> Ted Unangst wrote:
> > Otto Moerbeek wrote:
> > > Sparc64 and powerpc also have speculative execution, branch
> > > prediction and extensive caches. It is much wiser to assume they are
> > > also affected by (similar) bugs/explots or whatever you call it.
> >
> > A lot of the commonly available sparc64 gear, T2 and USIII, are in order.
>
> So that's not exactly right, since branch prediction and speculative exection
> are somewhat orthogonal to out of order execution. I'm assuming they are
> somewhat less affected because without all the fun out of order machinery the
> CPU won't go nearly so far down the branch.
>
> Specifically regarding meltdown, sparc64 doesn't map the kernel in userland,
> right?
sun4u have a split U vs S I/Dtlbs. Those TLB are software-updated.
I'm not entirely sure of the sun4v situation.
> Regarding spectre, the bounds check bypass requires executing past two
> branches. Can a sparc do that?
doubt it. Note since the tlb are software-updated if a branch
prefetched an unmapped page and that page translation didn't exist,
we'd have seen seen spurious speculative faults in the past, because
our software-update codepath isn't a willing party to such an
emmisions scandal.
No comments:
Post a Comment