On Nov 10, 11:31am, Max Matveev wrote:
> Subject: Re: Source level debugging with kdb
>
> Ray,
>
> First of all, I'll copy this replay to [EMAIL PROTECTED] - hopefully
> some(one) will add more to that I know.
> RB> Can some explain the advantages of doing remote debugging via kdb
> RB> instead of by kgdb?
> kdb can do backtraces without aid of frame pointers and it
> actually works as opposed to gdb (and kgdb, since kgdb is only a
> remote proxy for gdb). What alone is worth a lot.
I would say this is the big one. Also, with Keith's set up, you can access
the kdb commands via "mon" to get to things that the kernel would know about,
but gdb probably doesn't. An example might be certain MMU registers that can't
be read from user-land, or many of the different types of registers that are
only able to be accessed by the kernel in IA64.
I was talking to Keith's and Max's manager about this recently, too. He told
me that the stack tracebacks in the kernel might be strange enough that gdb
could not handle them as well as kdb, which would have more knowledge since it
is actually part of the kernel. He said this was especially true for IA64
tracebacks. Maybe Keith would know more here.
That's all I can think of off hand. Oh, there might be some issues with
modules. I think kdb understands modules, but I am not sure anymore. Maybe
Keith or Max can answer that. However, I believe newer versions of gdb also
understand modules, so that may be a moot point. I certainly know of people
using kgdb to debug modules.
Hope this helps,
Scott
--
Scott Foehner SGI
[EMAIL PROTECTED] Computer Systems Business Unit
650-933-3473 Core OS
