[PATCH v3 12/13] RISC-V/bfd: warn about non-boolean unaligned-access attribute

[Jan Beulich]

On 22.06.2026 11:02, Jiawei wrote:
> On 2026/6/19 19:52, Jan Beulich wrote:
>> The attribute being a boolean one, incoming values should be solely 0 or
>> 1. Convert other non-zero values to 1.
>> ---
>> v3: New.
>>
>> --- a/bfd/elfxx-riscv.c
>> +++ b/bfd/elfxx-riscv.c
>> @@ -3777,6 +3777,23 @@ riscv_merge_attributes (bfd *ibfd, struc
>>   	 initialized.  */
>>         out_attr[0].i = 1;
>>   
>> +      for (i = LEAST_KNOWN_OBJ_ATTRIBUTE; i < NUM_KNOWN_OBJ_ATTRIBUTES; i++)
>> +	{
>> +	  switch (i)
>> +	    {
>> +	    case Tag_RISCV_unaligned_access:
>> +	      if (out_attr[i].i <= 1)
>> +		break;
>> +
>> +	      _bfd_error_handler
>> +		(_("warning: %pB uses non-boolean `unaligned access' attribute; "
>> +		   "converting to boolean"),
>> +		 ibfd);
>> +	      out_attr[i].i = 1;
>> +	      break;
>> +	    }
>> +	}
>> +
>>         return true;
>>       }
>>   
>> @@ -3865,7 +3882,12 @@ riscv_merge_attributes (bfd *ibfd, struc
>>   	break;
>>   
>>         case Tag_RISCV_unaligned_access:
>> -	out_attr[i].i |= in_attr[i].i;
>> +	if (in_attr[i].i > 1)
>> +	  _bfd_error_handler
>> +	    (_("warning: %pB uses non-boolean `unaligned access' attribute; "
>> +	       "converting to boolean"),
>> +	     ibfd);
>> +	out_attr[i].i |= !!in_attr[i].i;
>>   	break;
>>   
>>         case Tag_RISCV_stack_align:
> 
> LGTM.
> 
> Checking the copied attributes for the first input
> object is necessary, since otherwise a non-boolean unaligned_access 
> value from
> the first object would be carried into the output unchanged.
> 
> One very small wording thought: the gas-side diagnostic uses
> `unaligned_access', while this one uses `unaligned access'. Either is fine
> with me, using the same spelling might make the diagnostics more
> consistent, but this does not need to block the patch.

Oh, thanks for noticing. I've added in the underscores, also for the subsequent
stack-align patch.

> Reviewed-by: Jiawei jiawei@iscas.ac.cn

Thanks.

Jan