[ On Sunday, June 10, 2001 at 23:53:36 (-0400), David Maxwell wrote: ]
> Subject: Re: ctrl+alt+del
>
> Well, that leaves open a whole discussion about where userland's border
> is, wrt 'init'. :-)

I don't think there's any question there.  Init is firmly and solidly a
100% user-land process.  (It just happens to be the "first" process but
from the time the "creates" process #1 and sets it up to exec /sbin/init
it never ever again runs in kernel space.)

> That would need to be settable as well. The people arguing for CAD
> to allow convienient shutdown via the remote-hands protocol, may not
> want to open their console to the messing that can be done with DDB. DDB
> may also not be compiled in.

If there's no DDB compiled in then there are a zillion and one other
ways to safely shut down the machine from user-land; and an order of
magnitude more ways to invoke them without having to have the console
driver learn any magic and/or send any signals to init.

Or are you saying instead that there should always be a way to trigger a
kernel sync and reboot from the console without going to userland?

I'm not entirely opposed to having stand-alone kernel-only reboot
trigger support in the console, just so long as it asks me nicely
whether I really mean it or not and will go back to normal operation if
I say "no" (i.e. just like BSD/OS does).  I just don't think it's really
that important an option given that 99.999% of the times it's needed it
can be provided by DDB, and I think it would have to be a kernel
compile-time _option_ anyway, even on i386.

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods@acm.org>     <woods@robohack.ca>
Planix, Inc. <woods@planix.com>;   Secrets of the Weird <woods@weird.com>