Add -z bndplt to generate BND prefix in PLT entries

This patch adds "-z bndplt" option Linux/x86-64 linker to generate BND
prefix in PLT entries.  It also updated Linux/x86-64 assembler not to
generate R_X86_64_PLT32_BND nor R_X86_64_PC32_BND relocations.

bfd/

2014-11-18  Igor Zamyatin  <igor.zamyatin@intel.com>

	* elf64-x86-64.c (elf_x86_64_check_relocs): Enable MPX PLT only
	for -z bndplt.

gas/

2014-11-18  Igor Zamyatin  <igor.zamyatin@intel.com>

	* config/tc-i386-intel.c (i386_operator): Remove last argument
	from lex_got call.
	* config/tc-i386.c (reloc): Remove bnd_prefix from parameters'
	list.  Return always BFD_RELOC_32_PCREL.
	* (output_branch): Remove condition for BFD_RELOC_X86_64_PC32_BND.
	* (output_jump): Update call to reloc accordingly.
	* (output_interseg_jump): Likewise.
	* (output_disp): Likewise.
	* (output_imm): Likewise.
	* (x86_cons_fix_new): Likewise.
	* (lex_got): Remove bnd_prefix from parameters' list in macro and
	declarations. Don't use BFD_RELOC_X86_64_PLT32_BND.
	* (x86_cons): Update call to lex_got accordingly.
	* (i386_immediate): Likewise.
	* (i386_displacement): Likewise.
	* (md_apply_fix): Don't use BFD_RELOC_X86_64_PLT32_BND nor
	BFD_RELOC_X86_64_PC32_BND.
	* (tc_gen_reloc): Likewise.

include/

2014-11-18  Igor Zamyatin  <igor.zamyatin@intel.com>

	* bfdlink.h (struct bfd_link_info): Add bndplt.

ld/

2014-11-18  Igor Zamyatin  <igor.zamyatin@intel.com>

	* emulparams/elf_x86_64.sh (BNDPLT): Set to yes for x86_64.
	* emultempl/elf32.em (gld${EMULATION_NAME}_handle_option): Handle
	"-z bndplt" if BNDPLT is yes.
	(gld${EMULATION_NAME}_list_options): Add "-z bndplt" entry.
	* ld.texinfo: Add description for bndplt.

ld/testsuite/

2014-11-18  Igor Zamyatin  <igor.zamyatin@intel.com>

	* testsuite/ld-x86-64/bnd-ifunc-1.d: Add bndplt option.
	* testsuite/ld-x86-64/bnd-ifunc-2.d: Likewise.
	* testsuite/ld-x86-64/bnd-plt-1.d: Likewise.  Update dissassembly
	sections.
	* testsuite/ld-x86-64/mpx.exp: Handle mpx3 and mpx4 tests.
	* testsuite/ld-x86-64/mpx1a.rd: Remove _BND from relocation name.
	* testsuite/ld-x86-64/mpx1c.rd: Likewise.
	* testsuite/ld-x86-64/mpx2a.rd: Likewise.
	* testsuite/ld-x86-64/mpx2c.rd: Likewise.
	* testsuite/ld-x86-64/mpx3.dd: New file.
	* testsuite/ld-x86-64/mpx3a.s: Likewise.
	* testsuite/ld-x86-64/mpx3b.s: Likewise.
	* testsuite/ld-x86-64/mpx4.dd: Likewise.
	* testsuite/ld-x86-64/mpx4a.s: Likewise.
	* testsuite/ld-x86-64/mpx4b.s: Likewise.

6 files changed, 58 insertions, 64 deletions

diff --git a/gas/ChangeLog b/gas/ChangeLog
index cd4467f..943c239 100644
--- a/gas/ChangeLog
+++ b/gas/ChangeLog
@@ -1,3 +1,24 @@
+2014-11-18  Igor Zamyatin  <igor.zamyatin@intel.com>
+
+	* config/tc-i386-intel.c (i386_operator): Remove last argument
+	from lex_got call.
+	* config/tc-i386.c (reloc): Remove bnd_prefix from parameters'
+	list.  Return always BFD_RELOC_32_PCREL.
+	* (output_branch): Remove condition for BFD_RELOC_X86_64_PC32_BND.
+	* (output_jump): Update call to reloc accordingly.
+	* (output_interseg_jump): Likewise.
+	* (output_disp): Likewise.
+	* (output_imm): Likewise.
+	* (x86_cons_fix_new): Likewise.
+	* (lex_got): Remove bnd_prefix from parameters' list in macro and
+	declarations. Don't use BFD_RELOC_X86_64_PLT32_BND.
+	* (x86_cons): Update call to lex_got accordingly.
+	* (i386_immediate): Likewise.
+	* (i386_displacement): Likewise.
+	* (md_apply_fix): Don't use BFD_RELOC_X86_64_PLT32_BND nor
+	BFD_RELOC_X86_64_PC32_BND.
+	* (tc_gen_reloc): Likewise.
+
 2014-11-18  Jan Beulich  <jbeulich@suse.com>
 
 	* config/tc-aarch64.c (s_aarch64_arch_extension): New.
diff --git a/gas/config/tc-i386-intel.c b/gas/config/tc-i386-intel.c
index b55d985..86b96eb 100644
--- a/gas/config/tc-i386-intel.c
+++ b/gas/config/tc-i386-intel.c
@@ -141,9 +141,7 @@ operatorT i386_operator (const char *name, unsigned int operands, char *pc)
 	      int adjust = 0;
 	      char *gotfree_input_line = lex_got (&i.reloc[this_operand],
 						  &adjust,
-						  &intel_state.reloc_types,
-						  (i.bnd_prefix != NULL
-						   || add_bnd_prefix));
+						  &intel_state.reloc_types);
 
 	      if (!gotfree_input_line)
 		break;
diff --git a/gas/config/tc-i386.c b/gas/config/tc-i386.c
index 97e326c..b6ac902 100644
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -2830,7 +2830,6 @@ static bfd_reloc_code_real_type
 reloc (unsigned int size,
        int pcrel,
        int sign,
-       int bnd_prefix,
        bfd_reloc_code_real_type other)
 {
   if (other != NO_RELOC)
@@ -2909,9 +2908,7 @@ reloc (unsigned int size,
 	{
 	case 1: return BFD_RELOC_8_PCREL;
 	case 2: return BFD_RELOC_16_PCREL;
-	case 4: return (bnd_prefix && object_64bit
-			? BFD_RELOC_X86_64_PC32_BND
-			: BFD_RELOC_32_PCREL);
+	case 4: return BFD_RELOC_32_PCREL;
 	case 8: return BFD_RELOC_64_PCREL;
 	}
       as_bad (_("cannot do %u byte pc-relative relocation"), size);
@@ -6776,13 +6773,7 @@ output_branch (void)
 
   /* 1 possible extra opcode + 4 byte displacement go in var part.
      Pass reloc in fr_var.  */
-  frag_var (rs_machine_dependent, 5,
-	    ((!object_64bit
-	      || i.reloc[0] != NO_RELOC 
-	      || (i.bnd_prefix == NULL && !add_bnd_prefix))
-	     ? i.reloc[0]
-	     : BFD_RELOC_X86_64_PC32_BND),
-	    subtype, sym, off, p);
+  frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
 }
 
 static void
@@ -6858,10 +6849,7 @@ output_jump (void)
     }
 
   fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
-		      i.op[0].disps, 1, reloc (size, 1, 1,
-					       (i.bnd_prefix != NULL
-						|| add_bnd_prefix),
-					       i.reloc[0]));
+		      i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
 
   /* All jumps handled here are signed, but don't use a signed limit
      check for 32 and 16 bit jumps as we want to allow wrap around at
@@ -6927,7 +6915,7 @@ output_interseg_jump (void)
     }
   else
     fix_new_exp (frag_now, p - frag_now->fr_literal, size,
-		 i.op[1].imms, 0, reloc (size, 0, 0, 0, i.reloc[1]));
+		 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
   if (i.op[0].imms->X_op != O_constant)
     as_bad (_("can't handle non absolute segment in `%s'"),
 	    i.tm.name);
@@ -7206,10 +7194,7 @@ output_disp (fragS *insn_start_frag, offsetT insn_start_off)
 		}
 
 	      p = frag_more (size);
-	      reloc_type = reloc (size, pcrel, sign,
-				  (i.bnd_prefix != NULL
-				   || add_bnd_prefix),
-				  i.reloc[n]);
+	      reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
 	      if (GOT_symbol
 		  && GOT_symbol == i.op[n].disps->X_add_symbol
 		  && (((reloc_type == BFD_RELOC_32
@@ -7300,7 +7285,7 @@ output_imm (fragS *insn_start_frag, offsetT insn_start_off)
 		sign = 0;
 
 	      p = frag_more (size);
-	      reloc_type = reloc (size, 0, sign, 0, i.reloc[n]);
+	      reloc_type = reloc (size, 0, sign, i.reloc[n]);
 
 	      /*   This is tough to explain.  We end up with this one if we
 	       * have operands that look like
@@ -7393,7 +7378,7 @@ void
 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
 		  expressionS *exp, bfd_reloc_code_real_type r)
 {
-  r = reloc (len, 0, cons_sign, 0, r);
+  r = reloc (len, 0, cons_sign, r);
 
 #ifdef TE_PE
   if (exp->X_op == O_secrel)
@@ -7419,7 +7404,7 @@ x86_address_bytes (void)
 
 #if !(defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined (OBJ_MACH_O)) \
     || defined (LEX_AT)
-# define lex_got(reloc, adjust, types, bnd_prefix) NULL
+# define lex_got(reloc, adjust, types) NULL
 #else
 /* Parse operands of the form
    <symbol>@GOTOFF+<nnn>
@@ -7433,8 +7418,7 @@ x86_address_bytes (void)
 static char *
 lex_got (enum bfd_reloc_code_real *rel,
 	 int *adjust,
-	 i386_operand_type *types,
-	 int bnd_prefix)
+	 i386_operand_type *types)
 {
   /* Some of the relocations depend on the size of what field is to
      be relocated.  But in our callers i386_immediate and i386_displacement
@@ -7569,8 +7553,6 @@ lex_got (enum bfd_reloc_code_real *rel,
 		*adjust = len;
 	      memcpy (tmpbuf + first, past_reloc, second);
 	      tmpbuf[first + second] = '\0';
-	      if (bnd_prefix && *rel == BFD_RELOC_X86_64_PLT32)
-		*rel = BFD_RELOC_X86_64_PLT32_BND;
 	      return tmpbuf;
 	    }
 
@@ -7603,8 +7585,7 @@ lex_got (enum bfd_reloc_code_real *rel,
 static char *
 lex_got (enum bfd_reloc_code_real *rel ATTRIBUTE_UNUSED,
 	 int *adjust ATTRIBUTE_UNUSED,
-	 i386_operand_type *types,
-	 int bnd_prefix ATTRIBUTE_UNUSED)
+	 i386_operand_type *types)
 {
   static const struct
   {
@@ -7705,7 +7686,7 @@ x86_cons (expressionS *exp, int size)
       int adjust = 0;
 
       save = input_line_pointer;
-      gotfree_input_line = lex_got (&got_reloc, &adjust, NULL, 0);
+      gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
       if (gotfree_input_line)
 	input_line_pointer = gotfree_input_line;
 
@@ -7939,9 +7920,7 @@ i386_immediate (char *imm_start)
   save_input_line_pointer = input_line_pointer;
   input_line_pointer = imm_start;
 
-  gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types,
-				(i.bnd_prefix != NULL
-				 || add_bnd_prefix));
+  gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
   if (gotfree_input_line)
     input_line_pointer = gotfree_input_line;
 
@@ -8198,9 +8177,7 @@ i386_displacement (char *disp_start, char *disp_end)
       *displacement_string_end = '0';
     }
 #endif
-  gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types,
-				(i.bnd_prefix != NULL
-				 || add_bnd_prefix));
+  gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
   if (gotfree_input_line)
     input_line_pointer = gotfree_input_line;
 
@@ -9160,8 +9137,7 @@ md_apply_fix (fixS *fixP, valueT *valP, segT seg ATTRIBUTE_UNUSED)
       && (fixP->fx_r_type == BFD_RELOC_32_PCREL
 	  || fixP->fx_r_type == BFD_RELOC_64_PCREL
 	  || fixP->fx_r_type == BFD_RELOC_16_PCREL
-	  || fixP->fx_r_type == BFD_RELOC_8_PCREL
-	  || fixP->fx_r_type == BFD_RELOC_X86_64_PC32_BND)
+	  || fixP->fx_r_type == BFD_RELOC_8_PCREL)
       && !use_rela_relocations)
     {
       /* This is a hack.  There should be a better way to handle this.
@@ -9230,7 +9206,6 @@ md_apply_fix (fixS *fixP, valueT *valP, segT seg ATTRIBUTE_UNUSED)
       {
       case BFD_RELOC_386_PLT32:
       case BFD_RELOC_X86_64_PLT32:
-      case BFD_RELOC_X86_64_PLT32_BND:
 	/* Make the jump instruction point to the address of the operand.  At
 	   runtime we merely add the offset to the actual PLT entry.  */
 	value = -4;
@@ -10354,7 +10329,6 @@ tc_gen_reloc (asection *section ATTRIBUTE_UNUSED, fixS *fixp)
 #endif
 
     case BFD_RELOC_X86_64_PLT32:
-    case BFD_RELOC_X86_64_PLT32_BND:
     case BFD_RELOC_X86_64_GOT32:
     case BFD_RELOC_X86_64_GOTPCREL:
     case BFD_RELOC_386_PLT32:
@@ -10415,10 +10389,7 @@ tc_gen_reloc (asection *section ATTRIBUTE_UNUSED, fixS *fixp)
 	      break;
 	    case 1: code = BFD_RELOC_8_PCREL;  break;
 	    case 2: code = BFD_RELOC_16_PCREL; break;
-	    case 4:
-	      code = (fixp->fx_r_type == BFD_RELOC_X86_64_PC32_BND
-		      ? fixp-> fx_r_type : BFD_RELOC_32_PCREL);
-	      break;
+	    case 4: code = BFD_RELOC_32_PCREL; break;
 #ifdef BFD64
 	    case 8: code = BFD_RELOC_64_PCREL; break;
 #endif
@@ -10511,7 +10482,6 @@ tc_gen_reloc (asection *section ATTRIBUTE_UNUSED, fixS *fixp)
 	switch (code)
 	  {
 	  case BFD_RELOC_X86_64_PLT32:
-	  case BFD_RELOC_X86_64_PLT32_BND:
 	  case BFD_RELOC_X86_64_GOT32:
 	  case BFD_RELOC_X86_64_GOTPCREL:
 	  case BFD_RELOC_X86_64_TLSGD:
diff --git a/gas/testsuite/ChangeLog b/gas/testsuite/ChangeLog
index bf37ba8..40989b3 100644
--- a/gas/testsuite/ChangeLog
+++ b/gas/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2014-11-18  Igor Zamyatin  <igor.zamyatin@intel.com>
+
+	* gas/i386/x86-64-mpx-branch-1.d: Don't use *_BND relocations.
+	* gas/i386/x86-64-mpx-branch-2.d: Likewise.
+
 2014-11-18  Jan Beulich  <jbeulich@suse.com>
 
 	* gas/aarch64/crc32-directive.d: New.
diff --git a/gas/testsuite/gas/i386/x86-64-mpx-branch-1.d b/gas/testsuite/gas/i386/x86-64-mpx-branch-1.d
index 5edb1c7..c070029 100644
--- a/gas/testsuite/gas/i386/x86-64-mpx-branch-1.d
+++ b/gas/testsuite/gas/i386/x86-64-mpx-branch-1.d
@@ -8,8 +8,8 @@
 Disassembly of section .text:
 
 0+ <foo1-0xc>:
-[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 6 <foo1-0x6>	2: R_X86_64_PC32_BND	\*ABS\*\+0x10003c
-[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq c <foo1>	8: R_X86_64_PC32_BND	\*ABS\*\+0x10003c
+[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 6 <foo1-0x6>	2: R_X86_64_PC32	\*ABS\*\+0x10003c
+[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq c <foo1>	8: R_X86_64_PC32	\*ABS\*\+0x10003c
 
 0+c <foo1>:
 [ 	]*[a-f0-9]+:	f2 eb fd             	bnd jmp c <foo1>
@@ -20,9 +20,9 @@ Disassembly of section .text:
 [ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 24 <foo2>
 
 0+24 <foo2>:
-[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq 2a <foo2\+0x6>	26: R_X86_64_PC32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 0f 82 00 00 00 00 	bnd jb 31 <foo2\+0xd>	2d: R_X86_64_PC32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 37 <foo2\+0x13>	33: R_X86_64_PC32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq 3d <foo2\+0x19>	39: R_X86_64_PLT32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 0f 82 00 00 00 00 	bnd jb 44 <foo2\+0x20>	40: R_X86_64_PLT32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 4a <foo2\+0x26>	46: R_X86_64_PLT32_BND	foo-0x4
+[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq 2a <foo2\+0x6>	26: R_X86_64_PC32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 0f 82 00 00 00 00 	bnd jb 31 <foo2\+0xd>	2d: R_X86_64_PC32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 37 <foo2\+0x13>	33: R_X86_64_PC32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq 3d <foo2\+0x19>	39: R_X86_64_PLT32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 0f 82 00 00 00 00 	bnd jb 44 <foo2\+0x20>	40: R_X86_64_PLT32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 4a <foo2\+0x26>	46: R_X86_64_PLT32	foo-0x4
diff --git a/gas/testsuite/gas/i386/x86-64-mpx-branch-2.d b/gas/testsuite/gas/i386/x86-64-mpx-branch-2.d
index 86fb360..5bb6a57 100644
--- a/gas/testsuite/gas/i386/x86-64-mpx-branch-2.d
+++ b/gas/testsuite/gas/i386/x86-64-mpx-branch-2.d
@@ -8,8 +8,8 @@
 Disassembly of section .text:
 
 0+ <foo1-0xc>:
-[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 6 <foo1-0x6>	2: R_X86_64_PC32_BND	\*ABS\*\+0x10003c
-[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq c <foo1>	8: R_X86_64_PC32_BND	\*ABS\*\+0x10003c
+[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 6 <foo1-0x6>	2: R_X86_64_PC32	\*ABS\*\+0x10003c
+[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq c <foo1>	8: R_X86_64_PC32	\*ABS\*\+0x10003c
 
 0+c <foo1>:
 [ 	]*[a-f0-9]+:	f2 eb fd             	bnd jmp c <foo1>
@@ -20,9 +20,9 @@ Disassembly of section .text:
 [ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 24 <foo2>
 
 0+24 <foo2>:
-[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq 2a <foo2\+0x6>	26: R_X86_64_PC32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 0f 82 00 00 00 00 	bnd jb 31 <foo2\+0xd>	2d: R_X86_64_PC32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 37 <foo2\+0x13>	33: R_X86_64_PC32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq 3d <foo2\+0x19>	39: R_X86_64_PLT32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 0f 82 00 00 00 00 	bnd jb 44 <foo2\+0x20>	40: R_X86_64_PLT32_BND	foo-0x4
-[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 4a <foo2\+0x26>	46: R_X86_64_PLT32_BND	foo-0x4
+[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq 2a <foo2\+0x6>	26: R_X86_64_PC32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 0f 82 00 00 00 00 	bnd jb 31 <foo2\+0xd>	2d: R_X86_64_PC32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 37 <foo2\+0x13>	33: R_X86_64_PC32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 e9 00 00 00 00    	bnd jmpq 3d <foo2\+0x19>	39: R_X86_64_PLT32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 0f 82 00 00 00 00 	bnd jb 44 <foo2\+0x20>	40: R_X86_64_PLT32	foo-0x4
+[ 	]*[a-f0-9]+:	f2 e8 00 00 00 00    	bnd callq 4a <foo2\+0x26>	46: R_X86_64_PLT32	foo-0x4