[OpenRISC] [PATCH 0/4] OpenRISC large binary fixes

[OpenRISC] [PATCH 0/4] OpenRISC large binary fixes

[Stafford Horne]

These are several fixes to OpenRISC BFD to handle issues mostly related to
linking large files.

There will be a GCC patch then depends on this, a new -mcmodel=large argument is
added to GCC to allow creating GOT relocation values larger than 64k.

The series fixes 3 issues:

 * PR 27624 - Assertion failure when dynamic symol is localized
 * PR 21464 - GOT relocation truncation.
 * PR 27746 - Issue with silent overlows in PLT code.

Stafford Horne (4):
  or1k: Fix issue with plt link failure for local calls
  or1k: Implement relocation R_OR1K_GOT_AHI16 for gotha()
  or1k: Support cmodel=large when generating plt code
  or1k: Fail link if the PLT relocation overflows

 bfd/bfd-in2.h                     |   1 +
 bfd/elf32-or1k.c                  | 182 ++++++++++++++++++++++--------
 bfd/libbfd.h                      |   1 +
 bfd/reloc.c                       |   2 +
 cpu/or1k.opc                      |   7 +-
 gas/testsuite/gas/or1k/reloc-1.d  |   4 +-
 gas/testsuite/gas/or1k/reloc-1.s  |   4 +
 include/elf/or1k.h                |   1 +
 ld/testsuite/ld-or1k/or1k.exp     |   8 ++
 ld/testsuite/ld-or1k/pltlarge1.dd |  38 +++++++
 ld/testsuite/ld-or1k/pltlarge1.s  |  24 ++++
 ld/testsuite/ld-or1k/pltlarge2.dd |  39 +++++++
 ld/testsuite/ld-or1k/pltlarge2.s  |  25 ++++
 ld/testsuite/ld-or1k/pltlib.s     |   1 +
 opcodes/or1k-asm.c                |   7 +-
 15 files changed, 296 insertions(+), 48 deletions(-)
 create mode 100644 ld/testsuite/ld-or1k/pltlarge1.dd
 create mode 100644 ld/testsuite/ld-or1k/pltlarge1.s
 create mode 100644 ld/testsuite/ld-or1k/pltlarge2.dd
 create mode 100644 ld/testsuite/ld-or1k/pltlarge2.s

-- 
2.26.2

[OpenRISC] [PATCH 1/4] or1k: Fix issue with plt link failure for local calls

[Stafford Horne]

When building protobuf we were seeing the assert failure:

    /home/giuliobenetti/git/upstream/or1k-binutils-2.36.1/host/lib/gcc/or1k-buildroot-linux-uclibc/9.3.0/../../../../or1k-buildroot-linux-uclibc/bin/ld:
    BFD (GNU Binutils) 2.36.1 assertion fail elf32-or1k.c:2377
    /home/giuliobenetti/git/upstream/or1k-binutils-2.36.1/host/lib/gcc/or1k-buildroot-linux-uclibc/9.3.0/../../../../or1k-buildroot-linux-uclibc/bin/ld:
    BFD (GNU Binutils) 2.36.1 assertion fail elf32-or1k.c:2377
    /home/giuliobenetti/git/upstream/or1k-binutils-2.36.1/host/lib/gcc/or1k-buildroot-linux-uclibc/9.3.0/../../../../or1k-buildroot-linux-uclibc/bin/ld:
    BFD (GNU Binutils) 2.36.1 assertion fail elf32-or1k.c:2377
    collect2: error: ld returned 1 exit status

This failure happens while writing out PLT entries, there is a check
"BFD_ASSERT (h->dynindx != -1)" to confirm all plt entries have dynamic
symbol attributes.  This was failing for symbols that were
"forced_local" in previous linking code.

The fix adds logic to or1k_elf_adjust_dynamic_symbol to identify
"forced_local" symbols and exclude them from the the PLT.

bfd/ChangeLog:

	PR 27624
	* elf32-or1k.c (or1k_elf_adjust_dynamic_symbol): Change
	condition used to cleanup plt entries to cleanup forced local
	entries.

Cc: Giulio Benetti <giulio.benetti@benettiengineering.com>
---
 bfd/elf32-or1k.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/bfd/elf32-or1k.c b/bfd/elf32-or1k.c
index 013deb406f8..cc6c51220e5 100644
--- a/bfd/elf32-or1k.c
+++ b/bfd/elf32-or1k.c
@@ -2566,11 +2566,10 @@ or1k_elf_adjust_dynamic_symbol (struct bfd_link_info *info,
   if (h->type == STT_FUNC
       || h->needs_plt)
     {
-      if (! bfd_link_pic (info)
-	  && !h->def_dynamic
-	  && !h->ref_dynamic
-	  && h->root.type != bfd_link_hash_undefweak
-	  && h->root.type != bfd_link_hash_undefined)
+      if (h->plt.refcount <= 0
+	  || (SYMBOL_CALLS_LOCAL (info, h)
+	  || (ELF_ST_VISIBILITY (h->other) != STV_DEFAULT
+	      && h->root.type == bfd_link_hash_undefweak)))
 	{
 	  /* This case can occur if we saw a PLT reloc in an input
 	     file, but the symbol was never referred to by a dynamic
-- 
2.26.2

[OpenRISC] [PATCH 2/4] or1k: Implement relocation R_OR1K_GOT_AHI16 for gotha()

[Stafford Horne]

The gotha() relocation mnemonic will be outputted by OpenRISC GCC when
using the -mcmodel=large option.  This relocation is used along with
got() to generate 32-bit GOT offsets.  This increases the previous GOT
offset limit from the previous 16-bit (64K) limit.

This is needed on large binaries where the GOT grows larger than 64k.

bfd/ChangeLog:

	PR 21464
	* bfd-in2.h: Add BFD_RELOC_OR1K_GOT_AHI16 relocation.
	* elf32-or1k.c (or1k_elf_howto_table, or1k_reloc_map): Likewise.
	(or1k_final_link_relocate, or1k_elf_relocate_section,
	or1k_elf_check_relocs): Likewise.
	* libbfd.h (bfd_reloc_code_real_names): Likewise.
	* reloc.c: Likewise.

cpu/ChangeLog:

	PR 21464
	* or1k.opc (or1k_imm16_relocs, parse_reloc): Define parse logic
	for gotha() relocation.

include/ChangeLog:

	PR 21464
	* elf/or1k.h (elf_or1k_reloc_type): Define R_OR1K_GOT_AHI16 number.

opcodes/ChangeLog:

	PR 21464
	* or1k-asm.c: Regenerate.

gas/ChangeLog:

	PR 21464
	* testsuite/gas/or1k/reloc-1.s: Add test for new relocation.
	* testsuite/gas/or1k/reloc-1.d: Add test result for new
	relocation.

Cc: Giulio Benetti <giulio.benetti@benettiengineering.com>

fixup reloc, add tests
---
 bfd/bfd-in2.h                    |  1 +
 bfd/elf32-or1k.c                 | 21 ++++++++++++++++++++-
 bfd/libbfd.h                     |  1 +
 bfd/reloc.c                      |  2 ++
 cpu/or1k.opc                     |  7 ++++++-
 gas/testsuite/gas/or1k/reloc-1.d |  4 +++-
 gas/testsuite/gas/or1k/reloc-1.s |  4 ++++
 include/elf/or1k.h               |  1 +
 opcodes/or1k-asm.c               |  7 ++++++-
 9 files changed, 44 insertions(+), 4 deletions(-)

diff --git a/bfd/bfd-in2.h b/bfd/bfd-in2.h
index cdfb933e30d..ca1b3a537de 100644
--- a/bfd/bfd-in2.h
+++ b/bfd/bfd-in2.h
@@ -5031,6 +5031,7 @@ then it may be truncated to 8 bits.  */
   BFD_RELOC_OR1K_TLS_TPOFF,
   BFD_RELOC_OR1K_TLS_DTPOFF,
   BFD_RELOC_OR1K_TLS_DTPMOD,
+  BFD_RELOC_OR1K_GOT_AHI16,
 
 /* H8 elf Relocations.  */
   BFD_RELOC_H8_DIR16A8,
diff --git a/bfd/elf32-or1k.c b/bfd/elf32-or1k.c
index cc6c51220e5..070397d5944 100644
--- a/bfd/elf32-or1k.c
+++ b/bfd/elf32-or1k.c
@@ -808,6 +808,20 @@ static reloc_howto_type or1k_elf_howto_table[] =
 	 0,			/* Source Mask.  */
 	 0x03ffffff,		/* Dest Mask.  */
 	 true),			/* PC relative offset?  */
+
+  HOWTO (R_OR1K_GOT_AHI16,	/* type */
+	 16,			/* rightshift */
+	 2,			/* size (0 = byte, 1 = short, 2 = long) */
+	 16,			/* bitsize */
+	 false,			/* pc_relative */
+	 0,			/* bitpos */
+	 complain_overflow_signed, /* complain_on_overflow */
+	 bfd_elf_generic_reloc, /* special_function */
+	 "R_OR1K_GOT_AHI16",	/* name */
+	 false,			/* partial_inplace */
+	 0,			/* src_mask */
+	 0xffff,		/* dst_mask */
+	 false),		/* pcrel_offset */
 };
 
 /* Map BFD reloc types to Or1k ELF reloc types.  */
@@ -871,6 +885,7 @@ static const struct or1k_reloc_map or1k_reloc_map[] =
   { BFD_RELOC_OR1K_TLS_IE_LO13,	R_OR1K_TLS_IE_LO13 },
   { BFD_RELOC_OR1K_SLO13,	R_OR1K_SLO13 },
   { BFD_RELOC_OR1K_PLTA26,	R_OR1K_PLTA26 },
+  { BFD_RELOC_OR1K_GOT_AHI16,	R_OR1K_GOT_AHI16 },
 };
 
 /* tls_type is a mask used to track how each symbol is accessed,
@@ -1111,6 +1126,7 @@ or1k_final_link_relocate (reloc_howto_type *howto, bfd *input_bfd,
   switch (howto->type)
     {
     case R_OR1K_AHI16:
+    case R_OR1K_GOT_AHI16:
     case R_OR1K_GOTOFF_AHI16:
     case R_OR1K_TLS_IE_AHI16:
     case R_OR1K_TLS_LE_AHI16:
@@ -1376,6 +1392,7 @@ or1k_elf_relocate_section (bfd *output_bfd,
 	case R_OR1K_GOT16:
 	case R_OR1K_GOT_PG21:
 	case R_OR1K_GOT_LO13:
+	case R_OR1K_GOT_AHI16:
 	  {
 	    bfd_vma off;
 
@@ -1464,7 +1481,8 @@ or1k_elf_relocate_section (bfd *output_bfd,
 	    /* The GOT_PG21 and GOT_LO13 relocs are pc-relative,
 	       while the GOT16 reloc is GOT relative.  */
 	    relocation = got_base + off;
-	    if (r_type == R_OR1K_GOT16)
+	    if (r_type == R_OR1K_GOT16
+		|| r_type == R_OR1K_GOT_AHI16)
 	      relocation -= got_sym_value;
 
 	  /* Addend should be zero.  */
@@ -1993,6 +2011,7 @@ or1k_elf_check_relocs (bfd *abfd,
 	case R_OR1K_GOT16:
 	case R_OR1K_GOT_PG21:
 	case R_OR1K_GOT_LO13:
+	case R_OR1K_GOT_AHI16:
 	case R_OR1K_TLS_GD_HI16:
 	case R_OR1K_TLS_GD_LO16:
 	case R_OR1K_TLS_GD_PG21:
diff --git a/bfd/libbfd.h b/bfd/libbfd.h
index f1d25d0ab1f..793437ef260 100644
--- a/bfd/libbfd.h
+++ b/bfd/libbfd.h
@@ -2771,6 +2771,7 @@ static const char *const bfd_reloc_code_real_names[] = { "@@uninitialized@@",
   "BFD_RELOC_OR1K_TLS_TPOFF",
   "BFD_RELOC_OR1K_TLS_DTPOFF",
   "BFD_RELOC_OR1K_TLS_DTPMOD",
+  "BFD_RELOC_OR1K_GOT_AHI16",
   "BFD_RELOC_H8_DIR16A8",
   "BFD_RELOC_H8_DIR16R8",
   "BFD_RELOC_H8_DIR24A8",
diff --git a/bfd/reloc.c b/bfd/reloc.c
index 2eb0758a82c..896c1723f8d 100644
--- a/bfd/reloc.c
+++ b/bfd/reloc.c
@@ -6204,6 +6204,8 @@ ENUMX
   BFD_RELOC_OR1K_GOTPC_HI16
 ENUMX
   BFD_RELOC_OR1K_GOTPC_LO16
+ENUMX
+  BFD_RELOC_OR1K_GOT_AHI16
 ENUMX
   BFD_RELOC_OR1K_GOT16
 ENUMX
diff --git a/cpu/or1k.opc b/cpu/or1k.opc
index f0adcbb00a5..5d20a1f33a7 100644
--- a/cpu/or1k.opc
+++ b/cpu/or1k.opc
@@ -193,7 +193,7 @@ static const bfd_reloc_code_real_type or1k_imm16_relocs[][6] = {
     BFD_RELOC_OR1K_GOT_LO13,
     BFD_RELOC_UNUSED,
     BFD_RELOC_UNUSED,
-    BFD_RELOC_UNUSED },
+    BFD_RELOC_OR1K_GOT_AHI16 },
   { BFD_RELOC_OR1K_GOTPC_LO16,
     BFD_RELOC_UNUSED,
     BFD_RELOC_UNUSED,
@@ -296,6 +296,11 @@ parse_reloc (const char **strp)
 	str += 5;
 	cls = RCLASS_TPOFF;
       }
+    else if (strncasecmp (str, "got", 3) == 0)
+      {
+	str += 3;
+	cls = RCLASS_GOT;
+      }
 
     if (strncasecmp (str, "hi(", 3) == 0)
       {
diff --git a/gas/testsuite/gas/or1k/reloc-1.d b/gas/testsuite/gas/or1k/reloc-1.d
index d1bcf5608bb..3a001c4ed99 100644
--- a/gas/testsuite/gas/or1k/reloc-1.d
+++ b/gas/testsuite/gas/or1k/reloc-1.d
@@ -68,5 +68,7 @@ OFFSET   TYPE              VALUE
 000000ec R_OR1K_LO13       x
 000000f0 R_OR1K_GOT_LO13   x
 000000f4 R_OR1K_SLO13      x
-
+000000f8 R_OR1K_GOT_AHI16  x
+000000fc R_OR1K_GOT_AHI16  x
+00000100 R_OR1K_GOT_AHI16  x
 
diff --git a/gas/testsuite/gas/or1k/reloc-1.s b/gas/testsuite/gas/or1k/reloc-1.s
index e76abef6532..562609aa869 100644
--- a/gas/testsuite/gas/or1k/reloc-1.s
+++ b/gas/testsuite/gas/or1k/reloc-1.s
@@ -74,3 +74,7 @@
 	l.lbz	r5,po(x)(r3)
 	l.lbz	r5,gotpo(x)(r3)
 	l.sb	po(x)(r3),r6
+
+	l.movhi	r4,gotha(x)
+	l.ori	r3,r4,gotha(x)
+	l.addi	r3,r4,gotha(x)
diff --git a/include/elf/or1k.h b/include/elf/or1k.h
index 7f281481061..69ab90584a2 100644
--- a/include/elf/or1k.h
+++ b/include/elf/or1k.h
@@ -77,6 +77,7 @@ START_RELOC_NUMBERS (elf_or1k_reloc_type)
   RELOC_NUMBER (R_OR1K_TLS_IE_LO13,   51)
   RELOC_NUMBER (R_OR1K_SLO13,         52)
   RELOC_NUMBER (R_OR1K_PLTA26,        53)
+  RELOC_NUMBER (R_OR1K_GOT_AHI16,     54)
 END_RELOC_NUMBERS (R_OR1K_max)
 
 #define EF_OR1K_NODELAY (1UL << 0)
diff --git a/opcodes/or1k-asm.c b/opcodes/or1k-asm.c
index 045ab6230b6..fdccf67f9e1 100644
--- a/opcodes/or1k-asm.c
+++ b/opcodes/or1k-asm.c
@@ -177,7 +177,7 @@ static const bfd_reloc_code_real_type or1k_imm16_relocs[][6] = {
     BFD_RELOC_OR1K_GOT_LO13,
     BFD_RELOC_UNUSED,
     BFD_RELOC_UNUSED,
-    BFD_RELOC_UNUSED },
+    BFD_RELOC_OR1K_GOT_AHI16 },
   { BFD_RELOC_OR1K_GOTPC_LO16,
     BFD_RELOC_UNUSED,
     BFD_RELOC_UNUSED,
@@ -280,6 +280,11 @@ parse_reloc (const char **strp)
 	str += 5;
 	cls = RCLASS_TPOFF;
       }
+    else if (strncasecmp (str, "got", 3) == 0)
+      {
+	str += 3;
+	cls = RCLASS_GOT;
+      }
 
     if (strncasecmp (str, "hi(", 3) == 0)
       {
-- 
2.26.2

[OpenRISC] [PATCH 3/4] or1k: Support cmodel=large when generating plt code

[Stafford Horne]

The current PLT generation code will generate invalid code when the PLT
relocation offset exceeds 64k.  This fixes the issue by taking advantage
of OpenRISC GCC's -mcmodel=large option.

When gcc is run with -mcmodel=large it will output gotha relocations to
handle GOT offsets larger than 64k.  Detect these relocations and use
them to trigger genration of PLT code that can handle large GOT offsets.

The "large" plt code needs 2 extra instructions to create 32-bit offsets.

I use a saw_gotha as its similar to what we did with saw_plta and it
seems to work.

bfd/ChangeLog:

	PR 27746
	* elf32-or1k.c (PLT_ENTRY_SIZE_LARGE, PLT_MAX_INSN_COUNT,
	OR1K_ADD, OR1K_ORI): New macros to help with plt creation.
	(elf_or1k_link_hash_table): New field saw_gotha.
	(elf_or1k_plt_entry_size): New functions.
	(or1k_write_plt_entry): Update to support variable size PLTs.
	(or1k_elf_finish_dynamic_sections): Support writing large PLT
	entries.
	(or1k_elf_finish_dynamic_symbol): Support generating PLT entries
	for large GOT offsets.
	(allocate_dynrelocs): Use elf_or1k_plt_entry_size to account for
	PLT size.

ld/ChangeLog:

	PR 27746
	testsuite/ld-or1k/or1k.exp (or1kplttests): Add tests for new PLT
	entry types.
	testsuite/ld-or1k/pltlarge1.dd: New file.
	testsuite/ld-or1k/pltlarge1.s: New file.
	testsuite/ld-or1k/pltlarge2.dd: New file.
	testsuite/ld-or1k/pltlarge2.s: New file
	testsuite/ld-or1k/pltlib.s (x): Define size to avoid link
	failure.
---
 bfd/elf32-or1k.c                  | 146 ++++++++++++++++++++++--------
 ld/testsuite/ld-or1k/or1k.exp     |   8 ++
 ld/testsuite/ld-or1k/pltlarge1.dd |  38 ++++++++
 ld/testsuite/ld-or1k/pltlarge1.s  |  24 +++++
 ld/testsuite/ld-or1k/pltlarge2.dd |  39 ++++++++
 ld/testsuite/ld-or1k/pltlarge2.s  |  25 +++++
 ld/testsuite/ld-or1k/pltlib.s     |   1 +
 7 files changed, 241 insertions(+), 40 deletions(-)
 create mode 100644 ld/testsuite/ld-or1k/pltlarge1.dd
 create mode 100644 ld/testsuite/ld-or1k/pltlarge1.s
 create mode 100644 ld/testsuite/ld-or1k/pltlarge2.dd
 create mode 100644 ld/testsuite/ld-or1k/pltlarge2.s

diff --git a/bfd/elf32-or1k.c b/bfd/elf32-or1k.c
index 070397d5944..1a5b3319646 100644
--- a/bfd/elf32-or1k.c
+++ b/bfd/elf32-or1k.c
@@ -30,10 +30,14 @@
 #define N_ONES(X)	(((bfd_vma)2 << (X)) - 1)
 
 #define PLT_ENTRY_SIZE 16
+#define PLT_ENTRY_SIZE_LARGE (6*4)
+#define PLT_MAX_INSN_COUNT 6
 
 #define OR1K_MOVHI(D)		(0x18000000 | (D << 21))
 #define OR1K_ADRP(D)		(0x08000000 | (D << 21))
 #define OR1K_LWZ(D,A)		(0x84000000 | (D << 21) | (A << 16))
+#define OR1K_ADD(D,A,B)		(0xE0000000 | (D << 21) | (A << 16) | (B << 11))
+#define OR1K_ORI(D,A)		(0xA8000000 | (D << 21) | (A << 16))
 #define OR1K_ORI0(D)		(0xA8000000 | (D << 21))
 #define OR1K_JR(B)		(0x44000000 | (B << 11))
 #define OR1K_NOP		0x15000000
@@ -932,8 +936,15 @@ struct elf_or1k_link_hash_table
   struct elf_link_hash_table root;
 
   bool saw_plta;
+  bool saw_gotha;
 };
 
+static size_t
+elf_or1k_plt_entry_size (struct elf_or1k_link_hash_table *htab)
+{
+  return htab->saw_gotha ? PLT_ENTRY_SIZE_LARGE : PLT_ENTRY_SIZE;
+}
+
 /* Get the ELF linker hash table from a link_info structure.  */
 #define or1k_elf_hash_table(p) \
   ((is_elf_hash_table ((p)->hash)					\
@@ -2008,10 +2019,12 @@ or1k_elf_check_relocs (bfd *abfd,
 	    }
 	  break;
 
+	case R_OR1K_GOT_AHI16:
+	  htab->saw_gotha = true;
+	  /* FALLTHRU */
 	case R_OR1K_GOT16:
 	case R_OR1K_GOT_PG21:
 	case R_OR1K_GOT_LO13:
-	case R_OR1K_GOT_AHI16:
 	case R_OR1K_TLS_GD_HI16:
 	case R_OR1K_TLS_GD_LO16:
 	case R_OR1K_TLS_GD_PG21:
@@ -2211,33 +2224,46 @@ or1k_elf_check_relocs (bfd *abfd,
 }
 
 static void
-or1k_write_plt_entry (bfd *output_bfd, bfd_byte *contents, unsigned insn1,
-		      unsigned insn2, unsigned insn3, unsigned insnj)
+or1k_write_plt_entry (bfd *output_bfd, bfd_byte *contents, unsigned insnj,
+		      unsigned insns[], size_t insn_count)
 {
   unsigned nodelay = elf_elfheader (output_bfd)->e_flags & EF_OR1K_NODELAY;
-  unsigned insn4;
+  unsigned output_insns[PLT_MAX_INSN_COUNT];
+
+  /* Copy instructions into the output buffer.  */
+  for (size_t i = 0; i < insn_count; i++)
+    output_insns[i] = insns[i];
 
   /* Honor the no-delay-slot setting.  */
-  if (insn3 == OR1K_NOP)
+  if (insns[insn_count-1] == OR1K_NOP)
     {
-      insn4 = insn3;
+      unsigned slot1, slot2;
+
       if (nodelay)
-	insn3 = insnj;
+	slot1 = insns[insn_count-2], slot2 = insnj;
       else
-	insn3 = insn2, insn2 = insnj;
+	slot1 = insnj, slot2 = insns[insn_count-2];
+
+      output_insns[insn_count-2] = slot1;
+      output_insns[insn_count-1] = slot2;
+      output_insns[insn_count]   = OR1K_NOP;
     }
   else
     {
+      unsigned slot1, slot2;
+
       if (nodelay)
-	insn4 = insnj;
+	slot1 = insns[insn_count-1], slot2 = insnj;
       else
-	insn4 = insn3, insn3 = insnj;
+	slot1 = insnj, slot2 = insns[insn_count-1];
+
+      output_insns[insn_count-1] = slot1;
+      output_insns[insn_count]   = slot2;
     }
 
-  bfd_put_32 (output_bfd, insn1, contents);
-  bfd_put_32 (output_bfd, insn2, contents + 4);
-  bfd_put_32 (output_bfd, insn3, contents + 8);
-  bfd_put_32 (output_bfd, insn4, contents + 12);
+  /* Write out the output buffer.  */
+  for (size_t i = 0; i < (insn_count+1); i++)
+    bfd_put_32 (output_bfd, output_insns[i], contents + (i*4));
 }
 
 /* Finish up the dynamic sections.  */
@@ -2304,7 +2330,8 @@ or1k_elf_finish_dynamic_sections (bfd *output_bfd,
       splt = htab->root.splt;
       if (splt && splt->size > 0)
 	{
-	  unsigned plt0, plt1, plt2;
+	  unsigned plt[PLT_MAX_INSN_COUNT];
+	  size_t plt_insn_count = 3;
 	  bfd_vma got_addr = sgot->output_section->vma + sgot->output_offset;
 
 	  /* Note we force 16 byte alignment on the .got, so that
@@ -2315,27 +2342,35 @@ or1k_elf_finish_dynamic_sections (bfd *output_bfd,
 	      bfd_vma pc = splt->output_section->vma + splt->output_offset;
 	      unsigned pa = ((got_addr >> 13) - (pc >> 13)) & 0x1fffff;
 	      unsigned po = got_addr & 0x1fff;
-	      plt0 = OR1K_ADRP(12) | pa;
-	      plt1 = OR1K_LWZ(15,12) | (po + 8);
-	      plt2 = OR1K_LWZ(12,12) | (po + 4);
+	      plt[0] = OR1K_ADRP(12) | pa;
+	      plt[1] = OR1K_LWZ(15,12) | (po + 8);
+	      plt[2] = OR1K_LWZ(12,12) | (po + 4);
 	    }
 	  else if (bfd_link_pic (info))
 	    {
-	      plt0 = OR1K_LWZ(15, 16) | 8;	/* .got+8 */
-	      plt1 = OR1K_LWZ(12, 16) | 4;	/* .got+4 */
-	      plt2 = OR1K_NOP;
+	      plt[0] = OR1K_LWZ(15, 16) | 8;	/* .got+8 */
+	      plt[1] = OR1K_LWZ(12, 16) | 4;	/* .got+4 */
+	      plt[2] = OR1K_NOP;
 	    }
 	  else
 	    {
 	      unsigned ha = ((got_addr + 0x8000) >> 16) & 0xffff;
 	      unsigned lo = got_addr & 0xffff;
-	      plt0 = OR1K_MOVHI(12) | ha;
-	      plt1 = OR1K_LWZ(15,12) | (lo + 8);
-	      plt2 = OR1K_LWZ(12,12) | (lo + 4);
+	      plt[0] = OR1K_MOVHI(12) | ha;
+	      plt[1] = OR1K_LWZ(15,12) | (lo + 8);
+	      plt[2] = OR1K_LWZ(12,12) | (lo + 4);
 	    }
 
-	  or1k_write_plt_entry (output_bfd, splt->contents,
-				plt0, plt1, plt2, OR1K_JR(15));
+	  /* Fill large entries with 2 extra nops.  */
+	  if (htab->saw_gotha)
+	    {
+	      plt[3] = OR1K_NOP;
+	      plt[4] = OR1K_NOP;
+	      plt_insn_count = 5;
+	    }
+
+	  or1k_write_plt_entry (output_bfd, splt->contents, OR1K_JR(15),
+				plt, plt_insn_count);
 
 	  elf_section_data (splt->output_section)->this_hdr.sh_entsize = 4;
 	}
@@ -2378,7 +2413,8 @@ or1k_elf_finish_dynamic_symbol (bfd *output_bfd,
 
   if (h->plt.offset != (bfd_vma) -1)
     {
-      unsigned int plt0, plt1, plt2;
+      unsigned int plt[PLT_MAX_INSN_COUNT];
+      size_t plt_insn_count = 3;
       asection *splt;
       asection *sgot;
       asection *srela;
@@ -2407,7 +2443,7 @@ or1k_elf_finish_dynamic_symbol (bfd *output_bfd,
 	 corresponds to this symbol.  This is the index of this symbol
 	 in all the symbols for which we are making plt entries.  The
 	 first entry in the procedure linkage table is reserved.  */
-      plt_index = h->plt.offset / PLT_ENTRY_SIZE - 1;
+      plt_index = h->plt.offset / elf_or1k_plt_entry_size (htab) - 1;
       plt_addr = plt_base_addr + h->plt.offset;
       plt_reloc = plt_index * sizeof (Elf32_External_Rela);
 
@@ -2422,27 +2458,57 @@ or1k_elf_finish_dynamic_symbol (bfd *output_bfd,
 	{
 	  unsigned pa = ((got_addr >> 13) - (plt_addr >> 13)) & 0x1fffff;
 	  unsigned po = (got_addr & 0x1fff);
-	  plt0 = OR1K_ADRP(12) | pa;
-	  plt1 = OR1K_LWZ(12,12) | po;
-	  plt2 = OR1K_ORI0(11) | plt_reloc;
+	  plt[0] = OR1K_ADRP(12) | pa;
+	  plt[1] = OR1K_LWZ(12,12) | po;
+	  plt[2] = OR1K_ORI0(11) | plt_reloc;
 	}
       else if (bfd_link_pic (info))
 	{
-	  plt0 = OR1K_LWZ(12,16) | got_offset;
-	  plt1 = OR1K_ORI0(11) | plt_reloc;
-	  plt2 = OR1K_NOP;
+	  if (htab->saw_gotha)
+	    {
+	      unsigned gotha = ((got_offset + 0x8000) >> 16) & 0xffff;
+	      unsigned got = got_offset & 0xffff;
+	      unsigned pltrelhi = (plt_reloc >> 16) & 0xffff;
+	      unsigned pltrello = plt_reloc & 0xffff;
+
+	      plt[0] = OR1K_MOVHI(12) | gotha;
+	      plt[1] = OR1K_ADD(12,12,16);
+	      plt[2] = OR1K_LWZ(12,12) | got;
+	      plt[3] = OR1K_MOVHI(11) | pltrelhi;
+	      plt[4] = OR1K_ORI(11,11) | pltrello;
+	      plt_insn_count = 5;
+	    }
+	  else
+	    {
+	      plt[0] = OR1K_LWZ(12,16) | got_offset;
+	      plt[1] = OR1K_ORI0(11) | plt_reloc;
+	      plt[2] = OR1K_NOP;
+	    }
 	}
       else
 	{
 	  unsigned ha = ((got_addr + 0x8000) >> 16) & 0xffff;
 	  unsigned lo = got_addr & 0xffff;
-	  plt0 = OR1K_MOVHI(12) | ha;
-	  plt1 = OR1K_LWZ(12,12) | lo;
-	  plt2 = OR1K_ORI0(11) | plt_reloc;
+	  plt[0] = OR1K_MOVHI(12) | ha;
+	  plt[1] = OR1K_LWZ(12,12) | lo;
+	  plt[2] = OR1K_ORI0(11) | plt_reloc;
+	}
+
+      /* For large code model we fixup the non-PIC PLT relocation instructions
+	 here.  */
+      if (htab->saw_gotha && !bfd_link_pic (info))
+	{
+	  unsigned pltrelhi = (plt_reloc >> 16) & 0xffff;
+	  unsigned pltrello = plt_reloc & 0xffff;
+
+	  plt[2] = OR1K_MOVHI(11) | pltrelhi;
+	  plt[3] = OR1K_ORI(11,11) | pltrello;
+	  plt[4] = OR1K_NOP;
+	  plt_insn_count = 5;
 	}
 
       or1k_write_plt_entry (output_bfd, splt->contents + h->plt.offset,
-			    plt0, plt1, plt2, OR1K_JR(12));
+			    OR1K_JR(12), plt, plt_insn_count);
 
       /* Fill in the entry in the global offset table.  We initialize it to
 	 point to the top of the plt.  This is done to lazy lookup the actual
@@ -2770,7 +2836,7 @@ allocate_dynrelocs (struct elf_link_hash_entry *h, void * inf)
 	  /* If this is the first .plt entry, make room for the special
 	     first entry.  */
 	  if (splt->size == 0)
-	    splt->size = PLT_ENTRY_SIZE;
+	    splt->size = elf_or1k_plt_entry_size (htab);
 
 	  h->plt.offset = splt->size;
 
@@ -2787,7 +2853,7 @@ allocate_dynrelocs (struct elf_link_hash_entry *h, void * inf)
 	    }
 
 	  /* Make room for this entry.  */
-	  splt->size += PLT_ENTRY_SIZE;
+	  splt->size += elf_or1k_plt_entry_size (htab);
 
 	  /* We also need to make an entry in the .got.plt section, which
 	     will be placed in the .got section by the linker script.  */
diff --git a/ld/testsuite/ld-or1k/or1k.exp b/ld/testsuite/ld-or1k/or1k.exp
index 8bb943aacc9..a17ce0eee8e 100644
--- a/ld/testsuite/ld-or1k/or1k.exp
+++ b/ld/testsuite/ld-or1k/or1k.exp
@@ -53,6 +53,14 @@ set or1kplttests {
      "" {plt1.s}
      {{objdump -dr plt1.x.dd}}
      "plt1.x"}
+    {"PLT exec large" "tmpdir/libpltlib.so" ""
+     "" {pltlarge1.s}
+     {{objdump -dr pltlarge1.dd}}
+     "pltlarge1.x"}
+    {"PLT large -fpic -shared" "-fpic -shared" ""
+     "" {pltlarge2.s}
+     {{objdump -dr pltlarge2.dd}}
+     "pltlarge2.x"}
 }
 
 # Not implemented yet
diff --git a/ld/testsuite/ld-or1k/pltlarge1.dd b/ld/testsuite/ld-or1k/pltlarge1.dd
new file mode 100644
index 00000000000..623dfa0557c
--- /dev/null
+++ b/ld/testsuite/ld-or1k/pltlarge1.dd
@@ -0,0 +1,38 @@
+
+.*\.x:     file format elf32-or1k
+
+
+Disassembly of section \.plt:
+
+00002170 <\.plt>:
+ +[0-9a-f]+:	19 80 00 00 	l\.movhi r12,0x0
+ +[0-9a-f]+:	85 ec [0-9a-f]+ [0-9a-f]+ 	l\.lwz r15,[0-9]+\(r12\)
+ +[0-9a-f]+:	85 8c [0-9a-f]+ [0-9a-f]+ 	l\.lwz r12,[0-9]+\(r12\)
+ +[0-9a-f]+:	44 00 78 00 	l\.jr r15
+ +[0-9a-f]+:	15 00 00 00 	l\.nop 0x0
+ +[0-9a-f]+:	15 00 00 00 	l\.nop 0x0
+ +[0-9a-f]+:	19 80 00 00 	l\.movhi r12,0x0
+ +[0-9a-f]+:	85 8c [0-9a-f]+ [0-9a-f]+ 	l\.lwz r12,[0-9]+\(r12\)
+ +[0-9a-f]+:	19 60 00 00 	l\.movhi r11,0x0
+ +[0-9a-f]+:	44 00 60 00 	l\.jr r12
+ +[0-9a-f]+:	a9 6b 00 00 	l\.ori r11,r11,0x0
+ +[0-9a-f]+:	15 00 00 00 	l\.nop 0x0
+
+Disassembly of section \.text:
+
+000021a0 <_start>:
+ +[0-9a-f]+:	9c 21 ff fc 	l\.addi r1,r1,-4
+ +[0-9a-f]+:	d4 01 48 00 	l\.sw 0\(r1\),r9
+ +[0-9a-f]+:	04 00 00 02 	l\.jal [0-9a-f]+ <_start\+0x10>
+ +[0-9a-f]+:	1a 60 00 00 	l\.movhi r19,0x0
+ +[0-9a-f]+:	aa 73 [0-9a-f]+ [0-9a-f]+ 	l\.ori r19,r19,0x[0-9a-f]+
+ +[0-9a-f]+:	e2 73 48 00 	l\.add r19,r19,r9
+ +[0-9a-f]+:	1a 20 00 00 	l\.movhi r17,0x0
+ +[0-9a-f]+:	e2 31 98 00 	l\.add r17,r17,r19
+ +[0-9a-f]+:	86 31 00 10 	l\.lwz r17,16\(r17\)
+ +[0-9a-f]+:	84 71 00 00 	l\.lwz r3,0\(r17\)
+ +[0-9a-f]+:	07 ff ff f0 	l\.jal [0-9a-f]+ <\.plt\+0x18>
+ +[0-9a-f]+:	15 00 00 00 	l\.nop 0x0
+ +[0-9a-f]+:	85 21 00 00 	l\.lwz r9,0\(r1\)
+ +[0-9a-f]+:	44 00 48 00 	l\.jr r9
+ +[0-9a-f]+:	9c 21 00 04 	l\.addi r1,r1,4
diff --git a/ld/testsuite/ld-or1k/pltlarge1.s b/ld/testsuite/ld-or1k/pltlarge1.s
new file mode 100644
index 00000000000..42b16db425c
--- /dev/null
+++ b/ld/testsuite/ld-or1k/pltlarge1.s
@@ -0,0 +1,24 @@
+	.data
+	.p2align 16
+
+	.text
+	.globl	_start
+_start:
+	l.addi	r1, r1, -4
+	l.sw	0(r1), r9
+
+	l.jal	8
+	 l.movhi	r19, gotpchi(_GLOBAL_OFFSET_TABLE_-4)
+	l.ori	r19, r19, gotpclo(_GLOBAL_OFFSET_TABLE_+0)
+	l.add	r19, r19, r9
+
+	l.movhi	r17, gotha(x)
+	l.add	r17, r17, r19
+	l.lwz	r17, got(x)(r17)
+	l.lwz	r3, 0(r17)
+
+	l.jal	plt(func)
+	 l.nop
+	l.lwz	r9, 0(r1)
+	l.jr	r9
+	 l.addi	r1, r1, 4
diff --git a/ld/testsuite/ld-or1k/pltlarge2.dd b/ld/testsuite/ld-or1k/pltlarge2.dd
new file mode 100644
index 00000000000..140cc96777c
--- /dev/null
+++ b/ld/testsuite/ld-or1k/pltlarge2.dd
@@ -0,0 +1,39 @@
+
+.*\.x:     file format elf32-or1k
+
+
+Disassembly of section \.plt:
+
+[0-9a-f]+ <\.plt>:
+ +[0-9a-f]+:	85 f0 00 08 	l\.lwz r15,8\(r16\)
+ +[0-9a-f]+:	85 90 00 04 	l\.lwz r12,4\(r16\)
+ +[0-9a-f]+:	15 00 00 00 	l\.nop 0x0
+ +[0-9a-f]+:	44 00 78 00 	l\.jr r15
+ +[0-9a-f]+:	15 00 00 00 	l\.nop 0x0
+ +[0-9a-f]+:	15 00 00 00 	l\.nop 0x0
+ +[0-9a-f]+:	19 80 00 00 	l\.movhi r12,0x0
+ +[0-9a-f]+:	e1 8c 80 00 	l\.add r12,r12,r16
+ +[0-9a-f]+:	85 8c 00 0c 	l\.lwz r12,12\(r12\)
+ +[0-9a-f]+:	19 60 00 00 	l\.movhi r11,0x0
+ +[0-9a-f]+:	44 00 60 00 	l\.jr r12
+ +[0-9a-f]+:	a9 6b 00 00 	l\.ori r11,r11,0x0
+
+Disassembly of section \.text:
+
+[0-9a-f]+ <test>:
+ +[0-9a-f]+:	9c 21 ff f8 	l\.addi r1,r1,-8
+ +[0-9a-f]+:	d4 01 80 00 	l\.sw 0\(r1\),r16
+ +[0-9a-f]+:	d4 01 48 04 	l\.sw 4\(r1\),r9
+ +[0-9a-f]+:	04 00 [0-9a-f]+ [0-9a-f]+ 	l\.jal [0-9a-f]+ <test\+0x14>
+ +[0-9a-f]+:	1a 00 00 00 	l\.movhi r16,0x0
+ +[0-9a-f]+:	aa 10 [0-9a-f]+ [0-9a-f]+ 	l\.ori r16,r16,0x[0-9a-f]+
+ +[0-9a-f]+:	e2 10 48 00 	l\.add r16,r16,r9
+ +[0-9a-f]+:	1a 20 00 00 	l\.movhi r17,0x0
+ +[0-9a-f]+:	e2 31 80 00 	l\.add r17,r17,r16
+ +[0-9a-f]+:	86 31 00 10 	l\.lwz r17,16\(r17\)
+ +[0-9a-f]+:	07 ff ff f0 	l\.jal [0-9a-f]+ <\.plt\+0x18>
+ +[0-9a-f]+:	84 71 00 00 	l\.lwz r3,0\(r17\)
+ +[0-9a-f]+:	85 21 00 04 	l\.lwz r9,4\(r1\)
+ +[0-9a-f]+:	86 01 00 00 	l\.lwz r16,0\(r1\)
+ +[0-9a-f]+:	44 00 48 00 	l\.jr r9
+ +[0-9a-f]+:	9c 21 00 08 	l\.addi r1,r1,8
diff --git a/ld/testsuite/ld-or1k/pltlarge2.s b/ld/testsuite/ld-or1k/pltlarge2.s
new file mode 100644
index 00000000000..665578c67c1
--- /dev/null
+++ b/ld/testsuite/ld-or1k/pltlarge2.s
@@ -0,0 +1,25 @@
+	.section	.text
+	.align 4
+	.global	test
+	.type	test, @function
+test:
+	l.addi	r1, r1, -8
+	l.sw	0(r1), r16
+	l.sw	4(r1), r9
+
+	l.jal	8
+	 l.movhi	r16, gotpchi(_GLOBAL_OFFSET_TABLE_-4)
+	l.ori	r16, r16, gotpclo(_GLOBAL_OFFSET_TABLE_+0)
+	l.add	r16, r16, r9
+
+	l.movhi	r17, gotha(i)
+	l.add	r17, r17, r16
+	l.lwz	r17, got(i)(r17)
+
+	l.jal	plt(func)
+	 l.lwz	r3, 0(r17)
+
+	l.lwz	r9, 4(r1)
+	l.lwz	r16, 0(r1)
+	l.jr	r9
+	 l.addi	r1, r1, 8
diff --git a/ld/testsuite/ld-or1k/pltlib.s b/ld/testsuite/ld-or1k/pltlib.s
index baf76ca1af7..8b4d7ba48fd 100644
--- a/ld/testsuite/ld-or1k/pltlib.s
+++ b/ld/testsuite/ld-or1k/pltlib.s
@@ -1,5 +1,6 @@
 	.section .data
 	.globl x, y
+	.size x, 4
 x:	.long 33
 y:	.long 44
 
-- 
2.26.2

[OpenRISC] [PATCH 4/4] or1k: Fail link if the PLT relocation overflows

[Stafford Horne]

bfd/ChangeLog:

	PR 27746
	* elf32-or1k.c (or1k_elf_finish_dynamic_symbol): Add condition
	to check and fail if the PLT relocation overflows.
---
 bfd/elf32-or1k.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/bfd/elf32-or1k.c b/bfd/elf32-or1k.c
index 1a5b3319646..6afacfa57cf 100644
--- a/bfd/elf32-or1k.c
+++ b/bfd/elf32-or1k.c
@@ -2453,6 +2453,14 @@ or1k_elf_finish_dynamic_symbol (bfd *output_bfd,
       got_offset = (plt_index + 3) * 4;
       got_addr = got_base_addr + got_offset;
 
+      if (!htab->saw_gotha && plt_reloc > 0xffff)
+	{
+	  _bfd_error_handler (_("%pB: plt relocation overflow when writing %s"),
+		              output_bfd, h->root.root.string);
+	  bfd_set_error (bfd_error_bad_value);
+	  return false;
+	}
+
       /* Fill in the entry in the procedure linkage table.  */
       if (htab->saw_plta)
 	{
-- 
2.26.2

[OpenRISC] [PATCH 1/4] or1k: Fix issue with plt link failure for local calls

[Giulio Benetti]

Hi Stafford, All,

I've tested this patch succesfully rebuilding protobuf with Buildroot.

Best regards
-- 
Giulio Benetti
Benetti Engineering sas

On 4/19/21 2:05 AM, Stafford Horne wrote:
> When building protobuf we were seeing the assert failure:
> 
>      /home/giuliobenetti/git/upstream/or1k-binutils-2.36.1/host/lib/gcc/or1k-buildroot-linux-uclibc/9.3.0/../../../../or1k-buildroot-linux-uclibc/bin/ld:
>      BFD (GNU Binutils) 2.36.1 assertion fail elf32-or1k.c:2377
>      /home/giuliobenetti/git/upstream/or1k-binutils-2.36.1/host/lib/gcc/or1k-buildroot-linux-uclibc/9.3.0/../../../../or1k-buildroot-linux-uclibc/bin/ld:
>      BFD (GNU Binutils) 2.36.1 assertion fail elf32-or1k.c:2377
>      /home/giuliobenetti/git/upstream/or1k-binutils-2.36.1/host/lib/gcc/or1k-buildroot-linux-uclibc/9.3.0/../../../../or1k-buildroot-linux-uclibc/bin/ld:
>      BFD (GNU Binutils) 2.36.1 assertion fail elf32-or1k.c:2377
>      collect2: error: ld returned 1 exit status
> 
> This failure happens while writing out PLT entries, there is a check
> "BFD_ASSERT (h->dynindx != -1)" to confirm all plt entries have dynamic
> symbol attributes.  This was failing for symbols that were
> "forced_local" in previous linking code.
> 
> The fix adds logic to or1k_elf_adjust_dynamic_symbol to identify
> "forced_local" symbols and exclude them from the the PLT.
> 
> bfd/ChangeLog:
> 
> 	PR 27624
> 	* elf32-or1k.c (or1k_elf_adjust_dynamic_symbol): Change
> 	condition used to cleanup plt entries to cleanup forced local
> 	entries.
> 
> Cc: Giulio Benetti <giulio.benetti@benettiengineering.com>
> ---
>   bfd/elf32-or1k.c | 9 ++++-----
>   1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/bfd/elf32-or1k.c b/bfd/elf32-or1k.c
> index 013deb406f8..cc6c51220e5 100644
> --- a/bfd/elf32-or1k.c
> +++ b/bfd/elf32-or1k.c
> @@ -2566,11 +2566,10 @@ or1k_elf_adjust_dynamic_symbol (struct bfd_link_info *info,
>     if (h->type == STT_FUNC
>         || h->needs_plt)
>       {
> -      if (! bfd_link_pic (info)
> -	  && !h->def_dynamic
> -	  && !h->ref_dynamic
> -	  && h->root.type != bfd_link_hash_undefweak
> -	  && h->root.type != bfd_link_hash_undefined)
> +      if (h->plt.refcount <= 0
> +	  || (SYMBOL_CALLS_LOCAL (info, h)
> +	  || (ELF_ST_VISIBILITY (h->other) != STV_DEFAULT
> +	      && h->root.type == bfd_link_hash_undefweak)))
>   	{
>   	  /* This case can occur if we saw a PLT reloc in an input
>   	     file, but the symbol was never referred to by a dynamic
> 

[OpenRISC] [PATCH 0/4] OpenRISC large binary fixes

[Stafford Horne]

On Mon, Apr 19, 2021 at 09:05:20AM +0900, Stafford Horne wrote:
> These are several fixes to OpenRISC BFD to handle issues mostly related to
> linking large files.
> 
> There will be a GCC patch then depends on this, a new -mcmodel=large argument is
> added to GCC to allow creating GOT relocation values larger than 64k.
> 
> The series fixes 3 issues:
> 
>  * PR 27624 - Assertion failure when dynamic symol is localized
>  * PR 21464 - GOT relocation truncation.
>  * PR 27746 - Issue with silent overlows in PLT code.

As discussed offline with Giulio, this series does not properly fix the PR 21464
issue.  I have fixed the issue and will post a v2 after I fix some other things
uncovered...

Instead of using large PLTs whenever we see gotha() relocations, I will try to
detect when the plt_relocation overflows and switch to use the large model then.

This is needed as the gcc crtstuff will now always have gotha() relocations, with
the current binutils this would cause the large PLT code to always be generated.
We don't want that.

-Stafford