Re: [9fans] porting p9p to Google's NaCl

["André Günther" <Andre.G@gmx.de>]

> Having looked into this a bit, it appears to be more like 9vx than  
> QEMU  (uses segmenting capabilities), that is, unless I've forgotten  
> how 9vx is set up.
>
> I believe some X11 stuff also uses segment registers, as well as TLS  
> on Linux.
>


TLS and the kind of segmentation 9vx and NaCl are using are two pairs  
of shoes.

For every memory related operation in 32bit x86 there's a specific  
register responsible for being part of the address translation. (eg.  
%ds for various mov instructions and %ss for stack operations...)
However there are two "free" segment registers, namely: %fs and %gs.
So how can we possible use them? The x86 has the ability to prefix  
operations in a way that it uses another segment register instead of  
the usual one for translation.
This is how TLS is done. Either %gs or %fs are setup so that a  
prefixed instruction will use the TLS address range.

However the way segmentation is used in 9vx and NaCl is a whole other  
story. (What I am now saying me be specific to 9vx and differ on NaCl,  
but the general principle should be the same)
Here the segment registers like %cs, %ds, %es and %ss are directly  
altered (and given another segment description) so most of the memory  
instructions can just be used without further host->guest code  
translation (as long as the guest code doesn't use special  
segmentation stuff himself).

The basic idea is: when segment translation is done anyway, let's  
exploit it and get guest code memory translation/access restriction  
for free)

Hope that clears it a bit and i wasn't too much of a noise for the list,
best regards,
André