On Thu, May 8, 2025 at 9:30 PM John Hubbard <jhubbard@xxxxxxxxxx> wrote: > > On 5/8/25 4:36 AM, Amir Goldstein wrote: > > On Thu, May 8, 2025 at 9:31 AM John Hubbard <jhubbard@xxxxxxxxxx> wrote: > >> On 5/7/25 1:42 PM, Amir Goldstein wrote: > > ... > >> Yes, syscalls are the weak point for this approach, and the above is > >> reasonable, given the situation, which is: we are not set up to recreate > >> per-arch syscall tables for kselftests to use. But this does leave the > >> other big arch out in the cold: arm64. > >> > >> It's easy to add, though, if and when someone wants it. > > > > I have no problem adding || defined(__arm64__) > > it's the same syscall numbers anyway. > > > > Or I could do > > #if !defined(__alpha__) && !defined(_MIPS_SIM) > > > > but I could not bring myself to do the re-definitions that Christian > > added in mount_setattr_test.c for > > __NR_mount_setattr, __NR_open_tree, __NR_move_mount > > > > Note that there are stale definitions for __ia64__ in that file > > and the stale definition for __NR_move_mount is even wrong ;) > > > > Christian, > > > > How about moving the definitions from mount_setattr_test.c into wrappers.h > > and leaving only the common !defined(__alpha__) && !defined(_MIPS_SIM) > > case? > > > > By the way, is this approach possibly something that the larger kselftests > could use (not in this series, of course)? I recall most of them are doing > something that is x86-only, as well. And so if you have made some observations > about syscall numbers such as "only alpha and mips are different" (?), I > could definitely take that and run with it for the overall kselftests. I don't think this is generally correct. It is correct for the syscall numbers in this file. Thanks, Amir.
