[Bug ld/18514] New: elfutils reports '.plt' size not multiple of entry size

[deller at gmx dot de]

https://sourceware.org/bugzilla/show_bug.cgi?id=18514

            Bug ID: 18514
           Summary: elfutils reports '.plt' size not multiple of entry
                    size
           Product: binutils
           Version: unspecified
               URL: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=7539
                    81
            Status: NEW
          Severity: normal
          Priority: P2
         Component: ld
          Assignee: unassigned at sourceware dot org
          Reporter: deller at gmx dot de
                CC: dave.anglin at bell dot net
  Target Milestone: ---
            Target: hppa

Created attachment 8357
  --> https://sourceware.org/bugzilla/attachment.cgi?id=8357&action=edit
possible patch

(This bug report is brought upstream because of debian bug
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=753981 )

The elfutils of the elfutils-0.159 package do a validation of the sh_entsize
field in sections.

This check is in [elfutils]/libelf/elf32_updatenull.c Line ~ 373:

              /* Check that the section size is actually a multiple of
                 the entry size.  */
              if (shdr->sh_entsize != 0
                  && unlikely (shdr->sh_size % shdr->sh_entsize != 0)
                  && (elf->flags & ELF_F_PERMISSIVE) == 0)
                {
                  __libelf_seterrno (ELF_E_INVALID_SHENTSIZE);
                  return -1;
                } 

with gdb I could get the values of the variables (the error happens twice, so I
put in two lines here):
(gdb) p *shdr
$3 = {sh_name = 133, sh_type = 8, sh_flags = 7, sh_addr = 113608, sh_offset =
352, sh_size = 1108, sh_link = 0, sh_info = 0, sh_addralign = 4, sh_entsize =
8}
(gdb) p *shdr
$1 = {sh_name = 249, sh_type = 1, sh_flags = 7, sh_addr = 113608, sh_offset =
48072, sh_size = 1108, sh_link = 0, sh_info = 0, sh_addralign = 4, sh_entsize =
8} 

sh_entsize gets set to 8 by binutils in /bfd/elf32-hppa.c via sh_entsize  =
PLT_ENTRY_SIZE  (==8).

The check above now starts to fail on hppa binaries, since sh_size (e.g. 1108) 
is not a multiple of sh_entsize (=8). Reason for that is that in
bfd/elf32-hppa.c adds plt stubs to the plt section which are not a multiple of
8 in size themselves.

This patch seems to resolve that issue:

diff -up ./bfd/elf32-hppa.c.org ./bfd/elf32-hppa.c
--- ./bfd/elf32-hppa.c.org      2015-06-10 14:53:04.458322013 +0200
+++ ./bfd/elf32-hppa.c  2015-06-10 18:03:27.216795894 +0200
@@ -4581,9 +4581,11 @@ elf32_hppa_finish_dynamic_sections (bfd

   if (htab->splt != NULL && htab->splt->size != 0)
     {
-      /* Set plt entry size.  */
+      /* Set plt entry size to 0 instead of PLT_ENTRY_SIZE, since we add the
+        plt stubs and as such the section does not hold a table of fixed-size
+        entries.  */
       elf_section_data (htab->splt->output_section)
-       ->this_hdr.sh_entsize = PLT_ENTRY_SIZE;
+       ->this_hdr.sh_entsize = 0;

       if (htab->need_plt_stub)
        {


Is this a reasonable patch ?
Or would the value "4" be more correct here?
Or should the check in elfutils better be revised?

Thanks,
Helge

-- 
You are receiving this mail because:
You are on the CC list for the bug.