> next is memset(), which is mostly being called from genfs_getpages()
> to zero pages we're allocating for the file.  this is pretty silly,
> but I have some ideas on how to avoid it.  I already knew about this
> and it's on my list of things to fix.

Page zeroing is the source of fluctuation in NetBSD/mips.  It's
impacted directly by processor cache design.  Tsubai once mentioned me
that he could drew a performance gain using particular PPC insns for
page zeroing.  It'd be benefitial somehow to have zero paging/copy
paging 'va' as well as 'pa' combined with a specification (or
suggestion) of cache attribute desirable for operation.

Tohru Nishimura