[Cbe-oss-dev] [PATCH] spufs: marker-based tracing facility

[Julio M. Merino Vidal]

On 06/12/2007, at 9:23, Christoph Hellwig wrote:

> On Mon, Dec 03, 2007 at 04:14:00PM +0100, Julio M. Merino Vidal wrote:
>> Does your approach have any advantage over using, e.g. SystemTap?  If
>> not, I'd advocate using this tool instead, given that you also
>> mention that using lttng in the long term should be preferred (but I
>> don't know how hard it would be to extend it to use these markers).
>
> My approach just works :)  Systemtap is an out of tree codebase  
> that is
> both a complete mess and poking into kernel internals so updating it
> to new or heavily patched kernels is a nightmare.  I have no  
> intentions
> of supporting it.  Long term I'd prefer to get rid of sputrace.c and
> replace it with something like lttng once that gets merged.


As far as I know, systemtap "just works" also.  It is true that to  
get the markers working you need to apply some patches, but you need  
those anyway for lttng.  In fact, the patches needed to get the  
markers working in systemtap come from lttng.  And, at last, it looks  
like they would like to merge those patches very soon now (if it is  
not 2.6.24, it will be 2.6.25), and things will be better in that  
case because tracking markers is much easier than poking into kernel  
internals.

But oh well, if you plan on sputrace.c to die anyway, this may be  
revised in the future when those patches are merged.

-- 
Julio M. Merino Vidal <jmerino at ac.upc.edu>