Elf_Internal_Sym *isymbuf = NULL;
struct elf32_mn10300_link_hash_table *hash_table;
asection *section = sec;
+ bfd_vma align_gap_adjustment;
/* Assume nothing changes. */
*again = FALSE;
if (internal_relocs == NULL)
goto error_return;
+ /* Scan for worst case alignment gap changes. Note that this logic
+ is not ideal; what we should do is run this scan for every
+ opcode/address range and adjust accordingly, but that's
+ expensive. Worst case is that for an alignment of N bytes, we
+ move by 2*N-N-1 bytes, assuming we have aligns of 1, 2, 4, 8, etc
+ all before it. Plus, this still doesn't cover cross-section
+ jumps with section alignment. */
+ irelend = internal_relocs + sec->reloc_count;
+ align_gap_adjustment = 0;
+ for (irel = internal_relocs; irel < irelend; irel++)
+ {
+ if (ELF32_R_TYPE (irel->r_info) == (int) R_MN10300_ALIGN)
+ {
+ bfd_vma adj = 1 << irel->r_addend;
+ bfd_vma aend = irel->r_offset;
+
+ aend = BFD_ALIGN (aend, 1 << irel->r_addend);
+ adj = 2*adj - adj - 1;
+
+ /* Record the biggest adjustmnet. Skip any alignment at the
+ end of our section. */
+ if (align_gap_adjustment < adj
+ && aend < sec->output_section->vma + sec->output_offset + sec->size)
+ align_gap_adjustment = adj;
+ }
+ }
+
/* Walk through them looking for relaxing opportunities. */
irelend = internal_relocs + sec->reloc_count;
for (irel = internal_relocs; irel < irelend; irel++)
/* See if the value will fit in 16 bits, note the high value is
0x7fff + 2 as the target will be two bytes closer if we are
able to relax. */
- if ((long) value < 0x8001 && (long) value > -0x8000)
+ /* Account for jumps across alignment boundaries using
+ align_gap_adjustment. */
+ if (value < 0x8001 - align_gap_adjustment
+ && ((bfd_signed_vma) value > -0x8000 + (bfd_signed_vma) align_gap_adjustment))
{
unsigned char code;
+2007-11-21 Nick Clifton <nickc@redhat.com>
+
+ * ld-mn10300/i135409-4.s: New test case. Check for relaxation to
+ a 16-bit jump instruction.
+ * ld-mn10300/i135409-4.t: Linker script for the new test.
+ * ld-mn10300/i135409-4.d: Expected disassembly of new test.
+ * ld-mn10300/mn10300.exp: Run the new test.
+
2007-11-20 Nick Clifton <nickc@redhat.com>
* lib/ld-lib.exp (check_gc_sections_available): New proc, based