Subject: Re: True 64bit OS ?
To: None <>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
List: current-users
Date: 07/07/1997 14:46:42
Ted Lemon writes:

>> I'm writing some info about NetBSD to my WWW page and am in tempting
>> to say NetBSD is true 64-bit on 64-bit architectures. Is this
>> statement true ? I feel the sentence "true 64-bit" a bit vague ;-)
>> What does it mean anyway besides ability to address more memory ? 

 >I think this is not such a great idea - it's sort of like bragging
 >that a car is a True Vehicle.  The Alpha is a 64-bit-only
 >architecture, so anything that runs on the Alpha is by definition a
 >64-bit OS.  This isn't a selling point, so why bring it up?

Err, NT on Alpha is limited to 32-bit addressing.  It does use 64-bit
registers, though.

 >The only platform I know of where this distinction might be
 >interesting is the MIPS R4000 and later processers, which actually
 >have 32- and 64-bit modes.

Mips CPUS (mips3 and newer) are even more complicated...

The FPU has a mode bit that says whether FPU registers are 32 or 64
bits.  IEEE doubles always use 64 bits, but in 32-bit mode they use a
register pair.  64-bit FPU mode can be turned on even when the integer
CPU is used in 32-bit mode, and GCC has an option to generate code
that uses this feature.

Support for 64-bit FPU (not integer!) registers _may_ get added for
NetBSD 1.3.  AFAIK, it's just allowing that bit to be set from
userspace, and doubling the size of the FPU context and FPU state in
setjmp() buffers.

I haven't even begun to look at either using 64-bit integer registers
in a 32-bit address space. TLB support for 64-bit virtual addresses.

 >Maybe Jonathan Stone or Per Fogelstrom
 >can speak to the state of the implementation on those CPUs...

NetBSD is still purely 32-bit mode.  I beleive OpenBSD is also, but
anyone who cares should really ask Per.