Re: [uml-devel] [PATCH 8/10] UML - Maintain own LDT entries

[Bodo Stroesser]

Blaisorblade wrote:
> On Monday 07 November 2005 13:20, Bodo Stroesser wrote:
>
> > Blaisorblade wrote:
> >
> > > On Monday 31 October 2005 05:39, Jeff Dike wrote:
> > >
> > > > From: Bodo Stroesser <bstroesser@xxxxxxxxxxxxxxxxxxx>
>
>
> > > Or at least so I think (I must still give a proper look afterwards, and
> > > I'll post patches). Actually it seems that this is done on purpose, but I
> > > don't agree too much on this. I will see.
>
>
> > From the beginning my new code for SKAS included the checks/buffering you
> > later inserted for TT and SKAS. So this patch is a second version adapted
> > to your changes. It shifts your improvements into TT path only (where I
> > didn't do any changes in my old patch), while it uses my own stuff for
> > SKAS. Thus the patch doesn't really revert your improvements, but restricts
> > it to TT. As in SKAS0 UML now holds its own LDT data, there is no need for
> > buffering in this case. So I think it makes sense to have separate code for
> > SKAS.
>
> Yep, ok - I'm undecided about the new code for SKAS3, but it may make sense 
> (i.e. no opinion).
>
> Instead, I have another question: is there a proper reason for using the 
> processor format for storing the info and translating it back to (struct 
> user_desc)? I am planning to avoid this double translation because I don't 
> like it. Any opinion?

In my opinion there is no reason to change the current implementation for SAKS3/0.
Note: if someone reads LDT via [sys_]modify_ldt(), he will receive the requested
data in "processor format", that is LDT-descriptors. He will receive a list of
descriptors starting at the first descriptor of the LDT, thus no entry number is
needed in the enties.
The only case that uses user_desc is when writing one desriptor via modify_ldt().
modify_ldt(WRITE) exactly writes one LDT-descriptor, so user_desc must contain
the number of the entry to write. Thus user_desc is bigger than LDT descriptor.
It also uses an other data layout resulting in double the size of LDT-descriptor.
So I think it doesn't make sense to store user_desc. We save memory storing the
resulting LDT-descriptors, which then are copied transparently on modify_ldt(READ).
Conversion between user_desc and LDT-entry is done on modify_ldt(WRITE) in SKAS0
only. No other conversions are done in UML.

	Bodo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/