PATCH: PR ld/5913: IA64 linker crashes on supported relocations

[Jim Wilson]

H.J. Lu wrote:
> IA64 Linker crashes on bad input in

It looks like every port with TLS support has this problem, except maybe 
the FRV port.  A bit more than half of them have
>   /* If tls_sec is NULL, we should have signalled an error already.  */
>   if (elf_hash_table (info)->tls_sec == NULL)
>     return 0;
When I tried pasting that into the IA-64 port, I got no error at all.  A 
bit less than half have
>   BFD_ASSERT (elf_hash_table (info)->tls_sec != NULL);
like the IA-64 port.  The elf32-ppc.c port has no checks for a NULL 
tls_sec.  And the FRV port does this differently, it has this code:
>               if (! elf_hash_table (info)->tls_sec)
>                 {
>                   (*info->callbacks->undefined_symbol)
>                     (info, "TLS section", elf_hash_table (info)->dynobj,
>                      frvfdpic_got_section (info), entry->tlsoff_entry, TRUE);
>                   return FALSE;
>                 }
So it looks like the FRV port is the only one that currently handles 
this correctly.

Should we be trying to fix every target here, instead of just the IA-64 
port?

I tried to create my own testcase for x86 to see what happens there, but 
neither the assembler nor objcopy will let me create a broken .o file, 
which make me wonder how you created your testcase.  If this situation 
is very hard to get, maybe we don't need to worry about fixing other 
targets.

As for the patch, it seems reasonable.  We could maybe eliminate the 
code duplication by making elfNN_ia64_tprel_base and 
elfNN_ia64_dtprel_base into macros.  Maybe the error message should say 
something about the missing tls section like the FRV port does?  The 
error message says that the relocation is unsupported, but doesn't give 
any clue as to why.

Jim