make_toc_relative(Target_powerpc<size, big_endian>* target,
Address* value);
+ bool
+ make_got_relative(Target_powerpc<size, big_endian>* target,
+ const Symbol_value<size>* psymval,
+ Address addend,
+ Address* value);
+
// Perform the Sized_relobj_file method, then set up opd info from
// .opd relocs.
void
return true;
}
+template<int size, bool big_endian>
+bool
+Powerpc_relobj<size, big_endian>::make_got_relative(
+ Target_powerpc<size, big_endian>* target,
+ const Symbol_value<size>* psymval,
+ Address addend,
+ Address* value)
+{
+ Address addr = psymval->value(this, addend);
+ Address got_base = (target->got_section()->output_section()->address()
+ + this->toc_base_offset());
+ addr -= got_base;
+ if (addr + 0x80008000 > 0xffffffff)
+ return false;
+
+ *value = addr;
+ return true;
+}
+
// Perform the Sized_relobj_file method, then set up opd info from
// .opd relocs.
return true;
}
+template<int size>
+static bool
+relative_value_is_known(const Sized_symbol<size>* gsym)
+{
+ if (gsym->type() == elfcpp::STT_GNU_IFUNC)
+ return false;
+
+ if (gsym->is_from_dynobj()
+ || gsym->is_undefined()
+ || gsym->is_preemptible())
+ return false;
+
+ if (gsym->is_absolute())
+ return !parameters->options().output_is_position_independent();
+
+ return true;
+}
+
+template<int size>
+static bool
+relative_value_is_known(const Symbol_value<size>* psymval)
+{
+ if (psymval->is_ifunc_symbol())
+ return false;
+
+ bool is_ordinary;
+ unsigned int shndx = psymval->input_shndx(&is_ordinary);
+
+ return is_ordinary && shndx != elfcpp::SHN_UNDEF;
+}
+
+// PCREL_OPT in one instance flags to the linker that a pair of insns:
+// pld ra,symbol@got@pcrel
+// load/store rt,0(ra)
+// or
+// pla ra,symbol@pcrel
+// load/store rt,0(ra)
+// may be translated to
+// pload/pstore rt,symbol@pcrel
+// nop.
+// This function returns true if the optimization is possible, placing
+// the prefix insn in *PINSN1 and a NOP in *PINSN2.
+//
+// On entry to this function, the linker has already determined that
+// the pld can be replaced with pla: *PINSN1 is that pla insn,
+// while *PINSN2 is the second instruction.
+
+inline bool
+xlate_pcrel_opt(uint64_t *pinsn1, uint64_t *pinsn2)
+{
+ uint32_t insn2 = *pinsn2 >> 32;
+ uint64_t i1new;
+
+ // Check that regs match.
+ if (((insn2 >> 16) & 31) != ((*pinsn1 >> 21) & 31))
+ return false;
+
+ switch ((insn2 >> 26) & 63)
+ {
+ default:
+ return false;
+
+ case 32: // lwz
+ case 34: // lbz
+ case 36: // stw
+ case 38: // stb
+ case 40: // lhz
+ case 42: // lha
+ case 44: // sth
+ case 48: // lfs
+ case 50: // lfd
+ case 52: // stfs
+ case 54: // stfd
+ // These are the PMLS cases, where we just need to tack a prefix
+ // on the insn. Check that the D field is zero.
+ if ((insn2 & 0xffff) != 0)
+ return false;
+ i1new = ((1ULL << 58) | (2ULL << 56) | (1ULL << 52)
+ | (insn2 & ((63ULL << 26) | (31ULL << 21))));
+ break;
+
+ case 58: // lwa, ld
+ if ((insn2 & 0xfffd) != 0)
+ return false;
+ i1new = ((1ULL << 58) | (1ULL << 52)
+ | (insn2 & 2 ? 41ULL << 26 : 57ULL << 26)
+ | (insn2 & (31ULL << 21)));
+ break;
+
+ case 57: // lxsd, lxssp
+ if ((insn2 & 0xfffc) != 0 || (insn2 & 3) < 2)
+ return false;
+ i1new = ((1ULL << 58) | (1ULL << 52)
+ | ((40ULL | (insn2 & 3)) << 26)
+ | (insn2 & (31ULL << 21)));
+ break;
+
+ case 61: // stxsd, stxssp, lxv, stxv
+ if ((insn2 & 3) == 0)
+ return false;
+ else if ((insn2 & 3) >= 2)
+ {
+ if ((insn2 & 0xfffc) != 0)
+ return false;
+ i1new = ((1ULL << 58) | (1ULL << 52)
+ | ((44ULL | (insn2 & 3)) << 26)
+ | (insn2 & (31ULL << 21)));
+ }
+ else
+ {
+ if ((insn2 & 0xfff0) != 0)
+ return false;
+ i1new = ((1ULL << 58) | (1ULL << 52)
+ | ((50ULL | (insn2 & 4) | ((insn2 & 8) >> 3)) << 26)
+ | (insn2 & (31ULL << 21)));
+ }
+ break;
+
+ case 56: // lq
+ if ((insn2 & 0xffff) != 0)
+ return false;
+ i1new = ((1ULL << 58) | (1ULL << 52)
+ | (insn2 & ((63ULL << 26) | (31ULL << 21))));
+ break;
+
+ case 62: // std, stq
+ if ((insn2 & 0xfffd) != 0)
+ return false;
+ i1new = ((1ULL << 58) | (1ULL << 52)
+ | ((insn2 & 2) == 0 ? 61ULL << 26 : 60ULL << 26)
+ | (insn2 & (31ULL << 21)));
+ break;
+ }
+
+ *pinsn1 = i1new;
+ *pinsn2 = (uint64_t) nop << 32;
+ return true;
+}
+
// Perform a relocation.
template<int size, bool big_endian>
break;
}
- if (size == 64)
+ if (size == 64
+ && (gsym
+ ? relative_value_is_known(gsym)
+ : relative_value_is_known(psymval)))
{
+ Insn* iview;
+ Insn* iview2;
+ Insn insn;
+ uint64_t pinsn, pinsn2;
+
switch (r_type)
{
default:
// and
// addis ra,r2,0; addi rb,ra,x@toc@l;
// to nop; addi rb,r2,x@toc;
- // FIXME: the @got sequence shown above is not yet
- // optimized. Note that gcc as of 2017-01-07 doesn't use
- // the ELF @got relocs except for TLS, instead using the
- // PowerOpen variant of a compiler managed GOT (called TOC).
- // The PowerOpen TOC sequence equivalent to the first
- // example is optimized.
case elfcpp::R_POWERPC_GOT_TLSLD16_HA:
case elfcpp::R_POWERPC_GOT_TLSGD16_HA:
case elfcpp::R_POWERPC_GOT_TPREL16_HA:
case elfcpp::R_PPC64_TOC16_HA:
if (parameters->options().toc_optimize())
{
- Insn* iview = reinterpret_cast<Insn*>(view - d_offset);
- Insn insn = elfcpp::Swap<32, big_endian>::readval(iview);
- if (r_type == elfcpp::R_PPC64_TOC16_HA
- && object->make_toc_relative(target, &value))
+ iview = reinterpret_cast<Insn*>(view - d_offset);
+ insn = elfcpp::Swap<32, big_endian>::readval(iview);
+ if ((r_type == elfcpp::R_PPC64_TOC16_HA
+ && object->make_toc_relative(target, &value))
+ || (r_type == elfcpp::R_POWERPC_GOT16_HA
+ && object->make_got_relative(target, psymval,
+ rela.get_r_addend(),
+ &value)))
{
gold_assert((insn & ((0x3f << 26) | 0x1f << 16))
== ((15u << 26) | (2 << 16)));
case elfcpp::R_PPC64_TOC16_LO_DS:
if (parameters->options().toc_optimize())
{
- Insn* iview = reinterpret_cast<Insn*>(view - d_offset);
- Insn insn = elfcpp::Swap<32, big_endian>::readval(iview);
+ iview = reinterpret_cast<Insn*>(view - d_offset);
+ insn = elfcpp::Swap<32, big_endian>::readval(iview);
bool changed = false;
- if (r_type == elfcpp::R_PPC64_TOC16_LO_DS
- && object->make_toc_relative(target, &value))
+ if ((r_type == elfcpp::R_PPC64_TOC16_LO_DS
+ && object->make_toc_relative(target, &value))
+ || (r_type == elfcpp::R_PPC64_GOT16_LO_DS
+ && object->make_got_relative(target, psymval,
+ rela.get_r_addend(),
+ &value)))
{
gold_assert ((insn & (0x3f << 26)) == 58u << 26 /* ld */);
insn ^= (14u << 26) ^ (58u << 26);
}
break;
+ case elfcpp::R_PPC64_GOT_PCREL34:
+ if (parameters->options().toc_optimize())
+ {
+ iview = reinterpret_cast<Insn*>(view);
+ pinsn = elfcpp::Swap<32, big_endian>::readval(iview);
+ pinsn <<= 32;
+ pinsn |= elfcpp::Swap<32, big_endian>::readval(iview + 1);
+ if ((pinsn & ((-1ULL << 50) | (63ULL << 26)))
+ != ((1ULL << 58) | (1ULL << 52) | (57ULL << 26) /* pld */))
+ break;
+
+ Address relval = psymval->value(object, rela.get_r_addend());
+ relval -= address;
+ if (relval + (1ULL << 33) < 1ULL << 34)
+ {
+ value = relval;
+ // Replace with paddi
+ pinsn += (2ULL << 56) + (14ULL << 26) - (57ULL << 26);
+ elfcpp::Swap<32, big_endian>::writeval(iview, pinsn >> 32);
+ elfcpp::Swap<32, big_endian>::writeval(iview + 1,
+ pinsn & 0xffffffff);
+ goto pcrelopt;
+ }
+ }
+ break;
+
+ case elfcpp::R_PPC64_PCREL34:
+ {
+ iview = reinterpret_cast<Insn*>(view);
+ pinsn = elfcpp::Swap<32, big_endian>::readval(iview);
+ pinsn <<= 32;
+ pinsn |= elfcpp::Swap<32, big_endian>::readval(iview + 1);
+ if ((pinsn & ((-1ULL << 50) | (63ULL << 26)))
+ != ((1ULL << 58) | (2ULL << 56) | (1ULL << 52)
+ | (14ULL << 26) /* paddi */))
+ break;
+
+ pcrelopt:
+ const int reloc_size = elfcpp::Elf_sizes<size>::rela_size;
+ elfcpp::Shdr<size, big_endian> shdr(relinfo->reloc_shdr);
+ size_t reloc_count = shdr.get_sh_size() / reloc_size;
+ if (relnum >= reloc_count - 1)
+ break;
+
+ Reltype next_rela(preloc + reloc_size);
+ if ((elfcpp::elf_r_type<size>(next_rela.get_r_info())
+ != elfcpp::R_PPC64_PCREL_OPT)
+ || next_rela.get_r_offset() != rela.get_r_offset())
+ break;
+
+ Address off = next_rela.get_r_addend();
+ if (off == 0)
+ off = 8; // zero means next insn.
+ if (off + rela.get_r_offset() + 4 > view_size)
+ break;
+
+ iview2 = reinterpret_cast<Insn*>(view + off);
+ pinsn2 = elfcpp::Swap<32, big_endian>::readval(iview2);
+ pinsn2 <<= 32;
+ if ((pinsn2 & (63ULL << 58)) == 1ULL << 58)
+ break;
+ if (xlate_pcrel_opt(&pinsn, &pinsn2))
+ {
+ elfcpp::Swap<32, big_endian>::writeval(iview, pinsn >> 32);
+ elfcpp::Swap<32, big_endian>::writeval(iview + 1,
+ pinsn & 0xffffffff);
+ elfcpp::Swap<32, big_endian>::writeval(iview2, pinsn2 >> 32);
+ }
+ }
+ break;
+
case elfcpp::R_POWERPC_TPREL16_HA:
if (parameters->options().tls_optimize() && value + 0x8000 < 0x10000)
{
projects / deliverable / binutils-gdb.git / blobdiff