Re: Kernel marker has no performance impact on ia64.

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Peter Zijlstra (peterz@infradead.org) wrote:
> On Thu, 2008-06-12 at 12:43 -0400, Frank Ch. Eigler wrote:
> 
> > This is another reason why markers are a nice solution.  They allow
> > passing of actual useful values: not just the %p pointers but the most
> > interesting derived values (e.g., prev->pid). 
> 
> Useful to whoem? stap isn't the holy grail of tracing and certainly not
> the only consumer of trace points, so restricting the information to
> just what stap needs is actually making trace points less useful.
> 

LTTng also currently relies on the markers identifying useful data and
ideally won't contain any tracepoint-specific code to extract subfields
from structures.

By the way, a point that should be clarified is that putting a 
"prev->pid" field in a marker does not make this variable live unless
the marker is activated. When disabled, the marker site jumps over the
stack setup, including any pointer dereference.

Mathieu

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68