// merge.cc -- handle section merging for gold
-// Copyright 2006, 2007 Free Software Foundation, Inc.
+// Copyright (C) 2006-2018 Free Software Foundation, Inc.
// Written by Ian Lance Taylor <iant@google.com>.
// This file is part of gold.
#include <algorithm>
#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. OUTPUT_OFFSET is relative to the start of the merged
- // data in the output section.
- void
- add_mapping(const Merge_map*, unsigned int shndx, section_offset_type offset,
- section_size_type length, section_offset_type 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. *OUTPUT_OFFSET is
- // relative to the start of the merged data in the output section.
- bool
- get_output_offset(const Merge_map*, unsigned int shndx,
- section_offset_type offset,
- section_offset_type *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.
- section_offset_type input_offset;
- // The length.
- section_size_type length;
- // The offset in the output section.
- section_offset_type 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 input section.
- struct Input_merge_map
- {
- // We store these with the Relobj, and we look them up by input
- // section. It is possible to have two different merge maps
- // associated with a single output section. For example, this
- // happens routinely with .rodata, when merged string constants
- // and merged fixed size constants are both put into .rodata. The
- // output offset that we store is not the offset from the start of
- // the output section; it is the offset from the start of the
- // merged data in the output section. That means that the caller
- // is going to add the offset of the merged data within the output
- // section, which means that the caller needs to know which set of
- // merged data it found the entry in. So it's not enough to find
- // this data based on the input section and the output section; we
- // also have to find it based on a set of merged data in the
- // output section. In order to verify that we are looking at the
- // right data, we store a pointer to the Merge_map here, and we
- // pass in a pointer when looking at the data. If we are asked to
- // look up information for a different Merge_map, we report that
- // we don't have it, rather than trying a lookup and returning an
- // answer which will receive the wrong offset.
- const Merge_map* merge_map;
- // The list of mappings.
- std::vector<Input_merge_entry> 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<unsigned int, Input_merge_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.
// Get the Input_merge_map to use for an input section, or NULL.
-Object_merge_map::Input_merge_map*
-Object_merge_map::get_input_merge_map(unsigned int shndx)
+const Object_merge_map::Input_merge_map*
+Object_merge_map::get_input_merge_map(unsigned int shndx) const
{
gold_assert(shndx != -1U);
- if (shndx == this->first_shnum_)
- return &this->first_map_;
- if (shndx == this->second_shnum_)
- return &this->second_map_;
- Section_merge_maps::const_iterator p = this->section_merge_maps_.find(shndx);
- if (p != this->section_merge_maps_.end())
- return p->second;
+ const Section_merge_maps &maps = this->section_merge_maps_;
+ for (Section_merge_maps::const_iterator i = maps.begin(), e = maps.end();
+ i != e; ++i)
+ {
+ if (i->first == shndx)
+ return i->second;
+ }
return NULL;
}
// Get or create the Input_merge_map to use for an input section.
Object_merge_map::Input_merge_map*
-Object_merge_map::get_or_make_input_merge_map(const Merge_map* merge_map,
- unsigned int shndx)
-{
+Object_merge_map::get_or_make_input_merge_map(
+ const Output_section_data* output_data, unsigned int shndx) {
Input_merge_map* map = this->get_input_merge_map(shndx);
if (map != NULL)
{
// For a given input section in a given object, every mapping
// must be done with the same Merge_map.
- gold_assert(map->merge_map == merge_map);
+ gold_assert(map->output_data == output_data);
return map;
}
- // We need to create a new entry.
- if (this->first_shnum_ == -1U)
- {
- this->first_shnum_ = shndx;
- this->first_map_.merge_map = merge_map;
- return &this->first_map_;
- }
- if (this->second_shnum_ == -1U)
- {
- this->second_shnum_ = shndx;
- this->second_map_.merge_map = merge_map;
- return &this->second_map_;
- }
-
Input_merge_map* new_map = new Input_merge_map;
- new_map->merge_map = merge_map;
- this->section_merge_maps_[shndx] = new_map;
+ new_map->output_data = output_data;
+ Section_merge_maps &maps = this->section_merge_maps_;
+ maps.push_back(std::make_pair(shndx, new_map));
return new_map;
}
// Add a mapping.
void
-Object_merge_map::add_mapping(const Merge_map* merge_map, unsigned int shndx,
+Object_merge_map::add_mapping(const Output_section_data* output_data,
+ unsigned int shndx,
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);
+ Input_merge_map* map = this->get_or_make_input_merge_map(output_data, shndx);
+ map->add_mapping(input_offset, length, output_offset);
+}
+void
+Object_merge_map::Input_merge_map::add_mapping(
+ section_offset_type input_offset, section_size_type length,
+ section_offset_type output_offset) {
// Try to merge the new entry in the last one we saw.
- if (!map->entries.empty())
+ if (!this->entries.empty())
{
- Input_merge_entry& entry(map->entries.back());
+ Input_merge_entry& entry(this->entries.back());
// Use section_size_type to avoid signed/unsigned warnings.
section_size_type input_offset_u = input_offset;
gold_assert(input_offset < entry.input_offset);
gold_assert(input_offset_u + length
<= static_cast<section_size_type>(entry.input_offset));
- map->sorted = false;
+ this->sorted = false;
}
else if (entry.input_offset + entry.length == input_offset_u
&& (output_offset == -1
entry.input_offset = input_offset;
entry.length = length;
entry.output_offset = output_offset;
- map->entries.push_back(entry);
+ this->entries.push_back(entry);
}
// Get the output offset for an input address.
-inline bool
-Object_merge_map::get_output_offset(const Merge_map* merge_map,
- unsigned int shndx,
+bool
+Object_merge_map::get_output_offset(unsigned int shndx,
section_offset_type input_offset,
- section_offset_type *output_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)
return false;
if (!map->sorted)
Input_merge_entry entry;
entry.input_offset = input_offset;
std::vector<Input_merge_entry>::const_iterator p =
- std::lower_bound(map->entries.begin(), map->entries.end(),
+ std::upper_bound(map->entries.begin(), map->entries.end(),
entry, Input_merge_compare());
- if (p == map->entries.end() || p->input_offset > input_offset)
- {
- if (p == map->entries.begin())
- return false;
- --p;
- gold_assert(p->input_offset <= input_offset);
- }
+ if (p == map->entries.begin())
+ return false;
+ --p;
+ gold_assert(p->input_offset <= input_offset);
if (input_offset - p->input_offset
>= static_cast<section_offset_type>(p->length))
return true;
}
-// Class Merge_map.
+// Return whether this is the merge map for section SHNDX.
+
+const Output_section_data*
+Object_merge_map::find_merge_section(unsigned int shndx) const {
+ const Object_merge_map::Input_merge_map* map =
+ this->get_input_merge_map(shndx);
+ if (map == NULL)
+ return NULL;
+ return map->output_data;
+}
-// Add a mapping for the bytes from OFFSET to OFFSET + LENGTH in input
-// section SHNDX in object OBJECT to an OUTPUT_OFFSET in merged data
-// in an output section.
+// Initialize a mapping from input offsets to output addresses.
+template<int size>
void
-Merge_map::add_mapping(Relobj* object, unsigned int shndx,
- section_offset_type offset, section_size_type length,
- section_offset_type output_offset)
+Object_merge_map::initialize_input_to_output_map(
+ unsigned int shndx,
+ typename elfcpp::Elf_types<size>::Elf_Addr starting_address,
+ Unordered_map<section_offset_type,
+ typename elfcpp::Elf_types<size>::Elf_Addr>* initialize_map)
{
- Object_merge_map* object_merge_map = object->merge_map();
- if (object_merge_map == NULL)
- {
- object_merge_map = new Object_merge_map();
- object->set_merge_map(object_merge_map);
- }
-
- object_merge_map->add_mapping(this, shndx, offset, length, output_offset);
-}
+ Input_merge_map* map = this->get_input_merge_map(shndx);
+ gold_assert(map != NULL);
-// 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 merged data in the output
-// section. This returns true if the mapping is known, false
-// otherwise.
+ 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);
-bool
-Merge_map::get_output_offset(const Relobj* object, unsigned int shndx,
- section_offset_type offset,
- section_offset_type* output_offset) const
-{
- Object_merge_map* object_merge_map = object->merge_map();
- if (object_merge_map == NULL)
- return false;
- return object_merge_map->get_output_offset(this, shndx, offset,
- output_offset);
+ 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 Output_merge_base.
section_offset_type offset,
section_offset_type* poutput) const
{
- return this->merge_map_.get_output_offset(object, shndx, offset, poutput);
+ return object->merge_output_offset(shndx, offset, poutput);
+}
+
+// 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<Input_sections::iterator, bool> 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.
Output_merge_data::do_add_input_section(Relobj* object, unsigned int shndx)
{
section_size_type len;
- const unsigned char* p = object->section_contents(shndx, &len, false);
+ bool is_new;
+ const unsigned char* p = object->decompressed_section_contents(shndx, &len,
+ &is_new);
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;
+ Object_merge_map* merge_map = object->get_or_create_merge_map();
+ Object_merge_map::Input_merge_map* input_merge_map =
+ merge_map->get_or_make_input_merge_map(this, shndx);
+
for (section_size_type i = 0; i < len; i += entsize, p += entsize)
{
// Add the constant to the section contents. If we find that it
}
// Record the offset of this constant in the output section.
- this->add_mapping(object, shndx, i, entsize, k);
+ input_merge_map->add_mapping(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;
}
{
// Release the memory we don't need.
this->p_ = static_cast<unsigned char*>(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_);
}
Output_merge_string<Char_type>::do_add_input_section(Relobj* object,
unsigned int shndx)
{
- section_size_type 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<const Char_type*>(pdata);
- const Char_type* pend = p + len;
+ 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.
section_size_type i = 0;
- while (i < len)
+
+ // 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<uintptr_t>(pdata)
+ & (this->addralign() - 1));
+ bool has_misaligned_strings = false;
+
+ while (p < pend)
{
- const Char_type* pl;
- for (pl = p; *pl != 0; ++pl)
- {
- if (pl >= pend)
- {
- object->error(_("entry in mergeable string section "
- "not null terminated"));
- break;
- }
- }
+ size_t len = p < pend0 ? string_length(p) : pend - p;
- Stringpool::Key key;
- const Char_type* str = this->stringpool_.add_with_length(p, pl - p, true,
- &key);
+ // Within merge input section each string must be aligned.
+ if (len != 0
+ && ((reinterpret_cast<uintptr_t>(p) & (this->addralign() - 1))
+ != init_align_modulo))
+ has_misaligned_strings = true;
- section_size_type bytelen_with_null = ((pl - p) + 1) * sizeof(Char_type);
- this->merged_strings_.push_back(Merged_string(object, shndx, i, str,
- bytelen_with_null, key));
+ Stringpool::Key key;
+ this->stringpool_.add_with_length(p, len, true, &key);
- p = pl + 1;
- i += bytelen_with_null;
- ++count;
+ merged_strings.push_back(Merged_string(i, key));
+ p += len + 1;
+ 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;
}
{
this->stringpool_.set_string_offsets();
- for (typename Merged_strings::const_iterator p =
- this->merged_strings_.begin();
- p != this->merged_strings_.end();
- ++p)
+ for (typename Merged_strings_lists::const_iterator l =
+ this->merged_strings_lists_.begin();
+ l != this->merged_strings_lists_.end();
+ ++l)
{
- section_offset_type offset =
- this->stringpool_.get_offset_from_key(p->stringpool_key);
- this->add_mapping(p->object, p->shndx, p->offset, p->length, offset);
+ section_offset_type last_input_offset = 0;
+ section_offset_type last_output_offset = 0;
+ Relobj *object = (*l)->object;
+ Object_merge_map* merge_map = object->get_or_create_merge_map();
+ Object_merge_map::Input_merge_map* input_merge_map =
+ merge_map->get_or_make_input_merge_map(this, (*l)->shndx);
+
+ 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)
+ input_merge_map->add_mapping(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;
}
- // Save some memory.
- this->merged_strings_.clear();
+ // 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();
return this->stringpool_.get_strtab_size();
}
{
char buf[200];
snprintf(buf, sizeof buf, "%s merged %s", section_name, this->string_name());
- fprintf(stderr, _("%s: %s input: %zu\n"),
+ 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);
}
template
class Output_merge_string<uint32_t>;
+#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<section_offset_type, elfcpp::Elf_types<32>::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<section_offset_type, elfcpp::Elf_types<64>::Elf_Addr>*);
+#endif
+
} // End namespace gold.