[Cbe-oss-dev] Refactored cell powerpc oprofile patch

[Geoff Levand]

Hi Carl,

Carl Love wrote:
> I have not tested any of this code on Cell yet.  Don't know if this
> breaks anything.  But I do have a number of questions/concerns.  In
> summary, it really seems like adding support for PS 3 should not require
> such massive changes to the existing code.  After all, it is the same
> processor and you are adding the support in the same directory as the
> Cell processor.  I am very confused as to why you chose to pull cell
> code from multiple files/directories and replicate the code verbatim for
> ps3.  Seems like you should be able to use the existing code without
> creating a second identical copy.  I point out specifically where I see
> this happening below.


As I mentioned in another post, I think we can simplify what Denis has,
but the the existing cell pm code has a lot of platform specifics
throughout it, and those need to be abstracted out.  The underlying
processor is the same, but the specifics of how to use the pm support
is quite different.


> At this point, the only issue I see with PS3 support using the existing
> CELL code is the FW calls.  The CELL code makes FW calls to setup the
> debug bus and do SPU cycle based profiling.


Yes, and the plan is to abstract those out.


> We started the process
> within IBM to get approval to share the FW code with Sony and Toshiba so
> that all three platforms can share the same FW interface.  This will
> eliminate adding any kernel support to access the debug bus registers.


An interesting idea, but unfortunately, I don't think its a practical
solution.  PS3 was released 16 months ago, and we plan to support those
units.  Also, as just my guess, I don't think there is any chance to
change the existing LV1 logical performance monitor.


-Geoff