4 ms·
You would need to block access to high-resolution time sources. Because all an attacker needs is accurate measurement of how long code ran.
by rocqua 4y ago
You would need to block access to high-resolution time sources. Because all an attacker needs is accurate measurement of how long code ran.
- nequo 4y ago> Because all an attacker needs is accurate measurement of how long code ran. So would it thwart the attack if the kernel added a random delay to the execution time?
- IncRnd 4y agoIt wouldn't, not by itself. The attack would take more measurements to create a profile, however. Extending the time required to mount an attack is probably not sufficient to thwart an attack. It could be for some workloads, but not for all.
- ibejoeb 4y agoI think the intention here would be to provoke random jitter. So rather than trying to fight it with constant-time algorithms that turn out not to be under certain conditions, we make all the timing unreliably measurable. I think the terminology is little awkward. It's not algorithmic constant time, and it's not wall-clock constant time, but, I suppose, clock rate-relative input-independent time. So the options are 1) don't change the frequency, which has systemic negative effects, or 2) start with input-independent timing and purposefully skew it.