On Fri, Jul 18, 2003 at 05:55:30PM -0700, Steven Dake wrote: > On Fri, 2003-07-18 at 17:40, [EMAIL PROTECTED] wrote: > > Hi Keith, > > > > I'm vaguely daydreaming about a few KDB enhancements, I was curious if > > you've thought about them: > > > > -- 'cat /proc/meminfo' Turns out one bug I was chasing was OOM on > > a machine with 32GB of RAM (!) and it took me days to even think > > of looking there. > > > > -- Support for examining structures symbolically. I've had to walk > > the buffer cache by hand in the past, I've walked task struct > > by hand yesterday, and they guy down the hall stares at network > > structs. > > > While we are dreaming :) complete source level debugging would be very > useful. I was thinking this would best be done over serial or ethernet > with a kdb engine running in the target kernel. This would allow a > debugger to use the kdb commands to symbolicly debug a running kernel.
Yeah, I admit, inverting assembly back into source is tedious. FYI, I'm working at a level where I can justify spending a few days here and there, so my dreams are necessarily limited. > Much better then kgdb, which lacks many of the functionalities required > by a good kernel debugger. Uhh... well, kgdb does have all this stuff for crawling through stabs, and symbol tables and etc. So if source debug is really the requirement then I'd think that adding the missing features to kgdb would be the simpler route. Am I wrong? --linas
