address AbstractInterpreterGenerator::generate_slow_signature_handler() {
address entry = __ pc();
__ andr(esp, esp, -16);
__ mov(c_rarg3, esp);
// rmethod
// rlocals
// c_rarg3: first stack arg - wordSize
// adjust sp
__ sub(sp, c_rarg3, 18 * wordSize);
__ str(lr, Address(__ pre(sp, -2 * wordSize)));
__ call_VM(noreg,
CAST_FROM_FN_PTR(address,
InterpreterRuntime::slow_signature_handler),
rmethod, rlocals, c_rarg3);
// r0: result handler
// Stack layout:
// rsp: return address <- sp
// 1 garbage
// 8 integer args (if static first is unused)
// 1 float/double identifiers
// 8 double args
// stack args <- esp
// garbage
// expression stack bottom
// bcp (NULL)
// ...
// Restore LR
__ ldr(lr, Address(__ post(sp, 2 * wordSize)));
// Do FP first so we can use c_rarg3 as temp
__ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
const FloatRegister r = as_FloatRegister(i);
Label d, done;
__ tbnz(c_rarg3, i, d);
__ ldrs(r, Address(sp, (10 + i) * wordSize));
__ b(done);
__ bind(d);
__ ldrd(r, Address(sp, (10 + i) * wordSize));
__ bind(done);
}
// c_rarg0 contains the result from the call of
// InterpreterRuntime::slow_signature_handler so we don't touch it
// here. It will be loaded with the JNIEnv* later.
__ ldr(c_rarg1, Address(sp, 1 * wordSize));
for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
Register rm = as_Register(i), rn = as_Register(i+1);
__ ldp(rm, rn, Address(sp, i * wordSize));
}
__ add(sp, sp, 18 * wordSize);
__ ret(lr);
return entry;
}
VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
const int aarch64_code_length = VtableStub::pd_code_size_limit(true);
VtableStub* s = new(aarch64_code_length) VtableStub(true, vtable_index);
ResourceMark rm;
CodeBuffer cb(s->entry_point(), aarch64_code_length);
MacroAssembler* masm = new MacroAssembler(&cb);
#ifndef PRODUCT
if (CountCompiledCalls) {
__ lea(r19, ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
__ incrementw(Address(r19));
}
#endif
// get receiver (need to skip return address on top of stack)
assert(VtableStub::receiver_location() == j_rarg0->as_VMReg(), "receiver expected in j_rarg0");
// get receiver klass
address npe_addr = __ pc();
__ load_klass(r19, j_rarg0);
#ifndef PRODUCT
if (DebugVtables) {
Label L;
// check offset vs vtable length
__ ldrw(rscratch1, Address(r19, Klass::vtable_length_offset()));
__ cmpw(rscratch1, vtable_index * vtableEntry::size());
__ br(Assembler::GT, L);
__ enter();
__ mov(r2, vtable_index);
__ call_VM(noreg,
CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), j_rarg0, r2);
__ leave();
__ bind(L);
}
#endif // PRODUCT
__ lookup_virtual_method(r19, vtable_index, rmethod);
if (DebugVtables) {
Label L;
__ cbz(rmethod, L);
__ ldr(rscratch1, Address(rmethod, Method::from_compiled_offset()));
__ cbnz(rscratch1, L);
__ stop("Vtable entry is NULL");
__ bind(L);
}
// r0: receiver klass
// rmethod: Method*
// r2: receiver
address ame_addr = __ pc();
__ ldr(rscratch1, Address(rmethod, Method::from_compiled_offset()));
__ br(rscratch1);
__ flush();
if (PrintMiscellaneous && (WizardMode || Verbose)) {
tty->print_cr("vtable #%d at " PTR_FORMAT "[%d] left over: %d",
vtable_index, p2i(s->entry_point()),
(int)(s->code_end() - s->entry_point()),
(int)(s->code_end() - __ pc()));
}
guarantee(__ pc() <= s->code_end(), "overflowed buffer");
s->set_exception_points(npe_addr, ame_addr);
return s;
}
address JNI_FastGetField::generate_fast_get_int_field0(BasicType type) {
const char *name;
switch (type) {
case T_BOOLEAN: name = "jni_fast_GetBooleanField"; break;
case T_BYTE: name = "jni_fast_GetByteField"; break;
case T_CHAR: name = "jni_fast_GetCharField"; break;
case T_SHORT: name = "jni_fast_GetShortField"; break;
case T_INT: name = "jni_fast_GetIntField"; break;
case T_LONG: name = "jni_fast_GetLongField"; break;
case T_FLOAT: name = "jni_fast_GetFloatField"; break;
case T_DOUBLE: name = "jni_fast_GetDoubleField"; break;
default: ShouldNotReachHere();
}
ResourceMark rm;
BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE);
CodeBuffer cbuf(blob);
MacroAssembler* masm = new MacroAssembler(&cbuf);
address fast_entry = __ pc();
Label slow;
unsigned long offset;
__ adrp(rcounter_addr,
SafepointSynchronize::safepoint_counter_addr(), offset);
Address safepoint_counter_addr(rcounter_addr, offset);
__ ldrw(rcounter, safepoint_counter_addr);
__ andw(rscratch1, rcounter, 1);
__ cbnzw(rscratch1, slow);
__ eor(robj, c_rarg1, rcounter);
__ eor(robj, robj, rcounter); // obj, since
// robj ^ rcounter ^ rcounter == robj
// robj is address dependent on rcounter.
__ ldr(robj, Address(robj, 0)); // *obj
__ lsr(roffset, c_rarg2, 2); // offset
assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
speculative_load_pclist[count] = __ pc(); // Used by the segfault handler
switch (type) {
case T_BOOLEAN: __ ldrb (result, Address(robj, roffset)); break;
case T_BYTE: __ ldrsb (result, Address(robj, roffset)); break;
case T_CHAR: __ ldrh (result, Address(robj, roffset)); break;
case T_SHORT: __ ldrsh (result, Address(robj, roffset)); break;
case T_FLOAT: __ ldrw (result, Address(robj, roffset)); break;
case T_INT: __ ldrsw (result, Address(robj, roffset)); break;
case T_DOUBLE:
case T_LONG: __ ldr (result, Address(robj, roffset)); break;
default: ShouldNotReachHere();
}
// counter_addr is address dependent on result.
__ eor(rcounter_addr, rcounter_addr, result);
__ eor(rcounter_addr, rcounter_addr, result);
__ ldrw(rscratch1, safepoint_counter_addr);
__ cmpw(rcounter, rscratch1);
__ br (Assembler::NE, slow);
switch (type) {
case T_FLOAT: __ fmovs(v0, result); break;
case T_DOUBLE: __ fmovd(v0, result); break;
default: __ mov(r0, result); break;
}
__ ret(lr);
slowcase_entry_pclist[count++] = __ pc();
__ bind(slow);
address slow_case_addr;
switch (type) {
case T_BOOLEAN: slow_case_addr = jni_GetBooleanField_addr(); break;
case T_BYTE: slow_case_addr = jni_GetByteField_addr(); break;
case T_CHAR: slow_case_addr = jni_GetCharField_addr(); break;
case T_SHORT: slow_case_addr = jni_GetShortField_addr(); break;
case T_INT: slow_case_addr = jni_GetIntField_addr(); break;
case T_LONG: slow_case_addr = jni_GetLongField_addr(); break;
case T_FLOAT: slow_case_addr = jni_GetFloatField_addr(); break;
case T_DOUBLE: slow_case_addr = jni_GetDoubleField_addr(); break;
default: ShouldNotReachHere();
}
{
__ enter();
__ lea(rscratch1, ExternalAddress(slow_case_addr));
__ blr(rscratch1);
__ maybe_isb();
__ leave();
__ ret(lr);
}
__ flush ();
return fast_entry;
}
static void SubstrateHookFunctionThumb(SubstrateProcessRef process, void *symbol, void *replace, void **result) {
if (symbol == NULL)
return;
uint16_t *area(reinterpret_cast<uint16_t *>(symbol));
unsigned align((reinterpret_cast<uintptr_t>(area) & 0x2) == 0 ? 0 : 1);
uint16_t *thumb(area + align);
uint32_t *arm(reinterpret_cast<uint32_t *>(thumb + 2));
uint16_t *trail(reinterpret_cast<uint16_t *>(arm + 2));
if (
(align == 0 || area[0] == T$nop) &&
thumb[0] == T$bx(A$pc) &&
thumb[1] == T$nop &&
arm[0] == A$ldr_rd_$rn_im$(A$pc, A$pc, 4 - 8)
) {
if (result != NULL)
*result = reinterpret_cast<void *>(arm[1]);
SubstrateHookMemory code(process, arm + 1, sizeof(uint32_t) * 1);
arm[1] = reinterpret_cast<uint32_t>(replace);
return;
}
size_t required((trail - area) * sizeof(uint16_t));
size_t used(0);
while (used < required)
used += MSGetInstructionWidthThumb(reinterpret_cast<uint8_t *>(area) + used);
used = (used + sizeof(uint16_t) - 1) / sizeof(uint16_t) * sizeof(uint16_t);
size_t blank((used - required) / sizeof(uint16_t));
uint16_t backup[used / sizeof(uint16_t)];
memcpy(backup, area, used);
if (MSDebug) {
char name[16];
sprintf(name, "%p", area);
MSLogHexEx(area, used + sizeof(uint16_t), 2, name);
}
if (result != NULL) {
size_t length(used);
for (unsigned offset(0); offset != used / sizeof(uint16_t); ++offset)
if (T$pcrel$ldr(backup[offset]))
length += 3 * sizeof(uint16_t);
else if (T$pcrel$b(backup[offset]))
length += 6 * sizeof(uint16_t);
else if (T2$pcrel$b(backup + offset)) {
length += 5 * sizeof(uint16_t);
++offset;
} else if (T$pcrel$bl(backup + offset)) {
length += 5 * sizeof(uint16_t);
++offset;
} else if (T$pcrel$cbz(backup[offset])) {
length += 16 * sizeof(uint16_t);
} else if (T$pcrel$ldrw(backup[offset])) {
length += 4 * sizeof(uint16_t);
++offset;
} else if (T$pcrel$add(backup[offset]))
length += 6 * sizeof(uint16_t);
else if (T$32bit$i(backup[offset]))
++offset;
unsigned pad((length & 0x2) == 0 ? 0 : 1);
length += (pad + 2) * sizeof(uint16_t) + 2 * sizeof(uint32_t);
uint16_t *buffer(reinterpret_cast<uint16_t *>(mmap(
NULL, length, PROT_READ | PROT_WRITE, MAP_ANON | MAP_PRIVATE, -1, 0
)));
if (buffer == MAP_FAILED) {
MSLog(MSLogLevelError, "MS:Error:mmap() = %d", errno);
*result = NULL;
return;
}
if (false) fail: {
munmap(buffer, length);
*result = NULL;
return;
}
size_t start(pad), end(length / sizeof(uint16_t));
uint32_t *trailer(reinterpret_cast<uint32_t *>(buffer + end));
for (unsigned offset(0); offset != used / sizeof(uint16_t); ++offset) {
if (T$pcrel$ldr(backup[offset])) {
union {
uint16_t value;
struct {
uint16_t immediate : 8;
uint16_t rd : 3;
uint16_t : 5;
};
} bits = {backup[offset+0]};
buffer[start+0] = T$ldr_rd_$pc_im_4$(bits.rd, T$Label(start+0, end-2) / 4);
buffer[start+1] = T$ldr_rd_$rn_im_4$(bits.rd, bits.rd, 0);
// XXX: this code "works", but is "wrong": the mechanism is more complex than this
*--trailer = ((reinterpret_cast<uint32_t>(area + offset) + 4) & ~0x2) + bits.immediate * 4;
start += 2;
end -= 2;
} else if (T$pcrel$b(backup[offset])) {
union {
uint16_t value;
struct {
uint16_t imm8 : 8;
uint16_t cond : 4;
uint16_t /*1101*/ : 4;
};
} bits = {backup[offset+0]};
intptr_t jump(bits.imm8 << 1);
jump |= 1;
jump <<= 23;
jump >>= 23;
buffer[start+0] = T$b$_$im(bits.cond, (end-6 - (start+0)) * 2 - 4);
*--trailer = reinterpret_cast<uint32_t>(area + offset) + 4 + jump;
*--trailer = A$ldr_rd_$rn_im$(A$pc, A$pc, 4 - 8);
*--trailer = T$nop << 16 | T$bx(A$pc);
start += 1;
end -= 6;
} else if (T2$pcrel$b(backup + offset)) {
union {
uint16_t value;
struct {
uint16_t imm6 : 6;
uint16_t cond : 4;
uint16_t s : 1;
uint16_t : 5;
};
} bits = {backup[offset+0]};
union {
uint16_t value;
struct {
uint16_t imm11 : 11;
uint16_t j2 : 1;
uint16_t a : 1;
uint16_t j1 : 1;
uint16_t : 2;
};
} exts = {backup[offset+1]};
intptr_t jump(1);
jump |= exts.imm11 << 1;
jump |= bits.imm6 << 12;
if (exts.a) {
jump |= bits.s << 24;
jump |= (~(bits.s ^ exts.j1) & 0x1) << 23;
jump |= (~(bits.s ^ exts.j2) & 0x1) << 22;
jump |= bits.cond << 18;
jump <<= 7;
jump >>= 7;
} else {