Bogus "final link failed" messages

This patch is a result of noticing messages like the following:
tmpdir/tls32.o: in function `_start':
(.text+0x1c): unresolvable R_PPC_REL24 relocation against symbol `__tls_get_addr_opt'
./ld-new: final link failed: symbol needs debug section which does not exist

The "needs debug section" comes from attempting to use debug info to
find source line information to print the first error message.  That
error isn't of interest to the user, and any previous bfd_error value
which might be of interest is overwritten.  So save and restore
bfd_error around the fancy error reporting code.

That still doesn't leave us with a clean bfd_error.  Now we get
./ld-new: final link failed: nonrepresentable section on output
An unresolvable relocation surely doesn't mean there is some bfd
section that ld doesn't know how to output!  Digging into that showed
a _bfd_elf_section_from_bfd_section failure attempting to find an elf
section correcsponding to ".interp".  So don't go looking for elf
sections on linker created bfd sections.

And then fix the linker testsuite which expected the bogus message..

bfd/
	* elflink.c (elf_fixup_link_order): Don't attempt to find
	an elf_section for linker created bfd sections.
ld/
	* ldmisc.c (vfinfo): Save and restore bfd_error around bfd
	function calls that might set it.
	* testsuite/ld-elf/indirect.exp: Don't expect "nonrepresentable
	section" message.
This commit is contained in:
Alan Modra 2019-10-07 13:37:23 +10:30
parent c3614cffc8
commit 847d518374
5 changed files with 17 additions and 3 deletions

View File

@ -1,3 +1,8 @@
2019-10-07 Alan Modra <amodra@gmail.com>
* elflink.c (elf_fixup_link_order): Don't attempt to find
an elf_section for linker created bfd sections.
2019-10-07 Alan Modra <amodra@gmail.com>
* elf32-ppc.c (ppc_elf_tls_optimize): Don't process R_PPC_TLSLD

View File

@ -11535,7 +11535,8 @@ elf_fixup_link_order (bfd *abfd, asection *o)
{
s = p->u.indirect.section;
sub = s->owner;
if (bfd_get_flavour (sub) == bfd_target_elf_flavour
if ((s->flags & SEC_LINKER_CREATED) == 0
&& bfd_get_flavour (sub) == bfd_target_elf_flavour
&& elf_elfheader (sub)->e_ident[EI_CLASS] == bed->s->elfclass
&& (elfsec = _bfd_elf_section_from_bfd_section (sub, s))
&& elfsec < elf_numsections (sub)

View File

@ -1,3 +1,10 @@
2019-10-07 Alan Modra <amodra@gmail.com>
* ldmisc.c (vfinfo): Save and restore bfd_error around bfd
function calls that might set it.
* testsuite/ld-elf/indirect.exp: Don't expect "nonrepresentable
section" message.
2019-10-07 Alan Modra <amodra@gmail.com>
* testsuite/ld-powerpc/tlsexe.r: Adjust for added TLSMARK symbol.

View File

@ -322,6 +322,7 @@ vfinfo (FILE *fp, const char *fmt, va_list ap, bfd_boolean is_warning)
unsigned int linenumber;
bfd_boolean discard_last;
bfd_boolean done;
bfd_error_type last_bfd_error = bfd_get_error ();
abfd = args[arg_no].reladdr.abfd;
section = args[arg_no].reladdr.sec;
@ -406,6 +407,7 @@ vfinfo (FILE *fp, const char *fmt, va_list ap, bfd_boolean is_warning)
}
if (!done)
lfinfo (fp, "(%pA+0x%v)", section, offset);
bfd_set_error (last_bfd_error);
if (discard_last)
{

View File

@ -123,11 +123,10 @@ set testname "Indirect symbol 1b"
set cmd "$ld -e start -o tmpdir/indirect1 tmpdir/indirect1a.o tmpdir/libindirect1c.so tmpdir/indirect1b.o"
check_link_message "$cmd" [list $string1 $string] "$testname"
set string ": final link failed: nonrepresentable section on output"
set string2 ": no symbol version section for versioned symbol \`foo@FOO\'"
set testname "Indirect symbol 2"
set cmd "$ld -shared -o tmpdir/indirect2.so tmpdir/indirect2.o"
check_link_message "$cmd" [list $string2 $string] "$testname"
check_link_message "$cmd" [list $string2] "$testname"
global NOPIE_CFLAGS NOPIE_LDFLAGS