2551 lines
82 KiB
C++
2551 lines
82 KiB
C++
/////////////////////////////////////////////////////////////////////////
|
|
// $Id: proc_ctrl.cc,v 1.196 2008-01-18 08:57:35 sshwarts Exp $
|
|
/////////////////////////////////////////////////////////////////////////
|
|
//
|
|
// Copyright (C) 2001 MandrakeSoft S.A.
|
|
//
|
|
// MandrakeSoft S.A.
|
|
// 43, rue d'Aboukir
|
|
// 75002 Paris - France
|
|
// http://www.linux-mandrake.com/
|
|
// http://www.mandrakesoft.com/
|
|
//
|
|
// This library is free software; you can redistribute it and/or
|
|
// modify it under the terms of the GNU Lesser General Public
|
|
// License as published by the Free Software Foundation; either
|
|
// version 2 of the License, or (at your option) any later version.
|
|
//
|
|
// This library 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
|
|
// Lesser General Public License for more details.
|
|
//
|
|
// You should have received a copy of the GNU Lesser General Public
|
|
// License along with this library; if not, write to the Free Software
|
|
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
//
|
|
/////////////////////////////////////////////////////////////////////////
|
|
|
|
#define NEED_CPU_REG_SHORTCUTS 1
|
|
#include "bochs.h"
|
|
#include "cpu.h"
|
|
#define LOG_THIS BX_CPU_THIS_PTR
|
|
|
|
#if BX_SUPPORT_X86_64==0
|
|
// Make life easier for merging code.
|
|
#define RAX EAX
|
|
#define RCX ECX
|
|
#define RDX EDX
|
|
#endif
|
|
|
|
void BX_CPU_C::UndefinedOpcode(bxInstruction_c *i)
|
|
{
|
|
BX_DEBUG(("UndefinedOpcode: 0x%02x causes exception 6", (unsigned) i->b1()));
|
|
exception(BX_UD_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
void BX_CPU_C::NOP(bxInstruction_c *i)
|
|
{
|
|
// No operation.
|
|
}
|
|
|
|
void BX_CPU_C::PREFETCH(bxInstruction_c *i)
|
|
{
|
|
#if BX_SUPPORT_3DNOW || BX_SUPPORT_SSE >= 1
|
|
if (i->modC0()) {
|
|
BX_ERROR(("PREFETCH: use of register is undefined opcode"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
|
|
#if BX_INSTRUMENTATION
|
|
BX_CPU_CALL_METHODR(i->ResolveModrm, (i));
|
|
BX_INSTR_PREFETCH_HINT(BX_CPU_ID, i->nnn(), i->seg(), RMAddr(i));
|
|
#endif
|
|
|
|
#else
|
|
BX_INFO(("PREFETCH: required SSE or 3DNOW support"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
//
|
|
// The shutdown state is very similar to the state following the exection
|
|
// if HLT instruction. In this mode the processor stops executing
|
|
// instructions until #NMI, #SMI, #RESET or #INIT is received. If
|
|
// shutdown occurs why in NMI interrupt handler or in SMM, a hardware
|
|
// reset must be used to restart the processor execution.
|
|
//
|
|
void BX_CPU_C::shutdown(void)
|
|
{
|
|
BX_PANIC(("Entering to shutdown state still not implemented"));
|
|
|
|
BX_CPU_THIS_PTR clear_IF();
|
|
|
|
// artificial trap bit, why use another variable.
|
|
BX_CPU_THIS_PTR debug_trap |= BX_DEBUG_TRAP_SHUTDOWN; // artificial trap
|
|
BX_CPU_THIS_PTR async_event = 1; // so processor knows to check
|
|
// Execution of this instruction completes. The processor
|
|
// will remain in a halt state until one of the above conditions
|
|
// is met.
|
|
|
|
BX_INSTR_HLT(BX_CPU_ID);
|
|
|
|
#if BX_USE_IDLE_HACK
|
|
bx_gui->sim_is_idle();
|
|
#endif
|
|
|
|
longjmp(BX_CPU_THIS_PTR jmp_buf_env, 1); // go back to main decode loop
|
|
}
|
|
|
|
void BX_CPU_C::HLT(bxInstruction_c *i)
|
|
{
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_DEBUG(("HLT: %s priveledge check failed, CPL=%d, generate #GP(0)",
|
|
cpu_mode_string(BX_CPU_THIS_PTR cpu_mode), CPL));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
if (! BX_CPU_THIS_PTR get_IF()) {
|
|
BX_INFO(("WARNING: HLT instruction with IF=0!"));
|
|
}
|
|
|
|
// stops instruction execution and places the processor in a
|
|
// HALT state. An enabled interrupt, NMI, or reset will resume
|
|
// execution. If interrupt (including NMI) is used to resume
|
|
// execution after HLT, the saved CS:eIP points to instruction
|
|
// following HLT.
|
|
|
|
// artificial trap bit, why use another variable.
|
|
BX_CPU_THIS_PTR debug_trap |= BX_DEBUG_TRAP_HALT; // artificial trap
|
|
BX_CPU_THIS_PTR async_event = 1; // so processor knows to check
|
|
// Execution of this instruction completes. The processor
|
|
// will remain in a halt state until one of the above conditions
|
|
// is met.
|
|
|
|
BX_INSTR_HLT(BX_CPU_ID);
|
|
|
|
#if BX_USE_IDLE_HACK
|
|
bx_gui->sim_is_idle();
|
|
#endif
|
|
}
|
|
|
|
void BX_CPU_C::CLTS(bxInstruction_c *i)
|
|
{
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("CLTS: priveledge check failed, generate #GP(0)"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
BX_CPU_THIS_PTR cr0.set_TS(0);
|
|
}
|
|
|
|
void BX_CPU_C::INVD(bxInstruction_c *i)
|
|
{
|
|
#if BX_CPU_LEVEL >= 4
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("INVD: priveledge check failed, generate #GP(0)"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
BX_DEBUG(("INVD: Flush internal caches !"));
|
|
BX_INSTR_CACHE_CNTRL(BX_CPU_ID, BX_INSTR_INVD);
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
flushICaches();
|
|
#endif
|
|
|
|
#else
|
|
BX_INFO(("INVD: required 486 support, use --enable-cpu-level=4 option"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
void BX_CPU_C::WBINVD(bxInstruction_c *i)
|
|
{
|
|
#if BX_CPU_LEVEL >= 4
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("WBINVD: priveledge check failed, generate #GP(0)"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
BX_DEBUG(("WBINVD: Flush internal caches !"));
|
|
BX_INSTR_CACHE_CNTRL(BX_CPU_ID, BX_INSTR_WBINVD);
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
flushICaches();
|
|
#endif
|
|
|
|
#else
|
|
BX_INFO(("WBINVD: required 486 support, use --enable-cpu-level=4 option"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
void BX_CPU_C::CLFLUSH(bxInstruction_c *i)
|
|
{
|
|
#if BX_SUPPORT_CLFLUSH
|
|
bx_segment_reg_t *seg = &BX_CPU_THIS_PTR sregs[i->seg()];
|
|
|
|
BX_CPU_CALL_METHODR(i->ResolveModrm, (i));
|
|
|
|
// check if we could access the memory segment
|
|
if ((seg->cache.valid & SegAccessROK4G) != SegAccessROK4G)
|
|
{
|
|
execute_virtual_checks(seg, RMAddr(i), 1);
|
|
}
|
|
|
|
bx_address laddr = BX_CPU_THIS_PTR get_segment_base(i->seg()) + RMAddr(i);
|
|
bx_phy_address paddr;
|
|
|
|
if (BX_CPU_THIS_PTR cr0.get_PG()) {
|
|
paddr = dtranslate_linear(laddr, CPL, BX_READ);
|
|
paddr = A20ADDR(paddr);
|
|
}
|
|
else
|
|
{
|
|
paddr = A20ADDR(laddr);
|
|
}
|
|
|
|
BX_INSTR_CLFLUSH(BX_CPU_ID, laddr, paddr);
|
|
|
|
#else
|
|
BX_INFO(("CLFLUSH: not supported, enable with SSE2"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
#if BX_CPU_LEVEL >= 3
|
|
void BX_CPU_C::MOV_DdRd(bxInstruction_c *i)
|
|
{
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("MOV_DdRd: CPL!=0 not in real mode"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
/* NOTES:
|
|
* 32bit operands always used
|
|
* r/m field specifies general register
|
|
* reg field specifies which special register
|
|
*/
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
/* This instruction is always treated as a register-to-register,
|
|
* regardless of the encoding of the MOD field in the MODRM byte.
|
|
*/
|
|
if (!i->modC0())
|
|
BX_PANIC(("MOV_DdRd(): rm field not a register!"));
|
|
|
|
Bit32u val_32 = BX_READ_32BIT_REG(i->rm());
|
|
|
|
switch (i->nnn()) {
|
|
case 0: // DR0
|
|
BX_CPU_THIS_PTR dr0 = val_32;
|
|
break;
|
|
case 1: // DR1
|
|
BX_CPU_THIS_PTR dr1 = val_32;
|
|
break;
|
|
case 2: // DR2
|
|
BX_CPU_THIS_PTR dr2 = val_32;
|
|
break;
|
|
case 3: // DR3
|
|
BX_CPU_THIS_PTR dr3 = val_32;
|
|
break;
|
|
|
|
case 4: // DR4
|
|
// DR4 aliased to DR6 by default. With Debug Extensions on,
|
|
// access to DR4 causes #UD
|
|
#if BX_CPU_LEVEL >= 4
|
|
if (BX_CPU_THIS_PTR cr4.get_DE()) {
|
|
// Debug extensions CR4.DE is ON
|
|
BX_INFO(("MOV_DdRd: access to DR4 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
#endif
|
|
// Fall through to DR6 case
|
|
case 6: // DR6
|
|
#if BX_CPU_LEVEL <= 4
|
|
// On 386/486 bit12 is settable
|
|
BX_CPU_THIS_PTR dr6 = (BX_CPU_THIS_PTR dr6 & 0xffff0ff0) |
|
|
(val_32 & 0x0000f00f);
|
|
#else
|
|
// On Pentium+, bit12 is always zero
|
|
BX_CPU_THIS_PTR dr6 = (BX_CPU_THIS_PTR dr6 & 0xffff0ff0) |
|
|
(val_32 & 0x0000e00f);
|
|
#endif
|
|
break;
|
|
|
|
case 5: // DR5
|
|
// DR5 aliased to DR7 by default. With Debug Extensions on,
|
|
// access to DR5 causes #UD
|
|
#if BX_CPU_LEVEL >= 4
|
|
if (BX_CPU_THIS_PTR cr4.get_DE()) {
|
|
// Debug extensions CR4.DE is ON
|
|
BX_INFO(("MOV_DdRd: access to DR5 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
#endif
|
|
// Fall through to DR7 case
|
|
case 7: // DR7
|
|
// Note: 486+ ignore GE and LE flags. On the 386, exact
|
|
// data breakpoint matching does not occur unless it is enabled
|
|
// by setting the LE and/or GE flags.
|
|
|
|
// Some sanity checks...
|
|
if (val_32 & 0x00002000) {
|
|
BX_INFO(("MOV_DdRd: GD bit not supported yet"));
|
|
// Note: processor clears GD upon entering debug exception
|
|
// handler, to allow access to the debug registers
|
|
}
|
|
if ( (((val_32>>16) & 3)==2) ||
|
|
(((val_32>>20) & 3)==2) ||
|
|
(((val_32>>24) & 3)==2) ||
|
|
(((val_32>>28) & 3)==2) ) {
|
|
// IO breakpoints (10b) are not yet supported.
|
|
BX_PANIC(("MOV_DdRd: write of %08x contains IO breakpoint", val_32));
|
|
}
|
|
if ( (((val_32>>18) & 3)==2) ||
|
|
(((val_32>>22) & 3)==2) ||
|
|
(((val_32>>26) & 3)==2) ||
|
|
(((val_32>>30) & 3)==2) ) {
|
|
// LEN0..3 contains undefined length specifier (10b)
|
|
BX_PANIC(("MOV_DdRd: write of %08x contains undefined LENx", val_32));
|
|
}
|
|
if ( ((((val_32>>16) & 3)==0) && (((val_32>>18) & 3)!=0)) ||
|
|
((((val_32>>20) & 3)==0) && (((val_32>>22) & 3)!=0)) ||
|
|
((((val_32>>24) & 3)==0) && (((val_32>>26) & 3)!=0)) ||
|
|
((((val_32>>28) & 3)==0) && (((val_32>>30) & 3)!=0)) )
|
|
{
|
|
// Instruction breakpoint with LENx not 00b (1-byte length)
|
|
BX_PANIC(("MOV_DdRd: write of %08x, R/W=00b LEN!=00b", val_32));
|
|
}
|
|
#if BX_CPU_LEVEL <= 4
|
|
// 386/486: you can play with all the bits except b10 is always 1
|
|
BX_CPU_THIS_PTR dr7 = val_32 | 0x00000400;
|
|
#else
|
|
// Pentium+: bits15,14,12 are hardwired to 0, rest are settable.
|
|
// Even bits 11,10 are changeable though reserved.
|
|
BX_CPU_THIS_PTR dr7 = (val_32 & 0xffff2fff) | 0x00000400;
|
|
#endif
|
|
// if we have breakpoints enabled then we must check
|
|
// breakpoints condition in cpu loop
|
|
if(BX_CPU_THIS_PTR dr7 & 0xff)
|
|
BX_CPU_THIS_PTR async_event = 1;
|
|
break;
|
|
|
|
default:
|
|
BX_ERROR(("MOV_DdRd: #UD - control register index out of range"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
}
|
|
|
|
void BX_CPU_C::MOV_RdDd(bxInstruction_c *i)
|
|
{
|
|
Bit32u val_32;
|
|
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("MOV_RdDd: CPL!=0 not in real mode"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
/* This instruction is always treated as a register-to-register,
|
|
* regardless of the encoding of the MOD field in the MODRM byte.
|
|
*/
|
|
if (!i->modC0())
|
|
BX_PANIC(("MOV_RdDd(): rm field not a register!"));
|
|
|
|
switch (i->nnn()) {
|
|
case 0: // DR0
|
|
val_32 = (Bit32u) BX_CPU_THIS_PTR dr0;
|
|
break;
|
|
case 1: // DR1
|
|
val_32 = (Bit32u) BX_CPU_THIS_PTR dr1;
|
|
break;
|
|
case 2: // DR2
|
|
val_32 = (Bit32u) BX_CPU_THIS_PTR dr2;
|
|
break;
|
|
case 3: // DR3
|
|
val_32 = (Bit32u) BX_CPU_THIS_PTR dr3;
|
|
break;
|
|
|
|
case 4: // DR4
|
|
// DR4 aliased to DR6 by default. With Debug Extensions on,
|
|
// access to DR4 causes #UD
|
|
#if BX_CPU_LEVEL >= 4
|
|
if (BX_CPU_THIS_PTR cr4.get_DE()) {
|
|
// Debug extensions CR4.DE is ON
|
|
BX_INFO(("MOV_RdDd: access to DR4 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
#endif
|
|
// Fall through to DR6 case
|
|
case 6: // DR6
|
|
val_32 = BX_CPU_THIS_PTR dr6;
|
|
break;
|
|
|
|
case 5: // DR5
|
|
// DR5 aliased to DR7 by default. With Debug Extensions on,
|
|
// access to DR5 causes #UD
|
|
#if BX_CPU_LEVEL >= 4
|
|
if (BX_CPU_THIS_PTR cr4.get_DE()) {
|
|
// Debug extensions CR4.DE is ON
|
|
BX_INFO(("MOV_RdDd: access to DR5 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
#endif
|
|
// Fall through to DR7 case
|
|
case 7: // DR7
|
|
val_32 = BX_CPU_THIS_PTR dr7;
|
|
break;
|
|
|
|
default:
|
|
BX_ERROR(("MOV_RdDd: #UD - control register index out of range"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
|
|
BX_WRITE_32BIT_REGZ(i->rm(), val_32);
|
|
}
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
void BX_CPU_C::MOV_DqRq(bxInstruction_c *i)
|
|
{
|
|
BX_ASSERT(protected_mode());
|
|
|
|
/* NOTES:
|
|
* 64bit operands always used
|
|
* r/m field specifies general register
|
|
* reg field specifies which special register
|
|
*/
|
|
|
|
/* #GP(0) if CPL is not 0 */
|
|
if (CPL != 0) {
|
|
BX_ERROR(("MOV_DqRq: #GP(0) if CPL is not 0"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
/* This instruction is always treated as a register-to-register,
|
|
* regardless of the encoding of the MOD field in the MODRM byte.
|
|
*/
|
|
if (!i->modC0())
|
|
BX_PANIC(("MOV_DqRq(): rm field not a register!"));
|
|
|
|
Bit64u val_64 = BX_READ_64BIT_REG(i->rm());
|
|
|
|
switch (i->nnn()) {
|
|
case 0: // DR0
|
|
BX_CPU_THIS_PTR dr0 = val_64;
|
|
break;
|
|
case 1: // DR1
|
|
BX_CPU_THIS_PTR dr1 = val_64;
|
|
break;
|
|
case 2: // DR2
|
|
BX_CPU_THIS_PTR dr2 = val_64;
|
|
break;
|
|
case 3: // DR3
|
|
BX_CPU_THIS_PTR dr3 = val_64;
|
|
break;
|
|
|
|
case 4: // DR4
|
|
// DR4 aliased to DR6 by default. With Debug Extensions on,
|
|
// access to DR4 causes #UD
|
|
if (BX_CPU_THIS_PTR cr4.get_DE()) {
|
|
// Debug extensions CR4.DE is ON
|
|
BX_INFO(("MOV_DqRq: access to DR4 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
// Fall through to DR6 case
|
|
case 6: // DR6
|
|
// On Pentium+, bit12 is always zero
|
|
BX_CPU_THIS_PTR dr6 = (BX_CPU_THIS_PTR dr6 & 0xffff0ff0) |
|
|
(val_64 & 0x0000e00f);
|
|
break;
|
|
|
|
case 5: // DR5
|
|
// DR5 aliased to DR7 by default. With Debug Extensions on,
|
|
// access to DR5 causes #UD
|
|
if (BX_CPU_THIS_PTR cr4.get_DE()) {
|
|
// Debug extensions CR4.DE is ON
|
|
BX_INFO(("MOV_DqRq: access to DR5 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
// Fall through to DR7 case
|
|
case 7: // DR7
|
|
// Note: 486+ ignore GE and LE flags. On the 386, exact
|
|
// data breakpoint matching does not occur unless it is enabled
|
|
// by setting the LE and/or GE flags.
|
|
|
|
// Some sanity checks...
|
|
if (val_64 & 0x00002000) {
|
|
BX_PANIC(("MOV_DqRq: GD bit not supported yet"));
|
|
// Note: processor clears GD upon entering debug exception
|
|
// handler, to allow access to the debug registers
|
|
}
|
|
if ( (((val_64>>16) & 3)==2) ||
|
|
(((val_64>>20) & 3)==2) ||
|
|
(((val_64>>24) & 3)==2) ||
|
|
(((val_64>>28) & 3)==2) )
|
|
{
|
|
// IO breakpoints (10b) are not yet supported.
|
|
BX_PANIC(("MOV_DqRq: write of %08x:%08x contains IO breakpoint",
|
|
(Bit32u)(val_64 >> 32), (Bit32u)(val_64 & 0xFFFFFFFF)));
|
|
}
|
|
if ( (((val_64>>18) & 3)==2) ||
|
|
(((val_64>>22) & 3)==2) ||
|
|
(((val_64>>26) & 3)==2) ||
|
|
(((val_64>>30) & 3)==2) )
|
|
{
|
|
// LEN0..3 contains undefined length specifier (10b)
|
|
BX_PANIC(("MOV_DqRq: write of %08x:%08x contains undefined LENx",
|
|
(Bit32u)(val_64 >> 32), (Bit32u)(val_64 & 0xFFFFFFFF)));
|
|
}
|
|
if ( ((((val_64>>16) & 3)==0) && (((val_64>>18) & 3)!=0)) ||
|
|
((((val_64>>20) & 3)==0) && (((val_64>>22) & 3)!=0)) ||
|
|
((((val_64>>24) & 3)==0) && (((val_64>>26) & 3)!=0)) ||
|
|
((((val_64>>28) & 3)==0) && (((val_64>>30) & 3)!=0)) )
|
|
{
|
|
// Instruction breakpoint with LENx not 00b (1-byte length)
|
|
BX_PANIC(("MOV_DqRq: write of %08x:%08x , R/W=00b LEN!=00b",
|
|
(Bit32u)(val_64 >> 32), (Bit32u)(val_64 & 0xFFFFFFFF)));
|
|
}
|
|
|
|
// Pentium+: bits15,14,12 are hardwired to 0, rest are settable.
|
|
// Even bits 11,10 are changeable though reserved.
|
|
BX_CPU_THIS_PTR dr7 = (val_64 & 0xffff2fff) | 0x00000400;
|
|
break;
|
|
|
|
default:
|
|
BX_ERROR(("MOV_DqRq: #UD - control register index out of range"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
}
|
|
|
|
void BX_CPU_C::MOV_RqDq(bxInstruction_c *i)
|
|
{
|
|
Bit64u val_64;
|
|
|
|
BX_ASSERT(protected_mode());
|
|
|
|
/* #GP(0) if CPL is not 0 */
|
|
if (CPL != 0) {
|
|
BX_ERROR(("MOV_RqDq: #GP(0) if CPL is not 0"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
/* This instruction is always treated as a register-to-register,
|
|
* regardless of the encoding of the MOD field in the MODRM byte.
|
|
*/
|
|
if (!i->modC0())
|
|
BX_PANIC(("MOV_RqDq(): rm field not a register!"));
|
|
|
|
switch (i->nnn()) {
|
|
case 0: // DR0
|
|
val_64 = BX_CPU_THIS_PTR dr0;
|
|
break;
|
|
case 1: // DR1
|
|
val_64 = BX_CPU_THIS_PTR dr1;
|
|
break;
|
|
case 2: // DR2
|
|
val_64 = BX_CPU_THIS_PTR dr2;
|
|
break;
|
|
case 3: // DR3
|
|
val_64 = BX_CPU_THIS_PTR dr3;
|
|
break;
|
|
|
|
case 4: // DR4
|
|
// DR4 aliased to DR6 by default. With Debug Extensions on,
|
|
// access to DR4 causes #UD
|
|
if (BX_CPU_THIS_PTR cr4.get_DE()) {
|
|
// Debug extensions CR4.DE is ON
|
|
BX_INFO(("MOV_RqDq: access to DR4 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
// Fall through to DR6 case
|
|
case 6: // DR6
|
|
val_64 = BX_CPU_THIS_PTR dr6;
|
|
break;
|
|
|
|
case 5: // DR5
|
|
// DR5 aliased to DR7 by default. With Debug Extensions on,
|
|
// access to DR5 causes #UD
|
|
if (BX_CPU_THIS_PTR cr4.get_DE()) {
|
|
// Debug extensions CR4.DE is ON
|
|
BX_INFO(("MOV_RqDq: access to DR5 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
// Fall through to DR7 case
|
|
case 7: // DR7
|
|
val_64 = BX_CPU_THIS_PTR dr7;
|
|
break;
|
|
|
|
default:
|
|
BX_ERROR(("MOV_DqRq: #UD - control register index out of range"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
|
|
BX_WRITE_64BIT_REG(i->rm(), val_64);
|
|
}
|
|
#endif // #if BX_SUPPORT_X86_64
|
|
|
|
void BX_CPU_C::MOV_CdRd(bxInstruction_c *i)
|
|
{
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("MOV_CdRd: CPL!=0 not in real mode"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
/* NOTES:
|
|
* 32bit operands always used
|
|
* r/m field specifies general register
|
|
* reg field specifies which special register
|
|
*/
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
/* This instruction is always treated as a register-to-register,
|
|
* regardless of the encoding of the MOD field in the MODRM byte.
|
|
*/
|
|
if (!i->modC0())
|
|
BX_PANIC(("MOV_CdRd(): rm field not a register!"));
|
|
|
|
Bit32u val_32 = BX_READ_32BIT_REG(i->rm());
|
|
|
|
switch (i->nnn()) {
|
|
case 0: // CR0 (MSW)
|
|
SetCR0(val_32);
|
|
break;
|
|
case 2: /* CR2 */
|
|
BX_DEBUG(("MOV_CdRd:CR2 = %08x", (unsigned) val_32));
|
|
BX_CPU_THIS_PTR cr2 = val_32;
|
|
break;
|
|
case 3: // CR3
|
|
BX_DEBUG(("MOV_CdRd:CR3 = %08x", (unsigned) val_32));
|
|
// Reserved bits take on value of MOV instruction
|
|
CR3_change(val_32);
|
|
BX_INSTR_TLB_CNTRL(BX_CPU_ID, BX_INSTR_MOV_CR3, val_32);
|
|
break;
|
|
case 4: // CR4
|
|
#if BX_CPU_LEVEL == 3
|
|
BX_PANIC(("MOV_CdRd: write to CR4 of 0x%08x on 386", val_32));
|
|
UndefinedOpcode(i);
|
|
#else
|
|
// Protected mode: #GP(0) if attempt to write a 1 to
|
|
// any reserved bit of CR4
|
|
if (! SetCR4(val_32))
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
#endif
|
|
break;
|
|
default:
|
|
BX_ERROR(("MOV_CdRd: #UD - control register index out of range"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
}
|
|
|
|
void BX_CPU_C::MOV_RdCd(bxInstruction_c *i)
|
|
{
|
|
// mov control register data to register
|
|
Bit32u val_32;
|
|
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("MOV_RdCd: CPL!=0 not in real mode"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
/* NOTES:
|
|
* 32bit operands always used
|
|
* r/m field specifies general register
|
|
* reg field specifies which special register
|
|
*/
|
|
|
|
/* This instruction is always treated as a register-to-register,
|
|
* regardless of the encoding of the MOD field in the MODRM byte.
|
|
*/
|
|
if (!i->modC0())
|
|
BX_PANIC(("MOV_RdCd(): rm field not a register!"));
|
|
|
|
switch (i->nnn()) {
|
|
case 0: // CR0 (MSW)
|
|
val_32 = BX_CPU_THIS_PTR cr0.val32;
|
|
break;
|
|
case 2: /* CR2 */
|
|
BX_DEBUG(("MOV_RdCd: reading CR2"));
|
|
val_32 = (Bit32u) BX_CPU_THIS_PTR cr2;
|
|
break;
|
|
case 3: // CR3
|
|
BX_DEBUG(("MOV_RdCd: reading CR3"));
|
|
val_32 = BX_CPU_THIS_PTR cr3;
|
|
break;
|
|
case 4: // CR4
|
|
#if BX_CPU_LEVEL < 4
|
|
val_32 = 0;
|
|
BX_INFO(("MOV_RdCd: read of CR4 causes #UD"));
|
|
UndefinedOpcode(i);
|
|
#else
|
|
BX_DEBUG(("MOV_RdCd: read of CR4"));
|
|
val_32 = BX_CPU_THIS_PTR cr4.getRegister();
|
|
#endif
|
|
break;
|
|
default:
|
|
BX_ERROR(("MOV_RdCd: #UD - control register index out of range"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
|
|
BX_WRITE_32BIT_REGZ(i->rm(), val_32);
|
|
}
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
void BX_CPU_C::MOV_CqRq(bxInstruction_c *i)
|
|
{
|
|
BX_ASSERT(protected_mode());
|
|
|
|
/* NOTES:
|
|
* 64bit operands always used
|
|
* r/m field specifies general register
|
|
* reg field specifies which special register
|
|
*/
|
|
|
|
/* #GP(0) if CPL is not 0 */
|
|
if (CPL!=0) {
|
|
BX_ERROR(("MOV_CqRq: #GP(0) if CPL is not 0"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
/* This instruction is always treated as a register-to-register,
|
|
* regardless of the encoding of the MOD field in the MODRM byte.
|
|
*/
|
|
if (!i->modC0())
|
|
BX_PANIC(("MOV_CqRq(): rm field not a register!"));
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
Bit64u val_64 = BX_READ_64BIT_REG(i->rm());
|
|
|
|
switch (i->nnn()) {
|
|
case 0: // CR0 (MSW)
|
|
SetCR0((Bit32u) val_64);
|
|
break;
|
|
case 2: /* CR2 */
|
|
BX_DEBUG(("MOV_CqRq: write to CR2 of %08x:%08x",
|
|
(Bit32u)(val_64 >> 32), (Bit32u)(val_64 & 0xFFFFFFFF)));
|
|
BX_CPU_THIS_PTR cr2 = val_64;
|
|
break;
|
|
case 3: // CR3
|
|
BX_DEBUG(("MOV_CqRq: write to CR3 of %08x:%08x",
|
|
(Bit32u)(val_64 >> 32), (Bit32u)(val_64 & 0xFFFFFFFF)));
|
|
if (val_64 & BX_CONST64(0xffffffff00000000)) {
|
|
BX_PANIC(("CR3 write: Only 32 bit physical address space is emulated !"));
|
|
}
|
|
// Reserved bits take on value of MOV instruction
|
|
CR3_change((bx_phy_address) val_64);
|
|
BX_INSTR_TLB_CNTRL(BX_CPU_ID, BX_INSTR_MOV_CR3, val_64);
|
|
break;
|
|
case 4: // CR4
|
|
// Protected mode: #GP(0) if attempt to write a 1 to
|
|
// any reserved bit of CR4
|
|
BX_DEBUG(("MOV_CqRq: write to CR4 of %08x:%08x",
|
|
(Bit32u)(val_64 >> 32), (Bit32u)(val_64 & 0xFFFFFFFF)));
|
|
if (! SetCR4((Bit32u) val_64))
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
break;
|
|
#if BX_SUPPORT_APIC
|
|
case 8: // CR8
|
|
// CR8 is aliased to APIC->TASK PRIORITY register
|
|
// APIC.TPR[7:4] = CR8[3:0]
|
|
// APIC.TPR[3:0] = 0
|
|
// Reads of CR8 return zero extended APIC.TPR[7:4]
|
|
// Write to CR8 update APIC.TPR[7:4]
|
|
BX_CPU_THIS_PTR local_apic.set_tpr((val_64 & 0xF) << 0x4);
|
|
break;
|
|
#endif
|
|
default:
|
|
BX_ERROR(("MOV_CqRq: #UD - control register index out of range"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
}
|
|
|
|
void BX_CPU_C::MOV_RqCq(bxInstruction_c *i)
|
|
{
|
|
// mov control register data to register
|
|
Bit64u val_64;
|
|
|
|
BX_ASSERT(protected_mode());
|
|
|
|
/* NOTES:
|
|
* 64bit operands always used
|
|
* r/m field specifies general register
|
|
* reg field specifies which special register
|
|
*/
|
|
|
|
/* #GP(0) if CPL is not 0 */
|
|
if (CPL!=0) {
|
|
BX_ERROR(("MOV_RqCq: #GP(0) if CPL is not 0"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
/* This instruction is always treated as a register-to-register,
|
|
* regardless of the encoding of the MOD field in the MODRM byte.
|
|
*/
|
|
if (!i->modC0())
|
|
BX_PANIC(("MOV_RqCq(): rm field not a register!"));
|
|
|
|
switch (i->nnn()) {
|
|
case 0: // CR0 (MSW)
|
|
val_64 = BX_CPU_THIS_PTR cr0.val32;
|
|
break;
|
|
case 2: /* CR2 */
|
|
BX_DEBUG(("MOV_RqCq: read of CR2"));
|
|
val_64 = BX_CPU_THIS_PTR cr2;
|
|
break;
|
|
case 3: // CR3
|
|
BX_DEBUG(("MOV_RqCq: read of CR3"));
|
|
val_64 = BX_CPU_THIS_PTR cr3;
|
|
break;
|
|
case 4: // CR4
|
|
BX_DEBUG(("MOV_RqCq: read of CR4"));
|
|
val_64 = BX_CPU_THIS_PTR cr4.getRegister();
|
|
break;
|
|
#if BX_SUPPORT_APIC
|
|
case 8: // CR8
|
|
// CR8 is aliased to APIC->TASK PRIORITY register
|
|
// APIC.TPR[7:4] = CR8[3:0]
|
|
// APIC.TPR[3:0] = 0
|
|
// Reads of CR8 return zero extended APIC.TPR[7:4]
|
|
// Write to CR8 update APIC.TPR[7:4]
|
|
val_64 = (BX_CPU_THIS_PTR local_apic.get_tpr() & 0xF) >> 4;
|
|
break;
|
|
#endif
|
|
default:
|
|
BX_ERROR(("MOV_RqCq: #UD - control register index out of range"));
|
|
UndefinedOpcode(i);
|
|
}
|
|
|
|
BX_WRITE_64BIT_REG(i->rm(), val_64);
|
|
}
|
|
#endif // #if BX_SUPPORT_X86_64
|
|
|
|
#endif // #if BX_CPU_LEVEL >= 3
|
|
|
|
#if BX_CPU_LEVEL >= 2
|
|
void BX_CPU_C::LMSW_Ew(bxInstruction_c *i)
|
|
{
|
|
Bit16u msw;
|
|
Bit32u cr0;
|
|
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("LMSW: CPL!=0 not in real mode"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
if (i->modC0()) {
|
|
msw = BX_READ_16BIT_REG(i->rm());
|
|
}
|
|
else {
|
|
BX_CPU_CALL_METHODR(i->ResolveModrm, (i));
|
|
/* pointer, segment address pair */
|
|
msw = read_virtual_word(i->seg(), RMAddr(i));
|
|
}
|
|
|
|
// LMSW does not affect PG,CD,NW,AM,WP,NE,ET bits, and cannot clear PE
|
|
|
|
// LMSW cannot clear PE
|
|
if (BX_CPU_THIS_PTR cr0.get_PE())
|
|
msw |= 0x0001; // adjust PE bit to current value of 1
|
|
|
|
msw &= 0x000f; // LMSW only affects last 4 flags
|
|
cr0 = (BX_CPU_THIS_PTR cr0.val32 & 0xfffffff0) | msw;
|
|
SetCR0(cr0);
|
|
}
|
|
|
|
void BX_CPU_C::SMSW_Ew(bxInstruction_c *i)
|
|
{
|
|
Bit16u msw;
|
|
|
|
#if BX_CPU_LEVEL == 2
|
|
msw = 0xfff0; /* 80286 init value */
|
|
msw |= BX_CPU_THIS_PTR cr0.val32 & 0x000f;
|
|
#else /* 386+ */
|
|
msw = BX_CPU_THIS_PTR cr0.val32 & 0xffff;
|
|
#endif
|
|
|
|
if (i->modC0()) {
|
|
if (i->os32L()) {
|
|
BX_WRITE_32BIT_REGZ(i->rm(), msw); // zeros out high 16bits
|
|
}
|
|
else {
|
|
BX_WRITE_16BIT_REG(i->rm(), msw);
|
|
}
|
|
}
|
|
else {
|
|
BX_CPU_CALL_METHODR(i->ResolveModrm, (i));
|
|
/* pointer, segment address pair */
|
|
write_virtual_word(i->seg(), RMAddr(i), msw);
|
|
}
|
|
}
|
|
#endif
|
|
|
|
void BX_CPU_C::MOV_TdRd(bxInstruction_c *i)
|
|
{
|
|
#if BX_CPU_LEVEL <= 4
|
|
BX_PANIC(("MOV_TdRd: Still not implemented"));
|
|
#else
|
|
// Pentium+ does not have TRx. They were redesigned using the MSRs.
|
|
BX_INFO(("MOV_TdRd: causes #UD"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
void BX_CPU_C::MOV_RdTd(bxInstruction_c *i)
|
|
{
|
|
#if BX_CPU_LEVEL <= 4
|
|
BX_PANIC(("MOV_RdTd: Still not implemented"));
|
|
#else
|
|
// Pentium+ does not have TRx. They were redesigned using the MSRs.
|
|
BX_INFO(("MOV_RdTd: causes #UD"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
#if BX_CPU_LEVEL == 2
|
|
void BX_CPU_C::LOADALL(bxInstruction_c *i)
|
|
{
|
|
Bit16u msw, tr, flags, ip, ldtr;
|
|
Bit16u ds_raw, ss_raw, cs_raw, es_raw;
|
|
Bit16u di, si, bp, sp, bx, dx, cx, ax;
|
|
Bit16u base_15_0, limit;
|
|
Bit8u base_23_16, access;
|
|
|
|
if (v8086_mode()) BX_PANIC(("proc_ctrl: LOADALL in v8086 mode unsupported"));
|
|
|
|
if (BX_CPU_THIS_PTR cr0.get_PE())
|
|
{
|
|
BX_PANIC(("LOADALL not yet supported for protected mode"));
|
|
}
|
|
|
|
BX_PANIC(("LOADALL: handle CR0.val32"));
|
|
/* MSW */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x806, 2, &msw);
|
|
BX_CPU_THIS_PTR cr0.set_PE(msw & 0x01); msw >>= 1;
|
|
BX_CPU_THIS_PTR cr0.set_MP(msw & 0x01); msw >>= 1;
|
|
BX_CPU_THIS_PTR cr0.set_EM(msw & 0x01); msw >>= 1;
|
|
BX_CPU_THIS_PTR cr0.set_TS(msw & 0x01);
|
|
|
|
if (BX_CPU_THIS_PTR cr0.get_PE() || BX_CPU_THIS_PTR cr0.get_MP() || BX_CPU_THIS_PTR cr0.get_EM() || BX_CPU_THIS_PTR cr0.get_TS())
|
|
BX_PANIC(("LOADALL set PE, MP, EM or TS bits in MSW!"));
|
|
|
|
/* TR */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x816, 2, &tr);
|
|
BX_CPU_THIS_PTR tr.selector.value = tr;
|
|
BX_CPU_THIS_PTR tr.selector.rpl = (tr & 0x03); tr >>= 2;
|
|
BX_CPU_THIS_PTR tr.selector.ti = (tr & 0x01); tr >>= 1;
|
|
BX_CPU_THIS_PTR tr.selector.index = tr;
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x860, 2, &base_15_0);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x862, 1, &base_23_16);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x863, 1, &access);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x864, 2, &limit);
|
|
|
|
BX_CPU_THIS_PTR tr.cache.valid =
|
|
BX_CPU_THIS_PTR tr.cache.p = (access & 0x80) >> 7;
|
|
BX_CPU_THIS_PTR tr.cache.dpl = (access & 0x60) >> 5;
|
|
BX_CPU_THIS_PTR tr.cache.segment = (access & 0x10) >> 4;
|
|
// don't allow busy bit in tr.cache.type, so bit 2 is masked away too.
|
|
BX_CPU_THIS_PTR tr.cache.type = (access & 0x0d);
|
|
BX_CPU_THIS_PTR tr.cache.u.system.base = (base_23_16 << 16) | base_15_0;
|
|
BX_CPU_THIS_PTR tr.cache.u.system.limit = limit;
|
|
|
|
if ((BX_CPU_THIS_PTR tr.selector.value & 0xfffc) == 0) {
|
|
BX_CPU_THIS_PTR tr.cache.valid = 0;
|
|
}
|
|
if (BX_CPU_THIS_PTR tr.cache.u.system.limit < 43 ||
|
|
BX_CPU_THIS_PTR tr.cache.type != BX_SYS_SEGMENT_AVAIL_286_TSS ||
|
|
BX_CPU_THIS_PTR tr.cache.segment)
|
|
{
|
|
BX_CPU_THIS_PTR tr.cache.valid = 0;
|
|
}
|
|
if (BX_CPU_THIS_PTR tr.cache.valid==0)
|
|
{
|
|
BX_CPU_THIS_PTR tr.selector.value = 0;
|
|
BX_CPU_THIS_PTR tr.selector.index = 0;
|
|
BX_CPU_THIS_PTR tr.selector.ti = 0;
|
|
BX_CPU_THIS_PTR tr.selector.rpl = 0;
|
|
BX_CPU_THIS_PTR tr.cache.u.system.base = 0;
|
|
BX_CPU_THIS_PTR tr.cache.u.system.limit = 0;
|
|
BX_CPU_THIS_PTR tr.cache.p = 0;
|
|
}
|
|
|
|
/* FLAGS */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x818, 2, &flags);
|
|
write_flags(flags, 1, 1);
|
|
|
|
/* IP */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x81a, 2, &ip);
|
|
IP = ip;
|
|
|
|
/* LDTR */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x81c, 2, &ldtr);
|
|
BX_CPU_THIS_PTR ldtr.selector.value = ldtr;
|
|
BX_CPU_THIS_PTR ldtr.selector.rpl = (ldtr & 0x03); ldtr >>= 2;
|
|
BX_CPU_THIS_PTR ldtr.selector.ti = (ldtr & 0x01); ldtr >>= 1;
|
|
BX_CPU_THIS_PTR ldtr.selector.index = ldtr;
|
|
if ((BX_CPU_THIS_PTR ldtr.selector.value & 0xfffc) == 0)
|
|
{
|
|
BX_CPU_THIS_PTR ldtr.cache.valid = 0;
|
|
BX_CPU_THIS_PTR ldtr.cache.p = 0;
|
|
BX_CPU_THIS_PTR ldtr.cache.segment = 0;
|
|
BX_CPU_THIS_PTR ldtr.cache.type = 0;
|
|
BX_CPU_THIS_PTR ldtr.cache.u.system.base = 0;
|
|
BX_CPU_THIS_PTR ldtr.cache.u.system.limit = 0;
|
|
BX_CPU_THIS_PTR ldtr.selector.value = 0;
|
|
BX_CPU_THIS_PTR ldtr.selector.index = 0;
|
|
BX_CPU_THIS_PTR ldtr.selector.ti = 0;
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x854, 2, &base_15_0);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x856, 1, &base_23_16);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x857, 1, &access);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x858, 2, &limit);
|
|
BX_CPU_THIS_PTR ldtr.cache.valid =
|
|
BX_CPU_THIS_PTR ldtr.cache.p = access >> 7;
|
|
BX_CPU_THIS_PTR ldtr.cache.dpl = (access >> 5) & 0x03;
|
|
BX_CPU_THIS_PTR ldtr.cache.segment = (access >> 4) & 0x01;
|
|
BX_CPU_THIS_PTR ldtr.cache.type = (access & 0x0f);
|
|
BX_CPU_THIS_PTR ldtr.cache.u.system.base = (base_23_16 << 16) | base_15_0;
|
|
BX_CPU_THIS_PTR ldtr.cache.u.system.limit = limit;
|
|
|
|
if (access == 0) {
|
|
BX_PANIC(("loadall: LDTR case access byte=0."));
|
|
}
|
|
if (BX_CPU_THIS_PTR ldtr.cache.valid==0) {
|
|
BX_PANIC(("loadall: ldtr.valid=0"));
|
|
}
|
|
if (BX_CPU_THIS_PTR ldtr.cache.segment) { /* not a system segment */
|
|
BX_INFO((" AR byte = %02x", (unsigned) access));
|
|
BX_PANIC(("loadall: LDTR descriptor cache loaded with non system segment"));
|
|
}
|
|
if (BX_CPU_THIS_PTR ldtr.cache.type != BX_SYS_SEGMENT_LDT) {
|
|
BX_PANIC(("loadall: LDTR.type(%u) != LDT", (unsigned) (access & 0x0f)));
|
|
}
|
|
}
|
|
|
|
/* DS */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x81e, 2, &ds_raw);
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].selector.value = ds_raw;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].selector.rpl = (ds_raw & 0x03); ds_raw >>= 2;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].selector.ti = (ds_raw & 0x01); ds_raw >>= 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].selector.index = ds_raw;
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x848, 2, &base_15_0);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x84a, 1, &base_23_16);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x84b, 1, &access);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x84c, 2, &limit);
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].cache.u.segment.base = (base_23_16 << 16) | base_15_0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].cache.u.segment.limit = limit;
|
|
set_ar_byte(BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].cache, access);
|
|
|
|
if ((BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].selector.value & 0xfffc) == 0) {
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].cache.valid = 0;
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].cache.valid = 1;
|
|
}
|
|
if (BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].cache.valid==0 ||
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].cache.segment==0)
|
|
{
|
|
BX_PANIC(("loadall: DS invalid"));
|
|
}
|
|
|
|
/* SS */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x820, 2, &ss_raw);
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector.value = ss_raw;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector.rpl = (ss_raw & 0x03); ss_raw >>= 2;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector.ti = (ss_raw & 0x01); ss_raw >>= 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector.index = ss_raw;
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x842, 2, &base_15_0);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x844, 1, &base_23_16);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x845, 1, &access);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x846, 2, &limit);
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.base = (base_23_16 << 16) | base_15_0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit = limit;
|
|
set_ar_byte(BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache, access);
|
|
|
|
if ((BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector.value & 0xfffc) == 0) {
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid = 0;
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid = 1;
|
|
}
|
|
if (BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid==0 ||
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.segment==0)
|
|
{
|
|
BX_PANIC(("loadall: SS invalid"));
|
|
}
|
|
|
|
/* CS */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x822, 2, &cs_raw);
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.value = cs_raw;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.rpl = (cs_raw & 0x03); cs_raw >>= 2;
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.ti = (cs_raw & 0x01); cs_raw >>= 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.index = cs_raw;
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x83c, 2, &base_15_0);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x83e, 1, &base_23_16);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x83f, 1, &access);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x840, 2, &limit);
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base = (base_23_16 << 16) | base_15_0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit = limit;
|
|
set_ar_byte(BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache, access);
|
|
|
|
if ((BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.value & 0xfffc) == 0) {
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 0;
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 1;
|
|
}
|
|
if (BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid==0 ||
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.segment==0)
|
|
{
|
|
BX_PANIC(("loadall: CS invalid"));
|
|
}
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
BX_CPU_THIS_PTR updateFetchModeMask();
|
|
#endif
|
|
|
|
/* ES */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x824, 2, &es_raw);
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].selector.value = es_raw;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].selector.rpl = (es_raw & 0x03); es_raw >>= 2;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].selector.ti = (es_raw & 0x01); es_raw >>= 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].selector.index = es_raw;
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x836, 2, &base_15_0);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x838, 1, &base_23_16);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x839, 1, &access);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x83a, 2, &limit);
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].cache.u.segment.base = (base_23_16 << 16) | base_15_0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].cache.u.segment.limit = limit;
|
|
set_ar_byte(BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].cache, access);
|
|
|
|
if ((BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].selector.value & 0xfffc) == 0) {
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].cache.valid = 0;
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].cache.valid = 1;
|
|
}
|
|
if (BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].cache.valid==0 ||
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].cache.segment==0)
|
|
{
|
|
BX_PANIC(("loadall: ES invalid"));
|
|
}
|
|
|
|
#if 0
|
|
BX_INFO(("cs.dpl = %02x", (unsigned) BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.dpl));
|
|
BX_INFO(("ss.dpl = %02x", (unsigned) BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.dpl));
|
|
BX_INFO(("BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].dpl = 0x%02x", (unsigned) BX_CPU_THIS_PTR ds.cache.dpl));
|
|
BX_INFO(("BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].dpl = 0x%02x", (unsigned) BX_CPU_THIS_PTR es.cache.dpl));
|
|
BX_INFO(("LOADALL: setting cs.selector.rpl to %u",
|
|
(unsigned) BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.rpl));
|
|
BX_INFO(("LOADALL: setting ss.selector.rpl to %u",
|
|
(unsigned) BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector.rpl));
|
|
BX_INFO(("LOADALL: setting ds.selector.rpl to %u",
|
|
(unsigned) BX_CPU_THIS_PTR sregs[BX_SEG_REG_DS].selector.rpl));
|
|
BX_INFO(("LOADALL: setting es.selector.rpl to %u",
|
|
(unsigned) BX_CPU_THIS_PTR sregs[BX_SEG_REG_ES].selector.rpl));
|
|
#endif
|
|
|
|
/* DI */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x826, 2, &di);
|
|
DI = di;
|
|
|
|
/* SI */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x828, 2, &si);
|
|
SI = si;
|
|
|
|
/* BP */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x82a, 2, &bp);
|
|
BP = bp;
|
|
|
|
/* SP */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x82c, 2, &sp);
|
|
SP = sp;
|
|
|
|
/* BX */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x82e, 2, &bx);
|
|
BX = bx;
|
|
|
|
/* DX */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x830, 2, &dx);
|
|
DX = dx;
|
|
|
|
/* CX */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x832, 2, &cx);
|
|
CX = cx;
|
|
|
|
/* AX */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x834, 2, &ax);
|
|
AX = ax;
|
|
|
|
/* GDTR */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x84e, 2, &base_15_0);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x850, 1, &base_23_16);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x851, 1, &access);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x852, 2, &limit);
|
|
BX_CPU_THIS_PTR gdtr.base = (base_23_16 << 16) | base_15_0;
|
|
BX_CPU_THIS_PTR gdtr.limit = limit;
|
|
|
|
#if 0
|
|
if (access)
|
|
BX_INFO(("LOADALL: GDTR access bits not 0 (%02x)", (unsigned) access));
|
|
#endif
|
|
|
|
/* IDTR */
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x85a, 2, &base_15_0);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x85c, 1, &base_23_16);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x85d, 1, &access);
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, 0x85e, 2, &limit);
|
|
BX_CPU_THIS_PTR idtr.base = (base_23_16 << 16) | base_15_0;
|
|
BX_CPU_THIS_PTR idtr.limit = limit;
|
|
}
|
|
#endif
|
|
|
|
void BX_CPU_C::handleCpuModeChange(void)
|
|
{
|
|
#if BX_SUPPORT_X86_64
|
|
if (BX_CPU_THIS_PTR efer.lma) {
|
|
if (! BX_CPU_THIS_PTR cr0.get_PE()) {
|
|
BX_PANIC(("change_cpu_mode: EFER.LMA is set when CR0.PE=0 !"));
|
|
}
|
|
if (BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.l) {
|
|
BX_CPU_THIS_PTR cpu_mode = BX_MODE_LONG_64;
|
|
BX_DEBUG(("Long Mode Activated"));
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR cpu_mode = BX_MODE_LONG_COMPAT;
|
|
if (BX_CPU_THIS_PTR eip_reg.dword.rip_upper != 0) {
|
|
BX_PANIC(("handleCpuModeChange: leaving long mode with RIP upper != 0 !"));
|
|
}
|
|
BX_DEBUG(("Compatibility Mode Activated"));
|
|
}
|
|
}
|
|
else
|
|
#endif
|
|
{
|
|
if (BX_CPU_THIS_PTR cr0.get_PE()) {
|
|
if (BX_CPU_THIS_PTR get_VM()) {
|
|
BX_CPU_THIS_PTR cpu_mode = BX_MODE_IA32_V8086;
|
|
BX_DEBUG(("VM8086 Mode Activated"));
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR cpu_mode = BX_MODE_IA32_PROTECTED;
|
|
BX_DEBUG(("Protected Mode Activated"));
|
|
}
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR cpu_mode = BX_MODE_IA32_REAL;
|
|
BX_DEBUG(("Real Mode Activated"));
|
|
}
|
|
}
|
|
}
|
|
|
|
#if BX_CPU_LEVEL >= 4 && BX_SUPPORT_ALIGNMENT_CHECK
|
|
void BX_CPU_C::handleAlignmentCheck(void)
|
|
{
|
|
if (CPL == 3 && BX_CPU_THIS_PTR cr0.get_AM() && BX_CPU_THIS_PTR get_AC()) {
|
|
BX_CPU_THIS_PTR alignment_check = 1;
|
|
BX_INFO(("Enable alignment check (#AC exception)"));
|
|
}
|
|
else {
|
|
BX_CPU_THIS_PTR alignment_check = 0;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
void BX_CPU_C::SetCR0(Bit32u val_32)
|
|
{
|
|
bx_bool pe = val_32 & 0x01;
|
|
bx_bool nw = (val_32 >> 29) & 0x01;
|
|
bx_bool cd = (val_32 >> 30) & 0x01;
|
|
bx_bool pg = (val_32 >> 31) & 0x01;
|
|
|
|
if (pg && !pe) {
|
|
BX_ERROR(("SetCR0: GP(0) when attempt to set CR0.PG with CR0.PE cleared !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
if (nw && !cd) {
|
|
BX_ERROR(("SetCR0: GP(0) when attempt to set CR0.NW with CR0.CD cleared !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
if (pe && BX_CPU_THIS_PTR get_VM()) BX_PANIC(("EFLAGS.VM=1, enter_PM"));
|
|
|
|
// from either MOV_CdRd() or debug functions
|
|
// protection checks made already or forcing from debug
|
|
Bit32u oldCR0 = BX_CPU_THIS_PTR cr0.val32;
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
bx_bool prev_pg = BX_CPU_THIS_PTR cr0.get_PG();
|
|
|
|
if (prev_pg==0 && pg) {
|
|
if (BX_CPU_THIS_PTR efer.lme) {
|
|
if (!BX_CPU_THIS_PTR cr4.get_PAE()) {
|
|
BX_ERROR(("SetCR0: attempt to enter x86-64 long mode without enabling CR4.PAE !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
if (BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.l) {
|
|
BX_ERROR(("SetCR0: attempt to enter x86-64 long mode with CS.L !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
BX_CPU_THIS_PTR efer.lma = 1;
|
|
}
|
|
}
|
|
else if (prev_pg==1 && ! pg) {
|
|
if (BX_CPU_THIS_PTR cpu_mode == BX_MODE_LONG_64) {
|
|
BX_ERROR(("SetCR0: attempt to leave 64 bit mode directly to legacy mode !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
if (BX_CPU_THIS_PTR efer.lma) {
|
|
if (BX_CPU_THIS_PTR eip_reg.dword.rip_upper != 0) {
|
|
BX_PANIC(("SetCR0: attempt to leave x86-64 LONG mode with RIP upper != 0 !!!"));
|
|
}
|
|
BX_CPU_THIS_PTR efer.lma = 0;
|
|
}
|
|
}
|
|
#endif // #if BX_SUPPORT_X86_64
|
|
|
|
// handle reserved bits behaviour
|
|
#if BX_CPU_LEVEL == 3
|
|
val_32 = val_32 | 0x7ffffff0;
|
|
#elif BX_CPU_LEVEL == 4
|
|
val_32 = (val_32 | 0x00000010) & 0xe005003f;
|
|
#elif BX_CPU_LEVEL == 5
|
|
val_32 = val_32 | 0x00000010;
|
|
#elif BX_CPU_LEVEL == 6
|
|
val_32 = (val_32 | 0x00000010) & 0xe005003f;
|
|
#else
|
|
#error "SetCR0: implement reserved bits behaviour for this CPU_LEVEL"
|
|
#endif
|
|
BX_CPU_THIS_PTR cr0.val32 = val_32;
|
|
|
|
#if BX_CPU_LEVEL >= 4 && BX_SUPPORT_ALIGNMENT_CHECK
|
|
handleAlignmentCheck();
|
|
#endif
|
|
|
|
handleCpuModeChange();
|
|
|
|
// Give the paging unit a chance to look for changes in bits
|
|
// it cares about, like {PG,PE}, so it can flush cache entries etc.
|
|
pagingCR0Changed(oldCR0, val_32);
|
|
}
|
|
|
|
#if BX_CPU_LEVEL >= 4
|
|
bx_bool BX_CPU_C::SetCR4(Bit32u val_32)
|
|
{
|
|
Bit32u oldCR4 = BX_CPU_THIS_PTR cr4.getRegister();
|
|
Bit32u allowMask = 0;
|
|
|
|
// CR4 bit definitions from AMD Hammer manual:
|
|
// [63-11] Reserved, Must be Zero
|
|
// [10] OSXMMEXCPT: Operating System Unmasked Exception Support R/W
|
|
// [9] OSFXSR: Operating System FXSAVE/FXRSTOR Support R/W
|
|
// [8] PCE: Performance-Monitoring Counter Enable R/W
|
|
// [7] PGE: Page-Global Enable R/W
|
|
// [6] MCE: Machine Check Enable R/W
|
|
// [5] PAE: Physical-Address Extension R/W
|
|
// [4] PSE: Page Size Extensions R/W
|
|
// [3] DE: Debugging Extensions R/W
|
|
// [2] TSD: Time Stamp Disable R/W
|
|
// [1] PVI: Protected-Mode Virtual Interrupts R/W
|
|
// [0] VME: Virtual-8086 Mode Extensions R/W
|
|
|
|
#if BX_SUPPORT_VME
|
|
allowMask |= (1<<0) | (1<<1); /* VME */
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 5
|
|
allowMask |= (1<<2); /* TSD */
|
|
#endif
|
|
|
|
allowMask |= (1<<3); /* DE */
|
|
|
|
#if BX_SUPPORT_LARGE_PAGES
|
|
allowMask |= (1<<4);
|
|
#endif
|
|
|
|
#if BX_SUPPORT_PAE
|
|
allowMask |= (1<<5);
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 5
|
|
// NOTE: exception 18 never appears in Bochs
|
|
allowMask |= (1<<6); /* MCE */
|
|
#endif
|
|
|
|
#if BX_SUPPORT_GLOBAL_PAGES
|
|
allowMask |= (1<<7);
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 6
|
|
allowMask |= (1<<8); /* PCE */
|
|
allowMask |= (1<<9); /* OSFXSR */
|
|
#endif
|
|
|
|
#if BX_SUPPORT_SSE
|
|
allowMask |= (1<<10); /* OSXMMECPT */
|
|
#endif
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
// need to GP(0) if LMA=1 and PAE=1->0
|
|
if ((BX_CPU_THIS_PTR efer.lma)
|
|
&& (!(val_32 >> 5) & 1)
|
|
&& (BX_CPU_THIS_PTR cr4.get_PAE()))
|
|
{
|
|
BX_ERROR(("SetCR4: attempt to change PAE when EFER.LMA=1"));
|
|
return 0;
|
|
}
|
|
#endif
|
|
|
|
// Need to GPF if trying to set undefined bits.
|
|
if (val_32 & ~allowMask) {
|
|
BX_ERROR(("#GP(0): SetCR4: Write of 0x%08x not supported (allowMask=0x%x)", val_32, allowMask));
|
|
return 0;
|
|
}
|
|
|
|
val_32 &= allowMask; // Screen out unsupported bits. (not needed, for good measure)
|
|
BX_CPU_THIS_PTR cr4.setRegister(val_32);
|
|
pagingCR4Changed(oldCR4, BX_CPU_THIS_PTR cr4.getRegister());
|
|
return 1;
|
|
}
|
|
#endif
|
|
|
|
void BX_CPU_C::RDPMC(bxInstruction_c *i)
|
|
{
|
|
/* We need to be Pentium with MMX or later */
|
|
#if ((BX_CPU_LEVEL >= 6) || (BX_SUPPORT_MMX && BX_CPU_LEVEL == 5))
|
|
bx_bool pce = BX_CPU_THIS_PTR cr4.get_PCE();
|
|
|
|
if ((pce==1) || (CPL==0) || real_mode())
|
|
{
|
|
/* According to manual, Pentium 4 has 18 counters,
|
|
* previous versions have two. And the P4 also can do
|
|
* short read-out (EDX always 0). Otherwise it is
|
|
* limited to 40 bits.
|
|
*/
|
|
|
|
#if (BX_CPU_LEVEL == 6 && BX_SUPPORT_SSE >= 2) // Pentium 4 processor (see cpuid.cc)
|
|
if ((ECX & 0x7fffffff) >= 18)
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
#else //
|
|
if ((ECX & 0xffffffff) >= 2)
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
#endif
|
|
// Most counters are for hardware specific details, which
|
|
// we anyhow do not emulate (like pipeline stalls etc)
|
|
|
|
// Could be interesting to count number of memory reads,
|
|
// writes. Misaligned etc... But to monitor bochs, this
|
|
// is easier done from the host.
|
|
|
|
RAX = 0;
|
|
RDX = 0; // if P4 and ECX & 0x10000000, then always 0 (short read 32 bits)
|
|
|
|
BX_ERROR(("RDPMC: Performance Counters Support not reasonably implemented yet"));
|
|
} else {
|
|
// not allowed to use RDPMC!
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
#else
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
#if BX_CPU_LEVEL >= 5
|
|
Bit64u BX_CPU_C::get_TSC ()
|
|
{
|
|
return bx_pc_system.time_ticks() - BX_CPU_THIS_PTR msr.tsc_last_reset;
|
|
}
|
|
|
|
void BX_CPU_C::set_TSC (Bit32u newval)
|
|
{
|
|
// compute the correct setting of tsc_last_reset so that a get_TSC()
|
|
// will return newval
|
|
BX_CPU_THIS_PTR msr.tsc_last_reset =
|
|
bx_pc_system.time_ticks() - (Bit64u) newval;
|
|
|
|
// verify
|
|
BX_ASSERT (get_TSC() == (Bit64u) newval);
|
|
}
|
|
#endif
|
|
|
|
void BX_CPU_C::RDTSC(bxInstruction_c *i)
|
|
{
|
|
#if BX_CPU_LEVEL >= 5
|
|
bx_bool tsd = BX_CPU_THIS_PTR cr4.get_TSD();
|
|
if ((tsd==0) || (tsd==1 && CPL==0)) {
|
|
// return ticks
|
|
Bit64u ticks = BX_CPU_THIS_PTR get_TSC();
|
|
RAX = (Bit32u) (ticks & 0xffffffff);
|
|
RDX = (Bit32u) ((ticks >> 32) & 0xffffffff);
|
|
} else {
|
|
// not allowed to use RDTSC!
|
|
BX_ERROR(("RDTSC: incorrect usage of RDTSC instruction !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
#else
|
|
BX_INFO(("RDTSC: Pentium CPU required, use --enable-cpu=5"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
void BX_CPU_C::RDTSCP(bxInstruction_c *i)
|
|
{
|
|
RDTSC(i);
|
|
RCX = MSR_TSC_AUX;
|
|
}
|
|
#endif
|
|
|
|
void BX_CPU_C::RDMSR(bxInstruction_c *i)
|
|
{
|
|
#if BX_CPU_LEVEL >= 5
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("RDMSR: CPL!=0 not in real mode"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
/* We have the requested MSR register in ECX */
|
|
switch(ECX) {
|
|
|
|
#if BX_SUPPORT_SEP
|
|
case BX_MSR_SYSENTER_CS:
|
|
RAX = BX_CPU_THIS_PTR msr.sysenter_cs_msr;
|
|
RDX = 0;
|
|
return;
|
|
|
|
case BX_MSR_SYSENTER_ESP:
|
|
RAX = BX_CPU_THIS_PTR msr.sysenter_esp_msr;
|
|
RDX = 0;
|
|
return;
|
|
|
|
case BX_MSR_SYSENTER_EIP:
|
|
RAX = BX_CPU_THIS_PTR msr.sysenter_eip_msr;
|
|
RDX = 0;
|
|
return;
|
|
#endif
|
|
|
|
#if BX_SUPPORT_MTRR
|
|
case BX_MSR_MTRRCAP: // read only MSR
|
|
RAX = 0x508;
|
|
RDX = 0;
|
|
return;
|
|
|
|
case BX_MSR_MTRRPHYSBASE0:
|
|
case BX_MSR_MTRRPHYSMASK0:
|
|
case BX_MSR_MTRRPHYSBASE1:
|
|
case BX_MSR_MTRRPHYSMASK1:
|
|
case BX_MSR_MTRRPHYSBASE2:
|
|
case BX_MSR_MTRRPHYSMASK2:
|
|
case BX_MSR_MTRRPHYSBASE3:
|
|
case BX_MSR_MTRRPHYSMASK3:
|
|
case BX_MSR_MTRRPHYSBASE4:
|
|
case BX_MSR_MTRRPHYSMASK4:
|
|
case BX_MSR_MTRRPHYSBASE5:
|
|
case BX_MSR_MTRRPHYSMASK5:
|
|
case BX_MSR_MTRRPHYSBASE6:
|
|
case BX_MSR_MTRRPHYSMASK6:
|
|
case BX_MSR_MTRRPHYSBASE7:
|
|
case BX_MSR_MTRRPHYSMASK7:
|
|
RAX = BX_CPU_THIS_PTR msr.mtrrphys[ECX - BX_MSR_MTRRPHYSBASE0] & 0xffffffff;
|
|
RDX = BX_CPU_THIS_PTR msr.mtrrphys[ECX - BX_MSR_MTRRPHYSBASE0] >> 32;
|
|
return;
|
|
|
|
case BX_MSR_MTRRFIX64K_00000:
|
|
RAX = BX_CPU_THIS_PTR msr.mtrrfix64k_00000 & 0xffffffff;
|
|
RDX = BX_CPU_THIS_PTR msr.mtrrfix64k_00000 >> 32;
|
|
return;
|
|
case BX_MSR_MTRRFIX16K_80000:
|
|
RAX = BX_CPU_THIS_PTR msr.mtrrfix16k_80000 & 0xffffffff;
|
|
RDX = BX_CPU_THIS_PTR msr.mtrrfix16k_80000 >> 32;
|
|
return;
|
|
case BX_MSR_MTRRFIX16K_A0000:
|
|
RAX = BX_CPU_THIS_PTR msr.mtrrfix16k_a0000 & 0xffffffff;
|
|
RAX = BX_CPU_THIS_PTR msr.mtrrfix16k_a0000 >> 32;
|
|
return;
|
|
|
|
case BX_MSR_MTRRFIX4K_C0000:
|
|
case BX_MSR_MTRRFIX4K_C8000:
|
|
case BX_MSR_MTRRFIX4K_D0000:
|
|
case BX_MSR_MTRRFIX4K_D8000:
|
|
case BX_MSR_MTRRFIX4K_E0000:
|
|
case BX_MSR_MTRRFIX4K_E8000:
|
|
case BX_MSR_MTRRFIX4K_F0000:
|
|
case BX_MSR_MTRRFIX4K_F8000:
|
|
RAX = BX_CPU_THIS_PTR msr.mtrrfix4k[ECX - BX_MSR_MTRRFIX4K_C0000] & 0xffffffff;
|
|
RDX = BX_CPU_THIS_PTR msr.mtrrfix4k[ECX - BX_MSR_MTRRFIX4K_C0000] >> 32;
|
|
return;
|
|
|
|
case BX_MSR_PAT:
|
|
RAX = BX_CPU_THIS_PTR msr.pat & 0xffffffff;
|
|
RDX = BX_CPU_THIS_PTR msr.pat >> 32;
|
|
return;
|
|
|
|
case BX_MSR_MTRR_DEFTYPE:
|
|
RAX = BX_CPU_THIS_PTR msr.mtrr_deftype;
|
|
RDX = 0;
|
|
return;
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL == 5
|
|
/* The following registers are defined for Pentium only */
|
|
case BX_MSR_P5_MC_ADDR:
|
|
case BX_MSR_MC_TYPE:
|
|
/* TODO */
|
|
return;
|
|
|
|
case BX_MSR_CESR:
|
|
/* TODO */
|
|
return;
|
|
#else
|
|
/* These are noops on i686... */
|
|
case BX_MSR_P5_MC_ADDR:
|
|
case BX_MSR_MC_TYPE:
|
|
/* do nothing */
|
|
return;
|
|
|
|
/* ... And these cause an exception on i686 */
|
|
case BX_MSR_CESR:
|
|
case BX_MSR_CTR0:
|
|
case BX_MSR_CTR1:
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
#endif /* BX_CPU_LEVEL == 5 */
|
|
|
|
case BX_MSR_TSC:
|
|
RDTSC(i);
|
|
return;
|
|
|
|
/* MSR_APICBASE
|
|
0:7 Reserved
|
|
8 This is set if its the BSP
|
|
9:10 Reserved
|
|
11 APIC Global Enable bit (1=enabled 0=disabled)
|
|
12:35 APIC Base Address
|
|
36:63 Reserved
|
|
*/
|
|
#if BX_SUPPORT_APIC
|
|
case BX_MSR_APICBASE:
|
|
RAX = BX_CPU_THIS_PTR msr.apicbase;
|
|
RDX = 0;
|
|
BX_INFO(("RDMSR: Read %08x:%08x from MSR_APICBASE", EDX, EAX));
|
|
return;
|
|
#endif
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
case BX_MSR_EFER:
|
|
RAX = (Bit64u) BX_CPU_THIS_PTR get_EFER();
|
|
RDX = 0;
|
|
return;
|
|
|
|
case BX_MSR_STAR:
|
|
RAX = MSR_STAR & 0xffffffff;
|
|
RDX = MSR_STAR >> 32;
|
|
return;
|
|
|
|
case BX_MSR_LSTAR:
|
|
RAX = MSR_LSTAR & 0xffffffff;
|
|
RDX = MSR_LSTAR >> 32;
|
|
return;
|
|
|
|
case BX_MSR_CSTAR:
|
|
RAX = MSR_CSTAR & 0xffffffff;
|
|
RDX = MSR_CSTAR >> 32;
|
|
return;
|
|
|
|
case BX_MSR_FMASK:
|
|
RAX = MSR_FMASK & 0xffffffff;
|
|
RDX = MSR_FMASK >> 32;
|
|
return;
|
|
|
|
case BX_MSR_FSBASE:
|
|
RAX = MSR_FSBASE & 0xffffffff;
|
|
RDX = MSR_FSBASE >> 32;
|
|
return;
|
|
|
|
case BX_MSR_GSBASE:
|
|
RAX = MSR_GSBASE & 0xffffffff;
|
|
RDX = MSR_GSBASE >> 32;
|
|
return;
|
|
|
|
case BX_MSR_KERNELGSBASE:
|
|
RAX = MSR_KERNELGSBASE & 0xffffffff;
|
|
RDX = MSR_KERNELGSBASE >> 32;
|
|
return;
|
|
|
|
case BX_MSR_TSC_AUX:
|
|
RAX = MSR_TSC_AUX; // 32 bit MSR
|
|
RDX = 0;
|
|
return;
|
|
#endif // #if BX_SUPPORT_X86_64
|
|
|
|
default:
|
|
BX_ERROR(("RDMSR: Unknown register %#x", ECX));
|
|
#if BX_IGNORE_BAD_MSR
|
|
RAX = 0;
|
|
RDX = 0;
|
|
return;
|
|
#endif
|
|
}
|
|
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
|
|
#else /* BX_CPU_LEVEL >= 5 */
|
|
BX_INFO(("RDMSR: Pentium CPU required, use --enable-cpu-level=5"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
void BX_CPU_C::WRMSR(bxInstruction_c *i)
|
|
{
|
|
#if BX_CPU_LEVEL >= 5
|
|
if (!real_mode() && CPL!=0) {
|
|
BX_ERROR(("WRMSR: CPL!=0 not in real mode"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
BX_INSTR_WRMSR(BX_CPU_ID, ECX, ((Bit64u) EDX << 32) + EAX);
|
|
|
|
/* ECX has the MSR to write to */
|
|
switch(ECX) {
|
|
|
|
#if BX_SUPPORT_SEP
|
|
case BX_MSR_SYSENTER_CS: {
|
|
BX_CPU_THIS_PTR msr.sysenter_cs_msr = EAX;
|
|
return;
|
|
}
|
|
case BX_MSR_SYSENTER_ESP:
|
|
BX_CPU_THIS_PTR msr.sysenter_esp_msr = EAX;
|
|
return;
|
|
case BX_MSR_SYSENTER_EIP:
|
|
BX_CPU_THIS_PTR msr.sysenter_eip_msr = EAX;
|
|
return;
|
|
#endif
|
|
|
|
#if BX_SUPPORT_MTRR
|
|
case BX_MSR_MTRRCAP:
|
|
BX_ERROR(("WRMSR: MTRRCAP is read only MSR"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
|
|
case BX_MSR_MTRRPHYSBASE0:
|
|
case BX_MSR_MTRRPHYSMASK0:
|
|
case BX_MSR_MTRRPHYSBASE1:
|
|
case BX_MSR_MTRRPHYSMASK1:
|
|
case BX_MSR_MTRRPHYSBASE2:
|
|
case BX_MSR_MTRRPHYSMASK2:
|
|
case BX_MSR_MTRRPHYSBASE3:
|
|
case BX_MSR_MTRRPHYSMASK3:
|
|
case BX_MSR_MTRRPHYSBASE4:
|
|
case BX_MSR_MTRRPHYSMASK4:
|
|
case BX_MSR_MTRRPHYSBASE5:
|
|
case BX_MSR_MTRRPHYSMASK5:
|
|
case BX_MSR_MTRRPHYSBASE6:
|
|
case BX_MSR_MTRRPHYSMASK6:
|
|
case BX_MSR_MTRRPHYSBASE7:
|
|
case BX_MSR_MTRRPHYSMASK7:
|
|
BX_CPU_THIS_PTR msr.mtrrphys[ECX - BX_MSR_MTRRPHYSBASE0] = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_MTRRFIX64K_00000:
|
|
BX_CPU_THIS_PTR msr.mtrrfix64k_00000 = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
case BX_MSR_MTRRFIX16K_80000:
|
|
BX_CPU_THIS_PTR msr.mtrrfix16k_80000 = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
case BX_MSR_MTRRFIX16K_A0000:
|
|
BX_CPU_THIS_PTR msr.mtrrfix16k_a0000 = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_MTRRFIX4K_C0000:
|
|
case BX_MSR_MTRRFIX4K_C8000:
|
|
case BX_MSR_MTRRFIX4K_D0000:
|
|
case BX_MSR_MTRRFIX4K_D8000:
|
|
case BX_MSR_MTRRFIX4K_E0000:
|
|
case BX_MSR_MTRRFIX4K_E8000:
|
|
case BX_MSR_MTRRFIX4K_F0000:
|
|
case BX_MSR_MTRRFIX4K_F8000:
|
|
BX_CPU_THIS_PTR msr.mtrrfix4k[ECX - BX_MSR_MTRRFIX4K_C0000] = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_PAT:
|
|
BX_CPU_THIS_PTR msr.pat = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_MTRR_DEFTYPE:
|
|
BX_CPU_THIS_PTR msr.mtrr_deftype = EAX;
|
|
return;
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL == 5
|
|
/* The following registers are defined for Pentium only */
|
|
case BX_MSR_P5_MC_ADDR:
|
|
case BX_MSR_MC_TYPE:
|
|
case BX_MSR_CESR:
|
|
/* TODO */
|
|
return;
|
|
#else
|
|
/* These are noops on i686... */
|
|
case BX_MSR_P5_MC_ADDR:
|
|
case BX_MSR_MC_TYPE:
|
|
/* do nothing */
|
|
return;
|
|
|
|
/* ... And these cause an exception on i686 */
|
|
case BX_MSR_CESR:
|
|
case BX_MSR_CTR0:
|
|
case BX_MSR_CTR1:
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
#endif /* BX_CPU_LEVEL == 5 */
|
|
|
|
case BX_MSR_TSC:
|
|
BX_CPU_THIS_PTR set_TSC(EAX); /* ignore the high 32bits */
|
|
BX_INFO(("WRMSR: wrote 0x%08x to MSR_TSC", EAX));
|
|
return;
|
|
|
|
/* MSR_APICBASE
|
|
0:7 Reserved
|
|
8 This is set if its the BSP
|
|
9:10 Reserved
|
|
11 APIC Global Enable bit (1=enabled 0=disabled)
|
|
12:35 APIC Base Address (in Bochs 12:31 because of 32-bit physical addr)
|
|
36:63 Reserved
|
|
*/
|
|
#if BX_SUPPORT_APIC
|
|
case BX_MSR_APICBASE:
|
|
if (BX_CPU_THIS_PTR msr.apicbase & 0x800) {
|
|
BX_INFO(("WRMSR: wrote %08x:%08x to MSR_APICBASE", EDX, EAX));
|
|
BX_CPU_THIS_PTR msr.apicbase = EAX; /* ignore the high 32bits */
|
|
if (EDX != 0) {
|
|
BX_PANIC(("MSR_APICBASE: Only 32 bit physical address space is emulated !"));
|
|
}
|
|
BX_CPU_THIS_PTR local_apic.set_base(BX_CPU_THIS_PTR msr.apicbase);
|
|
// TLB flush is required for emulation correctness
|
|
TLB_flush(1); // don't care about performance of apic relocation
|
|
}
|
|
else {
|
|
BX_INFO(("WRMSR: MSR_APICBASE APIC global enable bit cleared !"));
|
|
}
|
|
return;
|
|
#endif
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
case BX_MSR_EFER:
|
|
// #GP(0) if changing EFER.LME when cr0.pg = 1
|
|
if ((BX_CPU_THIS_PTR efer.lme != ((EAX >> 8) & 1)) &&
|
|
BX_CPU_THIS_PTR cr0.get_PG())
|
|
{
|
|
BX_ERROR(("WRMSR: attempt to change LME when CR0.PG=1"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
BX_CPU_THIS_PTR efer.sce = (EAX >> 0) & 1;
|
|
BX_CPU_THIS_PTR efer.lme = (EAX >> 8) & 1;
|
|
BX_CPU_THIS_PTR efer.nxe = (EAX >> 11) & 1;
|
|
BX_CPU_THIS_PTR efer.ffxsr = (EAX >> 14) & 1;
|
|
return;
|
|
|
|
case BX_MSR_STAR:
|
|
MSR_STAR = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_LSTAR:
|
|
MSR_LSTAR = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_CSTAR:
|
|
MSR_CSTAR = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_FMASK:
|
|
MSR_FMASK = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_FSBASE:
|
|
MSR_FSBASE = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_GSBASE:
|
|
MSR_GSBASE = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_KERNELGSBASE:
|
|
MSR_KERNELGSBASE = ((Bit64u) EDX << 32) + EAX;
|
|
return;
|
|
|
|
case BX_MSR_TSC_AUX:
|
|
MSR_TSC_AUX = EAX;
|
|
return;
|
|
#endif // #if BX_SUPPORT_X86_64
|
|
|
|
default:
|
|
BX_ERROR(("WRMSR: Unknown register %#x", ECX));
|
|
#if BX_IGNORE_BAD_MSR
|
|
return;
|
|
#endif
|
|
}
|
|
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
|
|
#else /* BX_CPU_LEVEL >= 5 */
|
|
BX_INFO(("WRMSR: Pentium CPU required, use --enable-cpu-level=5"));
|
|
UndefinedOpcode(i);
|
|
#endif
|
|
}
|
|
|
|
#if BX_SUPPORT_MONITOR_MWAIT
|
|
bx_bool BX_CPU_C::is_monitor(bx_phy_address begin_addr, unsigned len)
|
|
{
|
|
bx_phy_address end_addr = begin_addr + len;
|
|
if (begin_addr >= BX_CPU_THIS_PTR monitor.monitor_end || end_addr <= BX_CPU_THIS_PTR monitor.monitor_begin)
|
|
return 0;
|
|
else
|
|
return 1;
|
|
}
|
|
|
|
void BX_CPU_C::check_monitor(bx_phy_address begin_addr, unsigned len)
|
|
{
|
|
if (is_monitor(begin_addr, len)) {
|
|
// wakeup from MWAIT state
|
|
BX_ASSERT(BX_CPU_THIS_PTR debug_trap & BX_DEBUG_TRAP_MWAIT);
|
|
BX_CPU_THIS_PTR debug_trap &= ~BX_DEBUG_TRAP_SPECIAL;
|
|
// clear monitor
|
|
BX_CPU_THIS_PTR mem->clear_monitor(BX_CPU_THIS_PTR bx_cpuid);
|
|
}
|
|
}
|
|
#endif
|
|
|
|
void BX_CPU_C::MONITOR(bxInstruction_c *i)
|
|
{
|
|
#if BX_SUPPORT_MONITOR_MWAIT
|
|
// TODO: #UD when CPL > 0 and
|
|
// MSR 0xC0010015[MONITOR_MWAIT_USER_UNABLE] = 1
|
|
BX_DEBUG(("MONITOR instruction executed EAX = 0x08x", (unsigned) EAX));
|
|
|
|
if (RCX != 0) {
|
|
BX_ERROR(("MONITOR: no optional extensions supported"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
bx_address addr, laddr;
|
|
bx_phy_address paddr;
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
if (i->as64L()) {
|
|
addr = RAX;
|
|
}
|
|
else
|
|
#endif
|
|
if (i->as32L()) {
|
|
laddr = EAX;
|
|
}
|
|
else {
|
|
addr = AX;
|
|
}
|
|
|
|
read_virtual_checks(&BX_CPU_THIS_PTR sregs[i->seg()], addr, 1);
|
|
|
|
// set MONITOR
|
|
laddr = BX_CPU_THIS_PTR get_segment_base(i->seg()) + addr;
|
|
|
|
if (BX_CPU_THIS_PTR cr0.get_PG()) {
|
|
paddr = dtranslate_linear(laddr, CPL, BX_READ);
|
|
paddr = A20ADDR(paddr);
|
|
}
|
|
else
|
|
{
|
|
paddr = A20ADDR(laddr);
|
|
}
|
|
|
|
BX_CPU_THIS_PTR monitor.monitor_begin = paddr;
|
|
BX_CPU_THIS_PTR monitor.monitor_end = paddr + CACHE_LINE_SIZE;
|
|
|
|
#else
|
|
BX_INFO(("MONITOR: use --enable-monitor-mwait to enable MONITOR/MWAIT support"));
|
|
UndefinedOpcode (i);
|
|
#endif
|
|
}
|
|
|
|
void BX_CPU_C::MWAIT(bxInstruction_c *i)
|
|
{
|
|
#if BX_SUPPORT_MONITOR_MWAIT
|
|
// TODO: #UD when CPL > 0 and
|
|
// MSR 0xC0010015[MONITOR_MWAIT_USER_UNABLE] = 1
|
|
BX_DEBUG(("MWAIT instruction executed ECX = 0x%08x", ECX));
|
|
|
|
// only one extension is supported
|
|
// ECX[0] - interrupt MWAIT even if EFLAGS.IF = 0
|
|
if (RCX & ~(BX_CONST64(1))) {
|
|
BX_ERROR(("MWAIT: incorrect optional extensions in RCX"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
// Do not enter optimized state if MONITOR wasn't properly set
|
|
if (BX_CPU_THIS_PTR monitor.monitor_begin == BX_CPU_THIS_PTR monitor.monitor_end) {
|
|
BX_ERROR(("MWAIT: incorrect MONITOR settings"));
|
|
return;
|
|
}
|
|
|
|
bx_pc_system.invlpg(BX_CPU_THIS_PTR monitor.monitor_begin);
|
|
if ((BX_CPU_THIS_PTR monitor.monitor_end & ~0xfff) != (BX_CPU_THIS_PTR monitor.monitor_begin & ~0xfff))
|
|
bx_pc_system.invlpg(BX_CPU_THIS_PTR monitor.monitor_end);
|
|
BX_DEBUG(("MWAIT for phys_addr=%08x", BX_CPU_THIS_PTR monitor.monitor_begin));
|
|
BX_CPU_THIS_PTR mem->set_monitor(BX_CPU_THIS_PTR bx_cpuid);
|
|
|
|
// stops instruction execution and places the processor in a optimized
|
|
// state. Events that cause exit from MWAIT state are:
|
|
// A store from another processor to monitored range, any unmasked
|
|
// interrupt, including INTR, NMI, SMI, INIT or reset will resume
|
|
// the execution. Any far control transfer between MONITOR and MWAIT
|
|
// resets the monitoring logic.
|
|
|
|
// artificial trap bit, why use another variable.
|
|
BX_CPU_THIS_PTR debug_trap |= BX_DEBUG_TRAP_MWAIT; // artificial trap
|
|
if (ECX & 1)
|
|
BX_CPU_THIS_PTR debug_trap |= BX_DEBUG_TRAP_MWAIT_IF;
|
|
BX_CPU_THIS_PTR async_event = 1; // so processor knows to check
|
|
// Execution of this instruction completes. The processor
|
|
// will remain in a optimized state until one of the above
|
|
// conditions is met.
|
|
|
|
BX_INSTR_MWAIT(BX_CPU_ID, BX_CPU_THIS_PTR monitor.monitor_begin, CACHE_LINE_SIZE, ECX);
|
|
|
|
#if BX_USE_IDLE_HACK
|
|
bx_gui->sim_is_idle();
|
|
#endif
|
|
|
|
#else
|
|
BX_INFO(("MWAIT: use --enable-monitor-mwait to enable MONITOR/MWAIT support"));
|
|
UndefinedOpcode (i);
|
|
#endif
|
|
}
|
|
|
|
void BX_CPU_C::SYSENTER(bxInstruction_c *i)
|
|
{
|
|
#if BX_SUPPORT_SEP
|
|
if (!protected_mode()) {
|
|
BX_ERROR(("SYSENTER not from protected mode !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
if ((BX_CPU_THIS_PTR msr.sysenter_cs_msr & BX_SELECTOR_RPL_MASK) == 0) {
|
|
BX_ERROR(("SYSENTER with zero sysenter_cs_msr !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
BX_CPU_THIS_PTR clear_VM(); // do this just like the book says to do
|
|
BX_CPU_THIS_PTR clear_IF();
|
|
BX_CPU_THIS_PTR clear_RF();
|
|
|
|
parse_selector(BX_CPU_THIS_PTR msr.sysenter_cs_msr & BX_SELECTOR_RPL_MASK,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.dpl = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.type = BX_CODE_EXEC_READ_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base = 0; // base address
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit = 0xFFFF; // segment limit
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit_scaled = 0xFFFFFFFF; // scaled segment limit
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.g = 1; // 4k granularity
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b = 1; // 32-bit mode
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.avl = 0; // available for use by system
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
BX_CPU_THIS_PTR updateFetchModeMask();
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 4 && BX_SUPPORT_ALIGNMENT_CHECK
|
|
BX_CPU_THIS_PTR alignment_check = 0; // CPL=0
|
|
#endif
|
|
|
|
parse_selector((BX_CPU_THIS_PTR msr.sysenter_cs_msr + 8) & BX_SELECTOR_RPL_MASK,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.dpl = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.type = BX_DATA_READ_WRITE_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.base = 0; // base address
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit = 0xFFFF; // segment limit
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit_scaled = 0xFFFFFFFF; // scaled segment limit
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.g = 1; // 4k granularity
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.d_b = 1; // 32-bit mode
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.avl = 0; // available for use by system
|
|
|
|
ESP = BX_CPU_THIS_PTR msr.sysenter_esp_msr;
|
|
EIP = BX_CPU_THIS_PTR msr.sysenter_eip_msr;
|
|
|
|
BX_INSTR_FAR_BRANCH(BX_CPU_ID, BX_INSTR_IS_SYSENTER,
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.value, EIP);
|
|
#else
|
|
BX_INFO(("SYSENTER: use --enable-sep to enable SYSENTER/SYSEXIT support"));
|
|
UndefinedOpcode (i);
|
|
#endif
|
|
}
|
|
|
|
void BX_CPU_C::SYSEXIT(bxInstruction_c *i)
|
|
{
|
|
#if BX_SUPPORT_SEP
|
|
if (!protected_mode()) {
|
|
BX_ERROR(("SYSEXIT not from protected mode !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
if ((BX_CPU_THIS_PTR msr.sysenter_cs_msr & BX_SELECTOR_RPL_MASK) == 0) {
|
|
BX_ERROR(("SYSEXIT with zero sysenter_cs_msr !"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
if (CPL != 0) {
|
|
BX_ERROR(("SYSEXIT at non-zero cpl %u !", CPL));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
parse_selector((BX_CPU_THIS_PTR msr.sysenter_cs_msr + 16) | 3,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.dpl = 3;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.type = BX_CODE_EXEC_READ_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base = 0; // base address
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit = 0xFFFF; // segment limit
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit_scaled = 0xFFFFFFFF; // scaled segment limit
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.g = 1; // 4k granularity
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b = 1; // 32-bit mode
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.avl = 0; // available for use by system
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
BX_CPU_THIS_PTR updateFetchModeMask();
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 4 && BX_SUPPORT_ALIGNMENT_CHECK
|
|
handleAlignmentCheck(); // CPL was modified
|
|
#endif
|
|
|
|
parse_selector((BX_CPU_THIS_PTR msr.sysenter_cs_msr + 24) | 3,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.dpl = 3;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.type = BX_DATA_READ_WRITE_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.base = 0; // base address
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit = 0xFFFF; // segment limit
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit_scaled = 0xFFFFFFFF; // scaled segment limit
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.g = 1; // 4k granularity
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.d_b = 1; // 32-bit mode
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.avl = 0; // available for use by system
|
|
|
|
ESP = ECX;
|
|
EIP = EDX;
|
|
|
|
BX_INSTR_FAR_BRANCH(BX_CPU_ID, BX_INSTR_IS_SYSEXIT,
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.value, EIP);
|
|
#else
|
|
BX_INFO(("SYSEXIT: use --enable-sep to enable SYSENTER/SYSEXIT support"));
|
|
UndefinedOpcode (i);
|
|
#endif
|
|
}
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
void BX_CPU_C::SYSCALL(bxInstruction_c *i)
|
|
{
|
|
bx_address temp_RIP;
|
|
|
|
BX_DEBUG(("Execute SYSCALL instruction"));
|
|
|
|
if (!BX_CPU_THIS_PTR efer.sce) {
|
|
exception(BX_UD_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
if (BX_CPU_THIS_PTR efer.lma)
|
|
{
|
|
RCX = RIP;
|
|
R11 = read_eflags() & ~(EFlagsRFMask);
|
|
|
|
if (BX_CPU_THIS_PTR cpu_mode == BX_MODE_LONG_64) {
|
|
temp_RIP = MSR_LSTAR;
|
|
}
|
|
else {
|
|
temp_RIP = MSR_CSTAR;
|
|
}
|
|
|
|
// set up CS segment, flat, 64-bit DPL=0
|
|
parse_selector((MSR_STAR >> 32) & BX_SELECTOR_RPL_MASK,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.dpl = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.type = BX_CODE_EXEC_READ_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base = 0; /* base address */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit = 0xFFFF; /* segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit_scaled = 0xFFFFFFFF; /* scaled segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.g = 1; /* 4k granularity */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.l = 1; /* 64-bit code */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.avl = 0; /* available for use by system */
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
BX_CPU_THIS_PTR updateFetchModeMask();
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 4 && BX_SUPPORT_ALIGNMENT_CHECK
|
|
BX_CPU_THIS_PTR alignment_check = 0; // CPL=0
|
|
#endif
|
|
|
|
// set up SS segment, flat, 64-bit DPL=0
|
|
parse_selector(((MSR_STAR >> 32) + 8) & BX_SELECTOR_RPL_MASK,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.dpl = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.type = BX_DATA_READ_WRITE_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.base = 0; /* base address */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit = 0xFFFF; /* segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit_scaled = 0xFFFFFFFF; /* scaled segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.g = 1; /* 4k granularity */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.d_b = 1; /* 32 bit stack */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.l = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.avl = 0; /* available for use by system */
|
|
|
|
writeEFlags(read_eflags() & (~MSR_FMASK), EFlagsValidMask);
|
|
BX_CPU_THIS_PTR clear_RF();
|
|
RIP = temp_RIP;
|
|
}
|
|
else {
|
|
// legacy mode
|
|
|
|
ECX = EIP;
|
|
temp_RIP = MSR_STAR & 0xFFFFFFFF;
|
|
|
|
// set up CS segment, flat, 32-bit DPL=0
|
|
parse_selector((MSR_STAR >> 32) & BX_SELECTOR_RPL_MASK,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.dpl = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.type = BX_CODE_EXEC_READ_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base = 0; /* base address */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit = 0xFFFF; /* segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit_scaled = 0xFFFFFFFF; /* scaled segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.g = 1; /* 4k granularity */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.l = 0; /* 32-bit code */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.avl = 0; /* available for use by system */
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
BX_CPU_THIS_PTR updateFetchModeMask();
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 4 && BX_SUPPORT_ALIGNMENT_CHECK
|
|
BX_CPU_THIS_PTR alignment_check = 0; // CPL=0
|
|
#endif
|
|
|
|
// set up SS segment, flat, 32-bit DPL=0
|
|
parse_selector(((MSR_STAR >> 32) + 8) & BX_SELECTOR_RPL_MASK,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.dpl = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.type = BX_DATA_READ_WRITE_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.base = 0; /* base address */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit = 0xFFFF; /* segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.limit_scaled = 0xFFFFFFFF; /* scaled segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.g = 1; /* 4k granularity */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.d_b = 1; /* 32 bit stack */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.l = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.u.segment.avl = 0; /* available for use by system */
|
|
|
|
BX_CPU_THIS_PTR clear_VM();
|
|
BX_CPU_THIS_PTR clear_IF();
|
|
BX_CPU_THIS_PTR clear_RF();
|
|
RIP = temp_RIP;
|
|
}
|
|
|
|
BX_INSTR_FAR_BRANCH(BX_CPU_ID, BX_INSTR_IS_SYSCALL,
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.value, RIP);
|
|
}
|
|
|
|
void BX_CPU_C::SYSRET(bxInstruction_c *i)
|
|
{
|
|
bx_address temp_RIP;
|
|
|
|
BX_DEBUG(("Execute SYSRET instruction"));
|
|
|
|
if (!BX_CPU_THIS_PTR efer.sce) {
|
|
exception(BX_UD_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
if(real_mode() || CPL != 0) {
|
|
BX_ERROR(("SYSRET: priveledge check failed, generate #GP(0)"));
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
}
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
if (BX_CPU_THIS_PTR cpu_mode == BX_MODE_LONG_64)
|
|
{
|
|
if (i->os64L()) {
|
|
// Return to 64-bit mode, set up CS segment, flat, 64-bit DPL=3
|
|
parse_selector(((MSR_STAR >> 48) + 16) | 3,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.dpl = 3;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.type = BX_CODE_EXEC_READ_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base = 0; /* base address */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit = 0xFFFF; /* segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit_scaled = 0xFFFFFFFF; /* scaled segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.g = 1; /* 4k granularity */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b = 0;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.l = 1; /* 64-bit code */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.avl = 0; /* available for use by system */
|
|
|
|
temp_RIP = RCX;
|
|
}
|
|
else {
|
|
// Return to 32-bit compatibility mode, set up CS segment, flat, 32-bit DPL=3
|
|
parse_selector((MSR_STAR >> 48) | 3,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.dpl = 3;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.type = BX_CODE_EXEC_READ_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base = 0; /* base address */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit = 0xFFFF; /* segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit_scaled = 0xFFFFFFFF; /* scaled segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.g = 1; /* 4k granularity */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.l = 0; /* 32-bit code */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.avl = 0; /* available for use by system */
|
|
|
|
temp_RIP = ECX;
|
|
}
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
BX_CPU_THIS_PTR updateFetchModeMask();
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 4 && BX_SUPPORT_ALIGNMENT_CHECK
|
|
handleAlignmentCheck(); // CPL was modified
|
|
#endif
|
|
|
|
// SS base, limit, attributes unchanged
|
|
parse_selector((MSR_STAR >> 48) + 8,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.dpl = 3;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.type = BX_DATA_READ_WRITE_ACCESSED;
|
|
|
|
writeEFlags((Bit32u) R11, EFlagsValidMask);
|
|
}
|
|
else { // (!64BIT_MODE)
|
|
// Return to 32-bit legacy mode, set up CS segment, flat, 32-bit DPL=3
|
|
parse_selector((MSR_STAR >> 48) | 3,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.dpl = 3;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.type = BX_CODE_EXEC_READ_ACCESSED;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base = 0; /* base address */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit = 0xFFFF; /* segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit_scaled = 0xFFFFFFFF; /* scaled segment limit */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.g = 1; /* 4k granularity */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.l = 0; /* 32-bit code */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.avl = 0; /* available for use by system */
|
|
|
|
#if BX_SUPPORT_ICACHE
|
|
BX_CPU_THIS_PTR updateFetchModeMask();
|
|
#endif
|
|
|
|
#if BX_CPU_LEVEL >= 4 && BX_SUPPORT_ALIGNMENT_CHECK
|
|
handleAlignmentCheck(); // CPL was modified
|
|
#endif
|
|
|
|
// SS base, limit, attributes unchanged
|
|
parse_selector((MSR_STAR >> 48) + 8,
|
|
&BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].selector);
|
|
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.valid = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.p = 1;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.dpl = 3;
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.segment = 1; /* data/code segment */
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_SS].cache.type = BX_DATA_READ_WRITE_ACCESSED;
|
|
|
|
BX_CPU_THIS_PTR assert_IF();
|
|
temp_RIP = ECX;
|
|
}
|
|
|
|
RIP = temp_RIP;
|
|
|
|
BX_INSTR_FAR_BRANCH(BX_CPU_ID, BX_INSTR_IS_SYSRET,
|
|
BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].selector.value, RIP);
|
|
}
|
|
|
|
void BX_CPU_C::SWAPGS(bxInstruction_c *i)
|
|
{
|
|
Bit64u temp_GS_base;
|
|
|
|
BX_ASSERT(protected_mode());
|
|
|
|
if(CPL != 0)
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
|
|
|
temp_GS_base = MSR_GSBASE;
|
|
MSR_GSBASE = MSR_KERNELGSBASE;
|
|
MSR_KERNELGSBASE = temp_GS_base;
|
|
}
|
|
#endif
|
|
|
|
#if BX_X86_DEBUGGER
|
|
void BX_CPU_C::hwbreakpoint_match(bx_address laddr, unsigned len, unsigned rw)
|
|
{
|
|
if (BX_CPU_THIS_PTR dr7 & 0x000000ff) {
|
|
// Only compare debug registers if any breakpoints are enabled
|
|
unsigned opa, opb;
|
|
opa = BX_HWDebugMemRW; // Read or Write always compares vs 11b
|
|
if (rw==BX_READ) // only compares vs 11b
|
|
opb = opa;
|
|
else // BX_WRITE or BX_RW; also compare vs 01b
|
|
opb = BX_HWDebugMemW;
|
|
Bit32u dr6_bits = hwdebug_compare(laddr, len, opa, opb);
|
|
if (dr6_bits) {
|
|
BX_CPU_THIS_PTR debug_trap |= dr6_bits;
|
|
BX_CPU_THIS_PTR async_event = 1;
|
|
}
|
|
}
|
|
}
|
|
|
|
Bit32u BX_CPU_C::hwdebug_compare(bx_address laddr_0, unsigned size,
|
|
unsigned opa, unsigned opb)
|
|
{
|
|
// Support x86 hardware debug facilities (DR0..DR7)
|
|
Bit32u dr7 = BX_CPU_THIS_PTR dr7;
|
|
|
|
bx_bool ibpoint_found = 0;
|
|
bx_address laddr_n = laddr_0 + (size - 1);
|
|
static bx_address alignment_mask[4] =
|
|
// 00b=1 01b=2 10b=undef 11b=4
|
|
{ 0x0, 0x1, 0x0, 0x3 };
|
|
|
|
Bit32u len0 = (dr7>>18) & 3;
|
|
Bit32u len1 = (dr7>>22) & 3;
|
|
Bit32u len2 = (dr7>>26) & 3;
|
|
Bit32u len3 = (dr7>>30) & 3;
|
|
|
|
bx_address dr0 = (BX_CPU_THIS_PTR dr0) & ~(alignment_mask[len0]);
|
|
bx_address dr1 = (BX_CPU_THIS_PTR dr1) & ~(alignment_mask[len1]);
|
|
bx_address dr2 = (BX_CPU_THIS_PTR dr2) & ~(alignment_mask[len2]);
|
|
bx_address dr3 = (BX_CPU_THIS_PTR dr3) & ~(alignment_mask[len3]);
|
|
|
|
bx_address dr0_n = dr0 + len0;
|
|
bx_address dr1_n = dr1 + len1;
|
|
bx_address dr2_n = dr2 + len2;
|
|
bx_address dr3_n = dr3 + len3;
|
|
|
|
Bit32u dr0_op = (dr7>>16) & 3;
|
|
Bit32u dr1_op = (dr7>>20) & 3;
|
|
Bit32u dr2_op = (dr7>>24) & 3;
|
|
Bit32u dr3_op = (dr7>>28) & 3;
|
|
|
|
// See if this instruction address matches any breakpoints
|
|
if ((dr7 & 0x00000003)) {
|
|
if ((dr0_op==opa || dr0_op==opb) &&
|
|
(laddr_0 <= dr0_n) &&
|
|
(laddr_n >= dr0))
|
|
ibpoint_found = 1;
|
|
}
|
|
if ((dr7 & 0x0000000c)) {
|
|
if ((dr1_op==opa || dr1_op==opb) &&
|
|
(laddr_0 <= dr1_n) &&
|
|
(laddr_n >= dr1))
|
|
ibpoint_found = 1;
|
|
}
|
|
if ((dr7 & 0x00000030)) {
|
|
if ((dr2_op==opa || dr2_op==opb) &&
|
|
(laddr_0 <= dr2_n) &&
|
|
(laddr_n >= dr2))
|
|
ibpoint_found = 1;
|
|
}
|
|
if ((dr7 & 0x000000c0)) {
|
|
if ((dr3_op==opa || dr3_op==opb) &&
|
|
(laddr_0 <= dr3_n) &&
|
|
(laddr_n >= dr3))
|
|
ibpoint_found = 1;
|
|
}
|
|
|
|
// If *any* enabled breakpoints matched, then we need to
|
|
// set status bits for *all* breakpoints, even disabled ones,
|
|
// as long as they meet the other breakpoint criteria.
|
|
// This code is similar to that above, only without the
|
|
// breakpoint enabled check. Seems weird to duplicate effort,
|
|
// but its more efficient to do it this way.
|
|
if (ibpoint_found) {
|
|
// dr6_mask is the return value. These bits represent the bits to
|
|
// be OR'd into DR6 as a result of the debug event.
|
|
Bit32u dr6_mask=0;
|
|
if ((dr0_op==opa || dr0_op==opb) &&
|
|
(laddr_0 <= dr0_n) &&
|
|
(laddr_n >= dr0))
|
|
dr6_mask |= 0x01;
|
|
if ((dr1_op==opa || dr1_op==opb) &&
|
|
(laddr_0 <= dr1_n) &&
|
|
(laddr_n >= dr1))
|
|
dr6_mask |= 0x02;
|
|
if ((dr2_op==opa || dr2_op==opb) &&
|
|
(laddr_0 <= dr2_n) &&
|
|
(laddr_n >= dr2))
|
|
dr6_mask |= 0x04;
|
|
if ((dr3_op==opa || dr3_op==opb) &&
|
|
(laddr_0 <= dr3_n) &&
|
|
(laddr_n >= dr3))
|
|
dr6_mask |= 0x08;
|
|
return(dr6_mask);
|
|
}
|
|
|
|
return(0);
|
|
}
|
|
#endif
|
|
|
|
/*
|
|
void BX_CPU_C::LFENCE(bxInstruction_c *i) {}
|
|
void BX_CPU_C::MFENCE(bxInstruction_c *i) {}
|
|
void BX_CPU_C::SFENCE(bxInstruction_c *i) {}
|
|
*/
|