Re: SMP kernels and single Cyrix processor problem

Brian Gerst (bgerst@quark.vpplus.com)
Wed, 12 Nov 1997 16:09:39 -0500

Linux Developer wrote:
>
> On Tue, 11 Nov 1997, Brian Gerst wrote:
>
> > I've been having problems lately with forgetting to unset SMP=1 which is
> > enabled by default in the 2.1.x kernels, only to end up with an unusable
> > kernel on my cpu/motherboard. I have a Tyan Tomcat IV (SMP capable)
> > board, and a Cyrix 6x86L 166+ processor. When the SMP kernel starts it
> > immediately oopses after detecting that the motherboard supports SMP,
> > even though the processor does not. I have not been able to capture the
> > oopses as they scroll by too fast and syslog has not started yet. Is
> > there any way I can catch the oopses so I can figure out where it is
> > crashing? At least it should determine that the mobo/cpu cannot run an
> > SMP kernel and halt with a warning instead of producing the oopses.
>
> I have spoken with Cyrix about SMP on the 6x86 and 6x86MX series of CPUs a
> number of times in the past. Cyrix and AMD both use the OpenPic chipset
> for SMP. In order to use a single Cyrix on an SMP board, you must compile
> the kernel as single-CPU, and if necessary, set the motherboard to
> single-CPU mode. The Tyan Tomcat series shouldn't require this, but you DO
> need a single-CPU kernel. Either that or another Cyrix and a dual Socket7
> OpenPic chipset motherboard. ;)

I've found that adding 'nosmp' to loadlin or lilo will allow an SMP
kernel to work on a Cyrix. It's just getting quite irritating that SMP
is turned on by default in the 2.1.x kernels and it's not a normal
config option (out of sight out of mind...)

All the more reason to move SMP to Config.in

Interesting to note though... the SMP kernel still tries to detect a
second processor even if nosmp is specified. In my case, it finds a
phantom second processor and was probably trying to enable it (and
crashing). AFAIK there is no BIOS switch to enable/disable the second
cpu, but I do have the single cpu variant of the board (all of which
means the second zif socket was not soldered onto the board!).

I am going to try to make a patch to detect if the cpu is really an
Intel cpu and disable SMP if otherwise.

-- 


						Brian Gerst