ppc multlibs and BSP removal was Re: powerpc altivec support

Sergei Organov osv at topconrd.ru
Tue Feb 15 13:02:17 UTC 2005


Eric Valette <eric.valette at free.fr> writes:
> Sergei Organov wrote:
> 
> > Overall, there is nothing unique about PowerPC but the way things are
> > currently implemented.
> 
> I totally disagree : things are implemented the way they are today
> because old exception code was so broken and had so many feature in
> ifdefs (no more used and not even compiling e.g gcc272, ...) that
> modifying it for all platform was something I refused to do.

I said nothing about *why* the code is as it is now, so you obviously
"totally disagree" with somebody else. Do you insist PowerPC as
architecture has some unique [mis]features that make unified approach to
exception management impossible? If so, could you please name them?

As for your refusal to modify the old code, the result is that instead
of fixing existing problems you effectively replaced one set of problems
with another, -- that's usually the case when somebody blindly
re-implements something from scratch without even trying to understand
why things have been implemented this way in the first place. And no,
I'm not an author of initial PowerPC port, so there is nothing personal
here.

> Now, I think motorolla_shared supports by itself more platform than
> you listed for old exception processing.

Maybe the support for Motorola VME boards is perfect, I don't know and
don't care, but the lack of unified PPC support is still a problem
anyway. Having separate ppc_asm.S for every BSP is simply a disgrace
that has no excuses, sorry.

> And BTW, the same exception model is used on Ix86 and the code is
> shared.

Wow! If the code is indeed shared it's not the same model. The "new
exception model" as seen in PowerPC is just "every BSP does whatever it
wants w.r.t. exceptions". Effectively it makes every PowerPC BSP a
separate RTEMS port.

> The problem was that the powerpc code and the old BSP were to
> different to unify anything.

... and you threw out the child along with the bath :(

Your reluctance or inability or lack of time or whatever to implement
unified approach doesn't mean none is (was) possible.

> And BTW do anything you want because I will no more use RTEMS on PPC.

Cool! Isn't it because it's a pain to use and support? ;)

> So make any modification you want provided you support them.

Didn't know I needed your permission for modifications before you
stopped using it.

-- 
Sergei.




More information about the users mailing list