Re: [PATCH][GOLD] Fix PR20765 [2.27 Regression] gold internal error in fix_errata on aarch64-linux-gnu

[Peter Smith <peter dot smith at linaro dot org>]

Thanks very much for taking a look at the patches, Han I'm quite happy for
you to take this from here if you want to.

Peter

On 1 December 2017 at 01:40, Han Shen <shenhan@google.com> wrote:

> Hi Cary, thanks for pointing this out and providing a suggestion. Hi
> Peter, thanks for providing the fix. I'll take a look into this issue and
> report back. Thanks, Han
>
>
> On Thu, Nov 30, 2017 at 3:06 PM Cary Coutant <ccoutant@gmail.com> wrote:
>
>> > The full details are in the PR, to summarise:
>> > - The internal fault is caused when there is more than one stub table
>> > and the addresses in the second stub table are dependent on the size
>> > of the first stub table. As the erratum stub addresses are only set
>> > once, an internal fault is triggered at relocation time when the
>> > inconsistency is detected.
>> > - The patch is to update the erratum stub addresses on each relaxation
>> pass.
>>
>> First, thanks for investigating and finding the problem!
>>
>> I think the stub code needs a little bit of rework in order to avoid
>> this extra work on each relaxation pass.
>>
>> I'd prefer to see the erratum_address stored as an offset relative to
>> the output address of the input section; the actual address can be
>> easily computed at the one time it's needed in
>> fix_errata_and_relocate_erratum_stubs(). The destination_address isn't
>> actually needed at all for errata stubs -- it's always erratum_address
>> + BPI -- so I don't think it needs to be stored or readjusted. It
>> could be ignored for errata stubs, or could be moved out of Stub_base
>> and into a new derived class for relocation stubs.
>>
>> This is not a change I can make on my own without adequate testing,
>> and I'm not set up to do any native aarch64 testing. Since this
>> appears to be a significant problem, though, I don't want to delay a
>> fix, and I'm inclined to go ahead and apply the patch you've provided.
>> I'd like to ask either you or Han to take a look at the suggestions
>> I've made above and submit a clean-up patch at a later time.
>>
>> I've committed this with the following ChangeLog:
>>
>> 2017-11-30  Peter Smith  <peter.smith@linaro.org>
>>
>> gold/
>>         PR gold/20765
>>         * aarch64.cc (Aarch64_relobj::update_erratum_address): New
>> method.
>>         (AArch64_relobj::scan_errata): Update addresses in stub table
>> after
>>         relaxation pass.
>>
>> -cary
>>
> --
> Han Shen |  Software Engineer |  shenhan@google.com |  +1-650-440-3330
> <(650)%20440-3330>
>