> >Nevertheless, 2.1 freezes can be cleanly debugged via the 'ktrace' patch
[...]
> Does this work with cold freezes? I mean, 0-degree-kelvin cold (no
> log messages, no drive syncs, no LED light response, no packet
> response, nothing)? [...]
yeah, i've debugged cold EIDE PCI freezes with this feature.
it's pretty simple: it prints a log onto the _text screen_, without
scrolling of course, and by directly writing into video memory. The
ringlog will run pretty fast, system will be slower, but it works. If now
a cold freeze happens, all output will suddenly stop. So you've simply got
to write down the log from the screen, where the kernel was when the
lockup happened ;)
-- mingo