In message <20010105200140.A8175@localhost>, Bernd Sieker writes:
>On 05.01.01, 10:14:05, Jason R Thorpe wrote:
>> On Fri, Jan 05, 2001 at 12:03:04PM -0600, Peter Seebach wrote:
>> Console abort is only useful if it's provided by the firmware on the
>> system (i.e. Abort Switch on the AlphaServers, which traps into the
>> firmware).  If you have to rely on the (wedged) operating system to
>> provide the abort functionality, well ... that's not useful at all.

>Which leaves the question why we trust a wedged system, that has
>already panicked, to be in consistent enough a state to do a reliable
>dump of the main memory to the dump partition.

We don't *trust* it, but having the option is better than nothing.

>How can we be sure that it's not in a state where it has messed up the
>disklabel or whatever it uses to determine the dump partition location
>at that point, to not accidently dump onto another part of the disk?

If it's in such a state, the "abort" not working could be pretty bad too;
it's easy to be closer to the keyboard than to the computer.

In general, "system is wedged" is roughly equivalent to "program in foreground
is in I/O wait talking to a PCMCIA card but kernel is fine", at least, that's
where I tend to run into it.  ):

-s