[MIPS] Add generation of PLT entries with compact jumps for MIPS R6

Add a new option to get the linker to emit PLTs that use compact
branches instead of delay slot branches.

bfd/
	* elfxx-mips.c (LA25_BC): New macro.
	(mips_elf_link_hash_table)<compact_branches>: New field.
	(STUB_JALRC): New macro.
	(mipsr6_o32_exec_plt0_entry_compact): New array.
	(mipsr6_n32_exec_plt0_entry_compact): Likewise.
	(mipsr6_n64_exec_plt0_entry_compact): Likewise.
	(mipsr6_exec_plt_entry_compact): Likewise.
	(mips_elf_create_la25_stub): Use BC instead of J for stubs
	when compact_branches is true.
	(_bfd_mips_elf_finish_dynamic_symbol): Choose the compact
	PLT for MIPSR6 with compact_branches.  Do not reorder the
	compact branches PLT.  Switch the lazy stub for MIPSR6
	with compact_branches to use JALRC.
	(mips_finish_exec_plt): Choose the compact PLT0 for MIPSR6
	when compact_branches is true.
	(_bfd_mips_elf_compact_branches): New function.
	* elfxx-mips.h (_bfd_mips_elf_compact_branches): New prototype.

ld/
	* emultempl/mipself.em (compact_branches): New static variable.
	(mips_create_output_section_statements): Call
	_bfd_mips_elf_compact_branches.
	(PARSE_AND_LIST_PROLOGUE): Add OPTION_COMPACT_BRANCHES and
	OPTION_NO_COMPACT_BRANCHES.
	(PARSE_AND_LIST_LONGOPTS): Add compact-branches,
	no-compact-branches.
	(PARSE_AND_LIST_OPTIONS): Add --compact-branches,
	--no-compact-branches.
	(PARSE_AND_LIST_ARGS_CASES): Handle the above.
	* ld.texinfo: Document --compact-branches, --no-compact-branches.
	* testsuite/ld-mips-elf/pic-and-nonpic-1-r6.dd: New test.
	* testsuite/ld-mips-elf/pic-and-nonpic-1-r6.nd: New test.
	* testsuite/ld-mips-elf/pic-and-nonpic-3a-r6.dd: New test.
	* testsuite/ld-mips-elf/pic-and-nonpic-3a-r6.gd: New test.
	* testsuite/ld-mips-elf/pic-and-nonpic-1a-r6.s: New test source.
	* testsuite/ld-mips-elf/pic-and-nonpic-3a-r6.s: New test source.
	* testsuite/ld-mips-elf/mips-elf.exp: Run the new tests.

3 files changed, 138 insertions, 12 deletions

diff --git a/bfd/ChangeLog b/bfd/ChangeLog
index cbaeab6..deca778 100644
--- a/bfd/ChangeLog
+++ b/bfd/ChangeLog
@@ -1,3 +1,24 @@
+2019-05-21  Matthew Fortune  <matthew.fortune@mips.com>
+	    Faraz Shahbazker  <fshahbazker@wavecomp.com>
+
+	* elfxx-mips.c (LA25_BC): New macro.
+	(mips_elf_link_hash_table)<compact_branches>: New field.
+	(STUB_JALRC): New macro.
+	(mipsr6_o32_exec_plt0_entry_compact): New array.
+	(mipsr6_n32_exec_plt0_entry_compact): Likewise.
+	(mipsr6_n64_exec_plt0_entry_compact): Likewise.
+	(mipsr6_exec_plt_entry_compact): Likewise.
+	(mips_elf_create_la25_stub): Use BC instead of J for stubs
+	when compact_branches is true.
+	(_bfd_mips_elf_finish_dynamic_symbol): Choose the compact
+	PLT for MIPSR6 with compact_branches.  Do not reorder the
+	compact branches PLT.  Switch the lazy stub for MIPSR6
+	with compact_branches to use JALRC.
+	(mips_finish_exec_plt): Choose the compact PLT0 for MIPSR6
+	when compact_branches is true.
+	(_bfd_mips_elf_compact_branches): New function.
+	* elfxx-mips.h (_bfd_mips_elf_compact_branches): New prototype.
+
 2019-05-21  Tamar Christina  <tamar.christina@arm.com>
 
 	PR ld/24373
diff --git a/bfd/elfxx-mips.c b/bfd/elfxx-mips.c
index 74dadf4..8c1ad72 100644
--- a/bfd/elfxx-mips.c
+++ b/bfd/elfxx-mips.c
@@ -292,6 +292,7 @@ struct mips_elf_la25_stub {
 
 #define LA25_LUI(VAL) (0x3c190000 | (VAL))	/* lui t9,VAL */
 #define LA25_J(VAL) (0x08000000 | (((VAL) >> 2) & 0x3ffffff)) /* j VAL */
+#define LA25_BC(VAL) (0xc8000000 | (((VAL) >> 2) & 0x3ffffff)) /* bc VAL */
 #define LA25_ADDIU(VAL) (0x27390000 | (VAL))	/* addiu t9,t9,VAL */
 #define LA25_LUI_MICROMIPS(VAL)						\
   (0x41b90000 | (VAL))				/* lui t9,VAL */
@@ -449,6 +450,9 @@ struct mips_elf_link_hash_table
   /* True if we suppress checks for invalid branches between ISA modes.  */
   bfd_boolean ignore_branch_isa;
 
+  /* True if we are targetting R6 compact branches.  */
+  bfd_boolean compact_branches;
+
   /* True if we're generating code for VxWorks.  */
   bfd_boolean is_vxworks;
 
@@ -920,6 +924,7 @@ static bfd *reldyn_sorting_bfd;
 #define STUB_MOVE 0x03e07825			/* or t7,ra,zero */
 #define STUB_LUI(VAL) (0x3c180000 + (VAL))	/* lui t8,VAL */
 #define STUB_JALR 0x0320f809			/* jalr ra,t9 */
+#define STUB_JALRC 0xf8190000			/* jalrc ra,t9 */
 #define STUB_ORI(VAL) (0x37180000 + (VAL))	/* ori t8,t8,VAL */
 #define STUB_LI16U(VAL) (0x34180000 + (VAL))	/* ori t8,zero,VAL unsigned */
 #define STUB_LI16S(abfd, VAL)						\
@@ -1036,6 +1041,20 @@ static const bfd_vma mips_o32_exec_plt0_entry[] =
   0x2718fffe	/* subu $24, $24, 2					*/
 };
 
+/* The format of the first PLT entry in an O32 executable using compact
+   jumps.  */
+static const bfd_vma mipsr6_o32_exec_plt0_entry_compact[] =
+{
+  0x3c1c0000,	/* lui $28, %hi(&GOTPLT[0])				*/
+  0x8f990000,	/* lw $25, %lo(&GOTPLT[0])($28)				*/
+  0x279c0000,	/* addiu $28, $28, %lo(&GOTPLT[0])			*/
+  0x031cc023,	/* subu $24, $24, $28					*/
+  0x03e07821,	/* move $15, $31	# 32-bit move (addu)		*/
+  0x0018c082,	/* srl $24, $24, 2					*/
+  0x2718fffe,	/* subu $24, $24, 2					*/
+  0xf8190000	/* jalrc $25						*/
+};
+
 /* The format of the first PLT entry in an N32 executable.  Different
    because gp ($28) is not available; we use t2 ($14) instead.  */
 static const bfd_vma mips_n32_exec_plt0_entry[] =
@@ -1050,6 +1069,21 @@ static const bfd_vma mips_n32_exec_plt0_entry[] =
   0x2718fffe	/* subu $24, $24, 2					*/
 };
 
+/* The format of the first PLT entry in an N32 executable using compact
+   jumps.  Different because gp ($28) is not available; we use t2 ($14)
+   instead.  */
+static const bfd_vma mipsr6_n32_exec_plt0_entry_compact[] =
+{
+  0x3c0e0000,	/* lui $14, %hi(&GOTPLT[0])				*/
+  0x8dd90000,	/* lw $25, %lo(&GOTPLT[0])($14)				*/
+  0x25ce0000,	/* addiu $14, $14, %lo(&GOTPLT[0])			*/
+  0x030ec023,	/* subu $24, $24, $14					*/
+  0x03e07821,	/* move $15, $31	# 32-bit move (addu)		*/
+  0x0018c082,	/* srl $24, $24, 2					*/
+  0x2718fffe,	/* subu $24, $24, 2					*/
+  0xf8190000	/* jalrc $25						*/
+};
+
 /* The format of the first PLT entry in an N64 executable.  Different
    from N32 because of the increased size of GOT entries.  */
 static const bfd_vma mips_n64_exec_plt0_entry[] =
@@ -1064,6 +1098,22 @@ static const bfd_vma mips_n64_exec_plt0_entry[] =
   0x2718fffe	/* subu $24, $24, 2					*/
 };
 
+/* The format of the first PLT entry in an N64 executable using compact
+   jumps.  Different from N32 because of the increased size of GOT
+   entries.  */
+static const bfd_vma mipsr6_n64_exec_plt0_entry_compact[] =
+{
+  0x3c0e0000,	/* lui $14, %hi(&GOTPLT[0])				*/
+  0xddd90000,	/* ld $25, %lo(&GOTPLT[0])($14)				*/
+  0x25ce0000,	/* addiu $14, $14, %lo(&GOTPLT[0])			*/
+  0x030ec023,	/* subu $24, $24, $14					*/
+  0x03e0782d,	/* move $15, $31	# 64-bit move (daddu)		*/
+  0x0018c0c2,	/* srl $24, $24, 3					*/
+  0x2718fffe,	/* subu $24, $24, 2					*/
+  0xf8190000	/* jalrc $25						*/
+};
+
+
 /* The format of the microMIPS first PLT entry in an O32 executable.
    We rely on v0 ($2) rather than t8 ($24) to contain the address
    of the GOTPLT entry handled, so this stub may only be used when
@@ -1106,9 +1156,6 @@ static const bfd_vma mips_exec_plt_entry[] =
   0x03200008	/* jr $25					*/
 };
 
-/* In the following PLT entry the JR and ADDIU instructions will
-   be swapped in _bfd_mips_elf_finish_dynamic_symbol because
-   LOAD_INTERLOCKS_P will be true for MIPS R6.  */
 static const bfd_vma mipsr6_exec_plt_entry[] =
 {
   0x3c0f0000,	/* lui $15, %hi(.got.plt entry)			*/
@@ -1117,6 +1164,14 @@ static const bfd_vma mipsr6_exec_plt_entry[] =
   0x03200009	/* jr $25					*/
 };
 
+static const bfd_vma mipsr6_exec_plt_entry_compact[] =
+{
+  0x3c0f0000,	/* lui $15, %hi(.got.plt entry)			*/
+  0x01f90000,	/* l[wd] $25, %lo(.got.plt entry)($15)		*/
+  0x25f80000,	/* addiu $24, $15, %lo(.got.plt entry)		*/
+  0xd8190000	/* jic $25, 0					*/
+};
+
 /* The format of subsequent MIPS16 o32 PLT entries.  We use v0 ($2)
    and v1 ($3) as temporaries because t8 ($24) and t9 ($25) are not
    directly addressable.  */
@@ -10604,6 +10659,8 @@ mips_elf_create_la25_stub (void **slot, void *data)
   asection *s;
   bfd_byte *loc;
   bfd_vma offset, target, target_high, target_low;
+  bfd_vma branch_pc;
+  bfd_signed_vma pcrel_offset = 0;
 
   stub = (struct mips_elf_la25_stub *) *slot;
   hti = (struct mips_htab_traverse_info *) data;
@@ -10627,6 +10684,12 @@ mips_elf_create_la25_stub (void **slot, void *data)
   /* Work out where in the section this stub should go.  */
   offset = stub->offset;
 
+  /* We add 8 here to account for the LUI/ADDIU instructions
+     before the branch instruction.  This cannot be moved down to
+     where pcrel_offset is calculated as 's' is updated in
+     mips_elf_get_la25_target.  */
+  branch_pc = s->output_section->vma + s->output_offset + offset + 8;
+
   /* Work out the target address.  */
   target = mips_elf_get_la25_target (stub, &s);
   target += s->output_section->vma + s->output_offset;
@@ -10634,6 +10697,12 @@ mips_elf_create_la25_stub (void **slot, void *data)
   target_high = ((target + 0x8000) >> 16) & 0xffff;
   target_low = (target & 0xffff);
 
+  /* Calculate the PC of the compact branch instruction (for the case where
+     compact branches are used for either microMIPSR6 or MIPSR6 with
+     compact branches.  Add 4-bytes to account for BC using the PC of the
+     next instruction as the base.  */
+  pcrel_offset = target - (branch_pc + 4);
+
   if (stub->stub_section != htab->strampoline)
     {
       /* This is a simple LUI/ADDIU stub.  Zero out the beginning
@@ -10672,8 +10741,16 @@ mips_elf_create_la25_stub (void **slot, void *data)
       else
 	{
 	  bfd_put_32 (hti->output_bfd, LA25_LUI (target_high), loc);
-	  bfd_put_32 (hti->output_bfd, LA25_J (target), loc + 4);
-	  bfd_put_32 (hti->output_bfd, LA25_ADDIU (target_low), loc + 8);
+	  if (MIPSR6_P (hti->output_bfd) && htab->compact_branches)
+	    {
+	      bfd_put_32 (hti->output_bfd, LA25_ADDIU (target_low), loc + 4);
+	      bfd_put_32 (hti->output_bfd, LA25_BC (pcrel_offset), loc + 8);
+	    }
+	  else
+	    {
+	      bfd_put_32 (hti->output_bfd, LA25_J (target), loc + 4);
+	      bfd_put_32 (hti->output_bfd, LA25_ADDIU (target_low), loc + 8);
+	    }
 	  bfd_put_32 (hti->output_bfd, 0, loc + 12);
 	}
     }
@@ -10830,14 +10907,16 @@ _bfd_mips_elf_finish_dynamic_symbol (bfd *output_bfd,
 	  /* Fill in the PLT entry itself.  */
 
 	  if (MIPSR6_P (output_bfd))
-	    plt_entry = mipsr6_exec_plt_entry;
+	    plt_entry = htab->compact_branches ? mipsr6_exec_plt_entry_compact
+					       : mipsr6_exec_plt_entry;
 	  else
 	    plt_entry = mips_exec_plt_entry;
 	  bfd_put_32 (output_bfd, plt_entry[0] | got_address_high, loc);
 	  bfd_put_32 (output_bfd, plt_entry[1] | got_address_low | load,
 		      loc + 4);
 
-	  if (! LOAD_INTERLOCKS_P (output_bfd))
+	  if (! LOAD_INTERLOCKS_P (output_bfd)
+	      || (MIPSR6_P (output_bfd) && htab->compact_branches))
 	    {
 	      bfd_put_32 (output_bfd, plt_entry[2] | got_address_low, loc + 8);
 	      bfd_put_32 (output_bfd, plt_entry[3], loc + 12);
@@ -11041,8 +11120,12 @@ _bfd_mips_elf_finish_dynamic_symbol (bfd *output_bfd,
 			  stub + idx);
 	      idx += 4;
 	    }
-	  bfd_put_32 (output_bfd, STUB_JALR, stub + idx);
-	  idx += 4;
+
+	  if (!(MIPSR6_P (output_bfd) && htab->compact_branches))
+	    {
+	      bfd_put_32 (output_bfd, STUB_JALR, stub + idx);
+	      idx += 4;
+	    }
 
 	  /* If a large stub is not required and sign extension is not a
 	     problem, then use legacy code in the stub.  */
@@ -11055,6 +11138,10 @@ _bfd_mips_elf_finish_dynamic_symbol (bfd *output_bfd,
 	  else
 	    bfd_put_32 (output_bfd, STUB_LI16S (output_bfd, h->dynindx),
 			stub + idx);
+	  idx += 4;
+
+	  if (MIPSR6_P (output_bfd) && htab->compact_branches)
+	    bfd_put_32 (output_bfd, STUB_JALRC, stub + idx);
 	}
 
       BFD_ASSERT (h->plt.plist->stub_offset <= htab->sstubs->size);
@@ -11428,11 +11515,17 @@ mips_finish_exec_plt (bfd *output_bfd, struct bfd_link_info *info)
   BFD_ASSERT (htab != NULL);
 
   if (ABI_64_P (output_bfd))
-    plt_entry = mips_n64_exec_plt0_entry;
+    plt_entry = (htab->compact_branches
+		 ? mipsr6_n64_exec_plt0_entry_compact
+		 : mips_n64_exec_plt0_entry);
   else if (ABI_N32_P (output_bfd))
-    plt_entry = mips_n32_exec_plt0_entry;
+    plt_entry = (htab->compact_branches
+		 ? mipsr6_n32_exec_plt0_entry_compact
+		 : mips_n32_exec_plt0_entry);
   else if (!htab->plt_header_is_comp)
-    plt_entry = mips_o32_exec_plt0_entry;
+    plt_entry = (htab->compact_branches
+		 ? mipsr6_o32_exec_plt0_entry_compact
+		 : mips_o32_exec_plt0_entry);
   else if (htab->insn32)
     plt_entry = micromips_insn32_o32_exec_plt0_entry;
   else
@@ -14190,6 +14283,16 @@ _bfd_mips_elf_linker_flags (struct bfd_link_info *info, bfd_boolean insn32,
   mips_elf_hash_table (info)->ignore_branch_isa = ignore_branch_isa;
   mips_elf_hash_table (info)->gnu_target = gnu_target;
 }
+
+/* A function that the linker calls to enable use of compact branches in
+   linker generated code for MIPSR6.  */
+
+void
+_bfd_mips_elf_compact_branches (struct bfd_link_info *info, bfd_boolean on)
+{
+  mips_elf_hash_table (info)->compact_branches = on;
+}
+
 
 /* Structure for saying that BFD machine EXTENSION extends BASE.  */
 
diff --git a/bfd/elfxx-mips.h b/bfd/elfxx-mips.h
index de2c0e5..acd9d79 100644
--- a/bfd/elfxx-mips.h
+++ b/bfd/elfxx-mips.h
@@ -150,6 +150,8 @@ extern void _bfd_mips_elf_use_plts_and_copy_relocs
   (struct bfd_link_info *);
 extern void _bfd_mips_elf_linker_flags
   (struct bfd_link_info *, bfd_boolean, bfd_boolean, bfd_boolean);
+extern void _bfd_mips_elf_compact_branches
+  (struct bfd_link_info *, bfd_boolean);
 extern bfd_boolean _bfd_mips_elf_init_stubs
   (struct bfd_link_info *,
    asection *(*) (const char *, asection *, asection *));