scripts: Exclude Rust compilation units with pahole

[YakoYakoYokuYoku]

Version 1.24 of pahole has the capability to exclude compilation units (CUs) of specific languages with the --lang_exclude option.
Rust, as of writing, is not currently supported by pahole neither BTF and if it's used with a build that has DEBUG_INFO_BTF_MODULES enabled it results in erroneous binaries. So it's better for pahole to exclude Rust compilation units until support for it arrives.

Fixes #735.

[YakoYakoYokuYoku]

To test this change (as of 459035a) the flag --skip_encoding_btf_enum64 needs to be passed in scripts/pahole-flags.sh, though this change might not be necessary here as it's gonna land in upstream probably.

[ojeda]

To test this change (as of 459035a) the flag --skip_encoding_btf_enum64 needs to be passed in scripts/pahole-flags.sh, though this change might not be necessary here as it's gonna land in upstream probably.

Sorry, I got this swapped out from my brain -- could you please expand a bit on why the flag is needed for testing?

The PR otherwise looks good to me, though this probably should go through the BPF tree after the Rust merge. Did you mean that by "gonna land in upstream probably"?

[YakoYakoYokuYoku]

Sorry, I got this swapped out from my brain -- could you please expand a bit on why the flag is needed for testing?

The PR otherwise looks good to me, though this probably should go through the BPF tree after the Rust merge. Did you mean that by "gonna land in upstream probably"?

There was an issue related to 64 bit enums with the stable branches of the kernel. Though it's solved for both stable and 6.0+ because with the former they've added --skip_encoding_btf_enum64 to scripts/pahole-flags.sh to avoid the encoding and the latter supports those kinds of enums. In resumi, upstream issues had been fixed and there are no blockers for this.

[YakoYakoYokuYoku]

@ojeda I think this is now ready for inclusion.

[ericcurtin]

We really need this upstream if this does fix the bpf issue with Rust for Linux, is CONFIG_MODULE_ALLOW_BTF_MISMATCH=y required also or no?

[ericcurtin]

I can confirm this worked great for me, to get BPF back working with the Rust-based GPU driver on our Fedora Asahi kernel for the non-Rust parts:

https://gitlab.com/fedora-asahi/kernel-asahi/-/commit/cb4241a09a5eeb51d07455df2ca8a1a882990c6f

So...

Reviewed-by: Eric Curtin <ecurtin@redhat.com>
Tested-by: Eric Curtin <ecurtin@redhat.com>

[ericcurtin]

@marcan

[YakoYakoYokuYoku]

We really need this upstream if this does fix the bpf issue with Rust for Linux, is CONFIG_MODULE_ALLOW_BTF_MISMATCH=y required also or no?

I don't think so as CONFIG_PAHOLE_HAS_LANG_EXCLUDE=y already suffices the condition of having --lang_exclude=rust.

Also I've added those tags in the commit message to credit you, @ericcurtin, so thanks to you too for testing this PR.

[ericcurtin]

We really need this upstream if this does fix the bpf issue with Rust for Linux, is CONFIG_MODULE_ALLOW_BTF_MISMATCH=y required also or no?

I don't think so as CONFIG_PAHOLE_HAS_LANG_EXCLUDE=y already suffices the condition of having --lang_exclude=rust.

Also I've added those tags in the commit message to credit you, @ericcurtin, so thanks to you too for testing this PR.

Yup you are right, tested without it, works great!

[Conan-Kudo]

Verified as well:

Reviewed-by: Neal Gompa <neal@gompa.dev>
Tested-by: Neal Gompa <neal@gompa.dev>

[YakoYakoYokuYoku]

You were added too, @Conan-Kudo.

[ojeda]

Thanks everyone! The tags are very useful.

@YakoYakoYokuYoku For upstream, could you please submit the patch? If so, please Cc the BPF list & maintainers and probably the pahole one (@acmel) too, as well as the Rust for Linux list.

[YakoYakoYokuYoku]

There it goes, though I've fooled up and the message was sent twice. If I recall fixes should be sent upstream directly, right?

[ojeda]

There it goes, though I've fooled up and the message was sent twice.

Thanks, and no worries :)

If I recall fixes should be sent upstream directly, right?

At this point, as soon as something has all its dependencies already in mainline, it can be sent.

[YakoYakoYokuYoku]

v2 at LKML.

[YakoYakoYokuYoku]

v3 at LKML.

[YakoYakoYokuYoku]

@ojeda, can you merge this? This was accepted for bpf-next so this can be included unless we wait for it to be in torvalds and pulled from there.

[ojeda]

Sure (in any case, please note that the rust branch will eventually go away when we get it close to upstream)