Every time Windows 95 starts up, its real-mode loader performs the following CPU identification test.
&0654:121E 9C PUSHF &0654:121F 33C0 XOR AX,AX ; try to clear bit 15 of flags &0654:1221 50 PUSH AX &0654:1222 9D POPF &0654:1223 9C PUSHF &0654:1224 58 POP AX &0654:1225 A90080 TEST AX,8000 ; bit 15 of flags set anyway? &0654:1228 7543 JNZ 126D ; yes (must be an 8086/8086) &0654:122A B80070 MOV AX,7000 ; try to set bits 12-14 of flags &0654:122D 50 PUSH AX &0654:122E 9D POPF &0654:122F FB STI &0654:1230 9C PUSHF &0654:1231 58 POP AX &0654:1232 9D POPF &0654:1233 A90070 TEST AX,7000 ; any of bits 12-14 of flags set? &0654:1236 7436 JZ 126E ; no (must be an 80286) &0654:1238 51 PUSH CX &0654:1239 33C9 XOR CX,CX &0654:123B 8BC4 MOV AX,SP &0654:123D 83E003 AND AX,0003 &0654:1240 7404 JZ 1246 &0654:1242 8BC8 MOV CX,AX &0654:1244 2BE0 SUB SP,AX &0654:1246 669C PUSHFD &0654:1248 666800000400 PUSH 00040000 ; try to set the AC bit of flags &0654:124E 669D POPFD &0654:1250 669C PUSHFD &0654:1252 6658 POP EAX &0654:1254 669D POPFD &0654:1256 66A900000400 TEST EAX,00040000 ; AC bit set? &0654:125C 7508 JNZ 1266 ; yes (must be an 80486, or newer) &0654:125E 40 INC AX &0654:125F 03E1 ADD SP,CX &0654:1261 59 POP CX &0654:1262 0BC0 OR AX,AX &0654:1264 F9 STC ; return ZF clear and CF set to indicate 80386 &0654:1265 C3 RET &0654:1266 03E1 ADD SP,CX &0654:1268 59 POP CX &0654:1269 660BC0 OR EAX,EAX ; return ZF clear and CF clear to indicate 80486 &0654:126C C3 RET &0654:126D 9D POPF &0654:126E 33C0 XOR AX,AX ; return ZF set to indicate 80286 or older &0654:1270 C3 RET
If the above function returns ZF set, then the processor is an 80286 or older, so Windows 95 displays the following message and exits:
You need an 80386 processor to run Windows.
If the above function returns CF set, then the processor is an 80386, and if CF is clear, the processor is an 80486 or newer.
When CF is set, Windows 95 proceeds to the following 80386 stepping check, which attempts to execute an XBTS (Extract Bit String) instruction – an instruction that existed only on B0 and earlier 80386 steppings.
&0654:1299 53 PUSH BX &0654:129A 51 PUSH CX &0654:129B 52 PUSH DX &0654:129C B80635 MOV AX,3506 ; save the current "invalid opcode" handler (IVT entry #6) &0654:129F CD21 INT 21 &0654:12A1 8CC0 MOV AX,ES &0654:12A3 66C1E010 SHL EAX,10 &0654:12A7 8BC3 MOV AX,BX &0654:12A9 6650 PUSH EAX &0654:12AB 1E PUSH DS &0654:12AC BAE012 MOV DX,12E0 &0654:12AF 8CCB MOV BX,CS &0654:12B1 8EDB MOV DS,BX &0654:12B3 B80625 MOV AX,2506 ; temporarily install a new "invalid opcode" exception handler &0654:12B6 CD21 INT 21 &0654:12B8 1F POP DS &0654:12B9 33C0 XOR AX,AX &0654:12BB 8BD0 MOV DX,AX &0654:12BD B900FF MOV CX,FF00 &0654:12C0 0FA6CA XBTS CX,DX,AX,CL ; attempt to execute an XBTS instruction &0654:12C3 6658 POP EAX &0654:12C5 8BD0 MOV DX,AX &0654:12C7 66C1E810 SHR EAX,10 &0654:12CB 1E PUSH DS &0654:12CC 8ED8 MOV DS,AX &0654:12CE B80625 MOV AX,2506 ; restore the original "invalid opcode" exception handler &0654:12D1 CD21 INT 21 &0654:12D3 1F POP DS &0654:12D4 B8B000 MOV AX,00B0 &0654:12D7 0BC9 OR CX,CX ; did XBTS work (ie, did CX change)? &0654:12D9 7401 JZ 12DC ; yes, so we must have a B0 stepping or earlier &0654:12DB 40 INC AX ; no, so bump the stepping to B1 &0654:12DC 5A POP DX &0654:12DD 59 POP CX &0654:12DE 5B POP BX &0654:12DF C3 RET ; returns AX == 0xB0 if B0 stepping or earlier, 0xB1 if B1 stepping or later &0654:12E0 55 PUSH BP ; temporary "invalid opcode" handler &0654:12E1 8BEC MOV BP,SP &0654:12E3 83460203 ADD [BP+02],0003 ; advance IP past the 3-byte XBTS instruction &0654:12E7 5D POP BP &0654:12E8 CF IRET
If the above function returns 0xB0, then the 80386 is a B0 or earlier stepping, so Windows 95 displays the following message and aborts:
Windows may not run correctly with the 80386 processor that is installed in this computer. Upgrade your 80386 processor.
Otherwise, the 80386 is a B1 or later stepping, so Windows 95 next performs a multiplication test (a simplified version of the multiplication tests discussed in “Early 80386 CPUs”):
&0654:12E9 33C9 XOR CX,CX &0654:12EB 66BB81000000 MOV EBX,00000081 &0654:12F1 66B800A01704 MOV EAX,0417A000 &0654:12F7 66F7E3 MUL EBX &0654:12FA 6683FA02 CMP EDX,00000002 &0654:12FE 750B JNZ 130B &0654:1300 663D00A0E70F CMP EAX,0FE7A000 &0654:1306 7503 JNZ 130B &0654:1308 E2E1 LOOP 12EB &0654:130A C3 RET
If any of the 65,536 identical multiplications return an incorrect result, Windows 95 displays the following message:
WARNING: The 80386 processor in this computer may not reliably execute 32-bit multiplication. Windows may occasionally fail on this computer. You may want to replace your 80386 processor. Press any key to continue...Press a key to continue
A multiplication failure implies that the 80386 stepping is B1, because later steppings resolved the problem.
You may have heard that Windows 95 pulled support for the 80386 B1 stepping, and that’s true, but only insofar as Windows 95 SETUP is concerned. The following code is executed by WINSETUP.BIN, a 16-bit Windows component that manages the Windows 95 installation process:
#05C7:69E8 1E PUSH DS #05C7:69E9 07 POP ES #05C7:69EA 6657 PUSH EDI #05C7:69EC FD STD #05C7:69ED 66BF00000000 MOV EDI,00000000 #05C7:69F3 678A07 MOV AL,[EDI] #05C7:69F6 67AA STOSB #05C7:69F8 33C0 XOR AX,AX #05C7:69FA 6681FFFFFF0000 CMP EDI,0000FFFF #05C7:6A01 7501 JNZ 6A04 #05C7:6A03 40 INC AX #05C7:6A04 FC CLD #05C7:6A05 665F POP EDI #05C7:6A07 C3 RET
Ths above code checks for B1 stepping Errata #7: “Wrong Register Size for String Instructions in Mixed 16/32-bit Addressing Systems.” It returns AX == 0 if the STOSB instruction updated EDI correctly (0xFFFFFFFF) or AX == 1 if EDI is incorrect (0x0000FFFF).
If Errata #7 is detected, then Windows 95 SETUP displays the following message and aborts:
Setup Error B1: Setup has detected an 80386 processor that is not compatible with this version of Windows. Before you can run this version of Windows, you need to upgrade your processor. Contact your computer manufacturer for more information.
However, if you can get through SETUP, Windows 95 will still run on a B1 stepping. For example, if you installed Windows 95 using a newer 80386, and then later “downgraded” the CPU to a B1, Windows 95 would still run. If your B1 suffered from the multiplication flaw, you would see the 32-bit multiplication warning on start-up, but you could still continue to run, and if there was no multiplication problem, you would not see any message at all.
PCjs v1.20.0 now supports a “stepping” attribute on the <cpu> element, which you can use to simulate specific stepping behavior. For example, a machine.xml file with the following CPU definition:
<cpu id="cpu386" model="80386" stepping="b0"/>
will cause Windows 95 to abort exactly as described as above. Similarly, selecting a 80386 B1 stepping:
<cpu id="cpu386" model="80386" stepping="b1"/>
will cause Windows 95 to display the 32-bit multiplication warning shown above (PCjs deliberately fails the exact multiplication test that Windows 95 performs).
If you want to simulate a B1 stepping that does not have the 32-bit multiplication flaw, set the stepping to B2:
<cpu id="cpu386" model="80386" stepping="b2"/>
B2 was not an actual 80386 stepping; it is a pseudo-stepping that provides a simple way of specifying a B1 80386 that passes all 32-bit multiplication tests.
As previously discussed, Windows 95 SETUP will refuse to install on any “A” or “B” stepping, but if it’s already been installed, it will start up on a B1 stepping.
PCjs stepping support is extremely limited at this point. Here’s a summary:
- 80386 steppings A0-B0 provide limited support for the short-lived XBTS and IBTS instructions
- 80386 steppings A0-B1 enable Errata #7 for STOSB (as tested by Windows 95; see above)
- 80386 stepping B1 enables 32-bit multiplication errors (as tested by Windows 95; see above)
- 80386 stepping B2 includes all supported B1 errata, but without 32-bit multiplication errors
In addition, on 80386 reset, we set the CPU revision number in DX to the appropriate value for the specified stepping.
Support for additional 80286 and 80386 errata may be added over time, as interesting scenarios or test cases are discovered.
October 27, 2015