[PATCH] D28481: Respect section groups in GC

[Rui Ueyama via llvm-commits]

On Wed, Jan 11, 2017 at 12:01 PM, Peter Collingbourne <peter at pcc.me.uk>
wrote:

>
>
> On Wed, Jan 11, 2017 at 11:14 AM, Rafael Avila de Espindola <
> rafael.espindola at gmail.com> wrote:
>
>> Reid Kleckner via Phabricator via llvm-commits
>> <llvm-commits at lists.llvm.org> writes:
>>
>> > rnk added a comment.
>> >
>> > In https://reviews.llvm.org/D28481#640798, @ruiu wrote:
>> >
>> >> I wonder if there's a different way to keep them alive. One thing I
>> can think of is to add a dummy relocation (something like R_X86_64_NONE) to
>> a section so that if the section is alive, the other sections will become
>> alive, without actually doing any relocations. Does it work?
>> >
>> >
>> > Currently, linker GC retains all data in sections referenced by
>> __start_$section __stop_$section. We need to suppress that retention in
>> addition to establishing this inverse reference. How do you propose to do
>> that?
>>
>> See my first reply. Using LINK_ORDER might work. Or we might need to
>> standardize it somehow.
>>
>> Since this is a question more about ELF than about lld, I suggest we
>> move the discussion to
>> https://groups.google.com/forum/#!forum/generic-abi. In particular, I
>> would like to know the opinion of the developers of other linkers (Cary
>> Coutant in particular).
>>
>> So far the possibilities are:
>>
>> * Change lld and gold to be like bfd (and maybe clarify the spec).
>> * Spec that LINK_ORDER always has a reverse dependency.
>> * Add a section flag that says that relocation from that section have
>>   inverse dependency.
>>
>> OK if I start that thread and CC folks on this discussion?
>>
>
> Agreed that this would be worth discussing on generic-abi. Please do start
> the thread.
>
> Another possibility would be to introduce a section flag that means that a
> section should *not* be treated as a GC root. Then ASan can emit a
> R_X86_64_NONE relocation to introduce a dependency from the global section
> to the metadata section. We could even hypothetically teach LLVM to put
> that flag on .init_array sections for constructors that are known not to
> have side effects.
>

I don't know if it can be defined clearly because the notion of GC root is
not defined by the standard but by linker implementations (and probably not
even well documented). Defining the reverse of the ambiguous behavior seems
a bit hard.

Thanks,
> --
> --
> Peter
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170111/ac2dfc75/attachment.html>