[rtems-tools]rtems-ld issue?

[Chris Johns]

On 4/9/2023 4:50 pm, zhengxiaojun wrote:
> 
> 
> 在 2023/8/31 13:39, Chris Johns 写道:
>> On 31/8/2023 3:12 pm, zhengxiaojun wrote:
>>>
>>>
>>> 在 2023/8/31 6:05, Chris Johns 写道:
>>>> On 31/8/2023 12:48 am, zhengxiaojun wrote:
>>>>> Hi,
>>>>>      I use the latest rtems-tools to generate RAP file, I found the object
>>>>> file
>>>>> become too small(elf,2184bytes==>rap,736bytes) and the file loaded failed,
>>>>> error
>>>>> message like this "load app.out error:offset past end of file: offset=736
>>>>> size=736 error."
>>>>
>>>> What arch and BSP?
>>>>
>>>>>      I reverted the rtl-rap.cpp commit(
>>>>> 0ad4aaafc20afcb5aacb7a82b0b3a8150b638975
>>>>> linker/rap: Ignore relocation records with no section), the rap file can be
>>>>> loaded.
>>>>
>>>> I think the commit has the wrong ticket id, it should be 4069 ...
>>>>
>>>> https://devel.rtems.org/ticket/4069
>>>>
>>>> My guess is the length includes the relocs that have been dropped because they
>>>> do not have a symbol section. A reloc without a symbol section cannot be
>>>> located
>>>> because you do not know the section to locate it against.
>>>>
>>>> Are you able to review the rtems-ld code for the issue?
>>>>
>>> I found something suspicious, get_relocations(s) return all the count of
>>> relocation and write to rap file, but actual count is less, because ignore the
>>> reloc with section=0.
>>
>> Right, this aligns with my guess. That value should the number written out and
>> it should not assume all are written.
>>
>>> This issue is similar to ticket #4781, maybe the same.
>>
>> Yeah it could. It would be nice if it is.
>>
>>> reduce count or write all relocs in rap ? I do not know which is the right
>>> solution.
>>
>> I am not sure. If relocs are needed then we have to have them and the patch you
>> isolated is wrong or needs more understanding to know why it is there. I suspect
>> an arch had an issue and this was the solution but that is guess on my part.
> 
> I tried to make the length fit the relocs write to rap file, the file can not be
> loaded, it crashed. So the relocs(reloc.symsect == 0) can not be ignored, at
> least in arm.

Can you please open a ticket and can you please add the code to build into the
RAP with the commands you are using? A script would help.

> In addition, I saw a loader complaint "rtl: unsupported section:..." about
> .ARM.attributes， the arm linkcmds always keep it.
> Should it be dealt with?

The RTL may need to be taught to ignore the section.

Chris