Scanning through the changes, it doesn't look like they'll cause any problems
for us.  We might want to put some effort into getting the 
idle loop stuff cleaned up to support more CPU, too..

I'm not sure about the pmap.c change either.  I had similar code in at
one point, but never saw it get exercised.

	-castor

On Tue, 14 Sep 1999, Noriyuki Soda wrote:

> jonathan> Can you send me, Simon, and Michael a pointer to the patched
> jonathan> files, just to check that the known problems dont get any
> jonathan> worse?
> 
> I've put the patch to mips on
>    ftp://ftp.jp.netbsd.org/pub/NetBSD-jp/hpcmips/binary/mips-0914.diff.gz
> 
> This file contains a patch to pmap.c, but Takemura-san will not commit
> the part (patch to pmap.c) to NetBSD cvs tree, because he and
> Shinohara-san are not sure about the change.
> 
> You might also think that some #ifdefs should be changed from
> defined(__hpcmips__) to something more appropriate, but I think
> that changeing it after importing hpcmips is better to make
> the change consistent, because in that case, we can change both
> mips part and hpcmips part at once.
> (Actually, I'm the one who'd like to change a #ifdef-name in locore.h
>  to other name, because that is needed in arc-port, too.)
> 
> jonathan>3. Soda-San just told me this port fixes several bugs in handing
> jonathan>  L2-cache-less, virtuall-indexed-L1-cache, mips cores.
> jonathan>  *Very* good news -- Michael Hitch and  I have been
> jonathan>  wrangling with that for years, to no avail, since we didnt
> jonathan>  have a no-L2-cache machine up and running.  Are the changes
> jonathan>  known to still work on a mips3 which does have an L2 cache?
> 
> castor> We hadn't seen any problems here on a QED 523x, but our set of
> castor> device drivers were very primitive.  I never got further feedback
> castor> from Soda-san about where the problems might lie.
> 
> Oops, I'm sorry that I didn't send the patches to you.
> I think patches to locore.h, db_interface.c and mips_machdep.c are
> trivial, although the patch to mips_machdep.c about mips_CacheAliasMask
> is serious. (and this bug was made by me on last summer. :-<)
> Also, I believe the patch to trap.c about mips3_intr_cycle_count
> doesn't need #ifdef __hpcmips__, I suppose that the reason this line
> is removed on revision 1.99 of our tree is just accident, isn't it?
> 
> Please note that VR41xx's minimum pagesize is 1K byte (not 4K byte),
> #ifdefs about MIPS3_PG_SHIFT and default_pg_mask are needed because
> of this.
> 
> Could you examine the patch file until this weekend?
> --
> soda
>