/* $Id: IEMAllCImplVmxInstr.cpp 103609 2024-02-29 03:51:57Z vboxsync $ */ /** @file * IEM - VT-x instruction implementation. */ /* * Copyright (C) 2011-2023 Oracle and/or its affiliates. * * This file is part of VirtualBox base platform packages, as * available from https://www.virtualbox.org. * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public License * as published by the Free Software Foundation, in version 3 of the * License. * * This program is distributed in the hope that it will be useful, but * WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, see . * * SPDX-License-Identifier: GPL-3.0-only */ /********************************************************************************************************************************* * Header Files * *********************************************************************************************************************************/ #define LOG_GROUP LOG_GROUP_IEM_VMX #define VMCPU_INCL_CPUM_GST_CTX #include #include #include #include #include #include #include #include #ifdef VBOX_WITH_NESTED_HWVIRT_VMX # include #endif #include #include "IEMInternal.h" #include #include #include #include #include #include #include #include #include #include "IEMInline.h" /********************************************************************************************************************************* * Defined Constants And Macros * *********************************************************************************************************************************/ #ifdef VBOX_WITH_NESTED_HWVIRT_VMX /** * Gets the ModR/M, SIB and displacement byte(s) from decoded opcodes given their * relative offsets. */ # ifdef IEM_WITH_CODE_TLB /** @todo IEM TLB */ # define IEM_MODRM_GET_U8(a_pVCpu, a_bModRm, a_offModRm) do { a_bModRm = 0; RT_NOREF(a_offModRm); } while (0) # define IEM_SIB_GET_U8(a_pVCpu, a_bSib, a_offSib) do { a_bSib = 0; RT_NOREF(a_offSib); } while (0) # define IEM_DISP_GET_U16(a_pVCpu, a_u16Disp, a_offDisp) do { a_u16Disp = 0; RT_NOREF(a_offDisp); } while (0) # define IEM_DISP_GET_S8_SX_U16(a_pVCpu, a_u16Disp, a_offDisp) do { a_u16Disp = 0; RT_NOREF(a_offDisp); } while (0) # define IEM_DISP_GET_U32(a_pVCpu, a_u32Disp, a_offDisp) do { a_u32Disp = 0; RT_NOREF(a_offDisp); } while (0) # define IEM_DISP_GET_S8_SX_U32(a_pVCpu, a_u32Disp, a_offDisp) do { a_u32Disp = 0; RT_NOREF(a_offDisp); } while (0) # define IEM_DISP_GET_S32_SX_U64(a_pVCpu, a_u64Disp, a_offDisp) do { a_u64Disp = 0; RT_NOREF(a_offDisp); } while (0) # define IEM_DISP_GET_S8_SX_U64(a_pVCpu, a_u64Disp, a_offDisp) do { a_u64Disp = 0; RT_NOREF(a_offDisp); } while (0) # if 0 # error "Implement me: Getting ModR/M, SIB, displacement needs to work even when instruction crosses a page boundary." # endif # else /* !IEM_WITH_CODE_TLB */ # define IEM_MODRM_GET_U8(a_pVCpu, a_bModRm, a_offModRm) \ do \ { \ Assert((a_offModRm) < (a_pVCpu)->iem.s.cbOpcode); \ (a_bModRm) = (a_pVCpu)->iem.s.abOpcode[(a_offModRm)]; \ } while (0) # define IEM_SIB_GET_U8(a_pVCpu, a_bSib, a_offSib) IEM_MODRM_GET_U8(a_pVCpu, a_bSib, a_offSib) # define IEM_DISP_GET_U16(a_pVCpu, a_u16Disp, a_offDisp) \ do \ { \ Assert((a_offDisp) + 1 < (a_pVCpu)->iem.s.cbOpcode); \ uint8_t const bTmpLo = (a_pVCpu)->iem.s.abOpcode[(a_offDisp)]; \ uint8_t const bTmpHi = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 1]; \ (a_u16Disp) = RT_MAKE_U16(bTmpLo, bTmpHi); \ } while (0) # define IEM_DISP_GET_S8_SX_U16(a_pVCpu, a_u16Disp, a_offDisp) \ do \ { \ Assert((a_offDisp) < (a_pVCpu)->iem.s.cbOpcode); \ (a_u16Disp) = (int8_t)((a_pVCpu)->iem.s.abOpcode[(a_offDisp)]); \ } while (0) # define IEM_DISP_GET_U32(a_pVCpu, a_u32Disp, a_offDisp) \ do \ { \ Assert((a_offDisp) + 3 < (a_pVCpu)->iem.s.cbOpcode); \ uint8_t const bTmp0 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp)]; \ uint8_t const bTmp1 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 1]; \ uint8_t const bTmp2 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 2]; \ uint8_t const bTmp3 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 3]; \ (a_u32Disp) = RT_MAKE_U32_FROM_U8(bTmp0, bTmp1, bTmp2, bTmp3); \ } while (0) # define IEM_DISP_GET_S8_SX_U32(a_pVCpu, a_u32Disp, a_offDisp) \ do \ { \ Assert((a_offDisp) + 1 < (a_pVCpu)->iem.s.cbOpcode); \ (a_u32Disp) = (int8_t)((a_pVCpu)->iem.s.abOpcode[(a_offDisp)]); \ } while (0) # define IEM_DISP_GET_S8_SX_U64(a_pVCpu, a_u64Disp, a_offDisp) \ do \ { \ Assert((a_offDisp) + 1 < (a_pVCpu)->iem.s.cbOpcode); \ (a_u64Disp) = (int8_t)((a_pVCpu)->iem.s.abOpcode[(a_offDisp)]); \ } while (0) # define IEM_DISP_GET_S32_SX_U64(a_pVCpu, a_u64Disp, a_offDisp) \ do \ { \ Assert((a_offDisp) + 3 < (a_pVCpu)->iem.s.cbOpcode); \ uint8_t const bTmp0 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp)]; \ uint8_t const bTmp1 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 1]; \ uint8_t const bTmp2 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 2]; \ uint8_t const bTmp3 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 3]; \ (a_u64Disp) = (int32_t)RT_MAKE_U32_FROM_U8(bTmp0, bTmp1, bTmp2, bTmp3); \ } while (0) # endif /* !IEM_WITH_CODE_TLB */ /** Check for VMX instructions requiring to be in VMX operation. * @note Any changes here, check if IEMOP_HLP_IN_VMX_OPERATION needs updating. */ # define IEM_VMX_IN_VMX_OPERATION(a_pVCpu, a_szInstr, a_InsDiagPrefix) \ do \ { \ if (IEM_VMX_IS_ROOT_MODE(a_pVCpu)) \ { /* likely */ } \ else \ { \ Log((a_szInstr ": Not in VMX operation (root mode) -> #UD\n")); \ (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.enmDiag = a_InsDiagPrefix##_VmxRoot; \ return iemRaiseUndefinedOpcode(a_pVCpu); \ } \ } while (0) /** Marks a VM-entry failure with a diagnostic reason, logs and returns. */ # define IEM_VMX_VMENTRY_FAILED_RET(a_pVCpu, a_pszInstr, a_pszFailure, a_VmxDiag) \ do \ { \ LogRel(("%s: VM-entry failed! enmDiag=%u (%s) -> %s\n", (a_pszInstr), (a_VmxDiag), \ HMGetVmxDiagDesc(a_VmxDiag), (a_pszFailure))); \ (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.enmDiag = (a_VmxDiag); \ return VERR_VMX_VMENTRY_FAILED; \ } while (0) /** Marks a VM-entry failure with an return code, diagnostic reason, logs and * returns. */ # define IEM_VMX_VMENTRY_FAILED_RET_2(a_pVCpu, a_pszInstr, a_pszFailure, a_VmxDiag, a_rc) \ do \ { \ LogRel(("%s: VM-entry failed! rc=%Rrc enmDiag=%u (%s) -> %s\n", (a_pszInstr), (a_rc), (a_VmxDiag), \ HMGetVmxDiagDesc(a_VmxDiag), (a_pszFailure))); \ (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.enmDiag = (a_VmxDiag); \ return VERR_VMX_VMENTRY_FAILED; \ } while (0) /** Marks a VM-exit failure with a diagnostic reason and logs. */ # define IEM_VMX_VMEXIT_FAILED(a_pVCpu, a_uExitReason, a_pszFailure, a_VmxDiag) \ do \ { \ LogRel(("VM-exit failed! uExitReason=%u enmDiag=%u (%s) -> %s\n", (a_uExitReason), (a_VmxDiag), \ HMGetVmxDiagDesc(a_VmxDiag), (a_pszFailure))); \ (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.enmDiag = (a_VmxDiag); \ } while (0) /** Marks a VM-exit failure with a diagnostic reason, logs and returns. */ # define IEM_VMX_VMEXIT_FAILED_RET(a_pVCpu, a_uExitReason, a_pszFailure, a_VmxDiag) \ do \ { \ IEM_VMX_VMEXIT_FAILED(a_pVCpu, a_uExitReason, a_pszFailure, a_VmxDiag); \ return VERR_VMX_VMEXIT_FAILED; \ } while (0) /********************************************************************************************************************************* * Global Variables * *********************************************************************************************************************************/ /** @todo NSTVMX: The following VM-exit intercepts are pending: * VMX_EXIT_IO_SMI * VMX_EXIT_SMI * VMX_EXIT_GETSEC * VMX_EXIT_RSM * VMX_EXIT_MONITOR (APIC access VM-exit caused by MONITOR pending) * VMX_EXIT_ERR_MACHINE_CHECK (we never need to raise this?) * VMX_EXIT_VMFUNC * VMX_EXIT_ENCLS * VMX_EXIT_PML_FULL * VMX_EXIT_XSAVES * VMX_EXIT_XRSTORS */ /** * Map of VMCS field encodings to their virtual-VMCS structure offsets. * * The first array dimension is VMCS field encoding of Width OR'ed with Type and the * second dimension is the Index, see VMXVMCSFIELD. */ uint16_t const g_aoffVmcsMap[16][VMX_V_VMCS_MAX_INDEX + 1] = { /* VMX_VMCSFIELD_WIDTH_16BIT | VMX_VMCSFIELD_TYPE_CONTROL: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u16Vpid), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u16PostIntNotifyVector), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u16EptpIndex), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u16HlatPrefixSize), /* 4-11 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 12-19 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 20-27 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 28-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_16BIT | VMX_VMCSFIELD_TYPE_VMEXIT_INFO: */ { /* 0-7 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 8-15 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 16-23 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 24-31 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 32-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_16BIT | VMX_VMCSFIELD_TYPE_GUEST_STATE: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, GuestEs), /* 1 */ RT_UOFFSETOF(VMXVVMCS, GuestCs), /* 2 */ RT_UOFFSETOF(VMXVVMCS, GuestSs), /* 3 */ RT_UOFFSETOF(VMXVVMCS, GuestDs), /* 4 */ RT_UOFFSETOF(VMXVVMCS, GuestFs), /* 5 */ RT_UOFFSETOF(VMXVVMCS, GuestGs), /* 6 */ RT_UOFFSETOF(VMXVVMCS, GuestLdtr), /* 7 */ RT_UOFFSETOF(VMXVVMCS, GuestTr), /* 8 */ RT_UOFFSETOF(VMXVVMCS, u16GuestIntStatus), /* 9 */ RT_UOFFSETOF(VMXVVMCS, u16PmlIndex), /* 10-17 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 18-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 26-33 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 34 */ UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_16BIT | VMX_VMCSFIELD_TYPE_HOST_STATE: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, HostEs), /* 1 */ RT_UOFFSETOF(VMXVVMCS, HostCs), /* 2 */ RT_UOFFSETOF(VMXVVMCS, HostSs), /* 3 */ RT_UOFFSETOF(VMXVVMCS, HostDs), /* 4 */ RT_UOFFSETOF(VMXVVMCS, HostFs), /* 5 */ RT_UOFFSETOF(VMXVVMCS, HostGs), /* 6 */ RT_UOFFSETOF(VMXVVMCS, HostTr), /* 7-14 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 15-22 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 23-30 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 31-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_64BIT | VMX_VMCSFIELD_TYPE_CONTROL: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64AddrIoBitmapA), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64AddrIoBitmapB), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64AddrMsrBitmap), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64AddrExitMsrStore), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64AddrExitMsrLoad), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64AddrEntryMsrLoad), /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64ExecVmcsPtr), /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64AddrPml), /* 8 */ RT_UOFFSETOF(VMXVVMCS, u64TscOffset), /* 9 */ RT_UOFFSETOF(VMXVVMCS, u64AddrVirtApic), /* 10 */ RT_UOFFSETOF(VMXVVMCS, u64AddrApicAccess), /* 11 */ RT_UOFFSETOF(VMXVVMCS, u64AddrPostedIntDesc), /* 12 */ RT_UOFFSETOF(VMXVVMCS, u64VmFuncCtls), /* 13 */ RT_UOFFSETOF(VMXVVMCS, u64EptPtr), /* 14 */ RT_UOFFSETOF(VMXVVMCS, u64EoiExitBitmap0), /* 15 */ RT_UOFFSETOF(VMXVVMCS, u64EoiExitBitmap1), /* 16 */ RT_UOFFSETOF(VMXVVMCS, u64EoiExitBitmap2), /* 17 */ RT_UOFFSETOF(VMXVVMCS, u64EoiExitBitmap3), /* 18 */ RT_UOFFSETOF(VMXVVMCS, u64AddrEptpList), /* 19 */ RT_UOFFSETOF(VMXVVMCS, u64AddrVmreadBitmap), /* 20 */ RT_UOFFSETOF(VMXVVMCS, u64AddrVmwriteBitmap), /* 21 */ RT_UOFFSETOF(VMXVVMCS, u64AddrXcptVeInfo), /* 22 */ RT_UOFFSETOF(VMXVVMCS, u64XssExitBitmap), /* 23 */ RT_UOFFSETOF(VMXVVMCS, u64EnclsExitBitmap), /* 24 */ RT_UOFFSETOF(VMXVVMCS, u64SppTablePtr), /* 25 */ RT_UOFFSETOF(VMXVVMCS, u64TscMultiplier), /* 26 */ RT_UOFFSETOF(VMXVVMCS, u64ProcCtls3), /* 27 */ RT_UOFFSETOF(VMXVVMCS, u64EnclvExitBitmap), /* 28 */ UINT16_MAX, /* 29 */ UINT16_MAX, /* 30 */ UINT16_MAX, /* 31 */ RT_UOFFSETOF(VMXVVMCS, u64PconfigExitBitmap), /* 32 */ RT_UOFFSETOF(VMXVVMCS, u64HlatPtr), /* 33 */ UINT16_MAX, /* 34 */ RT_UOFFSETOF(VMXVVMCS, u64ExitCtls2) }, /* VMX_VMCSFIELD_WIDTH_64BIT | VMX_VMCSFIELD_TYPE_VMEXIT_INFO: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64RoGuestPhysAddr), /* 1-8 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 9-16 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 17-24 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 25-32 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 33-34*/ UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_64BIT | VMX_VMCSFIELD_TYPE_GUEST_STATE: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64VmcsLinkPtr), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64GuestDebugCtlMsr), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPatMsr), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64GuestEferMsr), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPerfGlobalCtlMsr), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPdpte0), /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPdpte1), /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPdpte2), /* 8 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPdpte3), /* 9 */ RT_UOFFSETOF(VMXVVMCS, u64GuestBndcfgsMsr), /* 10 */ RT_UOFFSETOF(VMXVVMCS, u64GuestRtitCtlMsr), /* 11 */ UINT16_MAX, /* 12 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPkrsMsr), /* 13-20 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 21-28 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 29-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_64BIT | VMX_VMCSFIELD_TYPE_HOST_STATE: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64HostPatMsr), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64HostEferMsr), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64HostPerfGlobalCtlMsr), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64HostPkrsMsr), /* 4-11 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 12-19 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 20-27 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 28-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_32BIT | VMX_VMCSFIELD_TYPE_CONTROL: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u32PinCtls), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u32ProcCtls), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u32XcptBitmap), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u32XcptPFMask), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u32XcptPFMatch), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u32Cr3TargetCount), /* 6 */ RT_UOFFSETOF(VMXVVMCS, u32ExitCtls), /* 7 */ RT_UOFFSETOF(VMXVVMCS, u32ExitMsrStoreCount), /* 8 */ RT_UOFFSETOF(VMXVVMCS, u32ExitMsrLoadCount), /* 9 */ RT_UOFFSETOF(VMXVVMCS, u32EntryCtls), /* 10 */ RT_UOFFSETOF(VMXVVMCS, u32EntryMsrLoadCount), /* 11 */ RT_UOFFSETOF(VMXVVMCS, u32EntryIntInfo), /* 12 */ RT_UOFFSETOF(VMXVVMCS, u32EntryXcptErrCode), /* 13 */ RT_UOFFSETOF(VMXVVMCS, u32EntryInstrLen), /* 14 */ RT_UOFFSETOF(VMXVVMCS, u32TprThreshold), /* 15 */ RT_UOFFSETOF(VMXVVMCS, u32ProcCtls2), /* 16 */ RT_UOFFSETOF(VMXVVMCS, u32PleGap), /* 17 */ RT_UOFFSETOF(VMXVVMCS, u32PleWindow), /* 18-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 26-33 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 34 */ UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_32BIT | VMX_VMCSFIELD_TYPE_VMEXIT_INFO: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u32RoVmInstrError), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitReason), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitIntInfo), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitIntErrCode), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u32RoIdtVectoringInfo), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u32RoIdtVectoringErrCode), /* 6 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitInstrLen), /* 7 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitInstrInfo), /* 8-15 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 16-23 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 24-31 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 32-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_32BIT | VMX_VMCSFIELD_TYPE_GUEST_STATE: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u32GuestEsLimit), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u32GuestCsLimit), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u32GuestSsLimit), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u32GuestDsLimit), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u32GuestFsLimit), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u32GuestGsLimit), /* 6 */ RT_UOFFSETOF(VMXVVMCS, u32GuestLdtrLimit), /* 7 */ RT_UOFFSETOF(VMXVVMCS, u32GuestTrLimit), /* 8 */ RT_UOFFSETOF(VMXVVMCS, u32GuestGdtrLimit), /* 9 */ RT_UOFFSETOF(VMXVVMCS, u32GuestIdtrLimit), /* 10 */ RT_UOFFSETOF(VMXVVMCS, u32GuestEsAttr), /* 11 */ RT_UOFFSETOF(VMXVVMCS, u32GuestCsAttr), /* 12 */ RT_UOFFSETOF(VMXVVMCS, u32GuestSsAttr), /* 13 */ RT_UOFFSETOF(VMXVVMCS, u32GuestDsAttr), /* 14 */ RT_UOFFSETOF(VMXVVMCS, u32GuestFsAttr), /* 15 */ RT_UOFFSETOF(VMXVVMCS, u32GuestGsAttr), /* 16 */ RT_UOFFSETOF(VMXVVMCS, u32GuestLdtrAttr), /* 17 */ RT_UOFFSETOF(VMXVVMCS, u32GuestTrAttr), /* 18 */ RT_UOFFSETOF(VMXVVMCS, u32GuestIntrState), /* 19 */ RT_UOFFSETOF(VMXVVMCS, u32GuestActivityState), /* 20 */ RT_UOFFSETOF(VMXVVMCS, u32GuestSmBase), /* 21 */ RT_UOFFSETOF(VMXVVMCS, u32GuestSysenterCS), /* 22 */ UINT16_MAX, /* 23 */ RT_UOFFSETOF(VMXVVMCS, u32PreemptTimer), /* 24-31 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 32-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_32BIT | VMX_VMCSFIELD_TYPE_HOST_STATE: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u32HostSysenterCs), /* 1-8 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 9-16 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 17-24 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 25-32 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 33-34 */ UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_NATURAL | VMX_VMCSFIELD_TYPE_CONTROL: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64Cr0Mask), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64Cr4Mask), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64Cr0ReadShadow), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64Cr4ReadShadow), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64Cr3Target0), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64Cr3Target1), /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64Cr3Target2), /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64Cr3Target3), /* 8-15 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 16-23 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 24-27 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 32-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_NATURAL | VMX_VMCSFIELD_TYPE_VMEXIT_INFO: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64RoExitQual), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64RoIoRcx), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64RoIoRsi), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64RoIoRdi), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64RoIoRip), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64RoGuestLinearAddr), /* 6-13 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 14-21 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 22-29 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 30-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_NATURAL | VMX_VMCSFIELD_TYPE_GUEST_STATE: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64GuestCr0), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64GuestCr3), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64GuestCr4), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64GuestEsBase), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64GuestCsBase), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64GuestSsBase), /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64GuestDsBase), /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64GuestFsBase), /* 8 */ RT_UOFFSETOF(VMXVVMCS, u64GuestGsBase), /* 9 */ RT_UOFFSETOF(VMXVVMCS, u64GuestLdtrBase), /* 10 */ RT_UOFFSETOF(VMXVVMCS, u64GuestTrBase), /* 11 */ RT_UOFFSETOF(VMXVVMCS, u64GuestGdtrBase), /* 12 */ RT_UOFFSETOF(VMXVVMCS, u64GuestIdtrBase), /* 13 */ RT_UOFFSETOF(VMXVVMCS, u64GuestDr7), /* 14 */ RT_UOFFSETOF(VMXVVMCS, u64GuestRsp), /* 15 */ RT_UOFFSETOF(VMXVVMCS, u64GuestRip), /* 16 */ RT_UOFFSETOF(VMXVVMCS, u64GuestRFlags), /* 17 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPendingDbgXcpts), /* 18 */ RT_UOFFSETOF(VMXVVMCS, u64GuestSysenterEsp), /* 19 */ RT_UOFFSETOF(VMXVVMCS, u64GuestSysenterEip), /* 20 */ RT_UOFFSETOF(VMXVVMCS, u64GuestSCetMsr), /* 21 */ RT_UOFFSETOF(VMXVVMCS, u64GuestSsp), /* 22 */ RT_UOFFSETOF(VMXVVMCS, u64GuestIntrSspTableAddrMsr), /* 23-27 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 31-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX }, /* VMX_VMCSFIELD_WIDTH_NATURAL | VMX_VMCSFIELD_TYPE_HOST_STATE: */ { /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64HostCr0), /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64HostCr3), /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64HostCr4), /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64HostFsBase), /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64HostGsBase), /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64HostTrBase), /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64HostGdtrBase), /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64HostIdtrBase), /* 8 */ RT_UOFFSETOF(VMXVVMCS, u64HostSysenterEsp), /* 9 */ RT_UOFFSETOF(VMXVVMCS, u64HostSysenterEip), /* 10 */ RT_UOFFSETOF(VMXVVMCS, u64HostRsp), /* 11 */ RT_UOFFSETOF(VMXVVMCS, u64HostRip), /* 12 */ RT_UOFFSETOF(VMXVVMCS, u64HostSCetMsr), /* 13 */ RT_UOFFSETOF(VMXVVMCS, u64HostSsp), /* 14 */ RT_UOFFSETOF(VMXVVMCS, u64HostIntrSspTableAddrMsr), /* 15-22 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 23-30 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, /* 31-34 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX } }; /** * Gets a host selector from the VMCS. * * @param pVmcs Pointer to the virtual VMCS. * @param iSelReg The index of the segment register (X86_SREG_XXX). */ DECLINLINE(RTSEL) iemVmxVmcsGetHostSelReg(PCVMXVVMCS pVmcs, uint8_t iSegReg) { Assert(iSegReg < X86_SREG_COUNT); RTSEL HostSel; uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_16BIT; uint8_t const uType = VMX_VMCSFIELD_TYPE_HOST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS16_HOST_ES_SEL, VMX_BF_VMCSFIELD_INDEX); Assert(uIndex <= VMX_V_VMCS_MAX_INDEX); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t const *pbVmcs = (uint8_t *)pVmcs; uint8_t const *pbField = pbVmcs + offField; HostSel = *(uint16_t *)pbField; return HostSel; } /** * Sets a guest segment register in the VMCS. * * @param pVmcs Pointer to the virtual VMCS. * @param iSegReg The index of the segment register (X86_SREG_XXX). * @param pSelReg Pointer to the segment register. */ static void iemVmxVmcsSetGuestSegReg(PCVMXVVMCS pVmcs, uint8_t iSegReg, PCCPUMSELREG pSelReg) RT_NOEXCEPT { Assert(pSelReg); Assert(iSegReg < X86_SREG_COUNT); /* Selector. */ { uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_16BIT; uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS16_GUEST_ES_SEL, VMX_BF_VMCSFIELD_INDEX); Assert(uIndex <= VMX_V_VMCS_MAX_INDEX); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t *pbVmcs = (uint8_t *)pVmcs; uint8_t *pbField = pbVmcs + offField; *(uint16_t *)pbField = pSelReg->Sel; } /* Limit. */ { uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_32BIT; uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS32_GUEST_ES_LIMIT, VMX_BF_VMCSFIELD_INDEX); Assert(uIndex <= VMX_V_VMCS_MAX_INDEX); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t *pbVmcs = (uint8_t *)pVmcs; uint8_t *pbField = pbVmcs + offField; *(uint32_t *)pbField = pSelReg->u32Limit; } /* Base. */ { uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_NATURAL; uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS_GUEST_ES_BASE, VMX_BF_VMCSFIELD_INDEX); Assert(uIndex <= VMX_V_VMCS_MAX_INDEX); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t const *pbVmcs = (uint8_t *)pVmcs; uint8_t const *pbField = pbVmcs + offField; *(uint64_t *)pbField = pSelReg->u64Base; } /* Attributes. */ { uint32_t const fValidAttrMask = X86DESCATTR_TYPE | X86DESCATTR_DT | X86DESCATTR_DPL | X86DESCATTR_P | X86DESCATTR_AVL | X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G | X86DESCATTR_UNUSABLE; uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_32BIT; uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, VMX_BF_VMCSFIELD_INDEX); Assert(uIndex <= VMX_V_VMCS_MAX_INDEX); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t *pbVmcs = (uint8_t *)pVmcs; uint8_t *pbField = pbVmcs + offField; *(uint32_t *)pbField = pSelReg->Attr.u & fValidAttrMask; } } /** * Gets a guest segment register from the VMCS. * * @returns VBox status code. * @param pVmcs Pointer to the virtual VMCS. * @param iSegReg The index of the segment register (X86_SREG_XXX). * @param pSelReg Where to store the segment register (only updated when * VINF_SUCCESS is returned). * * @remarks Warning! This does not validate the contents of the retrieved segment * register. */ static int iemVmxVmcsGetGuestSegReg(PCVMXVVMCS pVmcs, uint8_t iSegReg, PCPUMSELREG pSelReg) RT_NOEXCEPT { Assert(pSelReg); Assert(iSegReg < X86_SREG_COUNT); /* Selector. */ uint16_t u16Sel; { uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_16BIT; uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS16_GUEST_ES_SEL, VMX_BF_VMCSFIELD_INDEX); AssertReturn(uIndex <= VMX_V_VMCS_MAX_INDEX, VERR_IEM_IPE_3); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t const *pbVmcs = (uint8_t *)pVmcs; uint8_t const *pbField = pbVmcs + offField; u16Sel = *(uint16_t *)pbField; } /* Limit. */ uint32_t u32Limit; { uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_32BIT; uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS32_GUEST_ES_LIMIT, VMX_BF_VMCSFIELD_INDEX); AssertReturn(uIndex <= VMX_V_VMCS_MAX_INDEX, VERR_IEM_IPE_3); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t const *pbVmcs = (uint8_t *)pVmcs; uint8_t const *pbField = pbVmcs + offField; u32Limit = *(uint32_t *)pbField; } /* Base. */ uint64_t u64Base; { uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_NATURAL; uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS_GUEST_ES_BASE, VMX_BF_VMCSFIELD_INDEX); AssertReturn(uIndex <= VMX_V_VMCS_MAX_INDEX, VERR_IEM_IPE_3); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t const *pbVmcs = (uint8_t *)pVmcs; uint8_t const *pbField = pbVmcs + offField; u64Base = *(uint64_t *)pbField; /** @todo NSTVMX: Should we zero out high bits here for 32-bit virtual CPUs? */ } /* Attributes. */ uint32_t u32Attr; { uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_32BIT; uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE; uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, VMX_BF_VMCSFIELD_INDEX); AssertReturn(uIndex <= VMX_V_VMCS_MAX_INDEX, VERR_IEM_IPE_3); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; uint8_t const *pbVmcs = (uint8_t *)pVmcs; uint8_t const *pbField = pbVmcs + offField; u32Attr = *(uint32_t *)pbField; } pSelReg->Sel = u16Sel; pSelReg->ValidSel = u16Sel; pSelReg->fFlags = CPUMSELREG_FLAGS_VALID; pSelReg->u32Limit = u32Limit; pSelReg->u64Base = u64Base; pSelReg->Attr.u = u32Attr; return VINF_SUCCESS; } /** * Converts an IEM exception event type to a VMX event type. * * @returns The VMX event type. * @param uVector The interrupt / exception vector. * @param fFlags The IEM event flag (see IEM_XCPT_FLAGS_XXX). */ DECLINLINE(uint8_t) iemVmxGetEventType(uint32_t uVector, uint32_t fFlags) { /* Paranoia (callers may use these interchangeably). */ AssertCompile(VMX_EXIT_INT_INFO_TYPE_NMI == VMX_IDT_VECTORING_INFO_TYPE_NMI); AssertCompile(VMX_EXIT_INT_INFO_TYPE_HW_XCPT == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_EXT_INT == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_SW_XCPT == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_SW_INT == VMX_IDT_VECTORING_INFO_TYPE_SW_INT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_NMI == VMX_ENTRY_INT_INFO_TYPE_NMI); AssertCompile(VMX_EXIT_INT_INFO_TYPE_HW_XCPT == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_EXT_INT == VMX_ENTRY_INT_INFO_TYPE_EXT_INT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_SW_XCPT == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_SW_INT == VMX_ENTRY_INT_INFO_TYPE_SW_INT); AssertCompile(VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT == VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT); if (fFlags & IEM_XCPT_FLAGS_T_CPU_XCPT) { if (uVector == X86_XCPT_NMI) return VMX_EXIT_INT_INFO_TYPE_NMI; return VMX_EXIT_INT_INFO_TYPE_HW_XCPT; } if (fFlags & IEM_XCPT_FLAGS_T_SOFT_INT) { if (fFlags & (IEM_XCPT_FLAGS_BP_INSTR | IEM_XCPT_FLAGS_OF_INSTR)) return VMX_EXIT_INT_INFO_TYPE_SW_XCPT; if (fFlags & IEM_XCPT_FLAGS_ICEBP_INSTR) return VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT; return VMX_EXIT_INT_INFO_TYPE_SW_INT; } Assert(fFlags & IEM_XCPT_FLAGS_T_EXT_INT); return VMX_EXIT_INT_INFO_TYPE_EXT_INT; } /** * Determines whether the guest is using PAE paging given the VMCS. * * @returns @c true if PAE paging mode is used, @c false otherwise. * @param pVmcs Pointer to the virtual VMCS. * * @warning Only use this prior to switching the guest-CPU state with the * nested-guest CPU state! */ DECL_FORCE_INLINE(bool) iemVmxVmcsIsGuestPaePagingEnabled(PCVMXVVMCS pVmcs) { return ( !(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST) && (pVmcs->u64GuestCr4.u & X86_CR4_PAE) && (pVmcs->u64GuestCr0.u & X86_CR0_PG)); } /** * Sets the Exit qualification VMCS field. * * @param pVCpu The cross context virtual CPU structure. * @param u64ExitQual The Exit qualification. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetExitQual(PVMCPUCC pVCpu, uint64_t u64ExitQual) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64RoExitQual.u = u64ExitQual; } /** * Sets the VM-exit interruption information field. * * @param pVCpu The cross context virtual CPU structure. * @param uExitIntInfo The VM-exit interruption information. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetExitIntInfo(PVMCPUCC pVCpu, uint32_t uExitIntInfo) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32RoExitIntInfo = uExitIntInfo; } /** * Sets the VM-exit interruption error code. * * @param pVCpu The cross context virtual CPU structure. * @param uErrCode The error code. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetExitIntErrCode(PVMCPUCC pVCpu, uint32_t uErrCode) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32RoExitIntErrCode = uErrCode; } /** * Sets the IDT-vectoring information field. * * @param pVCpu The cross context virtual CPU structure. * @param uIdtVectorInfo The IDT-vectoring information. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetIdtVectoringInfo(PVMCPUCC pVCpu, uint32_t uIdtVectorInfo) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32RoIdtVectoringInfo = uIdtVectorInfo; } /** * Sets the IDT-vectoring error code field. * * @param pVCpu The cross context virtual CPU structure. * @param uErrCode The error code. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetIdtVectoringErrCode(PVMCPUCC pVCpu, uint32_t uErrCode) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32RoIdtVectoringErrCode = uErrCode; } /** * Sets the VM-exit guest-linear address VMCS field. * * @param pVCpu The cross context virtual CPU structure. * @param uGuestLinearAddr The VM-exit guest-linear address. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetExitGuestLinearAddr(PVMCPUCC pVCpu, uint64_t uGuestLinearAddr) { /* Bits 63:32 of guest-linear address MBZ if the guest isn't in long mode prior to the VM-exit. */ Assert(CPUMIsGuestInLongModeEx(IEM_GET_CTX(pVCpu)) || !(uGuestLinearAddr & UINT64_C(0xffffffff00000000))); pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64RoGuestLinearAddr.u = uGuestLinearAddr; } /** * Sets the VM-exit guest-physical address VMCS field. * * @param pVCpu The cross context virtual CPU structure. * @param uGuestPhysAddr The VM-exit guest-physical address. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetExitGuestPhysAddr(PVMCPUCC pVCpu, uint64_t uGuestPhysAddr) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64RoGuestPhysAddr.u = uGuestPhysAddr; } /** * Sets the VM-exit instruction length VMCS field. * * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The VM-exit instruction length in bytes. * * @remarks Callers may clear this field to 0. Hence, this function does not check * the validity of the instruction length. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetExitInstrLen(PVMCPUCC pVCpu, uint32_t cbInstr) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32RoExitInstrLen = cbInstr; } /** * Sets the VM-exit instruction info. VMCS field. * * @param pVCpu The cross context virtual CPU structure. * @param uExitInstrInfo The VM-exit instruction information. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetExitInstrInfo(PVMCPUCC pVCpu, uint32_t uExitInstrInfo) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32RoExitInstrInfo = uExitInstrInfo; } /** * Sets the guest pending-debug exceptions field. * * @param pVCpu The cross context virtual CPU structure. * @param uGuestPendingDbgXcpts The guest pending-debug exceptions. */ DECL_FORCE_INLINE(void) iemVmxVmcsSetGuestPendingDbgXcpts(PVMCPUCC pVCpu, uint64_t uGuestPendingDbgXcpts) { Assert(!(uGuestPendingDbgXcpts & VMX_VMCS_GUEST_PENDING_DEBUG_VALID_MASK)); pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64GuestPendingDbgXcpts.u = uGuestPendingDbgXcpts; } /** * Implements VMSucceed for VMX instruction success. * * @param pVCpu The cross context virtual CPU structure. */ DECL_FORCE_INLINE(void) iemVmxVmSucceed(PVMCPUCC pVCpu) { return CPUMSetGuestVmxVmSucceed(&pVCpu->cpum.GstCtx); } /** * Implements VMFailInvalid for VMX instruction failure. * * @param pVCpu The cross context virtual CPU structure. */ DECL_FORCE_INLINE(void) iemVmxVmFailInvalid(PVMCPUCC pVCpu) { return CPUMSetGuestVmxVmFailInvalid(&pVCpu->cpum.GstCtx); } /** * Implements VMFail for VMX instruction failure. * * @param pVCpu The cross context virtual CPU structure. * @param enmInsErr The VM instruction error. */ DECL_FORCE_INLINE(void) iemVmxVmFail(PVMCPUCC pVCpu, VMXINSTRERR enmInsErr) { return CPUMSetGuestVmxVmFail(&pVCpu->cpum.GstCtx, enmInsErr); } /** * Checks if the given auto-load/store MSR area count is valid for the * implementation. * * @returns @c true if it's within the valid limit, @c false otherwise. * @param pVCpu The cross context virtual CPU structure. * @param uMsrCount The MSR area count to check. */ DECL_FORCE_INLINE(bool) iemVmxIsAutoMsrCountValid(PCVMCPU pVCpu, uint32_t uMsrCount) { uint64_t const u64VmxMiscMsr = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Misc; uint32_t const cMaxSupportedMsrs = VMX_MISC_MAX_MSRS(u64VmxMiscMsr); Assert(cMaxSupportedMsrs <= VMX_V_AUTOMSR_AREA_SIZE / sizeof(VMXAUTOMSR)); if (uMsrCount <= cMaxSupportedMsrs) return true; return false; } /** * Flushes the current VMCS contents back to guest memory. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. */ DECL_FORCE_INLINE(int) iemVmxWriteCurrentVmcsToGstMem(PVMCPUCC pVCpu) { Assert(IEM_VMX_HAS_CURRENT_VMCS(pVCpu)); int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), IEM_VMX_GET_CURRENT_VMCS(pVCpu), &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs, sizeof(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs)); return rc; } /** * Populates the current VMCS contents from guest memory. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. */ DECL_FORCE_INLINE(int) iemVmxReadCurrentVmcsFromGstMem(PVMCPUCC pVCpu) { Assert(IEM_VMX_HAS_CURRENT_VMCS(pVCpu)); int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs, IEM_VMX_GET_CURRENT_VMCS(pVCpu), sizeof(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs)); return rc; } /** * Gets the instruction diagnostic for segment base checks during VM-entry of a * nested-guest. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegBase(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegBaseCs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegBaseDs; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegBaseEs; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegBaseFs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegBaseGs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegBaseSs; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_1); } } /** * Gets the instruction diagnostic for segment base checks during VM-entry of a * nested-guest that is in Virtual-8086 mode. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegBaseV86(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegBaseV86Cs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegBaseV86Ds; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegBaseV86Es; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegBaseV86Fs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegBaseV86Gs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegBaseV86Ss; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_2); } } /** * Gets the instruction diagnostic for segment limit checks during VM-entry of a * nested-guest that is in Virtual-8086 mode. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegLimitV86(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegLimitV86Cs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegLimitV86Ds; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegLimitV86Es; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegLimitV86Fs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegLimitV86Gs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegLimitV86Ss; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_3); } } /** * Gets the instruction diagnostic for segment attribute checks during VM-entry of a * nested-guest that is in Virtual-8086 mode. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegAttrV86(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrV86Cs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrV86Ds; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrV86Es; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrV86Fs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrV86Gs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrV86Ss; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_4); } } /** * Gets the instruction diagnostic for segment attributes reserved bits failure * during VM-entry of a nested-guest. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegAttrRsvd(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdCs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdDs; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrRsvdEs; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdFs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdGs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdSs; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_5); } } /** * Gets the instruction diagnostic for segment attributes descriptor-type * (code/segment or system) failure during VM-entry of a nested-guest. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegAttrDescType(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeCs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeDs; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeEs; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeFs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeGs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeSs; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_6); } } /** * Gets the instruction diagnostic for segment attributes descriptor-type * (code/segment or system) failure during VM-entry of a nested-guest. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegAttrPresent(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrPresentCs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrPresentDs; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrPresentEs; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrPresentFs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrPresentGs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrPresentSs; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_7); } } /** * Gets the instruction diagnostic for segment attribute granularity failure during * VM-entry of a nested-guest. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegAttrGran(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrGranCs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrGranDs; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrGranEs; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrGranFs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrGranGs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrGranSs; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_8); } } /** * Gets the instruction diagnostic for segment attribute DPL/RPL failure during * VM-entry of a nested-guest. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegAttrDplRpl(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplCs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplDs; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrDplRplEs; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplFs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplGs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplSs; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_9); } } /** * Gets the instruction diagnostic for segment attribute type accessed failure * during VM-entry of a nested-guest. * * @param iSegReg The segment index (X86_SREG_XXX). */ static VMXVDIAG iemVmxGetDiagVmentrySegAttrTypeAcc(unsigned iSegReg) RT_NOEXCEPT { switch (iSegReg) { case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccCs; case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccDs; case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccEs; case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccFs; case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccGs; case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccSs; IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_10); } } /** * Saves the guest control registers, debug registers and some MSRs are part of * VM-exit. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmexitSaveGuestControlRegsMsrs(PVMCPUCC pVCpu) RT_NOEXCEPT { /* * Saves the guest control registers, debug registers and some MSRs. * See Intel spec. 27.3.1 "Saving Control Registers, Debug Registers and MSRs". */ PVMXVVMCS pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* Save control registers. */ pVmcs->u64GuestCr0.u = pVCpu->cpum.GstCtx.cr0; pVmcs->u64GuestCr3.u = pVCpu->cpum.GstCtx.cr3; pVmcs->u64GuestCr4.u = pVCpu->cpum.GstCtx.cr4; /* Save SYSENTER CS, ESP, EIP. */ pVmcs->u32GuestSysenterCS = pVCpu->cpum.GstCtx.SysEnter.cs; if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { pVmcs->u64GuestSysenterEsp.u = pVCpu->cpum.GstCtx.SysEnter.esp; pVmcs->u64GuestSysenterEip.u = pVCpu->cpum.GstCtx.SysEnter.eip; } else { pVmcs->u64GuestSysenterEsp.s.Lo = pVCpu->cpum.GstCtx.SysEnter.esp; pVmcs->u64GuestSysenterEip.s.Lo = pVCpu->cpum.GstCtx.SysEnter.eip; } /* Save debug registers (DR7 and IA32_DEBUGCTL MSR). */ if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_DEBUG) { pVmcs->u64GuestDr7.u = pVCpu->cpum.GstCtx.dr[7]; /** @todo NSTVMX: Support IA32_DEBUGCTL MSR */ } /* Save PAT MSR. */ if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_PAT_MSR) pVmcs->u64GuestPatMsr.u = pVCpu->cpum.GstCtx.msrPAT; /* Save EFER MSR. */ if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_EFER_MSR) pVmcs->u64GuestEferMsr.u = pVCpu->cpum.GstCtx.msrEFER; /* We don't support clearing IA32_BNDCFGS MSR yet. */ Assert(!(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_CLEAR_BNDCFGS_MSR)); /* Nothing to do for SMBASE register - We don't support SMM yet. */ } /** * Saves the guest force-flags in preparation of entering the nested-guest. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmentrySaveNmiBlockingFF(PVMCPUCC pVCpu) RT_NOEXCEPT { /* We shouldn't be called multiple times during VM-entry. */ Assert(pVCpu->cpum.GstCtx.hwvirt.fSavedInhibit == 0); /* MTF should not be set outside VMX non-root mode. */ Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_MTF)); /* * Preserve the required force-flags. * * We cache and clear force-flags that would affect the execution of the * nested-guest. Cached flags are then restored while returning to the guest * if necessary. * * - VMCPU_FF_INHIBIT_INTERRUPTS need not be cached as it only affects * interrupts until the completion of the current VMLAUNCH/VMRESUME * instruction. Interrupt inhibition for any nested-guest instruction * is supplied by the guest-interruptibility state VMCS field and will * be set up as part of loading the guest state. Technically * blocking-by-STI is possible with VMLAUNCH/VMRESUME but we currently * disallow it since we can't distinguish it from blocking-by-MovSS * and no nested-hypervisor we care about uses STI immediately * followed by VMLAUNCH/VMRESUME. * * - VMCPU_FF_BLOCK_NMIS needs to be cached as VM-exits caused before * successful VM-entry (due to invalid guest-state) need to continue * blocking NMIs if it was in effect before VM-entry. * * - MTF need not be preserved as it's used only in VMX non-root mode and * is supplied through the VM-execution controls. * * The remaining FFs (e.g. timers, APIC updates) can stay in place so that * we will be able to generate interrupts that may cause VM-exits for * the nested-guest. */ pVCpu->cpum.GstCtx.hwvirt.fSavedInhibit = pVCpu->cpum.GstCtx.eflags.uBoth & CPUMCTX_INHIBIT_NMI; } /** * Restores the guest force-flags in preparation of exiting the nested-guest. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmexitRestoreNmiBlockingFF(PVMCPUCC pVCpu) RT_NOEXCEPT { /** @todo r=bird: why aren't we clearing the nested guest flags first here? * If there is some other code doing that already, it would be great * to point to it here... */ pVCpu->cpum.GstCtx.eflags.uBoth |= pVCpu->cpum.GstCtx.hwvirt.fSavedInhibit; pVCpu->cpum.GstCtx.hwvirt.fSavedInhibit = 0; } /** * Performs the VMX transition to/from VMX non-root mode. * * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The length of the current instruction. */ static int iemVmxTransition(PVMCPUCC pVCpu, uint8_t cbInstr) RT_NOEXCEPT { /* * Inform PGM about paging mode changes. * We include X86_CR0_PE because PGM doesn't handle paged-real mode yet, * see comment in iemMemPageTranslateAndCheckAccess(). */ int rc = PGMChangeMode(pVCpu, pVCpu->cpum.GstCtx.cr0 | X86_CR0_PE, pVCpu->cpum.GstCtx.cr4, pVCpu->cpum.GstCtx.msrEFER, true /* fForce */); if (RT_SUCCESS(rc)) { /* likely */ } else return rc; /* Invalidate IEM TLBs now that we've forced a PGM mode change. */ IEMTlbInvalidateAll(pVCpu); /* Inform CPUM (recompiler), can later be removed. */ CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_ALL); /* Re-initialize IEM cache/state after the drastic mode switch. */ iemReInitExec(pVCpu, cbInstr); return rc; } /** * Calculates the current VMX-preemption timer value. * * @returns The current VMX-preemption timer value. * @param pVCpu The cross context virtual CPU structure. */ static uint32_t iemVmxCalcPreemptTimer(PVMCPUCC pVCpu) RT_NOEXCEPT { /* * Assume the following: * PreemptTimerShift = 5 * VmcsPreemptTimer = 2 (i.e. need to decrement by 1 every 2 * RT_BIT(5) = 20000 TSC ticks) * EntryTick = 50000 (TSC at time of VM-entry) * * CurTick Delta PreemptTimerVal * ---------------------------------- * 60000 10000 2 * 80000 30000 1 * 90000 40000 0 -> VM-exit. * * If Delta >= VmcsPreemptTimer * RT_BIT(PreemptTimerShift) cause a VMX-preemption timer VM-exit. * The saved VMX-preemption timer value is calculated as follows: * PreemptTimerVal = VmcsPreemptTimer - (Delta / (VmcsPreemptTimer * RT_BIT(PreemptTimerShift))) * E.g.: * Delta = 10000 * Tmp = 10000 / (2 * 10000) = 0.5 * NewPt = 2 - 0.5 = 2 * Delta = 30000 * Tmp = 30000 / (2 * 10000) = 1.5 * NewPt = 2 - 1.5 = 1 * Delta = 40000 * Tmp = 40000 / 20000 = 2 * NewPt = 2 - 2 = 0 */ IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_HWVIRT); uint32_t const uVmcsPreemptVal = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32PreemptTimer; if (uVmcsPreemptVal > 0) { uint64_t const uCurTick = TMCpuTickGetNoCheck(pVCpu); uint64_t const uEntryTick = pVCpu->cpum.GstCtx.hwvirt.vmx.uEntryTick; uint64_t const uDelta = uCurTick - uEntryTick; uint32_t const uPreemptTimer = uVmcsPreemptVal - ASMDivU64ByU32RetU32(uDelta, uVmcsPreemptVal * RT_BIT(VMX_V_PREEMPT_TIMER_SHIFT)); return uPreemptTimer; } return 0; } /** * Saves guest segment registers, GDTR, IDTR, LDTR, TR as part of VM-exit. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmexitSaveGuestSegRegs(PVMCPUCC pVCpu) RT_NOEXCEPT { /* * Save guest segment registers, GDTR, IDTR, LDTR, TR. * See Intel spec 27.3.2 "Saving Segment Registers and Descriptor-Table Registers". */ /* CS, SS, ES, DS, FS, GS. */ PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; for (unsigned iSegReg = 0; iSegReg < X86_SREG_COUNT; iSegReg++) { PCCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg]; if (!pSelReg->Attr.n.u1Unusable) iemVmxVmcsSetGuestSegReg(pVmcs, iSegReg, pSelReg); else { /* * For unusable segments the attributes are undefined except for CS and SS. * For the rest we don't bother preserving anything but the unusable bit. */ switch (iSegReg) { case X86_SREG_CS: pVmcs->GuestCs = pSelReg->Sel; pVmcs->u64GuestCsBase.u = pSelReg->u64Base; pVmcs->u32GuestCsLimit = pSelReg->u32Limit; pVmcs->u32GuestCsAttr = pSelReg->Attr.u & ( X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G | X86DESCATTR_UNUSABLE); break; case X86_SREG_SS: pVmcs->GuestSs = pSelReg->Sel; if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) pVmcs->u64GuestSsBase.u &= UINT32_C(0xffffffff); pVmcs->u32GuestSsAttr = pSelReg->Attr.u & (X86DESCATTR_DPL | X86DESCATTR_UNUSABLE); break; case X86_SREG_DS: pVmcs->GuestDs = pSelReg->Sel; if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) pVmcs->u64GuestDsBase.u &= UINT32_C(0xffffffff); pVmcs->u32GuestDsAttr = X86DESCATTR_UNUSABLE; break; case X86_SREG_ES: pVmcs->GuestEs = pSelReg->Sel; if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) pVmcs->u64GuestEsBase.u &= UINT32_C(0xffffffff); pVmcs->u32GuestEsAttr = X86DESCATTR_UNUSABLE; break; case X86_SREG_FS: pVmcs->GuestFs = pSelReg->Sel; pVmcs->u64GuestFsBase.u = pSelReg->u64Base; pVmcs->u32GuestFsAttr = X86DESCATTR_UNUSABLE; break; case X86_SREG_GS: pVmcs->GuestGs = pSelReg->Sel; pVmcs->u64GuestGsBase.u = pSelReg->u64Base; pVmcs->u32GuestGsAttr = X86DESCATTR_UNUSABLE; break; } } } /* Segment attribute bits 31:17 and 11:8 MBZ. */ uint32_t const fValidAttrMask = X86DESCATTR_TYPE | X86DESCATTR_DT | X86DESCATTR_DPL | X86DESCATTR_P | X86DESCATTR_AVL | X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G | X86DESCATTR_UNUSABLE; /* LDTR. */ { PCCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.ldtr; pVmcs->GuestLdtr = pSelReg->Sel; pVmcs->u64GuestLdtrBase.u = pSelReg->u64Base; Assert(X86_IS_CANONICAL(pSelReg->u64Base)); pVmcs->u32GuestLdtrLimit = pSelReg->u32Limit; pVmcs->u32GuestLdtrAttr = pSelReg->Attr.u & fValidAttrMask; } /* TR. */ { PCCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.tr; pVmcs->GuestTr = pSelReg->Sel; pVmcs->u64GuestTrBase.u = pSelReg->u64Base; pVmcs->u32GuestTrLimit = pSelReg->u32Limit; pVmcs->u32GuestTrAttr = pSelReg->Attr.u & fValidAttrMask; } /* GDTR. */ pVmcs->u64GuestGdtrBase.u = pVCpu->cpum.GstCtx.gdtr.pGdt; pVmcs->u32GuestGdtrLimit = pVCpu->cpum.GstCtx.gdtr.cbGdt; /* IDTR. */ pVmcs->u64GuestIdtrBase.u = pVCpu->cpum.GstCtx.idtr.pIdt; pVmcs->u32GuestIdtrLimit = pVCpu->cpum.GstCtx.idtr.cbIdt; } /** * Saves guest non-register state as part of VM-exit. * * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason. */ static void iemVmxVmexitSaveGuestNonRegState(PVMCPUCC pVCpu, uint32_t uExitReason) RT_NOEXCEPT { /* * Save guest non-register state. * See Intel spec. 27.3.4 "Saving Non-Register State". */ PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* * Activity state. * Most VM-exits will occur in the active state. However, if the first instruction * following the VM-entry is a HLT instruction, and the MTF VM-execution control is set, * the VM-exit will be from the HLT activity state. * * See Intel spec. 25.5.2 "Monitor Trap Flag". */ /** @todo NSTVMX: Does triple-fault VM-exit reflect a shutdown activity state or * not? */ EMSTATE const enmActivityState = EMGetState(pVCpu); switch (enmActivityState) { case EMSTATE_HALTED: pVmcs->u32GuestActivityState = VMX_VMCS_GUEST_ACTIVITY_HLT; break; default: pVmcs->u32GuestActivityState = VMX_VMCS_GUEST_ACTIVITY_ACTIVE; break; } /* * Interruptibility-state. */ /* NMI. */ pVmcs->u32GuestIntrState = 0; if (pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI) { if (pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking) pVmcs->u32GuestIntrState |= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI; } else { if (CPUMAreInterruptsInhibitedByNmi(&pVCpu->cpum.GstCtx)) pVmcs->u32GuestIntrState |= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI; } /* Blocking-by-STI or blocking-by-MovSS. */ uint32_t fInhibitShw; if (!CPUMIsInInterruptShadowWithUpdateEx(&pVCpu->cpum.GstCtx, &fInhibitShw)) { /* probable */} else { if (pVCpu->cpum.GstCtx.rip == pVCpu->cpum.GstCtx.uRipInhibitInt) { /* * We must ensure only one of these bits is set. * Our emulation can have both set (perhaps because AMD doesn't distinguish * between the two?). Hence, the 'else' with blocking-by-MovSS taking priority * since it blocks more. Nested Ubuntu 22.04.2 running inside a Hyper-V enabled * Windows Server 2008 R2 guest runs into this issue. * * See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State". */ if (fInhibitShw & CPUMCTX_INHIBIT_SHADOW_SS) pVmcs->u32GuestIntrState |= VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS; else { Assert(fInhibitShw & CPUMCTX_INHIBIT_SHADOW_STI); pVmcs->u32GuestIntrState |= VMX_VMCS_GUEST_INT_STATE_BLOCK_STI; } } } /* Nothing to do for SMI/enclave. We don't support enclaves or SMM yet. */ /* * Pending debug exceptions. * * For VM-exits where it is not applicable, we can safely zero out the field. * For VM-exits where it is applicable, it's expected to be updated by the caller already. */ if ( uExitReason != VMX_EXIT_INIT_SIGNAL && uExitReason != VMX_EXIT_SMI && uExitReason != VMX_EXIT_ERR_MACHINE_CHECK && !VMXIsVmexitTrapLike(uExitReason)) { /** @todo NSTVMX: also must exclude VM-exits caused by debug exceptions when * block-by-MovSS is in effect. */ pVmcs->u64GuestPendingDbgXcpts.u = 0; } /* * Save the VMX-preemption timer value back into the VMCS if the feature is enabled. * * For VMX-preemption timer VM-exits, we should have already written back 0 if the * feature is supported back into the VMCS, and thus there is nothing further to do here. */ if ( uExitReason != VMX_EXIT_PREEMPT_TIMER && (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER)) pVmcs->u32PreemptTimer = iemVmxCalcPreemptTimer(pVCpu); /* * Save the guest PAE PDPTEs. */ if ( !CPUMIsGuestInPAEModeEx(&pVCpu->cpum.GstCtx) || !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT)) { /* * Without EPT or when the nested-guest is not using PAE paging, the values saved * in the VMCS during VM-exit are undefined. We zero them here for consistency. */ pVmcs->u64GuestPdpte0.u = 0; pVmcs->u64GuestPdpte1.u = 0; pVmcs->u64GuestPdpte2.u = 0; pVmcs->u64GuestPdpte3.u = 0; } else { /* * With EPT and when the nested-guest is using PAE paging, we update the PDPTEs from * the nested-guest CPU context. Both IEM (Mov CRx) and hardware-assisted execution * of the nested-guest is expected to have updated them. */ pVmcs->u64GuestPdpte0.u = pVCpu->cpum.GstCtx.aPaePdpes[0].u; pVmcs->u64GuestPdpte1.u = pVCpu->cpum.GstCtx.aPaePdpes[1].u; pVmcs->u64GuestPdpte2.u = pVCpu->cpum.GstCtx.aPaePdpes[2].u; pVmcs->u64GuestPdpte3.u = pVCpu->cpum.GstCtx.aPaePdpes[3].u; } /* Clear PGM's copy of the EPT pointer for added safety. */ if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT) PGMSetGuestEptPtr(pVCpu, 0 /* uEptPtr */); } /** * Saves the guest-state as part of VM-exit. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason. */ static void iemVmxVmexitSaveGuestState(PVMCPUCC pVCpu, uint32_t uExitReason) RT_NOEXCEPT { iemVmxVmexitSaveGuestControlRegsMsrs(pVCpu); iemVmxVmexitSaveGuestSegRegs(pVCpu); pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64GuestRip.u = pVCpu->cpum.GstCtx.rip; pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64GuestRsp.u = pVCpu->cpum.GstCtx.rsp; pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64GuestRFlags.u = pVCpu->cpum.GstCtx.rflags.u; /** @todo NSTVMX: Check RFLAGS.RF handling. */ iemVmxVmexitSaveGuestNonRegState(pVCpu, uExitReason); } /** * Saves the guest MSRs into the VM-exit MSR-store area as part of VM-exit. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason (for diagnostic purposes). */ static int iemVmxVmexitSaveGuestAutoMsrs(PVMCPUCC pVCpu, uint32_t uExitReason) RT_NOEXCEPT { /* * Save guest MSRs. * See Intel spec. 27.4 "Saving MSRs". */ PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char * const pszFailure = "VMX-abort"; /* * The VM-exit MSR-store area address need not be a valid guest-physical address if the * VM-exit MSR-store count is 0. If this is the case, bail early without reading it. * See Intel spec. 24.7.2 "VM-Exit Controls for MSRs". */ uint32_t const cMsrs = RT_MIN(pVmcs->u32ExitMsrStoreCount, RT_ELEMENTS(pVCpu->cpum.GstCtx.hwvirt.vmx.aExitMsrStoreArea)); if (!cMsrs) return VINF_SUCCESS; /* * Verify the MSR auto-store count. Physical CPUs can behave unpredictably if the count * is exceeded including possibly raising #MC exceptions during VMX transition. Our * implementation causes a VMX-abort followed by a triple-fault. */ bool const fIsMsrCountValid = iemVmxIsAutoMsrCountValid(pVCpu, cMsrs); if (fIsMsrCountValid) { /* likely */ } else IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrStoreCount); /* * Optimization if the nested hypervisor is using the same guest-physical page for both * the VM-entry MSR-load area as well as the VM-exit MSR store area. */ PVMXAUTOMSR pMsrArea; RTGCPHYS const GCPhysVmEntryMsrLoadArea = pVmcs->u64AddrEntryMsrLoad.u; RTGCPHYS const GCPhysVmExitMsrStoreArea = pVmcs->u64AddrExitMsrStore.u; if (GCPhysVmEntryMsrLoadArea == GCPhysVmExitMsrStoreArea) pMsrArea = pVCpu->cpum.GstCtx.hwvirt.vmx.aEntryMsrLoadArea; else { int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.aExitMsrStoreArea[0], GCPhysVmExitMsrStoreArea, cMsrs * sizeof(VMXAUTOMSR)); if (RT_SUCCESS(rc)) pMsrArea = pVCpu->cpum.GstCtx.hwvirt.vmx.aExitMsrStoreArea; else { AssertMsgFailed(("VM-exit: Failed to read MSR auto-store area at %#RGp, rc=%Rrc\n", GCPhysVmExitMsrStoreArea, rc)); IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrStorePtrReadPhys); } } /* * Update VM-exit MSR store area. */ PVMXAUTOMSR pMsr = pMsrArea; for (uint32_t idxMsr = 0; idxMsr < cMsrs; idxMsr++, pMsr++) { if ( !pMsr->u32Reserved && pMsr->u32Msr != MSR_IA32_SMBASE && pMsr->u32Msr >> 8 != MSR_IA32_X2APIC_START >> 8) { VBOXSTRICTRC rcStrict = CPUMQueryGuestMsr(pVCpu, pMsr->u32Msr, &pMsr->u64Value); if (rcStrict == VINF_SUCCESS) continue; /* * If we're in ring-0, we cannot handle returns to ring-3 at this point and continue VM-exit. * If any nested hypervisor loads MSRs that require ring-3 handling, we cause a VMX-abort * recording the MSR index in the auxiliary info. field and indicated further by our * own, specific diagnostic code. Later, we can try implement handling of the MSR in ring-0 * if possible, or come up with a better, generic solution. */ pVCpu->cpum.GstCtx.hwvirt.vmx.uAbortAux = pMsr->u32Msr; VMXVDIAG const enmDiag = rcStrict == VINF_CPUM_R3_MSR_READ ? kVmxVDiag_Vmexit_MsrStoreRing3 : kVmxVDiag_Vmexit_MsrStore; IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, enmDiag); } else { pVCpu->cpum.GstCtx.hwvirt.vmx.uAbortAux = pMsr->u32Msr; IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrStoreRsvd); } } /* * Commit the VM-exit MSR store are to guest memory. */ int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVmExitMsrStoreArea, pMsrArea, cMsrs * sizeof(VMXAUTOMSR)); if (RT_SUCCESS(rc)) return VINF_SUCCESS; NOREF(uExitReason); NOREF(pszFailure); AssertMsgFailed(("VM-exit: Failed to write MSR auto-store area at %#RGp, rc=%Rrc\n", GCPhysVmExitMsrStoreArea, rc)); IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrStorePtrWritePhys); } /** * Performs a VMX abort (due to an fatal error during VM-exit). * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param enmAbort The VMX abort reason. */ static VBOXSTRICTRC iemVmxAbort(PVMCPUCC pVCpu, VMXABORT enmAbort) RT_NOEXCEPT { /* * Perform the VMX abort. * See Intel spec. 27.7 "VMX Aborts". */ LogFunc(("enmAbort=%u (%s) -> RESET\n", enmAbort, VMXGetAbortDesc(enmAbort))); /* We don't support SMX yet. */ pVCpu->cpum.GstCtx.hwvirt.vmx.enmAbort = enmAbort; if (IEM_VMX_HAS_CURRENT_VMCS(pVCpu)) { RTGCPHYS const GCPhysVmcs = IEM_VMX_GET_CURRENT_VMCS(pVCpu); uint32_t const offVmxAbort = RT_UOFFSETOF(VMXVVMCS, enmVmxAbort); PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVmcs + offVmxAbort, &enmAbort, sizeof(enmAbort)); } return VINF_EM_TRIPLE_FAULT; } /** * Loads host control registers, debug registers and MSRs as part of VM-exit. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmexitLoadHostControlRegsMsrs(PVMCPUCC pVCpu) RT_NOEXCEPT { /* * Load host control registers, debug registers and MSRs. * See Intel spec. 27.5.1 "Loading Host Control Registers, Debug Registers, MSRs". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE); /* CR0. */ { /* Bits 63:32, 28:19, 17, 15:6, ET, CD, NW and CR0 fixed bits are not modified. */ uint64_t const fCr0IgnMask = VMX_EXIT_HOST_CR0_IGNORE_MASK; uint64_t const uHostCr0 = pVmcs->u64HostCr0.u; uint64_t const uGuestCr0 = pVCpu->cpum.GstCtx.cr0; uint64_t const uValidHostCr0 = (uHostCr0 & ~fCr0IgnMask) | (uGuestCr0 & fCr0IgnMask); /* Verify we have not modified CR0 fixed bits in VMX operation. */ #ifdef VBOX_STRICT uint64_t const uCr0Mb1 = iemVmxGetCr0Fixed0(pVCpu, true /* fVmxNonRootMode */); bool const fUx = RT_BOOL(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST); AssertMsg( (uValidHostCr0 & uCr0Mb1) == uCr0Mb1 && (uValidHostCr0 & ~VMX_V_CR0_FIXED1) == 0, ("host=%#RX64 guest=%#RX64 mb1=%#RX64 valid_host_cr0=%#RX64 fUx=%RTbool\n", uHostCr0, uGuestCr0, uCr0Mb1, uValidHostCr0, fUx)); #endif Assert(!(uValidHostCr0 >> 32)); CPUMSetGuestCR0(pVCpu, uValidHostCr0); } /* CR4. */ { /* CR4 fixed bits are not modified. */ uint64_t const uCr4Mb1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed0; uint64_t const uCr4Mb0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed1; uint64_t const uHostCr4 = pVmcs->u64HostCr4.u; uint64_t uValidHostCr4 = (uHostCr4 & uCr4Mb0) | uCr4Mb1; if (fHostInLongMode) uValidHostCr4 |= X86_CR4_PAE; else uValidHostCr4 &= ~(uint64_t)X86_CR4_PCIDE; /* Verify we have not modified CR4 fixed bits in VMX non-root operation. */ AssertMsg( (uValidHostCr4 & uCr4Mb1) == uCr4Mb1 && (uValidHostCr4 & ~uCr4Mb0) == 0, ("host=%#RX64 guest=%#RX64, uCr4Mb1=%#RX64 uCr4Mb0=%#RX64 valid_host_cr4=%#RX64\n", uHostCr4, pVCpu->cpum.GstCtx.cr4, uCr4Mb1, uCr4Mb0, uValidHostCr4)); CPUMSetGuestCR4(pVCpu, uValidHostCr4); } /* CR3 (host value validated while checking host-state during VM-entry). */ pVCpu->cpum.GstCtx.cr3 = pVmcs->u64HostCr3.u; /* DR7. */ pVCpu->cpum.GstCtx.dr[7] = X86_DR7_INIT_VAL; /** @todo NSTVMX: Support IA32_DEBUGCTL MSR */ /* Save SYSENTER CS, ESP, EIP (host value validated while checking host-state during VM-entry). */ pVCpu->cpum.GstCtx.SysEnter.eip = pVmcs->u64HostSysenterEip.u; pVCpu->cpum.GstCtx.SysEnter.esp = pVmcs->u64HostSysenterEsp.u; pVCpu->cpum.GstCtx.SysEnter.cs = pVmcs->u32HostSysenterCs; /* FS, GS bases are loaded later while we load host segment registers. */ /* EFER MSR (host value validated while checking host-state during VM-entry). */ if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR) pVCpu->cpum.GstCtx.msrEFER = pVmcs->u64HostEferMsr.u; else if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { if (fHostInLongMode) pVCpu->cpum.GstCtx.msrEFER |= (MSR_K6_EFER_LMA | MSR_K6_EFER_LME); else pVCpu->cpum.GstCtx.msrEFER &= ~(MSR_K6_EFER_LMA | MSR_K6_EFER_LME); } /* We don't support IA32_PERF_GLOBAL_CTRL MSR yet. */ /* PAT MSR (host value is validated while checking host-state during VM-entry). */ if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_PAT_MSR) pVCpu->cpum.GstCtx.msrPAT = pVmcs->u64HostPatMsr.u; /* We don't support IA32_BNDCFGS MSR yet. */ } /** * Loads host segment registers, GDTR, IDTR, LDTR and TR as part of VM-exit. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmexitLoadHostSegRegs(PVMCPUCC pVCpu) RT_NOEXCEPT { /* * Load host segment registers, GDTR, IDTR, LDTR and TR. * See Intel spec. 27.5.2 "Loading Host Segment and Descriptor-Table Registers". * * Warning! Be careful to not touch fields that are reserved by VT-x, * e.g. segment limit high bits stored in segment attributes (in bits 11:8). */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE); /* CS, SS, ES, DS, FS, GS. */ for (unsigned iSegReg = 0; iSegReg < X86_SREG_COUNT; iSegReg++) { RTSEL const HostSel = iemVmxVmcsGetHostSelReg(pVmcs, iSegReg); bool const fUnusable = RT_BOOL(HostSel == 0); PCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg]; /* Selector. */ pSelReg->Sel = HostSel; pSelReg->ValidSel = HostSel; pSelReg->fFlags = CPUMSELREG_FLAGS_VALID; /* Limit. */ pSelReg->u32Limit = 0xffffffff; /* Base. */ pSelReg->u64Base = 0; /* Attributes. */ if (iSegReg == X86_SREG_CS) { pSelReg->Attr.n.u4Type = X86_SEL_TYPE_CODE | X86_SEL_TYPE_READ | X86_SEL_TYPE_ACCESSED; pSelReg->Attr.n.u1DescType = 1; pSelReg->Attr.n.u2Dpl = 0; pSelReg->Attr.n.u1Present = 1; pSelReg->Attr.n.u1Long = fHostInLongMode; pSelReg->Attr.n.u1DefBig = !fHostInLongMode; pSelReg->Attr.n.u1Granularity = 1; Assert(!pSelReg->Attr.n.u1Unusable); Assert(!fUnusable); } else { pSelReg->Attr.n.u4Type = X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED; pSelReg->Attr.n.u1DescType = 1; pSelReg->Attr.n.u2Dpl = 0; pSelReg->Attr.n.u1Present = 1; pSelReg->Attr.n.u1DefBig = 1; pSelReg->Attr.n.u1Granularity = 1; pSelReg->Attr.n.u1Unusable = fUnusable; } } /* FS base. */ if ( !pVCpu->cpum.GstCtx.fs.Attr.n.u1Unusable || fHostInLongMode) { Assert(X86_IS_CANONICAL(pVmcs->u64HostFsBase.u)); pVCpu->cpum.GstCtx.fs.u64Base = pVmcs->u64HostFsBase.u; } /* GS base. */ if ( !pVCpu->cpum.GstCtx.gs.Attr.n.u1Unusable || fHostInLongMode) { Assert(X86_IS_CANONICAL(pVmcs->u64HostGsBase.u)); pVCpu->cpum.GstCtx.gs.u64Base = pVmcs->u64HostGsBase.u; } /* TR. */ Assert(X86_IS_CANONICAL(pVmcs->u64HostTrBase.u)); Assert(!pVCpu->cpum.GstCtx.tr.Attr.n.u1Unusable); pVCpu->cpum.GstCtx.tr.Sel = pVmcs->HostTr; pVCpu->cpum.GstCtx.tr.ValidSel = pVmcs->HostTr; pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID; pVCpu->cpum.GstCtx.tr.u32Limit = X86_SEL_TYPE_SYS_386_TSS_LIMIT_MIN; pVCpu->cpum.GstCtx.tr.u64Base = pVmcs->u64HostTrBase.u; pVCpu->cpum.GstCtx.tr.Attr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY; pVCpu->cpum.GstCtx.tr.Attr.n.u1DescType = 0; pVCpu->cpum.GstCtx.tr.Attr.n.u2Dpl = 0; pVCpu->cpum.GstCtx.tr.Attr.n.u1Present = 1; pVCpu->cpum.GstCtx.tr.Attr.n.u1DefBig = 0; pVCpu->cpum.GstCtx.tr.Attr.n.u1Granularity = 0; /* LDTR (Warning! do not touch the base and limits here). */ pVCpu->cpum.GstCtx.ldtr.Sel = 0; pVCpu->cpum.GstCtx.ldtr.ValidSel = 0; pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID; pVCpu->cpum.GstCtx.ldtr.Attr.u = X86DESCATTR_UNUSABLE; /* GDTR. */ Assert(X86_IS_CANONICAL(pVmcs->u64HostGdtrBase.u)); pVCpu->cpum.GstCtx.gdtr.pGdt = pVmcs->u64HostGdtrBase.u; pVCpu->cpum.GstCtx.gdtr.cbGdt = 0xffff; /* IDTR.*/ Assert(X86_IS_CANONICAL(pVmcs->u64HostIdtrBase.u)); pVCpu->cpum.GstCtx.idtr.pIdt = pVmcs->u64HostIdtrBase.u; pVCpu->cpum.GstCtx.idtr.cbIdt = 0xffff; } /** * Loads the host MSRs from the VM-exit MSR-load area as part of VM-exit. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VMX instruction name (for logging purposes). */ static int iemVmxVmexitLoadHostAutoMsrs(PVMCPUCC pVCpu, uint32_t uExitReason) RT_NOEXCEPT { /* * Load host MSRs. * See Intel spec. 27.6 "Loading MSRs". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char * const pszFailure = "VMX-abort"; /* * The VM-exit MSR-load area address need not be a valid guest-physical address if the * VM-exit MSR load count is 0. If this is the case, bail early without reading it. * See Intel spec. 24.7.2 "VM-Exit Controls for MSRs". */ uint32_t const cMsrs = RT_MIN(pVmcs->u32ExitMsrLoadCount, RT_ELEMENTS(pVCpu->cpum.GstCtx.hwvirt.vmx.aExitMsrLoadArea)); if (!cMsrs) return VINF_SUCCESS; /* * Verify the MSR auto-load count. Physical CPUs can behave unpredictably if the count * is exceeded including possibly raising #MC exceptions during VMX transition. Our * implementation causes a VMX-abort followed by a triple-fault. */ bool const fIsMsrCountValid = iemVmxIsAutoMsrCountValid(pVCpu, cMsrs); if (fIsMsrCountValid) { /* likely */ } else IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrLoadCount); RTGCPHYS const GCPhysVmExitMsrLoadArea = pVmcs->u64AddrExitMsrLoad.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.aExitMsrLoadArea[0], GCPhysVmExitMsrLoadArea, cMsrs * sizeof(VMXAUTOMSR)); if (RT_SUCCESS(rc)) { PCVMXAUTOMSR pMsr = pVCpu->cpum.GstCtx.hwvirt.vmx.aExitMsrLoadArea; for (uint32_t idxMsr = 0; idxMsr < cMsrs; idxMsr++, pMsr++) { if ( !pMsr->u32Reserved && pMsr->u32Msr != MSR_K8_FS_BASE && pMsr->u32Msr != MSR_K8_GS_BASE && pMsr->u32Msr != MSR_K6_EFER && pMsr->u32Msr != MSR_IA32_SMM_MONITOR_CTL && pMsr->u32Msr >> 8 != MSR_IA32_X2APIC_START >> 8) { VBOXSTRICTRC rcStrict = CPUMSetGuestMsr(pVCpu, pMsr->u32Msr, pMsr->u64Value); if (rcStrict == VINF_SUCCESS) continue; /* * If we're in ring-0, we cannot handle returns to ring-3 at this point and continue VM-exit. * If any nested hypervisor loads MSRs that require ring-3 handling, we cause a VMX-abort * recording the MSR index in the auxiliary info. field and indicated further by our * own, specific diagnostic code. Later, we can try implement handling of the MSR in ring-0 * if possible, or come up with a better, generic solution. */ pVCpu->cpum.GstCtx.hwvirt.vmx.uAbortAux = pMsr->u32Msr; VMXVDIAG const enmDiag = rcStrict == VINF_CPUM_R3_MSR_WRITE ? kVmxVDiag_Vmexit_MsrLoadRing3 : kVmxVDiag_Vmexit_MsrLoad; IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, enmDiag); } else IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrLoadRsvd); } } else { AssertMsgFailed(("VM-exit: Failed to read MSR auto-load area at %#RGp, rc=%Rrc\n", GCPhysVmExitMsrLoadArea, rc)); IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrLoadPtrReadPhys); } NOREF(uExitReason); NOREF(pszFailure); return VINF_SUCCESS; } /** * Loads the host state as part of VM-exit. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason (for logging purposes). */ static VBOXSTRICTRC iemVmxVmexitLoadHostState(PVMCPUCC pVCpu, uint32_t uExitReason) RT_NOEXCEPT { /* * Load host state. * See Intel spec. 27.5 "Loading Host State". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE); /* We cannot return from a long-mode guest to a host that is not in long mode. */ if ( CPUMIsGuestInLongMode(pVCpu) && !fHostInLongMode) { Log(("VM-exit from long-mode guest to host not in long-mode -> VMX-Abort\n")); return iemVmxAbort(pVCpu, VMXABORT_HOST_NOT_IN_LONG_MODE); } /* * Check host PAE PDPTEs prior to loading the host state. * See Intel spec. 26.5.4 "Checking and Loading Host Page-Directory-Pointer-Table Entries". */ if ( (pVmcs->u64HostCr4.u & X86_CR4_PAE) && !fHostInLongMode && ( !CPUMIsGuestInPAEModeEx(&pVCpu->cpum.GstCtx) || pVmcs->u64HostCr3.u != pVCpu->cpum.GstCtx.cr3)) { int const rc = PGMGstMapPaePdpesAtCr3(pVCpu, pVmcs->u64HostCr3.u); if (RT_SUCCESS(rc)) { /* likely*/ } else { IEM_VMX_VMEXIT_FAILED(pVCpu, uExitReason, "VMX-abort", kVmxVDiag_Vmexit_HostPdpte); return iemVmxAbort(pVCpu, VMXBOART_HOST_PDPTE); } } iemVmxVmexitLoadHostControlRegsMsrs(pVCpu); iemVmxVmexitLoadHostSegRegs(pVCpu); /* * Load host RIP, RSP and RFLAGS. * See Intel spec. 27.5.3 "Loading Host RIP, RSP and RFLAGS" */ pVCpu->cpum.GstCtx.rip = pVmcs->u64HostRip.u; pVCpu->cpum.GstCtx.rsp = pVmcs->u64HostRsp.u; pVCpu->cpum.GstCtx.rflags.u = X86_EFL_1; /* Clear address range monitoring. */ EMMonitorWaitClear(pVCpu); /* Perform the VMX transition (PGM updates). */ VBOXSTRICTRC rcStrict = iemVmxTransition(pVCpu, 0 /*cbInstr - whatever*/); if (rcStrict == VINF_SUCCESS) { /* likely */ } else if (RT_SUCCESS(rcStrict)) { Log3(("VM-exit: iemVmxTransition returns %Rrc (uExitReason=%u) -> Setting passup status\n", VBOXSTRICTRC_VAL(rcStrict), uExitReason)); rcStrict = iemSetPassUpStatus(pVCpu, rcStrict); } else { Log3(("VM-exit: iemVmxTransition failed! rc=%Rrc (uExitReason=%u)\n", VBOXSTRICTRC_VAL(rcStrict), uExitReason)); return VBOXSTRICTRC_VAL(rcStrict); } Assert(rcStrict == VINF_SUCCESS); /* Load MSRs from the VM-exit auto-load MSR area. */ int rc = iemVmxVmexitLoadHostAutoMsrs(pVCpu, uExitReason); if (RT_FAILURE(rc)) { Log(("VM-exit failed while loading host MSRs -> VMX-Abort\n")); return iemVmxAbort(pVCpu, VMXABORT_LOAD_HOST_MSR); } return VINF_SUCCESS; } /** * Gets VM-exit instruction information along with any displacement for an * instruction VM-exit. * * @returns The VM-exit instruction information. * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason. * @param uInstrId The VM-exit instruction identity (VMXINSTRID_XXX). * @param pGCPtrDisp Where to store the displacement field. Optional, can be * NULL. */ static uint32_t iemVmxGetExitInstrInfo(PVMCPUCC pVCpu, uint32_t uExitReason, VMXINSTRID uInstrId, PRTGCPTR pGCPtrDisp) RT_NOEXCEPT { RTGCPTR GCPtrDisp; VMXEXITINSTRINFO ExitInstrInfo; ExitInstrInfo.u = 0; /* * Get and parse the ModR/M byte from our decoded opcodes. */ uint8_t bRm; uint8_t const offModRm = pVCpu->iem.s.offModRm; IEM_MODRM_GET_U8(pVCpu, bRm, offModRm); if ((bRm & X86_MODRM_MOD_MASK) == (3 << X86_MODRM_MOD_SHIFT)) { /* * ModR/M indicates register addressing. * * The primary/secondary register operands are reported in the iReg1 or iReg2 * fields depending on whether it is a read/write form. */ uint8_t idxReg1; uint8_t idxReg2; if (!VMXINSTRID_IS_MODRM_PRIMARY_OP_W(uInstrId)) { idxReg1 = ((bRm >> X86_MODRM_REG_SHIFT) & X86_MODRM_REG_SMASK) | pVCpu->iem.s.uRexReg; idxReg2 = (bRm & X86_MODRM_RM_MASK) | pVCpu->iem.s.uRexB; } else { idxReg1 = (bRm & X86_MODRM_RM_MASK) | pVCpu->iem.s.uRexB; idxReg2 = ((bRm >> X86_MODRM_REG_SHIFT) & X86_MODRM_REG_SMASK) | pVCpu->iem.s.uRexReg; } ExitInstrInfo.All.u2Scaling = 0; ExitInstrInfo.All.iReg1 = idxReg1; ExitInstrInfo.All.u3AddrSize = pVCpu->iem.s.enmEffAddrMode; ExitInstrInfo.All.fIsRegOperand = 1; ExitInstrInfo.All.uOperandSize = pVCpu->iem.s.enmEffOpSize; ExitInstrInfo.All.iSegReg = 0; ExitInstrInfo.All.iIdxReg = 0; ExitInstrInfo.All.fIdxRegInvalid = 1; ExitInstrInfo.All.iBaseReg = 0; ExitInstrInfo.All.fBaseRegInvalid = 1; ExitInstrInfo.All.iReg2 = idxReg2; /* Displacement not applicable for register addressing. */ GCPtrDisp = 0; } else { /* * ModR/M indicates memory addressing. */ uint8_t uScale = 0; bool fBaseRegValid = false; bool fIdxRegValid = false; uint8_t iBaseReg = 0; uint8_t iIdxReg = 0; if (pVCpu->iem.s.enmEffAddrMode == IEMMODE_16BIT) { /* * Parse the ModR/M, displacement for 16-bit addressing mode. * See Intel instruction spec. Table 2-1. "16-Bit Addressing Forms with the ModR/M Byte". */ uint16_t u16Disp = 0; uint8_t const offDisp = offModRm + sizeof(bRm); if ((bRm & (X86_MODRM_MOD_MASK | X86_MODRM_RM_MASK)) == 6) { /* Displacement without any registers. */ IEM_DISP_GET_U16(pVCpu, u16Disp, offDisp); } else { /* Register (index and base). */ switch (bRm & X86_MODRM_RM_MASK) { case 0: fBaseRegValid = true; iBaseReg = X86_GREG_xBX; fIdxRegValid = true; iIdxReg = X86_GREG_xSI; break; case 1: fBaseRegValid = true; iBaseReg = X86_GREG_xBX; fIdxRegValid = true; iIdxReg = X86_GREG_xDI; break; case 2: fBaseRegValid = true; iBaseReg = X86_GREG_xBP; fIdxRegValid = true; iIdxReg = X86_GREG_xSI; break; case 3: fBaseRegValid = true; iBaseReg = X86_GREG_xBP; fIdxRegValid = true; iIdxReg = X86_GREG_xDI; break; case 4: fIdxRegValid = true; iIdxReg = X86_GREG_xSI; break; case 5: fIdxRegValid = true; iIdxReg = X86_GREG_xDI; break; case 6: fBaseRegValid = true; iBaseReg = X86_GREG_xBP; break; case 7: fBaseRegValid = true; iBaseReg = X86_GREG_xBX; break; } /* Register + displacement. */ switch ((bRm >> X86_MODRM_MOD_SHIFT) & X86_MODRM_MOD_SMASK) { case 0: break; case 1: IEM_DISP_GET_S8_SX_U16(pVCpu, u16Disp, offDisp); break; case 2: IEM_DISP_GET_U16(pVCpu, u16Disp, offDisp); break; default: { /* Register addressing, handled at the beginning. */ AssertMsgFailed(("ModR/M %#x implies register addressing, memory addressing expected!", bRm)); break; } } } Assert(!uScale); /* There's no scaling/SIB byte for 16-bit addressing. */ GCPtrDisp = (int16_t)u16Disp; /* Sign-extend the displacement. */ } else if (pVCpu->iem.s.enmEffAddrMode == IEMMODE_32BIT) { /* * Parse the ModR/M, SIB, displacement for 32-bit addressing mode. * See Intel instruction spec. Table 2-2. "32-Bit Addressing Forms with the ModR/M Byte". */ uint32_t u32Disp = 0; if ((bRm & (X86_MODRM_MOD_MASK | X86_MODRM_RM_MASK)) == 5) { /* Displacement without any registers. */ uint8_t const offDisp = offModRm + sizeof(bRm); IEM_DISP_GET_U32(pVCpu, u32Disp, offDisp); } else { /* Register (and perhaps scale, index and base). */ uint8_t offDisp = offModRm + sizeof(bRm); iBaseReg = (bRm & X86_MODRM_RM_MASK); if (iBaseReg == 4) { /* An SIB byte follows the ModR/M byte, parse it. */ uint8_t bSib; uint8_t const offSib = offModRm + sizeof(bRm); IEM_SIB_GET_U8(pVCpu, bSib, offSib); /* A displacement may follow SIB, update its offset. */ offDisp += sizeof(bSib); /* Get the scale. */ uScale = (bSib >> X86_SIB_SCALE_SHIFT) & X86_SIB_SCALE_SMASK; /* Get the index register. */ iIdxReg = (bSib >> X86_SIB_INDEX_SHIFT) & X86_SIB_INDEX_SMASK; fIdxRegValid = RT_BOOL(iIdxReg != 4); /* Get the base register. */ iBaseReg = bSib & X86_SIB_BASE_MASK; fBaseRegValid = true; if (iBaseReg == 5) { if ((bRm & X86_MODRM_MOD_MASK) == 0) { /* Mod is 0 implies a 32-bit displacement with no base. */ fBaseRegValid = false; IEM_DISP_GET_U32(pVCpu, u32Disp, offDisp); } else { /* Mod is not 0 implies an 8-bit/32-bit displacement (handled below) with an EBP base. */ iBaseReg = X86_GREG_xBP; } } } /* Register + displacement. */ switch ((bRm >> X86_MODRM_MOD_SHIFT) & X86_MODRM_MOD_SMASK) { case 0: /* Handled above */ break; case 1: IEM_DISP_GET_S8_SX_U32(pVCpu, u32Disp, offDisp); break; case 2: IEM_DISP_GET_U32(pVCpu, u32Disp, offDisp); break; default: { /* Register addressing, handled at the beginning. */ AssertMsgFailed(("ModR/M %#x implies register addressing, memory addressing expected!", bRm)); break; } } } GCPtrDisp = (int32_t)u32Disp; /* Sign-extend the displacement. */ } else { Assert(pVCpu->iem.s.enmEffAddrMode == IEMMODE_64BIT); /* * Parse the ModR/M, SIB, displacement for 64-bit addressing mode. * See Intel instruction spec. 2.2 "IA-32e Mode". */ uint64_t u64Disp = 0; bool const fRipRelativeAddr = RT_BOOL((bRm & (X86_MODRM_MOD_MASK | X86_MODRM_RM_MASK)) == 5); if (fRipRelativeAddr) { /* * RIP-relative addressing mode. * * The displacement is 32-bit signed implying an offset range of +/-2G. * See Intel instruction spec. 2.2.1.6 "RIP-Relative Addressing". */ uint8_t const offDisp = offModRm + sizeof(bRm); IEM_DISP_GET_S32_SX_U64(pVCpu, u64Disp, offDisp); } else { uint8_t offDisp = offModRm + sizeof(bRm); /* * Register (and perhaps scale, index and base). * * REX.B extends the most-significant bit of the base register. However, REX.B * is ignored while determining whether an SIB follows the opcode. Hence, we * shall OR any REX.B bit -after- inspecting for an SIB byte below. * * See Intel instruction spec. Table 2-5. "Special Cases of REX Encodings". */ iBaseReg = (bRm & X86_MODRM_RM_MASK); if (iBaseReg == 4) { /* An SIB byte follows the ModR/M byte, parse it. Displacement (if any) follows SIB. */ uint8_t bSib; uint8_t const offSib = offModRm + sizeof(bRm); IEM_SIB_GET_U8(pVCpu, bSib, offSib); /* Displacement may follow SIB, update its offset. */ offDisp += sizeof(bSib); /* Get the scale. */ uScale = (bSib >> X86_SIB_SCALE_SHIFT) & X86_SIB_SCALE_SMASK; /* Get the index. */ iIdxReg = ((bSib >> X86_SIB_INDEX_SHIFT) & X86_SIB_INDEX_SMASK) | pVCpu->iem.s.uRexIndex; fIdxRegValid = RT_BOOL(iIdxReg != 4); /* R12 -can- be used as an index register. */ /* Get the base. */ iBaseReg = (bSib & X86_SIB_BASE_MASK); fBaseRegValid = true; if (iBaseReg == 5) { if ((bRm & X86_MODRM_MOD_MASK) == 0) { /* Mod is 0 implies a signed 32-bit displacement with no base. */ IEM_DISP_GET_S32_SX_U64(pVCpu, u64Disp, offDisp); } else { /* Mod is non-zero implies an 8-bit/32-bit displacement (handled below) with RBP or R13 as base. */ iBaseReg = pVCpu->iem.s.uRexB ? X86_GREG_x13 : X86_GREG_xBP; } } } iBaseReg |= pVCpu->iem.s.uRexB; /* Register + displacement. */ switch ((bRm >> X86_MODRM_MOD_SHIFT) & X86_MODRM_MOD_SMASK) { case 0: /* Handled above */ break; case 1: IEM_DISP_GET_S8_SX_U64(pVCpu, u64Disp, offDisp); break; case 2: IEM_DISP_GET_S32_SX_U64(pVCpu, u64Disp, offDisp); break; default: { /* Register addressing, handled at the beginning. */ AssertMsgFailed(("ModR/M %#x implies register addressing, memory addressing expected!", bRm)); break; } } } GCPtrDisp = fRipRelativeAddr ? pVCpu->cpum.GstCtx.rip + u64Disp : u64Disp; } /* * The primary or secondary register operand is reported in iReg2 depending * on whether the primary operand is in read/write form. */ uint8_t idxReg2; if (!VMXINSTRID_IS_MODRM_PRIMARY_OP_W(uInstrId)) { idxReg2 = bRm & X86_MODRM_RM_MASK; if (pVCpu->iem.s.enmEffAddrMode == IEMMODE_64BIT) idxReg2 |= pVCpu->iem.s.uRexB; } else { idxReg2 = (bRm >> X86_MODRM_REG_SHIFT) & X86_MODRM_REG_SMASK; if (pVCpu->iem.s.enmEffAddrMode == IEMMODE_64BIT) idxReg2 |= pVCpu->iem.s.uRexReg; } ExitInstrInfo.All.u2Scaling = uScale; ExitInstrInfo.All.iReg1 = 0; /* Not applicable for memory addressing. */ ExitInstrInfo.All.u3AddrSize = pVCpu->iem.s.enmEffAddrMode; ExitInstrInfo.All.fIsRegOperand = 0; ExitInstrInfo.All.uOperandSize = pVCpu->iem.s.enmEffOpSize; ExitInstrInfo.All.iSegReg = pVCpu->iem.s.iEffSeg; ExitInstrInfo.All.iIdxReg = iIdxReg; ExitInstrInfo.All.fIdxRegInvalid = !fIdxRegValid; ExitInstrInfo.All.iBaseReg = iBaseReg; ExitInstrInfo.All.iIdxReg = !fBaseRegValid; ExitInstrInfo.All.iReg2 = idxReg2; } /* * Handle exceptions to the norm for certain instructions. * (e.g. some instructions convey an instruction identity in place of iReg2). */ switch (uExitReason) { case VMX_EXIT_GDTR_IDTR_ACCESS: { Assert(VMXINSTRID_IS_VALID(uInstrId)); Assert(VMXINSTRID_GET_ID(uInstrId) == (uInstrId & 0x3)); ExitInstrInfo.GdtIdt.u2InstrId = VMXINSTRID_GET_ID(uInstrId); ExitInstrInfo.GdtIdt.u2Undef0 = 0; break; } case VMX_EXIT_LDTR_TR_ACCESS: { Assert(VMXINSTRID_IS_VALID(uInstrId)); Assert(VMXINSTRID_GET_ID(uInstrId) == (uInstrId & 0x3)); ExitInstrInfo.LdtTr.u2InstrId = VMXINSTRID_GET_ID(uInstrId); ExitInstrInfo.LdtTr.u2Undef0 = 0; break; } case VMX_EXIT_RDRAND: case VMX_EXIT_RDSEED: { Assert(VMXINSTRID_IS_VALID(uInstrId)); Assert(VMXINSTRID_GET_ID(uInstrId) == (uInstrId & 0x3)); Assert(GCPtrDisp == 0); Assert(ExitInstrInfo.RdrandRdseed.u2OperandSize != 3); ExitInstrInfo.RdrandRdseed.u3Undef0 = 0; ExitInstrInfo.RdrandRdseed.u4Undef0 = 0; ExitInstrInfo.RdrandRdseed.u19Undef0 = 0; break; } } /* Update displacement and return the constructed VM-exit instruction information field. */ if (pGCPtrDisp) *pGCPtrDisp = GCPtrDisp; return ExitInstrInfo.u; } /** * VMX VM-exit handler. * * @returns Strict VBox status code. * @retval VINF_VMX_VMEXIT when the VM-exit is successful. * @retval VINF_EM_TRIPLE_FAULT when VM-exit is unsuccessful and leads to a * triple-fault. * * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason. * @param u64ExitQual The Exit qualification. * * @remarks We need not necessarily have completed VM-entry before a VM-exit is * called. Failures during VM-entry can cause VM-exits as well, so we * -cannot- assert we're in VMX non-root mode here. */ VBOXSTRICTRC iemVmxVmexit(PVMCPUCC pVCpu, uint32_t uExitReason, uint64_t u64ExitQual) RT_NOEXCEPT { # if defined(VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM) && !defined(IN_RING3) RT_NOREF3(pVCpu, uExitReason, u64ExitQual); AssertMsgFailed(("VM-exit should only be invoked from ring-3 when nested-guest executes only in ring-3!\n")); return VERR_IEM_IPE_7; # else PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* Just count this as an exit and be done with that. */ pVCpu->iem.s.cPotentialExits++; /* * Import all the guest-CPU state. * * HM on returning to guest execution would have to reset up a whole lot of state * anyway, (e.g., VM-entry/VM-exit controls) and we do not ever import a part of * the state and flag reloading the entire state on re-entry. So import the entire * state here, see HMNotifyVmxNstGstVmexit() for more comments. */ IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_ALL); /* * Ensure VM-entry interruption information valid bit is cleared. * * We do it here on every VM-exit so that even premature VM-exits (e.g. those caused * by invalid-guest state or machine-check exceptions) also clear this bit. * * See Intel spec. 27.2 "Recording VM-exit Information And Updating VM-entry control fields". */ if (VMX_ENTRY_INT_INFO_IS_VALID(pVmcs->u32EntryIntInfo)) pVmcs->u32EntryIntInfo &= ~VMX_ENTRY_INT_INFO_VALID; /* * Update the VM-exit reason and Exit qualification. * Other VMCS read-only data fields are expected to be updated by the caller already. */ pVmcs->u32RoExitReason = uExitReason; pVmcs->u64RoExitQual.u = u64ExitQual; Log2(("vmexit: reason=%u qual=%#RX64 cs:rip=%04x:%08RX64 cr0=%#RX64 cr3=%#RX64 cr4=%#RX64 eflags=%#RX32\n", uExitReason, pVmcs->u64RoExitQual.u, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.cr0, pVCpu->cpum.GstCtx.cr3, pVCpu->cpum.GstCtx.cr4, pVCpu->cpum.GstCtx.eflags.u)); /* * Update the IDT-vectoring information fields if the VM-exit is triggered during delivery of an event. * See Intel spec. 27.2.4 "Information for VM Exits During Event Delivery". */ { uint8_t uVector; uint32_t fFlags; uint32_t uErrCode; bool const fInEventDelivery = IEMGetCurrentXcpt(pVCpu, &uVector, &fFlags, &uErrCode, NULL /* puCr2 */); if (fInEventDelivery) { /* * A VM-exit is not considered to occur during event delivery when the VM-exit is * caused by a triple-fault or the original event results in a double-fault that * causes the VM exit directly (exception bitmap). Therefore, we must not set the * original event information into the IDT-vectoring information fields. * * See Intel spec. 27.2.4 "Information for VM Exits During Event Delivery". */ if ( uExitReason != VMX_EXIT_TRIPLE_FAULT && ( uExitReason != VMX_EXIT_XCPT_OR_NMI || !VMX_EXIT_INT_INFO_IS_XCPT_DF(pVmcs->u32RoExitIntInfo))) { uint8_t const uIdtVectoringType = iemVmxGetEventType(uVector, fFlags); uint8_t const fErrCodeValid = RT_BOOL(fFlags & IEM_XCPT_FLAGS_ERR); uint32_t const uIdtVectoringInfo = RT_BF_MAKE(VMX_BF_IDT_VECTORING_INFO_VECTOR, uVector) | RT_BF_MAKE(VMX_BF_IDT_VECTORING_INFO_TYPE, uIdtVectoringType) | RT_BF_MAKE(VMX_BF_IDT_VECTORING_INFO_ERR_CODE_VALID, fErrCodeValid) | RT_BF_MAKE(VMX_BF_IDT_VECTORING_INFO_VALID, 1); iemVmxVmcsSetIdtVectoringInfo(pVCpu, uIdtVectoringInfo); iemVmxVmcsSetIdtVectoringErrCode(pVCpu, uErrCode); Log2(("vmexit: idt_info=%#RX32 idt_err_code=%#RX32 cr2=%#RX64\n", uIdtVectoringInfo, uErrCode, pVCpu->cpum.GstCtx.cr2)); } } } /* The following VMCS fields should always be zero since we don't support injecting SMIs into a guest. */ Assert(pVmcs->u64RoIoRcx.u == 0); Assert(pVmcs->u64RoIoRsi.u == 0); Assert(pVmcs->u64RoIoRdi.u == 0); Assert(pVmcs->u64RoIoRip.u == 0); /* * Save the guest state back into the VMCS. * We only need to save the state when the VM-entry was successful. */ bool const fVmentryFailed = VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason); if (!fVmentryFailed) { /* We should not cause an NMI-window/interrupt-window VM-exit when injecting events as part of VM-entry. */ if (!CPUMIsGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx)) { Assert(uExitReason != VMX_EXIT_NMI_WINDOW); Assert(uExitReason != VMX_EXIT_INT_WINDOW); } /* For exception or NMI VM-exits, the VM-exit interruption info. field must be valid. */ Assert(uExitReason != VMX_EXIT_XCPT_OR_NMI || VMX_EXIT_INT_INFO_IS_VALID(pVmcs->u32RoExitIntInfo)); /* For external interrupts that occur while "acknowledge interrupt on exit" VM-exit is set, the VM-exit interruption info. field must be valid. */ Assert( uExitReason != VMX_EXIT_EXT_INT || !(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT) || VMX_EXIT_INT_INFO_IS_VALID(pVmcs->u32RoExitIntInfo)); /* * If we support storing EFER.LMA into IA32e-mode guest field on VM-exit, we need to do that now. * See Intel spec. 27.2 "Recording VM-exit Information And Updating VM-entry Control". * * It is not clear from the Intel spec. if this is done only when VM-entry succeeds. * If a VM-exit happens before loading guest EFER, we risk restoring the host EFER.LMA * as guest-CPU state would not been modified. Hence for now, we do this only when * the VM-entry succeeded. */ /** @todo r=ramshankar: Figure out if this bit gets set to host EFER.LMA on real * hardware when VM-exit fails during VM-entry (e.g. VERR_VMX_INVALID_GUEST_STATE). */ if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxExitSaveEferLma) { if (pVCpu->cpum.GstCtx.msrEFER & MSR_K6_EFER_LMA) pVmcs->u32EntryCtls |= VMX_ENTRY_CTLS_IA32E_MODE_GUEST; else pVmcs->u32EntryCtls &= ~VMX_ENTRY_CTLS_IA32E_MODE_GUEST; } /* * The rest of the high bits of the VM-exit reason are only relevant when the VM-exit * occurs in enclave mode/SMM which we don't support yet. * * If we ever add support for it, we can pass just the lower bits to the functions * below, till then an assert should suffice. */ Assert(!RT_HI_U16(uExitReason)); /* Save the guest state into the VMCS and restore guest MSRs from the auto-store guest MSR area. */ iemVmxVmexitSaveGuestState(pVCpu, uExitReason); int rc = iemVmxVmexitSaveGuestAutoMsrs(pVCpu, uExitReason); if (RT_SUCCESS(rc)) { /* likely */ } else return iemVmxAbort(pVCpu, VMXABORT_SAVE_GUEST_MSRS); /* Clear any saved NMI-blocking state so we don't assert on next VM-entry (if it was in effect on the previous one). */ pVCpu->cpum.GstCtx.hwvirt.fSavedInhibit &= ~CPUMCTX_INHIBIT_NMI; } else { /* Restore the NMI-blocking state if VM-entry failed due to invalid guest state or while loading MSRs. */ uint32_t const uExitReasonBasic = VMX_EXIT_REASON_BASIC(uExitReason); if ( uExitReasonBasic == VMX_EXIT_ERR_INVALID_GUEST_STATE || uExitReasonBasic == VMX_EXIT_ERR_MSR_LOAD) iemVmxVmexitRestoreNmiBlockingFF(pVCpu); } /* * Stop any running VMX-preemption timer if necessary. */ if (pVmcs->u32PinCtls & VMX_PIN_CTLS_PREEMPT_TIMER) CPUMStopGuestVmxPremptTimer(pVCpu); /* * Clear the state of "NMI unblocked due to IRET" as otherwise we risk * reporting a stale state on a subsequent VM-exit. This state will be * re-established while emulating IRET in VMX non-root mode. */ pVCpu->cpum.GstCtx.hwvirt.vmx.fNmiUnblockingIret = false; /* * Clear any pending VMX nested-guest force-flags. * These force-flags have no effect on (outer) guest execution and will * be re-evaluated and setup on the next nested-guest VM-entry. */ VMCPU_FF_CLEAR_MASK(pVCpu, VMCPU_FF_VMX_ALL_MASK); /* * We're no longer in nested-guest execution mode. * * It is important to do this prior to loading the host state because * PGM looks at fInVmxNonRootMode to determine if it needs to perform * second-level address translation while switching to host CR3. */ pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxNonRootMode = false; /* Restore the host (outer guest) state. */ VBOXSTRICTRC rcStrict = iemVmxVmexitLoadHostState(pVCpu, uExitReason); if (RT_SUCCESS(rcStrict)) { Assert(rcStrict == VINF_SUCCESS); rcStrict = VINF_VMX_VMEXIT; } else Log(("vmexit: Loading host-state failed. uExitReason=%u rc=%Rrc\n", uExitReason, VBOXSTRICTRC_VAL(rcStrict))); /* * Restore non-zero Secondary-processor based VM-execution controls * when the "activate secondary controls" bit was not set. */ if (pVmcs->u32RestoreProcCtls2) { Assert(!(pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)); pVmcs->u32ProcCtls2 = pVmcs->u32RestoreProcCtls2; pVmcs->u32RestoreProcCtls2 = 0; } if (VM_IS_HM_ENABLED(pVCpu->CTX_SUFF(pVM))) { /* Notify HM that the current VMCS fields have been modified. */ HMNotifyVmxNstGstCurrentVmcsChanged(pVCpu); /* Notify HM that we've completed the VM-exit. */ HMNotifyVmxNstGstVmexit(pVCpu); } # if defined(VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM) && defined(IN_RING3) /* Revert any IEM-only nested-guest execution policy, otherwise return rcStrict. */ Log(("vmexit: Disabling IEM-only EM execution policy!\n")); int rcSched = EMR3SetExecutionPolicy(pVCpu->CTX_SUFF(pVM)->pUVM, EMEXECPOLICY_IEM_ALL, false); if (rcSched != VINF_SUCCESS) iemSetPassUpStatus(pVCpu, rcSched); # endif return rcStrict; # endif } /** * VMX VM-exit handler for VM-exits due to instruction execution. * * This is intended for instructions where the caller provides all the relevant * VM-exit information. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param pExitInfo Pointer to the VM-exit information. */ static VBOXSTRICTRC iemVmxVmexitInstrWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { /* * For instructions where any of the following fields are not applicable: * - Exit qualification must be cleared. * - VM-exit instruction info. is undefined. * - Guest-linear address is undefined. * - Guest-physical address is undefined. * * The VM-exit instruction length is mandatory for all VM-exits that are caused by * instruction execution. For VM-exits that are not due to instruction execution this * field is undefined. * * In our implementation in IEM, all undefined fields are generally cleared. However, * if the caller supplies information (from say the physical CPU directly) it is * then possible that the undefined fields are not cleared. * * See Intel spec. 27.2.1 "Basic VM-Exit Information". * See Intel spec. 27.2.4 "Information for VM Exits Due to Instruction Execution". */ Assert(pExitInfo); AssertMsg(pExitInfo->uReason <= VMX_EXIT_MAX, ("uReason=%u\n", pExitInfo->uReason)); AssertMsg(pExitInfo->cbInstr >= 1 && pExitInfo->cbInstr <= 15, ("uReason=%u cbInstr=%u\n", pExitInfo->uReason, pExitInfo->cbInstr)); /* Update all the relevant fields from the VM-exit instruction information struct. */ iemVmxVmcsSetExitInstrInfo(pVCpu, pExitInfo->InstrInfo.u); iemVmxVmcsSetExitGuestLinearAddr(pVCpu, pExitInfo->u64GuestLinearAddr); iemVmxVmcsSetExitGuestPhysAddr(pVCpu, pExitInfo->u64GuestPhysAddr); iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr); /* Perform the VM-exit. */ return iemVmxVmexit(pVCpu, pExitInfo->uReason, pExitInfo->u64Qual); } /** * VMX VM-exit handler for VM-exits due to instruction execution. * * This is intended for instructions that only provide the VM-exit instruction * length. * * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstr(PVMCPUCC pVCpu, uint32_t uExitReason, uint8_t cbInstr) RT_NOEXCEPT { #ifdef VBOX_STRICT /* * To prevent us from shooting ourselves in the foot. * The follow instructions should convey more than just the instruction length. */ switch (uExitReason) { case VMX_EXIT_INVEPT: case VMX_EXIT_INVPCID: case VMX_EXIT_INVVPID: case VMX_EXIT_LDTR_TR_ACCESS: case VMX_EXIT_GDTR_IDTR_ACCESS: case VMX_EXIT_VMCLEAR: case VMX_EXIT_VMPTRLD: case VMX_EXIT_VMPTRST: case VMX_EXIT_VMREAD: case VMX_EXIT_VMWRITE: case VMX_EXIT_VMXON: case VMX_EXIT_XRSTORS: case VMX_EXIT_XSAVES: case VMX_EXIT_RDRAND: case VMX_EXIT_RDSEED: case VMX_EXIT_IO_INSTR: AssertMsgFailedReturn(("Use iemVmxVmexitInstrNeedsInfo for uExitReason=%u\n", uExitReason), VERR_IEM_IPE_5); break; } #endif VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_INSTR_LEN(uExitReason, cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } /** * Interface for HM and EM to emulate VM-exit due to a triple-fault. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitTripleFault(PVMCPUCC pVCpu) { VBOXSTRICTRC rcStrict = iemVmxVmexit(pVCpu, VMX_EXIT_TRIPLE_FAULT, 0 /* u64ExitQual */); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * Interface for HM and EM to emulate VM-exit due to startup-IPI (SIPI). * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param uVector The SIPI vector. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitStartupIpi(PVMCPUCC pVCpu, uint8_t uVector) { VBOXSTRICTRC rcStrict = iemVmxVmexit(pVCpu, VMX_EXIT_SIPI, uVector); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * Interface for HM and EM to emulate a VM-exit. * * If a specialized version of a VM-exit handler exists, that must be used instead. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param uExitReason The VM-exit reason. * @param u64ExitQual The Exit qualification. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexit(PVMCPUCC pVCpu, uint32_t uExitReason, uint64_t u64ExitQual) { VBOXSTRICTRC rcStrict = iemVmxVmexit(pVCpu, uExitReason, u64ExitQual); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * Interface for HM and EM to emulate a VM-exit due to an instruction. * * This is meant to be used for those instructions that VMX provides additional * decoding information beyond just the instruction length! * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitInstrWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { VBOXSTRICTRC rcStrict = iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * Interface for HM and EM to emulate a VM-exit due to an instruction. * * This is meant to be used for those instructions that VMX provides only the * instruction length. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @param cbInstr The instruction length in bytes. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitInstr(PVMCPUCC pVCpu, uint32_t uExitReason, uint8_t cbInstr) { VBOXSTRICTRC rcStrict = iemVmxVmexitInstr(pVCpu, uExitReason, cbInstr); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * VMX VM-exit handler for VM-exits due to instruction execution. * * This is intended for instructions that have a ModR/M byte and update the VM-exit * instruction information and Exit qualification fields. * * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason. * @param uInstrid The instruction identity (VMXINSTRID_XXX). * @param cbInstr The instruction length in bytes. * * @remarks Do not use this for INS/OUTS instruction. */ VBOXSTRICTRC iemVmxVmexitInstrNeedsInfo(PVMCPUCC pVCpu, uint32_t uExitReason, VMXINSTRID uInstrId, uint8_t cbInstr) RT_NOEXCEPT { #ifdef VBOX_STRICT /* * To prevent us from shooting ourselves in the foot. * The follow instructions convey specific info that require using their respective handlers. */ switch (uExitReason) { case VMX_EXIT_INVEPT: case VMX_EXIT_INVPCID: case VMX_EXIT_INVVPID: case VMX_EXIT_LDTR_TR_ACCESS: case VMX_EXIT_GDTR_IDTR_ACCESS: case VMX_EXIT_VMCLEAR: case VMX_EXIT_VMPTRLD: case VMX_EXIT_VMPTRST: case VMX_EXIT_VMREAD: case VMX_EXIT_VMWRITE: case VMX_EXIT_VMXON: case VMX_EXIT_XRSTORS: case VMX_EXIT_XSAVES: case VMX_EXIT_RDRAND: case VMX_EXIT_RDSEED: break; default: AssertMsgFailedReturn(("Use instruction-specific handler\n"), VERR_IEM_IPE_5); break; } #endif /* * Update the Exit qualification field with displacement bytes. * See Intel spec. 27.2.1 "Basic VM-Exit Information". */ /* Construct the VM-exit instruction information. */ RTGCPTR GCPtrDisp; uint32_t const uInstrInfo = iemVmxGetExitInstrInfo(pVCpu, uExitReason, uInstrId, &GCPtrDisp); VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_INFO(uExitReason, GCPtrDisp, uInstrInfo, cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } /** * VMX VM-exit handler for VM-exits due to INVLPG. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param GCPtrPage The guest-linear address of the page being invalidated. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrInvlpg(PVMCPUCC pVCpu, RTGCPTR GCPtrPage, uint8_t cbInstr) RT_NOEXCEPT { VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_INVLPG, GCPtrPage, cbInstr); Assert(IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode || !RT_HI_U32(ExitInfo.u64Qual)); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } /** * VMX VM-exit handler for VM-exits due to LMSW. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param uGuestCr0 The current guest CR0. * @param pu16NewMsw The machine-status word specified in LMSW's source * operand. This will be updated depending on the VMX * guest/host CR0 mask if LMSW is not intercepted. * @param GCPtrEffDst The guest-linear address of the source operand in case * of a memory operand. For register operand, pass * NIL_RTGCPTR. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrLmsw(PVMCPUCC pVCpu, uint32_t uGuestCr0, uint16_t *pu16NewMsw, RTGCPTR GCPtrEffDst, uint8_t cbInstr) RT_NOEXCEPT { Assert(pu16NewMsw); uint16_t const uNewMsw = *pu16NewMsw; if (CPUMIsGuestVmxLmswInterceptSet(&pVCpu->cpum.GstCtx, uNewMsw)) { Log2(("lmsw: Guest intercept -> VM-exit\n")); bool const fMemOperand = RT_BOOL(GCPtrEffDst != NIL_RTGCPTR); VMXVEXITINFO ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MOV_CRX, RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 0) /* CR0 */ | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_LMSW_OP, fMemOperand) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_LMSW_DATA, uNewMsw) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_LMSW), cbInstr); if (fMemOperand) { Assert(IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode || !RT_HI_U32(GCPtrEffDst)); ExitInfo.u64GuestLinearAddr = GCPtrEffDst; } return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } /* * If LMSW did not cause a VM-exit, any CR0 bits in the range 0:3 that is set in the * CR0 guest/host mask must be left unmodified. * * See Intel Spec. 25.3 "Changes To Instruction Behavior In VMX Non-root Operation". */ uint32_t const fGstHostMask = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64Cr0Mask.u; uint32_t const fGstHostLmswMask = fGstHostMask & (X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS); *pu16NewMsw = (uGuestCr0 & fGstHostLmswMask) | (uNewMsw & ~fGstHostLmswMask); return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to CLTS. * * @returns Strict VBox status code. * @retval VINF_VMX_MODIFIES_BEHAVIOR if the CLTS instruction did not cause a * VM-exit but must not modify the guest CR0.TS bit. * @retval VINF_VMX_INTERCEPT_NOT_ACTIVE if the CLTS instruction did not cause a * VM-exit and modification to the guest CR0.TS bit is allowed (subject to * CR0 fixed bits in VMX operation). * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrClts(PVMCPUCC pVCpu, uint8_t cbInstr) RT_NOEXCEPT { /* * If CR0.TS is owned by the host: * - If CR0.TS is set in the read-shadow, we must cause a VM-exit. * - If CR0.TS is cleared in the read-shadow, no VM-exit is caused and the * CLTS instruction completes without clearing CR0.TS. * * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally". */ uint32_t const fGstHostMask = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64Cr0Mask.u; if (fGstHostMask & X86_CR0_TS) { if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64Cr0ReadShadow.u & X86_CR0_TS) { Log2(("clts: Guest intercept -> VM-exit\n")); VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MOV_CRX, RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 0) /* CR0 */ | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_CLTS), cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } return VINF_VMX_MODIFIES_BEHAVIOR; } /* * If CR0.TS is not owned by the host, the CLTS instructions operates normally * and may modify CR0.TS (subject to CR0 fixed bits in VMX operation). */ return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to 'Mov CR0,GReg' and 'Mov CR4,GReg' * (CR0/CR4 write). * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param iCrReg The control register (either CR0 or CR4). * @param uGuestCrX The current guest CR0/CR4. * @param puNewCrX Pointer to the new CR0/CR4 value. Will be updated if no * VM-exit is caused. * @param iGReg The general register from which the CR0/CR4 value is being * loaded. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrMovToCr0Cr4(PVMCPUCC pVCpu, uint8_t iCrReg, uint64_t *puNewCrX, uint8_t iGReg, uint8_t cbInstr) RT_NOEXCEPT { Assert(puNewCrX); Assert(iCrReg == 0 || iCrReg == 4); Assert(iGReg < X86_GREG_COUNT); uint64_t const uNewCrX = *puNewCrX; if (CPUMIsGuestVmxMovToCr0Cr4InterceptSet(&pVCpu->cpum.GstCtx, iCrReg, uNewCrX)) { Log2(("mov_Cr_Rd: (CR%u) Guest intercept -> VM-exit\n", iCrReg)); VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MOV_CRX, RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, iCrReg) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_WRITE), cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } /* * If the Mov-to-CR0/CR4 did not cause a VM-exit, any bits owned by the host * must not be modified the instruction. * * See Intel Spec. 25.3 "Changes To Instruction Behavior In VMX Non-root Operation". */ uint64_t uGuestCrX; uint64_t fGstHostMask; if (iCrReg == 0) { IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0); uGuestCrX = pVCpu->cpum.GstCtx.cr0; fGstHostMask = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64Cr0Mask.u; } else { IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4); uGuestCrX = pVCpu->cpum.GstCtx.cr4; fGstHostMask = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64Cr4Mask.u; } *puNewCrX = (uGuestCrX & fGstHostMask) | (*puNewCrX & ~fGstHostMask); return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to 'Mov GReg,CR3' (CR3 read). * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param iGReg The general register to which the CR3 value is being stored. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrMovFromCr3(PVMCPUCC pVCpu, uint8_t iGReg, uint8_t cbInstr) RT_NOEXCEPT { Assert(iGReg < X86_GREG_COUNT); IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3); /* * If the CR3-store exiting control is set, we must cause a VM-exit. * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally". */ if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls & VMX_PROC_CTLS_CR3_STORE_EXIT) { Log2(("mov_Rd_Cr: (CR3) Guest intercept -> VM-exit\n")); VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MOV_CRX, RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 3) /* CR3 */ | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_READ), cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to 'Mov CR3,GReg' (CR3 write). * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param uNewCr3 The new CR3 value. * @param iGReg The general register from which the CR3 value is being * loaded. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrMovToCr3(PVMCPUCC pVCpu, uint64_t uNewCr3, uint8_t iGReg, uint8_t cbInstr) RT_NOEXCEPT { Assert(iGReg < X86_GREG_COUNT); /* * If the CR3-load exiting control is set and the new CR3 value does not * match any of the CR3-target values in the VMCS, we must cause a VM-exit. * * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally". */ if (CPUMIsGuestVmxMovToCr3InterceptSet(pVCpu, uNewCr3)) { Log2(("mov_Cr_Rd: (CR3) Guest intercept -> VM-exit\n")); VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MOV_CRX, RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 3) /* CR3 */ | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_WRITE), cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to 'Mov GReg,CR8' (CR8 read). * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param iGReg The general register to which the CR8 value is being stored. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrMovFromCr8(PVMCPUCC pVCpu, uint8_t iGReg, uint8_t cbInstr) RT_NOEXCEPT { Assert(iGReg < X86_GREG_COUNT); /* * If the CR8-store exiting control is set, we must cause a VM-exit. * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally". */ if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls & VMX_PROC_CTLS_CR8_STORE_EXIT) { Log2(("mov_Rd_Cr: (CR8) Guest intercept -> VM-exit\n")); VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MOV_CRX, RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 8) /* CR8 */ | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_READ), cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to 'Mov CR8,GReg' (CR8 write). * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param iGReg The general register from which the CR8 value is being * loaded. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrMovToCr8(PVMCPUCC pVCpu, uint8_t iGReg, uint8_t cbInstr) RT_NOEXCEPT { Assert(iGReg < X86_GREG_COUNT); /* * If the CR8-load exiting control is set, we must cause a VM-exit. * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally". */ if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls & VMX_PROC_CTLS_CR8_LOAD_EXIT) { Log2(("mov_Cr_Rd: (CR8) Guest intercept -> VM-exit\n")); VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MOV_CRX, RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 8) /* CR8 */ | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_WRITE), cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to 'Mov DRx,GReg' (DRx write) and 'Mov * GReg,DRx' (DRx read). * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param uInstrid The instruction identity (VMXINSTRID_MOV_TO_DRX or * VMXINSTRID_MOV_FROM_DRX). * @param iDrReg The debug register being accessed. * @param iGReg The general register to/from which the DRx value is being * store/loaded. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrMovDrX(PVMCPUCC pVCpu, VMXINSTRID uInstrId, uint8_t iDrReg, uint8_t iGReg, uint8_t cbInstr) RT_NOEXCEPT { Assert(iDrReg <= 7); Assert(uInstrId == VMXINSTRID_MOV_TO_DRX || uInstrId == VMXINSTRID_MOV_FROM_DRX); Assert(iGReg < X86_GREG_COUNT); if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT) { VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MOV_DRX, RT_BF_MAKE(VMX_BF_EXIT_QUAL_DRX_REGISTER, iDrReg) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_DRX_GENREG, iGReg) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_DRX_DIRECTION, uInstrId == VMXINSTRID_MOV_TO_DRX ? VMX_EXIT_QUAL_DRX_DIRECTION_WRITE : VMX_EXIT_QUAL_DRX_DIRECTION_READ), cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to I/O instructions (IN and OUT). * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param uInstrId The VM-exit instruction identity (VMXINSTRID_IO_IN or * VMXINSTRID_IO_OUT). * @param u16Port The I/O port being accessed. * @param fImm Whether the I/O port was encoded using an immediate operand * or the implicit DX register. * @param cbAccess The size of the I/O access in bytes (1, 2 or 4 bytes). * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrIo(PVMCPUCC pVCpu, VMXINSTRID uInstrId, uint16_t u16Port, bool fImm, uint8_t cbAccess, uint8_t cbInstr) RT_NOEXCEPT { Assert(uInstrId == VMXINSTRID_IO_IN || uInstrId == VMXINSTRID_IO_OUT); Assert(cbAccess == 1 || cbAccess == 2 || cbAccess == 4); if (CPUMIsGuestVmxIoInterceptSet(pVCpu, u16Port, cbAccess)) { VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_IO_INSTR, RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_WIDTH, cbAccess - 1) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_ENCODING, fImm) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_PORT, u16Port) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_DIRECTION, uInstrId == VMXINSTRID_IO_IN ? VMX_EXIT_QUAL_IO_DIRECTION_IN : VMX_EXIT_QUAL_IO_DIRECTION_OUT), cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to string I/O instructions (INS and OUTS). * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param uInstrId The VM-exit instruction identity (VMXINSTRID_IO_INS or * VMXINSTRID_IO_OUTS). * @param u16Port The I/O port being accessed. * @param cbAccess The size of the I/O access in bytes (1, 2 or 4 bytes). * @param fRep Whether the instruction has a REP prefix or not. * @param ExitInstrInfo The VM-exit instruction info. field. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrStrIo(PVMCPUCC pVCpu, VMXINSTRID uInstrId, uint16_t u16Port, uint8_t cbAccess, bool fRep, VMXEXITINSTRINFO ExitInstrInfo, uint8_t cbInstr) RT_NOEXCEPT { Assert(uInstrId == VMXINSTRID_IO_INS || uInstrId == VMXINSTRID_IO_OUTS); Assert(cbAccess == 1 || cbAccess == 2 || cbAccess == 4); Assert(ExitInstrInfo.StrIo.iSegReg < X86_SREG_COUNT); Assert(ExitInstrInfo.StrIo.u3AddrSize == 0 || ExitInstrInfo.StrIo.u3AddrSize == 1 || ExitInstrInfo.StrIo.u3AddrSize == 2); Assert(uInstrId != VMXINSTRID_IO_INS || ExitInstrInfo.StrIo.iSegReg == X86_SREG_ES); if (CPUMIsGuestVmxIoInterceptSet(pVCpu, u16Port, cbAccess)) { /* * Figure out the guest-linear address and the direction bit (INS/OUTS). */ /** @todo r=ramshankar: Is there something in IEM that already does this? */ static uint64_t const s_auAddrSizeMasks[] = { UINT64_C(0xffff), UINT64_C(0xffffffff), UINT64_C(0xffffffffffffffff) }; uint8_t const iSegReg = ExitInstrInfo.StrIo.iSegReg; uint8_t const uAddrSize = ExitInstrInfo.StrIo.u3AddrSize; uint64_t const uAddrSizeMask = s_auAddrSizeMasks[uAddrSize]; uint32_t uDirection; uint64_t uGuestLinearAddr; if (uInstrId == VMXINSTRID_IO_INS) { uDirection = VMX_EXIT_QUAL_IO_DIRECTION_IN; uGuestLinearAddr = pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base + (pVCpu->cpum.GstCtx.rdi & uAddrSizeMask); } else { uDirection = VMX_EXIT_QUAL_IO_DIRECTION_OUT; uGuestLinearAddr = pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base + (pVCpu->cpum.GstCtx.rsi & uAddrSizeMask); } /* * If the segment is unusable, the guest-linear address in undefined. * We shall clear it for consistency. * * See Intel spec. 27.2.1 "Basic VM-Exit Information". */ if (pVCpu->cpum.GstCtx.aSRegs[iSegReg].Attr.n.u1Unusable) uGuestLinearAddr = 0; VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_INFO_AND_LIN_ADDR(VMX_EXIT_IO_INSTR, RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_WIDTH, cbAccess - 1) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_DIRECTION, uDirection) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_IS_STRING, 1) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_IS_REP, fRep) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_ENCODING, VMX_EXIT_QUAL_IO_ENCODING_DX) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_PORT, u16Port), IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxInsOutInfo ? ExitInstrInfo.u : 0, cbInstr, uGuestLinearAddr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to MWAIT. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param fMonitorHwArmed Whether the address-range monitor hardware is armed. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitInstrMwait(PVMCPUCC pVCpu, bool fMonitorHwArmed, uint8_t cbInstr) RT_NOEXCEPT { VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_MWAIT, fMonitorHwArmed, cbInstr); return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo); } /** * VMX VM-exit handler for VM-exits due to PAUSE. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. */ static VBOXSTRICTRC iemVmxVmexitInstrPause(PVMCPUCC pVCpu, uint8_t cbInstr) RT_NOEXCEPT { PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* * The PAUSE VM-exit is controlled by the "PAUSE exiting" control and the * "PAUSE-loop exiting" control. * * The PLE-Gap is the maximum number of TSC ticks between two successive executions of * the PAUSE instruction before we cause a VM-exit. The PLE-Window is the maximum amount * of TSC ticks the guest is allowed to execute in a pause loop before we must cause * a VM-exit. * * See Intel spec. 24.6.13 "Controls for PAUSE-Loop Exiting". * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally". */ bool fIntercept = false; if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_PAUSE_EXIT) fIntercept = true; else if ( (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT) && IEM_GET_CPL(pVCpu) == 0) { IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_HWVIRT); /* * A previous-PAUSE-tick value of 0 is used to identify the first time * execution of a PAUSE instruction after VM-entry at CPL 0. We must * consider this to be the first execution of PAUSE in a loop according * to the Intel. * * All subsequent records for the previous-PAUSE-tick we ensure that it * cannot be zero by OR'ing 1 to rule out the TSC wrap-around cases at 0. */ uint64_t *puFirstPauseLoopTick = &pVCpu->cpum.GstCtx.hwvirt.vmx.uFirstPauseLoopTick; uint64_t *puPrevPauseTick = &pVCpu->cpum.GstCtx.hwvirt.vmx.uPrevPauseTick; uint64_t const uTick = TMCpuTickGet(pVCpu); uint32_t const uPleGap = pVmcs->u32PleGap; uint32_t const uPleWindow = pVmcs->u32PleWindow; if ( *puPrevPauseTick == 0 || uTick - *puPrevPauseTick > uPleGap) *puFirstPauseLoopTick = uTick; else if (uTick - *puFirstPauseLoopTick > uPleWindow) fIntercept = true; *puPrevPauseTick = uTick | 1; } if (fIntercept) return iemVmxVmexitInstr(pVCpu, VMX_EXIT_PAUSE, cbInstr); return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exits due to task switches. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param enmTaskSwitch The cause of the task switch. * @param SelNewTss The selector of the new TSS. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitTaskSwitch(PVMCPUCC pVCpu, IEMTASKSWITCH enmTaskSwitch, RTSEL SelNewTss, uint8_t cbInstr) RT_NOEXCEPT { /* * Task-switch VM-exits are unconditional and provide the Exit qualification. * * If the cause of the task switch is due to execution of CALL, IRET or the JMP * instruction or delivery of the exception generated by one of these instructions * lead to a task switch through a task gate in the IDT, we need to provide the * VM-exit instruction length. Any other means of invoking a task switch VM-exit * leaves the VM-exit instruction length field undefined. * * See Intel spec. 25.2 "Other Causes Of VM Exits". * See Intel spec. 27.2.4 "Information for VM Exits Due to Instruction Execution". */ Assert(cbInstr <= 15); uint8_t uType; switch (enmTaskSwitch) { case IEMTASKSWITCH_CALL: uType = VMX_EXIT_QUAL_TASK_SWITCH_TYPE_CALL; break; case IEMTASKSWITCH_IRET: uType = VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IRET; break; case IEMTASKSWITCH_JUMP: uType = VMX_EXIT_QUAL_TASK_SWITCH_TYPE_JMP; break; case IEMTASKSWITCH_INT_XCPT: uType = VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IDT; break; IEM_NOT_REACHED_DEFAULT_CASE_RET(); } uint64_t const u64ExitQual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_TASK_SWITCH_NEW_TSS, SelNewTss) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_TASK_SWITCH_SOURCE, uType); iemVmxVmcsSetExitInstrLen(pVCpu, cbInstr); return iemVmxVmexit(pVCpu, VMX_EXIT_TASK_SWITCH, u64ExitQual); } /** * VMX VM-exit handler for trap-like VM-exits. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. */ static VBOXSTRICTRC iemVmxVmexitTrapLikeWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { Assert(VMXIsVmexitTrapLike(pExitInfo->uReason)); iemVmxVmcsSetGuestPendingDbgXcpts(pVCpu, pExitInfo->u64GuestPendingDbgXcpts); return iemVmxVmexit(pVCpu, pExitInfo->uReason, pExitInfo->u64Qual); } /** * Interface for HM and EM to emulate a trap-like VM-exit (MTF, APIC-write, * Virtualized-EOI, TPR-below threshold). * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitTrapLike(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { Assert(pExitInfo); VBOXSTRICTRC rcStrict = iemVmxVmexitTrapLikeWithInfo(pVCpu, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * VMX VM-exit handler for VM-exits due to task switches. * * This is intended for task switches where the caller provides all the relevant * VM-exit information. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. */ static VBOXSTRICTRC iemVmxVmexitTaskSwitchWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo) RT_NOEXCEPT { Assert(pExitInfo->uReason == VMX_EXIT_TASK_SWITCH); iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr); iemVmxVmcsSetIdtVectoringInfo(pVCpu, pExitEventInfo->uIdtVectoringInfo); iemVmxVmcsSetIdtVectoringErrCode(pVCpu, pExitEventInfo->uIdtVectoringErrCode); return iemVmxVmexit(pVCpu, VMX_EXIT_TASK_SWITCH, pExitInfo->u64Qual); } /** * Interface for HM and EM to emulate a VM-exit due to a task switch. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitTaskSwitch(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo) { Assert(pExitInfo); Assert(pExitEventInfo); Assert(pExitInfo->uReason == VMX_EXIT_TASK_SWITCH); VBOXSTRICTRC rcStrict = iemVmxVmexitTaskSwitchWithInfo(pVCpu, pExitInfo, pExitEventInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * VMX VM-exit handler for VM-exits due to expiring of the preemption timer. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. */ VBOXSTRICTRC iemVmxVmexitPreemptTimer(PVMCPUCC pVCpu) RT_NOEXCEPT { Assert(VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER)); Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32PinCtls & VMX_PIN_CTLS_PREEMPT_TIMER); /* Import the hardware virtualization state (for nested-guest VM-entry TSC-tick). */ IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_HWVIRT); /* Save the VMX-preemption timer value (of 0) back in to the VMCS if the CPU supports this feature. */ if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ExitCtls & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER) pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32PreemptTimer = 0; /* Cause the VMX-preemption timer VM-exit. The Exit qualification MBZ. */ return iemVmxVmexit(pVCpu, VMX_EXIT_PREEMPT_TIMER, 0 /* u64ExitQual */); } /** * Interface for HM and EM to emulate VM-exit due to expiry of the preemption timer. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitPreemptTimer(PVMCPUCC pVCpu) { VBOXSTRICTRC rcStrict = iemVmxVmexitPreemptTimer(pVCpu); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * VMX VM-exit handler for VM-exits due to external interrupts. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param uVector The external interrupt vector (pass 0 if the interrupt * is still pending since we typically won't know the * vector). * @param fIntPending Whether the external interrupt is pending or * acknowledged in the interrupt controller. */ static VBOXSTRICTRC iemVmxVmexitExtInt(PVMCPUCC pVCpu, uint8_t uVector, bool fIntPending) RT_NOEXCEPT { Assert(!fIntPending || uVector == 0); /* The VM-exit is subject to "External interrupt exiting" being set. */ if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32PinCtls & VMX_PIN_CTLS_EXT_INT_EXIT) { if (fIntPending) { /* * If the interrupt is pending and we don't need to acknowledge the * interrupt on VM-exit, cause the VM-exit immediately. * * See Intel spec 25.2 "Other Causes Of VM Exits". */ if (!(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT)) return iemVmxVmexit(pVCpu, VMX_EXIT_EXT_INT, 0 /* u64ExitQual */); /* * If the interrupt is pending and we -do- need to acknowledge the interrupt * on VM-exit, postpone VM-exit till after the interrupt controller has been * acknowledged that the interrupt has been consumed. Callers would have to call * us again after getting the vector (and ofc, with fIntPending with false). */ return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /* * If the interrupt is no longer pending (i.e. it has been acknowledged) and the * "External interrupt exiting" and "Acknowledge interrupt on VM-exit" controls are * all set, we need to record the vector of the external interrupt in the * VM-exit interruption information field. Otherwise, mark this field as invalid. * * See Intel spec. 27.2.2 "Information for VM Exits Due to Vectored Events". */ uint32_t uExitIntInfo; if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT) { bool const fNmiUnblocking = pVCpu->cpum.GstCtx.hwvirt.vmx.fNmiUnblockingIret; uExitIntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, uVector) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_EXT_INT) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_NMI_UNBLOCK_IRET, fNmiUnblocking) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VALID, 1); } else uExitIntInfo = 0; iemVmxVmcsSetExitIntInfo(pVCpu, uExitIntInfo); /* * Cause the VM-exit whether or not the vector has been stored * in the VM-exit interruption-information field. */ return iemVmxVmexit(pVCpu, VMX_EXIT_EXT_INT, 0 /* u64ExitQual */); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * Interface for HM and EM to emulate VM-exit due to external interrupts. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param uVector The external interrupt vector (pass 0 if the external * interrupt is still pending). * @param fIntPending Whether the external interrupt is pending or * acknowledged in the interrupt controller. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitExtInt(PVMCPUCC pVCpu, uint8_t uVector, bool fIntPending) { VBOXSTRICTRC rcStrict = iemVmxVmexitExtInt(pVCpu, uVector, fIntPending); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * VMX VM-exit handler for VM-exits due to a double fault caused during delivery of * an event. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. */ VBOXSTRICTRC iemVmxVmexitEventDoubleFault(PVMCPUCC pVCpu) RT_NOEXCEPT { uint32_t const fXcptBitmap = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32XcptBitmap; if (fXcptBitmap & RT_BIT(X86_XCPT_DF)) { /* * The NMI-unblocking due to IRET field need not be set for double faults. * See Intel spec. 31.7.1.2 "Resuming Guest Software After Handling An Exception". */ uint32_t const uExitIntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, X86_XCPT_DF) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_ERR_CODE_VALID, 1) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_NMI_UNBLOCK_IRET, 0) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VALID, 1); iemVmxVmcsSetExitIntInfo(pVCpu, uExitIntInfo); return iemVmxVmexit(pVCpu, VMX_EXIT_XCPT_OR_NMI, 0 /* u64ExitQual */); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for VM-exit due to delivery of an events. * * This is intended for VM-exit due to exceptions or NMIs where the caller provides * all the relevant VM-exit information. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. */ static VBOXSTRICTRC iemVmxVmexitEventWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo) RT_NOEXCEPT { Assert(pExitInfo); Assert(pExitEventInfo); Assert(pExitInfo->uReason == VMX_EXIT_XCPT_OR_NMI); Assert(VMX_EXIT_INT_INFO_IS_VALID(pExitEventInfo->uExitIntInfo)); iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr); iemVmxVmcsSetExitIntInfo(pVCpu, pExitEventInfo->uExitIntInfo); iemVmxVmcsSetExitIntErrCode(pVCpu, pExitEventInfo->uExitIntErrCode); iemVmxVmcsSetIdtVectoringInfo(pVCpu, pExitEventInfo->uIdtVectoringInfo); iemVmxVmcsSetIdtVectoringErrCode(pVCpu, pExitEventInfo->uIdtVectoringErrCode); return iemVmxVmexit(pVCpu, VMX_EXIT_XCPT_OR_NMI, pExitInfo->u64Qual); } /** * Interface for HM and EM to emulate VM-exit due to NMIs. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitXcptNmi(PVMCPUCC pVCpu) { VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_ONLY_REASON(VMX_EXIT_XCPT_OR_NMI); VMXVEXITEVENTINFO const ExitEventInfo = VMXVEXITEVENTINFO_INIT_ONLY_INT( RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VALID, 1) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_NMI) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, X86_XCPT_NMI), 0); VBOXSTRICTRC rcStrict = iemVmxVmexitEventWithInfo(pVCpu, &ExitInfo, &ExitEventInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * Interface for HM and EM to emulate VM-exit due to exceptions. * * Exception includes NMIs, software exceptions (those generated by INT3 or * INTO) and privileged software exceptions (those generated by INT1/ICEBP). * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitXcpt(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo) { Assert(pExitInfo); Assert(pExitEventInfo); VBOXSTRICTRC rcStrict = iemVmxVmexitEventWithInfo(pVCpu, pExitInfo, pExitEventInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * VMX VM-exit handler for VM-exits due to delivery of an event. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param uVector The interrupt / exception vector. * @param fFlags The flags (see IEM_XCPT_FLAGS_XXX). * @param uErrCode The error code associated with the event. * @param uCr2 The CR2 value in case of a \#PF exception. * @param cbInstr The instruction length in bytes. */ VBOXSTRICTRC iemVmxVmexitEvent(PVMCPUCC pVCpu, uint8_t uVector, uint32_t fFlags, uint32_t uErrCode, uint64_t uCr2, uint8_t cbInstr) RT_NOEXCEPT { /* * If the event is being injected as part of VM-entry, it is -not- subject to event * intercepts in the nested-guest. However, secondary exceptions that occur during * injection of any event -are- subject to event interception. * * See Intel spec. 26.5.1.2 "VM Exits During Event Injection". */ if (!CPUMIsGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx)) { /* * If the event is a virtual-NMI (which is an NMI being inject during VM-entry) * virtual-NMI blocking must be set in effect rather than physical NMI blocking. * * See Intel spec. 24.6.1 "Pin-Based VM-Execution Controls". */ if ( uVector == X86_XCPT_NMI && (fFlags & IEM_XCPT_FLAGS_T_CPU_XCPT) && (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)) pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking = true; else Assert(!pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking); CPUMSetGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx, true); return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /* * We are injecting an external interrupt, check if we need to cause a VM-exit now. * If not, the caller will continue delivery of the external interrupt as it would * normally. The interrupt is no longer pending in the interrupt controller at this * point. */ if (fFlags & IEM_XCPT_FLAGS_T_EXT_INT) { Assert(!VMX_IDT_VECTORING_INFO_IS_VALID(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32RoIdtVectoringInfo)); return iemVmxVmexitExtInt(pVCpu, uVector, false /* fIntPending */); } /* * Evaluate intercepts for hardware exceptions, software exceptions (#BP, #OF), * and privileged software exceptions (#DB generated by INT1/ICEBP) and software * interrupts. */ Assert(fFlags & (IEM_XCPT_FLAGS_T_CPU_XCPT | IEM_XCPT_FLAGS_T_SOFT_INT)); bool fIntercept; if ( !(fFlags & IEM_XCPT_FLAGS_T_SOFT_INT) || (fFlags & (IEM_XCPT_FLAGS_BP_INSTR | IEM_XCPT_FLAGS_OF_INSTR | IEM_XCPT_FLAGS_ICEBP_INSTR))) fIntercept = CPUMIsGuestVmxXcptInterceptSet(&pVCpu->cpum.GstCtx, uVector, uErrCode); else { /* Software interrupts cannot be intercepted and therefore do not cause a VM-exit. */ fIntercept = false; } /* * Now that we've determined whether the event causes a VM-exit, we need to construct the * relevant VM-exit information and cause the VM-exit. */ if (fIntercept) { Assert(!(fFlags & IEM_XCPT_FLAGS_T_EXT_INT)); /* Construct the rest of the event related information fields and cause the VM-exit. */ uint64_t u64ExitQual; if (uVector == X86_XCPT_PF) { Assert(fFlags & IEM_XCPT_FLAGS_CR2); u64ExitQual = uCr2; } else if (uVector == X86_XCPT_DB) { IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_DR6); u64ExitQual = pVCpu->cpum.GstCtx.dr[6] & VMX_VMCS_EXIT_QUAL_VALID_MASK; } else u64ExitQual = 0; uint8_t const fNmiUnblocking = pVCpu->cpum.GstCtx.hwvirt.vmx.fNmiUnblockingIret; bool const fErrCodeValid = RT_BOOL(fFlags & IEM_XCPT_FLAGS_ERR); uint8_t const uIntInfoType = iemVmxGetEventType(uVector, fFlags); uint32_t const uExitIntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, uVector) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_TYPE, uIntInfoType) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_ERR_CODE_VALID, fErrCodeValid) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_NMI_UNBLOCK_IRET, fNmiUnblocking) | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VALID, 1); iemVmxVmcsSetExitIntInfo(pVCpu, uExitIntInfo); iemVmxVmcsSetExitIntErrCode(pVCpu, uErrCode); /* * For VM-exits due to software exceptions (those generated by INT3 or INTO) or privileged * software exceptions (those generated by INT1/ICEBP) we need to supply the VM-exit instruction * length. */ if ( (fFlags & IEM_XCPT_FLAGS_T_SOFT_INT) || (fFlags & (IEM_XCPT_FLAGS_BP_INSTR | IEM_XCPT_FLAGS_OF_INSTR | IEM_XCPT_FLAGS_ICEBP_INSTR))) iemVmxVmcsSetExitInstrLen(pVCpu, cbInstr); else iemVmxVmcsSetExitInstrLen(pVCpu, 0); return iemVmxVmexit(pVCpu, VMX_EXIT_XCPT_OR_NMI, u64ExitQual); } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * VMX VM-exit handler for EPT misconfiguration. * * @param pVCpu The cross context virtual CPU structure. * @param GCPhysAddr The physical address causing the EPT misconfiguration. * This need not be page aligned (e.g. nested-guest in real * mode). */ static VBOXSTRICTRC iemVmxVmexitEptMisconfig(PVMCPUCC pVCpu, RTGCPHYS GCPhysAddr) RT_NOEXCEPT { iemVmxVmcsSetExitGuestPhysAddr(pVCpu, GCPhysAddr); return iemVmxVmexit(pVCpu, VMX_EXIT_EPT_MISCONFIG, 0 /* u64ExitQual */); } /** * VMX VM-exit handler for EPT misconfiguration. * * This is intended for EPT misconfigurations where the caller provides all the * relevant VM-exit information. * * @param pVCpu The cross context virtual CPU structure. * @param GCPhysAddr The physical address causing the EPT misconfiguration. * This need not be page aligned (e.g. nested-guest in real * mode). * @param pExitEventInfo Pointer to the VM-exit event information. */ static VBOXSTRICTRC iemVmxVmexitEptMisconfigWithInfo(PVMCPUCC pVCpu, RTGCPHYS GCPhysAddr, PCVMXVEXITEVENTINFO pExitEventInfo) RT_NOEXCEPT { Assert(pExitEventInfo); Assert(!VMX_EXIT_INT_INFO_IS_VALID(pExitEventInfo->uExitIntInfo)); iemVmxVmcsSetIdtVectoringInfo(pVCpu, pExitEventInfo->uIdtVectoringInfo); iemVmxVmcsSetIdtVectoringErrCode(pVCpu, pExitEventInfo->uIdtVectoringErrCode); iemVmxVmcsSetExitGuestPhysAddr(pVCpu, GCPhysAddr); return iemVmxVmexit(pVCpu, VMX_EXIT_EPT_MISCONFIG, 0 /* u64ExitQual */); } /** * Interface for HM and EM to emulate a VM-exit due to an EPT misconfiguration. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param GCPhysAddr The nested-guest physical address causing the EPT * misconfiguration. * @param pExitEventInfo Pointer to the VM-exit event information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitEptMisconfig(PVMCPUCC pVCpu, RTGCPHYS GCPhysAddr, PCVMXVEXITEVENTINFO pExitEventInfo) { IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); iemInitExec(pVCpu, 0 /*fExecOpts*/); VBOXSTRICTRC rcStrict = iemVmxVmexitEptMisconfigWithInfo(pVCpu, GCPhysAddr, pExitEventInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * VMX VM-exit handler for EPT violation. * * @param pVCpu The cross context virtual CPU structure. * @param fAccess The access causing the EPT violation, IEM_ACCESS_XXX. * @param fSlatFail The SLAT failure info, IEM_SLAT_FAIL_XXX. * @param fEptAccess The EPT paging structure bits. * @param GCPhysAddr The physical address causing the EPT violation. This * need not be page aligned (e.g. nested-guest in real * mode). * @param fIsLinearAddrValid Whether translation of a linear address caused this * EPT violation. If @c false, GCPtrAddr must be 0. * @param GCPtrAddr The linear address causing the EPT violation. * @param cbInstr The VM-exit instruction length. */ static VBOXSTRICTRC iemVmxVmexitEptViolation(PVMCPUCC pVCpu, uint32_t fAccess, uint32_t fSlatFail, uint64_t fEptAccess, RTGCPHYS GCPhysAddr, bool fIsLinearAddrValid, uint64_t GCPtrAddr, uint8_t cbInstr) RT_NOEXCEPT { /* * If the linear address isn't valid (can happen when loading PDPTEs * as part of MOV CR execution) the linear address field is undefined. * While we can leave it this way, it's preferable to zero it for consistency. */ Assert(fIsLinearAddrValid || GCPtrAddr == 0); uint64_t const fCaps = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64EptVpidCaps; bool const fSupportsAccessDirty = RT_BOOL(fCaps & MSR_IA32_VMX_EPT_VPID_CAP_ACCESS_DIRTY); uint32_t const fDataRdMask = IEM_ACCESS_WHAT_MASK | IEM_ACCESS_TYPE_READ; uint32_t const fDataWrMask = IEM_ACCESS_WHAT_MASK | IEM_ACCESS_TYPE_WRITE; uint32_t const fInstrMask = IEM_ACCESS_WHAT_MASK | IEM_ACCESS_TYPE_EXEC; bool const fDataRead = ((fAccess & fDataRdMask) == IEM_ACCESS_DATA_R) | fSupportsAccessDirty; bool const fDataWrite = ((fAccess & fDataWrMask) == IEM_ACCESS_DATA_W) | fSupportsAccessDirty; bool const fInstrFetch = ((fAccess & fInstrMask) == IEM_ACCESS_INSTRUCTION); bool const fEptRead = RT_BOOL(fEptAccess & EPT_E_READ); bool const fEptWrite = RT_BOOL(fEptAccess & EPT_E_WRITE); bool const fEptExec = RT_BOOL(fEptAccess & EPT_E_EXECUTE); bool const fNmiUnblocking = pVCpu->cpum.GstCtx.hwvirt.vmx.fNmiUnblockingIret; bool const fIsLinearToPhysAddr = fIsLinearAddrValid & RT_BOOL(fSlatFail & IEM_SLAT_FAIL_LINEAR_TO_PHYS_ADDR); uint64_t const u64ExitQual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_ACCESS_READ, fDataRead) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_ACCESS_WRITE, fDataWrite) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_ACCESS_INSTR_FETCH, fInstrFetch) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_ENTRY_READ, fEptRead) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_ENTRY_WRITE, fEptWrite) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_ENTRY_EXECUTE, fEptExec) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_LINEAR_ADDR_VALID, fIsLinearAddrValid) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_LINEAR_TO_PHYS_ADDR, fIsLinearToPhysAddr) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_EPT_NMI_UNBLOCK_IRET, fNmiUnblocking); #ifdef VBOX_STRICT uint64_t const fMiscCaps = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Misc; uint32_t const fProcCtls2 = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2; Assert(!(fCaps & MSR_IA32_VMX_EPT_VPID_CAP_ADVEXITINFO_EPT_VIOLATION)); /* Advanced VM-exit info. not supported */ Assert(!(fCaps & MSR_IA32_VMX_EPT_VPID_CAP_SUPER_SHW_STACK)); /* Supervisor shadow stack control not supported. */ Assert(!(RT_BF_GET(fMiscCaps, VMX_BF_MISC_INTEL_PT))); /* Intel PT not supported. */ Assert(!(fProcCtls2 & VMX_PROC_CTLS2_MODE_BASED_EPT_PERM)); /* Mode-based execute control not supported. */ #endif iemVmxVmcsSetExitGuestPhysAddr(pVCpu, GCPhysAddr); iemVmxVmcsSetExitGuestLinearAddr(pVCpu, GCPtrAddr); iemVmxVmcsSetExitInstrLen(pVCpu, cbInstr); return iemVmxVmexit(pVCpu, VMX_EXIT_EPT_VIOLATION, u64ExitQual); } /** * VMX VM-exit handler for EPT violation. * * This is intended for EPT violations where the caller provides all the * relevant VM-exit information. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. */ static VBOXSTRICTRC iemVmxVmexitEptViolationWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo) RT_NOEXCEPT { Assert(pExitInfo); Assert(pExitEventInfo); Assert(pExitInfo->uReason == VMX_EXIT_EPT_VIOLATION); Assert(!VMX_EXIT_INT_INFO_IS_VALID(pExitEventInfo->uExitIntInfo)); iemVmxVmcsSetIdtVectoringInfo(pVCpu, pExitEventInfo->uIdtVectoringInfo); iemVmxVmcsSetIdtVectoringErrCode(pVCpu, pExitEventInfo->uIdtVectoringErrCode); iemVmxVmcsSetExitGuestPhysAddr(pVCpu, pExitInfo->u64GuestPhysAddr); if (pExitInfo->u64Qual & VMX_BF_EXIT_QUAL_EPT_LINEAR_ADDR_VALID_MASK) iemVmxVmcsSetExitGuestLinearAddr(pVCpu, pExitInfo->u64GuestLinearAddr); else iemVmxVmcsSetExitGuestLinearAddr(pVCpu, 0); iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr); return iemVmxVmexit(pVCpu, VMX_EXIT_EPT_VIOLATION, pExitInfo->u64Qual); } /** * Interface for HM and EM to emulate a VM-exit due to an EPT violation. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitEptViolation(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo) { IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); iemInitExec(pVCpu, 0 /*fExecOpts*/); VBOXSTRICTRC rcStrict = iemVmxVmexitEptViolationWithInfo(pVCpu, pExitInfo, pExitEventInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * VMX VM-exit handler for EPT-induced VM-exits. * * @param pVCpu The cross context virtual CPU structure. * @param pWalk The page walk info. * @param fAccess The access causing the EPT event, IEM_ACCESS_XXX. * @param fSlatFail Additional SLAT info, IEM_SLAT_FAIL_XXX. * @param cbInstr The VM-exit instruction length if applicable. Pass 0 if not * applicable. */ VBOXSTRICTRC iemVmxVmexitEpt(PVMCPUCC pVCpu, PPGMPTWALK pWalk, uint32_t fAccess, uint32_t fSlatFail, uint8_t cbInstr) RT_NOEXCEPT { Assert(pWalk->fIsSlat); Assert(pWalk->fFailed & PGM_WALKFAIL_EPT); Assert(!IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxEptXcptVe); /* #VE exceptions not supported. */ Assert(!(pWalk->fFailed & PGM_WALKFAIL_EPT_VIOLATION_CONVERTIBLE)); /* Without #VE, convertible violations not possible. */ if (pWalk->fFailed & PGM_WALKFAIL_EPT_VIOLATION) { LogFlow(("EptViolation: cs:rip=%04x:%08RX64 fAccess=%#RX32\n", pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, fAccess)); uint64_t const fEptAccess = (pWalk->fEffective & PGM_PTATTRS_EPT_MASK) >> PGM_PTATTRS_EPT_SHIFT; return iemVmxVmexitEptViolation(pVCpu, fAccess, fSlatFail, fEptAccess, pWalk->GCPhysNested, pWalk->fIsLinearAddrValid, pWalk->GCPtr, cbInstr); } LogFlow(("EptMisconfig: cs:rip=%04x:%08RX64 fAccess=%#RX32\n", pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, fAccess)); Assert(pWalk->fFailed & PGM_WALKFAIL_EPT_MISCONFIG); return iemVmxVmexitEptMisconfig(pVCpu, pWalk->GCPhysNested); } /** * VMX VM-exit handler for APIC accesses. * * @param pVCpu The cross context virtual CPU structure. * @param offAccess The offset of the register being accessed. * @param fAccess The type of access, see IEM_ACCESS_XXX. */ static VBOXSTRICTRC iemVmxVmexitApicAccess(PVMCPUCC pVCpu, uint16_t offAccess, uint32_t fAccess) RT_NOEXCEPT { VMXAPICACCESS enmAccess; bool const fInEventDelivery = IEMGetCurrentXcpt(pVCpu, NULL, NULL, NULL, NULL); if (fInEventDelivery) enmAccess = VMXAPICACCESS_LINEAR_EVENT_DELIVERY; else if ((fAccess & (IEM_ACCESS_WHAT_MASK | IEM_ACCESS_TYPE_MASK)) == IEM_ACCESS_INSTRUCTION) enmAccess = VMXAPICACCESS_LINEAR_INSTR_FETCH; else if (fAccess & IEM_ACCESS_TYPE_WRITE) enmAccess = VMXAPICACCESS_LINEAR_WRITE; else enmAccess = VMXAPICACCESS_LINEAR_READ; uint64_t const u64ExitQual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_APIC_ACCESS_OFFSET, offAccess) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_APIC_ACCESS_TYPE, enmAccess); return iemVmxVmexit(pVCpu, VMX_EXIT_APIC_ACCESS, u64ExitQual); } /** * VMX VM-exit handler for APIC accesses. * * This is intended for APIC accesses where the caller provides all the * relevant VM-exit information. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. */ static VBOXSTRICTRC iemVmxVmexitApicAccessWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo) RT_NOEXCEPT { /* VM-exit interruption information should not be valid for APIC-access VM-exits. */ Assert(!VMX_EXIT_INT_INFO_IS_VALID(pExitEventInfo->uExitIntInfo)); Assert(pExitInfo->uReason == VMX_EXIT_APIC_ACCESS); iemVmxVmcsSetExitIntInfo(pVCpu, 0); iemVmxVmcsSetExitIntErrCode(pVCpu, 0); iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr); iemVmxVmcsSetIdtVectoringInfo(pVCpu, pExitEventInfo->uIdtVectoringInfo); iemVmxVmcsSetIdtVectoringErrCode(pVCpu, pExitEventInfo->uIdtVectoringErrCode); return iemVmxVmexit(pVCpu, VMX_EXIT_APIC_ACCESS, pExitInfo->u64Qual); } /** * Interface for HM and EM to virtualize memory-mapped APIC accesses. * * @returns Strict VBox status code. * @retval VINF_VMX_MODIFIES_BEHAVIOR if the memory access was virtualized. * @retval VINF_VMX_VMEXIT if the access causes a VM-exit. * * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @param pExitEventInfo Pointer to the VM-exit event information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitApicAccess(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo) { Assert(pExitInfo); Assert(pExitEventInfo); VBOXSTRICTRC rcStrict = iemVmxVmexitApicAccessWithInfo(pVCpu, pExitInfo, pExitEventInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * VMX VM-exit handler for APIC-write VM-exits. * * @param pVCpu The cross context virtual CPU structure. * @param offApic The write to the virtual-APIC page offset that caused this * VM-exit. */ static VBOXSTRICTRC iemVmxVmexitApicWrite(PVMCPUCC pVCpu, uint16_t offApic) RT_NOEXCEPT { Assert(offApic < XAPIC_OFF_END + 4); /* Write only bits 11:0 of the APIC offset into the Exit qualification field. */ offApic &= UINT16_C(0xfff); return iemVmxVmexit(pVCpu, VMX_EXIT_APIC_WRITE, offApic); } /** * Clears any pending virtual-APIC write emulation. * * @returns The virtual-APIC offset that was written before clearing it. * @param pVCpu The cross context virtual CPU structure. */ DECLINLINE(uint16_t) iemVmxVirtApicClearPendingWrite(PVMCPUCC pVCpu) { IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_HWVIRT); uint16_t const offVirtApicWrite = pVCpu->cpum.GstCtx.hwvirt.vmx.offVirtApicWrite; pVCpu->cpum.GstCtx.hwvirt.vmx.offVirtApicWrite = 0; Assert(VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE)); VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_VMX_APIC_WRITE); return offVirtApicWrite; } /** * Reads a 32-bit register from the virtual-APIC page at the given offset. * * @returns The register from the virtual-APIC page. * @param pVCpu The cross context virtual CPU structure. * @param offReg The offset of the register being read. */ uint32_t iemVmxVirtApicReadRaw32(PVMCPUCC pVCpu, uint16_t offReg) RT_NOEXCEPT { Assert(offReg <= VMX_V_VIRT_APIC_SIZE - sizeof(uint32_t)); uint32_t uReg = 0; RTGCPHYS const GCPhysVirtApic = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64AddrVirtApic.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &uReg, GCPhysVirtApic + offReg, sizeof(uReg)); AssertMsgStmt(RT_SUCCESS(rc), ("Failed to read %u bytes at offset %#x of the virtual-APIC page at %#RGp: %Rrc\n", sizeof(uReg), offReg, GCPhysVirtApic, rc), uReg = 0); return uReg; } /** * Reads a 64-bit register from the virtual-APIC page at the given offset. * * @returns The register from the virtual-APIC page. * @param pVCpu The cross context virtual CPU structure. * @param offReg The offset of the register being read. */ static uint64_t iemVmxVirtApicReadRaw64(PVMCPUCC pVCpu, uint16_t offReg) RT_NOEXCEPT { Assert(offReg <= VMX_V_VIRT_APIC_SIZE - sizeof(uint64_t)); uint64_t uReg = 0; RTGCPHYS const GCPhysVirtApic = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64AddrVirtApic.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &uReg, GCPhysVirtApic + offReg, sizeof(uReg)); AssertMsgStmt(RT_SUCCESS(rc), ("Failed to read %u bytes at offset %#x of the virtual-APIC page at %#RGp: %Rrc\n", sizeof(uReg), offReg, GCPhysVirtApic, rc), uReg = 0); return uReg; } /** * Writes a 32-bit register to the virtual-APIC page at the given offset. * * @param pVCpu The cross context virtual CPU structure. * @param offReg The offset of the register being written. * @param uReg The register value to write. */ void iemVmxVirtApicWriteRaw32(PVMCPUCC pVCpu, uint16_t offReg, uint32_t uReg) RT_NOEXCEPT { Assert(offReg <= VMX_V_VIRT_APIC_SIZE - sizeof(uint32_t)); RTGCPHYS const GCPhysVirtApic = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64AddrVirtApic.u; int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic + offReg, &uReg, sizeof(uReg)); AssertMsgRC(rc, ("Failed to write %u bytes at offset %#x of the virtual-APIC page at %#RGp: %Rrc\n", sizeof(uReg), offReg, GCPhysVirtApic, rc)); } /** * Writes a 64-bit register to the virtual-APIC page at the given offset. * * @param pVCpu The cross context virtual CPU structure. * @param offReg The offset of the register being written. * @param uReg The register value to write. */ static void iemVmxVirtApicWriteRaw64(PVMCPUCC pVCpu, uint16_t offReg, uint64_t uReg) RT_NOEXCEPT { Assert(offReg <= VMX_V_VIRT_APIC_SIZE - sizeof(uint64_t)); RTGCPHYS const GCPhysVirtApic = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64AddrVirtApic.u; int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic + offReg, &uReg, sizeof(uReg)); AssertMsgRC(rc, ("Failed to write %u bytes at offset %#x of the virtual-APIC page at %#RGp: %Rrc\n", sizeof(uReg), offReg, GCPhysVirtApic, rc)); } /** * Sets the vector in a virtual-APIC 256-bit sparse register. * * @param pVCpu The cross context virtual CPU structure. * @param offReg The offset of the 256-bit spare register. * @param uVector The vector to set. * * @remarks This is based on our APIC device code. */ static void iemVmxVirtApicSetVectorInReg(PVMCPUCC pVCpu, uint16_t offReg, uint8_t uVector) RT_NOEXCEPT { /* Determine the vector offset within the chunk. */ uint16_t const offVector = (uVector & UINT32_C(0xe0)) >> 1; /* Read the chunk at the offset. */ uint32_t uReg; RTGCPHYS const GCPhysVirtApic = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64AddrVirtApic.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &uReg, GCPhysVirtApic + offReg + offVector, sizeof(uReg)); if (RT_SUCCESS(rc)) { /* Modify the chunk. */ uint16_t const idxVectorBit = uVector & UINT32_C(0x1f); uReg |= RT_BIT(idxVectorBit); /* Write the chunk. */ rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic + offReg + offVector, &uReg, sizeof(uReg)); AssertMsgRC(rc, ("Failed to set vector %#x in 256-bit register at %#x of the virtual-APIC page at %#RGp: %Rrc\n", uVector, offReg, GCPhysVirtApic, rc)); } else AssertMsgFailed(("Failed to get vector %#x in 256-bit register at %#x of the virtual-APIC page at %#RGp: %Rrc\n", uVector, offReg, GCPhysVirtApic, rc)); } /** * Clears the vector in a virtual-APIC 256-bit sparse register. * * @param pVCpu The cross context virtual CPU structure. * @param offReg The offset of the 256-bit spare register. * @param uVector The vector to clear. * * @remarks This is based on our APIC device code. */ static void iemVmxVirtApicClearVectorInReg(PVMCPUCC pVCpu, uint16_t offReg, uint8_t uVector) RT_NOEXCEPT { /* Determine the vector offset within the chunk. */ uint16_t const offVector = (uVector & UINT32_C(0xe0)) >> 1; /* Read the chunk at the offset. */ uint32_t uReg; RTGCPHYS const GCPhysVirtApic = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64AddrVirtApic.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &uReg, GCPhysVirtApic + offReg + offVector, sizeof(uReg)); if (RT_SUCCESS(rc)) { /* Modify the chunk. */ uint16_t const idxVectorBit = uVector & UINT32_C(0x1f); uReg &= ~RT_BIT(idxVectorBit); /* Write the chunk. */ rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic + offReg + offVector, &uReg, sizeof(uReg)); AssertMsgRC(rc, ("Failed to clear vector %#x in 256-bit register at %#x of the virtual-APIC page at %#RGp: %Rrc\n", uVector, offReg, GCPhysVirtApic, rc)); } else AssertMsgFailed(("Failed to get vector %#x in 256-bit register at %#x of the virtual-APIC page at %#RGp: %Rrc\n", uVector, offReg, GCPhysVirtApic, rc)); } /** * Checks if a memory access to the APIC-access page must causes an APIC-access * VM-exit. * * @param pVCpu The cross context virtual CPU structure. * @param offAccess The offset of the register being accessed. * @param cbAccess The size of the access in bytes. * @param fAccess The type of access, see IEM_ACCESS_XXX. * * @remarks This must not be used for MSR-based APIC-access page accesses! * @sa iemVmxVirtApicAccessMsrWrite, iemVmxVirtApicAccessMsrRead. */ static bool iemVmxVirtApicIsMemAccessIntercepted(PVMCPUCC pVCpu, uint16_t offAccess, size_t cbAccess, uint32_t fAccess) RT_NOEXCEPT { Assert(cbAccess > 0); PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* * We must cause a VM-exit if any of the following are true: * - TPR shadowing isn't active. * - The access size exceeds 32-bits. * - The access is not contained within low 4 bytes of a 16-byte aligned offset. * * See Intel spec. 29.4.2 "Virtualizing Reads from the APIC-Access Page". * See Intel spec. 29.4.3.1 "Determining Whether a Write Access is Virtualized". */ if ( !(pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW) || cbAccess > sizeof(uint32_t) || ((offAccess + cbAccess - 1) & 0xc) || offAccess >= XAPIC_OFF_END + 4) return true; /* * If the access is part of an operation where we have already * virtualized a virtual-APIC write, we must cause a VM-exit. */ if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE)) return true; /* * Check write accesses to the APIC-access page that cause VM-exits. */ if (fAccess & IEM_ACCESS_TYPE_WRITE) { if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT) { /* * With APIC-register virtualization, a write access to any of the * following registers are virtualized. Accessing any other register * causes a VM-exit. */ uint16_t const offAlignedAccess = offAccess & 0xfffc; switch (offAlignedAccess) { case XAPIC_OFF_ID: case XAPIC_OFF_TPR: case XAPIC_OFF_EOI: case XAPIC_OFF_LDR: case XAPIC_OFF_DFR: case XAPIC_OFF_SVR: case XAPIC_OFF_ESR: case XAPIC_OFF_ICR_LO: case XAPIC_OFF_ICR_HI: case XAPIC_OFF_LVT_TIMER: case XAPIC_OFF_LVT_THERMAL: case XAPIC_OFF_LVT_PERF: case XAPIC_OFF_LVT_LINT0: case XAPIC_OFF_LVT_LINT1: case XAPIC_OFF_LVT_ERROR: case XAPIC_OFF_TIMER_ICR: case XAPIC_OFF_TIMER_DCR: break; default: return true; } } else if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY) { /* * With virtual-interrupt delivery, a write access to any of the * following registers are virtualized. Accessing any other register * causes a VM-exit. * * Note! The specification does not allow writing to offsets in-between * these registers (e.g. TPR + 1 byte) unlike read accesses. */ switch (offAccess) { case XAPIC_OFF_TPR: case XAPIC_OFF_EOI: case XAPIC_OFF_ICR_LO: break; default: return true; } } else { /* * Without APIC-register virtualization or virtual-interrupt delivery, * only TPR accesses are virtualized. */ if (offAccess == XAPIC_OFF_TPR) { /* likely */ } else return true; } } else { /* * Check read accesses to the APIC-access page that cause VM-exits. */ if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT) { /* * With APIC-register virtualization, a read access to any of the * following registers are virtualized. Accessing any other register * causes a VM-exit. */ uint16_t const offAlignedAccess = offAccess & 0xfffc; switch (offAlignedAccess) { /** @todo r=ramshankar: What about XAPIC_OFF_LVT_CMCI? */ case XAPIC_OFF_ID: case XAPIC_OFF_VERSION: case XAPIC_OFF_TPR: case XAPIC_OFF_EOI: case XAPIC_OFF_LDR: case XAPIC_OFF_DFR: case XAPIC_OFF_SVR: case XAPIC_OFF_ISR0: case XAPIC_OFF_ISR1: case XAPIC_OFF_ISR2: case XAPIC_OFF_ISR3: case XAPIC_OFF_ISR4: case XAPIC_OFF_ISR5: case XAPIC_OFF_ISR6: case XAPIC_OFF_ISR7: case XAPIC_OFF_TMR0: case XAPIC_OFF_TMR1: case XAPIC_OFF_TMR2: case XAPIC_OFF_TMR3: case XAPIC_OFF_TMR4: case XAPIC_OFF_TMR5: case XAPIC_OFF_TMR6: case XAPIC_OFF_TMR7: case XAPIC_OFF_IRR0: case XAPIC_OFF_IRR1: case XAPIC_OFF_IRR2: case XAPIC_OFF_IRR3: case XAPIC_OFF_IRR4: case XAPIC_OFF_IRR5: case XAPIC_OFF_IRR6: case XAPIC_OFF_IRR7: case XAPIC_OFF_ESR: case XAPIC_OFF_ICR_LO: case XAPIC_OFF_ICR_HI: case XAPIC_OFF_LVT_TIMER: case XAPIC_OFF_LVT_THERMAL: case XAPIC_OFF_LVT_PERF: case XAPIC_OFF_LVT_LINT0: case XAPIC_OFF_LVT_LINT1: case XAPIC_OFF_LVT_ERROR: case XAPIC_OFF_TIMER_ICR: case XAPIC_OFF_TIMER_DCR: break; default: return true; } } else { /* Without APIC-register virtualization, only TPR accesses are virtualized. */ if (offAccess == XAPIC_OFF_TPR) { /* likely */ } else return true; } } /* The APIC access is virtualized, does not cause a VM-exit. */ return false; } /** * Virtualizes a memory-based APIC access by certain instructions even though they * do not use the address to access memory. * * This is for instructions like MONITOR, CLFLUSH, CLFLUSHOPT, ENTER which may cause * page-faults but do not use the address to access memory. * * @param pVCpu The cross context virtual CPU structure. * @param pGCPhysAccess Pointer to the guest-physical address accessed. * @param cbAccess The size of the access in bytes. * @param fAccess The type of access, see IEM_ACCESS_XXX. */ VBOXSTRICTRC iemVmxVirtApicAccessUnused(PVMCPUCC pVCpu, PRTGCPHYS pGCPhysAccess, size_t cbAccess, uint32_t fAccess) RT_NOEXCEPT { Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS); Assert(pGCPhysAccess); RTGCPHYS const GCPhysAccess = *pGCPhysAccess & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK; RTGCPHYS const GCPhysApic = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u64AddrApicAccess.u; Assert(!(GCPhysApic & GUEST_PAGE_OFFSET_MASK)); if (GCPhysAccess == GCPhysApic) { uint16_t const offAccess = *pGCPhysAccess & GUEST_PAGE_OFFSET_MASK; bool const fIntercept = iemVmxVirtApicIsMemAccessIntercepted(pVCpu, offAccess, cbAccess, fAccess); if (fIntercept) return iemVmxVmexitApicAccess(pVCpu, offAccess, fAccess); *pGCPhysAccess = GCPhysApic | offAccess; return VINF_VMX_MODIFIES_BEHAVIOR; } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * Virtualizes a memory-based APIC access. * * @returns VBox strict status code. * @retval VINF_VMX_MODIFIES_BEHAVIOR if the access was virtualized. * @retval VINF_VMX_VMEXIT if the access causes a VM-exit. * * @param pVCpu The cross context virtual CPU structure. * @param offAccess The offset of the register being accessed (within the * APIC-access page). * @param cbAccess The size of the access in bytes. * @param pvData Pointer to the data being written or where to store the data * being read. * @param fAccess The type of access, see IEM_ACCESS_XXX. */ static VBOXSTRICTRC iemVmxVirtApicAccessMem(PVMCPUCC pVCpu, uint16_t offAccess, size_t cbAccess, void *pvData, uint32_t fAccess) RT_NOEXCEPT { Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS); Assert(pvData); bool const fIntercept = iemVmxVirtApicIsMemAccessIntercepted(pVCpu, offAccess, cbAccess, fAccess); if (fIntercept) return iemVmxVmexitApicAccess(pVCpu, offAccess, fAccess); if (fAccess & IEM_ACCESS_TYPE_WRITE) { /* * A write access to the APIC-access page that is virtualized (rather than * causing a VM-exit) writes data to the virtual-APIC page. */ uint32_t const u32Data = *(uint32_t *)pvData; iemVmxVirtApicWriteRaw32(pVCpu, offAccess, u32Data); /* * Record the currently updated APIC offset, as we need this later for figuring * out whether to perform TPR, EOI or self-IPI virtualization as well as well * as for supplying the exit qualification when causing an APIC-write VM-exit. * * After completion of the current operation, we need to perform TPR virtualization, * EOI virtualization or APIC-write VM-exit depending on which register was written. * * The current operation may be a REP-prefixed string instruction, execution of any * other instruction, or delivery of an event through the IDT. * * Thus things like clearing bytes 3:1 of the VTPR, clearing VEOI are not to be * performed now but later after completion of the current operation. * * See Intel spec. 29.4.3.2 "APIC-Write Emulation". */ iemVmxVirtApicSetPendingWrite(pVCpu, offAccess); LogFlowFunc(("Write access at offset %#x not intercepted -> Wrote %#RX32\n", offAccess, u32Data)); } else { /* * A read access from the APIC-access page that is virtualized (rather than * causing a VM-exit) returns data from the virtual-APIC page. * * See Intel spec. 29.4.2 "Virtualizing Reads from the APIC-Access Page". */ Assert(fAccess & IEM_ACCESS_TYPE_READ); Assert(cbAccess <= 4); Assert(offAccess < XAPIC_OFF_END + 4); static uint32_t const s_auAccessSizeMasks[] = { 0, 0xff, 0xffff, 0xffffff, 0xffffffff }; uint32_t u32Data = iemVmxVirtApicReadRaw32(pVCpu, offAccess); u32Data &= s_auAccessSizeMasks[cbAccess]; *(uint32_t *)pvData = u32Data; LogFlowFunc(("Read access at offset %#x not intercepted -> Read %#RX32\n", offAccess, u32Data)); } return VINF_VMX_MODIFIES_BEHAVIOR; } /** * Virtualizes an MSR-based APIC read access. * * @returns VBox strict status code. * @retval VINF_VMX_MODIFIES_BEHAVIOR if the MSR read was virtualized. * @retval VINF_VMX_INTERCEPT_NOT_ACTIVE if the MSR read access must be * handled by the x2APIC device. * @retval VERR_OUT_RANGE if the MSR read was supposed to be virtualized but was * not within the range of valid MSRs, caller must raise \#GP(0). * @param pVCpu The cross context virtual CPU structure. * @param idMsr The x2APIC MSR being read. * @param pu64Value Where to store the read x2APIC MSR value (only valid when * VINF_VMX_MODIFIES_BEHAVIOR is returned). */ static VBOXSTRICTRC iemVmxVirtApicAccessMsrRead(PVMCPUCC pVCpu, uint32_t idMsr, uint64_t *pu64Value) RT_NOEXCEPT { Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_X2APIC_MODE); Assert(pu64Value); if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT) { if ( idMsr >= MSR_IA32_X2APIC_START && idMsr <= MSR_IA32_X2APIC_END) { uint16_t const offReg = (idMsr & 0xff) << 4; uint64_t const u64Value = iemVmxVirtApicReadRaw64(pVCpu, offReg); *pu64Value = u64Value; return VINF_VMX_MODIFIES_BEHAVIOR; } return VERR_OUT_OF_RANGE; } if (idMsr == MSR_IA32_X2APIC_TPR) { uint16_t const offReg = (idMsr & 0xff) << 4; uint64_t const u64Value = iemVmxVirtApicReadRaw64(pVCpu, offReg); *pu64Value = u64Value; return VINF_VMX_MODIFIES_BEHAVIOR; } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * Virtualizes an MSR-based APIC write access. * * @returns VBox strict status code. * @retval VINF_VMX_MODIFIES_BEHAVIOR if the MSR write was virtualized. * @retval VERR_OUT_RANGE if the MSR read was supposed to be virtualized but was * not within the range of valid MSRs, caller must raise \#GP(0). * @retval VINF_VMX_INTERCEPT_NOT_ACTIVE if the MSR must be written normally. * * @param pVCpu The cross context virtual CPU structure. * @param idMsr The x2APIC MSR being written. * @param u64Value The value of the x2APIC MSR being written. */ static VBOXSTRICTRC iemVmxVirtApicAccessMsrWrite(PVMCPUCC pVCpu, uint32_t idMsr, uint64_t u64Value) RT_NOEXCEPT { /* * Check if the access is to be virtualized. * See Intel spec. 29.5 "Virtualizing MSR-based APIC Accesses". */ if ( idMsr == MSR_IA32_X2APIC_TPR || ( (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY) && ( idMsr == MSR_IA32_X2APIC_EOI || idMsr == MSR_IA32_X2APIC_SELF_IPI))) { /* Validate the MSR write depending on the register. */ switch (idMsr) { case MSR_IA32_X2APIC_TPR: case MSR_IA32_X2APIC_SELF_IPI: { if (u64Value & UINT64_C(0xffffffffffffff00)) return VERR_OUT_OF_RANGE; break; } case MSR_IA32_X2APIC_EOI: { if (u64Value != 0) return VERR_OUT_OF_RANGE; break; } } /* Write the MSR to the virtual-APIC page. */ uint16_t const offReg = (idMsr & 0xff) << 4; iemVmxVirtApicWriteRaw64(pVCpu, offReg, u64Value); /* * Record the currently updated APIC offset, as we need this later for figuring * out whether to perform TPR, EOI or self-IPI virtualization as well as well * as for supplying the exit qualification when causing an APIC-write VM-exit. */ iemVmxVirtApicSetPendingWrite(pVCpu, offReg); return VINF_VMX_MODIFIES_BEHAVIOR; } return VINF_VMX_INTERCEPT_NOT_ACTIVE; } /** * Interface for HM and EM to virtualize x2APIC MSR accesses. * * @returns Strict VBox status code. * @retval VINF_VMX_MODIFIES_BEHAVIOR if the MSR access was virtualized. * @retval VINF_VMX_INTERCEPT_NOT_ACTIVE if the MSR access must be handled by * the x2APIC device. * @retval VERR_OUT_RANGE if the caller must raise \#GP(0). * * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param idMsr The MSR being read. * @param pu64Value Pointer to the value being written or where to store the * value being read. * @param fWrite Whether this is an MSR write or read access. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVirtApicAccessMsr(PVMCPUCC pVCpu, uint32_t idMsr, uint64_t *pu64Value, bool fWrite) { Assert(pu64Value); VBOXSTRICTRC rcStrict; if (fWrite) rcStrict = iemVmxVirtApicAccessMsrWrite(pVCpu, idMsr, *pu64Value); else rcStrict = iemVmxVirtApicAccessMsrRead(pVCpu, idMsr, pu64Value); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * Finds the most significant set bit in a virtual-APIC 256-bit sparse register. * * @returns VBox status code. * @retval VINF_SUCCESS when the highest set bit is found. * @retval VERR_NOT_FOUND when no bit is set. * * @param pVCpu The cross context virtual CPU structure. * @param offReg The offset of the APIC 256-bit sparse register. * @param pidxHighestBit Where to store the highest bit (most significant bit) * set in the register. Only valid when VINF_SUCCESS is * returned. * * @remarks The format of the 256-bit sparse register here mirrors that found in * real APIC hardware. */ static int iemVmxVirtApicGetHighestSetBitInReg(PVMCPUCC pVCpu, uint16_t offReg, uint8_t *pidxHighestBit) { Assert(offReg < XAPIC_OFF_END + 4); Assert(pidxHighestBit); /* * There are 8 contiguous fragments (of 16-bytes each) in the sparse register. * However, in each fragment only the first 4 bytes are used. */ uint8_t const cFrags = 8; for (int8_t iFrag = cFrags; iFrag >= 0; iFrag--) { uint16_t const offFrag = iFrag * 16; uint32_t const u32Frag = iemVmxVirtApicReadRaw32(pVCpu, offReg + offFrag); if (!u32Frag) continue; unsigned idxHighestBit = ASMBitLastSetU32(u32Frag); Assert(idxHighestBit > 0); --idxHighestBit; Assert(idxHighestBit <= UINT8_MAX); *pidxHighestBit = idxHighestBit; return VINF_SUCCESS; } return VERR_NOT_FOUND; } /** * Evaluates pending virtual interrupts. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxEvalPendingVirtIntrs(PVMCPUCC pVCpu) RT_NOEXCEPT { Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY); if (!(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT)) { uint8_t const uRvi = RT_LO_U8(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u16GuestIntStatus); uint8_t const uPpr = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_PPR); if ((uRvi >> 4) > (uPpr >> 4)) { Log2(("eval_virt_intrs: uRvi=%#x uPpr=%#x - Signalling pending interrupt\n", uRvi, uPpr)); VMCPU_FF_SET(pVCpu, VMCPU_FF_INTERRUPT_NESTED_GUEST); } else Log2(("eval_virt_intrs: uRvi=%#x uPpr=%#x - Nothing to do\n", uRvi, uPpr)); } } /** * Performs PPR virtualization. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxPprVirtualization(PVMCPUCC pVCpu) RT_NOEXCEPT { Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW); Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY); /* * PPR virtualization is caused in response to a VM-entry, TPR-virtualization, * or EOI-virtualization. * * See Intel spec. 29.1.3 "PPR Virtualization". */ uint8_t const uTpr = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_TPR); uint8_t const uSvi = RT_HI_U8(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u16GuestIntStatus) & 0xf0; uint32_t uPpr; if ((uTpr & 0xf0) >= uSvi) uPpr = uTpr; else uPpr = uSvi; Log2(("ppr_virt: uTpr=%#x uSvi=%#x uPpr=%#x\n", uTpr, uSvi, uPpr)); iemVmxVirtApicWriteRaw32(pVCpu, XAPIC_OFF_PPR, uPpr); } /** * Performs VMX TPR virtualization. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. */ static VBOXSTRICTRC iemVmxTprVirtualization(PVMCPUCC pVCpu) RT_NOEXCEPT { Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW); /* * We should have already performed the virtual-APIC write to the TPR offset * in the virtual-APIC page. We now perform TPR virtualization. * * See Intel spec. 29.1.2 "TPR Virtualization". */ if (!(pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)) { uint32_t const uTprThreshold = pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32TprThreshold; uint32_t const uTpr = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_TPR); /* * If the VTPR falls below the TPR threshold, we must cause a VM-exit. * See Intel spec. 29.1.2 "TPR Virtualization". */ if (((uTpr >> 4) & 0xf) < uTprThreshold) { Log2(("tpr_virt: uTpr=%u uTprThreshold=%u -> VM-exit\n", uTpr, uTprThreshold)); return iemVmxVmexit(pVCpu, VMX_EXIT_TPR_BELOW_THRESHOLD, 0 /* u64ExitQual */); } } else { iemVmxPprVirtualization(pVCpu); iemVmxEvalPendingVirtIntrs(pVCpu); } return VINF_SUCCESS; } /** * Checks whether an EOI write for the given interrupt vector causes a VM-exit or * not. * * @returns @c true if the EOI write is intercepted, @c false otherwise. * @param pVCpu The cross context virtual CPU structure. * @param uVector The interrupt that was acknowledged using an EOI. */ static bool iemVmxIsEoiInterceptSet(PCVMCPU pVCpu, uint8_t uVector) RT_NOEXCEPT { PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY); if (uVector < 64) return RT_BOOL(pVmcs->u64EoiExitBitmap0.u & RT_BIT_64(uVector)); if (uVector < 128) return RT_BOOL(pVmcs->u64EoiExitBitmap1.u & RT_BIT_64(uVector)); if (uVector < 192) return RT_BOOL(pVmcs->u64EoiExitBitmap2.u & RT_BIT_64(uVector)); return RT_BOOL(pVmcs->u64EoiExitBitmap3.u & RT_BIT_64(uVector)); } /** * Performs EOI virtualization. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. */ static VBOXSTRICTRC iemVmxEoiVirtualization(PVMCPUCC pVCpu) RT_NOEXCEPT { PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY); /* * Clear the interrupt guest-interrupt as no longer in-service (ISR) * and get the next guest-interrupt that's in-service (if any). * * See Intel spec. 29.1.4 "EOI Virtualization". */ uint8_t const uRvi = RT_LO_U8(pVmcs->u16GuestIntStatus); uint8_t const uSvi = RT_HI_U8(pVmcs->u16GuestIntStatus); Log2(("eoi_virt: uRvi=%#x uSvi=%#x\n", uRvi, uSvi)); uint8_t uVector = uSvi; iemVmxVirtApicClearVectorInReg(pVCpu, XAPIC_OFF_ISR0, uVector); uVector = 0; iemVmxVirtApicGetHighestSetBitInReg(pVCpu, XAPIC_OFF_ISR0, &uVector); if (uVector) Log2(("eoi_virt: next interrupt %#x\n", uVector)); else Log2(("eoi_virt: no interrupt pending in ISR\n")); /* Update guest-interrupt status SVI (leave RVI portion as it is) in the VMCS. */ pVmcs->u16GuestIntStatus = RT_MAKE_U16(uRvi, uVector); iemVmxPprVirtualization(pVCpu); if (iemVmxIsEoiInterceptSet(pVCpu, uVector)) return iemVmxVmexit(pVCpu, VMX_EXIT_VIRTUALIZED_EOI, uVector); iemVmxEvalPendingVirtIntrs(pVCpu); return VINF_SUCCESS; } /** * Performs self-IPI virtualization. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. */ static VBOXSTRICTRC iemVmxSelfIpiVirtualization(PVMCPUCC pVCpu) RT_NOEXCEPT { PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW); /* * We should have already performed the virtual-APIC write to the self-IPI offset * in the virtual-APIC page. We now perform self-IPI virtualization. * * See Intel spec. 29.1.5 "Self-IPI Virtualization". */ uint8_t const uVector = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_ICR_LO); Log2(("self_ipi_virt: uVector=%#x\n", uVector)); iemVmxVirtApicSetVectorInReg(pVCpu, XAPIC_OFF_IRR0, uVector); uint8_t const uRvi = RT_LO_U8(pVmcs->u16GuestIntStatus); uint8_t const uSvi = RT_HI_U8(pVmcs->u16GuestIntStatus); if (uVector > uRvi) pVmcs->u16GuestIntStatus = RT_MAKE_U16(uVector, uSvi); iemVmxEvalPendingVirtIntrs(pVCpu); return VINF_SUCCESS; } /** * Performs VMX APIC-write emulation. * * @returns VBox strict status code. * @param pVCpu The cross context virtual CPU structure. */ VBOXSTRICTRC iemVmxApicWriteEmulation(PVMCPUCC pVCpu) RT_NOEXCEPT { PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* Import the virtual-APIC write offset (part of the hardware-virtualization state). */ IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_HWVIRT); /* * Perform APIC-write emulation based on the virtual-APIC register written. * See Intel spec. 29.4.3.2 "APIC-Write Emulation". */ uint16_t const offApicWrite = iemVmxVirtApicClearPendingWrite(pVCpu); VBOXSTRICTRC rcStrict; switch (offApicWrite) { case XAPIC_OFF_TPR: { /* Clear bytes 3:1 of the VTPR and perform TPR virtualization. */ uint32_t uTpr = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_TPR); uTpr &= UINT32_C(0x000000ff); iemVmxVirtApicWriteRaw32(pVCpu, XAPIC_OFF_TPR, uTpr); Log2(("iemVmxApicWriteEmulation: TPR write %#x\n", uTpr)); rcStrict = iemVmxTprVirtualization(pVCpu); break; } case XAPIC_OFF_EOI: { if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY) { /* Clear VEOI and perform EOI virtualization. */ iemVmxVirtApicWriteRaw32(pVCpu, XAPIC_OFF_EOI, 0); Log2(("iemVmxApicWriteEmulation: EOI write\n")); rcStrict = iemVmxEoiVirtualization(pVCpu); } else rcStrict = iemVmxVmexitApicWrite(pVCpu, offApicWrite); break; } case XAPIC_OFF_ICR_LO: { if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY) { /* If the ICR_LO is valid, write it and perform self-IPI virtualization. */ uint32_t const uIcrLo = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_TPR); uint32_t const fIcrLoMb0 = UINT32_C(0xfffbb700); uint32_t const fIcrLoMb1 = UINT32_C(0x000000f0); if ( !(uIcrLo & fIcrLoMb0) && (uIcrLo & fIcrLoMb1)) { Log2(("iemVmxApicWriteEmulation: Self-IPI virtualization with vector %#x\n", (uIcrLo & 0xff))); rcStrict = iemVmxSelfIpiVirtualization(pVCpu); } else rcStrict = iemVmxVmexitApicWrite(pVCpu, offApicWrite); } else rcStrict = iemVmxVmexitApicWrite(pVCpu, offApicWrite); break; } case XAPIC_OFF_ICR_HI: { /* Clear bytes 2:0 of VICR_HI. No other virtualization or VM-exit must occur. */ uint32_t uIcrHi = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_ICR_HI); uIcrHi &= UINT32_C(0xff000000); iemVmxVirtApicWriteRaw32(pVCpu, XAPIC_OFF_ICR_HI, uIcrHi); rcStrict = VINF_SUCCESS; break; } default: { /* Writes to any other virtual-APIC register causes an APIC-write VM-exit. */ rcStrict = iemVmxVmexitApicWrite(pVCpu, offApicWrite); break; } } return rcStrict; } /** * Interface for HM and EM to perform an APIC-write emulation which may cause a * VM-exit. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecVmxVmexitApicWrite(PVMCPUCC pVCpu) { VBOXSTRICTRC rcStrict = iemVmxApicWriteEmulation(pVCpu); Assert(!pVCpu->iem.s.cActiveMappings); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /** * Checks guest control registers, debug registers and MSRs as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ DECLINLINE(int) iemVmxVmentryCheckGuestControlRegsMsrs(PVMCPUCC pVCpu, const char *pszInstr) { /* * Guest Control Registers, Debug Registers, and MSRs. * See Intel spec. 26.3.1.1 "Checks on Guest Control Registers, Debug Registers, and MSRs". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char * const pszFailure = "VM-exit"; bool const fUnrestrictedGuest = RT_BOOL(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST); /* CR0 reserved bits. */ { /* CR0 MB1 bits. */ uint64_t const u64Cr0Fixed0 = iemVmxGetCr0Fixed0(pVCpu, true /* fVmxNonRootMode */); if ((pVmcs->u64GuestCr0.u & u64Cr0Fixed0) == u64Cr0Fixed0) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr0Fixed0); /* CR0 MBZ bits. */ uint64_t const u64Cr0Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed1; if (!(pVmcs->u64GuestCr0.u & ~u64Cr0Fixed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr0Fixed1); /* Without unrestricted guest support, VT-x supports does not support unpaged protected mode. */ if ( !fUnrestrictedGuest && (pVmcs->u64GuestCr0.u & X86_CR0_PG) && !(pVmcs->u64GuestCr0.u & X86_CR0_PE)) IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr0PgPe); } /* CR4 reserved bits. */ { /* CR4 MB1 bits. */ uint64_t const u64Cr4Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed0; if ((pVmcs->u64GuestCr4.u & u64Cr4Fixed0) == u64Cr4Fixed0) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr4Fixed0); /* CR4 MBZ bits. */ uint64_t const u64Cr4Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed1; if (!(pVmcs->u64GuestCr4.u & ~u64Cr4Fixed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr4Fixed1); } /* DEBUGCTL MSR. */ if ( !(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG) || !(pVmcs->u64GuestDebugCtlMsr.u & ~MSR_IA32_DEBUGCTL_VALID_MASK_INTEL)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestDebugCtl); /* 64-bit CPU checks. */ bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST); if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { if (fGstInLongMode) { /* PAE must be set. */ if ( (pVmcs->u64GuestCr0.u & X86_CR0_PG) && (pVmcs->u64GuestCr0.u & X86_CR4_PAE)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPae); } else { /* PCIDE should not be set. */ if (!(pVmcs->u64GuestCr4.u & X86_CR4_PCIDE)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPcide); } /* CR3. */ if (!(pVmcs->u64GuestCr3.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxPhysAddrWidth)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr3); /* DR7. */ if ( !(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG) || !(pVmcs->u64GuestDr7.u & X86_DR7_MBZ_MASK)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestDr7); /* SYSENTER ESP and SYSENTER EIP. */ if ( X86_IS_CANONICAL(pVmcs->u64GuestSysenterEsp.u) && X86_IS_CANONICAL(pVmcs->u64GuestSysenterEip.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSysenterEspEip); } /* We don't support IA32_PERF_GLOBAL_CTRL MSR yet. */ Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR)); /* PAT MSR. */ if ( !(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR) || CPUMIsPatMsrValid(pVmcs->u64GuestPatMsr.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPatMsr); /* EFER MSR. */ if (pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR) { uint64_t const uValidEferMask = CPUMGetGuestEferMsrValidMask(pVCpu->CTX_SUFF(pVM)); if (!(pVmcs->u64GuestEferMsr.u & ~uValidEferMask)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestEferMsrRsvd); bool const fGstLma = RT_BOOL(pVmcs->u64GuestEferMsr.u & MSR_K6_EFER_LMA); bool const fGstLme = RT_BOOL(pVmcs->u64GuestEferMsr.u & MSR_K6_EFER_LME); if ( fGstLma == fGstInLongMode && ( !(pVmcs->u64GuestCr0.u & X86_CR0_PG) || fGstLma == fGstLme)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestEferMsr); } /* We don't support IA32_BNDCFGS MSR yet. */ Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_BNDCFGS_MSR)); NOREF(pszInstr); NOREF(pszFailure); return VINF_SUCCESS; } /** * Checks guest segment registers, LDTR and TR as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ DECLINLINE(int) iemVmxVmentryCheckGuestSegRegs(PVMCPUCC pVCpu, const char *pszInstr) { /* * Segment registers. * See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char * const pszFailure = "VM-exit"; bool const fGstInV86Mode = RT_BOOL(pVmcs->u64GuestRFlags.u & X86_EFL_VM); bool const fUnrestrictedGuest = RT_BOOL(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST); bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST); /* Selectors. */ if ( !fGstInV86Mode && !fUnrestrictedGuest && (pVmcs->GuestSs & X86_SEL_RPL) != (pVmcs->GuestCs & X86_SEL_RPL)) IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegSelCsSsRpl); for (unsigned iSegReg = 0; iSegReg < X86_SREG_COUNT; iSegReg++) { CPUMSELREG SelReg; int rc = iemVmxVmcsGetGuestSegReg(pVmcs, iSegReg, &SelReg); if (RT_LIKELY(rc == VINF_SUCCESS)) { /* likely */ } else return rc; /* * Virtual-8086 mode checks. */ if (fGstInV86Mode) { /* Base address. */ if (SelReg.u64Base == (uint64_t)SelReg.Sel << 4) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegBaseV86(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } /* Limit. */ if (SelReg.u32Limit == 0xffff) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegLimitV86(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } /* Attribute. */ if (SelReg.Attr.u == 0xf3) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrV86(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } /* We're done; move to checking the next segment. */ continue; } /* Checks done by 64-bit CPUs. */ if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { /* Base address. */ if ( iSegReg == X86_SREG_FS || iSegReg == X86_SREG_GS) { if (X86_IS_CANONICAL(SelReg.u64Base)) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegBase(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } } else if (iSegReg == X86_SREG_CS) { if (!RT_HI_U32(SelReg.u64Base)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegBaseCs); } else { if ( SelReg.Attr.n.u1Unusable || !RT_HI_U32(SelReg.u64Base)) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegBase(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } } } /* * Checks outside Virtual-8086 mode. */ uint8_t const uSegType = SelReg.Attr.n.u4Type; uint8_t const fCodeDataSeg = SelReg.Attr.n.u1DescType; uint8_t const fUsable = !SelReg.Attr.n.u1Unusable; uint8_t const uDpl = SelReg.Attr.n.u2Dpl; uint8_t const fPresent = SelReg.Attr.n.u1Present; uint8_t const uGranularity = SelReg.Attr.n.u1Granularity; uint8_t const uDefBig = SelReg.Attr.n.u1DefBig; uint8_t const fSegLong = SelReg.Attr.n.u1Long; /* Code or usable segment. */ if ( iSegReg == X86_SREG_CS || fUsable) { /* Reserved bits (bits 31:17 and bits 11:8). */ if (!(SelReg.Attr.u & 0xfffe0f00)) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrRsvd(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } /* Descriptor type. */ if (fCodeDataSeg) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrDescType(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } /* Present. */ if (fPresent) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrPresent(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } /* Granularity. */ if ( ((SelReg.u32Limit & 0x00000fff) == 0x00000fff || !uGranularity) && ((SelReg.u32Limit & 0xfff00000) == 0x00000000 || uGranularity)) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrGran(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } } if (iSegReg == X86_SREG_CS) { /* Segment Type and DPL. */ if ( uSegType == (X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED) && fUnrestrictedGuest) { if (uDpl == 0) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsDplZero); } else if ( uSegType == (X86_SEL_TYPE_CODE | X86_SEL_TYPE_ACCESSED) || uSegType == (X86_SEL_TYPE_CODE | X86_SEL_TYPE_READ | X86_SEL_TYPE_ACCESSED)) { X86DESCATTR AttrSs; AttrSs.u = pVmcs->u32GuestSsAttr; if (uDpl == AttrSs.n.u2Dpl) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsDplEqSs); } else if ((uSegType & (X86_SEL_TYPE_CODE | X86_SEL_TYPE_CONF | X86_SEL_TYPE_ACCESSED)) == (X86_SEL_TYPE_CODE | X86_SEL_TYPE_CONF | X86_SEL_TYPE_ACCESSED)) { X86DESCATTR AttrSs; AttrSs.u = pVmcs->u32GuestSsAttr; if (uDpl <= AttrSs.n.u2Dpl) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsDplLtSs); } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsType); /* Def/Big. */ if ( fGstInLongMode && fSegLong) { if (uDefBig == 0) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsDefBig); } } else if (iSegReg == X86_SREG_SS) { /* Segment Type. */ if ( !fUsable || uSegType == (X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED) || uSegType == (X86_SEL_TYPE_DOWN | X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrSsType); /* DPL. */ if (!fUnrestrictedGuest) { if (uDpl == (SelReg.Sel & X86_SEL_RPL)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrSsDplEqRpl); } X86DESCATTR AttrCs; AttrCs.u = pVmcs->u32GuestCsAttr; if ( AttrCs.n.u4Type == (X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED) || !(pVmcs->u64GuestCr0.u & X86_CR0_PE)) { if (uDpl == 0) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrSsDplZero); } } else { /* DS, ES, FS, GS. */ if (fUsable) { /* Segment type. */ if (uSegType & X86_SEL_TYPE_ACCESSED) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrTypeAcc(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } if ( !(uSegType & X86_SEL_TYPE_CODE) || (uSegType & X86_SEL_TYPE_READ)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsTypeRead); /* DPL. */ if ( !fUnrestrictedGuest && uSegType <= (X86_SEL_TYPE_CODE | X86_SEL_TYPE_READ | X86_SEL_TYPE_ACCESSED)) { if (uDpl >= (SelReg.Sel & X86_SEL_RPL)) { /* likely */ } else { VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrDplRpl(iSegReg); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } } } } } /* * LDTR. */ { CPUMSELREG Ldtr; Ldtr.Sel = pVmcs->GuestLdtr; Ldtr.u32Limit = pVmcs->u32GuestLdtrLimit; Ldtr.u64Base = pVmcs->u64GuestLdtrBase.u; Ldtr.Attr.u = pVmcs->u32GuestLdtrAttr; if (!Ldtr.Attr.n.u1Unusable) { /* Selector. */ if (!(Ldtr.Sel & X86_SEL_LDT)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegSelLdtr); /* Base. */ if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { if (X86_IS_CANONICAL(Ldtr.u64Base)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegBaseLdtr); } /* Attributes. */ /* Reserved bits (bits 31:17 and bits 11:8). */ if (!(Ldtr.Attr.u & 0xfffe0f00)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrRsvd); if (Ldtr.Attr.n.u4Type == X86_SEL_TYPE_SYS_LDT) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrType); if (!Ldtr.Attr.n.u1DescType) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrDescType); if (Ldtr.Attr.n.u1Present) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrPresent); if ( ((Ldtr.u32Limit & 0x00000fff) == 0x00000fff || !Ldtr.Attr.n.u1Granularity) && ((Ldtr.u32Limit & 0xfff00000) == 0x00000000 || Ldtr.Attr.n.u1Granularity)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrGran); } } /* * TR. */ { CPUMSELREG Tr; Tr.Sel = pVmcs->GuestTr; Tr.u32Limit = pVmcs->u32GuestTrLimit; Tr.u64Base = pVmcs->u64GuestTrBase.u; Tr.Attr.u = pVmcs->u32GuestTrAttr; /* Selector. */ if (!(Tr.Sel & X86_SEL_LDT)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegSelTr); /* Base. */ if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { if (X86_IS_CANONICAL(Tr.u64Base)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegBaseTr); } /* Attributes. */ /* Reserved bits (bits 31:17 and bits 11:8). */ if (!(Tr.Attr.u & 0xfffe0f00)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrRsvd); if (!Tr.Attr.n.u1Unusable) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrUnusable); if ( Tr.Attr.n.u4Type == X86_SEL_TYPE_SYS_386_TSS_BUSY || ( !fGstInLongMode && Tr.Attr.n.u4Type == X86_SEL_TYPE_SYS_286_TSS_BUSY)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrType); if (!Tr.Attr.n.u1DescType) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrDescType); if (Tr.Attr.n.u1Present) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrPresent); if ( ((Tr.u32Limit & 0x00000fff) == 0x00000fff || !Tr.Attr.n.u1Granularity) && ((Tr.u32Limit & 0xfff00000) == 0x00000000 || Tr.Attr.n.u1Granularity)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrGran); } NOREF(pszInstr); NOREF(pszFailure); return VINF_SUCCESS; } /** * Checks guest GDTR and IDTR as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ DECLINLINE(int) iemVmxVmentryCheckGuestGdtrIdtr(PVMCPUCC pVCpu, const char *pszInstr) { /* * GDTR and IDTR. * See Intel spec. 26.3.1.3 "Checks on Guest Descriptor-Table Registers". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char *const pszFailure = "VM-exit"; if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { /* Base. */ if (X86_IS_CANONICAL(pVmcs->u64GuestGdtrBase.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestGdtrBase); if (X86_IS_CANONICAL(pVmcs->u64GuestIdtrBase.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIdtrBase); } /* Limit. */ if (!RT_HI_U16(pVmcs->u32GuestGdtrLimit)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestGdtrLimit); if (!RT_HI_U16(pVmcs->u32GuestIdtrLimit)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIdtrLimit); NOREF(pszInstr); NOREF(pszFailure); return VINF_SUCCESS; } /** * Checks guest RIP and RFLAGS as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ DECLINLINE(int) iemVmxVmentryCheckGuestRipRFlags(PVMCPUCC pVCpu, const char *pszInstr) { /* * RIP and RFLAGS. * See Intel spec. 26.3.1.4 "Checks on Guest RIP and RFLAGS". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char *const pszFailure = "VM-exit"; bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST); /* RIP. */ if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { X86DESCATTR AttrCs; AttrCs.u = pVmcs->u32GuestCsAttr; if ( !fGstInLongMode || !AttrCs.n.u1Long) { if (!RT_HI_U32(pVmcs->u64GuestRip.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRipRsvd); } if ( fGstInLongMode && AttrCs.n.u1Long) { Assert(IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxLinearAddrWidth == 48); /* Canonical. */ if ( IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxLinearAddrWidth < 64 && X86_IS_CANONICAL(pVmcs->u64GuestRip.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRip); } } /* RFLAGS (bits 63:22 (or 31:22), bits 15, 5, 3 are reserved, bit 1 MB1). */ uint64_t const uGuestRFlags = IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode ? pVmcs->u64GuestRFlags.u : pVmcs->u64GuestRFlags.s.Lo; if ( !(uGuestRFlags & ~(X86_EFL_LIVE_MASK | X86_EFL_RA1_MASK)) && (uGuestRFlags & X86_EFL_RA1_MASK) == X86_EFL_RA1_MASK) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRFlagsRsvd); if (!(uGuestRFlags & X86_EFL_VM)) { /* likely */ } else { if ( fGstInLongMode || !(pVmcs->u64GuestCr0.u & X86_CR0_PE)) IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRFlagsVm); } if (VMX_ENTRY_INT_INFO_IS_EXT_INT(pVmcs->u32EntryIntInfo)) { if (uGuestRFlags & X86_EFL_IF) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRFlagsIf); } NOREF(pszInstr); NOREF(pszFailure); return VINF_SUCCESS; } /** * Checks guest non-register state as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ DECLINLINE(int) iemVmxVmentryCheckGuestNonRegState(PVMCPUCC pVCpu, const char *pszInstr) { /* * Guest non-register state. * See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State". */ PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char *const pszFailure = "VM-exit"; /* * Activity state. */ uint64_t const u64GuestVmxMiscMsr = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Misc; uint32_t const fActivityStateMask = RT_BF_GET(u64GuestVmxMiscMsr, VMX_BF_MISC_ACTIVITY_STATES); if (!(pVmcs->u32GuestActivityState & fActivityStateMask)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateRsvd); X86DESCATTR AttrSs; AttrSs.u = pVmcs->u32GuestSsAttr; if ( !AttrSs.n.u2Dpl || pVmcs->u32GuestActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateSsDpl); if ( pVmcs->u32GuestIntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_STI || pVmcs->u32GuestIntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS) { if (pVmcs->u32GuestActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateStiMovSs); } if (VMX_ENTRY_INT_INFO_IS_VALID(pVmcs->u32EntryIntInfo)) { uint8_t const uType = VMX_ENTRY_INT_INFO_TYPE(pVmcs->u32EntryIntInfo); uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(pVmcs->u32EntryIntInfo); AssertCompile(VMX_V_GUEST_ACTIVITY_STATE_MASK == (VMX_VMCS_GUEST_ACTIVITY_HLT | VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN)); switch (pVmcs->u32GuestActivityState) { case VMX_VMCS_GUEST_ACTIVITY_HLT: { if ( uType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT || uType == VMX_ENTRY_INT_INFO_TYPE_NMI || ( uType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT && ( uVector == X86_XCPT_DB || uVector == X86_XCPT_MC)) || ( uType == VMX_ENTRY_INT_INFO_TYPE_OTHER_EVENT && uVector == VMX_ENTRY_INT_INFO_VECTOR_MTF)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateHlt); break; } case VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN: { if ( uType == VMX_ENTRY_INT_INFO_TYPE_NMI || ( uType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT && uVector == X86_XCPT_MC)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateShutdown); break; } case VMX_VMCS_GUEST_ACTIVITY_ACTIVE: default: break; } } /* * Interruptibility state. */ if (!(pVmcs->u32GuestIntrState & ~VMX_VMCS_GUEST_INT_STATE_MASK)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateRsvd); if ((pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)) != (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateStiMovSs); if ( (pVmcs->u64GuestRFlags.u & X86_EFL_IF) || !(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateRFlagsSti); if (VMX_ENTRY_INT_INFO_IS_VALID(pVmcs->u32EntryIntInfo)) { uint8_t const uType = VMX_ENTRY_INT_INFO_TYPE(pVmcs->u32EntryIntInfo); if (uType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT) { if (!(pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateExtInt); } else if (uType == VMX_ENTRY_INT_INFO_TYPE_NMI) { if (!(pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))) { /* likely */ } else { /* * We don't support injecting NMIs when blocking-by-STI would be in effect. * We update the Exit qualification only when blocking-by-STI is set * without blocking-by-MovSS being set. Although in practise it does not * make much difference since the order of checks are implementation defined. */ if (!(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)) iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_NMI_INJECT); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateNmi); } if ( !(pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI) || !(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateVirtNmi); } } /* We don't support SMM yet. So blocking-by-SMIs must not be set. */ if (!(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateSmi); /* We don't support SGX yet. So enclave-interruption must not be set. */ if (!(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_ENCLAVE)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateEnclave); /* * Pending debug exceptions. */ uint64_t const uPendingDbgXcpts = IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode ? pVmcs->u64GuestPendingDbgXcpts.u : pVmcs->u64GuestPendingDbgXcpts.s.Lo; if (!(uPendingDbgXcpts & ~VMX_VMCS_GUEST_PENDING_DEBUG_VALID_MASK)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPndDbgXcptRsvd); if ( (pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)) || pVmcs->u32GuestActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT) { if ( (pVmcs->u64GuestRFlags.u & X86_EFL_TF) && !(pVmcs->u64GuestDebugCtlMsr.u & MSR_IA32_DEBUGCTL_BTF) && !(uPendingDbgXcpts & VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_BS)) IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPndDbgXcptBsTf); if ( ( !(pVmcs->u64GuestRFlags.u & X86_EFL_TF) || (pVmcs->u64GuestDebugCtlMsr.u & MSR_IA32_DEBUGCTL_BTF)) && (uPendingDbgXcpts & VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_BS)) IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPndDbgXcptBsNoTf); } /* We don't support RTM (Real-time Transactional Memory) yet. */ if (!(uPendingDbgXcpts & VMX_VMCS_GUEST_PENDING_DEBUG_RTM)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPndDbgXcptRtm); /* * VMCS link pointer. */ if (pVmcs->u64VmcsLinkPtr.u != UINT64_C(0xffffffffffffffff)) { RTGCPHYS const GCPhysShadowVmcs = pVmcs->u64VmcsLinkPtr.u; /* We don't support SMM yet (so VMCS link pointer cannot be the current VMCS). */ if (GCPhysShadowVmcs != IEM_VMX_GET_CURRENT_VMCS(pVCpu)) { /* likely */ } else { iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmcsLinkPtrCurVmcs); } /* Validate the address. */ if ( !(GCPhysShadowVmcs & X86_PAGE_4K_OFFSET_MASK) && !(GCPhysShadowVmcs >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysShadowVmcs)) { /* likely */ } else { iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrVmcsLinkPtr); } } NOREF(pszInstr); NOREF(pszFailure); return VINF_SUCCESS; } #ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT /** * Checks guest PDPTEs as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static int iemVmxVmentryCheckGuestPdptes(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { /* * Guest PDPTEs. * See Intel spec. 26.3.1.5 "Checks on Guest Page-Directory-Pointer-Table Entries". */ PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char * const pszFailure = "VM-exit"; /* * When EPT is used, we need to validate the PAE PDPTEs provided in the VMCS. * Otherwise, we load any PAE PDPTEs referenced by CR3 at a later point. */ if ( iemVmxVmcsIsGuestPaePagingEnabled(pVmcs) && (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT)) { /* Get PDPTEs from the VMCS. */ X86PDPE aPaePdptes[X86_PG_PAE_PDPE_ENTRIES]; aPaePdptes[0].u = pVmcs->u64GuestPdpte0.u; aPaePdptes[1].u = pVmcs->u64GuestPdpte1.u; aPaePdptes[2].u = pVmcs->u64GuestPdpte2.u; aPaePdptes[3].u = pVmcs->u64GuestPdpte3.u; /* Check validity of the PDPTEs. */ if (PGMGstArePaePdpesValid(pVCpu, &aPaePdptes[0])) { /* likely */ } else { iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_PDPTE); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPdpte); } } NOREF(pszFailure); NOREF(pszInstr); return VINF_SUCCESS; } #endif /* VBOX_WITH_NESTED_HWVIRT_VMX_EPT */ /** * Checks guest-state as part of VM-entry. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static int iemVmxVmentryCheckGuestState(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { int rc = iemVmxVmentryCheckGuestControlRegsMsrs(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { rc = iemVmxVmentryCheckGuestSegRegs(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { rc = iemVmxVmentryCheckGuestGdtrIdtr(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { rc = iemVmxVmentryCheckGuestRipRFlags(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { rc = iemVmxVmentryCheckGuestNonRegState(pVCpu, pszInstr); #ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT if (RT_SUCCESS(rc)) rc = iemVmxVmentryCheckGuestPdptes(pVCpu, pszInstr); #endif } } } } return rc; } /** * Checks host-state as part of VM-entry. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static int iemVmxVmentryCheckHostState(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { /* * Host Control Registers and MSRs. * See Intel spec. 26.2.2 "Checks on Host Control Registers and MSRs". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char * const pszFailure = "VMFail"; /* CR0 reserved bits. */ { /* CR0 MB1 bits. */ uint64_t const u64Cr0Fixed0 = iemVmxGetCr0Fixed0(pVCpu, true /* fVmxNonRootMode */); if ((pVmcs->u64HostCr0.u & u64Cr0Fixed0) == u64Cr0Fixed0) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr0Fixed0); /* CR0 MBZ bits. */ uint64_t const u64Cr0Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed1; if (!(pVmcs->u64HostCr0.u & ~u64Cr0Fixed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr0Fixed1); } /* CR4 reserved bits. */ { /* CR4 MB1 bits. */ uint64_t const u64Cr4Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed0; if ((pVmcs->u64HostCr4.u & u64Cr4Fixed0) == u64Cr4Fixed0) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr4Fixed0); /* CR4 MBZ bits. */ uint64_t const u64Cr4Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed1; if (!(pVmcs->u64HostCr4.u & ~u64Cr4Fixed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr4Fixed1); } if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { /* CR3 reserved bits. */ if (!(pVmcs->u64HostCr3.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxPhysAddrWidth)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr3); /* SYSENTER ESP and SYSENTER EIP. */ if ( X86_IS_CANONICAL(pVmcs->u64HostSysenterEsp.u) && X86_IS_CANONICAL(pVmcs->u64HostSysenterEip.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostSysenterEspEip); } /* We don't support IA32_PERF_GLOBAL_CTRL MSR yet. */ Assert(!(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_PERF_MSR)); /* PAT MSR. */ if ( !(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_PAT_MSR) || CPUMIsPatMsrValid(pVmcs->u64HostPatMsr.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostPatMsr); /* EFER MSR. */ bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE); uint64_t const uValidEferMask = CPUMGetGuestEferMsrValidMask(pVCpu->CTX_SUFF(pVM)); if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR) { if (!(pVmcs->u64HostEferMsr.u & ~uValidEferMask)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostEferMsrRsvd); bool const fHostLma = RT_BOOL(pVmcs->u64HostEferMsr.u & MSR_K6_EFER_LMA); bool const fHostLme = RT_BOOL(pVmcs->u64HostEferMsr.u & MSR_K6_EFER_LME); if ( fHostInLongMode == fHostLma && fHostInLongMode == fHostLme) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostEferMsr); } /* * Host Segment and Descriptor-Table Registers. * See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers". */ /* Selector RPL and TI. */ if ( !(pVmcs->HostCs & (X86_SEL_RPL | X86_SEL_LDT)) && !(pVmcs->HostSs & (X86_SEL_RPL | X86_SEL_LDT)) && !(pVmcs->HostDs & (X86_SEL_RPL | X86_SEL_LDT)) && !(pVmcs->HostEs & (X86_SEL_RPL | X86_SEL_LDT)) && !(pVmcs->HostFs & (X86_SEL_RPL | X86_SEL_LDT)) && !(pVmcs->HostGs & (X86_SEL_RPL | X86_SEL_LDT)) && !(pVmcs->HostTr & (X86_SEL_RPL | X86_SEL_LDT))) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostSel); /* CS and TR selectors cannot be 0. */ if ( pVmcs->HostCs && pVmcs->HostTr) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCsTr); /* SS cannot be 0 if 32-bit host. */ if ( fHostInLongMode || pVmcs->HostSs) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostSs); if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { /* FS, GS, GDTR, IDTR, TR base address. */ if ( X86_IS_CANONICAL(pVmcs->u64HostFsBase.u) && X86_IS_CANONICAL(pVmcs->u64HostGsBase.u) && X86_IS_CANONICAL(pVmcs->u64HostGdtrBase.u) && X86_IS_CANONICAL(pVmcs->u64HostIdtrBase.u) && X86_IS_CANONICAL(pVmcs->u64HostTrBase.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostSegBase); } /* * Host address-space size for 64-bit CPUs. * See Intel spec. 26.2.4 "Checks Related to Address-Space Size". */ bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST); if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { bool const fCpuInLongMode = CPUMIsGuestInLongMode(pVCpu); /* Logical processor in IA-32e mode. */ if (fCpuInLongMode) { if (fHostInLongMode) { /* PAE must be set. */ if (pVmcs->u64HostCr4.u & X86_CR4_PAE) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr4Pae); /* RIP must be canonical. */ if (X86_IS_CANONICAL(pVmcs->u64HostRip.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostRip); } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostLongMode); } else { /* Logical processor is outside IA-32e mode. */ if ( !fGstInLongMode && !fHostInLongMode) { /* PCIDE should not be set. */ if (!(pVmcs->u64HostCr4.u & X86_CR4_PCIDE)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr4Pcide); /* The high 32-bits of RIP MBZ. */ if (!pVmcs->u64HostRip.s.Hi) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostRipRsvd); } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostGuestLongMode); } } else { /* Host address-space size for 32-bit CPUs. */ if ( !fGstInLongMode && !fHostInLongMode) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostGuestLongModeNoCpu); } NOREF(pszInstr); NOREF(pszFailure); return VINF_SUCCESS; } #ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT /** * Checks the EPT pointer VMCS field as part of VM-entry. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param uEptPtr The EPT pointer to check. * @param penmVmxDiag Where to store the diagnostic reason on failure (not * updated on success). Optional, can be NULL. */ static int iemVmxVmentryCheckEptPtr(PVMCPUCC pVCpu, uint64_t uEptPtr, VMXVDIAG *penmVmxDiag) RT_NOEXCEPT { VMXVDIAG enmVmxDiag; /* Reserved bits. */ uint8_t const cMaxPhysAddrWidth = IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxPhysAddrWidth; uint64_t const fValidMask = VMX_EPTP_VALID_MASK & ~(UINT64_MAX << cMaxPhysAddrWidth); if (uEptPtr & fValidMask) { /* Memory Type. */ uint64_t const fCaps = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64EptVpidCaps; uint8_t const fMemType = RT_BF_GET(uEptPtr, VMX_BF_EPTP_MEMTYPE); if ( ( fMemType == VMX_EPTP_MEMTYPE_WB && RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_MEMTYPE_WB)) || ( fMemType == VMX_EPTP_MEMTYPE_UC && RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_MEMTYPE_UC))) { /* * Page walk length (PML4). * Intel used to specify bit 7 of IA32_VMX_EPT_VPID_CAP as page walk length * of 5 but that seems to be removed from the latest specs. leaving only PML4 * as the maximum supported page-walk level hence we hardcode it as 3 (1 less than 4) */ Assert(RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)); if (RT_BF_GET(uEptPtr, VMX_BF_EPTP_PAGE_WALK_LENGTH) == 3) { /* Access and dirty bits support in EPT structures. */ if ( !RT_BF_GET(uEptPtr, VMX_BF_EPTP_ACCESS_DIRTY) || RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_ACCESS_DIRTY)) return VINF_SUCCESS; enmVmxDiag = kVmxVDiag_Vmentry_EptpAccessDirty; } else enmVmxDiag = kVmxVDiag_Vmentry_EptpPageWalkLength; } else enmVmxDiag = kVmxVDiag_Vmentry_EptpMemType; } else enmVmxDiag = kVmxVDiag_Vmentry_EptpRsvd; if (penmVmxDiag) *penmVmxDiag = enmVmxDiag; return VERR_VMX_VMENTRY_FAILED; } #endif /** * Checks VMCS controls fields as part of VM-entry. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). * * @remarks This may update secondary-processor based VM-execution control fields * in the current VMCS if necessary. */ static int iemVmxVmentryCheckCtls(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { PVMXVVMCS pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char * const pszFailure = "VMFail"; bool const fVmxTrueMsrs = RT_BOOL(pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Basic & VMX_BF_BASIC_TRUE_CTLS_MASK); /* * VM-execution controls. * See Intel spec. 26.2.1.1 "VM-Execution Control Fields". */ { /* Pin-based VM-execution controls. */ { VMXCTLSMSR const PinCtls = fVmxTrueMsrs ? pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.TruePinCtls : pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.PinCtls; if (!(~pVmcs->u32PinCtls & PinCtls.n.allowed0)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_PinCtlsDisallowed0); if (!(pVmcs->u32PinCtls & ~PinCtls.n.allowed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_PinCtlsAllowed1); } /* Processor-based VM-execution controls. */ { VMXCTLSMSR const ProcCtls = fVmxTrueMsrs ? pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.TrueProcCtls : pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.ProcCtls; if (!(~pVmcs->u32ProcCtls & ProcCtls.n.allowed0)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ProcCtlsDisallowed0); if (!(pVmcs->u32ProcCtls & ~ProcCtls.n.allowed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ProcCtlsAllowed1); } /* Secondary processor-based VM-execution controls. */ if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS) { VMXCTLSMSR const ProcCtls2 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.ProcCtls2; if (!(~pVmcs->u32ProcCtls2 & ProcCtls2.n.allowed0)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ProcCtls2Disallowed0); if (!(pVmcs->u32ProcCtls2 & ~ProcCtls2.n.allowed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ProcCtls2Allowed1); } else if (pVmcs->u32ProcCtls2) { /* * If the "activate secondary controls" is clear, then the secondary processor-based VM-execution controls * is treated as 0. * * See Intel spec. 26.2.1.1 "VM-Execution Control Fields". * * Since this is a rather rare occurrence (only observed for a few VM-entries with Microsoft Hyper-V * enabled Windows Server 2008 R2 guest), it's not worth changing every place that reads this control to * also check the "activate secondary controls" bit. Instead, we temporarily save the guest programmed * control here, zero out the value the rest of our code uses and restore the guest programmed value * on VM-exit. */ pVmcs->u32RestoreProcCtls2 = pVmcs->u32ProcCtls2; pVmcs->u32ProcCtls2 = 0; } /* CR3-target count. */ if (pVmcs->u32Cr3TargetCount <= VMX_V_CR3_TARGET_COUNT) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_Cr3TargetCount); /* I/O bitmaps physical addresses. */ if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS) { RTGCPHYS const GCPhysIoBitmapA = pVmcs->u64AddrIoBitmapA.u; if ( !(GCPhysIoBitmapA & X86_PAGE_4K_OFFSET_MASK) && !(GCPhysIoBitmapA >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysIoBitmapA)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrIoBitmapA); RTGCPHYS const GCPhysIoBitmapB = pVmcs->u64AddrIoBitmapB.u; if ( !(GCPhysIoBitmapB & X86_PAGE_4K_OFFSET_MASK) && !(GCPhysIoBitmapB >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysIoBitmapB)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrIoBitmapB); } /* MSR bitmap physical address. */ if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS) { RTGCPHYS const GCPhysMsrBitmap = pVmcs->u64AddrMsrBitmap.u; if ( !(GCPhysMsrBitmap & X86_PAGE_4K_OFFSET_MASK) && !(GCPhysMsrBitmap >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysMsrBitmap)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrMsrBitmap); } /* TPR shadow related controls. */ if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW) { /* Virtual-APIC page physical address. */ RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u; if ( !(GCPhysVirtApic & X86_PAGE_4K_OFFSET_MASK) && !(GCPhysVirtApic >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrVirtApicPage); /* TPR threshold bits 31:4 MBZ without virtual-interrupt delivery. */ if ( !(pVmcs->u32TprThreshold & ~VMX_TPR_THRESHOLD_MASK) || (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_TprThresholdRsvd); /* The rest done XXX document */ } else { if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_X2APIC_MODE) && !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT) && !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)) { /* likely */ } else { if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_X2APIC_MODE) IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtX2ApicTprShadow); if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT) IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ApicRegVirt); Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtIntDelivery); } } /* NMI exiting and virtual-NMIs. */ if ( (pVmcs->u32PinCtls & VMX_PIN_CTLS_NMI_EXIT) || !(pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtNmi); /* Virtual-NMIs and NMI-window exiting. */ if ( (pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI) || !(pVmcs->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_NmiWindowExit); /* Virtualize APIC accesses. */ if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS) { /* APIC-access physical address. */ RTGCPHYS const GCPhysApicAccess = pVmcs->u64AddrApicAccess.u; if ( !(GCPhysApicAccess & X86_PAGE_4K_OFFSET_MASK) && !(GCPhysApicAccess >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysApicAccess)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrApicAccess); /* * Disallow APIC-access page and virtual-APIC page from being the same address. * Note! This is not an Intel requirement, but one imposed by our implementation. * This is done primarily to simplify recursion scenarios while redirecting accesses * between the APIC-access page and the virtual-APIC page. If any nested hypervisor * requires this, we can implement it later */ if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW) { RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u; if (GCPhysVirtApic != GCPhysApicAccess) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrApicAccessEqVirtApic); } } /* Virtualize-x2APIC mode is mutually exclusive with virtualize-APIC accesses. */ if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_X2APIC_MODE) || !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtX2ApicVirtApic); /* Virtual-interrupt delivery requires external interrupt exiting. */ if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY) || (pVmcs->u32PinCtls & VMX_PIN_CTLS_EXT_INT_EXIT)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtX2ApicVirtApic); /* VPID. */ if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VPID) || pVmcs->u16Vpid != 0) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_Vpid); #ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT /* Extended-Page-Table Pointer (EPTP). */ if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT) { VMXVDIAG enmVmxDiag; int const rc = iemVmxVmentryCheckEptPtr(pVCpu, pVmcs->u64EptPtr.u, &enmVmxDiag); if (RT_SUCCESS(rc)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, enmVmxDiag, rc); } #else Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT)); Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST)); #endif Assert(!(pVmcs->u32PinCtls & VMX_PIN_CTLS_POSTED_INT)); /* We don't support posted interrupts yet. */ Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_PML)); /* We don't support PML yet. */ Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VMFUNC)); /* We don't support VM functions yet. */ Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT_XCPT_VE)); /* We don't support EPT-violation #VE yet. */ Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_TSC_SCALING)); /* We don't support TSC-scaling yet. */ /* VMCS shadowing. */ if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING) { /* VMREAD-bitmap physical address. */ RTGCPHYS const GCPhysVmreadBitmap = pVmcs->u64AddrVmreadBitmap.u; if ( !(GCPhysVmreadBitmap & X86_PAGE_4K_OFFSET_MASK) && !(GCPhysVmreadBitmap >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmreadBitmap)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrVmreadBitmap); /* VMWRITE-bitmap physical address. */ RTGCPHYS const GCPhysVmwriteBitmap = pVmcs->u64AddrVmreadBitmap.u; if ( !(GCPhysVmwriteBitmap & X86_PAGE_4K_OFFSET_MASK) && !(GCPhysVmwriteBitmap >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmwriteBitmap)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrVmwriteBitmap); } } /* * VM-exit controls. * See Intel spec. 26.2.1.2 "VM-Exit Control Fields". */ { VMXCTLSMSR const ExitCtls = fVmxTrueMsrs ? pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.TrueExitCtls : pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.ExitCtls; if (!(~pVmcs->u32ExitCtls & ExitCtls.n.allowed0)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ExitCtlsDisallowed0); if (!(pVmcs->u32ExitCtls & ~ExitCtls.n.allowed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ExitCtlsAllowed1); /* Save preemption timer without activating it. */ if ( (pVmcs->u32PinCtls & VMX_PIN_CTLS_PREEMPT_TIMER) || !(pVmcs->u32ProcCtls & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_SavePreemptTimer); /* VM-exit MSR-store count and VM-exit MSR-store area address. */ if (pVmcs->u32ExitMsrStoreCount) { if ( !(pVmcs->u64AddrExitMsrStore.u & VMX_AUTOMSR_OFFSET_MASK) && !(pVmcs->u64AddrExitMsrStore.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), pVmcs->u64AddrExitMsrStore.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrExitMsrStore); } /* VM-exit MSR-load count and VM-exit MSR-load area address. */ if (pVmcs->u32ExitMsrLoadCount) { if ( !(pVmcs->u64AddrExitMsrLoad.u & VMX_AUTOMSR_OFFSET_MASK) && !(pVmcs->u64AddrExitMsrLoad.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), pVmcs->u64AddrExitMsrLoad.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrExitMsrLoad); } } /* * VM-entry controls. * See Intel spec. 26.2.1.3 "VM-Entry Control Fields". */ { VMXCTLSMSR const EntryCtls = fVmxTrueMsrs ? pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.TrueEntryCtls : pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.EntryCtls; if (!(~pVmcs->u32EntryCtls & EntryCtls.n.allowed0)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryCtlsDisallowed0); if (!(pVmcs->u32EntryCtls & ~EntryCtls.n.allowed1)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryCtlsAllowed1); /* Event injection. */ uint32_t const uIntInfo = pVmcs->u32EntryIntInfo; if (RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_VALID)) { /* Type and vector. */ uint8_t const uType = RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_TYPE); uint8_t const uVector = RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_VECTOR); uint8_t const uRsvd = RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_RSVD_12_30); if ( !uRsvd && VMXIsEntryIntInfoTypeValid(IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxMonitorTrapFlag, uType) && VMXIsEntryIntInfoVectorValid(uVector, uType)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryIntInfoTypeVecRsvd); /* Exception error code. */ if (RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID)) { /* Delivery possible only in Unrestricted-guest mode when CR0.PE is set. */ if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST) || (pVmcs->u64GuestCr0.s.Lo & X86_CR0_PE)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryIntInfoErrCodePe); /* Exceptions that provide an error code. */ if ( uType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT && ( uVector == X86_XCPT_DF || uVector == X86_XCPT_TS || uVector == X86_XCPT_NP || uVector == X86_XCPT_SS || uVector == X86_XCPT_GP || uVector == X86_XCPT_PF || uVector == X86_XCPT_AC)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryIntInfoErrCodeVec); /* Exception error-code reserved bits. */ if (!(pVmcs->u32EntryXcptErrCode & ~VMX_ENTRY_INT_XCPT_ERR_CODE_VALID_MASK)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryXcptErrCodeRsvd); /* Injecting a software interrupt, software exception or privileged software exception. */ if ( uType == VMX_ENTRY_INT_INFO_TYPE_SW_INT || uType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT || uType == VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT) { /* Instruction length must be in the range 0-15. */ if (pVmcs->u32EntryInstrLen <= VMX_ENTRY_INSTR_LEN_MAX) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryInstrLen); /* However, instruction length of 0 is allowed only when its CPU feature is present. */ if ( pVmcs->u32EntryInstrLen != 0 || IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxEntryInjectSoftInt) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryInstrLenZero); } } } /* VM-entry MSR-load count and VM-entry MSR-load area address. */ if (pVmcs->u32EntryMsrLoadCount) { if ( !(pVmcs->u64AddrEntryMsrLoad.u & VMX_AUTOMSR_OFFSET_MASK) && !(pVmcs->u64AddrEntryMsrLoad.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth) && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), pVmcs->u64AddrEntryMsrLoad.u)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrEntryMsrLoad); } Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)); /* We don't support SMM yet. */ Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON)); /* We don't support dual-monitor treatment yet. */ } NOREF(pszInstr); NOREF(pszFailure); return VINF_SUCCESS; } /** * Loads the guest control registers, debug register and some MSRs as part of * VM-entry. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmentryLoadGuestControlRegsMsrs(PVMCPUCC pVCpu) RT_NOEXCEPT { /* * Load guest control registers, debug registers and MSRs. * See Intel spec. 26.3.2.1 "Loading Guest Control Registers, Debug Registers and MSRs". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0); uint64_t const uGstCr0 = (pVmcs->u64GuestCr0.u & ~VMX_ENTRY_GUEST_CR0_IGNORE_MASK) | (pVCpu->cpum.GstCtx.cr0 & VMX_ENTRY_GUEST_CR0_IGNORE_MASK); pVCpu->cpum.GstCtx.cr0 = uGstCr0; pVCpu->cpum.GstCtx.cr4 = pVmcs->u64GuestCr4.u; pVCpu->cpum.GstCtx.cr3 = pVmcs->u64GuestCr3.u; if (pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG) pVCpu->cpum.GstCtx.dr[7] = (pVmcs->u64GuestDr7.u & ~VMX_ENTRY_GUEST_DR7_MBZ_MASK) | VMX_ENTRY_GUEST_DR7_MB1_MASK; pVCpu->cpum.GstCtx.SysEnter.eip = pVmcs->u64GuestSysenterEip.s.Lo; pVCpu->cpum.GstCtx.SysEnter.esp = pVmcs->u64GuestSysenterEsp.s.Lo; pVCpu->cpum.GstCtx.SysEnter.cs = pVmcs->u32GuestSysenterCS; if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode) { /* FS base and GS base are loaded while loading the rest of the guest segment registers. */ /* EFER MSR. */ if (!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)) { IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER); uint64_t const uHostEfer = pVCpu->cpum.GstCtx.msrEFER; bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST); bool const fGstPaging = RT_BOOL(uGstCr0 & X86_CR0_PG); if (fGstInLongMode) { /* If the nested-guest is in long mode, LMA and LME are both set. */ Assert(fGstPaging); pVCpu->cpum.GstCtx.msrEFER = uHostEfer | (MSR_K6_EFER_LMA | MSR_K6_EFER_LME); } else { /* * If the nested-guest is outside long mode: * - With paging: LMA is cleared, LME is cleared. * - Without paging: LMA is cleared, LME is left unmodified. */ uint64_t const fLmaLmeMask = MSR_K6_EFER_LMA | (fGstPaging ? MSR_K6_EFER_LME : 0); pVCpu->cpum.GstCtx.msrEFER = uHostEfer & ~fLmaLmeMask; } } /* else: see below. */ } /* PAT MSR. */ if (pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR) pVCpu->cpum.GstCtx.msrPAT = pVmcs->u64GuestPatMsr.u; /* EFER MSR. */ if (pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR) pVCpu->cpum.GstCtx.msrEFER = pVmcs->u64GuestEferMsr.u; /* We don't support IA32_PERF_GLOBAL_CTRL MSR yet. */ Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR)); /* We don't support IA32_BNDCFGS MSR yet. */ Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_BNDCFGS_MSR)); /* Nothing to do for SMBASE register - We don't support SMM yet. */ } /** * Loads the guest segment registers, GDTR, IDTR, LDTR and TR as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmentryLoadGuestSegRegs(PVMCPUCC pVCpu) RT_NOEXCEPT { /* * Load guest segment registers, GDTR, IDTR, LDTR and TR. * See Intel spec. 26.3.2.2 "Loading Guest Segment Registers and Descriptor-Table Registers". */ /* CS, SS, ES, DS, FS, GS. */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; for (unsigned iSegReg = 0; iSegReg < X86_SREG_COUNT; iSegReg++) { PCPUMSELREG pGstSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg]; CPUMSELREG VmcsSelReg; int rc = iemVmxVmcsGetGuestSegReg(pVmcs, iSegReg, &VmcsSelReg); AssertRC(rc); NOREF(rc); if (!(VmcsSelReg.Attr.u & X86DESCATTR_UNUSABLE)) { pGstSelReg->Sel = VmcsSelReg.Sel; pGstSelReg->ValidSel = VmcsSelReg.Sel; pGstSelReg->fFlags = CPUMSELREG_FLAGS_VALID; pGstSelReg->u64Base = VmcsSelReg.u64Base; pGstSelReg->u32Limit = VmcsSelReg.u32Limit; pGstSelReg->Attr.u = VmcsSelReg.Attr.u; } else { pGstSelReg->Sel = VmcsSelReg.Sel; pGstSelReg->ValidSel = VmcsSelReg.Sel; pGstSelReg->fFlags = CPUMSELREG_FLAGS_VALID; switch (iSegReg) { case X86_SREG_CS: pGstSelReg->u64Base = VmcsSelReg.u64Base; pGstSelReg->u32Limit = VmcsSelReg.u32Limit; pGstSelReg->Attr.u = VmcsSelReg.Attr.u; break; case X86_SREG_SS: pGstSelReg->u64Base = VmcsSelReg.u64Base & UINT32_C(0xfffffff0); pGstSelReg->u32Limit = 0; pGstSelReg->Attr.u = (VmcsSelReg.Attr.u & X86DESCATTR_DPL) | X86DESCATTR_D | X86DESCATTR_UNUSABLE; break; case X86_SREG_ES: case X86_SREG_DS: pGstSelReg->u64Base = 0; pGstSelReg->u32Limit = 0; pGstSelReg->Attr.u = X86DESCATTR_UNUSABLE; break; case X86_SREG_FS: case X86_SREG_GS: pGstSelReg->u64Base = VmcsSelReg.u64Base; pGstSelReg->u32Limit = 0; pGstSelReg->Attr.u = X86DESCATTR_UNUSABLE; break; } Assert(pGstSelReg->Attr.n.u1Unusable); } } /* LDTR. */ pVCpu->cpum.GstCtx.ldtr.Sel = pVmcs->GuestLdtr; pVCpu->cpum.GstCtx.ldtr.ValidSel = pVmcs->GuestLdtr; pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID; if (!(pVmcs->u32GuestLdtrAttr & X86DESCATTR_UNUSABLE)) { pVCpu->cpum.GstCtx.ldtr.u64Base = pVmcs->u64GuestLdtrBase.u; pVCpu->cpum.GstCtx.ldtr.u32Limit = pVmcs->u32GuestLdtrLimit; pVCpu->cpum.GstCtx.ldtr.Attr.u = pVmcs->u32GuestLdtrAttr; } else { pVCpu->cpum.GstCtx.ldtr.u64Base = 0; pVCpu->cpum.GstCtx.ldtr.u32Limit = 0; pVCpu->cpum.GstCtx.ldtr.Attr.u = X86DESCATTR_UNUSABLE; } /* TR. */ Assert(!(pVmcs->u32GuestTrAttr & X86DESCATTR_UNUSABLE)); pVCpu->cpum.GstCtx.tr.Sel = pVmcs->GuestTr; pVCpu->cpum.GstCtx.tr.ValidSel = pVmcs->GuestTr; pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID; pVCpu->cpum.GstCtx.tr.u64Base = pVmcs->u64GuestTrBase.u; pVCpu->cpum.GstCtx.tr.u32Limit = pVmcs->u32GuestTrLimit; pVCpu->cpum.GstCtx.tr.Attr.u = pVmcs->u32GuestTrAttr; /* GDTR. */ pVCpu->cpum.GstCtx.gdtr.cbGdt = pVmcs->u32GuestGdtrLimit; pVCpu->cpum.GstCtx.gdtr.pGdt = pVmcs->u64GuestGdtrBase.u; /* IDTR. */ pVCpu->cpum.GstCtx.idtr.cbIdt = pVmcs->u32GuestIdtrLimit; pVCpu->cpum.GstCtx.idtr.pIdt = pVmcs->u64GuestIdtrBase.u; } /** * Loads the guest MSRs from the VM-entry MSR-load area as part of VM-entry. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static int iemVmxVmentryLoadGuestAutoMsrs(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { /* * Load guest MSRs. * See Intel spec. 26.4 "Loading MSRs". */ PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; const char *const pszFailure = "VM-exit"; /* * The VM-entry MSR-load area address need not be a valid guest-physical address if the * VM-entry MSR load count is 0. If this is the case, bail early without reading it. * See Intel spec. 24.8.2 "VM-Entry Controls for MSRs". */ uint32_t const cMsrs = RT_MIN(pVmcs->u32EntryMsrLoadCount, RT_ELEMENTS(pVCpu->cpum.GstCtx.hwvirt.vmx.aEntryMsrLoadArea)); if (!cMsrs) return VINF_SUCCESS; /* * Verify the MSR auto-load count. Physical CPUs can behave unpredictably if the count is * exceeded including possibly raising #MC exceptions during VMX transition. Our * implementation shall fail VM-entry with an VMX_EXIT_ERR_MSR_LOAD VM-exit. */ bool const fIsMsrCountValid = iemVmxIsAutoMsrCountValid(pVCpu, cMsrs); if (fIsMsrCountValid) { /* likely */ } else { iemVmxVmcsSetExitQual(pVCpu, VMX_V_AUTOMSR_AREA_SIZE / sizeof(VMXAUTOMSR)); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_MsrLoadCount); } RTGCPHYS const GCPhysVmEntryMsrLoadArea = pVmcs->u64AddrEntryMsrLoad.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.aEntryMsrLoadArea[0], GCPhysVmEntryMsrLoadArea, cMsrs * sizeof(VMXAUTOMSR)); if (RT_SUCCESS(rc)) { PCVMXAUTOMSR pMsr = &pVCpu->cpum.GstCtx.hwvirt.vmx.aEntryMsrLoadArea[0]; for (uint32_t idxMsr = 0; idxMsr < cMsrs; idxMsr++, pMsr++) { if ( !pMsr->u32Reserved && pMsr->u32Msr != MSR_K8_FS_BASE && pMsr->u32Msr != MSR_K8_GS_BASE && pMsr->u32Msr != MSR_K6_EFER && pMsr->u32Msr != MSR_IA32_SMM_MONITOR_CTL && pMsr->u32Msr >> 8 != MSR_IA32_X2APIC_START >> 8) { VBOXSTRICTRC rcStrict = CPUMSetGuestMsr(pVCpu, pMsr->u32Msr, pMsr->u64Value); if (rcStrict == VINF_SUCCESS) continue; /* * If we're in ring-0, we cannot handle returns to ring-3 at this point and continue VM-entry. * If any nested hypervisor loads MSRs that require ring-3 handling, we cause a VM-entry failure * recording the MSR index in the Exit qualification (as per the Intel spec.) and indicated * further by our own, specific diagnostic code. Later, we can try implement handling of the * MSR in ring-0 if possible, or come up with a better, generic solution. */ iemVmxVmcsSetExitQual(pVCpu, idxMsr); VMXVDIAG const enmDiag = rcStrict == VINF_CPUM_R3_MSR_WRITE ? kVmxVDiag_Vmentry_MsrLoadRing3 : kVmxVDiag_Vmentry_MsrLoad; IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag); } else { iemVmxVmcsSetExitQual(pVCpu, idxMsr); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_MsrLoadRsvd); } } } else { AssertMsgFailed(("%s: Failed to read MSR auto-load area at %#RGp, rc=%Rrc\n", pszInstr, GCPhysVmEntryMsrLoadArea, rc)); IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_MsrLoadPtrReadPhys, rc); } NOREF(pszInstr); NOREF(pszFailure); return VINF_SUCCESS; } /** * Loads the guest-state non-register state as part of VM-entry. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). * * @remarks This must be called only after loading the nested-guest register state * (especially nested-guest RIP). */ static int iemVmxVmentryLoadGuestNonRegState(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { /* * Load guest non-register state. * See Intel spec. 26.6 "Special Features of VM Entry" */ const char *const pszFailure = "VM-exit"; PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* * If VM-entry is not vectoring, block-by-STI and block-by-MovSS state must be loaded. * If VM-entry is vectoring, there is no block-by-STI or block-by-MovSS. * * See Intel spec. 26.6.1 "Interruptibility State". */ bool const fEntryVectoring = VMXIsVmentryVectoring(pVmcs->u32EntryIntInfo, NULL /* puEntryIntInfoType */); if ( !fEntryVectoring && (pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI | VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))) CPUMSetInInterruptShadowEx(&pVCpu->cpum.GstCtx, pVmcs->u64GuestRip.u); else Assert(!CPUMIsInInterruptShadow(&pVCpu->cpum.GstCtx)); /* NMI blocking. */ if (pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI) { if (pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI) { pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking = true; Assert(!CPUMAreInterruptsInhibitedByNmi(&pVCpu->cpum.GstCtx)); } else { pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking = false; CPUMSetInterruptInhibitingByNmi(&pVCpu->cpum.GstCtx); } } else { pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking = false; Assert(!CPUMAreInterruptsInhibitedByNmi(&pVCpu->cpum.GstCtx)); } /* SMI blocking is irrelevant. We don't support SMIs yet. */ /* * Set PGM's copy of the EPT pointer. * The EPTP has already been validated while checking guest state. * * It is important to do this prior to mapping PAE PDPTEs (below). */ if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT) PGMSetGuestEptPtr(pVCpu, pVmcs->u64EptPtr.u); /* * Load the guest's PAE PDPTEs. */ if (!iemVmxVmcsIsGuestPaePagingEnabled(pVmcs)) { /* * When PAE paging is not used we clear the PAE PDPTEs for safety * in case we might be switching from a PAE host to a non-PAE guest. */ pVCpu->cpum.GstCtx.aPaePdpes[0].u = 0; pVCpu->cpum.GstCtx.aPaePdpes[1].u = 0; pVCpu->cpum.GstCtx.aPaePdpes[2].u = 0; pVCpu->cpum.GstCtx.aPaePdpes[3].u = 0; } else if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT) { /* * With EPT and the nested-guest using PAE paging, we've already validated the PAE PDPTEs * while checking the guest state. We can load them into the nested-guest CPU state now. * They'll later be used while mapping CR3 and the PAE PDPTEs. */ pVCpu->cpum.GstCtx.aPaePdpes[0].u = pVmcs->u64GuestPdpte0.u; pVCpu->cpum.GstCtx.aPaePdpes[1].u = pVmcs->u64GuestPdpte1.u; pVCpu->cpum.GstCtx.aPaePdpes[2].u = pVmcs->u64GuestPdpte2.u; pVCpu->cpum.GstCtx.aPaePdpes[3].u = pVmcs->u64GuestPdpte3.u; } else { /* * Without EPT and the nested-guest using PAE paging, we must load the PAE PDPTEs * referenced by CR3. This involves loading (and mapping) CR3 and validating them now. */ int const rc = PGMGstMapPaePdpesAtCr3(pVCpu, pVmcs->u64GuestCr3.u); if (RT_SUCCESS(rc)) { /* likely */ } else { iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_PDPTE); IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPdpte, rc); } } /* VPID is irrelevant. We don't support VPID yet. */ /* Clear address-range monitoring. */ EMMonitorWaitClear(pVCpu); return VINF_SUCCESS; } /** * Loads the guest VMCS referenced state (such as MSR bitmaps, I/O bitmaps etc). * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). * * @remarks This assumes various VMCS related data structure pointers have already * been verified prior to calling this function. */ static int iemVmxVmentryLoadGuestVmcsRefState(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { const char *const pszFailure = "VM-exit"; PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* * Virtualize APIC accesses. */ if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS) { /* APIC-access physical address. */ RTGCPHYS const GCPhysApicAccess = pVmcs->u64AddrApicAccess.u; /* * Register the handler for the APIC-access page. * * We don't deregister the APIC-access page handler during the VM-exit as a different * nested-VCPU might be using the same guest-physical address for its APIC-access page. * * We leave the page registered until the first access that happens outside VMX non-root * mode. Guest software is allowed to access structures such as the APIC-access page * only when no logical processor with a current VMCS references it in VMX non-root mode, * otherwise it can lead to unpredictable behavior including guest triple-faults. * * See Intel spec. 24.11.4 "Software Access to Related Structures". */ /** @todo r=bird: The lazy deregistration of the page is potentially slightly * problematic, as the guest may cause us to create lots of access * handler entries. However, any slowdown or similar effects should * only ever affect the guest itself, so not a big issue. Though, I * wish there was most recently used approach or something to tracking * these... */ PVMCC pVM = pVCpu->CTX_SUFF(pVM); int rc = PGMHandlerPhysicalRegisterVmxApicAccessPage(pVM, GCPhysApicAccess, pVM->iem.s.hVmxApicAccessPage); if (RT_SUCCESS(rc)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrApicAccessHandlerReg, rc); } /* * VMCS shadowing. */ if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING) { /* Read the VMREAD-bitmap. */ RTGCPHYS const GCPhysVmreadBitmap = pVmcs->u64AddrVmreadBitmap.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.abVmreadBitmap[0], GCPhysVmreadBitmap, sizeof(pVCpu->cpum.GstCtx.hwvirt.vmx.abVmreadBitmap)); if (RT_SUCCESS(rc)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmreadBitmapPtrReadPhys, rc); /* Read the VMWRITE-bitmap. */ RTGCPHYS const GCPhysVmwriteBitmap = pVmcs->u64AddrVmwriteBitmap.u; rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.abVmwriteBitmap[0], GCPhysVmwriteBitmap, sizeof(pVCpu->cpum.GstCtx.hwvirt.vmx.abVmwriteBitmap)); if (RT_SUCCESS(rc)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmwriteBitmapPtrReadPhys, rc); } /* * I/O bitmaps. */ if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS) { /* Read the IO bitmap A. */ RTGCPHYS const GCPhysIoBitmapA = pVmcs->u64AddrIoBitmapA.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.abIoBitmap[0], GCPhysIoBitmapA, VMX_V_IO_BITMAP_A_SIZE); if (RT_SUCCESS(rc)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_IoBitmapAPtrReadPhys, rc); /* Read the IO bitmap B. */ RTGCPHYS const GCPhysIoBitmapB = pVmcs->u64AddrIoBitmapB.u; rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.abIoBitmap[VMX_V_IO_BITMAP_A_SIZE], GCPhysIoBitmapB, VMX_V_IO_BITMAP_B_SIZE); if (RT_SUCCESS(rc)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_IoBitmapBPtrReadPhys, rc); } /* * TPR shadow and Virtual-APIC page. */ if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW) { /* Verify TPR threshold and VTPR when both virtualize-APIC accesses and virtual-interrupt delivery aren't used. */ if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS) && !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)) { /* Read the VTPR from the virtual-APIC page. */ RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u; uint8_t u8VTpr; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &u8VTpr, GCPhysVirtApic + XAPIC_OFF_TPR, sizeof(u8VTpr)); if (RT_SUCCESS(rc)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtApicPagePtrReadPhys, rc); /* Bits 3:0 of the TPR-threshold must not be greater than bits 7:4 of VTPR. */ if ((uint8_t)RT_BF_GET(pVmcs->u32TprThreshold, VMX_BF_TPR_THRESHOLD_TPR) <= (u8VTpr & 0xf0)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_TprThresholdVTpr); } } /* * VMCS link pointer. */ if (pVmcs->u64VmcsLinkPtr.u != UINT64_C(0xffffffffffffffff)) { /* Read the VMCS-link pointer from guest memory. */ RTGCPHYS const GCPhysShadowVmcs = pVmcs->u64VmcsLinkPtr.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.ShadowVmcs, GCPhysShadowVmcs, sizeof(pVCpu->cpum.GstCtx.hwvirt.vmx.ShadowVmcs)); if (RT_SUCCESS(rc)) { /* likely */ } else { iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR); IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmcsLinkPtrReadPhys, rc); } /* Verify the VMCS revision specified by the guest matches what we reported to the guest. */ if (pVCpu->cpum.GstCtx.hwvirt.vmx.ShadowVmcs.u32VmcsRevId.n.u31RevisionId == VMX_V_VMCS_REVISION_ID) { /* likely */ } else { iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmcsLinkPtrRevId); } /* Verify the shadow bit is set if VMCS shadowing is enabled . */ if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING) || pVCpu->cpum.GstCtx.hwvirt.vmx.ShadowVmcs.u32VmcsRevId.n.fIsShadowVmcs) { /* likely */ } else { iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR); IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmcsLinkPtrShadow); } /* Update our cache of the guest physical address of the shadow VMCS. */ pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysShadowVmcs = GCPhysShadowVmcs; } /* * MSR bitmap. */ if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS) { /* Read the MSR bitmap. */ RTGCPHYS const GCPhysMsrBitmap = pVmcs->u64AddrMsrBitmap.u; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap[0], GCPhysMsrBitmap, sizeof(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap)); if (RT_SUCCESS(rc)) { /* likely */ } else IEM_VMX_VMENTRY_FAILED_RET_2(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_MsrBitmapPtrReadPhys, rc); } NOREF(pszFailure); NOREF(pszInstr); return VINF_SUCCESS; } /** * Loads the guest-state as part of VM-entry. * * @returns VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). * * @remarks This must be done after all the necessary steps prior to loading of * guest-state (e.g. checking various VMCS state). */ static int iemVmxVmentryLoadGuestState(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { /* Load guest control registers, MSRs (that are directly part of the VMCS). */ iemVmxVmentryLoadGuestControlRegsMsrs(pVCpu); /* Load guest segment registers. */ iemVmxVmentryLoadGuestSegRegs(pVCpu); /* * Load guest RIP, RSP and RFLAGS. * See Intel spec. 26.3.2.3 "Loading Guest RIP, RSP and RFLAGS". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; pVCpu->cpum.GstCtx.rsp = pVmcs->u64GuestRsp.u; pVCpu->cpum.GstCtx.rip = pVmcs->u64GuestRip.u; pVCpu->cpum.GstCtx.rflags.u = pVmcs->u64GuestRFlags.u; /* Initialize the PAUSE-loop controls as part of VM-entry. */ pVCpu->cpum.GstCtx.hwvirt.vmx.uFirstPauseLoopTick = 0; pVCpu->cpum.GstCtx.hwvirt.vmx.uPrevPauseTick = 0; /* Load guest non-register state (such as interrupt shadows, NMI blocking etc). */ int rc = iemVmxVmentryLoadGuestNonRegState(pVCpu, pszInstr); if (rc == VINF_SUCCESS) { /* likely */ } else return rc; /* Load VMX related structures and state referenced by the VMCS. */ rc = iemVmxVmentryLoadGuestVmcsRefState(pVCpu, pszInstr); if (rc == VINF_SUCCESS) { /* likely */ } else return rc; NOREF(pszInstr); return VINF_SUCCESS; } /** * Returns whether there are is a pending debug exception on VM-entry. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static bool iemVmxVmentryIsPendingDebugXcpt(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { /* * Pending debug exceptions. * See Intel spec. 26.6.3 "Delivery of Pending Debug Exceptions after VM Entry". */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs); bool fPendingDbgXcpt = RT_BOOL(pVmcs->u64GuestPendingDbgXcpts.u & ( VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_BS | VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_EN_BP)); if (fPendingDbgXcpt) { uint8_t uEntryIntInfoType; bool const fEntryVectoring = VMXIsVmentryVectoring(pVmcs->u32EntryIntInfo, &uEntryIntInfoType); if (fEntryVectoring) { switch (uEntryIntInfoType) { case VMX_ENTRY_INT_INFO_TYPE_EXT_INT: case VMX_ENTRY_INT_INFO_TYPE_NMI: case VMX_ENTRY_INT_INFO_TYPE_HW_XCPT: case VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT: fPendingDbgXcpt = false; break; case VMX_ENTRY_INT_INFO_TYPE_SW_XCPT: { /* * Whether the pending debug exception for software exceptions other than * #BP and #OF is delivered after injecting the exception or is discard * is CPU implementation specific. We will discard them (easier). */ uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(pVmcs->u32EntryIntInfo); if ( uVector != X86_XCPT_BP && uVector != X86_XCPT_OF) fPendingDbgXcpt = false; RT_FALL_THRU(); } case VMX_ENTRY_INT_INFO_TYPE_SW_INT: { if (!(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)) fPendingDbgXcpt = false; break; } } } else { /* * When the VM-entry is not vectoring but there is blocking-by-MovSS, whether the * pending debug exception is held pending or is discarded is CPU implementation * specific. We will discard them (easier). */ if (pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS) fPendingDbgXcpt = false; /* There's no pending debug exception in the shutdown or wait-for-SIPI state. */ if (pVmcs->u32GuestActivityState & (VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN | VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT)) fPendingDbgXcpt = false; } } NOREF(pszInstr); return fPendingDbgXcpt; } /** * Set up the monitor-trap flag (MTF). * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static void iemVmxVmentrySetupMtf(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs); if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_MONITOR_TRAP_FLAG) { VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_MTF); Log(("%s: Monitor-trap flag set on VM-entry\n", pszInstr)); } else Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_MTF)); NOREF(pszInstr); } /** * Sets up NMI-window exiting. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static void iemVmxVmentrySetupNmiWindow(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs); if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT) { Assert(pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI); VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_NMI_WINDOW); Log(("%s: NMI-window set on VM-entry\n", pszInstr)); } else Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_NMI_WINDOW)); NOREF(pszInstr); } /** * Sets up interrupt-window exiting. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static void iemVmxVmentrySetupIntWindow(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs); if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT) { VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_INT_WINDOW); Log(("%s: Interrupt-window set on VM-entry\n", pszInstr)); } else Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_INT_WINDOW)); NOREF(pszInstr); } /** * Set up the VMX-preemption timer. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static void iemVmxVmentrySetupPreemptTimer(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs); if (pVmcs->u32PinCtls & VMX_PIN_CTLS_PREEMPT_TIMER) { /* * If the timer is 0, we must cause a VM-exit before executing the first * nested-guest instruction. So we can flag as though the timer has already * expired and we will check and cause a VM-exit at the right priority elsewhere * in the code. */ uint64_t uEntryTick; uint32_t const uPreemptTimer = pVmcs->u32PreemptTimer; if (uPreemptTimer) { int rc = CPUMStartGuestVmxPremptTimer(pVCpu, uPreemptTimer, VMX_V_PREEMPT_TIMER_SHIFT, &uEntryTick); AssertRC(rc); Log(("%s: VM-entry set up VMX-preemption timer at %#RX64\n", pszInstr, uEntryTick)); } else { uEntryTick = TMCpuTickGetNoCheck(pVCpu); VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER); Log(("%s: VM-entry set up VMX-preemption timer at %#RX64 to expire immediately!\n", pszInstr, uEntryTick)); } pVCpu->cpum.GstCtx.hwvirt.vmx.uEntryTick = uEntryTick; } else Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER)); NOREF(pszInstr); } /** * Injects an event using TRPM given a VM-entry interruption info and related * fields. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). * @param uEntryIntInfo The VM-entry interruption info. * @param uErrCode The error code associated with the event if any. * @param cbInstr The VM-entry instruction length (for software * interrupts and software exceptions). Pass 0 * otherwise. * @param GCPtrFaultAddress The guest CR2 if this is a \#PF event. */ static void iemVmxVmentryInjectTrpmEvent(PVMCPUCC pVCpu, const char *pszInstr, uint32_t uEntryIntInfo, uint32_t uErrCode, uint32_t cbInstr, RTGCUINTPTR GCPtrFaultAddress) RT_NOEXCEPT { Assert(VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo)); uint8_t const uType = VMX_ENTRY_INT_INFO_TYPE(uEntryIntInfo); uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(uEntryIntInfo); TRPMEVENT const enmTrpmEvent = HMVmxEventTypeToTrpmEventType(uEntryIntInfo); Assert(uType != VMX_ENTRY_INT_INFO_TYPE_OTHER_EVENT); int rc = TRPMAssertTrap(pVCpu, uVector, enmTrpmEvent); AssertRC(rc); Log(("%s: Injecting: vector=%#x type=%#x (%s)\n", pszInstr, uVector, uType, VMXGetEntryIntInfoTypeDesc(uType))); if (VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(uEntryIntInfo)) { TRPMSetErrorCode(pVCpu, uErrCode); Log(("%s: Injecting: err_code=%#x\n", pszInstr, uErrCode)); } if (VMX_ENTRY_INT_INFO_IS_XCPT_PF(uEntryIntInfo)) { TRPMSetFaultAddress(pVCpu, GCPtrFaultAddress); Log(("%s: Injecting: fault_addr=%RGp\n", pszInstr, GCPtrFaultAddress)); } else { switch (uType) { case VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT: TRPMSetTrapDueToIcebp(pVCpu); Log(("%s: Injecting: icebp\n", pszInstr)); RT_FALL_THRU(); case VMX_ENTRY_INT_INFO_TYPE_SW_INT: case VMX_ENTRY_INT_INFO_TYPE_SW_XCPT: TRPMSetInstrLength(pVCpu, cbInstr); Log(("%s: Injecting: instr_len=%u\n", pszInstr, cbInstr)); break; } } NOREF(pszInstr); } /** * Performs event injection (if any) as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. * @param pszInstr The VMX instruction name (for logging purposes). */ static void iemVmxVmentryInjectEvent(PVMCPUCC pVCpu, const char *pszInstr) RT_NOEXCEPT { PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; /* * Inject events. * The event that is going to be made pending for injection is not subject to VMX intercepts, * thus we flag ignoring of intercepts. However, recursive exceptions if any during delivery * of the current event -are- subject to intercepts, hence this flag will be flipped during * the actually delivery of this event. * * See Intel spec. 26.5 "Event Injection". */ uint32_t const uEntryIntInfo = pVmcs->u32EntryIntInfo; bool const fEntryIntInfoValid = VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo); CPUMSetGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx, !fEntryIntInfoValid); if (fEntryIntInfoValid) { if (VMX_ENTRY_INT_INFO_TYPE(uEntryIntInfo) != VMX_ENTRY_INT_INFO_TYPE_OTHER_EVENT) iemVmxVmentryInjectTrpmEvent(pVCpu, pszInstr, uEntryIntInfo, pVmcs->u32EntryXcptErrCode, pVmcs->u32EntryInstrLen, pVCpu->cpum.GstCtx.cr2); else { Assert(VMX_ENTRY_INT_INFO_VECTOR(uEntryIntInfo) == VMX_ENTRY_INT_INFO_VECTOR_MTF); VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_MTF); } /* * We need to clear the VM-entry interruption information field's valid bit on VM-exit. * * However, we do it here on VM-entry as well because while it isn't visible to guest * software until VM-exit, when and if HM looks at the VMCS to continue nested-guest * execution using hardware-assisted VMX, it will not try to inject the event again. * * See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection". */ pVmcs->u32EntryIntInfo &= ~VMX_ENTRY_INT_INFO_VALID; } else { /* * Inject any pending guest debug exception. * Unlike injecting events, this #DB injection on VM-entry is subject to #DB VMX intercept. * See Intel spec. 26.6.3 "Delivery of Pending Debug Exceptions after VM Entry". */ bool const fPendingDbgXcpt = iemVmxVmentryIsPendingDebugXcpt(pVCpu, pszInstr); if (fPendingDbgXcpt) { uint32_t const uDbgXcptInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DB) | RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT) | RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1); iemVmxVmentryInjectTrpmEvent(pVCpu, pszInstr, uDbgXcptInfo, 0 /* uErrCode */, pVmcs->u32EntryInstrLen, 0 /* GCPtrFaultAddress */); } } NOREF(pszInstr); } /** * Initializes all read-only VMCS fields as part of VM-entry. * * @param pVCpu The cross context virtual CPU structure. */ static void iemVmxVmentryInitReadOnlyFields(PVMCPUCC pVCpu) RT_NOEXCEPT { /* * Any VMCS field which we do not establish on every VM-exit but may potentially * be used on the VM-exit path of a nested hypervisor -and- is not explicitly * specified to be undefined, needs to be initialized here. * * Thus, it is especially important to clear the Exit qualification field * since it must be zero for VM-exits where it is not used. Similarly, the * VM-exit interruption information field's valid bit needs to be cleared for * the same reasons. */ PVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs); /* 16-bit (none currently). */ /* 32-bit. */ pVmcs->u32RoVmInstrError = 0; pVmcs->u32RoExitReason = 0; pVmcs->u32RoExitIntInfo = 0; pVmcs->u32RoExitIntErrCode = 0; pVmcs->u32RoIdtVectoringInfo = 0; pVmcs->u32RoIdtVectoringErrCode = 0; pVmcs->u32RoExitInstrLen = 0; pVmcs->u32RoExitInstrInfo = 0; /* 64-bit. */ pVmcs->u64RoGuestPhysAddr.u = 0; /* Natural-width. */ pVmcs->u64RoExitQual.u = 0; pVmcs->u64RoIoRcx.u = 0; pVmcs->u64RoIoRsi.u = 0; pVmcs->u64RoIoRdi.u = 0; pVmcs->u64RoIoRip.u = 0; pVmcs->u64RoGuestLinearAddr.u = 0; } /** * VMLAUNCH/VMRESUME instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param uInstrId The instruction identity (VMXINSTRID_VMLAUNCH or * VMXINSTRID_VMRESUME). * * @remarks Common VMX instruction checks are already expected to by the caller, * i.e. CR4.VMXE, Real/V86 mode, EFER/CS.L checks. */ static VBOXSTRICTRC iemVmxVmlaunchVmresume(PVMCPUCC pVCpu, uint8_t cbInstr, VMXINSTRID uInstrId) RT_NOEXCEPT { # if defined(VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM) && !defined(IN_RING3) RT_NOREF3(pVCpu, cbInstr, uInstrId); return VINF_EM_RAW_EMULATE_INSTR; # else Assert( uInstrId == VMXINSTRID_VMLAUNCH || uInstrId == VMXINSTRID_VMRESUME); const char * const pszInstr = uInstrId == VMXINSTRID_VMRESUME ? "vmresume" : "vmlaunch"; /* Nested-guest intercept. */ if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) return iemVmxVmexitInstr(pVCpu, uInstrId == VMXINSTRID_VMRESUME ? VMX_EXIT_VMRESUME : VMX_EXIT_VMLAUNCH, cbInstr); Assert(IEM_VMX_IS_ROOT_MODE(pVCpu)); /* * Basic VM-entry checks. * The order of the CPL, current and shadow VMCS and block-by-MovSS are important. * The checks following that do not have to follow a specific order. * * See Intel spec. 26.1 "Basic VM-entry Checks". */ /* CPL. */ if (IEM_GET_CPL(pVCpu) == 0) { /* likely */ } else { Log(("%s: CPL %u -> #GP(0)\n", pszInstr, IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_Cpl; return iemRaiseGeneralProtectionFault0(pVCpu); } /* Current VMCS valid. */ if (IEM_VMX_HAS_CURRENT_VMCS(pVCpu)) { /* likely */ } else { Log(("%s: VMCS pointer %#RGp invalid -> VMFailInvalid\n", pszInstr, IEM_VMX_GET_CURRENT_VMCS(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_PtrInvalid; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Current VMCS is not a shadow VMCS. */ PVMXVVMCS pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; if (!pVmcs->u32VmcsRevId.n.fIsShadowVmcs) { /* likely */ } else { Log(("%s: VMCS pointer %#RGp is a shadow VMCS -> VMFailInvalid\n", pszInstr, IEM_VMX_GET_CURRENT_VMCS(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_PtrShadowVmcs; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /** @todo Distinguish block-by-MovSS from block-by-STI. Currently we * use block-by-STI here which is not quite correct. */ if (!CPUMIsInInterruptShadowWithUpdate(&pVCpu->cpum.GstCtx)) { /* likely */ } else { Log(("%s: VM entry with events blocked by MOV SS -> VMFail\n", pszInstr)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_BlocKMovSS; iemVmxVmFail(pVCpu, VMXINSTRERR_VMENTRY_BLOCK_MOVSS); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } if (uInstrId == VMXINSTRID_VMLAUNCH) { /* VMLAUNCH with non-clear VMCS. */ if (pVmcs->fVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR) { /* likely */ } else if (pVmcs->fVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR_LEGACY) { /* Convert legacy launch-state value to current value, see @bugref{10318#c114} for reasons.*/ pVmcs->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR; Log(("vmlaunch: Updated legacy 'VMCLEAR' VMCS launch-state bit to current\n")); } else { Log(("vmlaunch: VMLAUNCH with non-clear VMCS %RGp -> VMFail\n", pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysVmcs)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_VmcsClear; iemVmxVmFail(pVCpu, VMXINSTRERR_VMLAUNCH_NON_CLEAR_VMCS); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } } else { /* VMRESUME with non-launched VMCS. */ if (pVmcs->fVmcsState == VMX_V_VMCS_LAUNCH_STATE_LAUNCHED) { /* likely */ } else { Log(("vmresume: VMRESUME with non-launched VMCS -> VMFail\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_VmcsLaunch; iemVmxVmFail(pVCpu, VMXINSTRERR_VMRESUME_NON_LAUNCHED_VMCS); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } } /* * We are allowed to cache VMCS related data structures (such as I/O bitmaps, MSR bitmaps) * while entering VMX non-root mode. We do some of this while checking VM-execution * controls. The nested hypervisor should not make assumptions and cannot expect * predictable behavior if changes to these structures are made in guest memory while * executing in VMX non-root mode. As far as VirtualBox is concerned, the guest cannot * modify them anyway as we cache them in host memory. * * See Intel spec. 24.11.4 "Software Access to Related Structures". */ Assert(IEM_VMX_HAS_CURRENT_VMCS(pVCpu)); int rc = iemVmxVmentryCheckCtls(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { rc = iemVmxVmentryCheckHostState(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { /* * Initialize read-only VMCS fields before VM-entry since we don't update all of them * for every VM-exit. This needs to be done before invoking a VM-exit (even those * ones that may occur during VM-entry below). */ iemVmxVmentryInitReadOnlyFields(pVCpu); /* * Blocking of NMIs need to be restored if VM-entry fails due to invalid-guest state. * So we save the VMCPU_FF_BLOCK_NMI force-flag here so we can restore it on * VM-exit when required. * See Intel spec. 26.7 "VM-entry Failures During or After Loading Guest State" */ iemVmxVmentrySaveNmiBlockingFF(pVCpu); rc = iemVmxVmentryCheckGuestState(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { /* * We've now entered nested-guest execution. * * It is important do this prior to loading the guest state because * as part of loading the guest state, PGM (and perhaps other components * in the future) relies on detecting whether VMX non-root mode has been * entered. */ pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxNonRootMode = true; rc = iemVmxVmentryLoadGuestState(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { rc = iemVmxVmentryLoadGuestAutoMsrs(pVCpu, pszInstr); if (RT_SUCCESS(rc)) { Assert(rc != VINF_CPUM_R3_MSR_WRITE); /* VMLAUNCH instruction must update the VMCS launch state. */ if (uInstrId == VMXINSTRID_VMLAUNCH) pVmcs->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_LAUNCHED; /* Perform the VMX transition (PGM updates). */ VBOXSTRICTRC rcStrict = iemVmxTransition(pVCpu, cbInstr); if (rcStrict == VINF_SUCCESS) { /* likely */ } else if (RT_SUCCESS(rcStrict)) { Log3(("%s: iemVmxTransition returns %Rrc -> Setting passup status\n", pszInstr, VBOXSTRICTRC_VAL(rcStrict))); rcStrict = iemSetPassUpStatus(pVCpu, rcStrict); } else { Log3(("%s: iemVmxTransition failed! rc=%Rrc\n", pszInstr, VBOXSTRICTRC_VAL(rcStrict))); return rcStrict; } /* Paranoia. */ Assert(rcStrict == VINF_SUCCESS); /* * The priority of potential VM-exits during VM-entry is important. * The priorities of VM-exits and events are listed from highest * to lowest as follows: * * 1. Event injection. * 2. Trap on task-switch (T flag set in TSS). * 3. TPR below threshold / APIC-write. * 4. SMI, INIT. * 5. MTF exit. * 6. Debug-trap exceptions (EFLAGS.TF), pending debug exceptions. * 7. VMX-preemption timer. * 9. NMI-window exit. * 10. NMI injection. * 11. Interrupt-window exit. * 12. Virtual-interrupt injection. * 13. Interrupt injection. * 14. Process next instruction (fetch, decode, execute). */ /* Setup VMX-preemption timer. */ iemVmxVmentrySetupPreemptTimer(pVCpu, pszInstr); /* Setup monitor-trap flag. */ iemVmxVmentrySetupMtf(pVCpu, pszInstr); /* Setup NMI-window exiting. */ iemVmxVmentrySetupNmiWindow(pVCpu, pszInstr); /* Setup interrupt-window exiting. */ iemVmxVmentrySetupIntWindow(pVCpu, pszInstr); /* * Inject any event that the nested hypervisor wants to inject. * Note! We cannot immediately perform the event injection here as we may have * pending PGM operations to perform due to switching page tables and/or * mode. */ iemVmxVmentryInjectEvent(pVCpu, pszInstr); # if defined(VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM) && defined(IN_RING3) /* Reschedule to IEM-only execution of the nested-guest. */ LogFlow(("%s: Enabling IEM-only EM execution policy!\n", pszInstr)); int rcSched = EMR3SetExecutionPolicy(pVCpu->CTX_SUFF(pVM)->pUVM, EMEXECPOLICY_IEM_ALL, true); if (rcSched != VINF_SUCCESS) iemSetPassUpStatus(pVCpu, rcSched); # endif /* Finally, done. */ Log2(("vmentry: %s: cs:rip=%04x:%08RX64 cr0=%#RX64 (%#RX64) cr4=%#RX64 (%#RX64) efer=%#RX64 (%#RX64)\n", pszInstr, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.cr0, pVmcs->u64Cr0ReadShadow.u, pVCpu->cpum.GstCtx.cr4, pVmcs->u64Cr4ReadShadow.u, pVCpu->cpum.GstCtx.msrEFER, pVmcs->u64GuestEferMsr.u)); return VINF_SUCCESS; } return iemVmxVmexit(pVCpu, VMX_EXIT_ERR_MSR_LOAD | VMX_EXIT_REASON_ENTRY_FAILED, pVmcs->u64RoExitQual.u); } } return iemVmxVmexit(pVCpu, VMX_EXIT_ERR_INVALID_GUEST_STATE | VMX_EXIT_REASON_ENTRY_FAILED, pVmcs->u64RoExitQual.u); } iemVmxVmFail(pVCpu, VMXINSTRERR_VMENTRY_INVALID_HOST_STATE); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } iemVmxVmFail(pVCpu, VMXINSTRERR_VMENTRY_INVALID_CTLS); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); # endif } /** * Interface for HM and EM to emulate the VMLAUNCH/VMRESUME instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param cbInstr The instruction length in bytes. * @param uInstrId The instruction ID (VMXINSTRID_VMLAUNCH or * VMXINSTRID_VMRESUME). * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedVmlaunchVmresume(PVMCPUCC pVCpu, uint8_t cbInstr, VMXINSTRID uInstrId) { IEMEXEC_ASSERT_INSTR_LEN_RETURN(cbInstr, 3); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK); iemInitExec(pVCpu, 0 /*fExecOpts*/); VBOXSTRICTRC rcStrict = iemVmxVmlaunchVmresume(pVCpu, cbInstr, uInstrId); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * Checks whether an RDMSR or WRMSR instruction for the given MSR is intercepted * (causes a VM-exit) or not. * * @returns @c true if the instruction is intercepted, @c false otherwise. * @param pVCpu The cross context virtual CPU structure. * @param uExitReason The VM-exit reason (VMX_EXIT_RDMSR or * VMX_EXIT_WRMSR). * @param idMsr The MSR. */ bool iemVmxIsRdmsrWrmsrInterceptSet(PCVMCPU pVCpu, uint32_t uExitReason, uint32_t idMsr) RT_NOEXCEPT { Assert(IEM_VMX_IS_NON_ROOT_MODE(pVCpu)); Assert( uExitReason == VMX_EXIT_RDMSR || uExitReason == VMX_EXIT_WRMSR); /* Consult the MSR bitmap if the feature is supported. */ PCVMXVVMCS const pVmcs = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs; Assert(pVmcs); if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS) { uint32_t const fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, idMsr); if (uExitReason == VMX_EXIT_RDMSR) return RT_BOOL(fMsrpm & VMXMSRPM_EXIT_RD); return RT_BOOL(fMsrpm & VMXMSRPM_EXIT_WR); } /* Without MSR bitmaps, all MSR accesses are intercepted. */ return true; } /** * VMREAD instruction execution worker that does not perform any validation checks. * * Callers are expected to have performed the necessary checks and to ensure the * VMREAD will succeed. * * @param pVmcs Pointer to the virtual VMCS. * @param pu64Dst Where to write the VMCS value. * @param u64VmcsField The VMCS field. * * @remarks May be called with interrupts disabled. */ static void iemVmxVmreadNoCheck(PCVMXVVMCS pVmcs, uint64_t *pu64Dst, uint64_t u64VmcsField) RT_NOEXCEPT { VMXVMCSFIELD VmcsField; VmcsField.u = u64VmcsField; uint8_t const uWidth = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_WIDTH); uint8_t const uType = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_TYPE); uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_INDEX); Assert(uIndex <= VMX_V_VMCS_MAX_INDEX); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; AssertMsg(offField < VMX_V_VMCS_SIZE, ("off=%u field=%#RX64 width=%#x type=%#x index=%#x (%u)\n", offField, u64VmcsField, uWidth, uType, uIndex, uIndex)); AssertCompile(VMX_V_SHADOW_VMCS_SIZE == VMX_V_VMCS_SIZE); /* * Read the VMCS component based on the field's effective width. * * The effective width is 64-bit fields adjusted to 32-bits if the access-type * indicates high bits (little endian). * * Note! The caller is responsible to trim the result and update registers * or memory locations are required. Here we just zero-extend to the largest * type (i.e. 64-bits). */ uint8_t const *pbVmcs = (uint8_t const *)pVmcs; uint8_t const *pbField = pbVmcs + offField; uint8_t const uEffWidth = VMXGetVmcsFieldWidthEff(VmcsField.u); switch (uEffWidth) { case VMX_VMCSFIELD_WIDTH_64BIT: case VMX_VMCSFIELD_WIDTH_NATURAL: *pu64Dst = *(uint64_t const *)pbField; break; case VMX_VMCSFIELD_WIDTH_32BIT: *pu64Dst = *(uint32_t const *)pbField; break; case VMX_VMCSFIELD_WIDTH_16BIT: *pu64Dst = *(uint16_t const *)pbField; break; } } /** * Interface for HM and EM to read a VMCS field from the nested-guest VMCS. * * It is ASSUMED the caller knows what they're doing. No VMREAD instruction checks * are performed. Bounds checks are strict builds only. * * @param pVmcs Pointer to the virtual VMCS. * @param u64VmcsField The VMCS field. * @param pu64Dst Where to store the VMCS value. * * @remarks May be called with interrupts disabled. * @todo This should probably be moved to CPUM someday. */ VMM_INT_DECL(void) IEMReadVmxVmcsField(PCVMXVVMCS pVmcs, uint64_t u64VmcsField, uint64_t *pu64Dst) { AssertPtr(pVmcs); AssertPtr(pu64Dst); iemVmxVmreadNoCheck(pVmcs, pu64Dst, u64VmcsField); } /** * VMREAD common (memory/register) instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param pu64Dst Where to write the VMCS value (only updated when * VINF_SUCCESS is returned). * @param u64VmcsField The VMCS field. * @param pExitInfo Pointer to the VM-exit information. Optional, can be * NULL. */ static VBOXSTRICTRC iemVmxVmreadCommon(PVMCPUCC pVCpu, uint8_t cbInstr, uint64_t *pu64Dst, uint64_t u64VmcsField, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { /* Nested-guest intercept. */ if ( IEM_VMX_IS_NON_ROOT_MODE(pVCpu) && CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, VMX_EXIT_VMREAD, u64VmcsField)) { if (pExitInfo) return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMREAD, VMXINSTRID_VMREAD, cbInstr); } /* CPL. */ if (IEM_GET_CPL(pVCpu) == 0) { /* likely */ } else { Log(("vmread: CPL %u -> #GP(0)\n", IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_Cpl; return iemRaiseGeneralProtectionFault0(pVCpu); } pVCpu->iem.s.cPotentialExits++; /* VMCS pointer in root mode. */ if ( !IEM_VMX_IS_ROOT_MODE(pVCpu) || IEM_VMX_HAS_CURRENT_VMCS(pVCpu)) { /* likely */ } else { Log(("vmread: VMCS pointer %#RGp invalid -> VMFailInvalid\n", IEM_VMX_GET_CURRENT_VMCS(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_PtrInvalid; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* VMCS-link pointer in non-root mode. */ if ( !IEM_VMX_IS_NON_ROOT_MODE(pVCpu) || IEM_VMX_HAS_SHADOW_VMCS(pVCpu)) { /* likely */ } else { Log(("vmread: VMCS-link pointer %#RGp invalid -> VMFailInvalid\n", IEM_VMX_GET_SHADOW_VMCS(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_LinkPtrInvalid; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Supported VMCS field. */ if (CPUMIsGuestVmxVmcsFieldValid(pVCpu->CTX_SUFF(pVM), u64VmcsField)) { /* likely */ } else { Log(("vmread: VMCS field %#RX64 invalid -> VMFail\n", u64VmcsField)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_FieldInvalid; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64VmcsField; iemVmxVmFail(pVCpu, VMXINSTRERR_VMREAD_INVALID_COMPONENT); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* * Reading from the current or shadow VMCS. */ PCVMXVVMCS pVmcs = !IEM_VMX_IS_NON_ROOT_MODE(pVCpu) ? &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs : &pVCpu->cpum.GstCtx.hwvirt.vmx.ShadowVmcs; iemVmxVmreadNoCheck(pVmcs, pu64Dst, u64VmcsField); Log4(("vmread %#RX64 => %#RX64\n", u64VmcsField, *pu64Dst)); return VINF_SUCCESS; } /** * VMREAD (64-bit register) instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param pu64Dst Where to store the VMCS field's value. * @param u64VmcsField The VMCS field. * @param pExitInfo Pointer to the VM-exit information. Optional, can be * NULL. */ static VBOXSTRICTRC iemVmxVmreadReg64(PVMCPUCC pVCpu, uint8_t cbInstr, uint64_t *pu64Dst, uint64_t u64VmcsField, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { VBOXSTRICTRC rcStrict = iemVmxVmreadCommon(pVCpu, cbInstr, pu64Dst, u64VmcsField, pExitInfo); if (rcStrict == VINF_SUCCESS) { iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } Log(("vmread/reg: iemVmxVmreadCommon failed rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict))); return rcStrict; } /** * VMREAD (32-bit register) instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param pu32Dst Where to store the VMCS field's value. * @param u32VmcsField The VMCS field. * @param pExitInfo Pointer to the VM-exit information. Optional, can be * NULL. */ static VBOXSTRICTRC iemVmxVmreadReg32(PVMCPUCC pVCpu, uint8_t cbInstr, uint32_t *pu32Dst, uint64_t u32VmcsField, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { uint64_t u64Dst; VBOXSTRICTRC rcStrict = iemVmxVmreadCommon(pVCpu, cbInstr, &u64Dst, u32VmcsField, pExitInfo); if (rcStrict == VINF_SUCCESS) { *pu32Dst = u64Dst; iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } Log(("vmread/reg: iemVmxVmreadCommon failed rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict))); return rcStrict; } /** * VMREAD (memory) instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param iEffSeg The effective segment register to use with @a u64Val. * Pass UINT8_MAX if it is a register access. * @param GCPtrDst The guest linear address to store the VMCS field's * value. * @param u64VmcsField The VMCS field. * @param pExitInfo Pointer to the VM-exit information. Optional, can be * NULL. */ static VBOXSTRICTRC iemVmxVmreadMem(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPTR GCPtrDst, uint64_t u64VmcsField, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { uint64_t u64Dst; VBOXSTRICTRC rcStrict = iemVmxVmreadCommon(pVCpu, cbInstr, &u64Dst, u64VmcsField, pExitInfo); if (rcStrict == VINF_SUCCESS) { /* * Write the VMCS field's value to the location specified in guest-memory. */ if (IEM_IS_64BIT_CODE(pVCpu)) rcStrict = iemMemStoreDataU64(pVCpu, iEffSeg, GCPtrDst, u64Dst); else rcStrict = iemMemStoreDataU32(pVCpu, iEffSeg, GCPtrDst, u64Dst); if (rcStrict == VINF_SUCCESS) { iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } Log(("vmread/mem: Failed to write to memory operand at %#RGv, rc=%Rrc\n", GCPtrDst, VBOXSTRICTRC_VAL(rcStrict))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_PtrMap; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrDst; return rcStrict; } Log(("vmread/mem: iemVmxVmreadCommon failed rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict))); return rcStrict; } /** * Interface for HM and EM to emulate the VMREAD instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedVmread(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { IEMEXEC_ASSERT_INSTR_LEN_RETURN(pExitInfo->cbInstr, 3); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); Assert(pExitInfo); iemInitExec(pVCpu, 0 /*fExecOpts*/); VBOXSTRICTRC rcStrict; uint8_t const cbInstr = pExitInfo->cbInstr; bool const fIs64BitMode = RT_BOOL(IEM_IS_64BIT_CODE(pVCpu)); uint64_t const u64FieldEnc = fIs64BitMode ? iemGRegFetchU64(pVCpu, pExitInfo->InstrInfo.VmreadVmwrite.iReg2) : iemGRegFetchU32(pVCpu, pExitInfo->InstrInfo.VmreadVmwrite.iReg2); if (pExitInfo->InstrInfo.VmreadVmwrite.fIsRegOperand) { if (fIs64BitMode) { uint64_t *pu64Dst = iemGRegRefU64(pVCpu, pExitInfo->InstrInfo.VmreadVmwrite.iReg1); rcStrict = iemVmxVmreadReg64(pVCpu, cbInstr, pu64Dst, u64FieldEnc, pExitInfo); } else { uint32_t *pu32Dst = iemGRegRefU32(pVCpu, pExitInfo->InstrInfo.VmreadVmwrite.iReg1); rcStrict = iemVmxVmreadReg32(pVCpu, cbInstr, pu32Dst, u64FieldEnc, pExitInfo); } } else { RTGCPTR const GCPtrDst = pExitInfo->GCPtrEffAddr; uint8_t const iEffSeg = pExitInfo->InstrInfo.VmreadVmwrite.iSegReg; rcStrict = iemVmxVmreadMem(pVCpu, cbInstr, iEffSeg, GCPtrDst, u64FieldEnc, pExitInfo); } Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * VMWRITE instruction execution worker that does not perform any validation * checks. * * Callers are expected to have performed the necessary checks and to ensure the * VMWRITE will succeed. * * @param pVmcs Pointer to the virtual VMCS. * @param u64Val The value to write. * @param u64VmcsField The VMCS field. * * @remarks May be called with interrupts disabled. */ static void iemVmxVmwriteNoCheck(PVMXVVMCS pVmcs, uint64_t u64Val, uint64_t u64VmcsField) RT_NOEXCEPT { VMXVMCSFIELD VmcsField; VmcsField.u = u64VmcsField; uint8_t const uWidth = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_WIDTH); uint8_t const uType = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_TYPE); uint8_t const uWidthType = (uWidth << 2) | uType; uint8_t const uIndex = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_INDEX); Assert(uIndex <= VMX_V_VMCS_MAX_INDEX); uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex]; Assert(offField < VMX_V_VMCS_SIZE); AssertCompile(VMX_V_SHADOW_VMCS_SIZE == VMX_V_VMCS_SIZE); /* * Write the VMCS component based on the field's effective width. * * The effective width is 64-bit fields adjusted to 32-bits if the access-type * indicates high bits (little endian). */ uint8_t *pbVmcs = (uint8_t *)pVmcs; uint8_t *pbField = pbVmcs + offField; uint8_t const uEffWidth = VMXGetVmcsFieldWidthEff(VmcsField.u); switch (uEffWidth) { case VMX_VMCSFIELD_WIDTH_64BIT: case VMX_VMCSFIELD_WIDTH_NATURAL: *(uint64_t *)pbField = u64Val; break; case VMX_VMCSFIELD_WIDTH_32BIT: *(uint32_t *)pbField = u64Val; break; case VMX_VMCSFIELD_WIDTH_16BIT: *(uint16_t *)pbField = u64Val; break; } } /** * Interface for HM and EM to write a VMCS field in the nested-guest VMCS. * * It is ASSUMED the caller knows what they're doing. No VMWRITE instruction checks * are performed. Bounds checks are strict builds only. * * @param pVmcs Pointer to the virtual VMCS. * @param u64VmcsField The VMCS field. * @param u64Val The value to write. * * @remarks May be called with interrupts disabled. * @todo This should probably be moved to CPUM someday. */ VMM_INT_DECL(void) IEMWriteVmxVmcsField(PVMXVVMCS pVmcs, uint64_t u64VmcsField, uint64_t u64Val) { AssertPtr(pVmcs); iemVmxVmwriteNoCheck(pVmcs, u64Val, u64VmcsField); } /** * VMWRITE instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param iEffSeg The effective segment register to use with @a u64Val. * Pass UINT8_MAX if it is a register access. * @param u64Val The value to write (or guest linear address to the * value), @a iEffSeg will indicate if it's a memory * operand. * @param u64VmcsField The VMCS field. * @param pExitInfo Pointer to the VM-exit information. Optional, can be * NULL. */ static VBOXSTRICTRC iemVmxVmwrite(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, uint64_t u64Val, uint64_t u64VmcsField, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { /* Nested-guest intercept. */ if ( IEM_VMX_IS_NON_ROOT_MODE(pVCpu) && CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, VMX_EXIT_VMWRITE, u64VmcsField)) { if (pExitInfo) return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMWRITE, VMXINSTRID_VMWRITE, cbInstr); } /* CPL. */ if (IEM_GET_CPL(pVCpu) == 0) { /* likely */ } else { Log(("vmwrite: CPL %u -> #GP(0)\n", IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_Cpl; return iemRaiseGeneralProtectionFault0(pVCpu); } pVCpu->iem.s.cPotentialExits++; /* VMCS pointer in root mode. */ if ( !IEM_VMX_IS_ROOT_MODE(pVCpu) || IEM_VMX_HAS_CURRENT_VMCS(pVCpu)) { /* likely */ } else { Log(("vmwrite: VMCS pointer %#RGp invalid -> VMFailInvalid\n", IEM_VMX_GET_CURRENT_VMCS(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_PtrInvalid; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* VMCS-link pointer in non-root mode. */ if ( !IEM_VMX_IS_NON_ROOT_MODE(pVCpu) || IEM_VMX_HAS_SHADOW_VMCS(pVCpu)) { /* likely */ } else { Log(("vmwrite: VMCS-link pointer %#RGp invalid -> VMFailInvalid\n", IEM_VMX_GET_SHADOW_VMCS(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_LinkPtrInvalid; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* If the VMWRITE instruction references memory, access the specified memory operand. */ bool const fIsRegOperand = iEffSeg == UINT8_MAX; if (!fIsRegOperand) { /* Read the value from the specified guest memory location. */ VBOXSTRICTRC rcStrict; RTGCPTR const GCPtrVal = u64Val; if (IEM_IS_64BIT_CODE(pVCpu)) rcStrict = iemMemFetchDataU64(pVCpu, &u64Val, iEffSeg, GCPtrVal); else rcStrict = iemMemFetchDataU32_ZX_U64(pVCpu, &u64Val, iEffSeg, GCPtrVal); if (RT_UNLIKELY(rcStrict != VINF_SUCCESS)) { Log(("vmwrite: Failed to read value from memory operand at %#RGv, rc=%Rrc\n", GCPtrVal, VBOXSTRICTRC_VAL(rcStrict))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_PtrMap; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVal; return rcStrict; } } else Assert(!pExitInfo || pExitInfo->InstrInfo.VmreadVmwrite.fIsRegOperand); /* Supported VMCS field. */ if (CPUMIsGuestVmxVmcsFieldValid(pVCpu->CTX_SUFF(pVM), u64VmcsField)) { /* likely */ } else { Log(("vmwrite: VMCS field %#RX64 invalid -> VMFail\n", u64VmcsField)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_FieldInvalid; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64VmcsField; iemVmxVmFail(pVCpu, VMXINSTRERR_VMWRITE_INVALID_COMPONENT); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Read-only VMCS field. */ bool const fIsFieldReadOnly = VMXIsVmcsFieldReadOnly(u64VmcsField); if ( !fIsFieldReadOnly || IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxVmwriteAll) { /* likely */ } else { Log(("vmwrite: Write to read-only VMCS component %#RX64 -> VMFail\n", u64VmcsField)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_FieldRo; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64VmcsField; iemVmxVmFail(pVCpu, VMXINSTRERR_VMWRITE_RO_COMPONENT); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* * Write to the current or shadow VMCS. */ bool const fInVmxNonRootMode = IEM_VMX_IS_NON_ROOT_MODE(pVCpu); PVMXVVMCS pVmcs = !fInVmxNonRootMode ? &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs : &pVCpu->cpum.GstCtx.hwvirt.vmx.ShadowVmcs; iemVmxVmwriteNoCheck(pVmcs, u64Val, u64VmcsField); Log4(("vmwrite %#RX64 <= %#RX64\n", u64VmcsField, u64Val)); if ( !fInVmxNonRootMode && VM_IS_HM_ENABLED(pVCpu->CTX_SUFF(pVM))) { /* Notify HM that the VMCS content might have changed. */ HMNotifyVmxNstGstCurrentVmcsChanged(pVCpu); } iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /** * Interface for HM and EM to emulate the VMWRITE instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedVmwrite(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { IEMEXEC_ASSERT_INSTR_LEN_RETURN(pExitInfo->cbInstr, 3); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); Assert(pExitInfo); iemInitExec(pVCpu, 0 /*fExecOpts*/); uint64_t u64Val; uint8_t iEffSeg; if (pExitInfo->InstrInfo.VmreadVmwrite.fIsRegOperand) { u64Val = iemGRegFetchU64(pVCpu, pExitInfo->InstrInfo.VmreadVmwrite.iReg1); iEffSeg = UINT8_MAX; } else { u64Val = pExitInfo->GCPtrEffAddr; iEffSeg = pExitInfo->InstrInfo.VmreadVmwrite.iSegReg; } uint8_t const cbInstr = pExitInfo->cbInstr; uint64_t const u64FieldEnc = IEM_IS_64BIT_CODE(pVCpu) ? iemGRegFetchU64(pVCpu, pExitInfo->InstrInfo.VmreadVmwrite.iReg2) : iemGRegFetchU32(pVCpu, pExitInfo->InstrInfo.VmreadVmwrite.iReg2); VBOXSTRICTRC rcStrict = iemVmxVmwrite(pVCpu, cbInstr, iEffSeg, u64Val, u64FieldEnc, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * VMCLEAR instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param iEffSeg The effective segment register to use with @a GCPtrVmcs. * @param GCPtrVmcs The linear address of the VMCS pointer. * @param pExitInfo Pointer to the VM-exit information. Optional, can be NULL. * * @remarks Common VMX instruction checks are already expected to by the caller, * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks. */ static VBOXSTRICTRC iemVmxVmclear(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPHYS GCPtrVmcs, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { /* Nested-guest intercept. */ if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) { if (pExitInfo) return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMCLEAR, VMXINSTRID_NONE, cbInstr); } Assert(IEM_VMX_IS_ROOT_MODE(pVCpu)); /* CPL. */ if (IEM_GET_CPL(pVCpu) == 0) { /* likely */ } else { Log(("vmclear: CPL %u -> #GP(0)\n", IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_Cpl; return iemRaiseGeneralProtectionFault0(pVCpu); } /* Get the VMCS pointer from the location specified by the source memory operand. */ RTGCPHYS GCPhysVmcs; VBOXSTRICTRC rcStrict = iemMemFetchDataU64(pVCpu, &GCPhysVmcs, iEffSeg, GCPtrVmcs); if (RT_LIKELY(rcStrict == VINF_SUCCESS)) { /* likely */ } else { Log(("vmclear: Failed to read VMCS physaddr from %#RGv, rc=%Rrc\n", GCPtrVmcs, VBOXSTRICTRC_VAL(rcStrict))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrMap; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVmcs; return rcStrict; } /* VMCS pointer alignment. */ if (!(GCPhysVmcs & X86_PAGE_4K_OFFSET_MASK)) { /* likely */ } else { Log(("vmclear: VMCS pointer not page-aligned -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrAlign; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; iemVmxVmFail(pVCpu, VMXINSTRERR_VMCLEAR_INVALID_PHYSADDR); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* VMCS physical-address width limits. */ if (!(GCPhysVmcs >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)) { /* likely */ } else { Log(("vmclear: VMCS pointer extends beyond physical-address width -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrWidth; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; iemVmxVmFail(pVCpu, VMXINSTRERR_VMCLEAR_INVALID_PHYSADDR); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* VMCS is not the VMXON region. */ if (GCPhysVmcs != pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysVmxon) { /* likely */ } else { Log(("vmclear: VMCS pointer cannot be identical to VMXON region pointer -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrVmxon; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; iemVmxVmFail(pVCpu, VMXINSTRERR_VMCLEAR_VMXON_PTR); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Ensure VMCS is not MMIO, ROM etc. This is not an Intel requirement but a restriction imposed by our implementation. */ if (PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmcs)) { /* likely */ } else { Log(("vmclear: VMCS not normal memory -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrAbnormal; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; iemVmxVmFail(pVCpu, VMXINSTRERR_VMCLEAR_INVALID_PHYSADDR); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* * VMCLEAR allows committing and clearing any valid VMCS pointer. * * If the current VMCS is the one being cleared, set its state to 'clear' and commit * to guest memory. Otherwise, set the state of the VMCS referenced in guest memory * to 'clear'. */ uint8_t const fVmcsLaunchStateClear = VMX_V_VMCS_LAUNCH_STATE_CLEAR; if ( IEM_VMX_HAS_CURRENT_VMCS(pVCpu) && IEM_VMX_GET_CURRENT_VMCS(pVCpu) == GCPhysVmcs) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.fVmcsState = fVmcsLaunchStateClear; iemVmxWriteCurrentVmcsToGstMem(pVCpu); IEM_VMX_CLEAR_CURRENT_VMCS(pVCpu); } else { AssertCompileMemberSize(VMXVVMCS, fVmcsState, sizeof(fVmcsLaunchStateClear)); rcStrict = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVmcs + RT_UOFFSETOF(VMXVVMCS, fVmcsState), (const void *)&fVmcsLaunchStateClear, sizeof(fVmcsLaunchStateClear)); if (RT_FAILURE(rcStrict)) return rcStrict; } iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /** * Interface for HM and EM to emulate the VMCLEAR instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedVmclear(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { Assert(pExitInfo); IEMEXEC_ASSERT_INSTR_LEN_RETURN(pExitInfo->cbInstr, 3); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); iemInitExec(pVCpu, 0 /*fExecOpts*/); uint8_t const iEffSeg = pExitInfo->InstrInfo.VmxXsave.iSegReg; uint8_t const cbInstr = pExitInfo->cbInstr; RTGCPTR const GCPtrVmcs = pExitInfo->GCPtrEffAddr; VBOXSTRICTRC rcStrict = iemVmxVmclear(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * VMPTRST instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param iEffSeg The effective segment register to use with @a GCPtrVmcs. * @param GCPtrVmcs The linear address of where to store the current VMCS * pointer. * @param pExitInfo Pointer to the VM-exit information. Optional, can be NULL. * * @remarks Common VMX instruction checks are already expected to by the caller, * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks. */ static VBOXSTRICTRC iemVmxVmptrst(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPHYS GCPtrVmcs, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { /* Nested-guest intercept. */ if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) { if (pExitInfo) return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMPTRST, VMXINSTRID_NONE, cbInstr); } Assert(IEM_VMX_IS_ROOT_MODE(pVCpu)); /* CPL. */ if (IEM_GET_CPL(pVCpu) == 0) { /* likely */ } else { Log(("vmptrst: CPL %u -> #GP(0)\n", IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrst_Cpl; return iemRaiseGeneralProtectionFault0(pVCpu); } /* Set the VMCS pointer to the location specified by the destination memory operand. */ AssertCompile(NIL_RTGCPHYS == ~(RTGCPHYS)0U); VBOXSTRICTRC rcStrict = iemMemStoreDataU64(pVCpu, iEffSeg, GCPtrVmcs, IEM_VMX_GET_CURRENT_VMCS(pVCpu)); if (RT_LIKELY(rcStrict == VINF_SUCCESS)) { iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } Log(("vmptrst: Failed to store VMCS pointer to memory at destination operand %#Rrc\n", VBOXSTRICTRC_VAL(rcStrict))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrst_PtrMap; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVmcs; return rcStrict; } /** * Interface for HM and EM to emulate the VMPTRST instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedVmptrst(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { Assert(pExitInfo); IEMEXEC_ASSERT_INSTR_LEN_RETURN(pExitInfo->cbInstr, 3); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); iemInitExec(pVCpu, 0 /*fExecOpts*/); uint8_t const iEffSeg = pExitInfo->InstrInfo.VmxXsave.iSegReg; uint8_t const cbInstr = pExitInfo->cbInstr; RTGCPTR const GCPtrVmcs = pExitInfo->GCPtrEffAddr; VBOXSTRICTRC rcStrict = iemVmxVmptrst(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * VMPTRLD instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param GCPtrVmcs The linear address of the current VMCS pointer. * @param pExitInfo Pointer to the VM-exit information. Optional, can be NULL. * * @remarks Common VMX instruction checks are already expected to by the caller, * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks. */ static VBOXSTRICTRC iemVmxVmptrld(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPHYS GCPtrVmcs, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { /* Nested-guest intercept. */ if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) { if (pExitInfo) return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMPTRLD, VMXINSTRID_NONE, cbInstr); } Assert(IEM_VMX_IS_ROOT_MODE(pVCpu)); /* CPL. */ if (IEM_GET_CPL(pVCpu) == 0) { /* likely */ } else { Log(("vmptrld: CPL %u -> #GP(0)\n", IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_Cpl; return iemRaiseGeneralProtectionFault0(pVCpu); } /* Get the VMCS pointer from the location specified by the source memory operand. */ RTGCPHYS GCPhysVmcs; VBOXSTRICTRC rcStrict = iemMemFetchDataU64(pVCpu, &GCPhysVmcs, iEffSeg, GCPtrVmcs); if (RT_LIKELY(rcStrict == VINF_SUCCESS)) { /* likely */ } else { Log(("vmptrld: Failed to read VMCS physaddr from %#RGv, rc=%Rrc\n", GCPtrVmcs, VBOXSTRICTRC_VAL(rcStrict))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrMap; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVmcs; return rcStrict; } /* VMCS pointer alignment. */ if (!(GCPhysVmcs & X86_PAGE_4K_OFFSET_MASK)) { /* likely */ } else { Log(("vmptrld: VMCS pointer not page-aligned -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrAlign; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INVALID_PHYSADDR); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* VMCS physical-address width limits. */ if (!(GCPhysVmcs >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)) { /* likely */ } else { Log(("vmptrld: VMCS pointer extends beyond physical-address width -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrWidth; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INVALID_PHYSADDR); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* VMCS is not the VMXON region. */ if (GCPhysVmcs != pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysVmxon) { /* likely */ } else { Log(("vmptrld: VMCS pointer cannot be identical to VMXON region pointer -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrVmxon; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_VMXON_PTR); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Ensure VMCS is not MMIO, ROM etc. This is not an Intel requirement but a restriction imposed by our implementation. */ if (PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmcs)) { /* likely */ } else { Log(("vmptrld: VMCS not normal memory -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrAbnormal; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INVALID_PHYSADDR); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Read just the VMCS revision from the VMCS. */ VMXVMCSREVID VmcsRevId; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &VmcsRevId, GCPhysVmcs, sizeof(VmcsRevId)); if (RT_SUCCESS(rc)) { /* likely */ } else { Log(("vmptrld: Failed to read revision identifier from VMCS at %#RGp, rc=%Rrc\n", GCPhysVmcs, rc)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_RevPtrReadPhys; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; return rc; } /* * Verify the VMCS revision specified by the guest matches what we reported to the guest. * Verify the VMCS is not a shadow VMCS, if the VMCS shadowing feature is supported. */ if ( VmcsRevId.n.u31RevisionId == VMX_V_VMCS_REVISION_ID && ( !VmcsRevId.n.fIsShadowVmcs || IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxVmcsShadowing)) { /* likely */ } else { if (VmcsRevId.n.u31RevisionId != VMX_V_VMCS_REVISION_ID) { Log(("vmptrld: VMCS revision mismatch, expected %#RX32 got %#RX32, GCPtrVmcs=%#RGv GCPhysVmcs=%#RGp -> VMFail()\n", VMX_V_VMCS_REVISION_ID, VmcsRevId.n.u31RevisionId, GCPtrVmcs, GCPhysVmcs)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_VmcsRevId; } else { Log(("vmptrld: Shadow VMCS -> VMFail()\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_ShadowVmcs; } iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INCORRECT_VMCS_REV); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* * We cache only the current VMCS in CPUMCTX. Therefore, VMPTRLD should always flush * the cache of an existing, current VMCS back to guest memory before loading a new, * different current VMCS. */ if (IEM_VMX_GET_CURRENT_VMCS(pVCpu) != GCPhysVmcs) { if (IEM_VMX_HAS_CURRENT_VMCS(pVCpu)) { iemVmxWriteCurrentVmcsToGstMem(pVCpu); IEM_VMX_CLEAR_CURRENT_VMCS(pVCpu); } /* Set the new VMCS as the current VMCS and read it from guest memory. */ IEM_VMX_SET_CURRENT_VMCS(pVCpu, GCPhysVmcs); rc = iemVmxReadCurrentVmcsFromGstMem(pVCpu); if (RT_SUCCESS(rc)) { /* Notify HM that a new, current VMCS is loaded. */ if (VM_IS_HM_ENABLED(pVCpu->CTX_SUFF(pVM))) HMNotifyVmxNstGstCurrentVmcsChanged(pVCpu); /* Convert legacy launch-state value to current value, see @bugref{10318#c114} for reasons.*/ if (pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.fVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR_LEGACY) { pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR; Log(("vmptrld: Updated legacy 'VMCLEAR' VMCS launch-state bit to current\n")); } } else { Log(("vmptrld: Failed to read VMCS at %#RGp, rc=%Rrc\n", GCPhysVmcs, rc)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrReadPhys; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs; return rc; } } else if ( IEM_VMX_HAS_CURRENT_VMCS(pVCpu) && pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.fVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR_LEGACY) { /* Convert legacy launch-state value to current value, see @bugref{10318#c114} for reasons.*/ pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR; Log(("vmptrld: Updated legacy VMCLEAR launch-state bit to current\n")); } Assert(IEM_VMX_HAS_CURRENT_VMCS(pVCpu)); iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /** * Interface for HM and EM to emulate the VMPTRLD instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedVmptrld(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { Assert(pExitInfo); IEMEXEC_ASSERT_INSTR_LEN_RETURN(pExitInfo->cbInstr, 3); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); iemInitExec(pVCpu, 0 /*fExecOpts*/); uint8_t const iEffSeg = pExitInfo->InstrInfo.VmxXsave.iSegReg; uint8_t const cbInstr = pExitInfo->cbInstr; RTGCPTR const GCPtrVmcs = pExitInfo->GCPtrEffAddr; VBOXSTRICTRC rcStrict = iemVmxVmptrld(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * INVVPID instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param iEffSeg The segment of the invvpid descriptor. * @param GCPtrInvvpidDesc The address of invvpid descriptor. * @param u64InvvpidType The invalidation type. * @param pExitInfo Pointer to the VM-exit information. Optional, can be * NULL. * * @remarks Common VMX instruction checks are already expected to by the caller, * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks. */ VBOXSTRICTRC iemVmxInvvpid(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPTR GCPtrInvvpidDesc, uint64_t u64InvvpidType, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { /* Check if INVVPID instruction is supported, otherwise raise #UD. */ if (!IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxVpid) return iemRaiseUndefinedOpcode(pVCpu); /* Nested-guest intercept. */ if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) { if (pExitInfo) return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_INVVPID, VMXINSTRID_NONE, cbInstr); } /* CPL. */ if (IEM_GET_CPL(pVCpu) != 0) { Log(("invvpid: CPL != 0 -> #GP(0)\n")); return iemRaiseGeneralProtectionFault0(pVCpu); } /* * Validate INVVPID invalidation type. * * The instruction specifies exactly ONE of the supported invalidation types. * * Each of the types has a bit in IA32_VMX_EPT_VPID_CAP MSR specifying if it is * supported. In theory, it's possible for a CPU to not support flushing individual * addresses but all the other types or any other combination. We do not take any * shortcuts here by assuming the types we currently expose to the guest. */ uint64_t const fCaps = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64EptVpidCaps; bool const fInvvpidSupported = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID); bool const fTypeIndivAddr = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID_INDIV_ADDR); bool const fTypeSingleCtx = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID_SINGLE_CTX); bool const fTypeAllCtx = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID_ALL_CTX); bool const fTypeSingleCtxRetainGlobals = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID_SINGLE_CTX_RETAIN_GLOBALS); bool afSupportedTypes[4]; afSupportedTypes[0] = fTypeIndivAddr; afSupportedTypes[1] = fTypeSingleCtx; afSupportedTypes[2] = fTypeAllCtx; afSupportedTypes[3] = fTypeSingleCtxRetainGlobals; if ( fInvvpidSupported && !(u64InvvpidType & ~(uint64_t)VMX_INVVPID_VALID_MASK) && afSupportedTypes[u64InvvpidType & 3]) { /* likely */ } else { Log(("invvpid: invalid/unsupported invvpid type %#x -> VMFail\n", u64InvvpidType)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_TypeInvalid; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64InvvpidType; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* * Fetch the invvpid descriptor from guest memory. */ RTUINT128U uDesc; VBOXSTRICTRC rcStrict = iemMemFetchDataU128(pVCpu, &uDesc, iEffSeg, GCPtrInvvpidDesc); if (rcStrict == VINF_SUCCESS) { /* * Validate the descriptor. */ if (uDesc.s.Lo <= 0xffff) { /* likely */ } else { Log(("invvpid: reserved bits set in invvpid descriptor %#RX64 -> #GP(0)\n", uDesc.s.Lo)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_DescRsvd; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = uDesc.s.Lo; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3); RTGCUINTPTR64 const GCPtrInvAddr = uDesc.s.Hi; uint16_t const uVpid = uDesc.Words.w0; uint64_t const uCr3 = pVCpu->cpum.GstCtx.cr3; switch (u64InvvpidType) { case VMXTLBFLUSHVPID_INDIV_ADDR: { if (uVpid != 0) { if (IEM_IS_CANONICAL(GCPtrInvAddr)) { /* Invalidate mappings for the linear address tagged with VPID. */ /** @todo PGM support for VPID? Currently just flush everything. */ PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */); iemVmxVmSucceed(pVCpu); } else { Log(("invvpid: invalidation address %#RGP is not canonical -> VMFail\n", GCPtrInvAddr)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_Type0InvalidAddr; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrInvAddr; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); } } else { Log(("invvpid: invalid VPID %#x for invalidation type %u -> VMFail\n", uVpid, u64InvvpidType)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_Type0InvalidVpid; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64InvvpidType; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); } break; } case VMXTLBFLUSHVPID_SINGLE_CONTEXT: { if (uVpid != 0) { /* Invalidate all mappings with VPID. */ /** @todo PGM support for VPID? Currently just flush everything. */ PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */); iemVmxVmSucceed(pVCpu); } else { Log(("invvpid: invalid VPID %#x for invalidation type %u -> VMFail\n", uVpid, u64InvvpidType)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_Type1InvalidVpid; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64InvvpidType; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); } break; } case VMXTLBFLUSHVPID_ALL_CONTEXTS: { /* Invalidate all mappings with non-zero VPIDs. */ /** @todo PGM support for VPID? Currently just flush everything. */ PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */); iemVmxVmSucceed(pVCpu); break; } case VMXTLBFLUSHVPID_SINGLE_CONTEXT_RETAIN_GLOBALS: { if (uVpid != 0) { /* Invalidate all mappings with VPID except global translations. */ /** @todo PGM support for VPID? Currently just flush everything. */ PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */); iemVmxVmSucceed(pVCpu); } else { Log(("invvpid: invalid VPID %#x for invalidation type %u -> VMFail\n", uVpid, u64InvvpidType)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_Type3InvalidVpid; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = uVpid; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); } break; } IEM_NOT_REACHED_DEFAULT_CASE_RET(); } rcStrict = iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } return rcStrict; } /** * Interface for HM and EM to emulate the INVVPID instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedInvvpid(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { IEMEXEC_ASSERT_INSTR_LEN_RETURN(pExitInfo->cbInstr, 4); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); Assert(pExitInfo); iemInitExec(pVCpu, 0 /*fExecOpts*/); uint8_t const iEffSeg = pExitInfo->InstrInfo.Inv.iSegReg; uint8_t const cbInstr = pExitInfo->cbInstr; RTGCPTR const GCPtrInvvpidDesc = pExitInfo->GCPtrEffAddr; uint64_t const u64InvvpidType = IEM_IS_64BIT_CODE(pVCpu) ? iemGRegFetchU64(pVCpu, pExitInfo->InstrInfo.Inv.iReg2) : iemGRegFetchU32(pVCpu, pExitInfo->InstrInfo.Inv.iReg2); VBOXSTRICTRC rcStrict = iemVmxInvvpid(pVCpu, cbInstr, iEffSeg, GCPtrInvvpidDesc, u64InvvpidType, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } #ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT /** * INVEPT instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param iEffSeg The segment of the invept descriptor. * @param GCPtrInveptDesc The address of invept descriptor. * @param u64InveptType The invalidation type. * @param pExitInfo Pointer to the VM-exit information. Optional, can be * NULL. * * @remarks Common VMX instruction checks are already expected to by the caller, * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks. */ static VBOXSTRICTRC iemVmxInvept(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPTR GCPtrInveptDesc, uint64_t u64InveptType, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { /* Check if EPT is supported, otherwise raise #UD. */ if (!IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxEpt) return iemRaiseUndefinedOpcode(pVCpu); /* Nested-guest intercept. */ if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) { if (pExitInfo) return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_INVEPT, VMXINSTRID_NONE, cbInstr); } /* CPL. */ if (IEM_GET_CPL(pVCpu) != 0) { Log(("invept: CPL != 0 -> #GP(0)\n")); return iemRaiseGeneralProtectionFault0(pVCpu); } /* * Validate INVEPT invalidation type. * * The instruction specifies exactly ONE of the supported invalidation types. * * Each of the types has a bit in IA32_VMX_EPT_VPID_CAP MSR specifying if it is * supported. In theory, it's possible for a CPU to not support flushing individual * addresses but all the other types or any other combination. We do not take any * shortcuts here by assuming the types we currently expose to the guest. */ uint64_t const fCaps = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64EptVpidCaps; bool const fInveptSupported = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVEPT); bool const fTypeSingleCtx = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVEPT_SINGLE_CTX); bool const fTypeAllCtx = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVEPT_ALL_CTX); bool afSupportedTypes[4]; afSupportedTypes[0] = false; afSupportedTypes[1] = fTypeSingleCtx; afSupportedTypes[2] = fTypeAllCtx; afSupportedTypes[3] = false; if ( fInveptSupported && !(u64InveptType & ~(uint64_t)VMX_INVEPT_VALID_MASK) && afSupportedTypes[u64InveptType & 3]) { /* likely */ } else { Log(("invept: invalid/unsupported invvpid type %#x -> VMFail\n", u64InveptType)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invept_TypeInvalid; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64InveptType; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* * Fetch the invept descriptor from guest memory. */ RTUINT128U uDesc; VBOXSTRICTRC rcStrict = iemMemFetchDataU128(pVCpu, &uDesc, iEffSeg, GCPtrInveptDesc); if (rcStrict == VINF_SUCCESS) { /* * Validate the descriptor. * * The Intel spec. does not explicit say the INVEPT instruction fails when reserved * bits in the descriptor are set, but it -does- for INVVPID. Until we test on real * hardware, it's assumed INVEPT behaves the same as INVVPID in this regard. It's * better to be strict in our emulation until proven otherwise. * * UPDATE: Hyper-V enabled Windows 10 Pro guests do NOT clear the reserved bits in * the descriptor. Hence, I've disabled this check for now, see @bugref{10318#c122}. */ #if 0 if (uDesc.s.Hi) { Log(("invept: reserved bits set in invept descriptor %#RX64 -> VMFail\n", uDesc.s.Hi)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invept_DescRsvd; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = uDesc.s.Hi; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } #endif /* * Flush TLB mappings based on the EPT type. */ if (u64InveptType == VMXTLBFLUSHEPT_SINGLE_CONTEXT) { uint64_t const GCPhysEptPtr = uDesc.s.Lo; int const rc = iemVmxVmentryCheckEptPtr(pVCpu, GCPhysEptPtr, NULL /* enmDiag */); if (RT_SUCCESS(rc)) { /* likely */ } else { Log(("invept: EPTP invalid %#RX64 -> VMFail\n", GCPhysEptPtr)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invept_EptpInvalid; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysEptPtr; iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } } /** @todo PGM support for EPT tags? Currently just flush everything. */ IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3); uint64_t const uCr3 = pVCpu->cpum.GstCtx.cr3; PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */); iemVmxVmSucceed(pVCpu); rcStrict = iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } return rcStrict; } /** * Interface for HM and EM to emulate the INVEPT instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedInvept(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { IEMEXEC_ASSERT_INSTR_LEN_RETURN(pExitInfo->cbInstr, 4); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); Assert(pExitInfo); iemInitExec(pVCpu, 0 /*fExecOpts*/); uint8_t const iEffSeg = pExitInfo->InstrInfo.Inv.iSegReg; uint8_t const cbInstr = pExitInfo->cbInstr; RTGCPTR const GCPtrInveptDesc = pExitInfo->GCPtrEffAddr; uint64_t const u64InveptType = IEM_IS_64BIT_CODE(pVCpu) ? iemGRegFetchU64(pVCpu, pExitInfo->InstrInfo.Inv.iReg2) : iemGRegFetchU32(pVCpu, pExitInfo->InstrInfo.Inv.iReg2); VBOXSTRICTRC rcStrict = iemVmxInvept(pVCpu, cbInstr, iEffSeg, GCPtrInveptDesc, u64InveptType, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } #endif /* VBOX_WITH_NESTED_HWVIRT_VMX_EPT */ /** * VMXON instruction execution worker. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure. * @param cbInstr The instruction length in bytes. * @param iEffSeg The effective segment register to use with @a * GCPtrVmxon. * @param GCPtrVmxon The linear address of the VMXON pointer. * @param pExitInfo Pointer to the VM-exit information. Optional, can be NULL. * * @remarks Common VMX instruction checks are already expected to by the caller, * i.e. CR4.VMXE, Real/V86 mode, EFER/CS.L checks. */ static VBOXSTRICTRC iemVmxVmxon(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPHYS GCPtrVmxon, PCVMXVEXITINFO pExitInfo) RT_NOEXCEPT { if (!IEM_VMX_IS_ROOT_MODE(pVCpu)) { /* CPL. */ if (IEM_GET_CPL(pVCpu) == 0) { /* likely */ } else { Log(("vmxon: CPL %u -> #GP(0)\n", IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cpl; return iemRaiseGeneralProtectionFault0(pVCpu); } /* A20M (A20 Masked) mode. */ if (PGMPhysIsA20Enabled(pVCpu)) { /* likely */ } else { Log(("vmxon: A20M mode -> #GP(0)\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_A20M; return iemRaiseGeneralProtectionFault0(pVCpu); } /* CR0. */ { /* * CR0 MB1 bits. * * We use VMX_V_CR0_FIXED0 below to ensure CR0.PE and CR0.PG are always set * while executing VMXON. CR0.PE and CR0.PG are only allowed to be clear * when the guest running in VMX non-root mode with unrestricted-guest control * enabled in the VMCS. */ uint64_t const uCr0Fixed0 = VMX_V_CR0_FIXED0; if ((pVCpu->cpum.GstCtx.cr0 & uCr0Fixed0) == uCr0Fixed0) { /* likely */ } else { Log(("vmxon: CR0 fixed0 bits cleared -> #GP(0)\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cr0Fixed0; return iemRaiseGeneralProtectionFault0(pVCpu); } /* CR0 MBZ bits. */ uint64_t const uCr0Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed1; if (!(pVCpu->cpum.GstCtx.cr0 & ~uCr0Fixed1)) { /* likely */ } else { Log(("vmxon: CR0 fixed1 bits set -> #GP(0)\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cr0Fixed1; return iemRaiseGeneralProtectionFault0(pVCpu); } } /* CR4. */ { /* CR4 MB1 bits. */ uint64_t const uCr4Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed0; if ((pVCpu->cpum.GstCtx.cr4 & uCr4Fixed0) == uCr4Fixed0) { /* likely */ } else { Log(("vmxon: CR4 fixed0 bits cleared -> #GP(0)\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cr4Fixed0; return iemRaiseGeneralProtectionFault0(pVCpu); } /* CR4 MBZ bits. */ uint64_t const uCr4Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed1; if (!(pVCpu->cpum.GstCtx.cr4 & ~uCr4Fixed1)) { /* likely */ } else { Log(("vmxon: CR4 fixed1 bits set -> #GP(0)\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cr4Fixed1; return iemRaiseGeneralProtectionFault0(pVCpu); } } /* Feature control MSR's LOCK and VMXON bits. */ uint64_t const uMsrFeatCtl = CPUMGetGuestIa32FeatCtrl(pVCpu); if ((uMsrFeatCtl & (MSR_IA32_FEATURE_CONTROL_LOCK | MSR_IA32_FEATURE_CONTROL_VMXON)) == (MSR_IA32_FEATURE_CONTROL_LOCK | MSR_IA32_FEATURE_CONTROL_VMXON)) { /* likely */ } else { Log(("vmxon: Feature control lock bit or VMXON bit cleared -> #GP(0)\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_MsrFeatCtl; return iemRaiseGeneralProtectionFault0(pVCpu); } /* Get the VMXON pointer from the location specified by the source memory operand. */ RTGCPHYS GCPhysVmxon; VBOXSTRICTRC rcStrict = iemMemFetchDataU64(pVCpu, &GCPhysVmxon, iEffSeg, GCPtrVmxon); if (RT_LIKELY(rcStrict == VINF_SUCCESS)) { /* likely */ } else { Log(("vmxon: Failed to read VMXON region physaddr from %#RGv, rc=%Rrc\n", GCPtrVmxon, VBOXSTRICTRC_VAL(rcStrict))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrMap; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVmxon; return rcStrict; } /* VMXON region pointer alignment. */ if (!(GCPhysVmxon & X86_PAGE_4K_OFFSET_MASK)) { /* likely */ } else { Log(("vmxon: VMXON region pointer not page-aligned -> VMFailInvalid\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrAlign; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmxon; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* VMXON physical-address width limits. */ if (!(GCPhysVmxon >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)) { /* likely */ } else { Log(("vmxon: VMXON region pointer extends beyond physical-address width -> VMFailInvalid\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrWidth; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmxon; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Ensure VMXON region is not MMIO, ROM etc. This is not an Intel requirement but a restriction imposed by our implementation. */ if (PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmxon)) { /* likely */ } else { Log(("vmxon: VMXON region not normal memory -> VMFailInvalid\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrAbnormal; pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmxon; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Read the VMCS revision ID from the VMXON region. */ VMXVMCSREVID VmcsRevId; int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &VmcsRevId, GCPhysVmxon, sizeof(VmcsRevId)); if (RT_SUCCESS(rc)) { /* likely */ } else { Log(("vmxon: Failed to read VMXON region at %#RGp, rc=%Rrc\n", GCPhysVmxon, rc)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrReadPhys; return rc; } /* Verify the VMCS revision specified by the guest matches what we reported to the guest. */ if (RT_LIKELY(VmcsRevId.u == VMX_V_VMCS_REVISION_ID)) { /* likely */ } else { /* Revision ID mismatch. */ if (!VmcsRevId.n.fIsShadowVmcs) { Log(("vmxon: VMCS revision mismatch, expected %#RX32 got %#RX32 -> VMFailInvalid\n", VMX_V_VMCS_REVISION_ID, VmcsRevId.n.u31RevisionId)); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_VmcsRevId; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Shadow VMCS disallowed. */ Log(("vmxon: Shadow VMCS -> VMFailInvalid\n")); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_ShadowVmcs; iemVmxVmFailInvalid(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* * Record that we're in VMX operation, block INIT, block and disable A20M. */ pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysVmxon = GCPhysVmxon; IEM_VMX_CLEAR_CURRENT_VMCS(pVCpu); pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxRootMode = true; /* Clear address-range monitoring. */ EMMonitorWaitClear(pVCpu); /** @todo NSTVMX: Intel PT. */ iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) { /* Nested-guest intercept. */ if (pExitInfo) return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo); return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMXON, VMXINSTRID_NONE, cbInstr); } Assert(IEM_VMX_IS_ROOT_MODE(pVCpu)); /* CPL. */ if (IEM_GET_CPL(pVCpu) > 0) { Log(("vmxon: In VMX root mode: CPL %u -> #GP(0)\n", IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_VmxRootCpl; return iemRaiseGeneralProtectionFault0(pVCpu); } /* VMXON when already in VMX root mode. */ iemVmxVmFail(pVCpu, VMXINSTRERR_VMXON_IN_VMXROOTMODE); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_VmxAlreadyRoot; return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /** * Interface for HM and EM to emulate the VMXON instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param pExitInfo Pointer to the VM-exit information. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedVmxon(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo) { Assert(pExitInfo); IEMEXEC_ASSERT_INSTR_LEN_RETURN(pExitInfo->cbInstr, 3); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); iemInitExec(pVCpu, 0 /*fExecOpts*/); uint8_t const iEffSeg = pExitInfo->InstrInfo.VmxXsave.iSegReg; uint8_t const cbInstr = pExitInfo->cbInstr; RTGCPTR const GCPtrVmxon = pExitInfo->GCPtrEffAddr; VBOXSTRICTRC rcStrict = iemVmxVmxon(pVCpu, cbInstr, iEffSeg, GCPtrVmxon, pExitInfo); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * Implements 'VMXOFF'. * * @remarks Common VMX instruction checks are already expected to by the caller, * i.e. CR4.VMXE, Real/V86 mode, EFER/CS.L checks. */ IEM_CIMPL_DEF_0(iemCImpl_vmxoff) { /* Nested-guest intercept. */ if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) return iemVmxVmexitInstr(pVCpu, VMX_EXIT_VMXOFF, cbInstr); /* CPL. */ if (IEM_GET_CPL(pVCpu) == 0) { /* likely */ } else { Log(("vmxoff: CPL %u -> #GP(0)\n", IEM_GET_CPL(pVCpu))); pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxoff_Cpl; return iemRaiseGeneralProtectionFault0(pVCpu); } /* Dual monitor treatment of SMIs and SMM. */ uint64_t const fSmmMonitorCtl = CPUMGetGuestIa32SmmMonitorCtl(pVCpu); if (!(fSmmMonitorCtl & MSR_IA32_SMM_MONITOR_VALID)) { /* likely */ } else { iemVmxVmFail(pVCpu, VMXINSTRERR_VMXOFF_DUAL_MON); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /* Record that we're no longer in VMX root operation, block INIT, block and disable A20M. */ pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxRootMode = false; Assert(!pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxNonRootMode); if (fSmmMonitorCtl & MSR_IA32_SMM_MONITOR_VMXOFF_UNBLOCK_SMI) { /** @todo NSTVMX: Unblock SMI. */ } EMMonitorWaitClear(pVCpu); /** @todo NSTVMX: Unblock and enable A20M. */ iemVmxVmSucceed(pVCpu); return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /** * Interface for HM and EM to emulate the VMXOFF instruction. * * @returns Strict VBox status code. * @param pVCpu The cross context virtual CPU structure of the calling EMT. * @param cbInstr The instruction length in bytes. * @thread EMT(pVCpu) */ VMM_INT_DECL(VBOXSTRICTRC) IEMExecDecodedVmxoff(PVMCPUCC pVCpu, uint8_t cbInstr) { IEMEXEC_ASSERT_INSTR_LEN_RETURN(cbInstr, 3); IEM_CTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | CPUMCTX_EXTRN_INHIBIT_INT | CPUMCTX_EXTRN_INHIBIT_NMI); iemInitExec(pVCpu, 0 /*fExecOpts*/); VBOXSTRICTRC rcStrict = IEM_CIMPL_CALL_0(iemCImpl_vmxoff); Assert(!pVCpu->iem.s.cActiveMappings); return iemUninitExecAndFiddleStatusAndMaybeReenter(pVCpu, rcStrict); } /** * Implements 'VMXON'. */ IEM_CIMPL_DEF_2(iemCImpl_vmxon, uint8_t, iEffSeg, RTGCPTR, GCPtrVmxon) { return iemVmxVmxon(pVCpu, cbInstr, iEffSeg, GCPtrVmxon, NULL /* pExitInfo */); } /** * Implements 'VMLAUNCH'. */ IEM_CIMPL_DEF_0(iemCImpl_vmlaunch) { return iemVmxVmlaunchVmresume(pVCpu, cbInstr, VMXINSTRID_VMLAUNCH); } /** * Implements 'VMRESUME'. */ IEM_CIMPL_DEF_0(iemCImpl_vmresume) { return iemVmxVmlaunchVmresume(pVCpu, cbInstr, VMXINSTRID_VMRESUME); } /** * Implements 'VMPTRLD'. */ IEM_CIMPL_DEF_2(iemCImpl_vmptrld, uint8_t, iEffSeg, RTGCPTR, GCPtrVmcs) { return iemVmxVmptrld(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, NULL /* pExitInfo */); } /** * Implements 'VMPTRST'. */ IEM_CIMPL_DEF_2(iemCImpl_vmptrst, uint8_t, iEffSeg, RTGCPTR, GCPtrVmcs) { return iemVmxVmptrst(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, NULL /* pExitInfo */); } /** * Implements 'VMCLEAR'. */ IEM_CIMPL_DEF_2(iemCImpl_vmclear, uint8_t, iEffSeg, RTGCPTR, GCPtrVmcs) { return iemVmxVmclear(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, NULL /* pExitInfo */); } /** * Implements 'VMWRITE' register. */ IEM_CIMPL_DEF_2(iemCImpl_vmwrite_reg, uint64_t, u64Val, uint64_t, u64VmcsField) { return iemVmxVmwrite(pVCpu, cbInstr, UINT8_MAX /* iEffSeg */, u64Val, u64VmcsField, NULL /* pExitInfo */); } /** * Implements 'VMWRITE' memory. */ IEM_CIMPL_DEF_3(iemCImpl_vmwrite_mem, uint8_t, iEffSeg, RTGCPTR, GCPtrVal, uint32_t, u64VmcsField) { return iemVmxVmwrite(pVCpu, cbInstr, iEffSeg, GCPtrVal, u64VmcsField, NULL /* pExitInfo */); } /** * Implements 'VMREAD' register (64-bit). */ IEM_CIMPL_DEF_2(iemCImpl_vmread_reg64, uint64_t *, pu64Dst, uint64_t, u64VmcsField) { return iemVmxVmreadReg64(pVCpu, cbInstr, pu64Dst, u64VmcsField, NULL /* pExitInfo */); } /** * Implements 'VMREAD' register (32-bit). */ IEM_CIMPL_DEF_2(iemCImpl_vmread_reg32, uint64_t *, pu64Dst, uint32_t, u32VmcsField) { VBOXSTRICTRC const rcStrict = iemVmxVmreadReg32(pVCpu, cbInstr, (uint32_t *)pu64Dst, u32VmcsField, NULL /* pExitInfo */); /* Zero the high part of the register on success. */ if (rcStrict == VINF_SUCCESS) *pu64Dst = (uint32_t)*pu64Dst; return rcStrict; } /** * Implements 'VMREAD' memory, 64-bit register. */ IEM_CIMPL_DEF_3(iemCImpl_vmread_mem_reg64, uint8_t, iEffSeg, RTGCPTR, GCPtrDst, uint32_t, u64VmcsField) { return iemVmxVmreadMem(pVCpu, cbInstr, iEffSeg, GCPtrDst, u64VmcsField, NULL /* pExitInfo */); } /** * Implements 'VMREAD' memory, 32-bit register. */ IEM_CIMPL_DEF_3(iemCImpl_vmread_mem_reg32, uint8_t, iEffSeg, RTGCPTR, GCPtrDst, uint32_t, u32VmcsField) { return iemVmxVmreadMem(pVCpu, cbInstr, iEffSeg, GCPtrDst, u32VmcsField, NULL /* pExitInfo */); } /** * Implements 'INVVPID'. */ IEM_CIMPL_DEF_3(iemCImpl_invvpid, uint8_t, iEffSeg, RTGCPTR, GCPtrInvvpidDesc, uint64_t, uInvvpidType) { return iemVmxInvvpid(pVCpu, cbInstr, iEffSeg, GCPtrInvvpidDesc, uInvvpidType, NULL /* pExitInfo */); } #if defined(VBOX_WITH_NESTED_HWVIRT_VMX_EPT) || defined(VBOX_WITH_IEM_RECOMPILER) /* HACK ALERT: Linking trick. */ /** * Implements 'INVEPT'. */ IEM_CIMPL_DEF_3(iemCImpl_invept, uint8_t, iEffSeg, RTGCPTR, GCPtrInveptDesc, uint64_t, uInveptType) { # ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT return iemVmxInvept(pVCpu, cbInstr, iEffSeg, GCPtrInveptDesc, uInveptType, NULL /* pExitInfo */); # else RT_NOREF(pVCpu, cbInstr, iEffSeg, GCPtrInveptDesc, uInveptType); AssertFailedReturn(VERR_IEM_ASPECT_NOT_IMPLEMENTED); # endif } #endif /** * Implements VMX's implementation of PAUSE. */ IEM_CIMPL_DEF_0(iemCImpl_vmx_pause) { if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) { VBOXSTRICTRC rcStrict = iemVmxVmexitInstrPause(pVCpu, cbInstr); if (rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE) return rcStrict; } /* * Outside VMX non-root operation or if the PAUSE instruction does not cause * a VM-exit, the instruction operates normally. */ return iemRegAddToRipAndFinishingClearingRF(pVCpu, cbInstr); } /** * @callback_method_impl{FNPGMPHYSHANDLER, VMX APIC-access page accesses} * * @remarks The @a uUser argument is currently unused. */ DECLCALLBACK(VBOXSTRICTRC) iemVmxApicAccessPageHandler(PVMCC pVM, PVMCPUCC pVCpu, RTGCPHYS GCPhysFault, void *pvPhys, void *pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, PGMACCESSORIGIN enmOrigin, uint64_t uUser) { RT_NOREF3(pvPhys, enmOrigin, uUser); RTGCPHYS const GCPhysAccessBase = GCPhysFault & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK; if (CPUMIsGuestInVmxNonRootMode(IEM_GET_CTX(pVCpu))) { Assert(CPUMIsGuestVmxProcCtls2Set(IEM_GET_CTX(pVCpu), VMX_PROC_CTLS2_VIRT_APIC_ACCESS)); Assert(CPUMGetGuestVmxApicAccessPageAddrEx(IEM_GET_CTX(pVCpu)) == GCPhysAccessBase); uint32_t const fAccess = enmAccessType == PGMACCESSTYPE_WRITE ? IEM_ACCESS_DATA_W : IEM_ACCESS_DATA_R; uint16_t const offAccess = GCPhysFault & GUEST_PAGE_OFFSET_MASK; LogFlowFunc(("Fault at %#RGp (cbBuf=%u fAccess=%#x)\n", GCPhysFault, cbBuf, fAccess)); VBOXSTRICTRC rcStrict = iemVmxVirtApicAccessMem(pVCpu, offAccess, cbBuf, pvBuf, fAccess); if (RT_FAILURE(rcStrict)) return rcStrict; /* Any access on this APIC-access page has been handled, caller should not carry out the access. */ return VINF_SUCCESS; } LogFunc(("Accessed outside VMX non-root mode, deregistering page handler for %#RGp\n", GCPhysAccessBase)); int rc = PGMHandlerPhysicalDeregister(pVM, GCPhysAccessBase); if (RT_FAILURE(rc)) return rc; /* Instruct the caller of this handler to perform the read/write as normal memory. */ return VINF_PGM_HANDLER_DO_DEFAULT; } # ifndef IN_RING3 /** * @callback_method_impl{FNPGMRZPHYSPFHANDLER, * \#PF access handler callback for guest VMX APIC-access page.} */ DECLCALLBACK(VBOXSTRICTRC) iemVmxApicAccessPagePfHandler(PVMCC pVM, PVMCPUCC pVCpu, RTGCUINT uErr, PCPUMCTX pCtx, RTGCPTR pvFault, RTGCPHYS GCPhysFault, uint64_t uUser) { RT_NOREF3(pVM, pCtx, uUser); /* * Handle the VMX APIC-access page only when the guest is in VMX non-root mode. * Otherwise we must deregister the page and allow regular RAM access. * Failing to do so lands us with endless EPT VM-exits. */ RTGCPHYS const GCPhysPage = GCPhysFault & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK; if (CPUMIsGuestInVmxNonRootMode(IEM_GET_CTX(pVCpu))) { Assert(CPUMIsGuestVmxProcCtls2Set(IEM_GET_CTX(pVCpu), VMX_PROC_CTLS2_VIRT_APIC_ACCESS)); Assert(CPUMGetGuestVmxApicAccessPageAddrEx(IEM_GET_CTX(pVCpu)) == GCPhysPage); /* * Check if the access causes an APIC-access VM-exit. */ uint32_t fAccess; if (uErr & X86_TRAP_PF_ID) fAccess = IEM_ACCESS_INSTRUCTION; else if (uErr & X86_TRAP_PF_RW) fAccess = IEM_ACCESS_DATA_W; else fAccess = IEM_ACCESS_DATA_R; RTGCPHYS const GCPhysNestedFault = (RTGCPHYS)pvFault; uint16_t const offAccess = GCPhysNestedFault & GUEST_PAGE_OFFSET_MASK; bool const fIntercept = iemVmxVirtApicIsMemAccessIntercepted(pVCpu, offAccess, 1 /* cbAccess */, fAccess); LogFlowFunc(("#PF at %#RGp (GCPhysNestedFault=%#RGp offAccess=%#x)\n", GCPhysFault, GCPhysNestedFault, offAccess)); if (fIntercept) { /* * Query the source VM-exit (from the execution engine) that caused this access * within the APIC-access page. Currently only HM is supported. */ AssertMsg(VM_IS_HM_ENABLED(pVM), ("VM-exit auxiliary info. fetching not supported for execution engine %d\n", pVM->bMainExecutionEngine)); HMEXITAUX HmExitAux; RT_ZERO(HmExitAux); int const rc = HMR0GetExitAuxInfo(pVCpu, &HmExitAux, HMVMX_READ_EXIT_INSTR_LEN | HMVMX_READ_EXIT_QUALIFICATION | HMVMX_READ_IDT_VECTORING_INFO | HMVMX_READ_IDT_VECTORING_ERROR_CODE); AssertRC(rc); /* * Verify the VM-exit reason must be an EPT violation. * Other accesses should go through the other handler (iemVmxApicAccessPageHandler). * Refer to @bugref{10092#c33s} for a more detailed explanation. */ AssertMsgReturn(HmExitAux.Vmx.uReason == VMX_EXIT_EPT_VIOLATION, ("Unexpected call to APIC-access page #PF handler for %#RGp offAcesss=%u uErr=%#RGx uReason=%u\n", GCPhysPage, offAccess, uErr, HmExitAux.Vmx.uReason), VERR_IEM_IPE_7); /* * Construct the virtual APIC-access VM-exit. */ VMXAPICACCESS enmAccess; if (HmExitAux.Vmx.u64Qual & VMX_EXIT_QUAL_EPT_LINEAR_ADDR_VALID) { if (VMX_IDT_VECTORING_INFO_IS_VALID(HmExitAux.Vmx.uIdtVectoringInfo)) enmAccess = VMXAPICACCESS_LINEAR_EVENT_DELIVERY; else if (fAccess == IEM_ACCESS_INSTRUCTION) enmAccess = VMXAPICACCESS_LINEAR_INSTR_FETCH; else if (fAccess & IEM_ACCESS_TYPE_WRITE) enmAccess = VMXAPICACCESS_LINEAR_WRITE; else enmAccess = VMXAPICACCESS_LINEAR_READ; /* For linear-address accesss the instruction length must be valid. */ AssertMsg(HmExitAux.Vmx.cbInstr > 0, ("Invalid APIC-access VM-exit instruction length. cbInstr=%u\n", HmExitAux.Vmx.cbInstr)); } else { if (VMX_IDT_VECTORING_INFO_IS_VALID(HmExitAux.Vmx.uIdtVectoringInfo)) enmAccess = VMXAPICACCESS_PHYSICAL_EVENT_DELIVERY; else { /** @todo How to distinguish between monitoring/trace vs other instructions * here? */ enmAccess = VMXAPICACCESS_PHYSICAL_INSTR; } /* For physical accesses the instruction length is undefined, we zero it for safety and consistency. */ HmExitAux.Vmx.cbInstr = 0; } /* * Raise the APIC-access VM-exit. */ LogFlowFunc(("Raising APIC-access VM-exit from #PF handler at offset %#x\n", offAccess)); VMXVEXITINFO const ExitInfo = VMXVEXITINFO_INIT_WITH_QUAL_AND_INSTR_LEN(VMX_EXIT_APIC_ACCESS, RT_BF_MAKE(VMX_BF_EXIT_QUAL_APIC_ACCESS_OFFSET, offAccess) | RT_BF_MAKE(VMX_BF_EXIT_QUAL_APIC_ACCESS_TYPE, enmAccess), HmExitAux.Vmx.cbInstr); VMXVEXITEVENTINFO const ExitEventInfo = VMXVEXITEVENTINFO_INIT_ONLY_IDT(HmExitAux.Vmx.uIdtVectoringInfo, HmExitAux.Vmx.uIdtVectoringErrCode); VBOXSTRICTRC const rcStrict = iemVmxVmexitApicAccessWithInfo(pVCpu, &ExitInfo, &ExitEventInfo); return iemExecStatusCodeFiddling(pVCpu, rcStrict); } /* * The access isn't intercepted, which means it needs to be virtualized. * * This requires emulating the instruction because we need the bytes being * read/written by the instruction not just the offset being accessed within * the APIC-access page (which we derive from the faulting address). */ LogFlowFunc(("Access at offset %#x not intercepted -> VINF_EM_RAW_EMULATE_INSTR\n", offAccess)); return VINF_EM_RAW_EMULATE_INSTR; } /** @todo This isn't ideal but works for now as nested-hypervisors generally play * nice because the spec states that this page should be modified only when * no CPU refers to it VMX non-root mode. Nonetheless, we could use an atomic * reference counter to ensure the aforementioned condition before * de-registering the page. */ LogFunc(("Accessed outside VMX non-root mode, deregistering page handler for %#RGp\n", GCPhysPage)); int const rc = PGMHandlerPhysicalDeregister(pVM, GCPhysPage); if (RT_FAILURE(rc)) return rc; return VINF_SUCCESS; } # endif /* !IN_RING3 */ #endif /* VBOX_WITH_NESTED_HWVIRT_VMX */ /** * Implements 'VMCALL'. */ IEM_CIMPL_DEF_0(iemCImpl_vmcall) { pVCpu->iem.s.cPotentialExits++; #ifdef VBOX_WITH_NESTED_HWVIRT_VMX /* Nested-guest intercept. */ if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu)) return iemVmxVmexitInstr(pVCpu, VMX_EXIT_VMCALL, cbInstr); #endif /* Join forces with vmmcall. */ return IEM_CIMPL_CALL_1(iemCImpl_Hypercall, OP_VMCALL); }