HWVMXR0.cpp revision 3214a3d99d71a0fb5953fdf35d9a8173f4c58a6d
/* $Id$ */
/** @file
* HM VMX (VT-x) - Host Context Ring-0.
*/
/*
* Copyright (C) 2006-2013 Oracle Corporation
*
* This file is part of VirtualBox Open Source Edition (OSE), as
* available from http://www.virtualbox.org. This file is free software;
* General Public License (GPL) as published by the Free Software
* Foundation, in version 2 as it comes in the "COPYING" file of the
* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
*/
/*******************************************************************************
* Header Files *
*******************************************************************************/
#define LOG_GROUP LOG_GROUP_HM
#include <iprt/asm-amd64-x86.h>
#ifdef VBOX_WITH_REM
#endif
#include "HMInternal.h"
#endif
#include "HWVMXR0.h"
/*******************************************************************************
* Defined Constants And Macros *
*******************************************************************************/
#if defined(RT_ARCH_AMD64)
# define VMX_IS_64BIT_HOST_MODE() (true)
#elif defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
# define VMX_IS_64BIT_HOST_MODE() (g_fVMXIs64bitHost != 0)
#else
# define VMX_IS_64BIT_HOST_MODE() (false)
#endif
do \
{ \
{ \
/* Must override this or else VT-x will fail with invalid guest state errors. */ \
/** @todo we shouldn't have to do this, if it is not 0xf3 it means we screwed up elsewhere (recompiler). */ \
/** @todo VT-x docs explicitly mentions 0xF3. Why not just val = 0xf3 ??. */ \
} \
else \
if ( CPUMIsGuestInRealModeEx(pCtx) \
{ \
/** @todo shouldn't the 'if' condition above check for 'pRealModeTSS' ? */ \
/* Must override this or else VT-x will fail with invalid guest state errors. */ \
val = 0xf3; \
} \
else \
) \
{ \
} \
else \
\
} while (0)
do \
{ \
} while (0)
/* Don't read from the cache in this macro; used only in case of failure where the cache is out of sync. */
do \
{ \
} while (0)
{ \
}
{ \
}
/*******************************************************************************
* Global Variables *
*******************************************************************************/
/* IO operation lookup arrays. */
/** See HMR0A.asm. */
#endif
/*******************************************************************************
* Local Functions *
*******************************************************************************/
/**
* Updates error from VMCS to HMCPU's lasterror record.
*
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param rc The error code.
*/
{
if ( rc == VERR_VMX_UNABLE_TO_START_VM
|| rc == VERR_VMX_INVALID_VMCS_FIELD)
{
}
}
/**
* Sets up and activates VT-x on the current CPU.
*
* @returns VBox status code.
* @param pCpu Pointer to the CPU info struct.
* @param pVM Pointer to the VM. (can be NULL after a resume!!)
* @param pvCpuPage Pointer to the global CPU page.
* @param HCPhysCpuPage Physical address of the global CPU page.
* @param fEnabledByHost Set if SUPR0EnableVTx or similar was used to enable
*/
VMMR0DECL(int) VMXR0EnableCpu(PHMGLOBLCPUINFO pCpu, PVM pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost)
{
if (!fEnabledByHost)
{
if (pVM)
{
/* Set revision dword at the beginning of the VMXON structure. */
}
/** @todo we should unmap the two pages from the virtual address space in order to prevent accidental corruption.
* (which can have very bad consequences!!!)
*/
/** @todo r=bird: Why is this code different than the probing code earlier
* on? It just sets VMXE if needed and doesn't check that it isn't
* set. Mac OS X host_vmxoff may leave this set and we'll fail here
* and debug-assert in the calling code. This is what caused the
* "regression" after backing out the SUPR0EnableVTx code hours before
* 4.2.0GA (reboot fixed the issue). I've changed here to do the same
* as the init code. */
if (!(uCr4 & X86_CR4_VMXE))
/*
* Enter VM root mode.
*/
if (RT_FAILURE(rc))
{
return VERR_VMX_VMXON_FAILED;
}
}
/*
* Flush all VPIDs (in case we or any other hypervisor have been using VPIDs) so that
* we can avoid an explicit flush while using new VPIDs. We would still need to flush
* each time while reusing a VPID after hitting the MaxASID limit once.
*/
if ( pVM
{
pCpu->fFlushAsidBeforeUse = false;
}
else
pCpu->fFlushAsidBeforeUse = true;
/*
* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}.
*/
++pCpu->cTlbFlushes;
return VINF_SUCCESS;
}
/**
* Deactivates VT-x on the current CPU.
*
* @returns VBox status code.
* @param pCpu Pointer to the CPU info struct.
* @param pvCpuPage Pointer to the global CPU page.
* @param HCPhysCpuPage Physical address of the global CPU page.
*/
{
/* If we're somehow not in VMX root mode, then we shouldn't dare leaving it. */
if (!(ASMGetCR4() & X86_CR4_VMXE))
return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
/* Leave VMX Root Mode. */
VMXDisable();
/* And clear the X86_CR4_VMXE bit. */
return VINF_SUCCESS;
}
VMMR0DECL(int) VMXR0GlobalInit(void)
{
/* Nothing to do. */
return VINF_SUCCESS;
}
VMMR0DECL(void) VMXR0GlobalTerm(void)
{
/* Nothing to do. */
}
/**
* Does Ring-0 per VM VT-x initialization.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
*/
{
int rc;
#ifdef LOG_ENABLED
#endif
{
/* Allocate one page for the APIC physical page (serves for filtering accesses). */
if (RT_FAILURE(rc))
return rc;
}
else
{
}
#ifdef VBOX_WITH_CRASHDUMP_MAGIC
{
if (RT_FAILURE(rc))
return rc;
}
#endif
/* Allocate VMCSs for all guest CPUs. */
{
/* Allocate one page for the VM control structure (VMCS). */
if (RT_FAILURE(rc))
return rc;
/* Allocate one page for the virtual APIC page for TPR caching. */
if (RT_FAILURE(rc))
return rc;
/* Allocate the MSR bitmap if this feature is supported. */
{
if (RT_FAILURE(rc))
return rc;
}
/* Allocate one page for the guest MSR load area (for preloading guest MSRs during the world switch). */
if (RT_FAILURE(rc))
return rc;
/* Allocate one page for the host MSR load area (for restoring host MSRs after the world switch back). */
if (RT_FAILURE(rc))
return rc;
#endif /* VBOX_WITH_AUTO_MSR_LOAD_RESTORE */
/* Current guest paging mode. */
#ifdef LOG_ENABLED
SUPR0Printf("VMXR0InitVM %x VMCS=%x (%x)\n", pVM, pVCpu->hm.s.vmx.pvVmcs, (uint32_t)pVCpu->hm.s.vmx.HCPhysVmcs);
#endif
}
return VINF_SUCCESS;
}
/**
* Does Ring-0 per VM VT-x termination.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
*/
{
{
{
}
{
}
{
}
{
}
{
}
#endif /* VBOX_WITH_AUTO_MSR_LOAD_RESTORE */
}
{
}
#ifdef VBOX_WITH_CRASHDUMP_MAGIC
{
}
#endif
return VINF_SUCCESS;
}
/**
* Sets up VT-x for the specified VM.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
*/
{
int rc = VINF_SUCCESS;
/* Initialize these always, see hmR3InitFinalizeR0().*/
/* Determine optimal flush type for EPT. */
{
{
else
{
/*
* Should never really happen. EPT is supported but no suitable flush types supported.
* We cannot ignore EPT at this point as we've already setup Unrestricted Guest execution.
*/
}
}
else
{
/*
* Should never really happen. EPT is supported but INVEPT instruction is not supported.
*/
}
}
/* Determine optimal flush type for VPID. */
{
{
else
{
/*
* Neither SINGLE nor ALL context flush types for VPID supported by the CPU.
* We do not handle other flush type combinations, ignore VPID capabilities.
*/
Log(("VMXR0SetupVM: Only VMX_FLUSH_VPID_INDIV_ADDR supported. Ignoring VPID.\n"));
if (pVM->hm.s.vmx.msr.vmx_ept_vpid_caps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
Log(("VMXR0SetupVM: Only VMX_FLUSH_VPID_SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
}
}
else
{
/*
* Should not really happen. EPT is supported but INVEPT is not supported.
* Ignore VPID capabilities as our code relies on using INVEPT for selective flushing.
*/
Log(("VMXR0SetupVM: VPID supported without INVEPT support. Ignoring VPID.\n"));
}
}
{
/* Set revision dword at the beginning of the VMCS structure. */
*(uint32_t *)pVCpu->hm.s.vmx.pvVmcs = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.msr.vmx_basic_info);
/*
* Clear and activate the VMCS.
*/
if (RT_FAILURE(rc))
goto vmx_end;
if (RT_FAILURE(rc))
goto vmx_end;
/*
* VMX_VMCS_CTRL_PIN_EXEC
* Set required bits to one and zero according to the MSR capabilities.
*/
| VMX_VMCS_CTRL_PIN_EXEC_NMI_EXIT; /* Non-maskable interrupts */
/*
* Enable the VMX preemption timer.
*/
/*
* VMX_VMCS_CTRL_PROC_EXEC
* Set required bits to one and zero according to the MSR capabilities.
*/
/* Program which event cause VM-exits and which features we want to use. */
| VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT; /* don't execute mwait or else we'll idle inside
the guest (host thinks the cpu load is high) */
/* Without nested paging we should intercept invlpg and cr3 mov instructions. */
{
}
/*
* VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT might cause a vmlaunch
* failure with an invalid control fields error. (combined with some other exit reasons)
*/
{
/* CR8 reads from the APIC shadow page; writes cause an exit is they lower the TPR below the threshold */
}
else
/* Exit on CR8 reads & writes in case the TPR shadow feature isn't present. */
{
}
/* We will use the secondary control if it's present. */
/* Mask away the bits that the CPU doesn't support */
/** @todo make sure they don't conflict with the above requirements. */
{
/*
* VMX_VMCS_CTRL_PROC_EXEC2
* Set required bits to one and zero according to the MSR capabilities.
*/
/* Mask away the bits that the CPU doesn't support */
/** @todo make sure they don't conflict with the above requirements. */
}
/*
* VMX_VMCS_CTRL_CR3_TARGET_COUNT
* Set required bits to one and zero according to the MSR capabilities.
*/
/*
* Forward all exception except #NM & #PF to the guest.
* We always need to check pagefaults since our shadow page table can be out of sync.
* And we always lazily sync the FPU & XMM state. .
*/
/** @todo Possible optimization:
* Keep the FPU and XMM state current in the EM thread. That way there's no need to
* lazily sync anything, but the downside is that we can't use the FPU stack or XMM
* registers ourselves of course.
*
* Note: only possible if the current state is actually ours (X86_CR0_TS flag)
*/
/*
* Don't filter page faults, all of them should cause a world switch.
*/
/*
* Set the MSR bitmap address.
*/
{
/*
*/
}
/*
*/
#endif /* VBOX_WITH_AUTO_MSR_LOAD_RESTORE */
{
/* Optional */
}
/* Set link pointer to -1. Not currently used. */
/*
* Clear VMCS, marking it inactive. Clear implementation specific data and writing back
* VMCS data back to memory.
*/
/*
* Configure the VMCS read cache.
*/
/*
* Status code VMCS reads.
*/
{
}
else
} /* for each VMCPU */
/*
* Setup the right TLB function based on CPU capabilities.
*/
else
return rc;
}
/**
* Sets the permission bits for the specified MSR.
*
* @param pVCpu Pointer to the VMCPU.
* @param ulMSR The MSR value.
* @param fRead Whether reading is allowed.
* @param fWrite Whether writing is allowed.
*/
{
unsigned ulBit;
/*
* Layout:
* 0x000 - 0x3ff - Low MSR read bits
* 0x400 - 0x7ff - High MSR read bits
* 0x800 - 0xbff - Low MSR write bits
* 0xc00 - 0xfff - High MSR write bits
*/
if (ulMSR <= 0x00001FFF)
{
/* Pentium-compatible MSRs */
}
else if ( ulMSR >= 0xC0000000
&& ulMSR <= 0xC0001FFF)
{
/* AMD Sixth Generation x86 Processor MSRs */
pvMsrBitmap += 0x400;
}
else
{
AssertFailed();
return;
}
if (fRead)
else
if (fWrite)
else
}
/**
* Injects an event (trap or external interrupt).
*
* @returns VBox status code. Note that it may return VINF_EM_RESET to
* indicate a triple fault when injecting X86_XCPT_DF.
*
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU Context.
* @param intInfo VMX interrupt info.
* @param cbInstr Opcode length of faulting instruction.
* @param errCode Error code (optional).
*/
static int hmR0VmxInjectEvent(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, uint32_t intInfo, uint32_t cbInstr, uint32_t errCode)
{
int rc;
#ifdef VBOX_WITH_STATISTICS
#endif
#ifdef VBOX_STRICT
if (iGate == 0xE)
{
LogFlow(("hmR0VmxInjectEvent: Injecting interrupt %d at %RGv error code=%08x CR2=%RGv intInfo=%08x\n", iGate,
}
else if (iGate < 0x20)
{
LogFlow(("hmR0VmxInjectEvent: Injecting interrupt %d at %RGv error code=%08x\n", iGate, (RTGCPTR)pCtx->rip,
errCode));
}
else
{
}
#endif
if ( CPUMIsGuestInRealModeEx(pCtx)
{
/*
* Injecting events doesn't work right with real mode emulation.
* (#GP if we try to inject external hardware interrupts)
* Inject the interrupt or trap directly instead.
*
* ASSUMES no access handlers for the bits we read or write below (should be safe).
*/
/*
* Check if the interrupt handler is present.
*/
{
Log(("IDT cbIdt violation\n"));
if (iGate != X86_XCPT_DF)
{
return hmR0VmxInjectEvent(pVM, pVCpu, pCtx, intInfo2, 0, 0 /* no error code according to the Intel docs */);
}
Log(("Triple fault -> reset the VM!\n"));
return VINF_EM_RESET;
}
|| iGate == 4)
{
}
else
/*
* Read the selector:offset pair of the interrupt handler.
*/
/*
* Construct the stack frame.
*/
/** @todo Check stack limit. */
rc = PGMPhysSimpleWriteGCPhys(pVM, pCtx->ss.u64Base + pCtx->sp, &pCtx->eflags, sizeof(uint16_t)); AssertRC(rc);
rc = PGMPhysSimpleWriteGCPhys(pVM, pCtx->ss.u64Base + pCtx->sp, &pCtx->cs, sizeof(uint16_t)); AssertRC(rc);
/*
* Update the CPU state for executing the handler.
*/
return VINF_SUCCESS;
}
/*
* Set event injection state.
*/
rc = VMXWriteVmcs(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, intInfo | (1 << VMX_EXIT_INTERRUPTION_INFO_VALID_SHIFT));
return rc;
}
/**
* Checks for pending guest interrupts and injects them.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU context.
*/
{
int rc;
/*
* Dispatch any pending interrupts (injected before, but a VM exit occurred prematurely).
*/
{
Log(("CPU%d: Reinjecting event %RX64 %08x at %RGv cr2=%RX64\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntrInfo,
rc = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, pVCpu->hm.s.Event.u64IntrInfo, 0, pVCpu->hm.s.Event.u32ErrCode);
return VINF_SUCCESS;
}
/*
* If an active trap is already pending, we must forward it first!
*/
if (!TRPMHasTrap(pVCpu))
{
{
return VINF_SUCCESS;
}
/** @todo SMI interrupts. */
/*
* When external interrupts are pending, we should exit the VM when IF is set.
*/
{
{
{
LogFlow(("Enable irq window exit!\n"));
}
/* else nothing to do but wait */
}
{
if (RT_SUCCESS(rc))
{
}
else
{
/* Can only happen in rare cases where a pending interrupt is cleared behind our back */
/* Just continue */
}
}
else
}
}
#ifdef VBOX_STRICT
if (TRPMHasTrap(pVCpu))
{
}
#endif
&& TRPMHasTrap(pVCpu)
)
{
/*
* If a new event is pending, dispatch it now.
*/
/*
* Clear the pending trap.
*/
{
switch (u8Vector)
{
case X86_XCPT_DF:
case X86_XCPT_TS:
case X86_XCPT_NP:
case X86_XCPT_SS:
case X86_XCPT_GP:
case X86_XCPT_PF:
case X86_XCPT_AC:
{
/** @todo r=ramshankar: setting this bit would blow up for real-mode guests with
* unrestricted guest execution. */
/* Valid error codes. */
break;
}
default:
break;
}
if ( u8Vector == X86_XCPT_BP
|| u8Vector == X86_XCPT_OF)
{
}
else
}
else
} /* if (interrupts can be dispatched) */
return VINF_SUCCESS;
}
/**
* Checks for pending VMX events and converts them to TRPM. Before we execute any instruction
* outside of VMX, any pending VMX event must be converted so that it can be delivered properly.
*
* @returns VBox status code.
* @param pVCpu Pointer to the VMCPU.
*/
{
{
/* If a trap was already pending, we did something wrong! */
/*
* Clear the pending event and move it over to TRPM for the rest
* of the world to see.
*/
{
break;
case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT: /** @todo Is classifying #BP, #OF as TRPM_SOFTWARE_INT correct? */
break;
break;
default:
AssertFailed();
}
TRPMAssertTrap(pVCpu, VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVCpu->hm.s.Event.u64IntrInfo), enmTrapType);
//@todo: Is there any situation where we need to call TRPMSetFaultAddress()?
}
return VINF_SUCCESS;
}
/**
* Save the host state into the VMCS.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
*/
{
int rc = VINF_SUCCESS;
/*
* Host CPU Context.
*/
{
/*
* Control registers.
*/
if (VMX_IS_64BIT_HOST_MODE())
{
cr3 = HMR0Get64bitCR3();
}
else
#endif
{
}
/*
* Selector registers.
*/
if (VMX_IS_64BIT_HOST_MODE())
{
}
else
{
/* sysenter loads LDT cs & ss, VMX doesn't like this. Load the GDT ones (safe). */
}
#else
#endif
/* Note: VMX is (again) very picky about the RPL of the selectors here; we'll restore them manually. */
#if HC_ARCH_BITS == 32
if (!VMX_IS_64BIT_HOST_MODE())
{
}
#endif
/*
* GDTR & IDTR.
*/
if (VMX_IS_64BIT_HOST_MODE())
{
}
else
#endif
{
ASMGetGDTR(&gdtr);
ASMGetIDTR(&idtr);
}
/*
* Save the base address of the TR selector.
*/
{
return VERR_VMX_INVALID_HOST_STATE;
}
if (VMX_IS_64BIT_HOST_MODE())
{
}
else
#endif
{
#if HC_ARCH_BITS == 64
#else
#endif
}
/*
* FS base and GS base.
*/
if (VMX_IS_64BIT_HOST_MODE())
{
}
#endif
/*
* Sysenter MSRs.
*/
/** @todo expensive!! */
if (VMX_IS_64BIT_HOST_MODE())
{
}
else
{
}
#else
#endif
/*
* Store all host MSRs in the VM-Exit load area, so they will be reloaded after
* the world switch back to the host.
*/
unsigned idxMsr = 0;
{
pMsr->u32Reserved = 0;
# if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
if (CPUMIsGuestInLongMode(pVCpu))
{
/* Must match the EFER value in our 64 bits switcher. */
}
else
# endif
}
if (VMX_IS_64BIT_HOST_MODE())
{
pMsr->u32Reserved = 0;
pMsr->u32Reserved = 0;
pMsr->u32Reserved = 0;
# if 0
pMsr->u32Reserved = 0;
# endif
}
# endif
{
pMsr->u32Reserved = 0;
}
/** @todo r=ramshankar: check IA32_VMX_MISC bits 27:25 for valid idxMsr
* range. */
#endif /* VBOX_WITH_AUTO_MSR_LOAD_RESTORE */
}
return rc;
}
/**
* Loads the 4 PDPEs into the guest state when nested paging is used and the
* guest operates in PAE mode.
*
* @returns VBox status code.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU context.
*/
{
if (CPUMIsGuestInPAEModeEx(pCtx))
{
}
return VINF_SUCCESS;
}
/**
* Saves the 4 PDPEs into the guest state when nested paging is used and the
* guest operates in PAE mode.
*
* @returns VBox status code.
* @param pVCpu Pointer to the VM CPU.
* @param pCtx Pointer to the guest CPU context.
*
* @remarks Tell PGM about CR3 changes before calling this helper.
*/
{
if (CPUMIsGuestInPAEModeEx(pCtx))
{
int rc;
}
return VINF_SUCCESS;
}
/**
* Update the exception bitmap according to the current CPU state.
*
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU context.
*/
{
/*
* Set up a mask for intercepting traps.
*/
/** @todo Do we really need to always intercept #DB? */
#ifdef VBOX_ALWAYS_TRAP_PF
#endif
#ifdef VBOX_STRICT
#endif
;
/*
* Without nested paging, #PF must be intercepted to implement shadow paging.
*/
/** @todo NP state won't change so maybe we should build the initial trap mask up front? */
/* Catch floating point exceptions if we need to report them to the guest in a different way. */
#ifdef VBOX_STRICT
#endif
/*
* Intercept all exceptions in real mode as none of them can be injected directly (#GP otherwise).
*/
/** @todo Despite the claim to intercept everything, with NP we do not intercept #PF. Should we? */
if ( CPUMIsGuestInRealModeEx(pCtx)
{
;
}
}
/**
* Loads a minimal guest state.
*
* NOTE: Don't do anything here that can cause a jump back to ring 3!!!!!
*
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU context.
*/
{
int rc;
/*
* Load EIP, ESP and EFLAGS.
*/
/*
* Bits 22-31, 15, 5 & 3 must be zero. Bit 1 must be 1.
*/
/*
* Check if real mode emulation using v86 mode.
*/
if ( CPUMIsGuestInRealModeEx(pCtx)
{
}
}
/**
* Sets up TSC offsetting and VMX preemption, if supported/configured/available.
*
* NOTE: This function reads the host TSC value. Therefore it must be executed very
* shortly before a VM entry and execution MUST NOT be rescheduled between a call to
* this function and a VM entry without calling this function again.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
*/
{
int rc = VINF_SUCCESS;
bool fOffsettedTsc;
{
uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVCpu, &fOffsettedTsc, &pVCpu->hm.s.vmx.u64TSCOffset);
/* Make sure the returned values have sane upper and lower boundaries. */
}
else
if (fOffsettedTsc)
{
{
/* Note: VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT takes precedence over TSC_OFFSET, applies to RDTSCP too. */
}
else
{
/* Fall back to rdtsc, rdtscp emulation as we would otherwise pass decreasing tsc values to the guest. */
TMCpuTickGetLastSeen(pVCpu), TMCpuTickGetLastSeen(pVCpu) - u64CurTSC - pVCpu->hm.s.vmx.u64TSCOffset,
TMCpuTickGet(pVCpu)));
}
}
else
{
}
return rc;
}
/**
* Loads the guest state.
*
* NOTE: Don't do anything here that can cause a jump back to ring 3!!!!!
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU context.
*/
{
int rc = VINF_SUCCESS;
/*
* VMX_VMCS_CTRL_ENTRY
* Set required bits to one and zero according to the MSR capabilities.
*/
/*
* Load guest debug controls (DR7 & IA32_DEBUGCTL_MSR).
* Forced to 1 on the 'first' VT-x capable CPUs; this actually includes the newest Nehalem CPUs
*/
if (CPUMIsGuestInLongModeEx(pCtx))
/* else Must be zero when AMD64 is not available. */
/*
* Mask away the bits that the CPU doesn't support.
*/
/*
* VMX_VMCS_CTRL_EXIT
* Set required bits to one and zero according to the MSR capabilities.
*/
/*
* Save debug controls (DR7 & IA32_DEBUGCTL_MSR)
* Forced to 1 on the 'first' VT-x capable CPUs; this actually includes the newest Nehalem CPUs
*/
if (VMX_IS_64BIT_HOST_MODE())
/* else Must be zero when AMD64 is not available. */
if (CPUMIsGuestInLongModeEx(pCtx))
else
#endif
/*
* Don't acknowledge external interrupts on VM-exit.
*/
/*
* Guest CPU context: ES, CS, SS, DS, FS, GS.
*/
{
{
{
/*
* Correct weird requirements for switching to protected mode.
*/
&& enmGuestMode >= PGMMODE_PROTECTED)
{
#ifdef VBOX_WITH_REM
/*
* Flush the recompiler code cache as it's not unlikely the guest will rewrite code
* it will later execute in real mode (OpenBSD 4.0 is one such example)
*/
#endif
/*
* DPL of all hidden selector registers must match the current CPL (0).
*/
}
}
}
}
/*
* Guest CPU context: LDTR.
*/
{
{
rc |= VMXWriteVmcs64(VMX_VMCS_GUEST_LDTR_BASE, 0); /* @todo removing "64" in the function should be the same. */
/* Note: vmlaunch will fail with 0 or just 0x02. No idea why. */
}
else
{
rc |= VMXWriteVmcs64(VMX_VMCS_GUEST_LDTR_BASE, pCtx->ldtr.u64Base); /* @todo removing "64" and it should be the same */
}
}
/*
* Guest CPU context: TR.
*/
{
/*
* Real mode emulation using v86 mode with CR4.VME (interrupt redirection
* using the int bitmap in the TSS).
*/
if ( CPUMIsGuestInRealModeEx(pCtx)
{
/* We convert it here every time as PCI regions could be reconfigured. */
attr.u = 0;
}
else
{
/* The TSS selector must be busy (REM bugs? see defect #XXXX). */
if (!(val & X86_SEL_TYPE_SYS_TSS_BUSY_MASK))
{
if (val & 0xf)
else
/* Default if no TR selector has been set (otherwise vmlaunch will fail!) */
}
AssertMsg((val & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY || (val & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY,
("%#x\n", val));
}
}
/*
* Guest CPU context: GDTR.
*/
{
}
/*
* Guest CPU context: IDTR.
*/
{
}
/*
* Sysenter MSRs.
*/
{
}
/*
* Guest CPU context: Control registers.
*/
{
if (CPUMIsGuestFPUStateActive(pVCpu) == false)
{
}
else
{
/** @todo check if we support the old style mess correctly. */
if (!(val & X86_CR0_NE))
Log(("Forcing X86_CR0_NE!!!\n"));
val |= X86_CR0_NE; /* always turn on the native mechanism to report FPU errors (old style uses interrupts) */
}
/* Protected mode & paging are always enabled; we use them for emulating real and protected mode without paging too. */
{
{
}
else
{
}
}
else
{
/* Note: We must also set this as we rely on protecting various pages for which supervisor writes must be caught. */
val |= X86_CR0_WP;
}
/* Always enable caching. */
/*
* CR0 flags owned by the host; if the guests attempts to change them, then the VM will exit.
*/
| X86_CR0_WP /* Must monitor this bit (it must always be enabled). */
| X86_CR0_PG /* Must monitor this bit (assumptions are made for real mode & protected mode without paging emulation) */
| X86_CR0_CD /* Bit not restored during VM-exit! */
| X86_CR0_NW /* Bit not restored during VM-exit! */
| X86_CR0_NE;
/*
* When the guest's FPU state is active, then we no longer care about the FPU related bits.
*/
if (CPUMIsGuestFPUStateActive(pVCpu) == false)
}
{
/* Set the required bits in cr4 too (currently X86_CR4_VMXE). */
{
{
case PGMMODE_REAL: /* Real mode -> emulated using v86 mode */
case PGMMODE_PROTECTED: /* Protected mode, no paging -> emulated using identity mapping. */
case PGMMODE_32_BIT: /* 32-bit paging. */
val &= ~X86_CR4_PAE;
break;
case PGMMODE_PAE: /* PAE paging. */
case PGMMODE_PAE_NX: /* PAE paging with NX enabled. */
/** Must use PAE paging as we could use physical memory > 4 GB */
val |= X86_CR4_PAE;
break;
case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
#ifdef VBOX_ENABLE_64_BITS_GUESTS
break;
#else
AssertFailed();
#endif
default: /* shut up gcc */
AssertFailed();
}
}
else if ( !CPUMIsGuestInPagedProtectedModeEx(pCtx)
{
/* We use 4 MB pages in our identity mapping page table for real and protected mode without paging. */
val |= X86_CR4_PSE;
/* Our identity mapping is a 32 bits page directory. */
val &= ~X86_CR4_PAE;
}
/*
* Turn off VME if we're in emulated real mode.
*/
if ( CPUMIsGuestInRealModeEx(pCtx)
{
val &= ~X86_CR4_VME;
}
/*
* CR4 flags owned by the host; if the guests attempts to change them, then the VM will exit.
*/
val = 0
| X86_CR4_VMXE;
}
#if 0
/* Enable single stepping if requested and CPU supports it. */
if (DBGFIsStepping(pVCpu))
{
}
#endif
{
{
/** @todo Check the IA32_VMX_EPT_VPID_CAP MSR for other supported memory types. */
{
/* We convert it here every time as PCI regions could be reconfigured. */
/*
* We use our identity mapping page table here as we need to map guest virtual to
* guest physical addresses; EPT will take care of the translation to host physical addresses.
*/
}
else
{
/* Save the real guest CR3 in VMX_VMCS_GUEST_CR3 */
}
}
else
{
}
/* Save our shadow CR3 register. */
}
/*
* Guest CPU context: Debug registers.
*/
{
/* Resync DR7 */
#ifdef DEBUG
/* Sync the hypervisor debug state now if any breakpoint is armed. */
&& !DBGFIsStepping(pVCpu))
{
/* Save the host and load the hypervisor debug state. */
/* DRx intercepts remain enabled. */
/* Override dr7 with the hypervisor value. */
}
else
#endif
/* Sync the debug state now if any breakpoint is armed. */
&& !DBGFIsStepping(pVCpu))
{
/* Disable DRx move intercepts. */
/* Save the host and load the guest debug state. */
}
/* IA32_DEBUGCTL MSR. */
/** @todo do we really ever need this? */
}
/*
* 64-bit guest mode.
*/
if (CPUMIsGuestInLongModeEx(pCtx))
{
#if !defined(VBOX_ENABLE_64_BITS_GUESTS)
#else
# ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
# endif
#endif
{
/* Update these as wrmsr might have changed them. */
}
}
else
{
}
/*
* Store all guest MSRs in the VM-entry load area, so they will be loaded
* during VM-entry and restored into the VM-exit store area during VM-exit.
*/
unsigned idxMsr = 0;
{
pMsr->u32Reserved = 0;
/* VT-x will complain if only MSR_K6_EFER_LME is set. */
if (!CPUMIsGuestInLongModeEx(pCtx))
{
pMsr->u32Reserved = 0;
pMsr->u32Reserved = 0;
pMsr->u32Reserved = 0;
#if 0
pMsr->u32Reserved = 0;
#endif
}
}
{
pMsr->u32Reserved = 0;
}
#endif /* VBOX_WITH_AUTO_MSR_LOAD_RESTORE */
/* Done with the major changes */
/* Minimal guest state update (ESP, EIP, EFLAGS mostly) */
return rc;
}
/**
* Syncs back the guest state from VMCS.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU context.
*/
{
int rc;
/* First sync back EIP, ESP, and EFLAGS. */
/* Take care of instruction fusing (sti, mov ss) */
if (uInterruptState != 0)
{
}
else
/* Control registers. */
/*
* No reason to sync back the CRx registers. They can't be changed by the guest unless in
* the nested paging case where CR3 & CR4 can be changed by the guest.
*/
&& CPUMIsGuestInPagedProtectedModeEx(pCtx)) /** @todo check if we will always catch mode switches and such... */
{
/* Can be updated behind our back in the nested paging case. */
{
}
}
/* Sync back DR7. */
/* Guest CPU context: ES, CS, SS, DS, FS, GS. */
/* System MSRs */
/* Misc. registers; must sync everything otherwise we can get out of sync when jumping to ring 3. */
/* Real mode emulation using v86 mode. */
if ( CPUMIsGuestInRealModeEx(pCtx)
{
/* Hide our emulation flags */
/* Restore original IOPL setting as we always use 0. */
/* Force a TR resync every time in case we switch modes. */
}
else
{
/* In real mode we have a fake TSS, so only sync it back when it's supposed to be valid. */
}
/*
* Save the possibly changed MSRs that we automatically restore and save during a world switch.
*/
{
pMsr += i;
switch (pMsr->u32IndexMSR)
{
case MSR_K8_LSTAR:
break;
case MSR_K6_STAR:
break;
case MSR_K8_SF_MASK:
break;
#if 0
case MSR_K8_KERNEL_GS_BASE:
break;
#endif
case MSR_K8_TSC_AUX:
break;
case MSR_K6_EFER:
/* EFER can't be changed without causing a VM-exit. */
/* Assert(pCtx->msrEFER == pMsr->u64Value); */
break;
default:
AssertFailed();
return VERR_HM_UNEXPECTED_LD_ST_MSR;
}
}
#endif /* VBOX_WITH_AUTO_MSR_LOAD_RESTORE */
return VINF_SUCCESS;
}
/**
* Dummy placeholder for TLB flush handling before VM-entry. Used in the case
* where neither EPT nor VPID is supported by the CPU.
*
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
*/
{
}
/**
* Setup the tagged TLB for EPT+VPID.
*
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
*/
{
pCpu = HMR0GetCurrentCpu();
/*
* Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last
* This can happen both for start & resume due to long jumps back to ring-3.
* If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
*/
bool fNewAsid = false;
{
fNewAsid = true;
}
/*
* Check for explicit TLB shootdowns.
*/
{
if (fNewAsid)
{
++pCpu->uCurrentAsid;
{
pCpu->cTlbFlushes++;
pCpu->fFlushAsidBeforeUse = true;
}
if (pCpu->fFlushAsidBeforeUse)
}
else
{
else
}
}
else
{
("hm->uCurrentAsid=%lu hm->cTlbFlushes=%lu cpu->uCurrentAsid=%lu cpu->cTlbFlushes=%lu\n",
/** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere so this path should
* not be executed. See hmQueueInvlPage() where it is commented
* out. Support individual entry flushing someday. */
{
/*
* Flush individual guest entries using VPID from the TLB or as little as possible with EPT
* as supported by the CPU.
*/
{
}
else
}
else
}
("Flush count mismatch for cpu %d (%x vs %x)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
/* Update VMCS with the VPID. */
}
/**
* Setup the tagged TLB for EPT only.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
*/
{
pCpu = HMR0GetCurrentCpu();
/*
* Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last
* This can happen both for start & resume due to long jumps back to ring-3.
*/
{
}
/*
* Check for explicit TLB shootdown flushes.
*/
else
{
/** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere so this path should
* not be executed. See hmQueueInvlPage() where it is commented
* out. Support individual entry flushing someday. */
{
/*
* We cannot flush individual entries without VPID support. Flush using EPT.
*/
}
}
#ifdef VBOX_WITH_STATISTICS
/** @todo r=ramshankar: this is not accurate anymore with the VPID+EPT
* handling. Should be fixed later. */
else
#endif
}
/**
* Setup the tagged TLB for VPID.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
*/
{
pCpu = HMR0GetCurrentCpu();
/*
* Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last
* This can happen both for start & resume due to long jumps back to ring-3.
* If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
*/
{
/* Force a TLB flush on VM entry. */
}
/*
* Check for explicit TLB shootdown flushes.
*/
{
++pCpu->uCurrentAsid;
{
pCpu->cTlbFlushes++;
pCpu->fFlushAsidBeforeUse = true;
}
if (pCpu->fFlushAsidBeforeUse)
}
else
{
("hm->uCurrentAsid=%lu hm->cTlbFlushes=%lu cpu->uCurrentAsid=%lu cpu->cTlbFlushes=%lu\n",
/** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere so this path should
* not be executed. See hmQueueInvlPage() where it is commented
* out. Support individual entry flushing someday. */
{
/*
* Flush individual guest entries using VPID from the TLB or as little as possible with EPT
* as supported by the CPU.
*/
{
}
else
}
}
("Flush count mismatch for cpu %d (%x vs %x)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
# ifdef VBOX_WITH_STATISTICS
/** @todo r=ramshankar: this is not accurate anymore with EPT+VPID handling.
* Should be fixed later. */
else
# endif
}
/**
* Runs guest code in a VT-x VM.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU context.
*/
{
int rc2;
bool fSetupTPRCaching = false;
bool fNeedTscSetup = true;
uint64_t u64OldLSTAR = 0;
unsigned cResume = 0;
#ifdef VBOX_STRICT
bool fWasInLongMode = false;
#endif
#endif
/*
* Check if we need to use TPR shadowing.
*/
if ( CPUMIsGuestInLongModeEx(pCtx)
)
{
fSetupTPRCaching = true;
}
Log2(("\n"));
/* This is not ideal, but if we don't clear the event injection in the VMCS right here,
* we may end up injecting some stale event into a VM, including injecting an event that
* originated before a VM reset *after* the VM has been reset. See @bugref{6220}.
*/
#ifdef VBOX_STRICT
{
/* allowed zero */
if ((val2 & pVM->hm.s.vmx.msr.vmx_pin_ctls.n.disallowed0) != pVM->hm.s.vmx.msr.vmx_pin_ctls.n.disallowed0)
Log(("Invalid VMX_VMCS_CTRL_PIN_EXEC: zero\n"));
/* allowed one */
Log(("Invalid VMX_VMCS_CTRL_PIN_EXEC: one\n"));
/*
* Must be set according to the MSR, but can be cleared if nested paging is used.
*/
{
}
/* allowed zero */
if ((val2 & pVM->hm.s.vmx.msr.vmx_proc_ctls.n.disallowed0) != pVM->hm.s.vmx.msr.vmx_proc_ctls.n.disallowed0)
Log(("Invalid VMX_VMCS_CTRL_PROC_EXEC: zero\n"));
/* allowed one */
Log(("Invalid VMX_VMCS_CTRL_PROC_EXEC: one\n"));
/* allowed zero */
if ((val2 & pVM->hm.s.vmx.msr.vmx_entry.n.disallowed0) != pVM->hm.s.vmx.msr.vmx_entry.n.disallowed0)
Log(("Invalid VMX_VMCS_CTRL_ENTRY: zero\n"));
/* allowed one */
Log(("Invalid VMX_VMCS_CTRL_ENTRY: one\n"));
/* allowed zero */
Log(("Invalid VMX_VMCS_CTRL_EXIT: zero\n"));
/* allowed one */
Log(("Invalid VMX_VMCS_CTRL_EXIT: one\n"));
}
#endif /* VBOX_STRICT */
#ifdef VBOX_WITH_CRASHDUMP_MAGIC
#endif
/*
* We can jump to this point to resume execution after determining that a VM-exit is innocent.
*/
("Expected %d, I'm %d; cResume=%d exitReason=%RGv exitQualification=%RGv\n",
Assert(!HMR0SuspendPending());
/* Not allowed to switch modes without reloading the host state (32->64 switcher)!! */
/*
* Safety precaution; looping for too long here can have a very bad effect on the host.
*/
{
goto end;
}
/*
* Check for IRQ inhibition due to instruction fusing (sti, mov ss).
*/
{
Log(("VM_FF_INHIBIT_INTERRUPTS at %RGv successor %RGv\n", (RTGCPTR)pCtx->rip, EMGetInhibitInterruptsPC(pVCpu)));
{
/*
* Note: we intentionally don't clear VM_FF_INHIBIT_INTERRUPTS here.
* Before we are able to execute this instruction in raw mode (iret to guest code) an external interrupt might
* force a world switch again. Possibly allowing a guest interrupt to be dispatched in the process. This could
* break the guest. Sounds very unlikely, but such timing sensitive problems are not as rare as you might think.
*/
/* Irq inhibition is no longer active; clear the corresponding VMX state. */
}
}
else
{
/* Irq inhibition is no longer active; clear the corresponding VMX state. */
}
{
{
}
}
#endif
/*
* Check for pending actions that force us to go back to ring-3.
*/
if ( VM_FF_ISPENDING(pVM, VM_FF_HM_TO_R3_MASK | VM_FF_REQUEST | VM_FF_PGM_POOL_FLUSH_PENDING | VM_FF_PDM_DMA)
|| VMCPU_FF_ISPENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK | VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL | VMCPU_FF_REQUEST))
{
/* Check if a sync operation is pending. */
{
rc = PGMSyncCR3(pVCpu, pCtx->cr0, pCtx->cr3, pCtx->cr4, VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
if (rc != VINF_SUCCESS)
{
goto end;
}
}
#ifdef DEBUG
/* Intercept X86_XCPT_DB if stepping is enabled */
if (!DBGFIsStepping(pVCpu))
#endif
{
{
rc = RT_UNLIKELY(VM_FF_ISPENDING(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_NO_MEMORY : VINF_EM_RAW_TO_R3;
goto end;
}
}
/* Pending request packets might contain actions that need immediate attention, such as pending hardware interrupts. */
{
goto end;
}
/* Check if a pgm pool flush is in progress. */
{
goto end;
}
/* Check if DMA work is pending (2nd+ run). */
{
goto end;
}
}
/*
* Exit to ring-3 preemption/work is pending.
*
* Interrupts are disabled before the call to make sure we don't miss any interrupt
* that would flag preemption (IPI, timer tick, ++). (Would've been nice to do this
* further down, but hmR0VmxCheckPendingInterrupt makes that impossible.)
*
* Note! Interrupts must be disabled done *before* we check for TLB flushes; TLB
* shootdowns rely on this.
*/
{
goto end;
}
#endif
/*
* When external interrupts are pending, we should exit the VM when IF is set.
* Note: *After* VM_FF_INHIBIT_INTERRUPTS check!
*/
if (RT_FAILURE(rc))
goto end;
/** @todo check timers?? */
/*
* TPR caching using CR8 is only available in 64-bit mode.
* Note: The 32-bit exception for AMD (X86_CPUID_AMD_FEATURE_ECX_CR8L), but this appears missing in Intel CPUs.
* Note: We can't do this in LoadGuestState() as PDMApicGetTPR can jump back to ring-3 (lock)!! (no longer true) .
*/
/** @todo query and update the TPR only when it could have been changed (mmio
* access & wrsmr (x2apic) */
if (fSetupTPRCaching)
{
/* TPR caching in CR8 */
bool fPending;
/* The TPR can be found at offset 0x80 in the APIC mmio page. */
/*
* Two options here:
* - external interrupt pending, but masked by the TPR value.
* -> a CR8 update that lower the current TPR value should cause an exit
* - no pending interrupts
* -> We don't need to be explicitely notified. There are enough world switches for detecting pending interrupts.
*/
/* cr8 bits 3-0 correspond to bits 7-4 of the task priority mmio register. */
{
/* Our patch code uses LSTAR for TPR caching. */
/** @todo r=ramshankar: we should check for MSR-bitmap support here. */
if (fPending)
{
/* A TPR change could activate a pending interrupt, so catch lstar writes. */
}
else
{
/*
* No interrupts are pending, so we don't need to be explicitely notified.
* There are enough world switches for detecting pending interrupts.
*/
}
}
}
#ifdef LOG_ENABLED
{
{
LogFlow(("Force TLB flush due to rescheduling to a different cpu (%d vs %d)\n", pVCpu->hm.s.idLastCpu,
}
{
pCpu->cTlbFlushes));
}
LogFlow(("Manual TLB flush\n"));
}
#endif
#endif
/*
* NOTE: DO NOT DO ANYTHING AFTER THIS POINT THAT MIGHT JUMP BACK TO RING-3!
* (until the actual world switch)
*/
#ifdef VBOX_STRICT
idCpuCheck = RTMpCpuId();
#endif
#ifdef LOG_ENABLED
#endif
/*
* Save the host state first.
*/
{
{
goto end;
}
}
/*
* Load the guest state.
*/
{
if (fNeedTscSetup)
{
fNeedTscSetup = false;
}
}
else
{
{
goto end;
}
}
/*
* Disable interrupts to make sure a poke will interrupt execution.
* This must be done *before* we check for TLB flushes; TLB shootdowns rely on this.
*/
#endif
/* Non-register state Guest Context */
/** @todo change me according to cpu state */
/* Set TLB flush state as checked until we return from the world switch. */
/* Deal with tagged TLB setup and invalidation. */
/*
* Manual save and restore:
* - General purpose registers except RIP, RSP
*
* Trashed:
* - CR2 (we don't care)
* - LDTR (reset to 0)
* - DRx (presumably not changed at all)
* - DR7 (reset to 0x400)
* - EFLAGS (reset to RT_BIT(1); not relevant)
*/
/* All done! Let's start VM execution. */
#ifdef VBOX_WITH_CRASHDUMP_MAGIC
#endif
/*
* Save the current TPR value in the LSTAR MSR so our patches can access it.
*/
{
}
#ifndef VBOX_WITH_AUTO_MSR_LOAD_RESTORE
/*
* Save the current Host TSC_AUX and write the guest TSC_AUX to the host, so that
* RDTSCPs (that don't cause exits) reads the guest MSR. See @bugref{3324}.
*/
{
uint64_t u64HostTscAux = 0;
}
#endif
#ifdef VBOX_WITH_KERNEL_USING_XMM
rc = HMR0VMXStartVMWrapXMM(pVCpu->hm.s.fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu, pVCpu->hm.s.vmx.pfnStartVM);
#else
rc = pVCpu->hm.s.vmx.pfnStartVM(pVCpu->hm.s.fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu);
#endif
/* Possibly the last TSC value seen by the guest (too high) (only when we're in TSC offset mode). */
{
#ifndef VBOX_WITH_AUTO_MSR_LOAD_RESTORE
/* Restore host's TSC_AUX. */
#endif
ASMReadTSC() + pVCpu->hm.s.vmx.u64TSCOffset - 0x400 /* guestimate of world switch overhead in clock ticks */);
}
/*
* Restore the host LSTAR MSR if the guest could have changed it.
*/
{
}
uOldEFlags = ~(RTCCUINTREG)0;
#endif
AssertMsg(!pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries, ("pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries=%d\n",
/* In case we execute a goto ResumeExecution later on. */
/*
* !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
* IMPORTANT: WE CAN'T DO ANY LOGGING OR OPERATIONS THAT CAN DO A LONGJMP BACK TO RING 3 *BEFORE* WE'VE SYNCED BACK (MOST OF) THE GUEST STATE
* !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
*/
{
goto end;
}
/* Success. Query the guest state and figure out what has happened. */
/* Investigate why there was a VM-exit. */
/* might not be valid; depends on VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID. */
/*
* Sync back the guest state.
*/
/* Note! NOW IT'S SAFE FOR LOGGING! */
#endif
/*
* Check if an injected event was interrupted prematurely.
*/
/* Ignore 'int xx' as they'll be restarted anyway. */
&& VMX_EXIT_INTERRUPTION_INFO_TYPE(pVCpu->hm.s.Event.u64IntrInfo) != VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT
/* Ignore software exceptions (such as int3) as they'll reoccur when we restart the instruction anyway. */
&& VMX_EXIT_INTERRUPTION_INFO_TYPE(pVCpu->hm.s.Event.u64IntrInfo) != VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT)
{
/* Error code present? */
{
Log(("Pending inject %RX64 at %RGv exit=%08x intInfo=%08x exitQualification=%RGv pending error=%RX64\n",
}
else
{
Log(("Pending inject %RX64 at %RGv exit=%08x intInfo=%08x exitQualification=%RGv\n", pVCpu->hm.s.Event.u64IntrInfo,
}
}
#ifdef VBOX_STRICT
/* Ignore software exceptions (such as int3) as they're reoccur when we restart the instruction anyway. */
&& VMX_EXIT_INTERRUPTION_INFO_TYPE(pVCpu->hm.s.Event.u64IntrInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT)
{
Log(("Ignore pending inject %RX64 at %RGv exit=%08x intInfo=%08x exitQualification=%RGv\n",
}
#endif
Log2(("instrInfo=%d instrError=%d instr length=%d\n", (uint32_t)instrInfo, (uint32_t)instrError, (uint32_t)cbInstr));
/*
* Sync back the TPR if it was changed.
*/
if ( fSetupTPRCaching
{
}
#ifdef DBGFTRACE_ENABLED /** @todo DTrace later. */
#endif
/* Some cases don't need a complete resync of the guest CPU state; handle them here. */
switch (exitReason)
{
case VMX_EXIT_XCPT_NMI: /* 0 Exception or non-maskable interrupt (NMI). */
case VMX_EXIT_EXT_INT: /* 1 External interrupt. */
{
{
/* External interrupt; leave to allow it to be dispatched again. */
break;
}
switch (VMX_EXIT_INTERRUPTION_INFO_TYPE(intInfo))
{
case VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI: /* Non-maskable interrupt. */
/* External interrupt; leave to allow it to be dispatched again. */
break;
case VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT: /* External hardware interrupt. */
AssertFailed(); /* can't come here; fails the first check. */
break;
case VMX_EXIT_INTERRUPTION_INFO_TYPE_DB_XCPT: /* Unknown why we get this type for #DB */
case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT: /* Software exception. (#BP or #OF) */
/* no break */
case VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT: /* Hardware exception. */
switch (vector)
{
case X86_XCPT_NM:
{
/** @todo don't intercept #NM exceptions anymore when we've activated the guest FPU state. */
/* If we sync the FPU/XMM state on-demand, then we can continue execution as if nothing has happened. */
if (rc == VINF_SUCCESS)
{
/* Continue execution. */
goto ResumeExecution;
}
Log(("Forward #NM fault to the guest\n"));
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
cbInstr, 0);
goto ResumeExecution;
}
case X86_XCPT_PF: /* Page fault */
{
#ifdef VBOX_ALWAYS_TRAP_PF
{
/*
* A genuine pagefault. Forward the trap to the guest by injecting the exception and resuming execution.
*/
Log(("Guest page fault at %RGv cr2=%RGv error code %RGv rsp=%RGv\n", (RTGCPTR)pCtx->rip, exitQualification,
/* Now we must update CR2. */
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
goto ResumeExecution;
}
#else
#endif
#ifdef VBOX_HM_WITH_GUEST_PATCHING
/* Shortcut for APIC TPR reads and writes; 32 bits guests only */
&& CPUMGetGuestCPL(pVCpu) == 0
{
if ( rc == VINF_SUCCESS
&& GCPhys == GCPhysApicBase)
{
/* Only attempt to patch the instruction once. */
if (!pPatch)
{
break;
}
}
}
#endif
/* Exit qualification contains the linear address of the page fault. */
/* Shortcut for APIC TPR reads and writes. */
{
if ( rc == VINF_SUCCESS
&& GCPhys == GCPhysApicBase)
{
Log(("Enable VT-x virtual APIC access filtering\n"));
X86_PTE_RW | X86_PTE_P);
}
}
/* Forward it to our trap handler first, in case our shadow pages are out of sync. */
if (rc == VINF_SUCCESS)
{ /* We've successfully synced our shadow pages, so let's just continue execution. */
Log2(("Shadow page fault at %RGv cr2=%RGv error code %x\n", (RTGCPTR)pCtx->rip, exitQualification ,errCode));
goto ResumeExecution;
}
else if (rc == VINF_EM_RAW_GUEST_TRAP)
{
/*
* A genuine pagefault. Forward the trap to the guest by injecting the exception and resuming execution.
*/
Log2(("Forward page fault to the guest\n"));
/* The error code might have been changed. */
/* Now we must update CR2. */
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
goto ResumeExecution;
}
#ifdef VBOX_STRICT
#endif
/* Need to go back to the recompiler to emulate the instruction. */
/* If event delivery caused the #PF (shadow or not), tell TRPM. */
break;
}
case X86_XCPT_MF: /* Floating point exception. */
{
{
/* old style FPU error reporting needs some extra work. */
/** @todo don't fall back to the recompiler, but do it manually. */
break;
}
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
goto ResumeExecution;
}
case X86_XCPT_DB: /* Debug exception. */
{
/*
* DR6, DR7.GD and IA32_DEBUGCTL.LBR are not updated yet.
*
* Exit qualification bits:
* 3:0 B0-B3 which breakpoint condition was met
* 12:4 Reserved (0)
* 13 BD - debug register access detected
* 14 BS - single step execution or branch taken
* 63:15 Reserved (0)
*/
/* Note that we don't support guest and host-initiated debugging at the same time. */
if (rc == VINF_EM_RAW_GUEST_TRAP)
{
/* Update DR6 here. */
/* Resync DR6 if the debug state is active. */
/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
/* Paranoia. */
/* Resync DR7 */
Log(("Trap %x (debug) at %RGv exit qualification %RX64 dr6=%x dr7=%x\n", vector, (RTGCPTR)pCtx->rip,
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
goto ResumeExecution;
}
/* Return to ring 3 to deal with the debug exit code. */
Log(("Debugger hardware BP at %04x:%RGv (rc=%Rrc)\n", pCtx->cs.Sel, pCtx->rip, VBOXSTRICTRC_VAL(rc)));
break;
}
case X86_XCPT_BP: /* Breakpoint. */
{
if (rc == VINF_EM_RAW_GUEST_TRAP)
{
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
goto ResumeExecution;
}
if (rc == VINF_SUCCESS)
{
goto ResumeExecution;
}
break;
}
case X86_XCPT_GP: /* General protection failure exception. */
{
#ifdef VBOX_STRICT
if ( !CPUMIsGuestInRealModeEx(pCtx)
{
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
goto ResumeExecution;
}
#endif
LogFlow(("Real mode X86_XCPT_GP instruction emulation at %x:%RGv\n", pCtx->cs.Sel, (RTGCPTR)pCtx->rip));
if (RT_SUCCESS(rc2))
{
bool fUpdateRIP = true;
rc = VINF_SUCCESS;
{
case OP_CLI:
break;
case OP_STI:
break;
case OP_HLT:
fUpdateRIP = false;
rc = VINF_EM_HALT;
break;
case OP_POPF:
{
{
cbParm = 4;
uMask = 0xffffffff;
}
else
{
cbParm = 2;
uMask = 0xffff;
}
if (RT_FAILURE(rc2))
{
break;
}
eflags.u = 0;
if (RT_FAILURE(rc2))
{
break;
}
/* RF cleared when popped in real mode; see pushf description in AMD manual. */
break;
}
case OP_PUSHF:
{
{
cbParm = 4;
uMask = 0xffffffff;
}
else
{
cbParm = 2;
uMask = 0xffff;
}
&GCPtrStack);
if (RT_FAILURE(rc2))
{
break;
}
/* RF & VM cleared when pushed in real mode; see pushf description in AMD manual. */
if (RT_FAILURE(rc2))
{
break;
}
break;
}
case OP_IRET:
{
{
break;
}
if (RT_FAILURE(rc2))
{
break;
}
if (RT_FAILURE(rc2))
{
break;
}
fUpdateRIP = false;
break;
}
case OP_INT:
{
fUpdateRIP = false;
break;
}
case OP_INTO:
{
{
LogFlow(("Realmode: INTO\n"));
fUpdateRIP = false;
}
break;
}
case OP_INT3:
{
LogFlow(("Realmode: INT 3\n"));
intInfo2 = 3;
fUpdateRIP = false;
break;
}
default:
fUpdateRIP = false;
break;
}
if (rc == VINF_SUCCESS)
{
if (fUpdateRIP)
/*
* LIDT, LGDT can end up here. In the future CRx changes as well. Just reload the
* whole context to be done with it.
*/
/* Only resume if successful. */
goto ResumeExecution;
}
}
else
break;
}
#ifdef VBOX_STRICT
case X86_XCPT_XF: /* SIMD exception. */
case X86_XCPT_DE: /* Divide error. */
case X86_XCPT_UD: /* Unknown opcode exception. */
case X86_XCPT_SS: /* Stack segment exception. */
case X86_XCPT_NP: /* Segment not present exception. */
{
switch (vector)
{
}
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
goto ResumeExecution;
}
#endif
default:
if ( CPUMIsGuestInRealModeEx(pCtx)
{
/* Go back to ring-3 in case of a triple fault. */
if ( vector == X86_XCPT_DF
&& rc == VINF_EM_RESET)
{
break;
}
goto ResumeExecution;
}
break;
} /* switch (vector) */
break;
default:
break;
}
break;
}
/*
* 48 EPT violation. An attempt to access memory with a guest-physical address was disallowed
* by the configuration of the EPT paging structures.
*/
case VMX_EXIT_EPT_VIOLATION:
{
/* Determine the kind of violation. */
errCode = 0;
/* If the page is present, then it's a page level protection fault. */
errCode |= X86_TRAP_PF_P;
else
{
/* Shortcut for APIC TPR reads and writes. */
{
{
Log(("Enable VT-x virtual APIC access filtering\n"));
X86_PTE_RW | X86_PTE_P);
}
}
}
/* GCPhys contains the guest physical address of the page fault. */
/* Handle the pagefault trap for the nested shadow table. */
/*
* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment below, @bugref{6043}.
*/
if ( rc == VINF_SUCCESS
|| rc == VERR_PAGE_NOT_PRESENT)
{
/* We've successfully synced our shadow pages, so let's just continue execution. */
Log2(("Shadow page fault at %RGv cr2=%RGp error code %x\n", (RTGCPTR)pCtx->rip, exitQualification , errCode));
goto ResumeExecution;
}
#ifdef VBOX_STRICT
if (rc != VINF_EM_RAW_EMULATE_INSTR)
LogFlow(("PGMTrap0eHandlerNestedPaging at %RGv failed with %Rrc\n", (RTGCPTR)pCtx->rip, VBOXSTRICTRC_VAL(rc)));
#endif
/* Need to go back to the recompiler to emulate the instruction. */
break;
}
case VMX_EXIT_EPT_MISCONFIG:
{
/* Shortcut for APIC TPR reads and writes. */
{
{
Log(("Enable VT-x virtual APIC access filtering\n"));
X86_PTE_RW | X86_PTE_P);
}
}
rc = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pCtx), GCPhys, UINT32_MAX);
/*
* If we succeed, resume execution.
* Or, if fail in interpreting the instruction because we couldn't get the guest physical address
* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
* weird case. See @bugref{6043}.
*/
if ( rc == VINF_SUCCESS
|| rc == VERR_PAGE_NOT_PRESENT)
{
goto ResumeExecution;
}
Log2(("PGMR0Trap0eHandlerNPMisconfig(,,,%RGp) at %RGv -> %Rrc\n", GCPhys, (RTGCPTR)pCtx->rip, VBOXSTRICTRC_VAL(rc)));
break;
}
case VMX_EXIT_INT_WINDOW: /* 7 Interrupt window exiting. */
/* Clear VM-exit on IF=1 change. */
VMCPU_FF_ISPENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC|VMCPU_FF_INTERRUPT_PIC)), pCtx->eflags.Bits.u1IF));
goto ResumeExecution; /* we check for pending guest interrupts there */
case VMX_EXIT_WBINVD: /* 54 Guest software attempted to execute WBINVD. (conditional) */
case VMX_EXIT_INVD: /* 13 Guest software attempted to execute INVD. (unconditional) */
/* Skip instruction and continue directly. */
/* Continue execution.*/
goto ResumeExecution;
case VMX_EXIT_CPUID: /* 10 Guest software attempted to execute CPUID. */
{
if (rc == VINF_SUCCESS)
{
/* Update EIP and continue execution. */
goto ResumeExecution;
}
break;
}
case VMX_EXIT_RDPMC: /* 15 Guest software attempted to execute RDPMC. */
{
if (rc == VINF_SUCCESS)
{
/* Update EIP and continue execution. */
goto ResumeExecution;
}
break;
}
case VMX_EXIT_RDTSC: /* 16 Guest software attempted to execute RDTSC. */
{
Log2(("VMX: Rdtsc\n"));
if (rc == VINF_SUCCESS)
{
/* Update EIP and continue execution. */
fNeedTscSetup = true; /* See @bugref{6634}. */
goto ResumeExecution;
}
break;
}
case VMX_EXIT_RDTSCP: /* 51 Guest software attempted to execute RDTSCP. */
{
Log2(("VMX: Rdtscp\n"));
if (rc == VINF_SUCCESS)
{
/* Update EIP and continue execution. */
fNeedTscSetup = true; /* See @bugref{6634}. */
goto ResumeExecution;
}
break;
}
case VMX_EXIT_INVLPG: /* 14 Guest software attempted to execute INVLPG. */
{
Log2(("VMX: invlpg\n"));
if (rc == VINF_SUCCESS)
{
/* Update EIP and continue execution. */
goto ResumeExecution;
}
AssertMsg(rc == VERR_EM_INTERPRETER, ("EMU: invlpg %RGv failed with %Rrc\n", exitQualification, VBOXSTRICTRC_VAL(rc)));
break;
}
case VMX_EXIT_MONITOR: /* 39 Guest software attempted to execute MONITOR. */
{
Log2(("VMX: monitor\n"));
if (rc == VINF_SUCCESS)
{
/* Update EIP and continue execution. */
goto ResumeExecution;
}
break;
}
case VMX_EXIT_WRMSR: /* 32 WRMSR. Guest software attempted to execute WRMSR. */
/* When an interrupt is pending, we'll let MSR_K8_LSTAR writes fault in our TPR patch code. */
{
{
/* Our patch code uses LSTAR for TPR caching. */
}
/* Skip the instruction and continue. */
/* Only resume if successful. */
goto ResumeExecution;
}
/* no break */
case VMX_EXIT_RDMSR: /* 31 RDMSR. Guest software attempted to execute RDMSR. */
{
STAM_COUNTER_INC((exitReason == VMX_EXIT_RDMSR) ? &pVCpu->hm.s.StatExitRdmsr : &pVCpu->hm.s.StatExitWrmsr);
if (rc == VINF_SUCCESS)
{
/* EIP has been updated already. */
/* Only resume if successful. */
goto ResumeExecution;
}
break;
}
case VMX_EXIT_MOV_CRX: /* 28 Control-register accesses. */
{
{
{
Log2(("VMX: %RGv mov cr%d, x\n", (RTGCPTR)pCtx->rip, VMX_EXIT_QUALIFICATION_CRX_REGISTER(exitQualification)));
STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxWrite[VMX_EXIT_QUALIFICATION_CRX_REGISTER(exitQualification)]);
{
case 0:
break;
case 2:
break;
case 3:
break;
case 4:
break;
case 8:
/* CR8 contains the APIC TPR */
break;
default:
AssertFailed();
break;
}
break;
}
{
Log2(("VMX: mov x, crx\n"));
STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxRead[VMX_EXIT_QUALIFICATION_CRX_REGISTER(exitQualification)]);
/* CR8 reads only cause an exit when the TPR shadow feature isn't present. */
break;
}
{
Log2(("VMX: clts\n"));
break;
}
{
rc = EMInterpretLMSW(pVM, pVCpu, CPUMCTX2CORE(pCtx), VMX_EXIT_QUALIFICATION_CRX_LMSW_DATA(exitQualification));
break;
}
}
/* Update EIP if no error occurred. */
if (RT_SUCCESS(rc))
if (rc == VINF_SUCCESS)
{
/* Only resume if successful. */
goto ResumeExecution;
}
break;
}
case VMX_EXIT_MOV_DRX: /* 29 Debug-register accesses. */
{
if ( !DBGFIsStepping(pVCpu)
{
/* Disable DRx move intercepts. */
/* Save the host and load the guest debug state. */
#ifdef LOG_ENABLED
if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(exitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
{
Log(("VMX_EXIT_MOV_DRX: write DR%d genreg %d\n", VMX_EXIT_QUALIFICATION_DRX_REGISTER(exitQualification),
}
else
#endif
#ifdef VBOX_WITH_STATISTICS
if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(exitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
else
#endif
goto ResumeExecution;
}
/** @todo clear VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT after the first
* time and restore DRx registers afterwards */
if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(exitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
{
}
else
{
Log2(("VMX: mov x, DRx\n"));
}
/* Update EIP if no error occurred. */
if (RT_SUCCESS(rc))
if (rc == VINF_SUCCESS)
{
/* Only resume if successful. */
goto ResumeExecution;
}
break;
}
/* Note: We'll get a #GP if the IO instruction isn't allowed (IOPL or TSS bitmap); no need to double check. */
case VMX_EXIT_IO_INSTR: /* 30 I/O instruction. */
{
bool fIOWrite = (VMX_EXIT_QUALIFICATION_IO_DIRECTION(exitQualification) == VMX_EXIT_QUALIFICATION_IO_DIRECTION_OUT);
/** @todo necessary to make the distinction? */
else
{
break;
}
{
/* Disassemble manually to deal with segment prefixes. */
/** @todo VMX_VMCS_RO_EXIT_GUEST_LINEAR_ADDR contains the flat pointer
* operand of the instruction. */
/** @todo VMX_VMCS32_RO_EXIT_INSTR_INFO also contains segment prefix info. */
if (RT_SUCCESS(rc))
{
if (fIOWrite)
{
}
else
{
}
}
else
}
else
{
if (fIOWrite)
{
if (rc == VINF_IOM_R3_IOPORT_WRITE)
}
else
{
if (IOM_SUCCESS(rc))
{
/* Write back to the EAX register. */
}
else
if (rc == VINF_IOM_R3_IOPORT_READ)
}
}
/*
* Handled the I/O return codes.
* (The unhandled cases end up with rc == VINF_EM_RAW_EMULATE_INSTR.)
*/
if (IOM_SUCCESS(rc))
{
/* Update EIP and continue execution. */
{
/* If any IO breakpoints are armed, then we should check if a debug trap needs to be generated. */
{
for (unsigned i = 0; i < 4; i++)
{
{
/* Clear all breakpoint status flags and set the one we just hit. */
/*
* Note: AMD64 Architecture Programmer's Manual 13.1:
* Bits 15:13 of the DR6 register is never cleared by the processor and must
* be cleared by software after the contents have been read.
*/
/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
/* Paranoia. */
/* Resync DR7 */
/* Construct inject info. */
rc2 = hmR0VmxInjectEvent(pVM, pVCpu, pCtx, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(intInfo),
0 /* cbInstr */, 0 /* errCode */);
goto ResumeExecution;
}
}
}
goto ResumeExecution;
}
break;
}
#ifdef VBOX_STRICT
if (rc == VINF_IOM_R3_IOPORT_READ)
else if (rc == VINF_IOM_R3_IOPORT_WRITE)
else
{
|| rc == VINF_EM_RAW_EMULATE_INSTR
|| rc == VINF_EM_RAW_GUEST_TRAP
}
#endif
break;
}
case VMX_EXIT_TPR_BELOW_THRESHOLD: /* 43 TPR below threshold. Guest software executed MOV to CR8. */
LogFlow(("VMX_EXIT_TPR_BELOW_THRESHOLD\n"));
/* RIP is already set to the next instruction and the TPR has been synced back. Just resume. */
goto ResumeExecution;
case VMX_EXIT_APIC_ACCESS: /* 44 APIC access. Guest software attempted to access memory at a physical address
on the APIC-access page. */
{
LogFlow(("VMX_EXIT_APIC_ACCESS\n"));
switch (uAccessType)
{
{
rc = IOMMMIOPhysHandler(pVM, pVCpu, (uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ) ? 0 : X86_TRAP_PF_RW,
if (rc == VINF_SUCCESS)
goto ResumeExecution; /* rip already updated */
break;
}
default:
break;
}
break;
}
case VMX_EXIT_PREEMPT_TIMER: /* 52 VMX-preemption timer expired. The preemption timer counted down to zero. */
goto ResumeExecution;
break;
default:
/* The rest is handled after syncing the entire CPU state. */
break;
}
/*
* Note: The guest state is not entirely synced back at this stage!
*/
/* Investigate why there was a VM-exit. (part 2) */
switch (exitReason)
{
case VMX_EXIT_XCPT_NMI: /* 0 Exception or non-maskable interrupt (NMI). */
case VMX_EXIT_EXT_INT: /* 1 External interrupt. */
case VMX_EXIT_EPT_VIOLATION:
case VMX_EXIT_EPT_MISCONFIG: /* 49 EPT misconfig is used by the PGM/MMIO optimizations. */
case VMX_EXIT_PREEMPT_TIMER: /* 52 VMX-preemption timer expired. The preemption timer counted down to zero. */
/* Already handled above. */
break;
case VMX_EXIT_TRIPLE_FAULT: /* 2 Triple fault. */
break;
case VMX_EXIT_INIT_SIGNAL: /* 3 INIT signal. */
case VMX_EXIT_SIPI: /* 4 Start-up IPI (SIPI). */
AssertFailed(); /* Can't happen. Yet. */
break;
case VMX_EXIT_IO_SMI: /* 5 I/O system-management interrupt (SMI). */
case VMX_EXIT_SMI: /* 6 Other SMI. */
AssertFailed(); /* Can't happen afaik. */
break;
case VMX_EXIT_TASK_SWITCH: /* 9 Task switch: too complicated to emulate, so fall back to the recompiler */
if ( (VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE(exitQualification) == VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE_IDT)
{
/* Caused by an injected interrupt. */
Log(("VMX_EXIT_TASK_SWITCH: reassert trap %d\n", VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVCpu->hm.s.Event.u64IntrInfo)));
//@todo: Why do we assume this had to be a hardware interrupt? What about software interrupts or exceptions?
rc2 = TRPMAssertTrap(pVCpu, VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVCpu->hm.s.Event.u64IntrInfo), TRPM_HARDWARE_INT);
}
/* else Exceptions and software interrupts can just be restarted. */
break;
case VMX_EXIT_HLT: /* 12 Guest software attempted to execute HLT. */
/* Check if external interrupts are pending; if so, don't switch back. */
goto ResumeExecution;
rc = VINF_EM_HALT;
break;
case VMX_EXIT_MWAIT: /* 36 Guest software executed MWAIT. */
Log2(("VMX: mwait\n"));
if ( rc == VINF_EM_HALT
|| rc == VINF_SUCCESS)
{
/* Update EIP and continue execution. */
/* Check if external interrupts are pending; if so, don't switch back. */
if ( rc == VINF_SUCCESS
|| ( rc == VINF_EM_HALT
)
goto ResumeExecution;
}
AssertMsg(rc == VERR_EM_INTERPRETER || rc == VINF_EM_HALT, ("EMU: mwait failed with %Rrc\n", VBOXSTRICTRC_VAL(rc)));
break;
case VMX_EXIT_RSM: /* 17 Guest software attempted to execute RSM in SMM. */
AssertFailed(); /* can't happen. */
break;
case VMX_EXIT_MTF: /* 37 Exit due to Monitor Trap Flag. */
#if 0
#endif
break;
case VMX_EXIT_VMCALL: /* 18 Guest software executed VMCALL. */
case VMX_EXIT_VMCLEAR: /* 19 Guest software executed VMCLEAR. */
case VMX_EXIT_VMLAUNCH: /* 20 Guest software executed VMLAUNCH. */
case VMX_EXIT_VMPTRLD: /* 21 Guest software executed VMPTRLD. */
case VMX_EXIT_VMPTRST: /* 22 Guest software executed VMPTRST. */
case VMX_EXIT_VMREAD: /* 23 Guest software executed VMREAD. */
case VMX_EXIT_VMRESUME: /* 24 Guest software executed VMRESUME. */
case VMX_EXIT_VMWRITE: /* 25 Guest software executed VMWRITE. */
case VMX_EXIT_VMXOFF: /* 26 Guest software executed VMXOFF. */
case VMX_EXIT_VMXON: /* 27 Guest software executed VMXON. */
/** @todo inject #UD immediately */
break;
case VMX_EXIT_CPUID: /* 10 Guest software attempted to execute CPUID. */
case VMX_EXIT_RDTSC: /* 16 Guest software attempted to execute RDTSC. */
case VMX_EXIT_INVLPG: /* 14 Guest software attempted to execute INVLPG. */
case VMX_EXIT_MOV_CRX: /* 28 Control-register accesses. */
case VMX_EXIT_MOV_DRX: /* 29 Debug-register accesses. */
case VMX_EXIT_IO_INSTR: /* 30 I/O instruction. */
case VMX_EXIT_RDPMC: /* 15 Guest software attempted to execute RDPMC. */
case VMX_EXIT_RDTSCP: /* 51 Guest software attempted to execute RDTSCP. */
/* already handled above */
|| rc == VINF_EM_RAW_INTERRUPT
|| rc == VERR_EM_INTERPRETER
|| rc == VINF_EM_RAW_EMULATE_INSTR
|| rc == VINF_PGM_SYNC_CR3
|| rc == VINF_IOM_R3_IOPORT_READ
|| rc == VINF_IOM_R3_IOPORT_WRITE
|| rc == VINF_EM_RAW_GUEST_TRAP
|| rc == VINF_TRPM_XCPT_DISPATCHED
|| rc == VINF_EM_RESCHEDULE_REM,
break;
case VMX_EXIT_TPR_BELOW_THRESHOLD: /* 43 TPR below threshold. Guest software executed MOV to CR8. */
case VMX_EXIT_RDMSR: /* 31 RDMSR. Guest software attempted to execute RDMSR. */
case VMX_EXIT_WRMSR: /* 32 WRMSR. Guest software attempted to execute WRMSR. */
case VMX_EXIT_PAUSE: /* 40 Guest software attempted to execute PAUSE. */
case VMX_EXIT_MONITOR: /* 39 Guest software attempted to execute MONITOR. */
case VMX_EXIT_APIC_ACCESS: /* 44 APIC access. Guest software attempted to access memory at a physical address
on the APIC-access page. */
{
/*
* If we decided to emulate them here, then we must sync the MSRs that could have been changed (sysenter, FS/GS base)
*/
break;
}
case VMX_EXIT_INT_WINDOW: /* 7 Interrupt window. */
break;
case VMX_EXIT_ERR_INVALID_GUEST_STATE: /* 33 VM-entry failure due to invalid guest state. */
{
#ifdef VBOX_STRICT
RTCCUINTREG val2 = 0;
Log(("VMX_EXIT_ERR_INVALID_GUEST_STATE\n"));
#endif /* VBOX_STRICT */
break;
}
case VMX_EXIT_ERR_MSR_LOAD: /* 34 VM-entry failure due to MSR loading. */
case VMX_EXIT_ERR_MACHINE_CHECK: /* 41 VM-entry failure due to machine-check. */
default:
break;
}
end:
/* We now going back to ring-3, so clear the action flag. */
/*
* Signal changes for the recompiler.
*/
/*
* If we executed vmlaunch/vmresume and an external IRQ was pending, then we don't have to do a full sync the next time.
*/
if ( exitReason == VMX_EXIT_EXT_INT
{
/* On the next entry we'll only sync the host context. */
}
else
{
/* On the next entry we'll sync everything. */
/** @todo we can do better than this */
/* Not in the VINF_PGM_CHANGE_MODE though! */
}
/* Translate into a less severe return code */
if (rc == VERR_EM_INTERPRETER)
else if (rc == VERR_VMX_INVALID_VMCS_PTR)
{
/* Try to extract more information about what might have gone wrong here. */
}
/* Just set the correct state here instead of trying to catch every goto above. */
/* Restore interrupts if we exited after disabling them. */
if (uOldEFlags != ~(RTCCUINTREG)0)
#endif
Log2(("X"));
return VBOXSTRICTRC_TODO(rc);
}
/**
* Enters the VT-x session.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCpu Pointer to the CPU info struct.
*/
{
if (!(cr4 & X86_CR4_VMXE))
{
AssertMsgFailed(("X86_CR4_VMXE should be set!\n"));
return VERR_VMX_X86_CR4_VMXE_CLEARED;
}
/* Activate the VMCS. */
if (RT_FAILURE(rc))
return rc;
return VINF_SUCCESS;
}
/**
* Leaves the VT-x session.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guests CPU context.
*/
{
#ifdef DEBUG
{
}
else
#endif
/*
* Save the guest debug state if necessary.
*/
{
/* Enable DRx move intercepts again. */
/* Resync the debug registers the next time. */
}
else
/*
* Clear VMCS, marking it inactive, clearing implementation-specific data and writing
* VMCS data back to memory.
*/
return VINF_SUCCESS;
}
/**
* Flush the TLB using EPT.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param enmFlush Type of flush.
*/
{
AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %x %RGv failed with %d\n", enmFlush, pVCpu->hm.s.vmx.HCPhysEPTP, rc));
#ifdef VBOX_WITH_STATISTICS
#endif
}
/**
* Flush the TLB using VPID.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU (can be NULL depending on @a
* enmFlush).
* @param enmFlush Type of flush.
* @param GCPtr Virtual address of the page to flush (can be 0 depending
* on @a enmFlush).
*/
{
if (enmFlush == VMX_FLUSH_VPID_ALL_CONTEXTS)
{
descriptor[0] = 0;
descriptor[1] = 0;
}
else
{
AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
}
("VMXR0InvVPID %x %x %RGv failed with %d\n", enmFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc));
#ifdef VBOX_WITH_STATISTICS
if (pVCpu)
#endif
}
/**
* Invalidates a guest page by guest virtual address. Only relevant for
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param GCVirt Guest virtual address of the page to invalidate.
*/
{
if (!fFlushPending)
{
/*
* We must invalidate the guest TLB entry in either case, we cannot ignore it even for the EPT case
* See @bugref{6043} and @bugref{6177}
*
* Set the VMCPU_FF_TLB_FLUSH force flag and flush before VMENTRY in hmR0VmxSetupTLB*() as this
* function maybe called in a loop with individual addresses.
*/
{
/* If we can flush just this page do it, otherwise flush as little as possible. */
else
}
}
return VINF_SUCCESS;
}
/**
* otherwise there is nothing really to invalidate.
*
* NOTE: Assumes the current instruction references this physical page though a virtual address!!
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param GCPhys Guest physical address of the page to invalidate.
*/
{
/*
* We cannot flush a page by guest-physical address. invvpid takes only a linear address
* while invept only flushes by EPT not individual addresses. We update the force flag here
* and flush before VMENTRY in hmR0VmxSetupTLB*(). This function might be called in a loop.
*/
return VINF_SUCCESS;
}
/**
* Report world switch error and dump some useful debug info.
*
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param rc Return code.
* @param pCtx Pointer to the current guest CPU context (not updated).
*/
{
switch (VBOXSTRICTRC_VAL(rc))
{
AssertFailed();
break;
{
int rc2;
if (rc2 == VINF_SUCCESS)
{
(uint32_t)instrError));
#ifdef VBOX_STRICT
ASMGetGDTR(&gdtr);
{
}
{
}
{
}
{
}
{
}
{
}
{
}
if (VMX_IS_64BIT_HOST_MODE())
{
}
# endif
#endif /* VBOX_STRICT */
}
break;
}
default:
/* impossible */
break;
}
}
#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
/**
* Prepares for and executes VMLAUNCH (64 bits guest mode).
*
* @returns VBox status code.
* @param pCtx Pointer to the guest CPU context.
* @param pCache Pointer to the VMCS cache.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
*/
DECLASM(int) VMXR0SwitcherStartVM64(RTHCUINT fResume, PCPUMCTX pCtx, PVMCSCACHE pCache, PVM pVM, PVMCPU pVCpu)
{
int rc;
pCpu = HMR0GetCurrentCpu();
#ifdef VBOX_WITH_CRASHDUMP_MAGIC
#endif
#ifdef DEBUG
#endif
aParam[3] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs >> 32); /* Param 2: VMCS physical address - Hi. */
aParam[5] = 0;
#ifdef VBOX_WITH_CRASHDUMP_MAGIC
#endif
#ifdef VBOX_WITH_CRASHDUMP_MAGIC
#endif
#ifdef DEBUG
AssertMsg(pCache->TestIn.HCPhysCpuPage== HCPhysCpuPage, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysCpuPage, HCPhysCpuPage));
AssertMsg(pCache->TestIn.HCPhysVmcs == pVCpu->hm.s.vmx.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
AssertMsg(pCache->TestIn.HCPhysVmcs == pCache->TestOut.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
AssertMsg(pCache->TestIn.pCache == pCache->TestOut.pCache, ("%RGv vs %RGv\n", pCache->TestIn.pCache,
("%RGv vs %RGv\n", pCache->TestIn.pCache, VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache)));
#endif
return rc;
}
#ifdef VBOX_STRICT
{
switch (idxField)
{
case VMX_VMCS_GUEST_RIP:
case VMX_VMCS_GUEST_RSP:
case VMX_VMCS_GUEST_RFLAGS:
case VMX_VMCS_GUEST_CR0:
case VMX_VMCS_GUEST_CR4:
case VMX_VMCS_GUEST_DR7:
case VMX_VMCS_GUEST_GDTR_BASE:
case VMX_VMCS_GUEST_IDTR_BASE:
case VMX_VMCS_GUEST_CS_BASE:
case VMX_VMCS_GUEST_DS_BASE:
case VMX_VMCS_GUEST_ES_BASE:
case VMX_VMCS_GUEST_FS_BASE:
case VMX_VMCS_GUEST_GS_BASE:
case VMX_VMCS_GUEST_SS_BASE:
case VMX_VMCS_GUEST_LDTR_BASE:
case VMX_VMCS_GUEST_TR_BASE:
case VMX_VMCS32_RO_IDT_INFO:
case VMX_VMCS_GUEST_CR3:
return true;
}
return false;
}
{
switch (idxField)
{
case VMX_VMCS_GUEST_LDTR_BASE:
case VMX_VMCS_GUEST_TR_BASE:
case VMX_VMCS_GUEST_GDTR_BASE:
case VMX_VMCS_GUEST_IDTR_BASE:
case VMX_VMCS_GUEST_CR0:
case VMX_VMCS_GUEST_CR4:
case VMX_VMCS_GUEST_CR3:
case VMX_VMCS_GUEST_DR7:
case VMX_VMCS_GUEST_RIP:
case VMX_VMCS_GUEST_RSP:
case VMX_VMCS_GUEST_CS_BASE:
case VMX_VMCS_GUEST_DS_BASE:
case VMX_VMCS_GUEST_ES_BASE:
case VMX_VMCS_GUEST_FS_BASE:
case VMX_VMCS_GUEST_GS_BASE:
case VMX_VMCS_GUEST_SS_BASE:
return true;
}
return false;
}
#endif /* VBOX_STRICT */
/**
* Executes the specified handler in 64-bit mode.
*
* @returns VBox status code.
* @param pVM Pointer to the VM.
* @param pVCpu Pointer to the VMCPU.
* @param pCtx Pointer to the guest CPU context.
* @param enmOp The operation to perform.
* @param cbParam Number of parameters.
* @param paParam Array of 32-bit parameters.
*/
VMMR0DECL(int) VMXR0Execute64BitsHandler(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, HM64ON32OP enmOp, uint32_t cbParam,
{
Assert(pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Write.aField));
Assert(pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Read.aField));
#ifdef VBOX_STRICT
#endif
/* Disable interrupts. */
#endif
pCpu = HMR0GetCurrentCpu();
/* Clear VMCS. Marking it inactive, clearing implementation-specific data and writing VMCS data back to memory. */
/* Leave VMX Root Mode. */
VMXDisable();
for (int i=(int)cbParam-1;i>=0;i--)
/* Call switcher. */
rc = pVM->hm.s.pfnHost32ToGuest64R0(pVM, RT_OFFSETOF(VM, aCpus[pVCpu->idCpu].cpum) - RT_OFFSETOF(VM, cpum));
/* Make sure the VMX instructions don't cause #UD faults. */
/* Enter VMX Root Mode */
if (RT_FAILURE(rc2))
{
return VERR_VMX_VMXON_FAILED;
}
return rc;
}
#endif /* HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) */
/**
* Executes VMWRITE.
*
* @returns VBox status code
* @param pVCpu Pointer to the VMCPU.
* @param idxField VMCS field index.
* @param u64Val 16, 32 or 64 bits value.
*/
{
int rc;
switch (idxField)
{
/* These fields consist of two parts, which are both writable in 32 bits mode. */
return rc;
case VMX_VMCS_GUEST_LDTR_BASE:
case VMX_VMCS_GUEST_TR_BASE:
case VMX_VMCS_GUEST_GDTR_BASE:
case VMX_VMCS_GUEST_IDTR_BASE:
case VMX_VMCS_GUEST_CR0:
case VMX_VMCS_GUEST_CR4:
case VMX_VMCS_GUEST_CR3:
case VMX_VMCS_GUEST_DR7:
case VMX_VMCS_GUEST_RIP:
case VMX_VMCS_GUEST_RSP:
case VMX_VMCS_GUEST_CS_BASE:
case VMX_VMCS_GUEST_DS_BASE:
case VMX_VMCS_GUEST_ES_BASE:
case VMX_VMCS_GUEST_FS_BASE:
case VMX_VMCS_GUEST_GS_BASE:
case VMX_VMCS_GUEST_SS_BASE:
/* Queue a 64 bits value as we can't set it in 32 bits host mode. */
if (u64Val >> 32ULL)
else
return rc;
default:
return VERR_INVALID_PARAMETER;
}
}
/**
* Cache VMCS writes for running 64 bits guests on 32 bits hosts.
*
* @param pVCpu Pointer to the VMCPU.
* @param idxField VMCS field index.
* @param u64Val 16, 32 or 64 bits value.
*/
{
/* Make sure there are no duplicates. */
{
{
return VINF_SUCCESS;
}
}
return VINF_SUCCESS;
}
#endif /* HC_ARCH_BITS == 32 && !VBOX_WITH_HYBRID_32BIT_KERNEL */