PowerPC64 GOT reloc optimisation

Alan Modra amodra@gmail.com
Thu Apr 16 06:14:07 GMT 2020


When the symbol referenced by a GOT reloc is an ifunc, we can't
optimise away the GOT indirection.  Well, we can, but only if a global
entry stub is created with the ifunc symbol redefined to the stub.
But that results in slower code and an indirection via the PLT so
there isn't much to like about that solution.

	* elf64-ppc.c (ppc64_elf_edit_toc): Exclude ifunc from GOT
	optimisation.
	(ppc64_elf_relocate_section): Likewise.

diff --git a/bfd/elf64-ppc.c b/bfd/elf64-ppc.c
index 945f83c7e6..25a3e468a7 100644
--- a/bfd/elf64-ppc.c
+++ b/bfd/elf64-ppc.c
@@ -9371,6 +9371,9 @@ ppc64_elf_edit_toc (struct bfd_link_info *info)
 		  || discarded_section (sym_sec))
 		continue;
 
+	      if ((h ? h->type : ELF_ST_TYPE (sym->st_info)) == STT_GNU_IFUNC)
+		continue;
+
 	      if (!SYMBOL_REFERENCES_LOCAL (info, h))
 		continue;
 
@@ -15886,6 +15889,8 @@ ppc64_elf_relocate_section (bfd *output_bfd,
 	  break;
 
 	case R_PPC64_GOT16_DS:
+	  if ((h ? h->elf.type : ELF_ST_TYPE (sym->st_info)) == STT_GNU_IFUNC)
+	    break;
 	  from = TOCstart + htab->sec_info[input_section->id].toc_off;
 	  if (relocation + addend - from + 0x8000 < 0x10000
 	      && SYMBOL_REFERENCES_LOCAL (info, &h->elf))
@@ -15903,6 +15908,8 @@ ppc64_elf_relocate_section (bfd *output_bfd,
 
 	case R_PPC64_GOT16_LO_DS:
 	case R_PPC64_GOT16_HA:
+	  if ((h ? h->elf.type : ELF_ST_TYPE (sym->st_info)) == STT_GNU_IFUNC)
+	    break;
 	  from = TOCstart + htab->sec_info[input_section->id].toc_off;
 	  if (relocation + addend - from + 0x80008000ULL < 0x100000000ULL
 	      && SYMBOL_REFERENCES_LOCAL (info, &h->elf))
@@ -15924,6 +15931,8 @@ ppc64_elf_relocate_section (bfd *output_bfd,
 	  break;
 
 	case R_PPC64_GOT_PCREL34:
+	  if ((h ? h->elf.type : ELF_ST_TYPE (sym->st_info)) == STT_GNU_IFUNC)
+	    break;
 	  from = (rel->r_offset
 		  + input_section->output_section->vma
 		  + input_section->output_offset);

-- 
Alan Modra
Australia Development Lab, IBM


More information about the Binutils mailing list