1
1
mirror of https://github.com/rui314/mold.git synced 2024-09-22 10:27:48 +03:00
mold/main.cc

1023 lines
28 KiB
C++
Raw Normal View History

2020-10-20 08:54:35 +03:00
#include "mold.h"
2020-10-02 07:28:26 +03:00
2020-11-03 14:49:30 +03:00
#include "llvm/BinaryFormat/Magic.h"
#include "llvm/Option/ArgList.h"
#include "llvm/Support/FileOutputBuffer.h"
2020-09-29 09:05:29 +03:00
#include <iostream>
2020-10-24 12:58:21 +03:00
using namespace llvm;
2020-10-21 05:28:43 +03:00
using namespace llvm::ELF;
2020-10-10 06:47:12 +03:00
using llvm::object::Archive;
2020-10-02 10:47:51 +03:00
using llvm::opt::InputArgList;
2020-10-02 07:28:26 +03:00
2020-11-08 02:39:13 +03:00
class MyTimer {
public:
MyTimer(StringRef name) {
timer = new Timer(name, name);
timer->startTimer();
}
MyTimer(StringRef name, llvm::TimerGroup &tg) {
timer = new Timer(name, name, tg);
timer->startTimer();
}
~MyTimer() { timer->stopTimer(); }
private:
llvm::Timer *timer;
};
2020-10-04 12:00:33 +03:00
//
// Command-line option processing
//
2020-10-02 07:28:26 +03:00
enum {
OPT_INVALID = 0,
#define OPTION(_1, _2, ID, _4, _5, _6, _7, _8, _9, _10, _11, _12) OPT_##ID,
#include "options.inc"
#undef OPTION
};
2020-10-02 10:47:51 +03:00
// Create prefix string literals used in Options.td
#define PREFIX(NAME, VALUE) const char *const NAME[] = VALUE;
#include "options.inc"
#undef PREFIX
// Create table mapping all options defined in Options.td
static const llvm::opt::OptTable::Info opt_info[] = {
#define OPTION(X1, X2, ID, KIND, GROUP, ALIAS, X7, X8, X9, X10, X11, X12) \
{X1, X2, X10, X11, OPT_##ID, llvm::opt::Option::KIND##Class, \
X9, X8, OPT_##GROUP, OPT_##ALIAS, X7, X12},
#include "options.inc"
#undef OPTION
};
class MyOptTable : llvm::opt::OptTable {
public:
MyOptTable() : OptTable(opt_info) {}
InputArgList parse(int argc, char **argv);
};
InputArgList MyOptTable::parse(int argc, char **argv) {
2020-11-01 02:55:13 +03:00
unsigned missing_index = 0;
unsigned missing_count = 0;
2020-10-02 10:47:51 +03:00
SmallVector<const char *, 256> vec(argv, argv + argc);
2020-11-01 02:55:13 +03:00
InputArgList args = this->ParseArgs(vec, missing_index, missing_count);
if (missing_count)
error(Twine(args.getArgString(missing_index)) + ": missing argument");
2020-10-02 10:47:51 +03:00
for (auto *arg : args.filtered(OPT_UNKNOWN))
error("unknown argument '" + arg->getAsString(args) + "'");
return args;
}
2020-10-04 12:00:33 +03:00
//
// Main
//
2020-10-14 13:36:06 +03:00
static std::vector<MemoryBufferRef> get_archive_members(MemoryBufferRef mb) {
2020-10-10 06:47:12 +03:00
std::unique_ptr<Archive> file =
CHECK(Archive::create(mb), mb.getBufferIdentifier() + ": failed to parse archive");
std::vector<MemoryBufferRef> vec;
Error err = Error::success();
for (const Archive::Child &c : file->children(err)) {
MemoryBufferRef mbref =
CHECK(c.getMemoryBufferRef(),
mb.getBufferIdentifier() +
": could not get the buffer for a child of the archive");
vec.push_back(mbref);
}
if (err)
error(mb.getBufferIdentifier() + ": Archive::children failed: " +
toString(std::move(err)));
2020-10-10 12:48:38 +03:00
file.release(); // leak
2020-10-10 06:47:12 +03:00
return vec;
}
2020-10-22 11:26:23 +03:00
static void read_file(std::vector<ObjectFile *> &files, StringRef path) {
2020-11-03 14:42:50 +03:00
auto mb_or_err = MemoryBuffer::getFile(path, -1, false);
if (auto ec = mb_or_err.getError())
error("cannot open " + path + ": " + ec.message());
2020-10-10 06:47:12 +03:00
2020-11-03 14:42:50 +03:00
std::unique_ptr<MemoryBuffer> &mb = *mb_or_err;
MemoryBufferRef mbref = mb->getMemBufferRef();
mb.release();
switch (identify_magic(mbref.getBuffer())) {
2020-10-10 06:47:12 +03:00
case file_magic::archive:
2020-11-03 14:42:50 +03:00
for (MemoryBufferRef member : get_archive_members(mbref))
2020-10-22 11:26:23 +03:00
files.push_back(new ObjectFile(member, path));
2020-10-10 06:47:12 +03:00
break;
case file_magic::elf_relocatable:
2020-11-04 09:39:31 +03:00
case file_magic::elf_shared_object:
2020-11-03 14:42:50 +03:00
files.push_back(new ObjectFile(mbref, ""));
2020-10-10 06:47:12 +03:00
break;
default:
error(path + ": unknown file type");
}
}
2020-10-28 07:42:05 +03:00
template <typename T>
static std::vector<ArrayRef<T>> split(const std::vector<T> &input, int unit) {
ArrayRef<T> arr(input);
std::vector<ArrayRef<T>> vec;
while (arr.size() >= unit) {
vec.push_back(arr.slice(0, unit));
arr = arr.slice(unit);
}
if (!arr.empty())
vec.push_back(arr);
return vec;
}
2020-11-07 15:53:21 +03:00
static void handle_mergeable_strings(std::vector<ObjectFile *> &files) {
2020-11-07 16:54:07 +03:00
static Counter counter("handle_merged_strings");
2020-11-07 16:07:19 +03:00
for (MergedSection *osec : MergedSection::instances)
counter.inc(osec->map.size());
2020-11-07 15:53:21 +03:00
// Resolve mergeable string pieces
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [](ObjectFile *file) {
2020-11-07 15:53:21 +03:00
for (InputSection *isec : file->mergeable_sections) {
for (StringPieceRef &ref : isec->pieces) {
2020-11-07 18:53:28 +03:00
InputSection *cur = ref.piece->isec;
2020-11-07 20:03:54 +03:00
while (!cur || cur->file->priority > isec->file->priority)
2020-11-07 15:53:21 +03:00
if (ref.piece->isec.compare_exchange_strong(cur, isec))
break;
}
}
});
// Calculate the total bytes of mergeable strings for each input section.
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [](ObjectFile *file) {
2020-11-07 15:53:21 +03:00
for (InputSection *isec : file->mergeable_sections) {
u32 offset = 0;
for (StringPieceRef &ref : isec->pieces) {
if (ref.piece->isec == isec) {
ref.piece->output_offset = offset;
2020-11-07 16:55:56 +03:00
offset += ref.piece->data.size() + 1;
2020-11-07 15:53:21 +03:00
}
}
isec->merged_size = offset;
}
});
// Assign each mergeable input section a unique index.
for (ObjectFile *file : files) {
for (InputSection *isec : file->mergeable_sections) {
MergedSection *osec = isec->merged_section;
isec->merged_offset = osec->shdr.sh_size;
osec->shdr.sh_size += isec->merged_size;
}
}
}
2020-10-27 18:21:41 +03:00
static void bin_sections(std::vector<ObjectFile *> &files) {
2020-10-28 08:06:35 +03:00
int unit = (files.size() + 127) / 128;
std::vector<ArrayRef<ObjectFile *>> slices = split(files, unit);
2020-10-26 07:36:56 +03:00
2020-11-08 04:05:59 +03:00
int num_osec = OutputSection::instances.size();
2020-10-28 08:22:25 +03:00
std::vector<std::vector<std::vector<InputSection *>>> groups(slices.size());
for (int i = 0; i < groups.size(); i++)
2020-11-08 04:05:59 +03:00
groups[i].resize(num_osec);
2020-10-28 08:06:35 +03:00
tbb::parallel_for(0, (int)slices.size(), [&](int i) {
for (ObjectFile *file : slices[i]) {
for (InputSection *isec : file->sections) {
if (!isec)
continue;
OutputSection *osec = isec->output_section;
2020-10-28 08:22:25 +03:00
groups[i][osec->idx].push_back(isec);
2020-10-28 08:06:35 +03:00
}
}
});
2020-11-08 04:05:59 +03:00
std::vector<int> sizes(num_osec);
2020-10-26 07:36:56 +03:00
2020-10-28 08:22:25 +03:00
for (ArrayRef<std::vector<InputSection *>> group : groups)
for (int i = 0; i < group.size(); i++)
sizes[i] += group[i].size();
2020-11-08 03:44:27 +03:00
2020-11-08 03:04:19 +03:00
tbb::parallel_for(0, (int)sizes.size(), [&](int i) {
2020-10-28 08:06:35 +03:00
OutputSection::instances[i]->sections.reserve(sizes[i]);
2020-11-08 03:04:19 +03:00
});
2020-10-28 08:06:35 +03:00
2020-11-08 04:05:59 +03:00
tbb::parallel_for(0, num_osec, [&](int j) {
for (int i = 0; i < groups.size(); i++) {
std::vector<InputSection *> &sections = OutputSection::instances[j]->sections;
sections.insert(sections.end(), groups[i][j].begin(), groups[i][j].end());
2020-10-28 08:06:35 +03:00
}
2020-11-08 04:05:59 +03:00
});
2020-10-26 07:36:56 +03:00
}
2020-10-22 17:19:48 +03:00
2020-10-26 08:18:00 +03:00
static void set_isec_offsets() {
2020-10-26 10:12:56 +03:00
#if 1
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(OutputSection::instances, [&](OutputSection *osec) {
2020-10-27 07:52:10 +03:00
if (osec->sections.empty())
return;
2020-10-28 07:42:05 +03:00
std::vector<ArrayRef<InputSection *>> slices = split(osec->sections, 100000);
std::vector<u64> size(slices.size());
std::vector<u32> alignments(slices.size());
2020-10-26 08:18:00 +03:00
2020-10-28 07:42:05 +03:00
tbb::parallel_for(0, (int)slices.size(), [&](int i) {
u64 off = 0;
u32 align = 1;
2020-10-26 10:12:35 +03:00
for (InputSection *isec : slices[i]) {
off = align_to(off, isec->shdr.sh_addralign);
isec->offset = off;
off += isec->shdr.sh_size;
align = std::max<u32>(align, isec->shdr.sh_addralign);
2020-10-26 10:12:35 +03:00
}
size[i] = off;
alignments[i] = align;
});
u32 align = *std::max_element(alignments.begin(), alignments.end());
2020-10-26 10:12:35 +03:00
std::vector<u64> start(slices.size());
2020-10-28 07:42:05 +03:00
for (int i = 1; i < slices.size(); i++)
2020-10-26 10:58:49 +03:00
start[i] = align_to(start[i - 1] + size[i], align);
2020-10-28 07:42:05 +03:00
tbb::parallel_for(1, (int)slices.size(), [&](int i) {
2020-10-26 10:12:35 +03:00
for (InputSection *isec : slices[i])
isec->offset += start[i];
});
osec->shdr.sh_size = start.back() + size.back();
2020-10-26 08:18:00 +03:00
osec->shdr.sh_addralign = align;
});
2020-10-26 10:12:56 +03:00
#else
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(OutputSection::instances, [&](OutputSection *osec) {
2020-10-27 07:52:10 +03:00
if (osec->sections.empty())
return;
u64 off = 0;
u32 align = 0;
2020-10-26 10:13:45 +03:00
for (InputSection *isec : osec->sections) {
off = align_to(off, isec->shdr.sh_addralign);
isec->offset = off;
off += isec->shdr.sh_size;
align = std::max<u32>(align, isec->shdr.sh_addralign);
2020-10-26 10:13:45 +03:00
}
osec->shdr.sh_size = off;
2020-10-26 15:57:25 +03:00
osec->shdr.sh_addralign = align;
});
2020-10-26 10:12:56 +03:00
#endif
2020-10-26 08:18:00 +03:00
}
2020-11-03 10:19:21 +03:00
static void scan_rels(ArrayRef<ObjectFile *> files) {
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [&](ObjectFile *file) { file->scan_relocations(); });
2020-11-06 06:01:52 +03:00
2020-11-03 10:18:30 +03:00
u32 got_offset = 0;
u32 gotplt_offset = 0;
u32 plt_offset = 0;
2020-11-04 14:11:32 +03:00
u32 relplt_offset = 0;
2020-11-06 06:32:38 +03:00
for (ObjectFile *file : files) {
file->got_offset = got_offset;
got_offset += file->num_got * 8;
2020-11-06 07:54:37 +03:00
file->gotplt_offset = gotplt_offset;
2020-11-06 06:32:38 +03:00
gotplt_offset += file->num_gotplt * 8;
2020-11-06 07:54:37 +03:00
file->plt_offset = plt_offset;
2020-11-06 06:32:38 +03:00
plt_offset += file->num_plt * 16;
2020-11-06 07:54:37 +03:00
file->relplt_offset = relplt_offset;
2020-11-06 06:32:38 +03:00
relplt_offset += file->num_relplt * sizeof(ELF64LE::Rela);
}
out::got->shdr.sh_size = got_offset;
out::gotplt->shdr.sh_size = gotplt_offset;
out::plt->shdr.sh_size = plt_offset;
out::relplt->shdr.sh_size = relplt_offset;
}
2020-11-06 06:42:03 +03:00
static void assign_got_offsets(ArrayRef<ObjectFile *> files) {
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [&](ObjectFile *file) {
2020-11-06 06:32:38 +03:00
u32 got_offset = file->got_offset;
u32 gotplt_offset = file->gotplt_offset;
u32 plt_offset = file->plt_offset;
u32 relplt_offset = file->relplt_offset;
2020-11-03 10:18:30 +03:00
for (Symbol *sym : file->symbols) {
if (sym->file != file)
continue;
2020-11-06 10:40:37 +03:00
u8 flags = sym->flags.load(std::memory_order_relaxed);
if (flags & Symbol::NEEDS_GOT) {
2020-11-03 10:18:30 +03:00
sym->got_offset = got_offset;
got_offset += 8;
}
2020-11-06 10:40:37 +03:00
if (flags & Symbol::NEEDS_GOTTP) {
2020-11-03 10:18:30 +03:00
sym->gottp_offset = got_offset;
got_offset += 8;
}
2020-11-06 10:40:37 +03:00
if (flags & Symbol::NEEDS_PLT) {
2020-11-04 14:11:32 +03:00
// Write a .got.plt entry
2020-11-03 10:18:30 +03:00
sym->gotplt_offset = gotplt_offset;
gotplt_offset += 8;
2020-11-04 14:11:32 +03:00
// Write a .plt entry
2020-11-03 10:18:30 +03:00
sym->plt_offset = plt_offset;
2020-11-06 06:42:03 +03:00
plt_offset += 16;
// Write a .rela.dyn entry
sym->relplt_offset = relplt_offset;
relplt_offset += sizeof(ELF64LE::Rela);
}
}
});
}
static void write_got(u8 *buf, ArrayRef<ObjectFile *> files) {
u8 *got = buf + out::got->shdr.sh_offset;
u8 *plt = buf + out::plt->shdr.sh_offset;
u8 *relplt = buf + out::relplt->shdr.sh_offset;
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [&](ObjectFile *file) {
2020-11-06 06:42:03 +03:00
for (Symbol *sym : file->symbols) {
if (sym->file != file)
continue;
2020-11-06 10:40:37 +03:00
u8 flags = sym->flags.load(std::memory_order_relaxed);
if (flags & Symbol::NEEDS_GOT)
2020-11-06 06:50:26 +03:00
*(u64 *)(got + sym->got_offset) = sym->get_addr();
2020-11-06 06:42:03 +03:00
2020-11-06 10:40:37 +03:00
if (flags & Symbol::NEEDS_GOTTP)
2020-11-06 06:50:26 +03:00
*(u64 *)(got + sym->gottp_offset) = sym->get_addr() - out::tls_end;
2020-11-06 06:42:03 +03:00
2020-11-06 10:40:37 +03:00
if (flags & Symbol::NEEDS_PLT) {
2020-11-06 06:42:03 +03:00
// Write a .plt entry
2020-11-04 14:11:32 +03:00
u64 S = out::gotplt->shdr.sh_addr + sym->gotplt_offset;
u64 P = out::plt->shdr.sh_addr + sym->plt_offset;
2020-11-06 06:42:03 +03:00
out::plt->write_entry(plt + sym->plt_offset, S - P - 6);
2020-11-04 14:11:32 +03:00
// Write a .rela.dyn entry
2020-11-06 06:42:03 +03:00
auto *rel = (ELF64LE::Rela *)(relplt + sym->relplt_offset);
2020-11-04 14:11:32 +03:00
rel->r_offset = out::gotplt->shdr.sh_addr + sym->gotplt_offset;
rel->setType(R_X86_64_IRELATIVE, false);
2020-11-06 06:50:26 +03:00
rel->r_addend = sym->get_addr();
2020-11-03 10:18:30 +03:00
}
}
2020-11-06 06:32:38 +03:00
});
2020-11-03 10:18:30 +03:00
}
2020-10-22 12:54:51 +03:00
// We want to sort output sections in the following order.
//
2020-10-22 17:19:48 +03:00
// alloc readonly data
// alloc readonly code
// alloc writable tdata
// alloc writable tbss
// alloc writable data
// alloc writable bss
// nonalloc
2020-10-29 12:31:06 +03:00
static int get_rank(const ELF64LE::Shdr shdr) {
bool alloc = shdr.sh_flags & SHF_ALLOC;
bool writable = shdr.sh_flags & SHF_WRITE;
bool exec = shdr.sh_flags & SHF_EXECINSTR;
bool tls = shdr.sh_flags & SHF_TLS;
2020-10-30 06:47:35 +03:00
bool nobits = shdr.sh_type == SHT_NOBITS;
2020-10-22 17:19:48 +03:00
return (alloc << 5) | (!writable << 4) | (!exec << 3) | (tls << 2) | !nobits;
2020-10-22 12:54:51 +03:00
}
2020-10-29 12:31:06 +03:00
static void sort_output_chunks(std::vector<OutputChunk *> &chunks) {
std::sort(chunks.begin(), chunks.end(), [](OutputChunk *a, OutputChunk *b) {
int x = get_rank(a->shdr);
int y = get_rank(b->shdr);
2020-10-22 13:03:31 +03:00
if (x != y)
return x > y;
// Tie-break to make output deterministic.
2020-10-25 07:17:10 +03:00
if (a->shdr.sh_flags != b->shdr.sh_flags)
return a->shdr.sh_flags < b->shdr.sh_flags;
if (a->shdr.sh_type != b->shdr.sh_type)
return a->shdr.sh_type < b->shdr.sh_type;
2020-10-22 13:03:31 +03:00
return a->name < b->name;
});
2020-10-22 12:54:51 +03:00
}
2020-10-22 14:01:10 +03:00
static std::vector<ELF64LE::Shdr *>
2020-11-03 14:33:16 +03:00
create_shdr(ArrayRef<OutputChunk *> output_chunks) {
2020-10-22 14:40:23 +03:00
static ELF64LE::Shdr null_entry = {};
2020-10-22 14:01:10 +03:00
std::vector<ELF64LE::Shdr *> vec;
vec.push_back(&null_entry);
2020-10-29 09:30:41 +03:00
int shndx = 1;
2020-10-27 11:44:05 +03:00
for (OutputChunk *chunk : output_chunks) {
if (!chunk->name.empty()) {
2020-10-25 07:17:10 +03:00
vec.push_back(&chunk->shdr);
2020-10-29 09:30:41 +03:00
chunk->shndx = shndx++;
2020-10-27 11:44:05 +03:00
}
}
2020-10-20 13:09:18 +03:00
return vec;
}
2020-11-03 14:26:22 +03:00
static u32 to_phdr_flags(u64 sh_flags) {
u32 ret = PF_R;
if (sh_flags & SHF_WRITE)
ret |= PF_W;
if (sh_flags & SHF_EXECINSTR)
ret |= PF_X;
return ret;
}
static std::vector<OutputPhdr::Entry>
2020-11-03 14:33:16 +03:00
create_phdr(ArrayRef<OutputChunk *> output_chunks) {
2020-11-03 14:26:22 +03:00
std::vector<OutputPhdr::Entry> entries;
auto add = [&](u32 type, u32 flags, u32 align, std::vector<OutputChunk *> members) {
ELF64LE::Phdr phdr = {};
phdr.p_type = type;
phdr.p_flags = flags;
phdr.p_align = align;
entries.push_back({phdr, members});
};
// Create a PT_PHDR for the program header itself.
add(PT_PHDR, PF_R, 8, {out::phdr});
// Create an PT_INTERP.
if (out::interp)
add(PT_INTERP, PF_R, 1, {out::interp});
// Create PT_LOAD segments.
bool first = true;
bool last_was_bss;
for (OutputChunk *chunk : output_chunks) {
if (!(chunk->shdr.sh_flags & SHF_ALLOC))
break;
u32 flags = to_phdr_flags(chunk->shdr.sh_flags);
bool this_is_bss =
(chunk->shdr.sh_type == SHT_NOBITS && !(chunk->shdr.sh_flags & SHF_TLS));
if (first) {
add(PT_LOAD, flags, PAGE_SIZE, {chunk});
last_was_bss = this_is_bss;
first = false;
continue;
}
if (entries.back().phdr.p_flags != flags || (last_was_bss && !this_is_bss))
add(PT_LOAD, flags, PAGE_SIZE, {chunk});
else
entries.back().members.push_back(chunk);
last_was_bss = this_is_bss;
}
// Create a PT_TLS.
for (int i = 0; i < output_chunks.size(); i++) {
if (output_chunks[i]->shdr.sh_flags & SHF_TLS) {
std::vector<OutputChunk *> vec = {output_chunks[i++]};
while (i < output_chunks.size() && (output_chunks[i]->shdr.sh_flags & SHF_TLS))
vec.push_back(output_chunks[i++]);
add(PT_TLS, to_phdr_flags(output_chunks[i]->shdr.sh_flags), 1, vec);
}
}
for (OutputPhdr::Entry &ent : entries)
for (OutputChunk *chunk : ent.members)
ent.phdr.p_align = std::max(ent.phdr.p_align, chunk->shdr.sh_addralign);
for (OutputPhdr::Entry &ent : entries)
if (ent.phdr.p_type == PT_LOAD)
ent.members.front()->starts_new_ptload = true;
return entries;
}
static u64 set_osec_offsets(ArrayRef<OutputChunk *> output_chunks) {
u64 fileoff = 0;
u64 vaddr = 0x200000;
2020-10-26 08:16:13 +03:00
for (OutputChunk *chunk : output_chunks) {
2020-10-30 05:40:38 +03:00
if (chunk->starts_new_ptload)
2020-10-26 08:16:13 +03:00
vaddr = align_to(vaddr, PAGE_SIZE);
2020-11-03 14:37:27 +03:00
bool is_bss = chunk->shdr.sh_type == SHT_NOBITS;
if (!is_bss) {
2020-10-30 05:45:10 +03:00
if (vaddr % PAGE_SIZE > fileoff % PAGE_SIZE)
fileoff += vaddr % PAGE_SIZE - fileoff % PAGE_SIZE;
else if (vaddr % PAGE_SIZE < fileoff % PAGE_SIZE)
fileoff = align_to(fileoff, PAGE_SIZE) + vaddr % PAGE_SIZE;
}
2020-10-30 05:40:38 +03:00
fileoff = align_to(fileoff, chunk->shdr.sh_addralign);
2020-10-26 08:16:13 +03:00
vaddr = align_to(vaddr, chunk->shdr.sh_addralign);
chunk->shdr.sh_offset = fileoff;
if (chunk->shdr.sh_flags & SHF_ALLOC)
chunk->shdr.sh_addr = vaddr;
2020-11-03 14:37:27 +03:00
if (!is_bss)
2020-11-03 14:13:03 +03:00
fileoff += chunk->shdr.sh_size;
2020-10-30 05:40:38 +03:00
2020-11-03 14:37:27 +03:00
bool is_tbss = is_bss && (chunk->shdr.sh_flags & SHF_TLS);
2020-10-30 05:40:38 +03:00
if (!is_tbss)
2020-11-03 14:13:03 +03:00
vaddr += chunk->shdr.sh_size;
2020-10-26 08:16:13 +03:00
}
return fileoff;
}
2020-11-04 08:23:39 +03:00
static void fix_synthetic_symbols(ArrayRef<OutputChunk *> output_chunks) {
auto start = [&](OutputChunk *chunk, Symbol *sym) {
if (sym) {
2020-11-04 08:41:40 +03:00
sym->shndx = chunk->shndx;
2020-11-06 06:52:16 +03:00
sym->value = chunk->shdr.sh_addr;
2020-11-04 08:23:39 +03:00
}
};
auto stop = [&](OutputChunk *chunk, Symbol *sym) {
if (sym) {
2020-11-04 08:41:40 +03:00
sym->shndx = chunk->shndx;
2020-11-06 06:52:16 +03:00
sym->value = chunk->shdr.sh_addr + chunk->shdr.sh_size;
2020-11-04 08:23:39 +03:00
}
};
// __bss_start
for (OutputChunk *chunk : output_chunks) {
if (chunk->name == ".bss" && !chunk->sections.empty()) {
start(chunk, out::__bss_start);
break;
}
}
// __ehdr_start
for (OutputChunk *chunk : output_chunks) {
if (chunk->shndx == 1) {
2020-11-04 08:41:40 +03:00
out::__ehdr_start->shndx = 1;
2020-11-06 06:52:16 +03:00
out::__ehdr_start->value = out::ehdr->shdr.sh_addr - chunk->shdr.sh_addr;
2020-11-04 08:23:39 +03:00
break;
}
}
// __rela_iplt_start and __rela_iplt_end
2020-11-05 15:47:07 +03:00
start(out::relplt, out::__rela_iplt_start);
stop(out::relplt, out::__rela_iplt_end);
2020-11-04 08:23:39 +03:00
// __{init,fini}_array_{start,end}
for (OutputChunk *chunk : output_chunks) {
switch (chunk->shdr.sh_type) {
case SHT_INIT_ARRAY:
start(chunk, out::__init_array_start);
stop(chunk, out::__init_array_end);
break;
case SHT_FINI_ARRAY:
start(chunk, out::__fini_array_start);
stop(chunk, out::__fini_array_end);
break;
}
}
// _end, end, _etext, etext, _edata and edata
for (OutputChunk *chunk : output_chunks) {
if (chunk->sections.empty())
continue;
if (chunk->shdr.sh_flags & SHF_ALLOC) {
stop(chunk, out::end);
stop(chunk, out::_end);
}
if (chunk->shdr.sh_flags & SHF_EXECINSTR) {
stop(chunk, out::etext);
stop(chunk, out::_etext);
}
if (chunk->shdr.sh_type != SHT_NOBITS && chunk->shdr.sh_flags & SHF_ALLOC) {
stop(chunk, out::edata);
stop(chunk, out::_edata);
}
}
// __start_ and __stop_ symbols
for (OutputChunk *chunk : output_chunks) {
if (!is_c_identifier(chunk->name))
continue;
start(chunk, Symbol::intern(("__start_" + chunk->name).str()));
stop(chunk, Symbol::intern(("__stop_" + chunk->name).str()));
}
}
2020-10-24 12:58:21 +03:00
static void unlink_async(tbb::task_group &tg, StringRef path) {
if (!sys::fs::exists(path) || !sys::fs::is_regular_file(path))
return;
int fd;
if (std::error_code ec = sys::fs::openFileForRead(path, fd))
return;
sys::fs::remove(path);
tg.run([=]() { close(fd); });
}
2020-11-03 14:29:24 +03:00
static FileOutputBuffer *open_output_file(u64 filesize) {
Expected<std::unique_ptr<FileOutputBuffer>> buf_or_err =
FileOutputBuffer::create(config.output, filesize, FileOutputBuffer::F_executable);
if (!buf_or_err)
error("failed to open " + config.output + ": " +
llvm::toString(buf_or_err.takeError()));
return std::move(*buf_or_err).release();
}
2020-10-29 16:32:55 +03:00
static void write_symtab(u8 *buf, std::vector<ObjectFile *> files) {
std::vector<u64> symtab_off(files.size() + 1);
std::vector<u64> strtab_off(files.size() + 1);
2020-11-04 01:14:49 +03:00
symtab_off[0] = sizeof(ELF64LE::Sym);
2020-10-27 18:21:41 +03:00
strtab_off[0] = 1;
2020-10-28 10:06:39 +03:00
for (int i = 1; i < files.size() + 1; i++) {
2020-10-27 18:21:41 +03:00
symtab_off[i] = symtab_off[i - 1] + files[i - 1]->local_symtab_size;
strtab_off[i] = strtab_off[i - 1] + files[i - 1]->local_strtab_size;
}
2020-10-28 10:01:14 +03:00
out::symtab->shdr.sh_info = symtab_off.back() / sizeof(ELF64LE::Sym);
2020-10-27 18:21:41 +03:00
tbb::parallel_for((size_t)0, files.size(),
[&](size_t i) {
files[i]->write_local_symtab(buf, symtab_off[i], strtab_off[i]);
});
2020-10-28 10:06:39 +03:00
symtab_off[0] = symtab_off.back();
strtab_off[0] = strtab_off.back();
2020-10-27 18:21:41 +03:00
2020-10-28 10:06:39 +03:00
for (int i = 1; i < files.size() + 1; i++) {
2020-10-27 18:21:41 +03:00
symtab_off[i] = symtab_off[i - 1] + files[i - 1]->global_symtab_size;
strtab_off[i] = strtab_off[i - 1] + files[i - 1]->global_strtab_size;
}
2020-11-03 14:01:45 +03:00
assert(symtab_off.back() == out::symtab->shdr.sh_size);
2020-11-03 14:03:05 +03:00
assert(strtab_off.back() == out::strtab->shdr.sh_size);
2020-10-27 18:21:41 +03:00
tbb::parallel_for((size_t)0, files.size(),
[&](size_t i) {
files[i]->write_global_symtab(buf, symtab_off[i], strtab_off[i]);
});
}
2020-10-30 07:47:51 +03:00
static int get_thread_count(InputArgList &args) {
if (auto *arg = args.getLastArg(OPT_thread_count)) {
int n;
if (!llvm::to_integer(arg->getValue(), n) || n <= 0)
error(arg->getSpelling() + ": expected a positive integer, but got '" +
arg->getValue() + "'");
return n;
}
return tbb::global_control::active_value(tbb::global_control::max_allowed_parallelism);
}
2020-10-23 06:09:27 +03:00
2020-10-30 07:47:51 +03:00
int main(int argc, char **argv) {
2020-10-14 13:59:51 +03:00
// Parse command line options
2020-10-02 10:47:51 +03:00
MyOptTable opt_table;
2020-10-09 15:10:12 +03:00
InputArgList args = opt_table.parse(argc - 1, argv + 1);
2020-10-02 10:47:51 +03:00
2020-10-30 07:47:51 +03:00
tbb::global_control tbb_cont(tbb::global_control::max_allowed_parallelism,
get_thread_count(args));
2020-11-03 12:02:28 +03:00
Counter::enabled = args.hasArg(OPT_stat);
2020-10-04 12:00:33 +03:00
if (auto *arg = args.getLastArg(OPT_o))
config.output = arg->getValue();
else
error("-o option is missing");
2020-10-29 06:24:54 +03:00
config.print_map = args.hasArg(OPT_print_map);
2020-11-04 12:47:13 +03:00
config.is_static = args.hasArg(OPT_static);
2020-10-29 06:24:54 +03:00
2020-11-05 02:31:32 +03:00
for (auto *arg : args.filtered(OPT_trace_symbol))
Symbol::intern(arg->getValue())->traced = true;
2020-10-09 14:47:45 +03:00
std::vector<ObjectFile *> files;
2020-11-07 17:00:01 +03:00
llvm::TimerGroup parse("parse", "parse");
2020-10-22 18:02:33 +03:00
llvm::TimerGroup before_copy("before_copy", "before_copy");
2020-11-07 18:16:23 +03:00
llvm::TimerGroup copy("copy", "copy");
2020-10-22 18:02:33 +03:00
2020-10-13 14:35:35 +03:00
// Open input files
2020-10-25 03:38:53 +03:00
{
2020-11-07 17:00:01 +03:00
MyTimer t("open", parse);
2020-10-25 03:38:53 +03:00
for (auto *arg : args)
if (arg->getOption().getID() == OPT_INPUT)
read_file(files, arg->getValue());
2020-11-07 17:00:01 +03:00
}
2020-10-25 03:38:53 +03:00
2020-11-07 17:00:01 +03:00
// Parse input files
{
MyTimer t("parse", parse);
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [](ObjectFile *file) { file->parse(); });
2020-10-25 03:38:53 +03:00
}
2020-10-18 13:17:44 +03:00
2020-11-07 12:06:09 +03:00
{
2020-11-07 17:00:01 +03:00
MyTimer t("merge", parse);
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [](ObjectFile *file) {
file->initialize_mergeable_sections();
});
2020-11-07 12:06:09 +03:00
}
2020-11-06 10:58:13 +03:00
Timer total_timer("total", "total");
total_timer.startTimer();
2020-10-18 13:05:28 +03:00
// Set priorities to files
2020-10-28 04:15:05 +03:00
int priority = 1;
for (ObjectFile *file : files)
2020-11-05 06:34:59 +03:00
if (!file->is_in_archive)
2020-10-28 04:15:05 +03:00
file->priority = priority++;
for (ObjectFile *file : files)
2020-11-05 06:34:59 +03:00
if (file->is_in_archive)
2020-10-28 04:15:05 +03:00
file->priority = priority++;
2020-10-18 13:05:28 +03:00
2020-10-18 15:03:51 +03:00
// Resolve symbols
2020-10-25 03:38:53 +03:00
{
2020-10-26 10:45:53 +03:00
MyTimer t("resolve_symbols", before_copy);
2020-10-28 13:27:23 +03:00
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [](ObjectFile *file) { file->resolve_symbols(); });
2020-10-28 13:27:23 +03:00
2020-11-05 06:20:16 +03:00
// Resolve symbols
2020-11-07 18:53:28 +03:00
std::vector<ObjectFile *> root;
2020-10-28 14:00:51 +03:00
for (ObjectFile *file : files)
2020-11-07 18:53:28 +03:00
if (file->is_alive)
root.push_back(file);
2020-10-28 14:00:51 +03:00
2020-11-05 06:20:16 +03:00
// Mark archive members we include into the final output.
2020-10-28 13:34:32 +03:00
tbb::parallel_do(
2020-11-07 18:53:28 +03:00
root,
2020-11-05 06:20:16 +03:00
[&](ObjectFile *file, tbb::parallel_do_feeder<ObjectFile *> &feeder) {
file->mark_live_archive_members(feeder);
2020-10-28 13:34:32 +03:00
});
2020-10-30 10:00:25 +03:00
2020-11-01 19:10:23 +03:00
// Eliminate unused archive members.
files.erase(std::remove_if(files.begin(), files.end(),
[](ObjectFile *file){ return !file->is_alive; }),
files.end());
// Convert weak symbols to absolute symbols with value 0.
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [](ObjectFile *file) {
file->hanlde_undefined_weak_symbols();
});
2020-10-25 03:38:53 +03:00
}
2020-10-19 15:50:33 +03:00
2020-11-05 03:24:47 +03:00
if (args.hasArg(OPT_trace))
for (ObjectFile *file : files)
llvm::outs() << toString(file) << "\n";
2020-10-19 15:50:33 +03:00
// Eliminate duplicate comdat groups.
2020-10-25 03:38:53 +03:00
{
MyTimer t("comdat", before_copy);
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [](ObjectFile *file) {
file->eliminate_duplicate_comdat_groups();
});
2020-10-25 03:38:53 +03:00
}
2020-10-10 06:47:12 +03:00
2020-11-07 14:29:06 +03:00
// Resolve mergeable strings
{
MyTimer t("resolve_strings", before_copy);
2020-11-07 15:53:21 +03:00
handle_mergeable_strings(files);
2020-11-07 14:29:06 +03:00
}
2020-10-27 06:50:25 +03:00
// Create .bss sections for common symbols.
{
MyTimer t("common", before_copy);
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files,
[](ObjectFile *file) { file->convert_common_symbols(); });
2020-10-27 06:50:25 +03:00
}
2020-10-26 05:34:26 +03:00
// Bin input sections into output sections
{
MyTimer t("bin_sections", before_copy);
2020-10-26 07:36:56 +03:00
bin_sections(files);
2020-10-26 05:34:26 +03:00
}
2020-10-23 04:27:11 +03:00
2020-10-29 12:19:10 +03:00
// Assign offsets within an output section to input sections.
2020-10-25 03:38:53 +03:00
{
2020-10-26 05:34:26 +03:00
MyTimer t("isec_offsets", before_copy);
2020-10-26 08:18:00 +03:00
set_isec_offsets();
2020-10-25 03:38:53 +03:00
}
2020-10-22 10:35:17 +03:00
2020-11-04 04:39:17 +03:00
std::vector<OutputChunk *> output_chunks;
2020-11-07 14:29:06 +03:00
2020-11-04 04:39:17 +03:00
for (OutputSection *osec : OutputSection::instances)
if (!osec->empty())
output_chunks.push_back(osec);
2020-11-07 15:53:21 +03:00
for (MergedSection *osec : MergedSection::instances)
2020-11-07 14:31:09 +03:00
if (osec->shdr.sh_size)
output_chunks.push_back(osec);
2020-11-04 04:39:17 +03:00
// Create a dummy file containing linker-synthesized symbols
// (e.g. `__bss_start`).
2020-11-04 08:49:30 +03:00
ObjectFile *internal_file = ObjectFile::create_internal_file(output_chunks);
internal_file->priority = priority++;
files.push_back(internal_file);
2020-11-04 04:39:17 +03:00
2020-10-29 12:19:10 +03:00
// Create linker-synthesized sections.
out::ehdr = new OutputEhdr;
out::phdr = new OutputPhdr;
out::shdr = new OutputShdr;
2020-11-04 12:47:13 +03:00
if (!config.is_static)
out::interp = new InterpSection;
2020-11-01 09:12:17 +03:00
out::got = new GotSection(".got");
out::gotplt = new GotSection(".got.plt");
2020-11-01 11:46:08 +03:00
out::plt = new PltSection;
2020-11-01 10:22:47 +03:00
out::relplt = new RelPltSection;
2020-10-29 12:19:10 +03:00
out::shstrtab = new ShstrtabSection;
out::symtab = new SymtabSection;
out::strtab = new StrtabSection;
2020-10-23 03:21:40 +03:00
// Scan relocations to fix the sizes of .got, .plt, .got.plt, .dynstr,
// .rela.dyn, .rela.plt.
2020-10-25 03:38:53 +03:00
{
2020-11-03 10:19:21 +03:00
MyTimer t("scan_rels", before_copy);
scan_rels(files);
2020-10-29 12:19:10 +03:00
}
2020-10-27 15:14:33 +03:00
// Compute .symtab and .strtab sizes
2020-10-27 14:24:28 +03:00
{
2020-10-27 14:58:28 +03:00
MyTimer t("symtab_size", before_copy);
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [](ObjectFile *file) { file->compute_symtab(); });
2020-10-27 14:34:45 +03:00
2020-10-27 14:58:28 +03:00
for (ObjectFile *file : files) {
2020-11-03 14:01:45 +03:00
out::symtab->shdr.sh_size += file->local_symtab_size + file->global_symtab_size;
2020-11-03 14:03:05 +03:00
out::strtab->shdr.sh_size += file->local_strtab_size + file->global_strtab_size;
2020-10-27 14:58:28 +03:00
}
2020-10-27 14:24:28 +03:00
}
2020-10-29 12:32:47 +03:00
// Add output sections.
2020-11-03 14:10:14 +03:00
if (out::got->shdr.sh_size)
2020-10-29 12:31:06 +03:00
output_chunks.push_back(out::got);
2020-11-03 14:10:14 +03:00
if (out::plt->shdr.sh_size)
2020-11-01 11:47:31 +03:00
output_chunks.push_back(out::plt);
2020-11-03 14:10:14 +03:00
if (out::gotplt->shdr.sh_size)
2020-11-01 09:27:04 +03:00
output_chunks.push_back(out::gotplt);
2020-11-03 14:10:14 +03:00
if (out::relplt->shdr.sh_size)
2020-11-01 11:52:11 +03:00
output_chunks.push_back(out::relplt);
2020-10-22 10:11:28 +03:00
2020-10-29 12:31:06 +03:00
sort_output_chunks(output_chunks);
// Add ELF header, program header and .interp to the output.
output_chunks.insert(output_chunks.begin(), out::ehdr);
output_chunks.insert(output_chunks.begin() + 1, out::phdr);
// output_chunks.insert(output_chunks.begin() + 2, out::interp);
2020-10-19 12:13:55 +03:00
2020-10-22 13:41:27 +03:00
// Add a string table for section names.
2020-10-22 16:43:23 +03:00
output_chunks.push_back(out::shstrtab);
2020-10-27 11:36:55 +03:00
// Add a section header.
output_chunks.push_back(out::shdr);
// Add .symtab and .strtab.
output_chunks.push_back(out::symtab);
output_chunks.push_back(out::strtab);
// Fix .shstrtab contents.
2020-10-22 14:53:20 +03:00
for (OutputChunk *chunk : output_chunks)
if (!chunk->name.empty())
2020-10-25 07:17:10 +03:00
chunk->shdr.sh_name = out::shstrtab->add_string(chunk->name);
2020-10-20 13:09:18 +03:00
2020-10-27 11:36:55 +03:00
// Create section header and program header contents.
2020-11-03 14:33:16 +03:00
out::shdr->set_entries(create_shdr(output_chunks));
out::phdr->set_entries(create_phdr(output_chunks));
2020-10-29 09:30:41 +03:00
out::symtab->shdr.sh_link = out::strtab->shndx;
2020-10-22 14:53:20 +03:00
2020-10-30 10:55:59 +03:00
// Assign offsets to output sections
u64 filesize = 0;
2020-10-25 03:38:53 +03:00
{
2020-10-26 08:36:39 +03:00
MyTimer t("osec_offset", before_copy);
2020-10-26 08:18:00 +03:00
filesize = set_osec_offsets(output_chunks);
2020-10-25 09:17:43 +03:00
}
2020-10-19 17:37:29 +03:00
2020-11-06 06:42:03 +03:00
// Assign symbols to GOT offsets
{
2020-11-06 06:42:50 +03:00
MyTimer t("assign_got_offsets", before_copy);
2020-11-06 06:42:03 +03:00
assign_got_offsets(files);
}
2020-11-03 10:51:28 +03:00
// Fix linker-synthesized symbol addresses.
2020-11-04 08:23:39 +03:00
fix_synthetic_symbols(output_chunks);
2020-11-01 07:05:51 +03:00
2020-11-06 06:50:26 +03:00
for (OutputChunk *chunk : output_chunks) {
ELF64LE::Shdr &shdr = chunk->shdr;
if (shdr.sh_flags & SHF_TLS)
out::tls_end = align_to(shdr.sh_addr + shdr.sh_size, shdr.sh_addralign);
}
2020-10-26 08:38:43 +03:00
2020-11-03 10:18:30 +03:00
tbb::task_group tg_unlink;
2020-10-25 03:38:53 +03:00
{
2020-10-26 08:35:13 +03:00
MyTimer t("unlink");
2020-11-03 10:18:30 +03:00
unlink_async(tg_unlink, config.output);
2020-10-25 03:38:53 +03:00
}
2020-10-24 12:58:21 +03:00
2020-10-20 03:20:52 +03:00
// Create an output file
2020-11-03 14:29:24 +03:00
FileOutputBuffer *output_buffer;
2020-10-20 03:20:52 +03:00
2020-10-30 05:40:38 +03:00
{
2020-11-07 20:15:29 +03:00
MyTimer t("open_output_file");
2020-11-03 14:29:24 +03:00
output_buffer = open_output_file(filesize);
2020-10-30 05:40:38 +03:00
}
2020-10-20 03:20:52 +03:00
2020-10-29 16:32:55 +03:00
u8 *buf = output_buffer->getBufferStart();
2020-10-20 03:20:52 +03:00
2020-11-06 04:12:05 +03:00
// Fill .symtab and .strtab
2020-11-06 03:12:49 +03:00
{
2020-11-07 18:16:23 +03:00
MyTimer t("write_symtab", copy);
2020-11-06 04:12:05 +03:00
write_symtab(buf, files);
}
// Copy input sections to the output file
{
2020-11-07 18:16:23 +03:00
MyTimer t("copy", copy);
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(output_chunks, [&](OutputChunk *chunk) {
chunk->copy_to(buf);
});
2020-10-25 03:38:53 +03:00
}
2020-10-20 04:13:46 +03:00
2020-11-06 06:42:03 +03:00
// Fill .plt, .got, got.plt and .rela.plt sections
{
2020-11-07 18:16:23 +03:00
MyTimer t("write_got", copy);
2020-11-06 06:42:03 +03:00
write_got(buf, files);
}
2020-11-07 16:54:07 +03:00
// Fill mergeable string sections
{
2020-11-07 18:16:23 +03:00
MyTimer t("write_merged_strings", copy);
2020-11-07 16:54:07 +03:00
2020-11-07 18:47:34 +03:00
tbb::parallel_for_each(files, [&](ObjectFile *file) {
2020-11-07 16:54:07 +03:00
for (InputSection *isec : file->mergeable_sections) {
MergedSection *osec = isec->merged_section;
u8 *base = buf + osec->shdr.sh_offset + isec->merged_offset;
for (StringPieceRef &ref : isec->pieces) {
StringPiece &piece = *ref.piece;
if (piece.isec == isec)
memcpy(base + piece.output_offset, piece.data.data(), piece.data.size());
}
}
});
}
2020-10-25 03:38:53 +03:00
{
2020-11-07 18:16:23 +03:00
MyTimer t("commit", copy);
2020-10-25 03:38:53 +03:00
if (auto e = output_buffer->commit())
error("failed to write to the output file: " + toString(std::move(e)));
}
2020-10-14 12:41:09 +03:00
2020-11-06 10:58:13 +03:00
total_timer.stopTimer();
2020-10-25 03:38:53 +03:00
{
2020-10-27 17:17:13 +03:00
MyTimer t("unlink_wait");
2020-11-03 10:18:30 +03:00
tg_unlink.wait();
2020-10-25 03:38:53 +03:00
}
2020-10-24 12:58:21 +03:00
2020-10-29 06:24:54 +03:00
if (config.print_map) {
MyTimer t("print_map");
2020-10-29 12:31:06 +03:00
print_map(files, output_chunks);
2020-10-29 06:24:54 +03:00
}
2020-10-28 13:34:32 +03:00
#if 0
2020-10-28 12:29:31 +03:00
for (ObjectFile *file : files)
for (InputSection *isec : file->sections)
if (isec)
llvm::outs() << toString(isec) << "\n";
2020-10-28 13:34:32 +03:00
#endif
2020-10-28 12:29:31 +03:00
2020-11-03 11:36:43 +03:00
// Show stat numbers
2020-11-03 11:54:40 +03:00
Counter num_input_sections("input_sections");
2020-11-03 11:36:43 +03:00
for (ObjectFile *file : files)
2020-11-03 11:54:40 +03:00
num_input_sections.inc(file->sections.size());
2020-11-03 11:36:43 +03:00
2020-11-03 11:43:06 +03:00
Counter num_output_chunks("output_chunks", output_chunks.size());
Counter num_files("files", files.size());
Counter filesize_counter("filesize", filesize);
2020-11-03 11:36:43 +03:00
2020-11-03 11:28:06 +03:00
Counter::print();
2020-10-18 10:21:17 +03:00
llvm::TimerGroup::printAll(llvm::outs());
2020-10-28 13:27:23 +03:00
llvm::outs().flush();
2020-10-18 10:21:17 +03:00
_exit(0);
2020-09-29 09:05:29 +03:00
}