architecture kernel problems

[Matthew Saltzman]

On Mon, 11 Dec 2006, James Wilkinson wrote:

> Les wrote:
>> I don't know what you are working on Mike, but if it helps, I installed
>> to a 433Mhz celeron with the i386 package and when I ran that command I
>> got:
>> kernel-2.6.18-1.2798.fc6.i586
>> kernel-2.6.18-1.2849.fc6.i586
>> kernel-devel-2.6.18-1.2798.fc6.i586
>> kernel-devel-2.6.18-1.2849.fc6.i586
>> kernel-headers-2.6.18-1.2849.fc6.i386
>>
>> I was sort of expecting them all to say i386???  Maybe someone can help
>> us understand what's happening.
>
> i586 will work, but an i686 kernel will work slightly better on a
> Celeron.
>
> The basic 32 bit set of instructions that x86 Linux uses were introduced
> with the Intel i386 back in 1986. Later Intel processors added extra
> instructions to help in specific cases. (Later on, various "multimedia"
> instructions were also added, but that's a separate discussion.)
>
> Most user-space programs don't actually need or gain from those extra
> instructions. So Fedora compiles most programs using only i386
> instructions -- hence the "i386" in most package names.
>
> Some packages (the kernel and glibc, for example) *can* make use of the
> extra instructions. So Fedora provides an i686 version of those
> packages, which do use the extra instructions. Unfortunately, there are
> still some processors which don't support the i686 level of instructions
> (Via only recently started supporting them, and AMD K6s are still used).
> For these processors, Fedora provides an i586 version. (i586
> instructions are still better than just the i386 instruction set in
> these cases).
>
> i586 programs will work on later processors, but i586 processors don't
> know how to handle i686 instructions. (If they did, they'd be i686
> processors).
>
> I'm not sure whether the kernel headers actually contain *any*
> instructions -- if they do, they'd be tiny portions of assembler. They
> don't use processor-specific instructions, so count as i386.
>
> Hope this helps,

IIRC, the i586 was peculiar to original Pentium and Pentium MMS.  I recall 
some discussion that optimal instruction ordering was different enough on 
i586 and i686 that running i586 kernels on i686 chips would result in more 
or less significant performance issues.

>
> James.
>
>

-- 
 		Matthew Saltzman

Clemson University Math Sciences
mjs AT clemson DOT edu
http://www.math.clemson.edu/~mjs