[PATCH] Add support for ELF notes

[Chris Johns]

On 30/3/2023 2:07 am, Sebastian Huber wrote:
> On 28.03.23 23:46, Chris Johns wrote:
>> On 29/3/2023 5:36 am, Sebastian Huber wrote:
>>> On 28.03.23 09:26, Chris Johns wrote:
>>>> On 28/3/2023 6:20 pm, Sebastian Huber wrote:
>>>>> On 28.03.23 09:03, Chris Johns wrote:
>>>>>> On 28/3/2023 5:48 pm, Sebastian Huber wrote:
>>>>>>> On 25.03.23 00:39, Chris Johns wrote:
>>>>>>>> On 24/3/2023 7:32 pm, Sebastian Huber wrote:
>>>>>>>>> On 23.03.23 20:07, Chris Johns wrote:
>>>>>>>>>> On 24/3/2023 3:57 am, Sebastian Huber wrote:
>>>>>>>>>>> On 23.03.23 17:52, Will wrote:
>>>>>>>>>>>> Great idea to store this information in the executable itself. Does
>>>>>>>>>>>> this
>>>>>>>>>>>> need
>>>>>>>>>>>> a RTEMS_TEST_STATE_LINK_ONLY test state or something similar for
>>>>>>>>>>>> minimum.exe?
>>>>>>>>>>> With the notes you can build the test. The test runner would look at the
>>>>>>>>>>> notes
>>>>>>>>>>> and then decide if it makes sense to run the test or not. It could still
>>>>>>>>>>> run the
>>>>>>>>>>> minimum.exe and see if it terminates.
>>>>>>>>>> I welcome notes support. Thanks for adding it.
>>>>>>>>>>
>>>>>>>>>> How will we control and manage the notes we support?
>>>>>>>>> In the new elfnote.h header file there are defines for the note type
>>>>>>>>> (domain-specific integer).
>>>>>>>> Great.
>>>>>>>>
>>>>>>>>>> Should we document the top level notes domains (?) with some we
>>>>>>>>>> control and
>>>>>>>>>> restrict and others users can use? For example `note.rtems.test`,
>>>>>>>>>> `note.rtems.kernel`, `note.rtems.bsp`, and `note.rtems.user`?
>>>>>>>>> The section name doesn't matter. You can divide the number space of the
>>>>>>>>> note
>>>>>>>>> type for this.
>>>>>>>> Does this mean we define the numbers or number ranges or is it left open?
>>>>>>>> I am
>>>>>>>> not sure I am following this bit.
>>>>>>> The patch has this:
>>>>>>>
>>>>>>> #define ELF_NOTE_RTEMS_TYPE( _index ) ( ( _index ) + 0x10000 )
>>>>>>>
>>>>>>> We basically have 32-bits available. We could also use something like this:
>>>>>>>
>>>>>>> #define ELF_NOTE_RTEMS_BASE_TYPE( _index ) ( ( _index ) + 0x10000000 )
>>>>>>>
>>>>>>> #define ELF_NOTE_RTEMS_TEST_TYPE( _index ) ( ( _index ) + 0x20000000 )
>>>>>>>
>>>>>>> #define ELF_NOTE_RTEMS_USER_TYPE( _index ) ( ( _index ) + 0x30000000 )
>>>>>>>
>>>>>>> ...
>>>>>>>
>>>>>> Nice.
>>>>> Maybe we should also encode the data type here, for example:
>>>>>
>>>>> #define ELF_NOTE_DATA_TYPE_BOOL 0
>>>>> #define ELF_NOTE_DATA_TYPE_INT8 1
>>>>> #define ELF_NOTE_DATA_TYPE_UINT8 2
>>>>> #define ELF_NOTE_DATA_TYPE_INT16 3
>>>>> #define ELF_NOTE_DATA_TYPE_UINT16 4
>>>>> #define ELF_NOTE_DATA_TYPE_INT32 5
>>>>> #define ELF_NOTE_DATA_TYPE_UINT32 6
>>>>> #define ELF_NOTE_DATA_TYPE_INT64 7
>>>>> #define ELF_NOTE_DATA_TYPE_UINT64 8
>>>>> #define ELF_NOTE_DATA_TYPE_FLOAT32 9
>>>>> #define ELF_NOTE_DATA_TYPE_FLOAT32 10
>>>>> #define ELF_NOTE_DATA_TYPE_FLOAT64 11
>>>>> #define ELF_NOTE_DATA_TYPE_FLOAT64 12
>>>>> #define ELF_NOTE_DATA_TYPE_STRING 13
>>>>> #define ELF_NOTE_DATA_TYPE_BINARY 14
>>>>>
>>>>> #define ELF_NOTE_RTEMS_BASE_TYPE( _data_type, _index ) \
>>>>>     ( ( ( _data_type ) << 28 ) | ( ( _index ) + 0x1000000 ) )
>>>>>
>>>>> This could be used by the host tool to more easily produce the right JSON
>>>>> data.
>>>>>
>>>> Good idea. I cannot think of anything else as scoping would tend to reduce to a
>>>> new type.
>>> Another option would be to add a GNU note type for JSON data. The note
>>> description would be an arbitrary JSON string.
>> Wooo that is interesting. JSON can be useful in these situations. I like it.
> 
> It looks like we are not the only ones needing such a thing. Nick Clifton
> suggested to use this:
> 
> https://systemd.io/ELF_PACKAGE_METADATA/
> 
> https://sourceware.org/pipermail/binutils/2023-March/126822.html

This is interesting and useful. I still think your notes support is needed.

Between the addition of considered notes and DWARF you can tell a lot about the
make up of an executable and that could form the base of a audit type check. For
example all objects are built with the same compiler and ABI options. I see no
point in saying a test has passed if the make up of the exe is not considered
"correct".

Chris