At Thu, 22 Feb 2001 23:32:02 +0900 (JST),
Izumi Tsutsui wrote:
> > I've put diffs to current here:
> > 	http://www.a-r.org/~ur/arcdiff0216.gz
> 
> Now I've checked this and have some comments again :-)
> 
> 1) I know there are a lot of work to make this really MI, but
>    it is still better to split this driver into attachment and core
>    like sys/arch/mvme68k/dev/{ncrsc_pcctwo,siop}.c .

Yup, it's on my TODO list.

> 2) You said it was difficult to share the script between 700 and 710,
>    but is it possible to share other part? If it is, `osiop' is
>    the best name for this device.

Other part is heavily depends on scripts.
Actually 710 may use 700's script, but it can't use any 710
enhanced features.

If we make different driver for 700 and 710, 'oosiop' is fine for 700?

> 3) There are some htole32()/le32toh() macros in arcsiop.c,
>    but current amiga and mvme68k siop.c have no byteswap instructions.

Ahh, byte swapping was not neeeded for scripts. I'll fix this.

>    It looks data bus of the devices on these 68k ports are hardwired
>    in "swapped" order, or lower 2 bit of address bus are reversed.
 :
> Anyway, we need re-think how to handle these endianness. Umm..
> 
> Maybe amiga and mvme68k require more complecated bus_space(9)
> structures in the perfect world ;-)

7xx doesn't swap byte lane on both endian mode.
I'm not sure about other bi-endian devices, but if it is 7xx unique,
I think driver should take care.

---
Shuichiro URATA
ur@a-r.org