got_(NULL), plt_(NULL), iplt_(NULL), brlt_section_(NULL),
glink_(NULL), rela_dyn_(NULL), copy_relocs_(elfcpp::R_POWERPC_COPY),
dynbss_(NULL), tlsld_got_offset_(-1U),
- stub_tables_(), branch_lookup_table_(), branch_info_()
+ stub_tables_(), branch_lookup_table_(), branch_info_(),
+ plt_thread_safe_(false)
{
}
}
}
+ bool
+ plt_thread_safe() const
+ { return this->plt_thread_safe_; }
+
private:
// The class which scans relocations.
typedef std::vector<Branch_info> Branches;
Branches branch_info_;
+
+ bool plt_thread_safe_;
};
template<>
if (gsym->is_from_dynobj())
return true;
- // If we are generating a shared object, and gsym symbol is
+ // If we are generating a shared object, and this symbol is
// undefined or preemptible, we need to use the PLT entry.
if (parameters->options().shared()
&& (gsym->is_undefined() || gsym->is_preemptible()))
return true;
- // If gsym is a call to a weak undefined symbol, we need to use
+ // If this is a call to a weak undefined symbol, we need to use
// the PLT entry; the symbol may be defined by a library loaded
// at runtime.
if ((flags & Symbol::FUNCTION_CALL) && gsym->is_weak_undefined())
symtab_(symtab), layout_(layout),
header_ent_cnt_(size == 32 ? 3 : 1),
header_index_(size == 32 ? 0x2000 : 0)
- {}
+ { }
class Got_entry;
{
unsigned int prev_brlt_size = 0;
if (pass == 1)
- this->group_sections(layout, task);
- else
{
- prev_brlt_size = this->branch_lookup_table_.size();
- this->branch_lookup_table_.clear();
- for (typename Stub_tables::iterator p = this->stub_tables_.begin();
- p != this->stub_tables_.end();
- ++p)
+ bool thread_safe = parameters->options().plt_thread_safe();
+ if (size == 64 && !parameters->options().user_set_plt_thread_safe())
{
- (*p)->clear_long_branch_stubs();
+ const char* const thread_starter[] =
+ {
+ "pthread_create",
+ /* libstdc++ */
+ "_ZNSt6thread15_M_start_threadESt10shared_ptrINS_10_Impl_baseEE",
+ /* librt */
+ "aio_init", "aio_read", "aio_write", "aio_fsync", "lio_listio",
+ "mq_notify", "create_timer",
+ /* libanl */
+ "getaddrinfo_a",
+ /* libgomp */
+ "GOMP_parallel_start",
+ "GOMP_parallel_loop_static_start",
+ "GOMP_parallel_loop_dynamic_start",
+ "GOMP_parallel_loop_guided_start",
+ "GOMP_parallel_loop_runtime_start",
+ "GOMP_parallel_sections_start",
+ };
+
+ for (unsigned int i = 0;
+ i < sizeof(thread_starter) / sizeof(thread_starter[0]);
+ i++)
+ {
+ Symbol* sym = symtab->lookup(thread_starter[i], NULL);
+ thread_safe = sym != NULL && sym->in_reg() && sym->in_real_elf();
+ if (thread_safe)
+ break;
+ }
}
+ this->plt_thread_safe_ = thread_safe;
+ this->group_sections(layout, task);
}
// We need address of stub tables valid for make_stub.
(*p)->set_address_and_size(os, off);
}
+ if (pass != 1)
+ {
+ // Clear plt call stubs, long branch stubs and branch lookup table.
+ prev_brlt_size = this->branch_lookup_table_.size();
+ this->branch_lookup_table_.clear();
+ for (typename Stub_tables::iterator p = this->stub_tables_.begin();
+ p != this->stub_tables_.end();
+ ++p)
+ {
+ (*p)->clear_stubs();
+ }
+ }
+
+ // Build all the stubs.
Stub_table<size, big_endian>* ifunc_stub_table
= this->stub_tables_.size() == 0 ? NULL : this->stub_tables_[0];
Stub_table<size, big_endian>* one_stub_table
b->make_stub(one_stub_table, ifunc_stub_table, symtab);
}
+ // Did anything change size?
unsigned int num_huge_branches = this->branch_lookup_table_.size();
bool again = num_huge_branches != prev_brlt_size;
if (size == 64 && num_huge_branches != 0)
}
}
+ // Set output section offsets for all input sections in an output
+ // section that just changed size. Anything past the stubs will
+ // need updating.
for (typename Output_sections::iterator p = os_need_update.begin();
p != os_need_update.end();
p++)
}
static const uint32_t add_0_11_11 = 0x7c0b5a14;
+static const uint32_t add_2_2_11 = 0x7c425a14;
static const uint32_t add_3_3_2 = 0x7c631214;
static const uint32_t add_3_3_13 = 0x7c636a14;
static const uint32_t add_11_0_11 = 0x7d605a14;
static const uint32_t add_12_2_11 = 0x7d825a14;
+static const uint32_t add_12_12_11 = 0x7d8c5a14;
static const uint32_t addi_11_11 = 0x396b0000;
static const uint32_t addi_12_12 = 0x398c0000;
static const uint32_t addi_2_2 = 0x38420000;
static const uint32_t bctr = 0x4e800420;
static const uint32_t blr = 0x4e800020;
static const uint32_t blrl = 0x4e800021;
+static const uint32_t bnectr_p4 = 0x4ce20420;
+static const uint32_t cmpldi_2_0 = 0x28220000;
static const uint32_t cror_15_15_15 = 0x4def7b82;
static const uint32_t cror_31_31_31 = 0x4ffffb82;
static const uint32_t ld_0_1 = 0xe8010000;
static const uint32_t stfd_0_1 = 0xd8010000;
static const uint32_t stvx_0_12_0 = 0x7c0c01ce;
static const uint32_t sub_11_11_12 = 0x7d6c5850;
+static const uint32_t xor_11_11_11 = 0x7d6b5a78;
// Write out the PLT.
}
}
+static inline uint32_t
+l(uint32_t a)
+{
+ return a & 0xffff;
+}
+
+static inline uint32_t
+hi(uint32_t a)
+{
+ return l(a >> 16);
+}
+
+static inline uint32_t
+ha(uint32_t a)
+{
+ return hi(a + 0x8000);
+}
+
+template<bool big_endian>
+static inline void
+write_insn(unsigned char* p, uint32_t v)
+{
+ elfcpp::Swap<32, big_endian>::writeval(p, v);
+}
+
// Stub_table holds information about plt and long branch stubs.
// Stubs are built in an area following some input section determined
// by group_sections(). This input section is converted to a relaxed
Stub_table(Target_powerpc<size, big_endian>* targ)
: Output_relaxed_input_section(NULL, 0, 0),
targ_(targ), plt_call_stubs_(), long_branch_stubs_(),
- orig_data_size_(0), plt_size_(0), branch_size_(0), prev_size_(0)
+ orig_data_size_(0), plt_size_(0), last_plt_size_(0),
+ branch_size_(0), last_branch_size_(0)
{ }
// Delayed Output_relaxed_input_section init.
find_long_branch_entry(const Powerpc_relobj<size, big_endian>*, Address);
void
- clear_long_branch_stubs()
+ clear_stubs()
{
+ this->plt_call_stubs_.clear();
+ this->plt_size_ = 0;
this->long_branch_stubs_.clear();
this->branch_size_ = 0;
}
// a suitably aligned address.
os->checkpoint_set_addralign(this->stub_align());
}
- if (this->prev_size_ != this->plt_size_ + this->branch_size_)
+ if (this->last_plt_size_ != this->plt_size_
+ || this->last_branch_size_ != this->branch_size_)
{
- this->prev_size_ = this->plt_size_ + this->branch_size_;
+ this->last_plt_size_ = this->plt_size_;
+ this->last_branch_size_ = this->branch_size_;
return true;
}
return false;
}
- section_size_type
- prev_size() const
- { return this->prev_size_; }
-
- void
- set_prev_size(section_size_type val)
- { this->prev_size_ = val; }
-
Target_powerpc<size, big_endian>*
targ() const
{ return targ_; }
private:
+ class Plt_stub_ent;
+ class Plt_stub_ent_hash;
+ typedef Unordered_map<Plt_stub_ent, unsigned int,
+ Plt_stub_ent_hash> Plt_stub_entries;
+
+ // Alignment of stub section.
unsigned int
- stub_align()
- { return size == 32 ? 16 : 32; }
+ stub_align() const
+ {
+ if (size == 32)
+ return 16;
+ unsigned int min_align = 32;
+ unsigned int user_align = 1 << parameters->options().plt_align();
+ return std::max(user_align, min_align);
+ }
- // We keep plt stubs aligned, so no fancy sizing.
+ // Size of a given plt call stub.
unsigned int
- plt_call_size() const
- { return size == 32 ? 16 : 32; }
+ plt_call_size(typename Plt_stub_entries::const_iterator p) const
+ {
+ if (size == 32)
+ return 16;
+
+ Address pltaddr = p->second;
+ if (p->first.sym_ == NULL
+ || (p->first.sym_->type() == elfcpp::STT_GNU_IFUNC
+ && p->first.sym_->can_use_relative_reloc(false)))
+ pltaddr += this->targ_->iplt_section()->address();
+ else
+ pltaddr += this->targ_->plt_section()->address();
+ Address tocbase = this->targ_->got_section()->output_section()->address();
+ const Powerpc_relobj<size, big_endian>* ppcobj = static_cast
+ <const Powerpc_relobj<size, big_endian>*>(p->first.object_);
+ tocbase += ppcobj->toc_base_offset();
+ Address off = pltaddr - tocbase;
+ bool static_chain = parameters->options().plt_static_chain();
+ bool thread_safe = this->targ_->plt_thread_safe();
+ unsigned int bytes = (4 * 5
+ + 4 * static_chain
+ + 8 * thread_safe
+ + 4 * (ha(off) != 0)
+ + 4 * (ha(off + 8 + 8 * static_chain) != ha(off)));
+ unsigned int align = 1 << parameters->options().plt_align();
+ if (align > 1)
+ bytes = (bytes + align - 1) & -align;
+ return bytes;
+ }
// Return long branch stub size.
unsigned int
branch_stub_size(Address to)
{
- Address loc = this->stub_address() + this->plt_size_ + this->branch_size_;
- if (loc - to + (1 << 25) < 2 << 25)
+ Address loc
+ = this->stub_address() + this->last_plt_size_ + this->branch_size_;
+ if (to - loc + (1 << 25) < 2 << 25)
return 4;
if (size == 64 || !parameters->options().output_is_position_independent())
return 16;
// In a sane world this would be a global.
Target_powerpc<size, big_endian>* targ_;
// Map sym/object/addend to stub offset.
- typedef Unordered_map<Plt_stub_ent, unsigned int,
- Plt_stub_ent_hash> Plt_stub_entries;
Plt_stub_entries plt_call_stubs_;
// Map destination address to stub offset.
typedef Unordered_map<Branch_stub_ent, unsigned int,
// size of input section
section_size_type orig_data_size_;
// size of stubs
- section_size_type plt_size_, branch_size_, prev_size_;
+ section_size_type plt_size_, last_plt_size_, branch_size_, last_branch_size_;
};
// Make a new stub table, and record.
{
Plt_stub_ent ent(object, gsym, r_type, addend);
Address off = this->plt_size_;
- if (this->plt_call_stubs_.insert(std::make_pair(ent, off)).second)
- this->plt_size_ = off + this->plt_call_size();
+ std::pair<typename Plt_stub_entries::iterator, bool> p
+ = this->plt_call_stubs_.insert(std::make_pair(ent, off));
+ if (p.second)
+ this->plt_size_ = off + this->plt_call_size(p.first);
}
template<int size, bool big_endian>
{
Plt_stub_ent ent(object, locsym_index, r_type, addend);
Address off = this->plt_size_;
- if (this->plt_call_stubs_.insert(std::make_pair(ent, off)).second)
- this->plt_size_ = off + this->plt_call_size();
+ std::pair<typename Plt_stub_entries::iterator, bool> p
+ = this->plt_call_stubs_.insert(std::make_pair(ent, off));
+ if (p.second)
+ this->plt_size_ = off + this->plt_call_size(p.first);
}
// Find a plt call stub.
this->set_data_size(total);
}
-static inline uint32_t
-l(uint32_t a)
-{
- return a & 0xffff;
-}
-
-static inline uint32_t
-hi(uint32_t a)
-{
- return l(a >> 16);
-}
-
-static inline uint32_t
-ha(uint32_t a)
-{
- return hi(a + 0x8000);
-}
-
-template<bool big_endian>
-static inline void
-write_insn(unsigned char* p, uint32_t v)
-{
- elfcpp::Swap<32, big_endian>::writeval(p, v);
-}
-
// Write out plt and long branch stub code.
template<int size, bool big_endian>
cs != this->plt_call_stubs_.end();
++cs)
{
- Address plt_addr;
+ Address pltoff;
bool is_ifunc;
const Symbol* gsym = cs->first.sym_;
if (gsym != NULL)
{
is_ifunc = (gsym->type() == elfcpp::STT_GNU_IFUNC
&& gsym->can_use_relative_reloc(false));
- plt_addr = gsym->plt_offset();
+ pltoff = gsym->plt_offset();
}
else
{
const Sized_relobj_file<size, big_endian>* relobj
= cs->first.object_;
unsigned int local_sym_index = cs->first.locsym_;
- plt_addr = relobj->local_plt_offset(local_sym_index);
+ pltoff = relobj->local_plt_offset(local_sym_index);
}
+ Address plt_addr = pltoff;
if (is_ifunc)
{
if (iplt_base == invalid_address)
const Powerpc_relobj<size, big_endian>* ppcobj = static_cast
<const Powerpc_relobj<size, big_endian>*>(cs->first.object_);
Address got_addr = got_os_addr + ppcobj->toc_base_offset();
- Address pltoff = plt_addr - got_addr;
+ Address off = plt_addr - got_addr;
- if (pltoff + 0x80008000 > 0xffffffff || (pltoff & 7) != 0)
+ if (off + 0x80008000 > 0xffffffff || (off & 7) != 0)
gold_error(_("%s: linkage table error against `%s'"),
cs->first.object_->name().c_str(),
cs->first.sym_->demangled_name().c_str());
+ bool static_chain = parameters->options().plt_static_chain();
+ bool thread_safe = this->targ_->plt_thread_safe();
+ bool use_fake_dep = false;
+ Address cmp_branch_off = 0;
+ if (thread_safe)
+ {
+ unsigned int pltindex
+ = ((pltoff - this->targ_->first_plt_entry_offset())
+ / this->targ_->plt_entry_size());
+ Address glinkoff
+ = (this->targ_->glink_section()->pltresolve_size
+ + pltindex * 8);
+ if (pltindex > 32768)
+ glinkoff += (pltindex - 32768) * 4;
+ Address to
+ = this->targ_->glink_section()->address() + glinkoff;
+ Address from
+ = (this->stub_address() + cs->second + 24
+ + 4 * (ha(off) != 0)
+ + 4 * (ha(off + 8 + 8 * static_chain) != ha(off))
+ + 4 * static_chain);
+ cmp_branch_off = to - from;
+ use_fake_dep = cmp_branch_off + (1 << 25) >= (1 << 26);
+ }
+
p = oview + cs->second;
- if (ha(pltoff) != 0)
+ if (ha(off) != 0)
{
- write_insn<big_endian>(p, addis_12_2 + ha(pltoff)), p += 4;
write_insn<big_endian>(p, std_2_1 + 40), p += 4;
- write_insn<big_endian>(p, ld_11_12 + l(pltoff)), p += 4;
- if (ha(pltoff + 16) != ha(pltoff))
+ write_insn<big_endian>(p, addis_12_2 + ha(off)), p += 4;
+ write_insn<big_endian>(p, ld_11_12 + l(off)), p += 4;
+ if (ha(off + 8 + 8 * static_chain) != ha(off))
{
- write_insn<big_endian>(p, addi_12_12 + l(pltoff)),p += 4;
- pltoff = 0;
+ write_insn<big_endian>(p, addi_12_12 + l(off)), p += 4;
+ off = 0;
}
write_insn<big_endian>(p, mtctr_11), p += 4;
- write_insn<big_endian>(p, ld_2_12 + l(pltoff + 8)), p += 4;
- write_insn<big_endian>(p, ld_11_12 + l(pltoff + 16)), p += 4;
- write_insn<big_endian>(p, bctr);
+ if (use_fake_dep)
+ {
+ write_insn<big_endian>(p, xor_11_11_11), p += 4;
+ write_insn<big_endian>(p, add_12_12_11), p += 4;
+ }
+ write_insn<big_endian>(p, ld_2_12 + l(off + 8)), p += 4;
+ if (static_chain)
+ write_insn<big_endian>(p, ld_11_12 + l(off + 16)), p += 4;
}
else
{
write_insn<big_endian>(p, std_2_1 + 40), p += 4;
- write_insn<big_endian>(p, ld_11_2 + l(pltoff)), p += 4;
- if (ha(pltoff + 16) != ha(pltoff))
+ write_insn<big_endian>(p, ld_11_2 + l(off)), p += 4;
+ if (ha(off + 8 + 8 * static_chain) != ha(off))
{
- write_insn<big_endian>(p, addi_2_2 + l(pltoff)), p += 4;
- pltoff = 0;
+ write_insn<big_endian>(p, addi_2_2 + l(off)), p += 4;
+ off = 0;
}
write_insn<big_endian>(p, mtctr_11), p += 4;
- write_insn<big_endian>(p, ld_11_2 + l(pltoff + 16)), p += 4;
- write_insn<big_endian>(p, ld_2_2 + l(pltoff + 8)), p += 4;
- write_insn<big_endian>(p, bctr);
+ if (use_fake_dep)
+ {
+ write_insn<big_endian>(p, xor_11_11_11), p += 4;
+ write_insn<big_endian>(p, add_2_2_11), p += 4;
+ }
+ if (static_chain)
+ write_insn<big_endian>(p, ld_11_2 + l(off + 16)), p += 4;
+ write_insn<big_endian>(p, ld_2_2 + l(off + 8)), p += 4;
}
+ if (thread_safe && !use_fake_dep)
+ {
+ write_insn<big_endian>(p, cmpldi_2_0), p += 4;
+ write_insn<big_endian>(p, bnectr_p4), p += 4;
+ write_insn<big_endian>(p, b | (cmp_branch_off & 0x3fffffc));
+ }
+ else
+ write_insn<big_endian>(p, bctr);
}
}
got_addr = g_o_t;
}
- Address pltoff = plt_addr - got_addr;
- if (ha(pltoff) == 0)
+ Address off = plt_addr - got_addr;
+ if (ha(off) == 0)
{
- write_insn<big_endian>(p + 0, lwz_11_30 + l(pltoff));
+ write_insn<big_endian>(p + 0, lwz_11_30 + l(off));
write_insn<big_endian>(p + 4, mtctr_11);
write_insn<big_endian>(p + 8, bctr);
}
else
{
- write_insn<big_endian>(p + 0, addis_11_30 + ha(pltoff));
- write_insn<big_endian>(p + 4, lwz_11_11 + l(pltoff));
+ write_insn<big_endian>(p + 0, addis_11_30 + ha(off));
+ write_insn<big_endian>(p + 4, lwz_11_11 + l(off));
write_insn<big_endian>(p + 8, mtctr_11);
write_insn<big_endian>(p + 12, bctr);
}
projects / binutils-gdb.git / commitdiff