Re: [PATCH 03/11] bpf: Add btf_ids object

From: Andrii Nakryiko <andrii.nakryiko@gmail.com>
To: Jiri Olsa <jolsa@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Networking <netdev@vger.kernel.org>, bpf <bpf@vger.kernel.org>,
	Song Liu <songliubraving@fb.com>, Yonghong Song <yhs@fb.com>,
	Martin KaFai Lau <kafai@fb.com>, David Miller <davem@redhat.com>,
	John Fastabend <john.fastabend@gmail.com>,
	Wenbo Zhang <ethercflow@gmail.com>,
	KP Singh <kpsingh@chromium.org>, Andrii Nakryiko <andriin@fb.com>,
	Brendan Gregg <bgregg@netflix.com>,
	Florent Revest <revest@chromium.org>,
	Al Viro <viro@zeniv.linux.org.uk>
Subject: Re: [PATCH 03/11] bpf: Add btf_ids object
Date: Thu, 18 Jun 2020 17:56:38 -0700	[thread overview]
Message-ID: <CAEf4BzZ=BN7zDU_8xMEEoF7khjC4bwGitU+iYf+6uFXPZ_=u-g@mail.gmail.com> (raw)
In-Reply-To: <20200616100512.2168860-4-jolsa@kernel.org>

On Tue, Jun 16, 2020 at 3:05 AM Jiri Olsa <jolsa@kernel.org> wrote:
>
> Adding support to generate .BTF_ids section that would
> hold various BTF IDs list for verifier.
>
> Adding macros help to define lists of BTF IDs placed in
> .BTF_ids section. They are initially filled with zeros
> (during compilation) and resolved later during the
> linking phase by btfid tool.
>
> Following defines list of one BTF ID that is accessible
> within kernel code as bpf_skb_output_btf_ids array.
>
>   extern int bpf_skb_output_btf_ids[];
>
>   BTF_ID_LIST(bpf_skb_output_btf_ids)
>   BTF_ID(struct, sk_buff)
>
> Suggested-by: Andrii Nakryiko <andrii.nakryiko@gmail.com>
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> ---
>  include/asm-generic/vmlinux.lds.h |  4 ++
>  kernel/bpf/Makefile               |  2 +-
>  kernel/bpf/btf_ids.c              |  3 ++
>  kernel/bpf/btf_ids.h              | 70 +++++++++++++++++++++++++++++++
>  4 files changed, 78 insertions(+), 1 deletion(-)
>  create mode 100644 kernel/bpf/btf_ids.c
>  create mode 100644 kernel/bpf/btf_ids.h
>

[...]

> +/*
> + * Following macros help to define lists of BTF IDs placed
> + * in .BTF_ids section. They are initially filled with zeros
> + * (during compilation) and resolved later during the
> + * linking phase by btfid tool.
> + *
> + * Any change in list layout must be reflected in btfid
> + * tool logic.
> + */
> +
> +#define SECTION ".BTF_ids"

nit: SECTION is super generic and non-greppable. BTF_IDS_SECTION?

> +
> +#define ____BTF_ID(symbol)                             \
> +asm(                                                   \
> +".pushsection " SECTION ",\"a\";               \n"     \

section should be also read-only? Either immediately here, of btfid
tool should mark it? Unless I missed that it's already doing it :)

> +".local " #symbol " ;                          \n"     \
> +".type  " #symbol ", @object;                  \n"     \
> +".size  " #symbol ", 4;                        \n"     \
> +#symbol ":                                     \n"     \
> +".zero 4                                       \n"     \
> +".popsection;                                  \n");
> +
> +#define __BTF_ID(...) \
> +       ____BTF_ID(__VA_ARGS__)

why varargs, if it's always a single argument? Or it's one of those
macro black magic things were it works only in this particular case,
but not others?

> +
> +#define __ID(prefix) \
> +       __PASTE(prefix, __COUNTER__)
> +
> +
> +/*
> + * The BTF_ID defines unique symbol for each ID pointing
> + * to 4 zero bytes.
> + */
> +#define BTF_ID(prefix, name) \
> +       __BTF_ID(__ID(__BTF_ID__##prefix##__##name##__))
> +
> +
> +/*
> + * The BTF_ID_LIST macro defines pure (unsorted) list
> + * of BTF IDs, with following layout:
> + *
> + * BTF_ID_LIST(list1)
> + * BTF_ID(type1, name1)
> + * BTF_ID(type2, name2)
> + *
> + * list1:
> + * __BTF_ID__type1__name1__1:
> + * .zero 4
> + * __BTF_ID__type2__name2__2:
> + * .zero 4
> + *
> + */
> +#define BTF_ID_LIST(name)                              \

nit: btw, you call it a list here, but btfids tool talks about
"sorts". Maybe stick to consistent naming. Either "list" or "set"
seems to be appropriate. Set implies a sorted aspect a bit more, IMO.

> +asm(                                                   \
> +".pushsection " SECTION ",\"a\";               \n"     \
> +".global " #name ";                            \n"     \

I was expecting to see reserved 4 bytes for list size? I also couldn't
find where btfids tool prepends it. From what I could understand, it
just assumed the first 4 bytes are the length prefix? Sorry if I'm
slow...

> +#name ":;                                      \n"     \
> +".popsection;                                  \n");
> +
> +#endif
> --
> 2.25.4
>