[PATCH] Don't check caches_rawsize in elf_link_input_bfd

[PATCH] Don't check caches_rawsize in elf_link_input_bfd

[H.J. Lu]

When backend caches the section contents, the .eh_frame section size
may be increased due to DW_EH_PE_pcrel address encoding conversion
even if caches_rawsize isn't set.  ld-elf/eh6.d is such an example.
We shouldn't check caches_rawsize to support section size increase.

OK for master?

H.J.
---
	* elflink.c (elf_link_input_bfd): Don't caches_rawsize to
	support section size increase.
---
 bfd/elflink.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/bfd/elflink.c b/bfd/elflink.c
index 021db68..20f75f6 100644
--- a/bfd/elflink.c
+++ b/bfd/elflink.c
@@ -10054,14 +10054,15 @@ elf_link_input_bfd (struct elf_final_link_info *flinfo, bfd *input_bfd)
 	}
 
       /* Get the contents of the section.  They have been cached by a
-	 relaxation routine.  Note that o is a section in an input
-	 file, so the contents field will not have been set by any of
-	 the routines which work on output files.  */
+	 relaxation or check_relocs routines.  The .eh_frame section
+	 size may be increased due to DW_EH_PE_pcrel address encoding
+	 conversion.  Note that o is a section in an input file, so the
+	 contents field will not have been set by any of the routines
+	 which work on output files.  */
       if (elf_section_data (o)->this_hdr.contents != NULL)
 	{
 	  contents = elf_section_data (o)->this_hdr.contents;
-	  if (bed->caches_rawsize
-	      && o->rawsize != 0
+	  if (o->rawsize != 0
 	      && o->rawsize < o->size)
 	    {
 	      memcpy (flinfo->contents, contents, o->rawsize);
-- 
2.5.5

Re: [PATCH] Don't check caches_rawsize in elf_link_input_bfd

[Alan Modra]

On Fri, Apr 29, 2016 at 04:10:04AM -0700, H.J. Lu wrote:
> When backend caches the section contents, the .eh_frame section size
> may be increased due to DW_EH_PE_pcrel address encoding conversion
> even if caches_rawsize isn't set.  ld-elf/eh6.d is such an example.
> We shouldn't check caches_rawsize to support section size increase.
> 
> OK for master?

No.  This would break targets that increase section size in
relax_section and realloc section contents.

-- 
Alan Modra
Australia Development Lab, IBM