Home Original page

Fix for cris-* regressions with "PATCH: PR ld/13177"

Tristan Gingold gingold@adacore.com
Fri Oct 14 08:07:00 GMT 2011 
On Oct 14, 2011, at 9:01 AM, Hans-Peter Nilsson wrote:

>> Date: Fri, 16 Sep 2011 09:24:22 -0700
>> From: "H.J. Lu" <hjl.tools@gmail.com>
> 
>>> cris-elf:
>>> Running /tmp/hpautotest-binutils/bsrc/src/ld/testsuite/ld-cris/cris.exp ...
>>> FAIL: ld-cris/pic-gc-72
>>> 
>>> ld.log:
>>> Executing on host: sh -c {/tmp/hpautotest-binutils/cris-axis-elf/ld/../binutils/objdump  -s -t -R -p -T tmpdir/dump > tmpdir/dump.out 2>ld.tmp}  /dev/null  (timeout = 300)
>>> exited abnormally with 0, output:BFD: tmpdir/dump(.rela.dyn): relocation 0 has invalid symbol index 16777215
>>> 
>>> ...
>> 
>> The problem is cris backend doesn't use plt.refcount nor got.refcount.
> 
> Wrong, it does, modulo bugs of course.  The problem is "just"
> mismatching bookkeeping between cris_elf_gc_sweep_hook and
> cris_elf_check_relocs.  Fixed thus, committed.
> 
> Yuck for that bookkeeping; a generally error-prone machinery.
> There has to be a better way to "count down".  Something like
> registering the "increment" for each reloc and then just apply a
> general method without having to inspect the reloc and the
> related symbol?  (Who said "C++"? :-)
> 
> Tristan, ok for 2.22 as well?

Yes.

>  What's policy there; are
> target-specific fixes up to target maintainers or is explicit
> approval required for that too?

Asking is very welcome because it would avoid possible collision with the creation of a release.

Tristan.

> 
> bfd:
> 	* elf32-cris.c (cris_elf_gc_sweep_hook) <R_CRIS_16_GOTPLT>
> 	<R_CRIS_32_GOTPLT>: Fix missing update of gotplt refcount for
> 	global symbols.
> 	<R_CRIS_8, R_CRIS_16, R_CRIS_32>: New cases for similar missing
> 	updates of the plt refcount.
> 	(elf_cris_adjust_gotplt_to_got): Assert integrity of the gotplt
> 	refcount in relation to the plt refcount.
> 
> Index: elf32-cris.c
> ===================================================================
> RCS file: /cvs/src/src/bfd/elf32-cris.c,v
> retrieving revision 1.115
> diff -p -u -r1.115 elf32-cris.c
> --- elf32-cris.c	13 Jun 2011 00:59:10 -0000	1.115
> +++ elf32-cris.c	14 Oct 2011 06:43:36 -0000
> @@ -2663,6 +2663,9 @@ cris_elf_gc_sweep_hook (bfd *abfd,
> 	  /* For local symbols, treat these like GOT relocs.  */
> 	  if (h == NULL)
> 	    goto local_got_reloc;
> +	  else
> +	    /* For global symbols, adjust the reloc-specific refcount.  */
> +	    elf_cris_hash_entry (h)->gotplt_refcount--;
> 	  /* Fall through.  */
> 
> 	case R_CRIS_32_PLT_GOTREL:
> @@ -2671,10 +2674,14 @@ cris_elf_gc_sweep_hook (bfd *abfd,
> 	    local_got_refcounts[-1]--;
> 	  /* Fall through.  */
> 
> +	case R_CRIS_8:
> +	case R_CRIS_16:
> +	case R_CRIS_32:
> 	case R_CRIS_8_PCREL:
> 	case R_CRIS_16_PCREL:
> 	case R_CRIS_32_PCREL:
> 	case R_CRIS_32_PLT_PCREL:
> +	  /* Negate the increment we did in cris_elf_check_relocs.  */
> 	  if (h != NULL)
> 	    {
> 	      if (ELF_ST_VISIBILITY (h->other) == STV_DEFAULT
> @@ -2733,6 +2740,11 @@ elf_cris_adjust_gotplt_to_got (h, p)
> {
>   struct bfd_link_info *info = (struct bfd_link_info *) p;
> 
> +  /* A GOTPLT reloc, when activated, is supposed to be included into
> +     the PLT refcount.  */
> +  BFD_ASSERT (h->gotplt_refcount == 0
> +	      || h->gotplt_refcount <= h->root.plt.refcount);
> +
>   /* If nobody wanted a GOTPLT with this symbol, we're done.  */
>   if (h->gotplt_refcount <= 0)
>     return TRUE;
> 
> 
> ld/testsuite:
> 
> 	* ld-cris/pic-gc-72.d: Adjust for dropping unused undefined
> 	dynamic symbol "dsofn".
> 
> Index: ld-cris/pic-gc-72.d
> ===================================================================
> RCS file: /cvs/src/src/ld/testsuite/ld-cris/pic-gc-72.d,v
> retrieving revision 1.1
> diff -p -u -r1.1 pic-gc-72.d
> --- ld-cris/pic-gc-72.d	25 Nov 2008 04:10:47 -0000	1.1
> +++ ld-cris/pic-gc-72.d	14 Oct 2011 06:41:41 -0000
> @@ -19,10 +19,11 @@ Contents of section .dynsym:
> Contents of section .dynstr:
> #...
> Contents of section .text:
> - 0188 0f050f05                             .*
> + 016e 0f050f05                             .*
> Contents of section .dynamic:
> + 2174 .*
> #...
> Contents of section .got:
> - 21e4 8c210000 00000000 00000000           .*
> + 21cc 74210000 00000000 00000000           .*
> Contents of section .data:
> - 21f0 00000000                             .*
> + 21d8 00000000                             .*
> 
> brgds, H-P
More information about the Binutils mailing list