2009-04-28 21:27:53

by Dragoslav Zaric

[permalink] [raw]
Subject: Debug only with printk ?

Hello,

I am relatively new in linux kernel and since debugging is key
element of programming, I am wondering is it really necessary
to use various debugging tools for kernel debugging ?

Is simple printing with printk enough to debug any kind of kernel problem,
and if not, what are the cases when printk is not enough ?

Also, is there possibility when using debugging tools to affect
kernel environment, because debugging tool will also take cpu time and
memory resources ?

thanks

Dragoslav Zaric
[Programmer; M Sc Astrophysics]


2009-04-29 03:15:50

by Valdis Klētnieks

[permalink] [raw]
Subject: Re: Debug only with printk ?

On Tue, 28 Apr 2009 23:27:42 +0200, Dragoslav Zaric said:
> I am relatively new in linux kernel and since debugging is key
> element of programming, I am wondering is it really necessary
> to use various debugging tools for kernel debugging ?

'git bisect' and friends are your friends if you're trying to track down
a bug in somebody else's code.

It's not *mandatory* to use other tools, but it certainly helps.

> Is simple printing with printk enough to debug any kind of kernel problem,
> and if not, what are the cases when printk is not enough ?

Usually, printk's are sufficient *if* you can use them. Times when this
may not be true:

1) Debugging an early-boot crash, before even the earlyprintk stuff is alive.

2) "Add a printk, recompile, reboot" gets pretty boring pretty fast, so
it may be a *lot* faster if you can use one of the kernel debugging/tracing
tools to do "Well, maybe we should check this... or this.. or wait, how
about *this*?" testing.

3) If you're trying to debug an issue on a production server, you may not
have the *option* to do a lot of compile/reboot cycles, and be forced to
use whatever debug/trace support is present in the already-running kernel.

> Also, is there possibility when using debugging tools to affect
> kernel environment, because debugging tool will also take cpu time and
> memory resources ?

If I had a <monetary unit> for every time when even adding a printk call
was enough to change the timing and cause a Heisenbug to change its behavior,
I'd have a lot of monetary units.

Yes, debugging tools and printks will change code and data offsets and timing
so data overlays and race conditions change or even dissapear/reappear.


Attachments:
(No filename) (226.00 B)

2009-04-29 17:16:12

by Lennart Sorensen

[permalink] [raw]
Subject: Re: Debug only with printk ?

On Tue, Apr 28, 2009 at 11:27:42PM +0200, Dragoslav Zaric wrote:
> I am relatively new in linux kernel and since debugging is key
> element of programming, I am wondering is it really necessary
> to use various debugging tools for kernel debugging ?
>
> Is simple printing with printk enough to debug any kind of kernel problem,
> and if not, what are the cases when printk is not enough ?
>
> Also, is there possibility when using debugging tools to affect
> kernel environment, because debugging tool will also take cpu time and
> memory resources ?

So far I have used printk's, parsing of the OOPS when I do something
stupid, educated guesing looking at the code following an OOPS or other
problem, and in a couple of cases using firescope. I have used firescope
to monitor the DMA ring buffer for the pcnet32 driver once.

--
Len Sorensen