On Sat, Jul 27, 2002 at 02:54:01PM +0200, Manuel Bouyer wrote:
> On Sat, Jul 27, 2002 at 01:33:06PM +0300, Jukka Marin wrote:
> > I guess 1.6 works Just Fine for everyone else?
> 
> No, 1.6 has problems on older sun4c (the ones with the "software flush"
> caches). There have been several threads here about it.

Ah.  Well, maybe it'll get fixed, then.  As it is, my SS1's are pretty much
useless.

> Attached is a patch that makes my sun4c solid again, but
> 1) I have no idea why cache_flush_segment() doesn't do its job here.
>    It could be a hardware bug with the ASI_FLUSHSEG instruction, but this
>    sounds a bit strange.

Is this a "new" thing, implemented after 1.3.3? :-)  Under 1.3.3, I had no
such problems.

> 2) The machine is really, really slow compiling things. It may be the new
>    gcc, but I will profile the kernel anyway, as the machine seems to
>    spend most of its time in kernel, according to top.

I noticed this, too.  (It's been two days and all I have been able to
build was tcsh.. gcc is dying and the machines crawl.. and top reports
the system is eating 50% of CPU time almost all the time.)

(Not to mention the time required to generate ssh keys.. not seconds, not
minutes, but more like hours.)

> You can cross-compile a kernel from another machine;

I guess I should learn to build sparc kernels on i386 (I have no sparcs
other than SS1's, sigh).

> but I've put the
> kernel I use on my ss1+, which should also work for you, at
> ftp://antioche.lip6.fr/pub/bouyer/netbsd-SUN4C.gz
> Note that this is not GENERIC, but a stripped down kernel.

Thanks, I'll give it a try!

  -jm