@readylayerone@alessandrod Frankendancer has been running XDP for 2 years
Agave improved replay quite a lot by copying a bunch of other FD features, like speculative execution (a bit incorrectly hence 4.0 patch releases on mainnet)
They have not yet caught up to full FD's performance but still good work
@deanmlittle IBT has a 0x3e notrack prefix that disables enforcement for individual jumps.
i.e. notrack jmp rax
Weakens the sandbox but at least the rest of the app gets function pointer integrity still.
Linux 7.0 still does not support Intel CET indirect branch tracking (IBT) in usermode.
Intel CPUs supporting IBT first appeared in 2021.
Here's my patch series: https://t.co/NoKvTvUgwU
This came up while adding control flow integrity to Firedancer's sandbox.
One of the best Linux x86 jails out there, enforcing process isolation, namespaces, strict seccomp, landlock, MPK (user protection keys), CFI, and soon more.
https://t.co/RH40zdEG8u
IBT enforces that indirect calls and jumps can't land on any arbitrary instruction, but one of these:
[f3 0f 1e fa] endbr64
"jmp rax" to elsewhere causes a segfault SEGV_CPERR, defending against control flow hijacking (e.g. corrupt function pointer).
@toly@SolanaSensei@solana@SolanaFndn@bw_solana 😴😴😴 Why do let the majority client pick protocol limits instead of giving validators a config option? Frankendancer has supported XDP for 2 years, so we're waiting on nothing. Just switch back to Agave once they sort out their performance issues
@JohnAlanWoods Solana (Sinon from @jump_firedancer) wrote *the* fastest Falcon implementation from scratch, using AVX512.
The author is upstreaming their contributions: https://t.co/nK6ZtclqP9
Expect more good OSS contributions from them 🙂