On Tue, May 20, 2025 at 09:59:27AM +0100, Alan Maguire wrote: [...] > > I discussed this with Jose, and the gcc behaviour with zero-sized > structs varies a bit between architectures. Given that complexity, my > inclination would be to class functions with 0-sized struct parameters > as having inconsistent representations. They can then be tackled by > adding location info on a per-site basis later as part of the > inline-related work. For now we would just not emit BTF for them, since > without that site-specific analysis we can't be sure from function > signature alone where parameters are stored. In practice this means > leaving one function out of kernel BTF. > > So long story short, I think it might make sense to withdraw this series > for now and see if we can tweak Tony's patch to class functions with > 0-sized parameters as inconsistent as the v1 version did, meaning they > don't get a BTF representation. Thanks! > > Alan > [...] Agreed that sounds reasonable, and I'd like to resolve the original problem on 32-bit, so will update my patch and resend. Thanks, Tony
