http://lists.openrisc.net/pipermail/openrisc/2011-December/000533.html where
I first propose changing the ISA as well as the implementation.


---Matthew Hicks

On Tue, Mar 20, 2012 at 2:48 PM, Stefan Kristiansson <
[email protected]> wrote:

>  On 03/20/2012 09:14 PM, Matthew Hicks wrote:
>
>
>
>  On Tue, Mar 20, 2012 at 12:52 PM, Jonas Bonn <[email protected]> wrote:
>
>>
>> On Tue, 2012-03-20 at 19:43 +0200, Stefan Kristiansson wrote:
>> > On 03/20/2012 03:50 PM, Jonas Bonn wrote:
>> > >
>>  > > I don't agree... in the same way that you don't have to respect any
>> ABI
>> > > as long as you aren't calling into "external" code, you are free to
>> > > fiddle with r0 as much as you want.  "r0 contains 0" is ABI, it is
>> not a
>> > > hardware detail.  A bit weird, I agree, but not fundamentally
>> incorrect.
>> > >
>> > > The real problem is that r0 isn't _necessarily_ writable... so there's
>> > > no guarantee that you can actually run a program that changes r0 on an
>> > > arbitrary implementation.
>> > >
>> > >
>> >
>> > No, the real problem is that the specification is not clear about
>> > whether it should be possible to do a write to r0 without any side
>> effects
>> > (regardless if the value actually will be written or not to r0).
>> >
>> > Like it is now it's freely up to the reader to do an implementation
>> > that locks up the cpu when someone tries to write 0 to r0.
>> > Although it perhaps wouldn't be a very good implementation ;)
>> >
>>
>>  Yeah, fair enough.  I guess the ambiguity kills whatever advantage the
>> flexibility may give you (dubious though it may be).
>>
>> R Diez's text is fine... might be even better to go a step further and
>> be explicit about the requirements/limitations of r0 (i.e. writing to r0
>> is a no-op; r0 is not guaranteed to contain 0 at startup; etc...)
>>
>
>  We had a discussion about this in December 2011.  I submitted a patch
> that would basically treat any write to r0 as if the value to be written is
> 0.  Most felt that changing the ISA would be an unacceptable ABI break and
> the patch was not accepted.
>
>
>
> That's a truth with modification, isn't it?
> You proposed changing an implementation (or1200) not the ISA,
> and most felt it wasn't worth it since it could as well (and should) be
> handled in software.
> I don't think there was any discussion about unacceptable ABI breaks and
> changing ISAs.
>
> Stefan
>
>
_______________________________________________
OpenRISC mailing list
[email protected]
http://lists.openrisc.net/listinfo/openrisc

Reply via email to