[gold][aarch64] Patch for erratum-843419 (1 of 2 - report erratum occurrences)

[Hán Shěn (沈涵)]

Hi Cary, thanks! (comments addressed and CL submitted)

Thanks,
Han

On Thu, Apr 16, 2015 at 2:37 PM, Cary Coutant <ccoutant@gmail.com> wrote:
>>> Given the constraints on the address, it's inefficient to loop over
>>> every address. I'd suggest instead something like this:
>>>
>>>   if (output_address & 0x03 != 0)
>>
>> 'output_address' is output section's output_address, so I use
>> 'output_address + span_start' here.
>
> Alternatively, you could do "output_address += span_start" at the top
> of the function, or pass "output_address + span_start" where you call
> the function.

Done.

>
>>>     return;
>>>   section_size_type offset = 0;
>>>   section_size_type span_length = span_end - span_start;
>>>   // The first instruction must be at page offset 0xFF8 or 0xFFC.
>>>   unsigned int page_offset = output_address & 0xfff;
>>
>> Same as above, 'output_address' is output section's output_address, so
>> I use 'output_address + span_start' here.
>> If there a consensus/guideline on whether to use '0xfff' or '0xFFF'?
>
> I find lots of both, but lower-case seems to predominate (I just find
> it easier to count upper-case F's in non-monospaced text like this
> email). Regardless, the 'x' should always be lower-case:
>
> +      page_offset = (output_address + span_start + offset) & 0XFFF;
> +      if (page_offset == 0XFF8)
> +        offset += 4;
> +      else  // (page_offset == 0xffc), we move to next page's 0xff8.
> +        offset += 0XFFC;
>
> I'd go with all lower-case (throughout this function).

Done with lower-case.

>
> This is OK with that change. Thanks!
>
> -cary


-- 
Han Shen