Fixed handling of redundant address/data override prefixes

This commit is contained in:
Jeff Parsons 2015-07-17 10:14:17 -07:00
commit b6e8eb8877
4 changed files with 68 additions and 19 deletions

View file

@ -3228,15 +3228,25 @@ if (DEBUGGER) {
var bOpcode = this.getByte(dbgAddr, 1);
/*
* Incorporate the following prefixes into the current instruction byte stream.
* TODO: Determine the actual effect of multiple OS (and/or multiple AS) prefixes.
* Incorporate OS and AS prefixes into the current instruction.
*
* And the verdict is in: redundant OS and AS prefixes must be ignored;
* see opOS() and opAS() for details. We limit the amount of redundancy
* to something reasonable (ie, 4).
*/
var cMax = 2; // let's make sure unfortunate memory contents don't screw us
var cMax = 4;
var fDataPrefix = false, fAddrPrefix = false;
while ((bOpcode == X86.OPCODE.OS || bOpcode == X86.OPCODE.AS) && cMax--) {
if (bOpcode == X86.OPCODE.OS) {
dbgAddr.fData32 = !dbgAddr.fData32;
if (!fDataPrefix) {
dbgAddr.fData32 = !dbgAddr.fData32;
fDataPrefix = true;
}
} else {
dbgAddr.fAddr32 = !dbgAddr.fAddr32;
if (!fAddrPrefix) {
dbgAddr.fAddr32 = !dbgAddr.fAddr32;
fAddrPrefix = true;
}
}
bOpcode = this.getByte(dbgAddr, 1);
}

View file

@ -58,8 +58,8 @@ if (!I386) {
}
} else {
/*
* These are the more general-purpose ModRM decoders, required for I386 suppport. The current addressing
* mode (16-bit or 32-bit) dynamically selects the appropriate byte and word decoders.
* These are the more general-purpose ModRM decoders, required for I386 support. The current addressing
* mode (16-bit or 32-bit) dynamically selects the appropriate set of decoders.
*/
if (typeof module !== 'undefined') {
var X86ModB16 = require("./x86modb16");
@ -141,9 +141,9 @@ function X86CPU(parmsCPU)
* if any function returns false, the software interrupt will be skipped (presumed to be emulated),
* and no further notification functions will be called.
*
* NOTE: Registered functions are called only for "INT N" instructions -- NOT "INT 3" or "INTO" or the
* "INT 0x00" generated by a divide-by-zero or any other kind of interrupt (nor any interrupt simulated
* with "PUSHF/CALLF").
* NOTE: Registered functions are called only for INT N instructions -- *not* INT 0x03 or INTO or the
* INT 0x00 generated by a divide-by-zero or any other kind of interrupt (nor any interrupt simulated
* with PUSHF/CALLF).
*
* aIntReturn is a hash of return address notifications set up by software interrupt notification
* functions that want to receive return notifications. A software interrupt function must call
@ -1163,7 +1163,7 @@ X86CPU.prototype.reset = function()
* the D0 stepping reported 0x0305; beyond that, it's not known exactly what revision numbers Intel used for all
* 80386 revisions.
*
* We define some additional "registers", such as regLIP. which mirrors the linear address corresponding to
* We define some additional "registers", such as regLIP, which mirrors the linear address corresponding to
* CS:IP (the address of the next opcode byte). In fact, regLIP functions as our internal IP register, so any
* code that needs the real IP must call getIP(). This, in turn, means that whenever CS or IP must be modified,
* regLIP must be recalculated, so you must use either setCSIP(), which takes both an offset and a segment,
@ -2791,6 +2791,11 @@ X86CPU.prototype.setBinding = function(sHTMLType, sBinding, control)
case "DS":
case "SS":
case "ES":
case "FS":
case "GS":
case "CR0":
case "CR2":
case "CR3":
case "PS": // this refers to "Processor Status", aka the 16-bit flags register (although DEBUG.COM refers to this as "PC", surprisingly)
case "C":
case "P":
@ -3822,6 +3827,8 @@ X86CPU.prototype.updateStatus = function(fForce)
this.updateReg("DS", this.getDS());
this.updateReg("SS", this.getSS());
this.updateReg("ES", this.getES());
this.updateReg("FS", this.getFS());
this.updateReg("GS", this.getGS());
this.updateReg("EIP", this.getIP());
var regPS = this.getPS();
this.updateReg("PS", regPS);
@ -3834,6 +3841,9 @@ X86CPU.prototype.updateStatus = function(fForce)
this.updateReg("A", (regPS & X86.PS.AF));
this.updateReg("P", (regPS & X86.PS.PF));
this.updateReg("C", (regPS & X86.PS.CF));
this.updateReg("CR0", this.regCR0);
this.updateReg("CR2", this.regCR2);
this.updateReg("CR3", this.regCR3);
}
}

View file

@ -3724,7 +3724,7 @@ X86.fnFaultMessage = function(nFault, nError, fHalt)
* However, the foregoing notwithstanding, if MESSAGE.HALT is enabled along with all the other required
* MESSAGE bits, then we want to halt regardless.
*/
if (this.messageEnabled(bitsMessage | Messages.HALT)) {
if (DEBUG && nFault == X86.EXCEPTION.GP_FAULT || this.messageEnabled(bitsMessage | Messages.HALT)) {
fHalt = true;
}

View file

@ -219,7 +219,7 @@ X86.opPUSHCS = function PUSHCS()
};
/**
* op=0x0F (POP CS) (undocumented on 8086/8088; replaced with opInvalid on 80186/80188, and op0F on 80286 and up)
* op=0x0F (POP CS) (undocumented on 8086/8088; replaced with opInvalid() on 80186/80188, and op0F() on 80286 and up)
*
* @this {X86CPU}
*/
@ -1375,10 +1375,18 @@ X86.opGS = function GS()
X86.opOS = function OS()
{
if (I386) {
/*
* See opAS() for a discussion of multiple prefixes, which applies equally to both
* operand-size and address-size prefixes.
*
* The simple fix here is to skip the bulk of the operation if the prefix is redundant.
*/
this.opFlags |= X86.OPFLAG.DATASIZE;
this.dataSize ^= 0x6; // that which is 2 shall become 4, and vice versa
this.dataMask ^= (0xffff0000|0); // that which is 0x0000ffff shall become 0xffffffff, and vice versa
this.updateDataSize();
if (!(this.opPrefixes & X86.OPFLAG.DATASIZE)) {
this.dataSize ^= 0x6; // that which is 2 shall become 4, and vice versa
this.dataMask ^= (0xffff0000|0); // that which is 0x0000ffff shall become 0xffffffff, and vice versa
this.updateDataSize();
}
this.nStepCycles -= this.cycleCounts.nOpCyclesPrefix;
}
};
@ -1393,10 +1401,31 @@ X86.opOS = function OS()
X86.opAS = function AS()
{
if (I386) {
/*
* Live and learn: multiple address-size prefixes can and do occur on a single instruction,
* and contrary to my original assumption that the prefixes act independently, they do not.
* During Windows 95 SETUP, the following instruction is executed:
*
* 06AF:1B4D 67672E CS:
* 06AF:1B50 FFA25A1B JMP [BP+SI+1B5A]
*
* which is in fact:
*
* 06AF:1B4D 67672E CS:
* 06AF:1B50 FFA25A1B0000 JMP [EDX+00001B5A]
*
* The other interesting question is: why/how did this instruction get encoded that way?
* All I can say is, there were no explicit prefixes in the source (BSG.ASM), so we'll chalk
* it up to a glitch in MASM.
*
* The simple fix here is to skip the bulk of the operation if the prefix is redundant.
*/
this.opFlags |= X86.OPFLAG.ADDRSIZE;
this.addrSize ^= 0x06; // that which is 2 shall become 4, and vice versa
this.addrMask ^= (0xffff0000|0); // that which is 0x0000ffff shall become 0xffffffff, and vice versa
this.updateAddrSize();
if (!(this.opPrefixes & X86.OPFLAG.ADDRSIZE)) {
this.addrSize ^= 0x06; // that which is 2 shall become 4, and vice versa
this.addrMask ^= (0xffff0000|0); // that which is 0x0000ffff shall become 0xffffffff, and vice versa
this.updateAddrSize();
}
this.nStepCycles -= this.cycleCounts.nOpCyclesPrefix;
}
};