URL
https://opencores.org/ocsvn/open8_urisc/open8_urisc/trunk
Subversion Repositories open8_urisc
[/] [open8_urisc/] [trunk/] [gnu/] [binutils/] [gold/] [arm.cc] - Rev 84
Go to most recent revision | Compare with Previous | Blame | View Log
// arm.cc -- arm target support for gold. // Copyright 2009, 2010 Free Software Foundation, Inc. // Written by Doug Kwan <dougkwan@google.com> based on the i386 code // by Ian Lance Taylor <iant@google.com>. // This file also contains borrowed and adapted code from // bfd/elf32-arm.c. // This file is part of gold. // This program is free software; you can redistribute it and/or modify // it under the terms of the GNU General Public License as published by // the Free Software Foundation; either version 3 of the License, or // (at your option) any later version. // This program is distributed in the hope that it will be useful, // but WITHOUT ANY WARRANTY; without even the implied warranty of // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the // GNU General Public License for more details. // You should have received a copy of the GNU General Public License // along with this program; if not, write to the Free Software // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, // MA 02110-1301, USA. #include "gold.h" #include <cstring> #include <limits> #include <cstdio> #include <string> #include <algorithm> #include <map> #include <utility> #include <set> #include "elfcpp.h" #include "parameters.h" #include "reloc.h" #include "arm.h" #include "object.h" #include "symtab.h" #include "layout.h" #include "output.h" #include "copy-relocs.h" #include "target.h" #include "target-reloc.h" #include "target-select.h" #include "tls.h" #include "defstd.h" #include "gc.h" #include "attributes.h" #include "arm-reloc-property.h" namespace { using namespace gold; template<bool big_endian> class Output_data_plt_arm; template<bool big_endian> class Stub_table; template<bool big_endian> class Arm_input_section; class Arm_exidx_cantunwind; class Arm_exidx_merged_section; class Arm_exidx_fixup; template<bool big_endian> class Arm_output_section; class Arm_exidx_input_section; template<bool big_endian> class Arm_relobj; template<bool big_endian> class Arm_relocate_functions; template<bool big_endian> class Arm_output_data_got; template<bool big_endian> class Target_arm; // For convenience. typedef elfcpp::Elf_types<32>::Elf_Addr Arm_address; // Maximum branch offsets for ARM, THUMB and THUMB2. const int32_t ARM_MAX_FWD_BRANCH_OFFSET = ((((1 << 23) - 1) << 2) + 8); const int32_t ARM_MAX_BWD_BRANCH_OFFSET = ((-((1 << 23) << 2)) + 8); const int32_t THM_MAX_FWD_BRANCH_OFFSET = ((1 << 22) -2 + 4); const int32_t THM_MAX_BWD_BRANCH_OFFSET = (-(1 << 22) + 4); const int32_t THM2_MAX_FWD_BRANCH_OFFSET = (((1 << 24) - 2) + 4); const int32_t THM2_MAX_BWD_BRANCH_OFFSET = (-(1 << 24) + 4); // Thread Control Block size. const size_t ARM_TCB_SIZE = 8; // The arm target class. // // This is a very simple port of gold for ARM-EABI. It is intended for // supporting Android only for the time being. // // TODOs: // - Implement all static relocation types documented in arm-reloc.def. // - Make PLTs more flexible for different architecture features like // Thumb-2 and BE8. // There are probably a lot more. // Ideally we would like to avoid using global variables but this is used // very in many places and sometimes in loops. If we use a function // returning a static instance of Arm_reloc_property_table, it will be very // slow in an threaded environment since the static instance needs to be // locked. The pointer is below initialized in the // Target::do_select_as_default_target() hook so that we do not spend time // building the table if we are not linking ARM objects. // // An alternative is to to process the information in arm-reloc.def in // compilation time and generate a representation of it in PODs only. That // way we can avoid initialization when the linker starts. Arm_reloc_property_table* arm_reloc_property_table = NULL; // Instruction template class. This class is similar to the insn_sequence // struct in bfd/elf32-arm.c. class Insn_template { public: // Types of instruction templates. enum Type { THUMB16_TYPE = 1, // THUMB16_SPECIAL_TYPE is used by sub-classes of Stub for instruction // templates with class-specific semantics. Currently this is used // only by the Cortex_a8_stub class for handling condition codes in // conditional branches. THUMB16_SPECIAL_TYPE, THUMB32_TYPE, ARM_TYPE, DATA_TYPE }; // Factory methods to create instruction templates in different formats. static const Insn_template thumb16_insn(uint32_t data) { return Insn_template(data, THUMB16_TYPE, elfcpp::R_ARM_NONE, 0); } // A Thumb conditional branch, in which the proper condition is inserted // when we build the stub. static const Insn_template thumb16_bcond_insn(uint32_t data) { return Insn_template(data, THUMB16_SPECIAL_TYPE, elfcpp::R_ARM_NONE, 1); } static const Insn_template thumb32_insn(uint32_t data) { return Insn_template(data, THUMB32_TYPE, elfcpp::R_ARM_NONE, 0); } static const Insn_template thumb32_b_insn(uint32_t data, int reloc_addend) { return Insn_template(data, THUMB32_TYPE, elfcpp::R_ARM_THM_JUMP24, reloc_addend); } static const Insn_template arm_insn(uint32_t data) { return Insn_template(data, ARM_TYPE, elfcpp::R_ARM_NONE, 0); } static const Insn_template arm_rel_insn(unsigned data, int reloc_addend) { return Insn_template(data, ARM_TYPE, elfcpp::R_ARM_JUMP24, reloc_addend); } static const Insn_template data_word(unsigned data, unsigned int r_type, int reloc_addend) { return Insn_template(data, DATA_TYPE, r_type, reloc_addend); } // Accessors. This class is used for read-only objects so no modifiers // are provided. uint32_t data() const { return this->data_; } // Return the instruction sequence type of this. Type type() const { return this->type_; } // Return the ARM relocation type of this. unsigned int r_type() const { return this->r_type_; } int32_t reloc_addend() const { return this->reloc_addend_; } // Return size of instruction template in bytes. size_t size() const; // Return byte-alignment of instruction template. unsigned alignment() const; private: // We make the constructor private to ensure that only the factory // methods are used. inline Insn_template(unsigned data, Type type, unsigned int r_type, int reloc_addend) : data_(data), type_(type), r_type_(r_type), reloc_addend_(reloc_addend) { } // Instruction specific data. This is used to store information like // some of the instruction bits. uint32_t data_; // Instruction template type. Type type_; // Relocation type if there is a relocation or R_ARM_NONE otherwise. unsigned int r_type_; // Relocation addend. int32_t reloc_addend_; }; // Macro for generating code to stub types. One entry per long/short // branch stub #define DEF_STUBS \ DEF_STUB(long_branch_any_any) \ DEF_STUB(long_branch_v4t_arm_thumb) \ DEF_STUB(long_branch_thumb_only) \ DEF_STUB(long_branch_v4t_thumb_thumb) \ DEF_STUB(long_branch_v4t_thumb_arm) \ DEF_STUB(short_branch_v4t_thumb_arm) \ DEF_STUB(long_branch_any_arm_pic) \ DEF_STUB(long_branch_any_thumb_pic) \ DEF_STUB(long_branch_v4t_thumb_thumb_pic) \ DEF_STUB(long_branch_v4t_arm_thumb_pic) \ DEF_STUB(long_branch_v4t_thumb_arm_pic) \ DEF_STUB(long_branch_thumb_only_pic) \ DEF_STUB(a8_veneer_b_cond) \ DEF_STUB(a8_veneer_b) \ DEF_STUB(a8_veneer_bl) \ DEF_STUB(a8_veneer_blx) \ DEF_STUB(v4_veneer_bx) // Stub types. #define DEF_STUB(x) arm_stub_##x, typedef enum { arm_stub_none, DEF_STUBS // First reloc stub type. arm_stub_reloc_first = arm_stub_long_branch_any_any, // Last reloc stub type. arm_stub_reloc_last = arm_stub_long_branch_thumb_only_pic, // First Cortex-A8 stub type. arm_stub_cortex_a8_first = arm_stub_a8_veneer_b_cond, // Last Cortex-A8 stub type. arm_stub_cortex_a8_last = arm_stub_a8_veneer_blx, // Last stub type. arm_stub_type_last = arm_stub_v4_veneer_bx } Stub_type; #undef DEF_STUB // Stub template class. Templates are meant to be read-only objects. // A stub template for a stub type contains all read-only attributes // common to all stubs of the same type. class Stub_template { public: Stub_template(Stub_type, const Insn_template*, size_t); ~Stub_template() { } // Return stub type. Stub_type type() const { return this->type_; } // Return an array of instruction templates. const Insn_template* insns() const { return this->insns_; } // Return size of template in number of instructions. size_t insn_count() const { return this->insn_count_; } // Return size of template in bytes. size_t size() const { return this->size_; } // Return alignment of the stub template. unsigned alignment() const { return this->alignment_; } // Return whether entry point is in thumb mode. bool entry_in_thumb_mode() const { return this->entry_in_thumb_mode_; } // Return number of relocations in this template. size_t reloc_count() const { return this->relocs_.size(); } // Return index of the I-th instruction with relocation. size_t reloc_insn_index(size_t i) const { gold_assert(i < this->relocs_.size()); return this->relocs_[i].first; } // Return the offset of the I-th instruction with relocation from the // beginning of the stub. section_size_type reloc_offset(size_t i) const { gold_assert(i < this->relocs_.size()); return this->relocs_[i].second; } private: // This contains information about an instruction template with a relocation // and its offset from start of stub. typedef std::pair<size_t, section_size_type> Reloc; // A Stub_template may not be copied. We want to share templates as much // as possible. Stub_template(const Stub_template&); Stub_template& operator=(const Stub_template&); // Stub type. Stub_type type_; // Points to an array of Insn_templates. const Insn_template* insns_; // Number of Insn_templates in insns_[]. size_t insn_count_; // Size of templated instructions in bytes. size_t size_; // Alignment of templated instructions. unsigned alignment_; // Flag to indicate if entry is in thumb mode. bool entry_in_thumb_mode_; // A table of reloc instruction indices and offsets. We can find these by // looking at the instruction templates but we pre-compute and then stash // them here for speed. std::vector<Reloc> relocs_; }; // // A class for code stubs. This is a base class for different type of // stubs used in the ARM target. // class Stub { private: static const section_offset_type invalid_offset = static_cast<section_offset_type>(-1); public: Stub(const Stub_template* stub_template) : stub_template_(stub_template), offset_(invalid_offset) { } virtual ~Stub() { } // Return the stub template. const Stub_template* stub_template() const { return this->stub_template_; } // Return offset of code stub from beginning of its containing stub table. section_offset_type offset() const { gold_assert(this->offset_ != invalid_offset); return this->offset_; } // Set offset of code stub from beginning of its containing stub table. void set_offset(section_offset_type offset) { this->offset_ = offset; } // Return the relocation target address of the i-th relocation in the // stub. This must be defined in a child class. Arm_address reloc_target(size_t i) { return this->do_reloc_target(i); } // Write a stub at output VIEW. BIG_ENDIAN select how a stub is written. void write(unsigned char* view, section_size_type view_size, bool big_endian) { this->do_write(view, view_size, big_endian); } // Return the instruction for THUMB16_SPECIAL_TYPE instruction template // for the i-th instruction. uint16_t thumb16_special(size_t i) { return this->do_thumb16_special(i); } protected: // This must be defined in the child class. virtual Arm_address do_reloc_target(size_t) = 0; // This may be overridden in the child class. virtual void do_write(unsigned char* view, section_size_type view_size, bool big_endian) { if (big_endian) this->do_fixed_endian_write<true>(view, view_size); else this->do_fixed_endian_write<false>(view, view_size); } // This must be overridden if a child class uses the THUMB16_SPECIAL_TYPE // instruction template. virtual uint16_t do_thumb16_special(size_t) { gold_unreachable(); } private: // A template to implement do_write. template<bool big_endian> void inline do_fixed_endian_write(unsigned char*, section_size_type); // Its template. const Stub_template* stub_template_; // Offset within the section of containing this stub. section_offset_type offset_; }; // Reloc stub class. These are stubs we use to fix up relocation because // of limited branch ranges. class Reloc_stub : public Stub { public: static const unsigned int invalid_index = static_cast<unsigned int>(-1); // We assume we never jump to this address. static const Arm_address invalid_address = static_cast<Arm_address>(-1); // Return destination address. Arm_address destination_address() const { gold_assert(this->destination_address_ != this->invalid_address); return this->destination_address_; } // Set destination address. void set_destination_address(Arm_address address) { gold_assert(address != this->invalid_address); this->destination_address_ = address; } // Reset destination address. void reset_destination_address() { this->destination_address_ = this->invalid_address; } // Determine stub type for a branch of a relocation of R_TYPE going // from BRANCH_ADDRESS to BRANCH_TARGET. If TARGET_IS_THUMB is set, // the branch target is a thumb instruction. TARGET is used for look // up ARM-specific linker settings. static Stub_type stub_type_for_reloc(unsigned int r_type, Arm_address branch_address, Arm_address branch_target, bool target_is_thumb); // Reloc_stub key. A key is logically a triplet of a stub type, a symbol // and an addend. Since we treat global and local symbol differently, we // use a Symbol object for a global symbol and a object-index pair for // a local symbol. class Key { public: // If SYMBOL is not null, this is a global symbol, we ignore RELOBJ and // R_SYM. Otherwise, this is a local symbol and RELOBJ must non-NULL // and R_SYM must not be invalid_index. Key(Stub_type stub_type, const Symbol* symbol, const Relobj* relobj, unsigned int r_sym, int32_t addend) : stub_type_(stub_type), addend_(addend) { if (symbol != NULL) { this->r_sym_ = Reloc_stub::invalid_index; this->u_.symbol = symbol; } else { gold_assert(relobj != NULL && r_sym != invalid_index); this->r_sym_ = r_sym; this->u_.relobj = relobj; } } ~Key() { } // Accessors: Keys are meant to be read-only object so no modifiers are // provided. // Return stub type. Stub_type stub_type() const { return this->stub_type_; } // Return the local symbol index or invalid_index. unsigned int r_sym() const { return this->r_sym_; } // Return the symbol if there is one. const Symbol* symbol() const { return this->r_sym_ == invalid_index ? this->u_.symbol : NULL; } // Return the relobj if there is one. const Relobj* relobj() const { return this->r_sym_ != invalid_index ? this->u_.relobj : NULL; } // Whether this equals to another key k. bool eq(const Key& k) const { return ((this->stub_type_ == k.stub_type_) && (this->r_sym_ == k.r_sym_) && ((this->r_sym_ != Reloc_stub::invalid_index) ? (this->u_.relobj == k.u_.relobj) : (this->u_.symbol == k.u_.symbol)) && (this->addend_ == k.addend_)); } // Return a hash value. size_t hash_value() const { return (this->stub_type_ ^ this->r_sym_ ^ gold::string_hash<char>( (this->r_sym_ != Reloc_stub::invalid_index) ? this->u_.relobj->name().c_str() : this->u_.symbol->name()) ^ this->addend_); } // Functors for STL associative containers. struct hash { size_t operator()(const Key& k) const { return k.hash_value(); } }; struct equal_to { bool operator()(const Key& k1, const Key& k2) const { return k1.eq(k2); } }; // Name of key. This is mainly for debugging. std::string name() const; private: // Stub type. Stub_type stub_type_; // If this is a local symbol, this is the index in the defining object. // Otherwise, it is invalid_index for a global symbol. unsigned int r_sym_; // If r_sym_ is an invalid index, this points to a global symbol. // Otherwise, it points to a relobj. We used the unsized and target // independent Symbol and Relobj classes instead of Sized_symbol<32> and // Arm_relobj, in order to avoid making the stub class a template // as most of the stub machinery is endianness-neutral. However, it // may require a bit of casting done by users of this class. union { const Symbol* symbol; const Relobj* relobj; } u_; // Addend associated with a reloc. int32_t addend_; }; protected: // Reloc_stubs are created via a stub factory. So these are protected. Reloc_stub(const Stub_template* stub_template) : Stub(stub_template), destination_address_(invalid_address) { } ~Reloc_stub() { } friend class Stub_factory; // Return the relocation target address of the i-th relocation in the // stub. Arm_address do_reloc_target(size_t i) { // All reloc stub have only one relocation. gold_assert(i == 0); return this->destination_address_; } private: // Address of destination. Arm_address destination_address_; }; // Cortex-A8 stub class. We need a Cortex-A8 stub to redirect any 32-bit // THUMB branch that meets the following conditions: // // 1. The branch straddles across a page boundary. i.e. lower 12-bit of // branch address is 0xffe. // 2. The branch target address is in the same page as the first word of the // branch. // 3. The branch follows a 32-bit instruction which is not a branch. // // To do the fix up, we need to store the address of the branch instruction // and its target at least. We also need to store the original branch // instruction bits for the condition code in a conditional branch. The // condition code is used in a special instruction template. We also want // to identify input sections needing Cortex-A8 workaround quickly. We store // extra information about object and section index of the code section // containing a branch being fixed up. The information is used to mark // the code section when we finalize the Cortex-A8 stubs. // class Cortex_a8_stub : public Stub { public: ~Cortex_a8_stub() { } // Return the object of the code section containing the branch being fixed // up. Relobj* relobj() const { return this->relobj_; } // Return the section index of the code section containing the branch being // fixed up. unsigned int shndx() const { return this->shndx_; } // Return the source address of stub. This is the address of the original // branch instruction. LSB is 1 always set to indicate that it is a THUMB // instruction. Arm_address source_address() const { return this->source_address_; } // Return the destination address of the stub. This is the branch taken // address of the original branch instruction. LSB is 1 if it is a THUMB // instruction address. Arm_address destination_address() const { return this->destination_address_; } // Return the instruction being fixed up. uint32_t original_insn() const { return this->original_insn_; } protected: // Cortex_a8_stubs are created via a stub factory. So these are protected. Cortex_a8_stub(const Stub_template* stub_template, Relobj* relobj, unsigned int shndx, Arm_address source_address, Arm_address destination_address, uint32_t original_insn) : Stub(stub_template), relobj_(relobj), shndx_(shndx), source_address_(source_address | 1U), destination_address_(destination_address), original_insn_(original_insn) { } friend class Stub_factory; // Return the relocation target address of the i-th relocation in the // stub. Arm_address do_reloc_target(size_t i) { if (this->stub_template()->type() == arm_stub_a8_veneer_b_cond) { // The conditional branch veneer has two relocations. gold_assert(i < 2); return i == 0 ? this->source_address_ + 4 : this->destination_address_; } else { // All other Cortex-A8 stubs have only one relocation. gold_assert(i == 0); return this->destination_address_; } } // Return an instruction for the THUMB16_SPECIAL_TYPE instruction template. uint16_t do_thumb16_special(size_t); private: // Object of the code section containing the branch being fixed up. Relobj* relobj_; // Section index of the code section containing the branch begin fixed up. unsigned int shndx_; // Source address of original branch. Arm_address source_address_; // Destination address of the original branch. Arm_address destination_address_; // Original branch instruction. This is needed for copying the condition // code from a condition branch to its stub. uint32_t original_insn_; }; // ARMv4 BX Rx branch relocation stub class. class Arm_v4bx_stub : public Stub { public: ~Arm_v4bx_stub() { } // Return the associated register. uint32_t reg() const { return this->reg_; } protected: // Arm V4BX stubs are created via a stub factory. So these are protected. Arm_v4bx_stub(const Stub_template* stub_template, const uint32_t reg) : Stub(stub_template), reg_(reg) { } friend class Stub_factory; // Return the relocation target address of the i-th relocation in the // stub. Arm_address do_reloc_target(size_t) { gold_unreachable(); } // This may be overridden in the child class. virtual void do_write(unsigned char* view, section_size_type view_size, bool big_endian) { if (big_endian) this->do_fixed_endian_v4bx_write<true>(view, view_size); else this->do_fixed_endian_v4bx_write<false>(view, view_size); } private: // A template to implement do_write. template<bool big_endian> void inline do_fixed_endian_v4bx_write(unsigned char* view, section_size_type) { const Insn_template* insns = this->stub_template()->insns(); elfcpp::Swap<32, big_endian>::writeval(view, (insns[0].data() + (this->reg_ << 16))); view += insns[0].size(); elfcpp::Swap<32, big_endian>::writeval(view, (insns[1].data() + this->reg_)); view += insns[1].size(); elfcpp::Swap<32, big_endian>::writeval(view, (insns[2].data() + this->reg_)); } // A register index (r0-r14), which is associated with the stub. uint32_t reg_; }; // Stub factory class. class Stub_factory { public: // Return the unique instance of this class. static const Stub_factory& get_instance() { static Stub_factory singleton; return singleton; } // Make a relocation stub. Reloc_stub* make_reloc_stub(Stub_type stub_type) const { gold_assert(stub_type >= arm_stub_reloc_first && stub_type <= arm_stub_reloc_last); return new Reloc_stub(this->stub_templates_[stub_type]); } // Make a Cortex-A8 stub. Cortex_a8_stub* make_cortex_a8_stub(Stub_type stub_type, Relobj* relobj, unsigned int shndx, Arm_address source, Arm_address destination, uint32_t original_insn) const { gold_assert(stub_type >= arm_stub_cortex_a8_first && stub_type <= arm_stub_cortex_a8_last); return new Cortex_a8_stub(this->stub_templates_[stub_type], relobj, shndx, source, destination, original_insn); } // Make an ARM V4BX relocation stub. // This method creates a stub from the arm_stub_v4_veneer_bx template only. Arm_v4bx_stub* make_arm_v4bx_stub(uint32_t reg) const { gold_assert(reg < 0xf); return new Arm_v4bx_stub(this->stub_templates_[arm_stub_v4_veneer_bx], reg); } private: // Constructor and destructor are protected since we only return a single // instance created in Stub_factory::get_instance(). Stub_factory(); // A Stub_factory may not be copied since it is a singleton. Stub_factory(const Stub_factory&); Stub_factory& operator=(Stub_factory&); // Stub templates. These are initialized in the constructor. const Stub_template* stub_templates_[arm_stub_type_last+1]; }; // A class to hold stubs for the ARM target. template<bool big_endian> class Stub_table : public Output_data { public: Stub_table(Arm_input_section<big_endian>* owner) : Output_data(), owner_(owner), reloc_stubs_(), reloc_stubs_size_(0), reloc_stubs_addralign_(1), cortex_a8_stubs_(), arm_v4bx_stubs_(0xf), prev_data_size_(0), prev_addralign_(1) { } ~Stub_table() { } // Owner of this stub table. Arm_input_section<big_endian>* owner() const { return this->owner_; } // Whether this stub table is empty. bool empty() const { return (this->reloc_stubs_.empty() && this->cortex_a8_stubs_.empty() && this->arm_v4bx_stubs_.empty()); } // Return the current data size. off_t current_data_size() const { return this->current_data_size_for_child(); } // Add a STUB using KEY. The caller is responsible for avoiding addition // if a STUB with the same key has already been added. void add_reloc_stub(Reloc_stub* stub, const Reloc_stub::Key& key) { const Stub_template* stub_template = stub->stub_template(); gold_assert(stub_template->type() == key.stub_type()); this->reloc_stubs_[key] = stub; // Assign stub offset early. We can do this because we never remove // reloc stubs and they are in the beginning of the stub table. uint64_t align = stub_template->alignment(); this->reloc_stubs_size_ = align_address(this->reloc_stubs_size_, align); stub->set_offset(this->reloc_stubs_size_); this->reloc_stubs_size_ += stub_template->size(); this->reloc_stubs_addralign_ = std::max(this->reloc_stubs_addralign_, align); } // Add a Cortex-A8 STUB that fixes up a THUMB branch at ADDRESS. // The caller is responsible for avoiding addition if a STUB with the same // address has already been added. void add_cortex_a8_stub(Arm_address address, Cortex_a8_stub* stub) { std::pair<Arm_address, Cortex_a8_stub*> value(address, stub); this->cortex_a8_stubs_.insert(value); } // Add an ARM V4BX relocation stub. A register index will be retrieved // from the stub. void add_arm_v4bx_stub(Arm_v4bx_stub* stub) { gold_assert(stub != NULL && this->arm_v4bx_stubs_[stub->reg()] == NULL); this->arm_v4bx_stubs_[stub->reg()] = stub; } // Remove all Cortex-A8 stubs. void remove_all_cortex_a8_stubs(); // Look up a relocation stub using KEY. Return NULL if there is none. Reloc_stub* find_reloc_stub(const Reloc_stub::Key& key) const { typename Reloc_stub_map::const_iterator p = this->reloc_stubs_.find(key); return (p != this->reloc_stubs_.end()) ? p->second : NULL; } // Look up an arm v4bx relocation stub using the register index. // Return NULL if there is none. Arm_v4bx_stub* find_arm_v4bx_stub(const uint32_t reg) const { gold_assert(reg < 0xf); return this->arm_v4bx_stubs_[reg]; } // Relocate stubs in this stub table. void relocate_stubs(const Relocate_info<32, big_endian>*, Target_arm<big_endian>*, Output_section*, unsigned char*, Arm_address, section_size_type); // Update data size and alignment at the end of a relaxation pass. Return // true if either data size or alignment is different from that of the // previous relaxation pass. bool update_data_size_and_addralign(); // Finalize stubs. Set the offsets of all stubs and mark input sections // needing the Cortex-A8 workaround. void finalize_stubs(); // Apply Cortex-A8 workaround to an address range. void apply_cortex_a8_workaround_to_address_range(Target_arm<big_endian>*, unsigned char*, Arm_address, section_size_type); protected: // Write out section contents. void do_write(Output_file*); // Return the required alignment. uint64_t do_addralign() const { return this->prev_addralign_; } // Reset address and file offset. void do_reset_address_and_file_offset() { this->set_current_data_size_for_child(this->prev_data_size_); } // Set final data size. void set_final_data_size() { this->set_data_size(this->current_data_size()); } private: // Relocate one stub. void relocate_stub(Stub*, const Relocate_info<32, big_endian>*, Target_arm<big_endian>*, Output_section*, unsigned char*, Arm_address, section_size_type); // Unordered map of relocation stubs. typedef Unordered_map<Reloc_stub::Key, Reloc_stub*, Reloc_stub::Key::hash, Reloc_stub::Key::equal_to> Reloc_stub_map; // List of Cortex-A8 stubs ordered by addresses of branches being // fixed up in output. typedef std::map<Arm_address, Cortex_a8_stub*> Cortex_a8_stub_list; // List of Arm V4BX relocation stubs ordered by associated registers. typedef std::vector<Arm_v4bx_stub*> Arm_v4bx_stub_list; // Owner of this stub table. Arm_input_section<big_endian>* owner_; // The relocation stubs. Reloc_stub_map reloc_stubs_; // Size of reloc stubs. off_t reloc_stubs_size_; // Maximum address alignment of reloc stubs. uint64_t reloc_stubs_addralign_; // The cortex_a8_stubs. Cortex_a8_stub_list cortex_a8_stubs_; // The Arm V4BX relocation stubs. Arm_v4bx_stub_list arm_v4bx_stubs_; // data size of this in the previous pass. off_t prev_data_size_; // address alignment of this in the previous pass. uint64_t prev_addralign_; }; // Arm_exidx_cantunwind class. This represents an EXIDX_CANTUNWIND entry // we add to the end of an EXIDX input section that goes into the output. class Arm_exidx_cantunwind : public Output_section_data { public: Arm_exidx_cantunwind(Relobj* relobj, unsigned int shndx) : Output_section_data(8, 4, true), relobj_(relobj), shndx_(shndx) { } // Return the object containing the section pointed by this. Relobj* relobj() const { return this->relobj_; } // Return the section index of the section pointed by this. unsigned int shndx() const { return this->shndx_; } protected: void do_write(Output_file* of) { if (parameters->target().is_big_endian()) this->do_fixed_endian_write<true>(of); else this->do_fixed_endian_write<false>(of); } // Write to a map file. void do_print_to_mapfile(Mapfile* mapfile) const { mapfile->print_output_data(this, _("** ARM cantunwind")); } private: // Implement do_write for a given endianness. template<bool big_endian> void inline do_fixed_endian_write(Output_file*); // The object containing the section pointed by this. Relobj* relobj_; // The section index of the section pointed by this. unsigned int shndx_; }; // During EXIDX coverage fix-up, we compact an EXIDX section. The // Offset map is used to map input section offset within the EXIDX section // to the output offset from the start of this EXIDX section. typedef std::map<section_offset_type, section_offset_type> Arm_exidx_section_offset_map; // Arm_exidx_merged_section class. This represents an EXIDX input section // with some of its entries merged. class Arm_exidx_merged_section : public Output_relaxed_input_section { public: // Constructor for Arm_exidx_merged_section. // EXIDX_INPUT_SECTION points to the unmodified EXIDX input section. // SECTION_OFFSET_MAP points to a section offset map describing how // parts of the input section are mapped to output. DELETED_BYTES is // the number of bytes deleted from the EXIDX input section. Arm_exidx_merged_section( const Arm_exidx_input_section& exidx_input_section, const Arm_exidx_section_offset_map& section_offset_map, uint32_t deleted_bytes); // Build output contents. void build_contents(const unsigned char*, section_size_type); // Return the original EXIDX input section. const Arm_exidx_input_section& exidx_input_section() const { return this->exidx_input_section_; } // Return the section offset map. const Arm_exidx_section_offset_map& section_offset_map() const { return this->section_offset_map_; } protected: // Write merged section into file OF. void do_write(Output_file* of); bool do_output_offset(const Relobj*, unsigned int, section_offset_type, section_offset_type*) const; private: // Original EXIDX input section. const Arm_exidx_input_section& exidx_input_section_; // Section offset map. const Arm_exidx_section_offset_map& section_offset_map_; // Merged section contents. We need to keep build the merged section // and save it here to avoid accessing the original EXIDX section when // we cannot lock the sections' object. unsigned char* section_contents_; }; // A class to wrap an ordinary input section containing executable code. template<bool big_endian> class Arm_input_section : public Output_relaxed_input_section { public: Arm_input_section(Relobj* relobj, unsigned int shndx) : Output_relaxed_input_section(relobj, shndx, 1), original_addralign_(1), original_size_(0), stub_table_(NULL), original_contents_(NULL) { } ~Arm_input_section() { delete[] this->original_contents_; } // Initialize. void init(); // Whether this is a stub table owner. bool is_stub_table_owner() const { return this->stub_table_ != NULL && this->stub_table_->owner() == this; } // Return the stub table. Stub_table<big_endian>* stub_table() const { return this->stub_table_; } // Set the stub_table. void set_stub_table(Stub_table<big_endian>* stub_table) { this->stub_table_ = stub_table; } // Downcast a base pointer to an Arm_input_section pointer. This is // not type-safe but we only use Arm_input_section not the base class. static Arm_input_section<big_endian>* as_arm_input_section(Output_relaxed_input_section* poris) { return static_cast<Arm_input_section<big_endian>*>(poris); } // Return the original size of the section. uint32_t original_size() const { return this->original_size_; } protected: // Write data to output file. void do_write(Output_file*); // Return required alignment of this. uint64_t do_addralign() const { if (this->is_stub_table_owner()) return std::max(this->stub_table_->addralign(), static_cast<uint64_t>(this->original_addralign_)); else return this->original_addralign_; } // Finalize data size. void set_final_data_size(); // Reset address and file offset. void do_reset_address_and_file_offset(); // Output offset. bool do_output_offset(const Relobj* object, unsigned int shndx, section_offset_type offset, section_offset_type* poutput) const { if ((object == this->relobj()) && (shndx == this->shndx()) && (offset >= 0) && (offset <= convert_types<section_offset_type, uint32_t>(this->original_size_))) { *poutput = offset; return true; } else return false; } private: // Copying is not allowed. Arm_input_section(const Arm_input_section&); Arm_input_section& operator=(const Arm_input_section&); // Address alignment of the original input section. uint32_t original_addralign_; // Section size of the original input section. uint32_t original_size_; // Stub table. Stub_table<big_endian>* stub_table_; // Original section contents. We have to make a copy here since the file // containing the original section may not be locked when we need to access // the contents. unsigned char* original_contents_; }; // Arm_exidx_fixup class. This is used to define a number of methods // and keep states for fixing up EXIDX coverage. class Arm_exidx_fixup { public: Arm_exidx_fixup(Output_section* exidx_output_section, bool merge_exidx_entries = true) : exidx_output_section_(exidx_output_section), last_unwind_type_(UT_NONE), last_inlined_entry_(0), last_input_section_(NULL), section_offset_map_(NULL), first_output_text_section_(NULL), merge_exidx_entries_(merge_exidx_entries) { } ~Arm_exidx_fixup() { delete this->section_offset_map_; } // Process an EXIDX section for entry merging. SECTION_CONTENTS points // to the EXIDX contents and SECTION_SIZE is the size of the contents. Return // number of bytes to be deleted in output. If parts of the input EXIDX // section are merged a heap allocated Arm_exidx_section_offset_map is store // in the located PSECTION_OFFSET_MAP. The caller owns the map and is // responsible for releasing it. template<bool big_endian> uint32_t process_exidx_section(const Arm_exidx_input_section* exidx_input_section, const unsigned char* section_contents, section_size_type section_size, Arm_exidx_section_offset_map** psection_offset_map); // Append an EXIDX_CANTUNWIND entry pointing at the end of the last // input section, if there is not one already. void add_exidx_cantunwind_as_needed(); // Return the output section for the text section which is linked to the // first exidx input in output. Output_section* first_output_text_section() const { return this->first_output_text_section_; } private: // Copying is not allowed. Arm_exidx_fixup(const Arm_exidx_fixup&); Arm_exidx_fixup& operator=(const Arm_exidx_fixup&); // Type of EXIDX unwind entry. enum Unwind_type { // No type. UT_NONE, // EXIDX_CANTUNWIND. UT_EXIDX_CANTUNWIND, // Inlined entry. UT_INLINED_ENTRY, // Normal entry. UT_NORMAL_ENTRY, }; // Process an EXIDX entry. We only care about the second word of the // entry. Return true if the entry can be deleted. bool process_exidx_entry(uint32_t second_word); // Update the current section offset map during EXIDX section fix-up. // If there is no map, create one. INPUT_OFFSET is the offset of a // reference point, DELETED_BYTES is the number of deleted by in the // section so far. If DELETE_ENTRY is true, the reference point and // all offsets after the previous reference point are discarded. void update_offset_map(section_offset_type input_offset, section_size_type deleted_bytes, bool delete_entry); // EXIDX output section. Output_section* exidx_output_section_; // Unwind type of the last EXIDX entry processed. Unwind_type last_unwind_type_; // Last seen inlined EXIDX entry. uint32_t last_inlined_entry_; // Last processed EXIDX input section. const Arm_exidx_input_section* last_input_section_; // Section offset map created in process_exidx_section. Arm_exidx_section_offset_map* section_offset_map_; // Output section for the text section which is linked to the first exidx // input in output. Output_section* first_output_text_section_; bool merge_exidx_entries_; }; // Arm output section class. This is defined mainly to add a number of // stub generation methods. template<bool big_endian> class Arm_output_section : public Output_section { public: typedef std::vector<std::pair<Relobj*, unsigned int> > Text_section_list; // We need to force SHF_LINK_ORDER in a SHT_ARM_EXIDX section. Arm_output_section(const char* name, elfcpp::Elf_Word type, elfcpp::Elf_Xword flags) : Output_section(name, type, (type == elfcpp::SHT_ARM_EXIDX ? flags | elfcpp::SHF_LINK_ORDER : flags)) { if (type == elfcpp::SHT_ARM_EXIDX) this->set_always_keeps_input_sections(); } ~Arm_output_section() { } // Group input sections for stub generation. void group_sections(section_size_type, bool, Target_arm<big_endian>*, const Task*); // Downcast a base pointer to an Arm_output_section pointer. This is // not type-safe but we only use Arm_output_section not the base class. static Arm_output_section<big_endian>* as_arm_output_section(Output_section* os) { return static_cast<Arm_output_section<big_endian>*>(os); } // Append all input text sections in this into LIST. void append_text_sections_to_list(Text_section_list* list); // Fix EXIDX coverage of this EXIDX output section. SORTED_TEXT_SECTION // is a list of text input sections sorted in ascending order of their // output addresses. void fix_exidx_coverage(Layout* layout, const Text_section_list& sorted_text_section, Symbol_table* symtab, bool merge_exidx_entries, const Task* task); // Link an EXIDX section into its corresponding text section. void set_exidx_section_link(); private: // For convenience. typedef Output_section::Input_section Input_section; typedef Output_section::Input_section_list Input_section_list; // Create a stub group. void create_stub_group(Input_section_list::const_iterator, Input_section_list::const_iterator, Input_section_list::const_iterator, Target_arm<big_endian>*, std::vector<Output_relaxed_input_section*>*, const Task* task); }; // Arm_exidx_input_section class. This represents an EXIDX input section. class Arm_exidx_input_section { public: static const section_offset_type invalid_offset = static_cast<section_offset_type>(-1); Arm_exidx_input_section(Relobj* relobj, unsigned int shndx, unsigned int link, uint32_t size, uint32_t addralign, uint32_t text_size) : relobj_(relobj), shndx_(shndx), link_(link), size_(size), addralign_(addralign), text_size_(text_size), has_errors_(false) { } ~Arm_exidx_input_section() { } // Accessors: This is a read-only class. // Return the object containing this EXIDX input section. Relobj* relobj() const { return this->relobj_; } // Return the section index of this EXIDX input section. unsigned int shndx() const { return this->shndx_; } // Return the section index of linked text section in the same object. unsigned int link() const { return this->link_; } // Return size of the EXIDX input section. uint32_t size() const { return this->size_; } // Return address alignment of EXIDX input section. uint32_t addralign() const { return this->addralign_; } // Return size of the associated text input section. uint32_t text_size() const { return this->text_size_; } // Whether there are any errors in the EXIDX input section. bool has_errors() const { return this->has_errors_; } // Set has-errors flag. void set_has_errors() { this->has_errors_ = true; } private: // Object containing this. Relobj* relobj_; // Section index of this. unsigned int shndx_; // text section linked to this in the same object. unsigned int link_; // Size of this. For ARM 32-bit is sufficient. uint32_t size_; // Address alignment of this. For ARM 32-bit is sufficient. uint32_t addralign_; // Size of associated text section. uint32_t text_size_; // Whether this has any errors. bool has_errors_; }; // Arm_relobj class. template<bool big_endian> class Arm_relobj : public Sized_relobj_file<32, big_endian> { public: static const Arm_address invalid_address = static_cast<Arm_address>(-1); Arm_relobj(const std::string& name, Input_file* input_file, off_t offset, const typename elfcpp::Ehdr<32, big_endian>& ehdr) : Sized_relobj_file<32, big_endian>(name, input_file, offset, ehdr), stub_tables_(), local_symbol_is_thumb_function_(), attributes_section_data_(NULL), mapping_symbols_info_(), section_has_cortex_a8_workaround_(NULL), exidx_section_map_(), output_local_symbol_count_needs_update_(false), merge_flags_and_attributes_(true) { } ~Arm_relobj() { delete this->attributes_section_data_; } // Return the stub table of the SHNDX-th section if there is one. Stub_table<big_endian>* stub_table(unsigned int shndx) const { gold_assert(shndx < this->stub_tables_.size()); return this->stub_tables_[shndx]; } // Set STUB_TABLE to be the stub_table of the SHNDX-th section. void set_stub_table(unsigned int shndx, Stub_table<big_endian>* stub_table) { gold_assert(shndx < this->stub_tables_.size()); this->stub_tables_[shndx] = stub_table; } // Whether a local symbol is a THUMB function. R_SYM is the symbol table // index. This is only valid after do_count_local_symbol is called. bool local_symbol_is_thumb_function(unsigned int r_sym) const { gold_assert(r_sym < this->local_symbol_is_thumb_function_.size()); return this->local_symbol_is_thumb_function_[r_sym]; } // Scan all relocation sections for stub generation. void scan_sections_for_stubs(Target_arm<big_endian>*, const Symbol_table*, const Layout*); // Convert regular input section with index SHNDX to a relaxed section. void convert_input_section_to_relaxed_section(unsigned shndx) { // The stubs have relocations and we need to process them after writing // out the stubs. So relocation now must follow section write. this->set_section_offset(shndx, -1ULL); this->set_relocs_must_follow_section_writes(); } // Downcast a base pointer to an Arm_relobj pointer. This is // not type-safe but we only use Arm_relobj not the base class. static Arm_relobj<big_endian>* as_arm_relobj(Relobj* relobj) { return static_cast<Arm_relobj<big_endian>*>(relobj); } // Processor-specific flags in ELF file header. This is valid only after // reading symbols. elfcpp::Elf_Word processor_specific_flags() const { return this->processor_specific_flags_; } // Attribute section data This is the contents of the .ARM.attribute section // if there is one. const Attributes_section_data* attributes_section_data() const { return this->attributes_section_data_; } // Mapping symbol location. typedef std::pair<unsigned int, Arm_address> Mapping_symbol_position; // Functor for STL container. struct Mapping_symbol_position_less { bool operator()(const Mapping_symbol_position& p1, const Mapping_symbol_position& p2) const { return (p1.first < p2.first || (p1.first == p2.first && p1.second < p2.second)); } }; // We only care about the first character of a mapping symbol, so // we only store that instead of the whole symbol name. typedef std::map<Mapping_symbol_position, char, Mapping_symbol_position_less> Mapping_symbols_info; // Whether a section contains any Cortex-A8 workaround. bool section_has_cortex_a8_workaround(unsigned int shndx) const { return (this->section_has_cortex_a8_workaround_ != NULL && (*this->section_has_cortex_a8_workaround_)[shndx]); } // Mark a section that has Cortex-A8 workaround. void mark_section_for_cortex_a8_workaround(unsigned int shndx) { if (this->section_has_cortex_a8_workaround_ == NULL) this->section_has_cortex_a8_workaround_ = new std::vector<bool>(this->shnum(), false); (*this->section_has_cortex_a8_workaround_)[shndx] = true; } // Return the EXIDX section of an text section with index SHNDX or NULL // if the text section has no associated EXIDX section. const Arm_exidx_input_section* exidx_input_section_by_link(unsigned int shndx) const { Exidx_section_map::const_iterator p = this->exidx_section_map_.find(shndx); return ((p != this->exidx_section_map_.end() && p->second->link() == shndx) ? p->second : NULL); } // Return the EXIDX section with index SHNDX or NULL if there is none. const Arm_exidx_input_section* exidx_input_section_by_shndx(unsigned shndx) const { Exidx_section_map::const_iterator p = this->exidx_section_map_.find(shndx); return ((p != this->exidx_section_map_.end() && p->second->shndx() == shndx) ? p->second : NULL); } // Whether output local symbol count needs updating. bool output_local_symbol_count_needs_update() const { return this->output_local_symbol_count_needs_update_; } // Set output_local_symbol_count_needs_update flag to be true. void set_output_local_symbol_count_needs_update() { this->output_local_symbol_count_needs_update_ = true; } // Update output local symbol count at the end of relaxation. void update_output_local_symbol_count(); // Whether we want to merge processor-specific flags and attributes. bool merge_flags_and_attributes() const { return this->merge_flags_and_attributes_; } // Export list of EXIDX section indices. void get_exidx_shndx_list(std::vector<unsigned int>* list) const { list->clear(); for (Exidx_section_map::const_iterator p = this->exidx_section_map_.begin(); p != this->exidx_section_map_.end(); ++p) { if (p->second->shndx() == p->first) list->push_back(p->first); } // Sort list to make result independent of implementation of map. std::sort(list->begin(), list->end()); } protected: // Post constructor setup. void do_setup() { // Call parent's setup method. Sized_relobj_file<32, big_endian>::do_setup(); // Initialize look-up tables. Stub_table_list empty_stub_table_list(this->shnum(), NULL); this->stub_tables_.swap(empty_stub_table_list); } // Count the local symbols. void do_count_local_symbols(Stringpool_template<char>*, Stringpool_template<char>*); void do_relocate_sections( const Symbol_table* symtab, const Layout* layout, const unsigned char* pshdrs, Output_file* of, typename Sized_relobj_file<32, big_endian>::Views* pivews); // Read the symbol information. void do_read_symbols(Read_symbols_data* sd); // Process relocs for garbage collection. void do_gc_process_relocs(Symbol_table*, Layout*, Read_relocs_data*); private: // Whether a section needs to be scanned for relocation stubs. bool section_needs_reloc_stub_scanning(const elfcpp::Shdr<32, big_endian>&, const Relobj::Output_sections&, const Symbol_table*, const unsigned char*); // Whether a section is a scannable text section. bool section_is_scannable(const elfcpp::Shdr<32, big_endian>&, unsigned int, const Output_section*, const Symbol_table*); // Whether a section needs to be scanned for the Cortex-A8 erratum. bool section_needs_cortex_a8_stub_scanning(const elfcpp::Shdr<32, big_endian>&, unsigned int, Output_section*, const Symbol_table*); // Scan a section for the Cortex-A8 erratum. void scan_section_for_cortex_a8_erratum(const elfcpp::Shdr<32, big_endian>&, unsigned int, Output_section*, Target_arm<big_endian>*); // Find the linked text section of an EXIDX section by looking at the // first relocation of the EXIDX section. PSHDR points to the section // headers of a relocation section and PSYMS points to the local symbols. // PSHNDX points to a location storing the text section index if found. // Return whether we can find the linked section. bool find_linked_text_section(const unsigned char* pshdr, const unsigned char* psyms, unsigned int* pshndx); // // Make a new Arm_exidx_input_section object for EXIDX section with // index SHNDX and section header SHDR. TEXT_SHNDX is the section // index of the linked text section. void make_exidx_input_section(unsigned int shndx, const elfcpp::Shdr<32, big_endian>& shdr, unsigned int text_shndx, const elfcpp::Shdr<32, big_endian>& text_shdr); // Return the output address of either a plain input section or a // relaxed input section. SHNDX is the section index. Arm_address simple_input_section_output_address(unsigned int, Output_section*); typedef std::vector<Stub_table<big_endian>*> Stub_table_list; typedef Unordered_map<unsigned int, const Arm_exidx_input_section*> Exidx_section_map; // List of stub tables. Stub_table_list stub_tables_; // Bit vector to tell if a local symbol is a thumb function or not. // This is only valid after do_count_local_symbol is called. std::vector<bool> local_symbol_is_thumb_function_; // processor-specific flags in ELF file header. elfcpp::Elf_Word processor_specific_flags_; // Object attributes if there is an .ARM.attributes section or NULL. Attributes_section_data* attributes_section_data_; // Mapping symbols information. Mapping_symbols_info mapping_symbols_info_; // Bitmap to indicate sections with Cortex-A8 workaround or NULL. std::vector<bool>* section_has_cortex_a8_workaround_; // Map a text section to its associated .ARM.exidx section, if there is one. Exidx_section_map exidx_section_map_; // Whether output local symbol count needs updating. bool output_local_symbol_count_needs_update_; // Whether we merge processor flags and attributes of this object to // output. bool merge_flags_and_attributes_; }; // Arm_dynobj class. template<bool big_endian> class Arm_dynobj : public Sized_dynobj<32, big_endian> { public: Arm_dynobj(const std::string& name, Input_file* input_file, off_t offset, const elfcpp::Ehdr<32, big_endian>& ehdr) : Sized_dynobj<32, big_endian>(name, input_file, offset, ehdr), processor_specific_flags_(0), attributes_section_data_(NULL) { } ~Arm_dynobj() { delete this->attributes_section_data_; } // Downcast a base pointer to an Arm_relobj pointer. This is // not type-safe but we only use Arm_relobj not the base class. static Arm_dynobj<big_endian>* as_arm_dynobj(Dynobj* dynobj) { return static_cast<Arm_dynobj<big_endian>*>(dynobj); } // Processor-specific flags in ELF file header. This is valid only after // reading symbols. elfcpp::Elf_Word processor_specific_flags() const { return this->processor_specific_flags_; } // Attributes section data. const Attributes_section_data* attributes_section_data() const { return this->attributes_section_data_; } protected: // Read the symbol information. void do_read_symbols(Read_symbols_data* sd); private: // processor-specific flags in ELF file header. elfcpp::Elf_Word processor_specific_flags_; // Object attributes if there is an .ARM.attributes section or NULL. Attributes_section_data* attributes_section_data_; }; // Functor to read reloc addends during stub generation. template<int sh_type, bool big_endian> struct Stub_addend_reader { // Return the addend for a relocation of a particular type. Depending // on whether this is a REL or RELA relocation, read the addend from a // view or from a Reloc object. elfcpp::Elf_types<32>::Elf_Swxword operator()( unsigned int /* r_type */, const unsigned char* /* view */, const typename Reloc_types<sh_type, 32, big_endian>::Reloc& /* reloc */) const; }; // Specialized Stub_addend_reader for SHT_REL type relocation sections. template<bool big_endian> struct Stub_addend_reader<elfcpp::SHT_REL, big_endian> { elfcpp::Elf_types<32>::Elf_Swxword operator()( unsigned int, const unsigned char*, const typename Reloc_types<elfcpp::SHT_REL, 32, big_endian>::Reloc&) const; }; // Specialized Stub_addend_reader for RELA type relocation sections. // We currently do not handle RELA type relocation sections but it is trivial // to implement the addend reader. This is provided for completeness and to // make it easier to add support for RELA relocation sections in the future. template<bool big_endian> struct Stub_addend_reader<elfcpp::SHT_RELA, big_endian> { elfcpp::Elf_types<32>::Elf_Swxword operator()( unsigned int, const unsigned char*, const typename Reloc_types<elfcpp::SHT_RELA, 32, big_endian>::Reloc& reloc) const { return reloc.get_r_addend(); } }; // Cortex_a8_reloc class. We keep record of relocation that may need // the Cortex-A8 erratum workaround. class Cortex_a8_reloc { public: Cortex_a8_reloc(Reloc_stub* reloc_stub, unsigned r_type, Arm_address destination) : reloc_stub_(reloc_stub), r_type_(r_type), destination_(destination) { } ~Cortex_a8_reloc() { } // Accessors: This is a read-only class. // Return the relocation stub associated with this relocation if there is // one. const Reloc_stub* reloc_stub() const { return this->reloc_stub_; } // Return the relocation type. unsigned int r_type() const { return this->r_type_; } // Return the destination address of the relocation. LSB stores the THUMB // bit. Arm_address destination() const { return this->destination_; } private: // Associated relocation stub if there is one, or NULL. const Reloc_stub* reloc_stub_; // Relocation type. unsigned int r_type_; // Destination address of this relocation. LSB is used to distinguish // ARM/THUMB mode. Arm_address destination_; }; // Arm_output_data_got class. We derive this from Output_data_got to add // extra methods to handle TLS relocations in a static link. template<bool big_endian> class Arm_output_data_got : public Output_data_got<32, big_endian> { public: Arm_output_data_got(Symbol_table* symtab, Layout* layout) : Output_data_got<32, big_endian>(), symbol_table_(symtab), layout_(layout) { } // Add a static entry for the GOT entry at OFFSET. GSYM is a global // symbol and R_TYPE is the code of a dynamic relocation that needs to be // applied in a static link. void add_static_reloc(unsigned int got_offset, unsigned int r_type, Symbol* gsym) { this->static_relocs_.push_back(Static_reloc(got_offset, r_type, gsym)); } // Add a static reloc for the GOT entry at OFFSET. RELOBJ is an object // defining a local symbol with INDEX. R_TYPE is the code of a dynamic // relocation that needs to be applied in a static link. void add_static_reloc(unsigned int got_offset, unsigned int r_type, Sized_relobj_file<32, big_endian>* relobj, unsigned int index) { this->static_relocs_.push_back(Static_reloc(got_offset, r_type, relobj, index)); } // Add a GOT pair for R_ARM_TLS_GD32. The creates a pair of GOT entries. // The first one is initialized to be 1, which is the module index for // the main executable and the second one 0. A reloc of the type // R_ARM_TLS_DTPOFF32 will be created for the second GOT entry and will // be applied by gold. GSYM is a global symbol. void add_tls_gd32_with_static_reloc(unsigned int got_type, Symbol* gsym); // Same as the above but for a local symbol in OBJECT with INDEX. void add_tls_gd32_with_static_reloc(unsigned int got_type, Sized_relobj_file<32, big_endian>* object, unsigned int index); protected: // Write out the GOT table. void do_write(Output_file*); private: // This class represent dynamic relocations that need to be applied by // gold because we are using TLS relocations in a static link. class Static_reloc { public: Static_reloc(unsigned int got_offset, unsigned int r_type, Symbol* gsym) : got_offset_(got_offset), r_type_(r_type), symbol_is_global_(true) { this->u_.global.symbol = gsym; } Static_reloc(unsigned int got_offset, unsigned int r_type, Sized_relobj_file<32, big_endian>* relobj, unsigned int index) : got_offset_(got_offset), r_type_(r_type), symbol_is_global_(false) { this->u_.local.relobj = relobj; this->u_.local.index = index; } // Return the GOT offset. unsigned int got_offset() const { return this->got_offset_; } // Relocation type. unsigned int r_type() const { return this->r_type_; } // Whether the symbol is global or not. bool symbol_is_global() const { return this->symbol_is_global_; } // For a relocation against a global symbol, the global symbol. Symbol* symbol() const { gold_assert(this->symbol_is_global_); return this->u_.global.symbol; } // For a relocation against a local symbol, the defining object. Sized_relobj_file<32, big_endian>* relobj() const { gold_assert(!this->symbol_is_global_); return this->u_.local.relobj; } // For a relocation against a local symbol, the local symbol index. unsigned int index() const { gold_assert(!this->symbol_is_global_); return this->u_.local.index; } private: // GOT offset of the entry to which this relocation is applied. unsigned int got_offset_; // Type of relocation. unsigned int r_type_; // Whether this relocation is against a global symbol. bool symbol_is_global_; // A global or local symbol. union { struct { // For a global symbol, the symbol itself. Symbol* symbol; } global; struct { // For a local symbol, the object defining object. Sized_relobj_file<32, big_endian>* relobj; // For a local symbol, the symbol index. unsigned int index; } local; } u_; }; // Symbol table of the output object. Symbol_table* symbol_table_; // Layout of the output object. Layout* layout_; // Static relocs to be applied to the GOT. std::vector<Static_reloc> static_relocs_; }; // The ARM target has many relocation types with odd-sizes or noncontiguous // bits. The default handling of relocatable relocation cannot process these // relocations. So we have to extend the default code. template<bool big_endian, int sh_type, typename Classify_reloc> class Arm_scan_relocatable_relocs : public Default_scan_relocatable_relocs<sh_type, Classify_reloc> { public: // Return the strategy to use for a local symbol which is a section // symbol, given the relocation type. inline Relocatable_relocs::Reloc_strategy local_section_strategy(unsigned int r_type, Relobj*) { if (sh_type == elfcpp::SHT_RELA) return Relocatable_relocs::RELOC_ADJUST_FOR_SECTION_RELA; else { if (r_type == elfcpp::R_ARM_TARGET1 || r_type == elfcpp::R_ARM_TARGET2) { const Target_arm<big_endian>* arm_target = Target_arm<big_endian>::default_target(); r_type = arm_target->get_real_reloc_type(r_type); } switch(r_type) { // Relocations that write nothing. These exclude R_ARM_TARGET1 // and R_ARM_TARGET2. case elfcpp::R_ARM_NONE: case elfcpp::R_ARM_V4BX: case elfcpp::R_ARM_TLS_GOTDESC: case elfcpp::R_ARM_TLS_CALL: case elfcpp::R_ARM_TLS_DESCSEQ: case elfcpp::R_ARM_THM_TLS_CALL: case elfcpp::R_ARM_GOTRELAX: case elfcpp::R_ARM_GNU_VTENTRY: case elfcpp::R_ARM_GNU_VTINHERIT: case elfcpp::R_ARM_THM_TLS_DESCSEQ16: case elfcpp::R_ARM_THM_TLS_DESCSEQ32: return Relocatable_relocs::RELOC_ADJUST_FOR_SECTION_0; // These should have been converted to something else above. case elfcpp::R_ARM_TARGET1: case elfcpp::R_ARM_TARGET2: gold_unreachable(); // Relocations that write full 32 bits. case elfcpp::R_ARM_ABS32: case elfcpp::R_ARM_REL32: case elfcpp::R_ARM_SBREL32: case elfcpp::R_ARM_GOTOFF32: case elfcpp::R_ARM_BASE_PREL: case elfcpp::R_ARM_GOT_BREL: case elfcpp::R_ARM_BASE_ABS: case elfcpp::R_ARM_ABS32_NOI: case elfcpp::R_ARM_REL32_NOI: case elfcpp::R_ARM_PLT32_ABS: case elfcpp::R_ARM_GOT_ABS: case elfcpp::R_ARM_GOT_PREL: case elfcpp::R_ARM_TLS_GD32: case elfcpp::R_ARM_TLS_LDM32: case elfcpp::R_ARM_TLS_LDO32: case elfcpp::R_ARM_TLS_IE32: case elfcpp::R_ARM_TLS_LE32: return Relocatable_relocs::RELOC_ADJUST_FOR_SECTION_4; default: // For all other static relocations, return RELOC_SPECIAL. return Relocatable_relocs::RELOC_SPECIAL; } } } }; // Utilities for manipulating integers of up to 32-bits namespace utils { // Sign extend an n-bit unsigned integer stored in an uint32_t into // an int32_t. NO_BITS must be between 1 to 32. template<int no_bits> static inline int32_t sign_extend(uint32_t bits) { gold_assert(no_bits >= 0 && no_bits <= 32); if (no_bits == 32) return static_cast<int32_t>(bits); uint32_t mask = (~((uint32_t) 0)) >> (32 - no_bits); bits &= mask; uint32_t top_bit = 1U << (no_bits - 1); int32_t as_signed = static_cast<int32_t>(bits); return (bits & top_bit) ? as_signed + (-top_bit * 2) : as_signed; } // Detects overflow of an NO_BITS integer stored in a uint32_t. template<int no_bits> static inline bool has_overflow(uint32_t bits) { gold_assert(no_bits >= 0 && no_bits <= 32); if (no_bits == 32) return false; int32_t max = (1 << (no_bits - 1)) - 1; int32_t min = -(1 << (no_bits - 1)); int32_t as_signed = static_cast<int32_t>(bits); return as_signed > max || as_signed < min; } // Detects overflow of an NO_BITS integer stored in a uint32_t when it // fits in the given number of bits as either a signed or unsigned value. // For example, has_signed_unsigned_overflow<8> would check // -128 <= bits <= 255 template<int no_bits> static inline bool has_signed_unsigned_overflow(uint32_t bits) { gold_assert(no_bits >= 2 && no_bits <= 32); if (no_bits == 32) return false; int32_t max = static_cast<int32_t>((1U << no_bits) - 1); int32_t min = -(1 << (no_bits - 1)); int32_t as_signed = static_cast<int32_t>(bits); return as_signed > max || as_signed < min; } // Select bits from A and B using bits in MASK. For each n in [0..31], // the n-th bit in the result is chosen from the n-th bits of A and B. // A zero selects A and a one selects B. static inline uint32_t bit_select(uint32_t a, uint32_t b, uint32_t mask) { return (a & ~mask) | (b & mask); } }; template<bool big_endian> class Target_arm : public Sized_target<32, big_endian> { public: typedef Output_data_reloc<elfcpp::SHT_REL, true, 32, big_endian> Reloc_section; // When were are relocating a stub, we pass this as the relocation number. static const size_t fake_relnum_for_stubs = static_cast<size_t>(-1); Target_arm() : Sized_target<32, big_endian>(&arm_info), got_(NULL), plt_(NULL), got_plt_(NULL), rel_dyn_(NULL), copy_relocs_(elfcpp::R_ARM_COPY), dynbss_(NULL), got_mod_index_offset_(-1U), tls_base_symbol_defined_(false), stub_tables_(), stub_factory_(Stub_factory::get_instance()), may_use_blx_(false), should_force_pic_veneer_(false), arm_input_section_map_(), attributes_section_data_(NULL), fix_cortex_a8_(false), cortex_a8_relocs_info_() { } // Virtual function which is set to return true by a target if // it can use relocation types to determine if a function's // pointer is taken. virtual bool can_check_for_function_pointers() const { return true; } // Whether a section called SECTION_NAME may have function pointers to // sections not eligible for safe ICF folding. virtual bool section_may_have_icf_unsafe_pointers(const char* section_name) const { return (!is_prefix_of(".ARM.exidx", section_name) && !is_prefix_of(".ARM.extab", section_name) && Target::section_may_have_icf_unsafe_pointers(section_name)); } // Whether we can use BLX. bool may_use_blx() const { return this->may_use_blx_; } // Set use-BLX flag. void set_may_use_blx(bool value) { this->may_use_blx_ = value; } // Whether we force PCI branch veneers. bool should_force_pic_veneer() const { return this->should_force_pic_veneer_; } // Set PIC veneer flag. void set_should_force_pic_veneer(bool value) { this->should_force_pic_veneer_ = value; } // Whether we use THUMB-2 instructions. bool using_thumb2() const { Object_attribute* attr = this->get_aeabi_object_attribute(elfcpp::Tag_CPU_arch); int arch = attr->int_value(); return arch == elfcpp::TAG_CPU_ARCH_V6T2 || arch >= elfcpp::TAG_CPU_ARCH_V7; } // Whether we use THUMB/THUMB-2 instructions only. bool using_thumb_only() const { Object_attribute* attr = this->get_aeabi_object_attribute(elfcpp::Tag_CPU_arch); if (attr->int_value() == elfcpp::TAG_CPU_ARCH_V6_M || attr->int_value() == elfcpp::TAG_CPU_ARCH_V6S_M) return true; if (attr->int_value() != elfcpp::TAG_CPU_ARCH_V7 && attr->int_value() != elfcpp::TAG_CPU_ARCH_V7E_M) return false; attr = this->get_aeabi_object_attribute(elfcpp::Tag_CPU_arch_profile); return attr->int_value() == 'M'; } // Whether we have an NOP instruction. If not, use mov r0, r0 instead. bool may_use_arm_nop() const { Object_attribute* attr = this->get_aeabi_object_attribute(elfcpp::Tag_CPU_arch); int arch = attr->int_value(); return (arch == elfcpp::TAG_CPU_ARCH_V6T2 || arch == elfcpp::TAG_CPU_ARCH_V6K || arch == elfcpp::TAG_CPU_ARCH_V7 || arch == elfcpp::TAG_CPU_ARCH_V7E_M); } // Whether we have THUMB-2 NOP.W instruction. bool may_use_thumb2_nop() const { Object_attribute* attr = this->get_aeabi_object_attribute(elfcpp::Tag_CPU_arch); int arch = attr->int_value(); return (arch == elfcpp::TAG_CPU_ARCH_V6T2 || arch == elfcpp::TAG_CPU_ARCH_V7 || arch == elfcpp::TAG_CPU_ARCH_V7E_M); } // Process the relocations to determine unreferenced sections for // garbage collection. void gc_process_relocs(Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* 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 scan_relocs(Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* 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); // Finalize the sections. void do_finalize_sections(Layout*, const Input_objects*, Symbol_table*); // Return the value to use for a dynamic symbol which requires special // treatment. uint64_t do_dynsym_value(const Symbol*) const; // Relocate a section. void relocate_section(const Relocate_info<32, big_endian>*, unsigned int sh_type, const unsigned char* prelocs, size_t reloc_count, Output_section* output_section, bool needs_special_offset_handling, unsigned char* view, Arm_address view_address, section_size_type view_size, const Reloc_symbol_changes*); // Scan the relocs during a relocatable link. void scan_relocatable_relocs(Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* 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*); // Relocate a section during a relocatable link. void relocate_for_relocatable(const Relocate_info<32, big_endian>*, 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*, unsigned char* view, Arm_address view_address, section_size_type view_size, unsigned char* reloc_view, section_size_type reloc_view_size); // Perform target-specific processing in a relocatable link. This is // only used if we use the relocation strategy RELOC_SPECIAL. void relocate_special_relocatable(const Relocate_info<32, big_endian>* relinfo, unsigned int sh_type, const unsigned char* preloc_in, size_t relnum, Output_section* output_section, off_t offset_in_output_section, unsigned char* view, typename elfcpp::Elf_types<32>::Elf_Addr view_address, section_size_type view_size, unsigned char* preloc_out); // Return whether SYM is defined by the ABI. bool do_is_defined_by_abi(Symbol* sym) const { return strcmp(sym->name(), "__tls_get_addr") == 0; } // Return whether there is a GOT section. bool has_got_section() const { return this->got_ != NULL; } // Return the size of the GOT section. section_size_type got_size() const { gold_assert(this->got_ != NULL); return this->got_->data_size(); } // Return the number of entries in the GOT. unsigned int got_entry_count() const { if (!this->has_got_section()) return 0; return this->got_size() / 4; } // Return the number of entries in the PLT. unsigned int plt_entry_count() const; // Return the offset of the first non-reserved PLT entry. unsigned int first_plt_entry_offset() const; // Return the size of each PLT entry. unsigned int plt_entry_size() const; // Map platform-specific reloc types static unsigned int get_real_reloc_type(unsigned int r_type); // // Methods to support stub-generations. // // Return the stub factory const Stub_factory& stub_factory() const { return this->stub_factory_; } // Make a new Arm_input_section object. Arm_input_section<big_endian>* new_arm_input_section(Relobj*, unsigned int); // Find the Arm_input_section object corresponding to the SHNDX-th input // section of RELOBJ. Arm_input_section<big_endian>* find_arm_input_section(Relobj* relobj, unsigned int shndx) const; // Make a new Stub_table Stub_table<big_endian>* new_stub_table(Arm_input_section<big_endian>*); // Scan a section for stub generation. void scan_section_for_stubs(const Relocate_info<32, big_endian>*, unsigned int, const unsigned char*, size_t, Output_section*, bool, const unsigned char*, Arm_address, section_size_type); // Relocate a stub. void relocate_stub(Stub*, const Relocate_info<32, big_endian>*, Output_section*, unsigned char*, Arm_address, section_size_type); // Get the default ARM target. static Target_arm<big_endian>* default_target() { gold_assert(parameters->target().machine_code() == elfcpp::EM_ARM && parameters->target().is_big_endian() == big_endian); return static_cast<Target_arm<big_endian>*>( parameters->sized_target<32, big_endian>()); } // Whether NAME belongs to a mapping symbol. static bool is_mapping_symbol_name(const char* name) { return (name && name[0] == '$' && (name[1] == 'a' || name[1] == 't' || name[1] == 'd') && (name[2] == '\0' || name[2] == '.')); } // Whether we work around the Cortex-A8 erratum. bool fix_cortex_a8() const { return this->fix_cortex_a8_; } // Whether we merge exidx entries in debuginfo. bool merge_exidx_entries() const { return parameters->options().merge_exidx_entries(); } // Whether we fix R_ARM_V4BX relocation. // 0 - do not fix // 1 - replace with MOV instruction (armv4 target) // 2 - make interworking veneer (>= armv4t targets only) General_options::Fix_v4bx fix_v4bx() const { return parameters->options().fix_v4bx(); } // Scan a span of THUMB code section for Cortex-A8 erratum. void scan_span_for_cortex_a8_erratum(Arm_relobj<big_endian>*, unsigned int, section_size_type, section_size_type, const unsigned char*, Arm_address); // Apply Cortex-A8 workaround to a branch. void apply_cortex_a8_workaround(const Cortex_a8_stub*, Arm_address, unsigned char*, Arm_address); protected: // Make an ELF object. Object* do_make_elf_object(const std::string&, Input_file*, off_t, const elfcpp::Ehdr<32, big_endian>& ehdr); Object* do_make_elf_object(const std::string&, Input_file*, off_t, const elfcpp::Ehdr<32, !big_endian>&) { gold_unreachable(); } Object* do_make_elf_object(const std::string&, Input_file*, off_t, const elfcpp::Ehdr<64, false>&) { gold_unreachable(); } Object* do_make_elf_object(const std::string&, Input_file*, off_t, const elfcpp::Ehdr<64, true>&) { gold_unreachable(); } // Make an output section. Output_section* do_make_output_section(const char* name, elfcpp::Elf_Word type, elfcpp::Elf_Xword flags) { return new Arm_output_section<big_endian>(name, type, flags); } void do_adjust_elf_header(unsigned char* view, int len) const; // We only need to generate stubs, and hence perform relaxation if we are // not doing relocatable linking. bool do_may_relax() const { return !parameters->options().relocatable(); } bool do_relax(int, const Input_objects*, Symbol_table*, Layout*, const Task*); // Determine whether an object attribute tag takes an integer, a // string or both. int do_attribute_arg_type(int tag) const; // Reorder tags during output. int do_attributes_order(int num) const; // This is called when the target is selected as the default. void do_select_as_default_target() { // No locking is required since there should only be one default target. // We cannot have both the big-endian and little-endian ARM targets // as the default. gold_assert(arm_reloc_property_table == NULL); arm_reloc_property_table = new Arm_reloc_property_table(); } private: // The class which scans relocations. class Scan { public: Scan() : issued_non_pic_error_(false) { } static inline int get_reference_flags(unsigned int r_type); inline void local(Symbol_table* symtab, Layout* layout, Target_arm* target, Sized_relobj_file<32, big_endian>* object, unsigned int data_shndx, Output_section* output_section, const elfcpp::Rel<32, big_endian>& reloc, unsigned int r_type, const elfcpp::Sym<32, big_endian>& lsym); inline void global(Symbol_table* symtab, Layout* layout, Target_arm* target, Sized_relobj_file<32, big_endian>* object, unsigned int data_shndx, Output_section* output_section, const elfcpp::Rel<32, big_endian>& reloc, unsigned int r_type, Symbol* gsym); inline bool local_reloc_may_be_function_pointer(Symbol_table* , Layout* , Target_arm* , Sized_relobj_file<32, big_endian>* , unsigned int , Output_section* , const elfcpp::Rel<32, big_endian>& , unsigned int , const elfcpp::Sym<32, big_endian>&); inline bool global_reloc_may_be_function_pointer(Symbol_table* , Layout* , Target_arm* , Sized_relobj_file<32, big_endian>* , unsigned int , Output_section* , const elfcpp::Rel<32, big_endian>& , unsigned int , Symbol*); private: static void unsupported_reloc_local(Sized_relobj_file<32, big_endian>*, unsigned int r_type); static void unsupported_reloc_global(Sized_relobj_file<32, big_endian>*, unsigned int r_type, Symbol*); void check_non_pic(Relobj*, unsigned int r_type); // Almost identical to Symbol::needs_plt_entry except that it also // handles STT_ARM_TFUNC. static bool symbol_needs_plt_entry(const Symbol* sym) { // An undefined symbol from an executable does not need a PLT entry. if (sym->is_undefined() && !parameters->options().shared()) return false; return (!parameters->doing_static_link() && (sym->type() == elfcpp::STT_FUNC || sym->type() == elfcpp::STT_ARM_TFUNC) && (sym->is_from_dynobj() || sym->is_undefined() || sym->is_preemptible())); } inline bool possible_function_pointer_reloc(unsigned int r_type); // Whether we have issued an error about a non-PIC compilation. bool issued_non_pic_error_; }; // The class which implements relocation. class Relocate { public: Relocate() { } ~Relocate() { } // Return whether the static relocation needs to be applied. inline bool should_apply_static_reloc(const Sized_symbol<32>* gsym, unsigned int r_type, bool is_32bit, Output_section* output_section); // Do a relocation. Return false if the caller should not issue // any warnings about this relocation. inline bool relocate(const Relocate_info<32, big_endian>*, Target_arm*, Output_section*, size_t relnum, const elfcpp::Rel<32, big_endian>&, unsigned int r_type, const Sized_symbol<32>*, const Symbol_value<32>*, unsigned char*, Arm_address, section_size_type); // Return whether we want to pass flag NON_PIC_REF for this // reloc. This means the relocation type accesses a symbol not via // GOT or PLT. static inline bool reloc_is_non_pic(unsigned int r_type) { switch (r_type) { // These relocation types reference GOT or PLT entries explicitly. case elfcpp::R_ARM_GOT_BREL: case elfcpp::R_ARM_GOT_ABS: case elfcpp::R_ARM_GOT_PREL: case elfcpp::R_ARM_GOT_BREL12: case elfcpp::R_ARM_PLT32_ABS: case elfcpp::R_ARM_TLS_GD32: case elfcpp::R_ARM_TLS_LDM32: case elfcpp::R_ARM_TLS_IE32: case elfcpp::R_ARM_TLS_IE12GP: // These relocate types may use PLT entries. case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_THM_JUMP24: case elfcpp::R_ARM_THM_JUMP19: case elfcpp::R_ARM_PLT32: case elfcpp::R_ARM_THM_XPC22: case elfcpp::R_ARM_PREL31: case elfcpp::R_ARM_SBREL31: return false; default: return true; } } private: // Do a TLS relocation. inline typename Arm_relocate_functions<big_endian>::Status relocate_tls(const Relocate_info<32, big_endian>*, Target_arm<big_endian>*, size_t, const elfcpp::Rel<32, big_endian>&, unsigned int, const Sized_symbol<32>*, const Symbol_value<32>*, unsigned char*, elfcpp::Elf_types<32>::Elf_Addr, section_size_type); }; // A class which returns the size required for a relocation type, // used while scanning relocs during a relocatable link. class Relocatable_size_for_reloc { public: unsigned int get_size_for_reloc(unsigned int, Relobj*); }; // Adjust TLS relocation type based on the options and whether this // is a local symbol. static tls::Tls_optimization optimize_tls_reloc(bool is_final, int r_type); // Get the GOT section, creating it if necessary. Arm_output_data_got<big_endian>* got_section(Symbol_table*, Layout*); // Get the GOT PLT section. Output_data_space* got_plt_section() const { gold_assert(this->got_plt_ != NULL); return this->got_plt_; } // Create a PLT entry for a global symbol. void make_plt_entry(Symbol_table*, Layout*, Symbol*); // Define the _TLS_MODULE_BASE_ symbol in the TLS segment. void define_tls_base_symbol(Symbol_table*, Layout*); // Create a GOT entry for the TLS module index. unsigned int got_mod_index_entry(Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* object); // Get the PLT section. const Output_data_plt_arm<big_endian>* plt_section() const { gold_assert(this->plt_ != NULL); return this->plt_; } // Get the dynamic reloc section, creating it if necessary. Reloc_section* rel_dyn_section(Layout*); // Get the section to use for TLS_DESC relocations. Reloc_section* rel_tls_desc_section(Layout*) const; // Return true if the symbol may need a COPY relocation. // References from an executable object to non-function symbols // defined in a dynamic object may need a COPY relocation. bool may_need_copy_reloc(Symbol* gsym) { return (gsym->type() != elfcpp::STT_ARM_TFUNC && gsym->may_need_copy_reloc()); } // Add a potential copy relocation. void copy_reloc(Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* object, unsigned int shndx, Output_section* output_section, Symbol* sym, const elfcpp::Rel<32, big_endian>& reloc) { this->copy_relocs_.copy_reloc(symtab, layout, symtab->get_sized_symbol<32>(sym), object, shndx, output_section, reloc, this->rel_dyn_section(layout)); } // Whether two EABI versions are compatible. static bool are_eabi_versions_compatible(elfcpp::Elf_Word v1, elfcpp::Elf_Word v2); // Merge processor-specific flags from input object and those in the ELF // header of the output. void merge_processor_specific_flags(const std::string&, elfcpp::Elf_Word); // Get the secondary compatible architecture. static int get_secondary_compatible_arch(const Attributes_section_data*); // Set the secondary compatible architecture. static void set_secondary_compatible_arch(Attributes_section_data*, int); static int tag_cpu_arch_combine(const char*, int, int*, int, int); // Helper to print AEABI enum tag value. static std::string aeabi_enum_name(unsigned int); // Return string value for TAG_CPU_name. static std::string tag_cpu_name_value(unsigned int); // Merge object attributes from input object and those in the output. void merge_object_attributes(const char*, const Attributes_section_data*); // Helper to get an AEABI object attribute Object_attribute* get_aeabi_object_attribute(int tag) const { Attributes_section_data* pasd = this->attributes_section_data_; gold_assert(pasd != NULL); Object_attribute* attr = pasd->get_attribute(Object_attribute::OBJ_ATTR_PROC, tag); gold_assert(attr != NULL); return attr; } // // Methods to support stub-generations. // // Group input sections for stub generation. void group_sections(Layout*, section_size_type, bool, const Task*); // Scan a relocation for stub generation. void scan_reloc_for_stub(const Relocate_info<32, big_endian>*, unsigned int, const Sized_symbol<32>*, unsigned int, const Symbol_value<32>*, elfcpp::Elf_types<32>::Elf_Swxword, Arm_address); // Scan a relocation section for stub. template<int sh_type> void scan_reloc_section_for_stubs( const Relocate_info<32, big_endian>* relinfo, const unsigned char* prelocs, size_t reloc_count, Output_section* output_section, bool needs_special_offset_handling, const unsigned char* view, elfcpp::Elf_types<32>::Elf_Addr view_address, section_size_type); // Fix .ARM.exidx section coverage. void fix_exidx_coverage(Layout*, const Input_objects*, Arm_output_section<big_endian>*, Symbol_table*, const Task*); // Functors for STL set. struct output_section_address_less_than { bool operator()(const Output_section* s1, const Output_section* s2) const { return s1->address() < s2->address(); } }; // Information about this specific target which we pass to the // general Target structure. static const Target::Target_info arm_info; // The types of GOT entries needed for this platform. // These values are exposed to the ABI in an incremental link. // Do not renumber existing values without changing the version // number of the .gnu_incremental_inputs section. enum Got_type { GOT_TYPE_STANDARD = 0, // GOT entry for a regular symbol GOT_TYPE_TLS_NOFFSET = 1, // GOT entry for negative TLS offset GOT_TYPE_TLS_OFFSET = 2, // GOT entry for positive TLS offset GOT_TYPE_TLS_PAIR = 3, // GOT entry for TLS module/offset pair GOT_TYPE_TLS_DESC = 4 // GOT entry for TLS_DESC pair }; typedef typename std::vector<Stub_table<big_endian>*> Stub_table_list; // Map input section to Arm_input_section. typedef Unordered_map<Section_id, Arm_input_section<big_endian>*, Section_id_hash> Arm_input_section_map; // Map output addresses to relocs for Cortex-A8 erratum. typedef Unordered_map<Arm_address, const Cortex_a8_reloc*> Cortex_a8_relocs_info; // The GOT section. Arm_output_data_got<big_endian>* got_; // The PLT section. Output_data_plt_arm<big_endian>* plt_; // The GOT PLT section. Output_data_space* got_plt_; // The dynamic reloc section. Reloc_section* rel_dyn_; // Relocs saved to avoid a COPY reloc. Copy_relocs<elfcpp::SHT_REL, 32, big_endian> 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_; // True if the _TLS_MODULE_BASE_ symbol has been defined. bool tls_base_symbol_defined_; // Vector of Stub_tables created. Stub_table_list stub_tables_; // Stub factory. const Stub_factory &stub_factory_; // Whether we can use BLX. bool may_use_blx_; // Whether we force PIC branch veneers. bool should_force_pic_veneer_; // Map for locating Arm_input_sections. Arm_input_section_map arm_input_section_map_; // Attributes section data in output. Attributes_section_data* attributes_section_data_; // Whether we want to fix code for Cortex-A8 erratum. bool fix_cortex_a8_; // Map addresses to relocs for Cortex-A8 erratum. Cortex_a8_relocs_info cortex_a8_relocs_info_; }; template<bool big_endian> const Target::Target_info Target_arm<big_endian>::arm_info = { 32, // size big_endian, // is_big_endian elfcpp::EM_ARM, // machine_code false, // has_make_symbol false, // has_resolve false, // has_code_fill true, // is_default_stack_executable '\0', // wrap_char "/usr/lib/libc.so.1", // dynamic_linker 0x8000, // default_text_segment_address 0x1000, // abi_pagesize (overridable by -z max-page-size) 0x1000, // common_pagesize (overridable by -z common-page-size) elfcpp::SHN_UNDEF, // small_common_shndx elfcpp::SHN_UNDEF, // large_common_shndx 0, // small_common_section_flags 0, // large_common_section_flags ".ARM.attributes", // attributes_section "aeabi" // attributes_vendor }; // Arm relocate functions class // template<bool big_endian> class Arm_relocate_functions : public Relocate_functions<32, big_endian> { public: typedef enum { STATUS_OKAY, // No error during relocation. STATUS_OVERFLOW, // Relocation overflow. STATUS_BAD_RELOC // Relocation cannot be applied. } Status; private: typedef Relocate_functions<32, big_endian> Base; typedef Arm_relocate_functions<big_endian> This; // Encoding of imm16 argument for movt and movw ARM instructions // from ARM ARM: // // imm16 := imm4 | imm12 // // f e d c b a 9 8 7 6 5 4 3 2 1 0 f e d c b a 9 8 7 6 5 4 3 2 1 0 // +-------+---------------+-------+-------+-----------------------+ // | | |imm4 | |imm12 | // +-------+---------------+-------+-------+-----------------------+ // Extract the relocation addend from VAL based on the ARM // instruction encoding described above. static inline typename elfcpp::Swap<32, big_endian>::Valtype extract_arm_movw_movt_addend( typename elfcpp::Swap<32, big_endian>::Valtype val) { // According to the Elf ABI for ARM Architecture the immediate // field is sign-extended to form the addend. return utils::sign_extend<16>(((val >> 4) & 0xf000) | (val & 0xfff)); } // Insert X into VAL based on the ARM instruction encoding described // above. static inline typename elfcpp::Swap<32, big_endian>::Valtype insert_val_arm_movw_movt( typename elfcpp::Swap<32, big_endian>::Valtype val, typename elfcpp::Swap<32, big_endian>::Valtype x) { val &= 0xfff0f000; val |= x & 0x0fff; val |= (x & 0xf000) << 4; return val; } // Encoding of imm16 argument for movt and movw Thumb2 instructions // from ARM ARM: // // imm16 := imm4 | i | imm3 | imm8 // // f e d c b a 9 8 7 6 5 4 3 2 1 0 f e d c b a 9 8 7 6 5 4 3 2 1 0 // +---------+-+-----------+-------++-+-----+-------+---------------+ // | |i| |imm4 || |imm3 | |imm8 | // +---------+-+-----------+-------++-+-----+-------+---------------+ // Extract the relocation addend from VAL based on the Thumb2 // instruction encoding described above. static inline typename elfcpp::Swap<32, big_endian>::Valtype extract_thumb_movw_movt_addend( typename elfcpp::Swap<32, big_endian>::Valtype val) { // According to the Elf ABI for ARM Architecture the immediate // field is sign-extended to form the addend. return utils::sign_extend<16>(((val >> 4) & 0xf000) | ((val >> 15) & 0x0800) | ((val >> 4) & 0x0700) | (val & 0x00ff)); } // Insert X into VAL based on the Thumb2 instruction encoding // described above. static inline typename elfcpp::Swap<32, big_endian>::Valtype insert_val_thumb_movw_movt( typename elfcpp::Swap<32, big_endian>::Valtype val, typename elfcpp::Swap<32, big_endian>::Valtype x) { val &= 0xfbf08f00; val |= (x & 0xf000) << 4; val |= (x & 0x0800) << 15; val |= (x & 0x0700) << 4; val |= (x & 0x00ff); return val; } // Calculate the smallest constant Kn for the specified residual. // (see (AAELF 4.6.1.4 Static ARM relocations, Group Relocations, p.32) static uint32_t calc_grp_kn(typename elfcpp::Swap<32, big_endian>::Valtype residual) { int32_t msb; if (residual == 0) return 0; // Determine the most significant bit in the residual and // align the resulting value to a 2-bit boundary. for (msb = 30; (msb >= 0) && !(residual & (3 << msb)); msb -= 2) ; // The desired shift is now (msb - 6), or zero, whichever // is the greater. return (((msb - 6) < 0) ? 0 : (msb - 6)); } // Calculate the final residual for the specified group index. // If the passed group index is less than zero, the method will return // the value of the specified residual without any change. // (see (AAELF 4.6.1.4 Static ARM relocations, Group Relocations, p.32) static typename elfcpp::Swap<32, big_endian>::Valtype calc_grp_residual(typename elfcpp::Swap<32, big_endian>::Valtype residual, const int group) { for (int n = 0; n <= group; n++) { // Calculate which part of the value to mask. uint32_t shift = calc_grp_kn(residual); // Calculate the residual for the next time around. residual &= ~(residual & (0xff << shift)); } return residual; } // Calculate the value of Gn for the specified group index. // We return it in the form of an encoded constant-and-rotation. // (see (AAELF 4.6.1.4 Static ARM relocations, Group Relocations, p.32) static typename elfcpp::Swap<32, big_endian>::Valtype calc_grp_gn(typename elfcpp::Swap<32, big_endian>::Valtype residual, const int group) { typename elfcpp::Swap<32, big_endian>::Valtype gn = 0; uint32_t shift = 0; for (int n = 0; n <= group; n++) { // Calculate which part of the value to mask. shift = calc_grp_kn(residual); // Calculate Gn in 32-bit as well as encoded constant-and-rotation form. gn = residual & (0xff << shift); // Calculate the residual for the next time around. residual &= ~gn; } // Return Gn in the form of an encoded constant-and-rotation. return ((gn >> shift) | ((gn <= 0xff ? 0 : (32 - shift) / 2) << 8)); } public: // Handle ARM long branches. static typename This::Status arm_branch_common(unsigned int, const Relocate_info<32, big_endian>*, unsigned char*, const Sized_symbol<32>*, const Arm_relobj<big_endian>*, unsigned int, const Symbol_value<32>*, Arm_address, Arm_address, bool); // Handle THUMB long branches. static typename This::Status thumb_branch_common(unsigned int, const Relocate_info<32, big_endian>*, unsigned char*, const Sized_symbol<32>*, const Arm_relobj<big_endian>*, unsigned int, const Symbol_value<32>*, Arm_address, Arm_address, bool); // Return the branch offset of a 32-bit THUMB branch. static inline int32_t thumb32_branch_offset(uint16_t upper_insn, uint16_t lower_insn) { // We use the Thumb-2 encoding (backwards compatible with Thumb-1) // involving the J1 and J2 bits. uint32_t s = (upper_insn & (1U << 10)) >> 10; uint32_t upper = upper_insn & 0x3ffU; uint32_t lower = lower_insn & 0x7ffU; uint32_t j1 = (lower_insn & (1U << 13)) >> 13; uint32_t j2 = (lower_insn & (1U << 11)) >> 11; uint32_t i1 = j1 ^ s ? 0 : 1; uint32_t i2 = j2 ^ s ? 0 : 1; return utils::sign_extend<25>((s << 24) | (i1 << 23) | (i2 << 22) | (upper << 12) | (lower << 1)); } // Insert OFFSET to a 32-bit THUMB branch and return the upper instruction. // UPPER_INSN is the original upper instruction of the branch. Caller is // responsible for overflow checking and BLX offset adjustment. static inline uint16_t thumb32_branch_upper(uint16_t upper_insn, int32_t offset) { uint32_t s = offset < 0 ? 1 : 0; uint32_t bits = static_cast<uint32_t>(offset); return (upper_insn & ~0x7ffU) | ((bits >> 12) & 0x3ffU) | (s << 10); } // Insert OFFSET to a 32-bit THUMB branch and return the lower instruction. // LOWER_INSN is the original lower instruction of the branch. Caller is // responsible for overflow checking and BLX offset adjustment. static inline uint16_t thumb32_branch_lower(uint16_t lower_insn, int32_t offset) { uint32_t s = offset < 0 ? 1 : 0; uint32_t bits = static_cast<uint32_t>(offset); return ((lower_insn & ~0x2fffU) | ((((bits >> 23) & 1) ^ !s) << 13) | ((((bits >> 22) & 1) ^ !s) << 11) | ((bits >> 1) & 0x7ffU)); } // Return the branch offset of a 32-bit THUMB conditional branch. static inline int32_t thumb32_cond_branch_offset(uint16_t upper_insn, uint16_t lower_insn) { uint32_t s = (upper_insn & 0x0400U) >> 10; uint32_t j1 = (lower_insn & 0x2000U) >> 13; uint32_t j2 = (lower_insn & 0x0800U) >> 11; uint32_t lower = (lower_insn & 0x07ffU); uint32_t upper = (s << 8) | (j2 << 7) | (j1 << 6) | (upper_insn & 0x003fU); return utils::sign_extend<21>((upper << 12) | (lower << 1)); } // Insert OFFSET to a 32-bit THUMB conditional branch and return the upper // instruction. UPPER_INSN is the original upper instruction of the branch. // Caller is responsible for overflow checking. static inline uint16_t thumb32_cond_branch_upper(uint16_t upper_insn, int32_t offset) { uint32_t s = offset < 0 ? 1 : 0; uint32_t bits = static_cast<uint32_t>(offset); return (upper_insn & 0xfbc0U) | (s << 10) | ((bits & 0x0003f000U) >> 12); } // Insert OFFSET to a 32-bit THUMB conditional branch and return the lower // instruction. LOWER_INSN is the original lower instruction of the branch. // The caller is responsible for overflow checking. static inline uint16_t thumb32_cond_branch_lower(uint16_t lower_insn, int32_t offset) { uint32_t bits = static_cast<uint32_t>(offset); uint32_t j2 = (bits & 0x00080000U) >> 19; uint32_t j1 = (bits & 0x00040000U) >> 18; uint32_t lo = (bits & 0x00000ffeU) >> 1; return (lower_insn & 0xd000U) | (j1 << 13) | (j2 << 11) | lo; } // R_ARM_ABS8: S + A static inline typename This::Status abs8(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval) { typedef typename elfcpp::Swap<8, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<32, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<8, big_endian>::readval(wv); Reltype addend = utils::sign_extend<8>(val); Reltype x = psymval->value(object, addend); val = utils::bit_select(val, x, 0xffU); elfcpp::Swap<8, big_endian>::writeval(wv, val); // R_ARM_ABS8 permits signed or unsigned results. int signed_x = static_cast<int32_t>(x); return ((signed_x < -128 || signed_x > 255) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_THM_ABS5: S + A static inline typename This::Status thm_abs5(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<32, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<16, big_endian>::readval(wv); Reltype addend = (val & 0x7e0U) >> 6; Reltype x = psymval->value(object, addend); val = utils::bit_select(val, x << 6, 0x7e0U); elfcpp::Swap<16, big_endian>::writeval(wv, val); // R_ARM_ABS16 permits signed or unsigned results. int signed_x = static_cast<int32_t>(x); return ((signed_x < -32768 || signed_x > 65535) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_ABS12: S + A static inline typename This::Status abs12(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<32, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<32, big_endian>::readval(wv); Reltype addend = val & 0x0fffU; Reltype x = psymval->value(object, addend); val = utils::bit_select(val, x, 0x0fffU); elfcpp::Swap<32, big_endian>::writeval(wv, val); return (utils::has_overflow<12>(x) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_ABS16: S + A static inline typename This::Status abs16(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<32, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<16, big_endian>::readval(wv); Reltype addend = utils::sign_extend<16>(val); Reltype x = psymval->value(object, addend); val = utils::bit_select(val, x, 0xffffU); elfcpp::Swap<16, big_endian>::writeval(wv, val); return (utils::has_signed_unsigned_overflow<16>(x) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_ABS32: (S + A) | T static inline typename This::Status abs32(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address thumb_bit) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype addend = elfcpp::Swap<32, big_endian>::readval(wv); Valtype x = psymval->value(object, addend) | thumb_bit; elfcpp::Swap<32, big_endian>::writeval(wv, x); return This::STATUS_OKAY; } // R_ARM_REL32: (S + A) | T - P static inline typename This::Status rel32(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address address, Arm_address thumb_bit) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype addend = elfcpp::Swap<32, big_endian>::readval(wv); Valtype x = (psymval->value(object, addend) | thumb_bit) - address; elfcpp::Swap<32, big_endian>::writeval(wv, x); return This::STATUS_OKAY; } // R_ARM_THM_JUMP24: (S + A) | T - P static typename This::Status thm_jump19(unsigned char* view, const Arm_relobj<big_endian>* object, const Symbol_value<32>* psymval, Arm_address address, Arm_address thumb_bit); // R_ARM_THM_JUMP6: S + A – P static inline typename This::Status thm_jump6(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address address) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<16, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<16, big_endian>::readval(wv); // bit[9]:bit[7:3]:’0’ (mask: 0x02f8) Reltype addend = (((val & 0x0200) >> 3) | ((val & 0x00f8) >> 2)); Reltype x = (psymval->value(object, addend) - address); val = (val & 0xfd07) | ((x & 0x0040) << 3) | ((val & 0x003e) << 2); elfcpp::Swap<16, big_endian>::writeval(wv, val); // CZB does only forward jumps. return ((x > 0x007e) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_THM_JUMP8: S + A – P static inline typename This::Status thm_jump8(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address address) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<16, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<16, big_endian>::readval(wv); Reltype addend = utils::sign_extend<8>((val & 0x00ff) << 1); Reltype x = (psymval->value(object, addend) - address); elfcpp::Swap<16, big_endian>::writeval(wv, (val & 0xff00) | ((x & 0x01fe) >> 1)); return (utils::has_overflow<8>(x) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_THM_JUMP11: S + A – P static inline typename This::Status thm_jump11(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address address) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<16, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<16, big_endian>::readval(wv); Reltype addend = utils::sign_extend<11>((val & 0x07ff) << 1); Reltype x = (psymval->value(object, addend) - address); elfcpp::Swap<16, big_endian>::writeval(wv, (val & 0xf800) | ((x & 0x0ffe) >> 1)); return (utils::has_overflow<11>(x) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_BASE_PREL: B(S) + A - P static inline typename This::Status base_prel(unsigned char* view, Arm_address origin, Arm_address address) { Base::rel32(view, origin - address); return STATUS_OKAY; } // R_ARM_BASE_ABS: B(S) + A static inline typename This::Status base_abs(unsigned char* view, Arm_address origin) { Base::rel32(view, origin); return STATUS_OKAY; } // R_ARM_GOT_BREL: GOT(S) + A - GOT_ORG static inline typename This::Status got_brel(unsigned char* view, typename elfcpp::Swap<32, big_endian>::Valtype got_offset) { Base::rel32(view, got_offset); return This::STATUS_OKAY; } // R_ARM_GOT_PREL: GOT(S) + A - P static inline typename This::Status got_prel(unsigned char* view, Arm_address got_entry, Arm_address address) { Base::rel32(view, got_entry - address); return This::STATUS_OKAY; } // R_ARM_PREL: (S + A) | T - P static inline typename This::Status prel31(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address address, Arm_address thumb_bit) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<32, big_endian>::readval(wv); Valtype addend = utils::sign_extend<31>(val); Valtype x = (psymval->value(object, addend) | thumb_bit) - address; val = utils::bit_select(val, x, 0x7fffffffU); elfcpp::Swap<32, big_endian>::writeval(wv, val); return (utils::has_overflow<31>(x) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_MOVW_ABS_NC: (S + A) | T (relative address base is ) // R_ARM_MOVW_PREL_NC: (S + A) | T - P // R_ARM_MOVW_BREL_NC: ((S + A) | T) - B(S) // R_ARM_MOVW_BREL: ((S + A) | T) - B(S) static inline typename This::Status movw(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address relative_address_base, Arm_address thumb_bit, bool check_overflow) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<32, big_endian>::readval(wv); Valtype addend = This::extract_arm_movw_movt_addend(val); Valtype x = ((psymval->value(object, addend) | thumb_bit) - relative_address_base); val = This::insert_val_arm_movw_movt(val, x); elfcpp::Swap<32, big_endian>::writeval(wv, val); return ((check_overflow && utils::has_overflow<16>(x)) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_MOVT_ABS: S + A (relative address base is 0) // R_ARM_MOVT_PREL: S + A - P // R_ARM_MOVT_BREL: S + A - B(S) static inline typename This::Status movt(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address relative_address_base) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<32, big_endian>::readval(wv); Valtype addend = This::extract_arm_movw_movt_addend(val); Valtype x = (psymval->value(object, addend) - relative_address_base) >> 16; val = This::insert_val_arm_movw_movt(val, x); elfcpp::Swap<32, big_endian>::writeval(wv, val); // FIXME: IHI0044D says that we should check for overflow. return This::STATUS_OKAY; } // R_ARM_THM_MOVW_ABS_NC: S + A | T (relative_address_base is 0) // R_ARM_THM_MOVW_PREL_NC: (S + A) | T - P // R_ARM_THM_MOVW_BREL_NC: ((S + A) | T) - B(S) // R_ARM_THM_MOVW_BREL: ((S + A) | T) - B(S) static inline typename This::Status thm_movw(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address relative_address_base, Arm_address thumb_bit, bool check_overflow) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<32, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Reltype val = (elfcpp::Swap<16, big_endian>::readval(wv) << 16) | elfcpp::Swap<16, big_endian>::readval(wv + 1); Reltype addend = This::extract_thumb_movw_movt_addend(val); Reltype x = (psymval->value(object, addend) | thumb_bit) - relative_address_base; val = This::insert_val_thumb_movw_movt(val, x); elfcpp::Swap<16, big_endian>::writeval(wv, val >> 16); elfcpp::Swap<16, big_endian>::writeval(wv + 1, val & 0xffff); return ((check_overflow && utils::has_overflow<16>(x)) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_THM_MOVT_ABS: S + A (relative address base is 0) // R_ARM_THM_MOVT_PREL: S + A - P // R_ARM_THM_MOVT_BREL: S + A - B(S) static inline typename This::Status thm_movt(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address relative_address_base) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<32, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Reltype val = (elfcpp::Swap<16, big_endian>::readval(wv) << 16) | elfcpp::Swap<16, big_endian>::readval(wv + 1); Reltype addend = This::extract_thumb_movw_movt_addend(val); Reltype x = (psymval->value(object, addend) - relative_address_base) >> 16; val = This::insert_val_thumb_movw_movt(val, x); elfcpp::Swap<16, big_endian>::writeval(wv, val >> 16); elfcpp::Swap<16, big_endian>::writeval(wv + 1, val & 0xffff); return This::STATUS_OKAY; } // R_ARM_THM_ALU_PREL_11_0: ((S + A) | T) - Pa (Thumb32) static inline typename This::Status thm_alu11(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address address, Arm_address thumb_bit) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<32, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Reltype insn = (elfcpp::Swap<16, big_endian>::readval(wv) << 16) | elfcpp::Swap<16, big_endian>::readval(wv + 1); // f e d c b|a|9|8 7 6 5|4|3 2 1 0||f|e d c|b a 9 8|7 6 5 4 3 2 1 0 // ----------------------------------------------------------------------- // ADD{S} 1 1 1 1 0|i|0|1 0 0 0|S|1 1 0 1||0|imm3 |Rd |imm8 // ADDW 1 1 1 1 0|i|1|0 0 0 0|0|1 1 0 1||0|imm3 |Rd |imm8 // ADR[+] 1 1 1 1 0|i|1|0 0 0 0|0|1 1 1 1||0|imm3 |Rd |imm8 // SUB{S} 1 1 1 1 0|i|0|1 1 0 1|S|1 1 0 1||0|imm3 |Rd |imm8 // SUBW 1 1 1 1 0|i|1|0 1 0 1|0|1 1 0 1||0|imm3 |Rd |imm8 // ADR[-] 1 1 1 1 0|i|1|0 1 0 1|0|1 1 1 1||0|imm3 |Rd |imm8 // Determine a sign for the addend. const int sign = ((insn & 0xf8ef0000) == 0xf0ad0000 || (insn & 0xf8ef0000) == 0xf0af0000) ? -1 : 1; // Thumb2 addend encoding: // imm12 := i | imm3 | imm8 int32_t addend = (insn & 0xff) | ((insn & 0x00007000) >> 4) | ((insn & 0x04000000) >> 15); // Apply a sign to the added. addend *= sign; int32_t x = (psymval->value(object, addend) | thumb_bit) - (address & 0xfffffffc); Reltype val = abs(x); // Mask out the value and a distinct part of the ADD/SUB opcode // (bits 7:5 of opword). insn = (insn & 0xfb0f8f00) | (val & 0xff) | ((val & 0x700) << 4) | ((val & 0x800) << 15); // Set the opcode according to whether the value to go in the // place is negative. if (x < 0) insn |= 0x00a00000; elfcpp::Swap<16, big_endian>::writeval(wv, insn >> 16); elfcpp::Swap<16, big_endian>::writeval(wv + 1, insn & 0xffff); return ((val > 0xfff) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_THM_PC8: S + A - Pa (Thumb) static inline typename This::Status thm_pc8(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address address) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<16, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype insn = elfcpp::Swap<16, big_endian>::readval(wv); Reltype addend = ((insn & 0x00ff) << 2); int32_t x = (psymval->value(object, addend) - (address & 0xfffffffc)); Reltype val = abs(x); insn = (insn & 0xff00) | ((val & 0x03fc) >> 2); elfcpp::Swap<16, big_endian>::writeval(wv, insn); return ((val > 0x03fc) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_THM_PC12: S + A - Pa (Thumb32) static inline typename This::Status thm_pc12(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, Arm_address address) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; typedef typename elfcpp::Swap<32, big_endian>::Valtype Reltype; Valtype* wv = reinterpret_cast<Valtype*>(view); Reltype insn = (elfcpp::Swap<16, big_endian>::readval(wv) << 16) | elfcpp::Swap<16, big_endian>::readval(wv + 1); // Determine a sign for the addend (positive if the U bit is 1). const int sign = (insn & 0x00800000) ? 1 : -1; int32_t addend = (insn & 0xfff); // Apply a sign to the added. addend *= sign; int32_t x = (psymval->value(object, addend) - (address & 0xfffffffc)); Reltype val = abs(x); // Mask out and apply the value and the U bit. insn = (insn & 0xff7ff000) | (val & 0xfff); // Set the U bit according to whether the value to go in the // place is positive. if (x >= 0) insn |= 0x00800000; elfcpp::Swap<16, big_endian>::writeval(wv, insn >> 16); elfcpp::Swap<16, big_endian>::writeval(wv + 1, insn & 0xffff); return ((val > 0xfff) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // R_ARM_V4BX static inline typename This::Status v4bx(const Relocate_info<32, big_endian>* relinfo, unsigned char* view, const Arm_relobj<big_endian>* object, const Arm_address address, const bool is_interworking) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<32, big_endian>::readval(wv); // Ensure that we have a BX instruction. gold_assert((val & 0x0ffffff0) == 0x012fff10); const uint32_t reg = (val & 0xf); if (is_interworking && reg != 0xf) { Stub_table<big_endian>* stub_table = object->stub_table(relinfo->data_shndx); gold_assert(stub_table != NULL); Arm_v4bx_stub* stub = stub_table->find_arm_v4bx_stub(reg); gold_assert(stub != NULL); int32_t veneer_address = stub_table->address() + stub->offset() - 8 - address; gold_assert((veneer_address <= ARM_MAX_FWD_BRANCH_OFFSET) && (veneer_address >= ARM_MAX_BWD_BRANCH_OFFSET)); // Replace with a branch to veneer (B <addr>) val = (val & 0xf0000000) | 0x0a000000 | ((veneer_address >> 2) & 0x00ffffff); } else { // Preserve Rm (lowest four bits) and the condition code // (highest four bits). Other bits encode MOV PC,Rm. val = (val & 0xf000000f) | 0x01a0f000; } elfcpp::Swap<32, big_endian>::writeval(wv, val); return This::STATUS_OKAY; } // R_ARM_ALU_PC_G0_NC: ((S + A) | T) - P // R_ARM_ALU_PC_G0: ((S + A) | T) - P // R_ARM_ALU_PC_G1_NC: ((S + A) | T) - P // R_ARM_ALU_PC_G1: ((S + A) | T) - P // R_ARM_ALU_PC_G2: ((S + A) | T) - P // R_ARM_ALU_SB_G0_NC: ((S + A) | T) - B(S) // R_ARM_ALU_SB_G0: ((S + A) | T) - B(S) // R_ARM_ALU_SB_G1_NC: ((S + A) | T) - B(S) // R_ARM_ALU_SB_G1: ((S + A) | T) - B(S) // R_ARM_ALU_SB_G2: ((S + A) | T) - B(S) static inline typename This::Status arm_grp_alu(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, const int group, Arm_address address, Arm_address thumb_bit, bool check_overflow) { gold_assert(group >= 0 && group < 3); typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype insn = elfcpp::Swap<32, big_endian>::readval(wv); // ALU group relocations are allowed only for the ADD/SUB instructions. // (0x00800000 - ADD, 0x00400000 - SUB) const Valtype opcode = insn & 0x01e00000; if (opcode != 0x00800000 && opcode != 0x00400000) return This::STATUS_BAD_RELOC; // Determine a sign for the addend. const int sign = (opcode == 0x00800000) ? 1 : -1; // shifter = rotate_imm * 2 const uint32_t shifter = (insn & 0xf00) >> 7; // Initial addend value. int32_t addend = insn & 0xff; // Rotate addend right by shifter. addend = (addend >> shifter) | (addend << (32 - shifter)); // Apply a sign to the added. addend *= sign; int32_t x = ((psymval->value(object, addend) | thumb_bit) - address); Valtype gn = Arm_relocate_functions::calc_grp_gn(abs(x), group); // Check for overflow if required if (check_overflow && (Arm_relocate_functions::calc_grp_residual(abs(x), group) != 0)) return This::STATUS_OVERFLOW; // Mask out the value and the ADD/SUB part of the opcode; take care // not to destroy the S bit. insn &= 0xff1ff000; // Set the opcode according to whether the value to go in the // place is negative. insn |= ((x < 0) ? 0x00400000 : 0x00800000); // Encode the offset (encoded Gn). insn |= gn; elfcpp::Swap<32, big_endian>::writeval(wv, insn); return This::STATUS_OKAY; } // R_ARM_LDR_PC_G0: S + A - P // R_ARM_LDR_PC_G1: S + A - P // R_ARM_LDR_PC_G2: S + A - P // R_ARM_LDR_SB_G0: S + A - B(S) // R_ARM_LDR_SB_G1: S + A - B(S) // R_ARM_LDR_SB_G2: S + A - B(S) static inline typename This::Status arm_grp_ldr(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, const int group, Arm_address address) { gold_assert(group >= 0 && group < 3); typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype insn = elfcpp::Swap<32, big_endian>::readval(wv); const int sign = (insn & 0x00800000) ? 1 : -1; int32_t addend = (insn & 0xfff) * sign; int32_t x = (psymval->value(object, addend) - address); // Calculate the relevant G(n-1) value to obtain this stage residual. Valtype residual = Arm_relocate_functions::calc_grp_residual(abs(x), group - 1); if (residual >= 0x1000) return This::STATUS_OVERFLOW; // Mask out the value and U bit. insn &= 0xff7ff000; // Set the U bit for non-negative values. if (x >= 0) insn |= 0x00800000; insn |= residual; elfcpp::Swap<32, big_endian>::writeval(wv, insn); return This::STATUS_OKAY; } // R_ARM_LDRS_PC_G0: S + A - P // R_ARM_LDRS_PC_G1: S + A - P // R_ARM_LDRS_PC_G2: S + A - P // R_ARM_LDRS_SB_G0: S + A - B(S) // R_ARM_LDRS_SB_G1: S + A - B(S) // R_ARM_LDRS_SB_G2: S + A - B(S) static inline typename This::Status arm_grp_ldrs(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, const int group, Arm_address address) { gold_assert(group >= 0 && group < 3); typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype insn = elfcpp::Swap<32, big_endian>::readval(wv); const int sign = (insn & 0x00800000) ? 1 : -1; int32_t addend = (((insn & 0xf00) >> 4) + (insn & 0xf)) * sign; int32_t x = (psymval->value(object, addend) - address); // Calculate the relevant G(n-1) value to obtain this stage residual. Valtype residual = Arm_relocate_functions::calc_grp_residual(abs(x), group - 1); if (residual >= 0x100) return This::STATUS_OVERFLOW; // Mask out the value and U bit. insn &= 0xff7ff0f0; // Set the U bit for non-negative values. if (x >= 0) insn |= 0x00800000; insn |= ((residual & 0xf0) << 4) | (residual & 0xf); elfcpp::Swap<32, big_endian>::writeval(wv, insn); return This::STATUS_OKAY; } // R_ARM_LDC_PC_G0: S + A - P // R_ARM_LDC_PC_G1: S + A - P // R_ARM_LDC_PC_G2: S + A - P // R_ARM_LDC_SB_G0: S + A - B(S) // R_ARM_LDC_SB_G1: S + A - B(S) // R_ARM_LDC_SB_G2: S + A - B(S) static inline typename This::Status arm_grp_ldc(unsigned char* view, const Sized_relobj_file<32, big_endian>* object, const Symbol_value<32>* psymval, const int group, Arm_address address) { gold_assert(group >= 0 && group < 3); typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype insn = elfcpp::Swap<32, big_endian>::readval(wv); const int sign = (insn & 0x00800000) ? 1 : -1; int32_t addend = ((insn & 0xff) << 2) * sign; int32_t x = (psymval->value(object, addend) - address); // Calculate the relevant G(n-1) value to obtain this stage residual. Valtype residual = Arm_relocate_functions::calc_grp_residual(abs(x), group - 1); if ((residual & 0x3) != 0 || residual >= 0x400) return This::STATUS_OVERFLOW; // Mask out the value and U bit. insn &= 0xff7fff00; // Set the U bit for non-negative values. if (x >= 0) insn |= 0x00800000; insn |= (residual >> 2); elfcpp::Swap<32, big_endian>::writeval(wv, insn); return This::STATUS_OKAY; } }; // Relocate ARM long branches. This handles relocation types // R_ARM_CALL, R_ARM_JUMP24, R_ARM_PLT32 and R_ARM_XPC25. // If IS_WEAK_UNDEFINED_WITH_PLT is true. The target symbol is weakly // undefined and we do not use PLT in this relocation. In such a case, // the branch is converted into an NOP. template<bool big_endian> typename Arm_relocate_functions<big_endian>::Status Arm_relocate_functions<big_endian>::arm_branch_common( unsigned int r_type, const Relocate_info<32, big_endian>* relinfo, unsigned char* view, const Sized_symbol<32>* gsym, const Arm_relobj<big_endian>* object, unsigned int r_sym, const Symbol_value<32>* psymval, Arm_address address, Arm_address thumb_bit, bool is_weakly_undefined_without_plt) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); Valtype val = elfcpp::Swap<32, big_endian>::readval(wv); bool insn_is_b = (((val >> 28) & 0xf) <= 0xe) && ((val & 0x0f000000UL) == 0x0a000000UL); bool insn_is_uncond_bl = (val & 0xff000000UL) == 0xeb000000UL; bool insn_is_cond_bl = (((val >> 28) & 0xf) < 0xe) && ((val & 0x0f000000UL) == 0x0b000000UL); bool insn_is_blx = (val & 0xfe000000UL) == 0xfa000000UL; bool insn_is_any_branch = (val & 0x0e000000UL) == 0x0a000000UL; // Check that the instruction is valid. if (r_type == elfcpp::R_ARM_CALL) { if (!insn_is_uncond_bl && !insn_is_blx) return This::STATUS_BAD_RELOC; } else if (r_type == elfcpp::R_ARM_JUMP24) { if (!insn_is_b && !insn_is_cond_bl) return This::STATUS_BAD_RELOC; } else if (r_type == elfcpp::R_ARM_PLT32) { if (!insn_is_any_branch) return This::STATUS_BAD_RELOC; } else if (r_type == elfcpp::R_ARM_XPC25) { // FIXME: AAELF document IH0044C does not say much about it other // than it being obsolete. if (!insn_is_any_branch) return This::STATUS_BAD_RELOC; } else gold_unreachable(); // A branch to an undefined weak symbol is turned into a jump to // the next instruction unless a PLT entry will be created. // Do the same for local undefined symbols. // The jump to the next instruction is optimized as a NOP depending // on the architecture. const Target_arm<big_endian>* arm_target = Target_arm<big_endian>::default_target(); if (is_weakly_undefined_without_plt) { gold_assert(!parameters->options().relocatable()); Valtype cond = val & 0xf0000000U; if (arm_target->may_use_arm_nop()) val = cond | 0x0320f000; else val = cond | 0x01a00000; // Using pre-UAL nop: mov r0, r0. elfcpp::Swap<32, big_endian>::writeval(wv, val); return This::STATUS_OKAY; } Valtype addend = utils::sign_extend<26>(val << 2); Valtype branch_target = psymval->value(object, addend); int32_t branch_offset = branch_target - address; // We need a stub if the branch offset is too large or if we need // to switch mode. bool may_use_blx = arm_target->may_use_blx(); Reloc_stub* stub = NULL; if (!parameters->options().relocatable() && (utils::has_overflow<26>(branch_offset) || ((thumb_bit != 0) && !(may_use_blx && r_type == elfcpp::R_ARM_CALL)))) { Valtype unadjusted_branch_target = psymval->value(object, 0); Stub_type stub_type = Reloc_stub::stub_type_for_reloc(r_type, address, unadjusted_branch_target, (thumb_bit != 0)); if (stub_type != arm_stub_none) { Stub_table<big_endian>* stub_table = object->stub_table(relinfo->data_shndx); gold_assert(stub_table != NULL); Reloc_stub::Key stub_key(stub_type, gsym, object, r_sym, addend); stub = stub_table->find_reloc_stub(stub_key); gold_assert(stub != NULL); thumb_bit = stub->stub_template()->entry_in_thumb_mode() ? 1 : 0; branch_target = stub_table->address() + stub->offset() + addend; branch_offset = branch_target - address; gold_assert(!utils::has_overflow<26>(branch_offset)); } } // At this point, if we still need to switch mode, the instruction // must either be a BLX or a BL that can be converted to a BLX. if (thumb_bit != 0) { // Turn BL to BLX. gold_assert(may_use_blx && r_type == elfcpp::R_ARM_CALL); val = (val & 0xffffff) | 0xfa000000 | ((branch_offset & 2) << 23); } val = utils::bit_select(val, (branch_offset >> 2), 0xffffffUL); elfcpp::Swap<32, big_endian>::writeval(wv, val); return (utils::has_overflow<26>(branch_offset) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // Relocate THUMB long branches. This handles relocation types // R_ARM_THM_CALL, R_ARM_THM_JUMP24 and R_ARM_THM_XPC22. // If IS_WEAK_UNDEFINED_WITH_PLT is true. The target symbol is weakly // undefined and we do not use PLT in this relocation. In such a case, // the branch is converted into an NOP. template<bool big_endian> typename Arm_relocate_functions<big_endian>::Status Arm_relocate_functions<big_endian>::thumb_branch_common( unsigned int r_type, const Relocate_info<32, big_endian>* relinfo, unsigned char* view, const Sized_symbol<32>* gsym, const Arm_relobj<big_endian>* object, unsigned int r_sym, const Symbol_value<32>* psymval, Arm_address address, Arm_address thumb_bit, bool is_weakly_undefined_without_plt) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); uint32_t upper_insn = elfcpp::Swap<16, big_endian>::readval(wv); uint32_t lower_insn = elfcpp::Swap<16, big_endian>::readval(wv + 1); // FIXME: These tests are too loose and do not take THUMB/THUMB-2 difference // into account. bool is_bl_insn = (lower_insn & 0x1000U) == 0x1000U; bool is_blx_insn = (lower_insn & 0x1000U) == 0x0000U; // Check that the instruction is valid. if (r_type == elfcpp::R_ARM_THM_CALL) { if (!is_bl_insn && !is_blx_insn) return This::STATUS_BAD_RELOC; } else if (r_type == elfcpp::R_ARM_THM_JUMP24) { // This cannot be a BLX. if (!is_bl_insn) return This::STATUS_BAD_RELOC; } else if (r_type == elfcpp::R_ARM_THM_XPC22) { // Check for Thumb to Thumb call. if (!is_blx_insn) return This::STATUS_BAD_RELOC; if (thumb_bit != 0) { gold_warning(_("%s: Thumb BLX instruction targets " "thumb function '%s'."), object->name().c_str(), (gsym ? gsym->name() : "(local)")); // Convert BLX to BL. lower_insn |= 0x1000U; } } else gold_unreachable(); // A branch to an undefined weak symbol is turned into a jump to // the next instruction unless a PLT entry will be created. // The jump to the next instruction is optimized as a NOP.W for // Thumb-2 enabled architectures. const Target_arm<big_endian>* arm_target = Target_arm<big_endian>::default_target(); if (is_weakly_undefined_without_plt) { gold_assert(!parameters->options().relocatable()); if (arm_target->may_use_thumb2_nop()) { elfcpp::Swap<16, big_endian>::writeval(wv, 0xf3af); elfcpp::Swap<16, big_endian>::writeval(wv + 1, 0x8000); } else { elfcpp::Swap<16, big_endian>::writeval(wv, 0xe000); elfcpp::Swap<16, big_endian>::writeval(wv + 1, 0xbf00); } return This::STATUS_OKAY; } int32_t addend = This::thumb32_branch_offset(upper_insn, lower_insn); Arm_address branch_target = psymval->value(object, addend); // For BLX, bit 1 of target address comes from bit 1 of base address. bool may_use_blx = arm_target->may_use_blx(); if (thumb_bit == 0 && may_use_blx) branch_target = utils::bit_select(branch_target, address, 0x2); int32_t branch_offset = branch_target - address; // We need a stub if the branch offset is too large or if we need // to switch mode. bool thumb2 = arm_target->using_thumb2(); if (!parameters->options().relocatable() && ((!thumb2 && utils::has_overflow<23>(branch_offset)) || (thumb2 && utils::has_overflow<25>(branch_offset)) || ((thumb_bit == 0) && (((r_type == elfcpp::R_ARM_THM_CALL) && !may_use_blx) || r_type == elfcpp::R_ARM_THM_JUMP24)))) { Arm_address unadjusted_branch_target = psymval->value(object, 0); Stub_type stub_type = Reloc_stub::stub_type_for_reloc(r_type, address, unadjusted_branch_target, (thumb_bit != 0)); if (stub_type != arm_stub_none) { Stub_table<big_endian>* stub_table = object->stub_table(relinfo->data_shndx); gold_assert(stub_table != NULL); Reloc_stub::Key stub_key(stub_type, gsym, object, r_sym, addend); Reloc_stub* stub = stub_table->find_reloc_stub(stub_key); gold_assert(stub != NULL); thumb_bit = stub->stub_template()->entry_in_thumb_mode() ? 1 : 0; branch_target = stub_table->address() + stub->offset() + addend; if (thumb_bit == 0 && may_use_blx) branch_target = utils::bit_select(branch_target, address, 0x2); branch_offset = branch_target - address; } } // At this point, if we still need to switch mode, the instruction // must either be a BLX or a BL that can be converted to a BLX. if (thumb_bit == 0) { gold_assert(may_use_blx && (r_type == elfcpp::R_ARM_THM_CALL || r_type == elfcpp::R_ARM_THM_XPC22)); // Make sure this is a BLX. lower_insn &= ~0x1000U; } else { // Make sure this is a BL. lower_insn |= 0x1000U; } // For a BLX instruction, make sure that the relocation is rounded up // to a word boundary. This follows the semantics of the instruction // which specifies that bit 1 of the target address will come from bit // 1 of the base address. if ((lower_insn & 0x5000U) == 0x4000U) gold_assert((branch_offset & 3) == 0); // Put BRANCH_OFFSET back into the insn. Assumes two's complement. // We use the Thumb-2 encoding, which is safe even if dealing with // a Thumb-1 instruction by virtue of our overflow check above. */ upper_insn = This::thumb32_branch_upper(upper_insn, branch_offset); lower_insn = This::thumb32_branch_lower(lower_insn, branch_offset); elfcpp::Swap<16, big_endian>::writeval(wv, upper_insn); elfcpp::Swap<16, big_endian>::writeval(wv + 1, lower_insn); gold_assert(!utils::has_overflow<25>(branch_offset)); return ((thumb2 ? utils::has_overflow<25>(branch_offset) : utils::has_overflow<23>(branch_offset)) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // Relocate THUMB-2 long conditional branches. // If IS_WEAK_UNDEFINED_WITH_PLT is true. The target symbol is weakly // undefined and we do not use PLT in this relocation. In such a case, // the branch is converted into an NOP. template<bool big_endian> typename Arm_relocate_functions<big_endian>::Status Arm_relocate_functions<big_endian>::thm_jump19( unsigned char* view, const Arm_relobj<big_endian>* object, const Symbol_value<32>* psymval, Arm_address address, Arm_address thumb_bit) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(view); uint32_t upper_insn = elfcpp::Swap<16, big_endian>::readval(wv); uint32_t lower_insn = elfcpp::Swap<16, big_endian>::readval(wv + 1); int32_t addend = This::thumb32_cond_branch_offset(upper_insn, lower_insn); Arm_address branch_target = psymval->value(object, addend); int32_t branch_offset = branch_target - address; // ??? Should handle interworking? GCC might someday try to // use this for tail calls. // FIXME: We do support thumb entry to PLT yet. if (thumb_bit == 0) { gold_error(_("conditional branch to PLT in THUMB-2 not supported yet.")); return This::STATUS_BAD_RELOC; } // Put RELOCATION back into the insn. upper_insn = This::thumb32_cond_branch_upper(upper_insn, branch_offset); lower_insn = This::thumb32_cond_branch_lower(lower_insn, branch_offset); // Put the relocated value back in the object file: elfcpp::Swap<16, big_endian>::writeval(wv, upper_insn); elfcpp::Swap<16, big_endian>::writeval(wv + 1, lower_insn); return (utils::has_overflow<21>(branch_offset) ? This::STATUS_OVERFLOW : This::STATUS_OKAY); } // Get the GOT section, creating it if necessary. template<bool big_endian> Arm_output_data_got<big_endian>* Target_arm<big_endian>::got_section(Symbol_table* symtab, Layout* layout) { if (this->got_ == NULL) { gold_assert(symtab != NULL && layout != NULL); this->got_ = new Arm_output_data_got<big_endian>(symtab, layout); layout->add_output_section_data(".got", elfcpp::SHT_PROGBITS, (elfcpp::SHF_ALLOC | elfcpp::SHF_WRITE), this->got_, ORDER_DATA, false); // The old GNU linker creates a .got.plt section. We just // create another set of data in the .got section. Note that we // always create a PLT if we create a GOT, although the PLT // might be empty. this->got_plt_ = new Output_data_space(4, "** GOT PLT"); layout->add_output_section_data(".got", elfcpp::SHT_PROGBITS, (elfcpp::SHF_ALLOC | elfcpp::SHF_WRITE), this->got_plt_, ORDER_DATA, false); // The first three entries are reserved. this->got_plt_->set_current_data_size(3 * 4); // Define _GLOBAL_OFFSET_TABLE_ at the start of the PLT. symtab->define_in_output_data("_GLOBAL_OFFSET_TABLE_", NULL, Symbol_table::PREDEFINED, this->got_plt_, 0, 0, elfcpp::STT_OBJECT, elfcpp::STB_LOCAL, elfcpp::STV_HIDDEN, 0, false, false); } return this->got_; } // Get the dynamic reloc section, creating it if necessary. template<bool big_endian> typename Target_arm<big_endian>::Reloc_section* Target_arm<big_endian>::rel_dyn_section(Layout* layout) { if (this->rel_dyn_ == NULL) { gold_assert(layout != NULL); this->rel_dyn_ = new Reloc_section(parameters->options().combreloc()); layout->add_output_section_data(".rel.dyn", elfcpp::SHT_REL, elfcpp::SHF_ALLOC, this->rel_dyn_, ORDER_DYNAMIC_RELOCS, false); } return this->rel_dyn_; } // Insn_template methods. // Return byte size of an instruction template. size_t Insn_template::size() const { switch (this->type()) { case THUMB16_TYPE: case THUMB16_SPECIAL_TYPE: return 2; case ARM_TYPE: case THUMB32_TYPE: case DATA_TYPE: return 4; default: gold_unreachable(); } } // Return alignment of an instruction template. unsigned Insn_template::alignment() const { switch (this->type()) { case THUMB16_TYPE: case THUMB16_SPECIAL_TYPE: case THUMB32_TYPE: return 2; case ARM_TYPE: case DATA_TYPE: return 4; default: gold_unreachable(); } } // Stub_template methods. Stub_template::Stub_template( Stub_type type, const Insn_template* insns, size_t insn_count) : type_(type), insns_(insns), insn_count_(insn_count), alignment_(1), entry_in_thumb_mode_(false), relocs_() { off_t offset = 0; // Compute byte size and alignment of stub template. for (size_t i = 0; i < insn_count; i++) { unsigned insn_alignment = insns[i].alignment(); size_t insn_size = insns[i].size(); gold_assert((offset & (insn_alignment - 1)) == 0); this->alignment_ = std::max(this->alignment_, insn_alignment); switch (insns[i].type()) { case Insn_template::THUMB16_TYPE: case Insn_template::THUMB16_SPECIAL_TYPE: if (i == 0) this->entry_in_thumb_mode_ = true; break; case Insn_template::THUMB32_TYPE: if (insns[i].r_type() != elfcpp::R_ARM_NONE) this->relocs_.push_back(Reloc(i, offset)); if (i == 0) this->entry_in_thumb_mode_ = true; break; case Insn_template::ARM_TYPE: // Handle cases where the target is encoded within the // instruction. if (insns[i].r_type() == elfcpp::R_ARM_JUMP24) this->relocs_.push_back(Reloc(i, offset)); break; case Insn_template::DATA_TYPE: // Entry point cannot be data. gold_assert(i != 0); this->relocs_.push_back(Reloc(i, offset)); break; default: gold_unreachable(); } offset += insn_size; } this->size_ = offset; } // Stub methods. // Template to implement do_write for a specific target endianness. template<bool big_endian> void inline Stub::do_fixed_endian_write(unsigned char* view, section_size_type view_size) { const Stub_template* stub_template = this->stub_template(); const Insn_template* insns = stub_template->insns(); // FIXME: We do not handle BE8 encoding yet. unsigned char* pov = view; for (size_t i = 0; i < stub_template->insn_count(); i++) { switch (insns[i].type()) { case Insn_template::THUMB16_TYPE: elfcpp::Swap<16, big_endian>::writeval(pov, insns[i].data() & 0xffff); break; case Insn_template::THUMB16_SPECIAL_TYPE: elfcpp::Swap<16, big_endian>::writeval( pov, this->thumb16_special(i)); break; case Insn_template::THUMB32_TYPE: { uint32_t hi = (insns[i].data() >> 16) & 0xffff; uint32_t lo = insns[i].data() & 0xffff; elfcpp::Swap<16, big_endian>::writeval(pov, hi); elfcpp::Swap<16, big_endian>::writeval(pov + 2, lo); } break; case Insn_template::ARM_TYPE: case Insn_template::DATA_TYPE: elfcpp::Swap<32, big_endian>::writeval(pov, insns[i].data()); break; default: gold_unreachable(); } pov += insns[i].size(); } gold_assert(static_cast<section_size_type>(pov - view) == view_size); } // Reloc_stub::Key methods. // Dump a Key as a string for debugging. std::string Reloc_stub::Key::name() const { if (this->r_sym_ == invalid_index) { // Global symbol key name // <stub-type>:<symbol name>:<addend>. const std::string sym_name = this->u_.symbol->name(); // We need to print two hex number and two colons. So just add 100 bytes // to the symbol name size. size_t len = sym_name.size() + 100; char* buffer = new char[len]; int c = snprintf(buffer, len, "%d:%s:%x", this->stub_type_, sym_name.c_str(), this->addend_); gold_assert(c > 0 && c < static_cast<int>(len)); delete[] buffer; return std::string(buffer); } else { // local symbol key name // <stub-type>:<object>:<r_sym>:<addend>. const size_t len = 200; char buffer[len]; int c = snprintf(buffer, len, "%d:%p:%u:%x", this->stub_type_, this->u_.relobj, this->r_sym_, this->addend_); gold_assert(c > 0 && c < static_cast<int>(len)); return std::string(buffer); } } // Reloc_stub methods. // Determine the type of stub needed, if any, for a relocation of R_TYPE at // LOCATION to DESTINATION. // This code is based on the arm_type_of_stub function in // bfd/elf32-arm.c. We have changed the interface a little to keep the Stub // class simple. Stub_type Reloc_stub::stub_type_for_reloc( unsigned int r_type, Arm_address location, Arm_address destination, bool target_is_thumb) { Stub_type stub_type = arm_stub_none; // This is a bit ugly but we want to avoid using a templated class for // big and little endianities. bool may_use_blx; bool should_force_pic_veneer; bool thumb2; bool thumb_only; if (parameters->target().is_big_endian()) { const Target_arm<true>* big_endian_target = Target_arm<true>::default_target(); may_use_blx = big_endian_target->may_use_blx(); should_force_pic_veneer = big_endian_target->should_force_pic_veneer(); thumb2 = big_endian_target->using_thumb2(); thumb_only = big_endian_target->using_thumb_only(); } else { const Target_arm<false>* little_endian_target = Target_arm<false>::default_target(); may_use_blx = little_endian_target->may_use_blx(); should_force_pic_veneer = little_endian_target->should_force_pic_veneer(); thumb2 = little_endian_target->using_thumb2(); thumb_only = little_endian_target->using_thumb_only(); } int64_t branch_offset; if (r_type == elfcpp::R_ARM_THM_CALL || r_type == elfcpp::R_ARM_THM_JUMP24) { // For THUMB BLX instruction, bit 1 of target comes from bit 1 of the // base address (instruction address + 4). if ((r_type == elfcpp::R_ARM_THM_CALL) && may_use_blx && !target_is_thumb) destination = utils::bit_select(destination, location, 0x2); branch_offset = static_cast<int64_t>(destination) - location; // Handle cases where: // - this call goes too far (different Thumb/Thumb2 max // distance) // - it's a Thumb->Arm call and blx is not available, or it's a // Thumb->Arm branch (not bl). A stub is needed in this case. if ((!thumb2 && (branch_offset > THM_MAX_FWD_BRANCH_OFFSET || (branch_offset < THM_MAX_BWD_BRANCH_OFFSET))) || (thumb2 && (branch_offset > THM2_MAX_FWD_BRANCH_OFFSET || (branch_offset < THM2_MAX_BWD_BRANCH_OFFSET))) || ((!target_is_thumb) && (((r_type == elfcpp::R_ARM_THM_CALL) && !may_use_blx) || (r_type == elfcpp::R_ARM_THM_JUMP24)))) { if (target_is_thumb) { // Thumb to thumb. if (!thumb_only) { stub_type = (parameters->options().shared() || should_force_pic_veneer) // PIC stubs. ? ((may_use_blx && (r_type == elfcpp::R_ARM_THM_CALL)) // V5T and above. Stub starts with ARM code, so // we must be able to switch mode before // reaching it, which is only possible for 'bl' // (ie R_ARM_THM_CALL relocation). ? arm_stub_long_branch_any_thumb_pic // On V4T, use Thumb code only. : arm_stub_long_branch_v4t_thumb_thumb_pic) // non-PIC stubs. : ((may_use_blx && (r_type == elfcpp::R_ARM_THM_CALL)) ? arm_stub_long_branch_any_any // V5T and above. : arm_stub_long_branch_v4t_thumb_thumb); // V4T. } else { stub_type = (parameters->options().shared() || should_force_pic_veneer) ? arm_stub_long_branch_thumb_only_pic // PIC stub. : arm_stub_long_branch_thumb_only; // non-PIC stub. } } else { // Thumb to arm. // FIXME: We should check that the input section is from an // object that has interwork enabled. stub_type = (parameters->options().shared() || should_force_pic_veneer) // PIC stubs. ? ((may_use_blx && (r_type == elfcpp::R_ARM_THM_CALL)) ? arm_stub_long_branch_any_arm_pic // V5T and above. : arm_stub_long_branch_v4t_thumb_arm_pic) // V4T. // non-PIC stubs. : ((may_use_blx && (r_type == elfcpp::R_ARM_THM_CALL)) ? arm_stub_long_branch_any_any // V5T and above. : arm_stub_long_branch_v4t_thumb_arm); // V4T. // Handle v4t short branches. if ((stub_type == arm_stub_long_branch_v4t_thumb_arm) && (branch_offset <= THM_MAX_FWD_BRANCH_OFFSET) && (branch_offset >= THM_MAX_BWD_BRANCH_OFFSET)) stub_type = arm_stub_short_branch_v4t_thumb_arm; } } } else if (r_type == elfcpp::R_ARM_CALL || r_type == elfcpp::R_ARM_JUMP24 || r_type == elfcpp::R_ARM_PLT32) { branch_offset = static_cast<int64_t>(destination) - location; if (target_is_thumb) { // Arm to thumb. // FIXME: We should check that the input section is from an // object that has interwork enabled. // We have an extra 2-bytes reach because of // the mode change (bit 24 (H) of BLX encoding). if (branch_offset > (ARM_MAX_FWD_BRANCH_OFFSET + 2) || (branch_offset < ARM_MAX_BWD_BRANCH_OFFSET) || ((r_type == elfcpp::R_ARM_CALL) && !may_use_blx) || (r_type == elfcpp::R_ARM_JUMP24) || (r_type == elfcpp::R_ARM_PLT32)) { stub_type = (parameters->options().shared() || should_force_pic_veneer) // PIC stubs. ? (may_use_blx ? arm_stub_long_branch_any_thumb_pic// V5T and above. : arm_stub_long_branch_v4t_arm_thumb_pic) // V4T stub. // non-PIC stubs. : (may_use_blx ? arm_stub_long_branch_any_any // V5T and above. : arm_stub_long_branch_v4t_arm_thumb); // V4T. } } else { // Arm to arm. if (branch_offset > ARM_MAX_FWD_BRANCH_OFFSET || (branch_offset < ARM_MAX_BWD_BRANCH_OFFSET)) { stub_type = (parameters->options().shared() || should_force_pic_veneer) ? arm_stub_long_branch_any_arm_pic // PIC stubs. : arm_stub_long_branch_any_any; /// non-PIC. } } } return stub_type; } // Cortex_a8_stub methods. // Return the instruction for a THUMB16_SPECIAL_TYPE instruction template. // I is the position of the instruction template in the stub template. uint16_t Cortex_a8_stub::do_thumb16_special(size_t i) { // The only use of this is to copy condition code from a conditional // branch being worked around to the corresponding conditional branch in // to the stub. gold_assert(this->stub_template()->type() == arm_stub_a8_veneer_b_cond && i == 0); uint16_t data = this->stub_template()->insns()[i].data(); gold_assert((data & 0xff00U) == 0xd000U); data |= ((this->original_insn_ >> 22) & 0xf) << 8; return data; } // Stub_factory methods. Stub_factory::Stub_factory() { // The instruction template sequences are declared as static // objects and initialized first time the constructor runs. // Arm/Thumb -> Arm/Thumb long branch stub. On V5T and above, use blx // to reach the stub if necessary. static const Insn_template elf32_arm_stub_long_branch_any_any[] = { Insn_template::arm_insn(0xe51ff004), // ldr pc, [pc, #-4] Insn_template::data_word(0, elfcpp::R_ARM_ABS32, 0), // dcd R_ARM_ABS32(X) }; // V4T Arm -> Thumb long branch stub. Used on V4T where blx is not // available. static const Insn_template elf32_arm_stub_long_branch_v4t_arm_thumb[] = { Insn_template::arm_insn(0xe59fc000), // ldr ip, [pc, #0] Insn_template::arm_insn(0xe12fff1c), // bx ip Insn_template::data_word(0, elfcpp::R_ARM_ABS32, 0), // dcd R_ARM_ABS32(X) }; // Thumb -> Thumb long branch stub. Used on M-profile architectures. static const Insn_template elf32_arm_stub_long_branch_thumb_only[] = { Insn_template::thumb16_insn(0xb401), // push {r0} Insn_template::thumb16_insn(0x4802), // ldr r0, [pc, #8] Insn_template::thumb16_insn(0x4684), // mov ip, r0 Insn_template::thumb16_insn(0xbc01), // pop {r0} Insn_template::thumb16_insn(0x4760), // bx ip Insn_template::thumb16_insn(0xbf00), // nop Insn_template::data_word(0, elfcpp::R_ARM_ABS32, 0), // dcd R_ARM_ABS32(X) }; // V4T Thumb -> Thumb long branch stub. Using the stack is not // allowed. static const Insn_template elf32_arm_stub_long_branch_v4t_thumb_thumb[] = { Insn_template::thumb16_insn(0x4778), // bx pc Insn_template::thumb16_insn(0x46c0), // nop Insn_template::arm_insn(0xe59fc000), // ldr ip, [pc, #0] Insn_template::arm_insn(0xe12fff1c), // bx ip Insn_template::data_word(0, elfcpp::R_ARM_ABS32, 0), // dcd R_ARM_ABS32(X) }; // V4T Thumb -> ARM long branch stub. Used on V4T where blx is not // available. static const Insn_template elf32_arm_stub_long_branch_v4t_thumb_arm[] = { Insn_template::thumb16_insn(0x4778), // bx pc Insn_template::thumb16_insn(0x46c0), // nop Insn_template::arm_insn(0xe51ff004), // ldr pc, [pc, #-4] Insn_template::data_word(0, elfcpp::R_ARM_ABS32, 0), // dcd R_ARM_ABS32(X) }; // V4T Thumb -> ARM short branch stub. Shorter variant of the above // one, when the destination is close enough. static const Insn_template elf32_arm_stub_short_branch_v4t_thumb_arm[] = { Insn_template::thumb16_insn(0x4778), // bx pc Insn_template::thumb16_insn(0x46c0), // nop Insn_template::arm_rel_insn(0xea000000, -8), // b (X-8) }; // ARM/Thumb -> ARM long branch stub, PIC. On V5T and above, use // blx to reach the stub if necessary. static const Insn_template elf32_arm_stub_long_branch_any_arm_pic[] = { Insn_template::arm_insn(0xe59fc000), // ldr r12, [pc] Insn_template::arm_insn(0xe08ff00c), // add pc, pc, ip Insn_template::data_word(0, elfcpp::R_ARM_REL32, -4), // dcd R_ARM_REL32(X-4) }; // ARM/Thumb -> Thumb long branch stub, PIC. On V5T and above, use // blx to reach the stub if necessary. We can not add into pc; // it is not guaranteed to mode switch (different in ARMv6 and // ARMv7). static const Insn_template elf32_arm_stub_long_branch_any_thumb_pic[] = { Insn_template::arm_insn(0xe59fc004), // ldr r12, [pc, #4] Insn_template::arm_insn(0xe08fc00c), // add ip, pc, ip Insn_template::arm_insn(0xe12fff1c), // bx ip Insn_template::data_word(0, elfcpp::R_ARM_REL32, 0), // dcd R_ARM_REL32(X) }; // V4T ARM -> ARM long branch stub, PIC. static const Insn_template elf32_arm_stub_long_branch_v4t_arm_thumb_pic[] = { Insn_template::arm_insn(0xe59fc004), // ldr ip, [pc, #4] Insn_template::arm_insn(0xe08fc00c), // add ip, pc, ip Insn_template::arm_insn(0xe12fff1c), // bx ip Insn_template::data_word(0, elfcpp::R_ARM_REL32, 0), // dcd R_ARM_REL32(X) }; // V4T Thumb -> ARM long branch stub, PIC. static const Insn_template elf32_arm_stub_long_branch_v4t_thumb_arm_pic[] = { Insn_template::thumb16_insn(0x4778), // bx pc Insn_template::thumb16_insn(0x46c0), // nop Insn_template::arm_insn(0xe59fc000), // ldr ip, [pc, #0] Insn_template::arm_insn(0xe08cf00f), // add pc, ip, pc Insn_template::data_word(0, elfcpp::R_ARM_REL32, -4), // dcd R_ARM_REL32(X) }; // Thumb -> Thumb long branch stub, PIC. Used on M-profile // architectures. static const Insn_template elf32_arm_stub_long_branch_thumb_only_pic[] = { Insn_template::thumb16_insn(0xb401), // push {r0} Insn_template::thumb16_insn(0x4802), // ldr r0, [pc, #8] Insn_template::thumb16_insn(0x46fc), // mov ip, pc Insn_template::thumb16_insn(0x4484), // add ip, r0 Insn_template::thumb16_insn(0xbc01), // pop {r0} Insn_template::thumb16_insn(0x4760), // bx ip Insn_template::data_word(0, elfcpp::R_ARM_REL32, 4), // dcd R_ARM_REL32(X) }; // V4T Thumb -> Thumb long branch stub, PIC. Using the stack is not // allowed. static const Insn_template elf32_arm_stub_long_branch_v4t_thumb_thumb_pic[] = { Insn_template::thumb16_insn(0x4778), // bx pc Insn_template::thumb16_insn(0x46c0), // nop Insn_template::arm_insn(0xe59fc004), // ldr ip, [pc, #4] Insn_template::arm_insn(0xe08fc00c), // add ip, pc, ip Insn_template::arm_insn(0xe12fff1c), // bx ip Insn_template::data_word(0, elfcpp::R_ARM_REL32, 0), // dcd R_ARM_REL32(X) }; // Cortex-A8 erratum-workaround stubs. // Stub used for conditional branches (which may be beyond +/-1MB away, // so we can't use a conditional branch to reach this stub). // original code: // // b<cond> X // after: // static const Insn_template elf32_arm_stub_a8_veneer_b_cond[] = { Insn_template::thumb16_bcond_insn(0xd001), // b<cond>.n true Insn_template::thumb32_b_insn(0xf000b800, -4), // b.w after Insn_template::thumb32_b_insn(0xf000b800, -4) // true: // b.w X }; // Stub used for b.w and bl.w instructions. static const Insn_template elf32_arm_stub_a8_veneer_b[] = { Insn_template::thumb32_b_insn(0xf000b800, -4) // b.w dest }; static const Insn_template elf32_arm_stub_a8_veneer_bl[] = { Insn_template::thumb32_b_insn(0xf000b800, -4) // b.w dest }; // Stub used for Thumb-2 blx.w instructions. We modified the original blx.w // instruction (which switches to ARM mode) to point to this stub. Jump to // the real destination using an ARM-mode branch. static const Insn_template elf32_arm_stub_a8_veneer_blx[] = { Insn_template::arm_rel_insn(0xea000000, -8) // b dest }; // Stub used to provide an interworking for R_ARM_V4BX relocation // (bx r[n] instruction). static const Insn_template elf32_arm_stub_v4_veneer_bx[] = { Insn_template::arm_insn(0xe3100001), // tst r<n>, #1 Insn_template::arm_insn(0x01a0f000), // moveq pc, r<n> Insn_template::arm_insn(0xe12fff10) // bx r<n> }; // Fill in the stub template look-up table. Stub templates are constructed // per instance of Stub_factory for fast look-up without locking // in a thread-enabled environment. this->stub_templates_[arm_stub_none] = new Stub_template(arm_stub_none, NULL, 0); #define DEF_STUB(x) \ do \ { \ size_t array_size \ = sizeof(elf32_arm_stub_##x) / sizeof(elf32_arm_stub_##x[0]); \ Stub_type type = arm_stub_##x; \ this->stub_templates_[type] = \ new Stub_template(type, elf32_arm_stub_##x, array_size); \ } \ while (0); DEF_STUBS #undef DEF_STUB } // Stub_table methods. // Remove all Cortex-A8 stub. template<bool big_endian> void Stub_table<big_endian>::remove_all_cortex_a8_stubs() { for (Cortex_a8_stub_list::iterator p = this->cortex_a8_stubs_.begin(); p != this->cortex_a8_stubs_.end(); ++p) delete p->second; this->cortex_a8_stubs_.clear(); } // Relocate one stub. This is a helper for Stub_table::relocate_stubs(). template<bool big_endian> void Stub_table<big_endian>::relocate_stub( Stub* stub, const Relocate_info<32, big_endian>* relinfo, Target_arm<big_endian>* arm_target, Output_section* output_section, unsigned char* view, Arm_address address, section_size_type view_size) { const Stub_template* stub_template = stub->stub_template(); if (stub_template->reloc_count() != 0) { // Adjust view to cover the stub only. section_size_type offset = stub->offset(); section_size_type stub_size = stub_template->size(); gold_assert(offset + stub_size <= view_size); arm_target->relocate_stub(stub, relinfo, output_section, view + offset, address + offset, stub_size); } } // Relocate all stubs in this stub table. template<bool big_endian> void Stub_table<big_endian>::relocate_stubs( const Relocate_info<32, big_endian>* relinfo, Target_arm<big_endian>* arm_target, Output_section* output_section, unsigned char* view, Arm_address address, section_size_type view_size) { // If we are passed a view bigger than the stub table's. we need to // adjust the view. gold_assert(address == this->address() && (view_size == static_cast<section_size_type>(this->data_size()))); // Relocate all relocation stubs. for (typename Reloc_stub_map::const_iterator p = this->reloc_stubs_.begin(); p != this->reloc_stubs_.end(); ++p) this->relocate_stub(p->second, relinfo, arm_target, output_section, view, address, view_size); // Relocate all Cortex-A8 stubs. for (Cortex_a8_stub_list::iterator p = this->cortex_a8_stubs_.begin(); p != this->cortex_a8_stubs_.end(); ++p) this->relocate_stub(p->second, relinfo, arm_target, output_section, view, address, view_size); // Relocate all ARM V4BX stubs. for (Arm_v4bx_stub_list::iterator p = this->arm_v4bx_stubs_.begin(); p != this->arm_v4bx_stubs_.end(); ++p) { if (*p != NULL) this->relocate_stub(*p, relinfo, arm_target, output_section, view, address, view_size); } } // Write out the stubs to file. template<bool big_endian> void Stub_table<big_endian>::do_write(Output_file* of) { 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); // Write relocation stubs. for (typename Reloc_stub_map::const_iterator p = this->reloc_stubs_.begin(); p != this->reloc_stubs_.end(); ++p) { Reloc_stub* stub = p->second; Arm_address address = this->address() + stub->offset(); gold_assert(address == align_address(address, stub->stub_template()->alignment())); stub->write(oview + stub->offset(), stub->stub_template()->size(), big_endian); } // Write Cortex-A8 stubs. for (Cortex_a8_stub_list::const_iterator p = this->cortex_a8_stubs_.begin(); p != this->cortex_a8_stubs_.end(); ++p) { Cortex_a8_stub* stub = p->second; Arm_address address = this->address() + stub->offset(); gold_assert(address == align_address(address, stub->stub_template()->alignment())); stub->write(oview + stub->offset(), stub->stub_template()->size(), big_endian); } // Write ARM V4BX relocation stubs. for (Arm_v4bx_stub_list::const_iterator p = this->arm_v4bx_stubs_.begin(); p != this->arm_v4bx_stubs_.end(); ++p) { if (*p == NULL) continue; Arm_address address = this->address() + (*p)->offset(); gold_assert(address == align_address(address, (*p)->stub_template()->alignment())); (*p)->write(oview + (*p)->offset(), (*p)->stub_template()->size(), big_endian); } of->write_output_view(this->offset(), oview_size, oview); } // Update the data size and address alignment of the stub table at the end // of a relaxation pass. Return true if either the data size or the // alignment changed in this relaxation pass. template<bool big_endian> bool Stub_table<big_endian>::update_data_size_and_addralign() { // Go over all stubs in table to compute data size and address alignment. off_t size = this->reloc_stubs_size_; unsigned addralign = this->reloc_stubs_addralign_; for (Cortex_a8_stub_list::const_iterator p = this->cortex_a8_stubs_.begin(); p != this->cortex_a8_stubs_.end(); ++p) { const Stub_template* stub_template = p->second->stub_template(); addralign = std::max(addralign, stub_template->alignment()); size = (align_address(size, stub_template->alignment()) + stub_template->size()); } for (Arm_v4bx_stub_list::const_iterator p = this->arm_v4bx_stubs_.begin(); p != this->arm_v4bx_stubs_.end(); ++p) { if (*p == NULL) continue; const Stub_template* stub_template = (*p)->stub_template(); addralign = std::max(addralign, stub_template->alignment()); size = (align_address(size, stub_template->alignment()) + stub_template->size()); } // Check if either data size or alignment changed in this pass. // Update prev_data_size_ and prev_addralign_. These will be used // as the current data size and address alignment for the next pass. bool changed = size != this->prev_data_size_; this->prev_data_size_ = size; if (addralign != this->prev_addralign_) changed = true; this->prev_addralign_ = addralign; return changed; } // Finalize the stubs. This sets the offsets of the stubs within the stub // table. It also marks all input sections needing Cortex-A8 workaround. template<bool big_endian> void Stub_table<big_endian>::finalize_stubs() { off_t off = this->reloc_stubs_size_; for (Cortex_a8_stub_list::const_iterator p = this->cortex_a8_stubs_.begin(); p != this->cortex_a8_stubs_.end(); ++p) { Cortex_a8_stub* stub = p->second; const Stub_template* stub_template = stub->stub_template(); uint64_t stub_addralign = stub_template->alignment(); off = align_address(off, stub_addralign); stub->set_offset(off); off += stub_template->size(); // Mark input section so that we can determine later if a code section // needs the Cortex-A8 workaround quickly. Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(stub->relobj()); arm_relobj->mark_section_for_cortex_a8_workaround(stub->shndx()); } for (Arm_v4bx_stub_list::const_iterator p = this->arm_v4bx_stubs_.begin(); p != this->arm_v4bx_stubs_.end(); ++p) { if (*p == NULL) continue; const Stub_template* stub_template = (*p)->stub_template(); uint64_t stub_addralign = stub_template->alignment(); off = align_address(off, stub_addralign); (*p)->set_offset(off); off += stub_template->size(); } gold_assert(off <= this->prev_data_size_); } // Apply Cortex-A8 workaround to an address range between VIEW_ADDRESS // and VIEW_ADDRESS + VIEW_SIZE - 1. VIEW points to the mapped address // of the address range seen by the linker. template<bool big_endian> void Stub_table<big_endian>::apply_cortex_a8_workaround_to_address_range( Target_arm<big_endian>* arm_target, unsigned char* view, Arm_address view_address, section_size_type view_size) { // Cortex-A8 stubs are sorted by addresses of branches being fixed up. for (Cortex_a8_stub_list::const_iterator p = this->cortex_a8_stubs_.lower_bound(view_address); ((p != this->cortex_a8_stubs_.end()) && (p->first < (view_address + view_size))); ++p) { // We do not store the THUMB bit in the LSB of either the branch address // or the stub offset. There is no need to strip the LSB. Arm_address branch_address = p->first; const Cortex_a8_stub* stub = p->second; Arm_address stub_address = this->address() + stub->offset(); // Offset of the branch instruction relative to this view. section_size_type offset = convert_to_section_size_type(branch_address - view_address); gold_assert((offset + 4) <= view_size); arm_target->apply_cortex_a8_workaround(stub, stub_address, view + offset, branch_address); } } // Arm_input_section methods. // Initialize an Arm_input_section. template<bool big_endian> void Arm_input_section<big_endian>::init() { Relobj* relobj = this->relobj(); unsigned int shndx = this->shndx(); // We have to cache original size, alignment and contents to avoid locking // the original file. this->original_addralign_ = convert_types<uint32_t, uint64_t>(relobj->section_addralign(shndx)); // This is not efficient but we expect only a small number of relaxed // input sections for stubs. section_size_type section_size; const unsigned char* section_contents = relobj->section_contents(shndx, §ion_size, false); this->original_size_ = convert_types<uint32_t, uint64_t>(relobj->section_size(shndx)); gold_assert(this->original_contents_ == NULL); this->original_contents_ = new unsigned char[section_size]; memcpy(this->original_contents_, section_contents, section_size); // We want to make this look like the original input section after // output sections are finalized. Output_section* os = relobj->output_section(shndx); off_t offset = relobj->output_section_offset(shndx); gold_assert(os != NULL && !relobj->is_output_section_offset_invalid(shndx)); this->set_address(os->address() + offset); this->set_file_offset(os->offset() + offset); this->set_current_data_size(this->original_size_); this->finalize_data_size(); } template<bool big_endian> void Arm_input_section<big_endian>::do_write(Output_file* of) { // We have to write out the original section content. gold_assert(this->original_contents_ != NULL); of->write(this->offset(), this->original_contents_, this->original_size_); // If this owns a stub table and it is not empty, write it. if (this->is_stub_table_owner() && !this->stub_table_->empty()) this->stub_table_->write(of); } // Finalize data size. template<bool big_endian> void Arm_input_section<big_endian>::set_final_data_size() { off_t off = convert_types<off_t, uint64_t>(this->original_size_); if (this->is_stub_table_owner()) { this->stub_table_->finalize_data_size(); off = align_address(off, this->stub_table_->addralign()); off += this->stub_table_->data_size(); } this->set_data_size(off); } // Reset address and file offset. template<bool big_endian> void Arm_input_section<big_endian>::do_reset_address_and_file_offset() { // Size of the original input section contents. off_t off = convert_types<off_t, uint64_t>(this->original_size_); // If this is a stub table owner, account for the stub table size. if (this->is_stub_table_owner()) { Stub_table<big_endian>* stub_table = this->stub_table_; // Reset the stub table's address and file offset. The // current data size for child will be updated after that. stub_table_->reset_address_and_file_offset(); off = align_address(off, stub_table_->addralign()); off += stub_table->current_data_size(); } this->set_current_data_size(off); } // Arm_exidx_cantunwind methods. // Write this to Output file OF for a fixed endianness. template<bool big_endian> void Arm_exidx_cantunwind::do_fixed_endian_write(Output_file* of) { off_t offset = this->offset(); const section_size_type oview_size = 8; unsigned char* const oview = of->get_output_view(offset, oview_size); typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(oview); Output_section* os = this->relobj_->output_section(this->shndx_); gold_assert(os != NULL); Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(this->relobj_); Arm_address output_offset = arm_relobj->get_output_section_offset(this->shndx_); Arm_address section_start; section_size_type section_size; // Find out the end of the text section referred by this. if (output_offset != Arm_relobj<big_endian>::invalid_address) { section_start = os->address() + output_offset; const Arm_exidx_input_section* exidx_input_section = arm_relobj->exidx_input_section_by_link(this->shndx_); gold_assert(exidx_input_section != NULL); section_size = convert_to_section_size_type(exidx_input_section->text_size()); } else { // Currently this only happens for a relaxed section. const Output_relaxed_input_section* poris = os->find_relaxed_input_section(this->relobj_, this->shndx_); gold_assert(poris != NULL); section_start = poris->address(); section_size = convert_to_section_size_type(poris->data_size()); } // We always append this to the end of an EXIDX section. Arm_address output_address = section_start + section_size; // Write out the entry. The first word either points to the beginning // or after the end of a text section. The second word is the special // EXIDX_CANTUNWIND value. uint32_t prel31_offset = output_address - this->address(); if (utils::has_overflow<31>(offset)) gold_error(_("PREL31 overflow in EXIDX_CANTUNWIND entry")); elfcpp::Swap<32, big_endian>::writeval(wv, prel31_offset & 0x7fffffffU); elfcpp::Swap<32, big_endian>::writeval(wv + 1, elfcpp::EXIDX_CANTUNWIND); of->write_output_view(this->offset(), oview_size, oview); } // Arm_exidx_merged_section methods. // Constructor for Arm_exidx_merged_section. // EXIDX_INPUT_SECTION points to the unmodified EXIDX input section. // SECTION_OFFSET_MAP points to a section offset map describing how // parts of the input section are mapped to output. DELETED_BYTES is // the number of bytes deleted from the EXIDX input section. Arm_exidx_merged_section::Arm_exidx_merged_section( const Arm_exidx_input_section& exidx_input_section, const Arm_exidx_section_offset_map& section_offset_map, uint32_t deleted_bytes) : Output_relaxed_input_section(exidx_input_section.relobj(), exidx_input_section.shndx(), exidx_input_section.addralign()), exidx_input_section_(exidx_input_section), section_offset_map_(section_offset_map) { // If we retain or discard the whole EXIDX input section, we would // not be here. gold_assert(deleted_bytes != 0 && deleted_bytes != this->exidx_input_section_.size()); // Fix size here so that we do not need to implement set_final_data_size. uint32_t size = exidx_input_section.size() - deleted_bytes; this->set_data_size(size); this->fix_data_size(); // Allocate buffer for section contents and build contents. this->section_contents_ = new unsigned char[size]; } // Build the contents of a merged EXIDX output section. void Arm_exidx_merged_section::build_contents( const unsigned char* original_contents, section_size_type original_size) { // Go over spans of input offsets and write only those that are not // discarded. section_offset_type in_start = 0; section_offset_type out_start = 0; section_offset_type in_max = convert_types<section_offset_type>(original_size); section_offset_type out_max = convert_types<section_offset_type>(this->data_size()); for (Arm_exidx_section_offset_map::const_iterator p = this->section_offset_map_.begin(); p != this->section_offset_map_.end(); ++p) { section_offset_type in_end = p->first; gold_assert(in_end >= in_start); section_offset_type out_end = p->second; size_t in_chunk_size = convert_types<size_t>(in_end - in_start + 1); if (out_end != -1) { size_t out_chunk_size = convert_types<size_t>(out_end - out_start + 1); gold_assert(out_chunk_size == in_chunk_size && in_end < in_max && out_end < out_max); memcpy(this->section_contents_ + out_start, original_contents + in_start, out_chunk_size); out_start += out_chunk_size; } in_start += in_chunk_size; } } // Given an input OBJECT, an input section index SHNDX within that // object, and an OFFSET relative to the start of that input // section, return whether or not the corresponding offset within // the output section is known. If this function returns true, it // sets *POUTPUT to the output offset. The value -1 indicates that // this input offset is being discarded. bool Arm_exidx_merged_section::do_output_offset( const Relobj* relobj, unsigned int shndx, section_offset_type offset, section_offset_type* poutput) const { // We only handle offsets for the original EXIDX input section. if (relobj != this->exidx_input_section_.relobj() || shndx != this->exidx_input_section_.shndx()) return false; section_offset_type section_size = convert_types<section_offset_type>(this->exidx_input_section_.size()); if (offset < 0 || offset >= section_size) // Input offset is out of valid range. *poutput = -1; else { // We need to look up the section offset map to determine the output // offset. Find the reference point in map that is first offset // bigger than or equal to this offset. Arm_exidx_section_offset_map::const_iterator p = this->section_offset_map_.lower_bound(offset); // The section offset maps are build such that this should not happen if // input offset is in the valid range. gold_assert(p != this->section_offset_map_.end()); // We need to check if this is dropped. section_offset_type ref = p->first; section_offset_type mapped_ref = p->second; if (mapped_ref != Arm_exidx_input_section::invalid_offset) // Offset is present in output. *poutput = mapped_ref + (offset - ref); else // Offset is discarded owing to EXIDX entry merging. *poutput = -1; } return true; } // Write this to output file OF. void Arm_exidx_merged_section::do_write(Output_file* of) { off_t offset = this->offset(); const section_size_type oview_size = this->data_size(); unsigned char* const oview = of->get_output_view(offset, oview_size); Output_section* os = this->relobj()->output_section(this->shndx()); gold_assert(os != NULL); memcpy(oview, this->section_contents_, oview_size); of->write_output_view(this->offset(), oview_size, oview); } // Arm_exidx_fixup methods. // Append an EXIDX_CANTUNWIND in the current output section if the last entry // is not an EXIDX_CANTUNWIND entry already. The new EXIDX_CANTUNWIND entry // points to the end of the last seen EXIDX section. void Arm_exidx_fixup::add_exidx_cantunwind_as_needed() { if (this->last_unwind_type_ != UT_EXIDX_CANTUNWIND && this->last_input_section_ != NULL) { Relobj* relobj = this->last_input_section_->relobj(); unsigned int text_shndx = this->last_input_section_->link(); Arm_exidx_cantunwind* cantunwind = new Arm_exidx_cantunwind(relobj, text_shndx); this->exidx_output_section_->add_output_section_data(cantunwind); this->last_unwind_type_ = UT_EXIDX_CANTUNWIND; } } // Process an EXIDX section entry in input. Return whether this entry // can be deleted in the output. SECOND_WORD in the second word of the // EXIDX entry. bool Arm_exidx_fixup::process_exidx_entry(uint32_t second_word) { bool delete_entry; if (second_word == elfcpp::EXIDX_CANTUNWIND) { // Merge if previous entry is also an EXIDX_CANTUNWIND. delete_entry = this->last_unwind_type_ == UT_EXIDX_CANTUNWIND; this->last_unwind_type_ = UT_EXIDX_CANTUNWIND; } else if ((second_word & 0x80000000) != 0) { // Inlined unwinding data. Merge if equal to previous. delete_entry = (merge_exidx_entries_ && this->last_unwind_type_ == UT_INLINED_ENTRY && this->last_inlined_entry_ == second_word); this->last_unwind_type_ = UT_INLINED_ENTRY; this->last_inlined_entry_ = second_word; } else { // Normal table entry. In theory we could merge these too, // but duplicate entries are likely to be much less common. delete_entry = false; this->last_unwind_type_ = UT_NORMAL_ENTRY; } return delete_entry; } // Update the current section offset map during EXIDX section fix-up. // If there is no map, create one. INPUT_OFFSET is the offset of a // reference point, DELETED_BYTES is the number of deleted by in the // section so far. If DELETE_ENTRY is true, the reference point and // all offsets after the previous reference point are discarded. void Arm_exidx_fixup::update_offset_map( section_offset_type input_offset, section_size_type deleted_bytes, bool delete_entry) { if (this->section_offset_map_ == NULL) this->section_offset_map_ = new Arm_exidx_section_offset_map(); section_offset_type output_offset; if (delete_entry) output_offset = Arm_exidx_input_section::invalid_offset; else output_offset = input_offset - deleted_bytes; (*this->section_offset_map_)[input_offset] = output_offset; } // Process EXIDX_INPUT_SECTION for EXIDX entry merging. Return the number of // bytes deleted. SECTION_CONTENTS points to the contents of the EXIDX // section and SECTION_SIZE is the number of bytes pointed by SECTION_CONTENTS. // If some entries are merged, also store a pointer to a newly created // Arm_exidx_section_offset_map object in *PSECTION_OFFSET_MAP. The caller // owns the map and is responsible for releasing it after use. template<bool big_endian> uint32_t Arm_exidx_fixup::process_exidx_section( const Arm_exidx_input_section* exidx_input_section, const unsigned char* section_contents, section_size_type section_size, Arm_exidx_section_offset_map** psection_offset_map) { Relobj* relobj = exidx_input_section->relobj(); unsigned shndx = exidx_input_section->shndx(); if ((section_size % 8) != 0) { // Something is wrong with this section. Better not touch it. gold_error(_("uneven .ARM.exidx section size in %s section %u"), relobj->name().c_str(), shndx); this->last_input_section_ = exidx_input_section; this->last_unwind_type_ = UT_NONE; return 0; } uint32_t deleted_bytes = 0; bool prev_delete_entry = false; gold_assert(this->section_offset_map_ == NULL); for (section_size_type i = 0; i < section_size; i += 8) { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; const Valtype* wv = reinterpret_cast<const Valtype*>(section_contents + i + 4); uint32_t second_word = elfcpp::Swap<32, big_endian>::readval(wv); bool delete_entry = this->process_exidx_entry(second_word); // Entry deletion causes changes in output offsets. We use a std::map // to record these. And entry (x, y) means input offset x // is mapped to output offset y. If y is invalid_offset, then x is // dropped in the output. Because of the way std::map::lower_bound // works, we record the last offset in a region w.r.t to keeping or // dropping. If there is no entry (x0, y0) for an input offset x0, // the output offset y0 of it is determined by the output offset y1 of // the smallest input offset x1 > x0 that there is an (x1, y1) entry // in the map. If y1 is not -1, then y0 = y1 + x0 - x1. Otherwise, y1 // y0 is also -1. if (delete_entry != prev_delete_entry && i != 0) this->update_offset_map(i - 1, deleted_bytes, prev_delete_entry); // Update total deleted bytes for this entry. if (delete_entry) deleted_bytes += 8; prev_delete_entry = delete_entry; } // If section offset map is not NULL, make an entry for the end of // section. if (this->section_offset_map_ != NULL) update_offset_map(section_size - 1, deleted_bytes, prev_delete_entry); *psection_offset_map = this->section_offset_map_; this->section_offset_map_ = NULL; this->last_input_section_ = exidx_input_section; // Set the first output text section so that we can link the EXIDX output // section to it. Ignore any EXIDX input section that is completely merged. if (this->first_output_text_section_ == NULL && deleted_bytes != section_size) { unsigned int link = exidx_input_section->link(); Output_section* os = relobj->output_section(link); gold_assert(os != NULL); this->first_output_text_section_ = os; } return deleted_bytes; } // Arm_output_section methods. // Create a stub group for input sections from BEGIN to END. OWNER // points to the input section to be the owner a new stub table. template<bool big_endian> void Arm_output_section<big_endian>::create_stub_group( Input_section_list::const_iterator begin, Input_section_list::const_iterator end, Input_section_list::const_iterator owner, Target_arm<big_endian>* target, std::vector<Output_relaxed_input_section*>* new_relaxed_sections, const Task* task) { // We use a different kind of relaxed section in an EXIDX section. // The static casting from Output_relaxed_input_section to // Arm_input_section is invalid in an EXIDX section. We are okay // because we should not be calling this for an EXIDX section. gold_assert(this->type() != elfcpp::SHT_ARM_EXIDX); // Currently we convert ordinary input sections into relaxed sections only // at this point but we may want to support creating relaxed input section // very early. So we check here to see if owner is already a relaxed // section. Arm_input_section<big_endian>* arm_input_section; if (owner->is_relaxed_input_section()) { arm_input_section = Arm_input_section<big_endian>::as_arm_input_section( owner->relaxed_input_section()); } else { gold_assert(owner->is_input_section()); // Create a new relaxed input section. We need to lock the original // file. Task_lock_obj<Object> tl(task, owner->relobj()); arm_input_section = target->new_arm_input_section(owner->relobj(), owner->shndx()); new_relaxed_sections->push_back(arm_input_section); } // Create a stub table. Stub_table<big_endian>* stub_table = target->new_stub_table(arm_input_section); arm_input_section->set_stub_table(stub_table); Input_section_list::const_iterator p = begin; Input_section_list::const_iterator prev_p; // Look for input sections or relaxed input sections in [begin ... end]. do { if (p->is_input_section() || p->is_relaxed_input_section()) { // The stub table information for input sections live // in their objects. Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(p->relobj()); arm_relobj->set_stub_table(p->shndx(), stub_table); } prev_p = p++; } while (prev_p != end); } // Group input sections for stub generation. GROUP_SIZE is roughly the limit // of stub groups. We grow a stub group by adding input section until the // size is just below GROUP_SIZE. The last input section will be converted // into a stub table. If STUB_ALWAYS_AFTER_BRANCH is false, we also add // input section after the stub table, effectively double the group size. // // This is similar to the group_sections() function in elf32-arm.c but is // implemented differently. template<bool big_endian> void Arm_output_section<big_endian>::group_sections( section_size_type group_size, bool stubs_always_after_branch, Target_arm<big_endian>* target, const Task* task) { // We only care about sections containing code. if ((this->flags() & elfcpp::SHF_EXECINSTR) == 0) return; // States for grouping. typedef enum { // No group is being built. NO_GROUP, // A group is being built but the stub table is not found yet. // We keep group a stub group until the size is just under GROUP_SIZE. // The last input section in the group will be used as the stub table. FINDING_STUB_SECTION, // A group is being built and we have already found a stub table. // We enter this state to grow a stub group by adding input section // after the stub table. This effectively doubles the group size. HAS_STUB_SECTION } State; // Any newly created relaxed sections are stored here. std::vector<Output_relaxed_input_section*> new_relaxed_sections; State state = NO_GROUP; section_size_type off = 0; section_size_type group_begin_offset = 0; section_size_type group_end_offset = 0; section_size_type stub_table_end_offset = 0; Input_section_list::const_iterator group_begin = this->input_sections().end(); Input_section_list::const_iterator stub_table = this->input_sections().end(); Input_section_list::const_iterator group_end = this->input_sections().end(); for (Input_section_list::const_iterator p = this->input_sections().begin(); p != this->input_sections().end(); ++p) { section_size_type section_begin_offset = align_address(off, p->addralign()); section_size_type section_end_offset = section_begin_offset + p->data_size(); // Check to see if we should group the previously seen sections. switch (state) { case NO_GROUP: break; case FINDING_STUB_SECTION: // Adding this section makes the group larger than GROUP_SIZE. if (section_end_offset - group_begin_offset >= group_size) { if (stubs_always_after_branch) { gold_assert(group_end != this->input_sections().end()); this->create_stub_group(group_begin, group_end, group_end, target, &new_relaxed_sections, task); state = NO_GROUP; } else { // But wait, there's more! Input sections up to // stub_group_size bytes after the stub table can be // handled by it too. state = HAS_STUB_SECTION; stub_table = group_end; stub_table_end_offset = group_end_offset; } } break; case HAS_STUB_SECTION: // Adding this section makes the post stub-section group larger // than GROUP_SIZE. if (section_end_offset - stub_table_end_offset >= group_size) { gold_assert(group_end != this->input_sections().end()); this->create_stub_group(group_begin, group_end, stub_table, target, &new_relaxed_sections, task); state = NO_GROUP; } break; default: gold_unreachable(); } // If we see an input section and currently there is no group, start // a new one. Skip any empty sections. We look at the data size // instead of calling p->relobj()->section_size() to avoid locking. if ((p->is_input_section() || p->is_relaxed_input_section()) && (p->data_size() != 0)) { if (state == NO_GROUP) { state = FINDING_STUB_SECTION; group_begin = p; group_begin_offset = section_begin_offset; } // Keep track of the last input section seen. group_end = p; group_end_offset = section_end_offset; } off = section_end_offset; } // Create a stub group for any ungrouped sections. if (state == FINDING_STUB_SECTION || state == HAS_STUB_SECTION) { gold_assert(group_end != this->input_sections().end()); this->create_stub_group(group_begin, group_end, (state == FINDING_STUB_SECTION ? group_end : stub_table), target, &new_relaxed_sections, task); } // Convert input section into relaxed input section in a batch. if (!new_relaxed_sections.empty()) this->convert_input_sections_to_relaxed_sections(new_relaxed_sections); // Update the section offsets for (size_t i = 0; i < new_relaxed_sections.size(); ++i) { Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj( new_relaxed_sections[i]->relobj()); unsigned int shndx = new_relaxed_sections[i]->shndx(); // Tell Arm_relobj that this input section is converted. arm_relobj->convert_input_section_to_relaxed_section(shndx); } } // Append non empty text sections in this to LIST in ascending // order of their position in this. template<bool big_endian> void Arm_output_section<big_endian>::append_text_sections_to_list( Text_section_list* list) { gold_assert((this->flags() & elfcpp::SHF_ALLOC) != 0); for (Input_section_list::const_iterator p = this->input_sections().begin(); p != this->input_sections().end(); ++p) { // We only care about plain or relaxed input sections. We also // ignore any merged sections. if ((p->is_input_section() || p->is_relaxed_input_section()) && p->data_size() != 0) list->push_back(Text_section_list::value_type(p->relobj(), p->shndx())); } } template<bool big_endian> void Arm_output_section<big_endian>::fix_exidx_coverage( Layout* layout, const Text_section_list& sorted_text_sections, Symbol_table* symtab, bool merge_exidx_entries, const Task* task) { // We should only do this for the EXIDX output section. gold_assert(this->type() == elfcpp::SHT_ARM_EXIDX); // We don't want the relaxation loop to undo these changes, so we discard // the current saved states and take another one after the fix-up. this->discard_states(); // Remove all input sections. uint64_t address = this->address(); typedef std::list<Output_section::Input_section> Input_section_list; Input_section_list input_sections; this->reset_address_and_file_offset(); this->get_input_sections(address, std::string(""), &input_sections); if (!this->input_sections().empty()) gold_error(_("Found non-EXIDX input sections in EXIDX output section")); // Go through all the known input sections and record them. typedef Unordered_set<Section_id, Section_id_hash> Section_id_set; typedef Unordered_map<Section_id, const Output_section::Input_section*, Section_id_hash> Text_to_exidx_map; Text_to_exidx_map text_to_exidx_map; for (Input_section_list::const_iterator p = input_sections.begin(); p != input_sections.end(); ++p) { // This should never happen. At this point, we should only see // plain EXIDX input sections. gold_assert(!p->is_relaxed_input_section()); text_to_exidx_map[Section_id(p->relobj(), p->shndx())] = &(*p); } Arm_exidx_fixup exidx_fixup(this, merge_exidx_entries); // Go over the sorted text sections. typedef Unordered_set<Section_id, Section_id_hash> Section_id_set; Section_id_set processed_input_sections; for (Text_section_list::const_iterator p = sorted_text_sections.begin(); p != sorted_text_sections.end(); ++p) { Relobj* relobj = p->first; unsigned int shndx = p->second; Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(relobj); const Arm_exidx_input_section* exidx_input_section = arm_relobj->exidx_input_section_by_link(shndx); // If this text section has no EXIDX section or if the EXIDX section // has errors, force an EXIDX_CANTUNWIND entry pointing to the end // of the last seen EXIDX section. if (exidx_input_section == NULL || exidx_input_section->has_errors()) { exidx_fixup.add_exidx_cantunwind_as_needed(); continue; } Relobj* exidx_relobj = exidx_input_section->relobj(); unsigned int exidx_shndx = exidx_input_section->shndx(); Section_id sid(exidx_relobj, exidx_shndx); Text_to_exidx_map::const_iterator iter = text_to_exidx_map.find(sid); if (iter == text_to_exidx_map.end()) { // This is odd. We have not seen this EXIDX input section before. // We cannot do fix-up. If we saw a SECTIONS clause in a script, // issue a warning instead. We assume the user knows what he // or she is doing. Otherwise, this is an error. if (layout->script_options()->saw_sections_clause()) gold_warning(_("unwinding may not work because EXIDX input section" " %u of %s is not in EXIDX output section"), exidx_shndx, exidx_relobj->name().c_str()); else gold_error(_("unwinding may not work because EXIDX input section" " %u of %s is not in EXIDX output section"), exidx_shndx, exidx_relobj->name().c_str()); exidx_fixup.add_exidx_cantunwind_as_needed(); continue; } // We need to access the contents of the EXIDX section, lock the // object here. Task_lock_obj<Object> tl(task, exidx_relobj); section_size_type exidx_size; const unsigned char* exidx_contents = exidx_relobj->section_contents(exidx_shndx, &exidx_size, false); // Fix up coverage and append input section to output data list. Arm_exidx_section_offset_map* section_offset_map = NULL; uint32_t deleted_bytes = exidx_fixup.process_exidx_section<big_endian>(exidx_input_section, exidx_contents, exidx_size, §ion_offset_map); if (deleted_bytes == exidx_input_section->size()) { // The whole EXIDX section got merged. Remove it from output. gold_assert(section_offset_map == NULL); exidx_relobj->set_output_section(exidx_shndx, NULL); // All local symbols defined in this input section will be dropped. // We need to adjust output local symbol count. arm_relobj->set_output_local_symbol_count_needs_update(); } else if (deleted_bytes > 0) { // Some entries are merged. We need to convert this EXIDX input // section into a relaxed section. gold_assert(section_offset_map != NULL); Arm_exidx_merged_section* merged_section = new Arm_exidx_merged_section(*exidx_input_section, *section_offset_map, deleted_bytes); merged_section->build_contents(exidx_contents, exidx_size); const std::string secname = exidx_relobj->section_name(exidx_shndx); this->add_relaxed_input_section(layout, merged_section, secname); arm_relobj->convert_input_section_to_relaxed_section(exidx_shndx); // All local symbols defined in discarded portions of this input // section will be dropped. We need to adjust output local symbol // count. arm_relobj->set_output_local_symbol_count_needs_update(); } else { // Just add back the EXIDX input section. gold_assert(section_offset_map == NULL); const Output_section::Input_section* pis = iter->second; gold_assert(pis->is_input_section()); this->add_script_input_section(*pis); } processed_input_sections.insert(Section_id(exidx_relobj, exidx_shndx)); } // Insert an EXIDX_CANTUNWIND entry at the end of output if necessary. exidx_fixup.add_exidx_cantunwind_as_needed(); // Remove any known EXIDX input sections that are not processed. for (Input_section_list::const_iterator p = input_sections.begin(); p != input_sections.end(); ++p) { if (processed_input_sections.find(Section_id(p->relobj(), p->shndx())) == processed_input_sections.end()) { // We discard a known EXIDX section because its linked // text section has been folded by ICF. We also discard an // EXIDX section with error, the output does not matter in this // case. We do this to avoid triggering asserts. Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(p->relobj()); const Arm_exidx_input_section* exidx_input_section = arm_relobj->exidx_input_section_by_shndx(p->shndx()); gold_assert(exidx_input_section != NULL); if (!exidx_input_section->has_errors()) { unsigned int text_shndx = exidx_input_section->link(); gold_assert(symtab->is_section_folded(p->relobj(), text_shndx)); } // Remove this from link. We also need to recount the // local symbols. p->relobj()->set_output_section(p->shndx(), NULL); arm_relobj->set_output_local_symbol_count_needs_update(); } } // Link exidx output section to the first seen output section and // set correct entry size. this->set_link_section(exidx_fixup.first_output_text_section()); this->set_entsize(8); // Make changes permanent. this->save_states(); this->set_section_offsets_need_adjustment(); } // Link EXIDX output sections to text output sections. template<bool big_endian> void Arm_output_section<big_endian>::set_exidx_section_link() { gold_assert(this->type() == elfcpp::SHT_ARM_EXIDX); if (!this->input_sections().empty()) { Input_section_list::const_iterator p = this->input_sections().begin(); Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(p->relobj()); unsigned exidx_shndx = p->shndx(); const Arm_exidx_input_section* exidx_input_section = arm_relobj->exidx_input_section_by_shndx(exidx_shndx); gold_assert(exidx_input_section != NULL); unsigned int text_shndx = exidx_input_section->link(); Output_section* os = arm_relobj->output_section(text_shndx); this->set_link_section(os); } } // Arm_relobj methods. // Determine if an input section is scannable for stub processing. SHDR is // the header of the section and SHNDX is the section index. OS is the output // section for the input section and SYMTAB is the global symbol table used to // look up ICF information. template<bool big_endian> bool Arm_relobj<big_endian>::section_is_scannable( const elfcpp::Shdr<32, big_endian>& shdr, unsigned int shndx, const Output_section* os, const Symbol_table* symtab) { // Skip any empty sections, unallocated sections or sections whose // type are not SHT_PROGBITS. if (shdr.get_sh_size() == 0 || (shdr.get_sh_flags() & elfcpp::SHF_ALLOC) == 0 || shdr.get_sh_type() != elfcpp::SHT_PROGBITS) return false; // Skip any discarded or ICF'ed sections. if (os == NULL || symtab->is_section_folded(this, shndx)) return false; // If this requires special offset handling, check to see if it is // a relaxed section. If this is not, then it is a merged section that // we cannot handle. if (this->is_output_section_offset_invalid(shndx)) { const Output_relaxed_input_section* poris = os->find_relaxed_input_section(this, shndx); if (poris == NULL) return false; } return true; } // Determine if we want to scan the SHNDX-th section for relocation stubs. // This is a helper for Arm_relobj::scan_sections_for_stubs() below. template<bool big_endian> bool Arm_relobj<big_endian>::section_needs_reloc_stub_scanning( const elfcpp::Shdr<32, big_endian>& shdr, const Relobj::Output_sections& out_sections, const Symbol_table* symtab, const unsigned char* pshdrs) { unsigned int sh_type = shdr.get_sh_type(); if (sh_type != elfcpp::SHT_REL && sh_type != elfcpp::SHT_RELA) return false; // Ignore empty section. off_t sh_size = shdr.get_sh_size(); if (sh_size == 0) return false; // Ignore reloc section with unexpected symbol table. The // error will be reported in the final link. if (this->adjust_shndx(shdr.get_sh_link()) != this->symtab_shndx()) return false; unsigned int reloc_size; if (sh_type == elfcpp::SHT_REL) reloc_size = elfcpp::Elf_sizes<32>::rel_size; else reloc_size = elfcpp::Elf_sizes<32>::rela_size; // Ignore reloc section with unexpected entsize or uneven size. // The error will be reported in the final link. if (reloc_size != shdr.get_sh_entsize() || sh_size % reloc_size != 0) return false; // Ignore reloc section with bad info. This error will be // reported in the final link. unsigned int index = this->adjust_shndx(shdr.get_sh_info()); if (index >= this->shnum()) return false; const unsigned int shdr_size = elfcpp::Elf_sizes<32>::shdr_size; const elfcpp::Shdr<32, big_endian> text_shdr(pshdrs + index * shdr_size); return this->section_is_scannable(text_shdr, index, out_sections[index], symtab); } // Return the output address of either a plain input section or a relaxed // input section. SHNDX is the section index. We define and use this // instead of calling Output_section::output_address because that is slow // for large output. template<bool big_endian> Arm_address Arm_relobj<big_endian>::simple_input_section_output_address( unsigned int shndx, Output_section* os) { if (this->is_output_section_offset_invalid(shndx)) { const Output_relaxed_input_section* poris = os->find_relaxed_input_section(this, shndx); // We do not handle merged sections here. gold_assert(poris != NULL); return poris->address(); } else return os->address() + this->get_output_section_offset(shndx); } // Determine if we want to scan the SHNDX-th section for non-relocation stubs. // This is a helper for Arm_relobj::scan_sections_for_stubs() below. template<bool big_endian> bool Arm_relobj<big_endian>::section_needs_cortex_a8_stub_scanning( const elfcpp::Shdr<32, big_endian>& shdr, unsigned int shndx, Output_section* os, const Symbol_table* symtab) { if (!this->section_is_scannable(shdr, shndx, os, symtab)) return false; // If the section does not cross any 4K-boundaries, it does not need to // be scanned. Arm_address address = this->simple_input_section_output_address(shndx, os); if ((address & ~0xfffU) == ((address + shdr.get_sh_size() - 1) & ~0xfffU)) return false; return true; } // Scan a section for Cortex-A8 workaround. template<bool big_endian> void Arm_relobj<big_endian>::scan_section_for_cortex_a8_erratum( const elfcpp::Shdr<32, big_endian>& shdr, unsigned int shndx, Output_section* os, Target_arm<big_endian>* arm_target) { // Look for the first mapping symbol in this section. It should be // at (shndx, 0). Mapping_symbol_position section_start(shndx, 0); typename Mapping_symbols_info::const_iterator p = this->mapping_symbols_info_.lower_bound(section_start); // There are no mapping symbols for this section. Treat it as a data-only // section. Issue a warning if section is marked as containing // instructions. if (p == this->mapping_symbols_info_.end() || p->first.first != shndx) { if ((this->section_flags(shndx) & elfcpp::SHF_EXECINSTR) != 0) gold_warning(_("cannot scan executable section %u of %s for Cortex-A8 " "erratum because it has no mapping symbols."), shndx, this->name().c_str()); return; } Arm_address output_address = this->simple_input_section_output_address(shndx, os); // Get the section contents. section_size_type input_view_size = 0; const unsigned char* input_view = this->section_contents(shndx, &input_view_size, false); // We need to go through the mapping symbols to determine what to // scan. There are two reasons. First, we should look at THUMB code and // THUMB code only. Second, we only want to look at the 4K-page boundary // to speed up the scanning. while (p != this->mapping_symbols_info_.end() && p->first.first == shndx) { typename Mapping_symbols_info::const_iterator next = this->mapping_symbols_info_.upper_bound(p->first); // Only scan part of a section with THUMB code. if (p->second == 't') { // Determine the end of this range. section_size_type span_start = convert_to_section_size_type(p->first.second); section_size_type span_end; if (next != this->mapping_symbols_info_.end() && next->first.first == shndx) span_end = convert_to_section_size_type(next->first.second); else span_end = convert_to_section_size_type(shdr.get_sh_size()); if (((span_start + output_address) & ~0xfffUL) != ((span_end + output_address - 1) & ~0xfffUL)) { arm_target->scan_span_for_cortex_a8_erratum(this, shndx, span_start, span_end, input_view, output_address); } } p = next; } } // Scan relocations for stub generation. template<bool big_endian> void Arm_relobj<big_endian>::scan_sections_for_stubs( Target_arm<big_endian>* arm_target, const Symbol_table* symtab, const Layout* layout) { unsigned int shnum = this->shnum(); const unsigned int shdr_size = elfcpp::Elf_sizes<32>::shdr_size; // Read the section headers. const unsigned char* pshdrs = this->get_view(this->elf_file()->shoff(), shnum * shdr_size, true, true); // To speed up processing, we set up hash tables for fast lookup of // input offsets to output addresses. this->initialize_input_to_output_maps(); const Relobj::Output_sections& out_sections(this->output_sections()); Relocate_info<32, big_endian> relinfo; relinfo.symtab = symtab; relinfo.layout = layout; relinfo.object = this; // Do relocation stubs scanning. const unsigned char* p = pshdrs + shdr_size; for (unsigned int i = 1; i < shnum; ++i, p += shdr_size) { const elfcpp::Shdr<32, big_endian> shdr(p); if (this->section_needs_reloc_stub_scanning(shdr, out_sections, symtab, pshdrs)) { unsigned int index = this->adjust_shndx(shdr.get_sh_info()); Arm_address output_offset = this->get_output_section_offset(index); Arm_address output_address; if (output_offset != invalid_address) output_address = out_sections[index]->address() + output_offset; else { // Currently this only happens for a relaxed section. const Output_relaxed_input_section* poris = out_sections[index]->find_relaxed_input_section(this, index); gold_assert(poris != NULL); output_address = poris->address(); } // Get the relocations. const unsigned char* prelocs = this->get_view(shdr.get_sh_offset(), shdr.get_sh_size(), true, false); // Get the section contents. This does work for the case in which // we modify the contents of an input section. We need to pass the // output view under such circumstances. section_size_type input_view_size = 0; const unsigned char* input_view = this->section_contents(index, &input_view_size, false); relinfo.reloc_shndx = i; relinfo.data_shndx = index; unsigned int sh_type = shdr.get_sh_type(); unsigned int reloc_size; if (sh_type == elfcpp::SHT_REL) reloc_size = elfcpp::Elf_sizes<32>::rel_size; else reloc_size = elfcpp::Elf_sizes<32>::rela_size; Output_section* os = out_sections[index]; arm_target->scan_section_for_stubs(&relinfo, sh_type, prelocs, shdr.get_sh_size() / reloc_size, os, output_offset == invalid_address, input_view, output_address, input_view_size); } } // Do Cortex-A8 erratum stubs scanning. This has to be done for a section // after its relocation section, if there is one, is processed for // relocation stubs. Merging this loop with the one above would have been // complicated since we would have had to make sure that relocation stub // scanning is done first. if (arm_target->fix_cortex_a8()) { const unsigned char* p = pshdrs + shdr_size; for (unsigned int i = 1; i < shnum; ++i, p += shdr_size) { const elfcpp::Shdr<32, big_endian> shdr(p); if (this->section_needs_cortex_a8_stub_scanning(shdr, i, out_sections[i], symtab)) this->scan_section_for_cortex_a8_erratum(shdr, i, out_sections[i], arm_target); } } // After we've done the relocations, we release the hash tables, // since we no longer need them. this->free_input_to_output_maps(); } // Count the local symbols. The ARM backend needs to know if a symbol // is a THUMB function or not. For global symbols, it is easy because // the Symbol object keeps the ELF symbol type. For local symbol it is // harder because we cannot access this information. So we override the // do_count_local_symbol in parent and scan local symbols to mark // THUMB functions. This is not the most efficient way but I do not want to // slow down other ports by calling a per symbol target hook inside // Sized_relobj_file<size, big_endian>::do_count_local_symbols. template<bool big_endian> void Arm_relobj<big_endian>::do_count_local_symbols( Stringpool_template<char>* pool, Stringpool_template<char>* dynpool) { // We need to fix-up the values of any local symbols whose type are // STT_ARM_TFUNC. // Ask parent to count the local symbols. Sized_relobj_file<32, big_endian>::do_count_local_symbols(pool, dynpool); const unsigned int loccount = this->local_symbol_count(); if (loccount == 0) return; // Initialize the thumb function bit-vector. std::vector<bool> empty_vector(loccount, false); this->local_symbol_is_thumb_function_.swap(empty_vector); // Read the symbol table section header. const unsigned int symtab_shndx = this->symtab_shndx(); elfcpp::Shdr<32, big_endian> symtabshdr(this, this->elf_file()->section_header(symtab_shndx)); gold_assert(symtabshdr.get_sh_type() == elfcpp::SHT_SYMTAB); // Read the local symbols. const int sym_size =elfcpp::Elf_sizes<32>::sym_size; gold_assert(loccount == symtabshdr.get_sh_info()); off_t locsize = loccount * sym_size; const unsigned char* psyms = this->get_view(symtabshdr.get_sh_offset(), locsize, true, true); // For mapping symbol processing, we need to read the symbol names. unsigned int strtab_shndx = this->adjust_shndx(symtabshdr.get_sh_link()); if (strtab_shndx >= this->shnum()) { this->error(_("invalid symbol table name index: %u"), strtab_shndx); return; } elfcpp::Shdr<32, big_endian> strtabshdr(this, this->elf_file()->section_header(strtab_shndx)); if (strtabshdr.get_sh_type() != elfcpp::SHT_STRTAB) { this->error(_("symbol table name section has wrong type: %u"), static_cast<unsigned int>(strtabshdr.get_sh_type())); return; } const char* pnames = reinterpret_cast<const char*>(this->get_view(strtabshdr.get_sh_offset(), strtabshdr.get_sh_size(), false, false)); // Loop over the local symbols and mark any local symbols pointing // to THUMB functions. // Skip the first dummy symbol. psyms += sym_size; typename Sized_relobj_file<32, big_endian>::Local_values* plocal_values = this->local_values(); for (unsigned int i = 1; i < loccount; ++i, psyms += sym_size) { elfcpp::Sym<32, big_endian> sym(psyms); elfcpp::STT st_type = sym.get_st_type(); Symbol_value<32>& lv((*plocal_values)[i]); Arm_address input_value = lv.input_value(); // Check to see if this is a mapping symbol. const char* sym_name = pnames + sym.get_st_name(); if (Target_arm<big_endian>::is_mapping_symbol_name(sym_name)) { bool is_ordinary; unsigned int input_shndx = this->adjust_sym_shndx(i, sym.get_st_shndx(), &is_ordinary); gold_assert(is_ordinary); // Strip of LSB in case this is a THUMB symbol. Mapping_symbol_position msp(input_shndx, input_value & ~1U); this->mapping_symbols_info_[msp] = sym_name[1]; } if (st_type == elfcpp::STT_ARM_TFUNC || (st_type == elfcpp::STT_FUNC && ((input_value & 1) != 0))) { // This is a THUMB function. Mark this and canonicalize the // symbol value by setting LSB. this->local_symbol_is_thumb_function_[i] = true; if ((input_value & 1) == 0) lv.set_input_value(input_value | 1); } } } // Relocate sections. template<bool big_endian> void Arm_relobj<big_endian>::do_relocate_sections( const Symbol_table* symtab, const Layout* layout, const unsigned char* pshdrs, Output_file* of, typename Sized_relobj_file<32, big_endian>::Views* pviews) { // Call parent to relocate sections. Sized_relobj_file<32, big_endian>::do_relocate_sections(symtab, layout, pshdrs, of, pviews); // We do not generate stubs if doing a relocatable link. if (parameters->options().relocatable()) return; // Relocate stub tables. unsigned int shnum = this->shnum(); Target_arm<big_endian>* arm_target = Target_arm<big_endian>::default_target(); Relocate_info<32, big_endian> relinfo; relinfo.symtab = symtab; relinfo.layout = layout; relinfo.object = this; for (unsigned int i = 1; i < shnum; ++i) { Arm_input_section<big_endian>* arm_input_section = arm_target->find_arm_input_section(this, i); if (arm_input_section != NULL && arm_input_section->is_stub_table_owner() && !arm_input_section->stub_table()->empty()) { // We cannot discard a section if it owns a stub table. Output_section* os = this->output_section(i); gold_assert(os != NULL); relinfo.reloc_shndx = elfcpp::SHN_UNDEF; relinfo.reloc_shdr = NULL; relinfo.data_shndx = i; relinfo.data_shdr = pshdrs + i * elfcpp::Elf_sizes<32>::shdr_size; gold_assert((*pviews)[i].view != NULL); // We are passed the output section view. Adjust it to cover the // stub table only. Stub_table<big_endian>* stub_table = arm_input_section->stub_table(); gold_assert((stub_table->address() >= (*pviews)[i].address) && ((stub_table->address() + stub_table->data_size()) <= (*pviews)[i].address + (*pviews)[i].view_size)); off_t offset = stub_table->address() - (*pviews)[i].address; unsigned char* view = (*pviews)[i].view + offset; Arm_address address = stub_table->address(); section_size_type view_size = stub_table->data_size(); stub_table->relocate_stubs(&relinfo, arm_target, os, view, address, view_size); } // Apply Cortex A8 workaround if applicable. if (this->section_has_cortex_a8_workaround(i)) { unsigned char* view = (*pviews)[i].view; Arm_address view_address = (*pviews)[i].address; section_size_type view_size = (*pviews)[i].view_size; Stub_table<big_endian>* stub_table = this->stub_tables_[i]; // Adjust view to cover section. Output_section* os = this->output_section(i); gold_assert(os != NULL); Arm_address section_address = this->simple_input_section_output_address(i, os); uint64_t section_size = this->section_size(i); gold_assert(section_address >= view_address && ((section_address + section_size) <= (view_address + view_size))); unsigned char* section_view = view + (section_address - view_address); // Apply the Cortex-A8 workaround to the output address range // corresponding to this input section. stub_table->apply_cortex_a8_workaround_to_address_range( arm_target, section_view, section_address, section_size); } } } // Find the linked text section of an EXIDX section by looking at the first // relocation. 4.4.1 of the EHABI specifications says that an EXIDX section // must be linked to its associated code section via the sh_link field of // its section header. However, some tools are broken and the link is not // always set. LD just drops such an EXIDX section silently, causing the // associated code not unwindabled. Here we try a little bit harder to // discover the linked code section. // // PSHDR points to the section header of a relocation section of an EXIDX // section. If we can find a linked text section, return true and // store the text section index in the location PSHNDX. Otherwise // return false. template<bool big_endian> bool Arm_relobj<big_endian>::find_linked_text_section( const unsigned char* pshdr, const unsigned char* psyms, unsigned int* pshndx) { elfcpp::Shdr<32, big_endian> shdr(pshdr); // If there is no relocation, we cannot find the linked text section. size_t reloc_size; if (shdr.get_sh_type() == elfcpp::SHT_REL) reloc_size = elfcpp::Elf_sizes<32>::rel_size; else reloc_size = elfcpp::Elf_sizes<32>::rela_size; size_t reloc_count = shdr.get_sh_size() / reloc_size; // Get the relocations. const unsigned char* prelocs = this->get_view(shdr.get_sh_offset(), shdr.get_sh_size(), true, false); // Find the REL31 relocation for the first word of the first EXIDX entry. for (size_t i = 0; i < reloc_count; ++i, prelocs += reloc_size) { Arm_address r_offset; typename elfcpp::Elf_types<32>::Elf_WXword r_info; if (shdr.get_sh_type() == elfcpp::SHT_REL) { typename elfcpp::Rel<32, big_endian> reloc(prelocs); r_info = reloc.get_r_info(); r_offset = reloc.get_r_offset(); } else { typename elfcpp::Rela<32, big_endian> reloc(prelocs); r_info = reloc.get_r_info(); r_offset = reloc.get_r_offset(); } unsigned int r_type = elfcpp::elf_r_type<32>(r_info); if (r_type != elfcpp::R_ARM_PREL31 && r_type != elfcpp::R_ARM_SBREL31) continue; unsigned int r_sym = elfcpp::elf_r_sym<32>(r_info); if (r_sym == 0 || r_sym >= this->local_symbol_count() || r_offset != 0) continue; // This is the relocation for the first word of the first EXIDX entry. // We expect to see a local section symbol. const int sym_size = elfcpp::Elf_sizes<32>::sym_size; elfcpp::Sym<32, big_endian> sym(psyms + r_sym * sym_size); if (sym.get_st_type() == elfcpp::STT_SECTION) { bool is_ordinary; *pshndx = this->adjust_sym_shndx(r_sym, sym.get_st_shndx(), &is_ordinary); gold_assert(is_ordinary); return true; } else return false; } return false; } // Make an EXIDX input section object for an EXIDX section whose index is // SHNDX. SHDR is the section header of the EXIDX section and TEXT_SHNDX // is the section index of the linked text section. template<bool big_endian> void Arm_relobj<big_endian>::make_exidx_input_section( unsigned int shndx, const elfcpp::Shdr<32, big_endian>& shdr, unsigned int text_shndx, const elfcpp::Shdr<32, big_endian>& text_shdr) { // Create an Arm_exidx_input_section object for this EXIDX section. Arm_exidx_input_section* exidx_input_section = new Arm_exidx_input_section(this, shndx, text_shndx, shdr.get_sh_size(), shdr.get_sh_addralign(), text_shdr.get_sh_size()); gold_assert(this->exidx_section_map_[shndx] == NULL); this->exidx_section_map_[shndx] = exidx_input_section; if (text_shndx == elfcpp::SHN_UNDEF || text_shndx >= this->shnum()) { gold_error(_("EXIDX section %s(%u) links to invalid section %u in %s"), this->section_name(shndx).c_str(), shndx, text_shndx, this->name().c_str()); exidx_input_section->set_has_errors(); } else if (this->exidx_section_map_[text_shndx] != NULL) { unsigned other_exidx_shndx = this->exidx_section_map_[text_shndx]->shndx(); gold_error(_("EXIDX sections %s(%u) and %s(%u) both link to text section" "%s(%u) in %s"), this->section_name(shndx).c_str(), shndx, this->section_name(other_exidx_shndx).c_str(), other_exidx_shndx, this->section_name(text_shndx).c_str(), text_shndx, this->name().c_str()); exidx_input_section->set_has_errors(); } else this->exidx_section_map_[text_shndx] = exidx_input_section; // Check section flags of text section. if ((text_shdr.get_sh_flags() & elfcpp::SHF_ALLOC) == 0) { gold_error(_("EXIDX section %s(%u) links to non-allocated section %s(%u) " " in %s"), this->section_name(shndx).c_str(), shndx, this->section_name(text_shndx).c_str(), text_shndx, this->name().c_str()); exidx_input_section->set_has_errors(); } else if ((text_shdr.get_sh_flags() & elfcpp::SHF_EXECINSTR) == 0) // I would like to make this an error but currently ld just ignores // this. gold_warning(_("EXIDX section %s(%u) links to non-executable section " "%s(%u) in %s"), this->section_name(shndx).c_str(), shndx, this->section_name(text_shndx).c_str(), text_shndx, this->name().c_str()); } // Read the symbol information. template<bool big_endian> void Arm_relobj<big_endian>::do_read_symbols(Read_symbols_data* sd) { // Call parent class to read symbol information. Sized_relobj_file<32, big_endian>::do_read_symbols(sd); // If this input file is a binary file, it has no processor // specific flags and attributes section. Input_file::Format format = this->input_file()->format(); if (format != Input_file::FORMAT_ELF) { gold_assert(format == Input_file::FORMAT_BINARY); this->merge_flags_and_attributes_ = false; return; } // Read processor-specific flags in ELF file header. const unsigned char* pehdr = this->get_view(elfcpp::file_header_offset, elfcpp::Elf_sizes<32>::ehdr_size, true, false); elfcpp::Ehdr<32, big_endian> ehdr(pehdr); this->processor_specific_flags_ = ehdr.get_e_flags(); // Go over the section headers and look for .ARM.attributes and .ARM.exidx // sections. std::vector<unsigned int> deferred_exidx_sections; const size_t shdr_size = elfcpp::Elf_sizes<32>::shdr_size; const unsigned char* pshdrs = sd->section_headers->data(); const unsigned char* ps = pshdrs + shdr_size; bool must_merge_flags_and_attributes = false; for (unsigned int i = 1; i < this->shnum(); ++i, ps += shdr_size) { elfcpp::Shdr<32, big_endian> shdr(ps); // Sometimes an object has no contents except the section name string // table and an empty symbol table with the undefined symbol. We // don't want to merge processor-specific flags from such an object. if (shdr.get_sh_type() == elfcpp::SHT_SYMTAB) { // Symbol table is not empty. const elfcpp::Elf_types<32>::Elf_WXword sym_size = elfcpp::Elf_sizes<32>::sym_size; if (shdr.get_sh_size() > sym_size) must_merge_flags_and_attributes = true; } else if (shdr.get_sh_type() != elfcpp::SHT_STRTAB) // If this is neither an empty symbol table nor a string table, // be conservative. must_merge_flags_and_attributes = true; if (shdr.get_sh_type() == elfcpp::SHT_ARM_ATTRIBUTES) { gold_assert(this->attributes_section_data_ == NULL); section_offset_type section_offset = shdr.get_sh_offset(); section_size_type section_size = convert_to_section_size_type(shdr.get_sh_size()); const unsigned char* view = this->get_view(section_offset, section_size, true, false); this->attributes_section_data_ = new Attributes_section_data(view, section_size); } else if (shdr.get_sh_type() == elfcpp::SHT_ARM_EXIDX) { unsigned int text_shndx = this->adjust_shndx(shdr.get_sh_link()); if (text_shndx == elfcpp::SHN_UNDEF) deferred_exidx_sections.push_back(i); else { elfcpp::Shdr<32, big_endian> text_shdr(pshdrs + text_shndx * shdr_size); this->make_exidx_input_section(i, shdr, text_shndx, text_shdr); } // EHABI 4.4.1 requires that SHF_LINK_ORDER flag to be set. if ((shdr.get_sh_flags() & elfcpp::SHF_LINK_ORDER) == 0) gold_warning(_("SHF_LINK_ORDER not set in EXIDX section %s of %s"), this->section_name(i).c_str(), this->name().c_str()); } } // This is rare. if (!must_merge_flags_and_attributes) { gold_assert(deferred_exidx_sections.empty()); this->merge_flags_and_attributes_ = false; return; } // Some tools are broken and they do not set the link of EXIDX sections. // We look at the first relocation to figure out the linked sections. if (!deferred_exidx_sections.empty()) { // We need to go over the section headers again to find the mapping // from sections being relocated to their relocation sections. This is // a bit inefficient as we could do that in the loop above. However, // we do not expect any deferred EXIDX sections normally. So we do not // want to slow down the most common path. typedef Unordered_map<unsigned int, unsigned int> Reloc_map; Reloc_map reloc_map; ps = pshdrs + shdr_size; for (unsigned int i = 1; i < this->shnum(); ++i, ps += shdr_size) { elfcpp::Shdr<32, big_endian> shdr(ps); elfcpp::Elf_Word sh_type = shdr.get_sh_type(); if (sh_type == elfcpp::SHT_REL || sh_type == elfcpp::SHT_RELA) { unsigned int info_shndx = this->adjust_shndx(shdr.get_sh_info()); if (info_shndx >= this->shnum()) gold_error(_("relocation section %u has invalid info %u"), i, info_shndx); Reloc_map::value_type value(info_shndx, i); std::pair<Reloc_map::iterator, bool> result = reloc_map.insert(value); if (!result.second) gold_error(_("section %u has multiple relocation sections " "%u and %u"), info_shndx, i, reloc_map[info_shndx]); } } // Read the symbol table section header. const unsigned int symtab_shndx = this->symtab_shndx(); elfcpp::Shdr<32, big_endian> symtabshdr(this, this->elf_file()->section_header(symtab_shndx)); gold_assert(symtabshdr.get_sh_type() == elfcpp::SHT_SYMTAB); // Read the local symbols. const int sym_size =elfcpp::Elf_sizes<32>::sym_size; const unsigned int loccount = this->local_symbol_count(); gold_assert(loccount == symtabshdr.get_sh_info()); off_t locsize = loccount * sym_size; const unsigned char* psyms = this->get_view(symtabshdr.get_sh_offset(), locsize, true, true); // Process the deferred EXIDX sections. for (unsigned int i = 0; i < deferred_exidx_sections.size(); ++i) { unsigned int shndx = deferred_exidx_sections[i]; elfcpp::Shdr<32, big_endian> shdr(pshdrs + shndx * shdr_size); unsigned int text_shndx = elfcpp::SHN_UNDEF; Reloc_map::const_iterator it = reloc_map.find(shndx); if (it != reloc_map.end()) find_linked_text_section(pshdrs + it->second * shdr_size, psyms, &text_shndx); elfcpp::Shdr<32, big_endian> text_shdr(pshdrs + text_shndx * shdr_size); this->make_exidx_input_section(shndx, shdr, text_shndx, text_shdr); } } } // Process relocations for garbage collection. The ARM target uses .ARM.exidx // sections for unwinding. These sections are referenced implicitly by // text sections linked in the section headers. If we ignore these implicit // references, the .ARM.exidx sections and any .ARM.extab sections they use // will be garbage-collected incorrectly. Hence we override the same function // in the base class to handle these implicit references. template<bool big_endian> void Arm_relobj<big_endian>::do_gc_process_relocs(Symbol_table* symtab, Layout* layout, Read_relocs_data* rd) { // First, call base class method to process relocations in this object. Sized_relobj_file<32, big_endian>::do_gc_process_relocs(symtab, layout, rd); // If --gc-sections is not specified, there is nothing more to do. // This happens when --icf is used but --gc-sections is not. if (!parameters->options().gc_sections()) return; unsigned int shnum = this->shnum(); const unsigned int shdr_size = elfcpp::Elf_sizes<32>::shdr_size; const unsigned char* pshdrs = this->get_view(this->elf_file()->shoff(), shnum * shdr_size, true, true); // Scan section headers for sections of type SHT_ARM_EXIDX. Add references // to these from the linked text sections. const unsigned char* ps = pshdrs + shdr_size; for (unsigned int i = 1; i < shnum; ++i, ps += shdr_size) { elfcpp::Shdr<32, big_endian> shdr(ps); if (shdr.get_sh_type() == elfcpp::SHT_ARM_EXIDX) { // Found an .ARM.exidx section, add it to the set of reachable // sections from its linked text section. unsigned int text_shndx = this->adjust_shndx(shdr.get_sh_link()); symtab->gc()->add_reference(this, text_shndx, this, i); } } } // Update output local symbol count. Owing to EXIDX entry merging, some local // symbols will be removed in output. Adjust output local symbol count // accordingly. We can only changed the static output local symbol count. It // is too late to change the dynamic symbols. template<bool big_endian> void Arm_relobj<big_endian>::update_output_local_symbol_count() { // Caller should check that this needs updating. We want caller checking // because output_local_symbol_count_needs_update() is most likely inlined. gold_assert(this->output_local_symbol_count_needs_update_); gold_assert(this->symtab_shndx() != -1U); if (this->symtab_shndx() == 0) { // This object has no symbols. Weird but legal. return; } // Read the symbol table section header. const unsigned int symtab_shndx = this->symtab_shndx(); elfcpp::Shdr<32, big_endian> symtabshdr(this, this->elf_file()->section_header(symtab_shndx)); gold_assert(symtabshdr.get_sh_type() == elfcpp::SHT_SYMTAB); // Read the local symbols. const int sym_size = elfcpp::Elf_sizes<32>::sym_size; const unsigned int loccount = this->local_symbol_count(); gold_assert(loccount == symtabshdr.get_sh_info()); off_t locsize = loccount * sym_size; const unsigned char* psyms = this->get_view(symtabshdr.get_sh_offset(), locsize, true, true); // Loop over the local symbols. typedef typename Sized_relobj_file<32, big_endian>::Output_sections Output_sections; const Output_sections& out_sections(this->output_sections()); unsigned int shnum = this->shnum(); unsigned int count = 0; // Skip the first, dummy, symbol. psyms += sym_size; for (unsigned int i = 1; i < loccount; ++i, psyms += sym_size) { elfcpp::Sym<32, big_endian> sym(psyms); Symbol_value<32>& lv((*this->local_values())[i]); // This local symbol was already discarded by do_count_local_symbols. if (lv.is_output_symtab_index_set() && !lv.has_output_symtab_entry()) continue; bool is_ordinary; unsigned int shndx = this->adjust_sym_shndx(i, sym.get_st_shndx(), &is_ordinary); if (shndx < shnum) { Output_section* os = out_sections[shndx]; // This local symbol no longer has an output section. Discard it. if (os == NULL) { lv.set_no_output_symtab_entry(); continue; } // Currently we only discard parts of EXIDX input sections. // We explicitly check for a merged EXIDX input section to avoid // calling Output_section_data::output_offset unless necessary. if ((this->get_output_section_offset(shndx) == invalid_address) && (this->exidx_input_section_by_shndx(shndx) != NULL)) { section_offset_type output_offset = os->output_offset(this, shndx, lv.input_value()); if (output_offset == -1) { // This symbol is defined in a part of an EXIDX input section // that is discarded due to entry merging. lv.set_no_output_symtab_entry(); continue; } } } ++count; } this->set_output_local_symbol_count(count); this->output_local_symbol_count_needs_update_ = false; } // Arm_dynobj methods. // Read the symbol information. template<bool big_endian> void Arm_dynobj<big_endian>::do_read_symbols(Read_symbols_data* sd) { // Call parent class to read symbol information. Sized_dynobj<32, big_endian>::do_read_symbols(sd); // Read processor-specific flags in ELF file header. const unsigned char* pehdr = this->get_view(elfcpp::file_header_offset, elfcpp::Elf_sizes<32>::ehdr_size, true, false); elfcpp::Ehdr<32, big_endian> ehdr(pehdr); this->processor_specific_flags_ = ehdr.get_e_flags(); // Read the attributes section if there is one. // We read from the end because gas seems to put it near the end of // the section headers. const size_t shdr_size = elfcpp::Elf_sizes<32>::shdr_size; const unsigned char* ps = sd->section_headers->data() + shdr_size * (this->shnum() - 1); for (unsigned int i = this->shnum(); i > 0; --i, ps -= shdr_size) { elfcpp::Shdr<32, big_endian> shdr(ps); if (shdr.get_sh_type() == elfcpp::SHT_ARM_ATTRIBUTES) { section_offset_type section_offset = shdr.get_sh_offset(); section_size_type section_size = convert_to_section_size_type(shdr.get_sh_size()); const unsigned char* view = this->get_view(section_offset, section_size, true, false); this->attributes_section_data_ = new Attributes_section_data(view, section_size); break; } } } // Stub_addend_reader methods. // Read the addend of a REL relocation of type R_TYPE at VIEW. template<bool big_endian> elfcpp::Elf_types<32>::Elf_Swxword Stub_addend_reader<elfcpp::SHT_REL, big_endian>::operator()( unsigned int r_type, const unsigned char* view, const typename Reloc_types<elfcpp::SHT_REL, 32, big_endian>::Reloc&) const { typedef struct Arm_relocate_functions<big_endian> RelocFuncs; switch (r_type) { case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_PLT32: { typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; const Valtype* wv = reinterpret_cast<const Valtype*>(view); Valtype val = elfcpp::Swap<32, big_endian>::readval(wv); return utils::sign_extend<26>(val << 2); } case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_THM_JUMP24: case elfcpp::R_ARM_THM_XPC22: { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; const Valtype* wv = reinterpret_cast<const Valtype*>(view); Valtype upper_insn = elfcpp::Swap<16, big_endian>::readval(wv); Valtype lower_insn = elfcpp::Swap<16, big_endian>::readval(wv + 1); return RelocFuncs::thumb32_branch_offset(upper_insn, lower_insn); } case elfcpp::R_ARM_THM_JUMP19: { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; const Valtype* wv = reinterpret_cast<const Valtype*>(view); Valtype upper_insn = elfcpp::Swap<16, big_endian>::readval(wv); Valtype lower_insn = elfcpp::Swap<16, big_endian>::readval(wv + 1); return RelocFuncs::thumb32_cond_branch_offset(upper_insn, lower_insn); } default: gold_unreachable(); } } // Arm_output_data_got methods. // Add a GOT pair for R_ARM_TLS_GD32. The creates a pair of GOT entries. // The first one is initialized to be 1, which is the module index for // the main executable and the second one 0. A reloc of the type // R_ARM_TLS_DTPOFF32 will be created for the second GOT entry and will // be applied by gold. GSYM is a global symbol. // template<bool big_endian> void Arm_output_data_got<big_endian>::add_tls_gd32_with_static_reloc( unsigned int got_type, Symbol* gsym) { if (gsym->has_got_offset(got_type)) return; // We are doing a static link. Just mark it as belong to module 1, // the executable. unsigned int got_offset = this->add_constant(1); gsym->set_got_offset(got_type, got_offset); got_offset = this->add_constant(0); this->static_relocs_.push_back(Static_reloc(got_offset, elfcpp::R_ARM_TLS_DTPOFF32, gsym)); } // Same as the above but for a local symbol. template<bool big_endian> void Arm_output_data_got<big_endian>::add_tls_gd32_with_static_reloc( unsigned int got_type, Sized_relobj_file<32, big_endian>* object, unsigned int index) { if (object->local_has_got_offset(index, got_type)) return; // We are doing a static link. Just mark it as belong to module 1, // the executable. unsigned int got_offset = this->add_constant(1); object->set_local_got_offset(index, got_type, got_offset); got_offset = this->add_constant(0); this->static_relocs_.push_back(Static_reloc(got_offset, elfcpp::R_ARM_TLS_DTPOFF32, object, index)); } template<bool big_endian> void Arm_output_data_got<big_endian>::do_write(Output_file* of) { // Call parent to write out GOT. Output_data_got<32, big_endian>::do_write(of); // We are done if there is no fix up. if (this->static_relocs_.empty()) return; gold_assert(parameters->doing_static_link()); 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_segment* tls_segment = this->layout_->tls_segment(); gold_assert(tls_segment != NULL); // The thread pointer $tp points to the TCB, which is followed by the // TLS. So we need to adjust $tp relative addressing by this amount. Arm_address aligned_tcb_size = align_address(ARM_TCB_SIZE, tls_segment->maximum_alignment()); for (size_t i = 0; i < this->static_relocs_.size(); ++i) { Static_reloc& reloc(this->static_relocs_[i]); Arm_address value; if (!reloc.symbol_is_global()) { Sized_relobj_file<32, big_endian>* object = reloc.relobj(); const Symbol_value<32>* psymval = reloc.relobj()->local_symbol(reloc.index()); // We are doing static linking. Issue an error and skip this // relocation if the symbol is undefined or in a discarded_section. bool is_ordinary; unsigned int shndx = psymval->input_shndx(&is_ordinary); if ((shndx == elfcpp::SHN_UNDEF) || (is_ordinary && shndx != elfcpp::SHN_UNDEF && !object->is_section_included(shndx) && !this->symbol_table_->is_section_folded(object, shndx))) { gold_error(_("undefined or discarded local symbol %u from " " object %s in GOT"), reloc.index(), reloc.relobj()->name().c_str()); continue; } value = psymval->value(object, 0); } else { const Symbol* gsym = reloc.symbol(); gold_assert(gsym != NULL); if (gsym->is_forwarder()) gsym = this->symbol_table_->resolve_forwards(gsym); // We are doing static linking. Issue an error and skip this // relocation if the symbol is undefined or in a discarded_section // unless it is a weakly_undefined symbol. if ((gsym->is_defined_in_discarded_section() || gsym->is_undefined()) && !gsym->is_weak_undefined()) { gold_error(_("undefined or discarded symbol %s in GOT"), gsym->name()); continue; } if (!gsym->is_weak_undefined()) { const Sized_symbol<32>* sym = static_cast<const Sized_symbol<32>*>(gsym); value = sym->value(); } else value = 0; } unsigned got_offset = reloc.got_offset(); gold_assert(got_offset < oview_size); typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(oview + got_offset); Valtype x; switch (reloc.r_type()) { case elfcpp::R_ARM_TLS_DTPOFF32: x = value; break; case elfcpp::R_ARM_TLS_TPOFF32: x = value + aligned_tcb_size; break; default: gold_unreachable(); } elfcpp::Swap<32, big_endian>::writeval(wv, x); } of->write_output_view(offset, oview_size, oview); } // A class to handle the PLT data. template<bool big_endian> class Output_data_plt_arm : public Output_section_data { public: typedef Output_data_reloc<elfcpp::SHT_REL, true, 32, big_endian> Reloc_section; Output_data_plt_arm(Layout*, Output_data_space*); // Add an entry to the PLT. void add_entry(Symbol* gsym); // Return the .rel.plt section data. const Reloc_section* rel_plt() const { return this->rel_; } // Return the number of PLT entries. unsigned int entry_count() const { return this->count_; } // Return the offset of the first non-reserved PLT entry. static unsigned int first_plt_entry_offset() { return sizeof(first_plt_entry); } // Return the size of a PLT entry. static unsigned int get_plt_entry_size() { return sizeof(plt_entry); } protected: 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")); } private: // Template for the first PLT entry. static const uint32_t first_plt_entry[5]; // Template for subsequent PLT entries. static const uint32_t plt_entry[3]; // Set the final size. void set_final_data_size() { this->set_data_size(sizeof(first_plt_entry) + this->count_ * sizeof(plt_entry)); } // Write out the PLT data. void do_write(Output_file*); // The reloc section. Reloc_section* rel_; // The .got.plt section. Output_data_space* got_plt_; // The number of PLT entries. unsigned int count_; }; // Create the PLT section. The ordinary .got section is an argument, // since we need to refer to the start. We also create our own .got // section just for PLT entries. template<bool big_endian> Output_data_plt_arm<big_endian>::Output_data_plt_arm(Layout* layout, Output_data_space* got_plt) : Output_section_data(4), got_plt_(got_plt), count_(0) { this->rel_ = new Reloc_section(false); layout->add_output_section_data(".rel.plt", elfcpp::SHT_REL, elfcpp::SHF_ALLOC, this->rel_, ORDER_DYNAMIC_PLT_RELOCS, false); } template<bool big_endian> void Output_data_plt_arm<big_endian>::do_adjust_output_section(Output_section* os) { os->set_entsize(0); } // Add an entry to the PLT. template<bool big_endian> void Output_data_plt_arm<big_endian>::add_entry(Symbol* gsym) { gold_assert(!gsym->has_plt_offset()); // Note that when setting the PLT offset we skip the initial // reserved PLT entry. gsym->set_plt_offset((this->count_) * sizeof(plt_entry) + sizeof(first_plt_entry)); ++this->count_; section_offset_type got_offset = this->got_plt_->current_data_size(); // Every PLT entry needs a GOT entry which points back to the PLT // entry (this will be changed by the dynamic linker, normally // lazily when the function is called). this->got_plt_->set_current_data_size(got_offset + 4); // Every PLT entry needs a reloc. gsym->set_needs_dynsym_entry(); this->rel_->add_global(gsym, elfcpp::R_ARM_JUMP_SLOT, this->got_plt_, got_offset); // Note that we don't need to save the symbol. The contents of the // PLT are independent of which symbols are used. The symbols only // appear in the relocations. } // ARM PLTs. // FIXME: This is not very flexible. Right now this has only been tested // on armv5te. If we are to support additional architecture features like // Thumb-2 or BE8, we need to make this more flexible like GNU ld. // The first entry in the PLT. template<bool big_endian> const uint32_t Output_data_plt_arm<big_endian>::first_plt_entry[5] = { 0xe52de004, // str lr, [sp, #-4]! 0xe59fe004, // ldr lr, [pc, #4] 0xe08fe00e, // add lr, pc, lr 0xe5bef008, // ldr pc, [lr, #8]! 0x00000000, // &GOT[0] - . }; // Subsequent entries in the PLT. template<bool big_endian> const uint32_t Output_data_plt_arm<big_endian>::plt_entry[3] = { 0xe28fc600, // add ip, pc, #0xNN00000 0xe28cca00, // add ip, ip, #0xNN000 0xe5bcf000, // ldr pc, [ip, #0xNNN]! }; // Write out the PLT. This uses the hand-coded instructions above, // and adjusts them as needed. This is all specified by the arm ELF // Processor Supplement. template<bool big_endian> void Output_data_plt_arm<big_endian>::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); const off_t got_file_offset = this->got_plt_->offset(); const section_size_type got_size = convert_to_section_size_type(this->got_plt_->data_size()); unsigned char* const got_view = of->get_output_view(got_file_offset, got_size); unsigned char* pov = oview; Arm_address plt_address = this->address(); Arm_address got_address = this->got_plt_->address(); // Write first PLT entry. All but the last word are constants. const size_t num_first_plt_words = (sizeof(first_plt_entry) / sizeof(plt_entry[0])); for (size_t i = 0; i < num_first_plt_words - 1; i++) elfcpp::Swap<32, big_endian>::writeval(pov + i * 4, first_plt_entry[i]); // Last word in first PLT entry is &GOT[0] - . elfcpp::Swap<32, big_endian>::writeval(pov + 16, got_address - (plt_address + 16)); pov += sizeof(first_plt_entry); unsigned char* got_pov = got_view; memset(got_pov, 0, 12); got_pov += 12; const int rel_size = elfcpp::Elf_sizes<32>::rel_size; unsigned int plt_offset = sizeof(first_plt_entry); unsigned int plt_rel_offset = 0; unsigned int got_offset = 12; const unsigned int count = this->count_; for (unsigned int i = 0; i < count; ++i, pov += sizeof(plt_entry), got_pov += 4, plt_offset += sizeof(plt_entry), plt_rel_offset += rel_size, got_offset += 4) { // Set and adjust the PLT entry itself. int32_t offset = ((got_address + got_offset) - (plt_address + plt_offset + 8)); gold_assert(offset >= 0 && offset < 0x0fffffff); uint32_t plt_insn0 = plt_entry[0] | ((offset >> 20) & 0xff); elfcpp::Swap<32, big_endian>::writeval(pov, plt_insn0); uint32_t plt_insn1 = plt_entry[1] | ((offset >> 12) & 0xff); elfcpp::Swap<32, big_endian>::writeval(pov + 4, plt_insn1); uint32_t plt_insn2 = plt_entry[2] | (offset & 0xfff); elfcpp::Swap<32, big_endian>::writeval(pov + 8, plt_insn2); // Set the entry in the GOT. elfcpp::Swap<32, big_endian>::writeval(got_pov, plt_address); } gold_assert(static_cast<section_size_type>(pov - oview) == oview_size); gold_assert(static_cast<section_size_type>(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); } // Create a PLT entry for a global symbol. template<bool big_endian> void Target_arm<big_endian>::make_plt_entry(Symbol_table* symtab, Layout* layout, Symbol* gsym) { if (gsym->has_plt_offset()) return; if (this->plt_ == NULL) { // Create the GOT sections first. this->got_section(symtab, layout); this->plt_ = new Output_data_plt_arm<big_endian>(layout, this->got_plt_); layout->add_output_section_data(".plt", elfcpp::SHT_PROGBITS, (elfcpp::SHF_ALLOC | elfcpp::SHF_EXECINSTR), this->plt_, ORDER_PLT, false); } this->plt_->add_entry(gsym); } // Return the number of entries in the PLT. template<bool big_endian> unsigned int Target_arm<big_endian>::plt_entry_count() const { if (this->plt_ == NULL) return 0; return this->plt_->entry_count(); } // Return the offset of the first non-reserved PLT entry. template<bool big_endian> unsigned int Target_arm<big_endian>::first_plt_entry_offset() const { return Output_data_plt_arm<big_endian>::first_plt_entry_offset(); } // Return the size of each PLT entry. template<bool big_endian> unsigned int Target_arm<big_endian>::plt_entry_size() const { return Output_data_plt_arm<big_endian>::get_plt_entry_size(); } // Get the section to use for TLS_DESC relocations. template<bool big_endian> typename Target_arm<big_endian>::Reloc_section* Target_arm<big_endian>::rel_tls_desc_section(Layout* layout) const { return this->plt_section()->rel_tls_desc(layout); } // Define the _TLS_MODULE_BASE_ symbol in the TLS segment. template<bool big_endian> void Target_arm<big_endian>::define_tls_base_symbol( Symbol_table* symtab, Layout* layout) { if (this->tls_base_symbol_defined_) return; Output_segment* tls_segment = layout->tls_segment(); if (tls_segment != NULL) { bool is_exec = parameters->options().output_is_executable(); symtab->define_in_output_segment("_TLS_MODULE_BASE_", NULL, Symbol_table::PREDEFINED, tls_segment, 0, 0, elfcpp::STT_TLS, elfcpp::STB_LOCAL, elfcpp::STV_HIDDEN, 0, (is_exec ? Symbol::SEGMENT_END : Symbol::SEGMENT_START), true); } this->tls_base_symbol_defined_ = true; } // Create a GOT entry for the TLS module index. template<bool big_endian> unsigned int Target_arm<big_endian>::got_mod_index_entry( Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* object) { if (this->got_mod_index_offset_ == -1U) { gold_assert(symtab != NULL && layout != NULL && object != NULL); Arm_output_data_got<big_endian>* got = this->got_section(symtab, layout); unsigned int got_offset; if (!parameters->doing_static_link()) { got_offset = got->add_constant(0); Reloc_section* rel_dyn = this->rel_dyn_section(layout); rel_dyn->add_local(object, 0, elfcpp::R_ARM_TLS_DTPMOD32, got, got_offset); } else { // We are doing a static link. Just mark it as belong to module 1, // the executable. got_offset = got->add_constant(1); } got->add_constant(0); this->got_mod_index_offset_ = got_offset; } return this->got_mod_index_offset_; } // Optimize the TLS relocation type based on what we know about the // symbol. IS_FINAL is true if the final address of this symbol is // known at link time. template<bool big_endian> tls::Tls_optimization Target_arm<big_endian>::optimize_tls_reloc(bool, int) { // FIXME: Currently we do not do any TLS optimization. return tls::TLSOPT_NONE; } // Get the Reference_flags for a particular relocation. template<bool big_endian> int Target_arm<big_endian>::Scan::get_reference_flags(unsigned int r_type) { switch (r_type) { case elfcpp::R_ARM_NONE: case elfcpp::R_ARM_V4BX: case elfcpp::R_ARM_GNU_VTENTRY: case elfcpp::R_ARM_GNU_VTINHERIT: // No symbol reference. return 0; case elfcpp::R_ARM_ABS32: case elfcpp::R_ARM_ABS16: case elfcpp::R_ARM_ABS12: case elfcpp::R_ARM_THM_ABS5: case elfcpp::R_ARM_ABS8: case elfcpp::R_ARM_BASE_ABS: case elfcpp::R_ARM_MOVW_ABS_NC: case elfcpp::R_ARM_MOVT_ABS: case elfcpp::R_ARM_THM_MOVW_ABS_NC: case elfcpp::R_ARM_THM_MOVT_ABS: case elfcpp::R_ARM_ABS32_NOI: return Symbol::ABSOLUTE_REF; case elfcpp::R_ARM_REL32: case elfcpp::R_ARM_LDR_PC_G0: case elfcpp::R_ARM_SBREL32: case elfcpp::R_ARM_THM_PC8: case elfcpp::R_ARM_BASE_PREL: case elfcpp::R_ARM_MOVW_PREL_NC: case elfcpp::R_ARM_MOVT_PREL: case elfcpp::R_ARM_THM_MOVW_PREL_NC: case elfcpp::R_ARM_THM_MOVT_PREL: case elfcpp::R_ARM_THM_ALU_PREL_11_0: case elfcpp::R_ARM_THM_PC12: case elfcpp::R_ARM_REL32_NOI: case elfcpp::R_ARM_ALU_PC_G0_NC: case elfcpp::R_ARM_ALU_PC_G0: case elfcpp::R_ARM_ALU_PC_G1_NC: case elfcpp::R_ARM_ALU_PC_G1: case elfcpp::R_ARM_ALU_PC_G2: case elfcpp::R_ARM_LDR_PC_G1: case elfcpp::R_ARM_LDR_PC_G2: case elfcpp::R_ARM_LDRS_PC_G0: case elfcpp::R_ARM_LDRS_PC_G1: case elfcpp::R_ARM_LDRS_PC_G2: case elfcpp::R_ARM_LDC_PC_G0: case elfcpp::R_ARM_LDC_PC_G1: case elfcpp::R_ARM_LDC_PC_G2: case elfcpp::R_ARM_ALU_SB_G0_NC: case elfcpp::R_ARM_ALU_SB_G0: case elfcpp::R_ARM_ALU_SB_G1_NC: case elfcpp::R_ARM_ALU_SB_G1: case elfcpp::R_ARM_ALU_SB_G2: case elfcpp::R_ARM_LDR_SB_G0: case elfcpp::R_ARM_LDR_SB_G1: case elfcpp::R_ARM_LDR_SB_G2: case elfcpp::R_ARM_LDRS_SB_G0: case elfcpp::R_ARM_LDRS_SB_G1: case elfcpp::R_ARM_LDRS_SB_G2: case elfcpp::R_ARM_LDC_SB_G0: case elfcpp::R_ARM_LDC_SB_G1: case elfcpp::R_ARM_LDC_SB_G2: case elfcpp::R_ARM_MOVW_BREL_NC: case elfcpp::R_ARM_MOVT_BREL: case elfcpp::R_ARM_MOVW_BREL: case elfcpp::R_ARM_THM_MOVW_BREL_NC: case elfcpp::R_ARM_THM_MOVT_BREL: case elfcpp::R_ARM_THM_MOVW_BREL: case elfcpp::R_ARM_GOTOFF32: case elfcpp::R_ARM_GOTOFF12: case elfcpp::R_ARM_SBREL31: return Symbol::RELATIVE_REF; case elfcpp::R_ARM_PLT32: case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_THM_JUMP24: case elfcpp::R_ARM_THM_JUMP19: case elfcpp::R_ARM_THM_JUMP6: case elfcpp::R_ARM_THM_JUMP11: case elfcpp::R_ARM_THM_JUMP8: // R_ARM_PREL31 is not used to relocate call/jump instructions but // in unwind tables. It may point to functions via PLTs. // So we treat it like call/jump relocations above. case elfcpp::R_ARM_PREL31: return Symbol::FUNCTION_CALL | Symbol::RELATIVE_REF; case elfcpp::R_ARM_GOT_BREL: case elfcpp::R_ARM_GOT_ABS: case elfcpp::R_ARM_GOT_PREL: // Absolute in GOT. return Symbol::ABSOLUTE_REF; case elfcpp::R_ARM_TLS_GD32: // Global-dynamic case elfcpp::R_ARM_TLS_LDM32: // Local-dynamic case elfcpp::R_ARM_TLS_LDO32: // Alternate local-dynamic case elfcpp::R_ARM_TLS_IE32: // Initial-exec case elfcpp::R_ARM_TLS_LE32: // Local-exec return Symbol::TLS_REF; case elfcpp::R_ARM_TARGET1: case elfcpp::R_ARM_TARGET2: case elfcpp::R_ARM_COPY: case elfcpp::R_ARM_GLOB_DAT: case elfcpp::R_ARM_JUMP_SLOT: case elfcpp::R_ARM_RELATIVE: case elfcpp::R_ARM_PC24: case elfcpp::R_ARM_LDR_SBREL_11_0_NC: case elfcpp::R_ARM_ALU_SBREL_19_12_NC: case elfcpp::R_ARM_ALU_SBREL_27_20_CK: default: // Not expected. We will give an error later. return 0; } } // Report an unsupported relocation against a local symbol. template<bool big_endian> void Target_arm<big_endian>::Scan::unsupported_reloc_local( Sized_relobj_file<32, big_endian>* object, unsigned int r_type) { gold_error(_("%s: unsupported reloc %u against local symbol"), object->name().c_str(), r_type); } // We are about to emit a dynamic relocation of type R_TYPE. If the // dynamic linker does not support it, issue an error. The GNU linker // only issues a non-PIC error for an allocated read-only section. // Here we know the section is allocated, but we don't know that it is // read-only. But we check for all the relocation types which the // glibc dynamic linker supports, so it seems appropriate to issue an // error even if the section is not read-only. template<bool big_endian> void Target_arm<big_endian>::Scan::check_non_pic(Relobj* object, unsigned int r_type) { switch (r_type) { // These are the relocation types supported by glibc for ARM. case elfcpp::R_ARM_RELATIVE: case elfcpp::R_ARM_COPY: case elfcpp::R_ARM_GLOB_DAT: case elfcpp::R_ARM_JUMP_SLOT: case elfcpp::R_ARM_ABS32: case elfcpp::R_ARM_ABS32_NOI: case elfcpp::R_ARM_PC24: // FIXME: The following 3 types are not supported by Android's dynamic // linker. case elfcpp::R_ARM_TLS_DTPMOD32: case elfcpp::R_ARM_TLS_DTPOFF32: case elfcpp::R_ARM_TLS_TPOFF32: return; default: { // This prevents us from issuing more than one error per reloc // section. But we can still wind up issuing more than one // error per object file. if (this->issued_non_pic_error_) return; const Arm_reloc_property* reloc_property = arm_reloc_property_table->get_reloc_property(r_type); gold_assert(reloc_property != NULL); object->error(_("requires unsupported dynamic reloc %s; " "recompile with -fPIC"), reloc_property->name().c_str()); this->issued_non_pic_error_ = true; return; } case elfcpp::R_ARM_NONE: gold_unreachable(); } } // Scan a relocation for a local symbol. // FIXME: This only handles a subset of relocation types used by Android // on ARM v5te devices. template<bool big_endian> inline void Target_arm<big_endian>::Scan::local(Symbol_table* symtab, Layout* layout, Target_arm* target, Sized_relobj_file<32, big_endian>* object, unsigned int data_shndx, Output_section* output_section, const elfcpp::Rel<32, big_endian>& reloc, unsigned int r_type, const elfcpp::Sym<32, big_endian>& lsym) { r_type = get_real_reloc_type(r_type); switch (r_type) { case elfcpp::R_ARM_NONE: case elfcpp::R_ARM_V4BX: case elfcpp::R_ARM_GNU_VTENTRY: case elfcpp::R_ARM_GNU_VTINHERIT: break; case elfcpp::R_ARM_ABS32: case elfcpp::R_ARM_ABS32_NOI: // If building a shared library (or a position-independent // executable), we need to create a dynamic relocation for // this location. The relocation applied at link time will // apply the link-time value, so we flag the location with // an R_ARM_RELATIVE relocation so the dynamic loader can // relocate it easily. if (parameters->options().output_is_position_independent()) { Reloc_section* rel_dyn = target->rel_dyn_section(layout); unsigned int r_sym = elfcpp::elf_r_sym<32>(reloc.get_r_info()); // If we are to add more other reloc types than R_ARM_ABS32, // we need to add check_non_pic(object, r_type) here. rel_dyn->add_local_relative(object, r_sym, elfcpp::R_ARM_RELATIVE, output_section, data_shndx, reloc.get_r_offset()); } break; case elfcpp::R_ARM_ABS16: case elfcpp::R_ARM_ABS12: case elfcpp::R_ARM_THM_ABS5: case elfcpp::R_ARM_ABS8: case elfcpp::R_ARM_BASE_ABS: case elfcpp::R_ARM_MOVW_ABS_NC: case elfcpp::R_ARM_MOVT_ABS: case elfcpp::R_ARM_THM_MOVW_ABS_NC: case elfcpp::R_ARM_THM_MOVT_ABS: // If building a shared library (or a position-independent // executable), we need to create a dynamic relocation for // this location. Because the addend needs to remain in the // data section, we need to be careful not to apply this // relocation statically. if (parameters->options().output_is_position_independent()) { check_non_pic(object, r_type); Reloc_section* rel_dyn = target->rel_dyn_section(layout); unsigned int r_sym = elfcpp::elf_r_sym<32>(reloc.get_r_info()); if (lsym.get_st_type() != elfcpp::STT_SECTION) rel_dyn->add_local(object, r_sym, r_type, output_section, data_shndx, reloc.get_r_offset()); 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, &is_ordinary); if (!is_ordinary) object->error(_("section symbol %u has bad shndx %u"), r_sym, shndx); else rel_dyn->add_local_section(object, shndx, r_type, output_section, data_shndx, reloc.get_r_offset()); } } break; case elfcpp::R_ARM_REL32: case elfcpp::R_ARM_LDR_PC_G0: case elfcpp::R_ARM_SBREL32: case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_THM_PC8: case elfcpp::R_ARM_BASE_PREL: case elfcpp::R_ARM_PLT32: case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_THM_JUMP24: case elfcpp::R_ARM_SBREL31: case elfcpp::R_ARM_PREL31: case elfcpp::R_ARM_MOVW_PREL_NC: case elfcpp::R_ARM_MOVT_PREL: case elfcpp::R_ARM_THM_MOVW_PREL_NC: case elfcpp::R_ARM_THM_MOVT_PREL: case elfcpp::R_ARM_THM_JUMP19: case elfcpp::R_ARM_THM_JUMP6: case elfcpp::R_ARM_THM_ALU_PREL_11_0: case elfcpp::R_ARM_THM_PC12: case elfcpp::R_ARM_REL32_NOI: case elfcpp::R_ARM_ALU_PC_G0_NC: case elfcpp::R_ARM_ALU_PC_G0: case elfcpp::R_ARM_ALU_PC_G1_NC: case elfcpp::R_ARM_ALU_PC_G1: case elfcpp::R_ARM_ALU_PC_G2: case elfcpp::R_ARM_LDR_PC_G1: case elfcpp::R_ARM_LDR_PC_G2: case elfcpp::R_ARM_LDRS_PC_G0: case elfcpp::R_ARM_LDRS_PC_G1: case elfcpp::R_ARM_LDRS_PC_G2: case elfcpp::R_ARM_LDC_PC_G0: case elfcpp::R_ARM_LDC_PC_G1: case elfcpp::R_ARM_LDC_PC_G2: case elfcpp::R_ARM_ALU_SB_G0_NC: case elfcpp::R_ARM_ALU_SB_G0: case elfcpp::R_ARM_ALU_SB_G1_NC: case elfcpp::R_ARM_ALU_SB_G1: case elfcpp::R_ARM_ALU_SB_G2: case elfcpp::R_ARM_LDR_SB_G0: case elfcpp::R_ARM_LDR_SB_G1: case elfcpp::R_ARM_LDR_SB_G2: case elfcpp::R_ARM_LDRS_SB_G0: case elfcpp::R_ARM_LDRS_SB_G1: case elfcpp::R_ARM_LDRS_SB_G2: case elfcpp::R_ARM_LDC_SB_G0: case elfcpp::R_ARM_LDC_SB_G1: case elfcpp::R_ARM_LDC_SB_G2: case elfcpp::R_ARM_MOVW_BREL_NC: case elfcpp::R_ARM_MOVT_BREL: case elfcpp::R_ARM_MOVW_BREL: case elfcpp::R_ARM_THM_MOVW_BREL_NC: case elfcpp::R_ARM_THM_MOVT_BREL: case elfcpp::R_ARM_THM_MOVW_BREL: case elfcpp::R_ARM_THM_JUMP11: case elfcpp::R_ARM_THM_JUMP8: // We don't need to do anything for a relative addressing relocation // against a local symbol if it does not reference the GOT. break; case elfcpp::R_ARM_GOTOFF32: case elfcpp::R_ARM_GOTOFF12: // We need a GOT section: target->got_section(symtab, layout); break; case elfcpp::R_ARM_GOT_BREL: case elfcpp::R_ARM_GOT_PREL: { // The symbol requires a GOT entry. Arm_output_data_got<big_endian>* got = target->got_section(symtab, layout); unsigned int r_sym = elfcpp::elf_r_sym<32>(reloc.get_r_info()); if (got->add_local(object, r_sym, GOT_TYPE_STANDARD)) { // If we are generating a shared object, we need to add a // dynamic RELATIVE relocation for this symbol's GOT entry. if (parameters->options().output_is_position_independent()) { Reloc_section* rel_dyn = target->rel_dyn_section(layout); unsigned int r_sym = elfcpp::elf_r_sym<32>(reloc.get_r_info()); rel_dyn->add_local_relative( object, r_sym, elfcpp::R_ARM_RELATIVE, got, object->local_got_offset(r_sym, GOT_TYPE_STANDARD)); } } } break; case elfcpp::R_ARM_TARGET1: case elfcpp::R_ARM_TARGET2: // This should have been mapped to another type already. // Fall through. case elfcpp::R_ARM_COPY: case elfcpp::R_ARM_GLOB_DAT: case elfcpp::R_ARM_JUMP_SLOT: case elfcpp::R_ARM_RELATIVE: // These are relocations which should only be seen by the // dynamic linker, and should never be seen here. gold_error(_("%s: unexpected reloc %u in object file"), object->name().c_str(), r_type); break; // These are initial TLS relocs, which are expected when // linking. case elfcpp::R_ARM_TLS_GD32: // Global-dynamic case elfcpp::R_ARM_TLS_LDM32: // Local-dynamic case elfcpp::R_ARM_TLS_LDO32: // Alternate local-dynamic case elfcpp::R_ARM_TLS_IE32: // Initial-exec case elfcpp::R_ARM_TLS_LE32: // Local-exec { bool output_is_shared = parameters->options().shared(); const tls::Tls_optimization optimized_type = Target_arm<big_endian>::optimize_tls_reloc(!output_is_shared, r_type); switch (r_type) { case elfcpp::R_ARM_TLS_GD32: // Global-dynamic if (optimized_type == tls::TLSOPT_NONE) { // Create a pair of GOT entries for the module index and // dtv-relative offset. Arm_output_data_got<big_endian>* got = target->got_section(symtab, layout); unsigned int r_sym = elfcpp::elf_r_sym<32>(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); break; } if (!parameters->doing_static_link()) got->add_local_pair_with_rel(object, r_sym, shndx, GOT_TYPE_TLS_PAIR, target->rel_dyn_section(layout), elfcpp::R_ARM_TLS_DTPMOD32, 0); else got->add_tls_gd32_with_static_reloc(GOT_TYPE_TLS_PAIR, object, r_sym); } else // FIXME: TLS optimization not supported yet. gold_unreachable(); break; case elfcpp::R_ARM_TLS_LDM32: // Local-dynamic if (optimized_type == tls::TLSOPT_NONE) { // Create a GOT entry for the module index. target->got_mod_index_entry(symtab, layout, object); } else // FIXME: TLS optimization not supported yet. gold_unreachable(); break; case elfcpp::R_ARM_TLS_LDO32: // Alternate local-dynamic break; case elfcpp::R_ARM_TLS_IE32: // Initial-exec layout->set_has_static_tls(); if (optimized_type == tls::TLSOPT_NONE) { // Create a GOT entry for the tp-relative offset. Arm_output_data_got<big_endian>* got = target->got_section(symtab, layout); unsigned int r_sym = elfcpp::elf_r_sym<32>(reloc.get_r_info()); if (!parameters->doing_static_link()) got->add_local_with_rel(object, r_sym, GOT_TYPE_TLS_OFFSET, target->rel_dyn_section(layout), elfcpp::R_ARM_TLS_TPOFF32); else if (!object->local_has_got_offset(r_sym, GOT_TYPE_TLS_OFFSET)) { got->add_local(object, r_sym, GOT_TYPE_TLS_OFFSET); unsigned int got_offset = object->local_got_offset(r_sym, GOT_TYPE_TLS_OFFSET); got->add_static_reloc(got_offset, elfcpp::R_ARM_TLS_TPOFF32, object, r_sym); } } else // FIXME: TLS optimization not supported yet. gold_unreachable(); break; case elfcpp::R_ARM_TLS_LE32: // Local-exec layout->set_has_static_tls(); if (output_is_shared) { // We need to create a dynamic relocation. gold_assert(lsym.get_st_type() != elfcpp::STT_SECTION); unsigned int r_sym = elfcpp::elf_r_sym<32>(reloc.get_r_info()); Reloc_section* rel_dyn = target->rel_dyn_section(layout); rel_dyn->add_local(object, r_sym, elfcpp::R_ARM_TLS_TPOFF32, output_section, data_shndx, reloc.get_r_offset()); } break; default: gold_unreachable(); } } break; case elfcpp::R_ARM_PC24: case elfcpp::R_ARM_LDR_SBREL_11_0_NC: case elfcpp::R_ARM_ALU_SBREL_19_12_NC: case elfcpp::R_ARM_ALU_SBREL_27_20_CK: default: unsupported_reloc_local(object, r_type); break; } } // Report an unsupported relocation against a global symbol. template<bool big_endian> void Target_arm<big_endian>::Scan::unsupported_reloc_global( Sized_relobj_file<32, big_endian>* object, unsigned int r_type, Symbol* gsym) { gold_error(_("%s: unsupported reloc %u against global symbol %s"), object->name().c_str(), r_type, gsym->demangled_name().c_str()); } template<bool big_endian> inline bool Target_arm<big_endian>::Scan::possible_function_pointer_reloc( unsigned int r_type) { switch (r_type) { case elfcpp::R_ARM_PC24: case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_PLT32: case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_THM_JUMP24: case elfcpp::R_ARM_SBREL31: case elfcpp::R_ARM_PREL31: case elfcpp::R_ARM_THM_JUMP19: case elfcpp::R_ARM_THM_JUMP6: case elfcpp::R_ARM_THM_JUMP11: case elfcpp::R_ARM_THM_JUMP8: // All the relocations above are branches except SBREL31 and PREL31. return false; default: // Be conservative and assume this is a function pointer. return true; } } template<bool big_endian> inline bool Target_arm<big_endian>::Scan::local_reloc_may_be_function_pointer( Symbol_table*, Layout*, Target_arm<big_endian>* target, Sized_relobj_file<32, big_endian>*, unsigned int, Output_section*, const elfcpp::Rel<32, big_endian>&, unsigned int r_type, const elfcpp::Sym<32, big_endian>&) { r_type = target->get_real_reloc_type(r_type); return possible_function_pointer_reloc(r_type); } template<bool big_endian> inline bool Target_arm<big_endian>::Scan::global_reloc_may_be_function_pointer( Symbol_table*, Layout*, Target_arm<big_endian>* target, Sized_relobj_file<32, big_endian>*, unsigned int, Output_section*, const elfcpp::Rel<32, big_endian>&, unsigned int r_type, Symbol* gsym) { // GOT is not a function. if (strcmp(gsym->name(), "_GLOBAL_OFFSET_TABLE_") == 0) return false; r_type = target->get_real_reloc_type(r_type); return possible_function_pointer_reloc(r_type); } // Scan a relocation for a global symbol. template<bool big_endian> inline void Target_arm<big_endian>::Scan::global(Symbol_table* symtab, Layout* layout, Target_arm* target, Sized_relobj_file<32, big_endian>* object, unsigned int data_shndx, Output_section* output_section, const elfcpp::Rel<32, big_endian>& reloc, unsigned int r_type, Symbol* gsym) { // A reference to _GLOBAL_OFFSET_TABLE_ implies that we need a got // section. We check here to avoid creating a dynamic reloc against // _GLOBAL_OFFSET_TABLE_. if (!target->has_got_section() && strcmp(gsym->name(), "_GLOBAL_OFFSET_TABLE_") == 0) target->got_section(symtab, layout); r_type = get_real_reloc_type(r_type); switch (r_type) { case elfcpp::R_ARM_NONE: case elfcpp::R_ARM_V4BX: case elfcpp::R_ARM_GNU_VTENTRY: case elfcpp::R_ARM_GNU_VTINHERIT: break; case elfcpp::R_ARM_ABS32: case elfcpp::R_ARM_ABS16: case elfcpp::R_ARM_ABS12: case elfcpp::R_ARM_THM_ABS5: case elfcpp::R_ARM_ABS8: case elfcpp::R_ARM_BASE_ABS: case elfcpp::R_ARM_MOVW_ABS_NC: case elfcpp::R_ARM_MOVT_ABS: case elfcpp::R_ARM_THM_MOVW_ABS_NC: case elfcpp::R_ARM_THM_MOVT_ABS: case elfcpp::R_ARM_ABS32_NOI: // Absolute addressing relocations. { // Make a PLT entry if necessary. if (this->symbol_needs_plt_entry(gsym)) { 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); } else if ((r_type == elfcpp::R_ARM_ABS32 || r_type == elfcpp::R_ARM_ABS32_NOI) && gsym->can_use_relative_reloc(false)) { Reloc_section* rel_dyn = target->rel_dyn_section(layout); rel_dyn->add_global_relative(gsym, elfcpp::R_ARM_RELATIVE, output_section, object, data_shndx, reloc.get_r_offset()); } else { check_non_pic(object, r_type); Reloc_section* rel_dyn = target->rel_dyn_section(layout); rel_dyn->add_global(gsym, r_type, output_section, object, data_shndx, reloc.get_r_offset()); } } } break; case elfcpp::R_ARM_GOTOFF32: case elfcpp::R_ARM_GOTOFF12: // We need a GOT section. target->got_section(symtab, layout); break; case elfcpp::R_ARM_REL32: case elfcpp::R_ARM_LDR_PC_G0: case elfcpp::R_ARM_SBREL32: case elfcpp::R_ARM_THM_PC8: case elfcpp::R_ARM_BASE_PREL: case elfcpp::R_ARM_MOVW_PREL_NC: case elfcpp::R_ARM_MOVT_PREL: case elfcpp::R_ARM_THM_MOVW_PREL_NC: case elfcpp::R_ARM_THM_MOVT_PREL: case elfcpp::R_ARM_THM_ALU_PREL_11_0: case elfcpp::R_ARM_THM_PC12: case elfcpp::R_ARM_REL32_NOI: case elfcpp::R_ARM_ALU_PC_G0_NC: case elfcpp::R_ARM_ALU_PC_G0: case elfcpp::R_ARM_ALU_PC_G1_NC: case elfcpp::R_ARM_ALU_PC_G1: case elfcpp::R_ARM_ALU_PC_G2: case elfcpp::R_ARM_LDR_PC_G1: case elfcpp::R_ARM_LDR_PC_G2: case elfcpp::R_ARM_LDRS_PC_G0: case elfcpp::R_ARM_LDRS_PC_G1: case elfcpp::R_ARM_LDRS_PC_G2: case elfcpp::R_ARM_LDC_PC_G0: case elfcpp::R_ARM_LDC_PC_G1: case elfcpp::R_ARM_LDC_PC_G2: case elfcpp::R_ARM_ALU_SB_G0_NC: case elfcpp::R_ARM_ALU_SB_G0: case elfcpp::R_ARM_ALU_SB_G1_NC: case elfcpp::R_ARM_ALU_SB_G1: case elfcpp::R_ARM_ALU_SB_G2: case elfcpp::R_ARM_LDR_SB_G0: case elfcpp::R_ARM_LDR_SB_G1: case elfcpp::R_ARM_LDR_SB_G2: case elfcpp::R_ARM_LDRS_SB_G0: case elfcpp::R_ARM_LDRS_SB_G1: case elfcpp::R_ARM_LDRS_SB_G2: case elfcpp::R_ARM_LDC_SB_G0: case elfcpp::R_ARM_LDC_SB_G1: case elfcpp::R_ARM_LDC_SB_G2: case elfcpp::R_ARM_MOVW_BREL_NC: case elfcpp::R_ARM_MOVT_BREL: case elfcpp::R_ARM_MOVW_BREL: case elfcpp::R_ARM_THM_MOVW_BREL_NC: case elfcpp::R_ARM_THM_MOVT_BREL: case elfcpp::R_ARM_THM_MOVW_BREL: // Relative addressing relocations. { // Make a dynamic relocation if necessary. if (gsym->needs_dynamic_reloc(Scan::get_reference_flags(r_type))) { if (target->may_need_copy_reloc(gsym)) { target->copy_reloc(symtab, layout, object, data_shndx, output_section, gsym, reloc); } else { check_non_pic(object, r_type); Reloc_section* rel_dyn = target->rel_dyn_section(layout); rel_dyn->add_global(gsym, r_type, output_section, object, data_shndx, reloc.get_r_offset()); } } } break; case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_PLT32: case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_THM_JUMP24: case elfcpp::R_ARM_SBREL31: case elfcpp::R_ARM_PREL31: case elfcpp::R_ARM_THM_JUMP19: case elfcpp::R_ARM_THM_JUMP6: case elfcpp::R_ARM_THM_JUMP11: case elfcpp::R_ARM_THM_JUMP8: // All the relocation above are branches except for the PREL31 ones. // A PREL31 relocation can point to a personality function in a shared // library. In that case we want to use a PLT because we want to // call the personality routine and the dynamic linkers we care about // do not support dynamic PREL31 relocations. An REL31 relocation may // point to a function whose unwinding behaviour is being described but // we will not mistakenly generate a PLT for that because we should use // a local section symbol. // If the symbol is fully resolved, this is just a relative // local reloc. Otherwise we need a PLT entry. if (gsym->final_value_is_known()) break; // If building a shared library, we can also skip the PLT entry // 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()) break; target->make_plt_entry(symtab, layout, gsym); break; case elfcpp::R_ARM_GOT_BREL: case elfcpp::R_ARM_GOT_ABS: case elfcpp::R_ARM_GOT_PREL: { // The symbol requires a GOT entry. Arm_output_data_got<big_endian>* got = target->got_section(symtab, layout); if (gsym->final_value_is_known()) got->add_global(gsym, GOT_TYPE_STANDARD); else { // If this symbol is not fully resolved, we need to add a // GOT entry with a dynamic relocation. Reloc_section* rel_dyn = target->rel_dyn_section(layout); if (gsym->is_from_dynobj() || gsym->is_undefined() || gsym->is_preemptible()) got->add_global_with_rel(gsym, GOT_TYPE_STANDARD, rel_dyn, elfcpp::R_ARM_GLOB_DAT); else { if (got->add_global(gsym, GOT_TYPE_STANDARD)) rel_dyn->add_global_relative( gsym, elfcpp::R_ARM_RELATIVE, got, gsym->got_offset(GOT_TYPE_STANDARD)); } } } break; case elfcpp::R_ARM_TARGET1: case elfcpp::R_ARM_TARGET2: // These should have been mapped to other types already. // Fall through. case elfcpp::R_ARM_COPY: case elfcpp::R_ARM_GLOB_DAT: case elfcpp::R_ARM_JUMP_SLOT: case elfcpp::R_ARM_RELATIVE: // These are relocations which should only be seen by the // dynamic linker, and should never be seen here. gold_error(_("%s: unexpected reloc %u in object file"), object->name().c_str(), r_type); break; // These are initial tls relocs, which are expected when // linking. case elfcpp::R_ARM_TLS_GD32: // Global-dynamic case elfcpp::R_ARM_TLS_LDM32: // Local-dynamic case elfcpp::R_ARM_TLS_LDO32: // Alternate local-dynamic case elfcpp::R_ARM_TLS_IE32: // Initial-exec case elfcpp::R_ARM_TLS_LE32: // Local-exec { const bool is_final = gsym->final_value_is_known(); const tls::Tls_optimization optimized_type = Target_arm<big_endian>::optimize_tls_reloc(is_final, r_type); switch (r_type) { case elfcpp::R_ARM_TLS_GD32: // Global-dynamic if (optimized_type == tls::TLSOPT_NONE) { // Create a pair of GOT entries for the module index and // dtv-relative offset. Arm_output_data_got<big_endian>* got = target->got_section(symtab, layout); if (!parameters->doing_static_link()) got->add_global_pair_with_rel(gsym, GOT_TYPE_TLS_PAIR, target->rel_dyn_section(layout), elfcpp::R_ARM_TLS_DTPMOD32, elfcpp::R_ARM_TLS_DTPOFF32); else got->add_tls_gd32_with_static_reloc(GOT_TYPE_TLS_PAIR, gsym); } else // FIXME: TLS optimization not supported yet. gold_unreachable(); break; case elfcpp::R_ARM_TLS_LDM32: // Local-dynamic if (optimized_type == tls::TLSOPT_NONE) { // Create a GOT entry for the module index. target->got_mod_index_entry(symtab, layout, object); } else // FIXME: TLS optimization not supported yet. gold_unreachable(); break; case elfcpp::R_ARM_TLS_LDO32: // Alternate local-dynamic break; case elfcpp::R_ARM_TLS_IE32: // Initial-exec layout->set_has_static_tls(); if (optimized_type == tls::TLSOPT_NONE) { // Create a GOT entry for the tp-relative offset. Arm_output_data_got<big_endian>* got = target->got_section(symtab, layout); if (!parameters->doing_static_link()) got->add_global_with_rel(gsym, GOT_TYPE_TLS_OFFSET, target->rel_dyn_section(layout), elfcpp::R_ARM_TLS_TPOFF32); else if (!gsym->has_got_offset(GOT_TYPE_TLS_OFFSET)) { got->add_global(gsym, GOT_TYPE_TLS_OFFSET); unsigned int got_offset = gsym->got_offset(GOT_TYPE_TLS_OFFSET); got->add_static_reloc(got_offset, elfcpp::R_ARM_TLS_TPOFF32, gsym); } } else // FIXME: TLS optimization not supported yet. gold_unreachable(); break; case elfcpp::R_ARM_TLS_LE32: // Local-exec layout->set_has_static_tls(); if (parameters->options().shared()) { // We need to create a dynamic relocation. Reloc_section* rel_dyn = target->rel_dyn_section(layout); rel_dyn->add_global(gsym, elfcpp::R_ARM_TLS_TPOFF32, output_section, object, data_shndx, reloc.get_r_offset()); } break; default: gold_unreachable(); } } break; case elfcpp::R_ARM_PC24: case elfcpp::R_ARM_LDR_SBREL_11_0_NC: case elfcpp::R_ARM_ALU_SBREL_19_12_NC: case elfcpp::R_ARM_ALU_SBREL_27_20_CK: default: unsupported_reloc_global(object, r_type, gsym); break; } } // Process relocations for gc. template<bool big_endian> void Target_arm<big_endian>::gc_process_relocs( Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* object, unsigned int data_shndx, unsigned int, 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) { typedef Target_arm<big_endian> Arm; typedef typename Target_arm<big_endian>::Scan Scan; gold::gc_process_relocs<32, big_endian, Arm, elfcpp::SHT_REL, Scan, typename Target_arm::Relocatable_size_for_reloc>( symtab, layout, this, object, data_shndx, prelocs, reloc_count, output_section, needs_special_offset_handling, local_symbol_count, plocal_symbols); } // Scan relocations for a section. template<bool big_endian> void Target_arm<big_endian>::scan_relocs(Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* 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) { typedef typename Target_arm<big_endian>::Scan Scan; if (sh_type == elfcpp::SHT_RELA) { gold_error(_("%s: unsupported RELA reloc section"), object->name().c_str()); return; } gold::scan_relocs<32, big_endian, Target_arm, elfcpp::SHT_REL, Scan>( symtab, layout, this, object, data_shndx, prelocs, reloc_count, output_section, needs_special_offset_handling, local_symbol_count, plocal_symbols); } // Finalize the sections. template<bool big_endian> void Target_arm<big_endian>::do_finalize_sections( Layout* layout, const Input_objects* input_objects, Symbol_table* symtab) { bool merged_any_attributes = false; // Merge processor-specific flags. for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(*p); if (arm_relobj->merge_flags_and_attributes()) { this->merge_processor_specific_flags( arm_relobj->name(), arm_relobj->processor_specific_flags()); this->merge_object_attributes(arm_relobj->name().c_str(), arm_relobj->attributes_section_data()); merged_any_attributes = true; } } for (Input_objects::Dynobj_iterator p = input_objects->dynobj_begin(); p != input_objects->dynobj_end(); ++p) { Arm_dynobj<big_endian>* arm_dynobj = Arm_dynobj<big_endian>::as_arm_dynobj(*p); this->merge_processor_specific_flags( arm_dynobj->name(), arm_dynobj->processor_specific_flags()); this->merge_object_attributes(arm_dynobj->name().c_str(), arm_dynobj->attributes_section_data()); merged_any_attributes = true; } // Create an empty uninitialized attribute section if we still don't have it // at this moment. This happens if there is no attributes sections in all // inputs. if (this->attributes_section_data_ == NULL) this->attributes_section_data_ = new Attributes_section_data(NULL, 0); // Check BLX use. const Object_attribute* cpu_arch_attr = this->get_aeabi_object_attribute(elfcpp::Tag_CPU_arch); if (cpu_arch_attr->int_value() > elfcpp::TAG_CPU_ARCH_V4) this->set_may_use_blx(true); // Check if we need to use Cortex-A8 workaround. if (parameters->options().user_set_fix_cortex_a8()) this->fix_cortex_a8_ = parameters->options().fix_cortex_a8(); else { // If neither --fix-cortex-a8 nor --no-fix-cortex-a8 is used, turn on // Cortex-A8 erratum workaround for ARMv7-A or ARMv7 with unknown // profile. const Object_attribute* cpu_arch_profile_attr = this->get_aeabi_object_attribute(elfcpp::Tag_CPU_arch_profile); this->fix_cortex_a8_ = (cpu_arch_attr->int_value() == elfcpp::TAG_CPU_ARCH_V7 && (cpu_arch_profile_attr->int_value() == 'A' || cpu_arch_profile_attr->int_value() == 0)); } // Check if we can use V4BX interworking. // The V4BX interworking stub contains BX instruction, // which is not specified for some profiles. if (this->fix_v4bx() == General_options::FIX_V4BX_INTERWORKING && !this->may_use_blx()) gold_error(_("unable to provide V4BX reloc interworking fix up; " "the target profile does not support BX instruction")); // Fill in some more dynamic tags. const Reloc_section* rel_plt = (this->plt_ == NULL ? NULL : this->plt_->rel_plt()); layout->add_target_dynamic_tags(true, this->got_plt_, rel_plt, this->rel_dyn_, true, false); // Emit any relocs we saved in an attempt to avoid generating COPY // relocs. if (this->copy_relocs_.any_saved_relocs()) this->copy_relocs_.emit(this->rel_dyn_section(layout)); // Handle the .ARM.exidx section. Output_section* exidx_section = layout->find_output_section(".ARM.exidx"); if (!parameters->options().relocatable()) { if (exidx_section != NULL && exidx_section->type() == elfcpp::SHT_ARM_EXIDX) { // Create __exidx_start and __exidx_end symbols. symtab->define_in_output_data("__exidx_start", NULL, Symbol_table::PREDEFINED, exidx_section, 0, 0, elfcpp::STT_OBJECT, elfcpp::STB_GLOBAL, elfcpp::STV_HIDDEN, 0, false, true); symtab->define_in_output_data("__exidx_end", NULL, Symbol_table::PREDEFINED, exidx_section, 0, 0, elfcpp::STT_OBJECT, elfcpp::STB_GLOBAL, elfcpp::STV_HIDDEN, 0, true, true); // For the ARM target, we need to add a PT_ARM_EXIDX segment for // the .ARM.exidx section. if (!layout->script_options()->saw_phdrs_clause()) { gold_assert(layout->find_output_segment(elfcpp::PT_ARM_EXIDX, 0, 0) == NULL); Output_segment* exidx_segment = layout->make_output_segment(elfcpp::PT_ARM_EXIDX, elfcpp::PF_R); exidx_segment->add_output_section_to_nonload(exidx_section, elfcpp::PF_R); } } else { symtab->define_as_constant("__exidx_start", NULL, Symbol_table::PREDEFINED, 0, 0, elfcpp::STT_OBJECT, elfcpp::STB_GLOBAL, elfcpp::STV_HIDDEN, 0, true, false); symtab->define_as_constant("__exidx_end", NULL, Symbol_table::PREDEFINED, 0, 0, elfcpp::STT_OBJECT, elfcpp::STB_GLOBAL, elfcpp::STV_HIDDEN, 0, true, false); } } // Create an .ARM.attributes section if we have merged any attributes // from inputs. if (merged_any_attributes) { Output_attributes_section_data* attributes_section = new Output_attributes_section_data(*this->attributes_section_data_); layout->add_output_section_data(".ARM.attributes", elfcpp::SHT_ARM_ATTRIBUTES, 0, attributes_section, ORDER_INVALID, false); } // Fix up links in section EXIDX headers. for (Layout::Section_list::const_iterator p = layout->section_list().begin(); p != layout->section_list().end(); ++p) if ((*p)->type() == elfcpp::SHT_ARM_EXIDX) { Arm_output_section<big_endian>* os = Arm_output_section<big_endian>::as_arm_output_section(*p); os->set_exidx_section_link(); } } // Return whether a direct absolute static relocation needs to be applied. // In cases where Scan::local() or Scan::global() has created // a dynamic relocation other than R_ARM_RELATIVE, the addend // of the relocation is carried in the data, and we must not // apply the static relocation. template<bool big_endian> inline bool Target_arm<big_endian>::Relocate::should_apply_static_reloc( const Sized_symbol<32>* gsym, unsigned int r_type, bool is_32bit, Output_section* output_section) { // If the output section is not allocated, then we didn't call // scan_relocs, we didn't create a dynamic reloc, and we must apply // the reloc here. if ((output_section->flags() & elfcpp::SHF_ALLOC) == 0) return true; int ref_flags = Scan::get_reference_flags(r_type); // For local symbols, we will have created a non-RELATIVE dynamic // relocation only if (a) the output is position independent, // (b) the relocation is absolute (not pc- or segment-relative), and // (c) the relocation is not 32 bits wide. if (gsym == NULL) return !(parameters->options().output_is_position_independent() && (ref_flags & Symbol::ABSOLUTE_REF) && !is_32bit); // For global symbols, we use the same helper routines used in the // scan pass. If we did not create a dynamic relocation, or if we // created a RELATIVE dynamic relocation, we should apply the static // relocation. bool has_dyn = gsym->needs_dynamic_reloc(ref_flags); bool is_rel = (ref_flags & Symbol::ABSOLUTE_REF) && gsym->can_use_relative_reloc(ref_flags & Symbol::FUNCTION_CALL); return !has_dyn || is_rel; } // Perform a relocation. template<bool big_endian> inline bool Target_arm<big_endian>::Relocate::relocate( const Relocate_info<32, big_endian>* relinfo, Target_arm* target, Output_section* output_section, size_t relnum, const elfcpp::Rel<32, big_endian>& rel, unsigned int r_type, const Sized_symbol<32>* gsym, const Symbol_value<32>* psymval, unsigned char* view, Arm_address address, section_size_type view_size) { typedef Arm_relocate_functions<big_endian> Arm_relocate_functions; r_type = get_real_reloc_type(r_type); const Arm_reloc_property* reloc_property = arm_reloc_property_table->get_implemented_static_reloc_property(r_type); if (reloc_property == NULL) { std::string reloc_name = arm_reloc_property_table->reloc_name_in_error_message(r_type); gold_error_at_location(relinfo, relnum, rel.get_r_offset(), _("cannot relocate %s in object file"), reloc_name.c_str()); return true; } const Arm_relobj<big_endian>* object = Arm_relobj<big_endian>::as_arm_relobj(relinfo->object); // If the final branch target of a relocation is THUMB instruction, this // is 1. Otherwise it is 0. Arm_address thumb_bit = 0; Symbol_value<32> symval; bool is_weakly_undefined_without_plt = false; bool have_got_offset = false; unsigned int got_offset = 0; // If the relocation uses the GOT entry of a symbol instead of the symbol // itself, we don't care about whether the symbol is defined or what kind // of symbol it is. if (reloc_property->uses_got_entry()) { // Get the GOT offset. // The GOT pointer points to the end of the GOT section. // We need to subtract the size of the GOT section to get // the actual offset to use in the relocation. // TODO: We should move GOT offset computing code in TLS relocations // to here. switch (r_type) { case elfcpp::R_ARM_GOT_BREL: case elfcpp::R_ARM_GOT_PREL: 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<32>(rel.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; default: break; } } else if (relnum != Target_arm<big_endian>::fake_relnum_for_stubs) { if (gsym != NULL) { // This is a global symbol. Determine if we use PLT and if the // final target is THUMB. if (gsym->use_plt_offset(Scan::get_reference_flags(r_type))) { // This uses a PLT, change the symbol value. symval.set_output_value(target->plt_section()->address() + gsym->plt_offset()); psymval = &symval; } else if (gsym->is_weak_undefined()) { // This is a weakly undefined symbol and we do not use PLT // for this relocation. A branch targeting this symbol will // be converted into an NOP. is_weakly_undefined_without_plt = true; } else if (gsym->is_undefined() && reloc_property->uses_symbol()) { // This relocation uses the symbol value but the symbol is // undefined. Exit early and have the caller reporting an // error. return true; } else { // Set thumb bit if symbol: // -Has type STT_ARM_TFUNC or // -Has type STT_FUNC, is defined and with LSB in value set. thumb_bit = (((gsym->type() == elfcpp::STT_ARM_TFUNC) || (gsym->type() == elfcpp::STT_FUNC && !gsym->is_undefined() && ((psymval->value(object, 0) & 1) != 0))) ? 1 : 0); } } else { // This is a local symbol. Determine if the final target is THUMB. // We saved this information when all the local symbols were read. elfcpp::Elf_types<32>::Elf_WXword r_info = rel.get_r_info(); unsigned int r_sym = elfcpp::elf_r_sym<32>(r_info); thumb_bit = object->local_symbol_is_thumb_function(r_sym) ? 1 : 0; } } else { // This is a fake relocation synthesized for a stub. It does not have // a real symbol. We just look at the LSB of the symbol value to // determine if the target is THUMB or not. thumb_bit = ((psymval->value(object, 0) & 1) != 0); } // Strip LSB if this points to a THUMB target. if (thumb_bit != 0 && reloc_property->uses_thumb_bit() && ((psymval->value(object, 0) & 1) != 0)) { Arm_address stripped_value = psymval->value(object, 0) & ~static_cast<Arm_address>(1); symval.set_output_value(stripped_value); psymval = &symval; } // To look up relocation stubs, we need to pass the symbol table index of // a local symbol. unsigned int r_sym = elfcpp::elf_r_sym<32>(rel.get_r_info()); // Get the addressing origin of the output segment defining the // symbol gsym if needed (AAELF 4.6.1.2 Relocation types). Arm_address sym_origin = 0; if (reloc_property->uses_symbol_base()) { if (r_type == elfcpp::R_ARM_BASE_ABS && gsym == NULL) // R_ARM_BASE_ABS with the NULL symbol will give the // absolute address of the GOT origin (GOT_ORG) (see ARM IHI // 0044C (AAELF): 4.6.1.8 Proxy generating relocations). sym_origin = target->got_plt_section()->address(); else if (gsym == NULL) sym_origin = 0; else if (gsym->source() == Symbol::IN_OUTPUT_SEGMENT) sym_origin = gsym->output_segment()->vaddr(); else if (gsym->source() == Symbol::IN_OUTPUT_DATA) sym_origin = gsym->output_data()->address(); // TODO: Assumes the segment base to be zero for the global symbols // till the proper support for the segment-base-relative addressing // will be implemented. This is consistent with GNU ld. } // For relative addressing relocation, find out the relative address base. Arm_address relative_address_base = 0; switch(reloc_property->relative_address_base()) { case Arm_reloc_property::RAB_NONE: // Relocations with relative address bases RAB_TLS and RAB_tp are // handled by relocate_tls. So we do not need to do anything here. case Arm_reloc_property::RAB_TLS: case Arm_reloc_property::RAB_tp: break; case Arm_reloc_property::RAB_B_S: relative_address_base = sym_origin; break; case Arm_reloc_property::RAB_GOT_ORG: relative_address_base = target->got_plt_section()->address(); break; case Arm_reloc_property::RAB_P: relative_address_base = address; break; case Arm_reloc_property::RAB_Pa: relative_address_base = address & 0xfffffffcU; break; default: gold_unreachable(); } typename Arm_relocate_functions::Status reloc_status = Arm_relocate_functions::STATUS_OKAY; bool check_overflow = reloc_property->checks_overflow(); switch (r_type) { case elfcpp::R_ARM_NONE: break; case elfcpp::R_ARM_ABS8: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::abs8(view, object, psymval); break; case elfcpp::R_ARM_ABS12: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::abs12(view, object, psymval); break; case elfcpp::R_ARM_ABS16: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::abs16(view, object, psymval); break; case elfcpp::R_ARM_ABS32: if (should_apply_static_reloc(gsym, r_type, true, output_section)) reloc_status = Arm_relocate_functions::abs32(view, object, psymval, thumb_bit); break; case elfcpp::R_ARM_ABS32_NOI: if (should_apply_static_reloc(gsym, r_type, true, output_section)) // No thumb bit for this relocation: (S + A) reloc_status = Arm_relocate_functions::abs32(view, object, psymval, 0); break; case elfcpp::R_ARM_MOVW_ABS_NC: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::movw(view, object, psymval, 0, thumb_bit, check_overflow); break; case elfcpp::R_ARM_MOVT_ABS: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::movt(view, object, psymval, 0); break; case elfcpp::R_ARM_THM_MOVW_ABS_NC: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::thm_movw(view, object, psymval, 0, thumb_bit, false); break; case elfcpp::R_ARM_THM_MOVT_ABS: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::thm_movt(view, object, psymval, 0); break; case elfcpp::R_ARM_MOVW_PREL_NC: case elfcpp::R_ARM_MOVW_BREL_NC: case elfcpp::R_ARM_MOVW_BREL: reloc_status = Arm_relocate_functions::movw(view, object, psymval, relative_address_base, thumb_bit, check_overflow); break; case elfcpp::R_ARM_MOVT_PREL: case elfcpp::R_ARM_MOVT_BREL: reloc_status = Arm_relocate_functions::movt(view, object, psymval, relative_address_base); break; case elfcpp::R_ARM_THM_MOVW_PREL_NC: case elfcpp::R_ARM_THM_MOVW_BREL_NC: case elfcpp::R_ARM_THM_MOVW_BREL: reloc_status = Arm_relocate_functions::thm_movw(view, object, psymval, relative_address_base, thumb_bit, check_overflow); break; case elfcpp::R_ARM_THM_MOVT_PREL: case elfcpp::R_ARM_THM_MOVT_BREL: reloc_status = Arm_relocate_functions::thm_movt(view, object, psymval, relative_address_base); break; case elfcpp::R_ARM_REL32: reloc_status = Arm_relocate_functions::rel32(view, object, psymval, address, thumb_bit); break; case elfcpp::R_ARM_THM_ABS5: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::thm_abs5(view, object, psymval); break; // Thumb long branches. case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_THM_XPC22: case elfcpp::R_ARM_THM_JUMP24: reloc_status = Arm_relocate_functions::thumb_branch_common( r_type, relinfo, view, gsym, object, r_sym, psymval, address, thumb_bit, is_weakly_undefined_without_plt); break; case elfcpp::R_ARM_GOTOFF32: { Arm_address got_origin; got_origin = target->got_plt_section()->address(); reloc_status = Arm_relocate_functions::rel32(view, object, psymval, got_origin, thumb_bit); } break; case elfcpp::R_ARM_BASE_PREL: gold_assert(gsym != NULL); reloc_status = Arm_relocate_functions::base_prel(view, sym_origin, address); break; case elfcpp::R_ARM_BASE_ABS: if (should_apply_static_reloc(gsym, r_type, false, output_section)) reloc_status = Arm_relocate_functions::base_abs(view, sym_origin); break; case elfcpp::R_ARM_GOT_BREL: gold_assert(have_got_offset); reloc_status = Arm_relocate_functions::got_brel(view, got_offset); break; case elfcpp::R_ARM_GOT_PREL: gold_assert(have_got_offset); // Get the address origin for GOT PLT, which is allocated right // after the GOT section, to calculate an absolute address of // the symbol GOT entry (got_origin + got_offset). Arm_address got_origin; got_origin = target->got_plt_section()->address(); reloc_status = Arm_relocate_functions::got_prel(view, got_origin + got_offset, address); break; case elfcpp::R_ARM_PLT32: case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_XPC25: gold_assert(gsym == NULL || gsym->has_plt_offset() || gsym->final_value_is_known() || (gsym->is_defined() && !gsym->is_from_dynobj() && !gsym->is_preemptible())); reloc_status = Arm_relocate_functions::arm_branch_common( r_type, relinfo, view, gsym, object, r_sym, psymval, address, thumb_bit, is_weakly_undefined_without_plt); break; case elfcpp::R_ARM_THM_JUMP19: reloc_status = Arm_relocate_functions::thm_jump19(view, object, psymval, address, thumb_bit); break; case elfcpp::R_ARM_THM_JUMP6: reloc_status = Arm_relocate_functions::thm_jump6(view, object, psymval, address); break; case elfcpp::R_ARM_THM_JUMP8: reloc_status = Arm_relocate_functions::thm_jump8(view, object, psymval, address); break; case elfcpp::R_ARM_THM_JUMP11: reloc_status = Arm_relocate_functions::thm_jump11(view, object, psymval, address); break; case elfcpp::R_ARM_PREL31: reloc_status = Arm_relocate_functions::prel31(view, object, psymval, address, thumb_bit); break; case elfcpp::R_ARM_V4BX: if (target->fix_v4bx() > General_options::FIX_V4BX_NONE) { const bool is_v4bx_interworking = (target->fix_v4bx() == General_options::FIX_V4BX_INTERWORKING); reloc_status = Arm_relocate_functions::v4bx(relinfo, view, object, address, is_v4bx_interworking); } break; case elfcpp::R_ARM_THM_PC8: reloc_status = Arm_relocate_functions::thm_pc8(view, object, psymval, address); break; case elfcpp::R_ARM_THM_PC12: reloc_status = Arm_relocate_functions::thm_pc12(view, object, psymval, address); break; case elfcpp::R_ARM_THM_ALU_PREL_11_0: reloc_status = Arm_relocate_functions::thm_alu11(view, object, psymval, address, thumb_bit); break; case elfcpp::R_ARM_ALU_PC_G0_NC: case elfcpp::R_ARM_ALU_PC_G0: case elfcpp::R_ARM_ALU_PC_G1_NC: case elfcpp::R_ARM_ALU_PC_G1: case elfcpp::R_ARM_ALU_PC_G2: case elfcpp::R_ARM_ALU_SB_G0_NC: case elfcpp::R_ARM_ALU_SB_G0: case elfcpp::R_ARM_ALU_SB_G1_NC: case elfcpp::R_ARM_ALU_SB_G1: case elfcpp::R_ARM_ALU_SB_G2: reloc_status = Arm_relocate_functions::arm_grp_alu(view, object, psymval, reloc_property->group_index(), relative_address_base, thumb_bit, check_overflow); break; case elfcpp::R_ARM_LDR_PC_G0: case elfcpp::R_ARM_LDR_PC_G1: case elfcpp::R_ARM_LDR_PC_G2: case elfcpp::R_ARM_LDR_SB_G0: case elfcpp::R_ARM_LDR_SB_G1: case elfcpp::R_ARM_LDR_SB_G2: reloc_status = Arm_relocate_functions::arm_grp_ldr(view, object, psymval, reloc_property->group_index(), relative_address_base); break; case elfcpp::R_ARM_LDRS_PC_G0: case elfcpp::R_ARM_LDRS_PC_G1: case elfcpp::R_ARM_LDRS_PC_G2: case elfcpp::R_ARM_LDRS_SB_G0: case elfcpp::R_ARM_LDRS_SB_G1: case elfcpp::R_ARM_LDRS_SB_G2: reloc_status = Arm_relocate_functions::arm_grp_ldrs(view, object, psymval, reloc_property->group_index(), relative_address_base); break; case elfcpp::R_ARM_LDC_PC_G0: case elfcpp::R_ARM_LDC_PC_G1: case elfcpp::R_ARM_LDC_PC_G2: case elfcpp::R_ARM_LDC_SB_G0: case elfcpp::R_ARM_LDC_SB_G1: case elfcpp::R_ARM_LDC_SB_G2: reloc_status = Arm_relocate_functions::arm_grp_ldc(view, object, psymval, reloc_property->group_index(), relative_address_base); break; // These are initial tls relocs, which are expected when // linking. case elfcpp::R_ARM_TLS_GD32: // Global-dynamic case elfcpp::R_ARM_TLS_LDM32: // Local-dynamic case elfcpp::R_ARM_TLS_LDO32: // Alternate local-dynamic case elfcpp::R_ARM_TLS_IE32: // Initial-exec case elfcpp::R_ARM_TLS_LE32: // Local-exec reloc_status = this->relocate_tls(relinfo, target, relnum, rel, r_type, gsym, psymval, view, address, view_size); break; // The known and unknown unsupported and/or deprecated relocations. case elfcpp::R_ARM_PC24: case elfcpp::R_ARM_LDR_SBREL_11_0_NC: case elfcpp::R_ARM_ALU_SBREL_19_12_NC: case elfcpp::R_ARM_ALU_SBREL_27_20_CK: default: // Just silently leave the method. We should get an appropriate error // message in the scan methods. break; } // Report any errors. switch (reloc_status) { case Arm_relocate_functions::STATUS_OKAY: break; case Arm_relocate_functions::STATUS_OVERFLOW: gold_error_at_location(relinfo, relnum, rel.get_r_offset(), _("relocation overflow in %s"), reloc_property->name().c_str()); break; case Arm_relocate_functions::STATUS_BAD_RELOC: gold_error_at_location( relinfo, relnum, rel.get_r_offset(), _("unexpected opcode while processing relocation %s"), reloc_property->name().c_str()); break; default: gold_unreachable(); } return true; } // Perform a TLS relocation. template<bool big_endian> inline typename Arm_relocate_functions<big_endian>::Status Target_arm<big_endian>::Relocate::relocate_tls( const Relocate_info<32, big_endian>* relinfo, Target_arm<big_endian>* target, size_t relnum, const elfcpp::Rel<32, big_endian>& rel, unsigned int r_type, const Sized_symbol<32>* gsym, const Symbol_value<32>* psymval, unsigned char* view, elfcpp::Elf_types<32>::Elf_Addr address, section_size_type /*view_size*/ ) { typedef Arm_relocate_functions<big_endian> ArmRelocFuncs; typedef Relocate_functions<32, big_endian> RelocFuncs; Output_segment* tls_segment = relinfo->layout->tls_segment(); const Sized_relobj_file<32, big_endian>* object = relinfo->object; elfcpp::Elf_types<32>::Elf_Addr value = psymval->value(object, 0); const bool is_final = (gsym == NULL ? !parameters->options().shared() : gsym->final_value_is_known()); const tls::Tls_optimization optimized_type = Target_arm<big_endian>::optimize_tls_reloc(is_final, r_type); switch (r_type) { case elfcpp::R_ARM_TLS_GD32: // Global-dynamic { unsigned int got_type = 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<32>(rel.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_NONE) { Arm_address got_entry = target->got_plt_section()->address() + got_offset; // Relocate the field with the PC relative offset of the pair of // GOT entries. RelocFuncs::pcrel32(view, got_entry, address); return ArmRelocFuncs::STATUS_OKAY; } } break; case elfcpp::R_ARM_TLS_LDM32: // Local-dynamic 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) - target->got_size()); Arm_address got_entry = target->got_plt_section()->address() + got_offset; // Relocate the field with the PC relative offset of the pair of // GOT entries. RelocFuncs::pcrel32(view, got_entry, address); return ArmRelocFuncs::STATUS_OKAY; } break; case elfcpp::R_ARM_TLS_LDO32: // Alternate local-dynamic RelocFuncs::rel32(view, value); return ArmRelocFuncs::STATUS_OKAY; case elfcpp::R_ARM_TLS_IE32: // Initial-exec 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_type = GOT_TYPE_TLS_OFFSET; unsigned int got_offset; if (gsym != NULL) { gold_assert(gsym->has_got_offset(got_type)); got_offset = gsym->got_offset(got_type); } else { unsigned int r_sym = elfcpp::elf_r_sym<32>(rel.get_r_info()); gold_assert(object->local_has_got_offset(r_sym, got_type)); got_offset = object->local_got_offset(r_sym, got_type); } // All GOT offsets are relative to the end of the GOT. got_offset -= target->got_size(); Arm_address got_entry = target->got_plt_section()->address() + got_offset; // Relocate the field with the PC relative offset of the GOT entry. RelocFuncs::pcrel32(view, got_entry, address); return ArmRelocFuncs::STATUS_OKAY; } break; case elfcpp::R_ARM_TLS_LE32: // Local-exec // If we're creating a shared library, a dynamic relocation will // have been created for this location, so do not apply it now. if (!parameters->options().shared()) { gold_assert(tls_segment != NULL); // $tp points to the TCB, which is followed by the TLS, so we // need to add TCB size to the offset. Arm_address aligned_tcb_size = align_address(ARM_TCB_SIZE, tls_segment->maximum_alignment()); RelocFuncs::rel32(view, value + aligned_tcb_size); } return ArmRelocFuncs::STATUS_OKAY; default: gold_unreachable(); } gold_error_at_location(relinfo, relnum, rel.get_r_offset(), _("unsupported reloc %u"), r_type); return ArmRelocFuncs::STATUS_BAD_RELOC; } // Relocate section data. template<bool big_endian> void Target_arm<big_endian>::relocate_section( const Relocate_info<32, big_endian>* relinfo, unsigned int sh_type, const unsigned char* prelocs, size_t reloc_count, Output_section* output_section, bool needs_special_offset_handling, unsigned char* view, Arm_address address, section_size_type view_size, const Reloc_symbol_changes* reloc_symbol_changes) { typedef typename Target_arm<big_endian>::Relocate Arm_relocate; gold_assert(sh_type == elfcpp::SHT_REL); // See if we are relocating a relaxed input section. If so, the view // covers the whole output section and we need to adjust accordingly. if (needs_special_offset_handling) { const Output_relaxed_input_section* poris = output_section->find_relaxed_input_section(relinfo->object, relinfo->data_shndx); if (poris != NULL) { Arm_address section_address = poris->address(); section_size_type section_size = poris->data_size(); gold_assert((section_address >= address) && ((section_address + section_size) <= (address + view_size))); off_t offset = section_address - address; view += offset; address += offset; view_size = section_size; } } gold::relocate_section<32, big_endian, Target_arm, elfcpp::SHT_REL, Arm_relocate>( relinfo, this, prelocs, reloc_count, output_section, needs_special_offset_handling, view, address, view_size, reloc_symbol_changes); } // Return the size of a relocation while scanning during a relocatable // link. template<bool big_endian> unsigned int Target_arm<big_endian>::Relocatable_size_for_reloc::get_size_for_reloc( unsigned int r_type, Relobj* object) { r_type = get_real_reloc_type(r_type); const Arm_reloc_property* arp = arm_reloc_property_table->get_implemented_static_reloc_property(r_type); if (arp != NULL) return arp->size(); else { std::string reloc_name = arm_reloc_property_table->reloc_name_in_error_message(r_type); gold_error(_("%s: unexpected %s in object file"), object->name().c_str(), reloc_name.c_str()); return 0; } } // Scan the relocs during a relocatable link. template<bool big_endian> void Target_arm<big_endian>::scan_relocatable_relocs( Symbol_table* symtab, Layout* layout, Sized_relobj_file<32, big_endian>* 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) { gold_assert(sh_type == elfcpp::SHT_REL); typedef Arm_scan_relocatable_relocs<big_endian, elfcpp::SHT_REL, Relocatable_size_for_reloc> Scan_relocatable_relocs; gold::scan_relocatable_relocs<32, big_endian, elfcpp::SHT_REL, Scan_relocatable_relocs>( symtab, layout, object, data_shndx, prelocs, reloc_count, output_section, needs_special_offset_handling, local_symbol_count, plocal_symbols, rr); } // Relocate a section during a relocatable link. template<bool big_endian> void Target_arm<big_endian>::relocate_for_relocatable( const Relocate_info<32, big_endian>* 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, unsigned char* view, Arm_address view_address, section_size_type view_size, unsigned char* reloc_view, section_size_type reloc_view_size) { gold_assert(sh_type == elfcpp::SHT_REL); gold::relocate_for_relocatable<32, big_endian, elfcpp::SHT_REL>( relinfo, prelocs, reloc_count, output_section, offset_in_output_section, rr, view, view_address, view_size, reloc_view, reloc_view_size); } // Perform target-specific processing in a relocatable link. This is // only used if we use the relocation strategy RELOC_SPECIAL. template<bool big_endian> void Target_arm<big_endian>::relocate_special_relocatable( const Relocate_info<32, big_endian>* relinfo, unsigned int sh_type, const unsigned char* preloc_in, size_t relnum, Output_section* output_section, off_t offset_in_output_section, unsigned char* view, elfcpp::Elf_types<32>::Elf_Addr view_address, section_size_type, unsigned char* preloc_out) { // We can only handle REL type relocation sections. gold_assert(sh_type == elfcpp::SHT_REL); typedef typename Reloc_types<elfcpp::SHT_REL, 32, big_endian>::Reloc Reltype; typedef typename Reloc_types<elfcpp::SHT_REL, 32, big_endian>::Reloc_write Reltype_write; const Arm_address invalid_address = static_cast<Arm_address>(0) - 1; const Arm_relobj<big_endian>* object = Arm_relobj<big_endian>::as_arm_relobj(relinfo->object); const unsigned int local_count = object->local_symbol_count(); Reltype reloc(preloc_in); Reltype_write reloc_write(preloc_out); elfcpp::Elf_types<32>::Elf_WXword r_info = reloc.get_r_info(); const unsigned int r_sym = elfcpp::elf_r_sym<32>(r_info); const unsigned int r_type = elfcpp::elf_r_type<32>(r_info); const Arm_reloc_property* arp = arm_reloc_property_table->get_implemented_static_reloc_property(r_type); gold_assert(arp != NULL); // Get the new symbol index. // We only use RELOC_SPECIAL strategy in local relocations. gold_assert(r_sym < local_count); // We are adjusting a section symbol. We need to find // the symbol table index of the section symbol for // the output section corresponding to input section // in which this symbol is defined. bool is_ordinary; unsigned int shndx = object->local_symbol_input_shndx(r_sym, &is_ordinary); gold_assert(is_ordinary); Output_section* os = object->output_section(shndx); gold_assert(os != NULL); gold_assert(os->needs_symtab_index()); unsigned int new_symndx = os->symtab_index(); // Get the new offset--the location in the output section where // this relocation should be applied. Arm_address offset = reloc.get_r_offset(); Arm_address new_offset; if (offset_in_output_section != invalid_address) new_offset = offset + offset_in_output_section; else { section_offset_type sot_offset = convert_types<section_offset_type, Arm_address>(offset); section_offset_type new_sot_offset = output_section->output_offset(object, relinfo->data_shndx, sot_offset); gold_assert(new_sot_offset != -1); new_offset = new_sot_offset; } // In an object file, r_offset is an offset within the section. // In an executable or dynamic object, generated by // --emit-relocs, r_offset is an absolute address. if (!parameters->options().relocatable()) { new_offset += view_address; if (offset_in_output_section != invalid_address) new_offset -= offset_in_output_section; } reloc_write.put_r_offset(new_offset); reloc_write.put_r_info(elfcpp::elf_r_info<32>(new_symndx, r_type)); // Handle the reloc addend. // The relocation uses a section symbol in the input file. // We are adjusting it to use a section symbol in the output // file. The input section symbol refers to some address in // the input section. We need the relocation in the output // file to refer to that same address. This adjustment to // the addend is the same calculation we use for a simple // absolute relocation for the input section symbol. const Symbol_value<32>* psymval = object->local_symbol(r_sym); // Handle THUMB bit. Symbol_value<32> symval; Arm_address thumb_bit = object->local_symbol_is_thumb_function(r_sym) ? 1 : 0; if (thumb_bit != 0 && arp->uses_thumb_bit() && ((psymval->value(object, 0) & 1) != 0)) { Arm_address stripped_value = psymval->value(object, 0) & ~static_cast<Arm_address>(1); symval.set_output_value(stripped_value); psymval = &symval; } unsigned char* paddend = view + offset; typename Arm_relocate_functions<big_endian>::Status reloc_status = Arm_relocate_functions<big_endian>::STATUS_OKAY; switch (r_type) { case elfcpp::R_ARM_ABS8: reloc_status = Arm_relocate_functions<big_endian>::abs8(paddend, object, psymval); break; case elfcpp::R_ARM_ABS12: reloc_status = Arm_relocate_functions<big_endian>::abs12(paddend, object, psymval); break; case elfcpp::R_ARM_ABS16: reloc_status = Arm_relocate_functions<big_endian>::abs16(paddend, object, psymval); break; case elfcpp::R_ARM_THM_ABS5: reloc_status = Arm_relocate_functions<big_endian>::thm_abs5(paddend, object, psymval); break; case elfcpp::R_ARM_MOVW_ABS_NC: case elfcpp::R_ARM_MOVW_PREL_NC: case elfcpp::R_ARM_MOVW_BREL_NC: case elfcpp::R_ARM_MOVW_BREL: reloc_status = Arm_relocate_functions<big_endian>::movw( paddend, object, psymval, 0, thumb_bit, arp->checks_overflow()); break; case elfcpp::R_ARM_THM_MOVW_ABS_NC: case elfcpp::R_ARM_THM_MOVW_PREL_NC: case elfcpp::R_ARM_THM_MOVW_BREL_NC: case elfcpp::R_ARM_THM_MOVW_BREL: reloc_status = Arm_relocate_functions<big_endian>::thm_movw( paddend, object, psymval, 0, thumb_bit, arp->checks_overflow()); break; case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_THM_XPC22: case elfcpp::R_ARM_THM_JUMP24: reloc_status = Arm_relocate_functions<big_endian>::thumb_branch_common( r_type, relinfo, paddend, NULL, object, 0, psymval, 0, thumb_bit, false); break; case elfcpp::R_ARM_PLT32: case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_XPC25: reloc_status = Arm_relocate_functions<big_endian>::arm_branch_common( r_type, relinfo, paddend, NULL, object, 0, psymval, 0, thumb_bit, false); break; case elfcpp::R_ARM_THM_JUMP19: reloc_status = Arm_relocate_functions<big_endian>::thm_jump19(paddend, object, psymval, 0, thumb_bit); break; case elfcpp::R_ARM_THM_JUMP6: reloc_status = Arm_relocate_functions<big_endian>::thm_jump6(paddend, object, psymval, 0); break; case elfcpp::R_ARM_THM_JUMP8: reloc_status = Arm_relocate_functions<big_endian>::thm_jump8(paddend, object, psymval, 0); break; case elfcpp::R_ARM_THM_JUMP11: reloc_status = Arm_relocate_functions<big_endian>::thm_jump11(paddend, object, psymval, 0); break; case elfcpp::R_ARM_PREL31: reloc_status = Arm_relocate_functions<big_endian>::prel31(paddend, object, psymval, 0, thumb_bit); break; case elfcpp::R_ARM_THM_PC8: reloc_status = Arm_relocate_functions<big_endian>::thm_pc8(paddend, object, psymval, 0); break; case elfcpp::R_ARM_THM_PC12: reloc_status = Arm_relocate_functions<big_endian>::thm_pc12(paddend, object, psymval, 0); break; case elfcpp::R_ARM_THM_ALU_PREL_11_0: reloc_status = Arm_relocate_functions<big_endian>::thm_alu11(paddend, object, psymval, 0, thumb_bit); break; // These relocation truncate relocation results so we cannot handle them // in a relocatable link. case elfcpp::R_ARM_MOVT_ABS: case elfcpp::R_ARM_THM_MOVT_ABS: case elfcpp::R_ARM_MOVT_PREL: case elfcpp::R_ARM_MOVT_BREL: case elfcpp::R_ARM_THM_MOVT_PREL: case elfcpp::R_ARM_THM_MOVT_BREL: case elfcpp::R_ARM_ALU_PC_G0_NC: case elfcpp::R_ARM_ALU_PC_G0: case elfcpp::R_ARM_ALU_PC_G1_NC: case elfcpp::R_ARM_ALU_PC_G1: case elfcpp::R_ARM_ALU_PC_G2: case elfcpp::R_ARM_ALU_SB_G0_NC: case elfcpp::R_ARM_ALU_SB_G0: case elfcpp::R_ARM_ALU_SB_G1_NC: case elfcpp::R_ARM_ALU_SB_G1: case elfcpp::R_ARM_ALU_SB_G2: case elfcpp::R_ARM_LDR_PC_G0: case elfcpp::R_ARM_LDR_PC_G1: case elfcpp::R_ARM_LDR_PC_G2: case elfcpp::R_ARM_LDR_SB_G0: case elfcpp::R_ARM_LDR_SB_G1: case elfcpp::R_ARM_LDR_SB_G2: case elfcpp::R_ARM_LDRS_PC_G0: case elfcpp::R_ARM_LDRS_PC_G1: case elfcpp::R_ARM_LDRS_PC_G2: case elfcpp::R_ARM_LDRS_SB_G0: case elfcpp::R_ARM_LDRS_SB_G1: case elfcpp::R_ARM_LDRS_SB_G2: case elfcpp::R_ARM_LDC_PC_G0: case elfcpp::R_ARM_LDC_PC_G1: case elfcpp::R_ARM_LDC_PC_G2: case elfcpp::R_ARM_LDC_SB_G0: case elfcpp::R_ARM_LDC_SB_G1: case elfcpp::R_ARM_LDC_SB_G2: gold_error(_("cannot handle %s in a relocatable link"), arp->name().c_str()); break; default: gold_unreachable(); } // Report any errors. switch (reloc_status) { case Arm_relocate_functions<big_endian>::STATUS_OKAY: break; case Arm_relocate_functions<big_endian>::STATUS_OVERFLOW: gold_error_at_location(relinfo, relnum, reloc.get_r_offset(), _("relocation overflow in %s"), arp->name().c_str()); break; case Arm_relocate_functions<big_endian>::STATUS_BAD_RELOC: gold_error_at_location(relinfo, relnum, reloc.get_r_offset(), _("unexpected opcode while processing relocation %s"), arp->name().c_str()); break; default: gold_unreachable(); } } // Return the value to use for a dynamic symbol which requires special // treatment. This is how we support equality comparisons of function // pointers across shared library boundaries, as described in the // processor specific ABI supplement. template<bool big_endian> uint64_t Target_arm<big_endian>::do_dynsym_value(const Symbol* gsym) const { gold_assert(gsym->is_from_dynobj() && gsym->has_plt_offset()); return this->plt_section()->address() + gsym->plt_offset(); } // Map platform-specific relocs to real relocs // template<bool big_endian> unsigned int Target_arm<big_endian>::get_real_reloc_type(unsigned int r_type) { switch (r_type) { case elfcpp::R_ARM_TARGET1: // This is either R_ARM_ABS32 or R_ARM_REL32; return elfcpp::R_ARM_ABS32; case elfcpp::R_ARM_TARGET2: // This can be any reloc type but usually is R_ARM_GOT_PREL return elfcpp::R_ARM_GOT_PREL; default: return r_type; } } // Whether if two EABI versions V1 and V2 are compatible. template<bool big_endian> bool Target_arm<big_endian>::are_eabi_versions_compatible( elfcpp::Elf_Word v1, elfcpp::Elf_Word v2) { // v4 and v5 are the same spec before and after it was released, // so allow mixing them. if ((v1 == elfcpp::EF_ARM_EABI_UNKNOWN || v2 == elfcpp::EF_ARM_EABI_UNKNOWN) || (v1 == elfcpp::EF_ARM_EABI_VER4 && v2 == elfcpp::EF_ARM_EABI_VER5) || (v1 == elfcpp::EF_ARM_EABI_VER5 && v2 == elfcpp::EF_ARM_EABI_VER4)) return true; return v1 == v2; } // Combine FLAGS from an input object called NAME and the processor-specific // flags in the ELF header of the output. Much of this is adapted from the // processor-specific flags merging code in elf32_arm_merge_private_bfd_data // in bfd/elf32-arm.c. template<bool big_endian> void Target_arm<big_endian>::merge_processor_specific_flags( const std::string& name, elfcpp::Elf_Word flags) { if (this->are_processor_specific_flags_set()) { elfcpp::Elf_Word out_flags = this->processor_specific_flags(); // Nothing to merge if flags equal to those in output. if (flags == out_flags) return; // Complain about various flag mismatches. elfcpp::Elf_Word version1 = elfcpp::arm_eabi_version(flags); elfcpp::Elf_Word version2 = elfcpp::arm_eabi_version(out_flags); if (!this->are_eabi_versions_compatible(version1, version2) && parameters->options().warn_mismatch()) gold_error(_("Source object %s has EABI version %d but output has " "EABI version %d."), name.c_str(), (flags & elfcpp::EF_ARM_EABIMASK) >> 24, (out_flags & elfcpp::EF_ARM_EABIMASK) >> 24); } else { // If the input is the default architecture and had the default // flags then do not bother setting the flags for the output // architecture, instead allow future merges to do this. If no // future merges ever set these flags then they will retain their // uninitialised values, which surprise surprise, correspond // to the default values. if (flags == 0) return; // This is the first time, just copy the flags. // We only copy the EABI version for now. this->set_processor_specific_flags(flags & elfcpp::EF_ARM_EABIMASK); } } // Adjust ELF file header. template<bool big_endian> void Target_arm<big_endian>::do_adjust_elf_header( unsigned char* view, int len) const { gold_assert(len == elfcpp::Elf_sizes<32>::ehdr_size); elfcpp::Ehdr<32, big_endian> ehdr(view); unsigned char e_ident[elfcpp::EI_NIDENT]; memcpy(e_ident, ehdr.get_e_ident(), elfcpp::EI_NIDENT); if (elfcpp::arm_eabi_version(this->processor_specific_flags()) == elfcpp::EF_ARM_EABI_UNKNOWN) e_ident[elfcpp::EI_OSABI] = elfcpp::ELFOSABI_ARM; else e_ident[elfcpp::EI_OSABI] = 0; e_ident[elfcpp::EI_ABIVERSION] = 0; // FIXME: Do EF_ARM_BE8 adjustment. elfcpp::Ehdr_write<32, big_endian> oehdr(view); oehdr.put_e_ident(e_ident); } // do_make_elf_object to override the same function in the base class. // We need to use a target-specific sub-class of // Sized_relobj_file<32, big_endian> to store ARM specific information. // Hence we need to have our own ELF object creation. template<bool big_endian> Object* Target_arm<big_endian>::do_make_elf_object( const std::string& name, Input_file* input_file, off_t offset, const elfcpp::Ehdr<32, big_endian>& ehdr) { int et = ehdr.get_e_type(); if (et == elfcpp::ET_REL) { Arm_relobj<big_endian>* obj = new Arm_relobj<big_endian>(name, input_file, offset, ehdr); obj->setup(); return obj; } else if (et == elfcpp::ET_DYN) { Sized_dynobj<32, big_endian>* obj = new Arm_dynobj<big_endian>(name, input_file, offset, ehdr); obj->setup(); return obj; } else { gold_error(_("%s: unsupported ELF file type %d"), name.c_str(), et); return NULL; } } // Read the architecture from the Tag_also_compatible_with attribute, if any. // Returns -1 if no architecture could be read. // This is adapted from get_secondary_compatible_arch() in bfd/elf32-arm.c. template<bool big_endian> int Target_arm<big_endian>::get_secondary_compatible_arch( const Attributes_section_data* pasd) { const Object_attribute* known_attributes = pasd->known_attributes(Object_attribute::OBJ_ATTR_PROC); // Note: the tag and its argument below are uleb128 values, though // currently-defined values fit in one byte for each. const std::string& sv = known_attributes[elfcpp::Tag_also_compatible_with].string_value(); if (sv.size() == 2 && sv.data()[0] == elfcpp::Tag_CPU_arch && (sv.data()[1] & 128) != 128) return sv.data()[1]; // This tag is "safely ignorable", so don't complain if it looks funny. return -1; } // Set, or unset, the architecture of the Tag_also_compatible_with attribute. // The tag is removed if ARCH is -1. // This is adapted from set_secondary_compatible_arch() in bfd/elf32-arm.c. template<bool big_endian> void Target_arm<big_endian>::set_secondary_compatible_arch( Attributes_section_data* pasd, int arch) { Object_attribute* known_attributes = pasd->known_attributes(Object_attribute::OBJ_ATTR_PROC); if (arch == -1) { known_attributes[elfcpp::Tag_also_compatible_with].set_string_value(""); return; } // Note: the tag and its argument below are uleb128 values, though // currently-defined values fit in one byte for each. char sv[3]; sv[0] = elfcpp::Tag_CPU_arch; gold_assert(arch != 0); sv[1] = arch; sv[2] = '\0'; known_attributes[elfcpp::Tag_also_compatible_with].set_string_value(sv); } // Combine two values for Tag_CPU_arch, taking secondary compatibility tags // into account. // This is adapted from tag_cpu_arch_combine() in bfd/elf32-arm.c. template<bool big_endian> int Target_arm<big_endian>::tag_cpu_arch_combine( const char* name, int oldtag, int* secondary_compat_out, int newtag, int secondary_compat) { #define T(X) elfcpp::TAG_CPU_ARCH_##X static const int v6t2[] = { T(V6T2), // PRE_V4. T(V6T2), // V4. T(V6T2), // V4T. T(V6T2), // V5T. T(V6T2), // V5TE. T(V6T2), // V5TEJ. T(V6T2), // V6. T(V7), // V6KZ. T(V6T2) // V6T2. }; static const int v6k[] = { T(V6K), // PRE_V4. T(V6K), // V4. T(V6K), // V4T. T(V6K), // V5T. T(V6K), // V5TE. T(V6K), // V5TEJ. T(V6K), // V6. T(V6KZ), // V6KZ. T(V7), // V6T2. T(V6K) // V6K. }; static const int v7[] = { T(V7), // PRE_V4. T(V7), // V4. T(V7), // V4T. T(V7), // V5T. T(V7), // V5TE. T(V7), // V5TEJ. T(V7), // V6. T(V7), // V6KZ. T(V7), // V6T2. T(V7), // V6K. T(V7) // V7. }; static const int v6_m[] = { -1, // PRE_V4. -1, // V4. T(V6K), // V4T. T(V6K), // V5T. T(V6K), // V5TE. T(V6K), // V5TEJ. T(V6K), // V6. T(V6KZ), // V6KZ. T(V7), // V6T2. T(V6K), // V6K. T(V7), // V7. T(V6_M) // V6_M. }; static const int v6s_m[] = { -1, // PRE_V4. -1, // V4. T(V6K), // V4T. T(V6K), // V5T. T(V6K), // V5TE. T(V6K), // V5TEJ. T(V6K), // V6. T(V6KZ), // V6KZ. T(V7), // V6T2. T(V6K), // V6K. T(V7), // V7. T(V6S_M), // V6_M. T(V6S_M) // V6S_M. }; static const int v7e_m[] = { -1, // PRE_V4. -1, // V4. T(V7E_M), // V4T. T(V7E_M), // V5T. T(V7E_M), // V5TE. T(V7E_M), // V5TEJ. T(V7E_M), // V6. T(V7E_M), // V6KZ. T(V7E_M), // V6T2. T(V7E_M), // V6K. T(V7E_M), // V7. T(V7E_M), // V6_M. T(V7E_M), // V6S_M. T(V7E_M) // V7E_M. }; static const int v4t_plus_v6_m[] = { -1, // PRE_V4. -1, // V4. T(V4T), // V4T. T(V5T), // V5T. T(V5TE), // V5TE. T(V5TEJ), // V5TEJ. T(V6), // V6. T(V6KZ), // V6KZ. T(V6T2), // V6T2. T(V6K), // V6K. T(V7), // V7. T(V6_M), // V6_M. T(V6S_M), // V6S_M. T(V7E_M), // V7E_M. T(V4T_PLUS_V6_M) // V4T plus V6_M. }; static const int* comb[] = { v6t2, v6k, v7, v6_m, v6s_m, v7e_m, // Pseudo-architecture. v4t_plus_v6_m }; // Check we've not got a higher architecture than we know about. if (oldtag > elfcpp::MAX_TAG_CPU_ARCH || newtag > elfcpp::MAX_TAG_CPU_ARCH) { gold_error(_("%s: unknown CPU architecture"), name); return -1; } // Override old tag if we have a Tag_also_compatible_with on the output. if ((oldtag == T(V6_M) && *secondary_compat_out == T(V4T)) || (oldtag == T(V4T) && *secondary_compat_out == T(V6_M))) oldtag = T(V4T_PLUS_V6_M); // And override the new tag if we have a Tag_also_compatible_with on the // input. if ((newtag == T(V6_M) && secondary_compat == T(V4T)) || (newtag == T(V4T) && secondary_compat == T(V6_M))) newtag = T(V4T_PLUS_V6_M); // Architectures before V6KZ add features monotonically. int tagh = std::max(oldtag, newtag); if (tagh <= elfcpp::TAG_CPU_ARCH_V6KZ) return tagh; int tagl = std::min(oldtag, newtag); int result = comb[tagh - T(V6T2)][tagl]; // Use Tag_CPU_arch == V4T and Tag_also_compatible_with (Tag_CPU_arch V6_M) // as the canonical version. if (result == T(V4T_PLUS_V6_M)) { result = T(V4T); *secondary_compat_out = T(V6_M); } else *secondary_compat_out = -1; if (result == -1) { gold_error(_("%s: conflicting CPU architectures %d/%d"), name, oldtag, newtag); return -1; } return result; #undef T } // Helper to print AEABI enum tag value. template<bool big_endian> std::string Target_arm<big_endian>::aeabi_enum_name(unsigned int value) { static const char* aeabi_enum_names[] = { "", "variable-size", "32-bit", "" }; const size_t aeabi_enum_names_size = sizeof(aeabi_enum_names) / sizeof(aeabi_enum_names[0]); if (value < aeabi_enum_names_size) return std::string(aeabi_enum_names[value]); else { char buffer[100]; sprintf(buffer, "<unknown value %u>", value); return std::string(buffer); } } // Return the string value to store in TAG_CPU_name. template<bool big_endian> std::string Target_arm<big_endian>::tag_cpu_name_value(unsigned int value) { static const char* name_table[] = { // These aren't real CPU names, but we can't guess // that from the architecture version alone. "Pre v4", "ARM v4", "ARM v4T", "ARM v5T", "ARM v5TE", "ARM v5TEJ", "ARM v6", "ARM v6KZ", "ARM v6T2", "ARM v6K", "ARM v7", "ARM v6-M", "ARM v6S-M", "ARM v7E-M" }; const size_t name_table_size = sizeof(name_table) / sizeof(name_table[0]); if (value < name_table_size) return std::string(name_table[value]); else { char buffer[100]; sprintf(buffer, "<unknown CPU value %u>", value); return std::string(buffer); } } // Merge object attributes from input file called NAME with those of the // output. The input object attributes are in the object pointed by PASD. template<bool big_endian> void Target_arm<big_endian>::merge_object_attributes( const char* name, const Attributes_section_data* pasd) { // Return if there is no attributes section data. if (pasd == NULL) return; // If output has no object attributes, just copy. const int vendor = Object_attribute::OBJ_ATTR_PROC; if (this->attributes_section_data_ == NULL) { this->attributes_section_data_ = new Attributes_section_data(*pasd); Object_attribute* out_attr = this->attributes_section_data_->known_attributes(vendor); // We do not output objects with Tag_MPextension_use_legacy - we move // the attribute's value to Tag_MPextension_use. */ if (out_attr[elfcpp::Tag_MPextension_use_legacy].int_value() != 0) { if (out_attr[elfcpp::Tag_MPextension_use].int_value() != 0 && out_attr[elfcpp::Tag_MPextension_use_legacy].int_value() != out_attr[elfcpp::Tag_MPextension_use].int_value()) { gold_error(_("%s has both the current and legacy " "Tag_MPextension_use attributes"), name); } out_attr[elfcpp::Tag_MPextension_use] = out_attr[elfcpp::Tag_MPextension_use_legacy]; out_attr[elfcpp::Tag_MPextension_use_legacy].set_type(0); out_attr[elfcpp::Tag_MPextension_use_legacy].set_int_value(0); } return; } const Object_attribute* in_attr = pasd->known_attributes(vendor); Object_attribute* out_attr = this->attributes_section_data_->known_attributes(vendor); // This needs to happen before Tag_ABI_FP_number_model is merged. */ if (in_attr[elfcpp::Tag_ABI_VFP_args].int_value() != out_attr[elfcpp::Tag_ABI_VFP_args].int_value()) { // Ignore mismatches if the object doesn't use floating point. */ if (out_attr[elfcpp::Tag_ABI_FP_number_model].int_value() == 0) out_attr[elfcpp::Tag_ABI_VFP_args].set_int_value( in_attr[elfcpp::Tag_ABI_VFP_args].int_value()); else if (in_attr[elfcpp::Tag_ABI_FP_number_model].int_value() != 0 && parameters->options().warn_mismatch()) gold_error(_("%s uses VFP register arguments, output does not"), name); } for (int i = 4; i < Vendor_object_attributes::NUM_KNOWN_ATTRIBUTES; ++i) { // Merge this attribute with existing attributes. switch (i) { case elfcpp::Tag_CPU_raw_name: case elfcpp::Tag_CPU_name: // These are merged after Tag_CPU_arch. break; case elfcpp::Tag_ABI_optimization_goals: case elfcpp::Tag_ABI_FP_optimization_goals: // Use the first value seen. break; case elfcpp::Tag_CPU_arch: { unsigned int saved_out_attr = out_attr->int_value(); // Merge Tag_CPU_arch and Tag_also_compatible_with. int secondary_compat = this->get_secondary_compatible_arch(pasd); int secondary_compat_out = this->get_secondary_compatible_arch( this->attributes_section_data_); out_attr[i].set_int_value( tag_cpu_arch_combine(name, out_attr[i].int_value(), &secondary_compat_out, in_attr[i].int_value(), secondary_compat)); this->set_secondary_compatible_arch(this->attributes_section_data_, secondary_compat_out); // Merge Tag_CPU_name and Tag_CPU_raw_name. if (out_attr[i].int_value() == saved_out_attr) ; // Leave the names alone. else if (out_attr[i].int_value() == in_attr[i].int_value()) { // The output architecture has been changed to match the // input architecture. Use the input names. out_attr[elfcpp::Tag_CPU_name].set_string_value( in_attr[elfcpp::Tag_CPU_name].string_value()); out_attr[elfcpp::Tag_CPU_raw_name].set_string_value( in_attr[elfcpp::Tag_CPU_raw_name].string_value()); } else { out_attr[elfcpp::Tag_CPU_name].set_string_value(""); out_attr[elfcpp::Tag_CPU_raw_name].set_string_value(""); } // If we still don't have a value for Tag_CPU_name, // make one up now. Tag_CPU_raw_name remains blank. if (out_attr[elfcpp::Tag_CPU_name].string_value() == "") { const std::string cpu_name = this->tag_cpu_name_value(out_attr[i].int_value()); // FIXME: If we see an unknown CPU, this will be set // to "<unknown CPU n>", where n is the attribute value. // This is different from BFD, which leaves the name alone. out_attr[elfcpp::Tag_CPU_name].set_string_value(cpu_name); } } break; case elfcpp::Tag_ARM_ISA_use: case elfcpp::Tag_THUMB_ISA_use: case elfcpp::Tag_WMMX_arch: case elfcpp::Tag_Advanced_SIMD_arch: // ??? Do Advanced_SIMD (NEON) and WMMX conflict? case elfcpp::Tag_ABI_FP_rounding: case elfcpp::Tag_ABI_FP_exceptions: case elfcpp::Tag_ABI_FP_user_exceptions: case elfcpp::Tag_ABI_FP_number_model: case elfcpp::Tag_VFP_HP_extension: case elfcpp::Tag_CPU_unaligned_access: case elfcpp::Tag_T2EE_use: case elfcpp::Tag_Virtualization_use: case elfcpp::Tag_MPextension_use: // Use the largest value specified. if (in_attr[i].int_value() > out_attr[i].int_value()) out_attr[i].set_int_value(in_attr[i].int_value()); break; case elfcpp::Tag_ABI_align8_preserved: case elfcpp::Tag_ABI_PCS_RO_data: // Use the smallest value specified. if (in_attr[i].int_value() < out_attr[i].int_value()) out_attr[i].set_int_value(in_attr[i].int_value()); break; case elfcpp::Tag_ABI_align8_needed: if ((in_attr[i].int_value() > 0 || out_attr[i].int_value() > 0) && (in_attr[elfcpp::Tag_ABI_align8_preserved].int_value() == 0 || (out_attr[elfcpp::Tag_ABI_align8_preserved].int_value() == 0))) { // This error message should be enabled once all non-conforming // binaries in the toolchain have had the attributes set // properly. // gold_error(_("output 8-byte data alignment conflicts with %s"), // name); } // Fall through. case elfcpp::Tag_ABI_FP_denormal: case elfcpp::Tag_ABI_PCS_GOT_use: { // These tags have 0 = don't care, 1 = strong requirement, // 2 = weak requirement. static const int order_021[3] = {0, 2, 1}; // Use the "greatest" from the sequence 0, 2, 1, or the largest // value if greater than 2 (for future-proofing). if ((in_attr[i].int_value() > 2 && in_attr[i].int_value() > out_attr[i].int_value()) || (in_attr[i].int_value() <= 2 && out_attr[i].int_value() <= 2 && (order_021[in_attr[i].int_value()] > order_021[out_attr[i].int_value()]))) out_attr[i].set_int_value(in_attr[i].int_value()); } break; case elfcpp::Tag_CPU_arch_profile: if (out_attr[i].int_value() != in_attr[i].int_value()) { // 0 will merge with anything. // 'A' and 'S' merge to 'A'. // 'R' and 'S' merge to 'R'. // 'M' and 'A|R|S' is an error. if (out_attr[i].int_value() == 0 || (out_attr[i].int_value() == 'S' && (in_attr[i].int_value() == 'A' || in_attr[i].int_value() == 'R'))) out_attr[i].set_int_value(in_attr[i].int_value()); else if (in_attr[i].int_value() == 0 || (in_attr[i].int_value() == 'S' && (out_attr[i].int_value() == 'A' || out_attr[i].int_value() == 'R'))) ; // Do nothing. else if (parameters->options().warn_mismatch()) { gold_error (_("conflicting architecture profiles %c/%c"), in_attr[i].int_value() ? in_attr[i].int_value() : '0', out_attr[i].int_value() ? out_attr[i].int_value() : '0'); } } break; case elfcpp::Tag_VFP_arch: { static const struct { int ver; int regs; } vfp_versions[7] = { {0, 0}, {1, 16}, {2, 16}, {3, 32}, {3, 16}, {4, 32}, {4, 16} }; // Values greater than 6 aren't defined, so just pick the // biggest. if (in_attr[i].int_value() > 6 && in_attr[i].int_value() > out_attr[i].int_value()) { *out_attr = *in_attr; break; } // The output uses the superset of input features // (ISA version) and registers. int ver = std::max(vfp_versions[in_attr[i].int_value()].ver, vfp_versions[out_attr[i].int_value()].ver); int regs = std::max(vfp_versions[in_attr[i].int_value()].regs, vfp_versions[out_attr[i].int_value()].regs); // This assumes all possible supersets are also a valid // options. int newval; for (newval = 6; newval > 0; newval--) { if (regs == vfp_versions[newval].regs && ver == vfp_versions[newval].ver) break; } out_attr[i].set_int_value(newval); } break; case elfcpp::Tag_PCS_config: if (out_attr[i].int_value() == 0) out_attr[i].set_int_value(in_attr[i].int_value()); else if (in_attr[i].int_value() != 0 && out_attr[i].int_value() != 0 && parameters->options().warn_mismatch()) { // It's sometimes ok to mix different configs, so this is only // a warning. gold_warning(_("%s: conflicting platform configuration"), name); } break; case elfcpp::Tag_ABI_PCS_R9_use: if (in_attr[i].int_value() != out_attr[i].int_value() && out_attr[i].int_value() != elfcpp::AEABI_R9_unused && in_attr[i].int_value() != elfcpp::AEABI_R9_unused && parameters->options().warn_mismatch()) { gold_error(_("%s: conflicting use of R9"), name); } if (out_attr[i].int_value() == elfcpp::AEABI_R9_unused) out_attr[i].set_int_value(in_attr[i].int_value()); break; case elfcpp::Tag_ABI_PCS_RW_data: if (in_attr[i].int_value() == elfcpp::AEABI_PCS_RW_data_SBrel && (in_attr[elfcpp::Tag_ABI_PCS_R9_use].int_value() != elfcpp::AEABI_R9_SB) && (out_attr[elfcpp::Tag_ABI_PCS_R9_use].int_value() != elfcpp::AEABI_R9_unused) && parameters->options().warn_mismatch()) { gold_error(_("%s: SB relative addressing conflicts with use " "of R9"), name); } // Use the smallest value specified. if (in_attr[i].int_value() < out_attr[i].int_value()) out_attr[i].set_int_value(in_attr[i].int_value()); break; case elfcpp::Tag_ABI_PCS_wchar_t: if (out_attr[i].int_value() && in_attr[i].int_value() && out_attr[i].int_value() != in_attr[i].int_value() && parameters->options().warn_mismatch() && parameters->options().wchar_size_warning()) { gold_warning(_("%s uses %u-byte wchar_t yet the output is to " "use %u-byte wchar_t; use of wchar_t values " "across objects may fail"), name, in_attr[i].int_value(), out_attr[i].int_value()); } else if (in_attr[i].int_value() && !out_attr[i].int_value()) out_attr[i].set_int_value(in_attr[i].int_value()); break; case elfcpp::Tag_ABI_enum_size: if (in_attr[i].int_value() != elfcpp::AEABI_enum_unused) { if (out_attr[i].int_value() == elfcpp::AEABI_enum_unused || out_attr[i].int_value() == elfcpp::AEABI_enum_forced_wide) { // The existing object is compatible with anything. // Use whatever requirements the new object has. out_attr[i].set_int_value(in_attr[i].int_value()); } else if (in_attr[i].int_value() != elfcpp::AEABI_enum_forced_wide && out_attr[i].int_value() != in_attr[i].int_value() && parameters->options().warn_mismatch() && parameters->options().enum_size_warning()) { unsigned int in_value = in_attr[i].int_value(); unsigned int out_value = out_attr[i].int_value(); gold_warning(_("%s uses %s enums yet the output is to use " "%s enums; use of enum values across objects " "may fail"), name, this->aeabi_enum_name(in_value).c_str(), this->aeabi_enum_name(out_value).c_str()); } } break; case elfcpp::Tag_ABI_VFP_args: // Already done. break; case elfcpp::Tag_ABI_WMMX_args: if (in_attr[i].int_value() != out_attr[i].int_value() && parameters->options().warn_mismatch()) { gold_error(_("%s uses iWMMXt register arguments, output does " "not"), name); } break; case Object_attribute::Tag_compatibility: // Merged in target-independent code. break; case elfcpp::Tag_ABI_HardFP_use: // 1 (SP) and 2 (DP) conflict, so combine to 3 (SP & DP). if ((in_attr[i].int_value() == 1 && out_attr[i].int_value() == 2) || (in_attr[i].int_value() == 2 && out_attr[i].int_value() == 1)) out_attr[i].set_int_value(3); else if (in_attr[i].int_value() > out_attr[i].int_value()) out_attr[i].set_int_value(in_attr[i].int_value()); break; case elfcpp::Tag_ABI_FP_16bit_format: if (in_attr[i].int_value() != 0 && out_attr[i].int_value() != 0) { if (in_attr[i].int_value() != out_attr[i].int_value() && parameters->options().warn_mismatch()) gold_error(_("fp16 format mismatch between %s and output"), name); } if (in_attr[i].int_value() != 0) out_attr[i].set_int_value(in_attr[i].int_value()); break; case elfcpp::Tag_DIV_use: // This tag is set to zero if we can use UDIV and SDIV in Thumb // mode on a v7-M or v7-R CPU; to one if we can not use UDIV or // SDIV at all; and to two if we can use UDIV or SDIV on a v7-A // CPU. We will merge as follows: If the input attribute's value // is one then the output attribute's value remains unchanged. If // the input attribute's value is zero or two then if the output // attribute's value is one the output value is set to the input // value, otherwise the output value must be the same as the // inputs. */ if (in_attr[i].int_value() != 1 && out_attr[i].int_value() != 1) { if (in_attr[i].int_value() != out_attr[i].int_value()) { gold_error(_("DIV usage mismatch between %s and output"), name); } } if (in_attr[i].int_value() != 1) out_attr[i].set_int_value(in_attr[i].int_value()); break; case elfcpp::Tag_MPextension_use_legacy: // We don't output objects with Tag_MPextension_use_legacy - we // move the value to Tag_MPextension_use. if (in_attr[i].int_value() != 0 && in_attr[elfcpp::Tag_MPextension_use].int_value() != 0) { if (in_attr[elfcpp::Tag_MPextension_use].int_value() != in_attr[i].int_value()) { gold_error(_("%s has has both the current and legacy " "Tag_MPextension_use attributes"), name); } } if (in_attr[i].int_value() > out_attr[elfcpp::Tag_MPextension_use].int_value()) out_attr[elfcpp::Tag_MPextension_use] = in_attr[i]; break; case elfcpp::Tag_nodefaults: // This tag is set if it exists, but the value is unused (and is // typically zero). We don't actually need to do anything here - // the merge happens automatically when the type flags are merged // below. break; case elfcpp::Tag_also_compatible_with: // Already done in Tag_CPU_arch. break; case elfcpp::Tag_conformance: // Keep the attribute if it matches. Throw it away otherwise. // No attribute means no claim to conform. if (in_attr[i].string_value() != out_attr[i].string_value()) out_attr[i].set_string_value(""); break; default: { const char* err_object = NULL; // The "known_obj_attributes" table does contain some undefined // attributes. Ensure that there are unused. if (out_attr[i].int_value() != 0 || out_attr[i].string_value() != "") err_object = "output"; else if (in_attr[i].int_value() != 0 || in_attr[i].string_value() != "") err_object = name; if (err_object != NULL && parameters->options().warn_mismatch()) { // Attribute numbers >=64 (mod 128) can be safely ignored. if ((i & 127) < 64) gold_error(_("%s: unknown mandatory EABI object attribute " "%d"), err_object, i); else gold_warning(_("%s: unknown EABI object attribute %d"), err_object, i); } // Only pass on attributes that match in both inputs. if (!in_attr[i].matches(out_attr[i])) { out_attr[i].set_int_value(0); out_attr[i].set_string_value(""); } } } // If out_attr was copied from in_attr then it won't have a type yet. if (in_attr[i].type() && !out_attr[i].type()) out_attr[i].set_type(in_attr[i].type()); } // Merge Tag_compatibility attributes and any common GNU ones. this->attributes_section_data_->merge(name, pasd); // Check for any attributes not known on ARM. typedef Vendor_object_attributes::Other_attributes Other_attributes; const Other_attributes* in_other_attributes = pasd->other_attributes(vendor); Other_attributes::const_iterator in_iter = in_other_attributes->begin(); Other_attributes* out_other_attributes = this->attributes_section_data_->other_attributes(vendor); Other_attributes::iterator out_iter = out_other_attributes->begin(); while (in_iter != in_other_attributes->end() || out_iter != out_other_attributes->end()) { const char* err_object = NULL; int err_tag = 0; // The tags for each list are in numerical order. // If the tags are equal, then merge. if (out_iter != out_other_attributes->end() && (in_iter == in_other_attributes->end() || in_iter->first > out_iter->first)) { // This attribute only exists in output. We can't merge, and we // don't know what the tag means, so delete it. err_object = "output"; err_tag = out_iter->first; int saved_tag = out_iter->first; delete out_iter->second; out_other_attributes->erase(out_iter); out_iter = out_other_attributes->upper_bound(saved_tag); } else if (in_iter != in_other_attributes->end() && (out_iter != out_other_attributes->end() || in_iter->first < out_iter->first)) { // This attribute only exists in input. We can't merge, and we // don't know what the tag means, so ignore it. err_object = name; err_tag = in_iter->first; ++in_iter; } else // The tags are equal. { // As present, all attributes in the list are unknown, and // therefore can't be merged meaningfully. err_object = "output"; err_tag = out_iter->first; // Only pass on attributes that match in both inputs. if (!in_iter->second->matches(*(out_iter->second))) { // No match. Delete the attribute. int saved_tag = out_iter->first; delete out_iter->second; out_other_attributes->erase(out_iter); out_iter = out_other_attributes->upper_bound(saved_tag); } else { // Matched. Keep the attribute and move to the next. ++out_iter; ++in_iter; } } if (err_object && parameters->options().warn_mismatch()) { // Attribute numbers >=64 (mod 128) can be safely ignored. */ if ((err_tag & 127) < 64) { gold_error(_("%s: unknown mandatory EABI object attribute %d"), err_object, err_tag); } else { gold_warning(_("%s: unknown EABI object attribute %d"), err_object, err_tag); } } } } // Stub-generation methods for Target_arm. // Make a new Arm_input_section object. template<bool big_endian> Arm_input_section<big_endian>* Target_arm<big_endian>::new_arm_input_section( Relobj* relobj, unsigned int shndx) { Section_id sid(relobj, shndx); Arm_input_section<big_endian>* arm_input_section = new Arm_input_section<big_endian>(relobj, shndx); arm_input_section->init(); // Register new Arm_input_section in map for look-up. std::pair<typename Arm_input_section_map::iterator, bool> ins = this->arm_input_section_map_.insert(std::make_pair(sid, arm_input_section)); // Make sure that it we have not created another Arm_input_section // for this input section already. gold_assert(ins.second); return arm_input_section; } // Find the Arm_input_section object corresponding to the SHNDX-th input // section of RELOBJ. template<bool big_endian> Arm_input_section<big_endian>* Target_arm<big_endian>::find_arm_input_section( Relobj* relobj, unsigned int shndx) const { Section_id sid(relobj, shndx); typename Arm_input_section_map::const_iterator p = this->arm_input_section_map_.find(sid); return (p != this->arm_input_section_map_.end()) ? p->second : NULL; } // Make a new stub table. template<bool big_endian> Stub_table<big_endian>* Target_arm<big_endian>::new_stub_table(Arm_input_section<big_endian>* owner) { Stub_table<big_endian>* stub_table = new Stub_table<big_endian>(owner); this->stub_tables_.push_back(stub_table); stub_table->set_address(owner->address() + owner->data_size()); stub_table->set_file_offset(owner->offset() + owner->data_size()); stub_table->finalize_data_size(); return stub_table; } // Scan a relocation for stub generation. template<bool big_endian> void Target_arm<big_endian>::scan_reloc_for_stub( const Relocate_info<32, big_endian>* relinfo, unsigned int r_type, const Sized_symbol<32>* gsym, unsigned int r_sym, const Symbol_value<32>* psymval, elfcpp::Elf_types<32>::Elf_Swxword addend, Arm_address address) { typedef typename Target_arm<big_endian>::Relocate Relocate; const Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(relinfo->object); bool target_is_thumb; Symbol_value<32> symval; if (gsym != NULL) { // This is a global symbol. Determine if we use PLT and if the // final target is THUMB. if (gsym->use_plt_offset(Scan::get_reference_flags(r_type))) { // This uses a PLT, change the symbol value. symval.set_output_value(this->plt_section()->address() + gsym->plt_offset()); psymval = &symval; target_is_thumb = false; } else if (gsym->is_undefined()) // There is no need to generate a stub symbol is undefined. return; else { target_is_thumb = ((gsym->type() == elfcpp::STT_ARM_TFUNC) || (gsym->type() == elfcpp::STT_FUNC && !gsym->is_undefined() && ((psymval->value(arm_relobj, 0) & 1) != 0))); } } else { // This is a local symbol. Determine if the final target is THUMB. target_is_thumb = arm_relobj->local_symbol_is_thumb_function(r_sym); } // Strip LSB if this points to a THUMB target. const Arm_reloc_property* reloc_property = arm_reloc_property_table->get_implemented_static_reloc_property(r_type); gold_assert(reloc_property != NULL); if (target_is_thumb && reloc_property->uses_thumb_bit() && ((psymval->value(arm_relobj, 0) & 1) != 0)) { Arm_address stripped_value = psymval->value(arm_relobj, 0) & ~static_cast<Arm_address>(1); symval.set_output_value(stripped_value); psymval = &symval; } // Get the symbol value. Symbol_value<32>::Value value = psymval->value(arm_relobj, 0); // Owing to pipelining, the PC relative branches below actually skip // two instructions when the branch offset is 0. Arm_address destination; switch (r_type) { case elfcpp::R_ARM_CALL: case elfcpp::R_ARM_JUMP24: case elfcpp::R_ARM_PLT32: // ARM branches. destination = value + addend + 8; break; case elfcpp::R_ARM_THM_CALL: case elfcpp::R_ARM_THM_XPC22: case elfcpp::R_ARM_THM_JUMP24: case elfcpp::R_ARM_THM_JUMP19: // THUMB branches. destination = value + addend + 4; break; default: gold_unreachable(); } Reloc_stub* stub = NULL; Stub_type stub_type = Reloc_stub::stub_type_for_reloc(r_type, address, destination, target_is_thumb); if (stub_type != arm_stub_none) { // Try looking up an existing stub from a stub table. Stub_table<big_endian>* stub_table = arm_relobj->stub_table(relinfo->data_shndx); gold_assert(stub_table != NULL); // Locate stub by destination. Reloc_stub::Key stub_key(stub_type, gsym, arm_relobj, r_sym, addend); // Create a stub if there is not one already stub = stub_table->find_reloc_stub(stub_key); if (stub == NULL) { // create a new stub and add it to stub table. stub = this->stub_factory().make_reloc_stub(stub_type); stub_table->add_reloc_stub(stub, stub_key); } // Record the destination address. stub->set_destination_address(destination | (target_is_thumb ? 1 : 0)); } // For Cortex-A8, we need to record a relocation at 4K page boundary. if (this->fix_cortex_a8_ && (r_type == elfcpp::R_ARM_THM_JUMP24 || r_type == elfcpp::R_ARM_THM_JUMP19 || r_type == elfcpp::R_ARM_THM_CALL || r_type == elfcpp::R_ARM_THM_XPC22) && (address & 0xfffU) == 0xffeU) { // Found a candidate. Note we haven't checked the destination is // within 4K here: if we do so (and don't create a record) we can't // tell that a branch should have been relocated when scanning later. this->cortex_a8_relocs_info_[address] = new Cortex_a8_reloc(stub, r_type, destination | (target_is_thumb ? 1 : 0)); } } // This function scans a relocation sections for stub generation. // The template parameter Relocate must be a class type which provides // a single function, relocate(), which implements the machine // specific part of a relocation. // BIG_ENDIAN is the endianness of the data. SH_TYPE is the section type: // SHT_REL or SHT_RELA. // PRELOCS points to the relocation data. RELOC_COUNT is the number // of relocs. OUTPUT_SECTION is the output section. // NEEDS_SPECIAL_OFFSET_HANDLING is true if input offsets need to be // mapped to output offsets. // VIEW is the section data, VIEW_ADDRESS is its memory address, and // VIEW_SIZE is the size. These refer to the input section, unless // NEEDS_SPECIAL_OFFSET_HANDLING is true, in which case they refer to // the output section. template<bool big_endian> template<int sh_type> void inline Target_arm<big_endian>::scan_reloc_section_for_stubs( const Relocate_info<32, big_endian>* relinfo, const unsigned char* prelocs, size_t reloc_count, Output_section* output_section, bool needs_special_offset_handling, const unsigned char* view, elfcpp::Elf_types<32>::Elf_Addr view_address, section_size_type) { typedef typename Reloc_types<sh_type, 32, big_endian>::Reloc Reltype; const int reloc_size = Reloc_types<sh_type, 32, big_endian>::reloc_size; Arm_relobj<big_endian>* arm_object = Arm_relobj<big_endian>::as_arm_relobj(relinfo->object); unsigned int local_count = arm_object->local_symbol_count(); Comdat_behavior comdat_behavior = CB_UNDETERMINED; for (size_t i = 0; i < reloc_count; ++i, prelocs += reloc_size) { Reltype reloc(prelocs); typename elfcpp::Elf_types<32>::Elf_WXword r_info = reloc.get_r_info(); unsigned int r_sym = elfcpp::elf_r_sym<32>(r_info); unsigned int r_type = elfcpp::elf_r_type<32>(r_info); r_type = this->get_real_reloc_type(r_type); // Only a few relocation types need stubs. if ((r_type != elfcpp::R_ARM_CALL) && (r_type != elfcpp::R_ARM_JUMP24) && (r_type != elfcpp::R_ARM_PLT32) && (r_type != elfcpp::R_ARM_THM_CALL) && (r_type != elfcpp::R_ARM_THM_XPC22) && (r_type != elfcpp::R_ARM_THM_JUMP24) && (r_type != elfcpp::R_ARM_THM_JUMP19) && (r_type != elfcpp::R_ARM_V4BX)) continue; section_offset_type offset = convert_to_section_size_type(reloc.get_r_offset()); if (needs_special_offset_handling) { offset = output_section->output_offset(relinfo->object, relinfo->data_shndx, offset); if (offset == -1) continue; } // Create a v4bx stub if --fix-v4bx-interworking is used. if (r_type == elfcpp::R_ARM_V4BX) { if (this->fix_v4bx() == General_options::FIX_V4BX_INTERWORKING) { // Get the BX instruction. typedef typename elfcpp::Swap<32, big_endian>::Valtype Valtype; const Valtype* wv = reinterpret_cast<const Valtype*>(view + offset); elfcpp::Elf_types<32>::Elf_Swxword insn = elfcpp::Swap<32, big_endian>::readval(wv); const uint32_t reg = (insn & 0xf); if (reg < 0xf) { // Try looking up an existing stub from a stub table. Stub_table<big_endian>* stub_table = arm_object->stub_table(relinfo->data_shndx); gold_assert(stub_table != NULL); if (stub_table->find_arm_v4bx_stub(reg) == NULL) { // create a new stub and add it to stub table. Arm_v4bx_stub* stub = this->stub_factory().make_arm_v4bx_stub(reg); gold_assert(stub != NULL); stub_table->add_arm_v4bx_stub(stub); } } } continue; } // Get the addend. Stub_addend_reader<sh_type, big_endian> stub_addend_reader; elfcpp::Elf_types<32>::Elf_Swxword addend = stub_addend_reader(r_type, view + offset, reloc); const Sized_symbol<32>* sym; Symbol_value<32> symval; const Symbol_value<32> *psymval; bool is_defined_in_discarded_section; unsigned int shndx; if (r_sym < local_count) { sym = NULL; psymval = arm_object->local_symbol(r_sym); // If the local symbol belongs to a section we are discarding, // and that section is a debug section, try to find the // corresponding kept section and map this symbol to its // counterpart in the kept section. The symbol must not // correspond to a section we are folding. bool is_ordinary; shndx = psymval->input_shndx(&is_ordinary); is_defined_in_discarded_section = (is_ordinary && shndx != elfcpp::SHN_UNDEF && !arm_object->is_section_included(shndx) && !relinfo->symtab->is_section_folded(arm_object, shndx)); // We need to compute the would-be final value of this local // symbol. if (!is_defined_in_discarded_section) { typedef Sized_relobj_file<32, big_endian> ObjType; typename ObjType::Compute_final_local_value_status status = arm_object->compute_final_local_value(r_sym, psymval, &symval, relinfo->symtab); if (status == ObjType::CFLV_OK) { // Currently we cannot handle a branch to a target in // a merged section. If this is the case, issue an error // and also free the merge symbol value. if (!symval.has_output_value()) { const std::string& section_name = arm_object->section_name(shndx); arm_object->error(_("cannot handle branch to local %u " "in a merged section %s"), r_sym, section_name.c_str()); } psymval = &symval; } else { // We cannot determine the final value. continue; } } } else { const Symbol* gsym; gsym = arm_object->global_symbol(r_sym); gold_assert(gsym != NULL); if (gsym->is_forwarder()) gsym = relinfo->symtab->resolve_forwards(gsym); sym = static_cast<const Sized_symbol<32>*>(gsym); if (sym->has_symtab_index() && sym->symtab_index() != -1U) symval.set_output_symtab_index(sym->symtab_index()); else symval.set_no_output_symtab_entry(); // We need to compute the would-be final value of this global // symbol. const Symbol_table* symtab = relinfo->symtab; const Sized_symbol<32>* sized_symbol = symtab->get_sized_symbol<32>(gsym); Symbol_table::Compute_final_value_status status; Arm_address value = symtab->compute_final_value<32>(sized_symbol, &status); // Skip this if the symbol has not output section. if (status == Symbol_table::CFVS_NO_OUTPUT_SECTION) continue; symval.set_output_value(value); if (gsym->type() == elfcpp::STT_TLS) symval.set_is_tls_symbol(); else if (gsym->type() == elfcpp::STT_GNU_IFUNC) symval.set_is_ifunc_symbol(); psymval = &symval; is_defined_in_discarded_section = (gsym->is_defined_in_discarded_section() && gsym->is_undefined()); shndx = 0; } Symbol_value<32> symval2; if (is_defined_in_discarded_section) { if (comdat_behavior == CB_UNDETERMINED) { std::string name = arm_object->section_name(relinfo->data_shndx); comdat_behavior = get_comdat_behavior(name.c_str()); } if (comdat_behavior == CB_PRETEND) { // FIXME: This case does not work for global symbols. // We have no place to store the original section index. // Fortunately this does not matter for comdat sections, // only for sections explicitly discarded by a linker // script. bool found; typename elfcpp::Elf_types<32>::Elf_Addr value = arm_object->map_to_kept_section(shndx, &found); if (found) symval2.set_output_value(value + psymval->input_value()); else symval2.set_output_value(0); } else { if (comdat_behavior == CB_WARNING) gold_warning_at_location(relinfo, i, offset, _("relocation refers to discarded " "section")); symval2.set_output_value(0); } symval2.set_no_output_symtab_entry(); psymval = &symval2; } // If symbol is a section symbol, we don't know the actual type of // destination. Give up. if (psymval->is_section_symbol()) continue; this->scan_reloc_for_stub(relinfo, r_type, sym, r_sym, psymval, addend, view_address + offset); } } // Scan an input section for stub generation. template<bool big_endian> void Target_arm<big_endian>::scan_section_for_stubs( const Relocate_info<32, big_endian>* relinfo, unsigned int sh_type, const unsigned char* prelocs, size_t reloc_count, Output_section* output_section, bool needs_special_offset_handling, const unsigned char* view, Arm_address view_address, section_size_type view_size) { if (sh_type == elfcpp::SHT_REL) this->scan_reloc_section_for_stubs<elfcpp::SHT_REL>( relinfo, prelocs, reloc_count, output_section, needs_special_offset_handling, view, view_address, view_size); else if (sh_type == elfcpp::SHT_RELA) // We do not support RELA type relocations yet. This is provided for // completeness. this->scan_reloc_section_for_stubs<elfcpp::SHT_RELA>( relinfo, prelocs, reloc_count, output_section, needs_special_offset_handling, view, view_address, view_size); else gold_unreachable(); } // Group input sections for stub generation. // // We group input sections in an output section so that the total size, // including any padding space due to alignment is smaller than GROUP_SIZE // unless the only input section in group is bigger than GROUP_SIZE already. // Then an ARM stub table is created to follow the last input section // in group. For each group an ARM stub table is created an is placed // after the last group. If STUB_ALWAYS_AFTER_BRANCH is false, we further // extend the group after the stub table. template<bool big_endian> void Target_arm<big_endian>::group_sections( Layout* layout, section_size_type group_size, bool stubs_always_after_branch, const Task* task) { // Group input sections and insert stub table Layout::Section_list section_list; layout->get_allocated_sections(§ion_list); for (Layout::Section_list::const_iterator p = section_list.begin(); p != section_list.end(); ++p) { Arm_output_section<big_endian>* output_section = Arm_output_section<big_endian>::as_arm_output_section(*p); output_section->group_sections(group_size, stubs_always_after_branch, this, task); } } // Relaxation hook. This is where we do stub generation. template<bool big_endian> bool Target_arm<big_endian>::do_relax( int pass, const Input_objects* input_objects, Symbol_table* symtab, Layout* layout, const Task* task) { // No need to generate stubs if this is a relocatable link. gold_assert(!parameters->options().relocatable()); // If this is the first pass, we need to group input sections into // stub groups. bool done_exidx_fixup = false; typedef typename Stub_table_list::iterator Stub_table_iterator; if (pass == 1) { // Determine the stub group size. The group size is the absolute // value of the parameter --stub-group-size. If --stub-group-size // is passed a negative value, we restrict stubs to be always after // the stubbed branches. int32_t stub_group_size_param = parameters->options().stub_group_size(); bool stubs_always_after_branch = stub_group_size_param < 0; section_size_type stub_group_size = abs(stub_group_size_param); if (stub_group_size == 1) { // Default value. // Thumb branch range is +-4MB has to be used as the default // maximum size (a given section can contain both ARM and Thumb // code, so the worst case has to be taken into account). If we are // fixing cortex-a8 errata, the branch range has to be even smaller, // since wide conditional branch has a range of +-1MB only. // // This value is 48K less than that, which allows for 4096 // 12-byte stubs. If we exceed that, then we will fail to link. // The user will have to relink with an explicit group size // option. stub_group_size = 4145152; } // The Cortex-A8 erratum fix depends on stubs not being in the same 4K // page as the first half of a 32-bit branch straddling two 4K pages. // This is a crude way of enforcing that. In addition, long conditional // branches of THUMB-2 have a range of +-1M. If we are fixing cortex-A8 // erratum, limit the group size to (1M - 12k) to avoid unreachable // cortex-A8 stubs from long conditional branches. if (this->fix_cortex_a8_) { stubs_always_after_branch = true; const section_size_type cortex_a8_group_size = 1024 * (1024 - 12); stub_group_size = std::max(stub_group_size, cortex_a8_group_size); } group_sections(layout, stub_group_size, stubs_always_after_branch, task); // Also fix .ARM.exidx section coverage. Arm_output_section<big_endian>* exidx_output_section = NULL; for (Layout::Section_list::const_iterator p = layout->section_list().begin(); p != layout->section_list().end(); ++p) if ((*p)->type() == elfcpp::SHT_ARM_EXIDX) { if (exidx_output_section == NULL) exidx_output_section = Arm_output_section<big_endian>::as_arm_output_section(*p); else // We cannot handle this now. gold_error(_("multiple SHT_ARM_EXIDX sections %s and %s in a " "non-relocatable link"), exidx_output_section->name(), (*p)->name()); } if (exidx_output_section != NULL) { this->fix_exidx_coverage(layout, input_objects, exidx_output_section, symtab, task); done_exidx_fixup = true; } } else { // If this is not the first pass, addresses and file offsets have // been reset at this point, set them here. for (Stub_table_iterator sp = this->stub_tables_.begin(); sp != this->stub_tables_.end(); ++sp) { Arm_input_section<big_endian>* owner = (*sp)->owner(); off_t off = align_address(owner->original_size(), (*sp)->addralign()); (*sp)->set_address_and_file_offset(owner->address() + off, owner->offset() + off); } } // The Cortex-A8 stubs are sensitive to layout of code sections. At the // beginning of each relaxation pass, just blow away all the stubs. // Alternatively, we could selectively remove only the stubs and reloc // information for code sections that have moved since the last pass. // That would require more book-keeping. if (this->fix_cortex_a8_) { // Clear all Cortex-A8 reloc information. for (typename Cortex_a8_relocs_info::const_iterator p = this->cortex_a8_relocs_info_.begin(); p != this->cortex_a8_relocs_info_.end(); ++p) delete p->second; this->cortex_a8_relocs_info_.clear(); // Remove all Cortex-A8 stubs. for (Stub_table_iterator sp = this->stub_tables_.begin(); sp != this->stub_tables_.end(); ++sp) (*sp)->remove_all_cortex_a8_stubs(); } // Scan relocs for relocation stubs for (Input_objects::Relobj_iterator op = input_objects->relobj_begin(); op != input_objects->relobj_end(); ++op) { Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(*op); // Lock the object so we can read from it. This is only called // single-threaded from Layout::finalize, so it is OK to lock. Task_lock_obj<Object> tl(task, arm_relobj); arm_relobj->scan_sections_for_stubs(this, symtab, layout); } // Check all stub tables to see if any of them have their data sizes // or addresses alignments changed. These are the only things that // matter. bool any_stub_table_changed = false; Unordered_set<const Output_section*> sections_needing_adjustment; for (Stub_table_iterator sp = this->stub_tables_.begin(); (sp != this->stub_tables_.end()) && !any_stub_table_changed; ++sp) { if ((*sp)->update_data_size_and_addralign()) { // Update data size of stub table owner. Arm_input_section<big_endian>* owner = (*sp)->owner(); uint64_t address = owner->address(); off_t offset = owner->offset(); owner->reset_address_and_file_offset(); owner->set_address_and_file_offset(address, offset); sections_needing_adjustment.insert(owner->output_section()); any_stub_table_changed = true; } } // Output_section_data::output_section() returns a const pointer but we // need to update output sections, so we record all output sections needing // update above and scan the sections here to find out what sections need // to be updated. for (Layout::Section_list::const_iterator p = layout->section_list().begin(); p != layout->section_list().end(); ++p) { if (sections_needing_adjustment.find(*p) != sections_needing_adjustment.end()) (*p)->set_section_offsets_need_adjustment(); } // Stop relaxation if no EXIDX fix-up and no stub table change. bool continue_relaxation = done_exidx_fixup || any_stub_table_changed; // Finalize the stubs in the last relaxation pass. if (!continue_relaxation) { for (Stub_table_iterator sp = this->stub_tables_.begin(); (sp != this->stub_tables_.end()) && !any_stub_table_changed; ++sp) (*sp)->finalize_stubs(); // Update output local symbol counts of objects if necessary. for (Input_objects::Relobj_iterator op = input_objects->relobj_begin(); op != input_objects->relobj_end(); ++op) { Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(*op); // Update output local symbol counts. We need to discard local // symbols defined in parts of input sections that are discarded by // relaxation. if (arm_relobj->output_local_symbol_count_needs_update()) { // We need to lock the object's file to update it. Task_lock_obj<Object> tl(task, arm_relobj); arm_relobj->update_output_local_symbol_count(); } } } return continue_relaxation; } // Relocate a stub. template<bool big_endian> void Target_arm<big_endian>::relocate_stub( Stub* stub, const Relocate_info<32, big_endian>* relinfo, Output_section* output_section, unsigned char* view, Arm_address address, section_size_type view_size) { Relocate relocate; const Stub_template* stub_template = stub->stub_template(); for (size_t i = 0; i < stub_template->reloc_count(); i++) { size_t reloc_insn_index = stub_template->reloc_insn_index(i); const Insn_template* insn = &stub_template->insns()[reloc_insn_index]; unsigned int r_type = insn->r_type(); section_size_type reloc_offset = stub_template->reloc_offset(i); section_size_type reloc_size = insn->size(); gold_assert(reloc_offset + reloc_size <= view_size); // This is the address of the stub destination. Arm_address target = stub->reloc_target(i) + insn->reloc_addend(); Symbol_value<32> symval; symval.set_output_value(target); // Synthesize a fake reloc just in case. We don't have a symbol so // we use 0. unsigned char reloc_buffer[elfcpp::Elf_sizes<32>::rel_size]; memset(reloc_buffer, 0, sizeof(reloc_buffer)); elfcpp::Rel_write<32, big_endian> reloc_write(reloc_buffer); reloc_write.put_r_offset(reloc_offset); reloc_write.put_r_info(elfcpp::elf_r_info<32>(0, r_type)); elfcpp::Rel<32, big_endian> rel(reloc_buffer); relocate.relocate(relinfo, this, output_section, this->fake_relnum_for_stubs, rel, r_type, NULL, &symval, view + reloc_offset, address + reloc_offset, reloc_size); } } // Determine whether an object attribute tag takes an integer, a // string or both. template<bool big_endian> int Target_arm<big_endian>::do_attribute_arg_type(int tag) const { if (tag == Object_attribute::Tag_compatibility) return (Object_attribute::ATTR_TYPE_FLAG_INT_VAL | Object_attribute::ATTR_TYPE_FLAG_STR_VAL); else if (tag == elfcpp::Tag_nodefaults) return (Object_attribute::ATTR_TYPE_FLAG_INT_VAL | Object_attribute::ATTR_TYPE_FLAG_NO_DEFAULT); else if (tag == elfcpp::Tag_CPU_raw_name || tag == elfcpp::Tag_CPU_name) return Object_attribute::ATTR_TYPE_FLAG_STR_VAL; else if (tag < 32) return Object_attribute::ATTR_TYPE_FLAG_INT_VAL; else return ((tag & 1) != 0 ? Object_attribute::ATTR_TYPE_FLAG_STR_VAL : Object_attribute::ATTR_TYPE_FLAG_INT_VAL); } // Reorder attributes. // // The ABI defines that Tag_conformance should be emitted first, and that // Tag_nodefaults should be second (if either is defined). This sets those // two positions, and bumps up the position of all the remaining tags to // compensate. template<bool big_endian> int Target_arm<big_endian>::do_attributes_order(int num) const { // Reorder the known object attributes in output. We want to move // Tag_conformance to position 4 and Tag_conformance to position 5 // and shift everything between 4 .. Tag_conformance - 1 to make room. if (num == 4) return elfcpp::Tag_conformance; if (num == 5) return elfcpp::Tag_nodefaults; if ((num - 2) < elfcpp::Tag_nodefaults) return num - 2; if ((num - 1) < elfcpp::Tag_conformance) return num - 1; return num; } // Scan a span of THUMB code for Cortex-A8 erratum. template<bool big_endian> void Target_arm<big_endian>::scan_span_for_cortex_a8_erratum( Arm_relobj<big_endian>* arm_relobj, unsigned int shndx, section_size_type span_start, section_size_type span_end, const unsigned char* view, Arm_address address) { // Scan for 32-bit Thumb-2 branches which span two 4K regions, where: // // The opcode is BLX.W, BL.W, B.W, Bcc.W // The branch target is in the same 4KB region as the // first half of the branch. // The instruction before the branch is a 32-bit // length non-branch instruction. section_size_type i = span_start; bool last_was_32bit = false; bool last_was_branch = false; while (i < span_end) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; const Valtype* wv = reinterpret_cast<const Valtype*>(view + i); uint32_t insn = elfcpp::Swap<16, big_endian>::readval(wv); bool is_blx = false, is_b = false; bool is_bl = false, is_bcc = false; bool insn_32bit = (insn & 0xe000) == 0xe000 && (insn & 0x1800) != 0x0000; if (insn_32bit) { // Load the rest of the insn (in manual-friendly order). insn = (insn << 16) | elfcpp::Swap<16, big_endian>::readval(wv + 1); // Encoding T4: B<c>.W. is_b = (insn & 0xf800d000U) == 0xf0009000U; // Encoding T1: BL<c>.W. is_bl = (insn & 0xf800d000U) == 0xf000d000U; // Encoding T2: BLX<c>.W. is_blx = (insn & 0xf800d000U) == 0xf000c000U; // Encoding T3: B<c>.W (not permitted in IT block). is_bcc = ((insn & 0xf800d000U) == 0xf0008000U && (insn & 0x07f00000U) != 0x03800000U); } bool is_32bit_branch = is_b || is_bl || is_blx || is_bcc; // If this instruction is a 32-bit THUMB branch that crosses a 4K // page boundary and it follows 32-bit non-branch instruction, // we need to work around. if (is_32bit_branch && ((address + i) & 0xfffU) == 0xffeU && last_was_32bit && !last_was_branch) { // Check to see if there is a relocation stub for this branch. bool force_target_arm = false; bool force_target_thumb = false; const Cortex_a8_reloc* cortex_a8_reloc = NULL; Cortex_a8_relocs_info::const_iterator p = this->cortex_a8_relocs_info_.find(address + i); if (p != this->cortex_a8_relocs_info_.end()) { cortex_a8_reloc = p->second; bool target_is_thumb = (cortex_a8_reloc->destination() & 1) != 0; if (cortex_a8_reloc->r_type() == elfcpp::R_ARM_THM_CALL && !target_is_thumb) force_target_arm = true; else if (cortex_a8_reloc->r_type() == elfcpp::R_ARM_THM_CALL && target_is_thumb) force_target_thumb = true; } off_t offset; Stub_type stub_type = arm_stub_none; // Check if we have an offending branch instruction. uint16_t upper_insn = (insn >> 16) & 0xffffU; uint16_t lower_insn = insn & 0xffffU; typedef struct Arm_relocate_functions<big_endian> RelocFuncs; if (cortex_a8_reloc != NULL && cortex_a8_reloc->reloc_stub() != NULL) // We've already made a stub for this instruction, e.g. // it's a long branch or a Thumb->ARM stub. Assume that // stub will suffice to work around the A8 erratum (see // setting of always_after_branch above). ; else if (is_bcc) { offset = RelocFuncs::thumb32_cond_branch_offset(upper_insn, lower_insn); stub_type = arm_stub_a8_veneer_b_cond; } else if (is_b || is_bl || is_blx) { offset = RelocFuncs::thumb32_branch_offset(upper_insn, lower_insn); if (is_blx) offset &= ~3; stub_type = (is_blx ? arm_stub_a8_veneer_blx : (is_bl ? arm_stub_a8_veneer_bl : arm_stub_a8_veneer_b)); } if (stub_type != arm_stub_none) { Arm_address pc_for_insn = address + i + 4; // The original instruction is a BL, but the target is // an ARM instruction. If we were not making a stub, // the BL would have been converted to a BLX. Use the // BLX stub instead in that case. if (this->may_use_blx() && force_target_arm && stub_type == arm_stub_a8_veneer_bl) { stub_type = arm_stub_a8_veneer_blx; is_blx = true; is_bl = false; } // Conversely, if the original instruction was // BLX but the target is Thumb mode, use the BL stub. else if (force_target_thumb && stub_type == arm_stub_a8_veneer_blx) { stub_type = arm_stub_a8_veneer_bl; is_blx = false; is_bl = true; } if (is_blx) pc_for_insn &= ~3; // If we found a relocation, use the proper destination, // not the offset in the (unrelocated) instruction. // Note this is always done if we switched the stub type above. if (cortex_a8_reloc != NULL) offset = (off_t) (cortex_a8_reloc->destination() - pc_for_insn); Arm_address target = (pc_for_insn + offset) | (is_blx ? 0 : 1); // Add a new stub if destination address in in the same page. if (((address + i) & ~0xfffU) == (target & ~0xfffU)) { Cortex_a8_stub* stub = this->stub_factory_.make_cortex_a8_stub(stub_type, arm_relobj, shndx, address + i, target, insn); Stub_table<big_endian>* stub_table = arm_relobj->stub_table(shndx); gold_assert(stub_table != NULL); stub_table->add_cortex_a8_stub(address + i, stub); } } } i += insn_32bit ? 4 : 2; last_was_32bit = insn_32bit; last_was_branch = is_32bit_branch; } } // Apply the Cortex-A8 workaround. template<bool big_endian> void Target_arm<big_endian>::apply_cortex_a8_workaround( const Cortex_a8_stub* stub, Arm_address stub_address, unsigned char* insn_view, Arm_address insn_address) { typedef typename elfcpp::Swap<16, big_endian>::Valtype Valtype; Valtype* wv = reinterpret_cast<Valtype*>(insn_view); Valtype upper_insn = elfcpp::Swap<16, big_endian>::readval(wv); Valtype lower_insn = elfcpp::Swap<16, big_endian>::readval(wv + 1); off_t branch_offset = stub_address - (insn_address + 4); typedef struct Arm_relocate_functions<big_endian> RelocFuncs; switch (stub->stub_template()->type()) { case arm_stub_a8_veneer_b_cond: // For a conditional branch, we re-write it to be an unconditional // branch to the stub. We use the THUMB-2 encoding here. upper_insn = 0xf000U; lower_insn = 0xb800U; // Fall through case arm_stub_a8_veneer_b: case arm_stub_a8_veneer_bl: case arm_stub_a8_veneer_blx: if ((lower_insn & 0x5000U) == 0x4000U) // For a BLX instruction, make sure that the relocation is // rounded up to a word boundary. This follows the semantics of // the instruction which specifies that bit 1 of the target // address will come from bit 1 of the base address. branch_offset = (branch_offset + 2) & ~3; // Put BRANCH_OFFSET back into the insn. gold_assert(!utils::has_overflow<25>(branch_offset)); upper_insn = RelocFuncs::thumb32_branch_upper(upper_insn, branch_offset); lower_insn = RelocFuncs::thumb32_branch_lower(lower_insn, branch_offset); break; default: gold_unreachable(); } // Put the relocated value back in the object file: elfcpp::Swap<16, big_endian>::writeval(wv, upper_insn); elfcpp::Swap<16, big_endian>::writeval(wv + 1, lower_insn); } template<bool big_endian> class Target_selector_arm : public Target_selector { public: Target_selector_arm() : Target_selector(elfcpp::EM_ARM, 32, big_endian, (big_endian ? "elf32-bigarm" : "elf32-littlearm")) { } Target* do_instantiate_target() { return new Target_arm<big_endian>(); } }; // Fix .ARM.exidx section coverage. template<bool big_endian> void Target_arm<big_endian>::fix_exidx_coverage( Layout* layout, const Input_objects* input_objects, Arm_output_section<big_endian>* exidx_section, Symbol_table* symtab, const Task* task) { // We need to look at all the input sections in output in ascending // order of of output address. We do that by building a sorted list // of output sections by addresses. Then we looks at the output sections // in order. The input sections in an output section are already sorted // by addresses within the output section. typedef std::set<Output_section*, output_section_address_less_than> Sorted_output_section_list; Sorted_output_section_list sorted_output_sections; // Find out all the output sections of input sections pointed by // EXIDX input sections. for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { Arm_relobj<big_endian>* arm_relobj = Arm_relobj<big_endian>::as_arm_relobj(*p); std::vector<unsigned int> shndx_list; arm_relobj->get_exidx_shndx_list(&shndx_list); for (size_t i = 0; i < shndx_list.size(); ++i) { const Arm_exidx_input_section* exidx_input_section = arm_relobj->exidx_input_section_by_shndx(shndx_list[i]); gold_assert(exidx_input_section != NULL); if (!exidx_input_section->has_errors()) { unsigned int text_shndx = exidx_input_section->link(); Output_section* os = arm_relobj->output_section(text_shndx); if (os != NULL && (os->flags() & elfcpp::SHF_ALLOC) != 0) sorted_output_sections.insert(os); } } } // Go over the output sections in ascending order of output addresses. typedef typename Arm_output_section<big_endian>::Text_section_list Text_section_list; Text_section_list sorted_text_sections; for (typename Sorted_output_section_list::iterator p = sorted_output_sections.begin(); p != sorted_output_sections.end(); ++p) { Arm_output_section<big_endian>* arm_output_section = Arm_output_section<big_endian>::as_arm_output_section(*p); arm_output_section->append_text_sections_to_list(&sorted_text_sections); } exidx_section->fix_exidx_coverage(layout, sorted_text_sections, symtab, merge_exidx_entries(), task); } Target_selector_arm<false> target_selector_arm; Target_selector_arm<true> target_selector_armbe; } // End anonymous namespace.
Go to most recent revision | Compare with Previous | Blame | View Log