On Mon 18 Aug 1997, Martin Husemann wrote:
> cgd wrote:
> 
> > In other words, if the device is logically presenting a 16-bit
> > register containing the value 0xabcd, it should show up as 0xabcd when
> > read with bus_space_read_2.
> 
> That sounds fine.
> 
> So the answers to my original question are: (1) no, I cannot use
> bus_space_read_4 at an arbitrary (unknown to me) position due to possible
> allignement constraints on some architecturs and (2) yes, I should use
> bus_space_read_2 to get the 16 bit signature value and expect it to be
> always the same.
> 
> Leo, I guess you should fix bus.h on port atari then?

I suppose so. At least for the isa-memory space, the I/O space is swapped
by the hardware. I postpone this until I actually have a card that needs
the byte-swaps. Both the Teles-ISDN and Accton card use 'pio'.

I also think that the documentation needs to be fixed. This whole discussion
shows that I'm not the only one who thought that endianness was not defined
in the bus-sceme. 

Leo.