AMD64: LDTSTOP vs hard reset

[Stefan Reinauer]

Hi,

In addition to lowering the number of resets it would also be possible
to use an LDTSTOP (as I implemented it in older versions of the CVS) 
at least when doing the hypertransport link changes. 
I know that this does not work with some preproduction CPUs, but I don't
think that would really matter. BUT: Could we get rid of all CPU resets
when doing an LDTSTOP assertion (It's a lot quicker than a hard reset)
or would we still need to reset the CPU for some other reasons?

Stefan

Stefan Reinauer  writes:
> Hi,
>
> In addition to lowering the number of resets it would also be possible
> to use an LDTSTOP (as I implemented it in older versions of the CVS) 
> at least when doing the hypertransport link changes. 
> I know that this does not work with some preproduction CPUs, but I don't
> think that would really matter. BUT: Could we get rid of all CPU resets
> when doing an LDTSTOP assertion (It's a lot quicker than a hard reset)
> or would we still need to reset the CPU for some other reasons?

When I was working with it I remember seeing some additional remaining
errata with regards to LDTSTOP.  So I decided not to mess with it.
There does seem to be a big delay in the phase that finds the rom chip
though.

Eric