Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 4 Apr 2006 22:48:45 +1000
From:      Andrew Reilly <andrew-freebsd@areilly.bpc-users.org>
To:        Scott Long <scottl@samsco.org>
Cc:        freebsd-amd64@freebsd.org
Subject:   Re: Status of NX bit support.
Message-ID:  <20060404124845.GA37816@gurney.reilly.home>
In-Reply-To: <44319240.8070203@samsco.org>
References:  <44301C6D.3010206@rogers.com> <200604031442.43477.jhb@freebsd.org> <44318E3F.6080808@rogers.com> <20060403211943.GA99241@troutmask.apl.washington.edu> <44319240.8070203@samsco.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Apr 03, 2006 at 03:23:12PM -0600, Scott Long wrote:
> Steve Kargl wrote:
> >You're joking, right?  How many registers are available for the
> >i386?  How many registers are available to an AMD64 cpu in 
> >64-bit mode?
> 
> You also get less efficient cache utilization due to the wider data
> types that are in use.  It seems to be mostly a wash between the
> advantages of more registers and the cost of lower cache efficiency.
> amd64 is nice when you need more kernel address space and/or more
> process address space.

I know I'm probably being barking mad, here, but what I'd really
like for my AMD64 workstation system is to be able to run it in
AMD64-mode, but with 32-bit pointers.  I vaguely remember
reading some early AMD bumpf that that could be a supported
configuration, somehow.  Seems like it should just be a compiler
switch, to specify 32-bit loads and stores for pointer values,
and probably some checking in the pmap system to make sure that
nothing is mapped outside the 32-bit range...

Anyone know if it's been done?  How it performs, if it's been
done?

I do a fair bit of DSP simulation work, and 64-bit long-longs
are very handy to have go fast, and more registers and
register-based calling conventions are always a good thing.  I
have no need for 64-bit addresses, at least at the process
level, and I suspect that most workstation users would be in
much the same boat...

Didn't MIPS, Sun and Apple support such an arrangement on
R4000+, SPARCv?8 and PowerPC970 (G5) systems?

Cheers,

-- 
Andrew



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20060404124845.GA37816>