ci: upgrade LLVM to 20.1, drop LLVM 11

[ti-mo]

LLVM 11 is getting a bit long in the tooth and doesn't support decl tags.

Target 14, 17 and the newly-released LLVM 20.

[dylandreimerink]

unknown insn class 6

That can't be good 🫣

[ti-mo]

unknown insn class 6

That can't be good 🫣

Looking into it, I think LLVM20 may be using ISAv3 by default.

[ti-mo]

Found it: llvm/llvm-project@7852ebc. -mcpu=v3 is now the default.

@yonghong-song This is slightly surprising given this breaks existing tools on kernels <5.1 if they don't specify -mcpu=v1/v2 explicitly. The kernel docs (https://docs.kernel.org/bpf/bpf_devel_QA.html#q-new-bpf-instruction-for-kernel-and-llvm) need to be updated as well:

By default the generic processor target is used, which is the base instruction set (v1) of BPF.

I'll set -mcpu=v2 here until we drop <5.1 in CI.

[yonghong-song]

Found it: llvm/llvm-project@7852ebc. -mcpu=v3 is now the default.

@yonghong-song This is slightly surprising given this breaks existing tools on kernels <5.1 if they don't specify -mcpu=v1/v2 explicitly. The kernel docs (https://docs.kernel.org/bpf/bpf_devel_QA.html#q-new-bpf-instruction-for-kernel-and-llvm) need to be updated as well:

By default the generic processor target is used, which is the base instruction set (v1) of BPF.

I'll set -mcpu=v2 here until we drop <5.1 in CI.

Yes, in llvm20 we move -mcpu=v3 as default to handle some atomic insns (esp. for arm64). Please to use -mcpu=v1/v2 as appropriate. Thanks.

[yonghong-song]

Found it: llvm/llvm-project@7852ebc. -mcpu=v3 is now the default.

@yonghong-song This is slightly surprising given this breaks existing tools on kernels <5.1 if they don't specify -mcpu=v1/v2 explicitly. The kernel docs (https://docs.kernel.org/bpf/bpf_devel_QA.html#q-new-bpf-instruction-for-kernel-and-llvm) need to be updated as well:

Will update the doc soon. thanks.