Handling of alignment exceptions of load/store multiples

[Becky Bruce]

Ahhhhh..... that explains it.  Leaving it in shouldn't hurt - I'll do 
as you suggest.

"Boundedly undefined" results - my favorite!

Thanks!
-B

On Oct 17, 2005, at 11:44 PM, Paul Mackerras wrote:

> Becky Bruce writes:
>
>> I'm currently in the process of merging align.c into
>> arch/powerpc/kernel.  I noticed that there's a difference in the
>> handling of alignment exceptions involving ld/st string instructions
>> and ld/st multiple instructions between the 2 architectures right now.
>> The 32-bit code does some handling of these, while the 64-bit version
>> currently just bails out.
>>
>> Should I try to adopt a handling model for these in the merged tree
>> like the 32-bit code, or is there a reason behind not attempting to
>> handle these on the 64-bit side that I'm not aware of?
>
> I think we might as well include the code for handling them, since we
> have it.  On the 64-bit side, I think they left that stuff out because
> (AFAICT) all the 64-bit cpus handle misaligned ld/st string in
> hardware, and misaligned ld/st multiple are deprecated by the
> architecture (implementations are allowed to produce boundedly
> undefined results).
>
> Regards,
> Paul.