Re: [ANNOUNCE] New utility: 'trace'

[Tom Zanussi]

On Wed, 2010-11-17 at 15:37 +0100, Thomas Gleixner wrote:
> On Wed, 17 Nov 2010, Peter Zijlstra wrote:
> 
> > On Wed, 2010-11-17 at 15:11 +0100, Peter Zijlstra wrote:
> > > The idea was to not let the filter engine work on the trace data (once
> > > its gathered) but on the trace argument right at the beginning of the
> > > tracepoint callchain, since some of the trace data is an expression of
> > > the trace argument (say next->prio instead of next), the trace
> > > expression wouldn't stay invariant, you'd have to write a different
> > > filter for the same effect.
> > > 
> > > So I think it would be wise to make this change sooner rather than
> > > later. 
> > 
> > Also, I see a lot of overlap with the dynamic probes stuff which needs
> > the help of magic dwarves to come up with the right data to gather.
> > 
> > Merging the dynamic tracepoint and filter stuff would be nice, there's
> > no way you can express next->prio without the help of these short
> > buggers with large axes.
> > 
> > The trouble is that the dynamic tracepoint stuff is privileged for good
> > reasons, try next+0x1000000 and you're out in the woods, priv. only
> > filters otoh just doesn't sound too hot.
> > 
> > Another nasty thing is that you actually need to have these dwarves
> > present, which means having the -debug package installed.
> 
> That sounds utterly insane for the basic use case where you trace
> application context. There is no point to filter out the tracepoints,
> really. Postprocessing can do that just fine.
> 
> I consider myself a power user of tracing, but hell I never used any
> of those filters in a real use case. awk, grep, scripting languages do
> just a better job and you don't miss any data just because you got
> your filter expression slightly wrong. Postprocessing always wins in
> that regard.
> 
> The only reason why these filters might make sense is to reduce the
> trace volume on system wide traces in production environments, but
> that's a completely different story. These scenarios can do with the
> dynamic tracepoint stuff and the custom filtering w/o putting the
> burden on the majority of users.
> 

Yeah, in my mind, the main point of the filters was to be a 'blunt
force' instrument preventing userspace from being overwhelmed by events.
The real filtering would happen in userspace with e.g. real scripting
languages.

Tom 

> Thanks,
> 
> 	tglx


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/