On Thu, Jul 19, 2012 at 03:52:15PM +0300, Avi Kivity wrote:
> On 07/19/2012 01:51 PM, Gleb Natapov wrote:
>
> >> > +int x86_linearize(struct x86_linearize_params *p, ulong *linear)
> >> > {
> >> > - struct desc_struct desc;
> >> > - bool usable;
> >> > ulong la;
> >> > u32 lim;
> >> > - u16 sel;
> >> > unsigned cpl, rpl;
> >> >
> >> > - la = seg_base(ctxt, addr.seg) + addr.ea;
> >> > - switch (ctxt->mode) {
> >> > + la = get_desc_base(&p->desc) + p->ea;
> >>
> >> This makes 64-bit mode slower, since before the patch it avoided reading
> >> the segment base for non-fs/gs segments, and only read the segment base
> >> for fs/gs. After the patch we always execute 4 VMREADs (and decode the
> >> results).
> >>
> > That's easy to fix by making caller prepare fake desc if mode is 64-bit
> > and segment is non-fs/gs. The question is if this even measurable?
>
> I'm sure it will be measurable, esp. on older processors. Why not
> measure it?
>
>
It is easier to just fix it :) Will do and resend if you are agree with
general approach.
--
Gleb.
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
