blob: 4197ca6a99577b8ed4058c222421e898e12bccad [file] [log] [blame]
/*
* Copyright (C) 2017-2019 Apple Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "config.h"
#include "CCallHelpers.h"
#include "CPU.h"
#include "FPRInfo.h"
#include "GPRInfo.h"
#include "InitializeThreading.h"
#include "LinkBuffer.h"
#include "ProbeContext.h"
#include "StackAlignment.h"
#include <limits>
#include <wtf/Compiler.h>
#include <wtf/DataLog.h>
#include <wtf/Function.h>
#include <wtf/Lock.h>
#include <wtf/NumberOfCores.h>
#include <wtf/PtrTag.h>
#include <wtf/Threading.h>
#include <wtf/text/StringCommon.h>
// We don't have a NO_RETURN_DUE_TO_EXIT, nor should we. That's ridiculous.
static bool hiddenTruthBecauseNoReturnIsStupid() { return true; }
static void usage()
{
dataLog("Usage: testmasm [<filter>]\n");
if (hiddenTruthBecauseNoReturnIsStupid())
exit(1);
}
#if ENABLE(JIT)
#if ENABLE(MASM_PROBE)
namespace WTF {
static void printInternal(PrintStream& out, void* value)
{
out.printf("%p", value);
}
} // namespace WTF
#endif // ENABLE(MASM_PROBE)
namespace JSC {
namespace Probe {
JS_EXPORT_PRIVATE void* probeStateForContext(Probe::Context&);
} // namespace Probe
} // namespace JSC
using namespace JSC;
namespace {
#if ENABLE(MASM_PROBE)
using CPUState = Probe::CPUState;
#endif
Lock crashLock;
typedef WTF::Function<void(CCallHelpers&)> Generator;
template<typename T> T nextID(T id) { return static_cast<T>(id + 1); }
#define TESTWORD64 0x0c0defefebeef000
#define TESTWORD32 0x0beef000
#define testWord32(x) (TESTWORD32 + static_cast<uint32_t>(x))
#define testWord64(x) (TESTWORD64 + static_cast<uint64_t>(x))
#if USE(JSVALUE64)
#define testWord(x) testWord64(x)
#else
#define testWord(x) testWord32(x)
#endif
// Nothing fancy for now; we just use the existing WTF assertion machinery.
#define CHECK_EQ(_actual, _expected) do { \
if ((_actual) == (_expected)) \
break; \
crashLock.lock(); \
dataLog("FAILED while testing " #_actual ": expected: ", _expected, ", actual: ", _actual, "\n"); \
WTFReportAssertionFailure(__FILE__, __LINE__, WTF_PRETTY_FUNCTION, "CHECK_EQ("#_actual ", " #_expected ")"); \
CRASH(); \
} while (false)
#define CHECK_NOT_EQ(_actual, _expected) do { \
if ((_actual) != (_expected)) \
break; \
crashLock.lock(); \
dataLog("FAILED while testing " #_actual ": expected not: ", _expected, ", actual: ", _actual, "\n"); \
WTFReportAssertionFailure(__FILE__, __LINE__, WTF_PRETTY_FUNCTION, "CHECK_NOT_EQ("#_actual ", " #_expected ")"); \
CRASH(); \
} while (false)
#if ENABLE(MASM_PROBE)
bool isPC(MacroAssembler::RegisterID id)
{
#if CPU(ARM_THUMB2)
return id == ARMRegisters::pc;
#else
UNUSED_PARAM(id);
return false;
#endif
}
bool isSP(MacroAssembler::RegisterID id)
{
return id == MacroAssembler::stackPointerRegister;
}
bool isFP(MacroAssembler::RegisterID id)
{
return id == MacroAssembler::framePointerRegister;
}
bool isSpecialGPR(MacroAssembler::RegisterID id)
{
if (isPC(id) || isSP(id) || isFP(id))
return true;
#if CPU(ARM64)
if (id == ARM64Registers::x18)
return true;
#elif CPU(MIPS)
if (id == MIPSRegisters::zero || id == MIPSRegisters::k0 || id == MIPSRegisters::k1)
return true;
#endif
return false;
}
#endif // ENABLE(MASM_PROBE)
MacroAssemblerCodeRef<JSEntryPtrTag> compile(Generator&& generate)
{
CCallHelpers jit;
generate(jit);
LinkBuffer linkBuffer(jit, nullptr);
return FINALIZE_CODE(linkBuffer, JSEntryPtrTag, "testmasm compilation");
}
template<typename T, typename... Arguments>
T invoke(const MacroAssemblerCodeRef<JSEntryPtrTag>& code, Arguments... arguments)
{
void* executableAddress = untagCFunctionPtr<JSEntryPtrTag>(code.code().executableAddress());
T (*function)(Arguments...) = bitwise_cast<T(*)(Arguments...)>(executableAddress);
return function(arguments...);
}
template<typename T, typename... Arguments>
T compileAndRun(Generator&& generator, Arguments... arguments)
{
return invoke<T>(compile(WTFMove(generator)), arguments...);
}
void emitFunctionPrologue(CCallHelpers& jit)
{
jit.emitFunctionPrologue();
#if CPU(ARM_THUMB2)
// MacroAssemblerARMv7 uses r6 as a temporary register, which is a
// callee-saved register, see 5.1.1 of the Procedure Call Standard for
// the ARM Architecture.
// http://infocenter.arm.com/help/topic/com.arm.doc.ihi0042f/IHI0042F_aapcs.pdf
jit.push(ARMRegisters::r6);
#endif
}
void emitFunctionEpilogue(CCallHelpers& jit)
{
#if CPU(ARM_THUMB2)
jit.pop(ARMRegisters::r6);
#endif
jit.emitFunctionEpilogue();
}
void testSimple()
{
CHECK_EQ(compileAndRun<int>([] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.move(CCallHelpers::TrustedImm32(42), GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
}), 42);
}
void testGetEffectiveAddress(size_t pointer, ptrdiff_t length, int32_t offset, CCallHelpers::Scale scale)
{
CHECK_EQ(compileAndRun<size_t>([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.move(CCallHelpers::TrustedImmPtr(bitwise_cast<void*>(pointer)), GPRInfo::regT0);
jit.move(CCallHelpers::TrustedImmPtr(bitwise_cast<void*>(length)), GPRInfo::regT1);
jit.getEffectiveAddress(CCallHelpers::BaseIndex(GPRInfo::regT0, GPRInfo::regT1, scale, offset), GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
}), pointer + offset + (static_cast<size_t>(1) << static_cast<int>(scale)) * length);
}
// branchTruncateDoubleToInt32(), when encountering Infinity, -Infinity or a
// Nan, should either yield 0 in dest or fail.
void testBranchTruncateDoubleToInt32(double val, int32_t expected)
{
const uint64_t valAsUInt = *reinterpret_cast<uint64_t*>(&val);
#if CPU(BIG_ENDIAN)
const bool isBigEndian = true;
#else
const bool isBigEndian = false;
#endif
CHECK_EQ(compileAndRun<int>([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.subPtr(CCallHelpers::TrustedImm32(stackAlignmentBytes()), MacroAssembler::stackPointerRegister);
if (isBigEndian) {
jit.store32(CCallHelpers::TrustedImm32(valAsUInt >> 32),
MacroAssembler::stackPointerRegister);
jit.store32(CCallHelpers::TrustedImm32(valAsUInt & 0xffffffff),
MacroAssembler::Address(MacroAssembler::stackPointerRegister, 4));
} else {
jit.store32(CCallHelpers::TrustedImm32(valAsUInt & 0xffffffff),
MacroAssembler::stackPointerRegister);
jit.store32(CCallHelpers::TrustedImm32(valAsUInt >> 32),
MacroAssembler::Address(MacroAssembler::stackPointerRegister, 4));
}
jit.loadDouble(MacroAssembler::stackPointerRegister, FPRInfo::fpRegT0);
MacroAssembler::Jump done;
done = jit.branchTruncateDoubleToInt32(FPRInfo::fpRegT0, GPRInfo::returnValueGPR, MacroAssembler::BranchIfTruncateSuccessful);
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::returnValueGPR);
done.link(&jit);
jit.addPtr(CCallHelpers::TrustedImm32(stackAlignmentBytes()), MacroAssembler::stackPointerRegister);
emitFunctionEpilogue(jit);
jit.ret();
}), expected);
}
static Vector<double> doubleOperands()
{
return Vector<double> {
0,
-0,
1,
-1,
42,
-42,
std::numeric_limits<double>::max(),
std::numeric_limits<double>::min(),
std::numeric_limits<double>::lowest(),
std::numeric_limits<double>::quiet_NaN(),
std::numeric_limits<double>::infinity(),
-std::numeric_limits<double>::infinity(),
};
}
#if CPU(X86) || CPU(X86_64) || CPU(ARM64)
static Vector<float> floatOperands()
{
return Vector<float> {
0,
-0,
1,
-1,
42,
-42,
std::numeric_limits<float>::max(),
std::numeric_limits<float>::min(),
std::numeric_limits<float>::lowest(),
std::numeric_limits<float>::quiet_NaN(),
std::numeric_limits<float>::infinity(),
-std::numeric_limits<float>::infinity(),
};
}
#endif
static Vector<int32_t> int32Operands()
{
return Vector<int32_t> {
0,
1,
-1,
2,
-2,
42,
-42,
64,
std::numeric_limits<int32_t>::max(),
std::numeric_limits<int32_t>::min(),
};
}
#if CPU(X86_64)
static Vector<int64_t> int64Operands()
{
return Vector<int64_t> {
0,
1,
-1,
2,
-2,
42,
-42,
64,
std::numeric_limits<int32_t>::max(),
std::numeric_limits<int32_t>::min(),
std::numeric_limits<int64_t>::max(),
std::numeric_limits<int64_t>::min(),
};
}
#endif
#if CPU(X86_64)
void testBranchTestBit32RegReg()
{
for (auto value : int32Operands()) {
auto test = compile([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
auto branch = jit.branchTestBit32(MacroAssembler::NonZero, GPRInfo::argumentGPR0, GPRInfo::argumentGPR1);
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::returnValueGPR);
auto done = jit.jump();
branch.link(&jit);
jit.move(CCallHelpers::TrustedImm32(1), GPRInfo::returnValueGPR);
done.link(&jit);
emitFunctionEpilogue(jit);
jit.ret();
});
for (auto value2 : int32Operands())
CHECK_EQ(invoke<int>(test, value, value2), (value>>(value2%32))&1);
}
}
void testBranchTestBit32RegImm()
{
for (auto value : int32Operands()) {
auto test = compile([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
auto branch = jit.branchTestBit32(MacroAssembler::NonZero, GPRInfo::argumentGPR0, CCallHelpers::TrustedImm32(value));
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::returnValueGPR);
auto done = jit.jump();
branch.link(&jit);
jit.move(CCallHelpers::TrustedImm32(1), GPRInfo::returnValueGPR);
done.link(&jit);
emitFunctionEpilogue(jit);
jit.ret();
});
for (auto value2 : int32Operands())
CHECK_EQ(invoke<int>(test, value2), (value2>>(value%32))&1);
}
}
void testBranchTestBit32AddrImm()
{
for (auto value : int32Operands()) {
auto test = compile([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
auto branch = jit.branchTestBit32(MacroAssembler::NonZero, MacroAssembler::Address(GPRInfo::argumentGPR0, 0), CCallHelpers::TrustedImm32(value));
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::returnValueGPR);
auto done = jit.jump();
branch.link(&jit);
jit.move(CCallHelpers::TrustedImm32(1), GPRInfo::returnValueGPR);
done.link(&jit);
emitFunctionEpilogue(jit);
jit.ret();
});
for (auto value2 : int32Operands())
CHECK_EQ(invoke<int>(test, &value2), (value2>>(value%32))&1);
}
}
void testBranchTestBit64RegReg()
{
for (auto value : int64Operands()) {
auto test = compile([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
auto branch = jit.branchTestBit64(MacroAssembler::NonZero, GPRInfo::argumentGPR0, GPRInfo::argumentGPR1);
jit.move(CCallHelpers::TrustedImm64(0), GPRInfo::returnValueGPR);
auto done = jit.jump();
branch.link(&jit);
jit.move(CCallHelpers::TrustedImm64(1), GPRInfo::returnValueGPR);
done.link(&jit);
emitFunctionEpilogue(jit);
jit.ret();
});
for (auto value2 : int64Operands())
CHECK_EQ(invoke<long int>(test, value, value2), (value>>(value2%64))&1);
}
}
void testBranchTestBit64RegImm()
{
for (auto value : int64Operands()) {
auto test = compile([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
auto branch = jit.branchTestBit64(MacroAssembler::NonZero, GPRInfo::argumentGPR0, CCallHelpers::TrustedImm32(value));
jit.move(CCallHelpers::TrustedImm64(0), GPRInfo::returnValueGPR);
auto done = jit.jump();
branch.link(&jit);
jit.move(CCallHelpers::TrustedImm64(1), GPRInfo::returnValueGPR);
done.link(&jit);
emitFunctionEpilogue(jit);
jit.ret();
});
for (auto value2 : int64Operands())
CHECK_EQ(invoke<long int>(test, value2), (value2>>(value%64))&1);
}
}
void testBranchTestBit64AddrImm()
{
for (auto value : int64Operands()) {
auto test = compile([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
auto branch = jit.branchTestBit64(MacroAssembler::NonZero, MacroAssembler::Address(GPRInfo::argumentGPR0, 0), CCallHelpers::TrustedImm32(value));
jit.move(CCallHelpers::TrustedImm64(0), GPRInfo::returnValueGPR);
auto done = jit.jump();
branch.link(&jit);
jit.move(CCallHelpers::TrustedImm64(1), GPRInfo::returnValueGPR);
done.link(&jit);
emitFunctionEpilogue(jit);
jit.ret();
});
for (auto value2 : int64Operands())
CHECK_EQ(invoke<long int>(test, &value2), (value2>>(value%64))&1);
}
}
#endif
void testCompareDouble(MacroAssembler::DoubleCondition condition)
{
double arg1 = 0;
double arg2 = 0;
auto compareDouble = compile([&, condition] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&arg1), FPRInfo::fpRegT0);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&arg2), FPRInfo::fpRegT1);
jit.move(CCallHelpers::TrustedImm32(-1), GPRInfo::returnValueGPR);
jit.compareDouble(condition, FPRInfo::fpRegT0, FPRInfo::fpRegT1, GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
auto compareDoubleGeneric = compile([&, condition] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&arg1), FPRInfo::fpRegT0);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&arg2), FPRInfo::fpRegT1);
jit.move(CCallHelpers::TrustedImm32(1), GPRInfo::returnValueGPR);
auto jump = jit.branchDouble(condition, FPRInfo::fpRegT0, FPRInfo::fpRegT1);
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::returnValueGPR);
jump.link(&jit);
emitFunctionEpilogue(jit);
jit.ret();
});
auto operands = doubleOperands();
for (auto a : operands) {
for (auto b : operands) {
arg1 = a;
arg2 = b;
CHECK_EQ(invoke<int>(compareDouble), invoke<int>(compareDoubleGeneric));
}
}
}
void testMul32WithImmediates()
{
for (auto immediate : int32Operands()) {
auto mul = compile([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.mul32(CCallHelpers::TrustedImm32(immediate), GPRInfo::argumentGPR0, GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
for (auto value : int32Operands())
CHECK_EQ(invoke<int>(mul, value), immediate * value);
}
}
#if CPU(ARM64)
void testMul32SignExtend()
{
for (auto value : int32Operands()) {
auto mul = compile([=] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.multiplySignExtend32(GPRInfo::argumentGPR0, GPRInfo::argumentGPR1, GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
for (auto value2 : int32Operands())
CHECK_EQ(invoke<long int>(mul, value, value2), ((long int) value) * ((long int) value2));
}
}
#endif
#if CPU(X86) || CPU(X86_64) || CPU(ARM64)
void testCompareFloat(MacroAssembler::DoubleCondition condition)
{
float arg1 = 0;
float arg2 = 0;
auto compareFloat = compile([&, condition] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.loadFloat(CCallHelpers::TrustedImmPtr(&arg1), FPRInfo::fpRegT0);
jit.loadFloat(CCallHelpers::TrustedImmPtr(&arg2), FPRInfo::fpRegT1);
jit.move(CCallHelpers::TrustedImm32(-1), GPRInfo::returnValueGPR);
jit.compareFloat(condition, FPRInfo::fpRegT0, FPRInfo::fpRegT1, GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
auto compareFloatGeneric = compile([&, condition] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.loadFloat(CCallHelpers::TrustedImmPtr(&arg1), FPRInfo::fpRegT0);
jit.loadFloat(CCallHelpers::TrustedImmPtr(&arg2), FPRInfo::fpRegT1);
jit.move(CCallHelpers::TrustedImm32(1), GPRInfo::returnValueGPR);
auto jump = jit.branchFloat(condition, FPRInfo::fpRegT0, FPRInfo::fpRegT1);
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::returnValueGPR);
jump.link(&jit);
emitFunctionEpilogue(jit);
jit.ret();
});
auto operands = floatOperands();
for (auto a : operands) {
for (auto b : operands) {
arg1 = a;
arg2 = b;
CHECK_EQ(invoke<int>(compareFloat), invoke<int>(compareFloatGeneric));
}
}
}
#endif
#if ENABLE(MASM_PROBE)
void testProbeReadsArgumentRegisters()
{
bool probeWasCalled = false;
compileAndRun<void>([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.pushPair(GPRInfo::argumentGPR0, GPRInfo::argumentGPR1);
jit.pushPair(GPRInfo::argumentGPR2, GPRInfo::argumentGPR3);
jit.move(CCallHelpers::TrustedImm32(testWord32(0)), GPRInfo::argumentGPR0);
jit.convertInt32ToDouble(GPRInfo::argumentGPR0, FPRInfo::fpRegT0);
jit.move(CCallHelpers::TrustedImm32(testWord32(1)), GPRInfo::argumentGPR0);
jit.convertInt32ToDouble(GPRInfo::argumentGPR0, FPRInfo::fpRegT1);
#if USE(JSVALUE64)
jit.move(CCallHelpers::TrustedImm64(testWord(0)), GPRInfo::argumentGPR0);
jit.move(CCallHelpers::TrustedImm64(testWord(1)), GPRInfo::argumentGPR1);
jit.move(CCallHelpers::TrustedImm64(testWord(2)), GPRInfo::argumentGPR2);
jit.move(CCallHelpers::TrustedImm64(testWord(3)), GPRInfo::argumentGPR3);
#else
jit.move(CCallHelpers::TrustedImm32(testWord(0)), GPRInfo::argumentGPR0);
jit.move(CCallHelpers::TrustedImm32(testWord(1)), GPRInfo::argumentGPR1);
jit.move(CCallHelpers::TrustedImm32(testWord(2)), GPRInfo::argumentGPR2);
jit.move(CCallHelpers::TrustedImm32(testWord(3)), GPRInfo::argumentGPR3);
#endif
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeWasCalled = true;
CHECK_EQ(cpu.gpr(GPRInfo::argumentGPR0), testWord(0));
CHECK_EQ(cpu.gpr(GPRInfo::argumentGPR1), testWord(1));
CHECK_EQ(cpu.gpr(GPRInfo::argumentGPR2), testWord(2));
CHECK_EQ(cpu.gpr(GPRInfo::argumentGPR3), testWord(3));
CHECK_EQ(cpu.fpr(FPRInfo::fpRegT0), testWord32(0));
CHECK_EQ(cpu.fpr(FPRInfo::fpRegT1), testWord32(1));
});
jit.popPair(GPRInfo::argumentGPR2, GPRInfo::argumentGPR3);
jit.popPair(GPRInfo::argumentGPR0, GPRInfo::argumentGPR1);
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(probeWasCalled, true);
}
void testProbeWritesArgumentRegisters()
{
// This test relies on testProbeReadsArgumentRegisters() having already validated
// that we can read from argument registers. We'll use that ability to validate
// that our writes did take effect.
unsigned probeCallCount = 0;
compileAndRun<void>([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.pushPair(GPRInfo::argumentGPR0, GPRInfo::argumentGPR1);
jit.pushPair(GPRInfo::argumentGPR2, GPRInfo::argumentGPR3);
// Pre-initialize with non-expected values.
#if USE(JSVALUE64)
jit.move(CCallHelpers::TrustedImm64(0), GPRInfo::argumentGPR0);
jit.move(CCallHelpers::TrustedImm64(0), GPRInfo::argumentGPR1);
jit.move(CCallHelpers::TrustedImm64(0), GPRInfo::argumentGPR2);
jit.move(CCallHelpers::TrustedImm64(0), GPRInfo::argumentGPR3);
#else
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::argumentGPR0);
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::argumentGPR1);
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::argumentGPR2);
jit.move(CCallHelpers::TrustedImm32(0), GPRInfo::argumentGPR3);
#endif
jit.convertInt32ToDouble(GPRInfo::argumentGPR0, FPRInfo::fpRegT0);
jit.convertInt32ToDouble(GPRInfo::argumentGPR0, FPRInfo::fpRegT1);
// Write expected values.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
cpu.gpr(GPRInfo::argumentGPR0) = testWord(0);
cpu.gpr(GPRInfo::argumentGPR1) = testWord(1);
cpu.gpr(GPRInfo::argumentGPR2) = testWord(2);
cpu.gpr(GPRInfo::argumentGPR3) = testWord(3);
cpu.fpr(FPRInfo::fpRegT0) = bitwise_cast<double>(testWord64(0));
cpu.fpr(FPRInfo::fpRegT1) = bitwise_cast<double>(testWord64(1));
});
// Validate that expected values were written.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
CHECK_EQ(cpu.gpr(GPRInfo::argumentGPR0), testWord(0));
CHECK_EQ(cpu.gpr(GPRInfo::argumentGPR1), testWord(1));
CHECK_EQ(cpu.gpr(GPRInfo::argumentGPR2), testWord(2));
CHECK_EQ(cpu.gpr(GPRInfo::argumentGPR3), testWord(3));
CHECK_EQ(cpu.fpr<uint64_t>(FPRInfo::fpRegT0), testWord64(0));
CHECK_EQ(cpu.fpr<uint64_t>(FPRInfo::fpRegT1), testWord64(1));
});
jit.popPair(GPRInfo::argumentGPR2, GPRInfo::argumentGPR3);
jit.popPair(GPRInfo::argumentGPR0, GPRInfo::argumentGPR1);
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(probeCallCount, 2);
}
static NEVER_INLINE NOT_TAIL_CALLED int testFunctionToTrashGPRs(int a, int b, int c, int d, int e, int f, int g, int h, int i, int j)
{
if (j > 0)
return testFunctionToTrashGPRs(a + 1, b + a, c + b, d + 5, e - a, f * 1.5, g ^ a, h - b, i, j - 1);
return a + 1;
}
static NEVER_INLINE NOT_TAIL_CALLED double testFunctionToTrashFPRs(double a, double b, double c, double d, double e, double f, double g, double h, double i, double j)
{
if (j > 0)
return testFunctionToTrashFPRs(a + 1, b + a, c + b, d + 5, e - a, f * 1.5, pow(g, a), h - b, i, j - 1);
return a + 1;
}
void testProbePreservesGPRS()
{
// This test relies on testProbeReadsArgumentRegisters() and testProbeWritesArgumentRegisters()
// having already validated that we can read and write from registers. We'll use these abilities
// to validate that the probe preserves register values.
unsigned probeCallCount = 0;
CPUState originalState;
compileAndRun<void>([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
// Write expected values into the registers (except for sp, fp, and pc).
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
originalState.gpr(id) = cpu.gpr(id);
if (isSpecialGPR(id))
continue;
cpu.gpr(id) = testWord(static_cast<int>(id));
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id)) {
originalState.fpr(id) = cpu.fpr(id);
cpu.fpr(id) = bitwise_cast<double>(testWord64(id));
}
});
// Invoke the probe to call a lot of functions and trash register values.
jit.probe([&] (Probe::Context&) {
probeCallCount++;
CHECK_EQ(testFunctionToTrashGPRs(0, 1, 2, 3, 4, 5, 6, 7, 8, 9), 10);
CHECK_EQ(testFunctionToTrashFPRs(0, 1, 2, 3, 4, 5, 6, 7, 8, 9), 10);
});
// Validate that the registers have the expected values.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
if (isSP(id) || isFP(id)) {
CHECK_EQ(cpu.gpr(id), originalState.gpr(id));
continue;
}
if (isSpecialGPR(id))
continue;
CHECK_EQ(cpu.gpr(id), testWord(id));
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id))
#if CPU(MIPS)
if (!(id & 1))
#endif
CHECK_EQ(cpu.fpr<uint64_t>(id), testWord64(id));
});
// Restore the original state.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
if (isSpecialGPR(id))
continue;
cpu.gpr(id) = originalState.gpr(id);
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id))
cpu.fpr(id) = originalState.fpr(id);
});
// Validate that the original state was restored.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
if (isSpecialGPR(id))
continue;
CHECK_EQ(cpu.gpr(id), originalState.gpr(id));
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id))
#if CPU(MIPS)
if (!(id & 1))
#endif
CHECK_EQ(cpu.fpr<uint64_t>(id), originalState.fpr<uint64_t>(id));
});
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(probeCallCount, 5);
}
void testProbeModifiesStackPointer(WTF::Function<void*(Probe::Context&)> computeModifiedStackPointer)
{
unsigned probeCallCount = 0;
CPUState originalState;
void* originalSP { nullptr };
void* modifiedSP { nullptr };
#if !(CPU(MIPS))
uintptr_t modifiedFlags { 0 };
#endif
#if CPU(X86) || CPU(X86_64)
auto flagsSPR = X86Registers::eflags;
uintptr_t flagsMask = 0xc5;
#elif CPU(ARM_THUMB2)
auto flagsSPR = ARMRegisters::apsr;
uintptr_t flagsMask = 0xf8000000;
#elif CPU(ARM64)
auto flagsSPR = ARM64Registers::nzcv;
uintptr_t flagsMask = 0xf0000000;
#endif
compileAndRun<void>([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
// Preserve original stack pointer and modify the sp, and
// write expected values into other registers (except for fp, and pc).
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
originalState.gpr(id) = cpu.gpr(id);
if (isSpecialGPR(id))
continue;
cpu.gpr(id) = testWord(static_cast<int>(id));
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id)) {
originalState.fpr(id) = cpu.fpr(id);
cpu.fpr(id) = bitwise_cast<double>(testWord64(id));
}
#if !(CPU(MIPS))
originalState.spr(flagsSPR) = cpu.spr(flagsSPR);
modifiedFlags = originalState.spr(flagsSPR) ^ flagsMask;
cpu.spr(flagsSPR) = modifiedFlags;
#endif
originalSP = cpu.sp();
modifiedSP = computeModifiedStackPointer(context);
cpu.sp() = modifiedSP;
});
// Validate that the registers have the expected values.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
if (isFP(id)) {
CHECK_EQ(cpu.gpr(id), originalState.gpr(id));
continue;
}
if (isSpecialGPR(id))
continue;
CHECK_EQ(cpu.gpr(id), testWord(id));
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id))
#if CPU(MIPS)
if (!(id & 1))
#endif
CHECK_EQ(cpu.fpr<uint64_t>(id), testWord64(id));
#if !(CPU(MIPS))
CHECK_EQ(cpu.spr(flagsSPR) & flagsMask, modifiedFlags & flagsMask);
#endif
CHECK_EQ(cpu.sp(), modifiedSP);
});
// Restore the original state.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
if (isSpecialGPR(id))
continue;
cpu.gpr(id) = originalState.gpr(id);
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id))
cpu.fpr(id) = originalState.fpr(id);
#if !(CPU(MIPS))
cpu.spr(flagsSPR) = originalState.spr(flagsSPR);
#endif
cpu.sp() = originalSP;
});
// Validate that the original state was restored.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
if (isSpecialGPR(id))
continue;
CHECK_EQ(cpu.gpr(id), originalState.gpr(id));
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id))
#if CPU(MIPS)
if (!(id & 1))
#endif
CHECK_EQ(cpu.fpr<uint64_t>(id), originalState.fpr<uint64_t>(id));
#if !(CPU(MIPS))
CHECK_EQ(cpu.spr(flagsSPR) & flagsMask, originalState.spr(flagsSPR) & flagsMask);
#endif
CHECK_EQ(cpu.sp(), originalSP);
});
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(probeCallCount, 4);
}
void testProbeModifiesStackPointerToInsideProbeStateOnStack()
{
size_t increment = sizeof(uintptr_t);
#if CPU(ARM64)
// The ARM64 probe uses ldp and stp which require 16 byte alignment.
increment = 2 * sizeof(uintptr_t);
#endif
for (size_t offset = 0; offset < sizeof(Probe::State); offset += increment) {
testProbeModifiesStackPointer([=] (Probe::Context& context) -> void* {
return reinterpret_cast<uint8_t*>(probeStateForContext(context)) + offset;
});
}
}
void testProbeModifiesStackPointerToNBytesBelowSP()
{
size_t increment = sizeof(uintptr_t);
#if CPU(ARM64)
// The ARM64 probe uses ldp and stp which require 16 byte alignment.
increment = 2 * sizeof(uintptr_t);
#endif
for (size_t offset = 0; offset < 1 * KB; offset += increment) {
testProbeModifiesStackPointer([=] (Probe::Context& context) -> void* {
return context.cpu.sp<uint8_t*>() - offset;
});
}
}
void testProbeModifiesProgramCounter()
{
// This test relies on testProbeReadsArgumentRegisters() and testProbeWritesArgumentRegisters()
// having already validated that we can read and write from registers. We'll use these abilities
// to validate that the probe preserves register values.
unsigned probeCallCount = 0;
bool continuationWasReached = false;
MacroAssemblerCodeRef<JSEntryPtrTag> continuation = compile([&] (CCallHelpers& jit) {
// Validate that we reached the continuation.
jit.probe([&] (Probe::Context&) {
probeCallCount++;
continuationWasReached = true;
});
emitFunctionEpilogue(jit);
jit.ret();
});
compileAndRun<void>([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
// Write expected values into the registers.
jit.probe([&] (Probe::Context& context) {
probeCallCount++;
context.cpu.pc() = untagCodePtr(continuation.code().executableAddress(), JSEntryPtrTag);
});
jit.breakpoint(); // We should never get here.
});
CHECK_EQ(probeCallCount, 2);
CHECK_EQ(continuationWasReached, true);
}
void testProbeModifiesStackValues()
{
unsigned probeCallCount = 0;
CPUState originalState;
void* originalSP { nullptr };
void* newSP { nullptr };
#if !CPU(MIPS)
uintptr_t modifiedFlags { 0 };
#endif
size_t numberOfExtraEntriesToWrite { 10 }; // ARM64 requires that this be 2 word aligned.
#if CPU(X86) || CPU(X86_64)
MacroAssembler::SPRegisterID flagsSPR = X86Registers::eflags;
uintptr_t flagsMask = 0xc5;
#elif CPU(ARM_THUMB2)
MacroAssembler::SPRegisterID flagsSPR = ARMRegisters::apsr;
uintptr_t flagsMask = 0xf8000000;
#elif CPU(ARM64)
MacroAssembler::SPRegisterID flagsSPR = ARM64Registers::nzcv;
uintptr_t flagsMask = 0xf0000000;
#endif
compileAndRun<void>([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
// Write expected values into the registers.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
auto& stack = context.stack();
probeCallCount++;
// Preserve the original CPU state.
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
originalState.gpr(id) = cpu.gpr(id);
if (isSpecialGPR(id))
continue;
cpu.gpr(id) = testWord(static_cast<int>(id));
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id)) {
originalState.fpr(id) = cpu.fpr(id);
cpu.fpr(id) = bitwise_cast<double>(testWord64(id));
}
#if !(CPU(MIPS))
originalState.spr(flagsSPR) = cpu.spr(flagsSPR);
modifiedFlags = originalState.spr(flagsSPR) ^ flagsMask;
cpu.spr(flagsSPR) = modifiedFlags;
#endif
// Ensure that we'll be writing over the regions of the stack where the Probe::State is.
originalSP = cpu.sp();
newSP = reinterpret_cast<uintptr_t*>(probeStateForContext(context)) - numberOfExtraEntriesToWrite;
cpu.sp() = newSP;
// Fill the stack with values.
uintptr_t* p = reinterpret_cast<uintptr_t*>(newSP);
int count = 0;
stack.set<double>(p++, 1.234567);
if (is32Bit())
p++; // On 32-bit targets, a double takes up 2 uintptr_t.
while (p < reinterpret_cast<uintptr_t*>(originalSP))
stack.set<uintptr_t>(p++, testWord(count++));
});
// Validate that the registers and stack have the expected values.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
auto& stack = context.stack();
probeCallCount++;
// Validate the register values.
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
if (isFP(id)) {
CHECK_EQ(cpu.gpr(id), originalState.gpr(id));
continue;
}
if (isSpecialGPR(id))
continue;
CHECK_EQ(cpu.gpr(id), testWord(id));
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id))
#if CPU(MIPS)
if (!(id & 1))
#endif
CHECK_EQ(cpu.fpr<uint64_t>(id), testWord64(id));
#if !(CPU(MIPS))
CHECK_EQ(cpu.spr(flagsSPR) & flagsMask, modifiedFlags & flagsMask);
#endif
CHECK_EQ(cpu.sp(), newSP);
// Validate the stack values.
uintptr_t* p = reinterpret_cast<uintptr_t*>(newSP);
int count = 0;
CHECK_EQ(stack.get<double>(p++), 1.234567);
if (is32Bit())
p++; // On 32-bit targets, a double takes up 2 uintptr_t.
while (p < reinterpret_cast<uintptr_t*>(originalSP))
CHECK_EQ(stack.get<uintptr_t>(p++), testWord(count++));
});
// Restore the original state.
jit.probe([&] (Probe::Context& context) {
auto& cpu = context.cpu;
probeCallCount++;
for (auto id = CCallHelpers::firstRegister(); id <= CCallHelpers::lastRegister(); id = nextID(id)) {
if (isSpecialGPR(id))
continue;
cpu.gpr(id) = originalState.gpr(id);
}
for (auto id = CCallHelpers::firstFPRegister(); id <= CCallHelpers::lastFPRegister(); id = nextID(id))
cpu.fpr(id) = originalState.fpr(id);
#if !(CPU(MIPS))
cpu.spr(flagsSPR) = originalState.spr(flagsSPR);
#endif
cpu.sp() = originalSP;
});
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(probeCallCount, 3);
}
#endif // ENABLE(MASM_PROBE)
void testByteSwap()
{
#if CPU(X86_64) || CPU(ARM64)
auto byteSwap16 = compile([] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.move(GPRInfo::argumentGPR0, GPRInfo::returnValueGPR);
jit.byteSwap16(GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(invoke<uint64_t>(byteSwap16, 0xaabbccddee001122), static_cast<uint64_t>(0x2211));
CHECK_EQ(invoke<uint64_t>(byteSwap16, 0xaabbccddee00ffaa), static_cast<uint64_t>(0xaaff));
auto byteSwap32 = compile([] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.move(GPRInfo::argumentGPR0, GPRInfo::returnValueGPR);
jit.byteSwap32(GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(invoke<uint64_t>(byteSwap32, 0xaabbccddee001122), static_cast<uint64_t>(0x221100ee));
CHECK_EQ(invoke<uint64_t>(byteSwap32, 0xaabbccddee00ffaa), static_cast<uint64_t>(0xaaff00ee));
auto byteSwap64 = compile([] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.move(GPRInfo::argumentGPR0, GPRInfo::returnValueGPR);
jit.byteSwap64(GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(invoke<uint64_t>(byteSwap64, 0xaabbccddee001122), static_cast<uint64_t>(0x221100eeddccbbaa));
CHECK_EQ(invoke<uint64_t>(byteSwap64, 0xaabbccddee00ffaa), static_cast<uint64_t>(0xaaff00eeddccbbaa));
#endif
}
void testMoveDoubleConditionally32()
{
#if CPU(X86_64) | CPU(ARM64)
double arg1 = 0;
double arg2 = 0;
const double zero = -0;
const double chosenDouble = 6.00000059604644775390625;
CHECK_EQ(static_cast<double>(static_cast<float>(chosenDouble)) == chosenDouble, false);
auto sel = compile([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&zero), FPRInfo::returnValueFPR);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&arg1), FPRInfo::fpRegT1);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&arg2), FPRInfo::fpRegT2);
jit.move(MacroAssembler::TrustedImm32(-1), GPRInfo::regT0);
jit.moveDoubleConditionally32(MacroAssembler::Equal, GPRInfo::regT0, GPRInfo::regT0, FPRInfo::fpRegT1, FPRInfo::fpRegT2, FPRInfo::returnValueFPR);
emitFunctionEpilogue(jit);
jit.ret();
});
arg1 = chosenDouble;
arg2 = 43;
CHECK_EQ(invoke<double>(sel), chosenDouble);
arg1 = 43;
arg2 = chosenDouble;
CHECK_EQ(invoke<double>(sel), 43.0);
#endif
}
void testMoveDoubleConditionally64()
{
#if CPU(X86_64) | CPU(ARM64)
double arg1 = 0;
double arg2 = 0;
const double zero = -0;
const double chosenDouble = 6.00000059604644775390625;
CHECK_EQ(static_cast<double>(static_cast<float>(chosenDouble)) == chosenDouble, false);
auto sel = compile([&] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&zero), FPRInfo::returnValueFPR);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&arg1), FPRInfo::fpRegT1);
jit.loadDouble(CCallHelpers::TrustedImmPtr(&arg2), FPRInfo::fpRegT2);
jit.move(MacroAssembler::TrustedImm64(-1), GPRInfo::regT0);
jit.moveDoubleConditionally64(MacroAssembler::Equal, GPRInfo::regT0, GPRInfo::regT0, FPRInfo::fpRegT1, FPRInfo::fpRegT2, FPRInfo::returnValueFPR);
emitFunctionEpilogue(jit);
jit.ret();
});
arg1 = chosenDouble;
arg2 = 43;
CHECK_EQ(invoke<double>(sel), chosenDouble);
arg1 = 43;
arg2 = chosenDouble;
CHECK_EQ(invoke<double>(sel), 43.0);
#endif
}
static void testCagePreservesPACFailureBit()
{
#if GIGACAGE_ENABLED
// Placate ASan builds and any environments that disables the Gigacage.
if (!Gigacage::shouldBeEnabled())
return;
RELEASE_ASSERT(!Gigacage::isDisablingPrimitiveGigacageForbidden());
auto cage = compile([] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.cageConditionally(Gigacage::Primitive, GPRInfo::argumentGPR0, GPRInfo::argumentGPR1, GPRInfo::argumentGPR2);
jit.move(GPRInfo::argumentGPR0, GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
void* ptr = Gigacage::tryMalloc(Gigacage::Primitive, 1);
void* taggedPtr = tagArrayPtr(ptr, 1);
RELEASE_ASSERT(hasOneBitSet(Gigacage::size(Gigacage::Primitive) << 2));
void* notCagedPtr = reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(ptr) + (Gigacage::size(Gigacage::Primitive) << 2));
CHECK_NOT_EQ(Gigacage::caged(Gigacage::Primitive, notCagedPtr), notCagedPtr);
void* taggedNotCagedPtr = tagArrayPtr(notCagedPtr, 1);
if (isARM64E()) {
// FIXME: This won't work if authentication failures trap but I don't know how to test for that right now.
CHECK_NOT_EQ(invoke<void*>(cage, taggedPtr, 2), ptr);
CHECK_EQ(invoke<void*>(cage, taggedNotCagedPtr, 1), untagArrayPtr(taggedPtr, 2));
} else
CHECK_EQ(invoke<void*>(cage, taggedPtr, 2), ptr);
CHECK_EQ(invoke<void*>(cage, taggedPtr, 1), ptr);
auto cageWithoutAuthentication = compile([] (CCallHelpers& jit) {
emitFunctionPrologue(jit);
jit.cageWithoutUntagging(Gigacage::Primitive, GPRInfo::argumentGPR0);
jit.move(GPRInfo::argumentGPR0, GPRInfo::returnValueGPR);
emitFunctionEpilogue(jit);
jit.ret();
});
CHECK_EQ(invoke<void*>(cageWithoutAuthentication, taggedPtr), taggedPtr);
if (isARM64E()) {
// FIXME: This won't work if authentication failures trap but I don't know how to test for that right now.
CHECK_NOT_EQ(invoke<void*>(cageWithoutAuthentication, taggedNotCagedPtr), taggedNotCagedPtr);
CHECK_NOT_EQ(untagArrayPtr(invoke<void*>(cageWithoutAuthentication, taggedNotCagedPtr), 1), notCagedPtr);
CHECK_NOT_EQ(invoke<void*>(cageWithoutAuthentication, taggedNotCagedPtr), taggedPtr);
CHECK_NOT_EQ(untagArrayPtr(invoke<void*>(cageWithoutAuthentication, taggedNotCagedPtr), 1), ptr);
}
Gigacage::free(Gigacage::Primitive, ptr);
#endif
}
#define RUN(test) do { \
if (!shouldRun(#test)) \
break; \
numberOfTests++; \
tasks.append( \
createSharedTask<void()>( \
[&] () { \
dataLog(#test "...\n"); \
test; \
dataLog(#test ": OK!\n"); \
})); \
} while (false);
void run(const char* filter)
{
JSC::initializeThreading();
unsigned numberOfTests = 0;
Deque<RefPtr<SharedTask<void()>>> tasks;
auto shouldRun = [&] (const char* testName) -> bool {
return !filter || WTF::findIgnoringASCIICaseWithoutLength(testName, filter) != WTF::notFound;
};
RUN(testSimple());
RUN(testGetEffectiveAddress(0xff00, 42, 8, CCallHelpers::TimesEight));
RUN(testGetEffectiveAddress(0xff00, -200, -300, CCallHelpers::TimesEight));
RUN(testBranchTruncateDoubleToInt32(0, 0));
RUN(testBranchTruncateDoubleToInt32(42, 42));
RUN(testBranchTruncateDoubleToInt32(42.7, 42));
RUN(testBranchTruncateDoubleToInt32(-1234, -1234));
RUN(testBranchTruncateDoubleToInt32(-1234.56, -1234));
RUN(testBranchTruncateDoubleToInt32(std::numeric_limits<double>::infinity(), 0));
RUN(testBranchTruncateDoubleToInt32(-std::numeric_limits<double>::infinity(), 0));
RUN(testBranchTruncateDoubleToInt32(std::numeric_limits<double>::quiet_NaN(), 0));
RUN(testBranchTruncateDoubleToInt32(std::numeric_limits<double>::signaling_NaN(), 0));
RUN(testBranchTruncateDoubleToInt32(std::numeric_limits<double>::max(), 0));
RUN(testBranchTruncateDoubleToInt32(-std::numeric_limits<double>::max(), 0));
// We run this last one to make sure that we don't use flags that were not
// reset to check a conversion result
RUN(testBranchTruncateDoubleToInt32(123, 123));
RUN(testCompareDouble(MacroAssembler::DoubleEqual));
RUN(testCompareDouble(MacroAssembler::DoubleNotEqual));
RUN(testCompareDouble(MacroAssembler::DoubleGreaterThan));
RUN(testCompareDouble(MacroAssembler::DoubleGreaterThanOrEqual));
RUN(testCompareDouble(MacroAssembler::DoubleLessThan));
RUN(testCompareDouble(MacroAssembler::DoubleLessThanOrEqual));
RUN(testCompareDouble(MacroAssembler::DoubleEqualOrUnordered));
RUN(testCompareDouble(MacroAssembler::DoubleNotEqualOrUnordered));
RUN(testCompareDouble(MacroAssembler::DoubleGreaterThanOrUnordered));
RUN(testCompareDouble(MacroAssembler::DoubleGreaterThanOrEqualOrUnordered));
RUN(testCompareDouble(MacroAssembler::DoubleLessThanOrUnordered));
RUN(testCompareDouble(MacroAssembler::DoubleLessThanOrEqualOrUnordered));
RUN(testMul32WithImmediates());
#if CPU(X86_64)
RUN(testBranchTestBit32RegReg());
RUN(testBranchTestBit32RegImm());
RUN(testBranchTestBit32AddrImm());
RUN(testBranchTestBit64RegReg());
RUN(testBranchTestBit64RegImm());
RUN(testBranchTestBit64AddrImm());
#endif
#if CPU(ARM64)
RUN(testMul32SignExtend());
#endif
#if CPU(X86) || CPU(X86_64) || CPU(ARM64)
RUN(testCompareFloat(MacroAssembler::DoubleEqual));
RUN(testCompareFloat(MacroAssembler::DoubleNotEqual));
RUN(testCompareFloat(MacroAssembler::DoubleGreaterThan));
RUN(testCompareFloat(MacroAssembler::DoubleGreaterThanOrEqual));
RUN(testCompareFloat(MacroAssembler::DoubleLessThan));
RUN(testCompareFloat(MacroAssembler::DoubleLessThanOrEqual));
RUN(testCompareFloat(MacroAssembler::DoubleEqualOrUnordered));
RUN(testCompareFloat(MacroAssembler::DoubleNotEqualOrUnordered));
RUN(testCompareFloat(MacroAssembler::DoubleGreaterThanOrUnordered));
RUN(testCompareFloat(MacroAssembler::DoubleGreaterThanOrEqualOrUnordered));
RUN(testCompareFloat(MacroAssembler::DoubleLessThanOrUnordered));
RUN(testCompareFloat(MacroAssembler::DoubleLessThanOrEqualOrUnordered));
#endif
#if ENABLE(MASM_PROBE)
RUN(testProbeReadsArgumentRegisters());
RUN(testProbeWritesArgumentRegisters());
RUN(testProbePreservesGPRS());
RUN(testProbeModifiesStackPointerToInsideProbeStateOnStack());
RUN(testProbeModifiesStackPointerToNBytesBelowSP());
RUN(testProbeModifiesProgramCounter());
RUN(testProbeModifiesStackValues());
#endif // ENABLE(MASM_PROBE)
RUN(testByteSwap());
RUN(testMoveDoubleConditionally32());
RUN(testMoveDoubleConditionally64());
RUN(testCagePreservesPACFailureBit());
if (tasks.isEmpty())
usage();
Lock lock;
Vector<Ref<Thread>> threads;
for (unsigned i = filter ? 1 : WTF::numberOfProcessorCores(); i--;) {
threads.append(
Thread::create(
"testmasm thread",
[&] () {
for (;;) {
RefPtr<SharedTask<void()>> task;
{
LockHolder locker(lock);
if (tasks.isEmpty())
return;
task = tasks.takeFirst();
}
task->run();
}
}));
}
for (auto& thread : threads)
thread->waitForCompletion();
crashLock.lock();
dataLog("Completed ", numberOfTests, " tests\n");
}
} // anonymous namespace
#else // not ENABLE(JIT)
static void run(const char*)
{
dataLog("JIT is not enabled.\n");
}
#endif // ENABLE(JIT)
int main(int argc, char** argv)
{
const char* filter = nullptr;
switch (argc) {
case 1:
break;
case 2:
filter = argv[1];
break;
default:
usage();
break;
}
run(filter);
return 0;
}
#if OS(WINDOWS)
extern "C" __declspec(dllexport) int WINAPI dllLauncherEntryPoint(int argc, const char* argv[])
{
return main(argc, const_cast<char**>(argv));
}
#endif