Re: [PATCH v2 1/1] libbpf: replace typeof with __typeof__

From: James Hilliard
Date: Thu Jun 09 2022 - 02:15:18 EST


On Wed, Jun 8, 2022 at 9:22 AM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote:
>
> On 6/8/22 3:04 PM, James Hilliard wrote:
> > On Wed, Jun 8, 2022 at 6:50 AM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote:
> >> On 6/8/22 8:40 AM, James Hilliard wrote:
> >>> It seems the gcc preprocessor breaks when typeof is used with
> >>> macros.
> >>>
> >>> Fixes errors like:
> >>> error: expected identifier or '(' before '#pragma'
> >>> 106 | SEC("cgroup/bind6")
> >>> | ^~~
> >>>
> >>> error: expected '=', ',', ';', 'asm' or '__attribute__' before '#pragma'
> >>> 114 | char _license[] SEC("license") = "GPL";
> >>> | ^~~
> >>>
> >>> Signed-off-by: James Hilliard <james.hilliard1@xxxxxxxxx>
> >>> ---
> >>> Changes v1 -> v2:
> >>> - replace typeof with __typeof__ instead of changing pragma macros
> >>> ---
> >>> tools/lib/bpf/bpf_core_read.h | 16 ++++++++--------
> >>> tools/lib/bpf/bpf_helpers.h | 4 ++--
> >>> tools/lib/bpf/bpf_tracing.h | 24 ++++++++++++------------
> >>> tools/lib/bpf/btf.h | 4 ++--
> >>> tools/lib/bpf/libbpf_internal.h | 6 +++---
> >>> tools/lib/bpf/usdt.bpf.h | 6 +++---
> >>> tools/lib/bpf/xsk.h | 12 ++++++------
> >>> 7 files changed, 36 insertions(+), 36 deletions(-)
> >>>
> >>> diff --git a/tools/lib/bpf/bpf_core_read.h b/tools/lib/bpf/bpf_core_read.h
> >>> index fd48b1ff59ca..d3a88721c9e7 100644
> >>> --- a/tools/lib/bpf/bpf_core_read.h
> >>> +++ b/tools/lib/bpf/bpf_core_read.h
> >>> @@ -111,7 +111,7 @@ enum bpf_enum_value_kind {
> >>> })
> >>>
> >>> #define ___bpf_field_ref1(field) (field)
> >>> -#define ___bpf_field_ref2(type, field) (((typeof(type) *)0)->field)
> >>> +#define ___bpf_field_ref2(type, field) (((__typeof__(type) *)0)->field)
> >>> #define ___bpf_field_ref(args...) \
> >>> ___bpf_apply(___bpf_field_ref, ___bpf_narg(args))(args)
> >>>
> >>
> >> Can't we just add the below?
> >>
> >> #ifndef typeof
> >> # define typeof __typeof__
> >> #endif

Ok, looks like this does appear to work, although just switching to __typeof__
may be preferable as it should work everywhere.

> >
> > From what I can tell it's not actually missing, but rather is
> > preprocessed differently
> > as the errors seem to be macro related.
>
> Are you saying that the above suggestion wouldn't work? Do you have some more
> details? I'm mainly wondering if there's a way where we could prevent letting
> typeof() usage slip through in future given from kernel side people are used
> to it.

I think my build env was using stale headers, I think I figured out
what's going on,
the typeof issue is triggered by building with -std=c17, the macro
issue is unrelated
and limited to the SEC macro it would appear.

I suppose running builds with -std=c17 would prevent typeof() usage from
slipping through.

>
> > I did also find this change which seems related:
> > https://github.com/torvalds/linux/commit/8faf7fc597d59b142af41ddd4a2d59485f75f88a
> >
> >>
> >> Thanks,
> >> Daniel
>