[parisc-linux] Strange delay loop calibration with SMP kernel on UP system

Thibaut VARENE varenet@esiee.fr
Sat, 12 Apr 2003 18:58:02 +0200


Hi pa-ckers!

As I was working on 2.5 SMPness on our A500, tausq suggested me to try SMP
kernel on a UP box.
So I deconfigured one CPU on the A500.

The kernel failed miserably, and I rebooted the box with its _original 2.4
SMP kernel_, _without bringing the second CPU back online_.

I repeat: i booted the box with the 2.4 SMP kernel it used to run with 2
CPUs. That's 2.4.20-pa28 fwiw.

What a surprise!

The delay loop calibration gave a completely different value than what I'd
have expected according to previous boots.

see:

A500 configured with 2 CPUs:
Calibrating delay loop... 1097.72 BogoMIPS

A500 configured with 1 CPU:
Calibrating delay loop... 366.18 BogoMIPS

Please mind I'm not talking about "total bogomips".
I'm talking about the calibration that takes place at the very early stage
of the boot process, on 1 CPU (the monarch on SMP systems).

So that has nothing to do with having half bogomips because there's half
processors.

That being said, one would have noticed that 1097.72/366.18 ~= 3

Now I wonder, is that normal (I couldn't check on another arch), or is
that some kind of new and unknown bug?

HTH,


Thibaut VARENE
The PA/Linux ESIEE Team
http://pateam.esiee.fr/