On Mon, 16 Dec 1996, Mark Brinicombe wrote:

> >Assuming the new kernel runs, will it be a bit faster on a StrongARM than
> >sa110-4740? In fact I feel that the context switch and especially the process
> >creation is extremely slow, running an sh script that used some [ ] (ie.
> >'test') was so slow to be completely unusable. I have to admit that RiscBSD
> >in general use seemed faster on my old ARM610 than on my new StrongARM...
> 
> Ok one of the problems with the earlier SA110 kernels was a bug in the pmap
> code that meant multiple mappings of a page could be uncached. This bug existed
> on the 610 but was not spotted. On the SA110 where the uncached performance is
> worse that ARM2 performance this bug became apparent.
> The result is that executing multiple copies of a binary simulatenously (inc
> forking) could result in various code pages being uncached.
> 

  Ahh!

  I wondered why when I was compiling two things simultaneously one of 
them always *crawled*!  Also playing two xanims simultaneously made one 
crawl along too!

   Didn't think anything of it at the time!

Ale.