Re: [PATCH bpf-next 9/9] docs/bpf: add documentation for BTF_KIND_TAG

[Yonghong Song <yhs@fb.com>]

On 9/8/21 10:42 PM, Andrii Nakryiko wrote:
> On Tue, Sep 7, 2021 at 4:01 PM Yonghong Song <yhs@fb.com> wrote:
>>
>> Add BTF_KIND_TAG documentation in btf.rst.
>>
>> Signed-off-by: Yonghong Song <yhs@fb.com>
>> ---
>>   Documentation/bpf/btf.rst | 30 ++++++++++++++++++++++++++++--
>>   1 file changed, 28 insertions(+), 2 deletions(-)
>>
> 
> [...]
> 
>> +2.2.17 BTF_KIND_TAG
>> +~~~~~~~~~~~~~~~~~~~
>> +
>> +``struct btf_type`` encoding requirement:
>> + * ``name_off``: offset to a non-empty string
>> + * ``info.kind_flag``: 0 for tagging ``type``, 1 for tagging member/argument of the ``type``
>> + * ``info.kind``: BTF_KIND_TAG
>> + * ``info.vlen``: 0
>> + * ``type``: ``struct``, ``union``, ``func`` or ``var``
>> +
>> +``btf_type`` is followed by ``struct btf_tag``.::
>> +
>> +    struct btf_tag {
>> +        __u32   comp_id;
>> +    };
>> +
>> +The ``name_off`` encodes btf_tag attribute string.
>> +If ``info.kind_flag`` is 1, the attribute is attached to the ``type``.
> 
> This contradicts "info.kind_flag" description above

will remove info.kind_flag stuff in the next revision.

> 
>> +If ``info.kind_flag`` is 0, the attribute is attached to either a
>> +``struct``/``union`` member or a ``func`` argument.
>> +Hence the ``type`` should be ``struct``, ``union`` or
>> +``func``, and ``btf_tag.comp_id``, starting from 0,
>> +indicates which member or argument is attached with
>> +the attribute.
> 
> Does the kernel validate this restriction for the VAR target type?
> I.e., if we have kind_flag == 0 (member of type), we should disallow
> VAR, right?

Yes, I even has a selftest for that.

> 
>> +
>>   3. BTF Kernel API
>>   *****************
>>
>> --
>> 2.30.2
>>