@paulmclellan "we don't know how to completely protect against this type of attack without reducing processor performance to a few percent (under 5%) of what it is today". That's completely wrong. That would be the simplest and worst way to fix it. We know how to do much better.
-
Show this thread
-
Completely protecting against this simply requires being able to roll back *every* effect of a bad speculation. Have a few extra cache lines, extra branch predictor counters, that are discarded if the speculation fails, but become the real ones on success. Stall if you run out.
1 reply 0 retweets 1 likeShow this thread -
Replying to @BruceHoult
That's just the easy stuff, but nowhere near sufficient to completely protect against speculation attacks. You also must not speculate at all on anything that effects any resource that is shared with another compute thread. No main memory access, no shared caches, ...
1 reply 0 retweets 0 likes -
Replying to @oe1cxw @BruceHoult
... no hyper-threading (because then everything would be shared), no sharing of FPUs etc between cores, and so on. Because roll back isn't helping you if another thread can already see the speculation while it is happening.
1 reply 0 retweets 0 likes
But nobody knows how to generate an exhaustive list of all relevant side channels and when nobody knows how to generate such an exhaustive list then nobody knows how to build a secure processor that speculates.
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.