Re: [PATCH 58/61] dynamic debug: combine dprintk and dynamic printk

[Jason Baron <jbaron@redhat.com>]

On Wed, Mar 25, 2009 at 11:50:23AM +0100, Ingo Molnar wrote:
> * Greg Kroah-Hartman <gregkh@suse.de> wrote:
> 
> > From: Jason Baron <jbaron@redhat.com>
> > 
> > This patch combines Greg Bank's dprintk() work with the existing 
> > dynamic printk patchset, we are now calling it 'dynamic debug'.
> > 
> > The new feature of this patchset is a richer /debugfs control file 
> > interface, (an example output from my system is at the bottom), 
> > which allows fined grained control over the the debug output. The 
> > output can be controlled by function, file, module, format string, 
> > and line number.
> 
> Hm, dunno this overlaps quite a bit with existing tracing 
> facilities.
> 
> There seems to be serious consolidation potential here, if you want 
> to reduce code and want to cooperate with existing tracing 
> facilities.
> 

we're definitely interested in consolidation...i had posted an optional
flag for toggling to use trace_printk() instead of printk() before...

> In particlar, this facility seems to be a subset of what is already 
> possible via trace_printk(). We could make every dprintk() site show 
> up in /debug/tracing/events/dprintk/, and make it possible to toggle 
> them on/off individually.
> 
> Furthermore it would also be possible to use the user-defined per 
> tracepoint filter expression facility. That way each dprintk site 
> can be (optionally) tailored individually.

we currently can toggle them individually, but not by pid.

> 
> For example we could enable such use:
> 
>   echo 'pid == 1234' > /debug/tracing/events/dprintk/fs/lockd/svc4proc.c:71/filter
>   echo 1             > /debug/tracing/events/dprintk/fs/lockd/svc4proc.c:71/enabled
> 
> To tailor the following dprintk() in fs/lockd/svc4proc.c (line 71):
> 
>   static __be32
>   nlm4svc_proc_null(struct svc_rqst *rqstp, void *argp, void *resp)
>   {
>           dprintk("lockd: NULL          called\n");
>           return rpc_success;
>   }
> 
> To only be printed for PID 1234.
> 
> Per subsystem filters would be available too. The subsystem 
> maintainers dont have to do anything but place dprintk() calls - 
> they'll show up under /debug/tracing/events/dprintk/ automatically.
> 
> Would you be interested in this?
> 

I don't think we want a tracepoint for each of these sites. note that
this facility ties into 'pr_debug' and 'dev_debug' as well...however i
think what you're suggesting is perhaps a new type of event that can
handle this case. Instead of register/unregistering to enable the event,
we can simply set a variable that is associated with each call site. is
this what you are suggesting? If so, i think this would be a nice way to
go...I would however, perhaps want an optional flag that makes these
into printk() and not trace_printk() for certain situations...

With respect to 2.6.30, I'm away next week, so maybe we can keep what we
already have for 2.6.30, and look at this integration for 2.6.31?

thanks,

-Jason