Re: Pentium DEATH and Cyrix COMA in user-mode

Andre Derrick Balsa (andrewbalsa@usa.net)
Mon, 10 Nov 1997 14:17:53 +0100


Hi Alan,

Alan Cox wrote:

> They are very similar it appears. One is "lock cmpxchg8" which causes a double
> lock cycle on the bus (I believe its to the bus), the other is a dual issue
> of an implicitly locked operation (again dual lock).
>

Yes, it is the CPU bus that gets locked i.e. interrupts are not serviced
and other bus masters are not granted access to the CPU bus.

Cyrix 6x86 CPUs have a NO_LOCK flag in a control register that disables
the implicit locking of xchg instructions (among other things).

This effectively prevents the Cyrix bug. I am still testing to check if
it has other negative effects.

On my two 6x86 Linux boxes it seems to be an acceptable solution.

If a similar solution could be found for the Pentium F0 bug, it could
save Intel a few billion dollars...

Cheers,

========================================================
Andrew D. Balsa
Home Page: http://www.tux.org/~balsa
andrewbalsa@usa.net
========================================================