From: Linus Torvalds
> Sent: 15 August 2018 21:19
...
> But if people run things on real machines, then BUG() is absolutely
> the last thing you EVER want to do for "debugging".
I'm not sure you want it on a live system either.
Live systems are where the 'hard' bugs show up.
I've just spent a couple of days pulling my hair out trying to work
out how to debug a customer system that was locking up solid when
running some new (and not completely testable by us) kernel code.
At 4am I suddenly realised that the distribution they are using
might be enabling 'panic_on_oops' by default.
Turning that off showed what was going wrong.
It wouldn't be as bad if Linux implemented 'dump to swap'.
For 'errors' that aren't completely fatal the system could
'fast shutdown' a lot of processes (maybe just refuse to schedule
them) while leaving enough running for fault diagnosis.
I'm not sure how you'd decide what to allow to run though.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT,
UK
Registration No: 1397386 (Wales)
