bdonlan

~~I have a PR prepared for this but need to wait for https://github.com/aws/s2n-quic/pull/1399 to land first.~~ Nevermind, problem wasn't as simple as I was expecting and I don't think I'll...

Can we confirm if this is the case for all three major OSes we target?

For consistency, we should possibly just use the nospec variant for all cursor advances, particularly if we standardize on using cursors for read and buffers for write.

The kernel only mitigates spectre gadgets within the kernel itself. Spectre gadgets in userspace require userspace mitigation.

Found the cause. There was nondeterminism, in the form of the JoinHandles coming out of a HashMap (where keys are of indeterminate order).

I did see that section, for my use case mutating global environment variables is not ideal.