Re: [patch] Nios II build_id fix

[Cesar Philippidis <cesar at codesourcery dot com>]

On 4/5/14, 10:12 PM, Alan Modra wrote:
> On Sat, Apr 05, 2014 at 09:16:10PM -0600, Sandra Loosemore wrote:
>> On 04/04/2014 07:31 PM, Alan Modra wrote:
>>> On Fri, Apr 04, 2014 at 10:52:22AM -0700, Cesar Philippidis wrote:
>>>> @@ -2016,6 +2016,10 @@ nios2_elf32_build_stubs (struct bfd_link
>>>>      {
>>>>        bfd_size_type size;
>>>>
>>>> +      /* Ignore non-stub sections.  */
>>>> +      if (!strstr (stub_sec->name, STUB_SUFFIX))
>>>> +	continue;
>>>
>>> Um, how do you manage to get non-stub sections in the stub bfd?  Hmm,
>>> via dynobj being set to the stub bfd, I expect.  If so, you could
>>> probably test (stub_sec->flags & SEC_LINKER_CREATED) == 0 instead.
>>
>> I had the same question when looking at this code.  I think this
>> also suffers from the same bug (PR 13049) mentioned in the comment
>> on STUB_SUFFIX in elf32-arm.c.
>>
>> Hmmmm.  I see that stub sections are being created without
>> SEC_LINKER_CREATED (ld/emultempl/nios2elf.em), while dynamic
>> sections created in dynobj do have that flag by default (from
>> ELF_DYNAMIC_SEC_FLAGS).  Other sections that end up in dynobj maybe
>> do, maybe don't
> 
> What other sections?  I don't think there will be any.  You'll either
> have stub sections or dynobj sections.  The latter all have
> SEC_LINKER_CREATED, the former doesn't.
> 
> I handle exactly the same situation in elf64-ppc.c, where dynobj is
> forced to be the stub bfd, without resorting to testing for
> STUB_SUFFIX in the name.
> 
> Note that elf32-arm.c only needs to test for the stub name because the
> ARM backend does create other sections (glue, veneer) in the stub bfd
> besides stub sections.  nios2 doesn't, as far as I can see.

Checking for dynobjs seemed to work out. I've tested this patch on both
ELF and Linux Nios II targets. Is it OK?

Thanks,
Cesar

Attachment: binutils_build-id_stubs.diff
Description: Text document