> Looks pretty good, though I haven't tried it.
> 
> One thing I've noticed by inspection is that we are using ldr(t) + masking 
> for in_short.  This may start to fail on future kernels 'cos we may well 
> turn on strict alignment checking.  Can you fix that bit to use ldrb(t) 
> plus an orr?
> 

Err, forget that.  It isn't the bit of code I thought it was ;-(

However, I wouldn't recommend the use of swp except when a locked transfer 
is really needed --  it can have nasty cache implications.

R.