URL
https://opencores.org/ocsvn/open8_urisc/open8_urisc/trunk
Subversion Repositories open8_urisc
[/] [open8_urisc/] [trunk/] [gnu/] [binutils/] [gold/] [gold.cc] - Rev 190
Go to most recent revision | Compare with Previous | Blame | View Log
// gold.cc -- main linker functions // Copyright 2006, 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc. // Written by Ian Lance Taylor <iant@google.com>. // 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 <cstdlib> #include <cstdio> #include <cstring> #include <unistd.h> #include <algorithm> #include "libiberty.h" #include "options.h" #include "target-select.h" #include "debug.h" #include "workqueue.h" #include "dirsearch.h" #include "readsyms.h" #include "symtab.h" #include "common.h" #include "object.h" #include "layout.h" #include "reloc.h" #include "defstd.h" #include "plugin.h" #include "gc.h" #include "icf.h" #include "incremental.h" #include "timer.h" namespace gold { class Object; const char* program_name; static Task* process_incremental_input(Incremental_binary*, unsigned int, Input_objects*, Symbol_table*, Layout*, Dirsearch*, Mapfile*, Task_token*, Task_token*); void gold_exit(Exit_status status) { if (parameters != NULL && parameters->options_valid() && parameters->options().has_plugins()) parameters->options().plugins()->cleanup(); if (status != GOLD_OK && parameters != NULL && parameters->options_valid()) unlink_if_ordinary(parameters->options().output_file_name()); exit(status); } void gold_nomem() { // We are out of memory, so try hard to print a reasonable message. // Note that we don't try to translate this message, since the // translation process itself will require memory. // LEN only exists to avoid a pointless warning when write is // declared with warn_use_result, as when compiling with // -D_USE_FORTIFY on GNU/Linux. Casting to void does not appear to // work, at least not with gcc 4.3.0. ssize_t len = write(2, program_name, strlen(program_name)); if (len >= 0) { const char* const s = ": out of memory\n"; len = write(2, s, strlen(s)); } gold_exit(GOLD_ERR); } // Handle an unreachable case. void do_gold_unreachable(const char* filename, int lineno, const char* function) { fprintf(stderr, _("%s: internal error in %s, at %s:%d\n"), program_name, function, filename, lineno); gold_exit(GOLD_ERR); } // This class arranges to run the functions done in the middle of the // link. It is just a closure. class Middle_runner : public Task_function_runner { public: Middle_runner(const General_options& options, const Input_objects* input_objects, Symbol_table* symtab, Layout* layout, Mapfile* mapfile) : options_(options), input_objects_(input_objects), symtab_(symtab), layout_(layout), mapfile_(mapfile) { } void run(Workqueue*, const Task*); private: const General_options& options_; const Input_objects* input_objects_; Symbol_table* symtab_; Layout* layout_; Mapfile* mapfile_; }; void Middle_runner::run(Workqueue* workqueue, const Task* task) { queue_middle_tasks(this->options_, task, this->input_objects_, this->symtab_, this->layout_, workqueue, this->mapfile_); } // This class arranges the tasks to process the relocs for garbage collection. class Gc_runner : public Task_function_runner { public: Gc_runner(const General_options& options, const Input_objects* input_objects, Symbol_table* symtab, Layout* layout, Mapfile* mapfile) : options_(options), input_objects_(input_objects), symtab_(symtab), layout_(layout), mapfile_(mapfile) { } void run(Workqueue*, const Task*); private: const General_options& options_; const Input_objects* input_objects_; Symbol_table* symtab_; Layout* layout_; Mapfile* mapfile_; }; void Gc_runner::run(Workqueue* workqueue, const Task* task) { queue_middle_gc_tasks(this->options_, task, this->input_objects_, this->symtab_, this->layout_, workqueue, this->mapfile_); } // Queue up the initial set of tasks for this link job. void queue_initial_tasks(const General_options& options, Dirsearch& search_path, const Command_line& cmdline, Workqueue* workqueue, Input_objects* input_objects, Symbol_table* symtab, Layout* layout, Mapfile* mapfile) { if (cmdline.begin() == cmdline.end()) { bool is_ok = false; if (options.printed_version()) is_ok = true; if (options.print_output_format()) { print_output_format(); is_ok = true; } if (is_ok) gold_exit(GOLD_OK); gold_fatal(_("no input files")); } int thread_count = options.thread_count_initial(); if (thread_count == 0) thread_count = cmdline.number_of_input_files(); workqueue->set_thread_count(thread_count); // For incremental links, the base output file. Incremental_binary* ibase = NULL; if (parameters->incremental_update()) { Output_file* of = new Output_file(options.output_file_name()); if (of->open_base_file(options.incremental_base(), true)) { ibase = open_incremental_binary(of); if (ibase != NULL && ibase->check_inputs(cmdline, layout->incremental_inputs())) ibase->init_layout(layout); else { delete ibase; ibase = NULL; of->close(); } } if (ibase == NULL) { if (set_parameters_incremental_full()) gold_info(_("linking with --incremental-full")); else gold_fallback(_("restart link with --incremental-full")); } } // Read the input files. We have to add the symbols to the symbol // table in order. We do this by creating a separate blocker for // each input file. We associate the blocker with the following // input file, to give us a convenient place to delete it. Task_token* this_blocker = NULL; if (ibase == NULL) { // Normal link. Queue a Read_symbols task for each input file // on the command line. for (Command_line::const_iterator p = cmdline.begin(); p != cmdline.end(); ++p) { Task_token* next_blocker = new Task_token(true); next_blocker->add_blocker(); workqueue->queue(new Read_symbols(input_objects, symtab, layout, &search_path, 0, mapfile, &*p, NULL, NULL, this_blocker, next_blocker)); this_blocker = next_blocker; } } else { // Incremental update link. Process the list of input files // stored in the base file, and queue a task for each file: // a Read_symbols task for a changed file, and an Add_symbols task // for an unchanged file. We need to mark all the space used by // unchanged files before we can start any tasks running. unsigned int input_file_count = ibase->input_file_count(); std::vector<Task*> tasks; tasks.reserve(input_file_count); for (unsigned int i = 0; i < input_file_count; ++i) { Task_token* next_blocker = new Task_token(true); next_blocker->add_blocker(); Task* t = process_incremental_input(ibase, i, input_objects, symtab, layout, &search_path, mapfile, this_blocker, next_blocker); tasks.push_back(t); this_blocker = next_blocker; } // Now we can queue the tasks. for (unsigned int i = 0; i < tasks.size(); i++) workqueue->queue(tasks[i]); } if (options.has_plugins()) { Task_token* next_blocker = new Task_token(true); next_blocker->add_blocker(); workqueue->queue(new Plugin_hook(options, input_objects, symtab, layout, &search_path, mapfile, this_blocker, next_blocker)); this_blocker = next_blocker; } if (options.relocatable() && (options.gc_sections() || options.icf_enabled())) gold_error(_("cannot mix -r with --gc-sections or --icf")); if (options.gc_sections() || options.icf_enabled()) { workqueue->queue(new Task_function(new Gc_runner(options, input_objects, symtab, layout, mapfile), this_blocker, "Task_function Gc_runner")); } else { workqueue->queue(new Task_function(new Middle_runner(options, input_objects, symtab, layout, mapfile), this_blocker, "Task_function Middle_runner")); } } // Process an incremental input file: if it is unchanged from the previous // link, return a task to add its symbols from the base file's incremental // info; if it has changed, return a normal Read_symbols task. We create a // task for every input file, if only to report the file for rebuilding the // incremental info. static Task* process_incremental_input(Incremental_binary* ibase, unsigned int input_file_index, Input_objects* input_objects, Symbol_table* symtab, Layout* layout, Dirsearch* search_path, Mapfile* mapfile, Task_token* this_blocker, Task_token* next_blocker) { const Incremental_binary::Input_reader* input_reader = ibase->get_input_reader(input_file_index); Incremental_input_type input_type = input_reader->type(); // Get the input argument corresponding to this input file, matching on // the argument serial number. If the input file cannot be matched // to an existing input argument, synthesize a new one. const Input_argument* input_argument = ibase->get_input_argument(input_file_index); if (input_argument == NULL) { Input_file_argument file(input_reader->filename(), Input_file_argument::INPUT_FILE_TYPE_FILE, "", false, parameters->options()); Input_argument* arg = new Input_argument(file); arg->set_script_info(ibase->get_script_info(input_file_index)); input_argument = arg; } gold_debug(DEBUG_INCREMENTAL, "Incremental object: %s, type %d", input_reader->filename(), input_type); if (input_type == INCREMENTAL_INPUT_SCRIPT) { // Incremental_binary::check_inputs should have cancelled the // incremental update if the script has changed. gold_assert(!ibase->file_has_changed(input_file_index)); return new Check_script(layout, ibase, input_file_index, input_reader, this_blocker, next_blocker); } if (input_type == INCREMENTAL_INPUT_ARCHIVE) { Incremental_library* lib = ibase->get_library(input_file_index); gold_assert(lib != NULL); if (lib->filename() == "/group/" || !ibase->file_has_changed(input_file_index)) { // Queue a task to check that no references have been added to any // of the library's unused symbols. return new Check_library(symtab, layout, ibase, input_file_index, input_reader, this_blocker, next_blocker); } else { // Queue a Read_symbols task to process the archive normally. return new Read_symbols(input_objects, symtab, layout, search_path, 0, mapfile, input_argument, NULL, NULL, this_blocker, next_blocker); } } if (input_type == INCREMENTAL_INPUT_ARCHIVE_MEMBER) { // For archive members, check the timestamp of the containing archive. Incremental_library* lib = ibase->get_library(input_file_index); gold_assert(lib != NULL); // Process members of a --start-lib/--end-lib group as normal objects. if (lib->filename() != "/group/") { if (ibase->file_has_changed(lib->input_file_index())) { return new Read_member(input_objects, symtab, layout, mapfile, input_reader, this_blocker, next_blocker); } else { // The previous contributions from this file will be kept. // Mark the pieces of output sections contributed by this // object. ibase->reserve_layout(input_file_index); Object* obj = make_sized_incremental_object(ibase, input_file_index, input_type, input_reader); return new Add_symbols(input_objects, symtab, layout, search_path, 0, mapfile, input_argument, obj, lib, NULL, this_blocker, next_blocker); } } } // Normal object file or shared library. Check if the file has changed // since the last incremental link. if (ibase->file_has_changed(input_file_index)) { return new Read_symbols(input_objects, symtab, layout, search_path, 0, mapfile, input_argument, NULL, NULL, this_blocker, next_blocker); } else { // The previous contributions from this file will be kept. // Mark the pieces of output sections contributed by this object. ibase->reserve_layout(input_file_index); Object* obj = make_sized_incremental_object(ibase, input_file_index, input_type, input_reader); return new Add_symbols(input_objects, symtab, layout, search_path, 0, mapfile, input_argument, obj, NULL, NULL, this_blocker, next_blocker); } } // Queue up a set of tasks to be done before queueing the middle set // of tasks. This is only necessary when garbage collection // (--gc-sections) of unused sections is desired. The relocs are read // and processed here early to determine the garbage sections before the // relocs can be scanned in later tasks. void queue_middle_gc_tasks(const General_options& options, const Task* , const Input_objects* input_objects, Symbol_table* symtab, Layout* layout, Workqueue* workqueue, Mapfile* mapfile) { // Read_relocs for all the objects must be done and processed to find // unused sections before any scanning of the relocs can take place. Task_token* this_blocker = NULL; for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { Task_token* next_blocker = new Task_token(true); next_blocker->add_blocker(); workqueue->queue(new Read_relocs(symtab, layout, *p, this_blocker, next_blocker)); this_blocker = next_blocker; } // If we are given only archives in input, we have no regular // objects and THIS_BLOCKER is NULL here. Create a dummy // blocker here so that we can run the middle tasks immediately. if (this_blocker == NULL) { gold_assert(input_objects->number_of_relobjs() == 0); this_blocker = new Task_token(true); } workqueue->queue(new Task_function(new Middle_runner(options, input_objects, symtab, layout, mapfile), this_blocker, "Task_function Middle_runner")); } // Queue up the middle set of tasks. These are the tasks which run // after all the input objects have been found and all the symbols // have been read, but before we lay out the output file. void queue_middle_tasks(const General_options& options, const Task* task, const Input_objects* input_objects, Symbol_table* symtab, Layout* layout, Workqueue* workqueue, Mapfile* mapfile) { Timer* timer = parameters->timer(); if (timer != NULL) timer->stamp(0); // Add any symbols named with -u options to the symbol table. symtab->add_undefined_symbols_from_command_line(layout); // If garbage collection was chosen, relocs have been read and processed // at this point by pre_middle_tasks. Layout can then be done for all // objects. if (parameters->options().gc_sections()) { // Find the start symbol if any. Symbol* start_sym = symtab->lookup(parameters->entry()); if (start_sym != NULL) { bool is_ordinary; unsigned int shndx = start_sym->shndx(&is_ordinary); if (is_ordinary) { symtab->gc()->worklist().push( Section_id(start_sym->object(), shndx)); } } // Symbols named with -u should not be considered garbage. symtab->gc_mark_undef_symbols(layout); gold_assert(symtab->gc() != NULL); // Do a transitive closure on all references to determine the worklist. symtab->gc()->do_transitive_closure(); } // If identical code folding (--icf) is chosen it makes sense to do it // only after garbage collection (--gc-sections) as we do not want to // be folding sections that will be garbage. if (parameters->options().icf_enabled()) { symtab->icf()->find_identical_sections(input_objects, symtab); } // Call Object::layout for the second time to determine the // output_sections for all referenced input sections. When // --gc-sections or --icf is turned on, Object::layout is // called twice. It is called the first time when the // symbols are added. if (parameters->options().gc_sections() || parameters->options().icf_enabled()) { for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { Task_lock_obj<Object> tlo(task, *p); (*p)->layout(symtab, layout, NULL); } } /* If plugins have specified a section order, re-arrange input sections according to a specified section order. If --section-ordering-file is also specified, do not do anything here. */ if (parameters->options().has_plugins() && layout->is_section_ordering_specified() && !parameters->options().section_ordering_file ()) { for (Layout::Section_list::const_iterator p = layout->section_list().begin(); p != layout->section_list().end(); ++p) (*p)->update_section_layout(layout->get_section_order_map()); } // Layout deferred objects due to plugins. if (parameters->options().has_plugins()) { Plugin_manager* plugins = parameters->options().plugins(); gold_assert(plugins != NULL); plugins->layout_deferred_objects(); } if (parameters->options().gc_sections() || parameters->options().icf_enabled()) { for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { // Update the value of output_section stored in rd. Read_relocs_data* rd = (*p)->get_relocs_data(); for (Read_relocs_data::Relocs_list::iterator q = rd->relocs.begin(); q != rd->relocs.end(); ++q) { q->output_section = (*p)->output_section(q->data_shndx); q->needs_special_offset_handling = (*p)->is_output_section_offset_invalid(q->data_shndx); } } } // We have to support the case of not seeing any input objects, and // generate an empty file. Existing builds depend on being able to // pass an empty archive to the linker and get an empty object file // out. In order to do this we need to use a default target. if (input_objects->number_of_input_objects() == 0 && layout->incremental_base() == NULL) parameters_force_valid_target(); int thread_count = options.thread_count_middle(); if (thread_count == 0) thread_count = std::max(2, input_objects->number_of_input_objects()); workqueue->set_thread_count(thread_count); // Now we have seen all the input files. const bool doing_static_link = (!input_objects->any_dynamic() && !parameters->options().output_is_position_independent()); set_parameters_doing_static_link(doing_static_link); if (!doing_static_link && options.is_static()) { // We print out just the first .so we see; there may be others. gold_assert(input_objects->dynobj_begin() != input_objects->dynobj_end()); gold_error(_("cannot mix -static with dynamic object %s"), (*input_objects->dynobj_begin())->name().c_str()); } if (!doing_static_link && parameters->options().relocatable()) gold_fatal(_("cannot mix -r with dynamic object %s"), (*input_objects->dynobj_begin())->name().c_str()); if (!doing_static_link && options.oformat_enum() != General_options::OBJECT_FORMAT_ELF) gold_fatal(_("cannot use non-ELF output format with dynamic object %s"), (*input_objects->dynobj_begin())->name().c_str()); if (parameters->options().relocatable()) { Input_objects::Relobj_iterator p = input_objects->relobj_begin(); if (p != input_objects->relobj_end()) { bool uses_split_stack = (*p)->uses_split_stack(); for (++p; p != input_objects->relobj_end(); ++p) { if ((*p)->uses_split_stack() != uses_split_stack) gold_fatal(_("cannot mix split-stack '%s' and " "non-split-stack '%s' when using -r"), (*input_objects->relobj_begin())->name().c_str(), (*p)->name().c_str()); } } } // For incremental updates, record the existing GOT and PLT entries, // and the COPY relocations. if (parameters->incremental_update()) { Incremental_binary* ibase = layout->incremental_base(); ibase->process_got_plt(symtab, layout); ibase->emit_copy_relocs(symtab); } if (is_debugging_enabled(DEBUG_SCRIPT)) layout->script_options()->print(stderr); // For each dynamic object, record whether we've seen all the // dynamic objects that it depends upon. input_objects->check_dynamic_dependencies(); // See if any of the input definitions violate the One Definition Rule. // TODO: if this is too slow, do this as a task, rather than inline. symtab->detect_odr_violations(task, options.output_file_name()); // Do the --no-undefined-version check. if (!parameters->options().undefined_version()) { Script_options* so = layout->script_options(); so->version_script_info()->check_unmatched_names(symtab); } // Create any automatic note sections. layout->create_notes(); // Create any output sections required by any linker script. layout->create_script_sections(); // Define some sections and symbols needed for a dynamic link. This // handles some cases we want to see before we read the relocs. layout->create_initial_dynamic_sections(symtab); // Define symbols from any linker scripts. layout->define_script_symbols(symtab); // Attach sections to segments. layout->attach_sections_to_segments(); if (!parameters->options().relocatable()) { // Predefine standard symbols. define_standard_symbols(symtab, layout); // Define __start and __stop symbols for output sections where // appropriate. layout->define_section_symbols(symtab); } // Make sure we have symbols for any required group signatures. layout->define_group_signatures(symtab); Task_token* this_blocker = NULL; // Allocate common symbols. We use a blocker to run this before the // Scan_relocs tasks, because it writes to the symbol table just as // they do. if (parameters->options().define_common()) { this_blocker = new Task_token(true); this_blocker->add_blocker(); workqueue->queue(new Allocate_commons_task(symtab, layout, mapfile, this_blocker)); } // If doing garbage collection, the relocations have already been read. // Otherwise, read and scan the relocations. if (parameters->options().gc_sections() || parameters->options().icf_enabled()) { for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { Task_token* next_blocker = new Task_token(true); next_blocker->add_blocker(); workqueue->queue(new Scan_relocs(symtab, layout, *p, (*p)->get_relocs_data(), this_blocker, next_blocker)); this_blocker = next_blocker; } } else { // Read the relocations of the input files. We do this to find // which symbols are used by relocations which require a GOT and/or // a PLT entry, or a COPY reloc. When we implement garbage // collection we will do it here by reading the relocations in a // breadth first search by references. // // We could also read the relocations during the first pass, and // mark symbols at that time. That is how the old GNU linker works. // Doing that is more complex, since we may later decide to discard // some of the sections, and thus change our minds about the types // of references made to the symbols. for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { Task_token* next_blocker = new Task_token(true); next_blocker->add_blocker(); workqueue->queue(new Read_relocs(symtab, layout, *p, this_blocker, next_blocker)); this_blocker = next_blocker; } } if (this_blocker == NULL) { if (input_objects->number_of_relobjs() == 0) { // If we are given only archives in input, we have no regular // objects and THIS_BLOCKER is NULL here. Create a dummy // blocker here so that we can run the layout task immediately. this_blocker = new Task_token(true); } else { // If we failed to open any input files, it's possible for // THIS_BLOCKER to be NULL here. There's no real point in // continuing if that happens. gold_assert(parameters->errors()->error_count() > 0); gold_exit(GOLD_ERR); } } // When all those tasks are complete, we can start laying out the // output file. // TODO(csilvers): figure out a more principled way to get the target Target* target = const_cast<Target*>(¶meters->target()); workqueue->queue(new Task_function(new Layout_task_runner(options, input_objects, symtab, target, layout, mapfile), this_blocker, "Task_function Layout_task_runner")); } // Queue up the final set of tasks. This is called at the end of // Layout_task. void queue_final_tasks(const General_options& options, const Input_objects* input_objects, const Symbol_table* symtab, Layout* layout, Workqueue* workqueue, Output_file* of) { Timer* timer = parameters->timer(); if (timer != NULL) timer->stamp(1); int thread_count = options.thread_count_final(); if (thread_count == 0) thread_count = std::max(2, input_objects->number_of_input_objects()); workqueue->set_thread_count(thread_count); bool any_postprocessing_sections = layout->any_postprocessing_sections(); // Use a blocker to wait until all the input sections have been // written out. Task_token* input_sections_blocker = NULL; if (!any_postprocessing_sections) { input_sections_blocker = new Task_token(true); input_sections_blocker->add_blockers(input_objects->number_of_relobjs()); } // Use a blocker to block any objects which have to wait for the // output sections to complete before they can apply relocations. Task_token* output_sections_blocker = new Task_token(true); output_sections_blocker->add_blocker(); // Use a blocker to block the final cleanup task. Task_token* final_blocker = new Task_token(true); // Write_symbols_task, Write_sections_task, Write_data_task, // Relocate_tasks. final_blocker->add_blockers(3); final_blocker->add_blockers(input_objects->number_of_relobjs()); if (!any_postprocessing_sections) final_blocker->add_blocker(); // Queue a task to write out the symbol table. workqueue->queue(new Write_symbols_task(layout, symtab, input_objects, layout->sympool(), layout->dynpool(), of, final_blocker)); // Queue a task to write out the output sections. workqueue->queue(new Write_sections_task(layout, of, output_sections_blocker, final_blocker)); // Queue a task to write out everything else. workqueue->queue(new Write_data_task(layout, symtab, of, final_blocker)); // Queue a task for each input object to relocate the sections and // write out the local symbols. for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) workqueue->queue(new Relocate_task(symtab, layout, *p, of, input_sections_blocker, output_sections_blocker, final_blocker)); // Queue a task to write out the output sections which depend on // input sections. If there are any sections which require // postprocessing, then we need to do this last, since it may resize // the output file. if (!any_postprocessing_sections) { Task* t = new Write_after_input_sections_task(layout, of, input_sections_blocker, final_blocker); workqueue->queue(t); } else { Task_token* new_final_blocker = new Task_token(true); new_final_blocker->add_blocker(); Task* t = new Write_after_input_sections_task(layout, of, final_blocker, new_final_blocker); workqueue->queue(t); final_blocker = new_final_blocker; } // Queue a task to close the output file. This will be blocked by // FINAL_BLOCKER. workqueue->queue(new Task_function(new Close_task_runner(&options, layout, of), final_blocker, "Task_function Close_task_runner")); } } // End namespace gold.
Go to most recent revision | Compare with Previous | Blame | View Log