> On Mon, 19 Jan 1998, Richard Guy Briggs wrote:
> > > Thats all I can think of right now.. I'm sure there are many more things
> > > that I would feel nessassary in a 3.0 release.. I'm sure others can add..
> > How about a kernel panic core memory dump to disk for forensic research?
>
> The current logic behind not doing this in linux is that since its
> impossible to do it perfectly (e.g. 100% foolproof under all situations),
> it shouldnt be done at all. (Still, BSD manages to do it).
>
A couple weeks ago, Linus stated quite clearly that he disliked crash
dumps because they led to a style of fault repair which often fixed the
symptom instead of fixing the problem. By giving less information, a
programmer was forced to understand the problem, instead of patching it
away.
I understand that viewpoint. I have seen my sahre of lousey fixes.
On the other side, a crash dump can be extremely useful in tracking down
ill-system situations. The system is running slower and slower... or
everything is stopped but the mouse... or a storage creep has eaten up
half the storage in a week. A crash dump, with suitable diagnostics, can
reveal problem situations which do not result in immediate faults.
I think a crash dump facility would help accelerate progress in solving
the more mature problems which are showing up recently.
john alvord
p.s. I spent years on a dump analysis tool in a former life.