CPUMAllRegs.cpp revision 437cb30cecc67158e35c5cc5d453cfe4eddb7dd2
/* $Id$ */
/** @file
* CPUM - CPU Monitor(/Manager) - Getters and Setters.
*/
/*
* Copyright (C) 2006-2007 Sun Microsystems, Inc.
*
* This file is part of VirtualBox Open Source Edition (OSE), as
* available from http://www.virtualbox.org. This file is free software;
* you can redistribute it and/or modify it under the terms of the GNU
* General Public License (GPL) as published by the Free Software
* Foundation, in version 2 as it comes in the "COPYING" file of the
* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
*
* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
* Clara, CA 95054 USA or visit http://www.sun.com if you need
* additional information or have any questions.
*/
/*******************************************************************************
* Header Files *
*******************************************************************************/
#define LOG_GROUP LOG_GROUP_CPUM
#include <VBox/cpum.h>
#include <VBox/patm.h>
#include <VBox/dbgf.h>
#include <VBox/mm.h>
#include "CPUMInternal.h"
#include <VBox/vm.h>
#include <VBox/err.h>
#include <VBox/dis.h>
#include <VBox/log.h>
#include <VBox/hwaccm.h>
#include <VBox/tm.h>
#include <iprt/assert.h>
#include <iprt/asm.h>
#ifdef IN_RING3
#include <iprt/thread.h>
#endif
/** Disable stack frame pointer generation here. */
#if defined(_MSC_VER) && !defined(DEBUG)
# pragma optimize("y", off)
#endif
/**
* Sets or resets an alternative hypervisor context core.
*
* This is called when we get a hypervisor trap set switch the context
* core with the trap frame on the stack. It is called again to reset
* back to the default context core when resuming hypervisor execution.
*
* @param pVCpu The VMCPU handle.
* @param pCtxCore Pointer to the alternative context core or NULL
* to go back to the default context core.
*/
VMMDECL(void) CPUMHyperSetCtxCore(PVMCPU pVCpu, PCPUMCTXCORE pCtxCore)
{
PVM pVM = pVCpu->CTX_SUFF(pVM);
LogFlow(("CPUMHyperSetCtxCore: %p/%p/%p -> %p\n", pVCpu->cpum.s.CTX_SUFF(pHyperCore), pCtxCore));
if (!pCtxCore)
{
pCtxCore = CPUMCTX2CORE(&pVCpu->cpum.s.Hyper);
pVCpu->cpum.s.pHyperCoreR3 = (R3PTRTYPE(PCPUMCTXCORE))VM_R3_ADDR(pVM, pCtxCore);
pVCpu->cpum.s.pHyperCoreR0 = (R0PTRTYPE(PCPUMCTXCORE))VM_R0_ADDR(pVM, pCtxCore);
pVCpu->cpum.s.pHyperCoreRC = (RCPTRTYPE(PCPUMCTXCORE))VM_RC_ADDR(pVM, pCtxCore);
}
else
{
pVCpu->cpum.s.pHyperCoreR3 = (R3PTRTYPE(PCPUMCTXCORE))MMHyperCCToR3(pVM, pCtxCore);
pVCpu->cpum.s.pHyperCoreR0 = (R0PTRTYPE(PCPUMCTXCORE))MMHyperCCToR0(pVM, pCtxCore);
pVCpu->cpum.s.pHyperCoreRC = (RCPTRTYPE(PCPUMCTXCORE))MMHyperCCToRC(pVM, pCtxCore);
}
}
/**
* Gets the pointer to the internal CPUMCTXCORE structure for the hypervisor.
* This is only for reading in order to save a few calls.
*
* @param pVM Handle to the virtual machine.
*/
VMMDECL(PCCPUMCTXCORE) CPUMGetHyperCtxCore(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore);
}
/**
* Queries the pointer to the internal CPUMCTX structure for the hypervisor.
*
* @returns VBox status code.
* @param pVM Handle to the virtual machine.
* @param ppCtx Receives the hyper CPUMCTX pointer when successful.
*
* @deprecated This will *not* (and has never) given the right picture of the
* hypervisor register state. With CPUMHyperSetCtxCore() this is
* getting much worse. So, use the individual functions for getting
* and esp. setting the hypervisor registers.
*/
VMMDECL(int) CPUMQueryHyperCtxPtr(PVMCPU pVCpu, PCPUMCTX *ppCtx)
{
*ppCtx = &pVCpu->cpum.s.Hyper;
return VINF_SUCCESS;
}
VMMDECL(void) CPUMSetHyperGDTR(PVMCPU pVCpu, uint32_t addr, uint16_t limit)
{
pVCpu->cpum.s.Hyper.gdtr.cbGdt = limit;
pVCpu->cpum.s.Hyper.gdtr.pGdt = addr;
pVCpu->cpum.s.Hyper.gdtrPadding = 0;
}
VMMDECL(void) CPUMSetHyperIDTR(PVMCPU pVCpu, uint32_t addr, uint16_t limit)
{
pVCpu->cpum.s.Hyper.idtr.cbIdt = limit;
pVCpu->cpum.s.Hyper.idtr.pIdt = addr;
pVCpu->cpum.s.Hyper.idtrPadding = 0;
}
VMMDECL(void) CPUMSetHyperCR3(PVMCPU pVCpu, uint32_t cr3)
{
pVCpu->cpum.s.Hyper.cr3 = cr3;
#ifdef IN_RC
/* Update the current CR3. */
ASMSetCR3(cr3);
#endif
}
VMMDECL(uint32_t) CPUMGetHyperCR3(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Hyper.cr3;
}
VMMDECL(void) CPUMSetHyperCS(PVMCPU pVCpu, RTSEL SelCS)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->cs = SelCS;
}
VMMDECL(void) CPUMSetHyperDS(PVMCPU pVCpu, RTSEL SelDS)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->ds = SelDS;
}
VMMDECL(void) CPUMSetHyperES(PVMCPU pVCpu, RTSEL SelES)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->es = SelES;
}
VMMDECL(void) CPUMSetHyperFS(PVMCPU pVCpu, RTSEL SelFS)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->fs = SelFS;
}
VMMDECL(void) CPUMSetHyperGS(PVMCPU pVCpu, RTSEL SelGS)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->gs = SelGS;
}
VMMDECL(void) CPUMSetHyperSS(PVMCPU pVCpu, RTSEL SelSS)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->ss = SelSS;
}
VMMDECL(void) CPUMSetHyperESP(PVMCPU pVCpu, uint32_t u32ESP)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->esp = u32ESP;
}
VMMDECL(int) CPUMSetHyperEFlags(PVMCPU pVCpu, uint32_t Efl)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->eflags.u32 = Efl;
return VINF_SUCCESS;
}
VMMDECL(void) CPUMSetHyperEIP(PVMCPU pVCpu, uint32_t u32EIP)
{
pVCpu->cpum.s.CTX_SUFF(pHyperCore)->eip = u32EIP;
}
VMMDECL(void) CPUMSetHyperTR(PVMCPU pVCpu, RTSEL SelTR)
{
pVCpu->cpum.s.Hyper.tr = SelTR;
}
VMMDECL(void) CPUMSetHyperLDTR(PVMCPU pVCpu, RTSEL SelLDTR)
{
pVCpu->cpum.s.Hyper.ldtr = SelLDTR;
}
VMMDECL(void) CPUMSetHyperDR0(PVMCPU pVCpu, RTGCUINTREG uDr0)
{
pVCpu->cpum.s.Hyper.dr[0] = uDr0;
/** @todo in GC we must load it! */
}
VMMDECL(void) CPUMSetHyperDR1(PVMCPU pVCpu, RTGCUINTREG uDr1)
{
pVCpu->cpum.s.Hyper.dr[1] = uDr1;
/** @todo in GC we must load it! */
}
VMMDECL(void) CPUMSetHyperDR2(PVMCPU pVCpu, RTGCUINTREG uDr2)
{
pVCpu->cpum.s.Hyper.dr[2] = uDr2;
/** @todo in GC we must load it! */
}
VMMDECL(void) CPUMSetHyperDR3(PVMCPU pVCpu, RTGCUINTREG uDr3)
{
pVCpu->cpum.s.Hyper.dr[3] = uDr3;
/** @todo in GC we must load it! */
}
VMMDECL(void) CPUMSetHyperDR6(PVMCPU pVCpu, RTGCUINTREG uDr6)
{
pVCpu->cpum.s.Hyper.dr[6] = uDr6;
/** @todo in GC we must load it! */
}
VMMDECL(void) CPUMSetHyperDR7(PVMCPU pVCpu, RTGCUINTREG uDr7)
{
pVCpu->cpum.s.Hyper.dr[7] = uDr7;
/** @todo in GC we must load it! */
}
VMMDECL(RTSEL) CPUMGetHyperCS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->cs;
}
VMMDECL(RTSEL) CPUMGetHyperDS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->ds;
}
VMMDECL(RTSEL) CPUMGetHyperES(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->es;
}
VMMDECL(RTSEL) CPUMGetHyperFS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->fs;
}
VMMDECL(RTSEL) CPUMGetHyperGS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->gs;
}
VMMDECL(RTSEL) CPUMGetHyperSS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->ss;
}
VMMDECL(uint32_t) CPUMGetHyperEAX(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->eax;
}
VMMDECL(uint32_t) CPUMGetHyperEBX(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->ebx;
}
VMMDECL(uint32_t) CPUMGetHyperECX(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->ecx;
}
VMMDECL(uint32_t) CPUMGetHyperEDX(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->edx;
}
VMMDECL(uint32_t) CPUMGetHyperESI(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->esi;
}
VMMDECL(uint32_t) CPUMGetHyperEDI(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->edi;
}
VMMDECL(uint32_t) CPUMGetHyperEBP(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->ebp;
}
VMMDECL(uint32_t) CPUMGetHyperESP(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->esp;
}
VMMDECL(uint32_t) CPUMGetHyperEFlags(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->eflags.u32;
}
VMMDECL(uint32_t) CPUMGetHyperEIP(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->eip;
}
VMMDECL(uint64_t) CPUMGetHyperRIP(PVMCPU pVCpu)
{
return pVCpu->cpum.s.CTX_SUFF(pHyperCore)->rip;
}
VMMDECL(uint32_t) CPUMGetHyperIDTR(PVMCPU pVCpu, uint16_t *pcbLimit)
{
if (pcbLimit)
*pcbLimit = pVCpu->cpum.s.Hyper.idtr.cbIdt;
return pVCpu->cpum.s.Hyper.idtr.pIdt;
}
VMMDECL(uint32_t) CPUMGetHyperGDTR(PVMCPU pVCpu, uint16_t *pcbLimit)
{
if (pcbLimit)
*pcbLimit = pVCpu->cpum.s.Hyper.gdtr.cbGdt;
return pVCpu->cpum.s.Hyper.gdtr.pGdt;
}
VMMDECL(RTSEL) CPUMGetHyperLDTR(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Hyper.ldtr;
}
VMMDECL(RTGCUINTREG) CPUMGetHyperDR0(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Hyper.dr[0];
}
VMMDECL(RTGCUINTREG) CPUMGetHyperDR1(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Hyper.dr[1];
}
VMMDECL(RTGCUINTREG) CPUMGetHyperDR2(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Hyper.dr[2];
}
VMMDECL(RTGCUINTREG) CPUMGetHyperDR3(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Hyper.dr[3];
}
VMMDECL(RTGCUINTREG) CPUMGetHyperDR6(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Hyper.dr[6];
}
VMMDECL(RTGCUINTREG) CPUMGetHyperDR7(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Hyper.dr[7];
}
/**
* Gets the pointer to the internal CPUMCTXCORE structure.
* This is only for reading in order to save a few calls.
*
* @param pVCpu Handle to the virtual cpu.
*/
VMMDECL(PCCPUMCTXCORE) CPUMGetGuestCtxCore(PVMCPU pVCpu)
{
return CPUMCTX2CORE(&pVCpu->cpum.s.Guest);
}
/**
* Sets the guest context core registers.
*
* @param pVCpu Handle to the virtual cpu.
* @param pCtxCore The new context core values.
*/
VMMDECL(void) CPUMSetGuestCtxCore(PVMCPU pVCpu, PCCPUMCTXCORE pCtxCore)
{
/** @todo #1410 requires selectors to be checked. (huh? 1410?) */
PCPUMCTXCORE pCtxCoreDst = CPUMCTX2CORE(&pVCpu->cpum.s.Guest);
*pCtxCoreDst = *pCtxCore;
/* Mask away invalid parts of the cpu context. */
if (!CPUMIsGuestInLongMode(pVCpu))
{
uint64_t u64Mask = UINT64_C(0xffffffff);
pCtxCoreDst->rip &= u64Mask;
pCtxCoreDst->rax &= u64Mask;
pCtxCoreDst->rbx &= u64Mask;
pCtxCoreDst->rcx &= u64Mask;
pCtxCoreDst->rdx &= u64Mask;
pCtxCoreDst->rsi &= u64Mask;
pCtxCoreDst->rdi &= u64Mask;
pCtxCoreDst->rbp &= u64Mask;
pCtxCoreDst->rsp &= u64Mask;
pCtxCoreDst->rflags.u &= u64Mask;
pCtxCoreDst->r8 = 0;
pCtxCoreDst->r9 = 0;
pCtxCoreDst->r10 = 0;
pCtxCoreDst->r11 = 0;
pCtxCoreDst->r12 = 0;
pCtxCoreDst->r13 = 0;
pCtxCoreDst->r14 = 0;
pCtxCoreDst->r15 = 0;
}
}
/**
* Queries the pointer to the internal CPUMCTX structure
*
* @returns The CPUMCTX pointer.
* @param pVCpu Handle to the virtual cpu.
*/
VMMDECL(PCPUMCTX) CPUMQueryGuestCtxPtr(PVMCPU pVCpu)
{
return &pVCpu->cpum.s.Guest;
}
VMMDECL(int) CPUMSetGuestGDTR(PVMCPU pVCpu, uint32_t addr, uint16_t limit)
{
pVCpu->cpum.s.Guest.gdtr.cbGdt = limit;
pVCpu->cpum.s.Guest.gdtr.pGdt = addr;
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_GDTR;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestIDTR(PVMCPU pVCpu, uint32_t addr, uint16_t limit)
{
pVCpu->cpum.s.Guest.idtr.cbIdt = limit;
pVCpu->cpum.s.Guest.idtr.pIdt = addr;
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_IDTR;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestTR(PVMCPU pVCpu, uint16_t tr)
{
AssertMsgFailed(("Need to load the hidden bits too!\n"));
pVCpu->cpum.s.Guest.tr = tr;
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_TR;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestLDTR(PVMCPU pVCpu, uint16_t ldtr)
{
pVCpu->cpum.s.Guest.ldtr = ldtr;
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_LDTR;
return VINF_SUCCESS;
}
/**
* Set the guest CR0.
*
* When called in GC, the hyper CR0 may be updated if that is
* required. The caller only has to take special action if AM,
* WP, PG or PE changes.
*
* @returns VINF_SUCCESS (consider it void).
* @param pVCpu Handle to the virtual cpu.
* @param cr0 The new CR0 value.
*/
VMMDECL(int) CPUMSetGuestCR0(PVMCPU pVCpu, uint64_t cr0)
{
#ifdef IN_RC
/*
* Check if we need to change hypervisor CR0 because
* of math stuff.
*/
if ( (cr0 & (X86_CR0_TS | X86_CR0_EM | X86_CR0_MP))
!= (pVCpu->cpum.s.Guest.cr0 & (X86_CR0_TS | X86_CR0_EM | X86_CR0_MP)))
{
if (!(pVCpu->cpum.s.fUseFlags & CPUM_USED_FPU))
{
/*
* We haven't saved the host FPU state yet, so TS and MT are both set
* and EM should be reflecting the guest EM (it always does this).
*/
if ((cr0 & X86_CR0_EM) != (pVCpu->cpum.s.Guest.cr0 & X86_CR0_EM))
{
uint32_t HyperCR0 = ASMGetCR0();
AssertMsg((HyperCR0 & (X86_CR0_TS | X86_CR0_MP)) == (X86_CR0_TS | X86_CR0_MP), ("%#x\n", HyperCR0));
AssertMsg((HyperCR0 & X86_CR0_EM) == (pVCpu->cpum.s.Guest.cr0 & X86_CR0_EM), ("%#x\n", HyperCR0));
HyperCR0 &= ~X86_CR0_EM;
HyperCR0 |= cr0 & X86_CR0_EM;
Log(("CPUM New HyperCR0=%#x\n", HyperCR0));
ASMSetCR0(HyperCR0);
}
# ifdef VBOX_STRICT
else
{
uint32_t HyperCR0 = ASMGetCR0();
AssertMsg((HyperCR0 & (X86_CR0_TS | X86_CR0_MP)) == (X86_CR0_TS | X86_CR0_MP), ("%#x\n", HyperCR0));
AssertMsg((HyperCR0 & X86_CR0_EM) == (pVCpu->cpum.s.Guest.cr0 & X86_CR0_EM), ("%#x\n", HyperCR0));
}
# endif
}
else
{
/*
* Already saved the state, so we're just mirroring
* the guest flags.
*/
uint32_t HyperCR0 = ASMGetCR0();
AssertMsg( (HyperCR0 & (X86_CR0_TS | X86_CR0_EM | X86_CR0_MP))
== (pVCpu->cpum.s.Guest.cr0 & (X86_CR0_TS | X86_CR0_EM | X86_CR0_MP)),
("%#x %#x\n", HyperCR0, pVCpu->cpum.s.Guest.cr0));
HyperCR0 &= ~(X86_CR0_TS | X86_CR0_EM | X86_CR0_MP);
HyperCR0 |= cr0 & (X86_CR0_TS | X86_CR0_EM | X86_CR0_MP);
Log(("CPUM New HyperCR0=%#x\n", HyperCR0));
ASMSetCR0(HyperCR0);
}
}
#endif /* IN_RC */
/*
* Check for changes causing TLB flushes (for REM).
* The caller is responsible for calling PGM when appropriate.
*/
if ( (cr0 & (X86_CR0_PG | X86_CR0_WP | X86_CR0_PE))
!= (pVCpu->cpum.s.Guest.cr0 & (X86_CR0_PG | X86_CR0_WP | X86_CR0_PE)))
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_GLOBAL_TLB_FLUSH;
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_CR0;
pVCpu->cpum.s.Guest.cr0 = cr0 | X86_CR0_ET;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestCR2(PVMCPU pVCpu, uint64_t cr2)
{
pVCpu->cpum.s.Guest.cr2 = cr2;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestCR3(PVMCPU pVCpu, uint64_t cr3)
{
pVCpu->cpum.s.Guest.cr3 = cr3;
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_CR3;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestCR4(PVMCPU pVCpu, uint64_t cr4)
{
if ( (cr4 & (X86_CR4_PGE | X86_CR4_PAE | X86_CR4_PSE))
!= (pVCpu->cpum.s.Guest.cr4 & (X86_CR4_PGE | X86_CR4_PAE | X86_CR4_PSE)))
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_GLOBAL_TLB_FLUSH;
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_CR4;
if (!CPUMSupportsFXSR(pVCpu->CTX_SUFF(pVM)))
cr4 &= ~X86_CR4_OSFSXR;
pVCpu->cpum.s.Guest.cr4 = cr4;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestEFlags(PVMCPU pVCpu, uint32_t eflags)
{
pVCpu->cpum.s.Guest.eflags.u32 = eflags;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestEIP(PVMCPU pVCpu, uint32_t eip)
{
pVCpu->cpum.s.Guest.eip = eip;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestEAX(PVMCPU pVCpu, uint32_t eax)
{
pVCpu->cpum.s.Guest.eax = eax;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestEBX(PVMCPU pVCpu, uint32_t ebx)
{
pVCpu->cpum.s.Guest.ebx = ebx;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestECX(PVMCPU pVCpu, uint32_t ecx)
{
pVCpu->cpum.s.Guest.ecx = ecx;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestEDX(PVMCPU pVCpu, uint32_t edx)
{
pVCpu->cpum.s.Guest.edx = edx;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestESP(PVMCPU pVCpu, uint32_t esp)
{
pVCpu->cpum.s.Guest.esp = esp;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestEBP(PVMCPU pVCpu, uint32_t ebp)
{
pVCpu->cpum.s.Guest.ebp = ebp;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestESI(PVMCPU pVCpu, uint32_t esi)
{
pVCpu->cpum.s.Guest.esi = esi;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestEDI(PVMCPU pVCpu, uint32_t edi)
{
pVCpu->cpum.s.Guest.edi = edi;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestSS(PVMCPU pVCpu, uint16_t ss)
{
pVCpu->cpum.s.Guest.ss = ss;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestCS(PVMCPU pVCpu, uint16_t cs)
{
pVCpu->cpum.s.Guest.cs = cs;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestDS(PVMCPU pVCpu, uint16_t ds)
{
pVCpu->cpum.s.Guest.ds = ds;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestES(PVMCPU pVCpu, uint16_t es)
{
pVCpu->cpum.s.Guest.es = es;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestFS(PVMCPU pVCpu, uint16_t fs)
{
pVCpu->cpum.s.Guest.fs = fs;
return VINF_SUCCESS;
}
VMMDECL(int) CPUMSetGuestGS(PVMCPU pVCpu, uint16_t gs)
{
pVCpu->cpum.s.Guest.gs = gs;
return VINF_SUCCESS;
}
VMMDECL(void) CPUMSetGuestEFER(PVMCPU pVCpu, uint64_t val)
{
pVCpu->cpum.s.Guest.msrEFER = val;
}
VMMDECL(uint64_t) CPUMGetGuestMsr(PVMCPU pVCpu, unsigned idMsr)
{
uint64_t u64 = 0;
switch (idMsr)
{
case MSR_IA32_TSC:
u64 = TMCpuTickGet(pVCpu);
break;
case MSR_IA32_CR_PAT:
u64 = pVCpu->cpum.s.Guest.msrPAT;
break;
case MSR_IA32_SYSENTER_CS:
u64 = pVCpu->cpum.s.Guest.SysEnter.cs;
break;
case MSR_IA32_SYSENTER_EIP:
u64 = pVCpu->cpum.s.Guest.SysEnter.eip;
break;
case MSR_IA32_SYSENTER_ESP:
u64 = pVCpu->cpum.s.Guest.SysEnter.esp;
break;
case MSR_K6_EFER:
u64 = pVCpu->cpum.s.Guest.msrEFER;
break;
case MSR_K8_SF_MASK:
u64 = pVCpu->cpum.s.Guest.msrSFMASK;
break;
case MSR_K6_STAR:
u64 = pVCpu->cpum.s.Guest.msrSTAR;
break;
case MSR_K8_LSTAR:
u64 = pVCpu->cpum.s.Guest.msrLSTAR;
break;
case MSR_K8_CSTAR:
u64 = pVCpu->cpum.s.Guest.msrCSTAR;
break;
case MSR_K8_KERNEL_GS_BASE:
u64 = pVCpu->cpum.s.Guest.msrKERNELGSBASE;
break;
case MSR_K8_TSC_AUX:
u64 = pVCpu->cpum.s.GuestMsr.msr.tscAux;
break;
case MSR_IA32_PERF_STATUS:
/** @todo: could really be not exactly correct, maybe use host's values */
/* Keep consistent with helper_rdmsr() in REM */
u64 = (1000ULL /* TSC increment by tick */)
| (((uint64_t)4ULL) << 40 /* CPU multiplier */ );
break;
/* fs & gs base skipped on purpose as the current context might not be up-to-date. */
default:
AssertFailed();
break;
}
return u64;
}
VMMDECL(void) CPUMSetGuestMsr(PVMCPU pVCpu, unsigned idMsr, uint64_t valMsr)
{
/* On purpose only a limited number of MSRs; use the emulation function to update the others. */
switch (idMsr)
{
case MSR_K8_TSC_AUX:
pVCpu->cpum.s.GuestMsr.msr.tscAux = valMsr;
break;
default:
AssertFailed();
break;
}
}
VMMDECL(RTGCPTR) CPUMGetGuestIDTR(PVMCPU pVCpu, uint16_t *pcbLimit)
{
if (pcbLimit)
*pcbLimit = pVCpu->cpum.s.Guest.idtr.cbIdt;
return pVCpu->cpum.s.Guest.idtr.pIdt;
}
VMMDECL(RTSEL) CPUMGetGuestTR(PVMCPU pVCpu, PCPUMSELREGHID pHidden)
{
if (pHidden)
*pHidden = pVCpu->cpum.s.Guest.trHid;
return pVCpu->cpum.s.Guest.tr;
}
VMMDECL(RTSEL) CPUMGetGuestCS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.cs;
}
VMMDECL(RTSEL) CPUMGetGuestDS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.ds;
}
VMMDECL(RTSEL) CPUMGetGuestES(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.es;
}
VMMDECL(RTSEL) CPUMGetGuestFS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.fs;
}
VMMDECL(RTSEL) CPUMGetGuestGS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.gs;
}
VMMDECL(RTSEL) CPUMGetGuestSS(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.ss;
}
VMMDECL(RTSEL) CPUMGetGuestLDTR(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.ldtr;
}
VMMDECL(uint64_t) CPUMGetGuestCR0(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.cr0;
}
VMMDECL(uint64_t) CPUMGetGuestCR2(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.cr2;
}
VMMDECL(uint64_t) CPUMGetGuestCR3(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.cr3;
}
VMMDECL(uint64_t) CPUMGetGuestCR4(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.cr4;
}
VMMDECL(void) CPUMGetGuestGDTR(PVMCPU pVCpu, PVBOXGDTR pGDTR)
{
*pGDTR = pVCpu->cpum.s.Guest.gdtr;
}
VMMDECL(uint32_t) CPUMGetGuestEIP(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.eip;
}
VMMDECL(uint64_t) CPUMGetGuestRIP(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.rip;
}
VMMDECL(uint32_t) CPUMGetGuestEAX(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.eax;
}
VMMDECL(uint32_t) CPUMGetGuestEBX(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.ebx;
}
VMMDECL(uint32_t) CPUMGetGuestECX(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.ecx;
}
VMMDECL(uint32_t) CPUMGetGuestEDX(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.edx;
}
VMMDECL(uint32_t) CPUMGetGuestESI(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.esi;
}
VMMDECL(uint32_t) CPUMGetGuestEDI(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.edi;
}
VMMDECL(uint32_t) CPUMGetGuestESP(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.esp;
}
VMMDECL(uint32_t) CPUMGetGuestEBP(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.ebp;
}
VMMDECL(uint32_t) CPUMGetGuestEFlags(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.eflags.u32;
}
///@todo: crx should be an array
VMMDECL(int) CPUMGetGuestCRx(PVMCPU pVCpu, unsigned iReg, uint64_t *pValue)
{
switch (iReg)
{
case USE_REG_CR0:
*pValue = pVCpu->cpum.s.Guest.cr0;
break;
case USE_REG_CR2:
*pValue = pVCpu->cpum.s.Guest.cr2;
break;
case USE_REG_CR3:
*pValue = pVCpu->cpum.s.Guest.cr3;
break;
case USE_REG_CR4:
*pValue = pVCpu->cpum.s.Guest.cr4;
break;
default:
return VERR_INVALID_PARAMETER;
}
return VINF_SUCCESS;
}
VMMDECL(uint64_t) CPUMGetGuestDR0(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.dr[0];
}
VMMDECL(uint64_t) CPUMGetGuestDR1(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.dr[1];
}
VMMDECL(uint64_t) CPUMGetGuestDR2(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.dr[2];
}
VMMDECL(uint64_t) CPUMGetGuestDR3(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.dr[3];
}
VMMDECL(uint64_t) CPUMGetGuestDR6(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.dr[6];
}
VMMDECL(uint64_t) CPUMGetGuestDR7(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.dr[7];
}
VMMDECL(int) CPUMGetGuestDRx(PVMCPU pVCpu, uint32_t iReg, uint64_t *pValue)
{
AssertReturn(iReg <= USE_REG_DR7, VERR_INVALID_PARAMETER);
/* DR4 is an alias for DR6, and DR5 is an alias for DR7. */
if (iReg == 4 || iReg == 5)
iReg += 2;
*pValue = pVCpu->cpum.s.Guest.dr[iReg];
return VINF_SUCCESS;
}
VMMDECL(uint64_t) CPUMGetGuestEFER(PVMCPU pVCpu)
{
return pVCpu->cpum.s.Guest.msrEFER;
}
/**
* Gets a CpuId leaf.
*
* @param pVCpu The VMCPU handle.
* @param iLeaf The CPUID leaf to get.
* @param pEax Where to store the EAX value.
* @param pEbx Where to store the EBX value.
* @param pEcx Where to store the ECX value.
* @param pEdx Where to store the EDX value.
*/
VMMDECL(void) CPUMGetGuestCpuId(PVMCPU pVCpu, uint32_t iLeaf, uint32_t *pEax, uint32_t *pEbx, uint32_t *pEcx, uint32_t *pEdx)
{
PVM pVM = pVCpu->CTX_SUFF(pVM);
PCCPUMCPUID pCpuId;
if (iLeaf < RT_ELEMENTS(pVM->cpum.s.aGuestCpuIdStd))
pCpuId = &pVM->cpum.s.aGuestCpuIdStd[iLeaf];
else if (iLeaf - UINT32_C(0x80000000) < RT_ELEMENTS(pVM->cpum.s.aGuestCpuIdExt))
pCpuId = &pVM->cpum.s.aGuestCpuIdExt[iLeaf - UINT32_C(0x80000000)];
else if (iLeaf - UINT32_C(0xc0000000) < RT_ELEMENTS(pVM->cpum.s.aGuestCpuIdCentaur))
pCpuId = &pVM->cpum.s.aGuestCpuIdCentaur[iLeaf - UINT32_C(0xc0000000)];
else
pCpuId = &pVM->cpum.s.GuestCpuIdDef;
uint32_t cCurrentCacheIndex = *pEcx;
*pEax = pCpuId->eax;
*pEbx = pCpuId->ebx;
*pEcx = pCpuId->ecx;
*pEdx = pCpuId->edx;
if ( iLeaf == 1
&& pVM->cCpus > 1)
{
/* Bits 31-24: Initial APIC ID */
Assert(pVCpu->idCpu <= 255);
*pEbx |= (pVCpu->idCpu << 24);
}
if ( iLeaf == 4
&& cCurrentCacheIndex < 3
&& pVM->cpum.s.enmGuestCpuVendor == CPUMCPUVENDOR_INTEL)
{
uint32_t type, level, sharing, linesize,
partitions, associativity, sets, cores;
/* For type: 1 - data cache, 2 - i-cache, 3 - unified */
partitions = 1;
/* Those are only to shut up compiler, as they will always
get overwritten, and compiler should be able to figure that out */
sets = associativity = sharing = level = 1;
cores = pVM->cCpus > 32 ? 32 : pVM->cCpus;
switch (cCurrentCacheIndex)
{
case 0:
type = 1;
level = 1;
sharing = 1;
linesize = 64;
associativity = 8;
sets = 64;
break;
case 1:
level = 1;
type = 2;
sharing = 1;
linesize = 64;
associativity = 8;
sets = 64;
break;
case 2:
level = 2;
type = 3;
sharing = 2;
linesize = 64;
associativity = 24;
sets = 4096;
break;
}
*pEax |= ((cores - 1) << 26) |
((sharing - 1) << 14) |
(level << 5) |
1;
*pEbx = (linesize - 1) |
((partitions - 1) << 12) |
((associativity - 1) << 22); /* -1 encoding */
*pEcx = sets - 1;
}
Log2(("CPUMGetGuestCpuId: iLeaf=%#010x %RX32 %RX32 %RX32 %RX32\n", iLeaf, *pEax, *pEbx, *pEcx, *pEdx));
}
/**
* Gets a number of standard CPUID leafs.
*
* @returns Number of leafs.
* @param pVM The VM handle.
* @remark Intended for PATM.
*/
VMMDECL(uint32_t) CPUMGetGuestCpuIdStdMax(PVM pVM)
{
return RT_ELEMENTS(pVM->cpum.s.aGuestCpuIdStd);
}
/**
* Gets a number of extended CPUID leafs.
*
* @returns Number of leafs.
* @param pVM The VM handle.
* @remark Intended for PATM.
*/
VMMDECL(uint32_t) CPUMGetGuestCpuIdExtMax(PVM pVM)
{
return RT_ELEMENTS(pVM->cpum.s.aGuestCpuIdExt);
}
/**
* Gets a number of centaur CPUID leafs.
*
* @returns Number of leafs.
* @param pVM The VM handle.
* @remark Intended for PATM.
*/
VMMDECL(uint32_t) CPUMGetGuestCpuIdCentaurMax(PVM pVM)
{
return RT_ELEMENTS(pVM->cpum.s.aGuestCpuIdCentaur);
}
/**
* Sets a CPUID feature bit.
*
* @param pVM The VM Handle.
* @param enmFeature The feature to set.
*/
VMMDECL(void) CPUMSetGuestCpuIdFeature(PVM pVM, CPUMCPUIDFEATURE enmFeature)
{
switch (enmFeature)
{
/*
* Set the APIC bit in both feature masks.
*/
case CPUMCPUIDFEATURE_APIC:
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].edx |= X86_CPUID_FEATURE_EDX_APIC;
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001
&& pVM->cpum.s.enmGuestCpuVendor == CPUMCPUVENDOR_AMD)
pVM->cpum.s.aGuestCpuIdExt[1].edx |= X86_CPUID_AMD_FEATURE_EDX_APIC;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled APIC\n"));
break;
/*
* Set the x2APIC bit in the standard feature mask.
*/
case CPUMCPUIDFEATURE_X2APIC:
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].ecx |= X86_CPUID_FEATURE_ECX_X2APIC;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled x2APIC\n"));
break;
/*
* Set the sysenter/sysexit bit in the standard feature mask.
* Assumes the caller knows what it's doing! (host must support these)
*/
case CPUMCPUIDFEATURE_SEP:
{
if (!(ASMCpuId_EDX(1) & X86_CPUID_FEATURE_EDX_SEP))
{
AssertMsgFailed(("ERROR: Can't turn on SEP when the host doesn't support it!!\n"));
return;
}
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].edx |= X86_CPUID_FEATURE_EDX_SEP;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled sysenter/exit\n"));
break;
}
/*
* Set the syscall/sysret bit in the extended feature mask.
* Assumes the caller knows what it's doing! (host must support these)
*/
case CPUMCPUIDFEATURE_SYSCALL:
{
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax < 0x80000001
|| !(ASMCpuId_EDX(0x80000001) & X86_CPUID_AMD_FEATURE_EDX_SEP))
{
#if HC_ARCH_BITS == 32
/* X86_CPUID_AMD_FEATURE_EDX_SEP not set it seems in 32 bits mode.
* Even when the cpu is capable of doing so in 64 bits mode.
*/
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax < 0x80000001
|| !(ASMCpuId_EDX(0x80000001) & X86_CPUID_AMD_FEATURE_EDX_LONG_MODE)
|| !(ASMCpuId_EDX(1) & X86_CPUID_FEATURE_EDX_SEP))
#endif
{
LogRel(("WARNING: Can't turn on SYSCALL/SYSRET when the host doesn't support it!!\n"));
return;
}
}
/* Valid for both Intel and AMD CPUs, although only in 64 bits mode for Intel. */
pVM->cpum.s.aGuestCpuIdExt[1].edx |= X86_CPUID_AMD_FEATURE_EDX_SEP;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled syscall/ret\n"));
break;
}
/*
* Set the PAE bit in both feature masks.
* Assumes the caller knows what it's doing! (host must support these)
*/
case CPUMCPUIDFEATURE_PAE:
{
if (!(ASMCpuId_EDX(1) & X86_CPUID_FEATURE_EDX_PAE))
{
LogRel(("WARNING: Can't turn on PAE when the host doesn't support it!!\n"));
return;
}
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].edx |= X86_CPUID_FEATURE_EDX_PAE;
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001
&& pVM->cpum.s.enmGuestCpuVendor == CPUMCPUVENDOR_AMD)
pVM->cpum.s.aGuestCpuIdExt[1].edx |= X86_CPUID_AMD_FEATURE_EDX_PAE;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled PAE\n"));
break;
}
/*
* Set the LONG MODE bit in the extended feature mask.
* Assumes the caller knows what it's doing! (host must support these)
*/
case CPUMCPUIDFEATURE_LONG_MODE:
{
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax < 0x80000001
|| !(ASMCpuId_EDX(0x80000001) & X86_CPUID_AMD_FEATURE_EDX_LONG_MODE))
{
LogRel(("WARNING: Can't turn on LONG MODE when the host doesn't support it!!\n"));
return;
}
/* Valid for both Intel and AMD. */
pVM->cpum.s.aGuestCpuIdExt[1].edx |= X86_CPUID_AMD_FEATURE_EDX_LONG_MODE;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled LONG MODE\n"));
break;
}
/*
* Set the NXE bit in the extended feature mask.
* Assumes the caller knows what it's doing! (host must support these)
*/
case CPUMCPUIDFEATURE_NXE:
{
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax < 0x80000001
|| !(ASMCpuId_EDX(0x80000001) & X86_CPUID_AMD_FEATURE_EDX_NX))
{
LogRel(("WARNING: Can't turn on NXE when the host doesn't support it!!\n"));
return;
}
/* Valid for both Intel and AMD. */
pVM->cpum.s.aGuestCpuIdExt[1].edx |= X86_CPUID_AMD_FEATURE_EDX_NX;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled NXE\n"));
break;
}
case CPUMCPUIDFEATURE_LAHF:
{
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax < 0x80000001
|| !(ASMCpuId_ECX(0x80000001) & X86_CPUID_AMD_FEATURE_ECX_LAHF_SAHF))
{
LogRel(("WARNING: Can't turn on LAHF/SAHF when the host doesn't support it!!\n"));
return;
}
pVM->cpum.s.aGuestCpuIdExt[1].ecx |= X86_CPUID_AMD_FEATURE_ECX_LAHF_SAHF;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled LAHF/SAHF\n"));
break;
}
case CPUMCPUIDFEATURE_PAT:
{
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].edx |= X86_CPUID_FEATURE_EDX_PAT;
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001
&& pVM->cpum.s.enmGuestCpuVendor == CPUMCPUVENDOR_AMD)
pVM->cpum.s.aGuestCpuIdExt[1].edx |= X86_CPUID_AMD_FEATURE_EDX_PAT;
LogRel(("CPUMClearGuestCpuIdFeature: Enabled PAT\n"));
break;
}
case CPUMCPUIDFEATURE_RDTSCP:
{
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax < 0x80000001
|| !(ASMCpuId_EDX(0x80000001) & X86_CPUID_AMD_FEATURE_EDX_RDTSCP))
{
LogRel(("WARNING: Can't turn on RDTSCP when the host doesn't support it!!\n"));
return;
}
/* Valid for AMD only (for now). */
pVM->cpum.s.aGuestCpuIdExt[1].edx |= X86_CPUID_AMD_FEATURE_EDX_RDTSCP;
LogRel(("CPUMSetGuestCpuIdFeature: Enabled RDTSCP.\n"));
break;
}
default:
AssertMsgFailed(("enmFeature=%d\n", enmFeature));
break;
}
for (VMCPUID i = 0; i < pVM->cCpus; i++)
{
PVMCPU pVCpu = &pVM->aCpus[i];
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_CPUID;
}
}
/**
* Queries a CPUID feature bit.
*
* @returns boolean for feature presence
* @param pVM The VM Handle.
* @param enmFeature The feature to query.
*/
VMMDECL(bool) CPUMGetGuestCpuIdFeature(PVM pVM, CPUMCPUIDFEATURE enmFeature)
{
switch (enmFeature)
{
case CPUMCPUIDFEATURE_PAE:
{
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
return !!(pVM->cpum.s.aGuestCpuIdStd[1].edx & X86_CPUID_FEATURE_EDX_PAE);
break;
}
case CPUMCPUIDFEATURE_RDTSCP:
{
if (pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001)
return !!(pVM->cpum.s.aGuestCpuIdExt[1].edx & X86_CPUID_AMD_FEATURE_EDX_RDTSCP);
break;
}
case CPUMCPUIDFEATURE_LONG_MODE:
{
if (pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001)
return !!(pVM->cpum.s.aGuestCpuIdExt[1].edx & X86_CPUID_AMD_FEATURE_EDX_LONG_MODE);
break;
}
default:
AssertMsgFailed(("enmFeature=%d\n", enmFeature));
break;
}
return false;
}
/**
* Clears a CPUID feature bit.
*
* @param pVM The VM Handle.
* @param enmFeature The feature to clear.
*/
VMMDECL(void) CPUMClearGuestCpuIdFeature(PVM pVM, CPUMCPUIDFEATURE enmFeature)
{
switch (enmFeature)
{
/*
* Set the APIC bit in both feature masks.
*/
case CPUMCPUIDFEATURE_APIC:
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].edx &= ~X86_CPUID_FEATURE_EDX_APIC;
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001
&& pVM->cpum.s.enmGuestCpuVendor == CPUMCPUVENDOR_AMD)
pVM->cpum.s.aGuestCpuIdExt[1].edx &= ~X86_CPUID_AMD_FEATURE_EDX_APIC;
Log(("CPUMSetGuestCpuIdFeature: Disabled APIC\n"));
break;
/*
* Clear the x2APIC bit in the standard feature mask.
*/
case CPUMCPUIDFEATURE_X2APIC:
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].ecx &= ~X86_CPUID_FEATURE_ECX_X2APIC;
LogRel(("CPUMSetGuestCpuIdFeature: Disabled x2APIC\n"));
break;
case CPUMCPUIDFEATURE_PAE:
{
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].edx &= ~X86_CPUID_FEATURE_EDX_PAE;
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001
&& pVM->cpum.s.enmGuestCpuVendor == CPUMCPUVENDOR_AMD)
pVM->cpum.s.aGuestCpuIdExt[1].edx &= ~X86_CPUID_AMD_FEATURE_EDX_PAE;
LogRel(("CPUMClearGuestCpuIdFeature: Disabled PAE!\n"));
break;
}
case CPUMCPUIDFEATURE_PAT:
{
if (pVM->cpum.s.aGuestCpuIdStd[0].eax >= 1)
pVM->cpum.s.aGuestCpuIdStd[1].edx &= ~X86_CPUID_FEATURE_EDX_PAT;
if ( pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001
&& pVM->cpum.s.enmGuestCpuVendor == CPUMCPUVENDOR_AMD)
pVM->cpum.s.aGuestCpuIdExt[1].edx &= ~X86_CPUID_AMD_FEATURE_EDX_PAT;
LogRel(("CPUMClearGuestCpuIdFeature: Disabled PAT!\n"));
break;
}
case CPUMCPUIDFEATURE_LONG_MODE:
{
if (pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001)
pVM->cpum.s.aGuestCpuIdExt[1].edx &= ~X86_CPUID_AMD_FEATURE_EDX_LONG_MODE;
break;
}
case CPUMCPUIDFEATURE_LAHF:
{
if (pVM->cpum.s.aGuestCpuIdExt[0].eax >= 0x80000001)
pVM->cpum.s.aGuestCpuIdExt[1].ecx &= ~X86_CPUID_AMD_FEATURE_ECX_LAHF_SAHF;
break;
}
default:
AssertMsgFailed(("enmFeature=%d\n", enmFeature));
break;
}
for (VMCPUID i = 0; i < pVM->cCpus; i++)
{
PVMCPU pVCpu = &pVM->aCpus[i];
pVCpu->cpum.s.fChanged |= CPUM_CHANGED_CPUID;
}
}
/**
* Gets the host CPU vendor
*
* @returns CPU vendor
* @param pVM The VM handle.
*/
VMMDECL(CPUMCPUVENDOR) CPUMGetHostCpuVendor(PVM pVM)
{
return pVM->cpum.s.enmHostCpuVendor;
}
/**
* Gets the CPU vendor
*
* @returns CPU vendor
* @param pVM The VM handle.
*/
VMMDECL(CPUMCPUVENDOR) CPUMGetGuestCpuVendor(PVM pVM)
{
return pVM->cpum.s.enmGuestCpuVendor;
}
VMMDECL(int) CPUMSetGuestDR0(PVMCPU pVCpu, uint64_t uDr0)
{
pVCpu->cpum.s.Guest.dr[0] = uDr0;
return CPUMRecalcHyperDRx(pVCpu);
}
VMMDECL(int) CPUMSetGuestDR1(PVMCPU pVCpu, uint64_t uDr1)
{
pVCpu->cpum.s.Guest.dr[1] = uDr1;
return CPUMRecalcHyperDRx(pVCpu);
}
VMMDECL(int) CPUMSetGuestDR2(PVMCPU pVCpu, uint64_t uDr2)
{
pVCpu->cpum.s.Guest.dr[2] = uDr2;
return CPUMRecalcHyperDRx(pVCpu);
}
VMMDECL(int) CPUMSetGuestDR3(PVMCPU pVCpu, uint64_t uDr3)
{
pVCpu->cpum.s.Guest.dr[3] = uDr3;
return CPUMRecalcHyperDRx(pVCpu);
}
VMMDECL(int) CPUMSetGuestDR6(PVMCPU pVCpu, uint64_t uDr6)
{
pVCpu->cpum.s.Guest.dr[6] = uDr6;
return CPUMRecalcHyperDRx(pVCpu);
}
VMMDECL(int) CPUMSetGuestDR7(PVMCPU pVCpu, uint64_t uDr7)
{
pVCpu->cpum.s.Guest.dr[7] = uDr7;
return CPUMRecalcHyperDRx(pVCpu);
}
VMMDECL(int) CPUMSetGuestDRx(PVMCPU pVCpu, uint32_t iReg, uint64_t Value)
{
AssertReturn(iReg <= USE_REG_DR7, VERR_INVALID_PARAMETER);
/* DR4 is an alias for DR6, and DR5 is an alias for DR7. */
if (iReg == 4 || iReg == 5)
iReg += 2;
pVCpu->cpum.s.Guest.dr[iReg] = Value;
return CPUMRecalcHyperDRx(pVCpu);
}
/**
* Recalculates the hypvervisor DRx register values based on
* current guest registers and DBGF breakpoints.
*
* This is called whenever a guest DRx register is modified and when DBGF
* sets a hardware breakpoint. In guest context this function will reload
* any (hyper) DRx registers which comes out with a different value.
*
* @returns VINF_SUCCESS.
* @param pVCpu The VMCPU handle.
*/
VMMDECL(int) CPUMRecalcHyperDRx(PVMCPU pVCpu)
{
PVM pVM = pVCpu->CTX_SUFF(pVM);
/*
* Compare the DR7s first.
*
* We only care about the enabled flags. The GE and LE flags are always
* set and we don't care if the guest doesn't set them. GD is virtualized
* when we dispatch #DB, we never enable it.
*/
const RTGCUINTREG uDbgfDr7 = DBGFBpGetDR7(pVM);
#ifdef CPUM_VIRTUALIZE_DRX
const RTGCUINTREG uGstDr7 = CPUMGetGuestDR7(pVCpu);
#else
const RTGCUINTREG uGstDr7 = 0;
#endif
if ((uGstDr7 | uDbgfDr7) & X86_DR7_ENABLED_MASK)
{
/*
* Ok, something is enabled. Recalc each of the breakpoints.
* Straight forward code, not optimized/minimized in any way.
*/
RTGCUINTREG uNewDr7 = X86_DR7_GE | X86_DR7_LE | X86_DR7_MB1_MASK;
/* bp 0 */
RTGCUINTREG uNewDr0;
if (uDbgfDr7 & (X86_DR7_L0 | X86_DR7_G0))
{
uNewDr7 |= uDbgfDr7 & (X86_DR7_L0 | X86_DR7_G0 | X86_DR7_RW0_MASK | X86_DR7_LEN0_MASK);
uNewDr0 = DBGFBpGetDR0(pVM);
}
else if (uGstDr7 & (X86_DR7_L0 | X86_DR7_G0))
{
uNewDr7 |= uGstDr7 & (X86_DR7_L0 | X86_DR7_G0 | X86_DR7_RW0_MASK | X86_DR7_LEN0_MASK);
uNewDr0 = CPUMGetGuestDR0(pVCpu);
}
else
uNewDr0 = pVCpu->cpum.s.Hyper.dr[0];
/* bp 1 */
RTGCUINTREG uNewDr1;
if (uDbgfDr7 & (X86_DR7_L1 | X86_DR7_G1))
{
uNewDr7 |= uDbgfDr7 & (X86_DR7_L1 | X86_DR7_G1 | X86_DR7_RW1_MASK | X86_DR7_LEN1_MASK);
uNewDr1 = DBGFBpGetDR1(pVM);
}
else if (uGstDr7 & (X86_DR7_L1 | X86_DR7_G1))
{
uNewDr7 |= uGstDr7 & (X86_DR7_L1 | X86_DR7_G1 | X86_DR7_RW1_MASK | X86_DR7_LEN1_MASK);
uNewDr1 = CPUMGetGuestDR1(pVCpu);
}
else
uNewDr1 = pVCpu->cpum.s.Hyper.dr[1];
/* bp 2 */
RTGCUINTREG uNewDr2;
if (uDbgfDr7 & (X86_DR7_L2 | X86_DR7_G2))
{
uNewDr7 |= uDbgfDr7 & (X86_DR7_L2 | X86_DR7_G2 | X86_DR7_RW2_MASK | X86_DR7_LEN2_MASK);
uNewDr2 = DBGFBpGetDR2(pVM);
}
else if (uGstDr7 & (X86_DR7_L2 | X86_DR7_G2))
{
uNewDr7 |= uGstDr7 & (X86_DR7_L2 | X86_DR7_G2 | X86_DR7_RW2_MASK | X86_DR7_LEN2_MASK);
uNewDr2 = CPUMGetGuestDR2(pVCpu);
}
else
uNewDr2 = pVCpu->cpum.s.Hyper.dr[2];
/* bp 3 */
RTGCUINTREG uNewDr3;
if (uDbgfDr7 & (X86_DR7_L3 | X86_DR7_G3))
{
uNewDr7 |= uDbgfDr7 & (X86_DR7_L3 | X86_DR7_G3 | X86_DR7_RW3_MASK | X86_DR7_LEN3_MASK);
uNewDr3 = DBGFBpGetDR3(pVM);
}
else if (uGstDr7 & (X86_DR7_L3 | X86_DR7_G3))
{
uNewDr7 |= uGstDr7 & (X86_DR7_L3 | X86_DR7_G3 | X86_DR7_RW3_MASK | X86_DR7_LEN3_MASK);
uNewDr3 = CPUMGetGuestDR3(pVCpu);
}
else
uNewDr3 = pVCpu->cpum.s.Hyper.dr[3];
/*
* Apply the updates.
*/
#ifdef IN_RC
if (!(pVCpu->cpum.s.fUseFlags & CPUM_USE_DEBUG_REGS))
{
/** @todo save host DBx registers. */
}
#endif
pVCpu->cpum.s.fUseFlags |= CPUM_USE_DEBUG_REGS;
if (uNewDr3 != pVCpu->cpum.s.Hyper.dr[3])
CPUMSetHyperDR3(pVCpu, uNewDr3);
if (uNewDr2 != pVCpu->cpum.s.Hyper.dr[2])
CPUMSetHyperDR2(pVCpu, uNewDr2);
if (uNewDr1 != pVCpu->cpum.s.Hyper.dr[1])
CPUMSetHyperDR1(pVCpu, uNewDr1);
if (uNewDr0 != pVCpu->cpum.s.Hyper.dr[0])
CPUMSetHyperDR0(pVCpu, uNewDr0);
if (uNewDr7 != pVCpu->cpum.s.Hyper.dr[7])
CPUMSetHyperDR7(pVCpu, uNewDr7);
}
else
{
#ifdef IN_RC
if (pVCpu->cpum.s.fUseFlags & CPUM_USE_DEBUG_REGS)
{
/** @todo restore host DBx registers. */
}
#endif
pVCpu->cpum.s.fUseFlags &= ~CPUM_USE_DEBUG_REGS;
}
Log2(("CPUMRecalcHyperDRx: fUseFlags=%#x %RGr %RGr %RGr %RGr %RGr %RGr\n",
pVCpu->cpum.s.fUseFlags, pVCpu->cpum.s.Hyper.dr[0], pVCpu->cpum.s.Hyper.dr[1],
pVCpu->cpum.s.Hyper.dr[2], pVCpu->cpum.s.Hyper.dr[3], pVCpu->cpum.s.Hyper.dr[6],
pVCpu->cpum.s.Hyper.dr[7]));
return VINF_SUCCESS;
}
/**
* Tests if the guest has No-Execute Page Protection Enabled (NXE).
*
* @returns true if in real mode, otherwise false.
* @param pVCpu The virtual CPU handle.
*/
VMMDECL(bool) CPUMIsGuestNXEnabled(PVMCPU pVCpu)
{
return !!(pVCpu->cpum.s.Guest.msrEFER & MSR_K6_EFER_NXE);
}
/**
* Tests if the guest is running in real mode or not.
*
* @returns true if in real mode, otherwise false.
* @param pVCpu The virtual CPU handle.
*/
VMMDECL(bool) CPUMIsGuestInRealMode(PVMCPU pVCpu)
{
return !(pVCpu->cpum.s.Guest.cr0 & X86_CR0_PE);
}
/**
* Tests if the guest is running in protected or not.
*
* @returns true if in protected mode, otherwise false.
* @param pVCpu The virtual CPU handle.
*/
VMMDECL(bool) CPUMIsGuestInProtectedMode(PVMCPU pVCpu)
{
return !!(pVCpu->cpum.s.Guest.cr0 & X86_CR0_PE);
}
/**
* Tests if the guest is running in paged protected or not.
*
* @returns true if in paged protected mode, otherwise false.
* @param pVCpu The virtual CPU handle.
*/
VMMDECL(bool) CPUMIsGuestInPagedProtectedMode(PVMCPU pVCpu)
{
return (pVCpu->cpum.s.Guest.cr0 & (X86_CR0_PE | X86_CR0_PG)) == (X86_CR0_PE | X86_CR0_PG);
}
/**
* Tests if the guest is running in long mode or not.
*
* @returns true if in long mode, otherwise false.
* @param pVCpu The virtual CPU handle.
*/
VMMDECL(bool) CPUMIsGuestInLongMode(PVMCPU pVCpu)
{
return (pVCpu->cpum.s.Guest.msrEFER & MSR_K6_EFER_LMA) == MSR_K6_EFER_LMA;
}
/**
* Tests if the guest is running in PAE mode or not.
*
* @returns true if in PAE mode, otherwise false.
* @param pVCpu The virtual CPU handle.
*/
VMMDECL(bool) CPUMIsGuestInPAEMode(PVMCPU pVCpu)
{
return (pVCpu->cpum.s.Guest.cr4 & X86_CR4_PAE)
&& (pVCpu->cpum.s.Guest.cr0 & (X86_CR0_PE | X86_CR0_PG)) == (X86_CR0_PE | X86_CR0_PG)
&& !(pVCpu->cpum.s.Guest.msrEFER & MSR_K6_EFER_LMA);
}
#ifndef IN_RING0 /** @todo I don't think we need this in R0, so move it to CPUMAll.cpp? */
/**
* Transforms the guest CPU state to raw-ring mode.
*
* This function will change the any of the cs and ss register with DPL=0 to DPL=1.
*
* @returns VBox status. (recompiler failure)
* @param pVCpu The VMCPU handle.
* @param pCtxCore The context core (for trap usage).
* @see @ref pg_raw
*/
VMMDECL(int) CPUMRawEnter(PVMCPU pVCpu, PCPUMCTXCORE pCtxCore)
{
PVM pVM = pVCpu->CTX_SUFF(pVM);
Assert(!pVM->cpum.s.fRawEntered);
if (!pCtxCore)
pCtxCore = CPUMCTX2CORE(&pVCpu->cpum.s.Guest);
/*
* Are we in Ring-0?
*/
if ( pCtxCore->ss && (pCtxCore->ss & X86_SEL_RPL) == 0
&& !pCtxCore->eflags.Bits.u1VM)
{
/*
* Enter execution mode.
*/
PATMRawEnter(pVM, pCtxCore);
/*
* Set CPL to Ring-1.
*/
pCtxCore->ss |= 1;
if (pCtxCore->cs && (pCtxCore->cs & X86_SEL_RPL) == 0)
pCtxCore->cs |= 1;
}
else
{
AssertMsg((pCtxCore->ss & X86_SEL_RPL) >= 2 || pCtxCore->eflags.Bits.u1VM,
("ring-1 code not supported\n"));
/*
* PATM takes care of IOPL and IF flags for Ring-3 and Ring-2 code as well.
*/
PATMRawEnter(pVM, pCtxCore);
}
/*
* Assert sanity.
*/
AssertMsg((pCtxCore->eflags.u32 & X86_EFL_IF), ("X86_EFL_IF is clear\n"));
AssertReleaseMsg( pCtxCore->eflags.Bits.u2IOPL < (unsigned)(pCtxCore->ss & X86_SEL_RPL)
|| pCtxCore->eflags.Bits.u1VM,
("X86_EFL_IOPL=%d CPL=%d\n", pCtxCore->eflags.Bits.u2IOPL, pCtxCore->ss & X86_SEL_RPL));
Assert((pVCpu->cpum.s.Guest.cr0 & (X86_CR0_PG | X86_CR0_WP | X86_CR0_PE)) == (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP));
pCtxCore->eflags.u32 |= X86_EFL_IF; /* paranoia */
pVM->cpum.s.fRawEntered = true;
return VINF_SUCCESS;
}
/**
* Transforms the guest CPU state from raw-ring mode to correct values.
*
* This function will change any selector registers with DPL=1 to DPL=0.
*
* @returns Adjusted rc.
* @param pVCpu The VMCPU handle.
* @param rc Raw mode return code
* @param pCtxCore The context core (for trap usage).
* @see @ref pg_raw
*/
VMMDECL(int) CPUMRawLeave(PVMCPU pVCpu, PCPUMCTXCORE pCtxCore, int rc)
{
PVM pVM = pVCpu->CTX_SUFF(pVM);
/*
* Don't leave if we've already left (in GC).
*/
Assert(pVM->cpum.s.fRawEntered);
if (!pVM->cpum.s.fRawEntered)
return rc;
pVM->cpum.s.fRawEntered = false;
PCPUMCTX pCtx = &pVCpu->cpum.s.Guest;
if (!pCtxCore)
pCtxCore = CPUMCTX2CORE(pCtx);
Assert(pCtxCore->eflags.Bits.u1VM || (pCtxCore->ss & X86_SEL_RPL));
AssertMsg(pCtxCore->eflags.Bits.u1VM || pCtxCore->eflags.Bits.u2IOPL < (unsigned)(pCtxCore->ss & X86_SEL_RPL),
("X86_EFL_IOPL=%d CPL=%d\n", pCtxCore->eflags.Bits.u2IOPL, pCtxCore->ss & X86_SEL_RPL));
/*
* Are we executing in raw ring-1?
*/
if ( (pCtxCore->ss & X86_SEL_RPL) == 1
&& !pCtxCore->eflags.Bits.u1VM)
{
/*
* Leave execution mode.
*/
PATMRawLeave(pVM, pCtxCore, rc);
/* Not quite sure if this is really required, but shouldn't harm (too much anyways). */
/** @todo See what happens if we remove this. */
if ((pCtxCore->ds & X86_SEL_RPL) == 1)
pCtxCore->ds &= ~X86_SEL_RPL;
if ((pCtxCore->es & X86_SEL_RPL) == 1)
pCtxCore->es &= ~X86_SEL_RPL;
if ((pCtxCore->fs & X86_SEL_RPL) == 1)
pCtxCore->fs &= ~X86_SEL_RPL;
if ((pCtxCore->gs & X86_SEL_RPL) == 1)
pCtxCore->gs &= ~X86_SEL_RPL;
/*
* Ring-1 selector => Ring-0.
*/
pCtxCore->ss &= ~X86_SEL_RPL;
if ((pCtxCore->cs & X86_SEL_RPL) == 1)
pCtxCore->cs &= ~X86_SEL_RPL;
}
else
{
/*
* PATM is taking care of the IOPL and IF flags for us.
*/
PATMRawLeave(pVM, pCtxCore, rc);
if (!pCtxCore->eflags.Bits.u1VM)
{
/** @todo See what happens if we remove this. */
if ((pCtxCore->ds & X86_SEL_RPL) == 1)
pCtxCore->ds &= ~X86_SEL_RPL;
if ((pCtxCore->es & X86_SEL_RPL) == 1)
pCtxCore->es &= ~X86_SEL_RPL;
if ((pCtxCore->fs & X86_SEL_RPL) == 1)
pCtxCore->fs &= ~X86_SEL_RPL;
if ((pCtxCore->gs & X86_SEL_RPL) == 1)
pCtxCore->gs &= ~X86_SEL_RPL;
}
}
return rc;
}
/**
* Updates the EFLAGS while we're in raw-mode.
*
* @param pVCpu The VMCPU handle.
* @param pCtxCore The context core.
* @param eflags The new EFLAGS value.
*/
VMMDECL(void) CPUMRawSetEFlags(PVMCPU pVCpu, PCPUMCTXCORE pCtxCore, uint32_t eflags)
{
PVM pVM = pVCpu->CTX_SUFF(pVM);
if (!pVM->cpum.s.fRawEntered)
{
pCtxCore->eflags.u32 = eflags;
return;
}
PATMRawSetEFlags(pVM, pCtxCore, eflags);
}
#endif /* !IN_RING0 */
/**
* Gets the EFLAGS while we're in raw-mode.
*
* @returns The eflags.
* @param pVCpu The VMCPU handle.
* @param pCtxCore The context core.
*/
VMMDECL(uint32_t) CPUMRawGetEFlags(PVMCPU pVCpu, PCPUMCTXCORE pCtxCore)
{
#ifdef IN_RING0
return pCtxCore->eflags.u32;
#else
PVM pVM = pVCpu->CTX_SUFF(pVM);
if (!pVM->cpum.s.fRawEntered)
return pCtxCore->eflags.u32;
return PATMRawGetEFlags(pVM, pCtxCore);
#endif
}
/**
* Gets and resets the changed flags (CPUM_CHANGED_*).
* Only REM should call this function.
*
* @returns The changed flags.
* @param pVCpu The VMCPU handle.
*/
VMMDECL(unsigned) CPUMGetAndClearChangedFlagsREM(PVMCPU pVCpu)
{
unsigned fFlags = pVCpu->cpum.s.fChanged;
pVCpu->cpum.s.fChanged = 0;
/** @todo change the switcher to use the fChanged flags. */
if (pVCpu->cpum.s.fUseFlags & CPUM_USED_FPU_SINCE_REM)
{
fFlags |= CPUM_CHANGED_FPU_REM;
pVCpu->cpum.s.fUseFlags &= ~CPUM_USED_FPU_SINCE_REM;
}
return fFlags;
}
/**
* Sets the specified changed flags (CPUM_CHANGED_*).
*
* @param pVCpu The VMCPU handle.
*/
VMMDECL(void) CPUMSetChangedFlags(PVMCPU pVCpu, uint32_t fChangedFlags)
{
pVCpu->cpum.s.fChanged |= fChangedFlags;
}
/**
* Checks if the CPU supports the FXSAVE and FXRSTOR instruction.
* @returns true if supported.
* @returns false if not supported.
* @param pVM The VM handle.
*/
VMMDECL(bool) CPUMSupportsFXSR(PVM pVM)
{
return pVM->cpum.s.CPUFeatures.edx.u1FXSR != 0;
}
/**
* Checks if the host OS uses the SYSENTER / SYSEXIT instructions.
* @returns true if used.
* @returns false if not used.
* @param pVM The VM handle.
*/
VMMDECL(bool) CPUMIsHostUsingSysEnter(PVM pVM)
{
return (pVM->cpum.s.fHostUseFlags & CPUM_USE_SYSENTER) != 0;
}
/**
* Checks if the host OS uses the SYSCALL / SYSRET instructions.
* @returns true if used.
* @returns false if not used.
* @param pVM The VM handle.
*/
VMMDECL(bool) CPUMIsHostUsingSysCall(PVM pVM)
{
return (pVM->cpum.s.fHostUseFlags & CPUM_USE_SYSCALL) != 0;
}
#ifndef IN_RING3
/**
* Lazily sync in the FPU/XMM state
*
* @returns VBox status code.
* @param pVCpu VMCPU handle
*/
VMMDECL(int) CPUMHandleLazyFPU(PVMCPU pVCpu)
{
return cpumHandleLazyFPUAsm(&pVCpu->cpum.s);
}
#endif /* !IN_RING3 */
/**
* Checks if we activated the FPU/XMM state of the guest OS
* @returns true if we did.
* @returns false if not.
* @param pVCpu The VMCPU handle.
*/
VMMDECL(bool) CPUMIsGuestFPUStateActive(PVMCPU pVCpu)
{
return (pVCpu->cpum.s.fUseFlags & CPUM_USED_FPU) != 0;
}
/**
* Deactivate the FPU/XMM state of the guest OS
* @param pVCpu The VMCPU handle.
*/
VMMDECL(void) CPUMDeactivateGuestFPUState(PVMCPU pVCpu)
{
pVCpu->cpum.s.fUseFlags &= ~CPUM_USED_FPU;
}
/**
* Checks if the guest debug state is active
*
* @returns boolean
* @param pVM VM handle.
*/
VMMDECL(bool) CPUMIsGuestDebugStateActive(PVMCPU pVCpu)
{
return (pVCpu->cpum.s.fUseFlags & CPUM_USE_DEBUG_REGS) != 0;
}
/**
* Checks if the hyper debug state is active
*
* @returns boolean
* @param pVM VM handle.
*/
VMMDECL(bool) CPUMIsHyperDebugStateActive(PVMCPU pVCpu)
{
return (pVCpu->cpum.s.fUseFlags & CPUM_USE_DEBUG_REGS_HYPER) != 0;
}
/**
* Mark the guest's debug state as inactive
*
* @returns boolean
* @param pVM VM handle.
*/
VMMDECL(void) CPUMDeactivateGuestDebugState(PVMCPU pVCpu)
{
pVCpu->cpum.s.fUseFlags &= ~CPUM_USE_DEBUG_REGS;
}
/**
* Mark the hypervisor's debug state as inactive
*
* @returns boolean
* @param pVM VM handle.
*/
VMMDECL(void) CPUMDeactivateHyperDebugState(PVMCPU pVCpu)
{
pVCpu->cpum.s.fUseFlags &= ~CPUM_USE_DEBUG_REGS_HYPER;
}
/**
* Checks if the hidden selector registers are valid
* @returns true if they are.
* @returns false if not.
* @param pVM The VM handle.
*/
VMMDECL(bool) CPUMAreHiddenSelRegsValid(PVM pVM)
{
return HWACCMIsEnabled(pVM);
}
/**
* Get the current privilege level of the guest.
*
* @returns cpl
* @param pVM VM Handle.
* @param pRegFrame Trap register frame.
*/
VMMDECL(uint32_t) CPUMGetGuestCPL(PVMCPU pVCpu, PCPUMCTXCORE pCtxCore)
{
uint32_t cpl;
if (CPUMAreHiddenSelRegsValid(pVCpu->CTX_SUFF(pVM)))
{
/*
* The hidden CS.DPL register is always equal to the CPL, it is
* not affected by loading a conforming coding segment.
*
* This only seems to apply to AMD-V; in the VT-x case we *do* need to look
* at SS. (ACP2 regression during install after a far call to ring 2)
*/
if (RT_LIKELY(pVCpu->cpum.s.Guest.cr0 & X86_CR0_PE))
cpl = pCtxCore->ssHid.Attr.n.u2Dpl;
else
cpl = 0; /* CPL set to 3 for VT-x real-mode emulation. */
}
else if (RT_LIKELY(pVCpu->cpum.s.Guest.cr0 & X86_CR0_PE))
{
if (RT_LIKELY(!pCtxCore->eflags.Bits.u1VM))
{
/*
* The SS RPL is always equal to the CPL, while the CS RPL
* isn't necessarily equal if the segment is conforming.
* See section 4.11.1 in the AMD manual.
*/
cpl = (pCtxCore->ss & X86_SEL_RPL);
#ifndef IN_RING0
if (cpl == 1)
cpl = 0;
#endif
}
else
cpl = 3;
}
else
cpl = 0; /* real mode; cpl is zero */
return cpl;
}
/**
* Gets the current guest CPU mode.
*
* If paging mode is what you need, check out PGMGetGuestMode().
*
* @returns The CPU mode.
* @param pVCpu The VMCPU handle.
*/
VMMDECL(CPUMMODE) CPUMGetGuestMode(PVMCPU pVCpu)
{
CPUMMODE enmMode;
if (!(pVCpu->cpum.s.Guest.cr0 & X86_CR0_PE))
enmMode = CPUMMODE_REAL;
else if (!(pVCpu->cpum.s.Guest.msrEFER & MSR_K6_EFER_LMA))
enmMode = CPUMMODE_PROTECTED;
else
enmMode = CPUMMODE_LONG;
return enmMode;
}