RE: [Xen-users] 32bit or 64bit?

["Petersson, Mats" <Mats.Petersson@xxxxxxx>]

> -----Original Message-----
> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx 
> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> Ulrich Windl
> Sent: 02 March 2007 08:27
> To: xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: RE: [Xen-users] 32bit or 64bit?
> 
> On 1 Mar 2007 at 15:11, Kraska, Joe A (US SSA) wrote:
> 
> > 
> > PAE (which all modern x86 cpus have) will support up to 
> 64GB physical.

In Xen, some of the memory address bits are used for other purposes,
which limits Xen's usable memory in PAE to 16GB [I'm not sure EXACTLY
what causes this restriction, but I'm 99.999% sure that this is the
limit for 32-bit PAE]. 

> > 
> > Main restriction is that for an application to support more than 4GB
> > of addressable memory on 32 bit PAE systems, they have to 
> be compiled
> > a special way. If you don't have any processes which want to address
> > more than 4GB, 32 bit PAE will work fine for you.
> 
> Unaddressable physical memory is quite uninteresting: You 
> can't mem-map a DVD 
> image in 32bit mode, PAE or not.

Yes, and as far as I know, there's no way around the 32-bit barrier -
and even this is more like a 2-3GB barrier, as some memory space HAS to
be mapped to the kernel-space so that interrupts and system calls will
work. 

A 32-bit APPLICATION in 64-bit kernel only looses about 1% of the memory
space to Kernel space (for the shim-layer to translate from 32 to 64 bit
data structures), which means that a single application can without
problems use 3.95GB of memory, even if the app is 32-bit. In 64-bit,
each application can address more memory than any 8-socket AMD64 machine
can harbour (each socket allows up to 8 sticks of memory -> 64 sticks,
64 x 4GB per sticks -> 256GB). 
> 
> > 
> > I don't think 64bit xen is considered fully baked yet, although it
> > certainly is workable for investigation purposes (from experience).
> 
> I run SLES10 x86_64 with less trouble than those trying to 
> mess with PAE on this 
> list it seems.

I've been using 64-bit Xen for the last 12 months or more, and whilst
running xen-unstable is always a bit of a challenge, I'd say most of the
bugs that I see are not 64-bit only - the code to deal with 64-bit only
is only a little bit (startup-code, page-table handling and instruction
decoding are the parts that are different - and that's a few thousand
lines in a project that is roughly 100k lines - and bugs are generally
relatively evenly distributed in the code, so percentage-wise the 64-bit
only bugs should be only a few). 

There are some problems running 64-bit HVM guests that have to do with
handling of 64-bit instructions and simply that 64-bit guests do things
different than 32-bit guests, but those are being ironed out as soon as
we can root-cause them (the hard part is to figure out which
code-sequence causes the problem, and what's wrong with the code, which
is particularly hard when the actual problem happens thousands of
instructions later, in a piece of code that we may not have source-code
for!)

--
Mats
> 
> Regards,
> Ulrich
> 
> 
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
> 
> 
> 



_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users