// merge.h -- handle section merging for gold -*- C++ -*-
+// Copyright 2006, 2007, 2008 Free Software Foundation, Inc.
+
+// 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.
+
#ifndef GOLD_MERGE_H
#define GOLD_MERGE_H
#include <climits>
+#include <map>
+#include <vector>
#include "stringpool.h"
#include "output.h"
namespace gold
{
-// A general class for SHF_MERGE data, to hold functions shared by
-// fixed-size constant data and string data.
+class Merge_map;
-class Output_merge_base : public Output_section_data
+// 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:
- Output_merge_base(uint64_t entsize)
- : Output_section_data(1), merge_map_(), entsize_(entsize)
+ Object_merge_map()
+ : first_shnum_(-1U), first_map_(),
+ second_shnum_(-1U), second_map_(),
+ section_merge_maps_()
{ }
- // Return the output address for an input address.
+ ~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. MERGE_MAP is the map
+ // we are looking for, or NULL if we don't care. 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
- do_output_address(const Relobj* object, unsigned int shndx, off_t offset,
- uint64_t output_section_address, uint64_t* poutput) const;
+ get_output_offset(const Merge_map*, unsigned int shndx,
+ section_offset_type offset,
+ section_offset_type *output_offset);
- protected:
- // Return the entry size.
- uint64_t
- entsize() const
- { return this->entsize_; }
+ // Return whether this is the merge map for section SHNDX.
+ bool
+ is_merge_section_for(const Merge_map*, unsigned int shndx);
- // Add a mapping from an OFFSET in input section SHNDX in object
- // OBJECT to an OUTPUT_OFFSET in the output section.
+ // Initialize an mapping from input offsets to output addresses for
+ // section SHNDX. STARTING_ADDRESS is the output address of the
+ // merged section.
+ template<int size>
void
- add_mapping(Relobj* object, unsigned int shndx, off_t offset,
- off_t output_offset);
+ 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>*);
private:
- // We build a mapping from OBJECT/SHNDX/OFFSET to an offset in the
- // output section.
- struct Merge_key
+ // 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
{
- const Relobj* object;
- unsigned int shndx;
- off_t offset;
+ // 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;
};
- struct Merge_key_less
+ // A less-than comparison routine for Input_merge_entry.
+ struct Input_merge_compare
{
bool
- operator()(const Merge_key&, const Merge_key&) const;
+ operator()(const Input_merge_entry& i1, const Input_merge_entry& i2) const
+ { return i1.input_offset < i2.input_offset; }
};
- typedef std::map<Merge_key, off_t, Merge_key_less> Merge_map;
+ // A list of entries for a particular input section.
+ struct Input_merge_map
+ {
+ typedef std::vector<Input_merge_entry> Entries;
+
+ // 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.
+ Entries 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_;
+};
+
+// This class manages mappings from input sections to offsets in an
+// output section. This is used where input sections are merged. The
+// actual data is stored in fields in Object.
+
+class Merge_map
+{
+ public:
+ Merge_map()
+ { }
+
+ // Add a mapping for the bytes from OFFSET to OFFSET + LENGTH in the
+ // input section SHNDX in object OBJECT to OUTPUT_OFFSET in the
+ // output section. An OUTPUT_OFFSET of -1 means that the bytes are
+ // discarded. OUTPUT_OFFSET is not the offset from the start of the
+ // output section, it is the offset from the start of the merged
+ // data within the output section.
+ void
+ add_mapping(Relobj* object, unsigned int shndx,
+ section_offset_type offset, section_size_type length,
+ section_offset_type output_offset);
+
+ // 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 will be
+ // -1 if the bytes are not being copied to the output. This returns
+ // true if the mapping is known, false otherwise. This returns the
+ // value stored by add_mapping, namely the offset from the start of
+ // the merged data within the output section.
+ bool
+ get_output_offset(const Relobj* object, unsigned int shndx,
+ section_offset_type offset,
+ section_offset_type *output_offset) const;
+
+ // Return whether this is the merge mapping for section SHNDX in
+ // OBJECT. This should return true when get_output_offset would
+ // return true for some input offset.
+ bool
+ is_merge_section_for(const Relobj* object, unsigned int shndx) const;
+};
+// A general class for SHF_MERGE data, to hold functions shared by
+// fixed-size constant data and string data.
+
+class Output_merge_base : public Output_section_data
+{
+ public:
+ Output_merge_base(uint64_t entsize, uint64_t addralign)
+ : Output_section_data(addralign), merge_map_(), entsize_(entsize)
+ { }
+
+ protected:
+ // Return the output offset for an input offset.
+ bool
+ do_output_offset(const Relobj* object, unsigned int shndx,
+ section_offset_type offset,
+ section_offset_type* poutput) const;
+
+ // Return whether this is the merge section for an input section.
+ bool
+ do_is_merge_section_for(const Relobj*, unsigned int shndx) const;
+
+ // Return the entry size.
+ uint64_t
+ entsize() const
+ { return this->entsize_; }
+
+ // Add a mapping from an OFFSET in input section SHNDX in object
+ // OBJECT to an OUTPUT_OFFSET in the output section. OUTPUT_OFFSET
+ // is the offset from the start of the merged data in the output
+ // section.
+ void
+ add_mapping(Relobj* object, unsigned int shndx, section_offset_type offset,
+ section_size_type length, section_offset_type output_offset)
+ {
+ this->merge_map_.add_mapping(object, shndx, offset, length, output_offset);
+ }
+
+ private:
// A mapping from input object/section/offset to offset in output
// section.
Merge_map merge_map_;
-
// The entry size. For fixed-size constants, this is the size of
// the constants. For strings, this is the size of a character.
uint64_t entsize_;
class Output_merge_data : public Output_merge_base
{
public:
- Output_merge_data(uint64_t entsize)
- : Output_merge_base(entsize), p_(NULL), len_(0), alc_(0),
+ Output_merge_data(uint64_t entsize, uint64_t addralign)
+ : Output_merge_base(entsize, addralign), p_(NULL), len_(0), alc_(0),
+ input_count_(0),
hashtable_(128, Merge_data_hash(this), Merge_data_eq(this))
{ }
+ protected:
// Add an input section.
bool
do_add_input_section(Relobj* object, unsigned int shndx);
// Set the final data size.
void
- do_set_address(uint64_t, off_t);
+ set_final_data_size();
// Write the data to the file.
void
do_write(Output_file*);
+ // Write the data to a buffer.
+ void
+ do_write_to_buffer(unsigned char*);
+
+ // Print merge stats to stderr.
+ void
+ do_print_merge_stats(const char* section_name);
+
private:
// We build a hash table of the fixed-size constants. Each constant
// is stored as a pointer into the section data we are accumulating.
// A key in the hash table. This is an offset in the section
// contents we are building.
- typedef off_t Merge_data_key;
+ typedef section_offset_type Merge_data_key;
// Compute the hash code. To do this we need a pointer back to the
// object holding the data.
const unsigned char*
constant(Merge_data_key k) const
{
- gold_assert(k >= 0 && k < this->len_);
+ gold_assert(k >= 0 && k < static_cast<section_offset_type>(this->len_));
return this->p_ + k;
}
// The accumulated data.
unsigned char* p_;
// The length of the accumulated data.
- off_t len_;
+ section_size_type len_;
// The size of the allocated buffer.
- size_t alc_;
+ section_size_type alc_;
+ // The number of entries seen in input files.
+ size_t input_count_;
// The hash table.
Merge_data_hashtable hashtable_;
};
class Output_merge_string : public Output_merge_base
{
public:
- Output_merge_string()
- : Output_merge_base(sizeof(Char_type)), stringpool_(), merged_strings_()
- { this->stringpool_.set_no_zero_null(); }
+ Output_merge_string(uint64_t addralign)
+ : Output_merge_base(sizeof(Char_type), addralign), stringpool_(),
+ merged_strings_(), input_count_(0)
+ {
+ gold_assert(addralign <= sizeof(Char_type));
+ this->stringpool_.set_no_zero_null();
+ }
+ protected:
// Add an input section.
bool
do_add_input_section(Relobj* object, unsigned int shndx);
+ // Do all the final processing after the input sections are read in.
+ // Returns the final data size.
+ section_size_type
+ finalize_merged_data();
+
// Set the final data size.
void
- do_set_address(uint64_t, off_t);
+ set_final_data_size();
// Write the data to the file.
void
do_write(Output_file*);
+ // Write the data to a buffer.
+ void
+ do_write_to_buffer(unsigned char*);
+
+ // Print merge stats to stderr.
+ void
+ do_print_merge_stats(const char* section_name);
+
+ // Writes the stringpool to a buffer.
+ void
+ stringpool_to_buffer(unsigned char* buffer, section_size_type buffer_size)
+ { this->stringpool_.write_to_buffer(buffer, buffer_size); }
+
+ // Clears all the data in the stringpool, to save on memory.
+ void
+ clear_stringpool()
+ { this->stringpool_.clear(); }
+
private:
+ // The name of the string type, for stats.
+ const char*
+ string_name();
+
// As we see input sections, we build a mapping from object, section
// index and offset to strings.
struct Merged_string
// The input section in the input object.
unsigned int shndx;
// The offset in the input section.
- off_t offset;
+ section_offset_type offset;
// The string itself, a pointer into a Stringpool.
const Char_type* string;
-
- Merged_string(Relobj *objecta, unsigned int shndxa, off_t offseta,
- const Char_type* stringa)
- : object(objecta), shndx(shndxa), offset(offseta), string(stringa)
+ // The length of the string in bytes, including the null terminator.
+ size_t length;
+ // The key in the Stringpool.
+ Stringpool::Key stringpool_key;
+
+ Merged_string(Relobj *objecta, unsigned int shndxa,
+ section_offset_type offseta, const Char_type* stringa,
+ size_t lengtha, Stringpool::Key stringpool_keya)
+ : object(objecta), shndx(shndxa), offset(offseta), string(stringa),
+ length(lengtha), stringpool_key(stringpool_keya)
{ }
};
// Map from a location in an input object to an entry in the
// Stringpool.
Merged_strings merged_strings_;
+ // The number of entries seen in input files.
+ size_t input_count_;
};
} // End namespace gold.
projects / binutils.git / blobdiff