Re: [PATCH] Athlon/Opteron Prefetch Fix for 2.6.0test5 + numbers

[Jamie Lokier]

Andi Kleen wrote:
> > My point is simply that non-zero base GDT segments are possible in
> > userspace, and whatever code you add later to add the base should
> > be aware of that.
> 
> I don't see how a non standard GDT is possible in user space. The GDT 
> is only managed by the kernel. 2.6 offers to change it for NPTL, but
> that only applies to data segments.

I don't see anything in set_thread_area() which restricts them to data
segments.  Btw, NPTL isn't the only userspace code which uses this
function.

Admittedly it would be quite odd to use set_thread_area() to create
and use a GDT code segment - but given that you can do it, it would be
_really_ odd if prefetch instructions gave spurious faults in these
segments yet were fixed up elsewhere.

> In vm86 mode the user can load a different base without LDT, but that
> should not matter here (although it may be better to check for VM86 mode
> too) 

Shouldn't spurious faults in prefetch instructions in vm86 mode be
fixed up too?

-- Jamie
-
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/