Tweet-sized explanation of Foreshadow: Intel chips with speculative execution erroneously use OS-private bits of not-present page table entries as physical addresses, allow reading L1-cached data from resulting location via timing side-channel.
-
-
There is also no way society is going to give up speed gains for the sake of security. As
@securelyfitz put it to me some months back. Spec ex is inherently a side channel; it's bad we took so long to realize it, but we have to live with it if we want forward progress. -
The speed gains are marginal, and negative if you're measuring performance per watt or per area. If the Centerton microarchitecture had not been abandoned and had been heavily invested in, we'd likely have in-order, non-speculative chips competitive with Spectre-hell.
-
The only real saving grace of your proposal is that in practice, in order CPUs can be usable desktop systems (source: I have an in order Atom netbook that's still useable). But usable != pleasant.
-
They'd be a lot better than just "usable" if all of Intel's, AMD's, and ARM's R&D budgets over the past 15 years had been poured into them rather than into speculative foolery.
End of conversation
New conversation -
-
-
That's the reality. If speculative execution can not be truly consequence free -- not just "consequence free" in terms of easy and obvious measurement/detection, then it probably has to go in the entirety. Now it's obvious that there's value in finding the non-obvious effects.
-
Of course... If it's truly consequence free, that essentially must mean that it doesn't speed anything up, either. Isn't any version of speeding things up via spec ex a consequence of spec ex? So we're probably back at simply: "It just has to die."
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.