Re: SpeedStep / Geyserville lockups?

[Heather <star@betelgeuse.starshine.org>]

>>>>>> Is anyone else out there seeing lockups under 2.2.x (2.2.18 here) with a
     > PIII SpeedStep (aka Geyserville) chip?
     >
     > I've had lockups over the past week I suspect are due to CPU step-speed
>>>>>> changes resulting from power source switching.

>> I believe that Speedstep machines are much more stable if they are always 
>> (re)booted while attached to wall power.   Suspends and resumes should behave
>> much better, but *those* depend on whether your APM BIOS is crappy :(
> 
>  Yes, that's what the guy who maintains APM in the kernel told me after his
> talk at OLS last year.  I told him about my laptop that lets you switch
> between 33 and 66 MHz with a keypress, and he said to always boot up at high
> speed because waiting too long is a lot safer than not waiting long enough
> when it comes to the micro-delays that are timed with bogomips.

Ahhh, thanks for confirming what I was thinking.  It's nice to hear it 
(albeit secondhand) from someone who's actually been down there at the code
level...

> > I personally suspect that some things may not be all that well behaved "at
> > the wrong speed" too but I have no direct experience with that.
> 
>  I would be surprised unless you make your clock speed slower by more like
> an order of magnitude.  (depending on the drivers you are using, of course).
> Most of the time, the kernel will wait only the minimum delay specified by
> the hardware.  If there is a maximum delay, it will be much longer than the
> minimum, so waiting longer won't hurt.

Allow me to add, then, that my personal laptop can step down to 12% of 
233 MHz, so yes, that's probably an order of magnitude.  I don't do that 
during networking usage though and haven't been quite that low when using
external hardware via PCMCIA  --  so I have the *potential* to encounter such
problems, but I haven't, though it may be a matter of chance.   Said laptop 
is ancient :)  but its APM support is excellent.

I also have not played with the most recent kernels on it, so I don't know
if the kernel has become more sensitive to this sort of abuse.

* Heather Stern * star@ many places...