Identification of 386-class CPUs; checking for 486+




Behavioral identification of 386-class CPUs; checking for 486+
Copyright 1996-2000 by Grzegorz Mazur
All the brand names used here belong to their
owners. 

Last updated: 2000-09-18.



Revision history: 2000-09-18 - more details 1996-03-25 - initial
version, not yet finished



This part of article describes the routines for recognizing various 386-class
machines. The test for 486 or better CPU is presented.



Some backgroundThere are several 386-class chips. Intel original 80386,
later renamed to 386DX has at least 3 easily distinguishable varieties. There is
also 386SX, with 16-bit bus, Chips and Technologies 38600DX and IBM 386SLC. In
general, late Intel and AMD chips are the same, several early Intel varieties
can be distinguished, and we can also recognize SX and C&T stuff.
NexGen Nx586 chips produced until late 1995 also fit into the 386 cathegory.
Although they are architecturally more advanced than Pentium, they support the
instruction set architecture of 386.
All the 386-class and better chips support the identification mechanism.
After hardware reset, the chip signature is placed by microcode in EDX register.
Unfortunately, there is no way other than reset to obtain it, so the mechanism
is not very useful, although it can usually be used in real mode. Follow this link for details.



386(and Nx586) vs. 486+80486 and newer CPUs support the alignment check
mechanism. The CPU, while executing in its least privilleged state, can be
forced to report exception when 32-bit data is accessed in memory at the address
not being the multiple of 4. Alignment check mode is enabled by setting bit 18
of EFLAGS register to one. There is no such bit on 386 CPUs, so it can not be
toggled... To check for AC bit presence, we must first save ESP into some other
register and align it, rounding its contents down to a multiple of 4. Then we
save EFLAGS, change AC bit, set EFLAGS, save it again and restore the previous
value. If the value of AC in both saved copies is different, we have 486+ CPU. Otherwise we
have some 386 or NexGen Nx586.



Checking for old 80386 (sigma-pi)Old 80386, also known as sigma-pi
chips, had famuous multiply bug. 32-bit multiply was returning incorrect
results.



Checking for real Intel(or AMD) 32-bit 386 (80386, 386DX)All the
classic 32-bit 386 chips have POPAD bug. The bug exhibits when instructions
using 32-bit indexed addressing mode immediately follows POPAD instruction. In
such case the content of EAX may not be correctly updated by POPAD. The test
should be executed several times in a loop, cause the error may be masked by
interrupt. If the chip does not exhibit POPAD bug, it is probably C&T
38600DX. I have no information on POPAD behavior of IBM 386SLC.



AMD 386DXLV & SXLV chipsThese may be probably detected by checking
for the presence of UMOV instruction. UMOV is a SMM instruction similar to MOV
and is described in AMD manuals.



Copyright 1996-2000 by Grzegorz
Mazur