Cleaned up conditional jump handling

modules/pcjs/bin/romtests.json

@@ -8,16 +8,16 @@
   "ram": [
     { "id": "pc386.ramLow",
       "name": "",
-      "addr": 0x00000,
+      "addr": 0,
-      "size": 0xa0000,
+      "size": 655360,
       "test": false
     }
   ],
   "rom": [
     { "id": "pc386.romTests",
       "name": "",
-      "addr": 0xf0000,
+      "addr": 983296,
-      "size": 0x10000,
+      "size": 65280,
       "file": "/tests/pc/80386/tests.json",
       "notify": ""
     }

modules/pcjs/lib/README.md

@@ -100,8 +100,8 @@ A BackTrack index is encoded as a 32-bit value with three parts:
 
 This represents a total of 31 bits, with bit 31 reserved.
 
-For example, look at one of the last things a ROM does during boot: load a disk sector into RAM.  It will be up to the
+For example, look at one of the last things a ROM does during boot: loading a disk sector into RAM.  It will be up to the
-disk controller (or DMA controller if one is used) to create a BackTrack object representing the sector that was read,
+disk controller (or DMA controller, if used) to create a BackTrack object representing the sector that was read,
 adding that object to the global BackTrack object array, and then associating the corresponding BackTrack index with
 the first byte of RAM where the sector was loaded.  Subsequent bytes of RAM containing the rest of the sector will refer
 to the same BackTrack object, using BackTrack indexes containing offsets 1-511.

modules/pcjs/lib/debugger.js

@@ -1603,7 +1603,7 @@ if (DEBUGGER) {
         /*
          * When we dump the EXT word, we mask off the LIMIT1619 and BASE2431 bits, because those have already
          * been incorporated into the limit and base properties of the segment register; all we care about here
-         * are whether EXT contains any of the AVAIL (0x10), BIG (0x40) or GRANULARITY (0x80) bits.
+         * are whether EXT contains any of the AVAIL (0x10), BIG (0x40) or LIMITPAGES (0x80) bits.
          */
         this.println(sDump + " dpl=" + str.toHexByte(seg.dpl) + " type=" + str.toHexByte(seg.type >> 8) + " (" + sType + ")" + " ext=" + str.toHexWord(seg.ext & ~(X86.DESC.EXT.LIMIT1619 | X86.DESC.EXT.BASE2431)));
     };

modules/pcjs/lib/x86.js

@@ -103,10 +103,10 @@ var X86 = {
         MASK:   0xfff8      // table index
     },
     DESC: {                 // Descriptor Table Entry
-        LIMIT: {
+        LIMIT: {            // LIMIT bits 0-15
             OFFSET:     0x0
         },
-        BASE: {
+        BASE: {             // BASE bits 0-15
             OFFSET:     0x2
         },
         ACC: {              // bit definitions for the access word (offset 0x4)
@@ -166,7 +166,7 @@ var X86 = {
              * is 0xffffffff instead of 0xffff.
              */
             BIG:                            0x0040,     // clear if default operand/address size is 16-bit, set if 32-bit
-            GRANULARITY:                    0x0080,     // clear if limit is bytes, set if limit is 4Kb pages
+            LIMITPAGES:                     0x0080,     // clear if limit granularity is bytes, set if limit granularity is 4Kb pages
             BASE2431:                       0xff00
         },
         INVALID: 0          // use X86.DESC.INVALID for invalid DESC values

modules/pcjs/lib/x86cpu.js

@@ -3197,18 +3197,6 @@ X86CPU.prototype.getIPDisp = function()
     return w;
 };
 
-/**
- * getIPDispWord()
- *
- * @this {X86CPU}
- * @return {number} sign-extended value from the word at the current IP; IP advanced by 2 or 4
- */
-X86CPU.prototype.getIPDispWord = function()
-{
-    var w = this.getIPWord();
-    return (this.dataSize == 2? ((w << 16) >> 16) : w);
-};
-
 /**
  * getSIBAddr(mod)
  *

modules/pcjs/lib/x86op0f.js

@@ -342,6 +342,23 @@ X86.opMOVcr = function MOVcr()
     }
 };
 
+/*
+ * NOTE: The following 16 new conditional jumps actually rely on the OPERAND override setting
+ * for determining whether a signed 16-bit or 32-bit displacement will be fetched, even though
+ * the ADDRESS override might seem more intuitive.  Think of them as instructions that are loading
+ * a new operand into IP/EIP.
+ *
+ * Also, in 16-bit code, even though a signed rel16 value would seem to imply a range of -32768
+ * to +32767, any location within a 64Kb code segment outside that range can be reached by choosing
+ * a displacement in the opposite direction, causing the 16-bit value in EIP to underflow or overflow;
+ * any underflow or overflow doesn't matter, because only the low 16 bits of EIP are used when a
+ * 16-bit OPERAND size is in effect.
+ *
+ * In fact, for 16-bit jumps, it's simpler to always think of rel16 as an UNSIGNED value added to
+ * the current EIP, where the result is then truncated to a 16-bit value.  This is why we don't have
+ * to sign-extend rel16 before adding it to the current EIP.
+ */
+
 /**
  * opJOw()
  *
@@ -351,7 +368,7 @@ X86.opMOVcr = function MOVcr()
  */
 X86.opJOw = function JOw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (this.getOF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -369,7 +386,7 @@ X86.opJOw = function JOw()
  */
 X86.opJNOw = function JNOw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getOF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -387,7 +404,7 @@ X86.opJNOw = function JNOw()
  */
 X86.opJCw = function JCw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (this.getCF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -405,7 +422,7 @@ X86.opJCw = function JCw()
  */
 X86.opJNCw = function JNCw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getCF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -423,7 +440,7 @@ X86.opJNCw = function JNCw()
  */
 X86.opJZw = function JZw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (this.getZF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -441,7 +458,7 @@ X86.opJZw = function JZw()
  */
 X86.opJNZw = function JNZw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getZF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -459,7 +476,7 @@ X86.opJNZw = function JNZw()
  */
 X86.opJBEw = function JBEw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (this.getCF() || this.getZF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -477,7 +494,7 @@ X86.opJBEw = function JBEw()
  */
 X86.opJNBEw = function JNBEw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getCF() && !this.getZF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -495,7 +512,7 @@ X86.opJNBEw = function JNBEw()
  */
 X86.opJSw = function JSw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (this.getSF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -513,7 +530,7 @@ X86.opJSw = function JSw()
  */
 X86.opJNSw = function JNSw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getSF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -531,7 +548,7 @@ X86.opJNSw = function JNSw()
  */
 X86.opJPw = function JPw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (this.getPF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -549,7 +566,7 @@ X86.opJPw = function JPw()
  */
 X86.opJNPw = function JNPw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getPF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -567,7 +584,7 @@ X86.opJNPw = function JNPw()
  */
 X86.opJLw = function JLw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getSF() != !this.getOF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -585,7 +602,7 @@ X86.opJLw = function JLw()
  */
 X86.opJNLw = function JNLw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getSF() == !this.getOF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -603,7 +620,7 @@ X86.opJNLw = function JNLw()
  */
 X86.opJLEw = function JLEw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (this.getZF() || !this.getSF() != !this.getOF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -621,7 +638,7 @@ X86.opJLEw = function JLEw()
  */
 X86.opJNLEw = function JNLEw()
 {
-    var disp = this.getIPDispWord();
+    var disp = this.getIPWord();
     if (!this.getZF() && !this.getSF() == !this.getOF()) {
         this.setIP(this.getIP() + disp);
         this.nStepCycles -= this.cycleCounts.nOpCyclesJmpC;
@@ -1301,7 +1318,7 @@ X86.aOps0F[0x06] = X86.opCLTS;
 
 /*
  * On all processors (except the 8086/8088, of course), X86.OPCODE.UD2 (0x0F,0x0B), aka "UD2", is an
- * instruction guaranteed to raise a #UD (Invalid Opcode) exception (INT 0x06) on all future x86 processors.
+ * instruction guaranteed to raise a #UD (Invalid Opcode) exception (INT 0x06) on all post-8086 processors.
  */
 X86.aOps0F[0x0B] = X86.opInvalid;
 

modules/pcjs/lib/x86ops.js

@@ -3497,6 +3497,10 @@ X86.opESC = function ESC()
 /**
  * op=0xE0 (LOOPNZ disp)
  *
+ * NOTE: All the instructions in this group (LOOPNZ, LOOPZ, LOOP, and JCXZ) actually
+ * rely on the ADDRESS override setting for determining whether CX or ECX will be used,
+ * even though it seems counter-intuitive; ditto for the REP prefix.
+ *
  * @this {X86CPU}
  */
 X86.opLOOPNZ = function LOOPNZ()
@@ -3513,6 +3517,10 @@ X86.opLOOPNZ = function LOOPNZ()
 /**
  * op=0xE1 (LOOPZ disp)
  *
+ * NOTE: All the instructions in this group (LOOPNZ, LOOPZ, LOOP, and JCXZ) actually
+ * rely on the ADDRESS override setting for determining whether CX or ECX will be used,
+ * even though it seems counter-intuitive; ditto for the REP prefix.
+ *
  * @this {X86CPU}
  */
 X86.opLOOPZ = function LOOPZ()
@@ -3529,6 +3537,10 @@ X86.opLOOPZ = function LOOPZ()
 /**
  * op=0xE2 (LOOP disp)
  *
+ * NOTE: All the instructions in this group (LOOPNZ, LOOPZ, LOOP, and JCXZ) actually
+ * rely on the ADDRESS override setting for determining whether CX or ECX will be used,
+ * even though it seems counter-intuitive; ditto for the REP prefix.
+ *
  * @this {X86CPU}
  */
 X86.opLOOP = function LOOP()
@@ -3543,7 +3555,11 @@ X86.opLOOP = function LOOP()
 };
 
 /**
- * op=0xE3 (JCXZ disp)
+ * op=0xE3 (JCXZ/JECXZ disp)
+ *
+ * NOTE: All the instructions in this group (LOOPNZ, LOOPZ, LOOP, and JCXZ) actually
+ * rely on the ADDRESS override setting for determining whether CX or ECX will be used,
+ * even though it seems counter-intuitive; ditto for the REP prefix.
  *
  * @this {X86CPU}
  */

modules/pcjs/lib/x86seg.js

@@ -588,7 +588,7 @@ X86Seg.prototype.loadDesc8 = function(addrDesc, sel, fSuppress)
     if (I386 && cpu.model >= X86.MODEL_80386) {
         base |= (ext & X86.DESC.EXT.BASE2431) << 16;
         limit |= (ext & X86.DESC.EXT.LIMIT1619) << 16;
-        if (ext & X86.DESC.EXT.GRANULARITY) limit = (limit << 12) | 0xfff;
+        if (ext & X86.DESC.EXT.LIMITPAGES) limit = (limit << 12) | 0xfff;
     }
 
     while (true) {

tests/pc/80386/makefile

@@ -1,7 +1,7 @@
 all: tests.json
 
-tests.rom: tests.nasm
+tests.com: tests.nasm
-	nasm -f bin tests.nasm -l tests.lst -o tests.rom
+	nasm -f bin tests.nasm -l tests.lst -o tests.com
 
-tests.json: tests.rom
+tests.json: tests.com
-	node ../../../modules/filedump/bin/filedump --file=tests.rom --output=tests.json --overwrite
+	node ../../../modules/filedump/bin/filedump --file=tests.com --output=tests.json --overwrite

tests/pc/80386/tests.nasm

@@ -1,21 +1,85 @@
 	cpu	386
-
+;
-	org	0x0000
+; This file is designed to run as a ROM replacement, but it has a .COM extension because it's
+; also designed to run as a COM file under DOS (hence the "org 0x100").
+;
+	org	0x100
 	section .text
 
 	bits	16
 
+DPL0	equ	0
+DPL1	equ	1
+DPL2	equ	2
+DPL3	equ	3
+
+ACC_TYPE_SEG			equ	0x1000
+ACC_TYPE_CODE			equ	0x0800
+ACC_TYPE_READABLE		equ	0x0200
+ACC_TYPE_WRITABLE		equ	0x0200
+ACC_TYPE_CODE_READABLE		equ	0x1a00
+ACC_TYPE_DATA_WRITABLE		equ	0x1200
+
+EXT_BIG				equ	0x0040
+
+SEG_CODE_REAL	equ	0xf000
+SEG_CODE_PROT	equ	0x0008
+SEG_DATA_PROT	equ	0x0010
+
+CR0_MSW_PE	equ	0x0001
+
+;
+; descDT defines a descriptor, given a base (%1), limit (%2), type (%3), dpl (%4), and bigness (%5)
+;
+%macro	defDesc	1-5 0,0,0,0
+	dw	(%2 & 0x0000ffff)
+	dw	(%1 & 0x0000ffff)
+	dw	((%1 & 0x00ff0000) >> 16) | %3 | (%4 << 13)
+	dw	((%2 & 0x000f0000) >> 16) | %5 | ((%1 & 0xff000000) >> 16)
+%endmacro
+
+
 start:	mov	eax,0x44332211
 	mov	ecx,0x88776655
 	mul	ecx
 	div	ecx
+	jnz	near goProt
 
-	call	dword 0xf000:start
+	times	32768 nop
 
-	times	0xfff0-($-$$) nop
+romGDT:	defDesc	0
+	defDesc	0x000f0000,0x0000ffff,ACC_TYPE_CODE_READABLE,0
+	defDesc	0x00000000,0x000fffff,ACC_TYPE_DATA_WRITABLE,0
+
+goProt:	lgdt	[cs:romGDT]
+	mov	eax,cr0
+	or	eax,CR0_MSW_PE
+	mov	cr0,eax
+	jmp	dword SEG_CODE_PROT:inProt
+
+inProt:	mov	ax,SEG_DATA_PROT
+	mov	ds,ax
+;
+; Do some protected-mode tests...
+;
+
+goReal:	mov	eax,cr0
+	and	eax,~CR0_MSW_PE
+	mov	cr0,eax
+	jmp	dword SEG_CODE_REAL:inReal
+
+inReal:	or	eax,1
+	jnz	start
+
+	;
+	; Fill the remaining space with NOPs until we get to target offset 0xFFF0.
+	;
+	; Note that we subtract 0x100 from the target offset because we're ORG'ed at 0x100.
+	;
+	times	0xfff0-0x100-($-$$) nop
 
 	bits	16
-	jmp	0xf000:start
+	jmp	SEG_CODE_REAL:start
 
 	db	0x20
 	db	'04/04/15'