Re: Linux kdb (was Re: first thread)

[Philipp Rumpf <prumpf@jcsbs.lanobis.de>]

On Fri, Jan 29, 1999 at 11:32:34AM -0500, Albert D. Cahalan wrote:
> > We could really do with a more user-friendly alternative,
> > which is what a dump partition, some utilities aso. could give us.
> 
> No, that is not user-friendly at all.

> If we put System.map into the kernel, we could get symbol info
> in the oops message. That would be very good to have, because
> people are less willing to copy down "random" numbers.

Is it user-friendly to have people copy down "random" C names ?

Oopses have a good chance of killing the machine, but the chance
they kill the SCSI layer or drivers or the IDE code are quite low.
(Even if the hard disk driver gets killed, one can mount the floppy
drive as oops / dump partition and it is not necessary to do hand-
copying).

So we can save the Oops and nobody has to write it down. We have a
greater chance of getting the Oops report if we lock the partition
and make noise when the user wants to mount it (I mean noise that
gets repeated until the partition is unlocked).

Everything else (dumping kernel memory, splitting up the partition,
mounting it as a "file system" so we can access the Oops as a text
file in some directory, dumping as SysRq option) can be done later.

> Don't mention klogd and ksymoops: you must assume that system logs are
> never checked and that a crash will get ignored unless it is easy to
> report. The kernel ought to print "mail all you see to ...".

I do not think someone will ignore clear instructions repeatedly hitting
his console.
-
Linux-future: thinking about the future of the Linux kernel
http://humbolt.nl.linux.org/lists/