This is the mail archive of the binutils@sourceware.org mailing list for the binutils project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [BFD][PR21703]Override the new defined symbol with the old normal symbol when --allow-multiple-definition is provided


Hi,

On 12/07/17 15:17, Renlin Li wrote:
Hi H.J,

On 11/07/17 19:04, H.J. Lu wrote:
On Tue, Jul 11, 2017 at 10:31 AM, Renlin Li <renlin.li@foss.arm.com> wrote:
ld/ChangeLog:

2017-07-11  Renlin Li  <renlin.li@arm.com>

         PR ld/21703
         * testsuite/ld-elf/elf.exp : Run new test case.
         * testsuite/ld-elf/pr21703-1.s: New.
         * testsuite/ld-elf/pr21703-2.s: New.
         * testsuite/ld-elf/pr21703.sd: New.

bfd/ChangeLog:

2017-07-11  Renlin Li  <renlin.li@arm.com>


         PR ld/21703
         * elflink.c (_bfd_elf_merge_symbol): Handle multiple symbol
definition
         case.



Please add some a test to cover versioned_sym in:

This is because _bfd_elf_merge_symbol will be called for versioned symbol with the short
name.

For example, ld/testsuite/ld-elf/pr18735.s
It defines foo and an alias foo@FOO.
in addition to that, _bfd_elf_add_default_symbol will be called for foo@FOO to create a
new symbol which short name (foo in this case). For the original condition, this will
generate a multiple definition error.

"matched" flag seems introduced by you. Here I use it to differentiate the versioned
symbol case from normal symbol definition.



   /* There are multiple definitions of a normal symbol.  */
+  if (olddef && !olddyn && !oldweak && newdef && !newdyn && !newweak
+      && !versioned_sym && h->def_regular)
+    {

What should happen if symbol is defined in linker script or command
line?
This is a good point. I missed this case.
I have read the code a little bit.

Multiple definition handling is broken for this case.

the following command throws error about multiple definitions.
ld --defsym foo=0x10 pr21703-1.o -o tmpdir/dump

The command links fine, generating foo symbol with *MERGED* information from two definition.
ld pr21703-1.o --defsym foo=0x10 -o tmpdir/dump
00000010     4 OBJECT  GLOBAL DEFAULT  ABS foo

The functions to process symbol definitions from object file and command line is different.
Multiple-definition is not handled for command line symbol definition. And the symbol
value is overwritten by the symbol absolute value. The definition section is overwritten
as well to ABS section.

It's intended. Alan had a patch to allow multiple symbol definition without provide
--allow-multiple-definition option.
https://sourceware.org/ml/binutils/2014-12/msg00247.html

The partial symbol information overwritten is the same problem as the original bug I
reported.

So there is another question. What's the expected override behavior for command line, linker script and object/(static/dynamic)library defined symbols?

Is there a defined rule for those cases that I have missed?

Thanks!
Renlin.



With or without the patch, the behavior remains the same.
The symbol defined in command line option don't set h->def_regular flag.
h->ldscript_def flag is set instead.

Regards,
Renlin



Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]