Skip to contentSkip to conversation

Compiled BPF

LSF/MM/BPF TOPIC Jose E. Marchesi 1 participant 1 message Feb 26 BPF

lore
Share & tools
lore
#0 lore raw

Hello people!

[Turns out I forgot to send the topic proposal to the mailing list once
the topic was proposed (in time) via the google form. But better late than never..]

Work on GCC BPF (also binutils, etc) is progressing at a good pace and there are a lot of updates and topics that would benefit from having some face-to-face discussion with the kernel hackers.

Some of these updates and topics are:

- Dealing with some pending differences in behavior between GCC and clang/llvm.

- Support of "optional" but important BPF features such as quick calls which are currently not supported by GCC. We are preparing an exhaustive support matrix to track this, kees-style.

- Documentation and maintenance of the ABI in general, and some specific ABI issues in particular that need agreement and definition at this point.

- We are finally ready to add -fverifiable, aiming at GCC 17 (GCC 16 is now in stage 4) and we would like to get feedback on some of the details.

- Handling CO-RE limitations and corner cases at compile time. The compilers should emit proper diagnostics and behave the same way.

- Testing: we have finally upstreamed the infrastructure that allows loading and running GCC BPF tests in kernel. Along with the support for building the kernel BPF selftests, the idea is to have strong coverage to detect regressions ASAP in both kernel and compiler side.

- More upstream BPF projects are adding support for using GCC BPF, and support in distros is improving.

- We recently got an important reinforcement for the developmet of GCC BPF, thanks to Meta: Vineet Gupta, an experienced GCC hacker, has joined the party and is already doing great work, speeding up the development.

It would be useful if we could have some time during LSFMMBPF to dedicate to these topics, if possible.

Salud!