[PATCH] RFC: don't aggravate kernel panics

[Werner Almesberger]

Andy Green wrote:
> What we could do with is this business of spawning a new kernel to
> report a panic in a controlled way.

Ah yes, I saw you mention that you want to explore the kdump concept.
You can do pretty cool things with this, but at the same time, it may
be a little overkill ;-)

> Another way to come at that would be stash the syslog somewhere safe
> near the end of memory, set a signature there and reboot off the panic
> into the recovery kernel, and show it there.

We could just always copy the dmesg buffer to a safe place on panic
or such and do a normal reboot. As long as we make sure the DRAM
gets refreshed, then the next system, which doesn't even have to be
anything special, can just pick it up.

Apropos recovery system, are you thinking of a "connect to Internet
and mail crash log to Andy" button in the extended boot loader ? ;-))

> Anyway this patch doesn't hurt anything since we need interrupts to do
> I2C stuff anyway, so I sent it on andy-tracking.

Thanks !

> But the git.openmoko.org box seems down at the moment.

Yeah, unscheduled extended lunch break :)

- Werner