On Sat, 29 Oct 2005 22:39:43 +0100
Steve Woodford <scw@netbsd.org> wrote:

> Hi,
> 
> Is there still a need for a per-process und_sp on our arm32 ports? As 
> far as I can tell, it is completely vestigial; all processes can share 
> the same und_sp set up by initarm().

I think it might be needed by ARMFPE, but then we don't really use ARMFPE any more.  (Is it worth keeping the ARMFPE stuff in tree?)

Is a shared und_sp safe for debugging?  (I guess it probably is, as I suspect we used a shared undefined stack for debugging on other platforms)

I presume you're looking for further ways to improve performance?  I know that I wondered if we could remove the und_sp stuff from cpuswitch.S, as we've got to change processor modes twice to set and read the und_sp.  Which I believe is quite slow.

Thanks,
Chris