Re: [RFC bpf-next 1/3] libbpf: update BPF_PROG2() to handle empty structs

[Alan Maguire <alan.maguire@oracle.com>]

On 08/05/2025 14:22, Alan Maguire wrote:
> In the kernel we occasionally find empty structs as parameters to
> functions, usually because some arch-specific field(s) are not present.
> Ensure that when such structs are used as parameters to functions we
> handle the fact that no registers are used in their representation.
> 
> Deliberately not using a Fixes: tag here because for this to be useful
> we need a more recent pahole with [1].
>

apologies, this isn't quite right; we had exceptions in pahole encoding
ensuring struct parameters didn't require strict parameter/register
matching, it's just that moving to a size-based determination in the v1
of Tony's patch left zero-sized structs out. So we should probably mention

Fixes: 34586d29f8df ("libbpf: Add new BPF_PROG2 macro")

> [1] https://lore.kernel.org/dwarves/20250502070318.1561924-1-tony.ambardar@gmail.com/
> 
> Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
> ---
>  tools/lib/bpf/bpf_tracing.h | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/lib/bpf/bpf_tracing.h b/tools/lib/bpf/bpf_tracing.h
> index a8f6cd4841b0..7629650251dc 100644
> --- a/tools/lib/bpf/bpf_tracing.h
> +++ b/tools/lib/bpf/bpf_tracing.h
> @@ -694,12 +694,13 @@ ____##name(unsigned long long *ctx, ##args)
>  #endif
>  
>  #define ___bpf_treg_cnt(t) \
> +	__builtin_choose_expr(sizeof(t) == 0, 0,	\
>  	__builtin_choose_expr(sizeof(t) == 1, 1,	\
>  	__builtin_choose_expr(sizeof(t) == 2, 1,	\
>  	__builtin_choose_expr(sizeof(t) == 4, 1,	\
>  	__builtin_choose_expr(sizeof(t) == 8, 1,	\
>  	__builtin_choose_expr(sizeof(t) == 16, 2,	\
> -			      (void)0)))))
> +			      (void)0))))))
>  
>  #define ___bpf_reg_cnt0()		(0)
>  #define ___bpf_reg_cnt1(t, x)		(___bpf_reg_cnt0() + ___bpf_treg_cnt(t))
> @@ -717,12 +718,13 @@ ____##name(unsigned long long *ctx, ##args)
>  #define ___bpf_reg_cnt(args...)	 ___bpf_apply(___bpf_reg_cnt, ___bpf_narg2(args))(args)
>  
>  #define ___bpf_union_arg(t, x, n) \
> +	__builtin_choose_expr(sizeof(t) == 0, ({ t ___t; ___t; }), \
>  	__builtin_choose_expr(sizeof(t) == 1, ({ union { __u8 z[1]; t x; } ___t = { .z = {ctx[n]}}; ___t.x; }), \
>  	__builtin_choose_expr(sizeof(t) == 2, ({ union { __u16 z[1]; t x; } ___t = { .z = {ctx[n]} }; ___t.x; }), \
>  	__builtin_choose_expr(sizeof(t) == 4, ({ union { __u32 z[1]; t x; } ___t = { .z = {ctx[n]} }; ___t.x; }), \
>  	__builtin_choose_expr(sizeof(t) == 8, ({ union { __u64 z[1]; t x; } ___t = {.z = {ctx[n]} }; ___t.x; }), \
>  	__builtin_choose_expr(sizeof(t) == 16, ({ union { __u64 z[2]; t x; } ___t = {.z = {ctx[n], ctx[n + 1]} }; ___t.x; }), \
> -			      (void)0)))))
> +			      (void)0))))))
>  
>  #define ___bpf_ctx_arg0(n, args...)
>  #define ___bpf_ctx_arg1(n, t, x)		, ___bpf_union_arg(t, x, n - ___bpf_reg_cnt1(t, x))