/* $Id: HMVMXR0.cpp 49664 2013-11-26 15:53:35Z vboxsync $ */ /** @file * HM VMX (Intel VT-x) - Host Context Ring-0. */ /* * Copyright (C) 2012-2013 Oracle Corporation * * This file is part of VirtualBox Open Source Edition (OSE), as * available from http://www.virtualbox.org. This file is free software; * you can redistribute it and/or modify it under the terms of the GNU * General Public License (GPL) as published by the Free Software * Foundation, in version 2 as it comes in the "COPYING" file of the * VirtualBox OSE distribution. VirtualBox OSE is distributed in the * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind. */ /******************************************************************************* * Header Files * *******************************************************************************/ #define LOG_GROUP LOG_GROUP_HM #include #include #include #include "HMInternal.h" #include #include "HMVMXR0.h" #include #include #include #include #include #include #ifdef VBOX_WITH_REM # include #endif #ifdef DEBUG_ramshankar # define HMVMX_SAVE_FULL_GUEST_STATE # define HMVMX_SYNC_FULL_GUEST_STATE # define HMVMX_ALWAYS_CHECK_GUEST_STATE # define HMVMX_ALWAYS_TRAP_ALL_XCPTS # define HMVMX_ALWAYS_TRAP_PF # define HMVMX_ALWAYS_SWAP_FPU_STATE # define HMVMX_ALWAYS_FLUSH_TLB #endif /******************************************************************************* * Defined Constants And Macros * *******************************************************************************/ #if defined(RT_ARCH_AMD64) # define HMVMX_IS_64BIT_HOST_MODE() (true) typedef RTHCUINTREG HMVMXHCUINTREG; #elif defined(VBOX_WITH_HYBRID_32BIT_KERNEL) extern "C" uint32_t g_fVMXIs64bitHost; # define HMVMX_IS_64BIT_HOST_MODE() (g_fVMXIs64bitHost != 0) typedef uint64_t HMVMXHCUINTREG; #else # define HMVMX_IS_64BIT_HOST_MODE() (false) typedef RTHCUINTREG HMVMXHCUINTREG; #endif /** Use the function table. */ #define HMVMX_USE_FUNCTION_TABLE /** Determine which tagged-TLB flush handler to use. */ #define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0 #define HMVMX_FLUSH_TAGGED_TLB_EPT 1 #define HMVMX_FLUSH_TAGGED_TLB_VPID 2 #define HMVMX_FLUSH_TAGGED_TLB_NONE 3 /** @name Updated-guest-state flags. * @{ */ #define HMVMX_UPDATED_GUEST_RIP RT_BIT(0) #define HMVMX_UPDATED_GUEST_RSP RT_BIT(1) #define HMVMX_UPDATED_GUEST_RFLAGS RT_BIT(2) #define HMVMX_UPDATED_GUEST_CR0 RT_BIT(3) #define HMVMX_UPDATED_GUEST_CR3 RT_BIT(4) #define HMVMX_UPDATED_GUEST_CR4 RT_BIT(5) #define HMVMX_UPDATED_GUEST_GDTR RT_BIT(6) #define HMVMX_UPDATED_GUEST_IDTR RT_BIT(7) #define HMVMX_UPDATED_GUEST_LDTR RT_BIT(8) #define HMVMX_UPDATED_GUEST_TR RT_BIT(9) #define HMVMX_UPDATED_GUEST_SEGMENT_REGS RT_BIT(10) #define HMVMX_UPDATED_GUEST_DEBUG RT_BIT(11) #define HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR RT_BIT(12) #define HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR RT_BIT(13) #define HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR RT_BIT(14) #define HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS RT_BIT(15) #define HMVMX_UPDATED_GUEST_LAZY_MSRS RT_BIT(16) #define HMVMX_UPDATED_GUEST_ACTIVITY_STATE RT_BIT(17) #define HMVMX_UPDATED_GUEST_APIC_STATE RT_BIT(18) #define HMVMX_UPDATED_GUEST_ALL ( HMVMX_UPDATED_GUEST_RIP \ | HMVMX_UPDATED_GUEST_RSP \ | HMVMX_UPDATED_GUEST_RFLAGS \ | HMVMX_UPDATED_GUEST_CR0 \ | HMVMX_UPDATED_GUEST_CR3 \ | HMVMX_UPDATED_GUEST_CR4 \ | HMVMX_UPDATED_GUEST_GDTR \ | HMVMX_UPDATED_GUEST_IDTR \ | HMVMX_UPDATED_GUEST_LDTR \ | HMVMX_UPDATED_GUEST_TR \ | HMVMX_UPDATED_GUEST_SEGMENT_REGS \ | HMVMX_UPDATED_GUEST_DEBUG \ | HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR \ | HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR \ | HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR \ | HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS \ | HMVMX_UPDATED_GUEST_LAZY_MSRS \ | HMVMX_UPDATED_GUEST_ACTIVITY_STATE \ | HMVMX_UPDATED_GUEST_APIC_STATE) /** @} */ /** @name * Flags to skip redundant reads of some common VMCS fields that are not part of * the guest-CPU state but are in the transient structure. */ #define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO RT_BIT(0) #define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE RT_BIT(1) #define HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION RT_BIT(2) #define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN RT_BIT(3) #define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO RT_BIT(4) #define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE RT_BIT(5) #define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO RT_BIT(6) /** @} */ /** @name * States of the VMCS. * * This does not reflect all possible VMCS states but currently only those * needed for maintaining the VMCS consistently even when thread-context hooks * are used. Maybe later this can be extended (i.e. Nested Virtualization). */ #define HMVMX_VMCS_STATE_CLEAR RT_BIT(0) #define HMVMX_VMCS_STATE_ACTIVE RT_BIT(1) #define HMVMX_VMCS_STATE_LAUNCHED RT_BIT(2) /** @} */ /** * Exception bitmap mask for real-mode guests (real-on-v86). * * We need to intercept all exceptions manually (except #PF). #NM is also * handled separately, see hmR0VmxLoadSharedCR0(). #PF need not be intercepted * even in real-mode if we have Nested Paging support. */ #define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) | RT_BIT(X86_XCPT_DB) | RT_BIT(X86_XCPT_NMI) \ | RT_BIT(X86_XCPT_BP) | RT_BIT(X86_XCPT_OF) | RT_BIT(X86_XCPT_BR) \ | RT_BIT(X86_XCPT_UD) /* RT_BIT(X86_XCPT_NM) */ | RT_BIT(X86_XCPT_DF) \ | RT_BIT(X86_XCPT_CO_SEG_OVERRUN) | RT_BIT(X86_XCPT_TS) | RT_BIT(X86_XCPT_NP) \ | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \ | RT_BIT(X86_XCPT_MF) | RT_BIT(X86_XCPT_AC) | RT_BIT(X86_XCPT_MC) \ | RT_BIT(X86_XCPT_XF)) /** * Exception bitmap mask for all contributory exceptions. * * Page fault is deliberately excluded here as it's conditional as to whether * it's contributory or benign. Page faults are handled separately. */ #define HMVMX_CONTRIBUTORY_XCPT_MASK ( RT_BIT(X86_XCPT_GP) | RT_BIT(X86_XCPT_NP) | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_TS) \ | RT_BIT(X86_XCPT_DE)) /** Maximum VM-instruction error number. */ #define HMVMX_INSTR_ERROR_MAX 28 /** Profiling macro. */ #ifdef HM_PROFILE_EXIT_DISPATCH # define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed) # define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed) #else # define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0) # define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0) #endif /** Assert that preemption is disabled or covered by thread-context hooks. */ #define HMVMX_ASSERT_PREEMPT_SAFE() Assert( VMMR0ThreadCtxHooksAreRegistered(pVCpu) \ || !RTThreadPreemptIsEnabled(NIL_RTTHREAD)); /** Assert that we haven't migrated CPUs when thread-context hooks are not * used. */ #define HMVMX_ASSERT_CPU_SAFE() AssertMsg( VMMR0ThreadCtxHooksAreRegistered(pVCpu) \ || pVCpu->hm.s.idEnteredCpu == RTMpCpuId(), \ ("Illegal migration! Entered on CPU %u Current %u\n", \ pVCpu->hm.s.idEnteredCpu, RTMpCpuId())); \ /** Helper macro for VM-exit handlers called unexpectedly. */ #define HMVMX_RETURN_UNEXPECTED_EXIT() \ do { \ pVCpu->hm.s.u32HMError = pVmxTransient->uExitReason; \ return VERR_VMX_UNEXPECTED_EXIT; \ } while (0) /******************************************************************************* * Structures and Typedefs * *******************************************************************************/ /** * VMX transient state. * * A state structure for holding miscellaneous information across * VMX non-root operation and restored after the transition. */ typedef struct VMXTRANSIENT { /** The host's rflags/eflags. */ RTCCUINTREG uEflags; #if HC_ARCH_BITS == 32 uint32_t u32Alignment0; #endif /** The guest's LSTAR MSR value used for TPR patching for 32-bit guests. */ uint64_t u64LStarMsr; /** The guest's TPR value used for TPR shadowing. */ uint8_t u8GuestTpr; /** Alignment. */ uint8_t abAlignment0[7]; /** The basic VM-exit reason. */ uint16_t uExitReason; /** Alignment. */ uint16_t u16Alignment0; /** The VM-exit interruption error code. */ uint32_t uExitIntErrorCode; /** The VM-exit exit qualification. */ uint64_t uExitQualification; /** The VM-exit interruption-information field. */ uint32_t uExitIntInfo; /** The VM-exit instruction-length field. */ uint32_t cbInstr; /** The VM-exit instruction-information field. */ union { /** Plain unsigned int representation. */ uint32_t u; /** INS and OUTS information. */ struct { uint32_t u6Reserved0 : 7; /** The address size; 0=16-bit, 1=32-bit, 2=64-bit, rest undefined. */ uint32_t u3AddrSize : 3; uint32_t u5Reserved1 : 5; /** The segment register (X86_SREG_XXX). */ uint32_t iSegReg : 3; uint32_t uReserved2 : 14; } StrIo; } ExitInstrInfo; /** Whether the VM-entry failed or not. */ bool fVMEntryFailed; /** Alignment. */ uint8_t abAlignment1[3]; /** The VM-entry interruption-information field. */ uint32_t uEntryIntInfo; /** The VM-entry exception error code field. */ uint32_t uEntryXcptErrorCode; /** The VM-entry instruction length field. */ uint32_t cbEntryInstr; /** IDT-vectoring information field. */ uint32_t uIdtVectoringInfo; /** IDT-vectoring error code. */ uint32_t uIdtVectoringErrorCode; /** Mask of currently read VMCS fields; HMVMX_UPDATED_TRANSIENT_*. */ uint32_t fVmcsFieldsRead; /** Whether the guest FPU was active at the time of VM-exit. */ bool fWasGuestFPUStateActive; /** Whether the guest debug state was active at the time of VM-exit. */ bool fWasGuestDebugStateActive; /** Whether the hyper debug state was active at the time of VM-exit. */ bool fWasHyperDebugStateActive; /** Whether TSC-offsetting should be setup before VM-entry. */ bool fUpdateTscOffsettingAndPreemptTimer; /** Whether the VM-exit was caused by a page-fault during delivery of a * contributory exception or a page-fault. */ bool fVectoringPF; } VMXTRANSIENT; AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, sizeof(uint64_t)); AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, sizeof(uint64_t)); AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, sizeof(uint64_t)); AssertCompileMemberAlignment(VMXTRANSIENT, fWasGuestFPUStateActive, sizeof(uint64_t)); AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t)); /** Pointer to VMX transient state. */ typedef VMXTRANSIENT *PVMXTRANSIENT; /** * MSR-bitmap read permissions. */ typedef enum VMXMSREXITREAD { /** Reading this MSR causes a VM-exit. */ VMXMSREXIT_INTERCEPT_READ = 0xb, /** Reading this MSR does not cause a VM-exit. */ VMXMSREXIT_PASSTHRU_READ } VMXMSREXITREAD; /** Pointer to MSR-bitmap read permissions. */ typedef VMXMSREXITREAD* PVMXMSREXITREAD; /** * MSR-bitmap write permissions. */ typedef enum VMXMSREXITWRITE { /** Writing to this MSR causes a VM-exit. */ VMXMSREXIT_INTERCEPT_WRITE = 0xd, /** Writing to this MSR does not cause a VM-exit. */ VMXMSREXIT_PASSTHRU_WRITE } VMXMSREXITWRITE; /** Pointer to MSR-bitmap write permissions. */ typedef VMXMSREXITWRITE* PVMXMSREXITWRITE; /** * VMX VM-exit handler. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required * fields before using them. * @param pVmxTransient Pointer to the VMX-transient structure. */ #ifndef HMVMX_USE_FUNCTION_TABLE typedef int FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); #else typedef DECLCALLBACK(int) FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); /** Pointer to VM-exit handler. */ typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER; #endif /******************************************************************************* * Internal Functions * *******************************************************************************/ static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMX_FLUSH_EPT enmFlush); static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMX_FLUSH_VPID enmFlush, RTGCPTR GCPtr); static int hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr, uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress, uint32_t *puIntState); #if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu); #endif #ifndef HMVMX_USE_FUNCTION_TABLE DECLINLINE(int) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason); # define HMVMX_EXIT_DECL static int #else # define HMVMX_EXIT_DECL static DECLCALLBACK(int) #endif /** @name VM-exit handlers. * @{ */ static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi; static FNVMXEXITHANDLER hmR0VmxExitExtInt; static FNVMXEXITHANDLER hmR0VmxExitTripleFault; static FNVMXEXITHANDLER hmR0VmxExitInitSignal; static FNVMXEXITHANDLER hmR0VmxExitSipi; static FNVMXEXITHANDLER hmR0VmxExitIoSmi; static FNVMXEXITHANDLER hmR0VmxExitSmi; static FNVMXEXITHANDLER hmR0VmxExitIntWindow; static FNVMXEXITHANDLER hmR0VmxExitNmiWindow; static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch; static FNVMXEXITHANDLER hmR0VmxExitCpuid; static FNVMXEXITHANDLER hmR0VmxExitGetsec; static FNVMXEXITHANDLER hmR0VmxExitHlt; static FNVMXEXITHANDLER hmR0VmxExitInvd; static FNVMXEXITHANDLER hmR0VmxExitInvlpg; static FNVMXEXITHANDLER hmR0VmxExitRdpmc; static FNVMXEXITHANDLER hmR0VmxExitRdtsc; static FNVMXEXITHANDLER hmR0VmxExitRsm; static FNVMXEXITHANDLER hmR0VmxExitSetPendingXcptUD; static FNVMXEXITHANDLER hmR0VmxExitMovCRx; static FNVMXEXITHANDLER hmR0VmxExitMovDRx; static FNVMXEXITHANDLER hmR0VmxExitIoInstr; static FNVMXEXITHANDLER hmR0VmxExitRdmsr; static FNVMXEXITHANDLER hmR0VmxExitWrmsr; static FNVMXEXITHANDLER hmR0VmxExitErrInvalidGuestState; static FNVMXEXITHANDLER hmR0VmxExitErrMsrLoad; static FNVMXEXITHANDLER hmR0VmxExitErrUndefined; static FNVMXEXITHANDLER hmR0VmxExitMwait; static FNVMXEXITHANDLER hmR0VmxExitMtf; static FNVMXEXITHANDLER hmR0VmxExitMonitor; static FNVMXEXITHANDLER hmR0VmxExitPause; static FNVMXEXITHANDLER hmR0VmxExitErrMachineCheck; static FNVMXEXITHANDLER hmR0VmxExitTprBelowThreshold; static FNVMXEXITHANDLER hmR0VmxExitApicAccess; static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess; static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess; static FNVMXEXITHANDLER hmR0VmxExitEptViolation; static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig; static FNVMXEXITHANDLER hmR0VmxExitRdtscp; static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer; static FNVMXEXITHANDLER hmR0VmxExitWbinvd; static FNVMXEXITHANDLER hmR0VmxExitXsetbv; static FNVMXEXITHANDLER hmR0VmxExitRdrand; static FNVMXEXITHANDLER hmR0VmxExitInvpcid; /** @} */ static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); #ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient); #endif static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx); /******************************************************************************* * Global Variables * *******************************************************************************/ #ifdef HMVMX_USE_FUNCTION_TABLE /** * VMX_EXIT dispatch table. */ static const PFNVMXEXITHANDLER g_apfnVMExitHandlers[VMX_EXIT_MAX + 1] = { /* 00 VMX_EXIT_XCPT_OR_NMI */ hmR0VmxExitXcptOrNmi, /* 01 VMX_EXIT_EXT_INT */ hmR0VmxExitExtInt, /* 02 VMX_EXIT_TRIPLE_FAULT */ hmR0VmxExitTripleFault, /* 03 VMX_EXIT_INIT_SIGNAL */ hmR0VmxExitInitSignal, /* 04 VMX_EXIT_SIPI */ hmR0VmxExitSipi, /* 05 VMX_EXIT_IO_SMI */ hmR0VmxExitIoSmi, /* 06 VMX_EXIT_SMI */ hmR0VmxExitSmi, /* 07 VMX_EXIT_INT_WINDOW */ hmR0VmxExitIntWindow, /* 08 VMX_EXIT_NMI_WINDOW */ hmR0VmxExitNmiWindow, /* 09 VMX_EXIT_TASK_SWITCH */ hmR0VmxExitTaskSwitch, /* 10 VMX_EXIT_CPUID */ hmR0VmxExitCpuid, /* 11 VMX_EXIT_GETSEC */ hmR0VmxExitGetsec, /* 12 VMX_EXIT_HLT */ hmR0VmxExitHlt, /* 13 VMX_EXIT_INVD */ hmR0VmxExitInvd, /* 14 VMX_EXIT_INVLPG */ hmR0VmxExitInvlpg, /* 15 VMX_EXIT_RDPMC */ hmR0VmxExitRdpmc, /* 16 VMX_EXIT_RDTSC */ hmR0VmxExitRdtsc, /* 17 VMX_EXIT_RSM */ hmR0VmxExitRsm, /* 18 VMX_EXIT_VMCALL */ hmR0VmxExitSetPendingXcptUD, /* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitSetPendingXcptUD, /* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitSetPendingXcptUD, /* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitSetPendingXcptUD, /* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitSetPendingXcptUD, /* 23 VMX_EXIT_VMREAD */ hmR0VmxExitSetPendingXcptUD, /* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitSetPendingXcptUD, /* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitSetPendingXcptUD, /* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitSetPendingXcptUD, /* 27 VMX_EXIT_VMXON */ hmR0VmxExitSetPendingXcptUD, /* 28 VMX_EXIT_MOV_CRX */ hmR0VmxExitMovCRx, /* 29 VMX_EXIT_MOV_DRX */ hmR0VmxExitMovDRx, /* 30 VMX_EXIT_IO_INSTR */ hmR0VmxExitIoInstr, /* 31 VMX_EXIT_RDMSR */ hmR0VmxExitRdmsr, /* 32 VMX_EXIT_WRMSR */ hmR0VmxExitWrmsr, /* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ hmR0VmxExitErrInvalidGuestState, /* 34 VMX_EXIT_ERR_MSR_LOAD */ hmR0VmxExitErrMsrLoad, /* 35 UNDEFINED */ hmR0VmxExitErrUndefined, /* 36 VMX_EXIT_MWAIT */ hmR0VmxExitMwait, /* 37 VMX_EXIT_MTF */ hmR0VmxExitMtf, /* 38 UNDEFINED */ hmR0VmxExitErrUndefined, /* 39 VMX_EXIT_MONITOR */ hmR0VmxExitMonitor, /* 40 UNDEFINED */ hmR0VmxExitPause, /* 41 VMX_EXIT_PAUSE */ hmR0VmxExitErrMachineCheck, /* 42 VMX_EXIT_ERR_MACHINE_CHECK */ hmR0VmxExitErrUndefined, /* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ hmR0VmxExitTprBelowThreshold, /* 44 VMX_EXIT_APIC_ACCESS */ hmR0VmxExitApicAccess, /* 45 UNDEFINED */ hmR0VmxExitErrUndefined, /* 46 VMX_EXIT_XDTR_ACCESS */ hmR0VmxExitXdtrAccess, /* 47 VMX_EXIT_TR_ACCESS */ hmR0VmxExitXdtrAccess, /* 48 VMX_EXIT_EPT_VIOLATION */ hmR0VmxExitEptViolation, /* 49 VMX_EXIT_EPT_MISCONFIG */ hmR0VmxExitEptMisconfig, /* 50 VMX_EXIT_INVEPT */ hmR0VmxExitSetPendingXcptUD, /* 51 VMX_EXIT_RDTSCP */ hmR0VmxExitRdtscp, /* 52 VMX_EXIT_PREEMPT_TIMER */ hmR0VmxExitPreemptTimer, /* 53 VMX_EXIT_INVVPID */ hmR0VmxExitSetPendingXcptUD, /* 54 VMX_EXIT_WBINVD */ hmR0VmxExitWbinvd, /* 55 VMX_EXIT_XSETBV */ hmR0VmxExitXsetbv, /* 56 UNDEFINED */ hmR0VmxExitErrUndefined, /* 57 VMX_EXIT_RDRAND */ hmR0VmxExitRdrand, /* 58 VMX_EXIT_INVPCID */ hmR0VmxExitInvpcid, /* 59 VMX_EXIT_VMFUNC */ hmR0VmxExitSetPendingXcptUD }; #endif /* HMVMX_USE_FUNCTION_TABLE */ #ifdef VBOX_STRICT static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] = { /* 0 */ "(Not Used)", /* 1 */ "VMCALL executed in VMX root operation.", /* 2 */ "VMCLEAR with invalid physical address.", /* 3 */ "VMCLEAR with VMXON pointer.", /* 4 */ "VMLAUNCH with non-clear VMCS.", /* 5 */ "VMRESUME with non-launched VMCS.", /* 6 */ "VMRESUME after VMXOFF", /* 7 */ "VM entry with invalid control fields.", /* 8 */ "VM entry with invalid host state fields.", /* 9 */ "VMPTRLD with invalid physical address.", /* 10 */ "VMPTRLD with VMXON pointer.", /* 11 */ "VMPTRLD with incorrect revision identifier.", /* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.", /* 13 */ "VMWRITE to read-only VMCS component.", /* 14 */ "(Not Used)", /* 15 */ "VMXON executed in VMX root operation.", /* 16 */ "VM entry with invalid executive-VMCS pointer.", /* 17 */ "VM entry with non-launched executing VMCS.", /* 18 */ "VM entry with executive-VMCS pointer not VMXON pointer.", /* 19 */ "VMCALL with non-clear VMCS.", /* 20 */ "VMCALL with invalid VM-exit control fields.", /* 21 */ "(Not Used)", /* 22 */ "VMCALL with incorrect MSEG revision identifier.", /* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.", /* 24 */ "VMCALL with invalid SMM-monitor features.", /* 25 */ "VM entry with invalid VM-execution control fields in executive VMCS.", /* 26 */ "VM entry with events blocked by MOV SS.", /* 27 */ "(Not Used)", /* 28 */ "Invalid operand to INVEPT/INVVPID." }; #endif /* VBOX_STRICT */ /** * Updates the VM's last error record. If there was a VMX instruction error, * reads the error data from the VMCS and updates VCPU's last error record as * well. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU (can be NULL if @a rc is not * VERR_VMX_UNABLE_TO_START_VM or * VERR_VMX_INVALID_VMCS_FIELD). * @param rc The error code. */ static void hmR0VmxUpdateErrorRecord(PVM pVM, PVMCPU pVCpu, int rc) { AssertPtr(pVM); if ( rc == VERR_VMX_INVALID_VMCS_FIELD || rc == VERR_VMX_UNABLE_TO_START_VM) { AssertPtrReturnVoid(pVCpu); VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError); } pVM->hm.s.lLastError = rc; } /** * Reads the VM-entry interruption-information field from the VMCS into the VMX * transient structure. * * @returns VBox status code. * @param pVmxTransient Pointer to the VMX transient structure. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient) { int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo); AssertRCReturn(rc, rc); return VINF_SUCCESS; } /** * Reads the VM-entry exception error code field from the VMCS into * the VMX transient structure. * * @returns VBox status code. * @param pVmxTransient Pointer to the VMX transient structure. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient) { int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode); AssertRCReturn(rc, rc); return VINF_SUCCESS; } /** * Reads the VM-entry exception error code field from the VMCS into * the VMX transient structure. * * @returns VBox status code. * @param pVmxTransient Pointer to the VMX transient structure. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient) { int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr); AssertRCReturn(rc, rc); return VINF_SUCCESS; } /** * Reads the VM-exit interruption-information field from the VMCS into the VMX * transient structure. * * @returns VBox status code. * @param pVmxTransient Pointer to the VMX transient structure. */ DECLINLINE(int) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient) { if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO)) { int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo); AssertRCReturn(rc, rc); pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO; } return VINF_SUCCESS; } /** * Reads the VM-exit interruption error code from the VMCS into the VMX * transient structure. * * @returns VBox status code. * @param pVmxTransient Pointer to the VMX transient structure. */ DECLINLINE(int) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient) { if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE)) { int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode); AssertRCReturn(rc, rc); pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE; } return VINF_SUCCESS; } /** * Reads the VM-exit instruction length field from the VMCS into the VMX * transient structure. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pVmxTransient Pointer to the VMX transient structure. */ DECLINLINE(int) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient) { if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN)) { int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbInstr); AssertRCReturn(rc, rc); pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN; } return VINF_SUCCESS; } /** * Reads the VM-exit instruction-information field from the VMCS into * the VMX transient structure. * * @returns VBox status code. * @param pVmxTransient Pointer to the VMX transient structure. */ DECLINLINE(int) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient) { if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO)) { int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u); AssertRCReturn(rc, rc); pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO; } return VINF_SUCCESS; } /** * Reads the exit qualification from the VMCS into the VMX transient structure. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU (required for the VMCS cache * case). * @param pVmxTransient Pointer to the VMX transient structure. */ DECLINLINE(int) hmR0VmxReadExitQualificationVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient) { if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION)) { int rc = VMXReadVmcsGstN(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQualification); NOREF(pVCpu); AssertRCReturn(rc, rc); pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION; } return VINF_SUCCESS; } /** * Reads the IDT-vectoring information field from the VMCS into the VMX * transient structure. * * @returns VBox status code. * @param pVmxTransient Pointer to the VMX transient structure. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient) { if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO)) { int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_INFO, &pVmxTransient->uIdtVectoringInfo); AssertRCReturn(rc, rc); pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO; } return VINF_SUCCESS; } /** * Reads the IDT-vectoring error code from the VMCS into the VMX * transient structure. * * @returns VBox status code. * @param pVmxTransient Pointer to the VMX transient structure. */ DECLINLINE(int) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient) { if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE)) { int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode); AssertRCReturn(rc, rc); pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE; } return VINF_SUCCESS; } /** * Enters VMX root mode operation on the current CPU. * * @returns VBox status code. * @param pVM Pointer to the VM (optional, can be NULL, after * a resume). * @param HCPhysCpuPage Physical address of the VMXON region. * @param pvCpuPage Pointer to the VMXON region. */ static int hmR0VmxEnterRootMode(PVM pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage) { AssertReturn(HCPhysCpuPage != 0 && HCPhysCpuPage != NIL_RTHCPHYS, VERR_INVALID_PARAMETER); AssertReturn(pvCpuPage, VERR_INVALID_PARAMETER); Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); if (pVM) { /* Write the VMCS revision dword to the VMXON region. */ *(uint32_t *)pvCpuPage = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo); } /* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */ RTCCUINTREG uEflags = ASMIntDisableFlags(); /* Enable the VMX bit in CR4 if necessary. */ RTCCUINTREG uCr4 = ASMGetCR4(); if (!(uCr4 & X86_CR4_VMXE)) ASMSetCR4(uCr4 | X86_CR4_VMXE); /* Enter VMX root mode. */ int rc = VMXEnable(HCPhysCpuPage); if (RT_FAILURE(rc)) ASMSetCR4(uCr4); /* Restore interrupts. */ ASMSetFlags(uEflags); return rc; } /** * Exits VMX root mode operation on the current CPU. * * @returns VBox status code. */ static int hmR0VmxLeaveRootMode(void) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); /* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */ RTCCUINTREG uEflags = ASMIntDisableFlags(); /* If we're for some reason not in VMX root mode, then don't leave it. */ RTCCUINTREG uHostCR4 = ASMGetCR4(); int rc; if (uHostCR4 & X86_CR4_VMXE) { /* Exit VMX root mode and clear the VMX bit in CR4. */ VMXDisable(); ASMSetCR4(uHostCR4 & ~X86_CR4_VMXE); rc = VINF_SUCCESS; } else rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE; /* Restore interrupts. */ ASMSetFlags(uEflags); return rc; } /** * Allocates and maps one physically contiguous page. The allocated page is * zero'd out. (Used by various VT-x structures). * * @returns IPRT status code. * @param pMemObj Pointer to the ring-0 memory object. * @param ppVirt Where to store the virtual address of the * allocation. * @param pPhys Where to store the physical address of the * allocation. */ DECLINLINE(int) hmR0VmxPageAllocZ(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys) { AssertPtrReturn(pMemObj, VERR_INVALID_PARAMETER); AssertPtrReturn(ppVirt, VERR_INVALID_PARAMETER); AssertPtrReturn(pHCPhys, VERR_INVALID_PARAMETER); int rc = RTR0MemObjAllocCont(pMemObj, PAGE_SIZE, false /* fExecutable */); if (RT_FAILURE(rc)) return rc; *ppVirt = RTR0MemObjAddress(*pMemObj); *pHCPhys = RTR0MemObjGetPagePhysAddr(*pMemObj, 0 /* iPage */); ASMMemZero32(*ppVirt, PAGE_SIZE); return VINF_SUCCESS; } /** * Frees and unmaps an allocated physical page. * * @param pMemObj Pointer to the ring-0 memory object. * @param ppVirt Where to re-initialize the virtual address of * allocation as 0. * @param pHCPhys Where to re-initialize the physical address of the * allocation as 0. */ DECLINLINE(void) hmR0VmxPageFree(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys) { AssertPtr(pMemObj); AssertPtr(ppVirt); AssertPtr(pHCPhys); if (*pMemObj != NIL_RTR0MEMOBJ) { int rc = RTR0MemObjFree(*pMemObj, true /* fFreeMappings */); AssertRC(rc); *pMemObj = NIL_RTR0MEMOBJ; *ppVirt = 0; *pHCPhys = 0; } } /** * Worker function to free VT-x related structures. * * @returns IPRT status code. * @param pVM Pointer to the VM. */ static void hmR0VmxStructsFree(PVM pVM) { for (VMCPUID i = 0; i < pVM->cCpus; i++) { PVMCPU pVCpu = &pVM->aCpus[i]; AssertPtr(pVCpu); hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr); hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr); if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS) hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap, &pVCpu->hm.s.vmx.HCPhysMsrBitmap); hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic, &pVCpu->hm.s.vmx.HCPhysVirtApic); hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs); } hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess, &pVM->hm.s.vmx.HCPhysApicAccess); #ifdef VBOX_WITH_CRASHDUMP_MAGIC hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch); #endif } /** * Worker function to allocate VT-x related VM structures. * * @returns IPRT status code. * @param pVM Pointer to the VM. */ static int hmR0VmxStructsAlloc(PVM pVM) { /* * Initialize members up-front so we can cleanup properly on allocation failure. */ #define VMXLOCAL_INIT_VM_MEMOBJ(a_Name, a_VirtPrefix) \ pVM->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \ pVM->hm.s.vmx.a_VirtPrefix##a_Name = 0; \ pVM->hm.s.vmx.HCPhys##a_Name = 0; #define VMXLOCAL_INIT_VMCPU_MEMOBJ(a_Name, a_VirtPrefix) \ pVCpu->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \ pVCpu->hm.s.vmx.a_VirtPrefix##a_Name = 0; \ pVCpu->hm.s.vmx.HCPhys##a_Name = 0; #ifdef VBOX_WITH_CRASHDUMP_MAGIC VMXLOCAL_INIT_VM_MEMOBJ(Scratch, pv); #endif VMXLOCAL_INIT_VM_MEMOBJ(ApicAccess, pb); AssertCompile(sizeof(VMCPUID) == sizeof(pVM->cCpus)); for (VMCPUID i = 0; i < pVM->cCpus; i++) { PVMCPU pVCpu = &pVM->aCpus[i]; VMXLOCAL_INIT_VMCPU_MEMOBJ(Vmcs, pv); VMXLOCAL_INIT_VMCPU_MEMOBJ(VirtApic, pb); VMXLOCAL_INIT_VMCPU_MEMOBJ(MsrBitmap, pv); VMXLOCAL_INIT_VMCPU_MEMOBJ(GuestMsr, pv); VMXLOCAL_INIT_VMCPU_MEMOBJ(HostMsr, pv); } #undef VMXLOCAL_INIT_VMCPU_MEMOBJ #undef VMXLOCAL_INIT_VM_MEMOBJ /* The VMCS size cannot be more than 4096 bytes. See Intel spec. Appendix A.1 "Basic VMX Information". */ AssertReturnStmt(MSR_IA32_VMX_BASIC_INFO_VMCS_SIZE(pVM->hm.s.vmx.Msrs.u64BasicInfo) <= PAGE_SIZE, (&pVM->aCpus[0])->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE, VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO); /* * Allocate all the VT-x structures. */ int rc = VINF_SUCCESS; #ifdef VBOX_WITH_CRASHDUMP_MAGIC rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch); if (RT_FAILURE(rc)) goto cleanup; strcpy((char *)pVM->hm.s.vmx.pbScratch, "SCRATCH Magic"); *(uint64_t *)(pVM->hm.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef); #endif /* Allocate the APIC-access page for trapping APIC accesses from the guest. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC) { rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess, &pVM->hm.s.vmx.HCPhysApicAccess); if (RT_FAILURE(rc)) goto cleanup; } /* * Initialize per-VCPU VT-x structures. */ for (VMCPUID i = 0; i < pVM->cCpus; i++) { PVMCPU pVCpu = &pVM->aCpus[i]; AssertPtr(pVCpu); /* Allocate the VM control structure (VMCS). */ rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs); if (RT_FAILURE(rc)) goto cleanup; /* Allocate the Virtual-APIC page for transparent TPR accesses. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW) { rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic, &pVCpu->hm.s.vmx.HCPhysVirtApic); if (RT_FAILURE(rc)) goto cleanup; } /* Allocate the MSR-bitmap if supported by the CPU. The MSR-bitmap is for transparent accesses of specific MSRs. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS) { rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap, &pVCpu->hm.s.vmx.HCPhysMsrBitmap); if (RT_FAILURE(rc)) goto cleanup; ASMMemFill32(pVCpu->hm.s.vmx.pvMsrBitmap, PAGE_SIZE, UINT32_C(0xffffffff)); } /* Allocate the VM-entry MSR-load and VM-exit MSR-store page for the guest MSRs. */ rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr); if (RT_FAILURE(rc)) goto cleanup; /* Allocate the VM-exit MSR-load page for the host MSRs. */ rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr); if (RT_FAILURE(rc)) goto cleanup; } return VINF_SUCCESS; cleanup: hmR0VmxStructsFree(pVM); return rc; } /** * Does global VT-x initialization (called during module initialization). * * @returns VBox status code. */ VMMR0DECL(int) VMXR0GlobalInit(void) { #ifdef HMVMX_USE_FUNCTION_TABLE AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_apfnVMExitHandlers)); # ifdef VBOX_STRICT for (unsigned i = 0; i < RT_ELEMENTS(g_apfnVMExitHandlers); i++) Assert(g_apfnVMExitHandlers[i]); # endif #endif return VINF_SUCCESS; } /** * Does global VT-x termination (called during module termination). */ VMMR0DECL(void) VMXR0GlobalTerm() { /* Nothing to do currently. */ } /** * Sets up and activates VT-x on the current CPU. * * @returns VBox status code. * @param pCpu Pointer to the global CPU info struct. * @param pVM Pointer to the VM (can be NULL after a host resume * operation). * @param pvCpuPage Pointer to the VMXON region (can be NULL if @a * fEnabledByHost is true). * @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if * @a fEnabledByHost is true). * @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to * enable VT-x on the host. * @param pvMsrs Opaque pointer to VMXMSRS struct. */ VMMR0DECL(int) VMXR0EnableCpu(PHMGLOBALCPUINFO pCpu, PVM pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost, void *pvMsrs) { AssertReturn(pCpu, VERR_INVALID_PARAMETER); AssertReturn(pvMsrs, VERR_INVALID_PARAMETER); Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); /* Enable VT-x if it's not already enabled by the host. */ if (!fEnabledByHost) { int rc = hmR0VmxEnterRootMode(pVM, HCPhysCpuPage, pvCpuPage); if (RT_FAILURE(rc)) return rc; } /* * Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been using EPTPs) so * we don't retain any stale guest-physical mappings which won't get invalidated when flushing by VPID. */ PVMXMSRS pMsrs = (PVMXMSRS)pvMsrs; if (pMsrs->u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS) { hmR0VmxFlushEpt(NULL /* pVCpu */, VMX_FLUSH_EPT_ALL_CONTEXTS); pCpu->fFlushAsidBeforeUse = false; } else pCpu->fFlushAsidBeforeUse = true; /* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */ ++pCpu->cTlbFlushes; return VINF_SUCCESS; } /** * Deactivates VT-x on the current CPU. * * @returns VBox status code. * @param pCpu Pointer to the global CPU info struct. * @param pvCpuPage Pointer to the VMXON region. * @param HCPhysCpuPage Physical address of the VMXON region. * * @remarks This function should never be called when SUPR0EnableVTx() or * similar was used to enable VT-x on the host. */ VMMR0DECL(int) VMXR0DisableCpu(PHMGLOBALCPUINFO pCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage) { NOREF(pCpu); NOREF(pvCpuPage); NOREF(HCPhysCpuPage); Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); return hmR0VmxLeaveRootMode(); } /** * Sets the permission bits for the specified MSR in the MSR bitmap. * * @param pVCpu Pointer to the VMCPU. * @param uMSR The MSR value. * @param enmRead Whether reading this MSR causes a VM-exit. * @param enmWrite Whether writing this MSR causes a VM-exit. */ static void hmR0VmxSetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, VMXMSREXITREAD enmRead, VMXMSREXITWRITE enmWrite) { int32_t iBit; uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.vmx.pvMsrBitmap; /* * Layout: * 0x000 - 0x3ff - Low MSR read bits * 0x400 - 0x7ff - High MSR read bits * 0x800 - 0xbff - Low MSR write bits * 0xc00 - 0xfff - High MSR write bits */ if (uMsr <= 0x00001FFF) iBit = uMsr; else if ( uMsr >= 0xC0000000 && uMsr <= 0xC0001FFF) { iBit = (uMsr - 0xC0000000); pbMsrBitmap += 0x400; } else { AssertMsgFailed(("hmR0VmxSetMsrPermission: Invalid MSR %#RX32\n", uMsr)); return; } Assert(iBit <= 0x1fff); if (enmRead == VMXMSREXIT_INTERCEPT_READ) ASMBitSet(pbMsrBitmap, iBit); else ASMBitClear(pbMsrBitmap, iBit); if (enmWrite == VMXMSREXIT_INTERCEPT_WRITE) ASMBitSet(pbMsrBitmap + 0x800, iBit); else ASMBitClear(pbMsrBitmap + 0x800, iBit); } #ifdef VBOX_STRICT /** * Gets the permission bits for the specified MSR in the MSR bitmap. * * @returns VBox status code. * @retval VINF_SUCCESS if the specified MSR is found. * @retval VERR_NOT_FOUND if the specified MSR is not found. * @retval VERR_NOT_SUPPORTED if VT-x doesn't allow the MSR. * * @param pVCpu Pointer to the VMCPU. * @param uMsr The MSR. * @param penmRead Where to store the read permissions. * @param penmWrite Where to store the write permissions. */ static int hmR0VmxGetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, PVMXMSREXITREAD penmRead, PVMXMSREXITWRITE penmWrite) { AssertPtrReturn(penmRead, VERR_INVALID_PARAMETER); AssertPtrReturn(penmWrite, VERR_INVALID_PARAMETER); int32_t iBit; uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.vmx.pvMsrBitmap; /* See hmR0VmxSetMsrPermission() for the layout. */ if (uMsr <= 0x00001FFF) iBit = uMsr; else if ( uMsr >= 0xC0000000 && uMsr <= 0xC0001FFF) { iBit = (uMsr - 0xC0000000); pbMsrBitmap += 0x400; } else { AssertMsgFailed(("hmR0VmxGetMsrPermission: Invalid MSR %#RX32\n", uMsr)); return VERR_NOT_SUPPORTED; } Assert(iBit <= 0x1fff); if (ASMBitTest(pbMsrBitmap, iBit)) *penmRead = VMXMSREXIT_INTERCEPT_READ; else *penmRead = VMXMSREXIT_PASSTHRU_READ; if (ASMBitTest(pbMsrBitmap + 0x800, iBit)) *penmWrite = VMXMSREXIT_INTERCEPT_WRITE; else *penmWrite = VMXMSREXIT_PASSTHRU_WRITE; return VINF_SUCCESS; } #endif /* VBOX_STRICT */ /** * Updates the VMCS with the number of effective MSRs in the auto-load/store MSR * area. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param cMsrs The number of MSRs. */ DECLINLINE(int) hmR0VmxSetAutoLoadStoreMsrCount(PVMCPU pVCpu, uint32_t cMsrs) { /* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */ uint32_t const cMaxSupportedMsrs = MSR_IA32_VMX_MISC_MAX_MSR(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc); if (RT_UNLIKELY(cMsrs > cMaxSupportedMsrs)) { LogRel(("CPU auto-load/store MSR count in VMCS exceeded cMsrs=%u Supported=%u.\n", cMsrs, cMaxSupportedMsrs)); pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } /* Update number of guest MSRs to load/store across the world-switch. */ int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs); AssertRCReturn(rc, rc); /* Update number of host MSRs to load after the world-switch. Identical to guest-MSR count as it's always paired. */ rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs); AssertRCReturn(rc, rc); /* Update the VCPU's copy of the MSR count. */ pVCpu->hm.s.vmx.cMsrs = cMsrs; return VINF_SUCCESS; } /** * Adds a new (or updates the value of an existing) guest/host MSR * pair to be swapped during the world-switch as part of the * auto-load/store MSR area in the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param uMsr The MSR. * @param uGuestMsr Value of the guest MSR. * @param fUpdateHostMsr Whether to update the value of the host MSR if * necessary. */ static int hmR0VmxAddAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr, uint64_t uGuestMsrValue, bool fUpdateHostMsr) { PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr; uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs; uint32_t i; for (i = 0; i < cMsrs; i++) { if (pGuestMsr->u32Msr == uMsr) break; pGuestMsr++; } bool fAdded = false; if (i == cMsrs) { ++cMsrs; int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs); AssertRCReturn(rc, rc); /* Now that we're swapping MSRs during the world-switch, allow the guest to read/write them without causing VM-exits. */ if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS) hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); fAdded = true; } /* Update the MSR values in the auto-load/store MSR area. */ pGuestMsr->u32Msr = uMsr; pGuestMsr->u64Value = uGuestMsrValue; /* Create/update the MSR slot in the host MSR area. */ PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr; pHostMsr += i; pHostMsr->u32Msr = uMsr; /* * Update the host MSR only when requested by the called AND when we're * adding it to the auto-load/store area. Otherwise, it would have been * updated by hmR0VmxSaveHostMsrs(). We do this for performance reasons. */ if ( fAdded && fUpdateHostMsr) { pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr); } return VINF_SUCCESS; } /** * Removes a guest/shost MSR pair to be swapped during the world-switch from the * auto-load/store MSR area in the VMCS. * * Does not fail if the MSR in @a uMsr is not found in the auto-load/store MSR * area. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param uMsr The MSR. */ static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr) { PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr; uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs; for (uint32_t i = 0; i < cMsrs; i++) { /* Find the MSR. */ if (pGuestMsr->u32Msr == uMsr) { /* If it's the last MSR, simply reduce the count. */ if (i == cMsrs - 1) { --cMsrs; break; } /* Remove it by swapping the last MSR in place of it, and reducing the count. */ PVMXAUTOMSR pLastGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr; pLastGuestMsr += cMsrs; pGuestMsr->u32Msr = pLastGuestMsr->u32Msr; pGuestMsr->u64Value = pLastGuestMsr->u64Value; PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr; PVMXAUTOMSR pLastHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr; pLastHostMsr += cMsrs; pHostMsr->u32Msr = pLastHostMsr->u32Msr; pHostMsr->u64Value = pLastHostMsr->u64Value; --cMsrs; break; } pGuestMsr++; } /* Update the VMCS if the count changed (meaning the MSR was found). */ if (cMsrs != pVCpu->hm.s.vmx.cMsrs) { int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs); AssertRCReturn(rc, rc); /* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */ if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS) hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE); } return VINF_SUCCESS; } /** * Checks if the specified guest MSR is part of the auto-load/store area in * the VMCS. * * @returns true if found, false otherwise. * @param pVCpu Pointer to the VMCPU. * @param uMsr The MSR to find. */ static bool hmR0VmxIsAutoLoadStoreGuestMsr(PVMCPU pVCpu, uint32_t uMsr) { PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr; uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs; for (uint32_t i = 0; i < cMsrs; i++, pGuestMsr++) { if (pGuestMsr->u32Msr == uMsr) return true; } return false; } /** * Updates the value of all host MSRs in the auto-load/store area in the VMCS. * * @param pVCpu Pointer to the VMCPU. * * @remarks No-long-jump zone!!! */ static void hmR0VmxUpdateAutoLoadStoreHostMsrs(PVMCPU pVCpu) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr; PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr; uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs; for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++) { AssertReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr); pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr); } pVCpu->hm.s.vmx.fUpdatedHostMsrs = true; } #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) /** * Saves a set of host MSRs to allow read/write passthru access to the guest and * perform lazy restoration of the host MSRs while leaving VT-x. * * @param pVCpu Pointer to the VMCPU. * * @remarks No-long-jump zone!!! */ static void hmR0VmxLazySaveHostMsrs(PVMCPU pVCpu) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); #define VMXLOCAL_SAVE_HOST_MSR(uMsr, a_HostMsrField, RestoreFlag) \ do { \ if (!(pVCpu->hm.s.vmx.fRestoreHostMsrs & (RestoreFlag))) \ { \ pVCpu->hm.s.vmx.u64Host##a_HostMsrField = ASMRdMsr(uMsr); \ Log4(("hmR0VmxLazySaveHostMsrs: uMsr=%#RX32 HostValue=%#RX64\n", (uMsr), pVCpu->hm.s.vmx.u64Host##a_HostMsrField)); \ } \ } while (0) /* * Note: If you're adding MSRs here, make sure to update the MSR-bitmap permissions in hmR0VmxSetupProcCtls(). */ VMXLOCAL_SAVE_HOST_MSR(MSR_K8_LSTAR, LStarMsr, VMX_RESTORE_HOST_MSR_LSTAR); VMXLOCAL_SAVE_HOST_MSR(MSR_K6_STAR, StarMsr, VMX_RESTORE_HOST_MSR_STAR); VMXLOCAL_SAVE_HOST_MSR(MSR_K8_SF_MASK, SFMaskMsr, VMX_RESTORE_HOST_MSR_SFMASK); VMXLOCAL_SAVE_HOST_MSR(MSR_K8_KERNEL_GS_BASE, KernelGSBaseMsr, VMX_RESTORE_HOST_MSR_KERNELGSBASE); #undef VMXLOCAL_SAVE_HOST_MSR } /** * Checks whether the MSR belongs to the set of guest MSRs that we restore * lazily while leaving VT-x. * * @returns true if it does, false otherwise. * @param pVCpu Pointer to the VMCPU. * @param uMsr The MSR to check. */ static bool hmR0VmxIsLazyGuestMsr(PVMCPU pVCpu, uint32_t uMsr) { NOREF(pVCpu); switch (uMsr) { case MSR_K8_LSTAR: case MSR_K6_STAR: case MSR_K8_SF_MASK: case MSR_K8_KERNEL_GS_BASE: return true; } return false; } /** * Saves a set of guests MSRs back into the guest-CPU context. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static void hmR0VmxLazySaveGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); #define VMXLOCAL_SAVE_GUEST_MSR(uMsr, a_GuestMsrField, RestoreFlag) \ do { \ if (pVCpu->hm.s.vmx.fRestoreHostMsrs & (RestoreFlag)) \ { \ pMixedCtx->msr##a_GuestMsrField = ASMRdMsr(uMsr); \ Log4(("hmR0VmxLazySaveGuestMsrs: uMsr=%#RX32 GuestValue=%#RX64\n", (uMsr), pMixedCtx->msr##a_GuestMsrField)); \ } \ } while (0) VMXLOCAL_SAVE_GUEST_MSR(MSR_K8_LSTAR, LSTAR, VMX_RESTORE_HOST_MSR_LSTAR); VMXLOCAL_SAVE_GUEST_MSR(MSR_K6_STAR, STAR, VMX_RESTORE_HOST_MSR_STAR); VMXLOCAL_SAVE_GUEST_MSR(MSR_K8_SF_MASK, SFMASK, VMX_RESTORE_HOST_MSR_SFMASK); VMXLOCAL_SAVE_GUEST_MSR(MSR_K8_KERNEL_GS_BASE, KERNELGSBASE, VMX_RESTORE_HOST_MSR_KERNELGSBASE); #undef VMXLOCAL_SAVE_GUEST_MSR } /** * Loads a set of guests MSRs to allow read/passthru to the guest. * * The name of this function is slightly confusing. This function does NOT * postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a * common prefix for functions dealing with "lazy restoration" of the shared * MSRs. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static void hmR0VmxLazyLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); #define VMXLOCAL_LOAD_GUEST_MSR(uMsr, a_GuestMsrField, a_HostMsrField, RestoreFlag) \ do { \ if ( (pVCpu->hm.s.vmx.fRestoreHostMsrs & (RestoreFlag)) \ || pMixedCtx->msr##a_GuestMsrField != pVCpu->hm.s.vmx.u64Host##a_HostMsrField) \ { \ ASMWrMsr((uMsr), pMixedCtx->msr##a_GuestMsrField); \ } \ pVCpu->hm.s.vmx.fRestoreHostMsrs |= (RestoreFlag); \ Log4(("Load: MSRSWAP uMsr=%#RX32 GuestValue=%#RX64\n", (uMsr), pMixedCtx->msr##a_GuestMsrField)); \ } while (0) VMXLOCAL_LOAD_GUEST_MSR(MSR_K8_LSTAR, LSTAR, LStarMsr, VMX_RESTORE_HOST_MSR_LSTAR); VMXLOCAL_LOAD_GUEST_MSR(MSR_K6_STAR, STAR, StarMsr, VMX_RESTORE_HOST_MSR_STAR); VMXLOCAL_LOAD_GUEST_MSR(MSR_K8_SF_MASK, SFMASK, SFMaskMsr, VMX_RESTORE_HOST_MSR_SFMASK); VMXLOCAL_LOAD_GUEST_MSR(MSR_K8_KERNEL_GS_BASE, KERNELGSBASE, KernelGSBaseMsr, VMX_RESTORE_HOST_MSR_KERNELGSBASE); #undef VMXLOCAL_LOAD_GUEST_MSR } /** * Performs lazy restoration of the set of host MSRs if they were previously * loaded with guest MSR values. * * @param pVCpu Pointer to the VMCPU. * * @remarks No-long-jump zone!!! * @remarks The guest MSRs should have been saved back into the guest-CPU * context by hmR0VmxSaveGuestLazyMsrs()!!! */ static void hmR0VmxLazyRestoreHostMsrs(PVMCPU pVCpu) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); #define VMXLOCAL_RESTORE_HOST_MSR(uMsr, a_HostMsrField, RestoreFlag) \ do { \ if (pVCpu->hm.s.vmx.fRestoreHostMsrs & (RestoreFlag)) \ { \ ASMWrMsr((uMsr), pVCpu->hm.s.vmx.u64Host##a_HostMsrField); \ pVCpu->hm.s.vmx.fRestoreHostMsrs &= ~(RestoreFlag); \ Log4(("hmR0VmxLazyRestoreHostMsrs: uMsr=%#RX32 HostValue=%#RX64\n", (uMsr), \ pVCpu->hm.s.vmx.u64Host##a_HostMsrField)); \ } \ } while (0) VMXLOCAL_RESTORE_HOST_MSR(MSR_K8_LSTAR, LStarMsr, VMX_RESTORE_HOST_MSR_LSTAR); VMXLOCAL_RESTORE_HOST_MSR(MSR_K6_STAR, StarMsr, VMX_RESTORE_HOST_MSR_STAR); VMXLOCAL_RESTORE_HOST_MSR(MSR_K8_SF_MASK, SFMaskMsr, VMX_RESTORE_HOST_MSR_SFMASK); VMXLOCAL_RESTORE_HOST_MSR(MSR_K8_KERNEL_GS_BASE, KernelGSBaseMsr, VMX_RESTORE_HOST_MSR_KERNELGSBASE); #undef VMXLOCAL_RESTORE_HOST_MSR } #endif /* HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) */ #ifdef VBOX_STRICT /** * Verifies whether the guest/host MSR pairs in the auto-load/store area in the * VMCS are correct. * * @param pVCpu Pointer to the VMCPU. */ static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPU pVCpu) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); /* Verify MSR counts in the VMCS are what we think it should be. */ uint32_t cMsrs; int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc); Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cMsrs); AssertRC(rc); Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc); Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs); PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr; PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr; for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++) { /* Verify that the MSRs are paired properly and that the host MSR has the correct value. */ AssertMsgReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr, ("HostMsr=%#RX32 GuestMsr=%#RX32\n", pHostMsr->u32Msr, pGuestMsr->u32Msr)); uint64_t u64Msr = ASMRdMsr(pHostMsr->u32Msr); AssertMsgReturnVoid(pHostMsr->u64Value == u64Msr, ("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64\n", pHostMsr->u32Msr, pHostMsr->u64Value, u64Msr)); /* Verify that the permissions are as expected in the MSR bitmap. */ if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS) { VMXMSREXITREAD enmRead; VMXMSREXITWRITE enmWrite; rc = hmR0VmxGetMsrPermission(pVCpu, pGuestMsr->u32Msr, &enmRead, &enmWrite); AssertMsgReturnVoid(rc == VINF_SUCCESS, ("hmR0VmxGetMsrPermission! failed. rc=%Rrc\n", rc)); AssertMsgReturnVoid(enmRead == VMXMSREXIT_PASSTHRU_READ, ("u32Msr=%#RX32 No passthru read permission!\n", pGuestMsr->u32Msr)); AssertMsgReturnVoid(enmWrite == VMXMSREXIT_PASSTHRU_WRITE, ("u32Msr=%#RX32 No passthru write permission!\n", pGuestMsr->u32Msr)); } } } # endif /* VBOX_STRICT */ /** * Flushes the TLB using EPT. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU (can be NULL depending on @a * enmFlush). * @param enmFlush Type of flush. * * @remarks Caller is responsible for making sure this function is called only * when NestedPaging is supported and providing @a enmFlush that is * supported by the CPU. * @remarks Can be called with interrupts disabled. */ static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMX_FLUSH_EPT enmFlush) { uint64_t au64Descriptor[2]; if (enmFlush == VMX_FLUSH_EPT_ALL_CONTEXTS) au64Descriptor[0] = 0; else { Assert(pVCpu); au64Descriptor[0] = pVCpu->hm.s.vmx.HCPhysEPTP; } au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */ int rc = VMXR0InvEPT(enmFlush, &au64Descriptor[0]); AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %RGv failed with %Rrc\n", enmFlush, pVCpu ? pVCpu->hm.s.vmx.HCPhysEPTP : 0, rc)); if ( RT_SUCCESS(rc) && pVCpu) { STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging); } } /** * Flushes the TLB using VPID. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU (can be NULL depending on @a * enmFlush). * @param enmFlush Type of flush. * @param GCPtr Virtual address of the page to flush (can be 0 depending * on @a enmFlush). * * @remarks Can be called with interrupts disabled. */ static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMX_FLUSH_VPID enmFlush, RTGCPTR GCPtr) { NOREF(pVM); AssertPtr(pVM); Assert(pVM->hm.s.vmx.fVpid); uint64_t au64Descriptor[2]; if (enmFlush == VMX_FLUSH_VPID_ALL_CONTEXTS) { au64Descriptor[0] = 0; au64Descriptor[1] = 0; } else { AssertPtr(pVCpu); AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid)); AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid)); au64Descriptor[0] = pVCpu->hm.s.uCurrentAsid; au64Descriptor[1] = GCPtr; } int rc = VMXR0InvVPID(enmFlush, &au64Descriptor[0]); NOREF(rc); AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvVPID %#x %u %RGv failed with %d\n", enmFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc)); if ( RT_SUCCESS(rc) && pVCpu) { STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid); } } /** * Invalidates a guest page by guest virtual address. Only relevant for * EPT/VPID, otherwise there is nothing really to invalidate. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param GCVirt Guest virtual address of the page to invalidate. */ VMMR0DECL(int) VMXR0InvalidatePage(PVM pVM, PVMCPU pVCpu, RTGCPTR GCVirt) { AssertPtr(pVM); AssertPtr(pVCpu); LogFlowFunc(("pVM=%p pVCpu=%p GCVirt=%RGv\n", pVM, pVCpu, GCVirt)); bool fFlushPending = VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_FLUSH); if (!fFlushPending) { /* * We must invalidate the guest TLB entry in either case, we cannot ignore it even for the EPT case * See @bugref{6043} and @bugref{6177}. * * Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*() as this * function maybe called in a loop with individual addresses. */ if (pVM->hm.s.vmx.fVpid) { if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR) { hmR0VmxFlushVpid(pVM, pVCpu, VMX_FLUSH_VPID_INDIV_ADDR, GCVirt); STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt); } else VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH); } else if (pVM->hm.s.fNestedPaging) VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH); } return VINF_SUCCESS; } /** * Invalidates a guest page by physical address. Only relevant for EPT/VPID, * otherwise there is nothing really to invalidate. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param GCPhys Guest physical address of the page to invalidate. */ VMMR0DECL(int) VMXR0InvalidatePhysPage(PVM pVM, PVMCPU pVCpu, RTGCPHYS GCPhys) { NOREF(pVM); NOREF(GCPhys); LogFlowFunc(("%RGp\n", GCPhys)); /* * We cannot flush a page by guest-physical address. invvpid takes only a linear address while invept only flushes * by EPT not individual addresses. We update the force flag here and flush before the next VM-entry in hmR0VmxFlushTLB*(). * This function might be called in a loop. This should cause a flush-by-EPT if EPT is in use. See @bugref{6568}. */ VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH); STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgPhys); return VINF_SUCCESS; } /** * Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the * case where neither EPT nor VPID is supported by the CPU. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCpu Pointer to the global HM struct. * * @remarks Called with interrupts disabled. */ static void hmR0VmxFlushTaggedTlbNone(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu) { AssertPtr(pVCpu); AssertPtr(pCpu); NOREF(pVM); VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH); /** @todo TLB shootdown is currently not used. See hmQueueInvlPage(). */ #if 0 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN); pVCpu->hm.s.TlbShootdown.cPages = 0; #endif pVCpu->hm.s.idLastCpu = pCpu->idCpu; pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes; pVCpu->hm.s.fForceTLBFlush = false; return; } /** * Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCpu Pointer to the global HM CPU struct. * @remarks All references to "ASID" in this function pertains to "VPID" in * Intel's nomenclature. The reason is, to avoid confusion in compare * statements since the host-CPU copies are named "ASID". * * @remarks Called with interrupts disabled. */ static void hmR0VmxFlushTaggedTlbBoth(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu) { #ifdef VBOX_WITH_STATISTICS bool fTlbFlushed = false; # define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0) # define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \ if (!fTlbFlushed) \ STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \ } while (0) #else # define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0) # define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0) #endif AssertPtr(pVM); AssertPtr(pCpu); AssertPtr(pVCpu); AssertMsg(pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled." "fNestedPaging=%RTbool fVpid=%RTbool", pVM->hm.s.fNestedPaging, pVM->hm.s.vmx.fVpid)); /* * Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last. * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB * or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore. */ if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes) { ++pCpu->uCurrentAsid; if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid) { pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */ pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */ pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */ } pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid; pVCpu->hm.s.idLastCpu = pCpu->idCpu; pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes; /* * Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also * invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}. */ hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt); STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch); HMVMX_SET_TAGGED_TLB_FLUSHED(); VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH); /* Already flushed-by-EPT, skip doing it again below. */ } /* Check for explicit TLB shootdowns. */ if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) { /* * Changes to the EPT paging structure by VMM requires flushing by EPT as the CPU creates * guest-physical (only EPT-tagged) mappings while traversing the EPT tables when EPT is in use. * Flushing by VPID will only flush linear (only VPID-tagged) and combined (EPT+VPID tagged) mappings * but not guest-physical mappings. * See Intel spec. 28.3.2 "Creating and Using Cached Translation Information". See @bugref{6568}. */ hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt); STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb); HMVMX_SET_TAGGED_TLB_FLUSHED(); } /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage() * where it is commented out. Support individual entry flushing * someday. */ #if 0 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN)) { STAM_COUNTER_INC(&pVCpu->hm.s.StatTlbShootdown); /* * Flush individual guest entries using VPID from the TLB or as little as possible with EPT * as supported by the CPU. */ if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR) { for (uint32_t i = 0; i < pVCpu->hm.s.TlbShootdown.cPages; i++) hmR0VmxFlushVpid(pVM, pVCpu, VMX_FLUSH_VPID_INDIV_ADDR, pVCpu->hm.s.TlbShootdown.aPages[i]); } else hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt); HMVMX_SET_TAGGED_TLB_FLUSHED(); VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN); pVCpu->hm.s.TlbShootdown.cPages = 0; } #endif pVCpu->hm.s.fForceTLBFlush = false; HMVMX_UPDATE_FLUSH_SKIPPED_STAT(); Assert(pVCpu->hm.s.idLastCpu == pCpu->idCpu); Assert(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes); AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes, ("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes)); AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid, ("cpu%d uCurrentAsid = %u\n", pCpu->idCpu, pCpu->uCurrentAsid)); AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid, ("cpu%d VM uCurrentAsid = %u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid)); /* Update VMCS with the VPID. */ int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, pVCpu->hm.s.uCurrentAsid); AssertRC(rc); #undef HMVMX_SET_TAGGED_TLB_FLUSHED } /** * Flushes the tagged-TLB entries for EPT CPUs as necessary. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCpu Pointer to the global HM CPU struct. * * @remarks Called with interrupts disabled. */ static void hmR0VmxFlushTaggedTlbEpt(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu) { AssertPtr(pVM); AssertPtr(pVCpu); AssertPtr(pCpu); AssertMsg(pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with NestedPaging disabled.")); AssertMsg(!pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID enabled.")); /* * Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last. * A change in the TLB flush count implies the host CPU is online after a suspend/resume. */ if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes) { pVCpu->hm.s.fForceTLBFlush = true; STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch); } /* Check for explicit TLB shootdown flushes. */ if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) { pVCpu->hm.s.fForceTLBFlush = true; STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb); } pVCpu->hm.s.idLastCpu = pCpu->idCpu; pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes; if (pVCpu->hm.s.fForceTLBFlush) { hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt); pVCpu->hm.s.fForceTLBFlush = false; } /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage() * where it is commented out. Support individual entry flushing * someday. */ #if 0 else { if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN)) { /* We cannot flush individual entries without VPID support. Flush using EPT. */ STAM_COUNTER_INC(&pVCpu->hm.s.StatTlbShootdown); hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt); } else STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); pVCpu->hm.s.TlbShootdown.cPages = 0; VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN); } #endif } /** * Flushes the tagged-TLB entries for VPID CPUs as necessary. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCpu Pointer to the global HM CPU struct. * * @remarks Called with interrupts disabled. */ static void hmR0VmxFlushTaggedTlbVpid(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu) { AssertPtr(pVM); AssertPtr(pVCpu); AssertPtr(pCpu); AssertMsg(pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked with VPID disabled.")); AssertMsg(!pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging enabled")); /* * Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last. * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB * or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore. */ if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes) { pVCpu->hm.s.fForceTLBFlush = true; STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch); } /* Check for explicit TLB shootdown flushes. */ if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) { /* * If we ever support VPID flush combinations other than ALL or SINGLE-context (see hmR0VmxSetupTaggedTlb()) * we would need to explicitly flush in this case (add an fExplicitFlush = true here and change the * pCpu->fFlushAsidBeforeUse check below to include fExplicitFlush's too) - an obscure corner case. */ pVCpu->hm.s.fForceTLBFlush = true; STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb); } pVCpu->hm.s.idLastCpu = pCpu->idCpu; if (pVCpu->hm.s.fForceTLBFlush) { ++pCpu->uCurrentAsid; if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid) { pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */ pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */ pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */ } pVCpu->hm.s.fForceTLBFlush = false; pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes; pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid; if (pCpu->fFlushAsidBeforeUse) hmR0VmxFlushVpid(pVM, pVCpu, pVM->hm.s.vmx.enmFlushVpid, 0 /* GCPtr */); } /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage() * where it is commented out. Support individual entry flushing * someday. */ #if 0 else { AssertMsg(pVCpu->hm.s.uCurrentAsid && pCpu->uCurrentAsid, ("hm->uCurrentAsid=%lu hm->cTlbFlushes=%lu cpu->uCurrentAsid=%lu cpu->cTlbFlushes=%lu\n", pVCpu->hm.s.uCurrentAsid, pVCpu->hm.s.cTlbFlushes, pCpu->uCurrentAsid, pCpu->cTlbFlushes)); if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN)) { /* Flush individual guest entries using VPID or as little as possible with EPT as supported by the CPU. */ if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR) { for (uint32_t i = 0; i < pVCpu->hm.s.TlbShootdown.cPages; i++) hmR0VmxFlushVpid(pVM, pVCpu, VMX_FLUSH_VPID_INDIV_ADDR, pVCpu->hm.s.TlbShootdown.aPages[i]); } else hmR0VmxFlushVpid(pVM, pVCpu, pVM->hm.s.vmx.enmFlushVpid, 0 /* GCPtr */); pVCpu->hm.s.TlbShootdown.cPages = 0; VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN); } else STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); } #endif AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes, ("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes)); AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid, ("cpu%d uCurrentAsid = %u\n", pCpu->idCpu, pCpu->uCurrentAsid)); AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid, ("cpu%d VM uCurrentAsid = %u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid)); int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, pVCpu->hm.s.uCurrentAsid); AssertRC(rc); } /** * Flushes the guest TLB entry based on CPU capabilities. * * @param pVCpu Pointer to the VMCPU. * @param pCpu Pointer to the global HM CPU struct. */ DECLINLINE(void) hmR0VmxFlushTaggedTlb(PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu) { #ifdef HMVMX_ALWAYS_FLUSH_TLB VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH); #endif PVM pVM = pVCpu->CTX_SUFF(pVM); switch (pVM->hm.s.vmx.uFlushTaggedTlb) { case HMVMX_FLUSH_TAGGED_TLB_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pVM, pVCpu, pCpu); break; case HMVMX_FLUSH_TAGGED_TLB_EPT: hmR0VmxFlushTaggedTlbEpt(pVM, pVCpu, pCpu); break; case HMVMX_FLUSH_TAGGED_TLB_VPID: hmR0VmxFlushTaggedTlbVpid(pVM, pVCpu, pCpu); break; case HMVMX_FLUSH_TAGGED_TLB_NONE: hmR0VmxFlushTaggedTlbNone(pVM, pVCpu, pCpu); break; default: AssertMsgFailed(("Invalid flush-tag function identifier\n")); break; } /* VMCPU_FF_TLB_SHOOTDOWN is unused. */ Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN)); /* Don't assert that VMCPU_FF_TLB_FLUSH should no longer pending. It can be set by other EMTs. */ } /** * Sets up the appropriate tagged TLB-flush level and handler for flushing guest * TLB entries from the host TLB before VM-entry. * * @returns VBox status code. * @param pVM Pointer to the VM. */ static int hmR0VmxSetupTaggedTlb(PVM pVM) { /* * Determine optimal flush type for Nested Paging. * We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup unrestricted * guest execution (see hmR3InitFinalizeR0()). */ if (pVM->hm.s.fNestedPaging) { if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT) { if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT) pVM->hm.s.vmx.enmFlushEpt = VMX_FLUSH_EPT_SINGLE_CONTEXT; else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS) pVM->hm.s.vmx.enmFlushEpt = VMX_FLUSH_EPT_ALL_CONTEXTS; else { /* Shouldn't happen. EPT is supported but no suitable flush-types supported. */ pVM->hm.s.vmx.enmFlushEpt = VMX_FLUSH_EPT_NOT_SUPPORTED; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } /* Make sure the write-back cacheable memory type for EPT is supported. */ if (!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB)) { LogRel(("hmR0VmxSetupTaggedTlb: Unsupported EPTP memory type %#x.\n", pVM->hm.s.vmx.Msrs.u64EptVpidCaps)); pVM->hm.s.vmx.enmFlushEpt = VMX_FLUSH_EPT_NOT_SUPPORTED; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } } else { /* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */ pVM->hm.s.vmx.enmFlushEpt = VMX_FLUSH_EPT_NOT_SUPPORTED; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } } /* * Determine optimal flush type for VPID. */ if (pVM->hm.s.vmx.fVpid) { if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID) { if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT) pVM->hm.s.vmx.enmFlushVpid = VMX_FLUSH_VPID_SINGLE_CONTEXT; else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS) pVM->hm.s.vmx.enmFlushVpid = VMX_FLUSH_VPID_ALL_CONTEXTS; else { /* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */ if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR) LogRel(("hmR0VmxSetupTaggedTlb: Only INDIV_ADDR supported. Ignoring VPID.\n")); if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS) LogRel(("hmR0VmxSetupTaggedTlb: Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n")); pVM->hm.s.vmx.enmFlushVpid = VMX_FLUSH_VPID_NOT_SUPPORTED; pVM->hm.s.vmx.fVpid = false; } } else { /* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */ Log4(("hmR0VmxSetupTaggedTlb: VPID supported without INVEPT support. Ignoring VPID.\n")); pVM->hm.s.vmx.enmFlushVpid = VMX_FLUSH_VPID_NOT_SUPPORTED; pVM->hm.s.vmx.fVpid = false; } } /* * Setup the handler for flushing tagged-TLBs. */ if (pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid) pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT_VPID; else if (pVM->hm.s.fNestedPaging) pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT; else if (pVM->hm.s.vmx.fVpid) pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_VPID; else pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_NONE; return VINF_SUCCESS; } /** * Sets up pin-based VM-execution controls in the VMCS. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. */ static int hmR0VmxSetupPinCtls(PVM pVM, PVMCPU pVCpu) { AssertPtr(pVM); AssertPtr(pVCpu); uint32_t val = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0; /* Bits set here must always be set. */ uint32_t zap = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1; /* Bits cleared here must always be cleared. */ val |= VMX_VMCS_CTRL_PIN_EXEC_EXT_INT_EXIT /* External interrupts causes a VM-exits. */ | VMX_VMCS_CTRL_PIN_EXEC_NMI_EXIT; /* Non-maskable interrupts causes a VM-exit. */ Assert(!(val & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)); /* Enable the VMX preemption timer. */ if (pVM->hm.s.vmx.fUsePreemptTimer) { Assert(pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER); val |= VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER; } if ((val & zap) != val) { LogRel(("hmR0VmxSetupPinCtls: invalid pin-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n", pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0, val, zap)); pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, val); AssertRCReturn(rc, rc); /* Update VCPU with the currently set pin-based VM-execution controls. */ pVCpu->hm.s.vmx.u32PinCtls = val; return rc; } /** * Sets up processor-based VM-execution controls in the VMCS. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVMCPU Pointer to the VMCPU. */ static int hmR0VmxSetupProcCtls(PVM pVM, PVMCPU pVCpu) { AssertPtr(pVM); AssertPtr(pVCpu); int rc = VERR_INTERNAL_ERROR_5; uint32_t val = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0; /* Bits set here must be set in the VMCS. */ uint32_t zap = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */ val |= VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT /* HLT causes a VM-exit. */ | VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING /* Use TSC-offsetting. */ | VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */ | VMX_VMCS_CTRL_PROC_EXEC_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */ | VMX_VMCS_CTRL_PROC_EXEC_RDPMC_EXIT /* RDPMC causes a VM-exit. */ | VMX_VMCS_CTRL_PROC_EXEC_MONITOR_EXIT /* MONITOR causes a VM-exit. */ | VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT; /* MWAIT causes a VM-exit. */ /* We toggle VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */ if ( !(pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT) || (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT)) { LogRel(("hmR0VmxSetupProcCtls: unsupported VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT combo!")); pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } /* Without Nested Paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */ if (!pVM->hm.s.fNestedPaging) { Assert(!pVM->hm.s.vmx.fUnrestrictedGuest); /* Paranoia. */ val |= VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT | VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT; } /* Use TPR shadowing if supported by the CPU. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW) { Assert(pVCpu->hm.s.vmx.HCPhysVirtApic); Assert(!(pVCpu->hm.s.vmx.HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */ rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, 0); rc |= VMXWriteVmcs64(VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL, pVCpu->hm.s.vmx.HCPhysVirtApic); AssertRCReturn(rc, rc); val |= VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */ /* CR8 writes causes a VM-exit based on TPR threshold. */ Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT)); Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT)); } else { /* * Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is invalid on 32-bit Intel CPUs. * Set this control only for 64-bit guests. */ if (pVM->hm.s.fAllow64BitGuests) { val |= VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT /* CR8 reads causes a VM-exit. */ | VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT; /* CR8 writes causes a VM-exit. */ } } /* Use MSR-bitmaps if supported by the CPU. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS) { val |= VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS; Assert(pVCpu->hm.s.vmx.HCPhysMsrBitmap); Assert(!(pVCpu->hm.s.vmx.HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */ rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, pVCpu->hm.s.vmx.HCPhysMsrBitmap); AssertRCReturn(rc, rc); /* * The guest can access the following MSRs (read, write) without causing VM-exits; they are loaded/stored * automatically as dedicated fields in the VMCS. */ hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_CS, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_ESP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_EIP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); hmR0VmxSetMsrPermission(pVCpu, MSR_K8_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); hmR0VmxSetMsrPermission(pVCpu, MSR_K8_FS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) /* * Set passthru permissions for the following MSRs (mandatory for VT-x) required for 64-bit guests. */ if ( HMVMX_IS_64BIT_HOST_MODE() && pVM->hm.s.fAllow64BitGuests) { hmR0VmxSetMsrPermission(pVCpu, MSR_K8_LSTAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); hmR0VmxSetMsrPermission(pVCpu, MSR_K6_STAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); hmR0VmxSetMsrPermission(pVCpu, MSR_K8_SF_MASK, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); hmR0VmxSetMsrPermission(pVCpu, MSR_K8_KERNEL_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE); } #endif } /* Use the secondary processor-based VM-execution controls if supported by the CPU. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL) val |= VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL; if ((val & zap) != val) { LogRel(("hmR0VmxSetupProcCtls: invalid processor-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n", pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0, val, zap)); pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, val); AssertRCReturn(rc, rc); /* Update VCPU with the currently set processor-based VM-execution controls. */ pVCpu->hm.s.vmx.u32ProcCtls = val; /* * Secondary processor-based VM-execution controls. */ if (RT_LIKELY(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)) { val = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0; /* Bits set here must be set in the VMCS. */ zap = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT) val |= VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT; /* WBINVD causes a VM-exit. */ if (pVM->hm.s.fNestedPaging) val |= VMX_VMCS_CTRL_PROC_EXEC2_EPT; /* Enable EPT. */ else { /* * Without Nested Paging, INVPCID should cause a VM-exit. Enabling this bit causes the CPU to refer to * VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT when INVPCID is executed by the guest. * See Intel spec. 25.4 "Changes to instruction behaviour in VMX non-root operation". */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_INVPCID) val |= VMX_VMCS_CTRL_PROC_EXEC2_INVPCID; } if (pVM->hm.s.vmx.fVpid) val |= VMX_VMCS_CTRL_PROC_EXEC2_VPID; /* Enable VPID. */ if (pVM->hm.s.vmx.fUnrestrictedGuest) val |= VMX_VMCS_CTRL_PROC_EXEC2_UNRESTRICTED_GUEST; /* Enable Unrestricted Execution. */ /* Enable Virtual-APIC page accesses if supported by the CPU. This is essentially where the TPR shadow resides. */ /** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be * done dynamically. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC) { Assert(pVM->hm.s.vmx.HCPhysApicAccess); Assert(!(pVM->hm.s.vmx.HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */ val |= VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC; /* Virtualize APIC accesses. */ rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, pVM->hm.s.vmx.HCPhysApicAccess); AssertRCReturn(rc, rc); } if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP) val |= VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP; /* Enable RDTSCP support. */ if ((val & zap) != val) { LogRel(("hmR0VmxSetupProcCtls: invalid secondary processor-based VM-execution controls combo! " "cpu=%#RX64 val=%#RX64 zap=%#RX64\n", pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0, val, zap)); return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, val); AssertRCReturn(rc, rc); /* Update VCPU with the currently set secondary processor-based VM-execution controls. */ pVCpu->hm.s.vmx.u32ProcCtls2 = val; } else if (RT_UNLIKELY(pVM->hm.s.vmx.fUnrestrictedGuest)) { LogRel(("hmR0VmxSetupProcCtls: Unrestricted Guest set as true when secondary processor-based VM-execution controls not " "available\n")); return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } return VINF_SUCCESS; } /** * Sets up miscellaneous (everything other than Pin & Processor-based * VM-execution) control fields in the VMCS. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. */ static int hmR0VmxSetupMiscCtls(PVM pVM, PVMCPU pVCpu) { NOREF(pVM); AssertPtr(pVM); AssertPtr(pVCpu); int rc = VERR_GENERAL_FAILURE; /* All fields are zero-initialized during allocation; but don't remove the commented block below. */ #if 0 /* All CR3 accesses cause VM-exits. Later we optimize CR3 accesses (see hmR0VmxLoadGuestCR3AndCR4())*/ rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, 0); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, 0); AssertRCReturn(rc, rc); /* * Set MASK & MATCH to 0. VMX checks if GuestPFErrCode & MASK == MATCH. If equal (in our case it always is) * and if the X86_XCPT_PF bit in the exception bitmap is set it causes a VM-exit, if clear doesn't cause an exit. * We thus use the exception bitmap to control it rather than use both. */ rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, 0); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, 0); AssertRCReturn(rc, rc); /** @todo Explore possibility of using IO-bitmaps. */ /* All IO & IOIO instructions cause VM-exits. */ rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_A_FULL, 0); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_B_FULL, 0); AssertRCReturn(rc, rc); /* Initialize the MSR-bitmap area. */ rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, 0); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, 0); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, 0); AssertRCReturn(rc, rc); #endif /* Setup MSR auto-load/store area. */ Assert(pVCpu->hm.s.vmx.HCPhysGuestMsr); Assert(!(pVCpu->hm.s.vmx.HCPhysGuestMsr & 0xf)); /* Lower 4 bits MBZ. */ rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr); AssertRCReturn(rc, rc); Assert(pVCpu->hm.s.vmx.HCPhysHostMsr); Assert(!(pVCpu->hm.s.vmx.HCPhysHostMsr & 0xf)); /* Lower 4 bits MBZ. */ rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysHostMsr); AssertRCReturn(rc, rc); /* Set VMCS link pointer. Reserved for future use, must be -1. Intel spec. 24.4 "Guest-State Area". */ rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, UINT64_C(0xffffffffffffffff)); AssertRCReturn(rc, rc); /* All fields are zero-initialized during allocation; but don't remove the commented block below. */ #if 0 /* Setup debug controls */ rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, 0); /** @todo We don't support IA32_DEBUGCTL MSR. Should we? */ AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0); AssertRCReturn(rc, rc); #endif return rc; } /** * Sets up the initial exception bitmap in the VMCS based on static conditions * (i.e. conditions that cannot ever change after starting the VM). * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. */ static int hmR0VmxInitXcptBitmap(PVM pVM, PVMCPU pVCpu) { AssertPtr(pVM); AssertPtr(pVCpu); LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu)); uint32_t u32XcptBitmap = 0; /* Without Nested Paging, #PF must cause a VM-exit so we can sync our shadow page tables. */ if (!pVM->hm.s.fNestedPaging) u32XcptBitmap |= RT_BIT(X86_XCPT_PF); pVCpu->hm.s.vmx.u32XcptBitmap = u32XcptBitmap; int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap); AssertRCReturn(rc, rc); return rc; } /** * Sets up the initial guest-state mask. The guest-state mask is consulted * before reading guest-state fields from the VMCS as VMREADs can be expensive * for the nested virtualization case (as it would cause a VM-exit). * * @param pVCpu Pointer to the VMCPU. */ static int hmR0VmxInitUpdatedGuestStateMask(PVMCPU pVCpu) { /* Initially the guest-state is up-to-date as there is nothing in the VMCS. */ pVCpu->hm.s.vmx.fUpdatedGuestState = HMVMX_UPDATED_GUEST_ALL; return VINF_SUCCESS; } /** * Does per-VM VT-x initialization. * * @returns VBox status code. * @param pVM Pointer to the VM. */ VMMR0DECL(int) VMXR0InitVM(PVM pVM) { LogFlowFunc(("pVM=%p\n", pVM)); int rc = hmR0VmxStructsAlloc(pVM); if (RT_FAILURE(rc)) { LogRel(("VMXR0InitVM: hmR0VmxStructsAlloc failed! rc=%Rrc\n", rc)); return rc; } return VINF_SUCCESS; } /** * Does per-VM VT-x termination. * * @returns VBox status code. * @param pVM Pointer to the VM. */ VMMR0DECL(int) VMXR0TermVM(PVM pVM) { LogFlowFunc(("pVM=%p\n", pVM)); #ifdef VBOX_WITH_CRASHDUMP_MAGIC if (pVM->hm.s.vmx.hMemObjScratch != NIL_RTR0MEMOBJ) ASMMemZero32(pVM->hm.s.vmx.pvScratch, PAGE_SIZE); #endif hmR0VmxStructsFree(pVM); return VINF_SUCCESS; } /** * Sets up the VM for execution under VT-x. * This function is only called once per-VM during initialization. * * @returns VBox status code. * @param pVM Pointer to the VM. */ VMMR0DECL(int) VMXR0SetupVM(PVM pVM) { AssertPtrReturn(pVM, VERR_INVALID_PARAMETER); Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); LogFlowFunc(("pVM=%p\n", pVM)); /* * Without UnrestrictedGuest, pRealModeTSS and pNonPagingModeEPTPageTable *must* always be allocated. * We no longer support the highly unlikely case of UnrestrictedGuest without pRealModeTSS. See hmR3InitFinalizeR0(). */ /* -XXX- change hmR3InitFinalizeR0Intel() to fail if pRealModeTSS alloc fails. */ if ( !pVM->hm.s.vmx.fUnrestrictedGuest && ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable || !pVM->hm.s.vmx.pRealModeTSS)) { LogRel(("VMXR0SetupVM: invalid real-on-v86 state.\n")); return VERR_INTERNAL_ERROR; } #ifdef VBOX_WITH_HYBRID_32BIT_KERNEL /* * This is for the darwin 32-bit/PAE kernels trying to execute 64-bit guests. We don't bother with * the 32<->64 switcher in this case. This is a rare, legacy use-case with barely any test coverage. */ if ( pVM->hm.s.fAllow64BitGuests && !HMVMX_IS_64BIT_HOST_MODE()) { LogRel(("VMXR0SetupVM: Unsupported guest and host paging mode combination.\n")); return VERR_PGM_UNSUPPORTED_HOST_PAGING_MODE; } #endif /* Initialize these always, see hmR3InitFinalizeR0().*/ pVM->hm.s.vmx.enmFlushEpt = VMX_FLUSH_EPT_NONE; pVM->hm.s.vmx.enmFlushVpid = VMX_FLUSH_VPID_NONE; /* Setup the tagged-TLB flush handlers. */ int rc = hmR0VmxSetupTaggedTlb(pVM); if (RT_FAILURE(rc)) { LogRel(("VMXR0SetupVM: hmR0VmxSetupTaggedTlb failed! rc=%Rrc\n", rc)); return rc; } for (VMCPUID i = 0; i < pVM->cCpus; i++) { PVMCPU pVCpu = &pVM->aCpus[i]; AssertPtr(pVCpu); AssertPtr(pVCpu->hm.s.vmx.pvVmcs); /* Log the VCPU pointers, useful for debugging SMP VMs. */ Log4(("VMXR0SetupVM: pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu)); /* Set revision dword at the beginning of the VMCS structure. */ *(uint32_t *)pVCpu->hm.s.vmx.pvVmcs = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo); /* Initialize our VMCS region in memory, set the VMCS launch state to "clear". */ rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); /* Load this VMCS as the current VMCS. */ rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXActivateVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); rc = hmR0VmxSetupPinCtls(pVM, pVCpu); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupPinCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); rc = hmR0VmxSetupProcCtls(pVM, pVCpu); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupProcCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); rc = hmR0VmxSetupMiscCtls(pVM, pVCpu); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupMiscCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); rc = hmR0VmxInitXcptBitmap(pVM, pVCpu); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitXcptBitmap failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); rc = hmR0VmxInitUpdatedGuestStateMask(pVCpu); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitUpdatedGuestStateMask failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); #if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) rc = hmR0VmxInitVmcsReadCache(pVM, pVCpu); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitVmcsReadCache failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); #endif /* Re-sync the CPU's internal data into our VMCS memory region & reset the launch state to "clear". */ rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs(2) failed! rc=%Rrc (pVM=%p)\n", rc, pVM), hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc); pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR; hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc); } return VINF_SUCCESS; } /** * Saves the host control registers (CR0, CR3, CR4) into the host-state area in * the VMCS. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. */ DECLINLINE(int) hmR0VmxSaveHostControlRegs(PVM pVM, PVMCPU pVCpu) { NOREF(pVM); NOREF(pVCpu); RTCCUINTREG uReg = ASMGetCR0(); int rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR0, uReg); AssertRCReturn(rc, rc); #ifdef VBOX_WITH_HYBRID_32BIT_KERNEL /* For the darwin 32-bit hybrid kernel, we need the 64-bit CR3 as it uses 64-bit paging. */ if (HMVMX_IS_64BIT_HOST_MODE()) { uint64_t uRegCR3 = HMR0Get64bitCR3(); rc = VMXWriteVmcs64(VMX_VMCS_HOST_CR3, uRegCR3); } else #endif { uReg = ASMGetCR3(); rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR3, uReg); } AssertRCReturn(rc, rc); uReg = ASMGetCR4(); rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR4, uReg); AssertRCReturn(rc, rc); return rc; } #if HC_ARCH_BITS == 64 /** * Macro for adjusting host segment selectors to satisfy VT-x's VM-entry * requirements. See hmR0VmxSaveHostSegmentRegs(). */ # define VMXLOCAL_ADJUST_HOST_SEG(seg, selValue) \ if ((selValue) & (X86_SEL_RPL | X86_SEL_LDT)) \ { \ bool fValidSelector = true; \ if ((selValue) & X86_SEL_LDT) \ { \ uint32_t uAttr = ASMGetSegAttr((selValue)); \ fValidSelector = RT_BOOL(uAttr != ~0U && (uAttr & X86_DESC_P)); \ } \ if (fValidSelector) \ { \ pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_SEL_##seg; \ pVCpu->hm.s.vmx.RestoreHost.uHostSel##seg = (selValue); \ } \ (selValue) = 0; \ } #endif /** * Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into * the host-state area in the VMCS. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. */ DECLINLINE(int) hmR0VmxSaveHostSegmentRegs(PVM pVM, PVMCPU pVCpu) { NOREF(pVM); int rc = VERR_INTERNAL_ERROR_5; /* * Host DS, ES, FS and GS segment registers. */ #if HC_ARCH_BITS == 64 RTSEL uSelDS = ASMGetDS(); RTSEL uSelES = ASMGetES(); RTSEL uSelFS = ASMGetFS(); RTSEL uSelGS = ASMGetGS(); #else RTSEL uSelDS = 0; RTSEL uSelES = 0; RTSEL uSelFS = 0; RTSEL uSelGS = 0; #endif /* Recalculate which host-state bits need to be manually restored. */ pVCpu->hm.s.vmx.fRestoreHostFlags = 0; /* * Host CS and SS segment registers. */ #ifdef VBOX_WITH_HYBRID_32BIT_KERNEL RTSEL uSelCS; RTSEL uSelSS; if (HMVMX_IS_64BIT_HOST_MODE()) { uSelCS = (RTSEL)(uintptr_t)&SUPR0Abs64bitKernelCS; uSelSS = (RTSEL)(uintptr_t)&SUPR0Abs64bitKernelSS; } else { /* Seems darwin uses the LDT (TI flag is set) in the CS & SS selectors which VT-x doesn't like. */ uSelCS = (RTSEL)(uintptr_t)&SUPR0AbsKernelCS; uSelSS = (RTSEL)(uintptr_t)&SUPR0AbsKernelSS; } #else RTSEL uSelCS = ASMGetCS(); RTSEL uSelSS = ASMGetSS(); #endif /* * Host TR segment register. */ RTSEL uSelTR = ASMGetTR(); #if HC_ARCH_BITS == 64 /* * Determine if the host segment registers are suitable for VT-x. Otherwise use zero to gain VM-entry and restore them * before we get preempted. See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers". */ VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS); VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES); VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS); VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS); # undef VMXLOCAL_ADJUST_HOST_SEG #endif /* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */ Assert(!(uSelCS & X86_SEL_RPL)); Assert(!(uSelCS & X86_SEL_LDT)); Assert(!(uSelSS & X86_SEL_RPL)); Assert(!(uSelSS & X86_SEL_LDT)); Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT)); Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT)); Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT)); Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT)); Assert(!(uSelTR & X86_SEL_RPL)); Assert(!(uSelTR & X86_SEL_LDT)); Assert(uSelCS); Assert(uSelTR); /* Assertion is right but we would not have updated u32ExitCtls yet. */ #if 0 if (!(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE)) Assert(uSelSS != 0); #endif /* Write these host selector fields into the host-state area in the VMCS. */ rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_CS, uSelCS); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_SS, uSelSS); AssertRCReturn(rc, rc); #if HC_ARCH_BITS == 64 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_DS, uSelDS); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_ES, uSelES); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_FS, uSelFS); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_GS, uSelGS); AssertRCReturn(rc, rc); #endif rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_TR, uSelTR); AssertRCReturn(rc, rc); /* * Host GDTR and IDTR. */ RTGDTR Gdtr; RT_ZERO(Gdtr); #ifdef VBOX_WITH_HYBRID_32BIT_KERNEL if (HMVMX_IS_64BIT_HOST_MODE()) { X86XDTR64 Gdtr64; X86XDTR64 Idtr64; HMR0Get64bitGdtrAndIdtr(&Gdtr64, &Idtr64); rc = VMXWriteVmcs64(VMX_VMCS_HOST_GDTR_BASE, Gdtr64.uAddr); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS_HOST_IDTR_BASE, Idtr64.uAddr); AssertRCReturn(rc, rc); Gdtr.cbGdt = Gdtr64.cb; Gdtr.pGdt = (uintptr_t)Gdtr64.uAddr; } else #endif { RTIDTR Idtr; ASMGetGDTR(&Gdtr); ASMGetIDTR(&Idtr); rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, Gdtr.pGdt); AssertRCReturn(rc, rc); rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, Idtr.pIdt); AssertRCReturn(rc, rc); #if HC_ARCH_BITS == 64 /* * Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps them to the * maximum limit (0xffff) on every VM-exit. */ if (Gdtr.cbGdt != 0xffff) { pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_GDTR; AssertCompile(sizeof(Gdtr) == sizeof(X86XDTR64)); memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64)); } /* * IDT limit is practically 0xfff. Therefore if the host has the limit as 0xfff, VT-x bloating the limit to 0xffff * is not a problem as it's not possible to get at them anyway. See Intel spec. 6.14.1 "64-Bit Mode IDT" and * Intel spec. 6.2 "Exception and Interrupt Vectors". */ if (Idtr.cbIdt < 0x0fff) { pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_IDTR; AssertCompile(sizeof(Idtr) == sizeof(X86XDTR64)); memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostIdtr, &Idtr, sizeof(X86XDTR64)); } #endif } /* * Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI and RPL bits * is effectively what the CPU does for "scaling by 8". TI is always 0 and RPL should be too in most cases. */ if ((uSelTR & X86_SEL_MASK) > Gdtr.cbGdt) { AssertMsgFailed(("hmR0VmxSaveHostSegmentRegs: TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, Gdtr.cbGdt)); return VERR_VMX_INVALID_HOST_STATE; } PCX86DESCHC pDesc = (PCX86DESCHC)(Gdtr.pGdt + (uSelTR & X86_SEL_MASK)); #ifdef VBOX_WITH_HYBRID_32BIT_KERNEL if (HMVMX_IS_64BIT_HOST_MODE()) { /* We need the 64-bit TR base for hybrid darwin. */ uint64_t u64TRBase = X86DESC64_BASE((PX86DESC64)pDesc); rc = VMXWriteVmcs64(VMX_VMCS_HOST_TR_BASE, u64TRBase); } else #endif { uintptr_t uTRBase; #if HC_ARCH_BITS == 64 uTRBase = X86DESC64_BASE(pDesc); /* * VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on all VM-exits. * The type is the same for 64-bit busy TSS[1]. The limit needs manual restoration if the host has something else. * Task switching is not supported in 64-bit mode[2], but the limit still matters as IOPM is supported in 64-bit mode. * Restoring the limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0. * * [1] See Intel spec. 3.5 "System Descriptor Types". * [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode". */ Assert(pDesc->System.u4Type == 11); if ( pDesc->System.u16LimitLow != 0x67 || pDesc->System.u4LimitHigh) { pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_SEL_TR; pVCpu->hm.s.vmx.RestoreHost.uHostSelTR = uSelTR; /* Store the GDTR here as we need it while restoring TR. */ memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64)); } #else uTRBase = X86DESC_BASE(pDesc); #endif rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_TR_BASE, uTRBase); } AssertRCReturn(rc, rc); /* * Host FS base and GS base. */ #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) { uint64_t u64FSBase = ASMRdMsr(MSR_K8_FS_BASE); uint64_t u64GSBase = ASMRdMsr(MSR_K8_GS_BASE); rc = VMXWriteVmcs64(VMX_VMCS_HOST_FS_BASE, u64FSBase); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS_HOST_GS_BASE, u64GSBase); AssertRCReturn(rc, rc); # if HC_ARCH_BITS == 64 /* Store the base if we have to restore FS or GS manually as we need to restore the base as well. */ if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_FS) pVCpu->hm.s.vmx.RestoreHost.uHostFSBase = u64FSBase; if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_GS) pVCpu->hm.s.vmx.RestoreHost.uHostGSBase = u64GSBase; # endif } #endif return rc; } /** * Saves certain host MSRs in the VM-Exit MSR-load area and some in the * host-state area of the VMCS. Theses MSRs will be automatically restored on * the host after every successful VM exit. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxSaveHostMsrs(PVM pVM, PVMCPU pVCpu) { NOREF(pVM); AssertPtr(pVCpu); AssertPtr(pVCpu->hm.s.vmx.pvHostMsr); int rc = VINF_SUCCESS; #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if ( HMVMX_IS_64BIT_HOST_MODE() && pVM->hm.s.fAllow64BitGuests) { hmR0VmxLazySaveHostMsrs(pVCpu); } #endif if (pVCpu->hm.s.vmx.cMsrs > 0) hmR0VmxUpdateAutoLoadStoreHostMsrs(pVCpu); /* * Host Sysenter MSRs. */ rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS)); AssertRCReturn(rc, rc); #ifdef VBOX_WITH_HYBRID_32BIT_KERNEL if (HMVMX_IS_64BIT_HOST_MODE()) { rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP)); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP)); } else { rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP)); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP)); } #elif HC_ARCH_BITS == 32 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP)); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP)); #else rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP)); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP)); #endif AssertRCReturn(rc, rc); /** @todo IA32_PERF_GLOBALCTRL, IA32_PAT, IA32_EFER, also see * hmR0VmxSetupExitCtls() !! */ return rc; } /** * Sets up VM-entry controls in the VMCS. These controls can affect things done * on VM-exit; e.g. "load debug controls", see Intel spec. 24.8.1 "VM-entry * controls". * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxLoadGuestEntryCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS)) { PVM pVM = pVCpu->CTX_SUFF(pVM); uint32_t val = pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0; /* Bits set here must be set in the VMCS. */ uint32_t zap = pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */ /* Load debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x capable CPUs only supports the 1-setting of this bit. */ val |= VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG; /* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry. */ if (CPUMIsGuestInLongModeEx(pMixedCtx)) val |= VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST; else Assert(!(val & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST)); /* * The following should -not- be set (since we're not in SMM mode): * - VMX_VMCS_CTRL_ENTRY_ENTRY_SMM * - VMX_VMCS_CTRL_ENTRY_DEACTIVATE_DUALMON */ /** @todo VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR, * VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR, * VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR */ if ((val & zap) != val) { LogRel(("hmR0VmxLoadGuestEntryCtls: invalid VM-entry controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n", pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0, val, zap)); pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, val); AssertRCReturn(rc, rc); /* Update VCPU with the currently set VM-exit controls. */ pVCpu->hm.s.vmx.u32EntryCtls = val; VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS); } return rc; } /** * Sets up the VM-exit controls in the VMCS. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks requires EFER. */ DECLINLINE(int) hmR0VmxLoadGuestExitCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); int rc = VINF_SUCCESS; if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_VMX_EXIT_CTLS)) { PVM pVM = pVCpu->CTX_SUFF(pVM); uint32_t val = pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0; /* Bits set here must be set in the VMCS. */ uint32_t zap = pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */ /* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only supported the 1-setting of this bit. */ val |= VMX_VMCS_CTRL_EXIT_SAVE_DEBUG; /* * Set the host long mode active (EFER.LMA) bit (which Intel calls "Host address-space size") if necessary. * On VM-exit, VT-x sets both the host EFER.LMA and EFER.LME bit to this value. See assertion in hmR0VmxSaveHostMsrs(). */ #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) val |= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE; else Assert(!(val & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE)); #elif HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) if (CPUMIsGuestInLongModeEx(pMixedCtx)) val |= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE; /* The switcher goes to long mode. */ else Assert(!(val & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE)); #endif /* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */ Assert(!(val & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT)); /** @todo VMX_VMCS_CTRL_EXIT_LOAD_PERF_MSR, * VMX_VMCS_CTRL_EXIT_SAVE_GUEST_PAT_MSR, * VMX_VMCS_CTRL_EXIT_LOAD_HOST_PAT_MSR, * VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR, * VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR. */ if (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER) val |= VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER; if ((val & zap) != val) { LogRel(("hmR0VmxSetupProcCtls: invalid VM-exit controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n", pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0, val, zap)); pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT; return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO; } rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, val); AssertRCReturn(rc, rc); /* Update VCPU with the currently set VM-exit controls. */ pVCpu->hm.s.vmx.u32ExitCtls = val; VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_VMX_EXIT_CTLS); } return rc; } /** * Loads the guest APIC and related state. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ DECLINLINE(int) hmR0VmxLoadGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); int rc = VINF_SUCCESS; if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE)) { /* Setup TPR shadowing. Also setup TPR patching for 32-bit guests. */ if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW) { Assert(pVCpu->hm.s.vmx.HCPhysVirtApic); bool fPendingIntr = false; uint8_t u8Tpr = 0; uint8_t u8PendingIntr = 0; rc = PDMApicGetTPR(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr); AssertRCReturn(rc, rc); /* * If there are external interrupts pending but masked by the TPR value, instruct VT-x to cause a VM-exit when * the guest lowers its TPR below the highest-priority pending interrupt and we can deliver the interrupt. * If there are no external interrupts pending, set threshold to 0 to not cause a VM-exit. We will eventually deliver * the interrupt when we VM-exit for other reasons. */ pVCpu->hm.s.vmx.pbVirtApic[0x80] = u8Tpr; /* Offset 0x80 is TPR in the APIC MMIO range. */ uint32_t u32TprThreshold = 0; if (fPendingIntr) { /* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR (which is the Task-Priority Class). */ const uint8_t u8PendingPriority = (u8PendingIntr >> 4) & 0xf; const uint8_t u8TprPriority = (u8Tpr >> 4) & 0xf; if (u8PendingPriority <= u8TprPriority) u32TprThreshold = u8PendingPriority; else u32TprThreshold = u8TprPriority; /* Required for Vista 64-bit guest, see @bugref{6398}. */ } Assert(!(u32TprThreshold & 0xfffffff0)); /* Bits 31:4 MBZ. */ rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold); AssertRCReturn(rc, rc); } VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE); } return rc; } /** * Gets the guest's interruptibility-state ("interrupt shadow" as AMD calls it). * * @returns Guest's interruptibility-state. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! * @remarks Has side-effects with VMCPU_FF_INHIBIT_INTERRUPTS force-flag. */ DECLINLINE(uint32_t) hmR0VmxGetGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { /* * Instructions like STI and MOV SS inhibit interrupts till the next instruction completes. Check if we should * inhibit interrupts or clear any existing interrupt-inhibition. */ uint32_t uIntrState = 0; if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)) { /* If inhibition is active, RIP & RFLAGS should've been accessed (i.e. read previously from the VMCS or from ring-3). */ AssertMsg((pVCpu->hm.s.vmx.fUpdatedGuestState & (HMVMX_UPDATED_GUEST_RIP | HMVMX_UPDATED_GUEST_RFLAGS)) == (HMVMX_UPDATED_GUEST_RIP | HMVMX_UPDATED_GUEST_RFLAGS), ("%#x\n", pVCpu->hm.s.vmx.fUpdatedGuestState)); if (pMixedCtx->rip != EMGetInhibitInterruptsPC(pVCpu)) { /* * We can clear the inhibit force flag as even if we go back to the recompiler without executing guest code in * VT-x, the flag's condition to be cleared is met and thus the cleared state is correct. */ VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS); } else if (pMixedCtx->eflags.Bits.u1IF) uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI; else uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS; } return uIntrState; } /** * Loads the guest's interruptibility-state into the guest-state area in the * VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param uIntrState The interruptibility-state to set. */ static int hmR0VmxLoadGuestIntrState(PVMCPU pVCpu, uint32_t uIntrState) { NOREF(pVCpu); AssertMsg(!(uIntrState & 0xfffffff0), ("%#x\n", uIntrState)); /* Bits 31:4 MBZ. */ Assert((uIntrState & 0x3) != 0x3); /* Block-by-STI and MOV SS cannot be simultaneously set. */ int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, uIntrState); AssertRCReturn(rc, rc); return rc; } /** * Loads the guest's RIP into the guest-state area in the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RIP)) { rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RIP, pMixedCtx->rip); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_RIP); Log4(("Load: VMX_VMCS_GUEST_RIP=%#RX64 fContextUseFlags=%#RX32\n", pMixedCtx->rip, VMCPU_HMCF_VALUE(pVCpu))); } return rc; } /** * Loads the guest's RSP into the guest-state area in the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RSP)) { rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RSP, pMixedCtx->rsp); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_RSP); Log4(("Load: VMX_VMCS_GUEST_RSP=%#RX64\n", pMixedCtx->rsp)); } return rc; } /** * Loads the guest's RFLAGS into the guest-state area in the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS)) { /* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ). Let us assert it as such and use 32-bit VMWRITE. */ Assert(!(pMixedCtx->rflags.u64 >> 32)); X86EFLAGS Eflags = pMixedCtx->eflags; Eflags.u32 &= VMX_EFLAGS_RESERVED_0; /* Bits 22-31, 15, 5 & 3 MBZ. */ Eflags.u32 |= VMX_EFLAGS_RESERVED_1; /* Bit 1 MB1. */ /* * If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so we can restore them on VM exit. * Modify the real-mode guest's eflags so that VT-x can run the real-mode guest code under Virtual 8086 mode. */ if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS); Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM))); pVCpu->hm.s.vmx.RealMode.Eflags.u32 = Eflags.u32; /* Save the original eflags of the real-mode guest. */ Eflags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */ Eflags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */ } rc = VMXWriteVmcs32(VMX_VMCS_GUEST_RFLAGS, Eflags.u32); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_RFLAGS); Log4(("Load: VMX_VMCS_GUEST_RFLAGS=%#RX32\n", Eflags.u32)); } return rc; } /** * Loads the guest RIP, RSP and RFLAGS into the guest-state area in the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxLoadGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); rc = hmR0VmxLoadGuestRsp(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); rc = hmR0VmxLoadGuestRflags(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); return rc; } /** * Loads the guest CR0 control register into the guest-state area in the VMCS. * CR0 is partially shared with the host and we have to consider the FPU bits. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadSharedCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { /* * Guest CR0. * Guest FPU. */ int rc = VINF_SUCCESS; if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0)) { Assert(!(pMixedCtx->cr0 >> 32)); uint32_t u32GuestCR0 = pMixedCtx->cr0; PVM pVM = pVCpu->CTX_SUFF(pVM); /* The guest's view (read access) of its CR0 is unblemished. */ rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, u32GuestCR0); AssertRCReturn(rc, rc); Log4(("Load: VMX_VMCS_CTRL_CR0_READ_SHADOW=%#RX32\n", u32GuestCR0)); /* Setup VT-x's view of the guest CR0. */ /* Minimize VM-exits due to CR3 changes when we have NestedPaging. */ if (pVM->hm.s.fNestedPaging) { if (CPUMIsGuestPagingEnabledEx(pMixedCtx)) { /* The guest has paging enabled, let it access CR3 without causing a VM exit if supported. */ pVCpu->hm.s.vmx.u32ProcCtls &= ~( VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT); } else { /* The guest doesn't have paging enabled, make CR3 access to cause VM exits to update our shadow. */ pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT; } /* If we have unrestricted guest execution, we never have to intercept CR3 reads. */ if (pVM->hm.s.vmx.fUnrestrictedGuest) pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT; rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRCReturn(rc, rc); } else u32GuestCR0 |= X86_CR0_WP; /* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */ /* * Guest FPU bits. * Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be set on the first * CPUs to support VT-x and no mention of with regards to UX in VM-entry checks. */ u32GuestCR0 |= X86_CR0_NE; bool fInterceptNM = false; if (CPUMIsGuestFPUStateActive(pVCpu)) { fInterceptNM = false; /* Guest FPU active, no need to VM-exit on #NM. */ /* The guest should still get #NM exceptions when it expects it to, so we should not clear TS & MP bits here. We're only concerned about -us- not intercepting #NMs when the guest-FPU is active. Not the guest itself! */ } else { fInterceptNM = true; /* Guest FPU inactive, VM-exit on #NM for lazy FPU loading. */ u32GuestCR0 |= X86_CR0_TS /* Guest can task switch quickly and do lazy FPU syncing. */ | X86_CR0_MP; /* FWAIT/WAIT should not ignore CR0.TS and should generate #NM. */ } /* Catch floating point exceptions if we need to report them to the guest in a different way. */ bool fInterceptMF = false; if (!(pMixedCtx->cr0 & X86_CR0_NE)) fInterceptMF = true; /* Finally, intercept all exceptions as we cannot directly inject them in real-mode, see hmR0VmxInjectEventVmcs(). */ if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { Assert(PDMVmmDevHeapIsEnabled(pVM)); Assert(pVM->hm.s.vmx.pRealModeTSS); pVCpu->hm.s.vmx.u32XcptBitmap |= HMVMX_REAL_MODE_XCPT_MASK; fInterceptNM = true; fInterceptMF = true; } else pVCpu->hm.s.vmx.u32XcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK; if (fInterceptNM) pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_NM); else pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_NM); if (fInterceptMF) pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_MF); else pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_MF); /* Additional intercepts for debugging, define these yourself explicitly. */ #ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS pVCpu->hm.s.vmx.u32XcptBitmap |= 0 | RT_BIT(X86_XCPT_BP) | RT_BIT(X86_XCPT_DB) | RT_BIT(X86_XCPT_DE) | RT_BIT(X86_XCPT_NM) | RT_BIT(X86_XCPT_UD) | RT_BIT(X86_XCPT_NP) | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_GP) | RT_BIT(X86_XCPT_PF) | RT_BIT(X86_XCPT_MF) ; #elif defined(HMVMX_ALWAYS_TRAP_PF) pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_PF); #endif Assert(pVM->hm.s.fNestedPaging || (pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT(X86_XCPT_PF))); /* Set/clear the CR0 specific bits along with their exceptions (PE, PG, CD, NW). */ uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1); uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr0Fixed1); if (pVM->hm.s.vmx.fUnrestrictedGuest) /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG). */ uSetCR0 &= ~(X86_CR0_PE | X86_CR0_PG); else Assert((uSetCR0 & (X86_CR0_PE | X86_CR0_PG)) == (X86_CR0_PE | X86_CR0_PG)); u32GuestCR0 |= uSetCR0; u32GuestCR0 &= uZapCR0; u32GuestCR0 &= ~(X86_CR0_CD | X86_CR0_NW); /* Always enable caching. */ /* Write VT-x's view of the guest CR0 into the VMCS and update the exception bitmap. */ rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR0, u32GuestCR0); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap); AssertRCReturn(rc, rc); Log4(("Load: VMX_VMCS_GUEST_CR0=%#RX32 (uSetCR0=%#RX32 uZapCR0=%#RX32)\n", u32GuestCR0, uSetCR0, uZapCR0)); /* * CR0 is shared between host and guest along with a CR0 read shadow. Therefore, certain bits must not be changed * by the guest because VT-x ignores saving/restoring them (namely CD, ET, NW) and for certain other bits * we want to be notified immediately of guest CR0 changes (e.g. PG to update our shadow page tables). */ uint32_t u32CR0Mask = 0; u32CR0Mask = X86_CR0_PE | X86_CR0_NE | X86_CR0_WP | X86_CR0_PG | X86_CR0_ET /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.ET */ | X86_CR0_CD /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.CD */ | X86_CR0_NW; /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.NW */ /** @todo Avoid intercepting CR0.PE with unrestricted guests. Fix PGM * enmGuestMode to be in-sync with the current mode. See @bugref{6398} * and @bugref{6944}. */ #if 0 if (pVM->hm.s.vmx.fUnrestrictedGuest) u32CR0Mask &= ~X86_CR0_PE; #endif if (pVM->hm.s.fNestedPaging) u32CR0Mask &= ~X86_CR0_WP; /* If the guest FPU state is active, don't need to VM-exit on writes to FPU related bits in CR0. */ if (fInterceptNM) { u32CR0Mask |= X86_CR0_TS | X86_CR0_MP; } /* Write the CR0 mask into the VMCS and update the VCPU's copy of the current CR0 mask. */ pVCpu->hm.s.vmx.u32CR0Mask = u32CR0Mask; rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_MASK, u32CR0Mask); AssertRCReturn(rc, rc); Log4(("Load: VMX_VMCS_CTRL_CR0_MASK=%#RX32\n", u32CR0Mask)); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR0); } return rc; } /** * Loads the guest control registers (CR3, CR4) into the guest-state area * in the VMCS. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadGuestCR3AndCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; PVM pVM = pVCpu->CTX_SUFF(pVM); /* * Guest CR2. * It's always loaded in the assembler code. Nothing to do here. */ /* * Guest CR3. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR3)) { RTGCPHYS GCPhysGuestCR3 = NIL_RTGCPHYS; if (pVM->hm.s.fNestedPaging) { pVCpu->hm.s.vmx.HCPhysEPTP = PGMGetHyperCR3(pVCpu); /* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */ Assert(pVCpu->hm.s.vmx.HCPhysEPTP); Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & UINT64_C(0xfff0000000000000))); Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & 0xfff)); /* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */ pVCpu->hm.s.vmx.HCPhysEPTP |= VMX_EPT_MEMTYPE_WB | (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT); /* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */ AssertMsg( ((pVCpu->hm.s.vmx.HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */ && ((pVCpu->hm.s.vmx.HCPhysEPTP >> 6) & 0x3f) == 0, /* Bits 6:11 MBZ. */ ("EPTP %#RX64\n", pVCpu->hm.s.vmx.HCPhysEPTP)); rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVCpu->hm.s.vmx.HCPhysEPTP); AssertRCReturn(rc, rc); Log4(("Load: VMX_VMCS64_CTRL_EPTP_FULL=%#RX64\n", pVCpu->hm.s.vmx.HCPhysEPTP)); if ( pVM->hm.s.vmx.fUnrestrictedGuest || CPUMIsGuestPagingEnabledEx(pMixedCtx)) { /* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */ if (CPUMIsGuestInPAEModeEx(pMixedCtx)) { rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u); AssertRCReturn(rc, rc); rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u); AssertRCReturn(rc, rc); } /* The guest's view of its CR3 is unblemished with Nested Paging when the guest is using paging or we have Unrestricted Execution to handle the guest when it's not using paging. */ GCPhysGuestCR3 = pMixedCtx->cr3; } else { /* * The guest is not using paging, but the CPU (VT-x) has to. While the guest thinks it accesses physical memory * directly, we use our identity-mapped page table to map guest-linear to guest-physical addresses. * EPT takes care of translating it to host-physical addresses. */ RTGCPHYS GCPhys; Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable); Assert(PDMVmmDevHeapIsEnabled(pVM)); /* We obtain it here every time as the guest could have relocated this PCI region. */ rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys); AssertRCReturn(rc, rc); GCPhysGuestCR3 = GCPhys; } Log4(("Load: VMX_VMCS_GUEST_CR3=%#RGv (GstN)\n", GCPhysGuestCR3)); rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_CR3, GCPhysGuestCR3); } else { /* Non-nested paging case, just use the hypervisor's CR3. */ RTHCPHYS HCPhysGuestCR3 = PGMGetHyperCR3(pVCpu); Log4(("Load: VMX_VMCS_GUEST_CR3=%#RHv (HstN)\n", HCPhysGuestCR3)); rc = VMXWriteVmcsHstN(VMX_VMCS_GUEST_CR3, HCPhysGuestCR3); } AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR3); } /* * Guest CR4. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR4)) { Assert(!(pMixedCtx->cr4 >> 32)); uint32_t u32GuestCR4 = pMixedCtx->cr4; /* The guest's view of its CR4 is unblemished. */ rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, u32GuestCR4); AssertRCReturn(rc, rc); Log4(("Load: VMX_VMCS_CTRL_CR4_READ_SHADOW=%#RX32\n", u32GuestCR4)); /* Setup VT-x's view of the guest CR4. */ /* * If we're emulating real-mode using virtual-8086 mode, we want to redirect software interrupts to the 8086 program * interrupt handler. Clear the VME bit (the interrupt redirection bitmap is already all 0, see hmR3InitFinalizeR0()) * See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode". */ if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { Assert(pVM->hm.s.vmx.pRealModeTSS); Assert(PDMVmmDevHeapIsEnabled(pVM)); u32GuestCR4 &= ~X86_CR4_VME; } if (pVM->hm.s.fNestedPaging) { if ( !CPUMIsGuestPagingEnabledEx(pMixedCtx) && !pVM->hm.s.vmx.fUnrestrictedGuest) { /* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */ u32GuestCR4 |= X86_CR4_PSE; /* Our identity mapping is a 32-bit page directory. */ u32GuestCR4 &= ~X86_CR4_PAE; } /* else use guest CR4.*/ } else { /* * The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host * paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables. */ switch (pVCpu->hm.s.enmShadowMode) { case PGMMODE_REAL: /* Real-mode. */ case PGMMODE_PROTECTED: /* Protected mode without paging. */ case PGMMODE_32_BIT: /* 32-bit paging. */ { u32GuestCR4 &= ~X86_CR4_PAE; break; } case PGMMODE_PAE: /* PAE paging. */ case PGMMODE_PAE_NX: /* PAE paging with NX. */ { u32GuestCR4 |= X86_CR4_PAE; break; } case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */ case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */ #ifdef VBOX_ENABLE_64_BITS_GUESTS break; #endif default: AssertFailed(); return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE; } } /* We need to set and clear the CR4 specific bits here (mainly the X86_CR4_VMXE bit). */ uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1); uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr4Fixed1); u32GuestCR4 |= uSetCR4; u32GuestCR4 &= uZapCR4; /* Write VT-x's view of the guest CR4 into the VMCS. */ Log4(("Load: VMX_VMCS_GUEST_CR4=%#RX32 (Set=%#RX32 Zap=%#RX32)\n", u32GuestCR4, uSetCR4, uZapCR4)); rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR4, u32GuestCR4); AssertRCReturn(rc, rc); /* Setup CR4 mask. CR4 flags owned by the host, if the guest attempts to change them, that would cause a VM exit. */ uint32_t u32CR4Mask = 0; u32CR4Mask = X86_CR4_VME | X86_CR4_PAE | X86_CR4_PGE | X86_CR4_PSE | X86_CR4_VMXE; pVCpu->hm.s.vmx.u32CR4Mask = u32CR4Mask; rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_MASK, u32CR4Mask); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR4); } return rc; } /** * Loads the guest debug registers into the guest-state area in the VMCS. * This also sets up whether #DB and MOV DRx accesses cause VM exits. * * The guest debug bits are partially shared with the host (e.g. DR6, DR0-3). * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadSharedDebugState(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { if (!VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG)) return VINF_SUCCESS; #ifdef VBOX_STRICT /* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */ if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG) { /* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */ Assert((pMixedCtx->dr[7] & (X86_DR7_MBZ_MASK | X86_DR7_RAZ_MASK)) == 0); /* Bits 63:32, 15, 14, 12, 11 are reserved. */ Assert((pMixedCtx->dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK); /* Bit 10 is reserved (RA1). */ } #endif int rc; PVM pVM = pVCpu->CTX_SUFF(pVM); bool fInterceptDB = false; bool fInterceptMovDRx = false; if ( pVCpu->hm.s.fSingleInstruction || DBGFIsStepping(pVCpu)) { /* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */ if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG) { pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG; rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRCReturn(rc, rc); Assert(fInterceptDB == false); } else { pMixedCtx->eflags.u32 |= X86_EFL_TF; pVCpu->hm.s.fClearTrapFlag = true; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS); fInterceptDB = true; } } if ( fInterceptDB || (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK)) { /* * Use the combined guest and host DRx values found in the hypervisor * register set because the debugger has breakpoints active or someone * is single stepping on the host side without a monitor trap flag. * * Note! DBGF expects a clean DR6 state before executing guest code. */ #if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if ( CPUMIsGuestInLongModeEx(pMixedCtx) && !CPUMIsHyperDebugStateActivePending(pVCpu)) { CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */); Assert(CPUMIsHyperDebugStateActivePending(pVCpu)); Assert(!CPUMIsGuestDebugStateActivePending(pVCpu)); } else #endif if (!CPUMIsHyperDebugStateActive(pVCpu)) { CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */); Assert(CPUMIsHyperDebugStateActive(pVCpu)); Assert(!CPUMIsGuestDebugStateActive(pVCpu)); } /* Update DR7. (The other DRx values are handled by CPUM one way or the other.) */ rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)CPUMGetHyperDR7(pVCpu)); AssertRCReturn(rc, rc); pVCpu->hm.s.fUsingHyperDR7 = true; fInterceptDB = true; fInterceptMovDRx = true; } else { /* * If the guest has enabled debug registers, we need to load them prior to * executing guest code so they'll trigger at the right time. */ if (pMixedCtx->dr[7] & (X86_DR7_ENABLED_MASK | X86_DR7_GD)) /** @todo Why GD? */ { #if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if ( CPUMIsGuestInLongModeEx(pMixedCtx) && !CPUMIsGuestDebugStateActivePending(pVCpu)) { CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */); Assert(CPUMIsGuestDebugStateActivePending(pVCpu)); Assert(!CPUMIsHyperDebugStateActivePending(pVCpu)); STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed); } else #endif if (!CPUMIsGuestDebugStateActive(pVCpu)) { CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */); Assert(CPUMIsGuestDebugStateActive(pVCpu)); Assert(!CPUMIsHyperDebugStateActive(pVCpu)); STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed); } Assert(!fInterceptDB); Assert(!fInterceptMovDRx); } /* * If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we * must intercept #DB in order to maintain a correct DR6 guest value. */ #if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) else if ( !CPUMIsGuestDebugStateActivePending(pVCpu) && !CPUMIsGuestDebugStateActive(pVCpu)) #else else if (!CPUMIsGuestDebugStateActive(pVCpu)) #endif { fInterceptMovDRx = true; fInterceptDB = true; } /* Update guest DR7. */ rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, pMixedCtx->dr[7]); AssertRCReturn(rc, rc); pVCpu->hm.s.fUsingHyperDR7 = false; } /* * Update the exception bitmap regarding intercepting #DB generated by the guest. */ if (fInterceptDB) pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_DB); else if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { #ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_DB); #endif } rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap); AssertRCReturn(rc, rc); /* * Update the processor-based VM-execution controls regarding intercepting MOV DRx instructions. */ if (fInterceptMovDRx) pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT; else pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT; rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_DEBUG); return VINF_SUCCESS; } #ifdef VBOX_STRICT /** * Strict function to validate segment registers. * * @remarks ASSUMES CR0 is up to date. */ static void hmR0VmxValidateSegmentRegs(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx) { /* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers". */ /* NOTE: The reason we check for attribute value 0 and not just the unusable bit here is because hmR0VmxWriteSegmentReg() * only updates the VMCS' copy of the value with the unusable bit and doesn't change the guest-context value. */ if ( !pVM->hm.s.vmx.fUnrestrictedGuest && ( !CPUMIsGuestInRealModeEx(pCtx) && !CPUMIsGuestInV86ModeEx(pCtx))) { /* Protected mode checks */ /* CS */ Assert(pCtx->cs.Attr.n.u1Present); Assert(!(pCtx->cs.Attr.u & 0xf00)); Assert(!(pCtx->cs.Attr.u & 0xfffe0000)); Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff || !(pCtx->cs.Attr.n.u1Granularity)); Assert( !(pCtx->cs.u32Limit & 0xfff00000) || (pCtx->cs.Attr.n.u1Granularity)); /* CS cannot be loaded with NULL in protected mode. */ Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS?!? */ if (pCtx->cs.Attr.n.u4Type == 9 || pCtx->cs.Attr.n.u4Type == 11) Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl); else if (pCtx->cs.Attr.n.u4Type == 13 || pCtx->cs.Attr.n.u4Type == 15) Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl); else AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl)); /* SS */ Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL)); Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL)); if ( !(pCtx->cr0 & X86_CR0_PE) || pCtx->cs.Attr.n.u4Type == 3) { Assert(!pCtx->ss.Attr.n.u2Dpl); } if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE)) { Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL)); Assert(pCtx->ss.Attr.n.u4Type == 3 || pCtx->ss.Attr.n.u4Type == 7); Assert(pCtx->ss.Attr.n.u1Present); Assert(!(pCtx->ss.Attr.u & 0xf00)); Assert(!(pCtx->ss.Attr.u & 0xfffe0000)); Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff || !(pCtx->ss.Attr.n.u1Granularity)); Assert( !(pCtx->ss.u32Limit & 0xfff00000) || (pCtx->ss.Attr.n.u1Granularity)); } /* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */ if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE)) { Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED); Assert(pCtx->ds.Attr.n.u1Present); Assert(pCtx->ds.Attr.n.u4Type > 11 || pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL)); Assert(!(pCtx->ds.Attr.u & 0xf00)); Assert(!(pCtx->ds.Attr.u & 0xfffe0000)); Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff || !(pCtx->ds.Attr.n.u1Granularity)); Assert( !(pCtx->ds.u32Limit & 0xfff00000) || (pCtx->ds.Attr.n.u1Granularity)); Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE) || (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ)); } if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE)) { Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED); Assert(pCtx->es.Attr.n.u1Present); Assert(pCtx->es.Attr.n.u4Type > 11 || pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL)); Assert(!(pCtx->es.Attr.u & 0xf00)); Assert(!(pCtx->es.Attr.u & 0xfffe0000)); Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff || !(pCtx->es.Attr.n.u1Granularity)); Assert( !(pCtx->es.u32Limit & 0xfff00000) || (pCtx->es.Attr.n.u1Granularity)); Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE) || (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ)); } if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE)) { Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED); Assert(pCtx->fs.Attr.n.u1Present); Assert(pCtx->fs.Attr.n.u4Type > 11 || pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL)); Assert(!(pCtx->fs.Attr.u & 0xf00)); Assert(!(pCtx->fs.Attr.u & 0xfffe0000)); Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff || !(pCtx->fs.Attr.n.u1Granularity)); Assert( !(pCtx->fs.u32Limit & 0xfff00000) || (pCtx->fs.Attr.n.u1Granularity)); Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE) || (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ)); } if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE)) { Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED); Assert(pCtx->gs.Attr.n.u1Present); Assert(pCtx->gs.Attr.n.u4Type > 11 || pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL)); Assert(!(pCtx->gs.Attr.u & 0xf00)); Assert(!(pCtx->gs.Attr.u & 0xfffe0000)); Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff || !(pCtx->gs.Attr.n.u1Granularity)); Assert( !(pCtx->gs.u32Limit & 0xfff00000) || (pCtx->gs.Attr.n.u1Granularity)); Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE) || (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ)); } /* 64-bit capable CPUs. */ # if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) Assert(!(pCtx->cs.u64Base >> 32)); Assert(!pCtx->ss.Attr.u || !(pCtx->ss.u64Base >> 32)); Assert(!pCtx->ds.Attr.u || !(pCtx->ds.u64Base >> 32)); Assert(!pCtx->es.Attr.u || !(pCtx->es.u64Base >> 32)); # endif } else if ( CPUMIsGuestInV86ModeEx(pCtx) || ( CPUMIsGuestInRealModeEx(pCtx) && !pVM->hm.s.vmx.fUnrestrictedGuest)) { /* Real and v86 mode checks. */ /* hmR0VmxWriteSegmentReg() writes the modified in VMCS. We want what we're feeding to VT-x. */ uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr; if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3; u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3; } else { u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u; } /* CS */ AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel)); Assert(pCtx->cs.u32Limit == 0xffff); Assert(u32CSAttr == 0xf3); /* SS */ Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4); Assert(pCtx->ss.u32Limit == 0xffff); Assert(u32SSAttr == 0xf3); /* DS */ Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4); Assert(pCtx->ds.u32Limit == 0xffff); Assert(u32DSAttr == 0xf3); /* ES */ Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4); Assert(pCtx->es.u32Limit == 0xffff); Assert(u32ESAttr == 0xf3); /* FS */ Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4); Assert(pCtx->fs.u32Limit == 0xffff); Assert(u32FSAttr == 0xf3); /* GS */ Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4); Assert(pCtx->gs.u32Limit == 0xffff); Assert(u32GSAttr == 0xf3); /* 64-bit capable CPUs. */ # if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) Assert(!(pCtx->cs.u64Base >> 32)); Assert(!u32SSAttr || !(pCtx->ss.u64Base >> 32)); Assert(!u32DSAttr || !(pCtx->ds.u64Base >> 32)); Assert(!u32ESAttr || !(pCtx->es.u64Base >> 32)); # endif } } #endif /* VBOX_STRICT */ /** * Writes a guest segment register into the guest-state area in the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param idxSel Index of the selector in the VMCS. * @param idxLimit Index of the segment limit in the VMCS. * @param idxBase Index of the segment base in the VMCS. * @param idxAccess Index of the access rights of the segment in the VMCS. * @param pSelReg Pointer to the segment selector. * * @remarks No-long-jump zone!!! */ static int hmR0VmxWriteSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase, uint32_t idxAccess, PCPUMSELREG pSelReg) { int rc = VMXWriteVmcs32(idxSel, pSelReg->Sel); /* 16-bit guest selector field. */ AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(idxLimit, pSelReg->u32Limit); /* 32-bit guest segment limit field. */ AssertRCReturn(rc, rc); rc = VMXWriteVmcsGstN(idxBase, pSelReg->u64Base); /* Natural width guest segment base field.*/ AssertRCReturn(rc, rc); uint32_t u32Access = pSelReg->Attr.u; if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { /* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */ u32Access = 0xf3; Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS); Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM))); } else { /* * The way to differentiate between whether this is really a null selector or was just a selector loaded with 0 in * real-mode is using the segment attributes. A selector loaded in real-mode with the value 0 is valid and usable in * protected-mode and we should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures NULL selectors * loaded in protected-mode have their attribute as 0. */ if (!u32Access) u32Access = X86DESCATTR_UNUSABLE; } /* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */ AssertMsg((u32Access & X86DESCATTR_UNUSABLE) || (u32Access & X86_SEL_TYPE_ACCESSED), ("Access bit not set for usable segment. idx=%#x sel=%#x attr %#x\n", idxBase, pSelReg, pSelReg->Attr.u)); rc = VMXWriteVmcs32(idxAccess, u32Access); /* 32-bit guest segment access-rights field. */ AssertRCReturn(rc, rc); return rc; } /** * Loads the guest segment registers, GDTR, IDTR, LDTR, (TR, FS and GS bases) * into the guest-state area in the VMCS. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCPU Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks ASSUMES pMixedCtx->cr0 is up to date (strict builds validation). * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VERR_INTERNAL_ERROR_5; PVM pVM = pVCpu->CTX_SUFF(pVM); /* * Guest Segment registers: CS, SS, DS, ES, FS, GS. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS)) { /* Save the segment attributes for real-on-v86 mode hack, so we can restore them on VM-exit. */ if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { pVCpu->hm.s.vmx.RealMode.AttrCS.u = pMixedCtx->cs.Attr.u; pVCpu->hm.s.vmx.RealMode.AttrSS.u = pMixedCtx->ss.Attr.u; pVCpu->hm.s.vmx.RealMode.AttrDS.u = pMixedCtx->ds.Attr.u; pVCpu->hm.s.vmx.RealMode.AttrES.u = pMixedCtx->es.Attr.u; pVCpu->hm.s.vmx.RealMode.AttrFS.u = pMixedCtx->fs.Attr.u; pVCpu->hm.s.vmx.RealMode.AttrGS.u = pMixedCtx->gs.Attr.u; } #ifdef VBOX_WITH_REM if (!pVM->hm.s.vmx.fUnrestrictedGuest) { Assert(pVM->hm.s.vmx.pRealModeTSS); AssertCompile(PGMMODE_REAL < PGMMODE_PROTECTED); if ( pVCpu->hm.s.vmx.fWasInRealMode && PGMGetGuestMode(pVCpu) >= PGMMODE_PROTECTED) { /* Signal that the recompiler must flush its code-cache as the guest -may- rewrite code it will later execute in real-mode (e.g. OpenBSD 4.0) */ REMFlushTBs(pVM); Log4(("Load: Switch to protected mode detected!\n")); pVCpu->hm.s.vmx.fWasInRealMode = false; } } #endif rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_CS, VMX_VMCS32_GUEST_CS_LIMIT, VMX_VMCS_GUEST_CS_BASE, VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS, &pMixedCtx->cs); AssertRCReturn(rc, rc); rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_SS, VMX_VMCS32_GUEST_SS_LIMIT, VMX_VMCS_GUEST_SS_BASE, VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS, &pMixedCtx->ss); AssertRCReturn(rc, rc); rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_DS, VMX_VMCS32_GUEST_DS_LIMIT, VMX_VMCS_GUEST_DS_BASE, VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS, &pMixedCtx->ds); AssertRCReturn(rc, rc); rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_ES, VMX_VMCS32_GUEST_ES_LIMIT, VMX_VMCS_GUEST_ES_BASE, VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, &pMixedCtx->es); AssertRCReturn(rc, rc); rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_FS, VMX_VMCS32_GUEST_FS_LIMIT, VMX_VMCS_GUEST_FS_BASE, VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS, &pMixedCtx->fs); AssertRCReturn(rc, rc); rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_GS, VMX_VMCS32_GUEST_GS_LIMIT, VMX_VMCS_GUEST_GS_BASE, VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS, &pMixedCtx->gs); AssertRCReturn(rc, rc); #ifdef VBOX_STRICT /* Validate. */ hmR0VmxValidateSegmentRegs(pVM, pVCpu, pMixedCtx); #endif VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS); Log4(("Load: CS=%#RX16 Base=%#RX64 Limit=%#RX32 Attr=%#RX32\n", pMixedCtx->cs.Sel, pMixedCtx->cs.u64Base, pMixedCtx->cs.u32Limit, pMixedCtx->cs.Attr.u)); } /* * Guest TR. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_TR)) { /* * Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is achieved * using the interrupt redirection bitmap (all bits cleared to let the guest handle INT-n's) in the TSS. * See hmR3InitFinalizeR0() to see how pRealModeTSS is setup. */ uint16_t u16Sel = 0; uint32_t u32Limit = 0; uint64_t u64Base = 0; uint32_t u32AccessRights = 0; if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { u16Sel = pMixedCtx->tr.Sel; u32Limit = pMixedCtx->tr.u32Limit; u64Base = pMixedCtx->tr.u64Base; u32AccessRights = pMixedCtx->tr.Attr.u; } else { Assert(pVM->hm.s.vmx.pRealModeTSS); Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMR3CanExecuteGuest() -XXX- what about inner loop changes? */ /* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */ RTGCPHYS GCPhys; rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys); AssertRCReturn(rc, rc); X86DESCATTR DescAttr; DescAttr.u = 0; DescAttr.n.u1Present = 1; DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY; u16Sel = 0; u32Limit = HM_VTX_TSS_SIZE; u64Base = GCPhys; /* in real-mode phys = virt. */ u32AccessRights = DescAttr.u; } /* Validate. */ Assert(!(u16Sel & RT_BIT(2))); AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY || (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights)); AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights)); Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/ Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/ Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */ Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */ Assert( (u32Limit & 0xfff) == 0xfff || !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */ Assert( !(pMixedCtx->tr.u32Limit & 0xfff00000) || (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */ rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_TR, u16Sel); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit); AssertRCReturn(rc, rc); rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_TR_BASE, u64Base); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_TR); Log4(("Load: VMX_VMCS_GUEST_TR_BASE=%#RX64\n", u64Base)); } /* * Guest GDTR. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_GDTR)) { rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pMixedCtx->gdtr.cbGdt); AssertRCReturn(rc, rc); rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, pMixedCtx->gdtr.pGdt); AssertRCReturn(rc, rc); /* Validate. */ Assert(!(pMixedCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */ VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_GDTR); Log4(("Load: VMX_VMCS_GUEST_GDTR_BASE=%#RX64\n", pMixedCtx->gdtr.pGdt)); } /* * Guest LDTR. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LDTR)) { /* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */ uint32_t u32Access = 0; if (!pMixedCtx->ldtr.Attr.u) u32Access = X86DESCATTR_UNUSABLE; else u32Access = pMixedCtx->ldtr.Attr.u; rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_LDTR, pMixedCtx->ldtr.Sel); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pMixedCtx->ldtr.u32Limit); AssertRCReturn(rc, rc); rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_LDTR_BASE, pMixedCtx->ldtr.u64Base); AssertRCReturn(rc, rc); rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access); AssertRCReturn(rc, rc); /* Validate. */ if (!(u32Access & X86DESCATTR_UNUSABLE)) { Assert(!(pMixedCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */ Assert(pMixedCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */ Assert(!pMixedCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */ Assert(pMixedCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */ Assert(!pMixedCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */ Assert(!(pMixedCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */ Assert( (pMixedCtx->ldtr.u32Limit & 0xfff) == 0xfff || !pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */ Assert( !(pMixedCtx->ldtr.u32Limit & 0xfff00000) || pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */ } VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_LDTR); Log4(("Load: VMX_VMCS_GUEST_LDTR_BASE=%#RX64\n", pMixedCtx->ldtr.u64Base)); } /* * Guest IDTR. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_IDTR)) { rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pMixedCtx->idtr.cbIdt); AssertRCReturn(rc, rc); rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, pMixedCtx->idtr.pIdt); AssertRCReturn(rc, rc); /* Validate. */ Assert(!(pMixedCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */ VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_IDTR); Log4(("Load: VMX_VMCS_GUEST_IDTR_BASE=%#RX64\n", pMixedCtx->idtr.pIdt)); } return VINF_SUCCESS; } /** * Loads certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store * areas. These MSRs will automatically be loaded to the host CPU on every * successful VM entry and stored from the host CPU on every successful VM exit. * * This also creates/updates MSR slots for the host MSRs. The actual host * MSR values are -not- updated here for performance reasons. See * hmR0VmxSaveHostMsrs(). * * Also loads the sysenter MSRs into the guest-state area in the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { AssertPtr(pVCpu); AssertPtr(pVCpu->hm.s.vmx.pvGuestMsr); /* * MSRs that we use the auto-load/store MSR area in the VMCS. */ PVM pVM = pVCpu->CTX_SUFF(pVM); if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS)) { if (pVM->hm.s.fAllow64BitGuests) { #if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_LSTAR, pMixedCtx->msrLSTAR, false /* fUpdateHostMsr */); hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_STAR, pMixedCtx->msrSTAR, false /* fUpdateHostMsr */); hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_SF_MASK, pMixedCtx->msrSFMASK, false /* fUpdateHostMsr */); hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_KERNEL_GS_BASE, pMixedCtx->msrKERNELGSBASE, false /* fUpdateHostMsr */); # ifdef DEBUG PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr; for (uint32_t i = 0; i < pVCpu->hm.s.vmx.cMsrs; i++, pMsr++) Log4(("MSR[%RU32]: u32Msr=%#RX32 u64Value=%#RX64\n", i, pMsr->u32Msr, pMsr->u64Value)); # endif #endif } VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS); } /* * Guest Sysenter MSRs. * These flags are only set when MSR-bitmaps are not supported by the CPU and we cause * VM-exits on WRMSRs for these MSRs. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR)) { int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pMixedCtx->SysEnter.cs); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR); } if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR)) { int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, pMixedCtx->SysEnter.eip); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); } if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR)) { int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, pMixedCtx->SysEnter.esp); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); } return VINF_SUCCESS; } /** * Loads the guest activity state into the guest-state area in the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadGuestActivityState(PVMCPU pVCpu, PCPUMCTX pCtx) { NOREF(pCtx); /** @todo See if we can make use of other states, e.g. * VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN or HLT. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE)) { int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, VMX_VMCS_GUEST_ACTIVITY_ACTIVE); AssertRCReturn(rc, rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE); } return VINF_SUCCESS; } /** * Sets up the appropriate function to run guest code. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSetupVMRunHandler(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { if (CPUMIsGuestInLongModeEx(pMixedCtx)) { #ifndef VBOX_ENABLE_64_BITS_GUESTS return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE; #endif Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); /* Guaranteed by hmR3InitFinalizeR0(). */ #if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) /* 32-bit host. We need to switch to 64-bit before running the 64-bit guest. */ if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0SwitcherStartVM64) { pVCpu->hm.s.vmx.pfnStartVM = VMXR0SwitcherStartVM64; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_VMX_EXIT_CTLS | HM_CHANGED_VMX_ENTRY_CTLS); } #else /* 64-bit host or hybrid host. */ pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM64; #endif } else { /* Guest is not in long mode, use the 32-bit handler. */ #if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0StartVM32) { pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_VMX_EXIT_CTLS | HM_CHANGED_VMX_ENTRY_CTLS); } #else pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32; #endif } Assert(pVCpu->hm.s.vmx.pfnStartVM); return VINF_SUCCESS; } /** * Wrapper for running the guest code in VT-x. * * @returns VBox strict status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCtx Pointer to the guest-CPU context. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx) { /* * 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses floating-point operations * using SSE instructions. Some XMM registers (XMM6-XMM15) are callee-saved and thus the need for this XMM wrapper. * Refer MSDN docs. "Configuring Programs for 64-bit / x64 Software Conventions / Register Usage" for details. */ const bool fResumeVM = RT_BOOL(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED); /** @todo Add stats for resume vs launch. */ #ifdef VBOX_WITH_KERNEL_USING_XMM return HMR0VMXStartVMWrapXMM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu, pVCpu->hm.s.vmx.pfnStartVM); #else return pVCpu->hm.s.vmx.pfnStartVM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu); #endif } /** * Reports world-switch error and dumps some useful debug info. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param rcVMRun The return code from VMLAUNCH/VMRESUME. * @param pCtx Pointer to the guest-CPU context. * @param pVmxTransient Pointer to the VMX transient structure (only * exitReason updated). */ static void hmR0VmxReportWorldSwitchError(PVM pVM, PVMCPU pVCpu, int rcVMRun, PCPUMCTX pCtx, PVMXTRANSIENT pVmxTransient) { Assert(pVM); Assert(pVCpu); Assert(pCtx); Assert(pVmxTransient); HMVMX_ASSERT_PREEMPT_SAFE(); Log4(("VM-entry failure: %Rrc\n", rcVMRun)); switch (rcVMRun) { case VERR_VMX_INVALID_VMXON_PTR: AssertFailed(); break; case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */ case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */ { int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason); rc |= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError); rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); AssertRC(rc); pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu; /* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted(). Cannot do it here as we may have been long preempted. */ #ifdef VBOX_STRICT Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason, pVmxTransient->uExitReason)); Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQualification)); Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError)); if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX) Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError])); else Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX)); Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu)); Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu)); /* VMX control bits. */ uint32_t u32Val; uint64_t u64Val; HMVMXHCUINTREG uHCReg; rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_PIN_EXEC %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_PROC_EXEC %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_PROC_EXEC2 %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_ENTRY %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_EXIT %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_CR3_TARGET_COUNT %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %u\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_TPR_THRESHOLD %u\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT %u (guest MSRs)\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT %u (host MSRs)\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT %u (guest MSRs)\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK %#RX32\n", u32Val)); rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, &u32Val); AssertRC(rc); Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH %#RX32\n", u32Val)); rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc); Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc); Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc); Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc); Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg)); rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc); Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val)); /* Guest bits. */ rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val); AssertRC(rc); Log4(("Old Guest Rip %#RX64 New %#RX64\n", pCtx->rip, u64Val)); rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val); AssertRC(rc); Log4(("Old Guest Rsp %#RX64 New %#RX64\n", pCtx->rsp, u64Val)); rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Val); AssertRC(rc); Log4(("Old Guest Rflags %#RX32 New %#RX32\n", pCtx->eflags.u32, u32Val)); rc = VMXReadVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, &u32Val); AssertRC(rc); Log4(("VMX_VMCS16_GUEST_FIELD_VPID %u\n", u32Val)); /* Host bits. */ rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR0, &uHCReg); AssertRC(rc); Log4(("Host CR0 %#RHr\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR3, &uHCReg); AssertRC(rc); Log4(("Host CR3 %#RHr\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR4, &uHCReg); AssertRC(rc); Log4(("Host CR4 %#RHr\n", uHCReg)); RTGDTR HostGdtr; PCX86DESCHC pDesc; ASMGetGDTR(&HostGdtr); rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_CS, &u32Val); AssertRC(rc); Log4(("Host CS %#08x\n", u32Val)); if (u32Val < HostGdtr.cbGdt) { pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK)); HMR0DumpDescriptor(pDesc, u32Val, "CS: "); } rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_DS, &u32Val); AssertRC(rc); Log4(("Host DS %#08x\n", u32Val)); if (u32Val < HostGdtr.cbGdt) { pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK)); HMR0DumpDescriptor(pDesc, u32Val, "DS: "); } rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_ES, &u32Val); AssertRC(rc); Log4(("Host ES %#08x\n", u32Val)); if (u32Val < HostGdtr.cbGdt) { pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK)); HMR0DumpDescriptor(pDesc, u32Val, "ES: "); } rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_FS, &u32Val); AssertRC(rc); Log4(("Host FS %#08x\n", u32Val)); if (u32Val < HostGdtr.cbGdt) { pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK)); HMR0DumpDescriptor(pDesc, u32Val, "FS: "); } rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_GS, &u32Val); AssertRC(rc); Log4(("Host GS %#08x\n", u32Val)); if (u32Val < HostGdtr.cbGdt) { pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK)); HMR0DumpDescriptor(pDesc, u32Val, "GS: "); } rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_SS, &u32Val); AssertRC(rc); Log4(("Host SS %#08x\n", u32Val)); if (u32Val < HostGdtr.cbGdt) { pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK)); HMR0DumpDescriptor(pDesc, u32Val, "SS: "); } rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_TR, &u32Val); AssertRC(rc); Log4(("Host TR %#08x\n", u32Val)); if (u32Val < HostGdtr.cbGdt) { pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK)); HMR0DumpDescriptor(pDesc, u32Val, "TR: "); } rc = VMXReadVmcsHstN(VMX_VMCS_HOST_TR_BASE, &uHCReg); AssertRC(rc); Log4(("Host TR Base %#RHv\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, &uHCReg); AssertRC(rc); Log4(("Host GDTR Base %#RHv\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, &uHCReg); AssertRC(rc); Log4(("Host IDTR Base %#RHv\n", uHCReg)); rc = VMXReadVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, &u32Val); AssertRC(rc); Log4(("Host SYSENTER CS %#08x\n", u32Val)); rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_EIP, &uHCReg); AssertRC(rc); Log4(("Host SYSENTER EIP %#RHv\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_ESP, &uHCReg); AssertRC(rc); Log4(("Host SYSENTER ESP %#RHv\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RSP, &uHCReg); AssertRC(rc); Log4(("Host RSP %#RHv\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RIP, &uHCReg); AssertRC(rc); Log4(("Host RIP %#RHv\n", uHCReg)); # if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) { Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER))); Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR))); Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR))); Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR))); Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK))); Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE))); } # endif #endif /* VBOX_STRICT */ break; } default: /* Impossible */ AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun)); break; } NOREF(pVM); NOREF(pCtx); } #if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) #ifndef VMX_USE_CACHED_VMCS_ACCESSES # error "VMX_USE_CACHED_VMCS_ACCESSES not defined when it should be!" #endif #ifdef VBOX_STRICT static bool hmR0VmxIsValidWriteField(uint32_t idxField) { switch (idxField) { case VMX_VMCS_GUEST_RIP: case VMX_VMCS_GUEST_RSP: case VMX_VMCS_GUEST_SYSENTER_EIP: case VMX_VMCS_GUEST_SYSENTER_ESP: case VMX_VMCS_GUEST_GDTR_BASE: case VMX_VMCS_GUEST_IDTR_BASE: case VMX_VMCS_GUEST_CS_BASE: case VMX_VMCS_GUEST_DS_BASE: case VMX_VMCS_GUEST_ES_BASE: case VMX_VMCS_GUEST_FS_BASE: case VMX_VMCS_GUEST_GS_BASE: case VMX_VMCS_GUEST_SS_BASE: case VMX_VMCS_GUEST_LDTR_BASE: case VMX_VMCS_GUEST_TR_BASE: case VMX_VMCS_GUEST_CR3: return true; } return false; } static bool hmR0VmxIsValidReadField(uint32_t idxField) { switch (idxField) { /* Read-only fields. */ case VMX_VMCS_RO_EXIT_QUALIFICATION: return true; } /* Remaining readable fields should also be writable. */ return hmR0VmxIsValidWriteField(idxField); } #endif /* VBOX_STRICT */ /** * Executes the specified handler in 64-bit mode. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCtx Pointer to the guest CPU context. * @param enmOp The operation to perform. * @param cbParam Number of parameters. * @param paParam Array of 32-bit parameters. */ VMMR0DECL(int) VMXR0Execute64BitsHandler(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, HM64ON32OP enmOp, uint32_t cbParam, uint32_t *paParam) { int rc, rc2; PHMGLOBALCPUINFO pCpu; RTHCPHYS HCPhysCpuPage; RTCCUINTREG uOldEflags; AssertReturn(pVM->hm.s.pfnHost32ToGuest64R0, VERR_HM_NO_32_TO_64_SWITCHER); Assert(enmOp > HM64ON32OP_INVALID && enmOp < HM64ON32OP_END); Assert(pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Write.aField)); Assert(pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Read.aField)); #ifdef VBOX_STRICT for (uint32_t i = 0; i < pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries; i++) Assert(hmR0VmxIsValidWriteField(pVCpu->hm.s.vmx.VMCSCache.Write.aField[i])); for (uint32_t i = 0; i hm.s.vmx.VMCSCache.Read.cValidEntries; i++) Assert(hmR0VmxIsValidReadField(pVCpu->hm.s.vmx.VMCSCache.Read.aField[i])); #endif /* Disable interrupts. */ uOldEflags = ASMIntDisableFlags(); #ifdef VBOX_WITH_VMMR0_DISABLE_LAPIC_NMI RTCPUID idHostCpu = RTMpCpuId(); CPUMR0SetLApic(pVCpu, idHostCpu); #endif pCpu = HMR0GetCurrentCpu(); HCPhysCpuPage = RTR0MemObjGetPagePhysAddr(pCpu->hMemObj, 0); /* Clear VMCS. Marking it inactive, clearing implementation-specific data and writing VMCS data back to memory. */ VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); /* Leave VMX Root Mode. */ VMXDisable(); ASMSetCR4(ASMGetCR4() & ~X86_CR4_VMXE); CPUMSetHyperESP(pVCpu, VMMGetStackRC(pVCpu)); CPUMSetHyperEIP(pVCpu, enmOp); for (int i = (int)cbParam - 1; i >= 0; i--) CPUMPushHyper(pVCpu, paParam[i]); STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatWorldSwitch3264, z); /* Call the switcher. */ rc = pVM->hm.s.pfnHost32ToGuest64R0(pVM, RT_OFFSETOF(VM, aCpus[pVCpu->idCpu].cpum) - RT_OFFSETOF(VM, cpum)); STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatWorldSwitch3264, z); /** @todo replace with hmR0VmxEnterRootMode() and hmR0VmxLeaveRootMode(). */ /* Make sure the VMX instructions don't cause #UD faults. */ ASMSetCR4(ASMGetCR4() | X86_CR4_VMXE); /* Re-enter VMX Root Mode */ rc2 = VMXEnable(HCPhysCpuPage); if (RT_FAILURE(rc2)) { ASMSetCR4(ASMGetCR4() & ~X86_CR4_VMXE); ASMSetFlags(uOldEflags); return rc2; } rc2 = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); AssertRC(rc2); Assert(!(ASMGetFlags() & X86_EFL_IF)); ASMSetFlags(uOldEflags); return rc; } /** * Prepares for and executes VMLAUNCH (64-bit guests) for 32-bit hosts * supporting 64-bit guests. * * @returns VBox status code. * @param fResume Whether to VMLAUNCH or VMRESUME. * @param pCtx Pointer to the guest-CPU context. * @param pCache Pointer to the VMCS cache. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. */ DECLASM(int) VMXR0SwitcherStartVM64(RTHCUINT fResume, PCPUMCTX pCtx, PVMCSCACHE pCache, PVM pVM, PVMCPU pVCpu) { uint32_t aParam[6]; PHMGLOBALCPUINFO pCpu = NULL; RTHCPHYS HCPhysCpuPage = 0; int rc = VERR_INTERNAL_ERROR_5; pCpu = HMR0GetCurrentCpu(); HCPhysCpuPage = RTR0MemObjGetPagePhysAddr(pCpu->hMemObj, 0); #ifdef VBOX_WITH_CRASHDUMP_MAGIC pCache->uPos = 1; pCache->interPD = PGMGetInterPaeCR3(pVM); pCache->pSwitcher = (uint64_t)pVM->hm.s.pfnHost32ToGuest64R0; #endif #if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES) pCache->TestIn.HCPhysCpuPage = 0; pCache->TestIn.HCPhysVmcs = 0; pCache->TestIn.pCache = 0; pCache->TestOut.HCPhysVmcs = 0; pCache->TestOut.pCache = 0; pCache->TestOut.pCtx = 0; pCache->TestOut.eflags = 0; #endif aParam[0] = (uint32_t)(HCPhysCpuPage); /* Param 1: VMXON physical address - Lo. */ aParam[1] = (uint32_t)(HCPhysCpuPage >> 32); /* Param 1: VMXON physical address - Hi. */ aParam[2] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs); /* Param 2: VMCS physical address - Lo. */ aParam[3] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs >> 32); /* Param 2: VMCS physical address - Hi. */ aParam[4] = VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache); aParam[5] = 0; #ifdef VBOX_WITH_CRASHDUMP_MAGIC pCtx->dr[4] = pVM->hm.s.vmx.pScratchPhys + 16 + 8; *(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) = 1; #endif rc = VMXR0Execute64BitsHandler(pVM, pVCpu, pCtx, HM64ON32OP_VMXRCStartVM64, 6, &aParam[0]); #ifdef VBOX_WITH_CRASHDUMP_MAGIC Assert(*(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) == 5); Assert(pCtx->dr[4] == 10); *(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) = 0xff; #endif #if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES) AssertMsg(pCache->TestIn.HCPhysCpuPage == HCPhysCpuPage, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysCpuPage, HCPhysCpuPage)); AssertMsg(pCache->TestIn.HCPhysVmcs == pVCpu->hm.s.vmx.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs, pVCpu->hm.s.vmx.HCPhysVmcs)); AssertMsg(pCache->TestIn.HCPhysVmcs == pCache->TestOut.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs, pCache->TestOut.HCPhysVmcs)); AssertMsg(pCache->TestIn.pCache == pCache->TestOut.pCache, ("%RGv vs %RGv\n", pCache->TestIn.pCache, pCache->TestOut.pCache)); AssertMsg(pCache->TestIn.pCache == VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache), ("%RGv vs %RGv\n", pCache->TestIn.pCache, VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache))); AssertMsg(pCache->TestIn.pCtx == pCache->TestOut.pCtx, ("%RGv vs %RGv\n", pCache->TestIn.pCtx, pCache->TestOut.pCtx)); Assert(!(pCache->TestOut.eflags & X86_EFL_IF)); #endif return rc; } /** * Initialize the VMCS-Read cache. The VMCS cache is used for 32-bit hosts * running 64-bit guests (except 32-bit Darwin which runs with 64-bit paging in * 32-bit mode) for 64-bit fields that cannot be accessed in 32-bit mode. Some * 64-bit fields -can- be accessed (those that have a 32-bit FULL & HIGH part). * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. */ static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu) { #define VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, idxField) \ { \ Assert(pCache->Read.aField[idxField##_CACHE_IDX] == 0); \ pCache->Read.aField[idxField##_CACHE_IDX] = idxField; \ pCache->Read.aFieldVal[idxField##_CACHE_IDX] = 0; \ ++cReadFields; \ } AssertPtr(pVM); AssertPtr(pVCpu); PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache; uint32_t cReadFields = 0; /* * Don't remove the #if 0'd fields in this code. They're listed here for consistency * and serve to indicate exceptions to the rules. */ /* Guest-natural selector base fields. */ #if 0 /* These are 32-bit in practice. See Intel spec. 2.5 "Control Registers". */ VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR0); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR4); #endif VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_ES_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CS_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SS_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_DS_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_FS_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GS_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_LDTR_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_TR_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GDTR_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_IDTR_BASE); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RSP); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RIP); #if 0 /* Unused natural width guest-state fields. */ VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3); /* Handled in Nested Paging case */ #endif VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_ESP); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_EIP); /* 64-bit guest-state fields; unused as we use two 32-bit VMREADs for these 64-bit fields (using "FULL" and "HIGH" fields). */ #if 0 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_DEBUGCTL_FULL); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PAT_FULL); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_EFER_FULL); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE0_FULL); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE1_FULL); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE2_FULL); VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE3_FULL); #endif /* Natural width guest-state fields. */ VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_QUALIFICATION); #if 0 /* Currently unused field. */ VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_GUEST_LINEAR_ADDR); #endif if (pVM->hm.s.fNestedPaging) { VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3); AssertMsg(cReadFields == VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields, VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX)); pCache->Read.cValidEntries = VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX; } else { AssertMsg(cReadFields == VMX_VMCS_MAX_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields, VMX_VMCS_MAX_CACHE_IDX)); pCache->Read.cValidEntries = VMX_VMCS_MAX_CACHE_IDX; } #undef VMXLOCAL_INIT_READ_CACHE_FIELD return VINF_SUCCESS; } /** * Writes a field into the VMCS. This can either directly invoke a VMWRITE or * queue up the VMWRITE by using the VMCS write cache (on 32-bit hosts, except * darwin, running 64-bit guests). * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param idxField The VMCS field encoding. * @param u64Val 16, 32 or 64-bit value. */ VMMR0DECL(int) VMXWriteVmcs64Ex(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val) { int rc; switch (idxField) { /* * These fields consists of a "FULL" and a "HIGH" part which can be written to individually. */ /* 64-bit Control fields. */ case VMX_VMCS64_CTRL_IO_BITMAP_A_FULL: case VMX_VMCS64_CTRL_IO_BITMAP_B_FULL: case VMX_VMCS64_CTRL_MSR_BITMAP_FULL: case VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL: case VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL: case VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL: case VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL: case VMX_VMCS64_CTRL_TSC_OFFSET_FULL: case VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL: case VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL: case VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL: case VMX_VMCS64_CTRL_EPTP_FULL: case VMX_VMCS64_CTRL_EPTP_LIST_FULL: /* 64-bit Guest-state fields. */ case VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL: case VMX_VMCS64_GUEST_DEBUGCTL_FULL: case VMX_VMCS64_GUEST_PAT_FULL: case VMX_VMCS64_GUEST_EFER_FULL: case VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL: case VMX_VMCS64_GUEST_PDPTE0_FULL: case VMX_VMCS64_GUEST_PDPTE1_FULL: case VMX_VMCS64_GUEST_PDPTE2_FULL: case VMX_VMCS64_GUEST_PDPTE3_FULL: /* 64-bit Host-state fields. */ case VMX_VMCS64_HOST_FIELD_PAT_FULL: case VMX_VMCS64_HOST_FIELD_EFER_FULL: case VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL: { rc = VMXWriteVmcs32(idxField, u64Val); rc |= VMXWriteVmcs32(idxField + 1, (uint32_t)(u64Val >> 32)); break; } /* * These fields do not have high and low parts. Queue up the VMWRITE by using the VMCS write-cache (for 64-bit * values). When we switch the host to 64-bit mode for running 64-bit guests, these VMWRITEs get executed then. */ /* Natural-width Guest-state fields. */ case VMX_VMCS_GUEST_CR3: case VMX_VMCS_GUEST_ES_BASE: case VMX_VMCS_GUEST_CS_BASE: case VMX_VMCS_GUEST_SS_BASE: case VMX_VMCS_GUEST_DS_BASE: case VMX_VMCS_GUEST_FS_BASE: case VMX_VMCS_GUEST_GS_BASE: case VMX_VMCS_GUEST_LDTR_BASE: case VMX_VMCS_GUEST_TR_BASE: case VMX_VMCS_GUEST_GDTR_BASE: case VMX_VMCS_GUEST_IDTR_BASE: case VMX_VMCS_GUEST_RSP: case VMX_VMCS_GUEST_RIP: case VMX_VMCS_GUEST_SYSENTER_ESP: case VMX_VMCS_GUEST_SYSENTER_EIP: { if (!(u64Val >> 32)) { /* If this field is 64-bit, VT-x will zero out the top bits. */ rc = VMXWriteVmcs32(idxField, (uint32_t)u64Val); } else { /* Assert that only the 32->64 switcher case should ever come here. */ Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); rc = VMXWriteCachedVmcsEx(pVCpu, idxField, u64Val); } break; } default: { AssertMsgFailed(("VMXWriteVmcs64Ex: Invalid field %#RX32 (pVCpu=%p u64Val=%#RX64)\n", idxField, pVCpu, u64Val)); rc = VERR_INVALID_PARAMETER; break; } } AssertRCReturn(rc, rc); return rc; } /** * Queue up a VMWRITE by using the VMCS write cache. This is only used on 32-bit * hosts (except darwin) for 64-bit guests. * * @param pVCpu Pointer to the VMCPU. * @param idxField The VMCS field encoding. * @param u64Val 16, 32 or 64-bit value. */ VMMR0DECL(int) VMXWriteCachedVmcsEx(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val) { AssertPtr(pVCpu); PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache; AssertMsgReturn(pCache->Write.cValidEntries < VMCSCACHE_MAX_ENTRY - 1, ("entries=%u\n", pCache->Write.cValidEntries), VERR_ACCESS_DENIED); /* Make sure there are no duplicates. */ for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++) { if (pCache->Write.aField[i] == idxField) { pCache->Write.aFieldVal[i] = u64Val; return VINF_SUCCESS; } } pCache->Write.aField[pCache->Write.cValidEntries] = idxField; pCache->Write.aFieldVal[pCache->Write.cValidEntries] = u64Val; pCache->Write.cValidEntries++; return VINF_SUCCESS; } /* Enable later when the assembly code uses these as callbacks. */ #if 0 /* * Loads the VMCS write-cache into the CPU (by executing VMWRITEs). * * @param pVCpu Pointer to the VMCPU. * @param pCache Pointer to the VMCS cache. * * @remarks No-long-jump zone!!! */ VMMR0DECL(void) VMXWriteCachedVmcsLoad(PVMCPU pVCpu, PVMCSCACHE pCache) { AssertPtr(pCache); for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++) { int rc = VMXWriteVmcs64(pCache->Write.aField[i], pCache->Write.aFieldVal[i]); AssertRC(rc); } pCache->Write.cValidEntries = 0; } /** * Stores the VMCS read-cache from the CPU (by executing VMREADs). * * @param pVCpu Pointer to the VMCPU. * @param pCache Pointer to the VMCS cache. * * @remarks No-long-jump zone!!! */ VMMR0DECL(void) VMXReadCachedVmcsStore(PVMCPU pVCpu, PVMCSCACHE pCache) { AssertPtr(pCache); for (uint32_t i = 0; i < pCache->Read.cValidEntries; i++) { int rc = VMXReadVmcs64(pCache->Read.aField[i], &pCache->Read.aFieldVal[i]); AssertRC(rc); } } #endif #endif /* HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) */ /** * Sets up the usage of TSC-offsetting and updates the VMCS. If offsetting is * not possible, cause VM-exits on RDTSC(P)s. Also sets up the VMX preemption * timer. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * * @remarks No-long-jump zone!!! */ static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVMCPU pVCpu) { int rc = VERR_INTERNAL_ERROR_5; bool fOffsettedTsc = false; PVM pVM = pVCpu->CTX_SUFF(pVM); if (pVM->hm.s.vmx.fUsePreemptTimer) { uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVCpu, &fOffsettedTsc, &pVCpu->hm.s.vmx.u64TSCOffset); /* Make sure the returned values have sane upper and lower boundaries. */ uint64_t u64CpuHz = SUPGetCpuHzFromGIP(g_pSUPGlobalInfoPage); cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second */ cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 2048); /* 1/2048th of a second */ cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift; uint32_t cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16); rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_PREEMPT_TIMER_VALUE, cPreemptionTickCount); AssertRC(rc); } else fOffsettedTsc = TMCpuTickCanUseRealTSC(pVCpu, &pVCpu->hm.s.vmx.u64TSCOffset); if (fOffsettedTsc) { uint64_t u64CurTSC = ASMReadTSC(); if (u64CurTSC + pVCpu->hm.s.vmx.u64TSCOffset >= TMCpuTickGetLastSeen(pVCpu)) { /* Note: VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT takes precedence over TSC_OFFSET, applies to RDTSCP too. */ rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, pVCpu->hm.s.vmx.u64TSCOffset); AssertRC(rc); pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT; rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc); STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset); } else { /* VM-exit on RDTSC(P) as we would otherwise pass decreasing TSC values to the guest. */ pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT; rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc); STAM_COUNTER_INC(&pVCpu->hm.s.StatTscInterceptOverFlow); } } else { /* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */ pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT; rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc); STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept); } } /** * Determines if an exception is a contributory exception. Contributory * exceptions are ones which can cause double-faults. Page-fault is * intentionally not included here as it's a conditional contributory exception. * * @returns true if the exception is contributory, false otherwise. * @param uVector The exception vector. */ DECLINLINE(bool) hmR0VmxIsContributoryXcpt(const uint32_t uVector) { switch (uVector) { case X86_XCPT_GP: case X86_XCPT_SS: case X86_XCPT_NP: case X86_XCPT_TS: case X86_XCPT_DE: return true; default: break; } return false; } /** * Sets an event as a pending event to be injected into the guest. * * @param pVCpu Pointer to the VMCPU. * @param u32IntInfo The VM-entry interruption-information field. * @param cbInstr The VM-entry instruction length in bytes (for software * interrupts, exceptions and privileged software * exceptions). * @param u32ErrCode The VM-entry exception error code. * @param GCPtrFaultAddress The fault-address (CR2) in case it's a * page-fault. * * @remarks Statistics counter assumes this is a guest event being injected or * re-injected into the guest, i.e. 'StatInjectPendingReflect' is * always incremented. */ DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPU pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode, RTGCUINTPTR GCPtrFaultAddress) { Assert(!pVCpu->hm.s.Event.fPending); pVCpu->hm.s.Event.fPending = true; pVCpu->hm.s.Event.u64IntInfo = u32IntInfo; pVCpu->hm.s.Event.u32ErrCode = u32ErrCode; pVCpu->hm.s.Event.cbInstr = cbInstr; pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress; STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect); } /** * Sets a double-fault (#DF) exception as pending-for-injection into the VM. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); uint32_t u32IntInfo = X86_XCPT_DF | VMX_EXIT_INTERRUPTION_INFO_VALID; u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID; hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */); } /** * Handle a condition that occurred while delivering an event through the guest * IDT. * * @returns VBox status code (informational error codes included). * @retval VINF_SUCCESS if we should continue handling the VM-exit. * @retval VINF_HM_DOUBLE_FAULT if a #DF condition was detected and we ought to * continue execution of the guest which will delivery the #DF. * @retval VINF_EM_RESET if we detected a triple-fault condition. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * @param pVmxTransient Pointer to the VMX transient structure. * * @remarks No-long-jump zone!!! */ static int hmR0VmxCheckExitDueToEventDelivery(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { int rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient); AssertRCReturn(rc, rc); if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo)) { rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient); AssertRCReturn(rc, rc); uint32_t uIntType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo); uint32_t uExitVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo); uint32_t uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo); typedef enum { VMXREFLECTXCPT_XCPT, /* Reflect the exception to the guest or for further evaluation by VMM. */ VMXREFLECTXCPT_DF, /* Reflect the exception as a double-fault to the guest. */ VMXREFLECTXCPT_TF, /* Indicate a triple faulted state to the VMM. */ VMXREFLECTXCPT_NONE /* Nothing to reflect. */ } VMXREFLECTXCPT; /* See Intel spec. 30.7.1.1 "Reflecting Exceptions to Guest Software". */ VMXREFLECTXCPT enmReflect = VMXREFLECTXCPT_NONE; if (VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo)) { if (uIntType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT) { enmReflect = VMXREFLECTXCPT_XCPT; #ifdef VBOX_STRICT if ( hmR0VmxIsContributoryXcpt(uIdtVector) && uExitVector == X86_XCPT_PF) { Log4(("IDT: vcpu[%RU32] Contributory #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2)); } #endif if ( uExitVector == X86_XCPT_PF && uIdtVector == X86_XCPT_PF) { pVmxTransient->fVectoringPF = true; Log4(("IDT: vcpu[%RU32] Vectoring #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2)); } else if ( (pVCpu->hm.s.vmx.u32XcptBitmap & HMVMX_CONTRIBUTORY_XCPT_MASK) && hmR0VmxIsContributoryXcpt(uExitVector) && ( hmR0VmxIsContributoryXcpt(uIdtVector) || uIdtVector == X86_XCPT_PF)) { enmReflect = VMXREFLECTXCPT_DF; } else if (uIdtVector == X86_XCPT_DF) enmReflect = VMXREFLECTXCPT_TF; } else if ( uIntType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT || uIntType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT || uIntType == VMX_IDT_VECTORING_INFO_TYPE_NMI) { /* * Ignore software interrupts (INT n), software exceptions (#BP, #OF) and privileged software exception * (whatever they are) as they reoccur when restarting the instruction. */ enmReflect = VMXREFLECTXCPT_XCPT; } } else { /* * If event delivery caused an EPT violation/misconfig or APIC access VM-exit, then the VM-exit * interruption-information will not be valid and we end up here. In such cases, it is sufficient to reflect the * original exception to the guest after handling the VM-exit. */ if ( uIntType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT || uIntType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT || uIntType == VMX_IDT_VECTORING_INFO_TYPE_NMI) { enmReflect = VMXREFLECTXCPT_XCPT; } } switch (enmReflect) { case VMXREFLECTXCPT_XCPT: { Assert( uIntType != VMX_IDT_VECTORING_INFO_TYPE_SW_INT && uIntType != VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT && uIntType != VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT); uint32_t u32ErrCode = 0; if (VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo)) { rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient); AssertRCReturn(rc, rc); u32ErrCode = pVmxTransient->uIdtVectoringErrorCode; } /* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF. See hmR0VmxExitXcptPF(). */ hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo), 0 /* cbInstr */, u32ErrCode, pMixedCtx->cr2); rc = VINF_SUCCESS; Log4(("IDT: vcpu[%RU32] Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.u32ErrCode)); break; } case VMXREFLECTXCPT_DF: { hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx); rc = VINF_HM_DOUBLE_FAULT; Log4(("IDT: vcpu[%RU32] Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo, uIdtVector, uExitVector)); break; } case VMXREFLECTXCPT_TF: { rc = VINF_EM_RESET; Log4(("IDT: vcpu[%RU32] Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", pVCpu->idCpu, uIdtVector, uExitVector)); break; } default: Assert(rc == VINF_SUCCESS); break; } } Assert(rc == VINF_SUCCESS || rc == VINF_HM_DOUBLE_FAULT || rc == VINF_EM_RESET); return rc; } /** * Saves the guest's CR0 register from the VMCS into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_CR0)) { uint32_t uVal = 0; int rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &uVal); AssertRCReturn(rc, rc); uint32_t uShadow = 0; rc = VMXReadVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uShadow); AssertRCReturn(rc, rc); uVal = (uShadow & pVCpu->hm.s.vmx.u32CR0Mask) | (uVal & ~pVCpu->hm.s.vmx.u32CR0Mask); CPUMSetGuestCR0(pVCpu, uVal); pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_CR0; } return VINF_SUCCESS; } /** * Saves the guest's CR4 register from the VMCS into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); int rc = VINF_SUCCESS; if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_CR4)) { uint32_t uVal = 0; uint32_t uShadow = 0; rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &uVal); AssertRCReturn(rc, rc); rc = VMXReadVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uShadow); AssertRCReturn(rc, rc); uVal = (uShadow & pVCpu->hm.s.vmx.u32CR4Mask) | (uVal & ~pVCpu->hm.s.vmx.u32CR4Mask); CPUMSetGuestCR4(pVCpu, uVal); pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_CR4; } return rc; } /** * Saves the guest's RIP register from the VMCS into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_RIP)) { uint64_t u64Val = 0; rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val); AssertRCReturn(rc, rc); pMixedCtx->rip = u64Val; pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_RIP; } return rc; } /** * Saves the guest's RSP register from the VMCS into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_RSP)) { uint64_t u64Val = 0; rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val); AssertRCReturn(rc, rc); pMixedCtx->rsp = u64Val; pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_RSP; } return rc; } /** * Saves the guest's RFLAGS from the VMCS into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_RFLAGS)) { uint32_t uVal = 0; int rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &uVal); AssertRCReturn(rc, rc); pMixedCtx->eflags.u32 = uVal; if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) /* Undo our real-on-v86-mode changes to eflags if necessary. */ { Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS); Log4(("Saving real-mode EFLAGS VT-x view=%#RX32\n", pMixedCtx->eflags.u32)); pMixedCtx->eflags.Bits.u1VM = 0; pMixedCtx->eflags.Bits.u2IOPL = pVCpu->hm.s.vmx.RealMode.Eflags.Bits.u2IOPL; } pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_RFLAGS; } return VINF_SUCCESS; } /** * Wrapper for saving the guest's RIP, RSP and RFLAGS from the VMCS into the * guest-CPU context. */ DECLINLINE(int) hmR0VmxSaveGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestRsp(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); return rc; } /** * Saves the guest's interruptibility-state ("interrupt shadow" as AMD calls it) * from the guest-state area in the VMCS. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static void hmR0VmxSaveGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { uint32_t uIntrState = 0; int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState); AssertRC(rc); if (!uIntrState) VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS); else { Assert( uIntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI || uIntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS); rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx); AssertRC(rc); rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* for hmR0VmxGetGuestIntrState(). */ AssertRC(rc); EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip); Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)); } } /** * Saves the guest's activity state. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestActivityState(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); /* Nothing to do for now until we make use of different guest-CPU activity state. Just update the flag. */ pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_ACTIVITY_STATE; return VINF_SUCCESS; } /** * Saves the guest SYSENTER MSRs (SYSENTER_CS, SYSENTER_EIP, SYSENTER_ESP) from * the current VMCS into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestSysenterMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR)) { uint32_t u32Val = 0; rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRCReturn(rc, rc); pMixedCtx->SysEnter.cs = u32Val; pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR; } uint64_t u64Val = 0; if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR)) { rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, &u64Val); AssertRCReturn(rc, rc); pMixedCtx->SysEnter.eip = u64Val; pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR; } if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR)) { rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, &u64Val); AssertRCReturn(rc, rc); pMixedCtx->SysEnter.esp = u64Val; pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR; } return rc; } /** * Saves the set of guest MSRs (that we restore lazily while leaving VT-x) from * the CPU back into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestLazyMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { if (pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_LAZY_MSRS) return VINF_SUCCESS; #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if ( HMVMX_IS_64BIT_HOST_MODE() && pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests) { /* We should not get preempted to a different CPU at this point while reading the MSRs. */ VMMRZCallRing3Disable(pVCpu); HM_DISABLE_PREEMPT_IF_NEEDED(); hmR0VmxLazySaveGuestMsrs(pVCpu, pMixedCtx); HM_RESTORE_PREEMPT_IF_NEEDED(); VMMRZCallRing3Enable(pVCpu); } #endif pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_LAZY_MSRS; return VINF_SUCCESS; } /** * Saves the auto load/store'd guest MSRs from the current VMCS into * the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestAutoLoadStoreMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { if (pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS) return VINF_SUCCESS; PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr; uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs; Log4(("hmR0VmxSaveGuestAutoLoadStoreMsrs: cMsrs=%u\n", cMsrs)); for (uint32_t i = 0; i < cMsrs; i++, pMsr++) { switch (pMsr->u32Msr) { case MSR_K8_TSC_AUX: CPUMSetGuestMsr(pVCpu, MSR_K8_TSC_AUX, pMsr->u64Value); break; case MSR_K8_LSTAR: pMixedCtx->msrLSTAR = pMsr->u64Value; break; case MSR_K6_STAR: pMixedCtx->msrSTAR = pMsr->u64Value; break; case MSR_K8_SF_MASK: pMixedCtx->msrSFMASK = pMsr->u64Value; break; case MSR_K8_KERNEL_GS_BASE: pMixedCtx->msrKERNELGSBASE = pMsr->u64Value; break; default: { AssertFailed(); return VERR_HM_UNEXPECTED_LD_ST_MSR; } } } pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS; return VINF_SUCCESS; } /** * Saves the guest control registers from the current VMCS into the guest-CPU * context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestControlRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { /* Guest CR0. Guest FPU. */ int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); /* Guest CR4. */ rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); /* Guest CR2 - updated always during the world-switch or in #PF. */ /* Guest CR3. Only changes with Nested Paging. This must be done -after- saving CR0 and CR4 from the guest! */ if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_CR3)) { Assert(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_CR0); Assert(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_CR4); PVM pVM = pVCpu->CTX_SUFF(pVM); if ( pVM->hm.s.vmx.fUnrestrictedGuest || ( pVM->hm.s.fNestedPaging && CPUMIsGuestPagingEnabledEx(pMixedCtx))) { uint64_t u64Val = 0; rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_CR3, &u64Val); if (pMixedCtx->cr3 != u64Val) { CPUMSetGuestCR3(pVCpu, u64Val); if (VMMRZCallRing3IsEnabled(pVCpu)) { PGMUpdateCR3(pVCpu, u64Val); Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3)); } else { /* Set the force flag to inform PGM about it when necessary. It is cleared by PGMUpdateCR3().*/ VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3); } } /* If the guest is in PAE mode, sync back the PDPE's into the guest state. */ if (CPUMIsGuestInPAEModeEx(pMixedCtx)) /* Reads CR0, CR4 and EFER MSR (EFER is always up-to-date). */ { rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u); AssertRCReturn(rc, rc); rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u); AssertRCReturn(rc, rc); rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u); AssertRCReturn(rc, rc); rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u); AssertRCReturn(rc, rc); if (VMMRZCallRing3IsEnabled(pVCpu)) { PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]); Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES)); } else { /* Set the force flag to inform PGM about it when necessary. It is cleared by PGMGstUpdatePaePdpes(). */ VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES); } } } pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_CR3; } /* * Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> hmR0VmxCallRing3Callback() * -> VMMRZCallRing3Disable() -> hmR0VmxSaveGuestState() -> Set VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp * -> continue with VM-exit handling -> hmR0VmxSaveGuestControlRegs() and here we are. * * The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus * if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that * calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should * -NOT- check if HMVMX_UPDATED_GUEST_CR3 is already set or not! * * The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here. */ if (VMMRZCallRing3IsEnabled(pVCpu)) { if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3)) PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu)); if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES)) PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]); Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3)); Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES)); } return rc; } /** * Reads a guest segment register from the current VMCS into the guest-CPU * context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param idxSel Index of the selector in the VMCS. * @param idxLimit Index of the segment limit in the VMCS. * @param idxBase Index of the segment base in the VMCS. * @param idxAccess Index of the access rights of the segment in the VMCS. * @param pSelReg Pointer to the segment selector. * * @remarks No-long-jump zone!!! * @remarks Never call this function directly!!! Use the VMXLOCAL_READ_SEG() * macro as that takes care of whether to read from the VMCS cache or * not. */ DECLINLINE(int) hmR0VmxReadSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase, uint32_t idxAccess, PCPUMSELREG pSelReg) { NOREF(pVCpu); uint32_t u32Val = 0; int rc = VMXReadVmcs32(idxSel, &u32Val); AssertRCReturn(rc, rc); pSelReg->Sel = (uint16_t)u32Val; pSelReg->ValidSel = (uint16_t)u32Val; pSelReg->fFlags = CPUMSELREG_FLAGS_VALID; rc = VMXReadVmcs32(idxLimit, &u32Val); AssertRCReturn(rc, rc); pSelReg->u32Limit = u32Val; uint64_t u64Val = 0; rc = VMXReadVmcsGstNByIdxVal(idxBase, &u64Val); AssertRCReturn(rc, rc); pSelReg->u64Base = u64Val; rc = VMXReadVmcs32(idxAccess, &u32Val); AssertRCReturn(rc, rc); pSelReg->Attr.u = u32Val; /* * If VT-x marks the segment as unusable, most other bits remain undefined: * - For CS the L, D and G bits have meaning. * - For SS the DPL has meaning (it -is- the CPL for Intel and VBox). * - For the remaining data segments no bits are defined. * * The present bit and the unusable bit has been observed to be set at the * same time (the selector was supposed to invalid as we started executing * a V8086 interrupt in ring-0). * * What should be important for the rest of the VBox code that the P bit is * cleared. Some of the other VBox code recognizes the unusable bit, but * AMD-V certainly don't, and REM doesn't really either. So, to be on the * safe side here, we'll strip off P and other bits we don't care about. If * any code breaks because Attr.u != 0 when Sel < 4, it should be fixed. * * See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers". */ if (pSelReg->Attr.u & X86DESCATTR_UNUSABLE) { Assert(idxSel != VMX_VMCS16_GUEST_FIELD_TR); /* TR is the only selector that can never be unusable. */ /* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */ pSelReg->Attr.u &= X86DESCATTR_UNUSABLE | X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G | X86DESCATTR_DPL | X86DESCATTR_TYPE | X86DESCATTR_DT; Log4(("hmR0VmxReadSegmentReg: Unusable idxSel=%#x attr=%#x -> %#x\n", idxSel, u32Val, pSelReg->Attr.u)); #ifdef DEBUG_bird AssertMsg((u32Val & ~X86DESCATTR_P) == pSelReg->Attr.u, ("%#x: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n", idxSel, u32Val, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit)); #endif } return VINF_SUCCESS; } #ifdef VMX_USE_CACHED_VMCS_ACCESSES # define VMXLOCAL_READ_SEG(Sel, CtxSel) \ hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_##Sel, VMX_VMCS32_GUEST_##Sel##_LIMIT, \ VMX_VMCS_GUEST_##Sel##_BASE_CACHE_IDX, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel) #else # define VMXLOCAL_READ_SEG(Sel, CtxSel) \ hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_##Sel, VMX_VMCS32_GUEST_##Sel##_LIMIT, \ VMX_VMCS_GUEST_##Sel##_BASE, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel) #endif /** * Saves the guest segment registers from the current VMCS into the guest-CPU * context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { /* Guest segment registers. */ if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_SEGMENT_REGS)) { int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); rc = VMXLOCAL_READ_SEG(CS, cs); AssertRCReturn(rc, rc); rc = VMXLOCAL_READ_SEG(SS, ss); AssertRCReturn(rc, rc); rc = VMXLOCAL_READ_SEG(DS, ds); AssertRCReturn(rc, rc); rc = VMXLOCAL_READ_SEG(ES, es); AssertRCReturn(rc, rc); rc = VMXLOCAL_READ_SEG(FS, fs); AssertRCReturn(rc, rc); rc = VMXLOCAL_READ_SEG(GS, gs); AssertRCReturn(rc, rc); /* Restore segment attributes for real-on-v86 mode hack. */ if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { pMixedCtx->cs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrCS.u; pMixedCtx->ss.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrSS.u; pMixedCtx->ds.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrDS.u; pMixedCtx->es.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrES.u; pMixedCtx->fs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrFS.u; pMixedCtx->gs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrGS.u; } pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_SEGMENT_REGS; } return VINF_SUCCESS; } /** * Saves the guest descriptor table registers and task register from the current * VMCS into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestTableRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { int rc = VINF_SUCCESS; /* Guest LDTR. */ if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_LDTR)) { rc = VMXLOCAL_READ_SEG(LDTR, ldtr); AssertRCReturn(rc, rc); pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_LDTR; } /* Guest GDTR. */ uint64_t u64Val = 0; uint32_t u32Val = 0; if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_GDTR)) { rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, &u64Val); AssertRCReturn(rc, rc); rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc); pMixedCtx->gdtr.pGdt = u64Val; pMixedCtx->gdtr.cbGdt = u32Val; pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_GDTR; } /* Guest IDTR. */ if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_IDTR)) { rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, &u64Val); AssertRCReturn(rc, rc); rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc); pMixedCtx->idtr.pIdt = u64Val; pMixedCtx->idtr.cbIdt = u32Val; pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_IDTR; } /* Guest TR. */ if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_TR)) { rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); /* For real-mode emulation using virtual-8086 mode we have the fake TSS (pRealModeTSS) in TR, don't save the fake one. */ if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { rc = VMXLOCAL_READ_SEG(TR, tr); AssertRCReturn(rc, rc); } pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_TR; } return rc; } #undef VMXLOCAL_READ_SEG /** * Saves the guest debug-register DR7 from the current VMCS into the guest-CPU * context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestDR7(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { if (!(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_DEBUG)) { if (!pVCpu->hm.s.fUsingHyperDR7) { /* Upper 32-bits are always zero. See Intel spec. 2.7.3 "Loading and Storing Debug Registers". */ uint32_t u32Val; int rc = VMXReadVmcs32(VMX_VMCS_GUEST_DR7, &u32Val); AssertRCReturn(rc, rc); pMixedCtx->dr[7] = u32Val; } pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_DEBUG; } return VINF_SUCCESS; } /** * Saves the guest APIC state from the current VMCS into the guest-CPU context. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); /* Updating TPR is already done in hmR0VmxPostRunGuest(). Just update the flag. */ pVCpu->hm.s.vmx.fUpdatedGuestState |= HMVMX_UPDATED_GUEST_APIC_STATE; return VINF_SUCCESS; } /** * Saves the entire guest state from the currently active VMCS into the * guest-CPU context. This essentially VMREADs all guest-data. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ static int hmR0VmxSaveGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { Assert(pVCpu); Assert(pMixedCtx); if (pVCpu->hm.s.vmx.fUpdatedGuestState == HMVMX_UPDATED_GUEST_ALL) return VINF_SUCCESS; /* Though we can longjmp to ring-3 due to log-flushes here and get recalled again on the ring-3 callback path, there is no real need to. */ if (VMMRZCallRing3IsEnabled(pVCpu)) VMMR0LogFlushDisable(pVCpu); else Assert(VMMR0IsLogFlushDisabled(pVCpu)); Log4Func(("vcpu[%RU32]\n", pVCpu->idCpu)); int rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestRipRspRflags failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestControlRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSegmentRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestTableRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestDR7 failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestSysenterMsrs(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSysenterMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestLazyMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestAutoLoadStoreMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestActivityState(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestActivityState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); rc = hmR0VmxSaveGuestApicState(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestApicState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc); AssertMsg(pVCpu->hm.s.vmx.fUpdatedGuestState == HMVMX_UPDATED_GUEST_ALL, ("Missed guest state bits while saving state; residue %RX32\n", pVCpu->hm.s.vmx.fUpdatedGuestState)); if (VMMRZCallRing3IsEnabled(pVCpu)) VMMR0LogFlushEnable(pVCpu); return rc; } /** * Check per-VM and per-VCPU force flag actions that require us to go back to * ring-3 for one reason or another. * * @returns VBox status code (information status code included). * @retval VINF_SUCCESS if we don't have any actions that require going back to * ring-3. * @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync. * @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware * interrupts) * @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires * all EMTs to be in ring-3. * @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests. * @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return * to the EM loop. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ static int hmR0VmxCheckForceFlags(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx) { Assert(VMMRZCallRing3IsEnabled(pVCpu)); if ( VM_FF_IS_PENDING(pVM, !pVCpu->hm.s.fSingleInstruction ? VM_FF_HP_R0_PRE_HM_MASK : VM_FF_HP_R0_PRE_HM_STEP_MASK) || VMCPU_FF_IS_PENDING(pVCpu, !pVCpu->hm.s.fSingleInstruction ? VMCPU_FF_HP_R0_PRE_HM_MASK : VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) ) { /* We need the control registers now, make sure the guest-CPU context is updated. */ int rc3 = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); AssertRCReturn(rc3, rc3); /* Pending HM CR3 sync. */ if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3)) { int rc2 = PGMUpdateCR3(pVCpu, pMixedCtx->cr3); AssertMsgReturn(rc2 == VINF_SUCCESS || rc2 == VINF_PGM_SYNC_CR3, ("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS); Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3)); } /* Pending HM PAE PDPEs. */ if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES)) { PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]); Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES)); } /* Pending PGM C3 sync. */ if (VMCPU_FF_IS_PENDING(pVCpu,VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL)) { int rc2 = PGMSyncCR3(pVCpu, pMixedCtx->cr0, pMixedCtx->cr3, pMixedCtx->cr4, VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_PGM_SYNC_CR3)); if (rc2 != VINF_SUCCESS) { AssertRC(rc2); Log4(("hmR0VmxCheckForceFlags: PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", rc2)); return rc2; } } /* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */ if ( VM_FF_IS_PENDING(pVM, VM_FF_HM_TO_R3_MASK) || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK)) { STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF); int rc2 = RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_NO_MEMORY : VINF_EM_RAW_TO_R3; Log4(("hmR0VmxCheckForceFlags: HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc2)); return rc2; } /* Pending VM request packets, such as hardware interrupts. */ if ( VM_FF_IS_PENDING(pVM, VM_FF_REQUEST) || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_REQUEST)) { Log4(("hmR0VmxCheckForceFlags: Pending VM request forcing us back to ring-3\n")); return VINF_EM_PENDING_REQUEST; } /* Pending PGM pool flushes. */ if (VM_FF_IS_PENDING(pVM, VM_FF_PGM_POOL_FLUSH_PENDING)) { Log4(("hmR0VmxCheckForceFlags: PGM pool flush pending forcing us back to ring-3\n")); return VINF_PGM_POOL_FLUSH_PENDING; } /* Pending DMA requests. */ if (VM_FF_IS_PENDING(pVM, VM_FF_PDM_DMA)) { Log4(("hmR0VmxCheckForceFlags: Pending DMA request forcing us back to ring-3\n")); return VINF_EM_RAW_TO_R3; } } return VINF_SUCCESS; } /** * Converts any TRPM trap into a pending HM event. This is typically used when * entering from ring-3 (not longjmp returns). * * @param pVCpu Pointer to the VMCPU. */ static void hmR0VmxTrpmTrapToPendingEvent(PVMCPU pVCpu) { Assert(TRPMHasTrap(pVCpu)); Assert(!pVCpu->hm.s.Event.fPending); uint8_t uVector; TRPMEVENT enmTrpmEvent; RTGCUINT uErrCode; RTGCUINTPTR GCPtrFaultAddress; uint8_t cbInstr; int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr); AssertRC(rc); /* Refer Intel spec. 24.8.3 "VM-entry Controls for Event Injection" for the format of u32IntInfo. */ uint32_t u32IntInfo = uVector | VMX_EXIT_INTERRUPTION_INFO_VALID; if (enmTrpmEvent == TRPM_TRAP) { switch (uVector) { case X86_XCPT_BP: case X86_XCPT_OF: u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); break; case X86_XCPT_PF: case X86_XCPT_DF: case X86_XCPT_TS: case X86_XCPT_NP: case X86_XCPT_SS: case X86_XCPT_GP: case X86_XCPT_AC: u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID; /* no break! */ default: u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); break; } } else if (enmTrpmEvent == TRPM_HARDWARE_INT) { if (uVector == X86_XCPT_NMI) u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); else u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); } else if (enmTrpmEvent == TRPM_SOFTWARE_INT) u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); else AssertMsgFailed(("Invalid TRPM event type %d\n", enmTrpmEvent)); rc = TRPMResetTrap(pVCpu); AssertRC(rc); Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n", u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress)); hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress); STAM_COUNTER_DEC(&pVCpu->hm.s.StatInjectPendingReflect); } /** * Converts any pending HM event into a TRPM trap. Typically used when leaving * VT-x to execute any instruction. * * @param pvCpu Pointer to the VMCPU. */ static void hmR0VmxPendingEventToTrpmTrap(PVMCPU pVCpu) { Assert(pVCpu->hm.s.Event.fPending); uint32_t uVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo); uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVCpu->hm.s.Event.u64IntInfo); bool fErrorCodeValid = !!VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVCpu->hm.s.Event.u64IntInfo); uint32_t uErrorCode = pVCpu->hm.s.Event.u32ErrCode; /* If a trap was already pending, we did something wrong! */ Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo */, NULL /* pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP); TRPMEVENT enmTrapType; switch (uVectorType) { case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT: case VMX_IDT_VECTORING_INFO_TYPE_NMI: enmTrapType = TRPM_HARDWARE_INT; break; case VMX_IDT_VECTORING_INFO_TYPE_SW_INT: enmTrapType = TRPM_SOFTWARE_INT; break; case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT: case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT: /* #BP and #OF */ case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT: enmTrapType = TRPM_TRAP; break; default: AssertMsgFailed(("Invalid trap type %#x\n", uVectorType)); enmTrapType = TRPM_32BIT_HACK; break; } Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType)); int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType); AssertRC(rc); if (fErrorCodeValid) TRPMSetErrorCode(pVCpu, uErrorCode); if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT && uVector == X86_XCPT_PF) { TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress); } else if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT || uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT || uVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT) { AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT || (uVector == X86_XCPT_BP || uVector == X86_XCPT_OF), ("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType)); TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr); } pVCpu->hm.s.Event.fPending = false; } /** * Does the necessary state syncing before returning to ring-3 for any reason * (longjmp, preemption, voluntary exits to ring-3) from VT-x. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may * be out-of-sync. Make sure to update the required * fields before using them. * @param fSaveGuestState Whether to save the guest state or not. * * @remarks If you modify code here, make sure to check whether * hmR0VmxCallRing3Callback() needs to be updated too!!! * @remarks No-long-jmp zone!!! */ static int hmR0VmxLeave(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fSaveGuestState) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); Assert(!VMMRZCallRing3IsEnabled(pVCpu)); RTCPUID idCpu = RTMpCpuId(); Log4Func(("HostCpuId=%u\n", idCpu)); /* Save the guest state if necessary. */ if ( fSaveGuestState && pVCpu->hm.s.vmx.fUpdatedGuestState != HMVMX_UPDATED_GUEST_ALL) { int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); Assert(pVCpu->hm.s.vmx.fUpdatedGuestState == HMVMX_UPDATED_GUEST_ALL); } /* Restore host FPU state if necessary and resync on next R0 reentry .*/ if (CPUMIsGuestFPUStateActive(pVCpu)) { /* We shouldn't reload CR0 without saving it first. */ if (!fSaveGuestState) { int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); } CPUMR0SaveGuestFPU(pVM, pVCpu, pMixedCtx); Assert(!CPUMIsGuestFPUStateActive(pVCpu)); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR0); } /* Restore host debug registers if necessary and resync on next R0 reentry. */ #ifdef VBOX_STRICT if (CPUMIsHyperDebugStateActive(pVCpu)) Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT); #endif if (CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */)) VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG); Assert(!CPUMIsGuestDebugStateActive(pVCpu) && !CPUMIsGuestDebugStateActivePending(pVCpu)); Assert(!CPUMIsHyperDebugStateActive(pVCpu) && !CPUMIsHyperDebugStateActivePending(pVCpu)); #if HC_ARCH_BITS == 64 /* Restore host-state bits that VT-x only restores partially. */ if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED) && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED)) { Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags, idCpu)); VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost); pVCpu->hm.s.vmx.fRestoreHostFlags = 0; } #endif #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) /* Restore the host MSRs as we're leaving VT-x context. */ if ( HMVMX_IS_64BIT_HOST_MODE() && pVM->hm.s.fAllow64BitGuests && pVCpu->hm.s.vmx.fRestoreHostMsrs) { /* We shouldn't reload the guest MSRs without saving it first. */ if (!fSaveGuestState) { int rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); } Assert(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_LAZY_MSRS); hmR0VmxLazyRestoreHostMsrs(pVCpu); Assert(!pVCpu->hm.s.vmx.fRestoreHostMsrs); } #endif STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry); STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatLoadGuestState); STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit1); STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit2); STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO); STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx); STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi); STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3); VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC); /** @todo This partially defeats the purpose of having preemption hooks. * The problem is, deregistering the hooks should be moved to a place that * lasts until the EMT is about to be destroyed not everytime while leaving HM * context. */ if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE) { int rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); AssertRCReturn(rc, rc); pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR; Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu)); } Assert(!(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED)); NOREF(idCpu); return VINF_SUCCESS; } /** * Leaves the VT-x session. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jmp zone!!! */ DECLINLINE(int) hmR0VmxLeaveSession(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx) { HM_DISABLE_PREEMPT_IF_NEEDED(); HMVMX_ASSERT_CPU_SAFE(); Assert(!VMMRZCallRing3IsEnabled(pVCpu)); Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); /* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before and done this from the VMXR0ThreadCtxCallback(). */ if (!pVCpu->hm.s.fLeaveDone) { int rc2 = hmR0VmxLeave(pVM, pVCpu, pMixedCtx, true /* fSaveGuestState */); AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT_IF_NEEDED(), rc2); pVCpu->hm.s.fLeaveDone = true; } /* Deregister hook now that we've left HM context before re-enabling preemption. */ /** @todo This is bad. Deregistering here means we need to VMCLEAR always * (longjmp/exit-to-r3) in VT-x which is not efficient. */ if (VMMR0ThreadCtxHooksAreRegistered(pVCpu)) VMMR0ThreadCtxHooksDeregister(pVCpu); /* Leave HM context. This takes care of local init (term). */ int rc = HMR0LeaveCpu(pVCpu); HM_RESTORE_PREEMPT_IF_NEEDED(); return rc; } /** * Does the necessary state syncing before doing a longjmp to ring-3. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jmp zone!!! */ DECLINLINE(int) hmR0VmxLongJmpToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx) { return hmR0VmxLeaveSession(pVM, pVCpu, pMixedCtx); } /** * Take necessary actions before going back to ring-3. * * An action requires us to go back to ring-3. This function does the necessary * steps before we can safely return to ring-3. This is not the same as longjmps * to ring-3, this is voluntary and prepares the guest so it may continue * executing outside HM (recompiler/IEM). * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * @param rcExit The reason for exiting to ring-3. Can be * VINF_VMM_UNKNOWN_RING3_CALL. */ static int hmR0VmxExitToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, int rcExit) { Assert(pVM); Assert(pVCpu); Assert(pMixedCtx); HMVMX_ASSERT_PREEMPT_SAFE(); if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR)) { VMXGetActivatedVmcs(&pVCpu->hm.s.vmx.LastError.u64VMCSPhys); pVCpu->hm.s.vmx.LastError.u32VMCSRevision = *(uint32_t *)pVCpu->hm.s.vmx.pvVmcs; pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu; /* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */ } /* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */ VMMRZCallRing3Disable(pVCpu); Log4(("hmR0VmxExitToRing3: pVCpu=%p idCpu=%RU32 rcExit=%d\n", pVCpu, pVCpu->idCpu, rcExit)); /* We need to do this only while truly exiting the "inner loop" back to ring-3 and -not- for any longjmp to ring3. */ if (pVCpu->hm.s.Event.fPending) { hmR0VmxPendingEventToTrpmTrap(pVCpu); Assert(!pVCpu->hm.s.Event.fPending); } /* Save guest state and restore host state bits. */ int rc = hmR0VmxLeaveSession(pVM, pVCpu, pMixedCtx); AssertRCReturn(rc, rc); STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3); /* Sync recompiler state. */ VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3); CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR | CPUM_CHANGED_LDTR | CPUM_CHANGED_GDTR | CPUM_CHANGED_IDTR | CPUM_CHANGED_TR | CPUM_CHANGED_HIDDEN_SEL_REGS); Assert(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_CR0); if ( pVM->hm.s.fNestedPaging && CPUMIsGuestPagingEnabledEx(pMixedCtx)) { CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH); } Assert(!pVCpu->hm.s.fClearTrapFlag); /* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */ if (rcExit != VINF_EM_RAW_INTERRUPT) VMCPU_HMCF_SET(pVCpu, HM_CHANGED_ALL_GUEST); STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3); /* We do -not- want any longjmp notifications after this! We must return to ring-3 ASAP. */ VMMRZCallRing3RemoveNotification(pVCpu); VMMRZCallRing3Enable(pVCpu); return rc; } /** * VMMRZCallRing3() callback wrapper which saves the guest state before we * longjump to ring-3 and possibly get preempted. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param enmOperation The operation causing the ring-3 longjump. * @param pvUser Opaque pointer to the guest-CPU context. The data * may be out-of-sync. Make sure to update the required * fields before using them. * * @remarks If you modify code here, make sure to check whether * hmR0VmxLeave() needs to be updated too!!! */ DECLCALLBACK(int) hmR0VmxCallRing3Callback(PVMCPU pVCpu, VMMCALLRING3 enmOperation, void *pvUser) { if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION) { /* If anything here asserts or fails, good luck. */ VMMRZCallRing3RemoveNotification(pVCpu); VMMRZCallRing3Disable(pVCpu); HM_DISABLE_PREEMPT_IF_NEEDED(); PVM pVM = pVCpu->CTX_SUFF(pVM); if (CPUMIsGuestFPUStateActive(pVCpu)) CPUMR0SaveGuestFPU(pVM, pVCpu, (PCPUMCTX)pvUser); CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */); #if HC_ARCH_BITS == 64 /* Restore host-state bits that VT-x only restores partially. */ if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED) && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED)) { VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost); pVCpu->hm.s.vmx.fRestoreHostFlags = 0; } #endif #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) /* Restore the host MSRs as we're leaving VT-x context. */ if ( HMVMX_IS_64BIT_HOST_MODE() && pVM->hm.s.fAllow64BitGuests && pVCpu->hm.s.vmx.fRestoreHostMsrs) { hmR0VmxLazyRestoreHostMsrs(pVCpu); } #endif VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC); if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE) { VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR; } if (VMMR0ThreadCtxHooksAreRegistered(pVCpu)) VMMR0ThreadCtxHooksDeregister(pVCpu); HMR0LeaveCpu(pVCpu); HM_RESTORE_PREEMPT_IF_NEEDED(); return VINF_SUCCESS; } Assert(pVCpu); Assert(pvUser); Assert(VMMRZCallRing3IsEnabled(pVCpu)); HMVMX_ASSERT_PREEMPT_SAFE(); VMMRZCallRing3Disable(pVCpu); Assert(VMMR0IsLogFlushDisabled(pVCpu)); Log4(("hmR0VmxCallRing3Callback->hmR0VmxLongJmpToRing3 pVCpu=%p idCpu=%RU32\n enmOperation=%d", pVCpu, pVCpu->idCpu, enmOperation)); int rc = hmR0VmxLongJmpToRing3(pVCpu->CTX_SUFF(pVM), pVCpu, (PCPUMCTX)pvUser); AssertRCReturn(rc, rc); VMMRZCallRing3Enable(pVCpu); return VINF_SUCCESS; } /** * Sets the interrupt-window exiting control in the VMCS which instructs VT-x to * cause a VM-exit as soon as the guest is in a state to receive interrupts. * * @param pVCpu Pointer to the VMCPU. */ DECLINLINE(void) hmR0VmxSetIntWindowExitVmcs(PVMCPU pVCpu) { if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT)) { if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT)) { pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT; int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc); } } /* else we will deliver interrupts whenever the guest exits next and is in a state to receive events. */ } /** * Evaluates the event to be delivered to the guest and sets it as the pending * event. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ static void hmR0VmxEvaluatePendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { Assert(!pVCpu->hm.s.Event.fPending); /* Get the current interruptibility-state of the guest and then figure out what can be injected. */ uint32_t uIntrState = hmR0VmxGetGuestIntrState(pVCpu, pMixedCtx); bool fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS); bool fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI); Assert(!fBlockSti || (pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_RFLAGS)); Assert( !(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI) /* We don't support block-by-NMI and SMI yet.*/ && !(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); Assert(!fBlockSti || pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */ Assert(!TRPMHasTrap(pVCpu)); /** @todo SMI. SMIs take priority over NMIs. */ if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_NMI)) /* NMI. NMIs take priority over regular interrupts . */ { /* On some CPUs block-by-STI also blocks NMIs. See Intel spec. 26.3.1.5 "Checks On Guest Non-Register State". */ if ( !fBlockMovSS && !fBlockSti) { /* On some CPUs block-by-STI also blocks NMIs. See Intel spec. 26.3.1.5 "Checks On Guest Non-Register State". */ Log4(("Pending NMI vcpu[%RU32]\n", pVCpu->idCpu)); uint32_t u32IntInfo = X86_XCPT_NMI | VMX_EXIT_INTERRUPTION_INFO_VALID; u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddres */); VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI); } else hmR0VmxSetIntWindowExitVmcs(pVCpu); } else if ( VMCPU_FF_IS_PENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC)) && !pVCpu->hm.s.fSingleInstruction) { /* * Check if the guest can receive external interrupts (PIC/APIC). Once we do PDMGetInterrupt() we -must- deliver * the interrupt ASAP. We must not execute any guest code until we inject the interrupt which is why it is * evaluated here and not set as pending, solely based on the force-flags. */ int rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); AssertRC(rc); const bool fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF); if ( !fBlockInt && !fBlockSti && !fBlockMovSS) { uint8_t u8Interrupt; rc = PDMGetInterrupt(pVCpu, &u8Interrupt); if (RT_SUCCESS(rc)) { Log4(("Pending interrupt vcpu[%RU32] u8Interrupt=%#x \n", pVCpu->idCpu, u8Interrupt)); uint32_t u32IntInfo = u8Interrupt | VMX_EXIT_INTERRUPTION_INFO_VALID; u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrfaultAddress */); } else { /** @todo Does this actually happen? If not turn it into an assertion. */ Assert(!VMCPU_FF_IS_PENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC))); STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq); } } else hmR0VmxSetIntWindowExitVmcs(pVCpu); } } /** * Injects any pending events into the guest if the guest is in a state to * receive them. * * @returns VBox status code (informational status codes included). * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ static int hmR0VmxInjectPendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { HMVMX_ASSERT_PREEMPT_SAFE(); Assert(VMMRZCallRing3IsEnabled(pVCpu)); /* Get the current interruptibility-state of the guest and then figure out what can be injected. */ uint32_t uIntrState = hmR0VmxGetGuestIntrState(pVCpu, pMixedCtx); bool fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS); bool fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI); Assert(!fBlockSti || (pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_RFLAGS)); Assert( !(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI) /* We don't support block-by-NMI and SMI yet.*/ && !(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); Assert(!fBlockSti || pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */ Assert(!TRPMHasTrap(pVCpu)); int rc = VINF_SUCCESS; if (pVCpu->hm.s.Event.fPending) { #if defined(VBOX_STRICT) || defined(VBOX_WITH_STATISTICS) uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo); if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT) { rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); const bool fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF); Assert(!fBlockInt); Assert(!fBlockSti); Assert(!fBlockMovSS); } else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI) { Assert(!fBlockSti); Assert(!fBlockMovSS); } #endif Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo)); rc = hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.cbInstr, pVCpu->hm.s.Event.u32ErrCode, pVCpu->hm.s.Event.GCPtrFaultAddress, &uIntrState); AssertRCReturn(rc, rc); /* Update the interruptibility-state as it could have been changed by hmR0VmxInjectEventVmcs() (e.g. real-on-v86 guest injecting software interrupts) */ fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS); fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI); #ifdef VBOX_WITH_STATISTICS if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT) STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt); else STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt); #endif } /* Delivery pending debug exception if the guest is single-stepping. Evaluate and set the BS bit. */ if ( !pVCpu->hm.s.fSingleInstruction && !DBGFIsStepping(pVCpu)) { int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); AssertRCReturn(rc2, rc2); if (pMixedCtx->eflags.Bits.u1TF) /* We don't have any IA32_DEBUGCTL MSR for guests. Treat as all bits 0. */ { /* * The pending-debug exceptions field is cleared on all VM-exits except VMX_EXIT_TPR_BELOW_THRESHOLD, * VMX_EXIT_MTF, VMX_EXIT_APIC_WRITE and VMX_EXIT_VIRTUALIZED_EOI. * See Intel spec. 27.3.4 "Saving Non-Register State". */ rc2 = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, VMX_VMCS_GUEST_DEBUG_EXCEPTIONS_BS); AssertRCReturn(rc2, rc2); } } else { /* We are single-stepping in the hypervisor debugger, clear interrupt inhibition as setting the BS bit would mean delivering a #DB to the guest upon VM-entry when it shouldn't be. */ uIntrState = 0; } /* * There's no need to clear the VM entry-interruption information field here if we're not injecting anything. * VT-x clears the valid bit on every VM-exit. See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection". */ int rc2 = hmR0VmxLoadGuestIntrState(pVCpu, uIntrState); AssertRC(rc2); Assert(rc == VINF_SUCCESS || rc == VINF_EM_RESET); NOREF(fBlockMovSS); NOREF(fBlockSti); return rc; } /** * Sets an invalid-opcode (#UD) exception as pending-for-injection into the VM. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); uint32_t u32IntInfo = X86_XCPT_UD | VMX_EXIT_INTERRUPTION_INFO_VALID; hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */); } /** * Injects a double-fault (#DF) exception into the VM. * * @returns VBox status code (informational status code included). * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ DECLINLINE(int) hmR0VmxInjectXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t *puIntrState) { uint32_t u32IntInfo = X86_XCPT_DF | VMX_EXIT_INTERRUPTION_INFO_VALID; u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID; return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */, puIntrState); } /** * Sets a debug (#DB) exception as pending-for-injection into the VM. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx) { NOREF(pMixedCtx); uint32_t u32IntInfo = X86_XCPT_DB | VMX_EXIT_INTERRUPTION_INFO_VALID; u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */); } /** * Sets an overflow (#OF) exception as pending-for-injection into the VM. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * @param cbInstr The value of RIP that is to be pushed on the guest * stack. */ DECLINLINE(void) hmR0VmxSetPendingXcptOF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t cbInstr) { NOREF(pMixedCtx); uint32_t u32IntInfo = X86_XCPT_OF | VMX_EXIT_INTERRUPTION_INFO_VALID; u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */); } /** * Injects a general-protection (#GP) fault into the VM. * * @returns VBox status code (informational status code included). * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * @param u32ErrorCode The error code associated with the #GP. */ DECLINLINE(int) hmR0VmxInjectXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fErrorCodeValid, uint32_t u32ErrorCode, uint32_t *puIntrState) { uint32_t u32IntInfo = X86_XCPT_GP | VMX_EXIT_INTERRUPTION_INFO_VALID; u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); if (fErrorCodeValid) u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID; return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr */, u32ErrorCode, 0 /* GCPtrFaultAddress */, puIntrState); } /** * Sets a software interrupt (INTn) as pending-for-injection into the VM. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * @param uVector The software interrupt vector number. * @param cbInstr The value of RIP that is to be pushed on the guest * stack. */ DECLINLINE(void) hmR0VmxSetPendingIntN(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint16_t uVector, uint32_t cbInstr) { NOREF(pMixedCtx); uint32_t u32IntInfo = uVector | VMX_EXIT_INTERRUPTION_INFO_VALID; if ( uVector == X86_XCPT_BP || uVector == X86_XCPT_OF) { u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); } else u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT); hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */); } /** * Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's * stack. * * @returns VBox status code (information status code included). * @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault. * @param pVM Pointer to the VM. * @param pMixedCtx Pointer to the guest-CPU context. * @param uValue The value to push to the guest stack. */ DECLINLINE(int) hmR0VmxRealModeGuestStackPush(PVM pVM, PCPUMCTX pMixedCtx, uint16_t uValue) { /* * The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in * virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers". * See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound". */ if (pMixedCtx->sp == 1) return VINF_EM_RESET; pMixedCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */ int rc = PGMPhysSimpleWriteGCPhys(pVM, pMixedCtx->ss.u64Base + pMixedCtx->sp, &uValue, sizeof(uint16_t)); AssertRCReturn(rc, rc); return rc; } /** * Injects an event into the guest upon VM-entry by updating the relevant fields * in the VM-entry area in the VMCS. * * @returns VBox status code (informational error codes included). * @retval VINF_SUCCESS if the event is successfully injected into the VMCS. * @retval VINF_EM_RESET if event injection resulted in a triple-fault. * * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may * be out-of-sync. Make sure to update the required * fields before using them. * @param u64IntInfo The VM-entry interruption-information field. * @param cbInstr The VM-entry instruction length in bytes (for * software interrupts, exceptions and privileged * software exceptions). * @param u32ErrCode The VM-entry exception error code. * @param GCPtrFaultAddress The page-fault address for #PF exceptions. * @param puIntrState Pointer to the current guest interruptibility-state. * This interruptibility-state will be updated if * necessary. This cannot not be NULL. * * @remarks Requires CR0! * @remarks No-long-jump zone!!! */ static int hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr, uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress, uint32_t *puIntrState) { /* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */ AssertMsg(u64IntInfo >> 32 == 0, ("%#RX64\n", u64IntInfo)); Assert(puIntrState); uint32_t u32IntInfo = (uint32_t)u64IntInfo; const uint32_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(u32IntInfo); const uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo); #ifdef VBOX_STRICT /* Validate the error-code-valid bit for hardware exceptions. */ if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT) { switch (uVector) { case X86_XCPT_PF: case X86_XCPT_DF: case X86_XCPT_TS: case X86_XCPT_NP: case X86_XCPT_SS: case X86_XCPT_GP: case X86_XCPT_AC: AssertMsg(VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo), ("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector)); /* fallthru */ default: break; } } #endif /* Cannot inject an NMI when block-by-MOV SS is in effect. */ Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI || !(*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS)); STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]); /* We require CR0 to check if the guest is in real-mode. */ int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); /* * Hardware interrupts & exceptions cannot be delivered through the software interrupt redirection bitmap to the real * mode task in virtual-8086 mode. We must jump to the interrupt handler in the (real-mode) guest. * See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode" for interrupt & exception classes. * See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling. */ if (CPUMIsGuestInRealModeEx(pMixedCtx)) { PVM pVM = pVCpu->CTX_SUFF(pVM); if (!pVM->hm.s.vmx.fUnrestrictedGuest) { Assert(PDMVmmDevHeapIsEnabled(pVM)); Assert(pVM->hm.s.vmx.pRealModeTSS); /* We require RIP, RSP, RFLAGS, CS, IDTR. Save the required ones from the VMCS. */ rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); Assert(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_RIP); /* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */ const size_t cbIdtEntry = sizeof(X86IDTR16); if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pMixedCtx->idtr.cbIdt) { /* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */ if (uVector == X86_XCPT_DF) return VINF_EM_RESET; else if (uVector == X86_XCPT_GP) { /* If we're injecting a #GP with no valid IDT entry, inject a double-fault. */ return hmR0VmxInjectXcptDF(pVCpu, pMixedCtx, puIntrState); } /* If we're injecting an interrupt/exception with no valid IDT entry, inject a general-protection fault. */ /* No error codes for exceptions in real-mode. See Intel spec. 20.1.4 "Interrupt and Exception Handling" */ return hmR0VmxInjectXcptGP(pVCpu, pMixedCtx, false /* fErrCodeValid */, 0 /* u32ErrCode */, puIntrState); } /* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */ uint16_t uGuestIp = pMixedCtx->ip; if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT) { Assert(uVector == X86_XCPT_BP || uVector == X86_XCPT_OF); /* #BP and #OF are both benign traps, we need to resume the next instruction. */ uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr; } else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT) uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr; /* Get the code segment selector and offset from the IDT entry for the interrupt handler. */ X86IDTR16 IdtEntry; RTGCPHYS GCPhysIdtEntry = (RTGCPHYS)pMixedCtx->idtr.pIdt + uVector * cbIdtEntry; rc = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry); AssertRCReturn(rc, rc); /* Construct the stack frame for the interrupt/exception handler. */ rc = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->eflags.u32); rc |= hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->cs.Sel); rc |= hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, uGuestIp); AssertRCReturn(rc, rc); /* Clear the required eflag bits and jump to the interrupt/exception handler. */ if (rc == VINF_SUCCESS) { pMixedCtx->eflags.u32 &= ~(X86_EFL_IF | X86_EFL_TF | X86_EFL_RF | X86_EFL_AC); pMixedCtx->rip = IdtEntry.offSel; pMixedCtx->cs.Sel = IdtEntry.uSel; pMixedCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry; if ( uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT && uVector == X86_XCPT_PF) { pMixedCtx->cr2 = GCPtrFaultAddress; } /* If any other guest-state bits are changed here, make sure to update hmR0VmxPreRunGuestCommitted() when thread-context hooks are used. */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS | HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS | HM_CHANGED_GUEST_RSP); /* We're clearing interrupts, which means no block-by-STI interrupt-inhibition. */ if (*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI) { Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI && uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT); Log4(("Clearing inhibition due to STI.\n")); *puIntrState &= ~VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI; } Log4(("Injecting real-mode: u32IntInfo=%#x u32ErrCode=%#x instrlen=%#x\n", u32IntInfo, u32ErrCode, cbInstr)); /* The event has been truly dispatched. Mark it as no longer pending so we don't attempt to 'undo' it, if we are returning to ring-3 before executing guest code. */ pVCpu->hm.s.Event.fPending = false; } Assert(rc == VINF_SUCCESS || rc == VINF_EM_RESET); return rc; } else { /* * For unrestricted execution enabled CPUs running real-mode guests, we must not set the deliver-error-code bit. * See Intel spec. 26.2.1.3 "VM-Entry Control Fields". */ u32IntInfo &= ~VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID; } } /* Validate. */ Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */ Assert(!VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK(u32IntInfo)); /* Bit 12 MBZ. */ Assert(!(u32IntInfo & 0x7ffff000)); /* Bits 30:12 MBZ. */ /* Inject. */ rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo); if (VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo)) rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode); rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr); if ( VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT && uVector == X86_XCPT_PF) { pMixedCtx->cr2 = GCPtrFaultAddress; } Log4(("Injecting vcpu[%RU32] u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x pMixedCtx->uCR2=%#RX64\n", pVCpu->idCpu, u32IntInfo, u32ErrCode, cbInstr, pMixedCtx->cr2)); AssertRCReturn(rc, rc); return rc; } /** * Clears the interrupt-window exiting control in the VMCS and if necessary * clears the current event in the VMCS as well. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * * @remarks Use this function only to clear events that have not yet been * delivered to the guest but are injected in the VMCS! * @remarks No-long-jump zone!!! */ static void hmR0VmxClearEventVmcs(PVMCPU pVCpu) { int rc; Log4Func(("vcpu[%d]\n", pVCpu->idCpu)); /* Clear interrupt-window exiting control. */ if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT) { pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT; rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc); } if (!pVCpu->hm.s.Event.fPending) return; #ifdef VBOX_STRICT uint32_t u32EntryInfo; rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo); AssertRC(rc); Assert(VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)); #endif /* Clear the entry-interruption field (including the valid bit). */ rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0); AssertRC(rc); /* Clear the pending debug exception field. */ rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0); AssertRC(rc); } /** * Enters the VT-x session. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCpu Pointer to the CPU info struct. */ VMMR0DECL(int) VMXR0Enter(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu) { AssertPtr(pVM); AssertPtr(pVCpu); Assert(pVM->hm.s.vmx.fSupported); Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); NOREF(pCpu); NOREF(pVM); LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu)); Assert(VMCPU_HMCF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE)); #ifdef VBOX_STRICT /* Make sure we're in VMX root mode. */ RTCCUINTREG u32HostCR4 = ASMGetCR4(); if (!(u32HostCR4 & X86_CR4_VMXE)) { LogRel(("VMXR0Enter: X86_CR4_VMXE bit in CR4 is not set!\n")); return VERR_VMX_X86_CR4_VMXE_CLEARED; } #endif /* * Load the VCPU's VMCS as the current (and active) one. */ Assert(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR); int rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); if (RT_FAILURE(rc)) return rc; pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE; pVCpu->hm.s.fLeaveDone = false; Log4Func(("Activated Vmcs. HostCpuId=%u\n", RTMpCpuId())); return VINF_SUCCESS; } /** * The thread-context callback (only on platforms which support it). * * @param enmEvent The thread-context event. * @param pVCpu Pointer to the VMCPU. * @param fGlobalInit Whether global VT-x/AMD-V init. was used. * @thread EMT(pVCpu) */ VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPU pVCpu, bool fGlobalInit) { NOREF(fGlobalInit); switch (enmEvent) { case RTTHREADCTXEVENT_PREEMPTING: { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); Assert(VMMR0ThreadCtxHooksAreRegistered(pVCpu)); VMCPU_ASSERT_EMT(pVCpu); PVM pVM = pVCpu->CTX_SUFF(pVM); PCPUMCTX pMixedCtx = CPUMQueryGuestCtxPtr(pVCpu); /* No longjmps (logger flushes, locks) in this fragile context. */ VMMRZCallRing3Disable(pVCpu); Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId())); /* * Restore host-state (FPU, debug etc.) */ if (!pVCpu->hm.s.fLeaveDone) { /* Do -not- save guest-state here as we might already be in the middle of saving it (esp. bad if we are holding the PGM lock while saving the guest state (see hmR0VmxSaveGuestControlRegs()). */ hmR0VmxLeave(pVM, pVCpu, pMixedCtx, false /* fSaveGuestState */); pVCpu->hm.s.fLeaveDone = true; } /* Leave HM context, takes care of local init (term). */ int rc = HMR0LeaveCpu(pVCpu); AssertRC(rc); NOREF(rc); /* Restore longjmp state. */ VMMRZCallRing3Enable(pVCpu); STAM_COUNTER_INC(&pVCpu->hm.s.StatPreemptPreempting); break; } case RTTHREADCTXEVENT_RESUMED: { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); Assert(VMMR0ThreadCtxHooksAreRegistered(pVCpu)); VMCPU_ASSERT_EMT(pVCpu); /* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */ VMMRZCallRing3Disable(pVCpu); Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId())); /* Initialize the bare minimum state required for HM. This takes care of initializing VT-x if necessary (onlined CPUs, local init etc.) */ int rc = HMR0EnterCpu(pVCpu); AssertRC(rc); Assert(VMCPU_HMCF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE)); /* Load the active VMCS as the current one. */ if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR) { rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs); AssertRC(rc); NOREF(rc); pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE; Log4Func(("Resumed: Activated Vmcs. HostCpuId=%u\n", RTMpCpuId())); } pVCpu->hm.s.fLeaveDone = false; /* Restore longjmp state. */ VMMRZCallRing3Enable(pVCpu); break; } default: break; } } /** * Saves the host state in the VMCS host-state. * Sets up the VM-exit MSR-load area. * * The CPU state will be loaded from these fields on every successful VM-exit. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * * @remarks No-long-jump zone!!! */ static int hmR0VmxSaveHostState(PVM pVM, PVMCPU pVCpu) { Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); if (!VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT)) return VINF_SUCCESS; int rc = hmR0VmxSaveHostControlRegs(pVM, pVCpu); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostControlRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); rc = hmR0VmxSaveHostSegmentRegs(pVM, pVCpu); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostSegmentRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); rc = hmR0VmxSaveHostMsrs(pVM, pVCpu); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostMsrs failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_HOST_CONTEXT); return rc; } /** * Saves the host state in the VMCS host-state. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * * @remarks No-long-jump zone!!! */ VMMR0DECL(int) VMXR0SaveHostState(PVM pVM, PVMCPU pVCpu) { AssertPtr(pVM); AssertPtr(pVCpu); LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu)); /* Save the host state here while entering HM context. When thread-context hooks are used, we might get preempted and have to resave the host state but most of the time we won't be, so do it here before we disable interrupts. */ Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); return hmR0VmxSaveHostState(pVM, pVCpu); } /** * Loads the guest state into the VMCS guest-state area. The CPU state will be * loaded from these fields on every successful VM-entry. * * Sets up the VM-entry MSR-load and VM-exit MSR-store areas. * Sets up the VM-entry controls. * Sets up the appropriate VMX non-root function to execute guest code based on * the guest CPU mode. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * * @remarks No-long-jump zone!!! */ static int hmR0VmxLoadGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx) { AssertPtr(pVM); AssertPtr(pVCpu); AssertPtr(pMixedCtx); HMVMX_ASSERT_PREEMPT_SAFE(); #ifdef LOG_ENABLED /** @todo r=ramshankar: I'm not able to use VMMRZCallRing3Disable() here, * probably not initialized yet? Anyway this will do for now. * * Update: Should be possible once VMXR0LoadGuestState() is removed as an * interface and disable ring-3 calls when thread-context hooks are not * available. */ bool fCallerDisabledLogFlush = VMMR0IsLogFlushDisabled(pVCpu); VMMR0LogFlushDisable(pVCpu); #endif LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu)); STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestState, x); /* Determine real-on-v86 mode. */ pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = false; if ( !pVM->hm.s.vmx.fUnrestrictedGuest && CPUMIsGuestInRealModeEx(pMixedCtx)) { pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = true; } /* * Load the guest-state into the VMCS. * Any ordering dependency among the sub-functions below must be explicitly stated using comments. * Ideally, assert that the cross-dependent bits are up to date at the point of using it. */ int rc = hmR0VmxSetupVMRunHandler(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupVMRunHandler! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); /* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-entry control updates. */ rc = hmR0VmxLoadGuestEntryCtls(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestEntryCtls! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); /* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-exit control updates. */ rc = hmR0VmxLoadGuestExitCtls(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupExitCtls failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); rc = hmR0VmxLoadGuestActivityState(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestActivityState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); rc = hmR0VmxLoadGuestCR3AndCR4(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestCR3AndCR4: rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); /* Assumes pMixedCtx->cr0 is up-to-date (strict builds require CR0 for segment register validation checks). */ rc = hmR0VmxLoadGuestSegmentRegs(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestSegmentRegs: rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); rc = hmR0VmxLoadGuestMsrs(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadSharedMsrs! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); rc = hmR0VmxLoadGuestApicState(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestApicState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); /* * Loading Rflags here is fine, even though Rflags.TF might depend on guest debug state (which is not loaded here). * It is re-evaluated and updated if necessary in hmR0VmxLoadSharedState(). */ rc = hmR0VmxLoadGuestRipRspRflags(pVCpu, pMixedCtx); AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestRipRspRflags! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc); /* Clear any unused and reserved bits. */ VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR2); #ifdef LOG_ENABLED /* Only reenable log-flushing if the caller has it enabled. */ if (!fCallerDisabledLogFlush) VMMR0LogFlushEnable(pVCpu); #endif STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestState, x); return rc; } /** * Loads the state shared between the host and guest into the VMCS. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCtx Pointer to the guest-CPU context. * * @remarks No-long-jump zone!!! */ static void hmR0VmxLoadSharedState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx) { NOREF(pVM); Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); Assert(!VMMRZCallRing3IsEnabled(pVCpu)); if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0)) { int rc = hmR0VmxLoadSharedCR0(pVCpu, pCtx); AssertRC(rc); } if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG)) { int rc = hmR0VmxLoadSharedDebugState(pVCpu, pCtx); AssertRC(rc); /* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS)) { rc = hmR0VmxLoadGuestRflags(pVCpu, pCtx); AssertRC(rc); } } if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS)) { #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if ( HMVMX_IS_64BIT_HOST_MODE() && pVM->hm.s.fAllow64BitGuests) { hmR0VmxLazyLoadGuestMsrs(pVCpu, pCtx); } #endif VMCPU_HMCF_CLEAR(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS); } AssertMsg(!VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE), ("fContextUseFlags=%#RX32\n", VMCPU_HMCF_VALUE(pVCpu))); } /** * Worker for loading the guest-state bits in the inner VT-x execution loop. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. */ DECLINLINE(void) hmR0VmxLoadGuestStateOptimal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx) { HMVMX_ASSERT_PREEMPT_SAFE(); Log5(("LoadFlags=%#RX32\n", VMCPU_HMCF_VALUE(pVCpu))); #ifdef HMVMX_SYNC_FULL_GUEST_STATE VMCPU_HMCF_SET(pVCpu, HM_CHANGED_ALL_GUEST); #endif if (VMCPU_HMCF_IS_SET_ONLY(pVCpu, HM_CHANGED_GUEST_RIP)) { int rc = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx); AssertRC(rc); STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadMinimal); } else if (VMCPU_HMCF_VALUE(pVCpu)) { int rc = hmR0VmxLoadGuestState(pVM, pVCpu, pMixedCtx); AssertRC(rc); STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadFull); } /* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */ AssertMsg( !VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_ALL_GUEST) || VMCPU_HMCF_IS_PENDING_ONLY(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE), ("fContextUseFlags=%#RX32\n", VMCPU_HMCF_VALUE(pVCpu))); #ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVM, pVCpu, pMixedCtx); if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND) Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason)); #endif } /** * Does the preparations before executing guest code in VT-x. * * This may cause longjmps to ring-3 and may even result in rescheduling to the * recompiler. We must be cautious what we do here regarding committing * guest-state information into the VMCS assuming we assuredly execute the * guest in VT-x mode. If we fall back to the recompiler after updating the VMCS * and clearing the common-state (TRPM/forceflags), we must undo those changes * so that the recompiler can (and should) use them when it resumes guest * execution. Otherwise such operations must be done when we can no longer * exit to ring-3. * * @returns Strict VBox status code. * @retval VINF_SUCCESS if we can proceed with running the guest, interrupts * have been disabled. * @retval VINF_EM_RESET if a triple-fault occurs while injecting a * double-fault into the guest. * @retval VINF_* scheduling changes, we have to go back to ring-3. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * @param pVmxTransient Pointer to the VMX transient structure. * * @remarks Called with preemption disabled. In the VINF_SUCCESS return case * interrupts will be disabled. */ static int hmR0VmxPreRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { Assert(VMMRZCallRing3IsEnabled(pVCpu)); #ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 PGMRZDynMapFlushAutoSet(pVCpu); #endif /* Check force flag actions that might require us to go back to ring-3. */ int rc = hmR0VmxCheckForceFlags(pVM, pVCpu, pMixedCtx); if (rc != VINF_SUCCESS) return rc; #ifndef IEM_VERIFICATION_MODE_FULL /* Setup the Virtualized APIC accesses. pMixedCtx->msrApicBase is always up-to-date. It's not part of the VMCS. */ if ( pVCpu->hm.s.vmx.u64MsrApicBase != pMixedCtx->msrApicBase && (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)) { Assert(pVM->hm.s.vmx.HCPhysApicAccess); RTGCPHYS GCPhysApicBase; GCPhysApicBase = pMixedCtx->msrApicBase; GCPhysApicBase &= PAGE_BASE_GC_MASK; /* Unalias any existing mapping. */ rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase); AssertRCReturn(rc, rc); /* Map the HC APIC-access page into the GC space, this also updates the shadow page tables if necessary. */ Log4(("Mapped HC APIC-access page into GC: GCPhysApicBase=%#RGv\n", GCPhysApicBase)); rc = IOMMMIOMapMMIOHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hm.s.vmx.HCPhysApicAccess, X86_PTE_RW | X86_PTE_P); AssertRCReturn(rc, rc); pVCpu->hm.s.vmx.u64MsrApicBase = pMixedCtx->msrApicBase; } #endif /* !IEM_VERIFICATION_MODE_FULL */ /* Load the guest state bits, we can handle longjmps/getting preempted here. */ hmR0VmxLoadGuestStateOptimal(pVM, pVCpu, pMixedCtx); /* * Evaluate events as pending-for-injection into the guest. Toggling of force-flags here is safe as long as * we update TRPM on premature exits to ring-3 before executing guest code. We must NOT restore the force-flags. */ if (TRPMHasTrap(pVCpu)) hmR0VmxTrpmTrapToPendingEvent(pVCpu); else if (!pVCpu->hm.s.Event.fPending) hmR0VmxEvaluatePendingEvent(pVCpu, pMixedCtx); /* * Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus needs to be done with * longjmps or interrupts + preemption enabled. Event injection might also result in triple-faulting the VM. */ rc = hmR0VmxInjectPendingEvent(pVCpu, pMixedCtx); if (RT_UNLIKELY(rc != VINF_SUCCESS)) { Assert(rc == VINF_EM_RESET); return rc; } /* * No longjmps to ring-3 from this point on!!! * Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic. * This also disables flushing of the R0-logger instance (if any). */ VMMRZCallRing3Disable(pVCpu); /* * We disable interrupts so that we don't miss any interrupts that would flag preemption (IPI/timers etc.) * when thread-context hooks aren't used and we've been running with preemption disabled for a while. * * We need to check for force-flags that could've possible been altered since we last checked them (e.g. * by PDMGetInterrupt() leaving the PDM critical section, see @bugref{6398}). * * We also check a couple of other force-flags as a last opportunity to get the EMT back to ring-3 before * executing guest code. */ pVmxTransient->uEflags = ASMIntDisableFlags(); if ( VM_FF_IS_PENDING(pVM, VM_FF_EMT_RENDEZVOUS | VM_FF_TM_VIRTUAL_SYNC) || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK)) { hmR0VmxClearEventVmcs(pVCpu); ASMSetFlags(pVmxTransient->uEflags); VMMRZCallRing3Enable(pVCpu); STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF); return VINF_EM_RAW_TO_R3; } if (RTThreadPreemptIsPending(NIL_RTTHREAD)) { hmR0VmxClearEventVmcs(pVCpu); ASMSetFlags(pVmxTransient->uEflags); VMMRZCallRing3Enable(pVCpu); STAM_COUNTER_INC(&pVCpu->hm.s.StatPendingHostIrq); return VINF_EM_RAW_INTERRUPT; } /* We've injected any pending events. This is really the point of no return (to ring-3). */ pVCpu->hm.s.Event.fPending = false; return VINF_SUCCESS; } /** * Prepares to run guest code in VT-x and we've committed to doing so. This * means there is no backing out to ring-3 or anywhere else at this * point. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data may be * out-of-sync. Make sure to update the required fields * before using them. * @param pVmxTransient Pointer to the VMX transient structure. * * @remarks Called with preemption disabled. * @remarks No-long-jump zone!!! */ static void hmR0VmxPreRunGuestCommitted(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { Assert(!VMMRZCallRing3IsEnabled(pVCpu)); Assert(VMMR0IsLogFlushDisabled(pVCpu)); Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD)); VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM); VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC); /* Indicate the start of guest execution. */ /* * If we are injecting events to a real-on-v86 mode guest, we may have to update * RIP and some other registers, i.e. hmR0VmxInjectPendingEvent()->hmR0VmxInjectEventVmcs(). * Reload only the necessary state, the assertion will catch if other parts of the code * change. */ if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { hmR0VmxLoadGuestRipRspRflags(pVCpu, pMixedCtx); hmR0VmxLoadGuestSegmentRegs(pVCpu, pMixedCtx); } #ifdef HMVMX_ALWAYS_SWAP_FPU_STATE if (!CPUMIsGuestFPUStateActive(pVCpu)) CPUMR0LoadGuestFPU(pVM, pVCpu, pMixedCtx); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR0); #endif if ( pVCpu->hm.s.fUseGuestFpu && !CPUMIsGuestFPUStateActive(pVCpu)) { CPUMR0LoadGuestFPU(pVM, pVCpu, pMixedCtx); Assert(pVCpu->hm.s.vmx.fUpdatedGuestState & HMVMX_UPDATED_GUEST_CR0); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR0); } /* * The host MSR values the very first time around won't be updated, so we need to * fill those values in. Subsequently, it's updated as part of the host state. */ if ( !pVCpu->hm.s.vmx.fUpdatedHostMsrs && pVCpu->hm.s.vmx.cMsrs > 0) { VMCPU_HMCF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT); } /* * Load the host state bits as we may've been preempted (only happens when * thread-context hooks are used or when hmR0VmxSetupVMRunHandler() changes pfnStartVM). */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT)) { /* This ASSUMES that pfnStartVM has been set up already. */ int rc = hmR0VmxSaveHostState(pVM, pVCpu); AssertRC(rc); STAM_COUNTER_INC(&pVCpu->hm.s.StatPreemptSaveHostState); } Assert(!VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT)); /* * Load the state shared between host and guest (FPU, debug, lazy MSRs). */ if (VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE)) hmR0VmxLoadSharedState(pVM, pVCpu, pMixedCtx); AssertMsg(!VMCPU_HMCF_VALUE(pVCpu), ("fContextUseFlags=%#RX32\n", VMCPU_HMCF_VALUE(pVCpu))); /* Store status of the shared guest-host state at the time of VM-entry. */ #if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (CPUMIsGuestInLongModeEx(pMixedCtx)) { pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActivePending(pVCpu); pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActivePending(pVCpu); } else #endif { pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu); pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu); } pVmxTransient->fWasGuestFPUStateActive = CPUMIsGuestFPUStateActive(pVCpu); /* * Cache the TPR-shadow for checking on every VM-exit if it might have changed. */ if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW) pVmxTransient->u8GuestTpr = pVCpu->hm.s.vmx.pbVirtApic[0x80]; PHMGLOBALCPUINFO pCpu = HMR0GetCurrentCpu(); RTCPUID idCurrentCpu = pCpu->idCpu; if ( pVmxTransient->fUpdateTscOffsettingAndPreemptTimer || idCurrentCpu != pVCpu->hm.s.idLastCpu) { hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVCpu); pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = false; } ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB-shootdowns, set this across the world switch. */ hmR0VmxFlushTaggedTlb(pVCpu, pCpu); /* Invalidate the appropriate guest entries from the TLB. */ Assert(idCurrentCpu == pVCpu->hm.s.idLastCpu); pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Update the error reporting info. with the current host CPU. */ STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x); TMNotifyStartOfExecution(pVCpu); /* Finally, notify TM to resume its clocks as we're about to start executing. */ /* * Load the TSC_AUX MSR when we are not intercepting RDTSCP. */ uint64_t uGuestTscAuxMsr; if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP) { if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT)) { int rc2 = CPUMQueryGuestMsr(pVCpu, MSR_K8_TSC_AUX, &uGuestTscAuxMsr); AssertRC(rc2); hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX, uGuestTscAuxMsr, true /* fUpdateHostMsr */); } else hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX); } #ifdef VBOX_STRICT hmR0VmxCheckAutoLoadStoreMsrs(pVCpu); #endif } /** * Performs some essential restoration of state after running guest code in * VT-x. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * @param pVmxTransient Pointer to the VMX transient structure. * @param rcVMRun Return code of VMLAUNCH/VMRESUME. * * @remarks Called with interrupts disabled, and returns with interrups enabled! * * @remarks No-long-jump zone!!! This function will however re-enable longjmps * unconditionally when it is safe to do so. */ static void hmR0VmxPostRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, int rcVMRun) { NOREF(pVM); Assert(!VMMRZCallRing3IsEnabled(pVCpu)); ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB-shootdowns. */ ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for TLB-shootdowns. */ pVCpu->hm.s.vmx.fUpdatedGuestState = 0; /* Exits/longjmps to ring-3 requires saving the guest state. */ pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */ pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */ if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT)) { /** @todo Find a way to fix hardcoding a guestimate. */ TMCpuTickSetLastSeen(pVCpu, ASMReadTSC() + pVCpu->hm.s.vmx.u64TSCOffset - 0x400 /* guestimate of world switch overhead in clock ticks */); } STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatExit1, x); TMNotifyEndOfExecution(pVCpu); /* Notify TM that the guest is no longer running. */ Assert(!(ASMGetFlags() & X86_EFL_IF)); VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM); #ifdef HMVMX_ALWAYS_SWAP_FPU_STATE if (CPUMIsGuestFPUStateActive(pVCpu)) { hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); CPUMR0SaveGuestFPU(pVM, pVCpu, pMixedCtx); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR0); } #endif pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_REQUIRED; /* Host state messed up by VT-x, we must restore. */ pVCpu->hm.s.vmx.uVmcsState |= HMVMX_VMCS_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */ ASMSetFlags(pVmxTransient->uEflags); /* Enable interrupts. */ VMMRZCallRing3Enable(pVCpu); /* It is now safe to do longjmps to ring-3!!! */ /* Save the basic VM-exit reason. Refer Intel spec. 24.9.1 "Basic VM-exit Information". */ uint32_t uExitReason; int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason); rc |= hmR0VmxReadEntryIntInfoVmcs(pVmxTransient); AssertRC(rc); pVmxTransient->uExitReason = (uint16_t)VMX_EXIT_REASON_BASIC(uExitReason); pVmxTransient->fVMEntryFailed = VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uEntryIntInfo); /* If the VMLAUNCH/VMRESUME failed, we can bail out early. This does -not- cover VMX_EXIT_ERR_*. */ if (RT_UNLIKELY(rcVMRun != VINF_SUCCESS)) { Log4(("VM-entry failure: pVCpu=%p idCpu=%RU32 rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", pVCpu, pVCpu->idCpu, rcVMRun, pVmxTransient->fVMEntryFailed)); return; } if (RT_LIKELY(!pVmxTransient->fVMEntryFailed)) { /* Update the guest interruptibility-state from the VMCS. */ hmR0VmxSaveGuestIntrState(pVCpu, pMixedCtx); #if defined(HMVMX_SYNC_FULL_GUEST_STATE) || defined(HMVMX_SAVE_FULL_GUEST_STATE) rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRC(rc); #endif /* * If the TPR was raised by the guest, it wouldn't cause a VM-exit immediately. Instead we sync the TPR lazily whenever * we eventually get a VM-exit for any reason. This maybe expensive as PDMApicSetTPR() can longjmp to ring-3 and which is * why it's done here as it's easier and no less efficient to deal with it here than making hmR0VmxSaveGuestState() * cope with longjmps safely (see VMCPU_FF_HM_UPDATE_CR3 handling). */ if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW) && pVmxTransient->u8GuestTpr != pVCpu->hm.s.vmx.pbVirtApic[0x80]) { rc = PDMApicSetTPR(pVCpu, pVCpu->hm.s.vmx.pbVirtApic[0x80]); AssertRC(rc); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE); } } } /** * Runs the guest code using VT-x the normal way. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCtx Pointer to the guest-CPU context. * * @note Mostly the same as hmR0VmxRunGuestCodeStep. * @remarks Called with preemption disabled. */ static int hmR0VmxRunGuestCodeNormal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx) { VMXTRANSIENT VmxTransient; VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true; int rc = VERR_INTERNAL_ERROR_5; uint32_t cLoops = 0; for (;; cLoops++) { Assert(!HMR0SuspendPending()); HMVMX_ASSERT_CPU_SAFE(); /* Preparatory work for running guest code, this may force us to return to ring-3. This bugger disables interrupts on VINF_SUCCESS! */ STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x); rc = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient); if (rc != VINF_SUCCESS) break; hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient); rc = hmR0VmxRunGuest(pVM, pVCpu, pCtx); /* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */ /* Restore any residual host-state and save any bits shared between host and guest into the guest-CPU state. Re-enables interrupts! */ hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rc); /* Check for errors with running the VM (VMLAUNCH/VMRESUME). */ if (RT_UNLIKELY(rc != VINF_SUCCESS)) { STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x); hmR0VmxReportWorldSwitchError(pVM, pVCpu, rc, pCtx, &VmxTransient); return rc; } /* Handle the VM-exit. */ AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason)); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll); STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]); STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x); HMVMX_START_EXIT_DISPATCH_PROF(); #ifdef HMVMX_USE_FUNCTION_TABLE rc = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, pCtx, &VmxTransient); #else rc = hmR0VmxHandleExit(pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason); #endif STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x); if (rc != VINF_SUCCESS) break; else if (cLoops > pVM->hm.s.cMaxResumeLoops) { STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMaxResume); rc = VINF_EM_RAW_INTERRUPT; break; } } STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x); return rc; } /** * Single steps guest code using VT-x. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCtx Pointer to the guest-CPU context. * * @note Mostly the same as hmR0VmxRunGuestCodeNormal. * @remarks Called with preemption disabled. */ static int hmR0VmxRunGuestCodeStep(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx) { VMXTRANSIENT VmxTransient; VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true; int rc = VERR_INTERNAL_ERROR_5; uint32_t cLoops = 0; uint16_t uCsStart = pCtx->cs.Sel; uint64_t uRipStart = pCtx->rip; for (;; cLoops++) { Assert(!HMR0SuspendPending()); HMVMX_ASSERT_CPU_SAFE(); /* Preparatory work for running guest code, this may force us to return to ring-3. This bugger disables interrupts on VINF_SUCCESS! */ STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x); rc = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient); if (rc != VINF_SUCCESS) break; hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient); rc = hmR0VmxRunGuest(pVM, pVCpu, pCtx); /* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */ /* Restore any residual host-state and save any bits shared between host and guest into the guest-CPU state. Re-enables interrupts! */ hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rc); /* Check for errors with running the VM (VMLAUNCH/VMRESUME). */ if (RT_UNLIKELY(rc != VINF_SUCCESS)) { STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x); hmR0VmxReportWorldSwitchError(pVM, pVCpu, rc, pCtx, &VmxTransient); return rc; } /* Handle the VM-exit. */ AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason)); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll); STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]); STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x); HMVMX_START_EXIT_DISPATCH_PROF(); #ifdef HMVMX_USE_FUNCTION_TABLE rc = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, pCtx, &VmxTransient); #else rc = hmR0VmxHandleExit(pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason); #endif STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x); if (rc != VINF_SUCCESS) break; else if (cLoops > pVM->hm.s.cMaxResumeLoops) { STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMaxResume); rc = VINF_EM_RAW_INTERRUPT; break; } /* * Did the RIP change, if so, consider it a single step. * Otherwise, make sure one of the TFs gets set. */ int rc2 = hmR0VmxSaveGuestRip(pVCpu, pCtx); rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pCtx); AssertRCReturn(rc2, rc2); if ( pCtx->rip != uRipStart || pCtx->cs.Sel != uCsStart) { rc = VINF_EM_DBG_STEPPED; break; } VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG); } /* * Clear the X86_EFL_TF if necessary. */ if (pVCpu->hm.s.fClearTrapFlag) { int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pCtx); AssertRCReturn(rc2, rc2); pVCpu->hm.s.fClearTrapFlag = false; pCtx->eflags.Bits.u1TF = 0; } /** @todo there seems to be issues with the resume flag when the monitor trap * flag is pending without being used. Seen early in bios init when * accessing APIC page in prot mode. */ STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x); return rc; } /** * Runs the guest code using VT-x. * * @returns VBox status code. * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCtx Pointer to the guest-CPU context. * * @remarks Called with preemption disabled. */ VMMR0DECL(int) VMXR0RunGuestCode(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx) { Assert(VMMRZCallRing3IsEnabled(pVCpu)); Assert(pVCpu->hm.s.vmx.fUpdatedGuestState == HMVMX_UPDATED_GUEST_ALL); HMVMX_ASSERT_PREEMPT_SAFE(); VMMRZCallRing3SetNotification(pVCpu, hmR0VmxCallRing3Callback, pCtx); int rc; if (!pVCpu->hm.s.fSingleInstruction && !DBGFIsStepping(pVCpu)) rc = hmR0VmxRunGuestCodeNormal(pVM, pVCpu, pCtx); else rc = hmR0VmxRunGuestCodeStep(pVM, pVCpu, pCtx); if (rc == VERR_EM_INTERPRETER) rc = VINF_EM_RAW_EMULATE_INSTR; else if (rc == VINF_EM_RESET) rc = VINF_EM_TRIPLE_FAULT; int rc2 = hmR0VmxExitToRing3(pVM, pVCpu, pCtx, rc); if (RT_FAILURE(rc2)) { pVCpu->hm.s.u32HMError = rc; rc = rc2; } Assert(!VMMRZCallRing3IsNotificationSet(pVCpu)); return rc; } #ifndef HMVMX_USE_FUNCTION_TABLE DECLINLINE(int) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason) { #ifdef DEBUG_ramshankar # define SVVMCS() do { int rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRC(rc2); } while (0) # define LDVMCS() do { VMCPU_HMCF_SET(pVCpu, HM_CHANGED_ALL_GUEST); } while (0) #endif int rc; switch (rcReason) { case VMX_EXIT_EPT_MISCONFIG: /* SVVMCS(); */ rc = hmR0VmxExitEptMisconfig(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_EPT_VIOLATION: /* SVVMCS(); */ rc = hmR0VmxExitEptViolation(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_IO_INSTR: /* SVVMCS(); */ rc = hmR0VmxExitIoInstr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_CPUID: /* SVVMCS(); */ rc = hmR0VmxExitCpuid(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_RDTSC: /* SVVMCS(); */ rc = hmR0VmxExitRdtsc(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_RDTSCP: /* SVVMCS(); */ rc = hmR0VmxExitRdtscp(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_APIC_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitApicAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_XCPT_OR_NMI: /* SVVMCS(); */ rc = hmR0VmxExitXcptOrNmi(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_MOV_CRX: /* SVVMCS(); */ rc = hmR0VmxExitMovCRx(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_EXT_INT: /* SVVMCS(); */ rc = hmR0VmxExitExtInt(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_INT_WINDOW: /* SVVMCS(); */ rc = hmR0VmxExitIntWindow(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_MWAIT: /* SVVMCS(); */ rc = hmR0VmxExitMwait(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_MONITOR: /* SVVMCS(); */ rc = hmR0VmxExitMonitor(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_TASK_SWITCH: /* SVVMCS(); */ rc = hmR0VmxExitTaskSwitch(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_PREEMPT_TIMER: /* SVVMCS(); */ rc = hmR0VmxExitPreemptTimer(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_RDMSR: /* SVVMCS(); */ rc = hmR0VmxExitRdmsr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_WRMSR: /* SVVMCS(); */ rc = hmR0VmxExitWrmsr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_MOV_DRX: /* SVVMCS(); */ rc = hmR0VmxExitMovDRx(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_TPR_BELOW_THRESHOLD: /* SVVMCS(); */ rc = hmR0VmxExitTprBelowThreshold(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_HLT: /* SVVMCS(); */ rc = hmR0VmxExitHlt(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_INVD: /* SVVMCS(); */ rc = hmR0VmxExitInvd(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_INVLPG: /* SVVMCS(); */ rc = hmR0VmxExitInvlpg(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_RSM: /* SVVMCS(); */ rc = hmR0VmxExitRsm(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_MTF: /* SVVMCS(); */ rc = hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_PAUSE: /* SVVMCS(); */ rc = hmR0VmxExitPause(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_XDTR_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_TR_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_WBINVD: /* SVVMCS(); */ rc = hmR0VmxExitWbinvd(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_XSETBV: /* SVVMCS(); */ rc = hmR0VmxExitXsetbv(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_RDRAND: /* SVVMCS(); */ rc = hmR0VmxExitRdrand(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_INVPCID: /* SVVMCS(); */ rc = hmR0VmxExitInvpcid(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_GETSEC: /* SVVMCS(); */ rc = hmR0VmxExitGetsec(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_RDPMC: /* SVVMCS(); */ rc = hmR0VmxExitRdpmc(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break; case VMX_EXIT_TRIPLE_FAULT: rc = hmR0VmxExitTripleFault(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_NMI_WINDOW: rc = hmR0VmxExitNmiWindow(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_INIT_SIGNAL: rc = hmR0VmxExitInitSignal(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_SIPI: rc = hmR0VmxExitSipi(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_IO_SMI: rc = hmR0VmxExitIoSmi(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_SMI: rc = hmR0VmxExitSmi(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_ERR_MSR_LOAD: rc = hmR0VmxExitErrMsrLoad(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_ERR_INVALID_GUEST_STATE: rc = hmR0VmxExitErrInvalidGuestState(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_ERR_MACHINE_CHECK: rc = hmR0VmxExitErrMachineCheck(pVCpu, pMixedCtx, pVmxTransient); break; case VMX_EXIT_VMCALL: case VMX_EXIT_VMCLEAR: case VMX_EXIT_VMLAUNCH: case VMX_EXIT_VMPTRLD: case VMX_EXIT_VMPTRST: case VMX_EXIT_VMREAD: case VMX_EXIT_VMRESUME: case VMX_EXIT_VMWRITE: case VMX_EXIT_VMXOFF: case VMX_EXIT_VMXON: case VMX_EXIT_INVEPT: case VMX_EXIT_INVVPID: case VMX_EXIT_VMFUNC: rc = hmR0VmxExitSetPendingXcptUD(pVCpu, pMixedCtx, pVmxTransient); break; default: rc = hmR0VmxExitErrUndefined(pVCpu, pMixedCtx, pVmxTransient); break; } return rc; } #endif #ifdef DEBUG /* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */ # define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \ RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId() # define HMVMX_ASSERT_PREEMPT_CPUID() \ do \ { \ RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \ AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \ } while (0) # define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \ do { \ AssertPtr(pVCpu); \ AssertPtr(pMixedCtx); \ AssertPtr(pVmxTransient); \ Assert(pVmxTransient->fVMEntryFailed == false); \ Assert(ASMIntAreEnabled()); \ HMVMX_ASSERT_PREEMPT_SAFE(); \ HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \ Log4Func(("vcpu[%RU32] -v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v\n", pVCpu->idCpu)); \ HMVMX_ASSERT_PREEMPT_SAFE(); \ if (VMMR0IsLogFlushDisabled(pVCpu)) \ HMVMX_ASSERT_PREEMPT_CPUID(); \ HMVMX_STOP_EXIT_DISPATCH_PROF(); \ } while (0) # define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() \ do { \ Log4Func(("\n")); \ } while (0) #else /* Release builds */ # define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \ do { \ HMVMX_STOP_EXIT_DISPATCH_PROF(); \ NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient); \ } while (0) # define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() do { } while (0) #endif /** * Advances the guest RIP after reading it from the VMCS. * * @returns VBox status code. * @param pVCpu Pointer to the VMCPU. * @param pMixedCtx Pointer to the guest-CPU context. The data maybe * out-of-sync. Make sure to update the required fields * before using them. * @param pVmxTransient Pointer to the VMX transient structure. * * @remarks No-long-jump zone!!! */ DECLINLINE(int) hmR0VmxAdvanceGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient); rc |= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); pMixedCtx->rip += pVmxTransient->cbInstr; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP); return rc; } /** * Tries to determine what part of the guest-state VT-x has deemed as invalid * and update error record fields accordingly. * * @return VMX_IGS_* return codes. * @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything * wrong with the guest state. * * @param pVM Pointer to the VM. * @param pVCpu Pointer to the VMCPU. * @param pCtx Pointer to the guest-CPU state. */ static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx) { #define HMVMX_ERROR_BREAK(err) { uError = (err); break; } #define HMVMX_CHECK_BREAK(expr, err) if (!(expr)) { \ uError = (err); \ break; \ } else do { } while (0) /* Duplicate of IEM_IS_CANONICAL(). */ #define HMVMX_IS_CANONICAL(a_u64Addr) ((uint64_t)(a_u64Addr) + UINT64_C(0x800000000000) < UINT64_C(0x1000000000000)) int rc; uint32_t uError = VMX_IGS_ERROR; uint32_t u32Val; bool fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuest; do { /* * CR0. */ uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1); uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr0Fixed1); /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG). See Intel spec. 26.3.1 "Checks on guest Guest Control Registers, Debug Registers and MSRs." */ if (fUnrestrictedGuest) uSetCR0 &= ~(X86_CR0_PE | X86_CR0_PG); uint32_t u32GuestCR0; rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32GuestCR0); AssertRCBreak(rc); HMVMX_CHECK_BREAK((u32GuestCR0 & uSetCR0) == uSetCR0, VMX_IGS_CR0_FIXED1); HMVMX_CHECK_BREAK(!(u32GuestCR0 & ~uZapCR0), VMX_IGS_CR0_FIXED0); if ( !fUnrestrictedGuest && (u32GuestCR0 & X86_CR0_PG) && !(u32GuestCR0 & X86_CR0_PE)) { HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO); } /* * CR4. */ uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1); uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr4Fixed1); uint32_t u32GuestCR4; rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &u32GuestCR4); AssertRCBreak(rc); HMVMX_CHECK_BREAK((u32GuestCR4 & uSetCR4) == uSetCR4, VMX_IGS_CR4_FIXED1); HMVMX_CHECK_BREAK(!(u32GuestCR4 & ~uZapCR4), VMX_IGS_CR4_FIXED0); /* * IA32_DEBUGCTL MSR. */ uint64_t u64Val; rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val); AssertRCBreak(rc); if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG) && (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */ { HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED); } uint64_t u64DebugCtlMsr = u64Val; #ifdef VBOX_STRICT rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val); AssertRCBreak(rc); Assert(u32Val == pVCpu->hm.s.vmx.u32EntryCtls); #endif bool const fLongModeGuest = RT_BOOL(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST); /* * RIP and RFLAGS. */ uint32_t u32Eflags; #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) { rc = VMXReadVmcs64(VMX_VMCS_GUEST_RIP, &u64Val); AssertRCBreak(rc); /* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */ if ( !fLongModeGuest || !pCtx->cs.Attr.n.u1Long) { HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID); } /** @todo If the processor supports N < 64 linear-address bits, bits 63:N * must be identical if the "IA32e mode guest" VM-entry control is 1 * and CS.L is 1. No check applies if the CPU supports 64 * linear-address bits. */ /* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */ rc = VMXReadVmcs64(VMX_VMCS_GUEST_RFLAGS, &u64Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */ VMX_IGS_RFLAGS_RESERVED); HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */ u32Eflags = u64Val; } else #endif { rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Eflags); AssertRCBreak(rc); HMVMX_CHECK_BREAK(!(u32Eflags & 0xffc08028), VMX_IGS_RFLAGS_RESERVED); /* Bit 31:22, Bit 15, 5, 3 MBZ. */ HMVMX_CHECK_BREAK((u32Eflags & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */ } if ( fLongModeGuest || ( fUnrestrictedGuest && !(u32GuestCR0 & X86_CR0_PE))) { HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID); } uint32_t u32EntryInfo; rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo); AssertRCBreak(rc); if ( VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo) && VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT) { HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID); } /* * 64-bit checks. */ #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) { if ( fLongModeGuest && !fUnrestrictedGuest) { HMVMX_CHECK_BREAK(u32GuestCR0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE); HMVMX_CHECK_BREAK(u32GuestCR4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE); } if ( !fLongModeGuest && (u32GuestCR4 & X86_CR4_PCIDE)) { HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE); } /** @todo CR3 field must be such that bits 63:52 and bits in the range * 51:32 beyond the processor's physical-address width are 0. */ if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG) && (pCtx->dr[7] & X86_DR7_MBZ_MASK)) { HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED); } rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL); rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL); } #endif /* * PERF_GLOBAL MSR. */ if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR) { rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)), VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */ } /* * PAT MSR. */ if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR) { rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED); for (unsigned i = 0; i < 8; i++) { uint8_t u8Val = (u64Val & 0x7); if ( u8Val != 0 /* UC */ || u8Val != 1 /* WC */ || u8Val != 4 /* WT */ || u8Val != 5 /* WP */ || u8Val != 6 /* WB */ || u8Val != 7 /* UC- */) { HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID); } u64Val >>= 3; } } /* * EFER MSR. */ if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR) { rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)), VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */ HMVMX_CHECK_BREAK((u64Val & MSR_K6_EFER_LMA) == (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST), VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH); HMVMX_CHECK_BREAK( fUnrestrictedGuest || (u64Val & MSR_K6_EFER_LMA) == (u32GuestCR0 & X86_CR0_PG), VMX_IGS_EFER_LMA_PG_MISMATCH); } /* * Segment registers. */ HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID); if (!(u32Eflags & X86_EFL_VM)) { /* CS */ HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID); HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED); HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED); HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff || !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000) || (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID); /* CS cannot be loaded with NULL in protected mode. */ HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE); HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID); if (pCtx->cs.Attr.n.u4Type == 9 || pCtx->cs.Attr.n.u4Type == 11) HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL); else if (pCtx->cs.Attr.n.u4Type == 13 || pCtx->cs.Attr.n.u4Type == 15) HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH); else if (pVM->hm.s.vmx.fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3) HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID); else HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID); /* SS */ HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest || (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL); HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL); if ( !(pCtx->cr0 & X86_CR0_PE) || pCtx->cs.Attr.n.u4Type == 3) { HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID); } if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE)) { HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 || pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID); HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID); HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED); HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED); HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff || !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000) || (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID); } /* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */ if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE)) { HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID); HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID); HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest || pCtx->ds.Attr.n.u4Type > 11 || pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL); HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED); HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED); HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff || !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000) || (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE) || (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID); } if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE)) { HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID); HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID); HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest || pCtx->es.Attr.n.u4Type > 11 || pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL); HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED); HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED); HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff || !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000) || (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE) || (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID); } if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE)) { HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID); HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID); HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest || pCtx->fs.Attr.n.u4Type > 11 || pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL); HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED); HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED); HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff || !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000) || (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE) || (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID); } if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE)) { HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID); HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID); HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest || pCtx->gs.Attr.n.u4Type > 11 || pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL); HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED); HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED); HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff || !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000) || (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE) || (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID); } /* 64-bit capable CPUs. */ #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) { HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL); HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL); HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE) || HMVMX_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL); HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID); HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ss.u64Base >> 32), VMX_IGS_LONGMODE_SS_BASE_INVALID); HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ds.u64Base >> 32), VMX_IGS_LONGMODE_DS_BASE_INVALID); HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->es.u64Base >> 32), VMX_IGS_LONGMODE_ES_BASE_INVALID); } #endif } else { /* V86 mode checks. */ uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr; if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3; u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3; } else { u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u; } /* CS */ HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID); HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID); HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID); /* SS */ HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID); HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID); HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID); /* DS */ HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID); HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID); HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID); /* ES */ HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID); HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID); HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID); /* FS */ HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID); HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID); HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID); /* GS */ HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID); HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID); HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID); /* 64-bit capable CPUs. */ #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) { HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL); HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL); HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE) || HMVMX_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL); HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID); HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ss.u64Base >> 32), VMX_IGS_LONGMODE_SS_BASE_INVALID); HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ds.u64Base >> 32), VMX_IGS_LONGMODE_DS_BASE_INVALID); HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->es.u64Base >> 32), VMX_IGS_LONGMODE_ES_BASE_INVALID); } #endif } /* * TR. */ HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID); /* 64-bit capable CPUs. */ #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) { HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL); } #endif if (fLongModeGuest) { HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */ VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID); } else { HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */ || pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/ VMX_IGS_TR_ATTR_TYPE_INVALID); } HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID); HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID); HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */ HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff || !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID); HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000) || (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID); HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE); /* * GDTR and IDTR. */ #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if (HMVMX_IS_64BIT_HOST_MODE()) { rc = VMXReadVmcs64(VMX_VMCS_GUEST_GDTR_BASE, &u64Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL); rc = VMXReadVmcs64(VMX_VMCS_GUEST_IDTR_BASE, &u64Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(HMVMX_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL); } #endif rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */ rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRCBreak(rc); HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */ /* * Guest Non-Register State. */ /* Activity State. */ uint32_t u32ActivityState; rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState); AssertRCBreak(rc); HMVMX_CHECK_BREAK( !u32ActivityState || (u32ActivityState & MSR_IA32_VMX_MISC_ACTIVITY_STATES(pVM->hm.s.vmx.Msrs.u64Misc)), VMX_IGS_ACTIVITY_STATE_INVALID); HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl) || u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID); uint32_t u32IntrState; rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &u32IntrState); AssertRCBreak(rc); if ( u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS || u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI) { HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID); } /** @todo Activity state and injecting interrupts. Left as a todo since we * currently don't use activity states but ACTIVE. */ HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM) || u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID); /* Guest interruptibility-state. */ HMVMX_CHECK_BREAK(!(u32IntrState & 0xfffffff0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED); HMVMX_CHECK_BREAK((u32IntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS)) != ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS), VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID); HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF) || !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI), VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID); if (VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)) { if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT) { HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI) && !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS), VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID); } else if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI) { HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS), VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID); HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI), VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID); } } /** @todo Assumes the processor is not in SMM. */ HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI), VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID); HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM) || (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI), VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID); if ( (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI) && VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo) && VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI) { HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI), VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID); } /* Pending debug exceptions. */ if (HMVMX_IS_64BIT_HOST_MODE()) { rc = VMXReadVmcs64(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u64Val); AssertRCBreak(rc); /* Bits 63:15, Bit 13, Bits 11:4 MBZ. */ HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED); u32Val = u64Val; /* For pending debug exceptions checks below. */ } else { rc = VMXReadVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u32Val); AssertRCBreak(rc); /* Bits 31:15, Bit 13, Bits 11:4 MBZ. */ HMVMX_CHECK_BREAK(!(u64Val & 0xffffaff0), VMX_IGS_PENDING_DEBUG_RESERVED); } if ( (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI) || (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS) || u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT) { if ( (u32Eflags & X86_EFL_TF) && !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */ { /* Bit 14 is PendingDebug.BS. */ HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET); } if ( !(u32Eflags & X86_EFL_TF) || (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */ { /* Bit 14 is PendingDebug.BS. */ HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR); } } /* VMCS link pointer. */ rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val); AssertRCBreak(rc); if (u64Val != UINT64_C(0xffffffffffffffff)) { HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED); /** @todo Bits beyond the processor's physical-address width MBZ. */ /** @todo 32-bit located in memory referenced by value of this field (as a * physical address) must contain the processor's VMCS revision ID. */ /** @todo SMM checks. */ } /** @todo Checks on Guest Page-Directory-Pointer-Table Entries. */ /* Shouldn't happen but distinguish it from AssertRCBreak() errors. */ if (uError == VMX_IGS_ERROR) uError = VMX_IGS_REASON_NOT_FOUND; } while (0); pVCpu->hm.s.u32HMError = uError; return uError; #undef HMVMX_ERROR_BREAK #undef HMVMX_CHECK_BREAK #undef HMVMX_IS_CANONICAL } /* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */ /* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */ /* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */ /** @name VM-exit handlers. * @{ */ /** * VM-exit handler for external interrupts (VMX_EXIT_EXT_INT). */ HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt); /* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */ if (VMMR0ThreadCtxHooksAreRegistered(pVCpu)) return VINF_SUCCESS; return VINF_EM_RAW_INTERRUPT; } /** * VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI). */ HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3); int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient); AssertRCReturn(rc, rc); uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo); Assert( !(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT) && uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT); Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo)); if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI) { /* * This cannot be a guest NMI as the only way for the guest to receive an NMI is if we injected it ourselves and * anything we inject is not going to cause a VM-exit directly for the event being injected. * See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery". * * Dispatch the NMI to the host. See Intel spec. 27.5.5 "Updating Non-Register State". */ VMXDispatchHostNmi(); STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC); STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3); return VINF_SUCCESS; } /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */ rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient); if (RT_UNLIKELY(rc == VINF_HM_DOUBLE_FAULT)) { STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3); return VINF_SUCCESS; } else if (RT_UNLIKELY(rc == VINF_EM_RESET)) { STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3); return rc; } uint32_t uExitIntInfo = pVmxTransient->uExitIntInfo; uint32_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(uExitIntInfo); switch (uIntType) { case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT: /* Software exception. (#BP or #OF) */ Assert(uVector == X86_XCPT_DB || uVector == X86_XCPT_BP || uVector == X86_XCPT_OF); /* no break */ case VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT: { switch (uVector) { case X86_XCPT_PF: rc = hmR0VmxExitXcptPF(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_GP: rc = hmR0VmxExitXcptGP(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_NM: rc = hmR0VmxExitXcptNM(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_MF: rc = hmR0VmxExitXcptMF(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_DB: rc = hmR0VmxExitXcptDB(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_BP: rc = hmR0VmxExitXcptBP(pVCpu, pMixedCtx, pVmxTransient); break; #ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF); rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE); rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD); rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS); rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break; case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP); rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break; #endif default: { rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk); if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS); Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM))); Assert(CPUMIsGuestInRealModeEx(pMixedCtx)); rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient); rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient); AssertRCReturn(rc, rc); hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(uExitIntInfo), pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */); AssertRCReturn(rc, rc); } else { AssertMsgFailed(("Unexpected VM-exit caused by exception %#x\n", uVector)); pVCpu->hm.s.u32HMError = uVector; rc = VERR_VMX_UNEXPECTED_EXCEPTION; } break; } } break; } default: { pVCpu->hm.s.u32HMError = uExitIntInfo; rc = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE; AssertMsgFailed(("Unexpected interruption info %#x\n", VMX_EXIT_INTERRUPTION_INFO_TYPE(uExitIntInfo))); break; } } STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3); return rc; } /** * VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW). */ HMVMX_EXIT_DECL hmR0VmxExitIntWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */ Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT); pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT; int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRCReturn(rc, rc); /* Deliver the pending interrupt via hmR0VmxPreRunGuest()->hmR0VmxInjectEvent() and resume guest execution. */ STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow); return VINF_SUCCESS; } /** * VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW). */ HMVMX_EXIT_DECL hmR0VmxExitNmiWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); AssertMsgFailed(("Unexpected NMI-window exit.\n")); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitWbinvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWbinvd); return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); } /** * VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitInvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvd); return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); } /** * VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); PVM pVM = pVCpu->CTX_SUFF(pVM); int rc = EMInterpretCpuId(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx)); if (RT_LIKELY(rc == VINF_SUCCESS)) { rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); Assert(pVmxTransient->cbInstr == 2); } else { AssertMsgFailed(("hmR0VmxExitCpuid: EMInterpretCpuId failed with %Rrc\n", rc)); rc = VERR_EM_INTERPRETER; } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCpuid); return rc; } /** * VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); if (pMixedCtx->cr4 & X86_CR4_SMXE) return VINF_EM_RAW_EMULATE_INSTR; AssertMsgFailed(("hmR0VmxExitGetsec: unexpected VM-exit when CR4.SMXE is 0.\n")); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */ AssertRCReturn(rc, rc); PVM pVM = pVCpu->CTX_SUFF(pVM); rc = EMInterpretRdtsc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx)); if (RT_LIKELY(rc == VINF_SUCCESS)) { rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); Assert(pVmxTransient->cbInstr == 2); /* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */ if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING) pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true; } else { AssertMsgFailed(("hmR0VmxExitRdtsc: EMInterpretRdtsc failed with %Rrc\n", rc)); rc = VERR_EM_INTERPRETER; } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc); return rc; } /** * VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */ rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx); /* For MSR_K8_TSC_AUX */ AssertRCReturn(rc, rc); PVM pVM = pVCpu->CTX_SUFF(pVM); rc = EMInterpretRdtscp(pVM, pVCpu, pMixedCtx); if (RT_LIKELY(rc == VINF_SUCCESS)) { rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); Assert(pVmxTransient->cbInstr == 3); /* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */ if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING) pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true; } else { AssertMsgFailed(("hmR0VmxExitRdtscp: EMInterpretRdtscp failed with %Rrc\n", rc)); rc = VERR_EM_INTERPRETER; } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc); return rc; } /** * VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */ rc |= hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); /** @todo review if CR0 is really required by EM. */ AssertRCReturn(rc, rc); PVM pVM = pVCpu->CTX_SUFF(pVM); rc = EMInterpretRdpmc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx)); if (RT_LIKELY(rc == VINF_SUCCESS)) { rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); Assert(pVmxTransient->cbInstr == 2); } else { AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc)); rc = VERR_EM_INTERPRETER; } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdpmc); return rc; } /** * VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); PVM pVM = pVCpu->CTX_SUFF(pVM); Assert(!pVM->hm.s.fNestedPaging); int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); VBOXSTRICTRC rc2 = EMInterpretInvlpg(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), pVmxTransient->uExitQualification); rc = VBOXSTRICTRC_VAL(rc2); if (RT_LIKELY(rc == VINF_SUCCESS)) rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); else { AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitInvlpg: EMInterpretInvlpg %#RX64 failed with %Rrc\n", pVmxTransient->uExitQualification, rc)); } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvlpg); return rc; } /** * VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); PVM pVM = pVCpu->CTX_SUFF(pVM); rc = EMInterpretMonitor(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx)); if (RT_LIKELY(rc == VINF_SUCCESS)) rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); else { AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMonitor: EMInterpretMonitor failed with %Rrc\n", rc)); rc = VERR_EM_INTERPRETER; } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMonitor); return rc; } /** * VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); PVM pVM = pVCpu->CTX_SUFF(pVM); VBOXSTRICTRC rc2 = EMInterpretMWait(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx)); rc = VBOXSTRICTRC_VAL(rc2); if (RT_LIKELY( rc == VINF_SUCCESS || rc == VINF_EM_HALT)) { int rc3 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); AssertRCReturn(rc3, rc3); if ( rc == VINF_EM_HALT && EMMonitorWaitShouldContinue(pVCpu, pMixedCtx)) { rc = VINF_SUCCESS; } } else { AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMwait: EMInterpretMWait failed with %Rrc\n", rc)); rc = VERR_EM_INTERPRETER; } AssertMsg(rc == VINF_SUCCESS || rc == VINF_EM_HALT || rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMwait: failed, invalid error code %Rrc\n", rc)); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMwait); return rc; } /** * VM-exit handler for RSM (VMX_EXIT_RSM). Unconditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitRsm(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { /* * Execution of RSM outside of SMM mode causes #UD regardless of VMX root or VMX non-root mode. In theory, we should never * get this VM-exit. This can happen only if dual-monitor treatment of SMI and VMX is enabled, which can (only?) be done by * executing VMCALL in VMX root operation. If we get here, something funny is going on. * See Intel spec. "33.15.5 Enabling the Dual-Monitor Treatment". */ HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); AssertMsgFailed(("Unexpected RSM VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for SMI (VMX_EXIT_SMI). Unconditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { /* * This can only happen if we support dual-monitor treatment of SMI, which can be activated by executing VMCALL in VMX * root operation. Only an STM (SMM transfer monitor) would get this exit when we (the executive monitor) execute a VMCALL * in VMX root mode or receive an SMI. If we get here, something funny is going on. * See Intel spec. "33.15.6 Activating the Dual-Monitor Treatment" and Intel spec. 25.3 "Other Causes of VM-Exits" */ HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); AssertMsgFailed(("Unexpected SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for IO SMI (VMX_EXIT_IO_SMI). Unconditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitIoSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { /* Same treatment as VMX_EXIT_SMI. See comment in hmR0VmxExitSmi(). */ HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); AssertMsgFailed(("Unexpected IO SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for SIPI (VMX_EXIT_SIPI). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitSipi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { /* * SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest activity state is used. We currently * don't make use of it (see hmR0VmxLoadGuestActivityState()) as our guests don't have direct access to the host LAPIC. * See Intel spec. 25.3 "Other Causes of VM-exits". */ HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); AssertMsgFailed(("Unexpected SIPI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for INIT signal (VMX_EXIT_INIT_SIGNAL). Unconditional * VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitInitSignal(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { /* * INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM. * See Intel spec. 33.14.1 Default Treatment of SMI Delivery" and Intel spec. 29.3 "VMX Instructions" for "VMXON". * * It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these VM-exits. * See Intel spec. "23.8 Restrictions on VMX operation". */ HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); return VINF_SUCCESS; } /** * VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional * VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); return VINF_EM_RESET; } /** * VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT); int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); pMixedCtx->rip++; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP); if (EMShouldContinueAfterHalt(pVCpu, pMixedCtx)) /* Requires eflags. */ rc = VINF_SUCCESS; else rc = VINF_EM_HALT; STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt); return rc; } /** * VM-exit handler for instructions that result in a #UD exception delivered to * the guest. */ HMVMX_EXIT_DECL hmR0VmxExitSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx); return VINF_SUCCESS; } /** * VM-exit handler for expiry of the VMX preemption timer. */ HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* If the preemption-timer has expired, reinitialize the preemption timer on next VM-entry. */ pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true; /* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */ PVM pVM = pVCpu->CTX_SUFF(pVM); bool fTimersPending = TMTimerPollBool(pVM, pVCpu); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer); return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS; } /** * VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* We expose XSETBV to the guest, fallback to the recompiler for emulation. */ /** @todo check if XSETBV is supported by the recompiler. */ return VERR_EM_INTERPRETER; } /** * VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* The guest should not invalidate the host CPU's TLBs, fallback to recompiler. */ /** @todo implement EMInterpretInvpcid() */ return VERR_EM_INTERPRETER; } /** * VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE). * Error VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx); NOREF(uInvalidReason); #ifdef VBOX_STRICT uint32_t uIntrState; HMVMXHCUINTREG uHCReg; uint64_t u64Val; uint32_t u32Val; rc = hmR0VmxReadEntryIntInfoVmcs(pVmxTransient); rc |= hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient); rc |= hmR0VmxReadEntryInstrLenVmcs(pVmxTransient); rc |= VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState); AssertRCReturn(rc, rc); Log4(("uInvalidReason %u\n", uInvalidReason)); Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo)); Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode)); Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr)); Log4(("VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE %#RX32\n", uIntrState)); rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32Val); AssertRC(rc); Log4(("VMX_VMCS_GUEST_CR0 %#RX32\n", u32Val)); rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc); Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc); Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc); Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg)); rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc); Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg)); rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc); Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val)); #else NOREF(pVmxTransient); #endif HMDumpRegs(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx); return VERR_VMX_INVALID_GUEST_STATE; } /** * VM-exit handler for VM-entry failure due to an MSR-load * (VMX_EXIT_ERR_MSR_LOAD). Error VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitErrMsrLoad(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { NOREF(pVmxTransient); AssertMsgFailed(("Unexpected MSR-load exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for VM-entry failure due to a machine-check event * (VMX_EXIT_ERR_MACHINE_CHECK). Error VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitErrMachineCheck(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { NOREF(pVmxTransient); AssertMsgFailed(("Unexpected machine-check event exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for all undefined reasons. Should never ever happen.. in * theory. */ HMVMX_EXIT_DECL hmR0VmxExitErrUndefined(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { AssertMsgFailed(("Huh!? Undefined VM-exit reason %d. pVCpu=%p pMixedCtx=%p\n", pVmxTransient->uExitReason, pVCpu, pMixedCtx)); NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient); return VERR_VMX_UNDEFINED_EXIT_CODE; } /** * VM-exit handler for XDTR (LGDT, SGDT, LIDT, SIDT) accesses * (VMX_EXIT_XDTR_ACCESS) and LDT and TR access (LLDT, LTR, SLDT, STR). * Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitXdtrAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT. */ STAM_COUNTER_INC(&pVCpu->hm.s.StatExitXdtrAccess); if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT) return VERR_EM_INTERPRETER; AssertMsgFailed(("Unexpected XDTR access. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for RDRAND (VMX_EXIT_RDRAND). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitRdrand(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT. */ STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdrand); if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT) return VERR_EM_INTERPRETER; AssertMsgFailed(("Unexpected RDRAND exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for RDMSR (VMX_EXIT_RDMSR). */ HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* EMInterpretRdmsr() requires CR0, Eflags and SS segment register. */ int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)) { rc |= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx); } AssertRCReturn(rc, rc); Log4(("CS:RIP=%04x:%#RX64 ECX=%X\n", pMixedCtx->cs.Sel, pMixedCtx->rip, pMixedCtx->ecx)); #ifdef VBOX_STRICT if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS) && hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx)) { AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } #endif PVM pVM = pVCpu->CTX_SUFF(pVM); rc = EMInterpretRdmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx)); AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER, ("hmR0VmxExitRdmsr: failed, invalid error code %Rrc\n", rc)); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr); if (RT_LIKELY(rc == VINF_SUCCESS)) { rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); Assert(pVmxTransient->cbInstr == 2); } return rc; } /** * VM-exit handler for WRMSR (VMX_EXIT_WRMSR). */ HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); PVM pVM = pVCpu->CTX_SUFF(pVM); int rc = VINF_SUCCESS; /* EMInterpretWrmsr() requires CR0, EFLAGS and SS segment register. */ rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)) { rc |= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx); } AssertRCReturn(rc, rc); Log4(("ecx=%#RX32\n", pMixedCtx->ecx)); rc = EMInterpretWrmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx)); AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER, ("hmR0VmxExitWrmsr: failed, invalid error code %Rrc\n", rc)); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr); if (RT_LIKELY(rc == VINF_SUCCESS)) { rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); /* If this is an X2APIC WRMSR access, update the APIC state as well. */ if ( pMixedCtx->ecx >= MSR_IA32_X2APIC_START && pMixedCtx->ecx <= MSR_IA32_X2APIC_END) { /* We've already saved the APIC related guest-state (TPR) in hmR0VmxPostRunGuest(). When full APIC register * virtualization is implemented we'll have to make sure APIC state is saved from the VMCS before EMInterpretWrmsr() changes it. */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE); } else if (pMixedCtx->ecx == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */ pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true; /* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not supported. */ if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)) { switch (pMixedCtx->ecx) { case MSR_IA32_SYSENTER_CS: VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break; case MSR_IA32_SYSENTER_EIP: VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break; case MSR_IA32_SYSENTER_ESP: VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break; case MSR_K8_FS_BASE: /* no break */ case MSR_K8_GS_BASE: VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS); break; default: { if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx)) VMCPU_HMCF_SET(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS); #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) else if ( HMVMX_IS_64BIT_HOST_MODE() && hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx)) { VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS); } #endif break; } } } #ifdef VBOX_STRICT else { /* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */ switch (pMixedCtx->ecx) { case MSR_IA32_SYSENTER_CS: case MSR_IA32_SYSENTER_EIP: case MSR_IA32_SYSENTER_ESP: case MSR_K8_FS_BASE: case MSR_K8_GS_BASE: { AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } /* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */ default: { if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx)) { AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } #if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) if ( HMVMX_IS_64BIT_HOST_MODE() && hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx)) { AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } #endif break; } } } #endif /* VBOX_STRICT */ } return rc; } /** * VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT. */ STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPause); if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT) return VERR_EM_INTERPRETER; AssertMsgFailed(("Unexpected PAUSE exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } /** * VM-exit handler for when the TPR value is lowered below the specified * threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitTprBelowThreshold(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW); /* * The TPR has already been updated, see hmR0VMXPostRunGuest(). RIP is also updated as part of the VM-exit by VT-x. Update * the threshold in the VMCS, deliver the pending interrupt via hmR0VmxPreRunGuest()->hmR0VmxInjectPendingEvent() and * resume guest execution. */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold); return VINF_SUCCESS; } /** * VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional * VM-exit. * * @retval VINF_SUCCESS when guest execution can continue. * @retval VINF_PGM_CHANGE_MODE when shadow paging mode changed, back to ring-3. * @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3. * @retval VERR_EM_INTERPRETER when something unexpected happened, fallback to * recompiler. */ HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2); int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); AssertRCReturn(rc, rc); const RTGCUINTPTR uExitQualification = pVmxTransient->uExitQualification; const uint32_t uAccessType = VMX_EXIT_QUALIFICATION_CRX_ACCESS(uExitQualification); PVM pVM = pVCpu->CTX_SUFF(pVM); switch (uAccessType) { case VMX_EXIT_QUALIFICATION_CRX_ACCESS_WRITE: /* MOV to CRx */ { #if 0 /* EMInterpretCRxWrite() references a lot of guest state (EFER, RFLAGS, Segment Registers, etc.) Sync entire state */ rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); #else rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); #endif AssertRCReturn(rc, rc); rc = EMInterpretCRxWrite(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification), VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification)); Assert(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER || rc == VINF_PGM_CHANGE_MODE || rc == VINF_PGM_SYNC_CR3); switch (VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)) { case 0: /* CR0 */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR0); Log4(("CRX CR0 write rc=%d CR0=%#RX64\n", rc, pMixedCtx->cr0)); break; case 2: /* CR2 */ /* Nothing to do here, CR2 it's not part of the VMCS. */ break; case 3: /* CR3 */ Assert(!pVM->hm.s.fNestedPaging || !CPUMIsGuestPagingEnabledEx(pMixedCtx)); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR3); Log4(("CRX CR3 write rc=%d CR3=%#RX64\n", rc, pMixedCtx->cr3)); break; case 4: /* CR4 */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR4); Log4(("CRX CR4 write rc=%d CR4=%#RX64\n", rc, pMixedCtx->cr4)); break; case 8: /* CR8 */ Assert(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)); /* CR8 contains the APIC TPR. Was updated by EMInterpretCRxWrite(). */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE); break; default: AssertMsgFailed(("Invalid CRx register %#x\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification))); break; } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxWrite[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]); break; } case VMX_EXIT_QUALIFICATION_CRX_ACCESS_READ: /* MOV from CRx */ { /* EMInterpretCRxRead() requires EFER MSR, CS. */ rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); Assert( !pVM->hm.s.fNestedPaging || !CPUMIsGuestPagingEnabledEx(pMixedCtx) || VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 3); /* CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */ Assert( VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 8 || !(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)); rc = EMInterpretCRxRead(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification), VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)); Assert(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxRead[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]); Log4(("CRX CR%d Read access rc=%d\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification), rc)); break; } case VMX_EXIT_QUALIFICATION_CRX_ACCESS_CLTS: /* CLTS (Clear Task-Switch Flag in CR0) */ { rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); rc = EMInterpretCLTS(pVM, pVCpu); AssertRCReturn(rc, rc); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR0); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts); Log4(("CRX CLTS write rc=%d\n", rc)); break; } case VMX_EXIT_QUALIFICATION_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */ { rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); rc = EMInterpretLMSW(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), VMX_EXIT_QUALIFICATION_CRX_LMSW_DATA(uExitQualification)); if (RT_LIKELY(rc == VINF_SUCCESS)) VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR0); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw); Log4(("CRX LMSW write rc=%d\n", rc)); break; } default: { AssertMsgFailed(("Invalid access-type in Mov CRx exit qualification %#x\n", uAccessType)); rc = VERR_VMX_UNEXPECTED_EXCEPTION; } } /* Validate possible error codes. */ Assert(rc == VINF_SUCCESS || rc == VINF_PGM_CHANGE_MODE || rc == VERR_EM_INTERPRETER || rc == VINF_PGM_SYNC_CR3 || rc == VERR_VMX_UNEXPECTED_EXCEPTION); if (RT_SUCCESS(rc)) { int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); AssertRCReturn(rc2, rc2); } STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2); return rc; } /** * VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional * VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1); int rc2 = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); rc2 |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient); rc2 |= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx); rc2 |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* Eflag checks in EMInterpretDisasCurrent(). */ rc2 |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); /* CR0 checks & PGM* in EMInterpretDisasCurrent(). */ rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); /* SELM checks in EMInterpretDisasCurrent(). */ /* EFER also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */ AssertRCReturn(rc2, rc2); /* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */ uint32_t uIOPort = VMX_EXIT_QUALIFICATION_IO_PORT(pVmxTransient->uExitQualification); uint8_t uIOWidth = VMX_EXIT_QUALIFICATION_IO_WIDTH(pVmxTransient->uExitQualification); bool fIOWrite = ( VMX_EXIT_QUALIFICATION_IO_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_IO_DIRECTION_OUT); bool fIOString = VMX_EXIT_QUALIFICATION_IO_IS_STRING(pVmxTransient->uExitQualification); AssertReturn(uIOWidth <= 3 && uIOWidth != 2, VERR_HMVMX_IPE_1); /* I/O operation lookup arrays. */ static const uint32_t s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses. */ static const uint32_t s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving the result (in AL/AX/EAX). */ VBOXSTRICTRC rcStrict; const uint32_t cbValue = s_aIOSizes[uIOWidth]; const uint32_t cbInstr = pVmxTransient->cbInstr; bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */ PVM pVM = pVCpu->CTX_SUFF(pVM); if (fIOString) { #if 0 /* Not yet ready. IEM gurus with debian 32-bit guest without NP (on ATA reads). See @bugref{5752#c158} */ /* * INS/OUTS - I/O String instruction. * * Use instruction-information if available, otherwise fall back on * interpreting the instruction. */ Log4(("CS:RIP=%04x:%#RX64 %#06x/%u %c str\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r')); AssertReturn(pMixedCtx->dx == uIOPort, VERR_HMVMX_IPE_2); if (MSR_IA32_VMX_BASIC_INFO_VMCS_INS_OUTS(pVM->hm.s.vmx.Msrs.u64BasicInfo)) { rc2 = hmR0VmxReadExitInstrInfoVmcs(pVmxTransient); /** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */ rc2 |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRCReturn(rc2, rc2); AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_HMVMX_IPE_3); AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2); IEMMODE enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize; bool fRep = VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification); if (fIOWrite) { rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, pVmxTransient->ExitInstrInfo.StrIo.iSegReg); } else { /* * The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES. * Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS. * See Intel Instruction spec. for "INS". * See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS". */ rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr); } } else { /** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */ rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRCReturn(rc2, rc2); rcStrict = IEMExecOne(pVCpu); } /** @todo IEM needs to be setting these flags somehow. */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP); fUpdateRipAlready = true; #else PDISCPUSTATE pDis = &pVCpu->hm.s.DisState; rcStrict = EMInterpretDisasCurrent(pVM, pVCpu, pDis, NULL /* pcbInstr */); if (RT_SUCCESS(rcStrict)) { if (fIOWrite) { rcStrict = IOMInterpretOUTSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix, (DISCPUMODE)pDis->uAddrMode, cbValue); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringWrite); } else { rcStrict = IOMInterpretINSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix, (DISCPUMODE)pDis->uAddrMode, cbValue); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringRead); } } else { AssertMsg(rcStrict == VERR_EM_INTERPRETER, ("rcStrict=%Rrc RIP %#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->rip)); rcStrict = VINF_EM_RAW_EMULATE_INSTR; } #endif } else { /* * IN/OUT - I/O instruction. */ Log4(("CS:RIP=%04x:%#RX64 %#06x/%u %c\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r')); const uint32_t uAndVal = s_aIOOpAnd[uIOWidth]; Assert(!VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification)); if (fIOWrite) { rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pMixedCtx->eax & uAndVal, cbValue); if (rcStrict == VINF_IOM_R3_IOPORT_WRITE) HMR0SavePendingIOPortWrite(pVCpu, pMixedCtx->rip, pMixedCtx->rip + cbInstr, uIOPort, uAndVal, cbValue); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite); } else { uint32_t u32Result = 0; rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue); if (IOM_SUCCESS(rcStrict)) { /* Save result of I/O IN instr. in AL/AX/EAX. */ pMixedCtx->eax = (pMixedCtx->eax & ~uAndVal) | (u32Result & uAndVal); } else if (rcStrict == VINF_IOM_R3_IOPORT_READ) HMR0SavePendingIOPortRead(pVCpu, pMixedCtx->rip, pMixedCtx->rip + cbInstr, uIOPort, uAndVal, cbValue); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead); } } if (IOM_SUCCESS(rcStrict)) { if (!fUpdateRipAlready) { pMixedCtx->rip += cbInstr; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP); } /* * INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru while booting Fedora 17 64-bit guest. * See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ. */ if (fIOString) VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS); /* * If any I/O breakpoints are armed, we need to check if one triggered * and take appropriate action. * Note that the I/O breakpoint type is undefined if CR4.DE is 0. */ rc2 = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx); AssertRCReturn(rc2, rc2); /** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the * execution engines about whether hyper BPs and such are pending. */ uint32_t const uDr7 = pMixedCtx->dr[7]; if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK) && X86_DR7_ANY_RW_IO(uDr7) && (pMixedCtx->cr4 & X86_CR4_DE)) || DBGFBpIsHwIoArmed(pVM))) { STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck); /* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */ VMMRZCallRing3Disable(pVCpu); HM_DISABLE_PREEMPT_IF_NEEDED(); bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /*fDr6*/); VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pMixedCtx, uIOPort, cbValue); if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP) { /* Raise #DB. */ if (fIsGuestDbgActive) ASMSetDR6(pMixedCtx->dr[6]); if (pMixedCtx->dr[7] != uDr7) VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG); hmR0VmxSetPendingXcptDB(pVCpu, pMixedCtx); } /* rcStrict is VINF_SUCCESS or in [VINF_EM_FIRST..VINF_EM_LAST]. */ else if ( rcStrict2 != VINF_SUCCESS && (rcStrict == VINF_SUCCESS || rcStrict2 < rcStrict)) rcStrict = rcStrict2; HM_RESTORE_PREEMPT_IF_NEEDED(); VMMRZCallRing3Enable(pVCpu); } } #ifdef DEBUG if (rcStrict == VINF_IOM_R3_IOPORT_READ) Assert(!fIOWrite); else if (rcStrict == VINF_IOM_R3_IOPORT_WRITE) Assert(fIOWrite); else { /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST * statuses, that the VMM device and some others may return. See * IOM_SUCCESS() for guidance. */ AssertMsg( RT_FAILURE(rcStrict) || rcStrict == VINF_SUCCESS || rcStrict == VINF_EM_RAW_EMULATE_INSTR || rcStrict == VINF_EM_DBG_BREAKPOINT || rcStrict == VINF_EM_RAW_GUEST_TRAP || rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict))); } #endif STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1); return VBOXSTRICTRC_TODO(rcStrict); } /** * VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional * VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* Check if this task-switch occurred while delivery an event through the guest IDT. */ int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); AssertRCReturn(rc, rc); if (VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE_IDT) { rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient); AssertRCReturn(rc, rc); if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo)) { uint32_t uIntType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo); /* Software interrupts and exceptions will be regenerated when the recompiler restarts the instruction. */ if ( uIntType != VMX_IDT_VECTORING_INFO_TYPE_SW_INT && uIntType != VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT && uIntType != VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT) { uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo); bool fErrorCodeValid = !!VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo); /* Save it as a pending event and it'll be converted to a TRPM event on the way out to ring-3. */ Assert(!pVCpu->hm.s.Event.fPending); pVCpu->hm.s.Event.fPending = true; pVCpu->hm.s.Event.u64IntInfo = pVmxTransient->uIdtVectoringInfo; rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient); AssertRCReturn(rc, rc); if (fErrorCodeValid) pVCpu->hm.s.Event.u32ErrCode = pVmxTransient->uIdtVectoringErrorCode; else pVCpu->hm.s.Event.u32ErrCode = 0; if ( uIntType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT && uVector == X86_XCPT_PF) { pVCpu->hm.s.Event.GCPtrFaultAddress = pMixedCtx->cr2; } Log4(("Pending event on TaskSwitch uIntType=%#x uVector=%#x\n", uIntType, uVector)); } } } /** @todo Emulate task switch someday, currently just going back to ring-3 for * emulation. */ STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch); return VERR_EM_INTERPRETER; } /** * VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG); pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG; int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRCReturn(rc, rc); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMtf); return VINF_EM_DBG_STEPPED; } /** * VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */ int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient); if (RT_UNLIKELY(rc == VINF_HM_DOUBLE_FAULT)) return VINF_SUCCESS; else if (RT_UNLIKELY(rc == VINF_EM_RESET)) return rc; #if 0 /** @todo Investigate if IOMMMIOPhysHandler() requires a lot of state, for now * just sync the whole thing. */ rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); #else /* Aggressive state sync. for now. */ rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); #endif rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); AssertRCReturn(rc, rc); /* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */ uint32_t uAccessType = VMX_EXIT_QUALIFICATION_APIC_ACCESS_TYPE(pVmxTransient->uExitQualification); switch (uAccessType) { case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE: case VMX_APIC_ACCESS_TYPE_LINEAR_READ: { if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW) && VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification) == 0x80) { AssertMsgFailed(("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n")); } RTGCPHYS GCPhys = pMixedCtx->msrApicBase; /* Always up-to-date, msrApicBase is not part of the VMCS. */ GCPhys &= PAGE_BASE_GC_MASK; GCPhys += VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification); PVM pVM = pVCpu->CTX_SUFF(pVM); Log4(("ApicAccess uAccessType=%#x GCPhys=%#RGv Off=%#x\n", uAccessType, GCPhys, VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification))); VBOXSTRICTRC rc2 = IOMMMIOPhysHandler(pVM, pVCpu, (uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ) ? 0 : X86_TRAP_PF_RW, CPUMCTX2CORE(pMixedCtx), GCPhys); rc = VBOXSTRICTRC_VAL(rc2); Log4(("ApicAccess rc=%d\n", rc)); if ( rc == VINF_SUCCESS || rc == VERR_PAGE_TABLE_NOT_PRESENT || rc == VERR_PAGE_NOT_PRESENT) { VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RSP | HM_CHANGED_GUEST_RFLAGS | HM_CHANGED_VMX_GUEST_APIC_STATE); rc = VINF_SUCCESS; } break; } default: Log4(("ApicAccess uAccessType=%#x\n", uAccessType)); rc = VINF_EM_RAW_EMULATE_INSTR; break; } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess); return rc; } /** * VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional * VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); /* We should -not- get this VM-exit if the guest's debug registers were active. */ if (pVmxTransient->fWasGuestDebugStateActive) { AssertMsgFailed(("Unexpected MOV DRx exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); HMVMX_RETURN_UNEXPECTED_EXIT(); } int rc = VERR_INTERNAL_ERROR_5; if ( !DBGFIsStepping(pVCpu) && !pVCpu->hm.s.fSingleInstruction && !pVmxTransient->fWasHyperDebugStateActive) { /* Don't intercept MOV DRx and #DB any more. */ pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT; rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRCReturn(rc, rc); if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { #ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_DB); rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap); AssertRCReturn(rc, rc); #endif } /* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */ VMMRZCallRing3Disable(pVCpu); HM_DISABLE_PREEMPT_IF_NEEDED(); /* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */ CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */); Assert(CPUMIsGuestDebugStateActive(pVCpu) || HC_ARCH_BITS == 32); HM_RESTORE_PREEMPT_IF_NEEDED(); VMMRZCallRing3Enable(pVCpu); #ifdef VBOX_WITH_STATISTICS rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); AssertRCReturn(rc, rc); if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE) STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite); else STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead); #endif STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch); return VINF_SUCCESS; } /* * EMInterpretDRx[Write|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER, CS. EFER is always up-to-date. * Update the segment registers and DR7 from the CPU. */ rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); Log4(("CS:RIP=%04x:%#RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip)); PVM pVM = pVCpu->CTX_SUFF(pVM); if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE) { rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification), VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification)); if (RT_SUCCESS(rc)) VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite); } else { rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification), VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification)); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead); } Assert(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER); if (RT_SUCCESS(rc)) { int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient); AssertRCReturn(rc2, rc2); } return rc; } /** * VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG). * Conditional VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging); /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */ int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient); if (RT_UNLIKELY(rc == VINF_HM_DOUBLE_FAULT)) return VINF_SUCCESS; else if (RT_UNLIKELY(rc == VINF_EM_RESET)) return rc; RTGCPHYS GCPhys = 0; rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys); #if 0 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */ #else /* Aggressive state sync. for now. */ rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); #endif AssertRCReturn(rc, rc); /* * If we succeed, resume guest execution. * If we fail in interpreting the instruction because we couldn't get the guest physical address * of the page containing the instruction via the guest's page tables (we would invalidate the guest page * in the host TLB), resume execution which would cause a guest page fault to let the guest handle this * weird case. See @bugref{6043}. */ PVM pVM = pVCpu->CTX_SUFF(pVM); VBOXSTRICTRC rc2 = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pMixedCtx), GCPhys, UINT32_MAX); rc = VBOXSTRICTRC_VAL(rc2); Log4(("EPT misconfig at %#RGv RIP=%#RX64 rc=%d\n", GCPhys, pMixedCtx->rip, rc)); if ( rc == VINF_SUCCESS || rc == VERR_PAGE_TABLE_NOT_PRESENT || rc == VERR_PAGE_NOT_PRESENT) { /* Successfully handled MMIO operation. */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RSP | HM_CHANGED_GUEST_RFLAGS | HM_CHANGED_VMX_GUEST_APIC_STATE); rc = VINF_SUCCESS; } return rc; } /** * VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional * VM-exit. */ HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(); Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging); /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */ int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient); if (RT_UNLIKELY(rc == VINF_HM_DOUBLE_FAULT)) return VINF_SUCCESS; else if (RT_UNLIKELY(rc == VINF_EM_RESET)) return rc; RTGCPHYS GCPhys = 0; rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys); rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); #if 0 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */ #else /* Aggressive state sync. for now. */ rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); #endif AssertRCReturn(rc, rc); /* Intel spec. Table 27-7 "Exit Qualifications for EPT violations". */ AssertMsg(((pVmxTransient->uExitQualification >> 7) & 3) != 2, ("%#RX64", pVmxTransient->uExitQualification)); RTGCUINT uErrorCode = 0; if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_INSTR_FETCH) uErrorCode |= X86_TRAP_PF_ID; if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_DATA_WRITE) uErrorCode |= X86_TRAP_PF_RW; if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_ENTRY_PRESENT) uErrorCode |= X86_TRAP_PF_P; TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode); Log4(("EPT violation %#x at %#RX64 ErrorCode %#x CS:EIP=%04x:%#RX64\n", pVmxTransient->uExitQualification, GCPhys, uErrorCode, pMixedCtx->cs.Sel, pMixedCtx->rip)); /* Handle the pagefault trap for the nested shadow table. */ PVM pVM = pVCpu->CTX_SUFF(pVM); rc = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pMixedCtx), GCPhys); TRPMResetTrap(pVCpu); /* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */ if ( rc == VINF_SUCCESS || rc == VERR_PAGE_TABLE_NOT_PRESENT || rc == VERR_PAGE_NOT_PRESENT) { /* Successfully synced our nested page tables. */ STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RSP | HM_CHANGED_GUEST_RFLAGS); return VINF_SUCCESS; } Log4(("EPT return to ring-3 rc=%Rrc\n", rc)); return rc; } /** @} */ /* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */ /* -=-=-=-=-=-=-=-=-=- VM-exit Exception Handlers -=-=-=-=-=-=-=-=-=-=- */ /* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */ /** @name VM-exit exception handlers. * @{ */ /** * VM-exit exception handler for #MF (Math Fault: floating point exception). */ static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF); int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); if (!(pMixedCtx->cr0 & X86_CR0_NE)) { /* Old-style FPU error reporting needs some extra work. */ /** @todo don't fall back to the recompiler, but do it manually. */ return VERR_EM_INTERPRETER; } hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */); return rc; } /** * VM-exit exception handler for #BP (Breakpoint exception). */ static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP); /** @todo Try optimize this by not saving the entire guest state unless * really needed. */ int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); PVM pVM = pVCpu->CTX_SUFF(pVM); rc = DBGFRZTrap03Handler(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx)); if (rc == VINF_EM_RAW_GUEST_TRAP) { rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient); rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient); rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient); AssertRCReturn(rc, rc); hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */); } Assert(rc == VINF_SUCCESS || rc == VINF_EM_RAW_GUEST_TRAP || rc == VINF_EM_DBG_BREAKPOINT); return rc; } /** * VM-exit exception handler for #DB (Debug exception). */ static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB); Log6(("XcptDB\n")); /* * Get the DR6-like values from the exit qualification and pass it to DBGF * for processing. */ int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); AssertRCReturn(rc, rc); /* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */ uint64_t uDR6 = X86_DR6_INIT_VAL; uDR6 |= ( pVmxTransient->uExitQualification & (X86_DR6_B0 | X86_DR6_B1 | X86_DR6_B2 | X86_DR6_B3 | X86_DR6_BD | X86_DR6_BS)); rc = DBGFRZTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pMixedCtx), uDR6, pVCpu->hm.s.fSingleInstruction); if (rc == VINF_EM_RAW_GUEST_TRAP) { /* * The exception was for the guest. Update DR6, DR7.GD and * IA32_DEBUGCTL.LBR before forwarding it. * (See Intel spec. 27.1 "Architectural State before a VM-Exit".) */ VMMRZCallRing3Disable(pVCpu); HM_DISABLE_PREEMPT_IF_NEEDED(); pMixedCtx->dr[6] &= ~X86_DR6_B_MASK; pMixedCtx->dr[6] |= uDR6; if (CPUMIsGuestDebugStateActive(pVCpu)) ASMSetDR6(pMixedCtx->dr[6]); HM_RESTORE_PREEMPT_IF_NEEDED(); VMMRZCallRing3Enable(pVCpu); rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); /* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */ pMixedCtx->dr[7] &= ~X86_DR7_GD; /* Paranoia. */ pMixedCtx->dr[7] &= ~X86_DR7_RAZ_MASK; pMixedCtx->dr[7] |= X86_DR7_RA1_MASK; rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)pMixedCtx->dr[7]); AssertRCReturn(rc, rc); /* * Raise #DB in the guest. */ rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient); rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient); rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient); AssertRCReturn(rc, rc); hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */); return VINF_SUCCESS; } /* * Not a guest trap, must be a hypervisor related debug event then. * Update DR6 in case someone is interested in it. */ AssertMsg(rc == VINF_EM_DBG_STEPPED || rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc)); AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5); CPUMSetHyperDR6(pVCpu, uDR6); return rc; } /** * VM-exit exception handler for #NM (Device-not-available exception: floating * point exception). */ static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(); /* We require CR0 and EFER. EFER is always up-to-date. */ int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); /* We're playing with the host CPU state here, have to disable preemption or longjmp. */ VMMRZCallRing3Disable(pVCpu); HM_DISABLE_PREEMPT_IF_NEEDED(); /* If the guest FPU was active at the time of the #NM exit, then it's a guest fault. */ if (pVmxTransient->fWasGuestFPUStateActive) { rc = VINF_EM_RAW_GUEST_TRAP; Assert(CPUMIsGuestFPUStateActive(pVCpu) || VMCPU_HMCF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0)); } else { #ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS Assert(!pVmxTransient->fWasGuestFPUStateActive); #endif rc = CPUMR0Trap07Handler(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx); Assert(rc == VINF_EM_RAW_GUEST_TRAP || (rc == VINF_SUCCESS && CPUMIsGuestFPUStateActive(pVCpu))); } HM_RESTORE_PREEMPT_IF_NEEDED(); VMMRZCallRing3Enable(pVCpu); if (rc == VINF_SUCCESS) { /* Guest FPU state was activated, we'll want to change CR0 FPU intercepts before the next VM-reentry. */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_CR0); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowNM); pVCpu->hm.s.fUseGuestFpu = true; } else { /* Forward #NM to the guest. */ Assert(rc == VINF_EM_RAW_GUEST_TRAP); rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient); AssertRCReturn(rc, rc); hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr, 0 /* error code */, 0 /* GCPtrFaultAddress */); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNM); } return VINF_SUCCESS; } /** * VM-exit exception handler for #GP (General-protection exception). * * @remarks Requires pVmxTransient->uExitIntInfo to be up-to-date. */ static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP); int rc = VERR_INTERNAL_ERROR_5; if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) { #ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS /* If the guest is not in real-mode or we have unrestricted execution support, reflect #GP to the guest. */ rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient); rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient); rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient); rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); Log4(("#GP Gst: RIP %#RX64 ErrorCode=%#x CR0=%#RX64 CPL=%u\n", pMixedCtx->rip, pVmxTransient->uExitIntErrorCode, pMixedCtx->cr0, CPUMGetGuestCPL(pVCpu))); hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */); return rc; #else /* We don't intercept #GP. */ AssertMsgFailed(("Unexpected VM-exit caused by #GP exception\n")); NOREF(pVmxTransient); return VERR_VMX_UNEXPECTED_EXCEPTION; #endif } Assert(CPUMIsGuestInRealModeEx(pMixedCtx)); Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest); /* EMInterpretDisasCurrent() requires a lot of the state, save the entire state. */ rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); PDISCPUSTATE pDis = &pVCpu->hm.s.DisState; uint32_t cbOp = 0; PVM pVM = pVCpu->CTX_SUFF(pVM); rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp); if (RT_SUCCESS(rc)) { rc = VINF_SUCCESS; Assert(cbOp == pDis->cbInstr); Log4(("#GP Disas OpCode=%u CS:EIP %04x:%#RX64\n", pDis->pCurInstr->uOpcode, pMixedCtx->cs.Sel, pMixedCtx->rip)); switch (pDis->pCurInstr->uOpcode) { case OP_CLI: { pMixedCtx->eflags.Bits.u1IF = 0; pMixedCtx->rip += pDis->cbInstr; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCli); break; } case OP_STI: { pMixedCtx->eflags.Bits.u1IF = 1; pMixedCtx->rip += pDis->cbInstr; EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip); Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitSti); break; } case OP_HLT: { rc = VINF_EM_HALT; pMixedCtx->rip += pDis->cbInstr; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt); break; } case OP_POPF: { Log4(("POPF CS:RIP %04x:%#RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip)); uint32_t cbParm = 0; uint32_t uMask = 0; if (pDis->fPrefix & DISPREFIX_OPSIZE) { cbParm = 4; uMask = 0xffffffff; } else { cbParm = 2; uMask = 0xffff; } /* Get the stack pointer & pop the contents of the stack onto Eflags. */ RTGCPTR GCPtrStack = 0; X86EFLAGS Eflags; rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0, &GCPtrStack); if (RT_SUCCESS(rc)) { Assert(sizeof(Eflags.u32) >= cbParm); Eflags.u32 = 0; rc = PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u32, cbParm); } if (RT_FAILURE(rc)) { rc = VERR_EM_INTERPRETER; break; } Log4(("POPF %#x -> %#RX64 mask=%#x RIP=%#RX64\n", Eflags.u, pMixedCtx->rsp, uMask, pMixedCtx->rip)); pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ~(X86_EFL_POPF_BITS & uMask)) | (Eflags.u32 & X86_EFL_POPF_BITS & uMask); pMixedCtx->eflags.Bits.u1RF = 0; /* The RF bit is always cleared by POPF; see Intel Instruction reference. */ pMixedCtx->esp += cbParm; pMixedCtx->esp &= uMask; pMixedCtx->rip += pDis->cbInstr; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RSP | HM_CHANGED_GUEST_RFLAGS); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPopf); break; } case OP_PUSHF: { uint32_t cbParm = 0; uint32_t uMask = 0; if (pDis->fPrefix & DISPREFIX_OPSIZE) { cbParm = 4; uMask = 0xffffffff; } else { cbParm = 2; uMask = 0xffff; } /* Get the stack pointer & push the contents of eflags onto the stack. */ RTGCPTR GCPtrStack = 0; rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), (pMixedCtx->esp - cbParm) & uMask, SELMTOFLAT_FLAGS_CPL0, &GCPtrStack); if (RT_FAILURE(rc)) { rc = VERR_EM_INTERPRETER; break; } X86EFLAGS Eflags = pMixedCtx->eflags; /* The RF & VM bits are cleared on image stored on stack; see Intel Instruction reference for PUSHF. */ Eflags.Bits.u1RF = 0; Eflags.Bits.u1VM = 0; rc = PGMPhysWrite(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u, cbParm); if (RT_FAILURE(rc)) { rc = VERR_EM_INTERPRETER; break; } Log4(("PUSHF %#x -> %#RGv\n", Eflags.u, GCPtrStack)); pMixedCtx->esp -= cbParm; pMixedCtx->esp &= uMask; pMixedCtx->rip += pDis->cbInstr; VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RSP); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPushf); break; } case OP_IRET: { /** @todo Handle 32-bit operand sizes and check stack limits. See Intel * instruction reference. */ RTGCPTR GCPtrStack = 0; uint32_t uMask = 0xffff; uint16_t aIretFrame[3]; if (pDis->fPrefix & (DISPREFIX_OPSIZE | DISPREFIX_ADDRSIZE)) { rc = VERR_EM_INTERPRETER; break; } rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0, &GCPtrStack); if (RT_SUCCESS(rc)) rc = PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &aIretFrame[0], sizeof(aIretFrame)); if (RT_FAILURE(rc)) { rc = VERR_EM_INTERPRETER; break; } pMixedCtx->eip = 0; pMixedCtx->ip = aIretFrame[0]; pMixedCtx->cs.Sel = aIretFrame[1]; pMixedCtx->cs.ValidSel = aIretFrame[1]; pMixedCtx->cs.u64Base = (uint64_t)pMixedCtx->cs.Sel << 4; pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ~(X86_EFL_POPF_BITS & uMask)) | (aIretFrame[2] & X86_EFL_POPF_BITS & uMask); pMixedCtx->sp += sizeof(aIretFrame); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_SEGMENT_REGS | HM_CHANGED_GUEST_RSP | HM_CHANGED_GUEST_RFLAGS); Log4(("IRET %#RX32 to %04x:%x\n", GCPtrStack, pMixedCtx->cs.Sel, pMixedCtx->ip)); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIret); break; } case OP_INT: { uint16_t uVector = pDis->Param1.uValue & 0xff; hmR0VmxSetPendingIntN(pVCpu, pMixedCtx, uVector, pDis->cbInstr); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt); break; } case OP_INTO: { if (pMixedCtx->eflags.Bits.u1OF) { hmR0VmxSetPendingXcptOF(pVCpu, pMixedCtx, pDis->cbInstr); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt); } break; } default: { VBOXSTRICTRC rc2 = EMInterpretInstructionDisasState(pVCpu, pDis, CPUMCTX2CORE(pMixedCtx), 0 /* pvFault */, EMCODETYPE_SUPERVISOR); rc = VBOXSTRICTRC_VAL(rc2); VMCPU_HMCF_SET(pVCpu, HM_CHANGED_ALL_GUEST); Log4(("#GP rc=%Rrc\n", rc)); break; } } } else rc = VERR_EM_INTERPRETER; AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER || rc == VINF_PGM_CHANGE_MODE || rc == VINF_EM_HALT, ("#GP Unexpected rc=%Rrc\n", rc)); return rc; } #ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS /** * VM-exit exception handler wrapper for generic exceptions. Simply re-injects * the exception reported in the VMX transient structure back into the VM. * * @remarks Requires uExitIntInfo in the VMX transient structure to be * up-to-date. */ static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(); /* Re-inject the exception into the guest. This cannot be a double-fault condition which would have been handled in hmR0VmxCheckExitDueToEventDelivery(). */ int rc = hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient); rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient); AssertRCReturn(rc, rc); Assert(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO); hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */); return VINF_SUCCESS; } #endif /** * VM-exit exception handler for #PF (Page-fault exception). */ static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient) { HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(); PVM pVM = pVCpu->CTX_SUFF(pVM); int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient); rc |= hmR0VmxReadExitIntInfoVmcs(pVmxTransient); rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient); AssertRCReturn(rc, rc); #if defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) || defined(HMVMX_ALWAYS_TRAP_PF) if (pVM->hm.s.fNestedPaging) { pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */ if (RT_LIKELY(!pVmxTransient->fVectoringPF)) { pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */ hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */, pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQualification); } else { /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */ hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx); Log4(("Pending #DF due to vectoring #PF. NP\n")); } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); return rc; } #else Assert(!pVM->hm.s.fNestedPaging); NOREF(pVM); #endif rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRCReturn(rc, rc); Log4(("#PF: cr2=%#RX64 cs:rip=%#04x:%#RX64 uErrCode %#RX32 cr3=%#RX64\n", pVmxTransient->uExitQualification, pMixedCtx->cs.Sel, pMixedCtx->rip, pVmxTransient->uExitIntErrorCode, pMixedCtx->cr3)); TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQualification, (RTGCUINT)pVmxTransient->uExitIntErrorCode); rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pMixedCtx), (RTGCPTR)pVmxTransient->uExitQualification); Log4(("#PF: rc=%Rrc\n", rc)); if (rc == VINF_SUCCESS) { /* Successfully synced shadow pages tables or emulated an MMIO instruction. */ /** @todo this isn't quite right, what if guest does lgdt with some MMIO * memory? We don't update the whole state here... */ VMCPU_HMCF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RSP | HM_CHANGED_GUEST_RFLAGS | HM_CHANGED_VMX_GUEST_APIC_STATE); TRPMResetTrap(pVCpu); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF); return rc; } else if (rc == VINF_EM_RAW_GUEST_TRAP) { if (!pVmxTransient->fVectoringPF) { /* It's a guest page fault and needs to be reflected to the guest. */ uint32_t uGstErrorCode = TRPMGetErrorCode(pVCpu); TRPMResetTrap(pVCpu); pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */ pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */ hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */, uGstErrorCode, pVmxTransient->uExitQualification); } else { /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */ TRPMResetTrap(pVCpu); pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */ hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx); Log4(("#PF: Pending #DF due to vectoring #PF\n")); } STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); return VINF_SUCCESS; } TRPMResetTrap(pVCpu); STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM); return rc; } /** @} */