Re: Linux spin_unlock debate (fwd)

From: Maciej W. Rozycki (
Date: Wed Apr 26 2000 - 04:47:27 EST

On Tue, 25 Apr 2000, Oliver Xymoron wrote:

> > Not every i486 and i386 MP was weakly ordered.
> > Most weren't.
> > A few were.
> > Intel had no control over which was and which wasn't,
> > because everything was at the mercy of the system integrator.
> The ones that were weakly ordered probably didn't match the Intel MP boot
> spec either and are unlikely to ever be supported. Not because it's not
> doable, but since it's been 5+ years and no one's cared enough to do it
> yet.

 You can choose strong spin_unlock conditionally when either CONFIG_M386
or CONFIG_M486 is defined. I think there is a possibility more and more
i486 MP systems will appear as their first-hand owners are getting rid of
them gradually. The second-hand owners may prefer to use Linux over
proprietary OSes these systems used before. I can't recall any failure
reports for i82489DX-based systems until last year, even though there used
to be a few quirks in the startup code preventing such systems from
booting till recently.

 There may actually be weakly ordered Pentium MP systems, too. Correct me
if I'm wrong, but I believe there are no cache snooping signals in Pentium
60/66 CPUs and also triple/quad/etc. P54C systems cannot make use of these
signals as they only support dual configurations. So we may even have to
use strong spin_unlock for CONFIG_M586 or better let the user choose
either an integrated or a standalone APIC configuration -- it seems most
(all?) triple and larger P54C configurations used i82489DX APICs (probably
because the i82093AA I/O APIC was not available at that time, yet).

 Given that i82489DXs (their bus protocol) were targeted to i486 and later
systems I would hardly believe there were MP-compliant i386 systems.
Unless they used discrete APIC implementations. ;-)

+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+        e-mail:, PGP key available        +

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to Please read the FAQ at

This archive was generated by hypermail 2b29 : Sun Apr 30 2000 - 21:00:11 EST