Re: [RFC bpf-next 0/3] bpf: handle 0-sized structs properly

[Tony Ambardar <tony.ambardar@gmail.com>]

On Fri, May 09, 2025 at 11:40:47AM -0700, Andrii Nakryiko wrote:
> On Thu, May 8, 2025 at 6:22 AM Alan Maguire <alan.maguire@oracle.com> wrote:
> >
> > When testing v1 of [1] we noticed that functions with 0-sized structs
> > as parameters were not part of BTF encoding; this was fixed in v2.
> > However we need to make sure we handle such zero-sized structs
> > correctly since they confound the calling convention expectations -
> > no registers are used for the empty struct so this has knock-on effects
> > for subsequent register-parameter matching.
> 
> Do you have a list (or at least an example) of the function we are
> talking about, just curious to see what's that.
> 

BTW, Alan shared an example in the other pahole patch thread:
https://lore.kernel.org/dwarves/07d92da1-36f3-44d2-a0a4-cf7dabf278c6@oracle.com/

> The question I have is whether it's safe to assume that regardless of
> architecture we can assume that zero-sized struct has no effect on
> register allocation (which would seem logical, but is that true for
> all ABIs).
> 
> BTW, while looking at patch #2, I noticed that
> btf_distill_func_proto() disallows functions returning
> struct-by-value, which seems overly aggressive, at least for structs
> of up to 8 bytes. So maybe if we can validate that both cases are not
> introducing any new quirks across all supported architectures, we can
> solve both limitations?
> 

Given pahole (and my related patch) assume pass-by-value for well-sized
structs, I'd like to see this too. But while the pahole patch works on
64/32-bit archs, I noticed from patch #1 that e.g. ___bpf_treg_cnt()
seems to hard-code a 64-bit register size. Perhaps we can fix that too? 

> P.S., oh, and s390x selftest (test_struct_args) isn't happy, please check.
> 
> 
> >
> > Patch 1 updates BPF_PROG2() to handle the zero-sized struct case.
> > Patch 2 makes 0-sized structs a special case, allowing them to exist
> > as parameter representations in BTF without failing verification.
> > Patch 3 is a selftest that ensures the parameters after the 0-sized
> > struct are represented correctly.
> >
> > [1] https://lore.kernel.org/dwarves/20250502070318.1561924-1-tony.ambardar@gmail.com/
> >
> > Alan Maguire (3):
> >   libbpf: update BPF_PROG2() to handle empty structs
> >   bpf: allow 0-sized structs as function parameters
> >   selftests/bpf: add 0-length struct testing to tracing_struct tests
> >
> >  kernel/bpf/btf.c                                     |  2 +-
> >  tools/lib/bpf/bpf_tracing.h                          |  6 ++++--
> >  .../selftests/bpf/prog_tests/tracing_struct.c        |  2 ++
> >  tools/testing/selftests/bpf/progs/tracing_struct.c   | 11 +++++++++++
> >  tools/testing/selftests/bpf/test_kmods/bpf_testmod.c | 12 ++++++++++++
> >  5 files changed, 30 insertions(+), 3 deletions(-)
> >
> > --
> > 2.39.3
> >