
4902         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4903         if (r < 0)
4904                 goto out;
4905         data = ~0;
4906         r = kvm_write_guest_page(kvm, fn, &data,
4907                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4908                                  sizeof(u8));
4909 out:
4910         srcu_read_unlock(&kvm->srcu, idx);
4911         return r;
4912 }
4913
4914 static int init_rmode_identity_map(struct kvm *kvm)
4915 {
4916         int i, idx, r = 0;
4917         kvm_pfn_t identity_map_pfn;
4918         u32 tmp;
4919
4920         if (!enable_ept)
4921                 return 0;
4922
4923         /* Protect kvm->arch.ept_identity_pagetable_done. */
4924         mutex_lock(&kvm->slots_lock);
4925
4926         if (likely(kvm->arch.ept_identity_pagetable_done))
4927                 goto out2;
4928
4929         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4930
4931         r = alloc_identity_pagetable(kvm);
4932         if (r < 0)
4933                 goto out2;
4934
4935         idx = srcu_read_lock(&kvm->srcu);
4936         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4937         if (r < 0)
4938                 goto out;
4939         /* Set up identity-mapping pagetable for EPT in real mode */
4940         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4941                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4942                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4943                 r = kvm_write_guest_page(kvm, identity_map_pfn,
4944                                 &tmp, i * sizeof(tmp), sizeof(tmp));
4945                 if (r < 0)
4946                         goto out;
4947         }
4948         kvm->arch.ept_identity_pagetable_done = true;
4949
4950 out:
4951         srcu_read_unlock(&kvm->srcu, idx);
4952
4953 out2:
4954         mutex_unlock(&kvm->slots_lock);
4955         return r;
4956 }
4957
4958 static void seg_setup(int seg)
4959 {
4960         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4961         unsigned int ar;
4962
4963         vmcs_write16(sf->selector, 0);
4964         vmcs_writel(sf->base, 0);
4965         vmcs_write32(sf->limit, 0xffff);
4966         ar = 0x93;
4967         if (seg == VCPU_SREG_CS)
4968                 ar |= 0x08; /* code segment */
4969
4970         vmcs_write32(sf->ar_bytes, ar);
4971 }
4972
4973 static int alloc_apic_access_page(struct kvm *kvm)
4974 {
4975         struct page *page;
4976         int r = 0;
4977
4978         mutex_lock(&kvm->slots_lock);
4979         if (kvm->arch.apic_access_page_done)
4980                 goto out;
4981         r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4982                                     APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4983         if (r)
4984                 goto out;
4985
4986         page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4987         if (is_error_page(page)) {
4988                 r = -EFAULT;
4989                 goto out;
4990         }
4991
4992         /*
4993          * Do not pin the page in memory, so that memory hot-unplug
4994          * is able to migrate it.
4995          */
4996         put_page(page);
4997         kvm->arch.apic_access_page_done = true;
4998 out:
4999         mutex_unlock(&kvm->slots_lock);
5000         return r;
5001 }
5002
5003 static int alloc_identity_pagetable(struct kvm *kvm)
5004 {
5005         /* Called with kvm->slots_lock held. */
5006
5007         int r = 0;
5008
5009         BUG_ON(kvm->arch.ept_identity_pagetable_done);
5010
5011         r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
5012                                     kvm->arch.ept_identity_map_addr, PAGE_SIZE);
5013
5014         return r;
5015 }
5016
5017 static int allocate_vpid(void)
5018 {
5019         int vpid;
5020
5021         if (!enable_vpid)
5022                 return 0;
5023         spin_lock(&vmx_vpid_lock);
5024         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
5025         if (vpid < VMX_NR_VPIDS)
5026                 __set_bit(vpid, vmx_vpid_bitmap);
5027         else
5028                 vpid = 0;
5029         spin_unlock(&vmx_vpid_lock);
5030         return vpid;
5031 }
5032
5033 static void free_vpid(int vpid)
5034 {
5035         if (!enable_vpid || vpid == 0)
5036                 return;
5037         spin_lock(&vmx_vpid_lock);
5038         __clear_bit(vpid, vmx_vpid_bitmap);
5039         spin_unlock(&vmx_vpid_lock);
5040 }
5041
5042 static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
5043                                                           u32 msr, int type)
5044 {
5045         int f = sizeof(unsigned long);
5046
5047         if (!cpu_has_vmx_msr_bitmap())
5048                 return;
5049
5050         /*
5051          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5052          * have the write-low and read-high bitmap offsets the wrong way round.
5053          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5054          */
5055         if (msr <= 0x1fff) {
5056                 if (type & MSR_TYPE_R)
5057                         /* read-low */
5058                         __clear_bit(msr, msr_bitmap + 0x000 / f);
5059
5060                 if (type & MSR_TYPE_W)
5061                         /* write-low */
5062                         __clear_bit(msr, msr_bitmap + 0x800 / f);
5063
5064         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5065                 msr &= 0x1fff;
5066                 if (type & MSR_TYPE_R)
5067                         /* read-high */
5068                         __clear_bit(msr, msr_bitmap + 0x400 / f);
5069
5070                 if (type & MSR_TYPE_W)
5071                         /* write-high */
5072                         __clear_bit(msr, msr_bitmap + 0xc00 / f);
5073
5074         }
5075 }
5076
5077 static void __always_inline vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
5078                                                          u32 msr, int type)
5079 {
5080         int f = sizeof(unsigned long);
5081
5082         if (!cpu_has_vmx_msr_bitmap())
5083                 return;
5084
5085         /*
5086          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5087          * have the write-low and read-high bitmap offsets the wrong way round.
5088          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5089          */
5090         if (msr <= 0x1fff) {
5091                 if (type & MSR_TYPE_R)
5092                         /* read-low */
5093                         __set_bit(msr, msr_bitmap + 0x000 / f);
5094
5095                 if (type & MSR_TYPE_W)
5096                         /* write-low */
5097                         __set_bit(msr, msr_bitmap + 0x800 / f);
5098
5099         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5100                 msr &= 0x1fff;
5101                 if (type & MSR_TYPE_R)
5102                         /* read-high */
5103                         __set_bit(msr, msr_bitmap + 0x400 / f);
5104
5105                 if (type & MSR_TYPE_W)
5106                         /* write-high */
5107                         __set_bit(msr, msr_bitmap + 0xc00 / f);
5108
5109         }
5110 }
5111
5112 static void __always_inline vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
5113                                                       u32 msr, int type, bool value)
5114 {
5115         if (value)
5116                 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
5117         else
5118                 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
5119 }
5120
5121 /*
5122  * If a msr is allowed by L0, we should check whether it is allowed by L1.
5123  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
5124  */
5125 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
5126                                                unsigned long *msr_bitmap_nested,
5127                                                u32 msr, int type)
5128 {
5129         int f = sizeof(unsigned long);
5130
5131         if (!cpu_has_vmx_msr_bitmap()) {
5132                 WARN_ON(1);
5133                 return;
5134         }
5135
5136         /*
5137          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5138          * have the write-low and read-high bitmap offsets the wrong way round.
5139          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5140          */
5141         if (msr <= 0x1fff) {
5142                 if (type & MSR_TYPE_R &&
5143                    !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
5144                         /* read-low */
5145                         __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
5146
5147                 if (type & MSR_TYPE_W &&
5148                    !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
5149                         /* write-low */
5150                         __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
5151
5152         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5153                 msr &= 0x1fff;
5154                 if (type & MSR_TYPE_R &&
5155                    !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
5156                         /* read-high */
5157                         __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
5158
5159                 if (type & MSR_TYPE_W &&
5160                    !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
5161                         /* write-high */
5162                         __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
5163
5164         }
5165 }
5166
5167 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
5168 {
5169         u8 mode = 0;
5170
5171         if (cpu_has_secondary_exec_ctrls() &&
5172             (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
5173              SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
5174                 mode |= MSR_BITMAP_MODE_X2APIC;
5175                 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
5176                         mode |= MSR_BITMAP_MODE_X2APIC_APICV;
5177         }
5178
5179         if (is_long_mode(vcpu))
5180                 mode |= MSR_BITMAP_MODE_LM;
5181
5182         return mode;
5183 }
5184
5185 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
5186
5187 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
5188                                          u8 mode)
5189 {
5190         int msr;
5191
5192         for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
5193                 unsigned word = msr / BITS_PER_LONG;
5194                 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
5195                 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
5196         }
5197
5198         if (mode & MSR_BITMAP_MODE_X2APIC) {
5199                 /*
5200                  * TPR reads and writes can be virtualized even if virtual interrupt
5201                  * delivery is not in use.
5202                  */
5203                 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
5204                 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
5205                         vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
5206                         vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
5207                         vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
5208                 }
5209         }
5210 }
5211
5212 static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
5213 {
5214         struct vcpu_vmx *vmx = to_vmx(vcpu);
5215         unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
5216         u8 mode = vmx_msr_bitmap_mode(vcpu);
5217         u8 changed = mode ^ vmx->msr_bitmap_mode;
5218
5219         if (!changed)
5220                 return;
5221
5222         vmx_set_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW,
5223                                   !(mode & MSR_BITMAP_MODE_LM));
5224
5225         if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
5226                 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
5227
5228         vmx->msr_bitmap_mode = mode;
5229 }
5230
5231 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
5232 {
5233         return enable_apicv;
5234 }
5235
5236 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
5237 {
5238         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5239         gfn_t gfn;
5240
5241         /*
5242          * Don't need to mark the APIC access page dirty; it is never
5243          * written to by the CPU during APIC virtualization.
5244          */
5245
5246         if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
5247                 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
5248                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5249         }
5250
5251         if (nested_cpu_has_posted_intr(vmcs12)) {
5252                 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
5253                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5254         }
5255 }
5256
5257
5258 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5259 {
5260         struct vcpu_vmx *vmx = to_vmx(vcpu);
5261         int max_irr;
5262         void *vapic_page;
5263         u16 status;
5264
5265         if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
5266                 return;
5267
5268         vmx->nested.pi_pending = false;
5269         if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5270                 return;
5271
5272         max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
5273         if (max_irr != 256) {
5274                 vapic_page = kmap(vmx->nested.virtual_apic_page);
5275                 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
5276                 kunmap(vmx->nested.virtual_apic_page);
5277
5278                 status = vmcs_read16(GUEST_INTR_STATUS);
5279                 if ((u8)max_irr > ((u8)status & 0xff)) {
5280                         status &= ~0xff;
5281                         status |= (u8)max_irr;
5282                         vmcs_write16(GUEST_INTR_STATUS, status);
5283                 }
5284         }
5285
5286         nested_mark_vmcs12_pages_dirty(vcpu);
5287 }
5288
5289 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5290                                                      bool nested)
5291 {
5292 #ifdef CONFIG_SMP
5293         int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5294
5295         if (vcpu->mode == IN_GUEST_MODE) {
5296                 /*
5297                  * The vector of interrupt to be delivered to vcpu had
5298                  * been set in PIR before this function.
5299                  *
5300                  * Following cases will be reached in this block, and
5301                  * we always send a notification event in all cases as
5302                  * explained below.
5303                  *
5304                  * Case 1: vcpu keeps in non-root mode. Sending a
5305                  * notification event posts the interrupt to vcpu.
5306                  *
5307                  * Case 2: vcpu exits to root mode and is still
5308                  * runnable. PIR will be synced to vIRR before the
5309                  * next vcpu entry. Sending a notification event in
5310                  * this case has no effect, as vcpu is not in root
5311                  * mode.
5312                  *
5313                  * Case 3: vcpu exits to root mode and is blocked.
5314                  * vcpu_block() has already synced PIR to vIRR and
5315                  * never blocks vcpu if vIRR is not cleared. Therefore,
5316                  * a blocked vcpu here does not wait for any requested
5317                  * interrupts in PIR, and sending a notification event
5318                  * which has no effect is safe here.
5319                  */
5320
5321                 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5322                 return true;
5323         }
5324 #endif
5325         return false;
5326 }
5327
5328 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5329                                                 int vector)
5330 {
5331         struct vcpu_vmx *vmx = to_vmx(vcpu);
5332
5333         if (is_guest_mode(vcpu) &&
5334             vector == vmx->nested.posted_intr_nv) {
5335                 /*
5336                  * If a posted intr is not recognized by hardware,
5337                  * we will accomplish it in the next vmentry.
5338                  */
5339                 vmx->nested.pi_pending = true;
5340                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5341                 /* the PIR and ON have been set by L1. */
5342                 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
5343                         kvm_vcpu_kick(vcpu);
5344                 return 0;
5345         }
5346         return -1;
5347 }
5348 /*
5349  * Send interrupt to vcpu via posted interrupt way.
5350  * 1. If target vcpu is running(non-root mode), send posted interrupt
5351  * notification to vcpu and hardware will sync PIR to vIRR atomically.
5352  * 2. If target vcpu isn't running(root mode), kick it to pick up the
5353  * interrupt from PIR in next vmentry.
5354  */
5355 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5356 {
5357         struct vcpu_vmx *vmx = to_vmx(vcpu);
5358         int r;
5359
5360         r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5361         if (!r)
5362                 return;
5363
5364         if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5365                 return;
5366
5367         /* If a previous notification has sent the IPI, nothing to do.  */
5368         if (pi_test_and_set_on(&vmx->pi_desc))
5369                 return;
5370
5371         if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5372                 kvm_vcpu_kick(vcpu);
5373 }
5374
5375 /*
5376  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5377  * will not change in the lifetime of the guest.
5378  * Note that host-state that does change is set elsewhere. E.g., host-state
5379  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5380  */
5381 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5382 {
5383         u32 low32, high32;
5384         unsigned long tmpl;
5385         struct desc_ptr dt;
5386         unsigned long cr0, cr3, cr4;
5387
5388         cr0 = read_cr0();
5389         WARN_ON(cr0 & X86_CR0_TS);
5390         vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
5391
5392         /*
5393          * Save the most likely value for this task's CR3 in the VMCS.
5394          * We can't use __get_current_cr3_fast() because we're not atomic.
5395          */
5396         cr3 = __read_cr3();
5397         vmcs_writel(HOST_CR3, cr3);             /* 22.2.3  FIXME: shadow tables */
5398         vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
5399
5400         /* Save the most likely value for this task's CR4 in the VMCS. */
5401         cr4 = cr4_read_shadow();
5402         vmcs_writel(HOST_CR4, cr4);                     /* 22.2.3, 22.2.5 */
5403         vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
5404
5405         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
5406 #ifdef CONFIG_X86_64
5407         /*
5408          * Load null selectors, so we can avoid reloading them in
5409          * __vmx_load_host_state(), in case userspace uses the null selectors
5410          * too (the expected case).
5411          */
5412         vmcs_write16(HOST_DS_SELECTOR, 0);
5413         vmcs_write16(HOST_ES_SELECTOR, 0);
5414 #else
5415         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5416         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5417 #endif
5418         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
5419         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
5420
5421         store_idt(&dt);
5422         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
5423         vmx->host_idt_base = dt.address;
5424
5425         vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5426
5427         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5428         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5429         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5430         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
5431
5432         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5433                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5434                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5435         }
5436 }
5437
5438 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5439 {
5440         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5441         if (enable_ept)
5442                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5443         if (is_guest_mode(&vmx->vcpu))
5444                 vmx->vcpu.arch.cr4_guest_owned_bits &=
5445                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5446         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5447 }
5448
5449 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5450 {
5451         u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5452
5453         if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5454                 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5455         /* Enable the preemption timer dynamically */
5456         pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5457         return pin_based_exec_ctrl;
5458 }
5459
5460 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5461 {
5462         struct vcpu_vmx *vmx = to_vmx(vcpu);
5463
5464         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5465         if (cpu_has_secondary_exec_ctrls()) {
5466                 if (kvm_vcpu_apicv_active(vcpu))
5467                         vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5468                                       SECONDARY_EXEC_APIC_REGISTER_VIRT |
5469                                       SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5470                 else
5471                         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5472                                         SECONDARY_EXEC_APIC_REGISTER_VIRT |
5473                                         SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5474         }
5475
5476         if (cpu_has_vmx_msr_bitmap())
5477                 vmx_update_msr_bitmap(vcpu);
5478 }
5479
5480 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5481 {
5482         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5483
5484         if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5485                 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5486
5487         if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5488                 exec_control &= ~CPU_BASED_TPR_SHADOW;
5489 #ifdef CONFIG_X86_64
5490                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5491                                 CPU_BASED_CR8_LOAD_EXITING;
5492 #endif
5493         }
5494         if (!enable_ept)
5495                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5496                                 CPU_BASED_CR3_LOAD_EXITING  |
5497                                 CPU_BASED_INVLPG_EXITING;
5498         return exec_control;
5499 }
5500
5501 static bool vmx_rdrand_supported(void)
5502 {
5503         return vmcs_config.cpu_based_2nd_exec_ctrl &
5504                 SECONDARY_EXEC_RDRAND;
5505 }
5506
5507 static bool vmx_rdseed_supported(void)
5508 {
5509         return vmcs_config.cpu_based_2nd_exec_ctrl &
5510                 SECONDARY_EXEC_RDSEED;
5511 }
5512
5513 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
5514 {
5515         struct kvm_vcpu *vcpu = &vmx->vcpu;
5516
5517         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5518         if (!cpu_need_virtualize_apic_accesses(vcpu))
5519                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5520         if (vmx->vpid == 0)
5521                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5522         if (!enable_ept) {
5523                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5524                 enable_unrestricted_guest = 0;
5525                 /* Enable INVPCID for non-ept guests may cause performance regression. */
5526                 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5527         }
5528         if (!enable_unrestricted_guest)
5529                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5530         if (!ple_gap)
5531                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5532         if (!kvm_vcpu_apicv_active(vcpu))
5533                 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5534                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5535         exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5536         /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5537            (handle_vmptrld).
5538            We can NOT enable shadow_vmcs here because we don't have yet
5539            a current VMCS12
5540         */
5541         exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5542
5543         if (!enable_pml)
5544                 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5545
5546         if (vmx_xsaves_supported()) {
5547                 /* Exposing XSAVES only when XSAVE is exposed */
5548                 bool xsaves_enabled =
5549                         guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
5550                         guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
5551
5552                 if (!xsaves_enabled)
5553                         exec_control &= ~SECONDARY_EXEC_XSAVES;
5554
5555                 if (nested) {
5556                         if (xsaves_enabled)
5557                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5558                                         SECONDARY_EXEC_XSAVES;
5559                         else
5560                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5561                                         ~SECONDARY_EXEC_XSAVES;
5562                 }
5563         }
5564
5565         if (vmx_rdtscp_supported()) {
5566                 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
5567                 if (!rdtscp_enabled)
5568                         exec_control &= ~SECONDARY_EXEC_RDTSCP;
5569
5570                 if (nested) {
5571                         if (rdtscp_enabled)
5572                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5573                                         SECONDARY_EXEC_RDTSCP;
5574                         else
5575                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5576                                         ~SECONDARY_EXEC_RDTSCP;
5577                 }
5578         }
5579
5580         if (vmx_invpcid_supported()) {
5581                 /* Exposing INVPCID only when PCID is exposed */
5582                 bool invpcid_enabled =
5583                         guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
5584                         guest_cpuid_has(vcpu, X86_FEATURE_PCID);
5585
5586                 if (!invpcid_enabled) {
5587                         exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5588                         guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
5589                 }
5590
5591                 if (nested) {
5592                         if (invpcid_enabled)
5593                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5594                                         SECONDARY_EXEC_ENABLE_INVPCID;
5595                         else
5596                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5597                                         ~SECONDARY_EXEC_ENABLE_INVPCID;
5598                 }
5599         }
5600
5601         if (vmx_rdrand_supported()) {
5602                 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
5603                 if (rdrand_enabled)
5604                         exec_control &= ~SECONDARY_EXEC_RDRAND;
5605
5606                 if (nested) {
5607                         if (rdrand_enabled)
5608                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5609                                         SECONDARY_EXEC_RDRAND;
5610                         else
5611                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5612                                         ~SECONDARY_EXEC_RDRAND;
5613                 }
5614         }
5615
5616         if (vmx_rdseed_supported()) {
5617                 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
5618                 if (rdseed_enabled)
5619                         exec_control &= ~SECONDARY_EXEC_RDSEED;
5620
5621                 if (nested) {
5622                         if (rdseed_enabled)
5623                                 vmx->nested.nested_vmx_secondary_ctls_high |=
5624                                         SECONDARY_EXEC_RDSEED;
5625                         else
5626                                 vmx->nested.nested_vmx_secondary_ctls_high &=
5627                                         ~SECONDARY_EXEC_RDSEED;
5628                 }
5629         }
5630
5631         vmx->secondary_exec_control = exec_control;
5632 }
5633
5634 static void ept_set_mmio_spte_mask(void)
5635 {
5636         /*
5637          * EPT Misconfigurations can be generated if the value of bits 2:0
5638          * of an EPT paging-structure entry is 110b (write/execute).
5639          */
5640         kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
5641                                    VMX_EPT_MISCONFIG_WX_VALUE);
5642 }
5643
5644 #define VMX_XSS_EXIT_BITMAP 0
5645 /*
5646  * Sets up the vmcs for emulated real mode.
5647  */
5648 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5649 {
5650 #ifdef CONFIG_X86_64
5651         unsigned long a;
5652 #endif
5653         int i;
5654
5655         /* I/O */
5656         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5657         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5658
5659         if (enable_shadow_vmcs) {
5660                 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5661                 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5662         }
5663         if (cpu_has_vmx_msr_bitmap())
5664                 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
5665
5666         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5667
5668         /* Control */
5669         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5670         vmx->hv_deadline_tsc = -1;
5671
5672         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5673
5674         if (cpu_has_secondary_exec_ctrls()) {
5675                 vmx_compute_secondary_exec_control(vmx);
5676                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5677                              vmx->secondary_exec_control);
5678         }
5679
5680         if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5681                 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5682                 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5683                 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5684                 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5685
5686                 vmcs_write16(GUEST_INTR_STATUS, 0);
5687
5688                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5689                 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5690         }
5691
5692         if (ple_gap) {
5693                 vmcs_write32(PLE_GAP, ple_gap);
5694                 vmx->ple_window = ple_window;
5695                 vmx->ple_window_dirty = true;
5696         }
5697
5698         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5699         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5700         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
5701
5702         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
5703         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
5704         vmx_set_constant_host_state(vmx);
5705 #ifdef CONFIG_X86_64
5706         rdmsrl(MSR_FS_BASE, a);
5707         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5708         rdmsrl(MSR_GS_BASE, a);
5709         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5710 #else
5711         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5712         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5713 #endif
5714
5715         if (cpu_has_vmx_vmfunc())
5716                 vmcs_write64(VM_FUNCTION_CONTROL, 0);
5717
5718         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5719         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5720         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5721         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5722         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5723
5724         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5725                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5726
5727         for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5728                 u32 index = vmx_msr_index[i];
5729                 u32 data_low, data_high;
5730                 int j = vmx->nmsrs;
5731
5732                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5733                         continue;
5734                 if (wrmsr_safe(index, data_low, data_high) < 0)
5735                         continue;
5736                 vmx->guest_msrs[j].index = i;
5737                 vmx->guest_msrs[j].data = 0;
5738                 vmx->guest_msrs[j].mask = -1ull;
5739                 ++vmx->nmsrs;
5740         }
5741
5742         if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
5743                 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, vmx->arch_capabilities);
5744
5745         vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5746
5747         /* 22.2.1, 20.8.1 */
5748         vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5749
5750         vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5751         vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5752
5753         set_cr4_guest_host_mask(vmx);
5754
5755         if (vmx_xsaves_supported())
5756                 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5757
5758         if (enable_pml) {
5759                 ASSERT(vmx->pml_pg);
5760                 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5761                 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5762         }
5763
5764         return 0;
5765 }
5766
5767 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5768 {
5769         struct vcpu_vmx *vmx = to_vmx(vcpu);
5770         struct msr_data apic_base_msr;
5771         u64 cr0;
5772
5773         vmx->rmode.vm86_active = 0;
5774         vmx->spec_ctrl = 0;
5775
5776         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5777         kvm_set_cr8(vcpu, 0);
5778
5779         if (!init_event) {
5780                 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5781                                      MSR_IA32_APICBASE_ENABLE;
5782                 if (kvm_vcpu_is_reset_bsp(vcpu))
5783                         apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5784                 apic_base_msr.host_initiated = true;
5785                 kvm_set_apic_base(vcpu, &apic_base_msr);
5786         }
5787
5788         vmx_segment_cache_clear(vmx);
5789
5790         seg_setup(VCPU_SREG_CS);
5791         vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5792         vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5793
5794         seg_setup(VCPU_SREG_DS);
5795         seg_setup(VCPU_SREG_ES);
5796         seg_setup(VCPU_SREG_FS);
5797         seg_setup(VCPU_SREG_GS);
5798         seg_setup(VCPU_SREG_SS);
5799
5800         vmcs_write16(GUEST_TR_SELECTOR, 0);
5801         vmcs_writel(GUEST_TR_BASE, 0);
5802         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5803         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5804
5805         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5806         vmcs_writel(GUEST_LDTR_BASE, 0);
5807         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5808         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5809
5810         if (!init_event) {
5811                 vmcs_write32(GUEST_SYSENTER_CS, 0);
5812                 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5813                 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5814                 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5815         }
5816
5817         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
5818         kvm_rip_write(vcpu, 0xfff0);
5819
5820         vmcs_writel(GUEST_GDTR_BASE, 0);
5821         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5822
5823         vmcs_writel(GUEST_IDTR_BASE, 0);
5824         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5825
5826         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5827         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5828         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5829
5830         setup_msrs(vmx);
5831
5832         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
5833
5834         if (cpu_has_vmx_tpr_shadow() && !init_event) {
5835                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5836                 if (cpu_need_tpr_shadow(vcpu))
5837                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5838                                      __pa(vcpu->arch.apic->regs));
5839                 vmcs_write32(TPR_THRESHOLD, 0);
5840         }
5841
5842         kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5843
5844         if (vmx->vpid != 0)
5845                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5846
5847         cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5848         vmx->vcpu.arch.cr0 = cr0;
5849         vmx_set_cr0(vcpu, cr0); /* enter rmode */
5850         vmx_set_cr4(vcpu, 0);
5851         vmx_set_efer(vcpu, 0);
5852
5853         update_exception_bitmap(vcpu);
5854
5855         vpid_sync_context(vmx->vpid);
5856 }
5857
5858 /*
5859  * In nested virtualization, check if L1 asked to exit on external interrupts.
5860  * For most existing hypervisors, this will always return true.
5861  */
5862 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5863 {
5864         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5865                 PIN_BASED_EXT_INTR_MASK;
5866 }
5867
5868 /*
5869  * In nested virtualization, check if L1 has set
5870  * VM_EXIT_ACK_INTR_ON_EXIT
5871  */
5872 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5873 {
5874         return get_vmcs12(vcpu)->vm_exit_controls &
5875                 VM_EXIT_ACK_INTR_ON_EXIT;
5876 }
5877
5878 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5879 {
5880         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5881                 PIN_BASED_NMI_EXITING;
5882 }
5883
5884 static void enable_irq_window(struct kvm_vcpu *vcpu)
5885 {
5886         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5887                       CPU_BASED_VIRTUAL_INTR_PENDING);
5888 }
5889
5890 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5891 {
5892         if (!cpu_has_virtual_nmis() ||
5893             vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5894                 enable_irq_window(vcpu);
5895                 return;
5896         }
5897
5898         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5899                       CPU_BASED_VIRTUAL_NMI_PENDING);
5900 }
5901
5902 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5903 {
5904         struct vcpu_vmx *vmx = to_vmx(vcpu);
5905         uint32_t intr;
5906         int irq = vcpu->arch.interrupt.nr;
5907
5908         trace_kvm_inj_virq(irq);
5909
5910         ++vcpu->stat.irq_injections;
5911         if (vmx->rmode.vm86_active) {
5912                 int inc_eip = 0;
5913                 if (vcpu->arch.interrupt.soft)
5914                         inc_eip = vcpu->arch.event_exit_inst_len;
5915                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5916                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5917                 return;
5918         }
5919         intr = irq | INTR_INFO_VALID_MASK;
5920         if (vcpu->arch.interrupt.soft) {
5921                 intr |= INTR_TYPE_SOFT_INTR;
5922                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5923                              vmx->vcpu.arch.event_exit_inst_len);
5924         } else
5925                 intr |= INTR_TYPE_EXT_INTR;
5926         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5927 }
5928
5929 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5930 {
5931         struct vcpu_vmx *vmx = to_vmx(vcpu);
5932
5933         if (!cpu_has_virtual_nmis()) {
5934                 /*
5935                  * Tracking the NMI-blocked state in software is built upon
5936                  * finding the next open IRQ window. This, in turn, depends on
5937                  * well-behaving guests: They have to keep IRQs disabled at
5938                  * least as long as the NMI handler runs. Otherwise we may
5939                  * cause NMI nesting, maybe breaking the guest. But as this is
5940                  * highly unlikely, we can live with the residual risk.
5941                  */
5942                 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
5943                 vmx->loaded_vmcs->vnmi_blocked_time = 0;
5944         }
5945
5946         ++vcpu->stat.nmi_injections;
5947         vmx->loaded_vmcs->nmi_known_unmasked = false;
5948
5949         if (vmx->rmode.vm86_active) {
5950                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5951                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5952                 return;
5953         }
5954
5955         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5956                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5957 }
5958
5959 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5960 {
5961         struct vcpu_vmx *vmx = to_vmx(vcpu);
5962         bool masked;
5963
5964         if (!cpu_has_virtual_nmis())
5965                 return vmx->loaded_vmcs->soft_vnmi_blocked;
5966         if (vmx->loaded_vmcs->nmi_known_unmasked)
5967                 return false;
5968         masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5969         vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5970         return masked;
5971 }
5972
5973 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5974 {
5975         struct vcpu_vmx *vmx = to_vmx(vcpu);
5976
5977         if (!cpu_has_virtual_nmis()) {
5978                 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
5979                         vmx->loaded_vmcs->soft_vnmi_blocked = masked;
5980                         vmx->loaded_vmcs->vnmi_blocked_time = 0;
5981                 }
5982         } else {
5983                 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5984                 if (masked)
5985                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5986                                       GUEST_INTR_STATE_NMI);
5987                 else
5988                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5989                                         GUEST_INTR_STATE_NMI);
5990         }
5991 }
5992
5993 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5994 {
5995         if (to_vmx(vcpu)->nested.nested_run_pending)
5996                 return 0;
5997
5998         if (!cpu_has_virtual_nmis() &&
5999             to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
6000                 return 0;
6001
6002         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
6003                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
6004                    | GUEST_INTR_STATE_NMI));
6005 }
6006
6007 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
6008 {
6009         return (!to_vmx(vcpu)->nested.nested_run_pending &&
6010                 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
6011                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
6012                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
6013 }
6014
6015 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
6016 {
6017         int ret;
6018
6019         ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
6020                                     PAGE_SIZE * 3);
6021         if (ret)
6022                 return ret;
6023         kvm->arch.tss_addr = addr;
6024         return init_rmode_tss(kvm);
6025 }
6026
6027 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
6028 {
6029         switch (vec) {
6030         case BP_VECTOR:
6031                 /*
6032                  * Update instruction length as we may reinject the exception
6033                  * from user space while in guest debugging mode.
6034                  */
6035                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
6036                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6037                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6038                         return false;
6039                 /* fall through */
6040         case DB_VECTOR:
6041                 if (vcpu->guest_debug &
6042                         (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6043                         return false;
6044                 /* fall through */
6045         case DE_VECTOR:
6046         case OF_VECTOR:
6047         case BR_VECTOR:
6048         case UD_VECTOR:
6049         case DF_VECTOR:
6050         case SS_VECTOR:
6051         case GP_VECTOR:
6052         case MF_VECTOR:
6053                 return true;
6054         break;
6055         }
6056         return false;
6057 }
6058
6059 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
6060                                   int vec, u32 err_code)
6061 {
6062         /*
6063          * Instruction with address size override prefix opcode 0x67
6064          * Cause the #SS fault with 0 error code in VM86 mode.
6065          */
6066         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
6067                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
6068                         if (vcpu->arch.halt_request) {
6069                                 vcpu->arch.halt_request = 0;
6070                                 return kvm_vcpu_halt(vcpu);
6071                         }
6072                         return 1;
6073                 }
6074                 return 0;
6075         }
6076
6077         /*
6078          * Forward all other exceptions that are valid in real mode.
6079          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
6080          *        the required debugging infrastructure rework.
6081          */
6082         kvm_queue_exception(vcpu, vec);
6083         return 1;
6084 }
6085
6086 /*
6087  * Trigger machine check on the host. We assume all the MSRs are already set up
6088  * by the CPU and that we still run on the same CPU as the MCE occurred on.
6089  * We pass a fake environment to the machine check handler because we want
6090  * the guest to be always treated like user space, no matter what context
6091  * it used internally.
6092  */
6093 static void kvm_machine_check(void)
6094 {
6095 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
6096         struct pt_regs regs = {
6097                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
6098                 .flags = X86_EFLAGS_IF,
6099         };
6100
6101         do_machine_check(&regs, 0);
6102 #endif
6103 }
6104
6105 static int handle_machine_check(struct kvm_vcpu *vcpu)
6106 {
6107         /* already handled by vcpu_run */
6108         return 1;
6109 }
6110
6111 static int handle_exception(struct kvm_vcpu *vcpu)
6112 {
6113         struct vcpu_vmx *vmx = to_vmx(vcpu);
6114         struct kvm_run *kvm_run = vcpu->run;
6115         u32 intr_info, ex_no, error_code;
6116         unsigned long cr2, rip, dr6;
6117         u32 vect_info;
6118         enum emulation_result er;
6119
6120         vect_info = vmx->idt_vectoring_info;
6121         intr_info = vmx->exit_intr_info;
6122
6123         if (is_machine_check(intr_info))
6124                 return handle_machine_check(vcpu);
6125
6126         if (is_nmi(intr_info))
6127                 return 1;  /* already handled by vmx_vcpu_run() */
6128
6129         if (is_invalid_opcode(intr_info)) {
6130                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
6131                 if (er == EMULATE_USER_EXIT)
6132                         return 0;
6133                 if (er != EMULATE_DONE)
6134                         kvm_queue_exception(vcpu, UD_VECTOR);
6135                 return 1;
6136         }
6137
6138         error_code = 0;
6139         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6140                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6141
6142         /*
6143          * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
6144          * MMIO, it is better to report an internal error.
6145          * See the comments in vmx_handle_exit.
6146          */
6147         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
6148             !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
6149                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6150                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
6151                 vcpu->run->internal.ndata = 3;
6152                 vcpu->run->internal.data[0] = vect_info;
6153                 vcpu->run->internal.data[1] = intr_info;
6154                 vcpu->run->internal.data[2] = error_code;
6155                 return 0;
6156         }
6157
6158         if (is_page_fault(intr_info)) {
6159                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
6160                 /* EPT won't cause page fault directly */
6161                 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
6162                 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0,
6163                                 true);
6164         }
6165
6166         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
6167
6168         if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
6169                 return handle_rmode_exception(vcpu, ex_no, error_code);
6170
6171         switch (ex_no) {
6172         case AC_VECTOR:
6173                 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
6174                 return 1;
6175         case DB_VECTOR:
6176                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
6177                 if (!(vcpu->guest_debug &
6178                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
6179                         vcpu->arch.dr6 &= ~15;
6180                         vcpu->arch.dr6 |= dr6 | DR6_RTM;
6181                         if (is_icebp(intr_info))
6182                                 skip_emulated_instruction(vcpu);
6183
6184                         kvm_queue_exception(vcpu, DB_VECTOR);
6185                         return 1;
6186                 }
6187                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
6188                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
6189                 /* fall through */
6190         case BP_VECTOR:
6191                 /*
6192                  * Update instruction length as we may reinject #BP from
6193                  * user space while in guest debugging mode. Reading it for
6194                  * #DB as well causes no harm, it is not used in that case.
6195                  */
6196                 vmx->vcpu.arch.event_exit_inst_len =
6197                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6198                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6199                 rip = kvm_rip_read(vcpu);
6200                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
6201                 kvm_run->debug.arch.exception = ex_no;
6202                 break;
6203         default:
6204                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
6205                 kvm_run->ex.exception = ex_no;
6206                 kvm_run->ex.error_code = error_code;
6207                 break;
6208         }
6209         return 0;
6210 }
6211
6212 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
6213 {
6214         ++vcpu->stat.irq_exits;
6215         return 1;
6216 }
6217
6218 static int handle_triple_fault(struct kvm_vcpu *vcpu)
6219 {
6220         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6221         vcpu->mmio_needed = 0;
6222         return 0;
6223 }
6224
6225 static int handle_io(struct kvm_vcpu *vcpu)
6226 {
6227         unsigned long exit_qualification;
6228         int size, in, string, ret;
6229         unsigned port;
6230
6231         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6232         string = (exit_qualification & 16) != 0;
6233         in = (exit_qualification & 8) != 0;
6234
6235         ++vcpu->stat.io_exits;
6236
6237         if (string || in)
6238                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6239
6240         port = exit_qualification >> 16;
6241         size = (exit_qualification & 7) + 1;
6242
6243         ret = kvm_skip_emulated_instruction(vcpu);
6244
6245         /*
6246          * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6247          * KVM_EXIT_DEBUG here.
6248          */
6249         return kvm_fast_pio_out(vcpu, size, port) && ret;
6250 }
6251
6252 static void
6253 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
6254 {
6255         /*
6256          * Patch in the VMCALL instruction:
6257          */
6258         hypercall[0] = 0x0f;
6259         hypercall[1] = 0x01;
6260         hypercall[2] = 0xc1;
6261 }
6262
6263 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
6264 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
6265 {
6266         if (is_guest_mode(vcpu)) {
6267                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6268                 unsigned long orig_val = val;
6269
6270                 /*
6271                  * We get here when L2 changed cr0 in a way that did not change
6272                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
6273                  * but did change L0 shadowed bits. So we first calculate the
6274                  * effective cr0 value that L1 would like to write into the
6275                  * hardware. It consists of the L2-owned bits from the new
6276                  * value combined with the L1-owned bits from L1's guest_cr0.
6277                  */
6278                 val = (val & ~vmcs12->cr0_guest_host_mask) |
6279                         (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
6280
6281                 if (!nested_guest_cr0_valid(vcpu, val))
6282                         return 1;
6283
6284                 if (kvm_set_cr0(vcpu, val))
6285                         return 1;
6286                 vmcs_writel(CR0_READ_SHADOW, orig_val);
6287                 return 0;
6288         } else {
6289                 if (to_vmx(vcpu)->nested.vmxon &&
6290                     !nested_host_cr0_valid(vcpu, val))
6291                         return 1;
6292
6293                 return kvm_set_cr0(vcpu, val);
6294         }
6295 }
6296
6297 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
6298 {
6299         if (is_guest_mode(vcpu)) {
6300                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6301                 unsigned long orig_val = val;
6302
6303                 /* analogously to handle_set_cr0 */
6304                 val = (val & ~vmcs12->cr4_guest_host_mask) |
6305                         (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
6306                 if (kvm_set_cr4(vcpu, val))
6307                         return 1;
6308                 vmcs_writel(CR4_READ_SHADOW, orig_val);
6309                 return 0;
6310         } else
6311                 return kvm_set_cr4(vcpu, val);
6312 }
6313
6314 static int handle_cr(struct kvm_vcpu *vcpu)
6315 {
6316         unsigned long exit_qualification, val;
6317         int cr;
6318         int reg;
6319         int err;
6320         int ret;
6321
6322         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6323         cr = exit_qualification & 15;
6324         reg = (exit_qualification >> 8) & 15;
6325         switch ((exit_qualification >> 4) & 3) {
6326         case 0: /* mov to cr */
6327                 val = kvm_register_readl(vcpu, reg);
6328                 trace_kvm_cr_write(cr, val);
6329                 switch (cr) {
6330                 case 0:
6331                         err = handle_set_cr0(vcpu, val);
6332                         return kvm_complete_insn_gp(vcpu, err);
6333                 case 3:
6334                         err = kvm_set_cr3(vcpu, val);
6335                         return kvm_complete_insn_gp(vcpu, err);
6336                 case 4:
6337                         err = handle_set_cr4(vcpu, val);
6338                         return kvm_complete_insn_gp(vcpu, err);
6339                 case 8: {
6340                                 u8 cr8_prev = kvm_get_cr8(vcpu);
6341                                 u8 cr8 = (u8)val;
6342                                 err = kvm_set_cr8(vcpu, cr8);
6343                                 ret = kvm_complete_insn_gp(vcpu, err);
6344                                 if (lapic_in_kernel(vcpu))
6345                                         return ret;
6346                                 if (cr8_prev <= cr8)
6347                                         return ret;
6348                                 /*
6349                                  * TODO: we might be squashing a
6350                                  * KVM_GUESTDBG_SINGLESTEP-triggered
6351                                  * KVM_EXIT_DEBUG here.
6352                                  */
6353                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
6354                                 return 0;
6355                         }
6356                 }
6357                 break;
6358         case 2: /* clts */
6359                 WARN_ONCE(1, "Guest should always own CR0.TS");
6360                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
6361                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
6362                 return kvm_skip_emulated_instruction(vcpu);
6363         case 1: /*mov from cr*/
6364                 switch (cr) {
6365                 case 3:
6366                         val = kvm_read_cr3(vcpu);
6367                         kvm_register_write(vcpu, reg, val);
6368                         trace_kvm_cr_read(cr, val);
6369                         return kvm_skip_emulated_instruction(vcpu);
6370                 case 8:
6371                         val = kvm_get_cr8(vcpu);
6372                         kvm_register_write(vcpu, reg, val);
6373                         trace_kvm_cr_read(cr, val);
6374                         return kvm_skip_emulated_instruction(vcpu);
6375                 }
6376                 break;
6377         case 3: /* lmsw */
6378                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6379                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
6380                 kvm_lmsw(vcpu, val);
6381
6382                 return kvm_skip_emulated_instruction(vcpu);
6383         default:
6384                 break;
6385         }
6386         vcpu->run->exit_reason = 0;
6387         vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6388                (int)(exit_qualification >> 4) & 3, cr);
6389         return 0;
6390 }
6391
6392 static int handle_dr(struct kvm_vcpu *vcpu)
6393 {
6394         unsigned long exit_qualification;
6395         int dr, dr7, reg;
6396
6397         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6398         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6399
6400         /* First, if DR does not exist, trigger UD */
6401         if (!kvm_require_dr(vcpu, dr))
6402                 return 1;
6403
6404         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6405         if (!kvm_require_cpl(vcpu, 0))
6406                 return 1;
6407         dr7 = vmcs_readl(GUEST_DR7);
6408         if (dr7 & DR7_GD) {
6409                 /*
6410                  * As the vm-exit takes precedence over the debug trap, we
6411                  * need to emulate the latter, either for the host or the
6412                  * guest debugging itself.
6413                  */
6414                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6415                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6416                         vcpu->run->debug.arch.dr7 = dr7;
6417                         vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6418                         vcpu->run->debug.arch.exception = DB_VECTOR;
6419                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6420                         return 0;
6421                 } else {
6422                         vcpu->arch.dr6 &= ~15;
6423                         vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6424                         kvm_queue_exception(vcpu, DB_VECTOR);
6425                         return 1;
6426                 }
6427         }
6428
6429         if (vcpu->guest_debug == 0) {
6430                 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6431                                 CPU_BASED_MOV_DR_EXITING);
6432
6433                 /*
6434                  * No more DR vmexits; force a reload of the debug registers
6435                  * and reenter on this instruction.  The next vmexit will
6436                  * retrieve the full state of the debug registers.
6437                  */
6438                 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
6439                 return 1;
6440         }
6441
6442         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
6443         if (exit_qualification & TYPE_MOV_FROM_DR) {
6444                 unsigned long val;
6445
6446                 if (kvm_get_dr(vcpu, dr, &val))
6447                         return 1;
6448                 kvm_register_write(vcpu, reg, val);
6449         } else
6450                 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
6451                         return 1;
6452
6453         return kvm_skip_emulated_instruction(vcpu);
6454 }
6455
6456 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
6457 {
6458         return vcpu->arch.dr6;
6459 }
6460
6461 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
6462 {
6463 }
6464
6465 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
6466 {
6467         get_debugreg(vcpu->arch.db[0], 0);
6468         get_debugreg(vcpu->arch.db[1], 1);
6469         get_debugreg(vcpu->arch.db[2], 2);
6470         get_debugreg(vcpu->arch.db[3], 3);
6471         get_debugreg(vcpu->arch.dr6, 6);
6472         vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6473
6474         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6475         vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6476 }
6477
6478 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6479 {
6480         vmcs_writel(GUEST_DR7, val);
6481 }
6482
6483 static int handle_cpuid(struct kvm_vcpu *vcpu)
6484 {
6485         return kvm_emulate_cpuid(vcpu);
6486 }
6487
6488 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6489 {
6490         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6491         struct msr_data msr_info;
6492
6493         msr_info.index = ecx;
6494         msr_info.host_initiated = false;
6495         if (vmx_get_msr(vcpu, &msr_info)) {
6496                 trace_kvm_msr_read_ex(ecx);
6497                 kvm_inject_gp(vcpu, 0);
6498                 return 1;
6499         }
6500
6501         trace_kvm_msr_read(ecx, msr_info.data);
6502
6503         /* FIXME: handling of bits 32:63 of rax, rdx */
6504         vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6505         vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6506         return kvm_skip_emulated_instruction(vcpu);
6507 }
6508
6509 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6510 {
6511         struct msr_data msr;
6512         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6513         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6514                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6515
6516         msr.data = data;
6517         msr.index = ecx;
6518         msr.host_initiated = false;
6519         if (kvm_set_msr(vcpu, &msr) != 0) {
6520                 trace_kvm_msr_write_ex(ecx, data);
6521                 kvm_inject_gp(vcpu, 0);
6522                 return 1;
6523         }
6524
6525         trace_kvm_msr_write(ecx, data);
6526         return kvm_skip_emulated_instruction(vcpu);
6527 }
6528
6529 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6530 {
6531         kvm_apic_update_ppr(vcpu);
6532         return 1;
6533 }
6534
6535 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6536 {
6537         vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6538                         CPU_BASED_VIRTUAL_INTR_PENDING);
6539
6540         kvm_make_request(KVM_REQ_EVENT, vcpu);
6541
6542         ++vcpu->stat.irq_window_exits;
6543         return 1;
6544 }
6545
6546 static int handle_halt(struct kvm_vcpu *vcpu)
6547 {
6548         return kvm_emulate_halt(vcpu);
6549 }
6550
6551 static int handle_vmcall(struct kvm_vcpu *vcpu)
6552 {
6553         return kvm_emulate_hypercall(vcpu);
6554 }
6555
6556 static int handle_invd(struct kvm_vcpu *vcpu)
6557 {
6558         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6559 }
6560
6561 static int handle_invlpg(struct kvm_vcpu *vcpu)
6562 {
6563         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6564
6565         kvm_mmu_invlpg(vcpu, exit_qualification);
6566         return kvm_skip_emulated_instruction(vcpu);
6567 }
6568
6569 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6570 {
6571         int err;
6572
6573         err = kvm_rdpmc(vcpu);
6574         return kvm_complete_insn_gp(vcpu, err);
6575 }
6576
6577 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6578 {
6579         return kvm_emulate_wbinvd(vcpu);
6580 }
6581
6582 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6583 {
6584         u64 new_bv = kvm_read_edx_eax(vcpu);
6585         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6586
6587         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6588                 return kvm_skip_emulated_instruction(vcpu);
6589         return 1;
6590 }
6591
6592 static int handle_xsaves(struct kvm_vcpu *vcpu)
6593 {
6594         kvm_skip_emulated_instruction(vcpu);
6595         WARN(1, "this should never happen\n");
6596         return 1;
6597 }
6598
6599 static int handle_xrstors(struct kvm_vcpu *vcpu)
6600 {
6601         kvm_skip_emulated_instruction(vcpu);
6602         WARN(1, "this should never happen\n");
6603         return 1;
6604 }
6605
6606 static int handle_apic_access(struct kvm_vcpu *vcpu)
6607 {
6608         if (likely(fasteoi)) {
6609                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6610                 int access_type, offset;
6611
6612                 access_type = exit_qualification & APIC_ACCESS_TYPE;
6613                 offset = exit_qualification & APIC_ACCESS_OFFSET;
6614                 /*
6615                  * Sane guest uses MOV to write EOI, with written value
6616                  * not cared. So make a short-circuit here by avoiding
6617                  * heavy instruction emulation.
6618                  */
6619                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6620                     (offset == APIC_EOI)) {
6621                         kvm_lapic_set_eoi(vcpu);
6622                         return kvm_skip_emulated_instruction(vcpu);
6623                 }
6624         }
6625         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6626 }
6627
6628 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6629 {
6630         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6631         int vector = exit_qualification & 0xff;
6632
6633         /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6634         kvm_apic_set_eoi_accelerated(vcpu, vector);
6635         return 1;
6636 }
6637
6638 static int handle_apic_write(struct kvm_vcpu *vcpu)
6639 {
6640         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6641         u32 offset = exit_qualification & 0xfff;
6642
6643         /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6644         kvm_apic_write_nodecode(vcpu, offset);
6645         return 1;
6646 }
6647
6648 static int handle_task_switch(struct kvm_vcpu *vcpu)
6649 {
6650         struct vcpu_vmx *vmx = to_vmx(vcpu);
6651         unsigned long exit_qualification;
6652         bool has_error_code = false;
6653         u32 error_code = 0;
6654         u16 tss_selector;
6655         int reason, type, idt_v, idt_index;
6656
6657         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6658         idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6659         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6660
6661         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6662
6663         reason = (u32)exit_qualification >> 30;
6664         if (reason == TASK_SWITCH_GATE && idt_v) {
6665                 switch (type) {
6666                 case INTR_TYPE_NMI_INTR:
6667                         vcpu->arch.nmi_injected = false;
6668                         vmx_set_nmi_mask(vcpu, true);
6669                         break;
6670                 case INTR_TYPE_EXT_INTR:
6671                 case INTR_TYPE_SOFT_INTR:
6672                         kvm_clear_interrupt_queue(vcpu);
6673                         break;
6674                 case INTR_TYPE_HARD_EXCEPTION:
6675                         if (vmx->idt_vectoring_info &
6676                             VECTORING_INFO_DELIVER_CODE_MASK) {
6677                                 has_error_code = true;
6678                                 error_code =
6679                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
6680                         }
6681                         /* fall through */
6682                 case INTR_TYPE_SOFT_EXCEPTION:
6683                         kvm_clear_exception_queue(vcpu);
6684                         break;
6685                 default:
6686                         break;
6687                 }
6688         }
6689         tss_selector = exit_qualification;
6690
6691         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6692                        type != INTR_TYPE_EXT_INTR &&
6693                        type != INTR_TYPE_NMI_INTR))
6694                 skip_emulated_instruction(vcpu);
6695
6696         if (kvm_task_switch(vcpu, tss_selector,
6697                             type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6698                             has_error_code, error_code) == EMULATE_FAIL) {
6699                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6700                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6701                 vcpu->run->internal.ndata = 0;
6702                 return 0;
6703         }
6704
6705         /*
6706          * TODO: What about debug traps on tss switch?
6707          *       Are we supposed to inject them and update dr6?
6708          */
6709
6710         return 1;
6711 }
6712
6713 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6714 {
6715         unsigned long exit_qualification;
6716         gpa_t gpa;
6717         u64 error_code;
6718
6719         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6720
6721         /*
6722          * EPT violation happened while executing iret from NMI,
6723          * "blocked by NMI" bit has to be set before next VM entry.
6724          * There are errata that may cause this bit to not be set:
6725          * AAK134, BY25.
6726          */
6727         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6728                         cpu_has_virtual_nmis() &&
6729                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6730                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6731
6732         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6733         trace_kvm_page_fault(gpa, exit_qualification);
6734
6735         /* Is it a read fault? */
6736         error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6737                      ? PFERR_USER_MASK : 0;
6738         /* Is it a write fault? */
6739         error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6740                       ? PFERR_WRITE_MASK : 0;
6741         /* Is it a fetch fault? */
6742         error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6743                       ? PFERR_FETCH_MASK : 0;
6744         /* ept page table entry is present? */
6745         error_code |= (exit_qualification &
6746                        (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6747                         EPT_VIOLATION_EXECUTABLE))
6748                       ? PFERR_PRESENT_MASK : 0;
6749
6750         error_code |= (exit_qualification & 0x100) != 0 ?
6751                PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
6752
6753         vcpu->arch.exit_qualification = exit_qualification;
6754         return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6755 }
6756
6757 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6758 {
6759         int ret;
6760         gpa_t gpa;
6761
6762         /*
6763          * A nested guest cannot optimize MMIO vmexits, because we have an
6764          * nGPA here instead of the required GPA.
6765          */
6766         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6767         if (!is_guest_mode(vcpu) &&
6768             !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6769                 trace_kvm_fast_mmio(gpa);
6770                 /*
6771                  * Doing kvm_skip_emulated_instruction() depends on undefined
6772                  * behavior: Intel's manual doesn't mandate
6773                  * VM_EXIT_INSTRUCTION_LEN to be set in VMCS when EPT MISCONFIG
6774                  * occurs and while on real hardware it was observed to be set,
6775                  * other hypervisors (namely Hyper-V) don't set it, we end up
6776                  * advancing IP with some random value. Disable fast mmio when
6777                  * running nested and keep it for real hardware in hope that
6778                  * VM_EXIT_INSTRUCTION_LEN will always be set correctly.
6779                  */
6780                 if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
6781                         return kvm_skip_emulated_instruction(vcpu);
6782                 else
6783                         return x86_emulate_instruction(vcpu, gpa, EMULTYPE_SKIP,
6784                                                        NULL, 0) == EMULATE_DONE;
6785         }
6786
6787         ret = kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
6788         if (ret >= 0)
6789                 return ret;
6790
6791         /* It is the real ept misconfig */
6792         WARN_ON(1);
6793
6794         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6795         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6796
6797         return 0;
6798 }
6799
6800 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6801 {
6802         vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6803                         CPU_BASED_VIRTUAL_NMI_PENDING);
6804         ++vcpu->stat.nmi_window_exits;
6805         kvm_make_request(KVM_REQ_EVENT, vcpu);
6806
6807         return 1;
6808 }
6809
6810 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6811 {
6812         struct vcpu_vmx *vmx = to_vmx(vcpu);
6813         enum emulation_result err = EMULATE_DONE;
6814         int ret = 1;
6815         u32 cpu_exec_ctrl;
6816         bool intr_window_requested;
6817         unsigned count = 130;
6818
6819         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6820         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6821
6822         while (vmx->emulation_required && count-- != 0) {
6823                 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6824                         return handle_interrupt_window(&vmx->vcpu);
6825
6826                 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
6827                         return 1;
6828
6829                 err = emulate_instruction(vcpu, 0);
6830
6831                 if (err == EMULATE_USER_EXIT) {
6832                         ++vcpu->stat.mmio_exits;
6833                         ret = 0;
6834                         goto out;
6835                 }
6836
6837                 if (err != EMULATE_DONE) {
6838                         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6839                         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6840                         vcpu->run->internal.ndata = 0;
6841                         return 0;
6842                 }
6843
6844                 if (vcpu->arch.halt_request) {
6845                         vcpu->arch.halt_request = 0;
6846                         ret = kvm_vcpu_halt(vcpu);
6847                         goto out;
6848                 }
6849
6850                 if (signal_pending(current))
6851                         goto out;
6852                 if (need_resched())
6853                         schedule();
6854         }
6855
6856 out:
6857         return ret;
6858 }
6859
6860 static int __grow_ple_window(int val)
6861 {
6862         if (ple_window_grow < 1)
6863                 return ple_window;
6864
6865         val = min(val, ple_window_actual_max);
6866
6867         if (ple_window_grow < ple_window)
6868                 val *= ple_window_grow;
6869         else
6870                 val += ple_window_grow;
6871
6872         return val;
6873 }
6874
6875 static int __shrink_ple_window(int val, int modifier, int minimum)
6876 {
6877         if (modifier < 1)
6878                 return ple_window;
6879
6880         if (modifier < ple_window)
6881                 val /= modifier;
6882         else
6883                 val -= modifier;
6884
6885         return max(val, minimum);
6886 }
6887
6888 static void grow_ple_window(struct kvm_vcpu *vcpu)
6889 {
6890         struct vcpu_vmx *vmx = to_vmx(vcpu);
6891         int old = vmx->ple_window;
6892
6893         vmx->ple_window = __grow_ple_window(old);
6894
6895         if (vmx->ple_window != old)
6896                 vmx->ple_window_dirty = true;
6897
6898         trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6899 }
6900
6901 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6902 {
6903         struct vcpu_vmx *vmx = to_vmx(vcpu);
6904         int old = vmx->ple_window;
6905
6906         vmx->ple_window = __shrink_ple_window(old,
6907                                               ple_window_shrink, ple_window);
6908
6909         if (vmx->ple_window != old)
6910                 vmx->ple_window_dirty = true;
6911
6912         trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6913 }
6914
6915 /*
6916  * ple_window_actual_max is computed to be one grow_ple_window() below
6917  * ple_window_max. (See __grow_ple_window for the reason.)
6918  * This prevents overflows, because ple_window_max is int.
6919  * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6920  * this process.
6921  * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6922  */
6923 static void update_ple_window_actual_max(void)
6924 {
6925         ple_window_actual_max =
6926                         __shrink_ple_window(max(ple_window_max, ple_window),
6927                                             ple_window_grow, INT_MIN);
6928 }
6929
6930 /*
6931  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6932  */
6933 static void wakeup_handler(void)
6934 {
6935         struct kvm_vcpu *vcpu;
6936         int cpu = smp_processor_id();
6937
6938         spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6939         list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6940                         blocked_vcpu_list) {
6941                 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6942
6943                 if (pi_test_on(pi_desc) == 1)
6944                         kvm_vcpu_kick(vcpu);
6945         }
6946         spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6947 }
6948
6949 void vmx_enable_tdp(void)
6950 {
6951         kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6952                 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6953                 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6954                 0ull, VMX_EPT_EXECUTABLE_MASK,
6955                 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6956                 VMX_EPT_RWX_MASK, 0ull);
6957
6958         ept_set_mmio_spte_mask();
6959         kvm_enable_tdp();
6960 }
6961
6962 static __init int hardware_setup(void)
6963 {
6964         int r = -ENOMEM, i;
6965
6966         rdmsrl_safe(MSR_EFER, &host_efer);
6967
6968         for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6969                 kvm_define_shared_msr(i, vmx_msr_index[i]);
6970
6971         for (i = 0; i < VMX_BITMAP_NR; i++) {
6972                 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6973                 if (!vmx_bitmap[i])
6974                         goto out;
6975         }
6976
6977         memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6978         memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6979
6980         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6981
6982         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6983
6984         if (setup_vmcs_config(&vmcs_config) < 0) {
6985                 r = -EIO;
6986                 goto out;
6987         }
6988
6989         if (boot_cpu_has(X86_FEATURE_NX))
6990                 kvm_enable_efer_bits(EFER_NX);
6991
6992         if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6993                 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6994                 enable_vpid = 0;
6995
6996         if (!cpu_has_vmx_shadow_vmcs())
6997                 enable_shadow_vmcs = 0;
6998         if (enable_shadow_vmcs)
6999                 init_vmcs_shadow_fields();
7000
7001         if (!cpu_has_vmx_ept() ||
7002             !cpu_has_vmx_ept_4levels() ||
7003             !cpu_has_vmx_ept_mt_wb()) {
7004                 enable_ept = 0;
7005                 enable_unrestricted_guest = 0;
7006                 enable_ept_ad_bits = 0;
7007         }
7008
7009         if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7010                 enable_ept_ad_bits = 0;
7011
7012         if (!cpu_has_vmx_unrestricted_guest())
7013                 enable_unrestricted_guest = 0;
7014
7015         if (!cpu_has_vmx_flexpriority())
7016                 flexpriority_enabled = 0;
7017
7018         /*
7019          * set_apic_access_page_addr() is used to reload apic access
7020          * page upon invalidation.  No need to do anything if not
7021          * using the APIC_ACCESS_ADDR VMCS field.
7022          */
7023         if (!flexpriority_enabled)
7024                 kvm_x86_ops->set_apic_access_page_addr = NULL;
7025
7026         if (!cpu_has_vmx_tpr_shadow())
7027                 kvm_x86_ops->update_cr8_intercept = NULL;
7028
7029         if (enable_ept && !cpu_has_vmx_ept_2m_page())
7030                 kvm_disable_largepages();
7031
7032         if (!cpu_has_vmx_ple())
7033                 ple_gap = 0;
7034
7035         if (!cpu_has_vmx_apicv()) {
7036                 enable_apicv = 0;
7037                 kvm_x86_ops->sync_pir_to_irr = NULL;
7038         }
7039
7040         if (cpu_has_vmx_tsc_scaling()) {
7041                 kvm_has_tsc_control = true;
7042                 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7043                 kvm_tsc_scaling_ratio_frac_bits = 48;
7044         }
7045
7046         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7047
7048         if (enable_ept)
7049                 vmx_enable_tdp();
7050         else
7051                 kvm_disable_tdp();
7052
7053         update_ple_window_actual_max();
7054
7055         /*
7056          * Only enable PML when hardware supports PML feature, and both EPT
7057          * and EPT A/D bit features are enabled -- PML depends on them to work.
7058          */
7059         if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7060                 enable_pml = 0;
7061
7062         if (!enable_pml) {
7063                 kvm_x86_ops->slot_enable_log_dirty = NULL;
7064                 kvm_x86_ops->slot_disable_log_dirty = NULL;
7065                 kvm_x86_ops->flush_log_dirty = NULL;
7066                 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7067         }
7068
7069         if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
7070                 u64 vmx_msr;
7071
7072                 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7073                 cpu_preemption_timer_multi =
7074                          vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7075         } else {
7076                 kvm_x86_ops->set_hv_timer = NULL;
7077                 kvm_x86_ops->cancel_hv_timer = NULL;
7078         }
7079
7080         kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7081
7082         kvm_mce_cap_supported |= MCG_LMCE_P;
7083
7084         return alloc_kvm_area();
7085
7086 out:
7087         for (i = 0; i < VMX_BITMAP_NR; i++)
7088                 free_page((unsigned long)vmx_bitmap[i]);
7089
7090     return r;
7091 }
7092
7093 static __exit void hardware_unsetup(void)
7094 {
7095         int i;
7096
7097         for (i = 0; i < VMX_BITMAP_NR; i++)
7098                 free_page((unsigned long)vmx_bitmap[i]);
7099
7100         free_kvm_area();
7101 }
7102
7103 /*
7104  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
7105  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
7106  */
7107 static int handle_pause(struct kvm_vcpu *vcpu)
7108 {
7109         if (ple_gap)
7110                 grow_ple_window(vcpu);
7111
7112         /*
7113          * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
7114          * VM-execution control is ignored if CPL > 0. OTOH, KVM
7115          * never set PAUSE_EXITING and just set PLE if supported,
7116          * so the vcpu must be CPL=0 if it gets a PAUSE exit.
7117          */
7118         kvm_vcpu_on_spin(vcpu, true);
7119         return kvm_skip_emulated_instruction(vcpu);
7120 }
7121
7122 static int handle_nop(struct kvm_vcpu *vcpu)
7123 {
7124         return kvm_skip_emulated_instruction(vcpu);
7125 }
7126
7127 static int handle_mwait(struct kvm_vcpu *vcpu)
7128 {
7129         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
7130         return handle_nop(vcpu);
7131 }
7132
7133 static int handle_invalid_op(struct kvm_vcpu *vcpu)
7134 {
7135         kvm_queue_exception(vcpu, UD_VECTOR);
7136         return 1;
7137 }
7138
7139 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
7140 {
7141         return 1;
7142 }
7143
7144 static int handle_monitor(struct kvm_vcpu *vcpu)
7145 {
7146         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
7147         return handle_nop(vcpu);
7148 }
7149
7150 /*
7151  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
7152  * set the success or error code of an emulated VMX instruction, as specified
7153  * by Vol 2B, VMX Instruction Reference, "Conventions".
7154  */
7155 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
7156 {
7157         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
7158                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7159                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
7160 }
7161
7162 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
7163 {
7164         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7165                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
7166                             X86_EFLAGS_SF | X86_EFLAGS_OF))
7167                         | X86_EFLAGS_CF);
7168 }
7169
7170 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
7171                                         u32 vm_instruction_error)
7172 {
7173         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
7174                 /*
7175                  * failValid writes the error number to the current VMCS, which
7176                  * can't be done there isn't a current VMCS.
7177                  */
7178                 nested_vmx_failInvalid(vcpu);
7179                 return;
7180         }
7181         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7182                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7183                             X86_EFLAGS_SF | X86_EFLAGS_OF))
7184                         | X86_EFLAGS_ZF);
7185         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
7186         /*
7187          * We don't need to force a shadow sync because
7188          * VM_INSTRUCTION_ERROR is not shadowed
7189          */
7190 }
7191
7192 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
7193 {
7194         /* TODO: not to reset guest simply here. */
7195         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7196         pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
7197 }
7198
7199 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
7200 {
7201         struct vcpu_vmx *vmx =
7202                 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
7203
7204         vmx->nested.preemption_timer_expired = true;
7205         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
7206         kvm_vcpu_kick(&vmx->vcpu);
7207
7208         return HRTIMER_NORESTART;
7209 }
7210
7211 /*
7212  * Decode the memory-address operand of a vmx instruction, as recorded on an
7213  * exit caused by such an instruction (run by a guest hypervisor).
7214  * On success, returns 0. When the operand is invalid, returns 1 and throws
7215  * #UD or #GP.
7216  */
7217 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
7218                                  unsigned long exit_qualification,
7219                                  u32 vmx_instruction_info, bool wr, gva_t *ret)
7220 {
7221         gva_t off;
7222         bool exn;
7223         struct kvm_segment s;
7224
7225         /*
7226          * According to Vol. 3B, "Information for VM Exits Due to Instruction
7227          * Execution", on an exit, vmx_instruction_info holds most of the
7228          * addressing components of the operand. Only the displacement part
7229          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
7230          * For how an actual address is calculated from all these components,
7231          * refer to Vol. 1, "Operand Addressing".
7232          */
7233         int  scaling = vmx_instruction_info & 3;
7234         int  addr_size = (vmx_instruction_info >> 7) & 7;
7235         bool is_reg = vmx_instruction_info & (1u << 10);
7236         int  seg_reg = (vmx_instruction_info >> 15) & 7;
7237         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
7238         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
7239         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
7240         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
7241
7242         if (is_reg) {
7243                 kvm_queue_exception(vcpu, UD_VECTOR);
7244                 return 1;
7245         }
7246
7247         /* Addr = segment_base + offset */
7248         /* offset = base + [index * scale] + displacement */
7249         off = exit_qualification; /* holds the displacement */
7250         if (base_is_valid)
7251                 off += kvm_register_read(vcpu, base_reg);
7252         if (index_is_valid)
7253                 off += kvm_register_read(vcpu, index_reg)<<scaling;
7254         vmx_get_segment(vcpu, &s, seg_reg);
7255         *ret = s.base + off;
7256
7257         if (addr_size == 1) /* 32 bit */
7258                 *ret &= 0xffffffff;
7259
7260         /* Checks for #GP/#SS exceptions. */
7261         exn = false;
7262         if (is_long_mode(vcpu)) {
7263                 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
7264                  * non-canonical form. This is the only check on the memory
7265                  * destination for long mode!
7266                  */
7267                 exn = is_noncanonical_address(*ret, vcpu);
7268         } else if (is_protmode(vcpu)) {
7269                 /* Protected mode: apply checks for segment validity in the
7270                  * following order:
7271                  * - segment type check (#GP(0) may be thrown)
7272                  * - usability check (#GP(0)/#SS(0))
7273                  * - limit check (#GP(0)/#SS(0))
7274                  */
7275                 if (wr)
7276                         /* #GP(0) if the destination operand is located in a
7277                          * read-only data segment or any code segment.
7278                          */
7279                         exn = ((s.type & 0xa) == 0 || (s.type & 8));
7280                 else
7281                         /* #GP(0) if the source operand is located in an
7282                          * execute-only code segment
7283                          */
7284                         exn = ((s.type & 0xa) == 8);
7285                 if (exn) {
7286                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7287                         return 1;
7288                 }
7289                 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
7290                  */
7291                 exn = (s.unusable != 0);
7292                 /* Protected mode: #GP(0)/#SS(0) if the memory
7293                  * operand is outside the segment limit.
7294                  */
7295                 exn = exn || (off + sizeof(u64) > s.limit);
7296         }
7297         if (exn) {
7298                 kvm_queue_exception_e(vcpu,
7299                                       seg_reg == VCPU_SREG_SS ?
7300                                                 SS_VECTOR : GP_VECTOR,
7301                                       0);
7302                 return 1;
7303         }
7304
7305         return 0;
7306 }
7307
7308 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
7309 {
7310         gva_t gva;
7311         struct x86_exception e;
7312
7313         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7314                         vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
7315                 return 1;
7316
7317         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, vmpointer,
7318                                 sizeof(*vmpointer), &e)) {
7319                 kvm_inject_page_fault(vcpu, &e);
7320                 return 1;
7321         }
7322
7323         return 0;
7324 }
7325
7326 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7327 {
7328         struct vcpu_vmx *vmx = to_vmx(vcpu);
7329         struct vmcs *shadow_vmcs;
7330         int r;
7331
7332         r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
7333         if (r < 0)
7334                 goto out_vmcs02;
7335
7336         vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7337         if (!vmx->nested.cached_vmcs12)
7338                 goto out_cached_vmcs12;
7339
7340         if (enable_shadow_vmcs) {
7341                 shadow_vmcs = alloc_vmcs();
7342                 if (!shadow_vmcs)
7343                         goto out_shadow_vmcs;
7344                 /* mark vmcs as shadow */
7345                 shadow_vmcs->revision_id |= (1u << 31);
7346                 /* init shadow vmcs */
7347                 vmcs_clear(shadow_vmcs);
7348                 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7349         }
7350
7351         hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7352                      HRTIMER_MODE_REL_PINNED);
7353         vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7354
7355         vmx->nested.vmxon = true;
7356         return 0;
7357
7358 out_shadow_vmcs:
7359         kfree(vmx->nested.cached_vmcs12);
7360
7361 out_cached_vmcs12:
7362         free_loaded_vmcs(&vmx->nested.vmcs02);
7363
7364 out_vmcs02:
7365         return -ENOMEM;
7366 }
7367
7368 /*
7369  * Emulate the VMXON instruction.
7370  * Currently, we just remember that VMX is active, and do not save or even
7371  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7372  * do not currently need to store anything in that guest-allocated memory
7373  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7374  * argument is different from the VMXON pointer (which the spec says they do).
7375  */
7376 static int handle_vmon(struct kvm_vcpu *vcpu)
7377 {
7378         int ret;
7379         gpa_t vmptr;
7380         struct page *page;
7381         struct vcpu_vmx *vmx = to_vmx(vcpu);
7382         const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7383                 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7384
7385         /*
7386          * The Intel VMX Instruction Reference lists a bunch of bits that are
7387          * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7388          * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7389          * Otherwise, we should fail with #UD.  But most faulting conditions
7390          * have already been checked by hardware, prior to the VM-exit for
7391          * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
7392          * that bit set to 1 in non-root mode.
7393          */
7394         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7395                 kvm_queue_exception(vcpu, UD_VECTOR);
7396                 return 1;
7397         }
7398
7399         if (vmx->nested.vmxon) {
7400                 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7401                 return kvm_skip_emulated_instruction(vcpu);
7402         }
7403
7404         if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7405                         != VMXON_NEEDED_FEATURES) {
7406                 kvm_inject_gp(vcpu, 0);
7407                 return 1;
7408         }
7409
7410         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7411                 return 1;
7412
7413         /*
7414          * SDM 3: 24.11.5
7415          * The first 4 bytes of VMXON region contain the supported
7416          * VMCS revision identifier
7417          *
7418          * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7419          * which replaces physical address width with 32
7420          */
7421         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7422                 nested_vmx_failInvalid(vcpu);
7423                 return kvm_skip_emulated_instruction(vcpu);
7424         }
7425
7426         page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7427         if (is_error_page(page)) {
7428                 nested_vmx_failInvalid(vcpu);
7429                 return kvm_skip_emulated_instruction(vcpu);
7430         }
7431         if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7432                 kunmap(page);
7433                 kvm_release_page_clean(page);
7434                 nested_vmx_failInvalid(vcpu);
7435                 return kvm_skip_emulated_instruction(vcpu);
7436         }
7437         kunmap(page);
7438         kvm_release_page_clean(page);
7439
7440         vmx->nested.vmxon_ptr = vmptr;
7441         ret = enter_vmx_operation(vcpu);
7442         if (ret)
7443                 return ret;
7444
7445         nested_vmx_succeed(vcpu);
7446         return kvm_skip_emulated_instruction(vcpu);
7447 }
7448
7449 /*
7450  * Intel's VMX Instruction Reference specifies a common set of prerequisites
7451  * for running VMX instructions (except VMXON, whose prerequisites are
7452  * slightly different). It also specifies what exception to inject otherwise.
7453  * Note that many of these exceptions have priority over VM exits, so they
7454  * don't have to be checked again here.
7455  */
7456 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7457 {
7458         if (!to_vmx(vcpu)->nested.vmxon) {
7459                 kvm_queue_exception(vcpu, UD_VECTOR);
7460                 return 0;
7461         }
7462         return 1;
7463 }
7464
7465 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
7466 {
7467         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
7468         vmcs_write64(VMCS_LINK_POINTER, -1ull);
7469 }
7470
7471 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7472 {
7473         if (vmx->nested.current_vmptr == -1ull)
7474                 return;
7475
7476         if (enable_shadow_vmcs) {
7477                 /* copy to memory all shadowed fields in case
7478                    they were modified */
7479                 copy_shadow_to_vmcs12(vmx);
7480                 vmx->nested.sync_shadow_vmcs = false;
7481                 vmx_disable_shadow_vmcs(vmx);
7482         }
7483         vmx->nested.posted_intr_nv = -1;
7484
7485         /* Flush VMCS12 to guest memory */
7486         kvm_vcpu_write_guest_page(&vmx->vcpu,
7487                                   vmx->nested.current_vmptr >> PAGE_SHIFT,
7488                                   vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
7489
7490         vmx->nested.current_vmptr = -1ull;
7491 }
7492
7493 /*
7494  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7495  * just stops using VMX.
7496  */
7497 static void free_nested(struct vcpu_vmx *vmx)
7498 {
7499         if (!vmx->nested.vmxon)
7500                 return;
7501
7502         vmx->nested.vmxon = false;
7503         free_vpid(vmx->nested.vpid02);
7504         vmx->nested.posted_intr_nv = -1;
7505         vmx->nested.current_vmptr = -1ull;
7506         if (enable_shadow_vmcs) {
7507                 vmx_disable_shadow_vmcs(vmx);
7508                 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7509                 free_vmcs(vmx->vmcs01.shadow_vmcs);
7510                 vmx->vmcs01.shadow_vmcs = NULL;
7511         }
7512         kfree(vmx->nested.cached_vmcs12);
7513         /* Unpin physical memory we referred to in the vmcs02 */
7514         if (vmx->nested.apic_access_page) {
7515                 kvm_release_page_dirty(vmx->nested.apic_access_page);
7516                 vmx->nested.apic_access_page = NULL;
7517         }
7518         if (vmx->nested.virtual_apic_page) {
7519                 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
7520                 vmx->nested.virtual_apic_page = NULL;
7521         }
7522         if (vmx->nested.pi_desc_page) {
7523                 kunmap(vmx->nested.pi_desc_page);
7524                 kvm_release_page_dirty(vmx->nested.pi_desc_page);
7525                 vmx->nested.pi_desc_page = NULL;
7526                 vmx->nested.pi_desc = NULL;
7527         }
7528
7529         free_loaded_vmcs(&vmx->nested.vmcs02);
7530 }
7531
7532 /* Emulate the VMXOFF instruction */
7533 static int handle_vmoff(struct kvm_vcpu *vcpu)
7534 {
7535         if (!nested_vmx_check_permission(vcpu))
7536                 return 1;
7537         free_nested(to_vmx(vcpu));
7538         nested_vmx_succeed(vcpu);
7539         return kvm_skip_emulated_instruction(vcpu);
7540 }
7541
7542 /* Emulate the VMCLEAR instruction */
7543 static int handle_vmclear(struct kvm_vcpu *vcpu)
7544 {
7545         struct vcpu_vmx *vmx = to_vmx(vcpu);
7546         u32 zero = 0;
7547         gpa_t vmptr;
7548
7549         if (!nested_vmx_check_permission(vcpu))
7550                 return 1;
7551
7552         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7553                 return 1;
7554
7555         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7556                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
7557                 return kvm_skip_emulated_instruction(vcpu);
7558         }
7559
7560         if (vmptr == vmx->nested.vmxon_ptr) {
7561                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
7562                 return kvm_skip_emulated_instruction(vcpu);
7563         }
7564
7565         if (vmptr == vmx->nested.current_vmptr)
7566                 nested_release_vmcs12(vmx);
7567
7568         kvm_vcpu_write_guest(vcpu,
7569                         vmptr + offsetof(struct vmcs12, launch_state),
7570                         &zero, sizeof(zero));
7571
7572         nested_vmx_succeed(vcpu);
7573         return kvm_skip_emulated_instruction(vcpu);
7574 }
7575
7576 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7577
7578 /* Emulate the VMLAUNCH instruction */
7579 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7580 {
7581         return nested_vmx_run(vcpu, true);
7582 }
7583
7584 /* Emulate the VMRESUME instruction */
7585 static int handle_vmresume(struct kvm_vcpu *vcpu)
7586 {
7587
7588         return nested_vmx_run(vcpu, false);
7589 }
7590
7591 /*
7592  * Read a vmcs12 field. Since these can have varying lengths and we return
7593  * one type, we chose the biggest type (u64) and zero-extend the return value
7594  * to that size. Note that the caller, handle_vmread, might need to use only
7595  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7596  * 64-bit fields are to be returned).
7597  */
7598 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7599                                   unsigned long field, u64 *ret)
7600 {
7601         short offset = vmcs_field_to_offset(field);
7602         char *p;
7603
7604         if (offset < 0)
7605                 return offset;
7606
7607         p = ((char *)(get_vmcs12(vcpu))) + offset;
7608
7609         switch (vmcs_field_type(field)) {
7610         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7611                 *ret = *((natural_width *)p);
7612                 return 0;
7613         case VMCS_FIELD_TYPE_U16:
7614                 *ret = *((u16 *)p);
7615                 return 0;
7616         case VMCS_FIELD_TYPE_U32:
7617                 *ret = *((u32 *)p);
7618                 return 0;
7619         case VMCS_FIELD_TYPE_U64:
7620                 *ret = *((u64 *)p);
7621                 return 0;
7622         default:
7623                 WARN_ON(1);
7624                 return -ENOENT;
7625         }
7626 }
7627
7628
7629 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7630                                    unsigned long field, u64 field_value){
7631         short offset = vmcs_field_to_offset(field);
7632         char *p = ((char *) get_vmcs12(vcpu)) + offset;
7633         if (offset < 0)
7634                 return offset;
7635
7636         switch (vmcs_field_type(field)) {
7637         case VMCS_FIELD_TYPE_U16:
7638                 *(u16 *)p = field_value;
7639                 return 0;
7640         case VMCS_FIELD_TYPE_U32:
7641                 *(u32 *)p = field_value;
7642                 return 0;
7643         case VMCS_FIELD_TYPE_U64:
7644                 *(u64 *)p = field_value;
7645                 return 0;
7646         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7647                 *(natural_width *)p = field_value;
7648                 return 0;
7649         default:
7650                 WARN_ON(1);
7651                 return -ENOENT;
7652         }
7653
7654 }
7655
7656 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7657 {
7658         int i;
7659         unsigned long field;
7660         u64 field_value;
7661         struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7662         const unsigned long *fields = shadow_read_write_fields;
7663         const int num_fields = max_shadow_read_write_fields;
7664
7665         preempt_disable();
7666
7667         vmcs_load(shadow_vmcs);
7668
7669         for (i = 0; i < num_fields; i++) {
7670                 field = fields[i];
7671                 switch (vmcs_field_type(field)) {
7672                 case VMCS_FIELD_TYPE_U16:
7673                         field_value = vmcs_read16(field);
7674                         break;
7675                 case VMCS_FIELD_TYPE_U32:
7676                         field_value = vmcs_read32(field);
7677                         break;
7678                 case VMCS_FIELD_TYPE_U64:
7679                         field_value = vmcs_read64(field);
7680                         break;
7681                 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7682                         field_value = vmcs_readl(field);
7683                         break;
7684                 default:
7685                         WARN_ON(1);
7686                         continue;
7687                 }
7688                 vmcs12_write_any(&vmx->vcpu, field, field_value);
7689         }
7690
7691         vmcs_clear(shadow_vmcs);
7692         vmcs_load(vmx->loaded_vmcs->vmcs);
7693
7694         preempt_enable();
7695 }
7696
7697 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7698 {
7699         const unsigned long *fields[] = {
7700                 shadow_read_write_fields,
7701                 shadow_read_only_fields
7702         };
7703         const int max_fields[] = {
7704                 max_shadow_read_write_fields,
7705                 max_shadow_read_only_fields
7706         };
7707         int i, q;
7708         unsigned long field;
7709         u64 field_value = 0;
7710         struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7711
7712         vmcs_load(shadow_vmcs);
7713
7714         for (q = 0; q < ARRAY_SIZE(fields); q++) {
7715                 for (i = 0; i < max_fields[q]; i++) {
7716                         field = fields[q][i];
7717                         vmcs12_read_any(&vmx->vcpu, field, &field_value);
7718
7719                         switch (vmcs_field_type(field)) {
7720                         case VMCS_FIELD_TYPE_U16:
7721                                 vmcs_write16(field, (u16)field_value);
7722                                 break;
7723                         case VMCS_FIELD_TYPE_U32:
7724                                 vmcs_write32(field, (u32)field_value);
7725                                 break;
7726                         case VMCS_FIELD_TYPE_U64:
7727                                 vmcs_write64(field, (u64)field_value);
7728                                 break;
7729                         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7730                                 vmcs_writel(field, (long)field_value);
7731                                 break;
7732                         default:
7733                                 WARN_ON(1);
7734                                 break;
7735                         }
7736                 }
7737         }
7738
7739         vmcs_clear(shadow_vmcs);
7740         vmcs_load(vmx->loaded_vmcs->vmcs);
7741 }
7742
7743 /*
7744  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7745  * used before) all generate the same failure when it is missing.
7746  */
7747 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7748 {
7749         struct vcpu_vmx *vmx = to_vmx(vcpu);
7750         if (vmx->nested.current_vmptr == -1ull) {
7751                 nested_vmx_failInvalid(vcpu);
7752                 return 0;
7753         }
7754         return 1;
7755 }
7756
7757 static int handle_vmread(struct kvm_vcpu *vcpu)
7758 {
7759         unsigned long field;
7760         u64 field_value;
7761         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7762         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7763         gva_t gva = 0;
7764
7765         if (!nested_vmx_check_permission(vcpu))
7766                 return 1;
7767
7768         if (!nested_vmx_check_vmcs12(vcpu))
7769                 return kvm_skip_emulated_instruction(vcpu);
7770
7771         /* Decode instruction info and find the field to read */
7772         field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7773         /* Read the field, zero-extended to a u64 field_value */
7774         if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7775                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7776                 return kvm_skip_emulated_instruction(vcpu);
7777         }
7778         /*
7779          * Now copy part of this value to register or memory, as requested.
7780          * Note that the number of bits actually copied is 32 or 64 depending
7781          * on the guest's mode (32 or 64 bit), not on the given field's length.
7782          */
7783         if (vmx_instruction_info & (1u << 10)) {
7784                 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7785                         field_value);
7786         } else {
7787                 if (get_vmx_mem_address(vcpu, exit_qualification,
7788                                 vmx_instruction_info, true, &gva))
7789                         return 1;
7790                 /* _system ok, as hardware has verified cpl=0 */
7791                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7792                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7793         }
7794
7795         nested_vmx_succeed(vcpu);
7796         return kvm_skip_emulated_instruction(vcpu);
7797 }
7798
7799
7800 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7801 {
7802         unsigned long field;
7803         gva_t gva;
7804         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7805         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7806         /* The value to write might be 32 or 64 bits, depending on L1's long
7807          * mode, and eventually we need to write that into a field of several
7808          * possible lengths. The code below first zero-extends the value to 64
7809          * bit (field_value), and then copies only the appropriate number of
7810          * bits into the vmcs12 field.
7811          */
7812         u64 field_value = 0;
7813         struct x86_exception e;
7814
7815         if (!nested_vmx_check_permission(vcpu))
7816                 return 1;
7817
7818         if (!nested_vmx_check_vmcs12(vcpu))
7819                 return kvm_skip_emulated_instruction(vcpu);
7820
7821         if (vmx_instruction_info & (1u << 10))
7822                 field_value = kvm_register_readl(vcpu,
7823                         (((vmx_instruction_info) >> 3) & 0xf));
7824         else {
7825                 if (get_vmx_mem_address(vcpu, exit_qualification,
7826                                 vmx_instruction_info, false, &gva))
7827                         return 1;
7828                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7829                            &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7830                         kvm_inject_page_fault(vcpu, &e);
7831                         return 1;
7832                 }
7833         }
7834
7835
7836         field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7837         if (vmcs_field_readonly(field)) {
7838                 nested_vmx_failValid(vcpu,
7839                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7840                 return kvm_skip_emulated_instruction(vcpu);
7841         }
7842
7843         if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7844                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7845                 return kvm_skip_emulated_instruction(vcpu);
7846         }
7847
7848         nested_vmx_succeed(vcpu);
7849         return kvm_skip_emulated_instruction(vcpu);
7850 }
7851
7852 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7853 {
7854         vmx->nested.current_vmptr = vmptr;
7855         if (enable_shadow_vmcs) {
7856                 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7857                               SECONDARY_EXEC_SHADOW_VMCS);
7858                 vmcs_write64(VMCS_LINK_POINTER,
7859                              __pa(vmx->vmcs01.shadow_vmcs));
7860                 vmx->nested.sync_shadow_vmcs = true;
7861         }
7862 }
7863
7864 /* Emulate the VMPTRLD instruction */
7865 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7866 {
7867         struct vcpu_vmx *vmx = to_vmx(vcpu);
7868         gpa_t vmptr;
7869
7870         if (!nested_vmx_check_permission(vcpu))
7871                 return 1;
7872
7873         if (nested_vmx_get_vmptr(vcpu, &vmptr))
7874                 return 1;
7875
7876         if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7877                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
7878                 return kvm_skip_emulated_instruction(vcpu);
7879         }
7880
7881         if (vmptr == vmx->nested.vmxon_ptr) {
7882                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
7883                 return kvm_skip_emulated_instruction(vcpu);
7884         }
7885
7886         if (vmx->nested.current_vmptr != vmptr) {
7887                 struct vmcs12 *new_vmcs12;
7888                 struct page *page;
7889                 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7890                 if (is_error_page(page)) {
7891                         nested_vmx_failInvalid(vcpu);
7892                         return kvm_skip_emulated_instruction(vcpu);
7893                 }
7894                 new_vmcs12 = kmap(page);
7895                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7896                         kunmap(page);
7897                         kvm_release_page_clean(page);
7898                         nested_vmx_failValid(vcpu,
7899                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7900                         return kvm_skip_emulated_instruction(vcpu);
7901                 }
7902
7903                 nested_release_vmcs12(vmx);
7904                 /*
7905                  * Load VMCS12 from guest memory since it is not already
7906                  * cached.
7907                  */
7908                 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
7909                 kunmap(page);
7910                 kvm_release_page_clean(page);
7911
7912                 set_current_vmptr(vmx, vmptr);
7913         }
7914
7915         nested_vmx_succeed(vcpu);
7916         return kvm_skip_emulated_instruction(vcpu);
7917 }
7918
7919 /* Emulate the VMPTRST instruction */
7920 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7921 {
7922         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7923         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7924         gva_t vmcs_gva;
7925         struct x86_exception e;
7926
7927         if (!nested_vmx_check_permission(vcpu))
7928                 return 1;
7929
7930         if (get_vmx_mem_address(vcpu, exit_qualification,
7931                         vmx_instruction_info, true, &vmcs_gva))
7932                 return 1;
7933         /* ok to use *_system, as hardware has verified cpl=0 */
7934         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7935                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
7936                                  sizeof(u64), &e)) {
7937                 kvm_inject_page_fault(vcpu, &e);
7938                 return 1;
7939         }
7940         nested_vmx_succeed(vcpu);
7941         return kvm_skip_emulated_instruction(vcpu);
7942 }
7943
7944 /* Emulate the INVEPT instruction */
7945 static int handle_invept(struct kvm_vcpu *vcpu)
7946 {
7947         struct vcpu_vmx *vmx = to_vmx(vcpu);
7948         u32 vmx_instruction_info, types;
7949         unsigned long type;
7950         gva_t gva;
7951         struct x86_exception e;
7952         struct {
7953                 u64 eptp, gpa;
7954         } operand;
7955
7956         if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7957               SECONDARY_EXEC_ENABLE_EPT) ||
7958             !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7959                 kvm_queue_exception(vcpu, UD_VECTOR);
7960                 return 1;
7961         }
7962
7963         if (!nested_vmx_check_permission(vcpu))
7964                 return 1;
7965
7966         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7967         type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7968
7969         types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7970
7971         if (type >= 32 || !(types & (1 << type))) {
7972                 nested_vmx_failValid(vcpu,
7973                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7974                 return kvm_skip_emulated_instruction(vcpu);
7975         }
7976
7977         /* According to the Intel VMX instruction reference, the memory
7978          * operand is read even if it isn't needed (e.g., for type==global)
7979          */
7980         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7981                         vmx_instruction_info, false, &gva))
7982                 return 1;
7983         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7984                                 sizeof(operand), &e)) {
7985                 kvm_inject_page_fault(vcpu, &e);
7986                 return 1;
7987         }
7988
7989         switch (type) {
7990         case VMX_EPT_EXTENT_GLOBAL:
7991         /*
7992          * TODO: track mappings and invalidate
7993          * single context requests appropriately
7994          */
7995         case VMX_EPT_EXTENT_CONTEXT:
7996                 kvm_mmu_sync_roots(vcpu);
7997                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7998                 nested_vmx_succeed(vcpu);
7999                 break;
8000         default:
8001                 BUG_ON(1);
8002                 break;
8003         }
8004
8005         return kvm_skip_emulated_instruction(vcpu);
8006 }
8007
8008 static int handle_invvpid(struct kvm_vcpu *vcpu)
8009 {
8010         struct vcpu_vmx *vmx = to_vmx(vcpu);
8011         u32 vmx_instruction_info;
8012         unsigned long type, types;
8013         gva_t gva;
8014         struct x86_exception e;
8015         struct {
8016                 u64 vpid;
8017                 u64 gla;
8018         } operand;
8019
8020         if (!(vmx->nested.nested_vmx_secondary_ctls_high &
8021               SECONDARY_EXEC_ENABLE_VPID) ||
8022                         !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
8023                 kvm_queue_exception(vcpu, UD_VECTOR);
8024                 return 1;
8025         }
8026
8027         if (!nested_vmx_check_permission(vcpu))
8028                 return 1;
8029
8030         vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8031         type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
8032
8033         types = (vmx->nested.nested_vmx_vpid_caps &
8034                         VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
8035
8036         if (type >= 32 || !(types & (1 << type))) {
8037                 nested_vmx_failValid(vcpu,
8038                         VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8039                 return kvm_skip_emulated_instruction(vcpu);
8040         }
8041
8042         /* according to the intel vmx instruction reference, the memory
8043          * operand is read even if it isn't needed (e.g., for type==global)
8044          */
8045         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
8046                         vmx_instruction_info, false, &gva))
8047                 return 1;
8048         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
8049                                 sizeof(operand), &e)) {
8050                 kvm_inject_page_fault(vcpu, &e);
8051                 return 1;
8052         }
8053         if (operand.vpid >> 16) {
8054                 nested_vmx_failValid(vcpu,
8055                         VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8056                 return kvm_skip_emulated_instruction(vcpu);
8057         }
8058
8059         switch (type) {
8060         case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
8061                 if (is_noncanonical_address(operand.gla, vcpu)) {
8062                         nested_vmx_failValid(vcpu,
8063                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8064                         return kvm_skip_emulated_instruction(vcpu);
8065                 }
8066                 /* fall through */
8067         case VMX_VPID_EXTENT_SINGLE_CONTEXT:
8068         case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
8069                 if (!operand.vpid) {
8070                         nested_vmx_failValid(vcpu,
8071                                 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8072                         return kvm_skip_emulated_instruction(vcpu);
8073                 }
8074                 break;
8075         case VMX_VPID_EXTENT_ALL_CONTEXT:
8076                 break;
8077         default:
8078                 WARN_ON_ONCE(1);
8079                 return kvm_skip_emulated_instruction(vcpu);
8080         }
8081
8082         __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
8083         nested_vmx_succeed(vcpu);
8084
8085         return kvm_skip_emulated_instruction(vcpu);
8086 }
8087
8088 static int handle_pml_full(struct kvm_vcpu *vcpu)
8089 {
8090         unsigned long exit_qualification;
8091
8092         trace_kvm_pml_full(vcpu->vcpu_id);
8093
8094         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8095
8096         /*
8097          * PML buffer FULL happened while executing iret from NMI,
8098          * "blocked by NMI" bit has to be set before next VM entry.
8099          */
8100         if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
8101                         cpu_has_virtual_nmis() &&
8102                         (exit_qualification & INTR_INFO_UNBLOCK_NMI))
8103                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8104                                 GUEST_INTR_STATE_NMI);
8105
8106         /*
8107          * PML buffer already flushed at beginning of VMEXIT. Nothing to do
8108          * here.., and there's no userspace involvement needed for PML.
8109          */
8110         return 1;
8111 }
8112
8113 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
8114 {
8115         kvm_lapic_expired_hv_timer(vcpu);
8116         return 1;
8117 }
8118
8119 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
8120 {
8121         struct vcpu_vmx *vmx = to_vmx(vcpu);
8122         int maxphyaddr = cpuid_maxphyaddr(vcpu);
8123
8124         /* Check for memory type validity */
8125         switch (address & VMX_EPTP_MT_MASK) {
8126         case VMX_EPTP_MT_UC:
8127                 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_UC_BIT))
8128                         return false;
8129                 break;
8130         case VMX_EPTP_MT_WB:
8131                 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_WB_BIT))
8132                         return false;
8133                 break;
8134         default:
8135                 return false;
8136         }
8137
8138         /* only 4 levels page-walk length are valid */
8139         if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
8140                 return false;
8141
8142         /* Reserved bits should not be set */
8143         if (address >> maxphyaddr || ((address >> 7) & 0x1f))
8144                 return false;
8145
8146         /* AD, if set, should be supported */
8147         if (address & VMX_EPTP_AD_ENABLE_BIT) {
8148                 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPT_AD_BIT))
8149                         return false;
8150         }
8151
8152         return true;
8153 }
8154
8155 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
8156                                      struct vmcs12 *vmcs12)
8157 {
8158         u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
8159         u64 address;
8160         bool accessed_dirty;
8161         struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
8162
8163         if (!nested_cpu_has_eptp_switching(vmcs12) ||
8164             !nested_cpu_has_ept(vmcs12))
8165                 return 1;
8166
8167         if (index >= VMFUNC_EPTP_ENTRIES)
8168                 return 1;
8169
8170
8171         if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
8172                                      &address, index * 8, 8))
8173                 return 1;
8174
8175         accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
8176
8177         /*
8178          * If the (L2) guest does a vmfunc to the currently
8179          * active ept pointer, we don't have to do anything else
8180          */
8181         if (vmcs12->ept_pointer != address) {
8182                 if (!valid_ept_address(vcpu, address))
8183                         return 1;
8184
8185                 kvm_mmu_unload(vcpu);
8186                 mmu->ept_ad = accessed_dirty;
8187                 mmu->base_role.ad_disabled = !accessed_dirty;
8188                 vmcs12->ept_pointer = address;
8189                 /*
8190                  * TODO: Check what's the correct approach in case
8191                  * mmu reload fails. Currently, we just let the next
8192                  * reload potentially fail
8193                  */
8194                 kvm_mmu_reload(vcpu);
8195         }
8196
8197         return 0;
8198 }
8199
8200 static int handle_vmfunc(struct kvm_vcpu *vcpu)
8201 {
8202         struct vcpu_vmx *vmx = to_vmx(vcpu);
8203         struct vmcs12 *vmcs12;
8204         u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
8205
8206         /*
8207          * VMFUNC is only supported for nested guests, but we always enable the
8208          * secondary control for simplicity; for non-nested mode, fake that we
8209          * didn't by injecting #UD.
8210          */
8211         if (!is_guest_mode(vcpu)) {
8212                 kvm_queue_exception(vcpu, UD_VECTOR);
8213                 return 1;
8214         }
8215
8216         vmcs12 = get_vmcs12(vcpu);
8217         if ((vmcs12->vm_function_control & (1 << function)) == 0)
8218                 goto fail;
8219
8220         switch (function) {
8221         case 0:
8222                 if (nested_vmx_eptp_switching(vcpu, vmcs12))
8223                         goto fail;
8224                 break;
8225         default:
8226                 goto fail;
8227         }
8228         return kvm_skip_emulated_instruction(vcpu);
8229
8230 fail:
8231         nested_vmx_vmexit(vcpu, vmx->exit_reason,
8232                           vmcs_read32(VM_EXIT_INTR_INFO),
8233                           vmcs_readl(EXIT_QUALIFICATION));
8234         return 1;
8235 }
8236
8237 /*
8238  * The exit handlers return 1 if the exit was handled fully and guest execution
8239  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
8240  * to be done to userspace and return 0.
8241  */
8242 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
8243         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
8244         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
8245         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
8246         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
8247         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
8248         [EXIT_REASON_CR_ACCESS]               = handle_cr,
8249         [EXIT_REASON_DR_ACCESS]               = handle_dr,
8250         [EXIT_REASON_CPUID]                   = handle_cpuid,
8251         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
8252         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
8253         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
8254         [EXIT_REASON_HLT]                     = handle_halt,
8255         [EXIT_REASON_INVD]                    = handle_invd,
8256         [EXIT_REASON_INVLPG]                  = handle_invlpg,
8257         [EXIT_REASON_RDPMC]                   = handle_rdpmc,
8258         [EXIT_REASON_VMCALL]                  = handle_vmcall,
8259         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
8260         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
8261         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
8262         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
8263         [EXIT_REASON_VMREAD]                  = handle_vmread,
8264         [EXIT_REASON_VMRESUME]                = handle_vmresume,
8265         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
8266         [EXIT_REASON_VMOFF]                   = handle_vmoff,
8267         [EXIT_REASON_VMON]                    = handle_vmon,
8268         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
8269         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
8270         [EXIT_REASON_APIC_WRITE]              = handle_apic_write,
8271         [EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
8272         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
8273         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
8274         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
8275         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
8276         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
8277         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
8278         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
8279         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_mwait,
8280         [EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
8281         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
8282         [EXIT_REASON_INVEPT]                  = handle_invept,
8283         [EXIT_REASON_INVVPID]                 = handle_invvpid,
8284         [EXIT_REASON_RDRAND]                  = handle_invalid_op,
8285         [EXIT_REASON_RDSEED]                  = handle_invalid_op,
8286         [EXIT_REASON_XSAVES]                  = handle_xsaves,
8287         [EXIT_REASON_XRSTORS]                 = handle_xrstors,
8288         [EXIT_REASON_PML_FULL]                = handle_pml_full,
8289         [EXIT_REASON_VMFUNC]                  = handle_vmfunc,
8290         [EXIT_REASON_PREEMPTION_TIMER]        = handle_preemption_timer,
8291 };
8292
8293 static const int kvm_vmx_max_exit_handlers =
8294         ARRAY_SIZE(kvm_vmx_exit_handlers);
8295
8296 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
8297                                        struct vmcs12 *vmcs12)
8298 {
8299         unsigned long exit_qualification;
8300         gpa_t bitmap, last_bitmap;
8301         unsigned int port;
8302         int size;
8303         u8 b;
8304
8305         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
8306                 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
8307
8308         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8309
8310         port = exit_qualification >> 16;
8311         size = (exit_qualification & 7) + 1;
8312
8313         last_bitmap = (gpa_t)-1;
8314         b = -1;
8315
8316         while (size > 0) {
8317                 if (port < 0x8000)
8318                         bitmap = vmcs12->io_bitmap_a;
8319                 else if (port < 0x10000)
8320                         bitmap = vmcs12->io_bitmap_b;
8321                 else
8322                         return true;
8323                 bitmap += (port & 0x7fff) / 8;
8324
8325                 if (last_bitmap != bitmap)
8326                         if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
8327                                 return true;
8328                 if (b & (1 << (port & 7)))
8329                         return true;
8330
8331                 port++;
8332                 size--;
8333                 last_bitmap = bitmap;
8334         }
8335
8336         return false;
8337 }
8338
8339 /*
8340  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
8341  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
8342  * disinterest in the current event (read or write a specific MSR) by using an
8343  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
8344  */
8345 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
8346         struct vmcs12 *vmcs12, u32 exit_reason)
8347 {
8348         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
8349         gpa_t bitmap;
8350
8351         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
8352                 return true;
8353
8354         /*
8355          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
8356          * for the four combinations of read/write and low/high MSR numbers.
8357          * First we need to figure out which of the four to use:
8358          */
8359         bitmap = vmcs12->msr_bitmap;
8360         if (exit_reason == EXIT_REASON_MSR_WRITE)
8361                 bitmap += 2048;
8362         if (msr_index >= 0xc0000000) {
8363                 msr_index -= 0xc0000000;
8364                 bitmap += 1024;
8365         }
8366
8367         /* Then read the msr_index'th bit from this bitmap: */
8368         if (msr_index < 1024*8) {
8369                 unsigned char b;
8370                 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
8371                         return true;
8372                 return 1 & (b >> (msr_index & 7));
8373         } else
8374                 return true; /* let L1 handle the wrong parameter */
8375 }
8376
8377 /*
8378  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
8379  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
8380  * intercept (via guest_host_mask etc.) the current event.
8381  */
8382 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
8383         struct vmcs12 *vmcs12)
8384 {
8385         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8386         int cr = exit_qualification & 15;
8387         int reg;
8388         unsigned long val;
8389
8390         switch ((exit_qualification >> 4) & 3) {
8391         case 0: /* mov to cr */
8392                 reg = (exit_qualification >> 8) & 15;
8393                 val = kvm_register_readl(vcpu, reg);
8394                 switch (cr) {
8395                 case 0:
8396                         if (vmcs12->cr0_guest_host_mask &
8397                             (val ^ vmcs12->cr0_read_shadow))
8398                                 return true;
8399                         break;
8400                 case 3:
8401                         if ((vmcs12->cr3_target_count >= 1 &&
8402                                         vmcs12->cr3_target_value0 == val) ||
8403                                 (vmcs12->cr3_target_count >= 2 &&
8404                                         vmcs12->cr3_target_value1 == val) ||
8405                                 (vmcs12->cr3_target_count >= 3 &&
8406                                         vmcs12->cr3_target_value2 == val) ||
8407                                 (vmcs12->cr3_target_count >= 4 &&
8408                                         vmcs12->cr3_target_value3 == val))
8409                                 return false;
8410                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8411                                 return true;
8412                         break;
8413                 case 4:
8414                         if (vmcs12->cr4_guest_host_mask &
8415                             (vmcs12->cr4_read_shadow ^ val))
8416                                 return true;
8417                         break;
8418                 case 8:
8419                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8420                                 return true;
8421                         break;
8422                 }
8423                 break;
8424         case 2: /* clts */
8425                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8426                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
8427                         return true;
8428                 break;
8429         case 1: /* mov from cr */
8430                 switch (cr) {
8431                 case 3:
8432                         if (vmcs12->cpu_based_vm_exec_control &
8433                             CPU_BASED_CR3_STORE_EXITING)
8434                                 return true;
8435                         break;
8436                 case 8:
8437                         if (vmcs12->cpu_based_vm_exec_control &
8438                             CPU_BASED_CR8_STORE_EXITING)
8439                                 return true;
8440                         break;
8441                 }
8442                 break;
8443         case 3: /* lmsw */
8444                 /*
8445                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8446                  * cr0. Other attempted changes are ignored, with no exit.
8447                  */
8448                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
8449                 if (vmcs12->cr0_guest_host_mask & 0xe &
8450                     (val ^ vmcs12->cr0_read_shadow))
8451                         return true;
8452                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8453                     !(vmcs12->cr0_read_shadow & 0x1) &&
8454                     (val & 0x1))
8455                         return true;
8456                 break;
8457         }
8458         return false;
8459 }
8460
8461 /*
8462  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8463  * should handle it ourselves in L0 (and then continue L2). Only call this
8464  * when in is_guest_mode (L2).
8465  */
8466 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
8467 {
8468         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8469         struct vcpu_vmx *vmx = to_vmx(vcpu);
8470         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8471
8472         if (vmx->nested.nested_run_pending)
8473                 return false;
8474
8475         if (unlikely(vmx->fail)) {
8476                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8477                                     vmcs_read32(VM_INSTRUCTION_ERROR));
8478                 return true;
8479         }
8480
8481         /*
8482          * The host physical addresses of some pages of guest memory
8483          * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
8484          * Page). The CPU may write to these pages via their host
8485          * physical address while L2 is running, bypassing any
8486          * address-translation-based dirty tracking (e.g. EPT write
8487          * protection).
8488          *
8489          * Mark them dirty on every exit from L2 to prevent them from
8490          * getting out of sync with dirty tracking.
8491          */
8492         nested_mark_vmcs12_pages_dirty(vcpu);
8493
8494         trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8495                                 vmcs_readl(EXIT_QUALIFICATION),
8496                                 vmx->idt_vectoring_info,
8497                                 intr_info,
8498                                 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8499                                 KVM_ISA_VMX);
8500
8501         switch (exit_reason) {
8502         case EXIT_REASON_EXCEPTION_NMI:
8503                 if (is_nmi(intr_info))
8504                         return false;
8505                 else if (is_page_fault(intr_info))
8506                         return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
8507                 else if (is_no_device(intr_info) &&
8508                          !(vmcs12->guest_cr0 & X86_CR0_TS))
8509                         return false;
8510                 else if (is_debug(intr_info) &&
8511                          vcpu->guest_debug &
8512                          (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8513                         return false;
8514                 else if (is_breakpoint(intr_info) &&
8515                          vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8516                         return false;
8517                 return vmcs12->exception_bitmap &
8518                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8519         case EXIT_REASON_EXTERNAL_INTERRUPT:
8520                 return false;
8521         case EXIT_REASON_TRIPLE_FAULT:
8522                 return true;
8523         case EXIT_REASON_PENDING_INTERRUPT:
8524                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8525         case EXIT_REASON_NMI_WINDOW:
8526                 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8527         case EXIT_REASON_TASK_SWITCH:
8528                 return true;
8529         case EXIT_REASON_CPUID:
8530                 return true;
8531         case EXIT_REASON_HLT:
8532                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8533         case EXIT_REASON_INVD:
8534                 return true;
8535         case EXIT_REASON_INVLPG:
8536                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8537         case EXIT_REASON_RDPMC:
8538                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8539         case EXIT_REASON_RDRAND:
8540                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND);
8541         case EXIT_REASON_RDSEED:
8542                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED);
8543         case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8544                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8545         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8546         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8547         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8548         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8549         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8550         case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8551                 /*
8552                  * VMX instructions trap unconditionally. This allows L1 to
8553                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
8554                  */
8555                 return true;
8556         case EXIT_REASON_CR_ACCESS:
8557                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8558         case EXIT_REASON_DR_ACCESS:
8559                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8560         case EXIT_REASON_IO_INSTRUCTION:
8561                 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8562         case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8563                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8564         case EXIT_REASON_MSR_READ:
8565         case EXIT_REASON_MSR_WRITE:
8566                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8567         case EXIT_REASON_INVALID_STATE:
8568                 return true;
8569         case EXIT_REASON_MWAIT_INSTRUCTION:
8570                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8571         case EXIT_REASON_MONITOR_TRAP_FLAG:
8572                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8573         case EXIT_REASON_MONITOR_INSTRUCTION:
8574                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8575         case EXIT_REASON_PAUSE_INSTRUCTION:
8576                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8577                         nested_cpu_has2(vmcs12,
8578                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8579         case EXIT_REASON_MCE_DURING_VMENTRY:
8580                 return false;
8581         case EXIT_REASON_TPR_BELOW_THRESHOLD:
8582                 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8583         case EXIT_REASON_APIC_ACCESS:
8584                 return nested_cpu_has2(vmcs12,
8585                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8586         case EXIT_REASON_APIC_WRITE:
8587         case EXIT_REASON_EOI_INDUCED:
8588                 /* apic_write and eoi_induced should exit unconditionally. */
8589                 return true;
8590         case EXIT_REASON_EPT_VIOLATION:
8591                 /*
8592                  * L0 always deals with the EPT violation. If nested EPT is
8593                  * used, and the nested mmu code discovers that the address is
8594                  * missing in the guest EPT table (EPT12), the EPT violation
8595                  * will be injected with nested_ept_inject_page_fault()
8596                  */
8597                 return false;
8598         case EXIT_REASON_EPT_MISCONFIG:
8599                 /*
8600                  * L2 never uses directly L1's EPT, but rather L0's own EPT
8601                  * table (shadow on EPT) or a merged EPT table that L0 built
8602                  * (EPT on EPT). So any problems with the structure of the
8603                  * table is L0's fault.
8604                  */
8605                 return false;
8606         case EXIT_REASON_INVPCID:
8607                 return
8608                         nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
8609                         nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8610         case EXIT_REASON_WBINVD:
8611                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8612         case EXIT_REASON_XSETBV:
8613                 return true;
8614         case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8615                 /*
8616                  * This should never happen, since it is not possible to
8617                  * set XSS to a non-zero value---neither in L1 nor in L2.
8618                  * If if it were, XSS would have to be checked against
8619                  * the XSS exit bitmap in vmcs12.
8620                  */
8621                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8622         case EXIT_REASON_PREEMPTION_TIMER:
8623                 return false;
8624         case EXIT_REASON_PML_FULL:
8625                 /* We emulate PML support to L1. */
8626                 return false;
8627         case EXIT_REASON_VMFUNC:
8628                 /* VM functions are emulated through L2->L0 vmexits. */
8629                 return false;
8630         default:
8631                 return true;
8632         }
8633 }
8634
8635 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
8636 {
8637         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8638
8639         /*
8640          * At this point, the exit interruption info in exit_intr_info
8641          * is only valid for EXCEPTION_NMI exits.  For EXTERNAL_INTERRUPT
8642          * we need to query the in-kernel LAPIC.
8643          */
8644         WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
8645         if ((exit_intr_info &
8646              (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8647             (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
8648                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8649                 vmcs12->vm_exit_intr_error_code =
8650                         vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8651         }
8652
8653         nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
8654                           vmcs_readl(EXIT_QUALIFICATION));
8655         return 1;
8656 }
8657
8658 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8659 {
8660         *info1 = vmcs_readl(EXIT_QUALIFICATION);
8661         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8662 }
8663
8664 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8665 {
8666         if (vmx->pml_pg) {
8667                 __free_page(vmx->pml_pg);
8668                 vmx->pml_pg = NULL;
8669         }
8670 }
8671
8672 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8673 {
8674         struct vcpu_vmx *vmx = to_vmx(vcpu);
8675         u64 *pml_buf;
8676         u16 pml_idx;
8677
8678         pml_idx = vmcs_read16(GUEST_PML_INDEX);
8679
8680         /* Do nothing if PML buffer is empty */
8681         if (pml_idx == (PML_ENTITY_NUM - 1))
8682                 return;
8683
8684         /* PML index always points to next available PML buffer entity */
8685         if (pml_idx >= PML_ENTITY_NUM)
8686                 pml_idx = 0;
8687         else
8688                 pml_idx++;
8689
8690         pml_buf = page_address(vmx->pml_pg);
8691         for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8692                 u64 gpa;
8693
8694                 gpa = pml_buf[pml_idx];
8695                 WARN_ON(gpa & (PAGE_SIZE - 1));
8696                 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8697         }
8698
8699         /* reset PML index */
8700         vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8701 }
8702
8703 /*
8704  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8705  * Called before reporting dirty_bitmap to userspace.
8706  */
8707 static void kvm_flush_pml_buffers(struct kvm *kvm)
8708 {
8709         int i;
8710         struct kvm_vcpu *vcpu;
8711         /*
8712          * We only need to kick vcpu out of guest mode here, as PML buffer
8713          * is flushed at beginning of all VMEXITs, and it's obvious that only
8714          * vcpus running in guest are possible to have unflushed GPAs in PML
8715          * buffer.
8716          */
8717         kvm_for_each_vcpu(i, vcpu, kvm)
8718                 kvm_vcpu_kick(vcpu);
8719 }
8720
8721 static void vmx_dump_sel(char *name, uint32_t sel)
8722 {
8723         pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8724                name, vmcs_read16(sel),
8725                vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8726                vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8727                vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8728 }
8729
8730 static void vmx_dump_dtsel(char *name, uint32_t limit)
8731 {
8732         pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
8733                name, vmcs_read32(limit),
8734                vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8735 }
8736
8737 static void dump_vmcs(void)
8738 {
8739         u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8740         u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8741         u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8742         u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8743         u32 secondary_exec_control = 0;
8744         unsigned long cr4 = vmcs_readl(GUEST_CR4);
8745         u64 efer = vmcs_read64(GUEST_IA32_EFER);
8746         int i, n;
8747
8748         if (cpu_has_secondary_exec_ctrls())
8749                 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8750
8751         pr_err("*** Guest State ***\n");
8752         pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8753                vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8754                vmcs_readl(CR0_GUEST_HOST_MASK));
8755         pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8756                cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8757         pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8758         if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8759             (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8760         {
8761                 pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
8762                        vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8763                 pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
8764                        vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8765         }
8766         pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
8767                vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8768         pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
8769                vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8770         pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8771                vmcs_readl(GUEST_SYSENTER_ESP),
8772                vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8773         vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
8774         vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
8775         vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
8776         vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
8777         vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
8778         vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
8779         vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8780         vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8781         vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8782         vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
8783         if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8784             (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8785                 pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
8786                        efer, vmcs_read64(GUEST_IA32_PAT));
8787         pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
8788                vmcs_read64(GUEST_IA32_DEBUGCTL),
8789                vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8790         if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8791                 pr_err("PerfGlobCtl = 0x%016llx\n",
8792                        vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8793         if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8794                 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8795         pr_err("Interruptibility = %08x  ActivityState = %08x\n",
8796                vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8797                vmcs_read32(GUEST_ACTIVITY_STATE));
8798         if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8799                 pr_err("InterruptStatus = %04x\n",
8800                        vmcs_read16(GUEST_INTR_STATUS));
8801
8802         pr_err("*** Host State ***\n");
8803         pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
8804                vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8805         pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8806                vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8807                vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8808                vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8809                vmcs_read16(HOST_TR_SELECTOR));
8810         pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8811                vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8812                vmcs_readl(HOST_TR_BASE));
8813         pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8814                vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8815         pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8816                vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8817                vmcs_readl(HOST_CR4));
8818         pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8819                vmcs_readl(HOST_IA32_SYSENTER_ESP),
8820                vmcs_read32(HOST_IA32_SYSENTER_CS),
8821                vmcs_readl(HOST_IA32_SYSENTER_EIP));
8822         if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8823                 pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
8824                        vmcs_read64(HOST_IA32_EFER),
8825                        vmcs_read64(HOST_IA32_PAT));
8826         if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8827                 pr_err("PerfGlobCtl = 0x%016llx\n",
8828                        vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8829
8830         pr_err("*** Control State ***\n");
8831         pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8832                pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8833         pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8834         pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8835                vmcs_read32(EXCEPTION_BITMAP),
8836                vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8837                vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8838         pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8839                vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8840                vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8841                vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8842         pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8843                vmcs_read32(VM_EXIT_INTR_INFO),
8844                vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8845                vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8846         pr_err("        reason=%08x qualification=%016lx\n",
8847                vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8848         pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8849                vmcs_read32(IDT_VECTORING_INFO_FIELD),
8850                vmcs_read32(IDT_VECTORING_ERROR_CODE));
8851         pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8852         if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8853                 pr_err("TSC Multiplier = 0x%016llx\n",
8854                        vmcs_read64(TSC_MULTIPLIER));
8855         if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8856                 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8857         if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8858                 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8859         if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8860                 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8861         n = vmcs_read32(CR3_TARGET_COUNT);
8862         for (i = 0; i + 1 < n; i += 4)
8863                 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8864                        i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8865                        i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8866         if (i < n)
8867                 pr_err("CR3 target%u=%016lx\n",
8868                        i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8869         if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8870                 pr_err("PLE Gap=%08x Window=%08x\n",
8871                        vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8872         if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8873                 pr_err("Virtual processor ID = 0x%04x\n",
8874                        vmcs_read16(VIRTUAL_PROCESSOR_ID));
8875 }
8876
8877 /*
8878  * The guest has exited.  See if we can fix it or if we need userspace
8879  * assistance.
8880  */
8881 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8882 {
8883         struct vcpu_vmx *vmx = to_vmx(vcpu);
8884         u32 exit_reason = vmx->exit_reason;
8885         u32 vectoring_info = vmx->idt_vectoring_info;
8886
8887         trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8888
8889         /*
8890          * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8891          * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8892          * querying dirty_bitmap, we only need to kick all vcpus out of guest
8893          * mode as if vcpus is in root mode, the PML buffer must has been
8894          * flushed already.
8895          */
8896         if (enable_pml)
8897                 vmx_flush_pml_buffer(vcpu);
8898
8899         /* If guest state is invalid, start emulating */
8900         if (vmx->emulation_required)
8901                 return handle_invalid_guest_state(vcpu);
8902
8903         if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
8904                 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
8905
8906         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8907                 dump_vmcs();
8908                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8909                 vcpu->run->fail_entry.hardware_entry_failure_reason
8910                         = exit_reason;
8911                 return 0;
8912         }
8913
8914         if (unlikely(vmx->fail)) {
8915                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8916                 vcpu->run->fail_entry.hardware_entry_failure_reason
8917                         = vmcs_read32(VM_INSTRUCTION_ERROR);
8918                 return 0;
8919         }
8920
8921         /*
8922          * Note:
8923          * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8924          * delivery event since it indicates guest is accessing MMIO.
8925          * The vm-exit can be triggered again after return to guest that
8926          * will cause infinite loop.
8927          */
8928         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8929                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8930                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
8931                         exit_reason != EXIT_REASON_PML_FULL &&
8932                         exit_reason != EXIT_REASON_TASK_SWITCH)) {
8933                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8934                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8935                 vcpu->run->internal.ndata = 3;
8936                 vcpu->run->internal.data[0] = vectoring_info;
8937                 vcpu->run->internal.data[1] = exit_reason;
8938                 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
8939                 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
8940                         vcpu->run->internal.ndata++;
8941                         vcpu->run->internal.data[3] =
8942                                 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
8943                 }
8944                 return 0;
8945         }
8946
8947         if (unlikely(!cpu_has_virtual_nmis() &&
8948                      vmx->loaded_vmcs->soft_vnmi_blocked)) {
8949                 if (vmx_interrupt_allowed(vcpu)) {
8950                         vmx->loaded_vmcs->soft_vnmi_blocked = 0;
8951                 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
8952                            vcpu->arch.nmi_pending) {
8953                         /*
8954                          * This CPU don't support us in finding the end of an
8955                          * NMI-blocked window if the guest runs with IRQs
8956                          * disabled. So we pull the trigger after 1 s of
8957                          * futile waiting, but inform the user about this.
8958                          */
8959                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8960                                "state on VCPU %d after 1 s timeout\n",
8961                                __func__, vcpu->vcpu_id);
8962                         vmx->loaded_vmcs->soft_vnmi_blocked = 0;
8963                 }
8964         }
8965
8966         if (exit_reason < kvm_vmx_max_exit_handlers
8967             && kvm_vmx_exit_handlers[exit_reason])
8968                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8969         else {
8970                 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8971                                 exit_reason);
8972                 kvm_queue_exception(vcpu, UD_VECTOR);
8973                 return 1;
8974         }
8975 }
8976
8977 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8978 {
8979         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8980
8981         if (is_guest_mode(vcpu) &&
8982                 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8983                 return;
8984
8985         if (irr == -1 || tpr < irr) {
8986                 vmcs_write32(TPR_THRESHOLD, 0);
8987                 return;
8988         }
8989
8990         vmcs_write32(TPR_THRESHOLD, irr);
8991 }
8992
8993 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8994 {
8995         u32 sec_exec_control;
8996
8997         /* Postpone execution until vmcs01 is the current VMCS. */
8998         if (is_guest_mode(vcpu)) {
8999                 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
9000                 return;
9001         }
9002
9003         if (!cpu_has_vmx_virtualize_x2apic_mode())
9004                 return;
9005
9006         if (!cpu_need_tpr_shadow(vcpu))
9007                 return;
9008
9009         sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9010
9011         if (set) {
9012                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9013                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
9014         } else {
9015                 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
9016                 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9017                 vmx_flush_tlb_ept_only(vcpu);
9018         }
9019         vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
9020
9021         vmx_update_msr_bitmap(vcpu);
9022 }
9023
9024 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
9025 {
9026         struct vcpu_vmx *vmx = to_vmx(vcpu);
9027
9028         /*
9029          * Currently we do not handle the nested case where L2 has an
9030          * APIC access page of its own; that page is still pinned.
9031          * Hence, we skip the case where the VCPU is in guest mode _and_
9032          * L1 prepared an APIC access page for L2.
9033          *
9034          * For the case where L1 and L2 share the same APIC access page
9035          * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
9036          * in the vmcs12), this function will only update either the vmcs01
9037          * or the vmcs02.  If the former, the vmcs02 will be updated by
9038          * prepare_vmcs02.  If the latter, the vmcs01 will be updated in
9039          * the next L2->L1 exit.
9040          */
9041         if (!is_guest_mode(vcpu) ||
9042             !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
9043                              SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9044                 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9045                 vmx_flush_tlb_ept_only(vcpu);
9046         }
9047 }
9048
9049 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
9050 {
9051         u16 status;
9052         u8 old;
9053
9054         if (max_isr == -1)
9055                 max_isr = 0;
9056
9057         status = vmcs_read16(GUEST_INTR_STATUS);
9058         old = status >> 8;
9059         if (max_isr != old) {
9060                 status &= 0xff;
9061                 status |= max_isr << 8;
9062                 vmcs_write16(GUEST_INTR_STATUS, status);
9063         }
9064 }
9065
9066 static void vmx_set_rvi(int vector)
9067 {
9068         u16 status;
9069         u8 old;
9070
9071         if (vector == -1)
9072                 vector = 0;
9073
9074         status = vmcs_read16(GUEST_INTR_STATUS);
9075         old = (u8)status & 0xff;
9076         if ((u8)vector != old) {
9077                 status &= ~0xff;
9078                 status |= (u8)vector;
9079                 vmcs_write16(GUEST_INTR_STATUS, status);
9080         }
9081 }
9082
9083 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
9084 {
9085         if (!is_guest_mode(vcpu)) {
9086                 vmx_set_rvi(max_irr);
9087                 return;
9088         }
9089
9090         if (max_irr == -1)
9091                 return;
9092
9093         /*
9094          * In guest mode.  If a vmexit is needed, vmx_check_nested_events
9095          * handles it.
9096          */
9097         if (nested_exit_on_intr(vcpu))
9098                 return;
9099
9100         /*
9101          * Else, fall back to pre-APICv interrupt injection since L2
9102          * is run without virtual interrupt delivery.
9103          */
9104         if (!kvm_event_needs_reinjection(vcpu) &&
9105             vmx_interrupt_allowed(vcpu)) {
9106                 kvm_queue_interrupt(vcpu, max_irr, false);
9107                 vmx_inject_irq(vcpu);
9108         }
9109 }
9110
9111 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
9112 {
9113         struct vcpu_vmx *vmx = to_vmx(vcpu);
9114         int max_irr;
9115
9116         WARN_ON(!vcpu->arch.apicv_active);
9117         if (pi_test_on(&vmx->pi_desc)) {
9118                 pi_clear_on(&vmx->pi_desc);
9119                 /*
9120                  * IOMMU can write to PIR.ON, so the barrier matters even on UP.
9121                  * But on x86 this is just a compiler barrier anyway.
9122                  */
9123                 smp_mb__after_atomic();
9124                 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
9125         } else {
9126                 max_irr = kvm_lapic_find_highest_irr(vcpu);
9127         }
9128         vmx_hwapic_irr_update(vcpu, max_irr);
9129         return max_irr;
9130 }
9131
9132 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
9133 {
9134         if (!kvm_vcpu_apicv_active(vcpu))
9135                 return;
9136
9137         vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
9138         vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
9139         vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
9140         vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
9141 }
9142
9143 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
9144 {
9145         struct vcpu_vmx *vmx = to_vmx(vcpu);
9146
9147         pi_clear_on(&vmx->pi_desc);
9148         memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
9149 }
9150
9151 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
9152 {
9153         u32 exit_intr_info = 0;
9154         u16 basic_exit_reason = (u16)vmx->exit_reason;
9155
9156         if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
9157               || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
9158                 return;
9159
9160         if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9161                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9162         vmx->exit_intr_info = exit_intr_info;
9163
9164         /* if exit due to PF check for async PF */
9165         if (is_page_fault(exit_intr_info))
9166                 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
9167
9168         /* Handle machine checks before interrupts are enabled */
9169         if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
9170             is_machine_check(exit_intr_info))
9171                 kvm_machine_check();
9172
9173         /* We need to handle NMIs before interrupts are enabled */
9174         if (is_nmi(exit_intr_info)) {
9175                 kvm_before_handle_nmi(&vmx->vcpu);
9176                 asm("int $2");
9177                 kvm_after_handle_nmi(&vmx->vcpu);
9178         }
9179 }
9180
9181 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
9182 {
9183         u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9184
9185         if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
9186                         == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
9187                 unsigned int vector;
9188                 unsigned long entry;
9189                 gate_desc *desc;
9190                 struct vcpu_vmx *vmx = to_vmx(vcpu);
9191 #ifdef CONFIG_X86_64
9192                 unsigned long tmp;
9193 #endif
9194
9195                 vector =  exit_intr_info & INTR_INFO_VECTOR_MASK;
9196                 desc = (gate_desc *)vmx->host_idt_base + vector;
9197                 entry = gate_offset(desc);
9198                 asm volatile(
9199 #ifdef CONFIG_X86_64
9200                         "mov %%" _ASM_SP ", %[sp]\n\t"
9201                         "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
9202                         "push $%c[ss]\n\t"
9203                         "push %[sp]\n\t"
9204 #endif
9205                         "pushf\n\t"
9206                         __ASM_SIZE(push) " $%c[cs]\n\t"
9207                         CALL_NOSPEC
9208                         :
9209 #ifdef CONFIG_X86_64
9210                         [sp]"=&r"(tmp),
9211 #endif
9212                         ASM_CALL_CONSTRAINT
9213                         :
9214                         THUNK_TARGET(entry),
9215                         [ss]"i"(__KERNEL_DS),
9216                         [cs]"i"(__KERNEL_CS)
9217                         );
9218         }
9219 }
9220 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
9221
9222 static bool vmx_has_high_real_mode_segbase(void)
9223 {
9224         return enable_unrestricted_guest || emulate_invalid_guest_state;
9225 }
9226
9227 static bool vmx_mpx_supported(void)
9228 {
9229         return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
9230                 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
9231 }
9232
9233 static bool vmx_xsaves_supported(void)
9234 {
9235         return vmcs_config.cpu_based_2nd_exec_ctrl &
9236                 SECONDARY_EXEC_XSAVES;
9237 }
9238
9239 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
9240 {
9241         u32 exit_intr_info;
9242         bool unblock_nmi;
9243         u8 vector;
9244         bool idtv_info_valid;
9245
9246         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9247
9248         if (cpu_has_virtual_nmis()) {
9249                 if (vmx->loaded_vmcs->nmi_known_unmasked)
9250                         return;
9251                 /*
9252                  * Can't use vmx->exit_intr_info since we're not sure what
9253                  * the exit reason is.
9254                  */
9255                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9256                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
9257                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9258                 /*
9259                  * SDM 3: 27.7.1.2 (September 2008)
9260                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
9261                  * a guest IRET fault.
9262                  * SDM 3: 23.2.2 (September 2008)
9263                  * Bit 12 is undefined in any of the following cases:
9264                  *  If the VM exit sets the valid bit in the IDT-vectoring
9265                  *   information field.
9266                  *  If the VM exit is due to a double fault.
9267                  */
9268                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
9269                     vector != DF_VECTOR && !idtv_info_valid)
9270                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
9271                                       GUEST_INTR_STATE_NMI);
9272                 else
9273                         vmx->loaded_vmcs->nmi_known_unmasked =
9274                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
9275                                   & GUEST_INTR_STATE_NMI);
9276         } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
9277                 vmx->loaded_vmcs->vnmi_blocked_time +=
9278                         ktime_to_ns(ktime_sub(ktime_get(),
9279                                               vmx->loaded_vmcs->entry_time));
9280 }
9281
9282 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
9283                                       u32 idt_vectoring_info,
9284                                       int instr_len_field,
9285                                       int error_code_field)
9286 {
9287         u8 vector;
9288         int type;
9289         bool idtv_info_valid;
9290
9291         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9292
9293         vcpu->arch.nmi_injected = false;
9294         kvm_clear_exception_queue(vcpu);
9295         kvm_clear_interrupt_queue(vcpu);
9296
9297         if (!idtv_info_valid)
9298                 return;
9299
9300         kvm_make_request(KVM_REQ_EVENT, vcpu);
9301
9302         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
9303         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
9304
9305         switch (type) {
9306         case INTR_TYPE_NMI_INTR:
9307                 vcpu->arch.nmi_injected = true;
9308                 /*
9309                  * SDM 3: 27.7.1.2 (September 2008)
9310                  * Clear bit "block by NMI" before VM entry if a NMI
9311                  * delivery faulted.
9312                  */
9313                 vmx_set_nmi_mask(vcpu, false);
9314                 break;
9315         case INTR_TYPE_SOFT_EXCEPTION:
9316                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9317                 /* fall through */
9318         case INTR_TYPE_HARD_EXCEPTION:
9319                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
9320                         u32 err = vmcs_read32(error_code_field);
9321                         kvm_requeue_exception_e(vcpu, vector, err);
9322                 } else
9323                         kvm_requeue_exception(vcpu, vector);
9324                 break;
9325         case INTR_TYPE_SOFT_INTR:
9326                 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9327                 /* fall through */
9328         case INTR_TYPE_EXT_INTR:
9329                 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
9330                 break;
9331         default:
9332                 break;
9333         }
9334 }
9335
9336 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
9337 {
9338         __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
9339                                   VM_EXIT_INSTRUCTION_LEN,
9340                                   IDT_VECTORING_ERROR_CODE);
9341 }
9342
9343 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
9344 {
9345         __vmx_complete_interrupts(vcpu,
9346                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
9347                                   VM_ENTRY_INSTRUCTION_LEN,
9348                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
9349
9350         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9351 }
9352
9353 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
9354 {
9355         int i, nr_msrs;
9356         struct perf_guest_switch_msr *msrs;
9357
9358         msrs = perf_guest_get_msrs(&nr_msrs);
9359
9360         if (!msrs)
9361                 return;
9362
9363         for (i = 0; i < nr_msrs; i++)
9364                 if (msrs[i].host == msrs[i].guest)
9365                         clear_atomic_switch_msr(vmx, msrs[i].msr);
9366                 else
9367                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
9368                                         msrs[i].host);
9369 }
9370
9371 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
9372 {
9373         struct vcpu_vmx *vmx = to_vmx(vcpu);
9374         u64 tscl;
9375         u32 delta_tsc;
9376
9377         if (vmx->hv_deadline_tsc == -1)
9378                 return;
9379
9380         tscl = rdtsc();
9381         if (vmx->hv_deadline_tsc > tscl)
9382                 /* sure to be 32 bit only because checked on set_hv_timer */
9383                 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
9384                         cpu_preemption_timer_multi);
9385         else
9386                 delta_tsc = 0;
9387
9388         vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
9389 }
9390
9391 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
9392 {
9393         struct vcpu_vmx *vmx = to_vmx(vcpu);
9394         unsigned long debugctlmsr, cr3, cr4;
9395
9396         /* Record the guest's net vcpu time for enforced NMI injections. */
9397         if (unlikely(!cpu_has_virtual_nmis() &&
9398                      vmx->loaded_vmcs->soft_vnmi_blocked))
9399                 vmx->loaded_vmcs->entry_time = ktime_get();
9400
9401         /* Don't enter VMX if guest state is invalid, let the exit handler
9402            start emulation until we arrive back to a valid state */
9403         if (vmx->emulation_required)
9404                 return;
9405
9406         if (vmx->ple_window_dirty) {
9407                 vmx->ple_window_dirty = false;
9408                 vmcs_write32(PLE_WINDOW, vmx->ple_window);
9409         }
9410
9411         if (vmx->nested.sync_shadow_vmcs) {
9412                 copy_vmcs12_to_shadow(vmx);
9413                 vmx->nested.sync_shadow_vmcs = false;
9414         }
9415
9416         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
9417                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
9418         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
9419                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
9420
9421         cr3 = __get_current_cr3_fast();
9422         if (unlikely(cr3 != vmx->loaded_vmcs->vmcs_host_cr3)) {
9423                 vmcs_writel(HOST_CR3, cr3);
9424                 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
9425         }
9426
9427         cr4 = cr4_read_shadow();
9428         if (unlikely(cr4 != vmx->loaded_vmcs->vmcs_host_cr4)) {
9429                 vmcs_writel(HOST_CR4, cr4);
9430                 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
9431         }
9432
9433         /* When single-stepping over STI and MOV SS, we must clear the
9434          * corresponding interruptibility bits in the guest state. Otherwise
9435          * vmentry fails as it then expects bit 14 (BS) in pending debug
9436          * exceptions being set, but that's not correct for the guest debugging
9437          * case. */
9438         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9439                 vmx_set_interrupt_shadow(vcpu, 0);
9440
9441         if (static_cpu_has(X86_FEATURE_PKU) &&
9442             kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
9443             vcpu->arch.pkru != vmx->host_pkru)
9444                 __write_pkru(vcpu->arch.pkru);
9445
9446         atomic_switch_perf_msrs(vmx);
9447         debugctlmsr = get_debugctlmsr();
9448
9449         vmx_arm_hv_timer(vcpu);
9450
9451         /*
9452          * If this vCPU has touched SPEC_CTRL, restore the guest's value if
9453          * it's non-zero. Since vmentry is serialising on affected CPUs, there
9454          * is no need to worry about the conditional branch over the wrmsr
9455          * being speculatively taken.
9456          */
9457         x86_spec_ctrl_set_guest(vmx->spec_ctrl);
9458
9459         vmx->__launched = vmx->loaded_vmcs->launched;
9460         asm(
9461                 /* Store host registers */
9462                 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
9463                 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
9464                 "push %%" _ASM_CX " \n\t"
9465                 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9466                 "je 1f \n\t"
9467                 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9468                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
9469                 "1: \n\t"
9470                 /* Reload cr2 if changed */
9471                 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9472                 "mov %%cr2, %%" _ASM_DX " \n\t"
9473                 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9474                 "je 2f \n\t"
9475                 "mov %%" _ASM_AX", %%cr2 \n\t"
9476                 "2: \n\t"
9477                 /* Check if vmlaunch of vmresume is needed */
9478                 "cmpl $0, %c[launched](%0) \n\t"
9479                 /* Load guest registers.  Don't clobber flags. */
9480                 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9481                 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9482                 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9483                 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9484                 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9485                 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9486 #ifdef CONFIG_X86_64
9487                 "mov %c[r8](%0),  %%r8  \n\t"
9488                 "mov %c[r9](%0),  %%r9  \n\t"
9489                 "mov %c[r10](%0), %%r10 \n\t"
9490                 "mov %c[r11](%0), %%r11 \n\t"
9491                 "mov %c[r12](%0), %%r12 \n\t"
9492                 "mov %c[r13](%0), %%r13 \n\t"
9493                 "mov %c[r14](%0), %%r14 \n\t"
9494                 "mov %c[r15](%0), %%r15 \n\t"
9495 #endif
9496                 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9497
9498                 /* Enter guest mode */
9499                 "jne 1f \n\t"
9500                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9501                 "jmp 2f \n\t"
9502                 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9503                 "2: "
9504                 /* Save guest registers, load host registers, keep flags */
9505                 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9506                 "pop %0 \n\t"
9507                 "setbe %c[fail](%0)\n\t"
9508                 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9509                 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9510                 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9511                 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9512                 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9513                 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9514                 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9515 #ifdef CONFIG_X86_64
9516                 "mov %%r8,  %c[r8](%0) \n\t"
9517                 "mov %%r9,  %c[r9](%0) \n\t"
9518                 "mov %%r10, %c[r10](%0) \n\t"
9519                 "mov %%r11, %c[r11](%0) \n\t"
9520                 "mov %%r12, %c[r12](%0) \n\t"
9521                 "mov %%r13, %c[r13](%0) \n\t"
9522                 "mov %%r14, %c[r14](%0) \n\t"
9523                 "mov %%r15, %c[r15](%0) \n\t"
9524                 "xor %%r8d,  %%r8d \n\t"
9525                 "xor %%r9d,  %%r9d \n\t"
9526                 "xor %%r10d, %%r10d \n\t"
9527                 "xor %%r11d, %%r11d \n\t"
9528                 "xor %%r12d, %%r12d \n\t"
9529                 "xor %%r13d, %%r13d \n\t"
9530                 "xor %%r14d, %%r14d \n\t"
9531                 "xor %%r15d, %%r15d \n\t"
9532 #endif
9533                 "mov %%cr2, %%" _ASM_AX "   \n\t"
9534                 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9535
9536                 "xor %%eax, %%eax \n\t"
9537                 "xor %%ebx, %%ebx \n\t"
9538                 "xor %%esi, %%esi \n\t"
9539                 "xor %%edi, %%edi \n\t"
9540                 "pop  %%" _ASM_BP "; pop  %%" _ASM_DX " \n\t"
9541                 ".pushsection .rodata \n\t"
9542                 ".global vmx_return \n\t"
9543                 "vmx_return: " _ASM_PTR " 2b \n\t"
9544                 ".popsection"
9545               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9546                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9547                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9548                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9549                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9550                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9551                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9552                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9553                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9554                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9555                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9556 #ifdef CONFIG_X86_64
9557                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9558                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9559                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9560                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9561                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9562                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9563                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9564                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9565 #endif
9566                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9567                 [wordsize]"i"(sizeof(ulong))
9568               : "cc", "memory"
9569 #ifdef CONFIG_X86_64
9570                 , "rax", "rbx", "rdi", "rsi"
9571                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9572 #else
9573                 , "eax", "ebx", "edi", "esi"
9574 #endif
9575               );
9576
9577         /*
9578          * We do not use IBRS in the kernel. If this vCPU has used the
9579          * SPEC_CTRL MSR it may have left it on; save the value and
9580          * turn it off. This is much more efficient than blindly adding
9581          * it to the atomic save/restore list. Especially as the former
9582          * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
9583          *
9584          * For non-nested case:
9585          * If the L01 MSR bitmap does not intercept the MSR, then we need to
9586          * save it.
9587          *
9588          * For nested case:
9589          * If the L02 MSR bitmap does not intercept the MSR, then we need to
9590          * save it.
9591          */
9592         if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
9593                 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
9594
9595         x86_spec_ctrl_restore_host(vmx->spec_ctrl);
9596
9597         /* Eliminate branch target predictions from guest mode */
9598         vmexit_fill_RSB();
9599
9600         /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9601         if (debugctlmsr)
9602                 update_debugctlmsr(debugctlmsr);
9603
9604 #ifndef CONFIG_X86_64
9605         /*
9606          * The sysexit path does not restore ds/es, so we must set them to
9607          * a reasonable value ourselves.
9608          *
9609          * We can't defer this to vmx_load_host_state() since that function
9610          * may be executed in interrupt context, which saves and restore segments
9611          * around it, nullifying its effect.
9612          */
9613         loadsegment(ds, __USER_DS);
9614         loadsegment(es, __USER_DS);
9615 #endif
9616
9617         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9618                                   | (1 << VCPU_EXREG_RFLAGS)
9619                                   | (1 << VCPU_EXREG_PDPTR)
9620                                   | (1 << VCPU_EXREG_SEGMENTS)
9621                                   | (1 << VCPU_EXREG_CR3));
9622         vcpu->arch.regs_dirty = 0;
9623
9624         /*
9625          * eager fpu is enabled if PKEY is supported and CR4 is switched
9626          * back on host, so it is safe to read guest PKRU from current
9627          * XSAVE.
9628          */
9629         if (static_cpu_has(X86_FEATURE_PKU) &&
9630             kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
9631                 vcpu->arch.pkru = __read_pkru();
9632                 if (vcpu->arch.pkru != vmx->host_pkru)
9633                         __write_pkru(vmx->host_pkru);
9634         }
9635
9636         /*
9637          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9638          * we did not inject a still-pending event to L1 now because of
9639          * nested_run_pending, we need to re-enable this bit.
9640          */
9641         if (vmx->nested.nested_run_pending)
9642                 kvm_make_request(KVM_REQ_EVENT, vcpu);
9643
9644         vmx->nested.nested_run_pending = 0;
9645         vmx->idt_vectoring_info = 0;
9646
9647         vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
9648         if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9649                 return;
9650
9651         vmx->loaded_vmcs->launched = 1;
9652         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9653
9654         vmx_complete_atomic_exit(vmx);
9655         vmx_recover_nmi_blocking(vmx);
9656         vmx_complete_interrupts(vmx);
9657 }
9658 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
9659
9660 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9661 {
9662         struct vcpu_vmx *vmx = to_vmx(vcpu);
9663         int cpu;
9664
9665         if (vmx->loaded_vmcs == vmcs)
9666                 return;
9667
9668         cpu = get_cpu();
9669         vmx->loaded_vmcs = vmcs;
9670         vmx_vcpu_put(vcpu);
9671         vmx_vcpu_load(vcpu, cpu);
9672         vcpu->cpu = cpu;
9673         put_cpu();
9674 }
9675
9676 /*
9677  * Ensure that the current vmcs of the logical processor is the
9678  * vmcs01 of the vcpu before calling free_nested().
9679  */
9680 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9681 {
9682        struct vcpu_vmx *vmx = to_vmx(vcpu);
9683        int r;
9684
9685        r = vcpu_load(vcpu);
9686        BUG_ON(r);
9687        vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9688        free_nested(vmx);
9689        vcpu_put(vcpu);
9690 }
9691
9692 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9693 {
9694         struct vcpu_vmx *vmx = to_vmx(vcpu);
9695
9696         if (enable_pml)
9697                 vmx_destroy_pml_buffer(vmx);
9698         free_vpid(vmx->vpid);
9699         leave_guest_mode(vcpu);
9700         vmx_free_vcpu_nested(vcpu);
9701         free_loaded_vmcs(vmx->loaded_vmcs);
9702         kfree(vmx->guest_msrs);
9703         kvm_vcpu_uninit(vcpu);
9704         kmem_cache_free(kvm_vcpu_cache, vmx);
9705 }
9706
9707 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9708 {
9709         int err;
9710         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9711         unsigned long *msr_bitmap;
9712         int cpu;
9713
9714         if (!vmx)
9715                 return ERR_PTR(-ENOMEM);
9716
9717         vmx->vpid = allocate_vpid();
9718
9719         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9720         if (err)
9721                 goto free_vcpu;
9722
9723         err = -ENOMEM;
9724
9725         /*
9726          * If PML is turned on, failure on enabling PML just results in failure
9727          * of creating the vcpu, therefore we can simplify PML logic (by
9728          * avoiding dealing with cases, such as enabling PML partially on vcpus
9729          * for the guest, etc.
9730          */
9731         if (enable_pml) {
9732                 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9733                 if (!vmx->pml_pg)
9734                         goto uninit_vcpu;
9735         }
9736
9737         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9738         BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9739                      > PAGE_SIZE);
9740
9741         if (!vmx->guest_msrs)
9742                 goto free_pml;
9743
9744         err = alloc_loaded_vmcs(&vmx->vmcs01);
9745         if (err < 0)
9746                 goto free_msrs;
9747
9748         msr_bitmap = vmx->vmcs01.msr_bitmap;
9749         vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
9750         vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
9751         vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
9752         vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
9753         vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
9754         vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
9755         vmx->msr_bitmap_mode = 0;
9756
9757         vmx->loaded_vmcs = &vmx->vmcs01;
9758         cpu = get_cpu();
9759         vmx_vcpu_load(&vmx->vcpu, cpu);
9760         vmx->vcpu.cpu = cpu;
9761         err = vmx_vcpu_setup(vmx);
9762         vmx_vcpu_put(&vmx->vcpu);
9763         put_cpu();
9764         if (err)
9765                 goto free_vmcs;
9766         if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9767                 err = alloc_apic_access_page(kvm);
9768                 if (err)
9769                         goto free_vmcs;
9770         }
9771
9772         if (enable_ept) {
9773                 if (!kvm->arch.ept_identity_map_addr)
9774                         kvm->arch.ept_identity_map_addr =
9775                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9776                 err = init_rmode_identity_map(kvm);
9777                 if (err)
9778                         goto free_vmcs;
9779         }
9780
9781         if (nested) {
9782                 nested_vmx_setup_ctls_msrs(vmx);
9783                 vmx->nested.vpid02 = allocate_vpid();
9784         }
9785
9786         vmx->nested.posted_intr_nv = -1;
9787         vmx->nested.current_vmptr = -1ull;
9788
9789         vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9790
9791         /*
9792          * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
9793          * or POSTED_INTR_WAKEUP_VECTOR.
9794          */
9795         vmx->pi_desc.nv = POSTED_INTR_VECTOR;
9796         vmx->pi_desc.sn = 1;
9797
9798         return &vmx->vcpu;
9799
9800 free_vmcs:
9801         free_vpid(vmx->nested.vpid02);
9802         free_loaded_vmcs(vmx->loaded_vmcs);
9803 free_msrs:
9804         kfree(vmx->guest_msrs);
9805 free_pml:
9806         vmx_destroy_pml_buffer(vmx);
9807 uninit_vcpu:
9808         kvm_vcpu_uninit(&vmx->vcpu);
9809 free_vcpu:
9810         free_vpid(vmx->vpid);
9811         kmem_cache_free(kvm_vcpu_cache, vmx);
9812         return ERR_PTR(err);
9813 }
9814
9815 static void __init vmx_check_processor_compat(void *rtn)
9816 {
9817         struct vmcs_config vmcs_conf;
9818
9819         *(int *)rtn = 0;
9820         if (setup_vmcs_config(&vmcs_conf) < 0)
9821                 *(int *)rtn = -EIO;
9822         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9823                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9824                                 smp_processor_id());
9825                 *(int *)rtn = -EIO;
9826         }
9827 }
9828
9829 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9830 {
9831         u8 cache;
9832         u64 ipat = 0;
9833
9834         /* For VT-d and EPT combination
9835          * 1. MMIO: always map as UC
9836          * 2. EPT with VT-d:
9837          *   a. VT-d without snooping control feature: can't guarantee the
9838          *      result, try to trust guest.
9839          *   b. VT-d with snooping control feature: snooping control feature of
9840          *      VT-d engine can guarantee the cache correctness. Just set it
9841          *      to WB to keep consistent with host. So the same as item 3.
9842          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9843          *    consistent with host MTRR
9844          */
9845         if (is_mmio) {
9846                 cache = MTRR_TYPE_UNCACHABLE;
9847                 goto exit;
9848         }
9849
9850         if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9851                 ipat = VMX_EPT_IPAT_BIT;
9852                 cache = MTRR_TYPE_WRBACK;
9853                 goto exit;
9854         }
9855
9856         if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9857                 ipat = VMX_EPT_IPAT_BIT;
9858                 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9859                         cache = MTRR_TYPE_WRBACK;
9860                 else
9861                         cache = MTRR_TYPE_UNCACHABLE;
9862                 goto exit;
9863         }
9864
9865         cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9866
9867 exit:
9868         return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9869 }
9870
9871 static int vmx_get_lpage_level(void)
9872 {
9873         if (enable_ept && !cpu_has_vmx_ept_1g_page())
9874                 return PT_DIRECTORY_LEVEL;
9875         else
9876                 /* For shadow and EPT supported 1GB page */
9877                 return PT_PDPE_LEVEL;
9878 }
9879
9880 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9881 {
9882         /*
9883          * These bits in the secondary execution controls field
9884          * are dynamic, the others are mostly based on the hypervisor
9885          * architecture and the guest's CPUID.  Do not touch the
9886          * dynamic bits.
9887          */
9888         u32 mask =
9889                 SECONDARY_EXEC_SHADOW_VMCS |
9890                 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9891                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9892
9893         u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9894
9895         vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9896                      (new_ctl & ~mask) | (cur_ctl & mask));
9897 }
9898
9899 /*
9900  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9901  * (indicating "allowed-1") if they are supported in the guest's CPUID.
9902  */
9903 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9904 {
9905         struct vcpu_vmx *vmx = to_vmx(vcpu);
9906         struct kvm_cpuid_entry2 *entry;
9907
9908         vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9909         vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9910
9911 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {            \
9912         if (entry && (entry->_reg & (_cpuid_mask)))                     \
9913                 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask);       \
9914 } while (0)
9915
9916         entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9917         cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
9918         cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
9919         cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
9920         cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
9921         cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
9922         cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
9923         cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
9924         cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
9925         cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
9926         cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9927         cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
9928         cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
9929         cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
9930         cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
9931
9932         entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9933         cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
9934         cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
9935         cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
9936         cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
9937         /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9938         cr4_fixed1_update(bit(11),            ecx, bit(2));
9939
9940 #undef cr4_fixed1_update
9941 }
9942
9943 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9944 {
9945         struct vcpu_vmx *vmx = to_vmx(vcpu);
9946
9947         if (cpu_has_secondary_exec_ctrls()) {
9948                 vmx_compute_secondary_exec_control(vmx);
9949                 vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
9950         }
9951
9952         if (nested_vmx_allowed(vcpu))
9953                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9954                         FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9955         else
9956                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9957                         ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9958
9959         if (nested_vmx_allowed(vcpu))
9960                 nested_vmx_cr_fixed1_bits_update(vcpu);
9961 }
9962
9963 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9964 {
9965         if (func == 1 && nested)
9966                 entry->ecx |= bit(X86_FEATURE_VMX);
9967 }
9968
9969 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9970                 struct x86_exception *fault)
9971 {
9972         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9973         struct vcpu_vmx *vmx = to_vmx(vcpu);
9974         u32 exit_reason;
9975         unsigned long exit_qualification = vcpu->arch.exit_qualification;
9976
9977         if (vmx->nested.pml_full) {
9978                 exit_reason = EXIT_REASON_PML_FULL;
9979                 vmx->nested.pml_full = false;
9980                 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
9981         } else if (fault->error_code & PFERR_RSVD_MASK)
9982                 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9983         else
9984                 exit_reason = EXIT_REASON_EPT_VIOLATION;
9985
9986         nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
9987         vmcs12->guest_physical_address = fault->address;
9988 }
9989
9990 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
9991 {
9992         return nested_ept_get_cr3(vcpu) & VMX_EPTP_AD_ENABLE_BIT;
9993 }
9994
9995 /* Callbacks for nested_ept_init_mmu_context: */
9996
9997 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9998 {
9999         /* return the page table to be shadowed - in our case, EPT12 */
10000         return get_vmcs12(vcpu)->ept_pointer;
10001 }
10002
10003 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
10004 {
10005         WARN_ON(mmu_is_nested(vcpu));
10006         if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
10007                 return 1;
10008
10009         kvm_mmu_unload(vcpu);
10010         kvm_init_shadow_ept_mmu(vcpu,
10011                         to_vmx(vcpu)->nested.nested_vmx_ept_caps &
10012                         VMX_EPT_EXECUTE_ONLY_BIT,
10013                         nested_ept_ad_enabled(vcpu));
10014         vcpu->arch.mmu.set_cr3           = vmx_set_cr3;
10015         vcpu->arch.mmu.get_cr3           = nested_ept_get_cr3;
10016         vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
10017
10018         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
10019         return 0;
10020 }
10021
10022 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
10023 {
10024         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
10025 }
10026
10027 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
10028                                             u16 error_code)
10029 {
10030         bool inequality, bit;
10031
10032         bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
10033         inequality =
10034                 (error_code & vmcs12->page_fault_error_code_mask) !=
10035                  vmcs12->page_fault_error_code_match;
10036         return inequality ^ bit;
10037 }
10038
10039 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
10040                 struct x86_exception *fault)
10041 {
10042         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10043
10044         WARN_ON(!is_guest_mode(vcpu));
10045
10046         if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
10047                 !to_vmx(vcpu)->nested.nested_run_pending) {
10048                 vmcs12->vm_exit_intr_error_code = fault->error_code;
10049                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10050                                   PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
10051                                   INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
10052                                   fault->address);
10053         } else {
10054                 kvm_inject_page_fault(vcpu, fault);
10055         }
10056 }
10057
10058 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
10059                                                struct vmcs12 *vmcs12);
10060
10061 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
10062                                         struct vmcs12 *vmcs12)
10063 {
10064         struct vcpu_vmx *vmx = to_vmx(vcpu);
10065         struct page *page;
10066         u64 hpa;
10067
10068         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10069                 /*
10070                  * Translate L1 physical address to host physical
10071                  * address for vmcs02. Keep the page pinned, so this
10072                  * physical address remains valid. We keep a reference
10073                  * to it so we can release it later.
10074                  */
10075                 if (vmx->nested.apic_access_page) { /* shouldn't happen */
10076                         kvm_release_page_dirty(vmx->nested.apic_access_page);
10077                         vmx->nested.apic_access_page = NULL;
10078                 }
10079                 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
10080                 /*
10081                  * If translation failed, no matter: This feature asks
10082                  * to exit when accessing the given address, and if it
10083                  * can never be accessed, this feature won't do
10084                  * anything anyway.
10085                  */
10086                 if (!is_error_page(page)) {
10087                         vmx->nested.apic_access_page = page;
10088                         hpa = page_to_phys(vmx->nested.apic_access_page);
10089                         vmcs_write64(APIC_ACCESS_ADDR, hpa);
10090                 } else {
10091                         vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
10092                                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
10093                 }
10094         } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
10095                    cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
10096                 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
10097                               SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
10098                 kvm_vcpu_reload_apic_access_page(vcpu);
10099         }
10100
10101         if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
10102                 if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
10103                         kvm_release_page_dirty(vmx->nested.virtual_apic_page);
10104                         vmx->nested.virtual_apic_page = NULL;
10105                 }
10106                 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
10107
10108                 /*
10109                  * If translation failed, VM entry will fail because
10110                  * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
10111                  * Failing the vm entry is _not_ what the processor
10112                  * does but it's basically the only possibility we
10113                  * have.  We could still enter the guest if CR8 load
10114                  * exits are enabled, CR8 store exits are enabled, and
10115                  * virtualize APIC access is disabled; in this case
10116                  * the processor would never use the TPR shadow and we
10117                  * could simply clear the bit from the execution
10118                  * control.  But such a configuration is useless, so
10119                  * let's keep the code simple.
10120                  */
10121                 if (!is_error_page(page)) {
10122                         vmx->nested.virtual_apic_page = page;
10123                         hpa = page_to_phys(vmx->nested.virtual_apic_page);
10124                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
10125                 }
10126         }
10127
10128         if (nested_cpu_has_posted_intr(vmcs12)) {
10129                 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
10130                         kunmap(vmx->nested.pi_desc_page);
10131                         kvm_release_page_dirty(vmx->nested.pi_desc_page);
10132                         vmx->nested.pi_desc_page = NULL;
10133                 }
10134                 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
10135                 if (is_error_page(page))
10136                         return;
10137                 vmx->nested.pi_desc_page = page;
10138                 vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
10139                 vmx->nested.pi_desc =
10140                         (struct pi_desc *)((void *)vmx->nested.pi_desc +
10141                         (unsigned long)(vmcs12->posted_intr_desc_addr &
10142                         (PAGE_SIZE - 1)));
10143                 vmcs_write64(POSTED_INTR_DESC_ADDR,
10144                         page_to_phys(vmx->nested.pi_desc_page) +
10145                         (unsigned long)(vmcs12->posted_intr_desc_addr &
10146                         (PAGE_SIZE - 1)));
10147         }
10148         if (cpu_has_vmx_msr_bitmap() &&
10149             nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
10150             nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
10151                 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
10152                               CPU_BASED_USE_MSR_BITMAPS);
10153         else
10154                 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
10155                                 CPU_BASED_USE_MSR_BITMAPS);
10156 }
10157
10158 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
10159 {
10160         u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
10161         struct vcpu_vmx *vmx = to_vmx(vcpu);
10162
10163         if (vcpu->arch.virtual_tsc_khz == 0)
10164                 return;
10165
10166         /* Make sure short timeouts reliably trigger an immediate vmexit.
10167          * hrtimer_start does not guarantee this. */
10168         if (preemption_timeout <= 1) {
10169                 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
10170                 return;
10171         }
10172
10173         preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10174         preemption_timeout *= 1000000;
10175         do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
10176         hrtimer_start(&vmx->nested.preemption_timer,
10177                       ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
10178 }
10179
10180 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
10181                                                struct vmcs12 *vmcs12)
10182 {
10183         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
10184                 return 0;
10185
10186         if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
10187             !page_address_valid(vcpu, vmcs12->io_bitmap_b))
10188                 return -EINVAL;
10189
10190         return 0;
10191 }
10192
10193 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
10194                                                 struct vmcs12 *vmcs12)
10195 {
10196         if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
10197                 return 0;
10198
10199         if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
10200                 return -EINVAL;
10201
10202         return 0;
10203 }
10204
10205 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
10206                                                 struct vmcs12 *vmcs12)
10207 {
10208         if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10209                 return 0;
10210
10211         if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
10212                 return -EINVAL;
10213
10214         return 0;
10215 }
10216
10217 /*
10218  * Merge L0's and L1's MSR bitmap, return false to indicate that
10219  * we do not use the hardware.
10220  */
10221 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
10222                                                struct vmcs12 *vmcs12)
10223 {
10224         int msr;
10225         struct page *page;
10226         unsigned long *msr_bitmap_l1;
10227         unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
10228         /*
10229          * pred_cmd & spec_ctrl are trying to verify two things:
10230          *
10231          * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
10232          *    ensures that we do not accidentally generate an L02 MSR bitmap
10233          *    from the L12 MSR bitmap that is too permissive.
10234          * 2. That L1 or L2s have actually used the MSR. This avoids
10235          *    unnecessarily merging of the bitmap if the MSR is unused. This
10236          *    works properly because we only update the L01 MSR bitmap lazily.
10237          *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
10238          *    updated to reflect this when L1 (or its L2s) actually write to
10239          *    the MSR.
10240          */
10241         bool pred_cmd = !msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD);
10242         bool spec_ctrl = !msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL);
10243
10244         if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10245             !pred_cmd && !spec_ctrl)
10246                 return false;
10247
10248         page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
10249         if (is_error_page(page))
10250                 return false;
10251         msr_bitmap_l1 = (unsigned long *)kmap(page);
10252
10253         memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
10254
10255         if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
10256                 if (nested_cpu_has_apic_reg_virt(vmcs12))
10257                         for (msr = 0x800; msr <= 0x8ff; msr++)
10258                                 nested_vmx_disable_intercept_for_msr(
10259                                         msr_bitmap_l1, msr_bitmap_l0,
10260                                         msr, MSR_TYPE_R);
10261
10262                 nested_vmx_disable_intercept_for_msr(
10263                                 msr_bitmap_l1, msr_bitmap_l0,
10264                                 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
10265                                 MSR_TYPE_R | MSR_TYPE_W);
10266
10267                 if (nested_cpu_has_vid(vmcs12)) {
10268                         nested_vmx_disable_intercept_for_msr(
10269                                 msr_bitmap_l1, msr_bitmap_l0,
10270                                 APIC_BASE_MSR + (APIC_EOI >> 4),
10271                                 MSR_TYPE_W);
10272                         nested_vmx_disable_intercept_for_msr(
10273                                 msr_bitmap_l1, msr_bitmap_l0,
10274                                 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
10275                                 MSR_TYPE_W);
10276                 }
10277         }
10278
10279         if (spec_ctrl)
10280                 nested_vmx_disable_intercept_for_msr(
10281                                         msr_bitmap_l1, msr_bitmap_l0,
10282                                         MSR_IA32_SPEC_CTRL,
10283                                         MSR_TYPE_R | MSR_TYPE_W);
10284
10285         if (pred_cmd)
10286                 nested_vmx_disable_intercept_for_msr(
10287                                         msr_bitmap_l1, msr_bitmap_l0,
10288                                         MSR_IA32_PRED_CMD,
10289                                         MSR_TYPE_W);
10290
10291         kunmap(page);
10292         kvm_release_page_clean(page);
10293
10294         return true;
10295 }
10296
10297 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
10298                                            struct vmcs12 *vmcs12)
10299 {
10300         if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10301             !nested_cpu_has_apic_reg_virt(vmcs12) &&
10302             !nested_cpu_has_vid(vmcs12) &&
10303             !nested_cpu_has_posted_intr(vmcs12))
10304                 return 0;
10305
10306         /*
10307          * If virtualize x2apic mode is enabled,
10308          * virtualize apic access must be disabled.
10309          */
10310         if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10311             nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
10312                 return -EINVAL;
10313
10314         /*
10315          * If virtual interrupt delivery is enabled,
10316          * we must exit on external interrupts.
10317          */
10318         if (nested_cpu_has_vid(vmcs12) &&
10319            !nested_exit_on_intr(vcpu))
10320                 return -EINVAL;
10321
10322         /*
10323          * bits 15:8 should be zero in posted_intr_nv,
10324          * the descriptor address has been already checked
10325          * in nested_get_vmcs12_pages.
10326          */
10327         if (nested_cpu_has_posted_intr(vmcs12) &&
10328            (!nested_cpu_has_vid(vmcs12) ||
10329             !nested_exit_intr_ack_set(vcpu) ||
10330             vmcs12->posted_intr_nv & 0xff00))
10331                 return -EINVAL;
10332
10333         /* tpr shadow is needed by all apicv features. */
10334         if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10335                 return -EINVAL;
10336
10337         return 0;
10338 }
10339
10340 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
10341                                        unsigned long count_field,
10342                                        unsigned long addr_field)
10343 {
10344         int maxphyaddr;
10345         u64 count, addr;
10346
10347         if (vmcs12_read_any(vcpu, count_field, &count) ||
10348             vmcs12_read_any(vcpu, addr_field, &addr)) {
10349                 WARN_ON(1);
10350                 return -EINVAL;
10351         }
10352         if (count == 0)
10353                 return 0;
10354         maxphyaddr = cpuid_maxphyaddr(vcpu);
10355         if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
10356             (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
10357                 pr_debug_ratelimited(
10358                         "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
10359                         addr_field, maxphyaddr, count, addr);
10360                 return -EINVAL;
10361         }
10362         return 0;
10363 }
10364
10365 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
10366                                                 struct vmcs12 *vmcs12)
10367 {
10368         if (vmcs12->vm_exit_msr_load_count == 0 &&
10369             vmcs12->vm_exit_msr_store_count == 0 &&
10370             vmcs12->vm_entry_msr_load_count == 0)
10371                 return 0; /* Fast path */
10372         if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
10373                                         VM_EXIT_MSR_LOAD_ADDR) ||
10374             nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
10375                                         VM_EXIT_MSR_STORE_ADDR) ||
10376             nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
10377                                         VM_ENTRY_MSR_LOAD_ADDR))
10378                 return -EINVAL;
10379         return 0;
10380 }
10381
10382 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
10383                                          struct vmcs12 *vmcs12)
10384 {
10385         u64 address = vmcs12->pml_address;
10386         int maxphyaddr = cpuid_maxphyaddr(vcpu);
10387
10388         if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
10389                 if (!nested_cpu_has_ept(vmcs12) ||
10390                     !IS_ALIGNED(address, 4096)  ||
10391                     address >> maxphyaddr)
10392                         return -EINVAL;
10393         }
10394
10395         return 0;
10396 }
10397
10398 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
10399                                        struct vmx_msr_entry *e)
10400 {
10401         /* x2APIC MSR accesses are not allowed */
10402         if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
10403                 return -EINVAL;
10404         if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
10405             e->index == MSR_IA32_UCODE_REV)
10406                 return -EINVAL;
10407         if (e->reserved != 0)
10408                 return -EINVAL;
10409         return 0;
10410 }
10411
10412 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
10413                                      struct vmx_msr_entry *e)
10414 {
10415         if (e->index == MSR_FS_BASE ||
10416             e->index == MSR_GS_BASE ||
10417             e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
10418             nested_vmx_msr_check_common(vcpu, e))
10419                 return -EINVAL;
10420         return 0;
10421 }
10422
10423 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
10424                                       struct vmx_msr_entry *e)
10425 {
10426         if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
10427             nested_vmx_msr_check_common(vcpu, e))
10428                 return -EINVAL;
10429         return 0;
10430 }
10431
10432 /*
10433  * Load guest's/host's msr at nested entry/exit.
10434  * return 0 for success, entry index for failure.
10435  */
10436 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10437 {
10438         u32 i;
10439         struct vmx_msr_entry e;
10440         struct msr_data msr;
10441
10442         msr.host_initiated = false;
10443         for (i = 0; i < count; i++) {
10444                 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
10445                                         &e, sizeof(e))) {
10446                         pr_debug_ratelimited(
10447                                 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10448                                 __func__, i, gpa + i * sizeof(e));
10449                         goto fail;
10450                 }
10451                 if (nested_vmx_load_msr_check(vcpu, &e)) {
10452                         pr_debug_ratelimited(
10453                                 "%s check failed (%u, 0x%x, 0x%x)\n",
10454                                 __func__, i, e.index, e.reserved);
10455                         goto fail;
10456                 }
10457                 msr.index = e.index;
10458                 msr.data = e.value;
10459                 if (kvm_set_msr(vcpu, &msr)) {
10460                         pr_debug_ratelimited(
10461                                 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10462                                 __func__, i, e.index, e.value);
10463                         goto fail;
10464                 }
10465         }
10466         return 0;
10467 fail:
10468         return i + 1;
10469 }
10470
10471 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10472 {
10473         u32 i;
10474         struct vmx_msr_entry e;
10475
10476         for (i = 0; i < count; i++) {
10477                 struct msr_data msr_info;
10478                 if (kvm_vcpu_read_guest(vcpu,
10479                                         gpa + i * sizeof(e),
10480                                         &e, 2 * sizeof(u32))) {
10481                         pr_debug_ratelimited(
10482                                 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10483                                 __func__, i, gpa + i * sizeof(e));
10484                         return -EINVAL;
10485                 }
10486                 if (nested_vmx_store_msr_check(vcpu, &e)) {
10487                         pr_debug_ratelimited(
10488                                 "%s check failed (%u, 0x%x, 0x%x)\n",
10489                                 __func__, i, e.index, e.reserved);
10490                         return -EINVAL;
10491                 }
10492                 msr_info.host_initiated = false;
10493                 msr_info.index = e.index;
10494                 if (kvm_get_msr(vcpu, &msr_info)) {
10495                         pr_debug_ratelimited(
10496                                 "%s cannot read MSR (%u, 0x%x)\n",
10497                                 __func__, i, e.index);
10498                         return -EINVAL;
10499                 }
10500                 if (kvm_vcpu_write_guest(vcpu,
10501                                          gpa + i * sizeof(e) +
10502                                              offsetof(struct vmx_msr_entry, value),
10503                                          &msr_info.data, sizeof(msr_info.data))) {
10504                         pr_debug_ratelimited(
10505                                 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10506                                 __func__, i, e.index, msr_info.data);
10507                         return -EINVAL;
10508                 }
10509         }
10510         return 0;
10511 }
10512
10513 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
10514 {
10515         unsigned long invalid_mask;
10516
10517         invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
10518         return (val & invalid_mask) == 0;
10519 }
10520
10521 /*
10522  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
10523  * emulating VM entry into a guest with EPT enabled.
10524  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10525  * is assigned to entry_failure_code on failure.
10526  */
10527 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
10528                                u32 *entry_failure_code)
10529 {
10530         if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
10531                 if (!nested_cr3_valid(vcpu, cr3)) {
10532                         *entry_failure_code = ENTRY_FAIL_DEFAULT;
10533                         return 1;
10534                 }
10535
10536                 /*
10537                  * If PAE paging and EPT are both on, CR3 is not used by the CPU and
10538                  * must not be dereferenced.
10539                  */
10540                 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
10541                     !nested_ept) {
10542                         if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
10543                                 *entry_failure_code = ENTRY_FAIL_PDPTE;
10544                                 return 1;
10545                         }
10546                 }
10547
10548                 vcpu->arch.cr3 = cr3;
10549                 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
10550         }
10551
10552         kvm_mmu_reset_context(vcpu);
10553         return 0;
10554 }
10555
10556 /*
10557  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
10558  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
10559  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
10560  * guest in a way that will both be appropriate to L1's requests, and our
10561  * needs. In addition to modifying the active vmcs (which is vmcs02), this
10562  * function also has additional necessary side-effects, like setting various
10563  * vcpu->arch fields.
10564  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10565  * is assigned to entry_failure_code on failure.
10566  */
10567 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10568                           bool from_vmentry, u32 *entry_failure_code)
10569 {
10570         struct vcpu_vmx *vmx = to_vmx(vcpu);
10571         u32 exec_control, vmcs12_exec_ctrl;
10572
10573         vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10574         vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10575         vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10576         vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10577         vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10578         vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10579         vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10580         vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10581         vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10582         vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10583         vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10584         vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10585         vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10586         vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10587         vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10588         vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10589         vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10590         vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10591         vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10592         vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10593         vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10594         vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10595         vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10596         vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10597         vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10598         vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10599         vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10600         vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10601         vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10602         vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10603         vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10604         vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10605         vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10606         vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10607         vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10608         vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10609
10610         if (from_vmentry &&
10611             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10612                 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10613                 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10614         } else {
10615                 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10616                 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10617         }
10618         if (from_vmentry) {
10619                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10620                              vmcs12->vm_entry_intr_info_field);
10621                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10622                              vmcs12->vm_entry_exception_error_code);
10623                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10624                              vmcs12->vm_entry_instruction_len);
10625                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10626                              vmcs12->guest_interruptibility_info);
10627                 vmx->loaded_vmcs->nmi_known_unmasked =
10628                         !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
10629         } else {
10630                 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10631         }
10632         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10633         vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10634         vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10635                 vmcs12->guest_pending_dbg_exceptions);
10636         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10637         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10638
10639         if (nested_cpu_has_xsaves(vmcs12))
10640                 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10641         vmcs_write64(VMCS_LINK_POINTER, -1ull);
10642
10643         exec_control = vmcs12->pin_based_vm_exec_control;
10644
10645         /* Preemption timer setting is only taken from vmcs01.  */
10646         exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10647         exec_control |= vmcs_config.pin_based_exec_ctrl;
10648         if (vmx->hv_deadline_tsc == -1)
10649                 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10650
10651         /* Posted interrupts setting is only taken from vmcs12.  */
10652         if (nested_cpu_has_posted_intr(vmcs12)) {
10653                 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10654                 vmx->nested.pi_pending = false;
10655                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
10656         } else {
10657                 exec_control &= ~PIN_BASED_POSTED_INTR;
10658         }
10659
10660         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10661
10662         vmx->nested.preemption_timer_expired = false;
10663         if (nested_cpu_has_preemption_timer(vmcs12))
10664                 vmx_start_preemption_timer(vcpu);
10665
10666         /*
10667          * Whether page-faults are trapped is determined by a combination of
10668          * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10669          * If enable_ept, L0 doesn't care about page faults and we should
10670          * set all of these to L1's desires. However, if !enable_ept, L0 does
10671          * care about (at least some) page faults, and because it is not easy
10672          * (if at all possible?) to merge L0 and L1's desires, we simply ask
10673          * to exit on each and every L2 page fault. This is done by setting
10674          * MASK=MATCH=0 and (see below) EB.PF=1.
10675          * Note that below we don't need special code to set EB.PF beyond the
10676          * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10677          * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10678          * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10679          */
10680         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10681                 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10682         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10683                 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10684
10685         if (cpu_has_secondary_exec_ctrls()) {
10686                 exec_control = vmx->secondary_exec_control;
10687
10688                 /* Take the following fields only from vmcs12 */
10689                 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10690                                   SECONDARY_EXEC_ENABLE_INVPCID |
10691                                   SECONDARY_EXEC_RDTSCP |
10692                                   SECONDARY_EXEC_XSAVES |
10693                                   SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10694                                   SECONDARY_EXEC_APIC_REGISTER_VIRT |
10695                                   SECONDARY_EXEC_ENABLE_VMFUNC);
10696                 if (nested_cpu_has(vmcs12,
10697                                    CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
10698                         vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
10699                                 ~SECONDARY_EXEC_ENABLE_PML;
10700                         exec_control |= vmcs12_exec_ctrl;
10701                 }
10702
10703                 /* All VMFUNCs are currently emulated through L0 vmexits.  */
10704                 if (exec_control & SECONDARY_EXEC_ENABLE_VMFUNC)
10705                         vmcs_write64(VM_FUNCTION_CONTROL, 0);
10706
10707                 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10708                         vmcs_write64(EOI_EXIT_BITMAP0,
10709                                 vmcs12->eoi_exit_bitmap0);
10710                         vmcs_write64(EOI_EXIT_BITMAP1,
10711                                 vmcs12->eoi_exit_bitmap1);
10712                         vmcs_write64(EOI_EXIT_BITMAP2,
10713                                 vmcs12->eoi_exit_bitmap2);
10714                         vmcs_write64(EOI_EXIT_BITMAP3,
10715                                 vmcs12->eoi_exit_bitmap3);
10716                         vmcs_write16(GUEST_INTR_STATUS,
10717                                 vmcs12->guest_intr_status);
10718                 }
10719
10720                 /*
10721                  * Write an illegal value to APIC_ACCESS_ADDR. Later,
10722                  * nested_get_vmcs12_pages will either fix it up or
10723                  * remove the VM execution control.
10724                  */
10725                 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10726                         vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10727
10728                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10729         }
10730
10731
10732         /*
10733          * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10734          * Some constant fields are set here by vmx_set_constant_host_state().
10735          * Other fields are different per CPU, and will be set later when
10736          * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10737          */
10738         vmx_set_constant_host_state(vmx);
10739
10740         /*
10741          * Set the MSR load/store lists to match L0's settings.
10742          */
10743         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10744         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10745         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10746         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10747         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10748
10749         /*
10750          * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10751          * entry, but only if the current (host) sp changed from the value
10752          * we wrote last (vmx->host_rsp). This cache is no longer relevant
10753          * if we switch vmcs, and rather than hold a separate cache per vmcs,
10754          * here we just force the write to happen on entry.
10755          */
10756         vmx->host_rsp = 0;
10757
10758         exec_control = vmx_exec_control(vmx); /* L0's desires */
10759         exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10760         exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10761         exec_control &= ~CPU_BASED_TPR_SHADOW;
10762         exec_control |= vmcs12->cpu_based_vm_exec_control;
10763
10764         /*
10765          * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10766          * nested_get_vmcs12_pages can't fix it up, the illegal value
10767          * will result in a VM entry failure.
10768          */
10769         if (exec_control & CPU_BASED_TPR_SHADOW) {
10770                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10771                 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10772         } else {
10773 #ifdef CONFIG_X86_64
10774                 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
10775                                 CPU_BASED_CR8_STORE_EXITING;
10776 #endif
10777         }
10778
10779         /*
10780          * Merging of IO bitmap not currently supported.
10781          * Rather, exit every time.
10782          */
10783         exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10784         exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10785
10786         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10787
10788         /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10789          * bitwise-or of what L1 wants to trap for L2, and what we want to
10790          * trap. Note that CR0.TS also needs updating - we do this later.
10791          */
10792         update_exception_bitmap(vcpu);
10793         vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10794         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10795
10796         /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10797          * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10798          * bits are further modified by vmx_set_efer() below.
10799          */
10800         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10801
10802         /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10803          * emulated by vmx_set_efer(), below.
10804          */
10805         vm_entry_controls_init(vmx,
10806                 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10807                         ~VM_ENTRY_IA32E_MODE) |
10808                 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10809
10810         if (from_vmentry &&
10811             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10812                 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10813                 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10814         } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10815                 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10816         }
10817
10818         set_cr4_guest_host_mask(vmx);
10819
10820         if (from_vmentry &&
10821             vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10822                 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10823
10824         if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10825                 vmcs_write64(TSC_OFFSET,
10826                         vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10827         else
10828                 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10829         if (kvm_has_tsc_control)
10830                 decache_tsc_multiplier(vmx);
10831
10832         if (cpu_has_vmx_msr_bitmap())
10833                 vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
10834
10835         if (enable_vpid) {
10836                 /*
10837                  * There is no direct mapping between vpid02 and vpid12, the
10838                  * vpid02 is per-vCPU for L0 and reused while the value of
10839                  * vpid12 is changed w/ one invvpid during nested vmentry.
10840                  * The vpid12 is allocated by L1 for L2, so it will not
10841                  * influence global bitmap(for vpid01 and vpid02 allocation)
10842                  * even if spawn a lot of nested vCPUs.
10843                  */
10844                 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10845                         vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10846                         if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10847                                 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10848                                 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10849                         }
10850                 } else {
10851                         vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10852                         vmx_flush_tlb(vcpu);
10853                 }
10854
10855         }
10856
10857         if (enable_pml) {
10858                 /*
10859                  * Conceptually we want to copy the PML address and index from
10860                  * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10861                  * since we always flush the log on each vmexit, this happens
10862                  * to be equivalent to simply resetting the fields in vmcs02.
10863                  */
10864                 ASSERT(vmx->pml_pg);
10865                 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
10866                 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
10867         }
10868
10869         if (nested_cpu_has_ept(vmcs12)) {
10870                 if (nested_ept_init_mmu_context(vcpu)) {
10871                         *entry_failure_code = ENTRY_FAIL_DEFAULT;
10872                         return 1;
10873                 }
10874         } else if (nested_cpu_has2(vmcs12,
10875                                    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10876                 vmx_flush_tlb_ept_only(vcpu);
10877         }
10878
10879         /*
10880          * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10881          * bits which we consider mandatory enabled.
10882          * The CR0_READ_SHADOW is what L2 should have expected to read given
10883          * the specifications by L1; It's not enough to take
10884          * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10885          * have more bits than L1 expected.
10886          */
10887         vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10888         vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10889
10890         vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10891         vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10892
10893         if (from_vmentry &&
10894             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10895                 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10896         else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10897                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10898         else
10899                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10900         /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10901         vmx_set_efer(vcpu, vcpu->arch.efer);
10902
10903         /* Shadow page tables on either EPT or shadow page tables. */
10904         if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10905                                 entry_failure_code))
10906                 return 1;
10907
10908         if (!enable_ept)
10909                 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10910
10911         /*
10912          * L1 may access the L2's PDPTR, so save them to construct vmcs12
10913          */
10914         if (enable_ept) {
10915                 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10916                 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10917                 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10918                 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10919         }
10920
10921         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10922         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10923         return 0;
10924 }
10925
10926 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10927 {
10928         struct vcpu_vmx *vmx = to_vmx(vcpu);
10929
10930         if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10931             vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10932                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10933
10934         if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
10935                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10936
10937         if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10938                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10939
10940         if (nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12))
10941                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10942
10943         if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10944                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10945
10946         if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10947                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10948
10949         if (nested_vmx_check_pml_controls(vcpu, vmcs12))
10950                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10951
10952         if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10953                                 vmx->nested.nested_vmx_procbased_ctls_low,
10954                                 vmx->nested.nested_vmx_procbased_ctls_high) ||
10955             (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
10956              !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10957                                  vmx->nested.nested_vmx_secondary_ctls_low,
10958                                  vmx->nested.nested_vmx_secondary_ctls_high)) ||
10959             !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10960                                 vmx->nested.nested_vmx_pinbased_ctls_low,
10961                                 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10962             !vmx_control_verify(vmcs12->vm_exit_controls,
10963                                 vmx->nested.nested_vmx_exit_ctls_low,
10964                                 vmx->nested.nested_vmx_exit_ctls_high) ||
10965             !vmx_control_verify(vmcs12->vm_entry_controls,
10966                                 vmx->nested.nested_vmx_entry_ctls_low,
10967                                 vmx->nested.nested_vmx_entry_ctls_high))
10968                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10969
10970         if (nested_cpu_has_vmfunc(vmcs12)) {
10971                 if (vmcs12->vm_function_control &
10972                     ~vmx->nested.nested_vmx_vmfunc_controls)
10973                         return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10974
10975                 if (nested_cpu_has_eptp_switching(vmcs12)) {
10976                         if (!nested_cpu_has_ept(vmcs12) ||
10977                             !page_address_valid(vcpu, vmcs12->eptp_list_address))
10978                                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10979                 }
10980         }
10981
10982         if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
10983                 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10984
10985         if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10986             !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10987             !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10988                 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10989
10990         return 0;
10991 }
10992
10993 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10994                                   u32 *exit_qual)
10995 {
10996         bool ia32e;
10997
10998         *exit_qual = ENTRY_FAIL_DEFAULT;
10999
11000         if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
11001             !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
11002                 return 1;
11003
11004         if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
11005             vmcs12->vmcs_link_pointer != -1ull) {
11006                 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
11007                 return 1;
11008         }
11009
11010         /*
11011          * If the load IA32_EFER VM-entry control is 1, the following checks
11012          * are performed on the field for the IA32_EFER MSR:
11013          * - Bits reserved in the IA32_EFER MSR must be 0.
11014          * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
11015          *   the IA-32e mode guest VM-exit control. It must also be identical
11016          *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
11017          *   CR0.PG) is 1.
11018          */
11019         if (to_vmx(vcpu)->nested.nested_run_pending &&
11020             (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
11021                 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
11022                 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
11023                     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
11024                     ((vmcs12->guest_cr0 & X86_CR0_PG) &&
11025                      ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
11026                         return 1;
11027         }
11028
11029         /*
11030          * If the load IA32_EFER VM-exit control is 1, bits reserved in the
11031          * IA32_EFER MSR must be 0 in the field for that register. In addition,
11032          * the values of the LMA and LME bits in the field must each be that of
11033          * the host address-space size VM-exit control.
11034          */
11035         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
11036                 ia32e = (vmcs12->vm_exit_controls &
11037                          VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
11038                 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
11039                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
11040                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
11041                         return 1;
11042         }
11043
11044         return 0;
11045 }
11046
11047 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
11048 {
11049         struct vcpu_vmx *vmx = to_vmx(vcpu);
11050         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11051         u32 msr_entry_idx;
11052         u32 exit_qual;
11053
11054         enter_guest_mode(vcpu);
11055
11056         if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
11057                 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11058
11059         vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
11060         vmx_segment_cache_clear(vmx);
11061
11062         if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
11063                 leave_guest_mode(vcpu);
11064                 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11065                 nested_vmx_entry_failure(vcpu, vmcs12,
11066                                          EXIT_REASON_INVALID_STATE, exit_qual);
11067                 return 1;
11068         }
11069
11070         nested_get_vmcs12_pages(vcpu, vmcs12);
11071
11072         msr_entry_idx = nested_vmx_load_msr(vcpu,
11073                                             vmcs12->vm_entry_msr_load_addr,
11074                                             vmcs12->vm_entry_msr_load_count);
11075         if (msr_entry_idx) {
11076                 leave_guest_mode(vcpu);
11077                 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11078                 nested_vmx_entry_failure(vcpu, vmcs12,
11079                                 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
11080                 return 1;
11081         }
11082
11083         /*
11084          * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
11085          * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
11086          * returned as far as L1 is concerned. It will only return (and set
11087          * the success flag) when L2 exits (see nested_vmx_vmexit()).
11088          */
11089         return 0;
11090 }
11091
11092 /*
11093  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
11094  * for running an L2 nested guest.
11095  */
11096 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
11097 {
11098         struct vmcs12 *vmcs12;
11099         struct vcpu_vmx *vmx = to_vmx(vcpu);
11100         u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
11101         u32 exit_qual;
11102         int ret;
11103
11104         if (!nested_vmx_check_permission(vcpu))
11105                 return 1;
11106
11107         if (!nested_vmx_check_vmcs12(vcpu))
11108                 goto out;
11109
11110         vmcs12 = get_vmcs12(vcpu);
11111
11112         if (enable_shadow_vmcs)
11113                 copy_shadow_to_vmcs12(vmx);
11114
11115         /*
11116          * The nested entry process starts with enforcing various prerequisites
11117          * on vmcs12 as required by the Intel SDM, and act appropriately when
11118          * they fail: As the SDM explains, some conditions should cause the
11119          * instruction to fail, while others will cause the instruction to seem
11120          * to succeed, but return an EXIT_REASON_INVALID_STATE.
11121          * To speed up the normal (success) code path, we should avoid checking
11122          * for misconfigurations which will anyway be caught by the processor
11123          * when using the merged vmcs02.
11124          */
11125         if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
11126                 nested_vmx_failValid(vcpu,
11127                                      VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
11128                 goto out;
11129         }
11130
11131         if (vmcs12->launch_state == launch) {
11132                 nested_vmx_failValid(vcpu,
11133                         launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
11134                                : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
11135                 goto out;
11136         }
11137
11138         ret = check_vmentry_prereqs(vcpu, vmcs12);
11139         if (ret) {
11140                 nested_vmx_failValid(vcpu, ret);
11141                 goto out;
11142         }
11143
11144         /*
11145          * After this point, the trap flag no longer triggers a singlestep trap
11146          * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
11147          * This is not 100% correct; for performance reasons, we delegate most
11148          * of the checks on host state to the processor.  If those fail,
11149          * the singlestep trap is missed.
11150          */
11151         skip_emulated_instruction(vcpu);
11152
11153         ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
11154         if (ret) {
11155                 nested_vmx_entry_failure(vcpu, vmcs12,
11156                                          EXIT_REASON_INVALID_STATE, exit_qual);
11157                 return 1;
11158         }
11159
11160         /*
11161          * We're finally done with prerequisite checking, and can start with
11162          * the nested entry.
11163          */
11164
11165         ret = enter_vmx_non_root_mode(vcpu, true);
11166         if (ret)
11167                 return ret;
11168
11169         if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
11170                 return kvm_vcpu_halt(vcpu);
11171
11172         vmx->nested.nested_run_pending = 1;
11173
11174         return 1;
11175
11176 out:
11177         return kvm_skip_emulated_instruction(vcpu);
11178 }
11179
11180 /*
11181  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
11182  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
11183  * This function returns the new value we should put in vmcs12.guest_cr0.
11184  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
11185  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
11186  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
11187  *     didn't trap the bit, because if L1 did, so would L0).
11188  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
11189  *     been modified by L2, and L1 knows it. So just leave the old value of
11190  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
11191  *     isn't relevant, because if L0 traps this bit it can set it to anything.
11192  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
11193  *     changed these bits, and therefore they need to be updated, but L0
11194  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
11195  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
11196  */
11197 static inline unsigned long
11198 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11199 {
11200         return
11201         /*1*/   (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
11202         /*2*/   (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
11203         /*3*/   (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
11204                         vcpu->arch.cr0_guest_owned_bits));
11205 }
11206
11207 static inline unsigned long
11208 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11209 {
11210         return
11211         /*1*/   (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
11212         /*2*/   (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
11213         /*3*/   (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
11214                         vcpu->arch.cr4_guest_owned_bits));
11215 }
11216
11217 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
11218                                        struct vmcs12 *vmcs12)
11219 {
11220         u32 idt_vectoring;
11221         unsigned int nr;
11222
11223         if (vcpu->arch.exception.injected) {
11224                 nr = vcpu->arch.exception.nr;
11225                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11226
11227                 if (kvm_exception_is_soft(nr)) {
11228                         vmcs12->vm_exit_instruction_len =
11229                                 vcpu->arch.event_exit_inst_len;
11230                         idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
11231                 } else
11232                         idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
11233
11234                 if (vcpu->arch.exception.has_error_code) {
11235                         idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
11236                         vmcs12->idt_vectoring_error_code =
11237                                 vcpu->arch.exception.error_code;
11238                 }
11239
11240                 vmcs12->idt_vectoring_info_field = idt_vectoring;
11241         } else if (vcpu->arch.nmi_injected) {
11242                 vmcs12->idt_vectoring_info_field =
11243                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
11244         } else if (vcpu->arch.interrupt.pending) {
11245                 nr = vcpu->arch.interrupt.nr;
11246                 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11247
11248                 if (vcpu->arch.interrupt.soft) {
11249                         idt_vectoring |= INTR_TYPE_SOFT_INTR;
11250                         vmcs12->vm_entry_instruction_len =
11251                                 vcpu->arch.event_exit_inst_len;
11252                 } else
11253                         idt_vectoring |= INTR_TYPE_EXT_INTR;
11254
11255                 vmcs12->idt_vectoring_info_field = idt_vectoring;
11256         }
11257 }
11258
11259 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
11260 {
11261         struct vcpu_vmx *vmx = to_vmx(vcpu);
11262         unsigned long exit_qual;
11263         bool block_nested_events =
11264             vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
11265
11266         if (vcpu->arch.exception.pending &&
11267                 nested_vmx_check_exception(vcpu, &exit_qual)) {
11268                 if (block_nested_events)
11269                         return -EBUSY;
11270                 nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
11271                 return 0;
11272         }
11273
11274         if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
11275             vmx->nested.preemption_timer_expired) {
11276                 if (block_nested_events)
11277                         return -EBUSY;
11278                 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
11279                 return 0;
11280         }
11281
11282         if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
11283                 if (block_nested_events)
11284                         return -EBUSY;
11285                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
11286                                   NMI_VECTOR | INTR_TYPE_NMI_INTR |
11287                                   INTR_INFO_VALID_MASK, 0);
11288                 /*
11289                  * The NMI-triggered VM exit counts as injection:
11290                  * clear this one and block further NMIs.
11291                  */
11292                 vcpu->arch.nmi_pending = 0;
11293                 vmx_set_nmi_mask(vcpu, true);
11294                 return 0;
11295         }
11296
11297         if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
11298             nested_exit_on_intr(vcpu)) {
11299                 if (block_nested_events)
11300                         return -EBUSY;
11301                 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
11302                 return 0;
11303         }
11304
11305         vmx_complete_nested_posted_interrupt(vcpu);
11306         return 0;
11307 }
11308
11309 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
11310 {
11311         ktime_t remaining =
11312                 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
11313         u64 value;
11314
11315         if (ktime_to_ns(remaining) <= 0)
11316                 return 0;
11317
11318         value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
11319         do_div(value, 1000000);
11320         return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
11321 }
11322
11323 /*
11324  * Update the guest state fields of vmcs12 to reflect changes that
11325  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
11326  * VM-entry controls is also updated, since this is really a guest
11327  * state bit.)
11328  */
11329 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11330 {
11331         vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
11332         vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
11333
11334         vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
11335         vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
11336         vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
11337
11338         vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
11339         vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
11340         vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
11341         vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
11342         vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
11343         vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
11344         vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
11345         vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
11346         vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
11347         vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
11348         vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
11349         vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
11350         vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
11351         vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
11352         vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
11353         vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
11354         vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
11355         vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
11356         vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
11357         vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
11358         vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
11359         vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
11360         vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
11361         vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
11362         vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
11363         vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
11364         vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
11365         vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
11366         vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
11367         vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
11368         vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
11369         vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
11370         vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
11371         vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
11372         vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
11373         vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
11374
11375         vmcs12->guest_interruptibility_info =
11376                 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
11377         vmcs12->guest_pending_dbg_exceptions =
11378                 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
11379         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
11380                 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
11381         else
11382                 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
11383
11384         if (nested_cpu_has_preemption_timer(vmcs12)) {
11385                 if (vmcs12->vm_exit_controls &
11386                     VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
11387                         vmcs12->vmx_preemption_timer_value =
11388                                 vmx_get_preemption_timer_value(vcpu);
11389                 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
11390         }
11391
11392         /*
11393          * In some cases (usually, nested EPT), L2 is allowed to change its
11394          * own CR3 without exiting. If it has changed it, we must keep it.
11395          * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
11396          * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
11397          *
11398          * Additionally, restore L2's PDPTR to vmcs12.
11399          */
11400         if (enable_ept) {
11401                 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
11402                 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
11403                 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
11404                 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
11405                 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
11406         }
11407
11408         vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
11409
11410         if (nested_cpu_has_vid(vmcs12))
11411                 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
11412
11413         vmcs12->vm_entry_controls =
11414                 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
11415                 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
11416
11417         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
11418                 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
11419                 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11420         }
11421
11422         /* TODO: These cannot have changed unless we have MSR bitmaps and
11423          * the relevant bit asks not to trap the change */
11424         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
11425                 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
11426         if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
11427                 vmcs12->guest_ia32_efer = vcpu->arch.efer;
11428         vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
11429         vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
11430         vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
11431         if (kvm_mpx_supported())
11432                 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
11433 }
11434
11435 /*
11436  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
11437  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
11438  * and this function updates it to reflect the changes to the guest state while
11439  * L2 was running (and perhaps made some exits which were handled directly by L0
11440  * without going back to L1), and to reflect the exit reason.
11441  * Note that we do not have to copy here all VMCS fields, just those that
11442  * could have changed by the L2 guest or the exit - i.e., the guest-state and
11443  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
11444  * which already writes to vmcs12 directly.
11445  */
11446 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11447                            u32 exit_reason, u32 exit_intr_info,
11448                            unsigned long exit_qualification)
11449 {
11450         /* update guest state fields: */
11451         sync_vmcs12(vcpu, vmcs12);
11452
11453         /* update exit information fields: */
11454
11455         vmcs12->vm_exit_reason = exit_reason;
11456         vmcs12->exit_qualification = exit_qualification;
11457         vmcs12->vm_exit_intr_info = exit_intr_info;
11458
11459         vmcs12->idt_vectoring_info_field = 0;
11460         vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
11461         vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
11462
11463         if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
11464                 vmcs12->launch_state = 1;
11465
11466                 /* vm_entry_intr_info_field is cleared on exit. Emulate this
11467                  * instead of reading the real value. */
11468                 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
11469
11470                 /*
11471                  * Transfer the event that L0 or L1 may wanted to inject into
11472                  * L2 to IDT_VECTORING_INFO_FIELD.
11473                  */
11474                 vmcs12_save_pending_event(vcpu, vmcs12);
11475         }
11476
11477         /*
11478          * Drop what we picked up for L2 via vmx_complete_interrupts. It is
11479          * preserved above and would only end up incorrectly in L1.
11480          */
11481         vcpu->arch.nmi_injected = false;
11482         kvm_clear_exception_queue(vcpu);
11483         kvm_clear_interrupt_queue(vcpu);
11484 }
11485
11486 static void load_vmcs12_mmu_host_state(struct kvm_vcpu *vcpu,
11487                         struct vmcs12 *vmcs12)
11488 {
11489         u32 entry_failure_code;
11490
11491         nested_ept_uninit_mmu_context(vcpu);
11492
11493         /*
11494          * Only PDPTE load can fail as the value of cr3 was checked on entry and
11495          * couldn't have changed.
11496          */
11497         if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
11498                 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
11499
11500         if (!enable_ept)
11501                 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
11502 }
11503
11504 /*
11505  * A part of what we need to when the nested L2 guest exits and we want to
11506  * run its L1 parent, is to reset L1's guest state to the host state specified
11507  * in vmcs12.
11508  * This function is to be called not only on normal nested exit, but also on
11509  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
11510  * Failures During or After Loading Guest State").
11511  * This function should be called when the active VMCS is L1's (vmcs01).
11512  */
11513 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
11514                                    struct vmcs12 *vmcs12)
11515 {
11516         struct kvm_segment seg;
11517
11518         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
11519                 vcpu->arch.efer = vmcs12->host_ia32_efer;
11520         else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11521                 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
11522         else
11523                 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
11524         vmx_set_efer(vcpu, vcpu->arch.efer);
11525
11526         kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
11527         kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
11528         vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
11529         /*
11530          * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
11531          * actually changed, because vmx_set_cr0 refers to efer set above.
11532          *
11533          * CR0_GUEST_HOST_MASK is already set in the original vmcs01
11534          * (KVM doesn't change it);
11535          */
11536         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
11537         vmx_set_cr0(vcpu, vmcs12->host_cr0);
11538
11539         /* Same as above - no reason to call set_cr4_guest_host_mask().  */
11540         vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
11541         vmx_set_cr4(vcpu, vmcs12->host_cr4);
11542
11543         load_vmcs12_mmu_host_state(vcpu, vmcs12);
11544
11545         if (enable_vpid) {
11546                 /*
11547                  * Trivially support vpid by letting L2s share their parent
11548                  * L1's vpid. TODO: move to a more elaborate solution, giving
11549                  * each L2 its own vpid and exposing the vpid feature to L1.
11550                  */
11551                 vmx_flush_tlb(vcpu);
11552         }
11553         /* Restore posted intr vector. */
11554         if (nested_cpu_has_posted_intr(vmcs12))
11555                 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
11556
11557         vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
11558         vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
11559         vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
11560         vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
11561         vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
11562         vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
11563         vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
11564
11565         /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
11566         if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
11567                 vmcs_write64(GUEST_BNDCFGS, 0);
11568
11569         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
11570                 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
11571                 vcpu->arch.pat = vmcs12->host_ia32_pat;
11572         }
11573         if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
11574                 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
11575                         vmcs12->host_ia32_perf_global_ctrl);
11576
11577         /* Set L1 segment info according to Intel SDM
11578             27.5.2 Loading Host Segment and Descriptor-Table Registers */
11579         seg = (struct kvm_segment) {
11580                 .base = 0,
11581                 .limit = 0xFFFFFFFF,
11582                 .selector = vmcs12->host_cs_selector,
11583                 .type = 11,
11584                 .present = 1,
11585                 .s = 1,
11586                 .g = 1
11587         };
11588         if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11589                 seg.l = 1;
11590         else
11591                 seg.db = 1;
11592         vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
11593         seg = (struct kvm_segment) {
11594                 .base = 0,
11595                 .limit = 0xFFFFFFFF,
11596                 .type = 3,
11597                 .present = 1,
11598                 .s = 1,
11599                 .db = 1,
11600                 .g = 1
11601         };
11602         seg.selector = vmcs12->host_ds_selector;
11603         vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
11604         seg.selector = vmcs12->host_es_selector;
11605         vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
11606         seg.selector = vmcs12->host_ss_selector;
11607         vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
11608         seg.selector = vmcs12->host_fs_selector;
11609         seg.base = vmcs12->host_fs_base;
11610         vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
11611         seg.selector = vmcs12->host_gs_selector;
11612         seg.base = vmcs12->host_gs_base;
11613         vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
11614         seg = (struct kvm_segment) {
11615                 .base = vmcs12->host_tr_base,
11616                 .limit = 0x67,
11617                 .selector = vmcs12->host_tr_selector,
11618                 .type = 11,
11619                 .present = 1
11620         };
11621         vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
11622
11623         kvm_set_dr(vcpu, 7, 0x400);
11624         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11625
11626         if (cpu_has_vmx_msr_bitmap())
11627                 vmx_update_msr_bitmap(vcpu);
11628
11629         if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11630                                 vmcs12->vm_exit_msr_load_count))
11631                 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11632 }
11633
11634 /*
11635  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11636  * and modify vmcs12 to make it see what it would expect to see there if
11637  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11638  */
11639 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11640                               u32 exit_intr_info,
11641                               unsigned long exit_qualification)
11642 {
11643         struct vcpu_vmx *vmx = to_vmx(vcpu);
11644         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11645
11646         /* trying to cancel vmlaunch/vmresume is a bug */
11647         WARN_ON_ONCE(vmx->nested.nested_run_pending);
11648
11649         /*
11650          * The only expected VM-instruction error is "VM entry with
11651          * invalid control field(s)." Anything else indicates a
11652          * problem with L0.
11653          */
11654         WARN_ON_ONCE(vmx->fail && (vmcs_read32(VM_INSTRUCTION_ERROR) !=
11655                                    VMXERR_ENTRY_INVALID_CONTROL_FIELD));
11656
11657         leave_guest_mode(vcpu);
11658
11659         if (likely(!vmx->fail)) {
11660                 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11661                                exit_qualification);
11662
11663                 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11664                                          vmcs12->vm_exit_msr_store_count))
11665                         nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11666         }
11667
11668         vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11669         vm_entry_controls_reset_shadow(vmx);
11670         vm_exit_controls_reset_shadow(vmx);
11671         vmx_segment_cache_clear(vmx);
11672
11673         /* Update any VMCS fields that might have changed while L2 ran */
11674         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11675         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11676         vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11677         if (vmx->hv_deadline_tsc == -1)
11678                 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11679                                 PIN_BASED_VMX_PREEMPTION_TIMER);
11680         else
11681                 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11682                               PIN_BASED_VMX_PREEMPTION_TIMER);
11683         if (kvm_has_tsc_control)
11684                 decache_tsc_multiplier(vmx);
11685
11686         if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11687                 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11688                 vmx_set_virtual_x2apic_mode(vcpu,
11689                                 vcpu->arch.apic_base & X2APIC_ENABLE);
11690         } else if (!nested_cpu_has_ept(vmcs12) &&
11691                    nested_cpu_has2(vmcs12,
11692                                    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11693                 vmx_flush_tlb_ept_only(vcpu);
11694         }
11695
11696         /* This is needed for same reason as it was needed in prepare_vmcs02 */
11697         vmx->host_rsp = 0;
11698
11699         /* Unpin physical memory we referred to in vmcs02 */
11700         if (vmx->nested.apic_access_page) {
11701                 kvm_release_page_dirty(vmx->nested.apic_access_page);
11702                 vmx->nested.apic_access_page = NULL;
11703         }
11704         if (vmx->nested.virtual_apic_page) {
11705                 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
11706                 vmx->nested.virtual_apic_page = NULL;
11707         }
11708         if (vmx->nested.pi_desc_page) {
11709                 kunmap(vmx->nested.pi_desc_page);
11710                 kvm_release_page_dirty(vmx->nested.pi_desc_page);
11711                 vmx->nested.pi_desc_page = NULL;
11712                 vmx->nested.pi_desc = NULL;
11713         }
11714
11715         /*
11716          * We are now running in L2, mmu_notifier will force to reload the
11717          * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11718          */
11719         kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11720
11721         if (enable_shadow_vmcs)
11722                 vmx->nested.sync_shadow_vmcs = true;
11723
11724         /* in case we halted in L2 */
11725         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11726
11727         if (likely(!vmx->fail)) {
11728                 /*
11729                  * TODO: SDM says that with acknowledge interrupt on
11730                  * exit, bit 31 of the VM-exit interrupt information
11731                  * (valid interrupt) is always set to 1 on
11732                  * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
11733                  * need kvm_cpu_has_interrupt().  See the commit
11734                  * message for details.
11735                  */
11736                 if (nested_exit_intr_ack_set(vcpu) &&
11737                     exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
11738                     kvm_cpu_has_interrupt(vcpu)) {
11739                         int irq = kvm_cpu_get_interrupt(vcpu);
11740                         WARN_ON(irq < 0);
11741                         vmcs12->vm_exit_intr_info = irq |
11742                                 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11743                 }
11744
11745                 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11746                                                vmcs12->exit_qualification,
11747                                                vmcs12->idt_vectoring_info_field,
11748                                                vmcs12->vm_exit_intr_info,
11749                                                vmcs12->vm_exit_intr_error_code,
11750                                                KVM_ISA_VMX);
11751
11752                 load_vmcs12_host_state(vcpu, vmcs12);
11753
11754                 return;
11755         }
11756
11757         /*
11758          * After an early L2 VM-entry failure, we're now back
11759          * in L1 which thinks it just finished a VMLAUNCH or
11760          * VMRESUME instruction, so we need to set the failure
11761          * flag and the VM-instruction error field of the VMCS
11762          * accordingly.
11763          */
11764         nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
11765
11766         load_vmcs12_mmu_host_state(vcpu, vmcs12);
11767
11768         /*
11769          * The emulated instruction was already skipped in
11770          * nested_vmx_run, but the updated RIP was never
11771          * written back to the vmcs01.
11772          */
11773         skip_emulated_instruction(vcpu);
11774         vmx->fail = 0;
11775 }
11776
11777 /*
11778  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11779  */
11780 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11781 {
11782         if (is_guest_mode(vcpu)) {
11783                 to_vmx(vcpu)->nested.nested_run_pending = 0;
11784                 nested_vmx_vmexit(vcpu, -1, 0, 0);
11785         }
11786         free_nested(to_vmx(vcpu));
11787 }
11788
11789 /*
11790  * L1's failure to enter L2 is a subset of a normal exit, as explained in
11791  * 23.7 "VM-entry failures during or after loading guest state" (this also
11792  * lists the acceptable exit-reason and exit-qualification parameters).
11793  * It should only be called before L2 actually succeeded to run, and when
11794  * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11795  */
11796 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11797                         struct vmcs12 *vmcs12,
11798                         u32 reason, unsigned long qualification)
11799 {
11800         load_vmcs12_host_state(vcpu, vmcs12);
11801         vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11802         vmcs12->exit_qualification = qualification;
11803         nested_vmx_succeed(vcpu);
11804         if (enable_shadow_vmcs)
11805                 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11806 }
11807
11808 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11809                                struct x86_instruction_info *info,
11810                                enum x86_intercept_stage stage)
11811 {
11812         return X86EMUL_CONTINUE;
11813 }
11814
11815 #ifdef CONFIG_X86_64
11816 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11817 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11818                                   u64 divisor, u64 *result)
11819 {
11820         u64 low = a << shift, high = a >> (64 - shift);
11821
11822         /* To avoid the overflow on divq */
11823         if (high >= divisor)
11824                 return 1;
11825
11826         /* Low hold the result, high hold rem which is discarded */
11827         asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11828             "rm" (divisor), "0" (low), "1" (high));
11829         *result = low;
11830
11831         return 0;
11832 }
11833
11834 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11835 {
11836         struct vcpu_vmx *vmx = to_vmx(vcpu);
11837         u64 tscl = rdtsc();
11838         u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11839         u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11840
11841         /* Convert to host delta tsc if tsc scaling is enabled */
11842         if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11843                         u64_shl_div_u64(delta_tsc,
11844                                 kvm_tsc_scaling_ratio_frac_bits,
11845                                 vcpu->arch.tsc_scaling_ratio,
11846                                 &delta_tsc))
11847                 return -ERANGE;
11848
11849         /*
11850          * If the delta tsc can't fit in the 32 bit after the multi shift,
11851          * we can't use the preemption timer.
11852          * It's possible that it fits on later vmentries, but checking
11853          * on every vmentry is costly so we just use an hrtimer.
11854          */
11855         if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11856                 return -ERANGE;
11857
11858         vmx->hv_deadline_tsc = tscl + delta_tsc;
11859         vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11860                         PIN_BASED_VMX_PREEMPTION_TIMER);
11861
11862         return delta_tsc == 0;
11863 }
11864
11865 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11866 {
11867         struct vcpu_vmx *vmx = to_vmx(vcpu);
11868         vmx->hv_deadline_tsc = -1;
11869         vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11870                         PIN_BASED_VMX_PREEMPTION_TIMER);
11871 }
11872 #endif
11873
11874 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11875 {
11876         if (ple_gap)
11877                 shrink_ple_window(vcpu);
11878 }
11879
11880 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11881                                      struct kvm_memory_slot *slot)
11882 {
11883         kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11884         kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11885 }
11886
11887 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11888                                        struct kvm_memory_slot *slot)
11889 {
11890         kvm_mmu_slot_set_dirty(kvm, slot);
11891 }
11892
11893 static void vmx_flush_log_dirty(struct kvm *kvm)
11894 {
11895         kvm_flush_pml_buffers(kvm);
11896 }
11897
11898 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
11899 {
11900         struct vmcs12 *vmcs12;
11901         struct vcpu_vmx *vmx = to_vmx(vcpu);
11902         gpa_t gpa;
11903         struct page *page = NULL;
11904         u64 *pml_address;
11905
11906         if (is_guest_mode(vcpu)) {
11907                 WARN_ON_ONCE(vmx->nested.pml_full);
11908
11909                 /*
11910                  * Check if PML is enabled for the nested guest.
11911                  * Whether eptp bit 6 is set is already checked
11912                  * as part of A/D emulation.
11913                  */
11914                 vmcs12 = get_vmcs12(vcpu);
11915                 if (!nested_cpu_has_pml(vmcs12))
11916                         return 0;
11917
11918                 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
11919                         vmx->nested.pml_full = true;
11920                         return 1;
11921                 }
11922
11923                 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
11924
11925                 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
11926                 if (is_error_page(page))
11927                         return 0;
11928
11929                 pml_address = kmap(page);
11930                 pml_address[vmcs12->guest_pml_index--] = gpa;
11931                 kunmap(page);
11932                 kvm_release_page_clean(page);
11933         }
11934
11935         return 0;
11936 }
11937
11938 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11939                                            struct kvm_memory_slot *memslot,
11940                                            gfn_t offset, unsigned long mask)
11941 {
11942         kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11943 }
11944
11945 static void __pi_post_block(struct kvm_vcpu *vcpu)
11946 {
11947         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11948         struct pi_desc old, new;
11949         unsigned int dest;
11950
11951         do {
11952                 old.control = new.control = pi_desc->control;
11953                 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
11954                      "Wakeup handler not enabled while the VCPU is blocked\n");
11955
11956                 dest = cpu_physical_id(vcpu->cpu);
11957
11958                 if (x2apic_enabled())
11959                         new.ndst = dest;
11960                 else
11961                         new.ndst = (dest << 8) & 0xFF00;
11962
11963                 /* set 'NV' to 'notification vector' */
11964                 new.nv = POSTED_INTR_VECTOR;
11965         } while (cmpxchg64(&pi_desc->control, old.control,
11966                            new.control) != old.control);
11967
11968         if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
11969                 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11970                 list_del(&vcpu->blocked_vcpu_list);
11971                 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11972                 vcpu->pre_pcpu = -1;
11973         }
11974 }
11975
11976 /*
11977  * This routine does the following things for vCPU which is going
11978  * to be blocked if VT-d PI is enabled.
11979  * - Store the vCPU to the wakeup list, so when interrupts happen
11980  *   we can find the right vCPU to wake up.
11981  * - Change the Posted-interrupt descriptor as below:
11982  *      'NDST' <-- vcpu->pre_pcpu
11983  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11984  * - If 'ON' is set during this process, which means at least one
11985  *   interrupt is posted for this vCPU, we cannot block it, in
11986  *   this case, return 1, otherwise, return 0.
11987  *
11988  */
11989 static int pi_pre_block(struct kvm_vcpu *vcpu)
11990 {
11991         unsigned int dest;
11992         struct pi_desc old, new;
11993         struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11994
11995         if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11996                 !irq_remapping_cap(IRQ_POSTING_CAP)  ||
11997                 !kvm_vcpu_apicv_active(vcpu))
11998                 return 0;
11999
12000         WARN_ON(irqs_disabled());
12001         local_irq_disable();
12002         if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
12003                 vcpu->pre_pcpu = vcpu->cpu;
12004                 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12005                 list_add_tail(&vcpu->blocked_vcpu_list,
12006                               &per_cpu(blocked_vcpu_on_cpu,
12007                                        vcpu->pre_pcpu));
12008                 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12009         }
12010
12011         do {
12012                 old.control = new.control = pi_desc->control;
12013
12014                 WARN((pi_desc->sn == 1),
12015                      "Warning: SN field of posted-interrupts "
12016                      "is set before blocking\n");
12017
12018                 /*
12019                  * Since vCPU can be preempted during this process,
12020                  * vcpu->cpu could be different with pre_pcpu, we
12021                  * need to set pre_pcpu as the destination of wakeup
12022                  * notification event, then we can find the right vCPU
12023                  * to wakeup in wakeup handler if interrupts happen
12024                  * when the vCPU is in blocked state.
12025                  */
12026                 dest = cpu_physical_id(vcpu->pre_pcpu);
12027
12028                 if (x2apic_enabled())
12029                         new.ndst = dest;
12030                 else
12031                         new.ndst = (dest << 8) & 0xFF00;
12032
12033                 /* set 'NV' to 'wakeup vector' */
12034                 new.nv = POSTED_INTR_WAKEUP_VECTOR;
12035         } while (cmpxchg64(&pi_desc->control, old.control,
12036                            new.control) != old.control);
12037
12038         /* We should not block the vCPU if an interrupt is posted for it.  */
12039         if (pi_test_on(pi_desc) == 1)
12040                 __pi_post_block(vcpu);
12041
12042         local_irq_enable();
12043         return (vcpu->pre_pcpu == -1);
12044 }
12045
12046 static int vmx_pre_block(struct kvm_vcpu *vcpu)
12047 {
12048         if (pi_pre_block(vcpu))
12049                 return 1;
12050
12051         if (kvm_lapic_hv_timer_in_use(vcpu))
12052                 kvm_lapic_switch_to_sw_timer(vcpu);
12053
12054         return 0;
12055 }
12056
12057 static void pi_post_block(struct kvm_vcpu *vcpu)
12058 {
12059         if (vcpu->pre_pcpu == -1)
12060                 return;
12061
12062         WARN_ON(irqs_disabled());
12063         local_irq_disable();
12064         __pi_post_block(vcpu);
12065         local_irq_enable();
12066 }
12067
12068 static void vmx_post_block(struct kvm_vcpu *vcpu)
12069 {
12070         if (kvm_x86_ops->set_hv_timer)
12071                 kvm_lapic_switch_to_hv_timer(vcpu);
12072
12073         pi_post_block(vcpu);
12074 }
12075
12076 /*
12077  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
12078  *
12079  * @kvm: kvm
12080  * @host_irq: host irq of the interrupt
12081  * @guest_irq: gsi of the interrupt
12082  * @set: set or unset PI
12083  * returns 0 on success, < 0 on failure
12084  */
12085 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
12086                               uint32_t guest_irq, bool set)
12087 {
12088         struct kvm_kernel_irq_routing_entry *e;
12089         struct kvm_irq_routing_table *irq_rt;
12090         struct kvm_lapic_irq irq;
12091         struct kvm_vcpu *vcpu;
12092         struct vcpu_data vcpu_info;
12093         int idx, ret = 0;
12094
12095         if (!kvm_arch_has_assigned_device(kvm) ||
12096                 !irq_remapping_cap(IRQ_POSTING_CAP) ||
12097                 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
12098                 return 0;
12099
12100         idx = srcu_read_lock(&kvm->irq_srcu);
12101         irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
12102         if (guest_irq >= irq_rt->nr_rt_entries ||
12103             hlist_empty(&irq_rt->map[guest_irq])) {
12104                 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
12105                              guest_irq, irq_rt->nr_rt_entries);
12106                 goto out;
12107         }
12108
12109         hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
12110                 if (e->type != KVM_IRQ_ROUTING_MSI)
12111                         continue;
12112                 /*
12113                  * VT-d PI cannot support posting multicast/broadcast
12114                  * interrupts to a vCPU, we still use interrupt remapping
12115                  * for these kind of interrupts.
12116                  *
12117                  * For lowest-priority interrupts, we only support
12118                  * those with single CPU as the destination, e.g. user
12119                  * configures the interrupts via /proc/irq or uses
12120                  * irqbalance to make the interrupts single-CPU.
12121                  *
12122                  * We will support full lowest-priority interrupt later.
12123                  */
12124
12125                 kvm_set_msi_irq(kvm, e, &irq);
12126                 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
12127                         /*
12128                          * Make sure the IRTE is in remapped mode if
12129                          * we don't handle it in posted mode.
12130                          */
12131                         ret = irq_set_vcpu_affinity(host_irq, NULL);
12132                         if (ret < 0) {
12133                                 printk(KERN_INFO
12134                                    "failed to back to remapped mode, irq: %u\n",
12135                                    host_irq);
12136                                 goto out;
12137                         }
12138
12139                         continue;
12140                 }
12141
12142                 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
12143                 vcpu_info.vector = irq.vector;
12144
12145                 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
12146                                 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
12147
12148                 if (set)
12149                         ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
12150                 else
12151                         ret = irq_set_vcpu_affinity(host_irq, NULL);
12152
12153                 if (ret < 0) {
12154                         printk(KERN_INFO "%s: failed to update PI IRTE\n",
12155                                         __func__);
12156                         goto out;
12157                 }
12158         }
12159
12160         ret = 0;
12161 out:
12162         srcu_read_unlock(&kvm->irq_srcu, idx);
12163         return ret;
12164 }
12165
12166 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
12167 {
12168         if (vcpu->arch.mcg_cap & MCG_LMCE_P)
12169                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
12170                         FEATURE_CONTROL_LMCE;
12171         else
12172                 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
12173                         ~FEATURE_CONTROL_LMCE;
12174 }
12175
12176 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
12177         .cpu_has_kvm_support = cpu_has_kvm_support,
12178         .disabled_by_bios = vmx_disabled_by_bios,
12179         .hardware_setup = hardware_setup,
12180         .hardware_unsetup = hardware_unsetup,
12181         .check_processor_compatibility = vmx_check_processor_compat,
12182         .hardware_enable = hardware_enable,
12183         .hardware_disable = hardware_disable,
12184         .cpu_has_accelerated_tpr = report_flexpriority,
12185         .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
12186
12187         .vcpu_create = vmx_create_vcpu,
12188         .vcpu_free = vmx_free_vcpu,
12189         .vcpu_reset = vmx_vcpu_reset,
12190
12191         .prepare_guest_switch = vmx_save_host_state,
12192         .vcpu_load = vmx_vcpu_load,
12193         .vcpu_put = vmx_vcpu_put,
12194
12195         .update_bp_intercept = update_exception_bitmap,
12196         .get_msr = vmx_get_msr,
12197         .set_msr = vmx_set_msr,
12198         .get_segment_base = vmx_get_segment_base,
12199         .get_segment = vmx_get_segment,
12200         .set_segment = vmx_set_segment,
12201         .get_cpl = vmx_get_cpl,
12202         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
12203         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
12204         .decache_cr3 = vmx_decache_cr3,
12205         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
12206         .set_cr0 = vmx_set_cr0,
12207         .set_cr3 = vmx_set_cr3,
12208         .set_cr4 = vmx_set_cr4,
12209         .set_efer = vmx_set_efer,
12210         .get_idt = vmx_get_idt,
12211         .set_idt = vmx_set_idt,
12212         .get_gdt = vmx_get_gdt,
12213         .set_gdt = vmx_set_gdt,
12214         .get_dr6 = vmx_get_dr6,
12215         .set_dr6 = vmx_set_dr6,
12216         .set_dr7 = vmx_set_dr7,
12217         .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
12218         .cache_reg = vmx_cache_reg,
12219         .get_rflags = vmx_get_rflags,
12220         .set_rflags = vmx_set_rflags,
12221
12222         .tlb_flush = vmx_flush_tlb,
12223
12224         .run = vmx_vcpu_run,
12225         .handle_exit = vmx_handle_exit,
12226         .skip_emulated_instruction = skip_emulated_instruction,
12227         .set_interrupt_shadow = vmx_set_interrupt_shadow,
12228         .get_interrupt_shadow = vmx_get_interrupt_shadow,
12229         .patch_hypercall = vmx_patch_hypercall,
12230         .set_irq = vmx_inject_irq,
12231         .set_nmi = vmx_inject_nmi,
12232         .queue_exception = vmx_queue_exception,
12233         .cancel_injection = vmx_cancel_injection,
12234         .interrupt_allowed = vmx_interrupt_allowed,
12235         .nmi_allowed = vmx_nmi_allowed,
12236         .get_nmi_mask = vmx_get_nmi_mask,
12237         .set_nmi_mask = vmx_set_nmi_mask,
12238         .enable_nmi_window = enable_nmi_window,
12239         .enable_irq_window = enable_irq_window,
12240         .update_cr8_intercept = update_cr8_intercept,
12241         .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
12242         .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
12243         .get_enable_apicv = vmx_get_enable_apicv,
12244         .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
12245         .load_eoi_exitmap = vmx_load_eoi_exitmap,
12246         .apicv_post_state_restore = vmx_apicv_post_state_restore,
12247         .hwapic_irr_update = vmx_hwapic_irr_update,
12248         .hwapic_isr_update = vmx_hwapic_isr_update,
12249         .sync_pir_to_irr = vmx_sync_pir_to_irr,
12250         .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
12251
12252         .set_tss_addr = vmx_set_tss_addr,
12253         .get_tdp_level = get_ept_level,
12254         .get_mt_mask = vmx_get_mt_mask,
12255
12256         .get_exit_info = vmx_get_exit_info,
12257
12258         .get_lpage_level = vmx_get_lpage_level,
12259
12260         .cpuid_update = vmx_cpuid_update,
12261
12262         .rdtscp_supported = vmx_rdtscp_supported,
12263         .invpcid_supported = vmx_invpcid_supported,
12264
12265         .set_supported_cpuid = vmx_set_supported_cpuid,
12266
12267         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
12268
12269         .write_tsc_offset = vmx_write_tsc_offset,
12270
12271         .set_tdp_cr3 = vmx_set_cr3,
12272
12273         .check_intercept = vmx_check_intercept,
12274         .handle_external_intr = vmx_handle_external_intr,
12275         .mpx_supported = vmx_mpx_supported,
12276         .xsaves_supported = vmx_xsaves_supported,
12277
12278         .check_nested_events = vmx_check_nested_events,
12279
12280         .sched_in = vmx_sched_in,
12281
12282         .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
12283         .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
12284         .flush_log_dirty = vmx_flush_log_dirty,
12285         .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
12286         .write_log_dirty = vmx_write_pml_buffer,
12287
12288         .pre_block = vmx_pre_block,
12289         .post_block = vmx_post_block,
12290
12291         .pmu_ops = &intel_pmu_ops,
12292
12293         .update_pi_irte = vmx_update_pi_irte,
12294
12295 #ifdef CONFIG_X86_64
12296         .set_hv_timer = vmx_set_hv_timer,
12297         .cancel_hv_timer = vmx_cancel_hv_timer,
12298 #endif
12299
12300         .setup_mce = vmx_setup_mce,
12301 };
12302
12303 static int __init vmx_init(void)
12304 {
12305         int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
12306                      __alignof__(struct vcpu_vmx), THIS_MODULE);
12307         if (r)
12308                 return r;
12309
12310 #ifdef CONFIG_KEXEC_CORE
12311         rcu_assign_pointer(crash_vmclear_loaded_vmcss,
12312                            crash_vmclear_local_loaded_vmcss);
12313 #endif
12314
12315         return 0;
12316 }
12317
12318 static void __exit vmx_exit(void)
12319 {
12320 #ifdef CONFIG_KEXEC_CORE
12321         RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
12322         synchronize_rcu();
12323 #endif
12324
12325         kvm_exit();
12326 }
12327
12328 module_init(vmx_init)
12329 module_exit(vmx_exit)