[PATCH v2] LoongArch: Add bad static relocation check and output more information to user

[mengqinggang]

在 2024/6/18 下午5:45, Xi Ruoyao 写道:
> On Fri, 2024-04-19 at 17:59 +0800, Lulu Cai wrote:
>>   	case R_LARCH_ABS_HI20:
>>   	case R_LARCH_SOP_PUSH_ABSOLUTE:
>> +	  if (bfd_link_pic (info))
>> +	    return bad_static_reloc (abfd, rel, sec, r_type, h,
>> isym);
> Hi,
>
> This has broken linking the object files created by gas-2.38 with ld
> from Binutils trunk.  In the era of gas-2.38 we used
> R_LARCH_SOP_PUSH_ABSOLUTE heavily even with -fPIC:
>
> 0000000000000000 <_start>:
>     0:	1c000004 	pcaddu12i   	$a0, 0
> 			0: R_LARCH_SOP_PUSH_PCREL	_start+0x800
> 			0: R_LARCH_SOP_PUSH_ABSOLUTE	*ABS*+0xc
> 			0: R_LARCH_SOP_SR	*ABS*
> 			0: R_LARCH_SOP_POP_32_S_5_20	*ABS*
>     4:	02c00084 	addi.d      	$a0, $a0, 0
> 			4: R_LARCH_SOP_PUSH_PCREL	_start+0x4
> 			4: R_LARCH_SOP_PUSH_PCREL	_start+0x804
> 			4: R_LARCH_SOP_PUSH_ABSOLUTE	*ABS*+0xc
> 			4: R_LARCH_SOP_SR	*ABS*
> 			4: R_LARCH_SOP_PUSH_ABSOLUTE	*ABS*+0xc
> 			4: R_LARCH_SOP_SL	*ABS*
> 			4: R_LARCH_SOP_SUB	*ABS*
> 			4: R_LARCH_SOP_POP_32_S_10_12	*ABS*
>     8:	4c000020 	jirl        	$zero, $ra, 0
>     c:	03400000 	andi        	$zero, $zero, 0x0
>
> Now to me maintaining the support of those stack-based reloc types has
> been a burden.  Should we just remove the support in the upcoming 2.43
> release, or make them deprecated (with a warning) in 2.43 and remove the
> support in 2.44?
>