9 ms·
"trivial" how do you figure? Remember these exploits bypass your own code's conditionals over a shockingly far duration. Unless you just mean for incredibly res
by kllrnohj 2y ago
"trivial" how do you figure? Remember these exploits bypass your own code's conditionals over a shockingly far duration. Unless you just mean for incredibly restrictive usages such as eBPF?
possible absent any performance concerns at all, yeah sure
- lxgr 2y ago> Unless you just mean for incredibly restrictive usages such as eBPF? I was actually thinking something more like a bytecode interpreter that runs one operation and then sleeps until the next full wall clock second, but yes, that's my point: If you don't care about performance, you can make process isolation safe very easily.
- tptacek 2y agoI think at the point where you're suggesting 1hz bytecode interpreters the onus is kind of on you to be clear you're not talking about plausible points in the design space.
- lxgr 2y ago1 Hz is probably a bit too slow for practical applications, but my point is that somewhere between that, and simulating a parallel universe at each data-dependent branch, is probably a reasonably-safe spot, or more likely a spectrum that application developers get to pick their tradeoffs from.
- tptacek 2y agoI agree that 1hz is probably too slow for practical applications.
- throneaway1799 2y agoAn identically useful comment would've been to place the bounds at 0 and infinity.
- wbl 2y agoThat is actually still not secure. The cache will happily retain the trace of the access forever.
- lxgr 2y agoYeah, but unless a cache miss has a performance penalty of roughly one second, you should be fine.