X-Git-Url: http://drtracing.org/?a=blobdiff_plain;f=gold%2Fmerge.cc;h=f370c9cb9bc499fdf39b45e3aee73ea9cc932813;hb=7903e5309890633911c91539e23a407e1f74b959;hp=d8648d6f0554c4f1a4446a6fd55b7dc43d7a70bc;hpb=4625f782a5e7744937b60b0421de3ff9f55346ec;p=deliverable%2Fbinutils-gdb.git diff --git a/gold/merge.cc b/gold/merge.cc index d8648d6f05..f370c9cb9b 100644 --- a/gold/merge.cc +++ b/gold/merge.cc @@ -1,6 +1,6 @@ // merge.cc -- handle section merging for gold -// Copyright 2006, 2007 Free Software Foundation, Inc. +// Copyright 2006, 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc. // Written by Ian Lance Taylor . // This file is part of gold. @@ -26,104 +26,12 @@ #include #include "merge.h" +#include "compressed_output.h" namespace gold { -// For each object with merge sections, we store an Object_merge_map. -// This is used to map locations in input sections to a merged output -// section. The output section itself is not recorded here--it can be -// found in the map_to_output_ field of the Object. - -class Object_merge_map -{ - public: - Object_merge_map() - : first_shnum_(-1U), first_map_(), - second_shnum_(-1U), second_map_(), - section_merge_maps_() - { } - - ~Object_merge_map(); - - // Add a mapping for MERGE_MAP, for the bytes from OFFSET to OFFSET - // + LENGTH in the input section SHNDX to OUTPUT_OFFSET in the - // output section. An OUTPUT_OFFSET of -1 means that the bytes are - // discarded. - void - add_mapping(const Merge_map*, unsigned int shndx, off_t offset, off_t length, - off_t output_offset); - - // Get the output offset for an input address in MERGE_MAP. The - // input address is at offset OFFSET in section SHNDX. This sets - // *OUTPUT_OFFSET to the offset in the output section; this will be - // -1 if the bytes are not being copied to the output. This returns - // true if the mapping is known, false otherwise. - bool - get_output_offset(const Merge_map*, unsigned int shndx, off_t offset, - off_t *output_offset); - - private: - // Map input section offsets to a length and an output section - // offset. An output section offset of -1 means that this part of - // the input section is being discarded. - struct Input_merge_entry - { - // The offset in the input section. - off_t input_offset; - // The length. - off_t length; - // The offset in the output section. - off_t output_offset; - }; - - // A less-than comparison routine for Input_merge_entry. - struct Input_merge_compare - { - bool - operator()(const Input_merge_entry& i1, const Input_merge_entry& i2) const - { return i1.input_offset < i2.input_offset; } - }; - - // A list of entries for a particular section. - struct Input_merge_map - { - // The Merge_map for this section. - const Merge_map* merge_map; - // The list of mappings. - std::vector entries; - // Whether the ENTRIES field is sorted by input_offset. - bool sorted; - - Input_merge_map() - : merge_map(NULL), entries(), sorted(true) - { } - }; - - // Map input section indices to merge maps. - typedef std::map Section_merge_maps; - - // Return a pointer to the Input_merge_map to use for the input - // section SHNDX, or NULL. - Input_merge_map* - get_input_merge_map(unsigned int shndx); - - // Get or make the the Input_merge_map to use for the section SHNDX - // with MERGE_MAP. - Input_merge_map* - get_or_make_input_merge_map(const Merge_map* merge_map, unsigned int shndx); - - // Any given object file will normally only have a couple of input - // sections with mergeable contents. So we keep the first two input - // section numbers inline, and push any further ones into a map. A - // value of -1U in first_shnum_ or second_shnum_ means that we don't - // have a corresponding entry. - unsigned int first_shnum_; - Input_merge_map first_map_; - unsigned int second_shnum_; - Input_merge_map second_map_; - Section_merge_maps section_merge_maps_; -}; +// Class Object_merge_map. // Destructor. @@ -161,7 +69,7 @@ Object_merge_map::get_or_make_input_merge_map(const Merge_map* merge_map, if (map != NULL) { // For a given input section in a given object, every mapping - // must be donw with the same Merge_map. + // must be done with the same Merge_map. gold_assert(map->merge_map == merge_map); return map; } @@ -190,8 +98,9 @@ Object_merge_map::get_or_make_input_merge_map(const Merge_map* merge_map, void Object_merge_map::add_mapping(const Merge_map* merge_map, unsigned int shndx, - off_t input_offset, off_t length, - off_t output_offset) + section_offset_type input_offset, + section_size_type length, + section_offset_type output_offset) { Input_merge_map* map = this->get_or_make_input_merge_map(merge_map, shndx); @@ -200,18 +109,23 @@ Object_merge_map::add_mapping(const Merge_map* merge_map, unsigned int shndx, { Input_merge_entry& entry(map->entries.back()); + // Use section_size_type to avoid signed/unsigned warnings. + section_size_type input_offset_u = input_offset; + section_size_type output_offset_u = output_offset; + // If this entry is not in order, we need to sort the vector // before looking anything up. - if (input_offset < entry.input_offset + entry.length) + if (input_offset_u < entry.input_offset + entry.length) { - gold_assert(input_offset < entry.input_offset - && input_offset + length <= entry.input_offset); + gold_assert(input_offset < entry.input_offset); + gold_assert(input_offset_u + length + <= static_cast(entry.input_offset)); map->sorted = false; } - else if (entry.input_offset + entry.length == input_offset + else if (entry.input_offset + entry.length == input_offset_u && (output_offset == -1 ? entry.output_offset == -1 - : entry.output_offset + entry.length == output_offset)) + : entry.output_offset + entry.length == output_offset_u)) { entry.length += length; return; @@ -229,11 +143,13 @@ Object_merge_map::add_mapping(const Merge_map* merge_map, unsigned int shndx, bool Object_merge_map::get_output_offset(const Merge_map* merge_map, - unsigned int shndx, off_t input_offset, - off_t *output_offset) + unsigned int shndx, + section_offset_type input_offset, + section_offset_type* output_offset) { Input_merge_map* map = this->get_input_merge_map(shndx); - if (map == NULL || map->merge_map != merge_map) + if (map == NULL + || (merge_map != NULL && map->merge_map != merge_map)) return false; if (!map->sorted) @@ -256,7 +172,8 @@ Object_merge_map::get_output_offset(const Merge_map* merge_map, gold_assert(p->input_offset <= input_offset); } - if (input_offset - p->input_offset >= p->length) + if (input_offset - p->input_offset + >= static_cast(p->length)) return false; *output_offset = p->output_offset; @@ -265,16 +182,67 @@ Object_merge_map::get_output_offset(const Merge_map* merge_map, return true; } +// Return whether this is the merge map for section SHNDX. + +inline bool +Object_merge_map::is_merge_section_for(const Merge_map* merge_map, + unsigned int shndx) +{ + Input_merge_map* map = this->get_input_merge_map(shndx); + return map != NULL && map->merge_map == merge_map; +} + +// Initialize a mapping from input offsets to output addresses. + +template +void +Object_merge_map::initialize_input_to_output_map( + unsigned int shndx, + typename elfcpp::Elf_types::Elf_Addr starting_address, + Unordered_map::Elf_Addr>* initialize_map) +{ + Input_merge_map* map = this->get_input_merge_map(shndx); + gold_assert(map != NULL); + + gold_assert(initialize_map->empty()); + // We know how many entries we are going to add. + // reserve_unordered_map takes an expected count of buckets, not a + // count of elements, so double it to try to reduce collisions. + reserve_unordered_map(initialize_map, map->entries.size() * 2); + + for (Input_merge_map::Entries::const_iterator p = map->entries.begin(); + p != map->entries.end(); + ++p) + { + section_offset_type output_offset = p->output_offset; + if (output_offset != -1) + output_offset += starting_address; + else + { + // If we see a relocation against an address we have chosen + // to discard, we relocate to zero. FIXME: We could also + // issue a warning in this case; that would require + // reporting this somehow and checking it in the routines in + // reloc.h. + output_offset = 0; + } + initialize_map->insert(std::make_pair(p->input_offset, output_offset)); + } +} + // Class Merge_map. // Add a mapping for the bytes from OFFSET to OFFSET + LENGTH in input -// section SHNDX in object OBJECT to an OUTPUT_OFFSET in a merged -// output section. +// section SHNDX in object OBJECT to an OUTPUT_OFFSET in merged data +// in an output section. void Merge_map::add_mapping(Relobj* object, unsigned int shndx, - off_t offset, off_t length, off_t output_offset) + section_offset_type offset, section_size_type length, + section_offset_type output_offset) { + gold_assert(object != NULL); Object_merge_map* object_merge_map = object->merge_map(); if (object_merge_map == NULL) { @@ -287,12 +255,14 @@ Merge_map::add_mapping(Relobj* object, unsigned int shndx, // Return the output offset for an input address. The input address // is at offset OFFSET in section SHNDX in OBJECT. This sets -// *OUTPUT_OFFSET to the offset in the output section. This returns -// true if the mapping is known, false otherwise. +// *OUTPUT_OFFSET to the offset in the merged data in the output +// section. This returns true if the mapping is known, false +// otherwise. bool Merge_map::get_output_offset(const Relobj* object, unsigned int shndx, - off_t offset, off_t* output_offset) const + section_offset_type offset, + section_offset_type* output_offset) const { Object_merge_map* object_merge_map = object->merge_map(); if (object_merge_map == NULL) @@ -301,6 +271,17 @@ Merge_map::get_output_offset(const Relobj* object, unsigned int shndx, output_offset); } +// Return whether this is the merge section for SHNDX in OBJECT. + +bool +Merge_map::is_merge_section_for(const Relobj* object, unsigned int shndx) const +{ + Object_merge_map* object_merge_map = object->merge_map(); + if (object_merge_map == NULL) + return false; + return object_merge_map->is_merge_section_for(this, shndx); +} + // Class Output_merge_base. // Return the output offset for an input offset. The input address is @@ -310,12 +291,41 @@ Merge_map::get_output_offset(const Relobj* object, unsigned int shndx, bool Output_merge_base::do_output_offset(const Relobj* object, unsigned int shndx, - off_t offset, - off_t* poutput) const + section_offset_type offset, + section_offset_type* poutput) const { return this->merge_map_.get_output_offset(object, shndx, offset, poutput); } +// Return whether this is the merge section for SHNDX in OBJECT. + +bool +Output_merge_base::do_is_merge_section_for(const Relobj* object, + unsigned int shndx) const +{ + return this->merge_map_.is_merge_section_for(object, shndx); +} + +// Record a merged input section for script processing. + +void +Output_merge_base::record_input_section(Relobj* relobj, unsigned int shndx) +{ + gold_assert(this->keeps_input_sections_ && relobj != NULL); + // If this is the first input section, record it. We need do this because + // this->input_sections_ is unordered. + if (this->first_relobj_ == NULL) + { + this->first_relobj_ = relobj; + this->first_shndx_ = shndx; + } + + std::pair result = + this->input_sections_.insert(Section_id(relobj, shndx)); + // We should insert a merge section once only. + gold_assert(result.second); +} + // Class Output_merge_data. // Compute the hash code for a fixed-size constant. @@ -324,13 +334,14 @@ size_t Output_merge_data::Merge_data_hash::operator()(Merge_data_key k) const { const unsigned char* p = this->pomd_->constant(k); - uint64_t entsize = this->pomd_->entsize(); + section_size_type entsize = + convert_to_section_size_type(this->pomd_->entsize()); // Fowler/Noll/Vo (FNV) hash (type FNV-1a). if (sizeof(size_t) == 8) { size_t result = static_cast(14695981039346656037ULL); - for (uint64_t i = 0; i < entsize; ++i) + for (section_size_type i = 0; i < entsize; ++i) { result &= (size_t) *p++; result *= 1099511628211ULL; @@ -340,7 +351,7 @@ Output_merge_data::Merge_data_hash::operator()(Merge_data_key k) const else { size_t result = 2166136261UL; - for (uint64_t i = 0; i < entsize; ++i) + for (section_size_type i = 0; i < entsize; ++i) { result ^= (size_t) *p++; result *= 16777619UL; @@ -365,8 +376,10 @@ Output_merge_data::Merge_data_eq::operator()(Merge_data_key k1, void Output_merge_data::add_constant(const unsigned char* p) { - uint64_t entsize = this->entsize(); - uint64_t addsize = std::max(entsize, this->addralign()); + section_size_type entsize = convert_to_section_size_type(this->entsize()); + section_size_type addralign = + convert_to_section_size_type(this->addralign()); + section_size_type addsize = std::max(entsize, addralign); if (this->len_ + addsize > this->alc_) { if (this->alc_ == 0) @@ -392,15 +405,23 @@ Output_merge_data::add_constant(const unsigned char* p) bool Output_merge_data::do_add_input_section(Relobj* object, unsigned int shndx) { - off_t len; - const unsigned char* p = object->section_contents(shndx, &len, false); + section_size_type len; + bool is_new; + const unsigned char* p = object->decompressed_section_contents(shndx, &len, + &is_new); - uint64_t entsize = this->entsize(); + section_size_type entsize = convert_to_section_size_type(this->entsize()); if (len % entsize != 0) - return false; + { + if (is_new) + delete[] p; + return false; + } + + this->input_count_ += len / entsize; - for (off_t i = 0; i < len; i += entsize, p += entsize) + for (section_size_type i = 0; i < len; i += entsize, p += entsize) { // Add the constant to the section contents. If we find that it // is already in the hash table, we will remove it again. @@ -421,6 +442,13 @@ Output_merge_data::do_add_input_section(Relobj* object, unsigned int shndx) this->add_mapping(object, shndx, i, entsize, k); } + // For script processing, we keep the input sections. + if (this->keeps_input_sections()) + record_input_section(object, shndx); + + if (is_new) + delete[] p; + return true; } @@ -428,11 +456,14 @@ Output_merge_data::do_add_input_section(Relobj* object, unsigned int shndx) // constants. void -Output_merge_data::do_set_address(uint64_t, off_t) +Output_merge_data::set_final_data_size() { // Release the memory we don't need. this->p_ = static_cast(realloc(this->p_, this->len_)); - gold_assert(this->p_ != NULL); + // An Output_merge_data object may be empty and realloc is allowed + // to return a NULL pointer in this case. An Output_merge_data is empty + // if all its input sections have sizes that are not multiples of entsize. + gold_assert(this->p_ != NULL || this->len_ == 0); this->set_data_size(this->len_); } @@ -445,6 +476,26 @@ Output_merge_data::do_write(Output_file* of) of->write(this->offset(), this->p_, this->len_); } +// Write the data to a buffer. + +void +Output_merge_data::do_write_to_buffer(unsigned char* buffer) +{ + memcpy(buffer, this->p_, this->len_); +} + +// Print merge stats to stderr. + +void +Output_merge_data::do_print_merge_stats(const char* section_name) +{ + fprintf(stderr, + _("%s: %s merged constants size: %lu; input: %zu; output: %zu\n"), + program_name, section_name, + static_cast(this->entsize()), + this->input_count_, this->hashtable_.size()); +} + // Class Output_merge_string. // Add an input section to a merged string section. @@ -454,69 +505,164 @@ bool Output_merge_string::do_add_input_section(Relobj* object, unsigned int shndx) { - off_t len; - const unsigned char* pdata = object->section_contents(shndx, &len, false); + section_size_type sec_len; + bool is_new; + const unsigned char* pdata = object->decompressed_section_contents(shndx, + &sec_len, + &is_new); const Char_type* p = reinterpret_cast(pdata); + const Char_type* pend = p + sec_len / sizeof(Char_type); + const Char_type* pend0 = pend; - if (len % sizeof(Char_type) != 0) + if (sec_len % sizeof(Char_type) != 0) { object->error(_("mergeable string section length not multiple of " "character size")); + if (is_new) + delete[] pdata; return false; } + if (pend[-1] != 0) + { + gold_warning(_("%s: last entry in mergeable string section '%s' " + "not null terminated"), + object->name().c_str(), + object->section_name(shndx).c_str()); + // Find the end of the last NULL-terminated string in the buffer. + while (pend0 > p && pend0[-1] != 0) + --pend0; + } + + Merged_strings_list* merged_strings_list = + new Merged_strings_list(object, shndx); + this->merged_strings_lists_.push_back(merged_strings_list); + Merged_strings& merged_strings = merged_strings_list->merged_strings; + + // Count the number of non-null strings in the section and size the list. + size_t count = 0; + const Char_type* pt = p; + while (pt < pend0) + { + size_t len = string_length(pt); + if (len != 0) + ++count; + pt += len + 1; + } + if (pend0 < pend) + ++count; + merged_strings.reserve(count + 1); + // The index I is in bytes, not characters. - off_t i = 0; - while (i < len) + section_size_type i = 0; + + // We assume here that the beginning of the section is correctly + // aligned, so each string within the section must retain the same + // modulo. + uintptr_t init_align_modulo = (reinterpret_cast(pdata) + & (this->addralign() - 1)); + bool has_misaligned_strings = false; + + while (p < pend0) { - off_t plen = 0; - for (const Char_type* pl = p; *pl != 0; ++pl) - { - // The length PLEN is in characters, not bytes. - ++plen; - if (i + plen * static_cast(sizeof(Char_type)) >= len) - { - object->error(_("entry in mergeable string section " - "not null terminated")); - break; - } - } + size_t len = string_length(p); - const Char_type* str = this->stringpool_.add(p, true, NULL); + // Within merge input section each string must be aligned. + if (len != 0 + && ((reinterpret_cast(p) & (this->addralign() - 1)) + != init_align_modulo)) + has_misaligned_strings = true; - off_t bytelen_with_null = (plen + 1) * sizeof(Char_type); - this->merged_strings_.push_back(Merged_string(object, shndx, i, str, - bytelen_with_null)); + Stringpool::Key key; + this->stringpool_.add_with_length(p, len, true, &key); - p += plen + 1; - i += bytelen_with_null; + merged_strings.push_back(Merged_string(i, key)); + p += len + 1; + i += (len + 1) * sizeof(Char_type); } + if (p < pend) + { + size_t len = pend - p; + + Stringpool::Key key; + this->stringpool_.add_with_length(p, len, true, &key); + + merged_strings.push_back(Merged_string(i, key)); + + i += (len + 1) * sizeof(Char_type); + } + + // Record the last offset in the input section so that we can + // compute the length of the last string. + merged_strings.push_back(Merged_string(i, 0)); + + this->input_count_ += count; + this->input_size_ += i; + + if (has_misaligned_strings) + gold_warning(_("%s: section %s contains incorrectly aligned strings;" + " the alignment of those strings won't be preserved"), + object->name().c_str(), + object->section_name(shndx).c_str()); + + // For script processing, we keep the input sections. + if (this->keeps_input_sections()) + record_input_section(object, shndx); + + if (is_new) + delete[] pdata; return true; } -// Set the final data size of a merged string section. This is where -// we finalize the mappings from the input sections to the output -// section. +// Finalize the mappings from the input sections to the output +// section, and return the final data size. template -void -Output_merge_string::do_set_address(uint64_t, off_t) +section_size_type +Output_merge_string::finalize_merged_data() { this->stringpool_.set_string_offsets(); - for (typename Merged_strings::const_iterator p = - this->merged_strings_.begin(); - p != this->merged_strings_.end(); - ++p) - this->add_mapping(p->object, p->shndx, p->offset, p->length, - this->stringpool_.get_offset(p->string)); + for (typename Merged_strings_lists::const_iterator l = + this->merged_strings_lists_.begin(); + l != this->merged_strings_lists_.end(); + ++l) + { + section_offset_type last_input_offset = 0; + section_offset_type last_output_offset = 0; + for (typename Merged_strings::const_iterator p = + (*l)->merged_strings.begin(); + p != (*l)->merged_strings.end(); + ++p) + { + section_size_type length = p->offset - last_input_offset; + if (length > 0) + this->add_mapping((*l)->object, (*l)->shndx, last_input_offset, + length, last_output_offset); + last_input_offset = p->offset; + if (p->stringpool_key != 0) + last_output_offset = + this->stringpool_.get_offset_from_key(p->stringpool_key); + } + delete *l; + } - this->set_data_size(this->stringpool_.get_strtab_size()); + // Save some memory. This also ensures that this function will work + // if called twice, as may happen if Layout::set_segment_offsets + // finds a better alignment. + this->merged_strings_lists_.clear(); - // Save some memory. - this->merged_strings_.clear(); + return this->stringpool_.get_strtab_size(); +} + +template +void +Output_merge_string::set_final_data_size() +{ + const off_t final_data_size = this->finalize_merged_data(); + this->set_data_size(final_data_size); } // Write out a merged string section. @@ -528,6 +674,62 @@ Output_merge_string::do_write(Output_file* of) this->stringpool_.write(of, this->offset()); } +// Write a merged string section to a buffer. + +template +void +Output_merge_string::do_write_to_buffer(unsigned char* buffer) +{ + this->stringpool_.write_to_buffer(buffer, this->data_size()); +} + +// Return the name of the types of string to use with +// do_print_merge_stats. + +template +const char* +Output_merge_string::string_name() +{ + gold_unreachable(); + return NULL; +} + +template<> +const char* +Output_merge_string::string_name() +{ + return "strings"; +} + +template<> +const char* +Output_merge_string::string_name() +{ + return "16-bit strings"; +} + +template<> +const char* +Output_merge_string::string_name() +{ + return "32-bit strings"; +} + +// Print merge stats to stderr. + +template +void +Output_merge_string::do_print_merge_stats(const char* section_name) +{ + char buf[200]; + snprintf(buf, sizeof buf, "%s merged %s", section_name, this->string_name()); + fprintf(stderr, _("%s: %s input bytes: %zu\n"), + program_name, buf, this->input_size_); + fprintf(stderr, _("%s: %s input strings: %zu\n"), + program_name, buf, this->input_count_); + this->stringpool_.print_stats(buf); +} + // Instantiate the templates we need. template @@ -539,4 +741,22 @@ class Output_merge_string; template class Output_merge_string; +#if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG) +template +void +Object_merge_map::initialize_input_to_output_map<32>( + unsigned int shndx, + elfcpp::Elf_types<32>::Elf_Addr starting_address, + Unordered_map::Elf_Addr>*); +#endif + +#if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG) +template +void +Object_merge_map::initialize_input_to_output_map<64>( + unsigned int shndx, + elfcpp::Elf_types<64>::Elf_Addr starting_address, + Unordered_map::Elf_Addr>*); +#endif + } // End namespace gold.