Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!usc!wuarchive!rex!uflorida!mailer.cc.fsu.edu!prism!bb16
From: bb16@prism.gatech.EDU (Scott Bostater)
Newsgroups: comp.binaries.ibm.pc.d
Subject: Re: v09i191: popadbug, test for 386 CPU bug (part 01/01)
Message-ID: <19130@hydra.gatech.EDU>
Date: 4 Jan 91 13:39:58 GMT
References: <336@aplcomm.JHUAPL.EDU>
Organization: Georgia Institute of Technology
Lines: 45

In article <336@aplcomm.JHUAPL.EDU> tcs@mailer.jhuapl.edu (Carl Schelin) writes:
>I ran this on my IBM Model 80 here at work and on a new clone
>386 25mhz (three months old) and they both failed. The one at home
>has a co-processor but the one here doesn't (80).
>
>Anyone know what this is supposed to do? I did a virus check before 
>running and rebooted immediately after running (just to make sure). 
>No problems so far.

We've run it on several 386's where I work and it fails on all of them.  It has
not failed on the one 486 that we have.  I haven't yet tried it on a 386sx 
(anybody out there with a SX tried it??).  It also shows up in both of the
Sun 386i's that I have access to.

As far as I can tell from running the code in Turbo Debug, the only error is
that the EAX register gets cleared.  As the code stated this can be corrected
by placing a NOP after the POPAD command  (or single stepping through the 
code :-).  Looks like a microcode timing error.

As far as ramifications of this bug, I personnally don't think its too major
of a problem unless you want to program assembly code.  In that case you'll 
just have to remember to add the NOP after the POPAD command.  When programming
in 286 assembly language I have had little reason to use the POPA command
since I hardly ever need to push/pop ALL the registers. Its normally faster to 
push/pop just the required registers. There would be even less justification to
push/pop all of the 32-bit registers.  Since there has been a moderate amount
of 32-bit code produced by now and this is the first that the bug has shown up,
I don't think the POPAD command is getting used. 

Not having access to a any 32 bit compilers, I don't know if any of them 
generate the POPAD command, but I would doubt it unless you were optimizing for
smallest code size (and we all optimize for fastest execution, right? :-)

IMHO, the bottom line is that 
  1) yes, Intel has another bug
  2) no, the average programmer probably doesn't need to worry about it
  3) writers of 32-bit compilers and OS's should be aware of it

Well, that's my $0.05 worth  ($0.02 plus overhead :-)

-- 
Scott Bostater      Georgia Tech Research Institute - Radar Systems Analysis
"My soul finds rest in God alone; my salvation comes from Him"  -Ps 62.1
uucp:     ...!{allegra,amd,hplabs,ut-ngp}!gatech!prism!bb16
Internet: bb16@prism.gatech.edu