Subject: Re: Prentium MMX for NetBSD-current
To: Jason Thorpe <firstname.lastname@example.org>
From: Chris G. Demetriou <cgd@CS.cmu.edu>
Date: 01/17/1997 16:15:12
> On Fri, 17 Jan 1997 13:47:49 -0500
> "Chris G. Demetriou" <cgd@CS.cmu.edu> wrote:
> > So, every time the FreeBSD VM system comes up, somebody from NetBSD
> > says that "it has portability problems," and invariably somebody from
> > FreeBSD says "no it doesn't."
> I am going on comments made to me by Matt Green, who did a fair
> amount of work to attempt a port of the FreeBSD VM code to NetBSD,
> and by others who have mentioned their intent to write
> a veneer layer to make MMUs on other architectures emulate
> the i386 MMU.
So, why don't they speak up and voice their specific concerns?
Whining about code and justifying it with hearsay is... well, not too
cool, in my opinion.
> The latter is what concerns me the most; in the
> Mach portability model, the pmap module is supposed to hide
> machine-dependencies. If this is the case, why bother emulating
> another MMU?
Uh, how about "because pmap modules are a bitch to write"?
If you're using that as a metric for concern, then you should be
concerned about NetBSD, as well. For instance, the NetBSD/alpha pmap
code effectively treats the alpha MMU like an '030 MMU (with a few
minor tweaks to deal with reference/modify emulation), because that
was (much!) easier than writing a "proper" pmap. (Eventually, i'll
get around to doing the right thing, but...)