Re: [PATCH] printk.caller

[Peter Zijlstra <peterz@infradead.org>]

On Thu, 2008-10-02 at 16:31 -0700, Andrew Morton wrote:
> On Thu,  2 Oct 2008 16:21:15 -0700 (PDT)
> Roland McGrath <roland@redhat.com> wrote:
> 
> > This adds the printk.caller=[0|1] boot parameter, default setting
> > controlled by CONFIG_PRINTK_CALLER.  (This is modelled on printk.time
> > and CONFIG_PRINTK_TIME.)
> > 
> > When this is set, each printk line is automagically prefixed with
> > "{0x123abc} " giving the PC address of that printk call (actually
> > the PC address just after the call).
> > 
> > As a kernel hacker, I always hate having to grep for some fragment
> > of a message to find the code that generated it.  But I always have
> > my -g vmlinux handy, so:
> > 	(gdb) info line *(0x123abc - 1)
> > is real handy (it pops the source up in an Emacs buffer).
> > 
> 
> hm.  What do others think?

git grep is usually plenty fast for me, but I guess different people,
different tastes.

Also, I always use addr2line instead of gdb,.. another not-to-the-point
difference ;-)

The only real downside to this patch for me is that it potentially
increases the length of lines which means I;d have to stretch my serial
console window, but I guess others might object to the puny increase in
object size.

Flip a coin.

One tiny nit though:

> +                               char pbuf[sizeof caller * 2 + sizeof "{0x} "];

I thought we did sizeof() in-kernel.