Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!milano!bigtex!james
From: james@bigtex.cactus.org (James Van Artsdalen)
Newsgroups: comp.unix.sysv386
Subject: CPU steppings (Re: Problems with AHA1542B ...)
Message-ID: <52096@bigtex.cactus.org>
Date: 2 Jan 91 00:21:33 GMT
References: <1220@aut.autelca.ascom.ch> <1yyJu4w163w@bluemoon.uucp> <507@zok.UUCP>
Reply-To: james@bigtex.cactus.org (James Van Artsdalen)
Organization: Institute of Applied Cosmology, Austin TX
Lines: 45

In <507@zok.UUCP>, mark@zok.UUCP (Mark W. Snitily) wrote:

> Evidently the new 2.2 release enables instruction pipelining in the
> 386 processor, something which ISC neglected to tell us poor souls.

I was not aware that instruction pipelining was under software
control.  How is this done?  Do you refer to the LE and GE bits of
DR7?

> Check your 386 processor, just below the i386 logo should be the
> processor version.  If the chip is the SX211 version, you're in
> trouble because there is a bug in processor regarding instruction
> pipelining [...]

That line does indeed tell the stepping of the processor (the date
code is on the bottom of the chip).  But it is not a one to one
mapping: a given stepping may have several SXnnn codes.

However, I do not believe that you are necessarily in trouble with an
SX211 chip.  My memory may be wrong, but I think that the instruction
pre-fetch bugs required address pipelining (and paging, and a TLB
miss, and #NA asserted, and a full moon...).  One would hope that the
board designer took this into account and didn't use pipelining in
situations where it didn't work.

> The solution is to replace the chip with the "SX219" version, the hang
> problem will vanish completely.

Word of caution: if the system is designed for the SX211, it may not
have fixes for bugs in the SX219.  Changing CPU steppings can cause
problems.

>    I advise everyone to check their processor version before upgrading or
>    purchasing 2.2, this problem will most likely affect *any* machine with
>    the SX211 version chip.

Again, this is not correct in all cases.  It's entirely possible that
work-arounds for the errata are on the motherboard, in which case an
SX-211 might work just fine (and an SX-219 might not work at all).

33MHz parts and 25MHz parts have different SXnnn codes, even if
they're otherwise the same.
-- 
James R. Van Artsdalen          james@bigtex.cactus.org   "Live Free or Die"
Dell Computer Co    9505 Arboretum Blvd Austin TX 78759         512-338-8789