On Thu, 2006-10-05 at 23:32 +0200, Andi Kleen wrote:
> On Thursday 05 October 2006 23:30, Piet Delaney wrote:
> > On Thu, 2006-10-05 at 12:17 -0700, Tom Rini wrote:
> > > On Thu, Oct 05, 2006 at 08:46:20AM -0700, George Anzinger wrote:
> > > > Tom Rini wrote:
> > > > >On Wed, Oct 04, 2006 at 08:42:04PM -0700, Piet Delaney wrote:
> > > > >
> > > > >
> > > > >>Ton, George, Amit, et. al:
> > > > >>
> > > > >>If CONFIG_KGDB isn't defined, the kernel should be exactly the
> > > > >>same as when kgdb isn't integrated. So shouldn't we should add 
> > > > >>#ifdef CONFIG_KGDB in the traps.c code where the traps are 
> > > > >>initialized and panic notify is registered. 
> > > > >
> > > > >
> > > > >No, because that makes the code look way too ugly.
> > 
> > I agree that it's ugly, but until stock kernel wants to
> > do it this way I suspect that we will have an easier time
> > in being assimilated into the kernel.org repository by
> > being non-invasive.
> 
> If you need any ifdefs in core code you're doing something
> wrong. Any needed support should be integrated in a clean
> non preprocessor way.

There are zillions of preprocessor hooks for debug printfs;
surely your not suggesting they be removed. IMHO we should
use preprocessor hooks more for debugging. For example converting
'static inlines' to static in subsystems your working on and
making the stack larger since taking out inlines can increase
the stack size.

Pascal was a language that came out at the same time as C
but didn't have preprocessor hooks. There was a Pascal
kernel but I never took to it. I liked the power that preprocessor
hooks provides. Gosling also intentionally didn't provide
preprocessor support in Java but the Java OS never really
took off at Sun.

I think debugging is an excellent case were the existence
of preprocessor can be made. We abuse the preprocessor
useing #defines where enumerations should be used and we
use empty structure declarations instead of getting our
#include files organized cleanly. I find these abuses
of the C language much more offensive than using cpp
for debugging hooks.

> 
> The only patch should be to Makefiles with an option 
> to build the whole thing out of tree for modular loading too.

Then you can't use kgdb for early startup. kgdb has for 25
years now been available for debugging the complete kernel.
I'm not convinced we should break with that tradition. In fact,
IMHO we should be supporting kgdb down in the hypervisor;
'crash' just isn't as useful as live debugging. Debuging with
printf's is 1960's technology.
 
-piet

> 
> -Andi
-- 
Piet Delaney                                    Phone: (408) 200-5256
Blue Lane Technologies                          Fax:   (408) 200-5299
10450 Bubb Rd.
Cupertino, Ca. 95014                            Email: [EMAIL PROTECTED]

Attachment: signature.asc
Description: This is a digitally signed message part

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Kgdb-bugreport mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/kgdb-bugreport

Reply via email to