[PATCH] Always define referenced __start_SECNAME/__stop_SECNAME

[H.J. Lu]

On Tue, Jun 13, 2017 at 5:26 AM, Alan Modra <amodra@gmail.com> wrote:
> On Mon, Jun 12, 2017 at 07:10:47PM -0700, H.J. Lu wrote:
>> On Mon, Jun 12, 2017 at 6:54 PM, Alan Modra <amodra@gmail.com> wrote:
>> > On Mon, Jun 12, 2017 at 05:18:58PM -0700, H.J. Lu wrote:
>> >> On Mon, Jun 12, 2017 at 4:32 PM, Alan Modra <amodra@gmail.com> wrote:
>> >> > On Sat, Jun 10, 2017 at 03:46:49PM -0700, H.J. Lu wrote:
>> >> >> This patch changes linker to always define referenced __start_SECNAME and
>> >> >> __stop_SECNAME if the input section name is the same output section name,
>> >> >> which is always true for orphaned sections, and SECNAME is a C identifier.
>> >> >
>> >> > I think this change is reasonable.
>> >> >
>> >> >> Also __start_SECNAME and __stop_SECNAME symbols are marked as hidden by
>> >> >> ELF linker.
>> >> >
>> >> > Why is this necessary?  Also, you make another change in behaviour
>> >>
>> >> It is to make sure that  __start_SECNAME and __stop_SECNAME
>> >> symbols     for section SECNAME in different modules are unique.
>> >
>> > I understand, and it would have been good if these symbols were
>> > defined that way in the beginning, but they weren't.  I'm concerned
>> > that if you make this change as well we will find some code that
>> > relies on the symbols being dynamic.
>>
>> My change fixes a real bug:
>>
>> https://sourceware.org/bugzilla/show_bug.cgi?id=20022
>
> Really, your change is a workaround not a fix.  You avoid exporting
> __start or __stop symbols because doing so shows a problem with gc.  A
> real fix for the PR would make ld do the right thing even in the
> presense of exported dynamic symbols.

Exporting __start_SECNAME and __stop_SECNAME as dynamic
symbols may lead to unexpected behavior.  Reference to __start_SECNAME
and __stop_SECNAME to section SECNAME within a DSO will be
resolved to __start_SECNAME and __stop_SECNAME in another DSO
or executable.  In most cases, it isn't the intended behavior and there are
workaround if it is needed.

>> We can suggest workarounds if some codes depend on that.
>>
>> >>
>> >> > that you don't mention:  __start and __stop symbols were previously
>> >> > defined by ld -Ur, not just at final link.  You'll need to modify
>> >>
>> >> Is there a testcase for this behavior?
>> >
>> > Not that I'm aware of.
>> >
>>
>> >From ld manual:
>>
>> '-Ur'
>>      For anything other than C++ programs, this option is equivalent to
>>      '-r': it generates relocatable output--i.e., an output file that
>>      can in turn serve as input to 'ld'.  When linking C++ programs,
>>      '-Ur' _does_ resolve references to constructors, unlike '-r'.  It
>>      does not work to use '-Ur' on files that were themselves linked
>>      with '-Ur'; once the constructor table has been built, it cannot be
>>      added to.  Use '-Ur' only for the last partial link, and '-r' for
>>      the others.
>>
>> It isn't intended for  __start_SECNAME and __stop_SECNAME.  Check
>> config.build_constructors for __start_SECNAME and __stop_SECNAME
>> was introduced by:
>>
>> commit 71bfc0aef6964c54b8e29466e97fb246cdeb2049
>> Author: Alan Modra <amodra@gmail.com>
>> Date:   Thu Sep 7 07:08:58 2000 +0000
>>
>>     Fix list handling for orphan section output statements.
>
> That's a long time ago.  Prior to that patch, ld defined __start and
> __stop symbols for ld -r, which can lead to unexpected results for the
> unwary.  I can't remember, but I suspect I changed ld to define the
> symbols for -Ur so that people who wanted the previous ld -r behaviour
> for __start and __stop could get that by using ld -Ur.
>

If it is the case, it isn't intended usage for ld -Ur and people shouldn't
depend on it.


-- 
H.J.