Re: Relicensing tracepoints and markers to Dual LGPL v2.1/GPL v2,headers to Dual BSD/GPL

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

* Arjan van de Ven (arjan@infradead.org) wrote:
> On Mon, 26 Oct 2009 09:17:49 -0400
> Pierre-Marc Fournier <pierre-marc.fournier@polymtl.ca> wrote:
> 
> > Ingo Molnar wrote:
> > > 
> > > But i also disagree with it on a technical level: code duplication
> > > is _bad_. Why does the code have to be duplicated in user-space
> > > like that? I'd like Linux tracing code to be in the kernel repo.
> > > Why isnt this done properly, as part of the kernel project - to
> > > make sure it all stays in sync?
> > > 
> > 
> > If you mean that this code should solely be used inside the kernel,
> > then what you propose technically does not work. There is a very high
> > cost to accessing kernel code from userspace. 
> 
> yeah 100 cycles is insanely high, that's at least the equivalent of...
> say one cache miss.
> 

Hi Arjan,

Maybe it's just me, but if you are talking of system calls performed
with sysenter, my own machine (with an Intel Xeon E5405 CPU) seems to
disagree with your estimations:

For a system call doing basically _nothing_:

Cache-cold system call:
Time for a system call (returns after a simple test) :
TSC diff : 6426 cycles
time diff : 3212.53 ns

Cache-hot system call:
Time for system call (10000 calls) (returns after a simple test) :
TSC diff : 599.893 cycles per call
time diff : 299.903 ns per call

Otherwise, if you are instead referring to a vDSO, then the contexts is
slightly different, and it becomes cheaper, but it does not allow to
share code as much between the in-kernel and vDSO implementations,
mainly due to different locking required. Moreover, I wonder about the
maximum vDSO size that can be ported across multiple architectures. It
has only been used for really tiny pieces of code so far.

Thanks,

Mathieu

> 
> -- 
> Arjan van de Ven 	Intel Open Source Technology Centre
> For development, discussion and tips for power savings, 
> visit http://www.lesswatts.org

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