Interesting that they're using compiler analysis to identify at-risk code and insert fences instead of using a mitigation like retpolineshttps://twitter.com/apardoe/status/953054508984156160 …
-
I think retpolines are used for Variant 2 to avoid branch *target* prediction hacks, but aren't relevant for the Variant 1 which doesn't involve *target* misprediction, and the blog post you linked is just about Variant 1.
This. Two different bugs. This mess is all because the late-to-the-game researchers from the university decided they deserved the naming rights for some reason and then called both "Spectre".
-
-
My conjecture is that Variant 1 cannot be automatically fixed in all cases without significant performance impact. However, subsets of it can be mitigated automatically. MSVC seems to be mitigating the most obvious one (range checks) with /Qspectre. There are more, though.
-
Variant 2 *can* be completely fixed with retpolines and/or the microcode/OS fixes (specific requirements vary by CPU). Variant 3 can be fixed with KPTI or equivalent. Both are also fixable in newer silicon.
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.