X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=gold%2Fx86_64.cc;h=6c511e2b6079ab43a5bf904efbb77eba2cada362;hb=3a4f096e5ff41e2a3b832cf6c815ad1f28437c63;hp=a15b6ae331144f84b85be820d85ff4deafcda149;hpb=7404fe1b8d20b30162a7e56307d5a2d400cb5645;p=binutils-gdb.git diff --git a/gold/x86_64.cc b/gold/x86_64.cc index a15b6ae3311..6c511e2b607 100644 --- a/gold/x86_64.cc +++ b/gold/x86_64.cc @@ -1,7 +1,6 @@ // x86_64.cc -- x86_64 target support for gold. -// Copyright 2006, 2007, 2008, 2009, 2010, 2011, 2012 -// Free Software Foundation, Inc. +// Copyright (C) 2006-2016 Free Software Foundation, Inc. // Written by Ian Lance Taylor . // This file is part of gold. @@ -49,6 +48,37 @@ namespace using namespace gold; +// A class to handle the .got.plt section. + +class Output_data_got_plt_x86_64 : public Output_section_data_build +{ + public: + Output_data_got_plt_x86_64(Layout* layout) + : Output_section_data_build(8), + layout_(layout) + { } + + Output_data_got_plt_x86_64(Layout* layout, off_t data_size) + : Output_section_data_build(data_size, 8), + layout_(layout) + { } + + protected: + // Write out the PLT data. + void + do_write(Output_file*); + + // Write to a map file. + void + do_print_to_mapfile(Mapfile* mapfile) const + { mapfile->print_output_data(this, "** GOT PLT"); } + + private: + // A pointer to the Layout class, so that we can find the .dynamic + // section when we write out the GOT PLT section. + Layout* layout_; +}; + // A class to handle the PLT data. // This is an abstract base class that handles most of the linker details // but does not know the actual contents of PLT entries. The derived @@ -62,9 +92,9 @@ class Output_data_plt_x86_64 : public Output_section_data Output_data_plt_x86_64(Layout* layout, uint64_t addralign, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative) - : Output_section_data(addralign), layout_(layout), tlsdesc_rel_(NULL), + : Output_section_data(addralign), tlsdesc_rel_(NULL), irelative_rel_(NULL), got_(got), got_plt_(got_plt), got_irelative_(got_irelative), count_(0), irelative_count_(0), tlsdesc_got_offset_(-1U), free_list_() @@ -72,12 +102,12 @@ class Output_data_plt_x86_64 : public Output_section_data Output_data_plt_x86_64(Layout* layout, uint64_t plt_entry_size, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative, unsigned int plt_count) : Output_section_data((plt_count + 1) * plt_entry_size, plt_entry_size, false), - layout_(layout), tlsdesc_rel_(NULL), irelative_rel_(NULL), got_(got), + tlsdesc_rel_(NULL), irelative_rel_(NULL), got_(got), got_plt_(got_plt), got_irelative_(got_irelative), count_(plt_count), irelative_count_(0), tlsdesc_got_offset_(-1U), free_list_() { @@ -269,9 +299,6 @@ class Output_data_plt_x86_64 : public Output_section_data void do_write(Output_file*); - // A pointer to the Layout class, so that we can find the .dynamic - // section when we write out the GOT PLT section. - Layout* layout_; // The reloc section. Reloc_section* rel_; // The TLSDESC relocs, if necessary. These must follow the regular @@ -283,7 +310,7 @@ class Output_data_plt_x86_64 : public Output_section_data // The .got section. Output_data_got<64, false>* got_; // The .got.plt section. - Output_data_space* got_plt_; + Output_data_got_plt_x86_64* got_plt_; // The part of the .got.plt section used for IRELATIVE relocs. Output_data_space* got_irelative_; // The number of PLT entries. @@ -304,7 +331,7 @@ class Output_data_plt_x86_64_standard : public Output_data_plt_x86_64 public: Output_data_plt_x86_64_standard(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative) : Output_data_plt_x86_64(layout, plt_entry_size, got, got_plt, got_irelative) @@ -312,7 +339,7 @@ class Output_data_plt_x86_64_standard : public Output_data_plt_x86_64 Output_data_plt_x86_64_standard(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative, unsigned int plt_count) : Output_data_plt_x86_64(layout, plt_entry_size, @@ -376,6 +403,33 @@ class Output_data_plt_x86_64_standard : public Output_data_plt_x86_64 static const unsigned char plt_eh_frame_fde[plt_eh_frame_fde_size]; }; +template +class Lazy_view +{ + public: + Lazy_view(Sized_relobj_file* object, unsigned int data_shndx) + : object_(object), data_shndx_(data_shndx), view_(NULL), view_size_(0) + { } + + inline unsigned char + operator[](size_t offset) + { + if (this->view_ == NULL) + this->view_ = this->object_->section_contents(this->data_shndx_, + &this->view_size_, + true); + if (offset >= this->view_size_) + return 0; + return this->view_[offset]; + } + + private: + Sized_relobj_file* object_; + unsigned int data_shndx_; + const unsigned char* view_; + section_size_type view_size_; +}; + // The x86_64 target class. // See the ABI at // http://www.x86-64.org/documentation/abi.pdf @@ -396,7 +450,7 @@ class Target_x86_64 : public Sized_target got_(NULL), plt_(NULL), got_plt_(NULL), got_irelative_(NULL), got_tlsdesc_(NULL), global_offset_table_(NULL), rela_dyn_(NULL), rela_irelative_(NULL), copy_relocs_(elfcpp::R_X86_64_COPY), - dynbss_(NULL), got_mod_index_offset_(-1U), tlsdesc_reloc_info_(), + got_mod_index_offset_(-1U), tlsdesc_reloc_info_(), tls_base_symbol_defined_(false) { } @@ -469,6 +523,21 @@ class Target_x86_64 : public Sized_target const unsigned char* plocal_symbols, Relocatable_relocs*); + // Scan the relocs for --emit-relocs. + void + emit_relocs_scan(Symbol_table* symtab, + Layout* layout, + Sized_relobj_file* object, + unsigned int data_shndx, + unsigned int sh_type, + const unsigned char* prelocs, + size_t reloc_count, + Output_section* output_section, + bool needs_special_offset_handling, + size_t local_symbol_count, + const unsigned char* plocal_syms, + Relocatable_relocs* rr); + // Emit relocations for a section. void relocate_relocs( @@ -477,8 +546,7 @@ class Target_x86_64 : public Sized_target const unsigned char* prelocs, size_t reloc_count, Output_section* output_section, - off_t offset_in_output_section, - const Relocatable_relocs*, + typename elfcpp::Elf_types::Elf_Off offset_in_output_section, unsigned char* view, typename elfcpp::Elf_types::Elf_Addr view_address, section_size_type view_size, @@ -535,6 +603,7 @@ class Target_x86_64 : public Sized_target void do_calls_non_split(Relobj* object, unsigned int shndx, section_offset_type fnoffset, section_size_type fnsize, + const unsigned char* prelocs, size_t reloc_count, unsigned char* view, section_size_type view_size, std::string* from, std::string* to) const; @@ -567,6 +636,11 @@ class Target_x86_64 : public Sized_target unsigned int plt_entry_size() const; + // Return the size of each GOT entry. + unsigned int + got_entry_size() const + { return 8; }; + // Create the GOT section for an incremental update. Output_data_got_base* init_got_plt_for_update(Symbol_table* symtab, @@ -619,7 +693,7 @@ class Target_x86_64 : public Sized_target Output_data_plt_x86_64* make_data_plt(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative) { return this->do_make_data_plt(layout, got, got_plt, got_irelative); @@ -628,7 +702,7 @@ class Target_x86_64 : public Sized_target Output_data_plt_x86_64* make_data_plt(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative, unsigned int plt_count) { @@ -639,7 +713,7 @@ class Target_x86_64 : public Sized_target virtual Output_data_plt_x86_64* do_make_data_plt(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative) { return new Output_data_plt_x86_64_standard(layout, got, got_plt, @@ -649,7 +723,7 @@ class Target_x86_64 : public Sized_target virtual Output_data_plt_x86_64* do_make_data_plt(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative, unsigned int plt_count) { @@ -676,7 +750,8 @@ class Target_x86_64 : public Sized_target unsigned int data_shndx, Output_section* output_section, const elfcpp::Rela& reloc, unsigned int r_type, - const elfcpp::Sym& lsym); + const elfcpp::Sym& lsym, + bool is_discarded); inline void global(Symbol_table* symtab, Layout* layout, Target_x86_64* target, @@ -749,11 +824,9 @@ class Target_x86_64 : public Sized_target // Do a relocation. Return false if the caller should not issue // any warnings about this relocation. inline bool - relocate(const Relocate_info*, Target_x86_64*, - Output_section*, - size_t relnum, const elfcpp::Rela&, - unsigned int r_type, const Sized_symbol*, - const Symbol_value*, + relocate(const Relocate_info*, unsigned int, + Target_x86_64*, Output_section*, size_t, const unsigned char*, + const Sized_symbol*, const Symbol_value*, unsigned char*, typename elfcpp::Elf_types::Elf_Addr, section_size_type); @@ -770,7 +843,6 @@ class Target_x86_64 : public Sized_target // Do a TLS General-Dynamic to Initial-Exec transition. inline void tls_gd_to_ie(const Relocate_info*, size_t relnum, - Output_segment* tls_segment, const elfcpp::Rela&, unsigned int r_type, typename elfcpp::Elf_types::Elf_Addr value, unsigned char* view, @@ -789,7 +861,6 @@ class Target_x86_64 : public Sized_target // Do a TLSDESC-style General-Dynamic to Initial-Exec transition. inline void tls_desc_gd_to_ie(const Relocate_info*, size_t relnum, - Output_segment* tls_segment, const elfcpp::Rela&, unsigned int r_type, typename elfcpp::Elf_types::Elf_Addr value, unsigned char* view, @@ -828,14 +899,67 @@ class Target_x86_64 : public Sized_target bool skip_call_tls_get_addr_; }; - // A class which returns the size required for a relocation type, - // used while scanning relocs during a relocatable link. - class Relocatable_size_for_reloc + // Check if relocation against this symbol is a candidate for + // conversion from + // mov foo@GOTPCREL(%rip), %reg + // to lea foo(%rip), %reg. + template + static inline bool + can_convert_mov_to_lea(const Symbol* gsym, unsigned int r_type, + size_t r_offset, View_type* view) { - public: - unsigned int - get_size_for_reloc(unsigned int, Relobj*); - }; + gold_assert(gsym != NULL); + // We cannot do the conversion unless it's one of these relocations. + if (r_type != elfcpp::R_X86_64_GOTPCREL + && r_type != elfcpp::R_X86_64_GOTPCRELX + && r_type != elfcpp::R_X86_64_REX_GOTPCRELX) + return false; + // We cannot convert references to IFUNC symbols, or to symbols that + // are not local to the current module. + if (gsym->type() == elfcpp::STT_GNU_IFUNC + || gsym->is_undefined () + || gsym->is_from_dynobj() + || gsym->is_preemptible()) + return false; + // If we are building a shared object and the symbol is protected, we may + // need to go through the GOT. + if (parameters->options().shared() + && gsym->visibility() == elfcpp::STV_PROTECTED) + return false; + // We cannot convert references to the _DYNAMIC symbol. + if (strcmp(gsym->name(), "_DYNAMIC") == 0) + return false; + // Check for a MOV opcode. + return (*view)[r_offset - 2] == 0x8b; + } + + // Convert + // callq *foo@GOTPCRELX(%rip) to + // addr32 callq foo + // and jmpq *foo@GOTPCRELX(%rip) to + // jmpq foo + // nop + template + static inline bool + can_convert_callq_to_direct(const Symbol* gsym, unsigned int r_type, + size_t r_offset, View_type* view) + { + gold_assert(gsym != NULL); + // We cannot do the conversion unless it's a GOTPCRELX relocation. + if (r_type != elfcpp::R_X86_64_GOTPCRELX) + return false; + // We cannot convert references to IFUNC symbols, or to symbols that + // are not local to the current module. + if (gsym->type() == elfcpp::STT_GNU_IFUNC + || gsym->is_undefined () + || gsym->is_from_dynobj() + || gsym->is_preemptible()) + return false; + // Check for a CALLQ or JMPQ opcode. + return ((*view)[r_offset - 2] == 0xff + && ((*view)[r_offset - 1] == 0x15 + || (*view)[r_offset - 1] == 0x25)); + } // Adjust TLS relocation type based on the options and whether this // is a local symbol. @@ -847,7 +971,7 @@ class Target_x86_64 : public Sized_target got_section(Symbol_table*, Layout*); // Get the GOT PLT section. - Output_data_space* + Output_data_got_plt_x86_64* got_plt_section() const { gold_assert(this->got_plt_ != NULL); @@ -916,10 +1040,13 @@ class Target_x86_64 : public Sized_target unsigned int shndx, Output_section* output_section, Symbol* sym, const elfcpp::Rela& reloc) { + unsigned int r_type = elfcpp::elf_r_type(reloc.get_r_info()); this->copy_relocs_.copy_reloc(symtab, layout, symtab->get_sized_symbol(sym), object, shndx, output_section, - reloc, this->rela_dyn_section(layout)); + r_type, reloc.get_r_offset(), + reloc.get_r_addend(), + this->rela_dyn_section(layout)); } // Information about this specific target which we pass to the @@ -958,7 +1085,7 @@ class Target_x86_64 : public Sized_target // The PLT section. Output_data_plt_x86_64* plt_; // The GOT PLT section. - Output_data_space* got_plt_; + Output_data_got_plt_x86_64* got_plt_; // The GOT section for IRELATIVE relocations. Output_data_space* got_irelative_; // The GOT section for TLSDESC relocations. @@ -971,8 +1098,6 @@ class Target_x86_64 : public Sized_target Reloc_section* rela_irelative_; // Relocs saved to avoid a COPY reloc. Copy_relocs copy_relocs_; - // Space for variables copied with a COPY reloc. - Output_data_space* dynbss_; // Offset of the GOT entry for the TLS module index. unsigned int got_mod_index_offset_; // We handle R_X86_64_TLSDESC against a local symbol as a target @@ -1006,7 +1131,9 @@ const Target::Target_info Target_x86_64<64>::x86_64_info = 0, // small_common_section_flags elfcpp::SHF_X86_64_LARGE, // large_common_section_flags NULL, // attributes_section - NULL // attributes_vendor + NULL, // attributes_vendor + "_start", // entry_symbol_name + 32, // hash_entry_size }; template<> @@ -1032,7 +1159,9 @@ const Target::Target_info Target_x86_64<32>::x86_64_info = 0, // small_common_section_flags elfcpp::SHF_X86_64_LARGE, // large_common_section_flags NULL, // attributes_section - NULL // attributes_vendor + NULL, // attributes_vendor + "_start", // entry_symbol_name + 32, // hash_entry_size }; // This is called when a new output section is created. This is where @@ -1074,7 +1203,7 @@ Target_x86_64::got_section(Symbol_table* symtab, Layout* layout) | elfcpp::SHF_WRITE), this->got_, got_order, true); - this->got_plt_ = new Output_data_space(8, "** GOT PLT"); + this->got_plt_ = new Output_data_got_plt_x86_64(layout); layout->add_output_section_data(".got.plt", elfcpp::SHT_PROGBITS, (elfcpp::SHF_ALLOC | elfcpp::SHF_WRITE), @@ -1163,6 +1292,27 @@ Target_x86_64::rela_irelative_section(Layout* layout) return this->rela_irelative_; } +// Write the first three reserved words of the .got.plt section. +// The remainder of the section is written while writing the PLT +// in Output_data_plt_i386::do_write. + +void +Output_data_got_plt_x86_64::do_write(Output_file* of) +{ + // The first entry in the GOT is the address of the .dynamic section + // aka the PT_DYNAMIC segment. The next two entries are reserved. + // We saved space for them when we created the section in + // Target_x86_64::got_section. + const off_t got_file_offset = this->offset(); + gold_assert(this->data_size() >= 24); + unsigned char* const got_view = of->get_output_view(got_file_offset, 24); + Output_section* dynamic = this->layout_->dynamic_section(); + uint64_t dynamic_addr = dynamic == NULL ? 0 : dynamic->address(); + elfcpp::Swap<64, false>::writeval(got_view, dynamic_addr); + memset(got_view + 8, 0, 16); + of->write_output_view(got_file_offset, 24, got_view); +} + // Initialize the PLT section. template @@ -1198,7 +1348,7 @@ Output_data_plt_x86_64::add_entry(Symbol_table* symtab, Layout* layout, unsigned int* pcount; unsigned int offset; unsigned int reserved; - Output_data_space* got; + Output_section_data_build* got; if (gsym->type() == elfcpp::STT_GNU_IFUNC && gsym->can_use_relative_reloc(false)) { @@ -1384,7 +1534,7 @@ Output_data_plt_x86_64::address_for_global(const Symbol* gsym) if (gsym->type() == elfcpp::STT_GNU_IFUNC && gsym->can_use_relative_reloc(false)) offset = (this->count_ + 1) * this->get_plt_entry_size(); - return this->address() + offset; + return this->address() + offset + gsym->plt_offset(); } // Return the PLT address to use for a local symbol. These are always @@ -1392,9 +1542,12 @@ Output_data_plt_x86_64::address_for_global(const Symbol* gsym) template uint64_t -Output_data_plt_x86_64::address_for_local(const Relobj*, unsigned int) +Output_data_plt_x86_64::address_for_local(const Relobj* object, + unsigned int r_sym) { - return this->address() + (this->count_ + 1) * this->get_plt_entry_size(); + return (this->address() + + (this->count_ + 1) * this->get_plt_entry_size() + + object->local_plt_offset(r_sym)); } // Set the final size. @@ -1464,11 +1617,16 @@ Output_data_plt_x86_64_standard::do_fill_plt_entry( unsigned int plt_offset, unsigned int plt_index) { + // Check PC-relative offset overflow in PLT entry. + uint64_t plt_got_pcrel_offset = (got_address + got_offset + - (plt_address + plt_offset + 6)); + if (Bits<32>::has_overflow(plt_got_pcrel_offset)) + gold_error(_("PC-relative offset overflow in PLT entry %d"), + plt_index + 1); + memcpy(pov, plt_entry, plt_entry_size); elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, - (got_address + got_offset - - (plt_address + plt_offset - + 6))); + plt_got_pcrel_offset); elfcpp::Swap_unaligned<32, false>::writeval(pov + 7, plt_index); elfcpp::Swap<32, false>::writeval(pov + 12, @@ -1602,18 +1760,9 @@ Output_data_plt_x86_64::do_write(Output_file* of) this->fill_first_plt_entry(pov, got_address, plt_address); pov += this->get_plt_entry_size(); - unsigned char* got_pov = got_view; - - // The first entry in the GOT is the address of the .dynamic section - // aka the PT_DYNAMIC segment. The next two entries are reserved. - // We saved space for them when we created the section in - // Target_x86_64::got_section. - Output_section* dynamic = this->layout_->dynamic_section(); - uint32_t dynamic_addr = dynamic == NULL ? 0 : dynamic->address(); - elfcpp::Swap<64, false>::writeval(got_pov, dynamic_addr); - got_pov += 8; - memset(got_pov, 0, 16); - got_pov += 16; + // The first three entries in the GOT are reserved, and are written + // by Output_data_got_plt_x86_64::do_write. + unsigned char* got_pov = got_view + 24; unsigned int plt_offset = this->get_plt_entry_size(); unsigned int got_offset = 24; @@ -1743,6 +1892,8 @@ template unsigned int Target_x86_64::first_plt_entry_offset() const { + if (this->plt_ == NULL) + return 0; return this->plt_->first_plt_entry_offset(); } @@ -1752,6 +1903,8 @@ template unsigned int Target_x86_64::plt_entry_size() const { + if (this->plt_ == NULL) + return 0; return this->plt_->get_plt_entry_size(); } @@ -1774,7 +1927,7 @@ Target_x86_64::init_got_plt_for_update(Symbol_table* symtab, true); // Add the three reserved entries. - this->got_plt_ = new Output_data_space((plt_count + 3) * 8, 8, "** GOT PLT"); + this->got_plt_ = new Output_data_got_plt_x86_64(layout, (plt_count + 3) * 8); layout->add_output_section_data(".got.plt", elfcpp::SHT_PROGBITS, (elfcpp::SHF_ALLOC | elfcpp::SHF_WRITE), @@ -2109,12 +2262,14 @@ Target_x86_64::Scan::get_reference_flags(unsigned int r_type) case elfcpp::R_X86_64_PC64: case elfcpp::R_X86_64_PC32: + case elfcpp::R_X86_64_PC32_BND: case elfcpp::R_X86_64_PC16: case elfcpp::R_X86_64_PC8: case elfcpp::R_X86_64_GOTOFF64: return Symbol::RELATIVE_REF; case elfcpp::R_X86_64_PLT32: + case elfcpp::R_X86_64_PLT32_BND: case elfcpp::R_X86_64_PLTOFF64: return Symbol::FUNCTION_CALL | Symbol::RELATIVE_REF; @@ -2122,6 +2277,8 @@ Target_x86_64::Scan::get_reference_flags(unsigned int r_type) case elfcpp::R_X86_64_GOT32: case elfcpp::R_X86_64_GOTPCREL64: case elfcpp::R_X86_64_GOTPCREL: + case elfcpp::R_X86_64_GOTPCRELX: + case elfcpp::R_X86_64_REX_GOTPCRELX: case elfcpp::R_X86_64_GOTPLT64: // Absolute in GOT. return Symbol::ABSOLUTE_REF; @@ -2196,6 +2353,7 @@ Target_x86_64::Scan::check_non_pic(Relobj* object, unsigned int r_type, // glibc supports these reloc types, but they can overflow. case elfcpp::R_X86_64_PC32: + case elfcpp::R_X86_64_PC32_BND: // A PC relative reference is OK against a local symbol or if // the symbol is defined locally. if (gsym == NULL @@ -2215,12 +2373,28 @@ Target_x86_64::Scan::check_non_pic(Relobj* object, unsigned int r_type, object->error(_("requires dynamic R_X86_64_32 reloc which may " "overflow at runtime; recompile with -fPIC")); else - object->error(_("requires dynamic %s reloc against '%s' which may " - "overflow at runtime; recompile with -fPIC"), - (r_type == elfcpp::R_X86_64_32 - ? "R_X86_64_32" - : "R_X86_64_PC32"), - gsym->name()); + { + const char *r_name; + switch (r_type) + { + case elfcpp::R_X86_64_32: + r_name = "R_X86_64_32"; + break; + case elfcpp::R_X86_64_PC32: + r_name = "R_X86_64_PC32"; + break; + case elfcpp::R_X86_64_PC32_BND: + r_name = "R_X86_64_PC32_BND"; + break; + default: + gold_unreachable(); + break; + } + object->error(_("requires dynamic %s reloc against '%s' " + "which may overflow at runtime; recompile " + "with -fPIC"), + r_name, gsym->name()); + } this->issued_non_pic_error_ = true; return; @@ -2270,8 +2444,12 @@ Target_x86_64::Scan::local(Symbol_table* symtab, Output_section* output_section, const elfcpp::Rela& reloc, unsigned int r_type, - const elfcpp::Sym& lsym) + const elfcpp::Sym& lsym, + bool is_discarded) { + if (is_discarded) + return; + // A local STT_GNU_IFUNC symbol may require a PLT entry. bool is_ifunc = lsym.get_st_type() == elfcpp::STT_GNU_IFUNC; if (is_ifunc && this->reloc_needs_plt_for_ifunc(object, r_type)) @@ -2299,7 +2477,7 @@ Target_x86_64::Scan::local(Symbol_table* symtab, unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); Reloc_section* rela_dyn = target->rela_dyn_section(layout); rela_dyn->add_local_relative(object, r_sym, - (size == 32 + (size == 32 ? elfcpp::R_X86_64_RELATIVE64 : elfcpp::R_X86_64_RELATIVE), output_section, data_shndx, @@ -2360,11 +2538,13 @@ Target_x86_64::Scan::local(Symbol_table* symtab, case elfcpp::R_X86_64_PC64: case elfcpp::R_X86_64_PC32: + case elfcpp::R_X86_64_PC32_BND: case elfcpp::R_X86_64_PC16: case elfcpp::R_X86_64_PC8: break; case elfcpp::R_X86_64_PLT32: + case elfcpp::R_X86_64_PLT32_BND: // Since we know this is a local symbol, we can handle this as a // PC32 reloc. break; @@ -2383,10 +2563,33 @@ Target_x86_64::Scan::local(Symbol_table* symtab, case elfcpp::R_X86_64_GOT32: case elfcpp::R_X86_64_GOTPCREL64: case elfcpp::R_X86_64_GOTPCREL: + case elfcpp::R_X86_64_GOTPCRELX: + case elfcpp::R_X86_64_REX_GOTPCRELX: case elfcpp::R_X86_64_GOTPLT64: { - // The symbol requires a GOT entry. + // The symbol requires a GOT section. Output_data_got<64, false>* got = target->got_section(symtab, layout); + + // If the relocation symbol isn't IFUNC, + // and is local, then we will convert + // mov foo@GOTPCREL(%rip), %reg + // to lea foo(%rip), %reg. + // in Relocate::relocate. + if ((r_type == elfcpp::R_X86_64_GOTPCREL + || r_type == elfcpp::R_X86_64_GOTPCRELX + || r_type == elfcpp::R_X86_64_REX_GOTPCRELX) + && reloc.get_r_offset() >= 2 + && !is_ifunc) + { + section_size_type stype; + const unsigned char* view = object->section_contents(data_shndx, + &stype, true); + if (view[reloc.get_r_offset() - 2] == 0x8b) + break; + } + + + // The symbol requires a GOT entry. unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); // For a STT_GNU_IFUNC symbol we want the PLT offset. That @@ -2477,7 +2680,7 @@ Target_x86_64::Scan::local(Symbol_table* symtab, shndx, GOT_TYPE_TLS_PAIR, target->rela_dyn_section(layout), - elfcpp::R_X86_64_DTPMOD64, 0); + elfcpp::R_X86_64_DTPMOD64); } else if (optimized_type != tls::TLSOPT_TO_LE) unsupported_reloc_local(object, r_type); @@ -2602,6 +2805,8 @@ Target_x86_64::Scan::possible_function_pointer_reloc(unsigned int r_type) case elfcpp::R_X86_64_GOT32: case elfcpp::R_X86_64_GOTPCREL64: case elfcpp::R_X86_64_GOTPCREL: + case elfcpp::R_X86_64_GOTPCRELX: + case elfcpp::R_X86_64_REX_GOTPCRELX: case elfcpp::R_X86_64_GOTPLT64: { return true; @@ -2706,7 +2911,8 @@ Target_x86_64::Scan::global(Symbol_table* symtab, // Make a dynamic relocation if necessary. if (gsym->needs_dynamic_reloc(Scan::get_reference_flags(r_type))) { - if (gsym->may_need_copy_reloc()) + if (!parameters->options().output_is_position_independent() + && gsym->may_need_copy_reloc()) { target->copy_reloc(symtab, layout, object, data_shndx, output_section, gsym, reloc); @@ -2732,7 +2938,8 @@ Target_x86_64::Scan::global(Symbol_table* symtab, reloc.get_r_offset(), reloc.get_r_addend()); } - else if (r_type == elfcpp::R_X86_64_64 + else if (((size == 64 && r_type == elfcpp::R_X86_64_64) + || (size == 32 && r_type == elfcpp::R_X86_64_32)) && gsym->can_use_relative_reloc(false)) { Reloc_section* rela_dyn = target->rela_dyn_section(layout); @@ -2756,6 +2963,7 @@ Target_x86_64::Scan::global(Symbol_table* symtab, case elfcpp::R_X86_64_PC64: case elfcpp::R_X86_64_PC32: + case elfcpp::R_X86_64_PC32_BND: case elfcpp::R_X86_64_PC16: case elfcpp::R_X86_64_PC8: { @@ -2765,7 +2973,8 @@ Target_x86_64::Scan::global(Symbol_table* symtab, // Make a dynamic relocation if necessary. if (gsym->needs_dynamic_reloc(Scan::get_reference_flags(r_type))) { - if (gsym->may_need_copy_reloc()) + if (parameters->options().output_is_executable() + && gsym->may_need_copy_reloc()) { target->copy_reloc(symtab, layout, object, data_shndx, output_section, gsym, reloc); @@ -2786,10 +2995,35 @@ Target_x86_64::Scan::global(Symbol_table* symtab, case elfcpp::R_X86_64_GOT32: case elfcpp::R_X86_64_GOTPCREL64: case elfcpp::R_X86_64_GOTPCREL: + case elfcpp::R_X86_64_GOTPCRELX: + case elfcpp::R_X86_64_REX_GOTPCRELX: case elfcpp::R_X86_64_GOTPLT64: { // The symbol requires a GOT entry. Output_data_got<64, false>* got = target->got_section(symtab, layout); + + // If we convert this from + // mov foo@GOTPCREL(%rip), %reg + // to lea foo(%rip), %reg. + // OR + // if we convert + // (callq|jmpq) *foo@GOTPCRELX(%rip) to + // (callq|jmpq) foo + // in Relocate::relocate, then there is nothing to do here. + + Lazy_view view(object, data_shndx); + size_t r_offset = reloc.get_r_offset(); + if (r_offset >= 2 + && Target_x86_64::can_convert_mov_to_lea(gsym, r_type, + r_offset, &view)) + break; + + if (r_offset >= 2 + && Target_x86_64::can_convert_callq_to_direct(gsym, r_type, + r_offset, + &view)) + break; + if (gsym->final_value_is_known()) { // For a STT_GNU_IFUNC symbol we want the PLT address. @@ -2851,15 +3085,11 @@ Target_x86_64::Scan::global(Symbol_table* symtab, } } } - // For GOTPLT64, we also need a PLT entry (but only if the - // symbol is not fully resolved). - if (r_type == elfcpp::R_X86_64_GOTPLT64 - && !gsym->final_value_is_known()) - target->make_plt_entry(symtab, layout, gsym); } break; case elfcpp::R_X86_64_PLT32: + case elfcpp::R_X86_64_PLT32_BND: // If the symbol is fully resolved, this is just a PC32 reloc. // Otherwise we need a PLT entry. if (gsym->final_value_is_known()) @@ -2910,7 +3140,12 @@ Target_x86_64::Scan::global(Symbol_table* symtab, case elfcpp::R_X86_64_GOTTPOFF: // Initial-exec case elfcpp::R_X86_64_TPOFF32: // Local-exec { - const bool is_final = gsym->final_value_is_known(); + // For the Initial-Exec model, we can treat undef symbols as final + // when building an executable. + const bool is_final = (gsym->final_value_is_known() || + (r_type == elfcpp::R_X86_64_GOTTPOFF && + gsym->is_undefined() && + parameters->options().output_is_executable())); const tls::Tls_optimization optimized_type = Target_x86_64::optimize_tls_reloc(is_final, r_type); switch (r_type) @@ -3039,15 +3274,16 @@ Target_x86_64::gc_process_relocs(Symbol_table* symtab, size_t local_symbol_count, const unsigned char* plocal_symbols) { + typedef gold::Default_classify_reloc + Classify_reloc; if (sh_type == elfcpp::SHT_REL) { return; } - gold::gc_process_relocs, elfcpp::SHT_RELA, - typename Target_x86_64::Scan, - typename Target_x86_64::Relocatable_size_for_reloc>( + gold::gc_process_relocs, Scan, + Classify_reloc>( symtab, layout, this, @@ -3077,6 +3313,9 @@ Target_x86_64::scan_relocs(Symbol_table* symtab, size_t local_symbol_count, const unsigned char* plocal_symbols) { + typedef gold::Default_classify_reloc + Classify_reloc; + if (sh_type == elfcpp::SHT_REL) { gold_error(_("%s: unsupported REL reloc section"), @@ -3084,8 +3323,7 @@ Target_x86_64::scan_relocs(Symbol_table* symtab, return; } - gold::scan_relocs, elfcpp::SHT_RELA, - typename Target_x86_64::Scan>( + gold::scan_relocs, Scan, Classify_reloc>( symtab, layout, this, @@ -3188,26 +3426,87 @@ Target_x86_64::do_finalize_sections( } } +// For x32, we need to handle PC-relative relocations using full 64-bit +// arithmetic, so that we can detect relocation overflows properly. +// This class overrides the pcrela32_check methods from the defaults in +// Relocate_functions in reloc.h. + +template +class X86_64_relocate_functions : public Relocate_functions +{ + public: + typedef Relocate_functions Base; + + // Do a simple PC relative relocation with the addend in the + // relocation. + static inline typename Base::Reloc_status + pcrela32_check(unsigned char* view, + typename elfcpp::Elf_types<64>::Elf_Addr value, + typename elfcpp::Elf_types<64>::Elf_Swxword addend, + typename elfcpp::Elf_types<64>::Elf_Addr address) + { + typedef typename elfcpp::Swap<32, false>::Valtype Valtype; + Valtype* wv = reinterpret_cast(view); + value = value + addend - address; + elfcpp::Swap<32, false>::writeval(wv, value); + return (Bits<32>::has_overflow(value) + ? Base::RELOC_OVERFLOW : Base::RELOC_OK); + } + + // Do a simple PC relative relocation with a Symbol_value with the + // addend in the relocation. + static inline typename Base::Reloc_status + pcrela32_check(unsigned char* view, + const Sized_relobj_file* object, + const Symbol_value* psymval, + typename elfcpp::Elf_types<64>::Elf_Swxword addend, + typename elfcpp::Elf_types<64>::Elf_Addr address) + { + typedef typename elfcpp::Swap<32, false>::Valtype Valtype; + Valtype* wv = reinterpret_cast(view); + typename elfcpp::Elf_types<64>::Elf_Addr value; + if (addend >= 0) + value = psymval->value(object, addend); + else + { + // For negative addends, get the symbol value without + // the addend, then add the addend using 64-bit arithmetic. + value = psymval->value(object, 0); + value += addend; + } + value -= address; + elfcpp::Swap<32, false>::writeval(wv, value); + return (Bits<32>::has_overflow(value) + ? Base::RELOC_OVERFLOW : Base::RELOC_OK); + } +}; + // Perform a relocation. template inline bool Target_x86_64::Relocate::relocate( const Relocate_info* relinfo, + unsigned int, Target_x86_64* target, Output_section*, size_t relnum, - const elfcpp::Rela& rela, - unsigned int r_type, + const unsigned char* preloc, const Sized_symbol* gsym, const Symbol_value* psymval, unsigned char* view, typename elfcpp::Elf_types::Elf_Addr address, section_size_type view_size) { + typedef X86_64_relocate_functions Reloc_funcs; + const elfcpp::Rela rela(preloc); + unsigned int r_type = elfcpp::elf_r_type(rela.get_r_info()); + if (this->skip_call_tls_get_addr_) { if ((r_type != elfcpp::R_X86_64_PLT32 + && r_type != elfcpp::R_X86_64_PLT32_BND + && r_type != elfcpp::R_X86_64_PC32_BND && r_type != elfcpp::R_X86_64_PC32) || gsym == NULL || strcmp(gsym->name(), "__tls_get_addr") != 0) @@ -3222,6 +3521,9 @@ Target_x86_64::Relocate::relocate( } } + if (view == NULL) + return true; + const Sized_relobj_file* object = relinfo->object; // Pick the value to use for symbols defined in the PLT. @@ -3229,8 +3531,7 @@ Target_x86_64::Relocate::relocate( if (gsym != NULL && gsym->use_plt_offset(Scan::get_reference_flags(r_type))) { - symval.set_output_value(target->plt_address_for_global(gsym) - + gsym->plt_offset()); + symval.set_output_value(target->plt_address_for_global(gsym)); psymval = &symval; } else if (gsym == NULL && psymval->is_ifunc_symbol()) @@ -3238,8 +3539,7 @@ Target_x86_64::Relocate::relocate( unsigned int r_sym = elfcpp::elf_r_sym(rela.get_r_info()); if (object->local_has_plt_offset(r_sym)) { - symval.set_output_value(target->plt_address_for_local(object, r_sym) - + object->local_plt_offset(r_sym)); + symval.set_output_value(target->plt_address_for_local(object, r_sym)); psymval = &symval; } } @@ -3251,13 +3551,14 @@ Target_x86_64::Relocate::relocate( // We need to subtract the size of the GOT section to get // the actual offset to use in the relocation. bool have_got_offset = false; - unsigned int got_offset = 0; + // Since the actual offset is always negative, we use signed int to + // support 64-bit GOT relocations. + int got_offset = 0; switch (r_type) { case elfcpp::R_X86_64_GOT32: case elfcpp::R_X86_64_GOT64: case elfcpp::R_X86_64_GOTPLT64: - case elfcpp::R_X86_64_GOTPCREL: case elfcpp::R_X86_64_GOTPCREL64: if (gsym != NULL) { @@ -3278,6 +3579,8 @@ Target_x86_64::Relocate::relocate( break; } + typename Reloc_funcs::Reloc_status rstatus = Reloc_funcs::RELOC_OK; + switch (r_type) { case elfcpp::R_X86_64_NONE: @@ -3286,53 +3589,48 @@ Target_x86_64::Relocate::relocate( break; case elfcpp::R_X86_64_64: - Relocate_functions::rela64(view, object, psymval, addend); + Reloc_funcs::rela64(view, object, psymval, addend); break; case elfcpp::R_X86_64_PC64: - Relocate_functions::pcrela64(view, object, psymval, addend, + Reloc_funcs::pcrela64(view, object, psymval, addend, address); break; case elfcpp::R_X86_64_32: - // FIXME: we need to verify that value + addend fits into 32 bits: - // uint64_t x = value + addend; - // x == static_cast(static_cast(x)) - // Likewise for other <=32-bit relocations (but see R_X86_64_32S). - Relocate_functions::rela32(view, object, psymval, addend); + rstatus = Reloc_funcs::rela32_check(view, object, psymval, addend, + Reloc_funcs::CHECK_UNSIGNED); break; case elfcpp::R_X86_64_32S: - // FIXME: we need to verify that value + addend fits into 32 bits: - // int64_t x = value + addend; // note this quantity is signed! - // x == static_cast(static_cast(x)) - Relocate_functions::rela32(view, object, psymval, addend); + rstatus = Reloc_funcs::rela32_check(view, object, psymval, addend, + Reloc_funcs::CHECK_SIGNED); break; case elfcpp::R_X86_64_PC32: - Relocate_functions::pcrela32(view, object, psymval, addend, - address); + case elfcpp::R_X86_64_PC32_BND: + rstatus = Reloc_funcs::pcrela32_check(view, object, psymval, addend, + address); break; case elfcpp::R_X86_64_16: - Relocate_functions::rela16(view, object, psymval, addend); + Reloc_funcs::rela16(view, object, psymval, addend); break; case elfcpp::R_X86_64_PC16: - Relocate_functions::pcrela16(view, object, psymval, addend, - address); + Reloc_funcs::pcrela16(view, object, psymval, addend, address); break; case elfcpp::R_X86_64_8: - Relocate_functions::rela8(view, object, psymval, addend); + Reloc_funcs::rela8(view, object, psymval, addend); break; case elfcpp::R_X86_64_PC8: - Relocate_functions::pcrela8(view, object, psymval, addend, - address); + Reloc_funcs::pcrela8(view, object, psymval, addend, address); break; case elfcpp::R_X86_64_PLT32: + case elfcpp::R_X86_64_PLT32_BND: gold_assert(gsym == NULL || gsym->has_plt_offset() || gsym->final_value_is_known() @@ -3342,8 +3640,8 @@ Target_x86_64::Relocate::relocate( // Note: while this code looks the same as for R_X86_64_PC32, it // behaves differently because psymval was set to point to // the PLT entry, rather than the symbol, in Scan::global(). - Relocate_functions::pcrela32(view, object, psymval, addend, - address); + rstatus = Reloc_funcs::pcrela32_check(view, object, psymval, addend, + address); break; case elfcpp::R_X86_64_PLTOFF64: @@ -3352,14 +3650,15 @@ Target_x86_64::Relocate::relocate( gold_assert(gsym->has_plt_offset() || gsym->final_value_is_known()); typename elfcpp::Elf_types::Elf_Addr got_address; - got_address = target->got_section(NULL, NULL)->address(); - Relocate_functions::rela64(view, object, psymval, - addend - got_address); + // This is the address of GLOBAL_OFFSET_TABLE. + got_address = target->got_plt_section()->address(); + Reloc_funcs::rela64(view, object, psymval, addend - got_address); } + break; case elfcpp::R_X86_64_GOT32: gold_assert(have_got_offset); - Relocate_functions::rela32(view, got_offset, addend); + Reloc_funcs::rela32(view, got_offset, addend); break; case elfcpp::R_X86_64_GOTPC32: @@ -3367,16 +3666,16 @@ Target_x86_64::Relocate::relocate( gold_assert(gsym); typename elfcpp::Elf_types::Elf_Addr value; value = target->got_plt_section()->address(); - Relocate_functions::pcrela32(view, value, addend, address); + Reloc_funcs::pcrela32_check(view, value, addend, address); } break; case elfcpp::R_X86_64_GOT64: - // The ABI doc says "Like GOT64, but indicates a PLT entry is needed." - // Since we always add a PLT entry, this is equivalent. case elfcpp::R_X86_64_GOTPLT64: + // R_X86_64_GOTPLT64 is obsolete and treated the the same as + // GOT64. gold_assert(have_got_offset); - Relocate_functions::rela64(view, got_offset, addend); + Reloc_funcs::rela64(view, got_offset, addend); break; case elfcpp::R_X86_64_GOTPC64: @@ -3384,7 +3683,7 @@ Target_x86_64::Relocate::relocate( gold_assert(gsym); typename elfcpp::Elf_types::Elf_Addr value; value = target->got_plt_section()->address(); - Relocate_functions::pcrela64(view, value, addend, address); + Reloc_funcs::pcrela64(view, value, addend, address); } break; @@ -3393,16 +3692,86 @@ Target_x86_64::Relocate::relocate( typename elfcpp::Elf_types::Elf_Addr value; value = (psymval->value(object, 0) - target->got_plt_section()->address()); - Relocate_functions::rela64(view, value, addend); + Reloc_funcs::rela64(view, value, addend); } break; case elfcpp::R_X86_64_GOTPCREL: + case elfcpp::R_X86_64_GOTPCRELX: + case elfcpp::R_X86_64_REX_GOTPCRELX: { - gold_assert(have_got_offset); - typename elfcpp::Elf_types::Elf_Addr value; - value = target->got_plt_section()->address() + got_offset; - Relocate_functions::pcrela32(view, value, addend, address); + // Convert + // mov foo@GOTPCREL(%rip), %reg + // to lea foo(%rip), %reg. + // if possible. + if ((gsym == NULL + && rela.get_r_offset() >= 2 + && view[-2] == 0x8b + && !psymval->is_ifunc_symbol()) + || (gsym != NULL + && rela.get_r_offset() >= 2 + && Target_x86_64::can_convert_mov_to_lea(gsym, r_type, + 0, &view))) + { + view[-2] = 0x8d; + Reloc_funcs::pcrela32(view, object, psymval, addend, address); + } + // Convert + // callq *foo@GOTPCRELX(%rip) to + // addr32 callq foo + // and jmpq *foo@GOTPCRELX(%rip) to + // jmpq foo + // nop + else if (gsym != NULL + && rela.get_r_offset() >= 2 + && Target_x86_64::can_convert_callq_to_direct(gsym, + r_type, + 0, &view)) + { + if (view[-1] == 0x15) + { + // Convert callq *foo@GOTPCRELX(%rip) to addr32 callq. + // Opcode of addr32 is 0x67 and opcode of direct callq is 0xe8. + view[-2] = 0x67; + view[-1] = 0xe8; + // Convert GOTPCRELX to 32-bit pc relative reloc. + Reloc_funcs::pcrela32(view, object, psymval, addend, address); + } + else + { + // Convert jmpq *foo@GOTPCRELX(%rip) to + // jmpq foo + // nop + // The opcode of direct jmpq is 0xe9. + view[-2] = 0xe9; + // The opcode of nop is 0x90. + view[3] = 0x90; + // Convert GOTPCRELX to 32-bit pc relative reloc. jmpq is rip + // relative and since the instruction following the jmpq is now + // the nop, offset the address by 1 byte. The start of the + // relocation also moves ahead by 1 byte. + Reloc_funcs::pcrela32(&view[-1], object, psymval, addend, + address - 1); + } + } + else + { + if (gsym != NULL) + { + gold_assert(gsym->has_got_offset(GOT_TYPE_STANDARD)); + got_offset = gsym->got_offset(GOT_TYPE_STANDARD) - target->got_size(); + } + else + { + unsigned int r_sym = elfcpp::elf_r_sym(rela.get_r_info()); + gold_assert(object->local_has_got_offset(r_sym, GOT_TYPE_STANDARD)); + got_offset = (object->local_got_offset(r_sym, GOT_TYPE_STANDARD) + - target->got_size()); + } + typename elfcpp::Elf_types::Elf_Addr value; + value = target->got_plt_section()->address() + got_offset; + Reloc_funcs::pcrela32_check(view, value, addend, address); + } } break; @@ -3411,7 +3780,7 @@ Target_x86_64::Relocate::relocate( gold_assert(have_got_offset); typename elfcpp::Elf_types::Elf_Addr value; value = target->got_plt_section()->address() + got_offset; - Relocate_functions::pcrela64(view, value, addend, address); + Reloc_funcs::pcrela64(view, value, addend, address); } break; @@ -3451,6 +3820,32 @@ Target_x86_64::Relocate::relocate( break; } + if (rstatus == Reloc_funcs::RELOC_OVERFLOW) + { + if (gsym == NULL) + { + unsigned int r_sym = elfcpp::elf_r_sym(rela.get_r_info()); + gold_error_at_location(relinfo, relnum, rela.get_r_offset(), + _("relocation overflow: " + "reference to local symbol %u in %s"), + r_sym, object->name().c_str()); + } + else if (gsym->is_defined() && gsym->source() == Symbol::FROM_OBJECT) + { + gold_error_at_location(relinfo, relnum, rela.get_r_offset(), + _("relocation overflow: " + "reference to '%s' defined in %s"), + gsym->name(), + gsym->object()->name().c_str()); + } + else + { + gold_error_at_location(relinfo, relnum, rela.get_r_offset(), + _("relocation overflow: reference to '%s'"), + gsym->name()); + } + } + return true; } @@ -3531,7 +3926,7 @@ Target_x86_64::Relocate::relocate_tls( if (optimized_type == tls::TLSOPT_TO_IE) { value = target->got_plt_section()->address() + got_offset; - this->tls_gd_to_ie(relinfo, relnum, tls_segment, rela, r_type, + this->tls_gd_to_ie(relinfo, relnum, rela, r_type, value, view, address, view_size); break; } @@ -3598,14 +3993,8 @@ Target_x86_64::Relocate::relocate_tls( } if (optimized_type == tls::TLSOPT_TO_IE) { - if (tls_segment == NULL) - { - gold_assert(parameters->errors()->error_count() > 0 - || issue_undefined_symbol_error(gsym)); - return; - } value = target->got_plt_section()->address() + got_offset; - this->tls_desc_gd_to_ie(relinfo, relnum, tls_segment, + this->tls_desc_gd_to_ie(relinfo, relnum, rela, r_type, value, view, address, view_size); break; @@ -3697,7 +4086,17 @@ Target_x86_64::Relocate::relocate_tls( break; case elfcpp::R_X86_64_GOTTPOFF: // Initial-exec - if (optimized_type == tls::TLSOPT_TO_LE) + if (gsym != NULL + && gsym->is_undefined() + && parameters->options().output_is_executable()) + { + Target_x86_64::Relocate::tls_ie_to_le(relinfo, relnum, + NULL, rela, + r_type, value, view, + view_size); + break; + } + else if (optimized_type == tls::TLSOPT_TO_LE) { if (tls_segment == NULL) { @@ -3761,7 +4160,6 @@ inline void Target_x86_64::Relocate::tls_gd_to_ie( const Relocate_info* relinfo, size_t relnum, - Output_segment*, const elfcpp::Rela& rela, unsigned int, typename elfcpp::Elf_types::Elf_Addr value, @@ -3873,7 +4271,6 @@ inline void Target_x86_64::Relocate::tls_desc_gd_to_ie( const Relocate_info* relinfo, size_t relnum, - Output_segment*, const elfcpp::Rela& rela, unsigned int r_type, typename elfcpp::Elf_types::Elf_Addr value, @@ -3959,8 +4356,12 @@ Target_x86_64::Relocate::tls_ld_to_le( section_size_type view_size) { // leaq foo@tlsld(%rip),%rdi; call __tls_get_addr@plt; + // For SIZE == 64: // ... leq foo@dtpoff(%rax),%reg // ==> .word 0x6666; .byte 0x66; movq %fs:0,%rax ... leaq x@tpoff(%rax),%rdx + // For SIZE == 32: + // ... leq foo@dtpoff(%rax),%reg + // ==> nopl 0x0(%rax); movl %fs:0,%eax ... leaq x@tpoff(%rax),%rdx tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, -3); tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, 9); @@ -3970,7 +4371,10 @@ Target_x86_64::Relocate::tls_ld_to_le( tls::check_tls(relinfo, relnum, rela.get_r_offset(), view[4] == 0xe8); - memcpy(view - 3, "\x66\x66\x66\x64\x48\x8b\x04\x25\0\0\0\0", 12); + if (size == 64) + memcpy(view - 3, "\x66\x66\x66\x64\x48\x8b\x04\x25\0\0\0\0", 12); + else + memcpy(view - 3, "\x0f\x1f\x40\x00\x64\x8b\x04\x25\0\0\0\0", 12); // The next reloc should be a PLT32 reloc against __tls_get_addr. // We can skip it. @@ -4011,6 +4415,8 @@ Target_x86_64::Relocate::tls_ie_to_le( // movq if (op1 == 0x4c) view[-3] = 0x49; + else if (size == 32 && op1 == 0x44) + view[-3] = 0x41; view[-2] = 0xc7; view[-1] = 0xc0 | reg; } @@ -4019,6 +4425,8 @@ Target_x86_64::Relocate::tls_ie_to_le( // Special handling for %rsp. if (op1 == 0x4c) view[-3] = 0x49; + else if (size == 32 && op1 == 0x44) + view[-3] = 0x41; view[-2] = 0x81; view[-1] = 0xc0 | reg; } @@ -4027,11 +4435,14 @@ Target_x86_64::Relocate::tls_ie_to_le( // addq if (op1 == 0x4c) view[-3] = 0x4d; + else if (size == 32 && op1 == 0x44) + view[-3] = 0x45; view[-2] = 0x8d; view[-1] = 0x80 | reg | (reg << 3); } - value -= tls_segment->memsz(); + if (tls_segment != NULL) + value -= tls_segment->memsz(); Relocate_functions::rela32(view, value, 0); } @@ -4051,10 +4462,13 @@ Target_x86_64::relocate_section( section_size_type view_size, const Reloc_symbol_changes* reloc_symbol_changes) { + typedef gold::Default_classify_reloc + Classify_reloc; + gold_assert(sh_type == elfcpp::SHT_RELA); - gold::relocate_section, elfcpp::SHT_RELA, - typename Target_x86_64::Relocate>( + gold::relocate_section, Relocate, + gold::Default_comdat_behavior, Classify_reloc>( relinfo, this, prelocs, @@ -4095,82 +4509,50 @@ Target_x86_64::apply_relocation( view_size); } -// Return the size of a relocation while scanning during a relocatable -// link. +// Scan the relocs during a relocatable link. template -unsigned int -Target_x86_64::Relocatable_size_for_reloc::get_size_for_reloc( - unsigned int r_type, - Relobj* object) +void +Target_x86_64::scan_relocatable_relocs( + Symbol_table* symtab, + Layout* layout, + Sized_relobj_file* object, + unsigned int data_shndx, + unsigned int sh_type, + const unsigned char* prelocs, + size_t reloc_count, + Output_section* output_section, + bool needs_special_offset_handling, + size_t local_symbol_count, + const unsigned char* plocal_symbols, + Relocatable_relocs* rr) { - switch (r_type) - { - case elfcpp::R_X86_64_NONE: - case elfcpp::R_X86_64_GNU_VTINHERIT: - case elfcpp::R_X86_64_GNU_VTENTRY: - case elfcpp::R_X86_64_TLSGD: // Global-dynamic - case elfcpp::R_X86_64_GOTPC32_TLSDESC: // Global-dynamic (from ~oliva url) - case elfcpp::R_X86_64_TLSDESC_CALL: - case elfcpp::R_X86_64_TLSLD: // Local-dynamic - case elfcpp::R_X86_64_DTPOFF32: - case elfcpp::R_X86_64_DTPOFF64: - case elfcpp::R_X86_64_GOTTPOFF: // Initial-exec - case elfcpp::R_X86_64_TPOFF32: // Local-exec - return 0; - - case elfcpp::R_X86_64_64: - case elfcpp::R_X86_64_PC64: - case elfcpp::R_X86_64_GOTOFF64: - case elfcpp::R_X86_64_GOTPC64: - case elfcpp::R_X86_64_PLTOFF64: - case elfcpp::R_X86_64_GOT64: - case elfcpp::R_X86_64_GOTPCREL64: - case elfcpp::R_X86_64_GOTPCREL: - case elfcpp::R_X86_64_GOTPLT64: - return 8; + typedef gold::Default_classify_reloc + Classify_reloc; + typedef gold::Default_scan_relocatable_relocs + Scan_relocatable_relocs; - case elfcpp::R_X86_64_32: - case elfcpp::R_X86_64_32S: - case elfcpp::R_X86_64_PC32: - case elfcpp::R_X86_64_PLT32: - case elfcpp::R_X86_64_GOTPC32: - case elfcpp::R_X86_64_GOT32: - return 4; - - case elfcpp::R_X86_64_16: - case elfcpp::R_X86_64_PC16: - return 2; - - case elfcpp::R_X86_64_8: - case elfcpp::R_X86_64_PC8: - return 1; - - case elfcpp::R_X86_64_COPY: - case elfcpp::R_X86_64_GLOB_DAT: - case elfcpp::R_X86_64_JUMP_SLOT: - case elfcpp::R_X86_64_RELATIVE: - case elfcpp::R_X86_64_IRELATIVE: - // These are outstanding tls relocs, which are unexpected when linking - case elfcpp::R_X86_64_TPOFF64: - case elfcpp::R_X86_64_DTPMOD64: - case elfcpp::R_X86_64_TLSDESC: - object->error(_("unexpected reloc %u in object file"), r_type); - return 0; + gold_assert(sh_type == elfcpp::SHT_RELA); - case elfcpp::R_X86_64_SIZE32: - case elfcpp::R_X86_64_SIZE64: - default: - object->error(_("unsupported reloc %u against local symbol"), r_type); - return 0; - } + gold::scan_relocatable_relocs( + symtab, + layout, + object, + data_shndx, + prelocs, + reloc_count, + output_section, + needs_special_offset_handling, + local_symbol_count, + plocal_symbols, + rr); } -// Scan the relocs during a relocatable link. +// Scan the relocs for --emit-relocs. template void -Target_x86_64::scan_relocatable_relocs( +Target_x86_64::emit_relocs_scan( Symbol_table* symtab, Layout* layout, Sized_relobj_file* object, @@ -4181,16 +4563,17 @@ Target_x86_64::scan_relocatable_relocs( Output_section* output_section, bool needs_special_offset_handling, size_t local_symbol_count, - const unsigned char* plocal_symbols, + const unsigned char* plocal_syms, Relocatable_relocs* rr) { - gold_assert(sh_type == elfcpp::SHT_RELA); + typedef gold::Default_classify_reloc + Classify_reloc; + typedef gold::Default_emit_relocs_strategy + Emit_relocs_strategy; - typedef gold::Default_scan_relocatable_relocs Scan_relocatable_relocs; + gold_assert(sh_type == elfcpp::SHT_RELA); - gold::scan_relocatable_relocs( + gold::scan_relocatable_relocs( symtab, layout, object, @@ -4200,7 +4583,7 @@ Target_x86_64::scan_relocatable_relocs( output_section, needs_special_offset_handling, local_symbol_count, - plocal_symbols, + plocal_syms, rr); } @@ -4214,23 +4597,24 @@ Target_x86_64::relocate_relocs( const unsigned char* prelocs, size_t reloc_count, Output_section* output_section, - off_t offset_in_output_section, - const Relocatable_relocs* rr, + typename elfcpp::Elf_types::Elf_Off offset_in_output_section, unsigned char* view, typename elfcpp::Elf_types::Elf_Addr view_address, section_size_type view_size, unsigned char* reloc_view, section_size_type reloc_view_size) { + typedef gold::Default_classify_reloc + Classify_reloc; + gold_assert(sh_type == elfcpp::SHT_RELA); - gold::relocate_relocs( + gold::relocate_relocs( relinfo, prelocs, reloc_count, output_section, offset_in_output_section, - rr, view, view_address, view_size, @@ -4248,7 +4632,7 @@ uint64_t Target_x86_64::do_dynsym_value(const Symbol* gsym) const { gold_assert(gsym->is_from_dynobj() && gsym->has_plt_offset()); - return this->plt_address_for_global(gsym) + gsym->plt_offset(); + return this->plt_address_for_global(gsym); } // Return a string used to fill a code section with nops to take up @@ -4365,35 +4749,60 @@ Target_x86_64::do_ehframe_datarel_base() const // code. We have to change the function so that it always ensures // that it has enough stack space to run some random function. +static const unsigned char cmp_insn_32[] = { 0x64, 0x3b, 0x24, 0x25 }; +static const unsigned char lea_r10_insn_32[] = { 0x44, 0x8d, 0x94, 0x24 }; +static const unsigned char lea_r11_insn_32[] = { 0x44, 0x8d, 0x9c, 0x24 }; + +static const unsigned char cmp_insn_64[] = { 0x64, 0x48, 0x3b, 0x24, 0x25 }; +static const unsigned char lea_r10_insn_64[] = { 0x4c, 0x8d, 0x94, 0x24 }; +static const unsigned char lea_r11_insn_64[] = { 0x4c, 0x8d, 0x9c, 0x24 }; + template void Target_x86_64::do_calls_non_split(Relobj* object, unsigned int shndx, section_offset_type fnoffset, section_size_type fnsize, + const unsigned char*, + size_t, unsigned char* view, section_size_type view_size, std::string* from, std::string* to) const { + const char* const cmp_insn = reinterpret_cast + (size == 32 ? cmp_insn_32 : cmp_insn_64); + const char* const lea_r10_insn = reinterpret_cast + (size == 32 ? lea_r10_insn_32 : lea_r10_insn_64); + const char* const lea_r11_insn = reinterpret_cast + (size == 32 ? lea_r11_insn_32 : lea_r11_insn_64); + + const size_t cmp_insn_len = + (size == 32 ? sizeof(cmp_insn_32) : sizeof(cmp_insn_64)); + const size_t lea_r10_insn_len = + (size == 32 ? sizeof(lea_r10_insn_32) : sizeof(lea_r10_insn_64)); + const size_t lea_r11_insn_len = + (size == 32 ? sizeof(lea_r11_insn_32) : sizeof(lea_r11_insn_64)); + const size_t nop_len = (size == 32 ? 7 : 8); + // The function starts with a comparison of the stack pointer and a // field in the TCB. This is followed by a jump. // cmp %fs:NN,%rsp - if (this->match_view(view, view_size, fnoffset, "\x64\x48\x3b\x24\x25", 5) - && fnsize > 9) + if (this->match_view(view, view_size, fnoffset, cmp_insn, cmp_insn_len) + && fnsize > nop_len + 1) { // We will call __morestack if the carry flag is set after this // comparison. We turn the comparison into an stc instruction // and some nops. view[fnoffset] = '\xf9'; - this->set_view_to_nop(view, view_size, fnoffset + 1, 8); + this->set_view_to_nop(view, view_size, fnoffset + 1, nop_len); } // lea NN(%rsp),%r10 // lea NN(%rsp),%r11 else if ((this->match_view(view, view_size, fnoffset, - "\x4c\x8d\x94\x24", 4) + lea_r10_insn, lea_r10_insn_len) || this->match_view(view, view_size, fnoffset, - "\x4c\x8d\x9c\x24", 4)) + lea_r11_insn, lea_r11_insn_len)) && fnsize > 8) { // This is loading an offset from the stack pointer for a @@ -4453,7 +4862,7 @@ class Output_data_plt_x86_64_nacl : public Output_data_plt_x86_64 public: Output_data_plt_x86_64_nacl(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative) : Output_data_plt_x86_64(layout, plt_entry_size, got, got_plt, got_irelative) @@ -4461,7 +4870,7 @@ class Output_data_plt_x86_64_nacl : public Output_data_plt_x86_64 Output_data_plt_x86_64_nacl(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative, unsigned int plt_count) : Output_data_plt_x86_64(layout, plt_entry_size, @@ -4534,7 +4943,7 @@ class Target_x86_64_nacl : public Target_x86_64 virtual Output_data_plt_x86_64* do_make_data_plt(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative) { return new Output_data_plt_x86_64_nacl(layout, got, got_plt, @@ -4544,7 +4953,7 @@ class Target_x86_64_nacl : public Target_x86_64 virtual Output_data_plt_x86_64* do_make_data_plt(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative, unsigned int plt_count) { @@ -4553,6 +4962,9 @@ class Target_x86_64_nacl : public Target_x86_64 plt_count); } + virtual std::string + do_code_fill(section_size_type length) const; + private: static const Target::Target_info x86_64_nacl_info; }; @@ -4580,7 +4992,9 @@ const Target::Target_info Target_x86_64_nacl<64>::x86_64_nacl_info = 0, // small_common_section_flags elfcpp::SHF_X86_64_LARGE, // large_common_section_flags NULL, // attributes_section - NULL // attributes_vendor + NULL, // attributes_vendor + "_start", // entry_symbol_name + 32, // hash_entry_size }; template<> @@ -4606,7 +5020,9 @@ const Target::Target_info Target_x86_64_nacl<32>::x86_64_nacl_info = 0, // small_common_section_flags elfcpp::SHF_X86_64_LARGE, // large_common_section_flags NULL, // attributes_section - NULL // attributes_vendor + NULL, // attributes_vendor + "_start", // entry_symbol_name + 32, // hash_entry_size }; #define NACLMASK 0xe0 // 32-byte alignment mask. @@ -4626,7 +5042,7 @@ Output_data_plt_x86_64_nacl::first_plt_entry[plt_entry_size] = 0x41, 0xff, 0xe3, // jmpq *%r11 // 9-byte nop sequence to pad out to the next 32-byte boundary. - 0x2e, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopl %cs:0x0(%rax,%rax,1) + 0x66, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopw 0x0(%rax,%rax,1) // 32 bytes of nop to pad out to the standard size 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, // excess data32 prefixes @@ -4777,6 +5193,16 @@ Output_data_plt_x86_64_nacl::plt_eh_frame_fde[plt_eh_frame_fde_size] = elfcpp::DW_CFA_nop }; +// Return a string used to fill a code section with nops. +// For NaCl, long NOPs are only valid if they do not cross +// bundle alignment boundaries, so keep it simple with one-byte NOPs. +template +std::string +Target_x86_64_nacl::do_code_fill(section_size_type length) const +{ + return std::string(length, static_cast(0x90)); +} + // The selector for x86_64-nacl object files. template