[RFC] libdl: Make Elf_Sym::st_other available

[Chris Johns]

On 1/3/2024 6:05 pm, Sebastian Huber wrote:
> On 29.02.24 00:27, Chris Johns wrote:
>> On 27/2/2024 6:46 pm, Sebastian Huber wrote:
>>> The 64-bit PowerPC ELFv2 relocation support needs access to the
>>> Elf_Sym::st_other symbol information.  The machine-specific relocation handler
>>> had only access to the Elf_Sym::st_info symbol information.  This change
>>> extends the 8-bit syminfo parameter to 16-bit and uses the additional
>>> 8-bits to provide Elf_Sym::st_other.  Another approach could be to pass
>>> a pointer to an Elf_Sym object instead of symname, syminfo, and
>>> symvalue.
>>
>> I think symname and symvalue have to stay or the code needed to support them
>> moves out to all reloc handlers [1]. I agree there is a limit to the number args
>> to keep adding. If there is a need for more fields then it may pay to pass in
>> Elf_Sym* or rtems_rtl_obj_sym* which is the symbol table reference?
>>
>> Chris
>>
>> [1] https://git.rtems.org/rtems/tree/cpukit/libdl/rtl-elf.c#n429
> 
> What do you prefer, a new st_other parameter, use Elf_Sym*, or use
> rtems_rtl_obj_sym*?
> 
> The
> 
> /**
>  * An object file symbol.
>  */
> typedef struct rtems_rtl_obj_sym
> {
>   rtems_chain_node node;    /**< The node's link in the chain. */
>   const char*      name;    /**< The symbol's name. */
>   void*            value;   /**< The value of the symbol. */
>   uint32_t         data;    /**< Format specific data. */
> } rtems_rtl_obj_sym;
> 
> has no st_info and st_other members. I tend to pass a Elf_Sym* pointer.

Ah thanks. I think Elf_Sym* as changes to sizeof(rtems_rtl_obj_sym) effects the
size of the runtime symbol table. The data you need is only for resolving that
obj's relocs and nothing more if I understand things correctly.

Chris