Ronald G Minnich <[EMAIL PROTECTED]> writes:
> yeah, it might be. On the other hand, we're fooling around with seeing how
> much stuff we can yank into linux user commands, such as over-clocking,
> watching fans, and so on.
Cool. How far have you gotten?
Anything you can do in a BIOS should be possible from linux...
>
> Also, some kind of less ad-hoc way to get into reboot would be nice. I
> spent a lot of time looking at the linux reboot code, and at some point I
> realized what a hack it is. Also, it for the most part barely works. All
> my cluster nodes appear to reboot via the triple-fault path. yuck.
Hmm. I'm not certain which path is best. But they all seem to
simply simulate a cpu startup. The address jumped to is exactly
the address an 8086 jumps to on startup. And other cpus simulate
this with some help of the chipset.
With respect to rebooting there are 3 wasy to accomplish this:
1) Unplug and plug in the machine. (A very hard reboot)
2) Use software to toggle the machines power status. (A hard reboot)
3) Use software to simulate a machine powering up (A soft reboot)
It appears linux bios handles 1 & 2 just fine since it doesn't need
to do anything. Why doesn't 3 work?
>
> > And wouldn't access to reboot from /proc mean that the system would
> > shutdown prematurely? Since init will call the reboot syscall when
> > it's time to reboot, why would you want to add an alternate mechanism?
>
> actually, it's useful. It's pretty much a nicer version of the "Magic
> SysReq key". If you're running ext3 then an instantaneous reboot is kind
> of handy, and you're not going to be hurt by doing it.
Though you might loose some data...
I sort of agree. But if you want something like this generate a new
minimal fs.
Eric
