X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=gold%2Fx86_64.cc;h=8b6027e72c309ffa96f8364f4bb112d7b4c35e84;hb=ac1bbaca10666fc85572a6deeaa6f1debcd4c129;hp=029f89c008e702edefa4de0a8c48b2c971745736;hpb=1bae613c85d3266ff261e36da1625582dd5d2655;p=binutils-gdb.git diff --git a/gold/x86_64.cc b/gold/x86_64.cc index 029f89c008e..8b6027e72c3 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) 2006-2020 Free Software Foundation, Inc. // Written by Ian Lance Taylor . // This file is part of gold. @@ -40,6 +39,7 @@ #include "target-select.h" #include "tls.h" #include "freebsd.h" +#include "nacl.h" #include "gc.h" #include "icf.h" @@ -48,7 +48,41 @@ 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 +// classes below fill in those details. template class Output_data_plt_x86_64 : public Output_section_data @@ -56,21 +90,24 @@ class Output_data_plt_x86_64 : public Output_section_data public: typedef Output_data_reloc Reloc_section; - Output_data_plt_x86_64(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_plt_x86_64(Layout* layout, uint64_t addralign, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, Output_data_space* got_irelative) - : Output_section_data(16), 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_() { this->init(layout); } - Output_data_plt_x86_64(Layout* layout, Output_data_got<64, false>* got, - Output_data_space* got_plt, + Output_data_plt_x86_64(Layout* layout, uint64_t plt_entry_size, + Output_data_got<64, false>* got, + 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, 16, false), - layout_(layout), tlsdesc_rel_(NULL), irelative_rel_(NULL), got_(got), + : Output_section_data((plt_count + 1) * plt_entry_size, + plt_entry_size, false), + 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_() { @@ -118,7 +155,10 @@ class Output_data_plt_x86_64 : public Output_section_data // Return the offset of the reserved TLSDESC_PLT entry. unsigned int get_tlsdesc_plt_offset() const - { return (this->count_ + this->irelative_count_ + 1) * plt_entry_size; } + { + return ((this->count_ + this->irelative_count_ + 1) + * this->get_plt_entry_size()); + } // Return the .rela.plt section data. Reloc_section* @@ -139,67 +179,145 @@ class Output_data_plt_x86_64 : public Output_section_data has_irelative_section() const { return this->irelative_rel_ != NULL; } - // Return the number of PLT entries. + // Get count of regular PLT entries. + unsigned int + regular_count() const + { return this->count_; } + + // Return the total number of PLT entries. unsigned int entry_count() const { return this->count_ + this->irelative_count_; } // Return the offset of the first non-reserved PLT entry. - static unsigned int + unsigned int first_plt_entry_offset() - { return plt_entry_size; } + { return this->get_plt_entry_size(); } // Return the size of a PLT entry. - static unsigned int - get_plt_entry_size() - { return plt_entry_size; } + unsigned int + get_plt_entry_size() const + { return this->do_get_plt_entry_size(); } // Reserve a slot in the PLT for an existing symbol in an incremental update. void reserve_slot(unsigned int plt_index) { - this->free_list_.remove((plt_index + 1) * plt_entry_size, - (plt_index + 2) * plt_entry_size); + this->free_list_.remove((plt_index + 1) * this->get_plt_entry_size(), + (plt_index + 2) * this->get_plt_entry_size()); } // Return the PLT address to use for a global symbol. uint64_t - address_for_global(const Symbol*); + address_for_global(const Symbol* sym) + { return do_address_for_global(sym); } // Return the PLT address to use for a local symbol. uint64_t - address_for_local(const Relobj*, unsigned int symndx); + address_for_local(const Relobj* obj, unsigned int symndx) + { return do_address_for_local(obj, symndx); } + + // Add .eh_frame information for the PLT. + void + add_eh_frame(Layout* layout) + { this->do_add_eh_frame(layout); } protected: + Output_data_got<64, false>* + got() const + { return this->got_; } + + Output_data_got_plt_x86_64* + got_plt() const + { return this->got_plt_; } + + Output_data_space* + got_irelative() const + { return this->got_irelative_; } + + // Fill in the first PLT entry. void - do_adjust_output_section(Output_section* os); + fill_first_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address) + { this->do_fill_first_plt_entry(pov, got_address, plt_address); } - // Write to a map file. + // Fill in a normal PLT entry. Returns the offset into the entry that + // should be the initial GOT slot value. + unsigned int + fill_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index) + { + return this->do_fill_plt_entry(pov, got_address, plt_address, + got_offset, plt_offset, plt_index); + } + + // Fill in the reserved TLSDESC PLT entry. void - do_print_to_mapfile(Mapfile* mapfile) const - { mapfile->print_output_data(this, _("** PLT")); } + fill_tlsdesc_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + typename elfcpp::Elf_types::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset) + { + this->do_fill_tlsdesc_entry(pov, got_address, plt_address, got_base, + tlsdesc_got_offset, plt_offset); + } - private: - // The size of an entry in the PLT. - static const int plt_entry_size = 16; + virtual unsigned int + do_get_plt_entry_size() const = 0; + + virtual void + do_fill_first_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_addr, + typename elfcpp::Elf_types::Elf_Addr plt_addr) + = 0; + + virtual unsigned int + do_fill_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index) = 0; + + virtual void + do_fill_tlsdesc_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + typename elfcpp::Elf_types::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset) = 0; - // The first entry in the PLT. - // From the AMD64 ABI: "Unlike Intel386 ABI, this ABI uses the same - // procedure linkage table for both programs and shared objects." - static const unsigned char first_plt_entry[plt_entry_size]; + // Return the PLT address to use for a global symbol. + virtual uint64_t + do_address_for_global(const Symbol* sym); - // Other entries in the PLT for an executable. - static const unsigned char plt_entry[plt_entry_size]; + // Return the PLT address to use for a local symbol. + virtual uint64_t + do_address_for_local(const Relobj* obj, unsigned int symndx); - // The reserved TLSDESC entry in the PLT for an executable. - static const unsigned char tlsdesc_plt_entry[plt_entry_size]; + virtual void + do_add_eh_frame(Layout* layout) = 0; - // The .eh_frame unwind information for the PLT. + void + do_adjust_output_section(Output_section* os); + + // Write to a map file. + void + do_print_to_mapfile(Mapfile* mapfile) const + { mapfile->print_output_data(this, _("** PLT")); } + + // The CIE of the .eh_frame unwind information for the PLT. static const int plt_eh_frame_cie_size = 16; - static const int plt_eh_frame_fde_size = 32; static const unsigned char plt_eh_frame_cie[plt_eh_frame_cie_size]; - static const unsigned char plt_eh_frame_fde[plt_eh_frame_fde_size]; + private: // Set the final size. void set_final_data_size(); @@ -208,9 +326,6 @@ 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 @@ -222,7 +337,7 @@ 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. @@ -237,6 +352,338 @@ class Output_data_plt_x86_64 : public Output_section_data Free_list free_list_; }; +template +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_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) + { } + + Output_data_plt_x86_64_standard(Layout* layout, + Output_data_got<64, false>* got, + 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, + got, got_plt, got_irelative, + plt_count) + { } + + protected: + virtual unsigned int + do_get_plt_entry_size() const + { return plt_entry_size; } + + virtual void + do_add_eh_frame(Layout* layout) + { + layout->add_eh_frame_for_plt(this, + this->plt_eh_frame_cie, + this->plt_eh_frame_cie_size, + plt_eh_frame_fde, + plt_eh_frame_fde_size); + } + + virtual void + do_fill_first_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_addr, + typename elfcpp::Elf_types::Elf_Addr plt_addr); + + virtual unsigned int + do_fill_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index); + + virtual void + do_fill_tlsdesc_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + typename elfcpp::Elf_types::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset); + + private: + // The size of an entry in the PLT. + static const int plt_entry_size = 16; + + // The first entry in the PLT. + // From the AMD64 ABI: "Unlike Intel386 ABI, this ABI uses the same + // procedure linkage table for both programs and shared objects." + static const unsigned char first_plt_entry[plt_entry_size]; + + // Other entries in the PLT for an executable. + static const unsigned char plt_entry[plt_entry_size]; + + // The reserved TLSDESC entry in the PLT for an executable. + static const unsigned char tlsdesc_plt_entry[plt_entry_size]; + + // The .eh_frame unwind information for the PLT. + static const int plt_eh_frame_fde_size = 32; + static const unsigned char plt_eh_frame_fde[plt_eh_frame_fde_size]; +}; + +class Output_data_plt_x86_64_bnd : public Output_data_plt_x86_64<64> +{ + public: + Output_data_plt_x86_64_bnd(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative) + : Output_data_plt_x86_64<64>(layout, plt_entry_size, + got, got_plt, got_irelative), + aplt_offset_(0) + { } + + Output_data_plt_x86_64_bnd(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative, + unsigned int plt_count) + : Output_data_plt_x86_64<64>(layout, plt_entry_size, + got, got_plt, got_irelative, + plt_count), + aplt_offset_(0) + { } + + protected: + virtual unsigned int + do_get_plt_entry_size() const + { return plt_entry_size; } + + // Return the PLT address to use for a global symbol. + uint64_t + do_address_for_global(const Symbol*); + + // Return the PLT address to use for a local symbol. + uint64_t + do_address_for_local(const Relobj*, unsigned int symndx); + + virtual void + do_add_eh_frame(Layout* layout) + { + layout->add_eh_frame_for_plt(this, + this->plt_eh_frame_cie, + this->plt_eh_frame_cie_size, + plt_eh_frame_fde, + plt_eh_frame_fde_size); + } + + virtual void + do_fill_first_plt_entry(unsigned char* pov, + elfcpp::Elf_types<64>::Elf_Addr got_addr, + elfcpp::Elf_types<64>::Elf_Addr plt_addr); + + virtual unsigned int + do_fill_plt_entry(unsigned char* pov, + elfcpp::Elf_types<64>::Elf_Addr got_address, + elfcpp::Elf_types<64>::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index); + + virtual void + do_fill_tlsdesc_entry(unsigned char* pov, + elfcpp::Elf_types<64>::Elf_Addr got_address, + elfcpp::Elf_types<64>::Elf_Addr plt_address, + elfcpp::Elf_types<64>::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset); + + void + fill_aplt_entry(unsigned char* pov, + elfcpp::Elf_types<64>::Elf_Addr got_address, + elfcpp::Elf_types<64>::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index); + + private: + // Set the final size. + void + set_final_data_size(); + + // Write out the BND PLT data. + void + do_write(Output_file*); + + // Offset of the Additional PLT (if using -z bndplt). + unsigned int aplt_offset_; + + // The size of an entry in the PLT. + static const int plt_entry_size = 16; + + // The size of an entry in the additional PLT. + static const int aplt_entry_size = 8; + + // The first entry in the PLT. + // From the AMD64 ABI: "Unlike Intel386 ABI, this ABI uses the same + // procedure linkage table for both programs and shared objects." + static const unsigned char first_plt_entry[plt_entry_size]; + + // Other entries in the PLT for an executable. + static const unsigned char plt_entry[plt_entry_size]; + + // Entries in the additional PLT. + static const unsigned char aplt_entry[aplt_entry_size]; + + // The reserved TLSDESC entry in the PLT for an executable. + static const unsigned char tlsdesc_plt_entry[plt_entry_size]; + + // The .eh_frame unwind information for the PLT. + static const int plt_eh_frame_fde_size = 32; + static const unsigned char plt_eh_frame_fde[plt_eh_frame_fde_size]; +}; + +// We use this PLT when Indirect Branch Tracking (IBT) is enabled. + +template +class Output_data_plt_x86_64_ibt : public Output_data_plt_x86_64 +{ + public: + Output_data_plt_x86_64_ibt(Layout* layout, + Output_data_got<64, false>* got, + 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), + aplt_offset_(0) + { } + + Output_data_plt_x86_64_ibt(Layout* layout, + Output_data_got<64, false>* got, + 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, + got, got_plt, got_irelative, + plt_count), + aplt_offset_(0) + { } + + protected: + virtual unsigned int + do_get_plt_entry_size() const + { return plt_entry_size; } + + // Return the PLT address to use for a global symbol. + uint64_t + do_address_for_global(const Symbol*); + + // Return the PLT address to use for a local symbol. + uint64_t + do_address_for_local(const Relobj*, unsigned int symndx); + + virtual void + do_add_eh_frame(Layout* layout) + { + layout->add_eh_frame_for_plt(this, + this->plt_eh_frame_cie, + this->plt_eh_frame_cie_size, + plt_eh_frame_fde, + plt_eh_frame_fde_size); + } + + virtual void + do_fill_first_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_addr, + typename elfcpp::Elf_types::Elf_Addr plt_addr); + + virtual unsigned int + do_fill_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index); + + virtual void + do_fill_tlsdesc_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + typename elfcpp::Elf_types::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset); + + void + fill_aplt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index); + + private: + // Set the final size. + void + set_final_data_size(); + + // Write out the BND PLT data. + void + do_write(Output_file*); + + // Offset of the Additional PLT (if using -z bndplt). + unsigned int aplt_offset_; + + // The size of an entry in the PLT. + static const int plt_entry_size = 16; + + // The size of an entry in the additional PLT. + static const int aplt_entry_size = 16; + + // The first entry in the PLT. + // From the AMD64 ABI: "Unlike Intel386 ABI, this ABI uses the same + // procedure linkage table for both programs and shared objects." + static const unsigned char first_plt_entry[plt_entry_size]; + + // Other entries in the PLT for an executable. + static const unsigned char plt_entry[plt_entry_size]; + + // Entries in the additional PLT. + static const unsigned char aplt_entry[aplt_entry_size]; + + // The reserved TLSDESC entry in the PLT for an executable. + static const unsigned char tlsdesc_plt_entry[plt_entry_size]; + + // The .eh_frame unwind information for the PLT. + static const int plt_eh_frame_fde_size = 32; + 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 @@ -252,13 +699,15 @@ class Target_x86_64 : public Sized_target // uses only Elf64_Rela relocation entries with explicit addends." typedef Output_data_reloc Reloc_section; - Target_x86_64() - : Sized_target(&x86_64_info), + Target_x86_64(const Target::Target_info* info = &x86_64_info) + : Sized_target(info), 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_(), - tls_base_symbol_defined_(false) + got_mod_index_offset_(-1U), tlsdesc_reloc_info_(), + tls_base_symbol_defined_(false), isa_1_used_(0), isa_1_needed_(0), + feature_1_(0), object_isa_1_used_(0), object_feature_1_(0), + seen_first_object_(false) { } // Hook for a new output section. @@ -268,16 +717,16 @@ class Target_x86_64 : public Sized_target // Scan the relocations to look for symbol adjustments. void gc_process_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); + 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); // Scan the relocations to look for symbol adjustments. void @@ -330,16 +779,30 @@ class Target_x86_64 : public Sized_target const unsigned char* plocal_symbols, Relocatable_relocs*); - // Relocate a section during a relocatable link. + // Scan the relocs for --emit-relocs. void - relocate_for_relocatable( + 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( const Relocate_info*, unsigned int sh_type, 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, @@ -383,10 +846,13 @@ class Target_x86_64 : public Sized_target // and global_reloc_may_be_function_pointer) // if a function's pointer is taken. ICF uses this in safe mode to only // fold those functions whose pointer is defintely not taken. For x86_64 - // pie binaries, safe ICF cannot be done by looking at relocation types. + // pie binaries, safe ICF cannot be done by looking at only relocation + // types, and for certain cases (e.g. R_X86_64_PC32), the instruction + // opcode is checked as well to distinguish a function call from taking + // a function's pointer. bool do_can_check_for_function_pointers() const - { return !parameters->options().pie(); } + { return true; } // Return the base for a DW_EH_PE_datarel encoding. uint64_t @@ -396,6 +862,7 @@ 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; @@ -428,6 +895,11 @@ 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, @@ -439,7 +911,7 @@ class Target_x86_64 : public Sized_target // necessary dynamic relocations. void reserve_local_got_entry(unsigned int got_index, - Sized_relobj* obj, + Sized_relobj* obj, unsigned int r_sym, unsigned int got_type); @@ -477,6 +949,39 @@ class Target_x86_64 : public Sized_target return this->tlsdesc_reloc_info_.size() - 1; } + Output_data_plt_x86_64* + make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + 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); + } + + Output_data_plt_x86_64* + make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative, + unsigned int plt_count) + { + return this->do_make_data_plt(layout, got, got_plt, got_irelative, + plt_count); + } + + virtual Output_data_plt_x86_64* + do_make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative); + + virtual Output_data_plt_x86_64* + do_make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative, + unsigned int plt_count); + private: // The class which scans relocations. class Scan @@ -495,7 +1000,8 @@ 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, @@ -508,22 +1014,22 @@ class Target_x86_64 : public Sized_target inline bool local_reloc_may_be_function_pointer(Symbol_table* symtab, Layout* layout, Target_x86_64* target, - Sized_relobj_file* object, - unsigned int data_shndx, - Output_section* output_section, - const elfcpp::Rela& reloc, + Sized_relobj_file* object, + unsigned int data_shndx, + Output_section* output_section, + const elfcpp::Rela& reloc, unsigned int r_type, - const elfcpp::Sym& lsym); + const elfcpp::Sym& lsym); inline bool global_reloc_may_be_function_pointer(Symbol_table* symtab, Layout* layout, - Target_x86_64* target, - Sized_relobj_file* object, - unsigned int data_shndx, - Output_section* output_section, - const elfcpp::Rela& reloc, + Target_x86_64* target, + Sized_relobj_file* object, + unsigned int data_shndx, + Output_section* output_section, + const elfcpp::Rela& reloc, unsigned int r_type, - Symbol* gsym); + Symbol* gsym); private: static void @@ -538,7 +1044,10 @@ class Target_x86_64 : public Sized_target check_non_pic(Relobj*, unsigned int r_type, Symbol*); inline bool - possible_function_pointer_reloc(unsigned int r_type); + possible_function_pointer_reloc(Sized_relobj_file* src_obj, + unsigned int src_indx, + unsigned int r_offset, + unsigned int r_type); bool reloc_needs_plt_for_ifunc(Sized_relobj_file*, @@ -568,11 +1077,9 @@ 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); @@ -580,7 +1087,7 @@ class Target_x86_64 : public Sized_target // Do a TLS relocation. inline void relocate_tls(const Relocate_info*, Target_x86_64*, - size_t relnum, const elfcpp::Rela&, + size_t relnum, const elfcpp::Rela&, unsigned int r_type, const Sized_symbol*, const Symbol_value*, unsigned char*, typename elfcpp::Elf_types::Elf_Addr, @@ -589,7 +1096,6 @@ 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, @@ -608,7 +1114,6 @@ 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, @@ -647,14 +1152,70 @@ 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. + // We can't do predefined symbols because they may become undefined + // (e.g., __ehdr_start when the headers aren't mapped to a segment). + if (gsym->type() == elfcpp::STT_GNU_IFUNC + || gsym->is_undefined() + || gsym->is_predefined() + || 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. @@ -666,7 +1227,7 @@ 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); @@ -731,19 +1292,37 @@ class Target_x86_64 : public Sized_target // Add a potential copy relocation. void copy_reloc(Symbol_table* symtab, Layout* layout, - Sized_relobj_file* object, + Sized_relobj_file* object, 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 - // general Target structure. - static const Target::Target_info x86_64_info; + // Record a target-specific program property in the .note.gnu.property + // section. + void + record_gnu_property(unsigned int, unsigned int, size_t, + const unsigned char*, const Object*); + + // Merge the target-specific program properties from the current object. + void + merge_gnu_properties(const Object*); + + // Finalize the target-specific program properties and add them back to + // the layout. + void + do_finalize_gnu_properties(Layout*) const; + + // Information about this specific target which we pass to the + // general Target structure. + static const Target::Target_info x86_64_info; // The types of GOT entries needed for this platform. // These values are exposed to the ABI in an incremental link. @@ -777,7 +1356,7 @@ 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. @@ -790,8 +1369,6 @@ 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 @@ -800,6 +1377,22 @@ class Target_x86_64 : public Sized_target std::vector tlsdesc_reloc_info_; // True if the _TLS_MODULE_BASE_ symbol has been defined. bool tls_base_symbol_defined_; + // Target-specific program properties, from .note.gnu.property section. + // Each bit represents a specific feature. + uint32_t isa_1_used_; + uint32_t isa_1_needed_; + uint32_t feature_1_; + // Target-specific properties from the current object. + // These bits get ORed into ISA_1_USED_ after all properties for the object + // have been processed. But if either is all zeroes (as when the property + // is absent from an object), the result should be all zeroes. + // (See PR ld/23486.) + uint32_t object_isa_1_used_; + // These bits get ANDed into FEATURE_1_ after all properties for the object + // have been processed. + uint32_t object_feature_1_; + // Whether we have seen our first object, for use in initializing FEATURE_1_. + bool seen_first_object_; }; template<> @@ -818,12 +1411,17 @@ const Target::Target_info Target_x86_64<64>::x86_64_info = 0x400000, // default_text_segment_address 0x1000, // abi_pagesize (overridable by -z max-page-size) 0x1000, // common_pagesize (overridable by -z common-page-size) + false, // isolate_execinstr + 0, // rosegment_gap elfcpp::SHN_UNDEF, // small_common_shndx elfcpp::SHN_X86_64_LCOMMON, // large_common_shndx 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 + elfcpp::SHT_X86_64_UNWIND, // unwind_section_type }; template<> @@ -842,12 +1440,17 @@ const Target::Target_info Target_x86_64<32>::x86_64_info = 0x400000, // default_text_segment_address 0x1000, // abi_pagesize (overridable by -z max-page-size) 0x1000, // common_pagesize (overridable by -z common-page-size) + false, // isolate_execinstr + 0, // rosegment_gap elfcpp::SHN_UNDEF, // small_common_shndx elfcpp::SHN_X86_64_LCOMMON, // large_common_shndx 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 + elfcpp::SHT_X86_64_UNWIND, // unwind_section_type }; // This is called when a new output section is created. This is where @@ -889,7 +1492,7 @@ 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), @@ -978,6 +1581,124 @@ Target_x86_64::rela_irelative_section(Layout* layout) return this->rela_irelative_; } +// Record a target-specific program property from the .note.gnu.property +// section. +template +void +Target_x86_64::record_gnu_property( + unsigned int, unsigned int pr_type, + size_t pr_datasz, const unsigned char* pr_data, + const Object* object) +{ + uint32_t val = 0; + + switch (pr_type) + { + case elfcpp::GNU_PROPERTY_X86_ISA_1_USED: + case elfcpp::GNU_PROPERTY_X86_ISA_1_NEEDED: + case elfcpp::GNU_PROPERTY_X86_FEATURE_1_AND: + if (pr_datasz != 4) + { + gold_warning(_("%s: corrupt .note.gnu.property section " + "(pr_datasz for property %d is not 4)"), + object->name().c_str(), pr_type); + return; + } + val = elfcpp::Swap<32, false>::readval(pr_data); + break; + default: + gold_warning(_("%s: unknown program property type 0x%x " + "in .note.gnu.property section"), + object->name().c_str(), pr_type); + break; + } + + switch (pr_type) + { + case elfcpp::GNU_PROPERTY_X86_ISA_1_USED: + this->object_isa_1_used_ |= val; + break; + case elfcpp::GNU_PROPERTY_X86_ISA_1_NEEDED: + this->isa_1_needed_ |= val; + break; + case elfcpp::GNU_PROPERTY_X86_FEATURE_1_AND: + // If we see multiple feature props in one object, OR them together. + this->object_feature_1_ |= val; + break; + } +} + +// Merge the target-specific program properties from the current object. +template +void +Target_x86_64::merge_gnu_properties(const Object*) +{ + if (this->seen_first_object_) + { + // If any object is missing the ISA_1_USED property, we must omit + // it from the output file. + if (this->object_isa_1_used_ == 0) + this->isa_1_used_ = 0; + else if (this->isa_1_used_ != 0) + this->isa_1_used_ |= this->object_isa_1_used_; + this->feature_1_ &= this->object_feature_1_; + } + else + { + this->isa_1_used_ = this->object_isa_1_used_; + this->feature_1_ = this->object_feature_1_; + this->seen_first_object_ = true; + } + this->object_isa_1_used_ = 0; + this->object_feature_1_ = 0; +} + +static inline void +add_property(Layout* layout, unsigned int pr_type, uint32_t val) +{ + unsigned char buf[4]; + elfcpp::Swap<32, false>::writeval(buf, val); + layout->add_gnu_property(elfcpp::NT_GNU_PROPERTY_TYPE_0, pr_type, 4, buf); +} + +// Finalize the target-specific program properties and add them back to +// the layout. +template +void +Target_x86_64::do_finalize_gnu_properties(Layout* layout) const +{ + if (this->isa_1_used_ != 0) + add_property(layout, elfcpp::GNU_PROPERTY_X86_ISA_1_USED, + this->isa_1_used_); + if (this->isa_1_needed_ != 0) + add_property(layout, elfcpp::GNU_PROPERTY_X86_ISA_1_NEEDED, + this->isa_1_needed_); + if (this->feature_1_ != 0) + add_property(layout, elfcpp::GNU_PROPERTY_X86_FEATURE_1_AND, + this->feature_1_); +} + +// 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 @@ -988,18 +1709,13 @@ Output_data_plt_x86_64::init(Layout* layout) layout->add_output_section_data(".rela.plt", elfcpp::SHT_RELA, elfcpp::SHF_ALLOC, this->rel_, ORDER_DYNAMIC_PLT_RELOCS, false); - - // Add unwind information if requested. - if (parameters->options().ld_generated_unwind_info()) - layout->add_eh_frame_for_plt(this, plt_eh_frame_cie, plt_eh_frame_cie_size, - plt_eh_frame_fde, plt_eh_frame_fde_size); } template void Output_data_plt_x86_64::do_adjust_output_section(Output_section* os) { - os->set_entsize(plt_entry_size); + os->set_entsize(this->get_plt_entry_size()); } // Add an entry to the PLT. @@ -1018,7 +1734,7 @@ 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)) { @@ -1040,7 +1756,7 @@ Output_data_plt_x86_64::add_entry(Symbol_table* symtab, Layout* layout, // Note that when setting the PLT offset for a non-IRELATIVE // entry we skip the initial reserved PLT entry. plt_index = *pcount + offset; - plt_offset = plt_index * plt_entry_size; + plt_offset = plt_index * this->get_plt_entry_size(); ++*pcount; @@ -1057,7 +1773,8 @@ Output_data_plt_x86_64::add_entry(Symbol_table* symtab, Layout* layout, // FIXME: This is probably not correct for IRELATIVE relocs. // For incremental updates, find an available slot. - plt_offset = this->free_list_.allocate(plt_entry_size, plt_entry_size, 0); + plt_offset = this->free_list_.allocate(this->get_plt_entry_size(), + this->get_plt_entry_size(), 0); if (plt_offset == -1) gold_fallback(_("out of patch space (PLT);" " relink with --incremental-full")); @@ -1065,7 +1782,7 @@ Output_data_plt_x86_64::add_entry(Symbol_table* symtab, Layout* layout, // The GOT and PLT entries have a 1-1 correspondance, so the GOT offset // can be calculated from the PLT index, adjusting for the three // reserved entries at the beginning of the GOT. - plt_index = plt_offset / plt_entry_size - 1; + plt_index = plt_offset / this->get_plt_entry_size() - 1; got_offset = (plt_index - offset + reserved) * 8; } @@ -1090,7 +1807,7 @@ Output_data_plt_x86_64::add_local_ifunc_entry( Sized_relobj_file* relobj, unsigned int local_sym_index) { - unsigned int plt_offset = this->irelative_count_ * plt_entry_size; + unsigned int plt_offset = this->irelative_count_ * this->get_plt_entry_size(); ++this->irelative_count_; section_offset_type got_offset = this->got_irelative_->current_data_size(); @@ -1197,13 +1914,13 @@ Output_data_plt_x86_64::rela_irelative(Symbol_table* symtab, template uint64_t -Output_data_plt_x86_64::address_for_global(const Symbol* gsym) +Output_data_plt_x86_64::do_address_for_global(const Symbol* gsym) { uint64_t offset = 0; if (gsym->type() == elfcpp::STT_GNU_IFUNC && gsym->can_use_relative_reloc(false)) - offset = (this->count_ + 1) * plt_entry_size; - return this->address() + offset; + offset = (this->count_ + 1) * this->get_plt_entry_size(); + return this->address() + offset + gsym->plt_offset(); } // Return the PLT address to use for a local symbol. These are always @@ -1211,9 +1928,12 @@ 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::do_address_for_local(const Relobj* object, + unsigned int r_sym) { - return this->address() + (this->count_ + 1) * plt_entry_size; + return (this->address() + + (this->count_ + 1) * this->get_plt_entry_size() + + object->local_plt_offset(r_sym)); } // Set the final size. @@ -1221,17 +1941,19 @@ template void Output_data_plt_x86_64::set_final_data_size() { - unsigned int count = this->count_ + this->irelative_count_; + // Number of regular and IFUNC PLT entries, plus the first entry. + unsigned int count = this->count_ + this->irelative_count_ + 1; + // Count the TLSDESC entry, if present. if (this->has_tlsdesc_entry()) ++count; - this->set_data_size((count + 1) * plt_entry_size); + this->set_data_size(count * this->get_plt_entry_size()); } // The first entry in the PLT for an executable. template const unsigned char -Output_data_plt_x86_64::first_plt_entry[plt_entry_size] = +Output_data_plt_x86_64_standard::first_plt_entry[plt_entry_size] = { // From AMD64 ABI Draft 0.98, page 76 0xff, 0x35, // pushq contents of memory address @@ -1241,11 +1963,28 @@ Output_data_plt_x86_64::first_plt_entry[plt_entry_size] = 0x90, 0x90, 0x90, 0x90 // noop (x4) }; +template +void +Output_data_plt_x86_64_standard::do_fill_first_plt_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address) +{ + memcpy(pov, first_plt_entry, plt_entry_size); + // We do a jmp relative to the PC at the end of this instruction. + elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, + (got_address + 8 + - (plt_address + 6))); + elfcpp::Swap<32, false>::writeval(pov + 8, + (got_address + 16 + - (plt_address + 12))); +} + // Subsequent entries in the PLT for an executable. template const unsigned char -Output_data_plt_x86_64::plt_entry[plt_entry_size] = +Output_data_plt_x86_64_standard::plt_entry[plt_entry_size] = { // From AMD64 ABI Draft 0.98, page 76 0xff, 0x25, // jmpq indirect @@ -1256,11 +1995,39 @@ Output_data_plt_x86_64::plt_entry[plt_entry_size] = 0, 0, 0, 0 // replaced with offset to start of .plt }; +template +unsigned int +Output_data_plt_x86_64_standard::do_fill_plt_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + 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, + plt_got_pcrel_offset); + + elfcpp::Swap_unaligned<32, false>::writeval(pov + 7, plt_index); + elfcpp::Swap<32, false>::writeval(pov + 12, + - (plt_offset + plt_entry_size)); + + return 6; +} + // The reserved TLSDESC entry in the PLT for an executable. template const unsigned char -Output_data_plt_x86_64::tlsdesc_plt_entry[plt_entry_size] = +Output_data_plt_x86_64_standard::tlsdesc_plt_entry[plt_entry_size] = { // From Alexandre Oliva, "Thread-Local Storage Descriptors for IA32 // and AMD64/EM64T", Version 0.9.4 (2005-10-10). @@ -1272,10 +2039,440 @@ Output_data_plt_x86_64::tlsdesc_plt_entry[plt_entry_size] = 0x40, 0 }; +template +void +Output_data_plt_x86_64_standard::do_fill_tlsdesc_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + typename elfcpp::Elf_types::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset) +{ + memcpy(pov, tlsdesc_plt_entry, plt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, + (got_address + 8 + - (plt_address + plt_offset + + 6))); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 8, + (got_base + + tlsdesc_got_offset + - (plt_address + plt_offset + + 12))); +} + +// Return the APLT address to use for a global symbol (for -z bndplt). + +uint64_t +Output_data_plt_x86_64_bnd::do_address_for_global(const Symbol* gsym) +{ + uint64_t offset = this->aplt_offset_; + // Convert the PLT offset into an APLT offset. + unsigned int plt_offset = gsym->plt_offset(); + if (gsym->type() == elfcpp::STT_GNU_IFUNC + && gsym->can_use_relative_reloc(false)) + offset += this->regular_count() * aplt_entry_size; + else + plt_offset -= plt_entry_size; + plt_offset = plt_offset / (plt_entry_size / aplt_entry_size); + return this->address() + offset + plt_offset; +} + +// Return the PLT address to use for a local symbol. These are always +// IRELATIVE relocs. + +uint64_t +Output_data_plt_x86_64_bnd::do_address_for_local(const Relobj* object, + unsigned int r_sym) +{ + // Convert the PLT offset into an APLT offset. + unsigned int plt_offset = ((object->local_plt_offset(r_sym) - plt_entry_size) + / (plt_entry_size / aplt_entry_size)); + return (this->address() + + this->aplt_offset_ + + this->regular_count() * aplt_entry_size + + plt_offset); +} + +// Set the final size. +void +Output_data_plt_x86_64_bnd::set_final_data_size() +{ + // Number of regular and IFUNC PLT entries. + unsigned int count = this->entry_count(); + // Count the first entry and the TLSDESC entry, if present. + unsigned int extra = this->has_tlsdesc_entry() ? 2 : 1; + unsigned int plt_size = (count + extra) * plt_entry_size; + // Offset of the APLT. + this->aplt_offset_ = plt_size; + // Size of the APLT. + plt_size += count * aplt_entry_size; + this->set_data_size(plt_size); +} + +// The first entry in the BND PLT. + +const unsigned char +Output_data_plt_x86_64_bnd::first_plt_entry[plt_entry_size] = +{ + // From AMD64 ABI Draft 0.98, page 76 + 0xff, 0x35, // pushq contents of memory address + 0, 0, 0, 0, // replaced with address of .got + 8 + 0xf2, 0xff, 0x25, // bnd jmp indirect + 0, 0, 0, 0, // replaced with address of .got + 16 + 0x0f, 0x1f, 0x00 // nop +}; + +void +Output_data_plt_x86_64_bnd::do_fill_first_plt_entry( + unsigned char* pov, + elfcpp::Elf_types<64>::Elf_Addr got_address, + elfcpp::Elf_types<64>::Elf_Addr plt_address) +{ + memcpy(pov, first_plt_entry, plt_entry_size); + // We do a jmp relative to the PC at the end of this instruction. + elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, + (got_address + 8 + - (plt_address + 6))); + elfcpp::Swap<32, false>::writeval(pov + 9, + (got_address + 16 + - (plt_address + 13))); +} + +// Subsequent entries in the BND PLT. + +const unsigned char +Output_data_plt_x86_64_bnd::plt_entry[plt_entry_size] = +{ + // From AMD64 ABI Draft 0.99.8, page 139 + 0x68, // pushq immediate + 0, 0, 0, 0, // replaced with offset into relocation table + 0xf2, 0xe9, // bnd jmpq relative + 0, 0, 0, 0, // replaced with offset to start of .plt + 0x0f, 0x1f, 0x44, 0, 0 // nop +}; + +// Entries in the BND Additional PLT. + +const unsigned char +Output_data_plt_x86_64_bnd::aplt_entry[aplt_entry_size] = +{ + // From AMD64 ABI Draft 0.99.8, page 139 + 0xf2, 0xff, 0x25, // bnd jmpq indirect + 0, 0, 0, 0, // replaced with address of symbol in .got + 0x90, // nop +}; + +unsigned int +Output_data_plt_x86_64_bnd::do_fill_plt_entry( + unsigned char* pov, + elfcpp::Elf_types<64>::Elf_Addr, + elfcpp::Elf_types<64>::Elf_Addr, + unsigned int, + unsigned int plt_offset, + unsigned int plt_index) +{ + memcpy(pov, plt_entry, plt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 1, plt_index); + elfcpp::Swap<32, false>::writeval(pov + 7, -(plt_offset + 11)); + return 0; +} + +void +Output_data_plt_x86_64_bnd::fill_aplt_entry( + unsigned char* pov, + elfcpp::Elf_types<64>::Elf_Addr got_address, + elfcpp::Elf_types<64>::Elf_Addr plt_address, + unsigned int got_offset, + 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 + 7)); + if (Bits<32>::has_overflow(plt_got_pcrel_offset)) + gold_error(_("PC-relative offset overflow in APLT entry %d"), + plt_index + 1); + + memcpy(pov, aplt_entry, aplt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 3, plt_got_pcrel_offset); +} + +// The reserved TLSDESC entry in the PLT for an executable. + +const unsigned char +Output_data_plt_x86_64_bnd::tlsdesc_plt_entry[plt_entry_size] = +{ + // From Alexandre Oliva, "Thread-Local Storage Descriptors for IA32 + // and AMD64/EM64T", Version 0.9.4 (2005-10-10). + 0xff, 0x35, // pushq x(%rip) + 0, 0, 0, 0, // replaced with address of linkmap GOT entry (at PLTGOT + 8) + 0xf2, 0xff, 0x25, // jmpq *y(%rip) + 0, 0, 0, 0, // replaced with offset of reserved TLSDESC_GOT entry + 0x0f, 0x1f, 0 // nop +}; + +void +Output_data_plt_x86_64_bnd::do_fill_tlsdesc_entry( + unsigned char* pov, + elfcpp::Elf_types<64>::Elf_Addr got_address, + elfcpp::Elf_types<64>::Elf_Addr plt_address, + elfcpp::Elf_types<64>::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset) +{ + memcpy(pov, tlsdesc_plt_entry, plt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, + (got_address + 8 + - (plt_address + plt_offset + + 6))); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 9, + (got_base + + tlsdesc_got_offset + - (plt_address + plt_offset + + 13))); +} + +// Return the APLT address to use for a global symbol (for IBT). + +template +uint64_t +Output_data_plt_x86_64_ibt::do_address_for_global(const Symbol* gsym) +{ + uint64_t offset = this->aplt_offset_; + // Convert the PLT offset into an APLT offset. + unsigned int plt_offset = gsym->plt_offset(); + if (gsym->type() == elfcpp::STT_GNU_IFUNC + && gsym->can_use_relative_reloc(false)) + offset += this->regular_count() * aplt_entry_size; + else + plt_offset -= plt_entry_size; + plt_offset = plt_offset / (plt_entry_size / aplt_entry_size); + return this->address() + offset + plt_offset; +} + +// Return the PLT address to use for a local symbol. These are always +// IRELATIVE relocs. + +template +uint64_t +Output_data_plt_x86_64_ibt::do_address_for_local(const Relobj* object, + unsigned int r_sym) +{ + // Convert the PLT offset into an APLT offset. + unsigned int plt_offset = ((object->local_plt_offset(r_sym) - plt_entry_size) + / (plt_entry_size / aplt_entry_size)); + return (this->address() + + this->aplt_offset_ + + this->regular_count() * aplt_entry_size + + plt_offset); +} + +// Set the final size. + +template +void +Output_data_plt_x86_64_ibt::set_final_data_size() +{ + // Number of regular and IFUNC PLT entries. + unsigned int count = this->entry_count(); + // Count the first entry and the TLSDESC entry, if present. + unsigned int extra = this->has_tlsdesc_entry() ? 2 : 1; + unsigned int plt_size = (count + extra) * plt_entry_size; + // Offset of the APLT. + this->aplt_offset_ = plt_size; + // Size of the APLT. + plt_size += count * aplt_entry_size; + this->set_data_size(plt_size); +} + +// The first entry in the IBT PLT. + +template<> +const unsigned char +Output_data_plt_x86_64_ibt<32>::first_plt_entry[plt_entry_size] = +{ + // MPX isn't supported for x32, so we don't need the BND prefix. + // From AMD64 ABI Draft 0.98, page 76 + 0xff, 0x35, // pushq contents of memory address + 0, 0, 0, 0, // replaced with address of .got + 8 + 0xff, 0x25, // jmp indirect + 0, 0, 0, 0, // replaced with address of .got + 16 + 0x90, 0x90, 0x90, 0x90 // noop (x4) +}; + +template<> +const unsigned char +Output_data_plt_x86_64_ibt<64>::first_plt_entry[plt_entry_size] = +{ + // Use the BND prefix so that IBT is compatible with MPX. + 0xff, 0x35, // pushq contents of memory address + 0, 0, 0, 0, // replaced with address of .got + 8 + 0xf2, 0xff, 0x25, // bnd jmp indirect + 0, 0, 0, 0, // replaced with address of .got + 16 + 0x0f, 0x1f, 0x00 // nop +}; + +template +void +Output_data_plt_x86_64_ibt::do_fill_first_plt_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address) +{ + // Offsets to the addresses needing relocation. + const unsigned int roff1 = 2; + const unsigned int roff2 = (size == 32) ? 8 : 9; + + memcpy(pov, first_plt_entry, plt_entry_size); + // We do a jmp relative to the PC at the end of this instruction. + elfcpp::Swap_unaligned<32, false>::writeval(pov + roff1, + (got_address + 8 + - (plt_address + roff1 + 4))); + elfcpp::Swap<32, false>::writeval(pov + roff2, + (got_address + 16 + - (plt_address + roff2 + 4))); +} + +// Subsequent entries in the IBT PLT. + +template<> +const unsigned char +Output_data_plt_x86_64_ibt<32>::plt_entry[plt_entry_size] = +{ + // From AMD64 ABI Draft 1.0-rc1, Chapter 13. + 0xf3, 0x0f, 0x1e, 0xfa, // endbr64 + 0x68, // pushq immediate + 0, 0, 0, 0, // replaced with offset into relocation table + 0xe9, // jmpq relative + 0, 0, 0, 0, // replaced with offset to start of .plt + 0x90, 0x90 // nop +}; + +template<> +const unsigned char +Output_data_plt_x86_64_ibt<64>::plt_entry[plt_entry_size] = +{ + // From AMD64 ABI Draft 1.0-rc1, Chapter 13. + 0xf3, 0x0f, 0x1e, 0xfa, // endbr64 + 0x68, // pushq immediate + 0, 0, 0, 0, // replaced with offset into relocation table + 0xf2, 0xe9, // bnd jmpq relative + 0, 0, 0, 0, // replaced with offset to start of .plt + 0x90 // nop +}; + +// Entries in the IBT Additional PLT. + +template<> +const unsigned char +Output_data_plt_x86_64_ibt<32>::aplt_entry[aplt_entry_size] = +{ + // From AMD64 ABI Draft 1.0-rc1, Chapter 13. + 0xf3, 0x0f, 0x1e, 0xfa, // endbr64 + 0xff, 0x25, // jmpq indirect + 0, 0, 0, 0, // replaced with address of symbol in .got + 0x0f, 0x1f, 0x04, 0x00, // nop + 0x90, 0x90 // nop +}; + +template<> +const unsigned char +Output_data_plt_x86_64_ibt<64>::aplt_entry[aplt_entry_size] = +{ + // From AMD64 ABI Draft 1.0-rc1, Chapter 13. + 0xf3, 0x0f, 0x1e, 0xfa, // endbr64 + 0xf2, 0xff, 0x25, // bnd jmpq indirect + 0, 0, 0, 0, // replaced with address of symbol in .got + 0x0f, 0x1f, 0x04, 0x00, // nop + 0x90, // nop +}; + +template +unsigned int +Output_data_plt_x86_64_ibt::do_fill_plt_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr, + typename elfcpp::Elf_types::Elf_Addr, + unsigned int, + unsigned int plt_offset, + unsigned int plt_index) +{ + // Offsets to the addresses needing relocation. + const unsigned int roff1 = 5; + const unsigned int roff2 = (size == 32) ? 10 : 11; + + memcpy(pov, plt_entry, plt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + roff1, plt_index); + elfcpp::Swap<32, false>::writeval(pov + roff2, -(plt_offset + roff2 + 4)); + return 0; +} + +template +void +Output_data_plt_x86_64_ibt::fill_aplt_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index) +{ + // Offset to the address needing relocation. + const unsigned int roff = (size == 32) ? 6 : 7; + + // Check PC-relative offset overflow in PLT entry. + uint64_t plt_got_pcrel_offset = (got_address + got_offset + - (plt_address + plt_offset + roff + 4)); + if (Bits<32>::has_overflow(plt_got_pcrel_offset)) + gold_error(_("PC-relative offset overflow in APLT entry %d"), + plt_index + 1); + + memcpy(pov, aplt_entry, aplt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + roff, plt_got_pcrel_offset); +} + +// The reserved TLSDESC entry in the IBT PLT for an executable. + +template +const unsigned char +Output_data_plt_x86_64_ibt::tlsdesc_plt_entry[plt_entry_size] = +{ + // From Alexandre Oliva, "Thread-Local Storage Descriptors for IA32 + // and AMD64/EM64T", Version 0.9.4 (2005-10-10). + 0xff, 0x35, // pushq x(%rip) + 0, 0, 0, 0, // replaced with address of linkmap GOT entry (at PLTGOT + 8) + 0xf2, 0xff, 0x25, // jmpq *y(%rip) + 0, 0, 0, 0, // replaced with offset of reserved TLSDESC_GOT entry + 0x0f, 0x1f, 0 // nop +}; + +template +void +Output_data_plt_x86_64_ibt::do_fill_tlsdesc_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + typename elfcpp::Elf_types::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset) +{ + memcpy(pov, tlsdesc_plt_entry, plt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, + (got_address + 8 + - (plt_address + plt_offset + + 6))); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 9, + (got_base + + tlsdesc_got_offset + - (plt_address + plt_offset + + 13))); +} + // The .eh_frame unwind information for the PLT. template -const unsigned char +const unsigned char Output_data_plt_x86_64::plt_eh_frame_cie[plt_eh_frame_cie_size] = { 1, // CIE version. @@ -1296,7 +2493,7 @@ Output_data_plt_x86_64::plt_eh_frame_cie[plt_eh_frame_cie_size] = template const unsigned char -Output_data_plt_x86_64::plt_eh_frame_fde[plt_eh_frame_fde_size] = +Output_data_plt_x86_64_standard::plt_eh_frame_fde[plt_eh_frame_fde_size] = { 0, 0, 0, 0, // Replaced with offset to .plt. 0, 0, 0, 0, // Replaced with size of .plt. @@ -1322,9 +2519,66 @@ Output_data_plt_x86_64::plt_eh_frame_fde[plt_eh_frame_fde_size] = elfcpp::DW_CFA_nop }; -// Write out the PLT. This uses the hand-coded instructions above, -// and adjusts them as needed. This is specified by the AMD64 ABI. - +// The .eh_frame unwind information for the BND PLT. +const unsigned char +Output_data_plt_x86_64_bnd::plt_eh_frame_fde[plt_eh_frame_fde_size] = +{ + 0, 0, 0, 0, // Replaced with offset to .plt. + 0, 0, 0, 0, // Replaced with size of .plt. + 0, // Augmentation size. + elfcpp::DW_CFA_def_cfa_offset, 16, // DW_CFA_def_cfa_offset: 16. + elfcpp::DW_CFA_advance_loc + 6, // Advance 6 to __PLT__ + 6. + elfcpp::DW_CFA_def_cfa_offset, 24, // DW_CFA_def_cfa_offset: 24. + elfcpp::DW_CFA_advance_loc + 10, // Advance 10 to __PLT__ + 16. + elfcpp::DW_CFA_def_cfa_expression, // DW_CFA_def_cfa_expression. + 11, // Block length. + elfcpp::DW_OP_breg7, 8, // Push %rsp + 8. + elfcpp::DW_OP_breg16, 0, // Push %rip. + elfcpp::DW_OP_lit15, // Push 0xf. + elfcpp::DW_OP_and, // & (%rip & 0xf). + elfcpp::DW_OP_lit5, // Push 5. + elfcpp::DW_OP_ge, // >= ((%rip & 0xf) >= 5) + elfcpp::DW_OP_lit3, // Push 3. + elfcpp::DW_OP_shl, // << (((%rip & 0xf) >= 5) << 3) + elfcpp::DW_OP_plus, // + ((((%rip&0xf)>=5)<<3)+%rsp+8 + elfcpp::DW_CFA_nop, // Align to 32 bytes. + elfcpp::DW_CFA_nop, + elfcpp::DW_CFA_nop, + elfcpp::DW_CFA_nop +}; + +// The .eh_frame unwind information for the BND PLT. +template +const unsigned char +Output_data_plt_x86_64_ibt::plt_eh_frame_fde[plt_eh_frame_fde_size] = +{ + 0, 0, 0, 0, // Replaced with offset to .plt. + 0, 0, 0, 0, // Replaced with size of .plt. + 0, // Augmentation size. + elfcpp::DW_CFA_def_cfa_offset, 16, // DW_CFA_def_cfa_offset: 16. + elfcpp::DW_CFA_advance_loc + 6, // Advance 6 to __PLT__ + 6. + elfcpp::DW_CFA_def_cfa_offset, 24, // DW_CFA_def_cfa_offset: 24. + elfcpp::DW_CFA_advance_loc + 10, // Advance 10 to __PLT__ + 16. + elfcpp::DW_CFA_def_cfa_expression, // DW_CFA_def_cfa_expression. + 11, // Block length. + elfcpp::DW_OP_breg7, 8, // Push %rsp + 8. + elfcpp::DW_OP_breg16, 0, // Push %rip. + elfcpp::DW_OP_lit15, // Push 0xf. + elfcpp::DW_OP_and, // & (%rip & 0xf). + elfcpp::DW_OP_lit9, // Push 9. + elfcpp::DW_OP_ge, // >= ((%rip & 0xf) >= 9) + elfcpp::DW_OP_lit3, // Push 3. + elfcpp::DW_OP_shl, // << (((%rip & 0xf) >= 9) << 3) + elfcpp::DW_OP_plus, // + ((((%rip&0xf)>=9)<<3)+%rsp+8 + elfcpp::DW_CFA_nop, // Align to 32 bytes. + elfcpp::DW_CFA_nop, + elfcpp::DW_CFA_nop, + elfcpp::DW_CFA_nop +}; + +// Write out the PLT. This uses the hand-coded instructions above, +// and adjusts them as needed. This is specified by the AMD64 ABI. + template void Output_data_plt_x86_64::do_write(Output_file* of) @@ -1356,32 +2610,96 @@ Output_data_plt_x86_64::do_write(Output_file* of) typename elfcpp::Elf_types::Elf_Addr got_address = this->got_plt_->address(); - memcpy(pov, first_plt_entry, plt_entry_size); - // We do a jmp relative to the PC at the end of this instruction. - elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, - (got_address + 8 - - (plt_address + 6))); - elfcpp::Swap<32, false>::writeval(pov + 8, - (got_address + 16 - - (plt_address + 12))); - pov += plt_entry_size; + this->fill_first_plt_entry(pov, got_address, plt_address); + pov += this->get_plt_entry_size(); - unsigned char* got_pov = got_view; + // 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; - // 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; + unsigned int plt_offset = this->get_plt_entry_size(); + unsigned int got_offset = 24; + const unsigned int count = this->count_ + this->irelative_count_; + for (unsigned int plt_index = 0; + plt_index < count; + ++plt_index, + pov += this->get_plt_entry_size(), + got_pov += 8, + plt_offset += this->get_plt_entry_size(), + got_offset += 8) + { + // Set and adjust the PLT entry itself. + unsigned int lazy_offset = this->fill_plt_entry(pov, + got_address, plt_address, + got_offset, plt_offset, + plt_index); + + // Set the entry in the GOT. + elfcpp::Swap<64, false>::writeval(got_pov, + plt_address + plt_offset + lazy_offset); + } + + if (this->has_tlsdesc_entry()) + { + // Set and adjust the reserved TLSDESC PLT entry. + unsigned int tlsdesc_got_offset = this->get_tlsdesc_got_offset(); + this->fill_tlsdesc_entry(pov, got_address, plt_address, got_base, + tlsdesc_got_offset, plt_offset); + pov += this->get_plt_entry_size(); + } + + gold_assert(static_cast(pov - oview) == oview_size); + gold_assert(static_cast(got_pov - got_view) == got_size); + + of->write_output_view(offset, oview_size, oview); + of->write_output_view(got_file_offset, got_size, got_view); +} + +// Write out the BND PLT. + +void +Output_data_plt_x86_64_bnd::do_write(Output_file* of) +{ + const off_t offset = this->offset(); + const section_size_type oview_size = + convert_to_section_size_type(this->data_size()); + unsigned char* const oview = of->get_output_view(offset, oview_size); + + Output_data_got<64, false>* got = this->got(); + Output_data_got_plt_x86_64* got_plt = this->got_plt(); + Output_data_space* got_irelative = this->got_irelative(); + + const off_t got_file_offset = got_plt->offset(); + gold_assert(parameters->incremental_update() + || (got_file_offset + got_plt->data_size() + == got_irelative->offset())); + const section_size_type got_size = + convert_to_section_size_type(got_plt->data_size() + + got_irelative->data_size()); + unsigned char* const got_view = of->get_output_view(got_file_offset, + got_size); + + unsigned char* pov = oview; + + // The base address of the .plt section. + elfcpp::Elf_types<64>::Elf_Addr plt_address = this->address(); + // The base address of the .got section. + elfcpp::Elf_types<64>::Elf_Addr got_base = got->address(); + // The base address of the PLT portion of the .got section, + // which is where the GOT pointer will point, and where the + // three reserved GOT entries are located. + elfcpp::Elf_types<64>::Elf_Addr got_address = got_plt->address(); + + this->fill_first_plt_entry(pov, got_address, plt_address); + pov += plt_entry_size; + + // 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 = plt_entry_size; unsigned int got_offset = 24; - const unsigned int count = this->count_ + this->irelative_count_; + const unsigned int count = this->entry_count(); for (unsigned int plt_index = 0; plt_index < count; ++plt_index, @@ -1391,35 +2709,132 @@ Output_data_plt_x86_64::do_write(Output_file* of) got_offset += 8) { // Set and adjust the PLT entry itself. - memcpy(pov, plt_entry, plt_entry_size); - elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, - (got_address + got_offset - - (plt_address + plt_offset - + 6))); + unsigned int lazy_offset = this->fill_plt_entry(pov, + got_address, plt_address, + got_offset, plt_offset, + plt_index); + + // Set the entry in the GOT. + elfcpp::Swap<64, false>::writeval(got_pov, + plt_address + plt_offset + lazy_offset); + } + + if (this->has_tlsdesc_entry()) + { + // Set and adjust the reserved TLSDESC PLT entry. + unsigned int tlsdesc_got_offset = this->get_tlsdesc_got_offset(); + this->fill_tlsdesc_entry(pov, got_address, plt_address, got_base, + tlsdesc_got_offset, plt_offset); + pov += this->get_plt_entry_size(); + } + + // Write the additional PLT. + got_offset = 24; + for (unsigned int plt_index = 0; + plt_index < count; + ++plt_index, + pov += aplt_entry_size, + plt_offset += aplt_entry_size, + got_offset += 8) + { + // Set and adjust the APLT entry. + this->fill_aplt_entry(pov, got_address, plt_address, got_offset, + plt_offset, plt_index); + } + + gold_assert(static_cast(pov - oview) == oview_size); + gold_assert(static_cast(got_pov - got_view) == got_size); + + of->write_output_view(offset, oview_size, oview); + of->write_output_view(got_file_offset, got_size, got_view); +} + +// Write out the IBT PLT. + +template +void +Output_data_plt_x86_64_ibt::do_write(Output_file* of) +{ + const off_t offset = this->offset(); + const section_size_type oview_size = + convert_to_section_size_type(this->data_size()); + unsigned char* const oview = of->get_output_view(offset, oview_size); + + Output_data_got<64, false>* got = this->got(); + Output_data_got_plt_x86_64* got_plt = this->got_plt(); + Output_data_space* got_irelative = this->got_irelative(); + + const off_t got_file_offset = got_plt->offset(); + gold_assert(parameters->incremental_update() + || (got_file_offset + got_plt->data_size() + == got_irelative->offset())); + const section_size_type got_size = + convert_to_section_size_type(got_plt->data_size() + + got_irelative->data_size()); + unsigned char* const got_view = of->get_output_view(got_file_offset, + got_size); + + unsigned char* pov = oview; + + // The base address of the .plt section. + elfcpp::Elf_types<64>::Elf_Addr plt_address = this->address(); + // The base address of the .got section. + elfcpp::Elf_types<64>::Elf_Addr got_base = got->address(); + // The base address of the PLT portion of the .got section, + // which is where the GOT pointer will point, and where the + // three reserved GOT entries are located. + elfcpp::Elf_types<64>::Elf_Addr got_address = got_plt->address(); + + this->fill_first_plt_entry(pov, got_address, plt_address); + pov += plt_entry_size; + + // 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; - elfcpp::Swap_unaligned<32, false>::writeval(pov + 7, plt_index); - elfcpp::Swap<32, false>::writeval(pov + 12, - - (plt_offset + plt_entry_size)); + unsigned int plt_offset = plt_entry_size; + unsigned int got_offset = 24; + const unsigned int count = this->entry_count(); + for (unsigned int plt_index = 0; + plt_index < count; + ++plt_index, + pov += plt_entry_size, + got_pov += 8, + plt_offset += plt_entry_size, + got_offset += 8) + { + // Set and adjust the PLT entry itself. + unsigned int lazy_offset = this->fill_plt_entry(pov, + got_address, plt_address, + got_offset, plt_offset, + plt_index); // Set the entry in the GOT. - elfcpp::Swap<64, false>::writeval(got_pov, plt_address + plt_offset + 6); + elfcpp::Swap<64, false>::writeval(got_pov, + plt_address + plt_offset + lazy_offset); } if (this->has_tlsdesc_entry()) { // Set and adjust the reserved TLSDESC PLT entry. unsigned int tlsdesc_got_offset = this->get_tlsdesc_got_offset(); - memcpy(pov, tlsdesc_plt_entry, plt_entry_size); - elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, - (got_address + 8 - - (plt_address + plt_offset - + 6))); - elfcpp::Swap_unaligned<32, false>::writeval(pov + 8, - (got_base - + tlsdesc_got_offset - - (plt_address + plt_offset - + 12))); - pov += plt_entry_size; + this->fill_tlsdesc_entry(pov, got_address, plt_address, got_base, + tlsdesc_got_offset, plt_offset); + pov += this->get_plt_entry_size(); + } + + // Write the additional PLT. + got_offset = 24; + for (unsigned int plt_index = 0; + plt_index < count; + ++plt_index, + pov += aplt_entry_size, + plt_offset += aplt_entry_size, + got_offset += 8) + { + // Set and adjust the APLT entry. + this->fill_aplt_entry(pov, got_address, plt_address, got_offset, + plt_offset, plt_index); } gold_assert(static_cast(pov - oview) == oview_size); @@ -1440,9 +2855,13 @@ Target_x86_64::make_plt_section(Symbol_table* symtab, Layout* layout) // Create the GOT sections first. this->got_section(symtab, layout); - this->plt_ = new Output_data_plt_x86_64(layout, this->got_, - this->got_plt_, - this->got_irelative_); + this->plt_ = this->make_data_plt(layout, this->got_, this->got_plt_, + this->got_irelative_); + + // Add unwind information if requested. + if (parameters->options().ld_generated_unwind_info()) + this->plt_->add_eh_frame(layout); + layout->add_output_section_data(".plt", elfcpp::SHT_PROGBITS, (elfcpp::SHF_ALLOC | elfcpp::SHF_EXECINSTR), @@ -1454,6 +2873,73 @@ Target_x86_64::make_plt_section(Symbol_table* symtab, Layout* layout) } } +template<> +Output_data_plt_x86_64<32>* +Target_x86_64<32>::do_make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative) +{ + if (this->feature_1_ & elfcpp::GNU_PROPERTY_X86_FEATURE_1_IBT) + return new Output_data_plt_x86_64_ibt<32>(layout, got, got_plt, + got_irelative); + return new Output_data_plt_x86_64_standard<32>(layout, got, got_plt, + got_irelative); +} + +template<> +Output_data_plt_x86_64<64>* +Target_x86_64<64>::do_make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative) +{ + if (this->feature_1_ & elfcpp::GNU_PROPERTY_X86_FEATURE_1_IBT) + return new Output_data_plt_x86_64_ibt<64>(layout, got, got_plt, + got_irelative); + else if (parameters->options().bndplt()) + return new Output_data_plt_x86_64_bnd(layout, got, got_plt, + got_irelative); + else + return new Output_data_plt_x86_64_standard<64>(layout, got, got_plt, + got_irelative); +} + +template<> +Output_data_plt_x86_64<32>* +Target_x86_64<32>::do_make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative, + unsigned int plt_count) +{ + if (this->feature_1_ & elfcpp::GNU_PROPERTY_X86_FEATURE_1_IBT) + return new Output_data_plt_x86_64_ibt<32>(layout, got, got_plt, + got_irelative, plt_count); + return new Output_data_plt_x86_64_standard<32>(layout, got, got_plt, + got_irelative, plt_count); +} + +template<> +Output_data_plt_x86_64<64>* +Target_x86_64<64>::do_make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative, + unsigned int plt_count) +{ + if (this->feature_1_ & elfcpp::GNU_PROPERTY_X86_FEATURE_1_IBT) + return new Output_data_plt_x86_64_ibt<64>(layout, got, got_plt, + got_irelative, plt_count); + else if (parameters->options().bndplt()) + return new Output_data_plt_x86_64_bnd(layout, got, got_plt, + got_irelative, plt_count); + else + return new Output_data_plt_x86_64_standard<64>(layout, got, got_plt, + got_irelative, + plt_count); +} + // Return the section for TLSDESC relocations. template @@ -1515,7 +3001,9 @@ template unsigned int Target_x86_64::first_plt_entry_offset() const { - return Output_data_plt_x86_64::first_plt_entry_offset(); + if (this->plt_ == NULL) + return 0; + return this->plt_->first_plt_entry_offset(); } // Return the size of each PLT entry. @@ -1524,7 +3012,9 @@ template unsigned int Target_x86_64::plt_entry_size() const { - return Output_data_plt_x86_64::get_plt_entry_size(); + if (this->plt_ == NULL) + return 0; + return this->plt_->get_plt_entry_size(); } // Create the GOT and PLT sections for an incremental update. @@ -1546,7 +3036,7 @@ 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), @@ -1581,10 +3071,15 @@ Target_x86_64::init_got_plt_for_update(Symbol_table* symtab, ORDER_NON_RELRO_FIRST, false); // Create the PLT section. - this->plt_ = new Output_data_plt_x86_64(layout, this->got_, - this->got_plt_, - this->got_irelative_, - plt_count); + this->plt_ = this->make_data_plt(layout, this->got_, + this->got_plt_, + this->got_irelative_, + plt_count); + + // Add unwind information if requested. + if (parameters->options().ld_generated_unwind_info()) + this->plt_->add_eh_frame(layout); + layout->add_output_section_data(".plt", elfcpp::SHT_PROGBITS, elfcpp::SHF_ALLOC | elfcpp::SHF_EXECINSTR, this->plt_, ORDER_PLT, false); @@ -1667,24 +3162,24 @@ Target_x86_64::reserve_global_got_entry(unsigned int got_index, this->got_, got_offset, 0); else rela_dyn->add_global_relative(gsym, elfcpp::R_X86_64_RELATIVE, - this->got_, got_offset, 0); + this->got_, got_offset, 0, false); } break; case GOT_TYPE_TLS_OFFSET: rela_dyn->add_global_relative(gsym, elfcpp::R_X86_64_TPOFF64, - this->got_, got_offset, 0); + this->got_, got_offset, 0, false); break; case GOT_TYPE_TLS_PAIR: this->got_->reserve_slot(got_index + 1); rela_dyn->add_global_relative(gsym, elfcpp::R_X86_64_DTPMOD64, - this->got_, got_offset, 0); + this->got_, got_offset, 0, false); rela_dyn->add_global_relative(gsym, elfcpp::R_X86_64_DTPOFF64, - this->got_, got_offset + 8, 0); + this->got_, got_offset + 8, 0, false); break; case GOT_TYPE_TLS_DESC: this->got_->reserve_slot(got_index + 1); rela_dyn->add_global_relative(gsym, elfcpp::R_X86_64_TLSDESC, - this->got_, got_offset, 0); + this->got_, got_offset, 0, false); break; default: gold_unreachable(); @@ -1758,7 +3253,7 @@ Target_x86_64::define_tls_base_symbol(Symbol_table* symtab, template void Target_x86_64::reserve_tlsdesc_entries(Symbol_table* symtab, - Layout* layout) + Layout* layout) { if (this->plt_ == NULL) this->make_plt_section(symtab, layout); @@ -1788,7 +3283,7 @@ Target_x86_64::got_mod_index_entry(Symbol_table* symtab, Layout* layout, Output_data_got<64, false>* got = this->got_section(symtab, layout); unsigned int got_offset = got->add_constant(0); rela_dyn->add_local(object, 0, elfcpp::R_X86_64_DTPMOD64, got, - got_offset, 0); + got_offset, 0); got->add_constant(0); this->got_mod_index_offset_ = got_offset; } @@ -1876,12 +3371,14 @@ 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; @@ -1889,6 +3386,8 @@ 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; @@ -1963,6 +3462,7 @@ 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 @@ -1970,7 +3470,7 @@ Target_x86_64::Scan::check_non_pic(Relobj* object, unsigned int r_type, && !gsym->is_undefined() && !gsym->is_preemptible())) return; - /* Fall through. */ + // Fall through. case elfcpp::R_X86_64_32: // R_X86_64_32 is OK for x32. if (size == 32 && r_type == elfcpp::R_X86_64_32) @@ -1982,12 +3482,28 @@ 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; @@ -1996,10 +3512,10 @@ Target_x86_64::Scan::check_non_pic(Relobj* object, unsigned int r_type, // section. But we can still wind up issuing more than one // error per object file. if (this->issued_non_pic_error_) - return; + return; gold_assert(parameters->options().output_is_position_independent()); object->error(_("requires unsupported dynamic reloc %u; " - "recompile with -fPIC"), + "recompile with -fPIC"), r_type); this->issued_non_pic_error_ = true; return; @@ -2021,7 +3537,7 @@ Target_x86_64::Scan::reloc_needs_plt_for_ifunc( int flags = Scan::get_reference_flags(r_type); if (flags & Symbol::TLS_REF) gold_error(_("%s: unsupported TLS reloc %u for IFUNC symbol"), - object->name().c_str(), r_type); + object->name().c_str(), r_type); return flags != 0; } @@ -2037,8 +3553,12 @@ 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)) @@ -2062,15 +3582,17 @@ Target_x86_64::Scan::local(Symbol_table* symtab, // R_X86_64_RELATIVE relocation so the dynamic loader can // relocate it easily. if (parameters->options().output_is_position_independent()) - { - unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); - Reloc_section* rela_dyn = target->rela_dyn_section(layout); + { + 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, - elfcpp::R_X86_64_RELATIVE, + (size == 32 + ? elfcpp::R_X86_64_RELATIVE64 + : elfcpp::R_X86_64_RELATIVE), output_section, data_shndx, reloc.get_r_offset(), reloc.get_r_addend(), is_ifunc); - } + } break; case elfcpp::R_X86_64_32: @@ -2082,7 +3604,7 @@ Target_x86_64::Scan::local(Symbol_table* symtab, // location. We can't use an R_X86_64_RELATIVE relocation // because that is always a 64-bit relocation. if (parameters->options().output_is_position_independent()) - { + { // Use R_X86_64_RELATIVE relocation for R_X86_64_32 under x32. if (size == 32 && r_type == elfcpp::R_X86_64_32) { @@ -2096,17 +3618,17 @@ Target_x86_64::Scan::local(Symbol_table* symtab, break; } - this->check_non_pic(object, r_type, NULL); + this->check_non_pic(object, r_type, NULL); - Reloc_section* rela_dyn = target->rela_dyn_section(layout); + Reloc_section* rela_dyn = target->rela_dyn_section(layout); unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); - if (lsym.get_st_type() != elfcpp::STT_SECTION) + if (lsym.get_st_type() != elfcpp::STT_SECTION) rela_dyn->add_local(object, r_sym, r_type, output_section, data_shndx, reloc.get_r_offset(), reloc.get_r_addend()); - else - { - gold_assert(lsym.get_st_value() == 0); + else + { + gold_assert(lsym.get_st_value() == 0); unsigned int shndx = lsym.get_st_shndx(); bool is_ordinary; shndx = object->adjust_sym_shndx(r_sym, shndx, @@ -2119,17 +3641,19 @@ Target_x86_64::Scan::local(Symbol_table* symtab, r_type, output_section, data_shndx, reloc.get_r_offset(), reloc.get_r_addend()); - } - } + } + } break; 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; @@ -2148,11 +3672,34 @@ 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. - Output_data_got<64, false>* got = target->got_section(symtab, layout); - unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); + // 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 (!parameters->incremental() + && (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 // lets function pointers compare correctly with shared @@ -2162,13 +3709,13 @@ Target_x86_64::Scan::local(Symbol_table* symtab, is_new = got->add_local_plt(object, r_sym, GOT_TYPE_STANDARD); else is_new = got->add_local(object, r_sym, GOT_TYPE_STANDARD); - if (is_new) - { - // If we are generating a shared object, we need to add a - // dynamic relocation for this symbol's GOT entry. - if (parameters->options().output_is_position_independent()) - { - Reloc_section* rela_dyn = target->rela_dyn_section(layout); + if (is_new) + { + // If we are generating a shared object, we need to add a + // dynamic relocation for this symbol's GOT entry. + if (parameters->options().output_is_position_independent()) + { + Reloc_section* rela_dyn = target->rela_dyn_section(layout); // R_X86_64_RELATIVE assumes a 64-bit relocation. if (r_type != elfcpp::R_X86_64_GOT32) { @@ -2178,19 +3725,19 @@ Target_x86_64::Scan::local(Symbol_table* symtab, elfcpp::R_X86_64_RELATIVE, got, got_offset, 0, is_ifunc); } - else - { - this->check_non_pic(object, r_type, NULL); - - gold_assert(lsym.get_st_type() != elfcpp::STT_SECTION); - rela_dyn->add_local( - object, r_sym, r_type, got, - object->local_got_offset(r_sym, GOT_TYPE_STANDARD), 0); - } - } - } - // For GOTPLT64, we'd normally want a PLT section, but since - // we know this is a local symbol, no PLT is needed. + else + { + this->check_non_pic(object, r_type, NULL); + + gold_assert(lsym.get_st_type() != elfcpp::STT_SECTION); + rela_dyn->add_local( + object, r_sym, r_type, got, + object->local_got_offset(r_sym, GOT_TYPE_STANDARD), 0); + } + } + } + // For GOTPLT64, we'd normally want a PLT section, but since + // we know this is a local symbol, no PLT is needed. } break; @@ -2219,50 +3766,50 @@ Target_x86_64::Scan::local(Symbol_table* symtab, { bool output_is_shared = parameters->options().shared(); const tls::Tls_optimization optimized_type - = Target_x86_64::optimize_tls_reloc(!output_is_shared, + = Target_x86_64::optimize_tls_reloc(!output_is_shared, r_type); switch (r_type) { - case elfcpp::R_X86_64_TLSGD: // General-dynamic - if (optimized_type == tls::TLSOPT_NONE) - { - // Create a pair of GOT entries for the module index and - // dtv-relative offset. - Output_data_got<64, false>* got - = target->got_section(symtab, layout); - unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); + case elfcpp::R_X86_64_TLSGD: // General-dynamic + if (optimized_type == tls::TLSOPT_NONE) + { + // Create a pair of GOT entries for the module index and + // dtv-relative offset. + Output_data_got<64, false>* got + = target->got_section(symtab, layout); + unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); unsigned int shndx = lsym.get_st_shndx(); bool is_ordinary; shndx = object->adjust_sym_shndx(r_sym, shndx, &is_ordinary); if (!is_ordinary) object->error(_("local symbol %u has bad shndx %u"), r_sym, shndx); - else + else got->add_local_pair_with_rel(object, r_sym, shndx, GOT_TYPE_TLS_PAIR, target->rela_dyn_section(layout), - elfcpp::R_X86_64_DTPMOD64, 0); - } - else if (optimized_type != tls::TLSOPT_TO_LE) + elfcpp::R_X86_64_DTPMOD64); + } + else if (optimized_type != tls::TLSOPT_TO_LE) unsupported_reloc_local(object, r_type); - break; + break; - case elfcpp::R_X86_64_GOTPC32_TLSDESC: - target->define_tls_base_symbol(symtab, layout); + case elfcpp::R_X86_64_GOTPC32_TLSDESC: + target->define_tls_base_symbol(symtab, layout); if (optimized_type == tls::TLSOPT_NONE) { - // Create reserved PLT and GOT entries for the resolver. - target->reserve_tlsdesc_entries(symtab, layout); - - // Generate a double GOT entry with an - // R_X86_64_TLSDESC reloc. The R_X86_64_TLSDESC reloc - // is resolved lazily, so the GOT entry needs to be in - // an area in .got.plt, not .got. Call got_section to - // make sure the section has been created. + // Create reserved PLT and GOT entries for the resolver. + target->reserve_tlsdesc_entries(symtab, layout); + + // Generate a double GOT entry with an + // R_X86_64_TLSDESC reloc. The R_X86_64_TLSDESC reloc + // is resolved lazily, so the GOT entry needs to be in + // an area in .got.plt, not .got. Call got_section to + // make sure the section has been created. target->got_section(symtab, layout); - Output_data_got<64, false>* got = target->got_tlsdesc_section(); - unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); + Output_data_got<64, false>* got = target->got_tlsdesc_section(); + unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); if (!object->local_has_got_offset(r_sym, GOT_TYPE_TLS_DESC)) { unsigned int got_offset = got->add_constant(0); @@ -2283,47 +3830,47 @@ Target_x86_64::Scan::local(Symbol_table* symtab, unsupported_reloc_local(object, r_type); break; - case elfcpp::R_X86_64_TLSDESC_CALL: + case elfcpp::R_X86_64_TLSDESC_CALL: break; - case elfcpp::R_X86_64_TLSLD: // Local-dynamic + case elfcpp::R_X86_64_TLSLD: // Local-dynamic if (optimized_type == tls::TLSOPT_NONE) { - // Create a GOT entry for the module index. - target->got_mod_index_entry(symtab, layout, object); + // Create a GOT entry for the module index. + target->got_mod_index_entry(symtab, layout, object); } else if (optimized_type != tls::TLSOPT_TO_LE) unsupported_reloc_local(object, r_type); break; - case elfcpp::R_X86_64_DTPOFF32: - case elfcpp::R_X86_64_DTPOFF64: + case elfcpp::R_X86_64_DTPOFF32: + case elfcpp::R_X86_64_DTPOFF64: break; - case elfcpp::R_X86_64_GOTTPOFF: // Initial-exec + case elfcpp::R_X86_64_GOTTPOFF: // Initial-exec layout->set_has_static_tls(); - if (optimized_type == tls::TLSOPT_NONE) - { - // Create a GOT entry for the tp-relative offset. - Output_data_got<64, false>* got - = target->got_section(symtab, layout); - unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); - got->add_local_with_rel(object, r_sym, GOT_TYPE_TLS_OFFSET, + if (optimized_type == tls::TLSOPT_NONE) + { + // Create a GOT entry for the tp-relative offset. + Output_data_got<64, false>* got + = target->got_section(symtab, layout); + unsigned int r_sym = elfcpp::elf_r_sym(reloc.get_r_info()); + got->add_local_with_rel(object, r_sym, GOT_TYPE_TLS_OFFSET, target->rela_dyn_section(layout), elfcpp::R_X86_64_TPOFF64); - } - else if (optimized_type != tls::TLSOPT_TO_LE) - unsupported_reloc_local(object, r_type); - break; - - case elfcpp::R_X86_64_TPOFF32: // Local-exec - layout->set_has_static_tls(); - if (output_is_shared) - unsupported_reloc_local(object, r_type); + } + else if (optimized_type != tls::TLSOPT_TO_LE) + unsupported_reloc_local(object, r_type); + break; + + case elfcpp::R_X86_64_TPOFF32: // Local-exec + layout->set_has_static_tls(); + if (output_is_shared) + unsupported_reloc_local(object, r_type); break; - default: - gold_unreachable(); + default: + gold_unreachable(); } } break; @@ -2354,7 +3901,11 @@ Target_x86_64::Scan::unsupported_reloc_global( // Returns true if this relocation type could be that of a function pointer. template inline bool -Target_x86_64::Scan::possible_function_pointer_reloc(unsigned int r_type) +Target_x86_64::Scan::possible_function_pointer_reloc( + Sized_relobj_file* src_obj, + unsigned int src_indx, + unsigned int r_offset, + unsigned int r_type) { switch (r_type) { @@ -2367,8 +3918,45 @@ 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; + } + case elfcpp::R_X86_64_PC32: + { + // This relocation may be used both for function calls and + // for taking address of a function. We distinguish between + // them by checking the opcodes. + uint64_t sh_flags = src_obj->section_flags(src_indx); + bool is_executable = (sh_flags & elfcpp::SHF_EXECINSTR) != 0; + if (is_executable) + { + section_size_type stype; + const unsigned char* view = src_obj->section_contents(src_indx, + &stype, + true); + + // call + if (r_offset >= 1 + && view[r_offset - 1] == 0xe8) + return false; + + // jmp + if (r_offset >= 1 + && view[r_offset - 1] == 0xe9) + return false; + + // jo/jno/jb/jnb/je/jne/jna/ja/js/jns/jp/jnp/jl/jge/jle/jg + if (r_offset >= 2 + && view[r_offset - 2] == 0x0f + && view[r_offset - 1] >= 0x80 + && view[r_offset - 1] <= 0x8f) + return false; + } + + // Be conservative and treat all others as function pointers. return true; } } @@ -2385,18 +3973,21 @@ Target_x86_64::Scan::local_reloc_may_be_function_pointer( Symbol_table* , Layout* , Target_x86_64* , - Sized_relobj_file* , - unsigned int , + Sized_relobj_file* src_obj, + unsigned int src_indx, Output_section* , - const elfcpp::Rela& , + const elfcpp::Rela& reloc, unsigned int r_type, const elfcpp::Sym&) { // When building a shared library, do not fold any local symbols as it is // not possible to distinguish pointer taken versus a call by looking at // the relocation types. - return (parameters->options().shared() - || possible_function_pointer_reloc(r_type)); + if (parameters->options().shared()) + return true; + + return possible_function_pointer_reloc(src_obj, src_indx, + reloc.get_r_offset(), r_type); } // For safe ICF, scan a relocation for a global symbol to check if it @@ -2409,20 +4000,23 @@ Target_x86_64::Scan::global_reloc_may_be_function_pointer( Symbol_table*, Layout* , Target_x86_64* , - Sized_relobj_file* , - unsigned int , + Sized_relobj_file* src_obj, + unsigned int src_indx, Output_section* , - const elfcpp::Rela& , + const elfcpp::Rela& reloc, unsigned int r_type, Symbol* gsym) { // When building a shared library, do not fold symbols whose visibility // is hidden, internal or protected. - return ((parameters->options().shared() - && (gsym->visibility() == elfcpp::STV_INTERNAL - || gsym->visibility() == elfcpp::STV_PROTECTED - || gsym->visibility() == elfcpp::STV_HIDDEN)) - || possible_function_pointer_reloc(r_type)); + if (parameters->options().shared() + && (gsym->visibility() == elfcpp::STV_INTERNAL + || gsym->visibility() == elfcpp::STV_PROTECTED + || gsym->visibility() == elfcpp::STV_HIDDEN)) + return true; + + return possible_function_pointer_reloc(src_obj, src_indx, + reloc.get_r_offset(), r_type); } // Scan a relocation for a global symbol. @@ -2430,14 +4024,14 @@ Target_x86_64::Scan::global_reloc_may_be_function_pointer( template inline void Target_x86_64::Scan::global(Symbol_table* symtab, - Layout* layout, - Target_x86_64* target, - Sized_relobj_file* object, - unsigned int data_shndx, - Output_section* output_section, - const elfcpp::Rela& reloc, - unsigned int r_type, - Symbol* gsym) + Layout* layout, + Target_x86_64* target, + Sized_relobj_file* object, + unsigned int data_shndx, + Output_section* output_section, + const elfcpp::Rela& reloc, + unsigned int r_type, + Symbol* gsym) { // A STT_GNU_IFUNC symbol may require a PLT entry. if (gsym->type() == elfcpp::STT_GNU_IFUNC @@ -2457,25 +4051,26 @@ Target_x86_64::Scan::global(Symbol_table* symtab, case elfcpp::R_X86_64_16: case elfcpp::R_X86_64_8: { - // Make a PLT entry if necessary. - if (gsym->needs_plt_entry()) - { - target->make_plt_entry(symtab, layout, gsym); - // Since this is not a PC-relative relocation, we may be - // taking the address of a function. In that case we need to - // set the entry in the dynamic symbol table to the address of - // the PLT entry. - if (gsym->is_from_dynobj() && !parameters->options().shared()) - gsym->set_needs_dynsym_value(); - } - // Make a dynamic relocation if necessary. - if (gsym->needs_dynamic_reloc(Scan::get_reference_flags(r_type))) - { - if (gsym->may_need_copy_reloc()) - { - target->copy_reloc(symtab, layout, object, - data_shndx, output_section, gsym, reloc); - } + // Make a PLT entry if necessary. + if (gsym->needs_plt_entry()) + { + target->make_plt_entry(symtab, layout, gsym); + // Since this is not a PC-relative relocation, we may be + // taking the address of a function. In that case we need to + // set the entry in the dynamic symbol table to the address of + // the PLT entry. + if (gsym->is_from_dynobj() && !parameters->options().shared()) + gsym->set_needs_dynsym_value(); + } + // Make a dynamic relocation if necessary. + if (gsym->needs_dynamic_reloc(Scan::get_reference_flags(r_type))) + { + if (!parameters->options().output_is_position_independent() + && gsym->may_need_copy_reloc()) + { + target->copy_reloc(symtab, layout, object, + data_shndx, output_section, gsym, reloc); + } else if (((size == 64 && r_type == elfcpp::R_X86_64_64) || (size == 32 && r_type == elfcpp::R_X86_64_32)) && gsym->type() == elfcpp::STT_GNU_IFUNC @@ -2497,53 +4092,56 @@ 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 - && gsym->can_use_relative_reloc(false)) - { - Reloc_section* rela_dyn = target->rela_dyn_section(layout); + 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); rela_dyn->add_global_relative(gsym, elfcpp::R_X86_64_RELATIVE, output_section, object, data_shndx, reloc.get_r_offset(), - reloc.get_r_addend()); - } - else - { - this->check_non_pic(object, r_type, gsym); - Reloc_section* rela_dyn = target->rela_dyn_section(layout); - rela_dyn->add_global(gsym, r_type, output_section, object, - data_shndx, reloc.get_r_offset(), - reloc.get_r_addend()); - } - } + reloc.get_r_addend(), false); + } + else + { + this->check_non_pic(object, r_type, gsym); + Reloc_section* rela_dyn = target->rela_dyn_section(layout); + rela_dyn->add_global(gsym, r_type, output_section, object, + data_shndx, reloc.get_r_offset(), + reloc.get_r_addend()); + } + } } break; 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: { - // Make a PLT entry if necessary. - if (gsym->needs_plt_entry()) - target->make_plt_entry(symtab, layout, gsym); - // Make a dynamic relocation if necessary. - if (gsym->needs_dynamic_reloc(Scan::get_reference_flags(r_type))) - { - if (gsym->may_need_copy_reloc()) - { - target->copy_reloc(symtab, layout, object, - data_shndx, output_section, gsym, reloc); - } - else - { - this->check_non_pic(object, r_type, gsym); - Reloc_section* rela_dyn = target->rela_dyn_section(layout); - rela_dyn->add_global(gsym, r_type, output_section, object, - data_shndx, reloc.get_r_offset(), - reloc.get_r_addend()); - } - } + // Make a PLT entry if necessary. + if (gsym->needs_plt_entry()) + target->make_plt_entry(symtab, layout, gsym); + // Make a dynamic relocation if necessary. + if (gsym->needs_dynamic_reloc(Scan::get_reference_flags(r_type))) + { + if (parameters->options().output_is_executable() + && gsym->may_need_copy_reloc()) + { + target->copy_reloc(symtab, layout, object, + data_shndx, output_section, gsym, reloc); + } + else + { + this->check_non_pic(object, r_type, gsym); + Reloc_section* rela_dyn = target->rela_dyn_section(layout); + rela_dyn->add_global(gsym, r_type, output_section, object, + data_shndx, reloc.get_r_offset(), + reloc.get_r_addend()); + } + } } break; @@ -2551,11 +4149,42 @@ 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 (gsym->final_value_is_known()) + // 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. + // We cannot make these optimizations in incremental linking mode, + // because we look at the opcode to decide whether or not to make + // change, and during an incremental update, the change may have + // already been applied. + + Lazy_view view(object, data_shndx); + size_t r_offset = reloc.get_r_offset(); + if (!parameters->incremental() + && r_offset >= 2 + && Target_x86_64::can_convert_mov_to_lea(gsym, r_type, + r_offset, &view)) + break; + + if (!parameters->incremental() + && 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. if (gsym->type() == elfcpp::STT_GNU_IFUNC) @@ -2563,11 +4192,11 @@ Target_x86_64::Scan::global(Symbol_table* symtab, else got->add_global(gsym, GOT_TYPE_STANDARD); } - else - { - // If this symbol is not fully resolved, we need to add a - // dynamic relocation for it. - Reloc_section* rela_dyn = target->rela_dyn_section(layout); + else + { + // If this symbol is not fully resolved, we need to add a + // dynamic relocation for it. + Reloc_section* rela_dyn = target->rela_dyn_section(layout); // Use a GLOB_DAT rather than a RELATIVE reloc if: // @@ -2588,10 +4217,10 @@ Target_x86_64::Scan::global(Symbol_table* symtab, && parameters->options().shared()) || (gsym->type() == elfcpp::STT_GNU_IFUNC && parameters->options().output_is_position_independent())) - got->add_global_with_rel(gsym, GOT_TYPE_STANDARD, rela_dyn, + got->add_global_with_rel(gsym, GOT_TYPE_STANDARD, rela_dyn, elfcpp::R_X86_64_GLOB_DAT); - else - { + else + { // For a STT_GNU_IFUNC symbol we want to write the PLT // offset into the GOT, so that function pointer // comparisons work correctly. @@ -2607,24 +4236,20 @@ Target_x86_64::Scan::global(Symbol_table* symtab, && !parameters->options().shared()) gsym->set_needs_dynsym_value(); } - if (is_new) + if (is_new) { unsigned int got_off = gsym->got_offset(GOT_TYPE_STANDARD); rela_dyn->add_global_relative(gsym, elfcpp::R_X86_64_RELATIVE, - got, got_off, 0); + got, got_off, 0, false); } - } - } - // 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()) @@ -2633,8 +4258,8 @@ Target_x86_64::Scan::global(Symbol_table* symtab, // if the symbol is defined in the output file and is protected // or hidden. if (gsym->is_defined() - && !gsym->is_from_dynobj() - && !gsym->is_preemptible()) + && !gsym->is_from_dynobj() + && !gsym->is_preemptible()) break; target->make_plt_entry(symtab, layout, gsym); break; @@ -2675,29 +4300,34 @@ 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); + = Target_x86_64::optimize_tls_reloc(is_final, r_type); switch (r_type) { - case elfcpp::R_X86_64_TLSGD: // General-dynamic + case elfcpp::R_X86_64_TLSGD: // General-dynamic if (optimized_type == tls::TLSOPT_NONE) { - // Create a pair of GOT entries for the module index and - // dtv-relative offset. - Output_data_got<64, false>* got - = target->got_section(symtab, layout); - got->add_global_pair_with_rel(gsym, GOT_TYPE_TLS_PAIR, + // Create a pair of GOT entries for the module index and + // dtv-relative offset. + Output_data_got<64, false>* got + = target->got_section(symtab, layout); + got->add_global_pair_with_rel(gsym, GOT_TYPE_TLS_PAIR, target->rela_dyn_section(layout), elfcpp::R_X86_64_DTPMOD64, elfcpp::R_X86_64_DTPOFF64); } else if (optimized_type == tls::TLSOPT_TO_IE) { - // Create a GOT entry for the tp-relative offset. - Output_data_got<64, false>* got - = target->got_section(symtab, layout); - got->add_global_with_rel(gsym, GOT_TYPE_TLS_OFFSET, + // Create a GOT entry for the tp-relative offset. + Output_data_got<64, false>* got + = target->got_section(symtab, layout); + got->add_global_with_rel(gsym, GOT_TYPE_TLS_OFFSET, target->rela_dyn_section(layout), elfcpp::R_X86_64_TPOFF64); } @@ -2705,30 +4335,30 @@ Target_x86_64::Scan::global(Symbol_table* symtab, unsupported_reloc_global(object, r_type, gsym); break; - case elfcpp::R_X86_64_GOTPC32_TLSDESC: - target->define_tls_base_symbol(symtab, layout); + case elfcpp::R_X86_64_GOTPC32_TLSDESC: + target->define_tls_base_symbol(symtab, layout); if (optimized_type == tls::TLSOPT_NONE) { - // Create reserved PLT and GOT entries for the resolver. - target->reserve_tlsdesc_entries(symtab, layout); - - // Create a double GOT entry with an R_X86_64_TLSDESC - // reloc. The R_X86_64_TLSDESC reloc is resolved - // lazily, so the GOT entry needs to be in an area in - // .got.plt, not .got. Call got_section to make sure - // the section has been created. + // Create reserved PLT and GOT entries for the resolver. + target->reserve_tlsdesc_entries(symtab, layout); + + // Create a double GOT entry with an R_X86_64_TLSDESC + // reloc. The R_X86_64_TLSDESC reloc is resolved + // lazily, so the GOT entry needs to be in an area in + // .got.plt, not .got. Call got_section to make sure + // the section has been created. target->got_section(symtab, layout); - Output_data_got<64, false>* got = target->got_tlsdesc_section(); + Output_data_got<64, false>* got = target->got_tlsdesc_section(); Reloc_section* rt = target->rela_tlsdesc_section(layout); - got->add_global_pair_with_rel(gsym, GOT_TYPE_TLS_DESC, rt, + got->add_global_pair_with_rel(gsym, GOT_TYPE_TLS_DESC, rt, elfcpp::R_X86_64_TLSDESC, 0); } else if (optimized_type == tls::TLSOPT_TO_IE) { - // Create a GOT entry for the tp-relative offset. - Output_data_got<64, false>* got - = target->got_section(symtab, layout); - got->add_global_with_rel(gsym, GOT_TYPE_TLS_OFFSET, + // Create a GOT entry for the tp-relative offset. + Output_data_got<64, false>* got + = target->got_section(symtab, layout); + got->add_global_with_rel(gsym, GOT_TYPE_TLS_OFFSET, target->rela_dyn_section(layout), elfcpp::R_X86_64_TPOFF64); } @@ -2736,46 +4366,46 @@ Target_x86_64::Scan::global(Symbol_table* symtab, unsupported_reloc_global(object, r_type, gsym); break; - case elfcpp::R_X86_64_TLSDESC_CALL: + case elfcpp::R_X86_64_TLSDESC_CALL: break; - case elfcpp::R_X86_64_TLSLD: // Local-dynamic + case elfcpp::R_X86_64_TLSLD: // Local-dynamic if (optimized_type == tls::TLSOPT_NONE) { - // Create a GOT entry for the module index. - target->got_mod_index_entry(symtab, layout, object); + // Create a GOT entry for the module index. + target->got_mod_index_entry(symtab, layout, object); } else if (optimized_type != tls::TLSOPT_TO_LE) unsupported_reloc_global(object, r_type, gsym); break; - case elfcpp::R_X86_64_DTPOFF32: - case elfcpp::R_X86_64_DTPOFF64: + case elfcpp::R_X86_64_DTPOFF32: + case elfcpp::R_X86_64_DTPOFF64: break; - case elfcpp::R_X86_64_GOTTPOFF: // Initial-exec + case elfcpp::R_X86_64_GOTTPOFF: // Initial-exec layout->set_has_static_tls(); - if (optimized_type == tls::TLSOPT_NONE) - { - // Create a GOT entry for the tp-relative offset. - Output_data_got<64, false>* got - = target->got_section(symtab, layout); - got->add_global_with_rel(gsym, GOT_TYPE_TLS_OFFSET, + if (optimized_type == tls::TLSOPT_NONE) + { + // Create a GOT entry for the tp-relative offset. + Output_data_got<64, false>* got + = target->got_section(symtab, layout); + got->add_global_with_rel(gsym, GOT_TYPE_TLS_OFFSET, target->rela_dyn_section(layout), elfcpp::R_X86_64_TPOFF64); - } - else if (optimized_type != tls::TLSOPT_TO_LE) - unsupported_reloc_global(object, r_type, gsym); - break; + } + else if (optimized_type != tls::TLSOPT_TO_LE) + unsupported_reloc_global(object, r_type, gsym); + break; - case elfcpp::R_X86_64_TPOFF32: // Local-exec + case elfcpp::R_X86_64_TPOFF32: // Local-exec layout->set_has_static_tls(); - if (parameters->options().shared()) - unsupported_reloc_local(object, r_type); + if (parameters->options().shared()) + unsupported_reloc_global(object, r_type, gsym); break; - default: - gold_unreachable(); + default: + gold_unreachable(); } } break; @@ -2785,7 +4415,7 @@ Target_x86_64::Scan::global(Symbol_table* symtab, default: gold_error(_("%s: unsupported reloc %u against global symbol %s"), object->name().c_str(), r_type, - gsym->demangled_name().c_str()); + gsym->demangled_name().c_str()); break; } } @@ -2804,15 +4434,16 @@ 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, @@ -2824,7 +4455,7 @@ Target_x86_64::gc_process_relocs(Symbol_table* symtab, needs_special_offset_handling, local_symbol_count, plocal_symbols); - + } // Scan relocations for a section. @@ -2842,6 +4473,9 @@ 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"), @@ -2849,8 +4483,7 @@ 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, @@ -2878,7 +4511,7 @@ Target_x86_64::do_finalize_sections( : this->plt_->rela_plt()); layout->add_target_dynamic_tags(false, this->got_plt_, rel_plt, this->rela_dyn_, true, false); - + // Fill in some more dynamic tags. Output_data_dynamic* const odyn = layout->dynamic_data(); if (odyn != NULL) @@ -2953,32 +4586,96 @@ 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_PC32) + && r_type != elfcpp::R_X86_64_GOTPCREL + && r_type != elfcpp::R_X86_64_GOTPCRELX + && 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) { gold_error_at_location(relinfo, relnum, rela.get_r_offset(), _("missing expected TLS relocation")); + this->skip_call_tls_get_addr_ = false; } else { @@ -2987,6 +4684,9 @@ 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. @@ -2994,8 +4694,7 @@ 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()) @@ -3003,8 +4702,7 @@ 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; } } @@ -3016,26 +4714,27 @@ 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) - { - gold_assert(gsym->has_got_offset(GOT_TYPE_STANDARD)); - got_offset = gsym->got_offset(GOT_TYPE_STANDARD) - target->got_size(); - } + { + 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()); - } + { + 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()); + } have_got_offset = true; break; @@ -3043,6 +4742,8 @@ 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: @@ -3051,55 +4752,50 @@ 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, - address); + 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->has_plt_offset() || gsym->final_value_is_known() || (gsym->is_defined() && !gsym->is_from_dynobj() @@ -3107,76 +4803,150 @@ 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: { - gold_assert(gsym); - gold_assert(gsym->has_plt_offset() - || gsym->final_value_is_known()); + gold_assert(gsym); + 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: { - gold_assert(gsym); + 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 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: { - gold_assert(gsym); + 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; case elfcpp::R_X86_64_GOTOFF64: { - typename elfcpp::Elf_types::Elf_Addr value; - value = (psymval->value(object, 0) - - target->got_plt_section()->address()); - Relocate_functions::rela64(view, value, addend); + typename elfcpp::Elf_types::Elf_Addr reladdr; + reladdr = target->got_plt_section()->address(); + Reloc_funcs::pcrela64(view, object, psymval, addend, reladdr); } 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 (!parameters->incremental() + && ((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 (!parameters->incremental() + && 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; case elfcpp::R_X86_64_GOTPCREL64: { - 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); + gold_assert(have_got_offset); + typename elfcpp::Elf_types::Elf_Addr value; + value = target->got_plt_section()->address() + got_offset; + Reloc_funcs::pcrela64(view, value, addend, address); } break; @@ -3204,7 +4974,7 @@ Target_x86_64::Relocate::relocate( case elfcpp::R_X86_64_GOTTPOFF: // Initial-exec case elfcpp::R_X86_64_TPOFF32: // Local-exec this->relocate_tls(relinfo, target, relnum, rela, r_type, gsym, psymval, - view, address, view_size); + view, address, view_size); break; case elfcpp::R_X86_64_SIZE32: @@ -3216,6 +4986,32 @@ 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; } @@ -3276,40 +5072,40 @@ Target_x86_64::Relocate::relocate_tls( break; } else - { - unsigned int got_type = (optimized_type == tls::TLSOPT_TO_IE - ? GOT_TYPE_TLS_OFFSET - : GOT_TYPE_TLS_PAIR); - unsigned int got_offset; - if (gsym != NULL) - { - gold_assert(gsym->has_got_offset(got_type)); - got_offset = gsym->got_offset(got_type) - 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)); - got_offset = (object->local_got_offset(r_sym, got_type) - - target->got_size()); - } - 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, - value, view, address, view_size); - break; - } - else if (optimized_type == tls::TLSOPT_NONE) - { - // Relocate the field with the offset of the pair of GOT - // entries. + { + unsigned int got_type = (optimized_type == tls::TLSOPT_TO_IE + ? GOT_TYPE_TLS_OFFSET + : GOT_TYPE_TLS_PAIR); + unsigned int got_offset; + if (gsym != NULL) + { + gold_assert(gsym->has_got_offset(got_type)); + got_offset = gsym->got_offset(got_type) - 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)); + got_offset = (object->local_got_offset(r_sym, got_type) + - target->got_size()); + } + if (optimized_type == tls::TLSOPT_TO_IE) + { + value = target->got_plt_section()->address() + got_offset; + this->tls_gd_to_ie(relinfo, relnum, rela, r_type, + value, view, address, view_size); + break; + } + else if (optimized_type == tls::TLSOPT_NONE) + { + // Relocate the field with the offset of the pair of GOT + // entries. value = target->got_plt_section()->address() + got_offset; - Relocate_functions::pcrela32(view, value, addend, + Relocate_functions::pcrela32(view, value, addend, address); - break; - } - } + break; + } + } gold_error_at_location(relinfo, relnum, rela.get_r_offset(), _("unsupported reloc %u"), r_type); break; @@ -3330,16 +5126,16 @@ Target_x86_64::Relocate::relocate_tls( return; } this->tls_desc_gd_to_le(relinfo, relnum, tls_segment, - rela, r_type, value, view, - view_size); + rela, r_type, value, view, + view_size); break; } else - { - unsigned int got_type = (optimized_type == tls::TLSOPT_TO_IE - ? GOT_TYPE_TLS_OFFSET - : GOT_TYPE_TLS_DESC); - unsigned int got_offset = 0; + { + unsigned int got_type = (optimized_type == tls::TLSOPT_TO_IE + ? GOT_TYPE_TLS_OFFSET + : GOT_TYPE_TLS_DESC); + unsigned int got_offset = 0; if (r_type == elfcpp::R_X86_64_GOTPC32_TLSDESC && optimized_type == tls::TLSOPT_NONE) { @@ -3349,45 +5145,39 @@ Target_x86_64::Relocate::relocate_tls( got_offset = (target->got_size() + target->got_plt_section()->data_size()); } - if (gsym != NULL) - { - gold_assert(gsym->has_got_offset(got_type)); - got_offset += gsym->got_offset(got_type) - 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)); - got_offset += (object->local_got_offset(r_sym, got_type) + if (gsym != NULL) + { + gold_assert(gsym->has_got_offset(got_type)); + got_offset += gsym->got_offset(got_type) - 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)); + got_offset += (object->local_got_offset(r_sym, got_type) - target->got_size()); - } - if (optimized_type == tls::TLSOPT_TO_IE) - { - if (tls_segment == NULL) + } + if (optimized_type == tls::TLSOPT_TO_IE) + { + value = target->got_plt_section()->address() + got_offset; + this->tls_desc_gd_to_ie(relinfo, relnum, + rela, r_type, value, view, address, + view_size); + break; + } + else if (optimized_type == tls::TLSOPT_NONE) + { + if (r_type == elfcpp::R_X86_64_GOTPC32_TLSDESC) { - 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, - rela, r_type, value, view, address, - view_size); - break; - } - else if (optimized_type == tls::TLSOPT_NONE) - { - if (r_type == elfcpp::R_X86_64_GOTPC32_TLSDESC) - { - // Relocate the field with the offset of the pair of GOT - // entries. - value = target->got_plt_section()->address() + got_offset; - Relocate_functions::pcrela32(view, value, addend, + // Relocate the field with the offset of the pair of GOT + // entries. + value = target->got_plt_section()->address() + got_offset; + Relocate_functions::pcrela32(view, value, addend, address); - } - break; - } - } + } + break; + } + } gold_error_at_location(relinfo, relnum, rela.get_r_offset(), _("unsupported reloc %u"), r_type); break; @@ -3399,7 +5189,7 @@ Target_x86_64::Relocate::relocate_tls( optimized_type = tls::TLSOPT_NONE; } if (optimized_type == tls::TLSOPT_TO_LE) - { + { if (tls_segment == NULL) { gold_assert(parameters->errors()->error_count() > 0 @@ -3409,19 +5199,19 @@ Target_x86_64::Relocate::relocate_tls( this->tls_ld_to_le(relinfo, relnum, tls_segment, rela, r_type, value, view, view_size); break; - } + } else if (optimized_type == tls::TLSOPT_NONE) - { - // Relocate the field with the offset of the GOT entry for - // the module index. - unsigned int got_offset; - got_offset = (target->got_mod_index_entry(NULL, NULL, NULL) + { + // Relocate the field with the offset of the GOT entry for + // the module index. + unsigned int got_offset; + got_offset = (target->got_mod_index_entry(NULL, NULL, NULL) - target->got_size()); value = target->got_plt_section()->address() + got_offset; - Relocate_functions::pcrela32(view, value, addend, + Relocate_functions::pcrela32(view, value, addend, address); - break; - } + break; + } gold_error_at_location(relinfo, relnum, rela.get_r_offset(), _("unsupported reloc %u"), r_type); break; @@ -3462,7 +5252,17 @@ 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) { @@ -3477,29 +5277,29 @@ Target_x86_64::Relocate::relocate_tls( break; } else if (optimized_type == tls::TLSOPT_NONE) - { - // Relocate the field with the offset of the GOT entry for - // the tp-relative offset of the symbol. - unsigned int got_offset; - if (gsym != NULL) - { - gold_assert(gsym->has_got_offset(GOT_TYPE_TLS_OFFSET)); - got_offset = (gsym->got_offset(GOT_TYPE_TLS_OFFSET) - - 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_TLS_OFFSET)); - got_offset = (object->local_got_offset(r_sym, GOT_TYPE_TLS_OFFSET) - - target->got_size()); - } + { + // Relocate the field with the offset of the GOT entry for + // the tp-relative offset of the symbol. + unsigned int got_offset; + if (gsym != NULL) + { + gold_assert(gsym->has_got_offset(GOT_TYPE_TLS_OFFSET)); + got_offset = (gsym->got_offset(GOT_TYPE_TLS_OFFSET) + - 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_TLS_OFFSET)); + got_offset = (object->local_got_offset(r_sym, GOT_TYPE_TLS_OFFSET) + - target->got_size()); + } value = target->got_plt_section()->address() + got_offset; - Relocate_functions::pcrela32(view, value, addend, + Relocate_functions::pcrela32(view, value, addend, address); - break; - } + break; + } gold_error_at_location(relinfo, relnum, rela.get_r_offset(), _("unsupported reloc type %u"), r_type); @@ -3526,7 +5326,6 @@ 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, @@ -3534,19 +5333,44 @@ Target_x86_64::Relocate::tls_gd_to_ie( typename elfcpp::Elf_types::Elf_Addr address, section_size_type view_size) { - // .byte 0x66; leaq foo@tlsgd(%rip),%rdi; - // .word 0x6666; rex64; call __tls_get_addr - // ==> movq %fs:0,%rax; addq x@gottpoff(%rip),%rax + // For SIZE == 64: + // .byte 0x66; leaq foo@tlsgd(%rip),%rdi; + // .word 0x6666; rex64; call __tls_get_addr@PLT + // ==> movq %fs:0,%rax; addq x@gottpoff(%rip),%rax + // .byte 0x66; leaq foo@tlsgd(%rip),%rdi; + // .word 0x66; rex64; call *__tls_get_addr@GOTPCREL(%rip) + // ==> movq %fs:0,%rax; addq x@gottpoff(%rip),%rax + // For SIZE == 32: + // leaq foo@tlsgd(%rip),%rdi; + // .word 0x6666; rex64; call __tls_get_addr@PLT + // ==> movl %fs:0,%eax; addq x@gottpoff(%rip),%rax + // leaq foo@tlsgd(%rip),%rdi; + // .word 0x66; rex64; call *__tls_get_addr@GOTPCREL(%rip) + // ==> movl %fs:0,%eax; addq x@gottpoff(%rip),%rax - tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, -4); tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, 12); - tls::check_tls(relinfo, relnum, rela.get_r_offset(), - (memcmp(view - 4, "\x66\x48\x8d\x3d", 4) == 0)); - tls::check_tls(relinfo, relnum, rela.get_r_offset(), - (memcmp(view + 4, "\x66\x66\x48\xe8", 4) == 0)); + (memcmp(view + 4, "\x66\x66\x48\xe8", 4) == 0 + || memcmp(view + 4, "\x66\x48\xff", 3) == 0)); - memcpy(view - 4, "\x64\x48\x8b\x04\x25\0\0\0\0\x48\x03\x05\0\0\0\0", 16); + if (size == 64) + { + tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, + -4); + tls::check_tls(relinfo, relnum, rela.get_r_offset(), + (memcmp(view - 4, "\x66\x48\x8d\x3d", 4) == 0)); + memcpy(view - 4, "\x64\x48\x8b\x04\x25\0\0\0\0\x48\x03\x05\0\0\0\0", + 16); + } + else + { + tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, + -3); + tls::check_tls(relinfo, relnum, rela.get_r_offset(), + (memcmp(view - 3, "\x48\x8d\x3d", 3) == 0)); + memcpy(view - 3, "\x64\x8b\x04\x25\0\0\0\0\x48\x03\x05\0\0\0\0", + 15); + } const elfcpp::Elf_Xword addend = rela.get_r_addend(); Relocate_functions::pcrela32(view + 8, value, addend - 8, @@ -3572,19 +5396,45 @@ Target_x86_64::Relocate::tls_gd_to_le( unsigned char* view, section_size_type view_size) { - // .byte 0x66; leaq foo@tlsgd(%rip),%rdi; - // .word 0x6666; rex64; call __tls_get_addr - // ==> movq %fs:0,%rax; leaq x@tpoff(%rax),%rax + // For SIZE == 64: + // .byte 0x66; leaq foo@tlsgd(%rip),%rdi; + // .word 0x6666; rex64; call __tls_get_addr@PLT + // ==> movq %fs:0,%rax; leaq x@tpoff(%rax),%rax + // .byte 0x66; leaq foo@tlsgd(%rip),%rdi; + // .word 0x66; rex64; call *__tls_get_addr@GOTPCREL(%rip) + // ==> movq %fs:0,%rax; leaq x@tpoff(%rax),%rax + // For SIZE == 32: + // leaq foo@tlsgd(%rip),%rdi; + // .word 0x6666; rex64; call __tls_get_addr@PLT + // ==> movl %fs:0,%eax; leaq x@tpoff(%rax),%rax + // leaq foo@tlsgd(%rip),%rdi; + // .word 0x66; rex64; call *__tls_get_addr@GOTPCREL(%rip) + // ==> movl %fs:0,%eax; leaq x@tpoff(%rax),%rax - tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, -4); tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, 12); - tls::check_tls(relinfo, relnum, rela.get_r_offset(), - (memcmp(view - 4, "\x66\x48\x8d\x3d", 4) == 0)); - tls::check_tls(relinfo, relnum, rela.get_r_offset(), - (memcmp(view + 4, "\x66\x66\x48\xe8", 4) == 0)); + (memcmp(view + 4, "\x66\x66\x48\xe8", 4) == 0 + || memcmp(view + 4, "\x66\x48\xff", 3) == 0)); - memcpy(view - 4, "\x64\x48\x8b\x04\x25\0\0\0\0\x48\x8d\x80\0\0\0\0", 16); + if (size == 64) + { + tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, + -4); + tls::check_tls(relinfo, relnum, rela.get_r_offset(), + (memcmp(view - 4, "\x66\x48\x8d\x3d", 4) == 0)); + memcpy(view - 4, "\x64\x48\x8b\x04\x25\0\0\0\0\x48\x8d\x80\0\0\0\0", + 16); + } + else + { + tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, + -3); + tls::check_tls(relinfo, relnum, rela.get_r_offset(), + (memcmp(view - 3, "\x48\x8d\x3d", 3) == 0)); + + memcpy(view - 3, "\x64\x8b\x04\x25\0\0\0\0\x48\x8d\x80\0\0\0\0", + 15); + } value -= tls_segment->memsz(); Relocate_functions::rela32(view + 8, value, 0); @@ -3601,7 +5451,6 @@ 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, @@ -3616,7 +5465,7 @@ Target_x86_64::Relocate::tls_desc_gd_to_ie( tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, -3); tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, 4); tls::check_tls(relinfo, relnum, rela.get_r_offset(), - view[-3] == 0x48 && view[-2] == 0x8d && view[-1] == 0x05); + view[-3] == 0x48 && view[-2] == 0x8d && view[-1] == 0x05); view[-2] = 0x8b; const elfcpp::Elf_Xword addend = rela.get_r_addend(); Relocate_functions::pcrela32(view, value, addend, address); @@ -3628,7 +5477,7 @@ Target_x86_64::Relocate::tls_desc_gd_to_ie( gold_assert(r_type == elfcpp::R_X86_64_TLSDESC_CALL); tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, 2); tls::check_tls(relinfo, relnum, rela.get_r_offset(), - view[0] == 0xff && view[1] == 0x10); + view[0] == 0xff && view[1] == 0x10); view[0] = 0x66; view[1] = 0x90; } @@ -3655,7 +5504,7 @@ Target_x86_64::Relocate::tls_desc_gd_to_le( tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, -3); tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, 4); tls::check_tls(relinfo, relnum, rela.get_r_offset(), - view[-3] == 0x48 && view[-2] == 0x8d && view[-1] == 0x05); + view[-3] == 0x48 && view[-2] == 0x8d && view[-1] == 0x05); view[-2] = 0xc7; view[-1] = 0xc0; value -= tls_segment->memsz(); @@ -3668,7 +5517,7 @@ Target_x86_64::Relocate::tls_desc_gd_to_le( gold_assert(r_type == elfcpp::R_X86_64_TLSDESC_CALL); tls::check_range(relinfo, relnum, rela.get_r_offset(), view_size, 2); tls::check_tls(relinfo, relnum, rela.get_r_offset(), - view[0] == 0xff && view[1] == 0x10); + view[0] == 0xff && view[1] == 0x10); view[0] = 0x66; view[1] = 0x90; } @@ -3687,18 +5536,45 @@ 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 + // leaq foo@tlsld(%rip),%rdi; call *__tls_get_addr@GOTPCREL(%rip) + // For SIZE == 64: + // ... leq foo@dtpoff(%rax),%reg + // ==> .word 0x6666; .byte 0x6666; movq %fs:0,%rax ... leaq x@tpoff(%rax),%rdx + // For SIZE == 32: + // ... leq foo@dtpoff(%rax),%reg + // ==> nopw 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); tls::check_tls(relinfo, relnum, rela.get_r_offset(), - view[-3] == 0x48 && view[-2] == 0x8d && view[-1] == 0x3d); + view[-3] == 0x48 && view[-2] == 0x8d && view[-1] == 0x3d); - tls::check_tls(relinfo, relnum, rela.get_r_offset(), view[4] == 0xe8); + tls::check_tls(relinfo, relnum, rela.get_r_offset(), + view[4] == 0xe8 || view[4] == 0xff); - memcpy(view - 3, "\x66\x66\x66\x64\x48\x8b\x04\x25\0\0\0\0", 12); + if (view[4] == 0xe8) + { + 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); + } + else + { + if (size == 64) + memcpy(view - 3, "\x66\x66\x66\x66\x64\x48\x8b\x04\x25\0\0\0\0", + 13); + else + memcpy(view - 3, "\x66\x0f\x1f\x40\x00\x64\x8b\x04\x25\0\0\0\0", + 13); + } // The next reloc should be a PLT32 reloc against __tls_get_addr. // We can skip it. @@ -3738,7 +5614,9 @@ Target_x86_64::Relocate::tls_ie_to_le( { // movq if (op1 == 0x4c) - view[-3] = 0x49; + view[-3] = 0x49; + else if (size == 32 && op1 == 0x44) + view[-3] = 0x41; view[-2] = 0xc7; view[-1] = 0xc0 | reg; } @@ -3746,7 +5624,9 @@ Target_x86_64::Relocate::tls_ie_to_le( { // Special handling for %rsp. if (op1 == 0x4c) - view[-3] = 0x49; + view[-3] = 0x49; + else if (size == 32 && op1 == 0x44) + view[-3] = 0x41; view[-2] = 0x81; view[-1] = 0xc0 | reg; } @@ -3754,12 +5634,15 @@ Target_x86_64::Relocate::tls_ie_to_le( { // addq if (op1 == 0x4c) - view[-3] = 0x4d; + 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); } @@ -3779,10 +5662,13 @@ 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, @@ -3823,82 +5709,50 @@ 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; + typedef gold::Default_classify_reloc + Classify_reloc; + typedef gold::Default_scan_relocatable_relocs + Scan_relocatable_relocs; - 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; - - 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, @@ -3909,16 +5763,17 @@ 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, @@ -3928,7 +5783,7 @@ Target_x86_64::scan_relocatable_relocs( output_section, needs_special_offset_handling, local_symbol_count, - plocal_symbols, + plocal_syms, rr); } @@ -3936,29 +5791,30 @@ Target_x86_64::scan_relocatable_relocs( template void -Target_x86_64::relocate_for_relocatable( +Target_x86_64::relocate_relocs( const Relocate_info* relinfo, unsigned int sh_type, 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_for_relocatable( + gold::relocate_relocs( relinfo, prelocs, reloc_count, output_section, offset_in_output_section, - rr, view, view_address, view_size, @@ -3976,7 +5832,7 @@ 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 @@ -3993,7 +5849,7 @@ Target_x86_64::do_code_fill(section_size_type length) const jmp[0] = 0xe9; elfcpp::Swap_unaligned<32, false>::writeval(jmp + 1, length - 5); return (std::string(reinterpret_cast(&jmp[0]), 5) - + std::string(length - 5, static_cast(0x90))); + + std::string(length - 5, static_cast(0x90))); } // Nop sequences of various lengths. @@ -4001,47 +5857,47 @@ Target_x86_64::do_code_fill(section_size_type length) const const char nop2[2] = { '\x66', '\x90' }; // xchg %ax %ax const char nop3[3] = { '\x0f', '\x1f', '\x00' }; // nop (%rax) const char nop4[4] = { '\x0f', '\x1f', '\x40', // nop 0(%rax) - '\x00'}; + '\x00'}; const char nop5[5] = { '\x0f', '\x1f', '\x44', // nop 0(%rax,%rax,1) '\x00', '\x00' }; const char nop6[6] = { '\x66', '\x0f', '\x1f', // nopw 0(%rax,%rax,1) - '\x44', '\x00', '\x00' }; + '\x44', '\x00', '\x00' }; const char nop7[7] = { '\x0f', '\x1f', '\x80', // nopl 0L(%rax) - '\x00', '\x00', '\x00', + '\x00', '\x00', '\x00', '\x00' }; const char nop8[8] = { '\x0f', '\x1f', '\x84', // nopl 0L(%rax,%rax,1) - '\x00', '\x00', '\x00', + '\x00', '\x00', '\x00', '\x00', '\x00' }; const char nop9[9] = { '\x66', '\x0f', '\x1f', // nopw 0L(%rax,%rax,1) - '\x84', '\x00', '\x00', + '\x84', '\x00', '\x00', '\x00', '\x00', '\x00' }; const char nop10[10] = { '\x66', '\x2e', '\x0f', // nopw %cs:0L(%rax,%rax,1) - '\x1f', '\x84', '\x00', + '\x1f', '\x84', '\x00', '\x00', '\x00', '\x00', '\x00' }; const char nop11[11] = { '\x66', '\x66', '\x2e', // data16 - '\x0f', '\x1f', '\x84', // nopw %cs:0L(%rax,%rax,1) + '\x0f', '\x1f', '\x84', // nopw %cs:0L(%rax,%rax,1) '\x00', '\x00', '\x00', '\x00', '\x00' }; const char nop12[12] = { '\x66', '\x66', '\x66', // data16; data16 - '\x2e', '\x0f', '\x1f', // nopw %cs:0L(%rax,%rax,1) + '\x2e', '\x0f', '\x1f', // nopw %cs:0L(%rax,%rax,1) '\x84', '\x00', '\x00', '\x00', '\x00', '\x00' }; const char nop13[13] = { '\x66', '\x66', '\x66', // data16; data16; data16 - '\x66', '\x2e', '\x0f', // nopw %cs:0L(%rax,%rax,1) + '\x66', '\x2e', '\x0f', // nopw %cs:0L(%rax,%rax,1) '\x1f', '\x84', '\x00', '\x00', '\x00', '\x00', - '\x00' }; + '\x00' }; const char nop14[14] = { '\x66', '\x66', '\x66', // data16; data16; data16 - '\x66', '\x66', '\x2e', // data16 + '\x66', '\x66', '\x2e', // data16 '\x0f', '\x1f', '\x84', // nopw %cs:0L(%rax,%rax,1) '\x00', '\x00', '\x00', - '\x00', '\x00' }; + '\x00', '\x00' }; const char nop15[15] = { '\x66', '\x66', '\x66', // data16; data16; data16 - '\x66', '\x66', '\x66', // data16; data16 + '\x66', '\x66', '\x66', // data16; data16 '\x2e', '\x0f', '\x1f', // nopw %cs:0L(%rax,%rax,1) '\x84', '\x00', '\x00', - '\x00', '\x00', '\x00' }; + '\x00', '\x00', '\x00' }; const char* nops[16] = { NULL, @@ -4093,35 +5949,60 @@ 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 @@ -4150,7 +6031,8 @@ Target_x86_64::do_calls_non_split(Relobj* object, unsigned int shndx, *to = "__morestack_non_split"; } -// The selector for x86_64 object files. +// The selector for x86_64 object files. Note this is never instantiated +// directly. It's only used in Target_selector_x86_64_nacl, below. template class Target_selector_x86_64 : public Target_selector_freebsd @@ -4158,9 +6040,9 @@ class Target_selector_x86_64 : public Target_selector_freebsd public: Target_selector_x86_64() : Target_selector_freebsd(elfcpp::EM_X86_64, size, false, - (size == 64 + (size == 64 ? "elf64-x86-64" : "elf32-x86-64"), - (size == 64 + (size == 64 ? "elf64-x86-64-freebsd" : "elf32-x86-64-freebsd"), (size == 64 ? "elf_x86_64" : "elf32_x86_64")) @@ -4172,7 +6054,378 @@ public: }; -Target_selector_x86_64<64> target_selector_x86_64; -Target_selector_x86_64<32> target_selector_x32; +// NaCl variant. It uses different PLT contents. + +template +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_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) + { } + + Output_data_plt_x86_64_nacl(Layout* layout, + Output_data_got<64, false>* got, + 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, + got, got_plt, got_irelative, + plt_count) + { } + + protected: + virtual unsigned int + do_get_plt_entry_size() const + { return plt_entry_size; } + + virtual void + do_add_eh_frame(Layout* layout) + { + layout->add_eh_frame_for_plt(this, + this->plt_eh_frame_cie, + this->plt_eh_frame_cie_size, + plt_eh_frame_fde, + plt_eh_frame_fde_size); + } + + virtual void + do_fill_first_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_addr, + typename elfcpp::Elf_types::Elf_Addr plt_addr); + + virtual unsigned int + do_fill_plt_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index); + + virtual void + do_fill_tlsdesc_entry(unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + typename elfcpp::Elf_types::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset); + + private: + // The size of an entry in the PLT. + static const int plt_entry_size = 64; + + // The first entry in the PLT. + static const unsigned char first_plt_entry[plt_entry_size]; + + // Other entries in the PLT for an executable. + static const unsigned char plt_entry[plt_entry_size]; + + // The reserved TLSDESC entry in the PLT for an executable. + static const unsigned char tlsdesc_plt_entry[plt_entry_size]; + + // The .eh_frame unwind information for the PLT. + static const int plt_eh_frame_fde_size = 32; + static const unsigned char plt_eh_frame_fde[plt_eh_frame_fde_size]; +}; + +template +class Target_x86_64_nacl : public Target_x86_64 +{ + public: + Target_x86_64_nacl() + : Target_x86_64(&x86_64_nacl_info) + { } + + virtual Output_data_plt_x86_64* + do_make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + 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, + got_irelative); + } + + virtual Output_data_plt_x86_64* + do_make_data_plt(Layout* layout, + Output_data_got<64, false>* got, + Output_data_got_plt_x86_64* got_plt, + Output_data_space* got_irelative, + unsigned int plt_count) + { + return new Output_data_plt_x86_64_nacl(layout, got, got_plt, + got_irelative, + plt_count); + } + + virtual std::string + do_code_fill(section_size_type length) const; + + private: + static const Target::Target_info x86_64_nacl_info; +}; + +template<> +const Target::Target_info Target_x86_64_nacl<64>::x86_64_nacl_info = +{ + 64, // size + false, // is_big_endian + elfcpp::EM_X86_64, // machine_code + false, // has_make_symbol + false, // has_resolve + true, // has_code_fill + true, // is_default_stack_executable + true, // can_icf_inline_merge_sections + '\0', // wrap_char + "/lib64/ld-nacl-x86-64.so.1", // dynamic_linker + 0x20000, // default_text_segment_address + 0x10000, // abi_pagesize (overridable by -z max-page-size) + 0x10000, // common_pagesize (overridable by -z common-page-size) + true, // isolate_execinstr + 0x10000000, // rosegment_gap + elfcpp::SHN_UNDEF, // small_common_shndx + elfcpp::SHN_X86_64_LCOMMON, // large_common_shndx + 0, // small_common_section_flags + elfcpp::SHF_X86_64_LARGE, // large_common_section_flags + NULL, // attributes_section + NULL, // attributes_vendor + "_start", // entry_symbol_name + 32, // hash_entry_size + elfcpp::SHT_X86_64_UNWIND, // unwind_section_type +}; + +template<> +const Target::Target_info Target_x86_64_nacl<32>::x86_64_nacl_info = +{ + 32, // size + false, // is_big_endian + elfcpp::EM_X86_64, // machine_code + false, // has_make_symbol + false, // has_resolve + true, // has_code_fill + true, // is_default_stack_executable + true, // can_icf_inline_merge_sections + '\0', // wrap_char + "/lib/ld-nacl-x86-64.so.1", // dynamic_linker + 0x20000, // default_text_segment_address + 0x10000, // abi_pagesize (overridable by -z max-page-size) + 0x10000, // common_pagesize (overridable by -z common-page-size) + true, // isolate_execinstr + 0x10000000, // rosegment_gap + elfcpp::SHN_UNDEF, // small_common_shndx + elfcpp::SHN_X86_64_LCOMMON, // large_common_shndx + 0, // small_common_section_flags + elfcpp::SHF_X86_64_LARGE, // large_common_section_flags + NULL, // attributes_section + NULL, // attributes_vendor + "_start", // entry_symbol_name + 32, // hash_entry_size + elfcpp::SHT_X86_64_UNWIND, // unwind_section_type +}; + +#define NACLMASK 0xe0 // 32-byte alignment mask. + +// The first entry in the PLT. + +template +const unsigned char +Output_data_plt_x86_64_nacl::first_plt_entry[plt_entry_size] = +{ + 0xff, 0x35, // pushq contents of memory address + 0, 0, 0, 0, // replaced with address of .got + 8 + 0x4c, 0x8b, 0x1d, // mov GOT+16(%rip), %r11 + 0, 0, 0, 0, // replaced with address of .got + 16 + 0x41, 0x83, 0xe3, NACLMASK, // and $-32, %r11d + 0x4d, 0x01, 0xfb, // add %r15, %r11 + 0x41, 0xff, 0xe3, // jmpq *%r11 + + // 9-byte nop sequence to pad out to the next 32-byte boundary. + 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 + 0x2e, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopw %cs:0x0(%rax,%rax,1) + 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, // excess data32 prefixes + 0x2e, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopw %cs:0x0(%rax,%rax,1) + 0x66, // excess data32 prefix + 0x90 // nop +}; + +template +void +Output_data_plt_x86_64_nacl::do_fill_first_plt_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address) +{ + memcpy(pov, first_plt_entry, plt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, + (got_address + 8 + - (plt_address + 2 + 4))); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 9, + (got_address + 16 + - (plt_address + 9 + 4))); +} + +// Subsequent entries in the PLT. + +template +const unsigned char +Output_data_plt_x86_64_nacl::plt_entry[plt_entry_size] = +{ + 0x4c, 0x8b, 0x1d, // mov name@GOTPCREL(%rip),%r11 + 0, 0, 0, 0, // replaced with address of symbol in .got + 0x41, 0x83, 0xe3, NACLMASK, // and $-32, %r11d + 0x4d, 0x01, 0xfb, // add %r15, %r11 + 0x41, 0xff, 0xe3, // jmpq *%r11 + + // 15-byte nop sequence to pad out to the next 32-byte boundary. + 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, // excess data32 prefixes + 0x2e, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopw %cs:0x0(%rax,%rax,1) + + // Lazy GOT entries point here (32-byte aligned). + 0x68, // pushq immediate + 0, 0, 0, 0, // replaced with index into relocation table + 0xe9, // jmp relative + 0, 0, 0, 0, // replaced with offset to start of .plt0 + + // 22 bytes of nop to pad out to the standard size. + 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, // excess data32 prefixes + 0x2e, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopw %cs:0x0(%rax,%rax,1) + 0x0f, 0x1f, 0x80, 0, 0, 0, 0, // nopl 0x0(%rax) +}; + +template +unsigned int +Output_data_plt_x86_64_nacl::do_fill_plt_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + unsigned int got_offset, + unsigned int plt_offset, + unsigned int plt_index) +{ + memcpy(pov, plt_entry, plt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 3, + (got_address + got_offset + - (plt_address + plt_offset + + 3 + 4))); + + elfcpp::Swap_unaligned<32, false>::writeval(pov + 33, plt_index); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 38, + - (plt_offset + 38 + 4)); + + return 32; +} + +// The reserved TLSDESC entry in the PLT. + +template +const unsigned char +Output_data_plt_x86_64_nacl::tlsdesc_plt_entry[plt_entry_size] = +{ + 0xff, 0x35, // pushq x(%rip) + 0, 0, 0, 0, // replaced with address of linkmap GOT entry (at PLTGOT + 8) + 0x4c, 0x8b, 0x1d, // mov y(%rip),%r11 + 0, 0, 0, 0, // replaced with offset of reserved TLSDESC_GOT entry + 0x41, 0x83, 0xe3, NACLMASK, // and $-32, %r11d + 0x4d, 0x01, 0xfb, // add %r15, %r11 + 0x41, 0xff, 0xe3, // jmpq *%r11 + + // 41 bytes of nop to pad out to the standard size. + 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, // excess data32 prefixes + 0x2e, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopw %cs:0x0(%rax,%rax,1) + 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, // excess data32 prefixes + 0x2e, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopw %cs:0x0(%rax,%rax,1) + 0x66, 0x66, // excess data32 prefixes + 0x2e, 0x0f, 0x1f, 0x84, 0, 0, 0, 0, 0, // nopw %cs:0x0(%rax,%rax,1) +}; + +template +void +Output_data_plt_x86_64_nacl::do_fill_tlsdesc_entry( + unsigned char* pov, + typename elfcpp::Elf_types::Elf_Addr got_address, + typename elfcpp::Elf_types::Elf_Addr plt_address, + typename elfcpp::Elf_types::Elf_Addr got_base, + unsigned int tlsdesc_got_offset, + unsigned int plt_offset) +{ + memcpy(pov, tlsdesc_plt_entry, plt_entry_size); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 2, + (got_address + 8 + - (plt_address + plt_offset + + 2 + 4))); + elfcpp::Swap_unaligned<32, false>::writeval(pov + 9, + (got_base + + tlsdesc_got_offset + - (plt_address + plt_offset + + 9 + 4))); +} + +// The .eh_frame unwind information for the PLT. + +template +const unsigned char +Output_data_plt_x86_64_nacl::plt_eh_frame_fde[plt_eh_frame_fde_size] = +{ + 0, 0, 0, 0, // Replaced with offset to .plt. + 0, 0, 0, 0, // Replaced with size of .plt. + 0, // Augmentation size. + elfcpp::DW_CFA_def_cfa_offset, 16, // DW_CFA_def_cfa_offset: 16. + elfcpp::DW_CFA_advance_loc + 6, // Advance 6 to __PLT__ + 6. + elfcpp::DW_CFA_def_cfa_offset, 24, // DW_CFA_def_cfa_offset: 24. + elfcpp::DW_CFA_advance_loc + 58, // Advance 58 to __PLT__ + 64. + elfcpp::DW_CFA_def_cfa_expression, // DW_CFA_def_cfa_expression. + 13, // Block length. + elfcpp::DW_OP_breg7, 8, // Push %rsp + 8. + elfcpp::DW_OP_breg16, 0, // Push %rip. + elfcpp::DW_OP_const1u, 63, // Push 0x3f. + elfcpp::DW_OP_and, // & (%rip & 0x3f). + elfcpp::DW_OP_const1u, 37, // Push 0x25. + elfcpp::DW_OP_ge, // >= ((%rip & 0x3f) >= 0x25) + elfcpp::DW_OP_lit3, // Push 3. + elfcpp::DW_OP_shl, // << (((%rip & 0x3f) >= 0x25) << 3) + elfcpp::DW_OP_plus, // + ((((%rip&0x3f)>=0x25)<<3)+%rsp+8 + elfcpp::DW_CFA_nop, // Align to 32 bytes. + 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 +class Target_selector_x86_64_nacl + : public Target_selector_nacl, + Target_x86_64_nacl > +{ + public: + Target_selector_x86_64_nacl() + : Target_selector_nacl, + Target_x86_64_nacl >("x86-64", + size == 64 + ? "elf64-x86-64-nacl" + : "elf32-x86-64-nacl", + size == 64 + ? "elf_x86_64_nacl" + : "elf32_x86_64_nacl") + { } +}; + +Target_selector_x86_64_nacl<64> target_selector_x86_64; +Target_selector_x86_64_nacl<32> target_selector_x32; } // End anonymous namespace.