[PATCH v2] LoongArch: Fix the infinite loop caused by calling undefweak symbol

[Lulu Cai]

On 11/20/24 10:22 AM, mengqinggang wrote:
>
> 在 2024/11/20 上午9:53, Xi Ruoyao 写道:
>> On Wed, 2024-11-20 at 09:28 +0800, Lulu Cai wrote:
>>> The undefweak symbol value of non-default visibility is 0 and does
>>> not use plt entry, and will not be relocated in the relocate_secion
>>> function. As a result, an infinite loop is generated because
>>> bl %plt(sym) => bl 0.
>>>
>>> Fix this by converting the bl/call36 instructions to nop or removing
>>> them.
>> IMO it's better to convert it to "jirl $ra, $zero, 0". Semantically an
>> undefined weak symbol should be resolved to address 0 so we should make
>> a call to address 0, instead of silently ignore the call.
>>
>> Also "b %plt(sym)" should be turned to "jirl $zero, $zero, 0" as well
>> when sym is non-default visibility and weak undefined.
>
> AArch64 optimizes  calls to undefined weak symbols to a NOP. It maybe 
> a better choice.
>

Each port handles calls to non-default visibility undefweak symbols 
slightly differently.
x86-64 and riscv preserve the semantics of the symbol using "call 0" and 
"jalr 0", while
aarc64 optimizes the call to "nop". It is unclear why aarch64 optimizes 
instead of preserving
semantics.

> In bfd/elfnn-aarch64.c:
>
>  6046         /* A call to an undefined weak symbol is converted to a 
> jump to
>  6047            the next instruction unless a PLT entry will be created.
>  6048            The jump to the next instruction is optimized as a NOP.
>  6049            Do the same for local undefined symbols. */
>  6050         if (weak_undef_p && ! via_plt_p)
>  6051 {
>  6052             bfd_putl32 (INSN_NOP, hit_data);
>  6053             return bfd_reloc_ok;
>  6054           }