1
0
Fork 0
virtualbox/include/VBox/vmm/vm.h
Daniel Baumann df1bda4fe9
Adding upstream version 7.0.20-dfsg.
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
2025-06-22 09:56:04 +02:00

1530 lines
58 KiB
C

/** @file
* VM - The Virtual Machine, data.
*/
/*
* Copyright (C) 2006-2023 Oracle and/or its affiliates.
*
* This file is part of VirtualBox base platform packages, as
* available from https://www.virtualbox.org.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation, in version 3 of the
* License.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, see <https://www.gnu.org/licenses>.
*
* The contents of this file may alternatively be used under the terms
* of the Common Development and Distribution License Version 1.0
* (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
* in the VirtualBox distribution, in which case the provisions of the
* CDDL are applicable instead of those of the GPL.
*
* You may elect to license modified versions of this file under the
* terms and conditions of either the GPL or the CDDL or both.
*
* SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
*/
#ifndef VBOX_INCLUDED_vmm_vm_h
#define VBOX_INCLUDED_vmm_vm_h
#ifndef RT_WITHOUT_PRAGMA_ONCE
# pragma once
#endif
#ifndef VBOX_FOR_DTRACE_LIB
# ifndef USING_VMM_COMMON_DEFS
# error "Compile job does not include VMM_COMMON_DEFS from src/VBox/VMM/Config.kmk - make sure you really need to include this file!"
# endif
# include <iprt/param.h>
# include <VBox/param.h>
# include <VBox/types.h>
# include <VBox/vmm/cpum.h>
# include <VBox/vmm/stam.h>
# include <VBox/vmm/vmapi.h>
# include <VBox/vmm/vmm.h>
# include <VBox/sup.h>
#else
# pragma D depends_on library vbox-types.d
# pragma D depends_on library CPUMInternal.d
# define VMM_INCLUDED_SRC_include_CPUMInternal_h
#endif
/** @defgroup grp_vm The Virtual Machine
* @ingroup grp_vmm
* @{
*/
/**
* The state of a Virtual CPU.
*
* The basic state indicated here is whether the CPU has been started or not. In
* addition, there are sub-states when started for assisting scheduling (GVMM
* mostly).
*
* The transition out of the STOPPED state is done by a vmR3PowerOn.
* The transition back to the STOPPED state is done by vmR3PowerOff.
*
* (Alternatively we could let vmR3PowerOn start CPU 0 only and let the SPIP
* handling switch on the other CPUs. Then vmR3Reset would stop all but CPU 0.)
*/
typedef enum VMCPUSTATE
{
/** The customary invalid zero. */
VMCPUSTATE_INVALID = 0,
/** Virtual CPU has not yet been started. */
VMCPUSTATE_STOPPED,
/** CPU started. */
VMCPUSTATE_STARTED,
/** CPU started in HM context. */
VMCPUSTATE_STARTED_HM,
/** Executing guest code and can be poked (RC or STI bits of HM). */
VMCPUSTATE_STARTED_EXEC,
/** Executing guest code using NEM. */
VMCPUSTATE_STARTED_EXEC_NEM,
VMCPUSTATE_STARTED_EXEC_NEM_WAIT,
VMCPUSTATE_STARTED_EXEC_NEM_CANCELED,
/** Halted. */
VMCPUSTATE_STARTED_HALTED,
/** The end of valid virtual CPU states. */
VMCPUSTATE_END,
/** Ensure 32-bit type. */
VMCPUSTATE_32BIT_HACK = 0x7fffffff
} VMCPUSTATE;
/** Enables 64-bit FFs. */
#define VMCPU_WITH_64_BIT_FFS
/**
* The cross context virtual CPU structure.
*
* Run 'kmk run-struct-tests' (from src/VBox/VMM if you like) after updating!
*/
typedef struct VMCPU
{
/** @name Volatile per-cpu data.
* @{ */
/** Per CPU forced action.
* See the VMCPU_FF_* \#defines. Updated atomically. */
#ifdef VMCPU_WITH_64_BIT_FFS
uint64_t volatile fLocalForcedActions;
#else
uint32_t volatile fLocalForcedActions;
uint32_t fForLocalForcedActionsExpansion;
#endif
/** The CPU state. */
VMCPUSTATE volatile enmState;
/** Padding up to 64 bytes. */
uint8_t abAlignment0[64 - 12];
/** @} */
/** IEM part.
* @remarks This comes first as it allows the use of 8-bit immediates for the
* first 64 bytes of the structure, reducing code size a wee bit. */
#ifdef VMM_INCLUDED_SRC_include_IEMInternal_h /* For PDB hacking. */
union VMCPUUNIONIEMFULL
#else
union VMCPUUNIONIEMSTUB
#endif
{
#ifdef VMM_INCLUDED_SRC_include_IEMInternal_h
struct IEMCPU s;
#endif
uint8_t padding[32832]; /* multiple of 64 */
} iem;
/** @name Static per-cpu data.
* (Putting this after IEM, hoping that it's less frequently used than it.)
* @{ */
/** Ring-3 Host Context VM Pointer. */
PVMR3 pVMR3;
/** Ring-0 Host Context VM Pointer, currently used by VTG/dtrace. */
RTR0PTR pVCpuR0ForVtg;
/** Raw-mode Context VM Pointer. */
uint32_t pVMRC;
/** Padding for new raw-mode (long mode). */
uint32_t pVMRCPadding;
/** Pointer to the ring-3 UVMCPU structure. */
PUVMCPU pUVCpu;
/** The native thread handle. */
RTNATIVETHREAD hNativeThread;
/** The native R0 thread handle. (different from the R3 handle!) */
RTNATIVETHREAD hNativeThreadR0;
/** The IPRT thread handle (for VMMDevTesting). */
RTTHREAD hThread;
/** The CPU ID.
* This is the index into the VM::aCpu array. */
#ifdef IN_RING0
VMCPUID idCpuUnsafe;
#else
VMCPUID idCpu;
#endif
/** Align the structures below bit on a 64-byte boundary and make sure it starts
* at the same offset in both 64-bit and 32-bit builds.
*
* @remarks The alignments of the members that are larger than 48 bytes should be
* 64-byte for cache line reasons. structs containing small amounts of
* data could be lumped together at the end with a < 64 byte padding
* following it (to grow into and align the struct size).
*/
uint8_t abAlignment1[64 - 6 * (HC_ARCH_BITS == 32 ? 4 : 8) - 8 - 4];
/** @} */
/** HM part. */
union VMCPUUNIONHM
{
#ifdef VMM_INCLUDED_SRC_include_HMInternal_h
struct HMCPU s;
#endif
uint8_t padding[9984]; /* multiple of 64 */
} hm;
/** NEM part. */
union VMCPUUNIONNEM
{
#ifdef VMM_INCLUDED_SRC_include_NEMInternal_h
struct NEMCPU s;
#endif
uint8_t padding[4608]; /* multiple of 64 */
} nem;
/** TRPM part. */
union VMCPUUNIONTRPM
{
#ifdef VMM_INCLUDED_SRC_include_TRPMInternal_h
struct TRPMCPU s;
#endif
uint8_t padding[128]; /* multiple of 64 */
} trpm;
/** TM part. */
union VMCPUUNIONTM
{
#ifdef VMM_INCLUDED_SRC_include_TMInternal_h
struct TMCPU s;
#endif
uint8_t padding[5760]; /* multiple of 64 */
} tm;
/** VMM part. */
union VMCPUUNIONVMM
{
#ifdef VMM_INCLUDED_SRC_include_VMMInternal_h
struct VMMCPU s;
#endif
uint8_t padding[9536]; /* multiple of 64 */
} vmm;
/** PDM part. */
union VMCPUUNIONPDM
{
#ifdef VMM_INCLUDED_SRC_include_PDMInternal_h
struct PDMCPU s;
#endif
uint8_t padding[256]; /* multiple of 64 */
} pdm;
/** IOM part. */
union VMCPUUNIONIOM
{
#ifdef VMM_INCLUDED_SRC_include_IOMInternal_h
struct IOMCPU s;
#endif
uint8_t padding[512]; /* multiple of 64 */
} iom;
/** DBGF part.
* @todo Combine this with other tiny structures. */
union VMCPUUNIONDBGF
{
#ifdef VMM_INCLUDED_SRC_include_DBGFInternal_h
struct DBGFCPU s;
#endif
uint8_t padding[512]; /* multiple of 64 */
} dbgf;
/** GIM part. */
union VMCPUUNIONGIM
{
#ifdef VMM_INCLUDED_SRC_include_GIMInternal_h
struct GIMCPU s;
#endif
uint8_t padding[512]; /* multiple of 64 */
} gim;
/** APIC part. */
union VMCPUUNIONAPIC
{
#ifdef VMM_INCLUDED_SRC_include_APICInternal_h
struct APICCPU s;
#endif
uint8_t padding[3840]; /* multiple of 64 */
} apic;
/*
* Some less frequently used global members that doesn't need to take up
* precious space at the head of the structure.
*/
/** Trace groups enable flags. */
uint32_t fTraceGroups; /* 64 / 44 */
/** Number of collisions hashing the ring-0 EMT handle. */
uint8_t cEmtHashCollisions;
uint8_t abAdHoc[3];
/** Profiling samples for use by ad hoc profiling. */
STAMPROFILEADV aStatAdHoc[8]; /* size: 40*8 = 320 */
/** Align the following members on page boundary. */
uint8_t abAlignment2[696];
/** PGM part. */
union VMCPUUNIONPGM
{
#ifdef VMM_INCLUDED_SRC_include_PGMInternal_h
struct PGMCPU s;
#endif
uint8_t padding[4096 + 28672]; /* multiple of 4096 */
} pgm;
/** CPUM part. */
union VMCPUUNIONCPUM
{
#ifdef VMM_INCLUDED_SRC_include_CPUMInternal_h
struct CPUMCPU s;
#endif
#ifdef VMCPU_INCL_CPUM_GST_CTX
/** The guest CPUM context for direct use by execution engines.
* This is not for general consumption, but for HM, REM, IEM, and maybe a few
* others. The rest will use the function based CPUM API. */
CPUMCTX GstCtx;
#endif
uint8_t padding[102400]; /* multiple of 4096 */
} cpum;
/** EM part. */
union VMCPUUNIONEM
{
#ifdef VMM_INCLUDED_SRC_include_EMInternal_h
struct EMCPU s;
#endif
uint8_t padding[40960]; /* multiple of 4096 */
} em;
} VMCPU;
#ifndef VBOX_FOR_DTRACE_LIB
/* Make sure the structure size is aligned on a 16384 boundary for arm64 purposes. */
AssertCompileSizeAlignment(VMCPU, 16384);
/** @name Operations on VMCPU::enmState
* @{ */
/** Gets the VMCPU state. */
#define VMCPU_GET_STATE(pVCpu) ( (pVCpu)->enmState )
/** Sets the VMCPU state. */
#define VMCPU_SET_STATE(pVCpu, enmNewState) \
ASMAtomicWriteU32((uint32_t volatile *)&(pVCpu)->enmState, (enmNewState))
/** Cmpares and sets the VMCPU state. */
#define VMCPU_CMPXCHG_STATE(pVCpu, enmNewState, enmOldState) \
ASMAtomicCmpXchgU32((uint32_t volatile *)&(pVCpu)->enmState, (enmNewState), (enmOldState))
/** Checks the VMCPU state. */
#ifdef VBOX_STRICT
# define VMCPU_ASSERT_STATE(pVCpu, enmExpectedState) \
do { \
VMCPUSTATE enmState = VMCPU_GET_STATE(pVCpu); \
AssertMsg(enmState == (enmExpectedState), \
("enmState=%d enmExpectedState=%d idCpu=%u\n", \
enmState, enmExpectedState, (pVCpu)->idCpu)); \
} while (0)
# define VMCPU_ASSERT_STATE_2(pVCpu, enmExpectedState, a_enmExpectedState2) \
do { \
VMCPUSTATE enmState = VMCPU_GET_STATE(pVCpu); \
AssertMsg( enmState == (enmExpectedState) \
|| enmState == (a_enmExpectedState2), \
("enmState=%d enmExpectedState=%d enmExpectedState2=%d idCpu=%u\n", \
enmState, enmExpectedState, a_enmExpectedState2, (pVCpu)->idCpu)); \
} while (0)
#else
# define VMCPU_ASSERT_STATE(pVCpu, enmExpectedState) do { } while (0)
# define VMCPU_ASSERT_STATE_2(pVCpu, enmExpectedState, a_enmExpectedState2) do { } while (0)
#endif
/** Tests if the state means that the CPU is started. */
#define VMCPUSTATE_IS_STARTED(enmState) ( (enmState) > VMCPUSTATE_STOPPED )
/** Tests if the state means that the CPU is stopped. */
#define VMCPUSTATE_IS_STOPPED(enmState) ( (enmState) == VMCPUSTATE_STOPPED )
/** @} */
/** The name of the raw-mode context VMM Core module. */
#define VMMRC_MAIN_MODULE_NAME "VMMRC.rc"
/** The name of the ring-0 context VMM Core module. */
#define VMMR0_MAIN_MODULE_NAME "VMMR0.r0"
/** VM Forced Action Flags.
*
* Use the VM_FF_SET() and VM_FF_CLEAR() macros to change the force
* action mask of a VM.
*
* Available VM bits:
* 0, 1, 5, 6, 7, 13, 14, 15, 16, 17, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30
*
*
* Available VMCPU bits:
* 14, 15, 36 to 63
*
* @todo If we run low on VMCPU, we may consider merging the SELM bits
*
* @{
*/
/** The virtual sync clock has been stopped, go to TM until it has been
* restarted... */
#define VM_FF_TM_VIRTUAL_SYNC RT_BIT_32(VM_FF_TM_VIRTUAL_SYNC_BIT)
#define VM_FF_TM_VIRTUAL_SYNC_BIT 2
/** PDM Queues are pending. */
#define VM_FF_PDM_QUEUES RT_BIT_32(VM_FF_PDM_QUEUES_BIT)
/** The bit number for VM_FF_PDM_QUEUES. */
#define VM_FF_PDM_QUEUES_BIT 3
/** PDM DMA transfers are pending. */
#define VM_FF_PDM_DMA RT_BIT_32(VM_FF_PDM_DMA_BIT)
/** The bit number for VM_FF_PDM_DMA. */
#define VM_FF_PDM_DMA_BIT 4
/** This action forces the VM to call DBGF so DBGF can service debugger
* requests in the emulation thread.
* This action flag stays asserted till DBGF clears it.*/
#define VM_FF_DBGF RT_BIT_32(VM_FF_DBGF_BIT)
/** The bit number for VM_FF_DBGF. */
#define VM_FF_DBGF_BIT 8
/** This action forces the VM to service pending requests from other
* thread or requests which must be executed in another context. */
#define VM_FF_REQUEST RT_BIT_32(VM_FF_REQUEST_BIT)
#define VM_FF_REQUEST_BIT 9
/** Check for VM state changes and take appropriate action. */
#define VM_FF_CHECK_VM_STATE RT_BIT_32(VM_FF_CHECK_VM_STATE_BIT)
/** The bit number for VM_FF_CHECK_VM_STATE. */
#define VM_FF_CHECK_VM_STATE_BIT 10
/** Reset the VM. (postponed) */
#define VM_FF_RESET RT_BIT_32(VM_FF_RESET_BIT)
/** The bit number for VM_FF_RESET. */
#define VM_FF_RESET_BIT 11
/** EMT rendezvous in VMM. */
#define VM_FF_EMT_RENDEZVOUS RT_BIT_32(VM_FF_EMT_RENDEZVOUS_BIT)
/** The bit number for VM_FF_EMT_RENDEZVOUS. */
#define VM_FF_EMT_RENDEZVOUS_BIT 12
/** PGM needs to allocate handy pages. */
#define VM_FF_PGM_NEED_HANDY_PAGES RT_BIT_32(VM_FF_PGM_NEED_HANDY_PAGES_BIT)
#define VM_FF_PGM_NEED_HANDY_PAGES_BIT 18
/** PGM is out of memory.
* Abandon all loops and code paths which can be resumed and get up to the EM
* loops. */
#define VM_FF_PGM_NO_MEMORY RT_BIT_32(VM_FF_PGM_NO_MEMORY_BIT)
#define VM_FF_PGM_NO_MEMORY_BIT 19
/** PGM is about to perform a lightweight pool flush
* Guest SMP: all EMT threads should return to ring 3
*/
#define VM_FF_PGM_POOL_FLUSH_PENDING RT_BIT_32(VM_FF_PGM_POOL_FLUSH_PENDING_BIT)
#define VM_FF_PGM_POOL_FLUSH_PENDING_BIT 20
/** Suspend the VM - debug only. */
#define VM_FF_DEBUG_SUSPEND RT_BIT_32(VM_FF_DEBUG_SUSPEND_BIT)
#define VM_FF_DEBUG_SUSPEND_BIT 31
/** This action forces the VM to check any pending interrupts on the APIC. */
#define VMCPU_FF_INTERRUPT_APIC RT_BIT_64(VMCPU_FF_INTERRUPT_APIC_BIT)
#define VMCPU_FF_INTERRUPT_APIC_BIT 0
/** This action forces the VM to check any pending interrups on the PIC. */
#define VMCPU_FF_INTERRUPT_PIC RT_BIT_64(VMCPU_FF_INTERRUPT_PIC_BIT)
#define VMCPU_FF_INTERRUPT_PIC_BIT 1
/** This action forces the VM to schedule and run pending timer (TM).
* @remarks Don't move - PATM compatibility. */
#define VMCPU_FF_TIMER RT_BIT_64(VMCPU_FF_TIMER_BIT)
#define VMCPU_FF_TIMER_BIT 2
/** This action forces the VM to check any pending NMIs. */
#define VMCPU_FF_INTERRUPT_NMI RT_BIT_64(VMCPU_FF_INTERRUPT_NMI_BIT)
#define VMCPU_FF_INTERRUPT_NMI_BIT 3
/** This action forces the VM to check any pending SMIs. */
#define VMCPU_FF_INTERRUPT_SMI RT_BIT_64(VMCPU_FF_INTERRUPT_SMI_BIT)
#define VMCPU_FF_INTERRUPT_SMI_BIT 4
/** PDM critical section unlocking is pending, process promptly upon return to R3. */
#define VMCPU_FF_PDM_CRITSECT RT_BIT_64(VMCPU_FF_PDM_CRITSECT_BIT)
#define VMCPU_FF_PDM_CRITSECT_BIT 5
/** Special EM internal force flag that is used by EMUnhaltAndWakeUp() to force
* the virtual CPU out of the next (/current) halted state. It is not processed
* nor cleared by emR3ForcedActions (similar to VMCPU_FF_BLOCK_NMIS), instead it
* is cleared the next time EM leaves the HALTED state. */
#define VMCPU_FF_UNHALT RT_BIT_64(VMCPU_FF_UNHALT_BIT)
#define VMCPU_FF_UNHALT_BIT 6
/** Pending IEM action (mask). */
#define VMCPU_FF_IEM RT_BIT_64(VMCPU_FF_IEM_BIT)
/** Pending IEM action (bit number). */
#define VMCPU_FF_IEM_BIT 7
/** Pending APIC action (bit number). */
#define VMCPU_FF_UPDATE_APIC_BIT 8
/** This action forces the VM to update APIC's asynchronously arrived
* interrupts as pending interrupts. */
#define VMCPU_FF_UPDATE_APIC RT_BIT_64(VMCPU_FF_UPDATE_APIC_BIT)
/** This action forces the VM to service pending requests from other
* thread or requests which must be executed in another context. */
#define VMCPU_FF_REQUEST RT_BIT_64(VMCPU_FF_REQUEST_BIT)
#define VMCPU_FF_REQUEST_BIT 9
/** Pending DBGF event (alternative to passing VINF_EM_DBG_EVENT around). */
#define VMCPU_FF_DBGF RT_BIT_64(VMCPU_FF_DBGF_BIT)
/** The bit number for VMCPU_FF_DBGF. */
#define VMCPU_FF_DBGF_BIT 10
/** Hardware virtualized nested-guest interrupt pending. */
#define VMCPU_FF_INTERRUPT_NESTED_GUEST RT_BIT_64(VMCPU_FF_INTERRUPT_NESTED_GUEST_BIT)
#define VMCPU_FF_INTERRUPT_NESTED_GUEST_BIT 11
/** This action forces PGM to update changes to CR3 when the guest was in HM mode
* (when using nested paging). */
#define VMCPU_FF_HM_UPDATE_CR3 RT_BIT_64(VMCPU_FF_HM_UPDATE_CR3_BIT)
#define VMCPU_FF_HM_UPDATE_CR3_BIT 12
/* Bit 13 used to be VMCPU_FF_HM_UPDATE_PAE_PDPES. */
/** This action forces the VM to resync the page tables before going
* back to execute guest code. (GLOBAL FLUSH) */
#define VMCPU_FF_PGM_SYNC_CR3 RT_BIT_64(VMCPU_FF_PGM_SYNC_CR3_BIT)
#define VMCPU_FF_PGM_SYNC_CR3_BIT 16
/** Same as VM_FF_PGM_SYNC_CR3 except that global pages can be skipped.
* (NON-GLOBAL FLUSH) */
#define VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL RT_BIT_64(VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL_BIT)
#define VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL_BIT 17
/** Check for pending TLB shootdown actions (deprecated)
* Reserved for future HM re-use if necessary / safe.
* Consumer: HM */
#define VMCPU_FF_TLB_SHOOTDOWN_UNUSED RT_BIT_64(VMCPU_FF_TLB_SHOOTDOWN_UNUSED_BIT)
#define VMCPU_FF_TLB_SHOOTDOWN_UNUSED_BIT 18
/** Check for pending TLB flush action.
* Consumer: HM
* @todo rename to VMCPU_FF_HM_TLB_FLUSH */
#define VMCPU_FF_TLB_FLUSH RT_BIT_64(VMCPU_FF_TLB_FLUSH_BIT)
/** The bit number for VMCPU_FF_TLB_FLUSH. */
#define VMCPU_FF_TLB_FLUSH_BIT 19
/* 20 used to be VMCPU_FF_TRPM_SYNC_IDT (raw-mode only). */
/* 21 used to be VMCPU_FF_SELM_SYNC_TSS (raw-mode only). */
/* 22 used to be VMCPU_FF_SELM_SYNC_GDT (raw-mode only). */
/* 23 used to be VMCPU_FF_SELM_SYNC_LDT (raw-mode only). */
/* 24 used to be VMCPU_FF_INHIBIT_INTERRUPTS, which moved to CPUMCTX::eflags.uBoth in v7.0.4. */
/* 25 used to be VMCPU_FF_BLOCK_NMIS, which moved to CPUMCTX::eflags.uBoth in v7.0.4. */
/** Force return to Ring-3. */
#define VMCPU_FF_TO_R3 RT_BIT_64(VMCPU_FF_TO_R3_BIT)
#define VMCPU_FF_TO_R3_BIT 28
/** Force return to ring-3 to service pending I/O or MMIO write.
* This is a backup for mechanism VINF_IOM_R3_IOPORT_COMMIT_WRITE and
* VINF_IOM_R3_MMIO_COMMIT_WRITE, allowing VINF_EM_DBG_BREAKPOINT and similar
* status codes to be propagated at the same time without loss. */
#define VMCPU_FF_IOM RT_BIT_64(VMCPU_FF_IOM_BIT)
#define VMCPU_FF_IOM_BIT 29
/* 30 used to be VMCPU_FF_CPUM */
/** VMX-preemption timer expired. */
#define VMCPU_FF_VMX_PREEMPT_TIMER RT_BIT_64(VMCPU_FF_VMX_PREEMPT_TIMER_BIT)
#define VMCPU_FF_VMX_PREEMPT_TIMER_BIT 31
/** Pending MTF (Monitor Trap Flag) event. */
#define VMCPU_FF_VMX_MTF RT_BIT_64(VMCPU_FF_VMX_MTF_BIT)
#define VMCPU_FF_VMX_MTF_BIT 32
/** VMX APIC-write emulation pending.
* @todo possible candidate for internal EFLAGS, or maybe just a summary bit
* (see also VMCPU_FF_VMX_INT_WINDOW). */
#define VMCPU_FF_VMX_APIC_WRITE RT_BIT_64(VMCPU_FF_VMX_APIC_WRITE_BIT)
#define VMCPU_FF_VMX_APIC_WRITE_BIT 33
/** VMX interrupt-window event pending.
*
* "Pending" is misleading here, it would be better to say that the event need
* to be generated at the next opportunity and that this flag causes it to be
* polled for on every instruction boundrary and such.
*
* @todo Change the IEM side of this to not poll but to track down the places
* where it can be generated and set an internal EFLAGS bit that causes it
* to be checked out when finishing the current instruction. */
#define VMCPU_FF_VMX_INT_WINDOW RT_BIT_64(VMCPU_FF_VMX_INT_WINDOW_BIT)
#define VMCPU_FF_VMX_INT_WINDOW_BIT 34
/** VMX NMI-window event pending.
* Same "pending" comment and todo in VMCPU_FF_VMX_INT_WINDOW. */
#define VMCPU_FF_VMX_NMI_WINDOW RT_BIT_64(VMCPU_FF_VMX_NMI_WINDOW_BIT)
#define VMCPU_FF_VMX_NMI_WINDOW_BIT 35
/** Externally VM forced actions. Used to quit the idle/wait loop. */
#define VM_FF_EXTERNAL_SUSPENDED_MASK ( VM_FF_CHECK_VM_STATE | VM_FF_DBGF | VM_FF_REQUEST | VM_FF_EMT_RENDEZVOUS )
/** Externally VMCPU forced actions. Used to quit the idle/wait loop. */
#define VMCPU_FF_EXTERNAL_SUSPENDED_MASK ( VMCPU_FF_REQUEST | VMCPU_FF_DBGF )
/** Externally forced VM actions. Used to quit the idle/wait loop. */
#define VM_FF_EXTERNAL_HALTED_MASK ( VM_FF_CHECK_VM_STATE | VM_FF_DBGF | VM_FF_REQUEST \
| VM_FF_PDM_QUEUES | VM_FF_PDM_DMA | VM_FF_EMT_RENDEZVOUS )
/** Externally forced VMCPU actions. Used to quit the idle/wait loop. */
#define VMCPU_FF_EXTERNAL_HALTED_MASK ( VMCPU_FF_UPDATE_APIC | VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC \
| VMCPU_FF_REQUEST | VMCPU_FF_INTERRUPT_NMI | VMCPU_FF_INTERRUPT_SMI \
| VMCPU_FF_UNHALT | VMCPU_FF_TIMER | VMCPU_FF_DBGF \
| VMCPU_FF_INTERRUPT_NESTED_GUEST)
/** High priority VM pre-execution actions. */
#define VM_FF_HIGH_PRIORITY_PRE_MASK ( VM_FF_CHECK_VM_STATE | VM_FF_DBGF | VM_FF_TM_VIRTUAL_SYNC \
| VM_FF_DEBUG_SUSPEND | VM_FF_PGM_NEED_HANDY_PAGES | VM_FF_PGM_NO_MEMORY \
| VM_FF_EMT_RENDEZVOUS )
/** High priority VMCPU pre-execution actions. */
#define VMCPU_FF_HIGH_PRIORITY_PRE_MASK ( VMCPU_FF_TIMER | VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC \
| VMCPU_FF_UPDATE_APIC | VMCPU_FF_DBGF \
| VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL \
| VMCPU_FF_INTERRUPT_NESTED_GUEST | VMCPU_FF_VMX_MTF | VMCPU_FF_VMX_APIC_WRITE \
| VMCPU_FF_VMX_PREEMPT_TIMER | VMCPU_FF_VMX_NMI_WINDOW | VMCPU_FF_VMX_INT_WINDOW )
/** High priority VM pre raw-mode execution mask. */
#define VM_FF_HIGH_PRIORITY_PRE_RAW_MASK ( VM_FF_PGM_NEED_HANDY_PAGES | VM_FF_PGM_NO_MEMORY )
/** High priority VMCPU pre raw-mode execution mask. */
#define VMCPU_FF_HIGH_PRIORITY_PRE_RAW_MASK ( VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL )
/** High priority post-execution actions. */
#define VM_FF_HIGH_PRIORITY_POST_MASK ( VM_FF_PGM_NO_MEMORY )
/** High priority post-execution actions. */
#define VMCPU_FF_HIGH_PRIORITY_POST_MASK ( VMCPU_FF_PDM_CRITSECT | VMCPU_FF_HM_UPDATE_CR3 | VMCPU_FF_IEM | VMCPU_FF_IOM )
/** Normal priority VM post-execution actions. */
#define VM_FF_NORMAL_PRIORITY_POST_MASK ( VM_FF_CHECK_VM_STATE | VM_FF_DBGF | VM_FF_RESET \
| VM_FF_PGM_NO_MEMORY | VM_FF_EMT_RENDEZVOUS)
/** Normal priority VMCPU post-execution actions. */
#define VMCPU_FF_NORMAL_PRIORITY_POST_MASK ( VMCPU_FF_DBGF )
/** Normal priority VM actions. */
#define VM_FF_NORMAL_PRIORITY_MASK ( VM_FF_REQUEST | VM_FF_PDM_QUEUES | VM_FF_PDM_DMA | VM_FF_EMT_RENDEZVOUS)
/** Normal priority VMCPU actions. */
#define VMCPU_FF_NORMAL_PRIORITY_MASK ( VMCPU_FF_REQUEST )
/** Flags to clear before resuming guest execution. */
#define VMCPU_FF_RESUME_GUEST_MASK ( VMCPU_FF_TO_R3 )
/** VM flags that cause the REP[|NE|E] STRINS loops to yield immediately. */
#define VM_FF_HIGH_PRIORITY_POST_REPSTR_MASK ( VM_FF_TM_VIRTUAL_SYNC | VM_FF_PGM_NEED_HANDY_PAGES | VM_FF_PGM_NO_MEMORY \
| VM_FF_EMT_RENDEZVOUS | VM_FF_PGM_POOL_FLUSH_PENDING | VM_FF_RESET)
/** VM flags that cause the REP[|NE|E] STRINS loops to yield. */
#define VM_FF_YIELD_REPSTR_MASK ( VM_FF_HIGH_PRIORITY_POST_REPSTR_MASK \
| VM_FF_PDM_QUEUES | VM_FF_PDM_DMA | VM_FF_DBGF | VM_FF_DEBUG_SUSPEND )
/** VMCPU flags that cause the REP[|NE|E] STRINS loops to yield immediately. */
#ifdef IN_RING3
# define VMCPU_FF_HIGH_PRIORITY_POST_REPSTR_MASK ( VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL | VMCPU_FF_DBGF \
| VMCPU_FF_VMX_MTF )
#else
# define VMCPU_FF_HIGH_PRIORITY_POST_REPSTR_MASK ( VMCPU_FF_TO_R3 | VMCPU_FF_IEM | VMCPU_FF_IOM | VMCPU_FF_PGM_SYNC_CR3 \
| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL | VMCPU_FF_DBGF | VMCPU_FF_VMX_MTF )
#endif
/** VMCPU flags that cause the REP[|NE|E] STRINS loops to yield, interrupts
* enabled. */
#define VMCPU_FF_YIELD_REPSTR_MASK ( VMCPU_FF_HIGH_PRIORITY_POST_REPSTR_MASK \
| VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_UPDATE_APIC | VMCPU_FF_INTERRUPT_PIC \
| VMCPU_FF_INTERRUPT_NMI | VMCPU_FF_INTERRUPT_SMI | VMCPU_FF_PDM_CRITSECT \
| VMCPU_FF_TIMER | VMCPU_FF_REQUEST \
| VMCPU_FF_INTERRUPT_NESTED_GUEST )
/** VMCPU flags that cause the REP[|NE|E] STRINS loops to yield, interrupts
* disabled. */
#define VMCPU_FF_YIELD_REPSTR_NOINT_MASK ( VMCPU_FF_YIELD_REPSTR_MASK \
& ~( VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_UPDATE_APIC | VMCPU_FF_INTERRUPT_PIC \
| VMCPU_FF_INTERRUPT_NESTED_GUEST) )
/** VM Flags that cause the HM loops to go back to ring-3. */
#define VM_FF_HM_TO_R3_MASK ( VM_FF_TM_VIRTUAL_SYNC | VM_FF_PGM_NEED_HANDY_PAGES | VM_FF_PGM_NO_MEMORY \
| VM_FF_PDM_QUEUES | VM_FF_EMT_RENDEZVOUS)
/** VMCPU Flags that cause the HM loops to go back to ring-3. */
#define VMCPU_FF_HM_TO_R3_MASK ( VMCPU_FF_TO_R3 | VMCPU_FF_TIMER | VMCPU_FF_PDM_CRITSECT \
| VMCPU_FF_IEM | VMCPU_FF_IOM)
/** High priority ring-0 VM pre HM-mode execution mask. */
#define VM_FF_HP_R0_PRE_HM_MASK (VM_FF_HM_TO_R3_MASK | VM_FF_REQUEST | VM_FF_PGM_POOL_FLUSH_PENDING | VM_FF_PDM_DMA)
/** High priority ring-0 VMCPU pre HM-mode execution mask. */
#define VMCPU_FF_HP_R0_PRE_HM_MASK ( VMCPU_FF_HM_TO_R3_MASK | VMCPU_FF_PGM_SYNC_CR3 \
| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL | VMCPU_FF_REQUEST \
| VMCPU_FF_VMX_APIC_WRITE | VMCPU_FF_VMX_MTF | VMCPU_FF_VMX_PREEMPT_TIMER)
/** High priority ring-0 VM pre HM-mode execution mask, single stepping. */
#define VM_FF_HP_R0_PRE_HM_STEP_MASK (VM_FF_HP_R0_PRE_HM_MASK & ~( VM_FF_TM_VIRTUAL_SYNC | VM_FF_PDM_QUEUES \
| VM_FF_EMT_RENDEZVOUS | VM_FF_REQUEST \
| VM_FF_PDM_DMA) )
/** High priority ring-0 VMCPU pre HM-mode execution mask, single stepping. */
#define VMCPU_FF_HP_R0_PRE_HM_STEP_MASK (VMCPU_FF_HP_R0_PRE_HM_MASK & ~( VMCPU_FF_TO_R3 | VMCPU_FF_TIMER \
| VMCPU_FF_PDM_CRITSECT | VMCPU_FF_REQUEST) )
/** All the VMX nested-guest flags. */
#define VMCPU_FF_VMX_ALL_MASK ( VMCPU_FF_VMX_PREEMPT_TIMER | VMCPU_FF_VMX_MTF | VMCPU_FF_VMX_APIC_WRITE \
| VMCPU_FF_VMX_INT_WINDOW | VMCPU_FF_VMX_NMI_WINDOW )
/** All the forced VM flags. */
#define VM_FF_ALL_MASK (UINT32_MAX)
/** All the forced VMCPU flags. */
#define VMCPU_FF_ALL_MASK (UINT32_MAX)
/** All the forced VM flags except those related to raw-mode and hardware
* assisted execution. */
#define VM_FF_ALL_REM_MASK (~(VM_FF_HIGH_PRIORITY_PRE_RAW_MASK) | VM_FF_PGM_NEED_HANDY_PAGES | VM_FF_PGM_NO_MEMORY)
/** All the forced VMCPU flags except those related to raw-mode and hardware
* assisted execution. */
#define VMCPU_FF_ALL_REM_MASK (~(VMCPU_FF_HIGH_PRIORITY_PRE_RAW_MASK | VMCPU_FF_PDM_CRITSECT | VMCPU_FF_TLB_FLUSH))
/** @} */
/** @def VM_FF_SET
* Sets a single force action flag.
*
* @param pVM The cross context VM structure.
* @param fFlag The flag to set.
*/
#define VM_FF_SET(pVM, fFlag) do { \
AssertCompile(RT_IS_POWER_OF_TWO(fFlag)); \
AssertCompile((fFlag) == RT_BIT_32(fFlag##_BIT)); \
ASMAtomicOrU32(&(pVM)->fGlobalForcedActions, (fFlag)); \
} while (0)
/** @def VMCPU_FF_SET
* Sets a single force action flag for the given VCPU.
*
* @param pVCpu The cross context virtual CPU structure.
* @param fFlag The flag to set.
* @sa VMCPU_FF_SET_MASK
*/
#ifdef VMCPU_WITH_64_BIT_FFS
# define VMCPU_FF_SET(pVCpu, fFlag) do { \
AssertCompile(RT_IS_POWER_OF_TWO(fFlag)); \
AssertCompile((fFlag) == RT_BIT_64(fFlag##_BIT)); \
ASMAtomicBitSet(&(pVCpu)->fLocalForcedActions, fFlag##_BIT); \
} while (0)
#else
# define VMCPU_FF_SET(pVCpu, fFlag) do { \
AssertCompile(RT_IS_POWER_OF_TWO(fFlag)); \
AssertCompile((fFlag) == RT_BIT_32(fFlag##_BIT)); \
ASMAtomicOrU32(&(pVCpu)->fLocalForcedActions, (fFlag)); \
} while (0)
#endif
/** @def VMCPU_FF_SET_MASK
* Sets a two or more force action flag for the given VCPU.
*
* @param pVCpu The cross context virtual CPU structure.
* @param fFlags The flags to set.
* @sa VMCPU_FF_SET
*/
#ifdef VMCPU_WITH_64_BIT_FFS
# if ARCH_BITS > 32
# define VMCPU_FF_SET_MASK(pVCpu, fFlags) \
do { ASMAtomicOrU64(&pVCpu->fLocalForcedActions, (fFlags)); } while (0)
# else
# define VMCPU_FF_SET_MASK(pVCpu, fFlags) do { \
if (!((fFlags) >> 32)) ASMAtomicOrU32((uint32_t volatile *)&pVCpu->fLocalForcedActions, (uint32_t)(fFlags)); \
else ASMAtomicOrU64(&pVCpu->fLocalForcedActions, (fFlags)); \
} while (0)
# endif
#else
# define VMCPU_FF_SET_MASK(pVCpu, fFlags) \
do { ASMAtomicOrU32(&pVCpu->fLocalForcedActions, (fFlags)); } while (0)
#endif
/** @def VM_FF_CLEAR
* Clears a single force action flag.
*
* @param pVM The cross context VM structure.
* @param fFlag The flag to clear.
*/
#define VM_FF_CLEAR(pVM, fFlag) do { \
AssertCompile(RT_IS_POWER_OF_TWO(fFlag)); \
AssertCompile((fFlag) == RT_BIT_32(fFlag##_BIT)); \
ASMAtomicAndU32(&(pVM)->fGlobalForcedActions, ~(fFlag)); \
} while (0)
/** @def VMCPU_FF_CLEAR
* Clears a single force action flag for the given VCPU.
*
* @param pVCpu The cross context virtual CPU structure.
* @param fFlag The flag to clear.
*/
#ifdef VMCPU_WITH_64_BIT_FFS
# define VMCPU_FF_CLEAR(pVCpu, fFlag) do { \
AssertCompile(RT_IS_POWER_OF_TWO(fFlag)); \
AssertCompile((fFlag) == RT_BIT_64(fFlag##_BIT)); \
ASMAtomicBitClear(&(pVCpu)->fLocalForcedActions, fFlag##_BIT); \
} while (0)
#else
# define VMCPU_FF_CLEAR(pVCpu, fFlag) do { \
AssertCompile(RT_IS_POWER_OF_TWO(fFlag)); \
AssertCompile((fFlag) == RT_BIT_32(fFlag##_BIT)); \
ASMAtomicAndU32(&(pVCpu)->fLocalForcedActions, ~(fFlag)); \
} while (0)
#endif
/** @def VMCPU_FF_CLEAR_MASK
* Clears two or more force action flags for the given VCPU.
*
* @param pVCpu The cross context virtual CPU structure.
* @param fFlags The flags to clear.
*/
#ifdef VMCPU_WITH_64_BIT_FFS
# if ARCH_BITS > 32
# define VMCPU_FF_CLEAR_MASK(pVCpu, fFlags) \
do { ASMAtomicAndU64(&(pVCpu)->fLocalForcedActions, ~(fFlags)); } while (0)
# else
# define VMCPU_FF_CLEAR_MASK(pVCpu, fFlags) do { \
if (!((fFlags) >> 32)) ASMAtomicAndU32((uint32_t volatile *)&(pVCpu)->fLocalForcedActions, ~(uint32_t)(fFlags)); \
else ASMAtomicAndU64(&(pVCpu)->fLocalForcedActions, ~(fFlags)); \
} while (0)
# endif
#else
# define VMCPU_FF_CLEAR_MASK(pVCpu, fFlags) \
do { ASMAtomicAndU32(&(pVCpu)->fLocalForcedActions, ~(fFlags)); } while (0)
#endif
/** @def VM_FF_IS_SET
* Checks if single a force action flag is set.
*
* @param pVM The cross context VM structure.
* @param fFlag The flag to check.
* @sa VM_FF_IS_ANY_SET
*/
#if !defined(VBOX_STRICT) || !defined(RT_COMPILER_SUPPORTS_LAMBDA)
# define VM_FF_IS_SET(pVM, fFlag) RT_BOOL((pVM)->fGlobalForcedActions & (fFlag))
#else
# define VM_FF_IS_SET(pVM, fFlag) \
([](PVM a_pVM) -> bool \
{ \
AssertCompile(RT_IS_POWER_OF_TWO(fFlag)); \
AssertCompile((fFlag) == RT_BIT_32(fFlag##_BIT)); \
return RT_BOOL(a_pVM->fGlobalForcedActions & (fFlag)); \
}(pVM))
#endif
/** @def VMCPU_FF_IS_SET
* Checks if a single force action flag is set for the given VCPU.
*
* @param pVCpu The cross context virtual CPU structure.
* @param fFlag The flag to check.
* @sa VMCPU_FF_IS_ANY_SET
*/
#if !defined(VBOX_STRICT) || !defined(RT_COMPILER_SUPPORTS_LAMBDA)
# define VMCPU_FF_IS_SET(pVCpu, fFlag) RT_BOOL((pVCpu)->fLocalForcedActions & (fFlag))
#else
# define VMCPU_FF_IS_SET(pVCpu, fFlag) \
([](PCVMCPU a_pVCpu) -> bool \
{ \
AssertCompile(RT_IS_POWER_OF_TWO(fFlag)); \
AssertCompile((fFlag) == RT_BIT_64(fFlag##_BIT)); \
return RT_BOOL(a_pVCpu->fLocalForcedActions & (fFlag)); \
}(pVCpu))
#endif
/** @def VM_FF_IS_ANY_SET
* Checks if one or more force action in the specified set is pending.
*
* @param pVM The cross context VM structure.
* @param fFlags The flags to check for.
* @sa VM_FF_IS_SET
*/
#define VM_FF_IS_ANY_SET(pVM, fFlags) RT_BOOL((pVM)->fGlobalForcedActions & (fFlags))
/** @def VMCPU_FF_IS_ANY_SET
* Checks if two or more force action flags in the specified set is set for the given VCPU.
*
* @param pVCpu The cross context virtual CPU structure.
* @param fFlags The flags to check for.
* @sa VMCPU_FF_IS_SET
*/
#define VMCPU_FF_IS_ANY_SET(pVCpu, fFlags) RT_BOOL((pVCpu)->fLocalForcedActions & (fFlags))
/** @def VM_FF_TEST_AND_CLEAR
* Checks if one (!) force action in the specified set is pending and clears it atomically
*
* @returns true if the bit was set.
* @returns false if the bit was clear.
* @param pVM The cross context VM structure.
* @param fFlag Flag constant to check and clear (_BIT is appended).
*/
#define VM_FF_TEST_AND_CLEAR(pVM, fFlag) (ASMAtomicBitTestAndClear(&(pVM)->fGlobalForcedActions, fFlag##_BIT))
/** @def VMCPU_FF_TEST_AND_CLEAR
* Checks if one (!) force action in the specified set is pending and clears it atomically
*
* @returns true if the bit was set.
* @returns false if the bit was clear.
* @param pVCpu The cross context virtual CPU structure.
* @param fFlag Flag constant to check and clear (_BIT is appended).
*/
#define VMCPU_FF_TEST_AND_CLEAR(pVCpu, fFlag) (ASMAtomicBitTestAndClear(&(pVCpu)->fLocalForcedActions, fFlag##_BIT))
/** @def VM_FF_IS_PENDING_EXCEPT
* Checks if one or more force action in the specified set is pending while one
* or more other ones are not.
*
* @param pVM The cross context VM structure.
* @param fFlags The flags to check for.
* @param fExcpt The flags that should not be set.
*/
#define VM_FF_IS_PENDING_EXCEPT(pVM, fFlags, fExcpt) \
( ((pVM)->fGlobalForcedActions & (fFlags)) && !((pVM)->fGlobalForcedActions & (fExcpt)) )
/** @def VM_IS_EMT
* Checks if the current thread is the emulation thread (EMT).
*
* @remark The ring-0 variation will need attention if we expand the ring-0
* code to let threads other than EMT mess around with the VM.
*/
#ifdef IN_RC
# define VM_IS_EMT(pVM) true
#else
# define VM_IS_EMT(pVM) (VMMGetCpu(pVM) != NULL)
#endif
/** @def VMCPU_IS_EMT
* Checks if the current thread is the emulation thread (EMT) for the specified
* virtual CPU.
*/
#ifdef IN_RC
# define VMCPU_IS_EMT(pVCpu) true
#else
# define VMCPU_IS_EMT(pVCpu) ((pVCpu) && ((pVCpu) == VMMGetCpu((pVCpu)->CTX_SUFF(pVM))))
#endif
/** @def VM_ASSERT_EMT
* Asserts that the current thread IS the emulation thread (EMT).
*/
#ifdef IN_RC
# define VM_ASSERT_EMT(pVM) Assert(VM_IS_EMT(pVM))
#elif defined(IN_RING0)
# define VM_ASSERT_EMT(pVM) Assert(VM_IS_EMT(pVM))
#else
# define VM_ASSERT_EMT(pVM) \
AssertMsg(VM_IS_EMT(pVM), \
("Not emulation thread! Thread=%RTnthrd ThreadEMT=%RTnthrd\n", RTThreadNativeSelf(), VMR3GetVMCPUNativeThread(pVM)))
#endif
/** @def VMCPU_ASSERT_EMT
* Asserts that the current thread IS the emulation thread (EMT) of the
* specified virtual CPU.
*/
#ifdef IN_RC
# define VMCPU_ASSERT_EMT(pVCpu) Assert(VMCPU_IS_EMT(pVCpu))
#elif defined(IN_RING0)
# define VMCPU_ASSERT_EMT(pVCpu) AssertMsg(VMCPU_IS_EMT(pVCpu), \
("Not emulation thread! Thread=%RTnthrd ThreadEMT=%RTnthrd idCpu=%u\n", \
RTThreadNativeSelf(), (pVCpu) ? (pVCpu)->hNativeThreadR0 : 0, \
(pVCpu) ? (pVCpu)->idCpu : 0))
#else
# define VMCPU_ASSERT_EMT(pVCpu) AssertMsg(VMCPU_IS_EMT(pVCpu), \
("Not emulation thread! Thread=%RTnthrd ThreadEMT=%RTnthrd idCpu=%#x\n", \
RTThreadNativeSelf(), (pVCpu)->hNativeThread, (pVCpu)->idCpu))
#endif
/** @def VM_ASSERT_EMT_RETURN
* Asserts that the current thread IS the emulation thread (EMT) and returns if it isn't.
*/
#ifdef IN_RC
# define VM_ASSERT_EMT_RETURN(pVM, rc) AssertReturn(VM_IS_EMT(pVM), (rc))
#elif defined(IN_RING0)
# define VM_ASSERT_EMT_RETURN(pVM, rc) AssertReturn(VM_IS_EMT(pVM), (rc))
#else
# define VM_ASSERT_EMT_RETURN(pVM, rc) \
AssertMsgReturn(VM_IS_EMT(pVM), \
("Not emulation thread! Thread=%RTnthrd ThreadEMT=%RTnthrd\n", RTThreadNativeSelf(), VMR3GetVMCPUNativeThread(pVM)), \
(rc))
#endif
/** @def VMCPU_ASSERT_EMT_RETURN
* Asserts that the current thread IS the emulation thread (EMT) and returns if it isn't.
*/
#ifdef IN_RC
# define VMCPU_ASSERT_EMT_RETURN(pVCpu, rc) AssertReturn(VMCPU_IS_EMT(pVCpu), (rc))
#elif defined(IN_RING0)
# define VMCPU_ASSERT_EMT_RETURN(pVCpu, rc) AssertReturn(VMCPU_IS_EMT(pVCpu), (rc))
#else
# define VMCPU_ASSERT_EMT_RETURN(pVCpu, rc) \
AssertMsgReturn(VMCPU_IS_EMT(pVCpu), \
("Not emulation thread! Thread=%RTnthrd ThreadEMT=%RTnthrd idCpu=%#x\n", \
RTThreadNativeSelf(), (pVCpu)->hNativeThread, (pVCpu)->idCpu), \
(rc))
#endif
/** @def VMCPU_ASSERT_EMT_OR_GURU
* Asserts that the current thread IS the emulation thread (EMT) of the
* specified virtual CPU.
*/
#if defined(IN_RC) || defined(IN_RING0)
# define VMCPU_ASSERT_EMT_OR_GURU(pVCpu) Assert( VMCPU_IS_EMT(pVCpu) \
|| pVCpu->CTX_SUFF(pVM)->enmVMState == VMSTATE_GURU_MEDITATION \
|| pVCpu->CTX_SUFF(pVM)->enmVMState == VMSTATE_GURU_MEDITATION_LS )
#else
# define VMCPU_ASSERT_EMT_OR_GURU(pVCpu) \
AssertMsg( VMCPU_IS_EMT(pVCpu) \
|| pVCpu->CTX_SUFF(pVM)->enmVMState == VMSTATE_GURU_MEDITATION \
|| pVCpu->CTX_SUFF(pVM)->enmVMState == VMSTATE_GURU_MEDITATION_LS, \
("Not emulation thread! Thread=%RTnthrd ThreadEMT=%RTnthrd idCpu=%#x\n", \
RTThreadNativeSelf(), (pVCpu)->hNativeThread, (pVCpu)->idCpu))
#endif
/** @def VMCPU_ASSERT_EMT_OR_NOT_RUNNING
* Asserts that the current thread IS the emulation thread (EMT) of the
* specified virtual CPU or the VM is not running.
*/
#if defined(IN_RC) || defined(IN_RING0)
# define VMCPU_ASSERT_EMT_OR_NOT_RUNNING(pVCpu) \
Assert( VMCPU_IS_EMT(pVCpu) \
|| !VM_IS_RUNNING_FOR_ASSERTIONS_ONLY((pVCpu)->CTX_SUFF(pVM)) )
#else
# define VMCPU_ASSERT_EMT_OR_NOT_RUNNING(pVCpu) \
AssertMsg( VMCPU_IS_EMT(pVCpu) \
|| !VM_IS_RUNNING_FOR_ASSERTIONS_ONLY((pVCpu)->CTX_SUFF(pVM)), \
("Not emulation thread! Thread=%RTnthrd ThreadEMT=%RTnthrd idCpu=%#x\n", \
RTThreadNativeSelf(), (pVCpu)->hNativeThread, (pVCpu)->idCpu))
#endif
/** @def VMSTATE_IS_RUNNING
* Checks if the given state indicates a running VM.
*/
#define VMSTATE_IS_RUNNING(a_enmVMState) \
( (a_enmVMState) == VMSTATE_RUNNING \
|| (a_enmVMState) == VMSTATE_RUNNING_LS )
/** @def VM_IS_RUNNING_FOR_ASSERTIONS_ONLY
* Checks if the VM is running.
* @note This is only for pure debug assertions. No AssertReturn or similar!
* @sa VMSTATE_IS_RUNNING
*/
#define VM_IS_RUNNING_FOR_ASSERTIONS_ONLY(pVM) \
( (pVM)->enmVMState == VMSTATE_RUNNING \
|| (pVM)->enmVMState == VMSTATE_RUNNING_LS )
/** @def VMSTATE_IS_POWERED_ON
* Checks if the given state indicates the VM is powered on.
*
* @note Excludes all error states, so a powered on VM that hit a fatal error,
* guru meditation, state load failure or similar will not be considered
* powered on by this test.
*/
#define VMSTATE_IS_POWERED_ON(a_enmVMState) \
( (a_enmVMState) >= VMSTATE_RESUMING && (a_enmVMState) < VMSTATE_POWERING_OFF )
/** @def VM_ASSERT_IS_NOT_RUNNING
* Asserts that the VM is not running.
*/
#if defined(IN_RC) || defined(IN_RING0)
#define VM_ASSERT_IS_NOT_RUNNING(pVM) Assert(!VM_IS_RUNNING_FOR_ASSERTIONS_ONLY(pVM))
#else
#define VM_ASSERT_IS_NOT_RUNNING(pVM) AssertMsg(!VM_IS_RUNNING_FOR_ASSERTIONS_ONLY(pVM), \
("VM is running. enmVMState=%d\n", (pVM)->enmVMState))
#endif
/** @def VM_ASSERT_EMT0
* Asserts that the current thread IS emulation thread \#0 (EMT0).
*/
#ifdef IN_RING3
# define VM_ASSERT_EMT0(a_pVM) VMCPU_ASSERT_EMT((a_pVM)->apCpusR3[0])
#else
# define VM_ASSERT_EMT0(a_pVM) VMCPU_ASSERT_EMT(&(a_pVM)->aCpus[0])
#endif
/** @def VM_ASSERT_EMT0_RETURN
* Asserts that the current thread IS emulation thread \#0 (EMT0) and returns if
* it isn't.
*/
#ifdef IN_RING3
# define VM_ASSERT_EMT0_RETURN(pVM, rc) VMCPU_ASSERT_EMT_RETURN((pVM)->apCpusR3[0], (rc))
#else
# define VM_ASSERT_EMT0_RETURN(pVM, rc) VMCPU_ASSERT_EMT_RETURN(&(pVM)->aCpus[0], (rc))
#endif
/**
* Asserts that the current thread is NOT the emulation thread.
*/
#define VM_ASSERT_OTHER_THREAD(pVM) \
AssertMsg(!VM_IS_EMT(pVM), ("Not other thread!!\n"))
/** @def VM_ASSERT_STATE
* Asserts a certain VM state.
*/
#define VM_ASSERT_STATE(pVM, _enmState) \
AssertMsg((pVM)->enmVMState == (_enmState), \
("state %s, expected %s\n", VMGetStateName((pVM)->enmVMState), VMGetStateName(_enmState)))
/** @def VM_ASSERT_STATE_RETURN
* Asserts a certain VM state and returns if it doesn't match.
*/
#define VM_ASSERT_STATE_RETURN(pVM, _enmState, rc) \
AssertMsgReturn((pVM)->enmVMState == (_enmState), \
("state %s, expected %s\n", VMGetStateName((pVM)->enmVMState), VMGetStateName(_enmState)), \
(rc))
/** @def VM_IS_VALID_EXT
* Asserts a the VM handle is valid for external access, i.e. not being destroy
* or terminated. */
#define VM_IS_VALID_EXT(pVM) \
( RT_VALID_ALIGNED_PTR(pVM, PAGE_SIZE) \
&& ( (unsigned)(pVM)->enmVMState < (unsigned)VMSTATE_DESTROYING \
|| ( (unsigned)(pVM)->enmVMState == (unsigned)VMSTATE_DESTROYING \
&& VM_IS_EMT(pVM))) )
/** @def VM_ASSERT_VALID_EXT_RETURN
* Asserts a the VM handle is valid for external access, i.e. not being
* destroy or terminated.
*/
#define VM_ASSERT_VALID_EXT_RETURN(pVM, rc) \
AssertMsgReturn(VM_IS_VALID_EXT(pVM), \
("pVM=%p state %s\n", (pVM), RT_VALID_ALIGNED_PTR(pVM, PAGE_SIZE) \
? VMGetStateName(pVM->enmVMState) : ""), \
(rc))
/** @def VMCPU_ASSERT_VALID_EXT_RETURN
* Asserts a the VMCPU handle is valid for external access, i.e. not being
* destroy or terminated.
*/
#define VMCPU_ASSERT_VALID_EXT_RETURN(pVCpu, rc) \
AssertMsgReturn( RT_VALID_ALIGNED_PTR(pVCpu, 64) \
&& RT_VALID_ALIGNED_PTR((pVCpu)->CTX_SUFF(pVM), PAGE_SIZE) \
&& (unsigned)(pVCpu)->CTX_SUFF(pVM)->enmVMState < (unsigned)VMSTATE_DESTROYING, \
("pVCpu=%p pVM=%p state %s\n", (pVCpu), RT_VALID_ALIGNED_PTR(pVCpu, 64) ? (pVCpu)->CTX_SUFF(pVM) : NULL, \
RT_VALID_ALIGNED_PTR(pVCpu, 64) && RT_VALID_ALIGNED_PTR((pVCpu)->CTX_SUFF(pVM), PAGE_SIZE) \
? VMGetStateName((pVCpu)->pVMR3->enmVMState) : ""), \
(rc))
#endif /* !VBOX_FOR_DTRACE_LIB */
/**
* Helper that HM and NEM uses for safely modifying VM::bMainExecutionEngine.
*
* ONLY HM and NEM MAY USE THIS!
*
* @param a_pVM The cross context VM structure.
* @param a_bValue The new value.
* @internal
*/
#define VM_SET_MAIN_EXECUTION_ENGINE(a_pVM, a_bValue) \
do { \
*const_cast<uint8_t *>(&(a_pVM)->bMainExecutionEngine) = (a_bValue); \
ASMCompilerBarrier(); /* just to be on the safe side */ \
} while (0)
/**
* Checks whether iem-executes-all-mode is used.
*
* @retval true if IEM is used.
* @retval false if not.
*
* @param a_pVM The cross context VM structure.
* @sa VM_IS_HM_OR_NEM_ENABLED, VM_IS_HM_ENABLED, VM_IS_NEM_ENABLED.
* @internal
*/
#define VM_IS_EXEC_ENGINE_IEM(a_pVM) ((a_pVM)->bMainExecutionEngine == VM_EXEC_ENGINE_IEM)
/**
* Checks whether HM (VT-x/AMD-V) or NEM is being used by this VM.
*
* @retval true if either is used.
* @retval false if software virtualization (raw-mode) is used.
*
* @param a_pVM The cross context VM structure.
* @sa VM_IS_EXEC_ENGINE_IEM, VM_IS_HM_ENABLED, VM_IS_NEM_ENABLED.
* @internal
*/
#define VM_IS_HM_OR_NEM_ENABLED(a_pVM) ((a_pVM)->bMainExecutionEngine != VM_EXEC_ENGINE_IEM)
/**
* Checks whether HM is being used by this VM.
*
* @retval true if HM (VT-x/AMD-v) is used.
* @retval false if not.
*
* @param a_pVM The cross context VM structure.
* @sa VM_IS_NEM_ENABLED, VM_IS_EXEC_ENGINE_IEM, VM_IS_HM_OR_NEM_ENABLED.
* @internal
*/
#define VM_IS_HM_ENABLED(a_pVM) ((a_pVM)->bMainExecutionEngine == VM_EXEC_ENGINE_HW_VIRT)
/**
* Checks whether NEM is being used by this VM.
*
* @retval true if a native hypervisor API is used.
* @retval false if not.
*
* @param a_pVM The cross context VM structure.
* @sa VM_IS_HM_ENABLED, VM_IS_EXEC_ENGINE_IEM, VM_IS_HM_OR_NEM_ENABLED.
* @internal
*/
#define VM_IS_NEM_ENABLED(a_pVM) ((a_pVM)->bMainExecutionEngine == VM_EXEC_ENGINE_NATIVE_API)
/**
* The cross context VM structure.
*
* It contains all the VM data which have to be available in all contexts.
* Even if it contains all the data the idea is to use APIs not to modify all
* the members all around the place. Therefore we make use of unions to hide
* everything which isn't local to the current source module. This means we'll
* have to pay a little bit of attention when adding new members to structures
* in the unions and make sure to keep the padding sizes up to date.
*
* Run 'kmk run-struct-tests' (from src/VBox/VMM if you like) after updating!
*/
typedef struct VM
{
/** The state of the VM.
* This field is read only to everyone except the VM and EM. */
VMSTATE volatile enmVMState;
/** Forced action flags.
* See the VM_FF_* \#defines. Updated atomically.
*/
volatile uint32_t fGlobalForcedActions;
/** Pointer to the array of page descriptors for the VM structure allocation. */
R3PTRTYPE(PSUPPAGE) paVMPagesR3;
/** Session handle. For use when calling SUPR0 APIs. */
#ifdef IN_RING0
PSUPDRVSESSION pSessionUnsafe;
#else
PSUPDRVSESSION pSession;
#endif
/** Pointer to the ring-3 VM structure. */
PUVM pUVM;
/** Ring-3 Host Context VM Pointer. */
#ifdef IN_RING0
R3PTRTYPE(struct VM *) pVMR3Unsafe;
#else
R3PTRTYPE(struct VM *) pVMR3;
#endif
/** Ring-0 Host Context VM pointer for making ring-0 calls. */
R0PTRTYPE(struct VM *) pVMR0ForCall;
/** Raw-mode Context VM Pointer. */
uint32_t pVMRC;
/** Padding for new raw-mode (long mode). */
uint32_t pVMRCPadding;
/** The GVM VM handle. Only the GVM should modify this field. */
#ifdef IN_RING0
uint32_t hSelfUnsafe;
#else
uint32_t hSelf;
#endif
/** Number of virtual CPUs. */
#ifdef IN_RING0
uint32_t cCpusUnsafe;
#else
uint32_t cCpus;
#endif
/** CPU excution cap (1-100) */
uint32_t uCpuExecutionCap;
/** Size of the VM structure. */
uint32_t cbSelf;
/** Size of the VMCPU structure. */
uint32_t cbVCpu;
/** Structure version number (TBD). */
uint32_t uStructVersion;
/** @name Various items that are frequently accessed.
* @{ */
/** The main execution engine, VM_EXEC_ENGINE_XXX.
* This is set early during vmR3InitRing3 by HM or NEM. */
uint8_t const bMainExecutionEngine;
/** Hardware VM support is available and enabled.
* Determined very early during init.
* This is placed here for performance reasons.
* @todo obsoleted by bMainExecutionEngine, eliminate. */
bool fHMEnabled;
/** @} */
/** Alignment padding. */
uint8_t uPadding1[6];
/** @name Debugging
* @{ */
/** Ring-3 Host Context VM Pointer. */
R3PTRTYPE(RTTRACEBUF) hTraceBufR3;
/** Ring-0 Host Context VM Pointer. */
R0PTRTYPE(RTTRACEBUF) hTraceBufR0;
/** @} */
/** Max EMT hash lookup collisions (in GVMM). */
uint8_t cMaxEmtHashCollisions;
/** Padding - the unions must be aligned on a 64 bytes boundary. */
uint8_t abAlignment3[HC_ARCH_BITS == 64 ? 23 : 51];
/** CPUM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_CPUMInternal_h
struct CPUM s;
#endif
#ifdef VBOX_INCLUDED_vmm_cpum_h
/** Read only info exposed about the host and guest CPUs. */
struct
{
/** Padding for hidden fields. */
uint8_t abHidden0[64 + 48];
/** Guest CPU feature information. */
CPUMFEATURES GuestFeatures;
} const ro;
#endif
/** @todo this is rather bloated because of static MSR range allocation.
* Probably a good idea to move it to a separate R0 allocation... */
uint8_t padding[8832 + 128*8192 + 0x1d00]; /* multiple of 64 */
} cpum;
/** PGM part.
* @note 16384 aligned for zero and mmio page storage. */
union
{
#ifdef VMM_INCLUDED_SRC_include_PGMInternal_h
struct PGM s;
#endif
uint8_t padding[53888]; /* multiple of 64 */
} pgm;
/** VMM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_VMMInternal_h
struct VMM s;
#endif
uint8_t padding[1600]; /* multiple of 64 */
} vmm;
/** HM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_HMInternal_h
struct HM s;
#endif
uint8_t padding[5504]; /* multiple of 64 */
} hm;
/** TRPM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_TRPMInternal_h
struct TRPM s;
#endif
uint8_t padding[2048]; /* multiple of 64 */
} trpm;
/** SELM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_SELMInternal_h
struct SELM s;
#endif
uint8_t padding[768]; /* multiple of 64 */
} selm;
/** MM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_MMInternal_h
struct MM s;
#endif
uint8_t padding[192]; /* multiple of 64 */
} mm;
/** PDM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_PDMInternal_h
struct PDM s;
#endif
uint8_t padding[22400]; /* multiple of 64 */
} pdm;
/** IOM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_IOMInternal_h
struct IOM s;
#endif
uint8_t padding[1152]; /* multiple of 64 */
} iom;
/** EM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_EMInternal_h
struct EM s;
#endif
uint8_t padding[256]; /* multiple of 64 */
} em;
/** NEM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_NEMInternal_h
struct NEM s;
#endif
uint8_t padding[4608]; /* multiple of 64 */
} nem;
/** TM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_TMInternal_h
struct TM s;
#endif
uint8_t padding[10112]; /* multiple of 64 */
} tm;
/** DBGF part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_DBGFInternal_h
struct DBGF s;
#endif
#ifdef VBOX_INCLUDED_vmm_dbgf_h
/** Read only info exposed about interrupt breakpoints and selected events. */
struct
{
/** Bitmap of enabled hardware interrupt breakpoints. */
uint32_t bmHardIntBreakpoints[256 / 32];
/** Bitmap of enabled software interrupt breakpoints. */
uint32_t bmSoftIntBreakpoints[256 / 32];
/** Bitmap of selected events.
* This includes non-selectable events too for simplicity, we maintain the
* state for some of these, as it may come in handy. */
uint64_t bmSelectedEvents[(DBGFEVENT_END + 63) / 64];
/** Enabled hardware interrupt breakpoints. */
uint32_t cHardIntBreakpoints;
/** Enabled software interrupt breakpoints. */
uint32_t cSoftIntBreakpoints;
/** The number of selected events. */
uint32_t cSelectedEvents;
/** The number of enabled hardware breakpoints. */
uint8_t cEnabledHwBreakpoints;
/** The number of enabled hardware I/O breakpoints. */
uint8_t cEnabledHwIoBreakpoints;
uint8_t au8Alignment1[2]; /**< Alignment padding. */
/** The number of enabled INT3 breakpoints. */
uint32_t volatile cEnabledInt3Breakpoints;
} const ro;
#endif
uint8_t padding[2432]; /* multiple of 64 */
} dbgf;
/** SSM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_SSMInternal_h
struct SSM s;
#endif
uint8_t padding[128]; /* multiple of 64 */
} ssm;
union
{
#ifdef VMM_INCLUDED_SRC_include_GIMInternal_h
struct GIM s;
#endif
uint8_t padding[448]; /* multiple of 64 */
} gim;
union
{
#ifdef VMM_INCLUDED_SRC_include_APICInternal_h
struct APIC s;
#endif
uint8_t padding[128]; /* multiple of 8 */
} apic;
/* ---- begin small stuff ---- */
/** VM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_VMInternal_h
struct VMINT s;
#endif
uint8_t padding[32]; /* multiple of 8 */
} vm;
/** CFGM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_CFGMInternal_h
struct CFGM s;
#endif
uint8_t padding[8]; /* multiple of 8 */
} cfgm;
/** IEM part. */
union
{
#ifdef VMM_INCLUDED_SRC_include_IEMInternal_h
struct IEM s;
#endif
uint8_t padding[16]; /* multiple of 8 */
} iem;
/** Statistics for ring-0 only components. */
struct
{
/** GMMR0 stats. */
struct
{
/** Chunk TLB hits. */
uint64_t cChunkTlbHits;
/** Chunk TLB misses. */
uint64_t cChunkTlbMisses;
} gmm;
uint64_t au64Padding[6]; /* probably more comming here... */
} R0Stats;
union
{
#ifdef VMM_INCLUDED_SRC_include_GCMInternal_h
struct GCM s;
#endif
uint8_t padding[32]; /* multiple of 8 */
} gcm;
/** Padding for aligning the structure size on a page boundrary. */
uint8_t abAlignment2[8872 - sizeof(PVMCPUR3) * VMM_MAX_CPU_COUNT];
/* ---- end small stuff ---- */
/** Array of VMCPU ring-3 pointers. */
PVMCPUR3 apCpusR3[VMM_MAX_CPU_COUNT];
/* This point is aligned on a 16384 boundrary (for arm64 purposes). */
} VM;
#ifndef VBOX_FOR_DTRACE_LIB
//AssertCompileSizeAlignment(VM, 16384);
#endif
#ifdef IN_RC
RT_C_DECLS_BEGIN
/** The VM structure.
* This is imported from the VMMRCBuiltin module, i.e. it's a one of those magic
* globals which we should avoid using.
*/
extern DECLIMPORT(VM) g_VM;
/** The VMCPU structure for virtual CPU \#0.
* This is imported from the VMMRCBuiltin module, i.e. it's a one of those magic
* globals which we should avoid using.
*/
extern DECLIMPORT(VMCPU) g_VCpu0;
RT_C_DECLS_END
#endif
/** @} */
#endif /* !VBOX_INCLUDED_vmm_vm_h */