2000-10-27 07:35:16

by H. Peter Anvin

[permalink] [raw]
Subject: Re: missing mxcsr initialization

MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 8bit
X-Comment-To: Alan Cox <[email protected]>
Disclaimer: Not speaking for Transmeta in any way, shape, or form.
Copyright: Copyright 2000 H. Peter Anvin - All Rights Reserved

Followup to: <[email protected]>
By author: Alan Cox <[email protected]>
In newsgroup: linux.dev.kernel
> > > corrected for include the facts that the XMM feature bit is an Intel specific
> > > bit that other vendors may use for other things, so you need to test vendor ==
> > ^^^
> > Note that they shouldn't do that! I would consider a very bad thing if they
> > goes out of sync on those bits.
> CPUID is vendor specific. Every bit in those fields is vendor specific. Every
> piece of documentation tells you to check the CPU vendor. Every time we didnt
> bother we got burned.
> I keep hearing people saying things like 'bad thing' 'assume standards'. Well
> all I can say is cite a vendor issued document which says 'dont bother checking
> the vendor'.

Intel does it because they want every other chip out there to act like
a 486.

> And when you can't find that document, put the checks in so we dont crash on
> an Athlon or when using MTRR on a Cyrix III etc

Chips that don't implement what they claim to implement are buggy and
should be treated as such. SPECIAL-CASE THE BUGGY CHIPS, NOT THE


<[email protected]> at work, <[email protected]> in private!
"Unix gives you enough rope to shoot yourself in the foot."