11:12 <manospitsid> trying to cross build from host amd64 to arm64 with llvm 17 and BTF enabled, getting this error:

11:12 <manospitsid> libbpf: BTF header not found

11:12 <manospitsid> pahole: .tmp_vmlinux.btf: Invalid argument

11:13 <manospitsid> gcc build is successful, I only pass LLVM= so something with the llvm obj utils must be going wrong?

11:31 <dmart> mrutland: ^ I think the original issue was to avoid misidentifying the calling frame when the caller has called a noreturn function and the compiler threw away the caller's epilogue. It feels like some padding is needed between functions in order to distinguish these cases -- a nop would serve.

11:32 <dmart> mrutland: If we must add something though, might as well make it a brk?

12:01 <HdkR> Is TPIDRRO_EL0 codified to mean anything in Linux these days?

12:01 <HdkR> I checked a bit ago and it seems like it is always set to zero atm

13:14 <linusw__> arnd: ardb and I were discussing the need for SMP_ON_UP, do we still need this after removing ARM11 and v6k?

13:14 <arnd> yes, unfortunately there are some single-core Cortex-A9 chips that break without it, notably some omap4/dra7 variants

13:15 <arnd> it's possible that we no longer need all of the changes that SMP_ON_UP does, I'd have to check how many things get patched there. I think it's just the barriers that are affected here

14:15 <linusw__> I suppose OMAP3 with it's Cortex-A8 is also technically multi_v7?

15:54 <javierm> linusw__: it is, yeah

15:55 <javierm> linusw__: and also am33xx AFAIK

19:11 <arnd> linusw__: I'm pretty sure that one doesn't need SMP_ON_UP though, as the barriers just work as expected