David Vernet <void@xxxxxxxxxxxxx> wrote:
> We had a productive BPF working group meeting at IETF 118, and we
> wanted to provide a summary to recap what was discussed.
David, this is a most excellent summary, thank you.
(I wasn't there, I had a conflict)
> subset. Existing language MMs do not properly handle control
> dependencies, and suffer from issues such as OOTA (Out-of-Thin-Air)
> reads. The presentation outlined the control dependencies proposed for
> various types of BPF instructions, such as atomics, jumps, etc.
I didn't know what an OOTA read was, but I think that hte first answer at:
https://stackoverflow.com/questions/42588079/what-is-out-of-thin-air-safety
explained it, but I just wnated to be sure that this was the same term.
I think that we don't expect compilers emitting BPF code to do things like
the tearing example:
void threadA() {
g = 0xAB00; // "tearing"
g += 0x00CD;
}
but an underlying BPF JIT might do this kind of thing, and of course, the
reads could be anywhere, and the variable could be written by C-code (or
RUST!).
--
Michael Richardson <mcr+IETF@xxxxxxxxxxxx>, Sandelman Software Works
-= IPv6 IoT consulting =- *I*LIKE*TRAINS*
Attachment:
signature.asc
Description: PGP signature
