On Wed, Aug 24, 2022 at 3:30 PM Donald Hunter <donald.hunter@xxxxxxxxx> wrote:
>
> Extend the libbpf documentation with a table of program types,
> attach points and ELF section names. This table uses data from
> program_types.csv which is generated from tools/lib/bpf/libbpf.c
> during the documentation build.
>
> Signed-off-by: Donald Hunter <donald.hunter@xxxxxxxxx>
> ---
> Documentation/Makefile | 3 +-
> Documentation/bpf/libbpf/Makefile | 36 ++++++++++++++++++++++
> Documentation/bpf/libbpf/index.rst | 3 ++
> Documentation/bpf/libbpf/program_types.rst | 18 +++++++++++
> Documentation/bpf/programs.rst | 3 ++
> 5 files changed, 62 insertions(+), 1 deletion(-)
> create mode 100644 Documentation/bpf/libbpf/Makefile
> create mode 100644 Documentation/bpf/libbpf/program_types.rst
>
> diff --git a/Documentation/Makefile b/Documentation/Makefile
> index 8a63ef2dcd1c..f007314770e1 100644
> --- a/Documentation/Makefile
> +++ b/Documentation/Makefile
> @@ -66,7 +66,8 @@ I18NSPHINXOPTS = $(PAPEROPT_$(PAPER)) $(SPHINXOPTS) .
> loop_cmd = $(echo-cmd) $(cmd_$(1)) || exit;
>
> BUILD_SUBDIRS = \
> - Documentation/userspace-api/media
> + Documentation/userspace-api/media \
> + Documentation/bpf/libbpf
>
> quiet_cmd_build_subdir = SUBDIR $2
> cmd_build_subdir = $(MAKE) BUILDDIR=$(abspath $(BUILDDIR)) $(build)=$2 $3
> diff --git a/Documentation/bpf/libbpf/Makefile b/Documentation/bpf/libbpf/Makefile
> new file mode 100644
> index 000000000000..c0c2811c4dd6
> --- /dev/null
> +++ b/Documentation/bpf/libbpf/Makefile
> @@ -0,0 +1,36 @@
> +# SPDX-License-Identifier: GPL-2.0
> +#
> +# Rules to convert BPF program types in tools/lib/bpf/libbpf.c
> +# into a .csv file
> +
> +FILES = program_types.csv
> +
> +TARGETS := $(addprefix $(BUILDDIR)/, $(FILES))
> +
> +$(BUILDDIR)/program_types.csv: $(srctree)/tools/lib/bpf/libbpf.c
> + $(Q)awk -F'[",[:space:]]+' \
> + 'BEGIN { print "Program Type,Attach Type,ELF Section Name,Sleepable" } \
> + /SEC_DEF\(\"/ && !/SEC_DEPRECATED/ { \
> + type = "``BPF_PROG_TYPE_" $$4 "``"; \
> + attach = index($$5, "0") ? "" : "``" $$5 "``"; \
> + section = "``" $$3 "``"; \
> + sleepable = index($$0, "SEC_SLEEPABLE") ? "Yes" : ""; \
> + print type "," attach "," section "," sleepable }' \
> + $< > $@
> +
> +.PHONY: all html epub xml latex linkcheck clean
> +
> +all: $(BUILDDIR) ${TARGETS}
> + @:
> +
> +html: all
> +epub: all
> +xml: all
> +latex: all
> +linkcheck:
> +
> +clean:
> + -$(Q)rm -f ${TARGETS} 2>/dev/null
> +
> +$(BUILDDIR):
> + $(Q)mkdir -p $@
> diff --git a/Documentation/bpf/libbpf/index.rst b/Documentation/bpf/libbpf/index.rst
> index 3722537d1384..2c04a9b3aa1f 100644
> --- a/Documentation/bpf/libbpf/index.rst
> +++ b/Documentation/bpf/libbpf/index.rst
> @@ -1,5 +1,7 @@
> .. SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
>
> +.. _libbpf:
> +
> libbpf
> ======
>
> @@ -9,6 +11,7 @@ libbpf
> API Documentation <https://libbpf.readthedocs.io/en/latest/api.html>
I'd put program_types here, it's more relevant and important than
libbpf naming conventions
> libbpf_naming_convention
> libbpf_build
> + program_types
>
> This is documentation for libbpf, a userspace library for loading and
> interacting with bpf programs.
> diff --git a/Documentation/bpf/libbpf/program_types.rst b/Documentation/bpf/libbpf/program_types.rst
> new file mode 100644
> index 000000000000..dc65ede09eef
> --- /dev/null
> +++ b/Documentation/bpf/libbpf/program_types.rst
> @@ -0,0 +1,18 @@
> +.. SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
> +
> +.. _program_types_and_elf:
> +
> +Program Types and ELF Sections
nit: two spaces?
> +===============================
> +
> +The table below lists the program types, their attach types where relevant and the ELF section
> +names supported by libbpf for them. The ELF section names follow these rules:
> +
> +- ``type`` is an exact match, e.g. ``SEC("socket")``
> +- ``type+`` means it can be either exact ``SEC("type")`` or well-formed ``SEC("type/extras")``
> + with a ‘``/``’ separator, e.g. ``SEC("tracepoint/syscalls/sys_enter_open")``
'/' is always going to be a type and "extras" separator, but extra
section format is not formalized. We have cases where it's all '/'s
(like tracepoint you mentioned), but newer and more complicated format
uses ':' as separator, e.g.
SEC("usdt/<path-to-binary>:<usdt_provide>:<usdt_name>") (let's mention
the latter as well to not create false impression of only ever having
'/' as separator)
> +
> +.. csv-table:: Program Types and Their ELF Section Names
> + :file: ../../output/program_types.csv
> + :widths: 40 30 20 10
> + :header-rows: 1
it would be helpful to include a short snippet from generated CSV file
to give a general idea of the output
> diff --git a/Documentation/bpf/programs.rst b/Documentation/bpf/programs.rst
> index 620eb667ac7a..c99000ab6d9b 100644
> --- a/Documentation/bpf/programs.rst
> +++ b/Documentation/bpf/programs.rst
> @@ -7,3 +7,6 @@ Program Types
> :glob:
>
> prog_*
> +
> +For a list of all program types, see :ref:`program_types_and_elf` in
> +the :ref:`libbpf` documentation.
> --
> 2.35.1
>
