Date:   Mon, 15 Mar 2021 21:41:56 -0700
From:   Andrii Nakryiko <andrii.nakryiko@...il.com>
To:     Florent Revest <revest@...omium.org>
Cc:     bpf <bpf@...r.kernel.org>, Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Andrii Nakryiko <andrii@...nel.org>,
        Yonghong Song <yhs@...com>, KP Singh <kpsingh@...nel.org>,
        Brendan Jackman <jackmanb@...omium.org>,
        open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH bpf-next 3/5] libbpf: Initialize the bpf_seq_printf
 parameters array field by field

On Mon, Mar 15, 2021 at 9:36 PM Andrii Nakryiko
<andrii.nakryiko@...il.com> wrote:
>
> On Wed, Mar 10, 2021 at 2:02 PM Florent Revest <revest@...omium.org> wrote:
> >
> > When initializing the __param array with a one liner, if all args are
> > const, the initial array value will be placed in the rodata section but
> > because libbpf does not support relocation in the rodata section, any
> > pointer in this array will stay NULL.
> >
> > This is a workaround, ideally the rodata relocation should be supported
> > by libbpf but this would require a disproportionate amount of work given
> > the actual usecases. (it is very unlikely that one uses a const array of
> > relocated addresses)

Can you please drop this paragraph? This is not a workaround, it's a
completely working code that should continue working. And this is not
something that libbpf doesn't support, there is no kernel interface to
make it work at all.

Please add Fixes: tag as well.

> >
> > Signed-off-by: Florent Revest <revest@...omium.org>
> > ---
> >  tools/lib/bpf/bpf_tracing.h | 30 +++++++++++++++++++++++++++++-
> >  1 file changed, 29 insertions(+), 1 deletion(-)
> >
> > diff --git a/tools/lib/bpf/bpf_tracing.h b/tools/lib/bpf/bpf_tracing.h
> > index f9ef37707888..f6a2deb3cd5b 100644
> > --- a/tools/lib/bpf/bpf_tracing.h
> > +++ b/tools/lib/bpf/bpf_tracing.h
> > @@ -413,6 +413,34 @@ typeof(name(0)) name(struct pt_regs *ctx)                              \
> >  }                                                                          \
> >  static __always_inline typeof(name(0)) ____##name(struct pt_regs *ctx, ##args)
> >
> > +#define ___bpf_build_param0(narg, x)
> > +#define ___bpf_build_param1(narg, x) ___param[narg - 1] = x
> > +#define ___bpf_build_param2(narg, x, args...) ___param[narg - 2] = x; \
> > +                                             ___bpf_build_param1(narg, args)
> > +#define ___bpf_build_param3(narg, x, args...) ___param[narg - 3] = x; \
> > +                                             ___bpf_build_param2(narg, args)
> > +#define ___bpf_build_param4(narg, x, args...) ___param[narg - 4] = x; \
> > +                                             ___bpf_build_param3(narg, args)
> > +#define ___bpf_build_param5(narg, x, args...) ___param[narg - 5] = x; \
> > +                                             ___bpf_build_param4(narg, args)
> > +#define ___bpf_build_param6(narg, x, args...) ___param[narg - 6] = x; \
> > +                                             ___bpf_build_param5(narg, args)
> > +#define ___bpf_build_param7(narg, x, args...) ___param[narg - 7] = x; \
> > +                                             ___bpf_build_param6(narg, args)
> > +#define ___bpf_build_param8(narg, x, args...) ___param[narg - 8] = x; \
> > +                                             ___bpf_build_param7(narg, args)
> > +#define ___bpf_build_param9(narg, x, args...) ___param[narg - 9] = x; \
> > +                                             ___bpf_build_param8(narg, args)
> > +#define ___bpf_build_param10(narg, x, args...) ___param[narg - 10] = x; \
> > +                                              ___bpf_build_param9(narg, args)
> > +#define ___bpf_build_param11(narg, x, args...) ___param[narg - 11] = x; \
> > +                                              ___bpf_build_param10(narg, args)
> > +#define ___bpf_build_param12(narg, x, args...) ___param[narg - 12] = x; \
> > +                                              ___bpf_build_param11(narg, args)
>
> took me some time to get why the [narg - 12] :) it makes sense, but
> then I started wondering why not
>
> #define ___bpf_build_param12(narg, x, args...)
> ___bpf_build_param11(narg, args); ___param[11] = x
>
> ? seems more straightforward, no?
>
> also please keep all of them on single line. And to make lines
> shorter, let's call it ___bpf_fillX? I also don't like hard-coded
> ___param, which is both inflexible and is obscure at the point of use
> of this macro. So let's pass it as the first argument?
>
> > +#define ___bpf_build_param(args...) \
> > +       unsigned long long ___param[___bpf_narg(args)];                 \
> > +       ___bpf_apply(___bpf_build_param, ___bpf_narg(args))(___bpf_narg(args), args)
> > +
>
> And here I'd pass array as a parameter and let caller define it, so
> macro is literally just filling the array elements, not defining the
> array itself and what's the type of elements
>
> >  /*
> >   * BPF_SEQ_PRINTF to wrap bpf_seq_printf to-be-printed values
> >   * in a structure.
> > @@ -422,7 +450,7 @@ static __always_inline typeof(name(0)) ____##name(struct pt_regs *ctx, ##args)
> >                 _Pragma("GCC diagnostic push")                              \
> >                 _Pragma("GCC diagnostic ignored \"-Wint-conversion\"")      \
> >                 static const char ___fmt[] = fmt;                           \
> > -               unsigned long long ___param[] = { args };                   \
> > +               ___bpf_build_param(args);                                   \
> >                 _Pragma("GCC diagnostic pop")                               \
> >                 int ___ret = bpf_seq_printf(seq, ___fmt, sizeof(___fmt),    \
> >                                             ___param, sizeof(___param));    \
>
> here you are violating separation of variables and code,
> ___bpf_build_param is defining a variable, then has code statements,
> then you are declaring ___ret after the code. So please split ___ret
> definition,
>
> > --
> > 2.30.1.766.gb4fecdf3b7-goog
> >

Powered by blists - more mailing lists