Skip to content

Commit

Permalink
[Mach-O] Merge all S_CSTRING_LITERALS sections
Browse files Browse the repository at this point in the history 
__TEXT,__cstring is one of such sections.
  • Loading branch information
@rui314
rui314 committed Jul 7, 2022
1 parent a4cc7f3 commit b843b2f
Show file tree
Hide file tree
Showing 3 changed files with 115 additions and 111 deletions.
17 changes: 10 additions & 7 deletions macho/input-files.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -166,7 +166,7 @@ split_regular_sections(Context<E> &ctx, ObjectFile<E> &file) {

for (i64 i = 0; i < file.sections.size(); i++)
if (InputSection<E> *isec = file.sections[i].get())
if (!isec->hdr.match("__TEXT", "__cstring"))
if (isec->hdr.type != S_CSTRING_LITERALS)
vec[i].isec = isec;

// Find all symbols whose type is N_SECT.
Expand Down Expand Up @@ -226,12 +226,14 @@ template <typename E>
void ObjectFile<E>::split_subsections_via_symbols(Context<E> &ctx) {
sym_to_subsec.resize(mach_syms.size());

auto add = [&](InputSection<E> &isec, u32 offset, u32 size, u8 p2align) {
auto add = [&](InputSection<E> &isec, u32 offset, u32 size, u8 p2align,
bool is_cstring) {
Subsection<E> *subsec = new Subsection<E>{
.isec = isec,
.input_offset = offset,
.input_size = size,
.input_addr = (u32)(isec.hdr.addr + offset),
.is_cstring = is_cstring,
.p2align = p2align,
};

Expand All @@ -244,22 +246,22 @@ void ObjectFile<E>::split_subsections_via_symbols(Context<E> &ctx) {
InputSection<E> &isec = *info.isec;
for (SplitRegion &r : info.regions) {
if (!r.is_alt_entry)
add(isec, r.offset, r.size, isec.hdr.p2align);
add(isec, r.offset, r.size, isec.hdr.p2align, false);
if (r.symidx != -1)
sym_to_subsec[r.symidx] = subsections.back();
}
}

// Split __cstring section.
for (std::unique_ptr<InputSection<E>> &isec : sections) {
if (isec && isec->hdr.match("__TEXT", "__cstring")) {
if (isec && isec->hdr.type == S_CSTRING_LITERALS) {
std::string_view str = isec->contents;
size_t pos = 0;

while (pos < str.size()) {
size_t end = str.find('\0', pos);
if (end == str.npos)
Fatal(ctx) << *this << " corruupted __TEXT,__cstring";
Fatal(ctx) << *this << " corruupted cstring section: " << *isec;

end = str.find_first_not_of('\0', end);
if (end == str.npos)
Expand All @@ -268,7 +270,7 @@ void ObjectFile<E>::split_subsections_via_symbols(Context<E> &ctx) {
// A constant string in __cstring has no alignment info, so we
// need to infer it.
u8 p2align = std::min<u8>(isec->hdr.p2align, std::countr_zero(pos));
add(*isec, pos, end - pos, p2align);
add(*isec, pos, end - pos, p2align, true);
pos = end;
}
}
Expand Down Expand Up @@ -422,7 +424,8 @@ void ObjectFile<E>::parse_compact_unwind(Context<E> &ctx, MachSection &hdr) {
UnwindRecord<E> &dst = unwind_records[idx];

auto error = [&] {
Fatal(ctx) << *this << ": __compact_unwind: unsupported relocation: " << i;
Fatal(ctx) << *this << ": __compact_unwind: unsupported relocation: " << i
<< " " << *this->syms[r.idx];
};

if (r.is_pcrel || r.p2size != 3 || r.type)
Expand Down
205 changes: 104 additions & 101 deletions macho/main.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -287,105 +287,6 @@ static void claim_unresolved_symbols(Context<E> &ctx) {
});
}

template <typename E>
static void merge_cstring_sections(Context<E> &ctx) {
Timer t(ctx, "merge_cstring_sections");

struct Entry {
Entry(Subsection<E> *subsec) : owner(subsec) {}

Entry(const Entry &other) :
owner(other.owner.load()), p2align(other.p2align.load()) {}

std::atomic<Subsection<E> *> owner = nullptr;
std::atomic_uint8_t p2align = 0;
};

struct SubsecRef {
Subsection<E> &subsec;
u64 hash = 0;
Entry *ent = nullptr;
};

std::vector<std::vector<SubsecRef>> vec(ctx.objs.size());

// Estimate the number of unique strings.
HyperLogLog estimator;

tbb::parallel_for((i64)0, (i64)ctx.objs.size(), [&](i64 i) {
ObjectFile<E> *file = ctx.objs[i];
HyperLogLog e;

for (Subsection<E> *subsec : file->subsections) {
if (&subsec->isec.osec == ctx.cstring) {
std::string_view str = subsec->get_contents();
u64 h = hash_string(str);
vec[i].push_back({*subsec, h, nullptr});
estimator.insert(h);
}
}
estimator.merge(e);
});

// Create a hash map large enough to hold all strings.
ConcurrentMap<Entry> map(estimator.get_cardinality() * 3 / 2);

// Insert all strings into the hash table.
tbb::parallel_for((i64)0, (i64)ctx.objs.size(), [&](i64 i) {
ObjectFile<E> *file = ctx.objs[i];

for (i64 j = 0; j < vec[i].size(); j++) {
SubsecRef &ref = vec[i][j];
std::string_view s = ref.subsec.get_contents();
ref.ent = map.insert(s, ref.hash, {&ref.subsec}).first;

Subsection<E> *existing = ref.ent->owner;
while (existing->isec.file.priority < file->priority &&
!ref.ent->owner.compare_exchange_weak(existing, &ref.subsec));

update_maximum(ref.ent->p2align, ref.subsec.p2align.load());
}
});

// Decide who will become the owner for each subsection.
tbb::parallel_for((i64)0, (i64)ctx.objs.size(), [&](i64 i) {
for (i64 j = 0; j < vec[i].size(); j++) {
SubsecRef &ref = vec[i][j];
if (ref.ent->owner != &ref.subsec) {
ref.subsec.is_coalesced = true;
ref.subsec.replacer = ref.ent->owner;

static Counter counter("num_merged_strings");
counter++;
}
}
});

// Merge strings
tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
for (std::unique_ptr<InputSection<E>> &isec : file->sections)
if (isec)
for (Relocation<E> &r : isec->rels)
if (r.subsec && r.subsec->is_coalesced)
r.subsec = r.subsec->replacer;
});

auto replace = [&](InputFile<E> *file) {
for (Symbol<E> *sym : file->syms)
if (sym->subsec && sym->subsec->is_coalesced)
sym->subsec = sym->subsec->replacer;
};

tbb::parallel_for_each(ctx.objs, replace);
tbb::parallel_for_each(ctx.dylibs, replace);

tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
std::erase_if(file->subsections, [](Subsection<E> *subsec) {
return subsec->is_coalesced;
});
});
}

template <typename E>
static Chunk<E> *find_section(Context<E> &ctx, std::string_view segname,
std::string_view sectname) {
Expand Down Expand Up @@ -446,6 +347,109 @@ static void create_synthetic_chunks(Context<E> &ctx) {
sort(seg->chunks, compare_chunks<E>);
}

template <typename E>
static void uniquify_cstrings(Context<E> &ctx, OutputSection<E> &osec) {
Timer t(ctx, "uniquify_cstrings");

struct Entry {
Entry(Subsection<E> *subsec) : owner(subsec) {}

Entry(const Entry &other) :
owner(other.owner.load()), p2align(other.p2align.load()) {}

std::atomic<Subsection<E> *> owner = nullptr;
std::atomic_uint8_t p2align = 0;
};

struct SubsecRef {
Subsection<E> *subsec = nullptr;
u64 hash = 0;
Entry *ent = nullptr;
};

std::vector<SubsecRef> vec(osec.members.size());

// Estimate the number of unique strings.
tbb::enumerable_thread_specific<HyperLogLog> estimators;

tbb::parallel_for((i64)0, (i64)osec.members.size(), [&](i64 i) {
Subsection<E> *subsec = osec.members[i];
if (subsec->is_cstring) {
u64 h = hash_string(subsec->get_contents());
vec[i].subsec = subsec;
vec[i].hash = h;
estimators.local().insert(h);
}
});

HyperLogLog estimator;
for (HyperLogLog &e : estimators)
estimator.merge(e);

// Create a hash map large enough to hold all strings.
ConcurrentMap<Entry> map(estimator.get_cardinality() * 3 / 2);

// Insert all strings into the hash table.
tbb::parallel_for_each(vec, [&](SubsecRef &ref) {
if (ref.subsec) {
std::string_view s = ref.subsec->get_contents();
ref.ent = map.insert(s, ref.hash, {ref.subsec}).first;

Subsection<E> *existing = ref.ent->owner;
while (existing->isec.file.priority < ref.subsec->isec.file.priority &&
!ref.ent->owner.compare_exchange_weak(existing, ref.subsec));

update_maximum(ref.ent->p2align, ref.subsec->p2align.load());
}
});

// Decide who will become the owner for each subsection.
tbb::parallel_for_each(vec, [&](SubsecRef &ref) {
if (ref.subsec && ref.subsec != ref.ent->owner) {
ref.subsec->is_coalesced = true;
ref.subsec->replacer = ref.ent->owner;
}
});

static Counter counter("num_merged_strings");
counter += std::erase_if(osec.members, [](Subsection<E> *subsec) {
return subsec->is_coalesced;
});
}

template <typename E>
static void merge_cstring_sections(Context<E> &ctx) {
Timer t(ctx, "merge_cstring_sections");

for (Chunk<E> *chunk : ctx.chunks)
if (chunk->is_output_section && chunk->hdr.type == S_CSTRING_LITERALS)
uniquify_cstrings(ctx, *(OutputSection<E> *)chunk);

// Rewrite relocations and symbols.
tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
for (std::unique_ptr<InputSection<E>> &isec : file->sections)
if (isec)
for (Relocation<E> &r : isec->rels)
if (r.subsec && r.subsec->is_coalesced)
r.subsec = r.subsec->replacer;
});

auto replace = [&](InputFile<E> *file) {
for (Symbol<E> *sym : file->syms)
if (sym && sym->subsec && sym->subsec->is_coalesced)
sym->subsec = sym->subsec->replacer;
};

tbb::parallel_for_each(ctx.objs, replace);
tbb::parallel_for_each(ctx.dylibs, replace);

tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
std::erase_if(file->subsections, [](Subsection<E> *subsec) {
return subsec->is_coalesced;
});
});
}

template <typename E>
static void scan_unwind_info(Context<E> &ctx) {
tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
Expand Down Expand Up @@ -930,12 +934,11 @@ static int do_main(int argc, char **argv) {

claim_unresolved_symbols(ctx);

merge_cstring_sections(ctx);

if (ctx.arg.dead_strip)
dead_strip(ctx);

create_synthetic_chunks(ctx);
merge_cstring_sections(ctx);

for (ObjectFile<E> *file : ctx.objs)
file->check_duplicate_symbols(ctx);
Expand Down
4 changes: 1 addition & 3 deletions macho/mold.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -243,6 +243,7 @@ class Subsection {

std::atomic_uint8_t p2align = 0;
std::atomic_bool is_alive = true;
bool is_cstring : 1 = false;
bool is_coalesced : 1 = false;
bool added_to_osec : 1 = false;
};
Expand Down Expand Up @@ -821,11 +822,9 @@ struct Context {
text = OutputSection<E>::get_instance(*this, "__TEXT", "__text");
data = OutputSection<E>::get_instance(*this, "__DATA", "__data");
bss = OutputSection<E>::get_instance(*this, "__DATA", "__bss");
cstring = OutputSection<E>::get_instance(*this, "__TEXT", "__cstring");
common = OutputSection<E>::get_instance(*this, "__DATA", "__common");

bss->hdr.type = S_ZEROFILL;
cstring->hdr.type = S_CSTRING_LITERALS;
common->hdr.type = S_ZEROFILL;
}

Expand Down Expand Up @@ -959,7 +958,6 @@ struct Context {
OutputSection<E> *text = nullptr;
OutputSection<E> *data = nullptr;
OutputSection<E> *bss = nullptr;
OutputSection<E> *cstring = nullptr;
OutputSection<E> *common = nullptr;
};

Expand Down

0 comments on commit b843b2f

Please sign in to comment.