Bounded loops in BPF programs [LWN.net]

Bounded loops in BPF programs

Posted Dec 4, 2018 12:20 UTC (Tue) by nix (subscriber, #2304)
In reply to: Bounded loops in BPF programs by rvolgers
Parent article: Bounded loops in BPF programs

That would mean redoing all the JITters and everything, and breaking userspace and everything else that's grown up aroud the BPF world unless you also implement a BPF->kWASM translator. Honestly, compared to adding a few instructions (that the verifier replaces, so they don't even need JIT or interpreter support) this seems like a ridiculously heavy hammer.

(And why would Google have a WASM JITter suitable for running in *kernelspace* anyway?)

to post comments

Bounded loops in BPF programs

Posted Dec 4, 2018 12:31 UTC (Tue) by roc (subscriber, #30627) [Link] (2 responses)

There's a pathological pattern that might be at work here.

1) We have a framework X.
2) X is missing a small feature. Framework Y has that feature, and is overall great, but it's more complex than we need right now, and besides the cost of migrating from X to Y for just this one feature can't be justified.
3) Add feature to X.
4) Goto 1.

Duplicating effort along the path of least resistance, one feature at a time.

"duplicating effort along the path of least resistance"

Posted Dec 4, 2018 14:30 UTC (Tue) by Herve5 (guest, #115399) [Link]

Indeed, this was a perfect comment, that I can see applying in a dozen industrial areas other than software, just around myself... I'll retain your wording :-D

Bounded loops in BPF programs

Posted Dec 4, 2018 20:02 UTC (Tue) by simcop2387 (subscriber, #101710) [Link]

This raises the question, and it's a big one. Does framework Y have things we don't want in framework X? If that answer is yes then that's a reason we can't use it directly. And at least with WebAssembly and BPF that answer is probably yes, there's a lot of foreign call support in webassembly that'd have to be removed and then other bits that would have to be reimplemented from BPF to make it work for the use cases where BPF is used in the kernel. I think in the end you'd end up with an incompatible with either hybrid framework Z that's a mishmash of X and Y, and I'm not sure that's better.