Merge pull request #15547 from ipsilon/eof-rjumps

eof: Support relative jumps

libevmasm/Assembly.cpp

@@ -1396,6 +1396,7 @@ LinkerObject const& Assembly::assembleEOF() const
 
 	m_tagPositionsInBytecode = std::vector<size_t>(m_usedTags, std::numeric_limits<size_t>::max());
 	std::map<size_t, uint16_t> dataSectionRef;
+	std::map<size_t, size_t> tagRef;
 
 	for (auto&& [codeSectionIndex, codeSection]: m_codeSections | ranges::views::enumerate)
 	{
@@ -1412,7 +1413,9 @@ LinkerObject const& Assembly::assembleEOF() const
 				solAssert(
 					item.instruction() != Instruction::DATALOADN &&
 					item.instruction() != Instruction::RETURNCONTRACT &&
-					item.instruction() != Instruction::EOFCREATE
+					item.instruction() != Instruction::EOFCREATE &&
+					item.instruction() != Instruction::RJUMP &&
+					item.instruction() != Instruction::RJUMPI
 				);
 				solAssert(!(item.instruction() >= Instruction::PUSH0 && item.instruction() <= Instruction::PUSH32));
 				ret.bytecode += assembleOperation(item);
@@ -1427,6 +1430,14 @@ LinkerObject const& Assembly::assembleEOF() const
 				ret.linkReferences.insert(linkRef);
 				break;
 			}
+			case RelativeJump:
+			case ConditionalRelativeJump:
+			{
+				ret.bytecode.push_back(static_cast<uint8_t>(item.instruction()));
+				tagRef[ret.bytecode.size()] = item.relativeJumpTagID();
+				appendBigEndianUint16(ret.bytecode, 0u);
+				break;
+			}
 			case EOFCreate:
 			{
 				ret.bytecode.push_back(static_cast<uint8_t>(Instruction::EOFCREATE));
@@ -1465,6 +1476,18 @@ LinkerObject const& Assembly::assembleEOF() const
 		setBigEndianUint16(ret.bytecode, codeSectionSizePositions[codeSectionIndex], ret.bytecode.size() - sectionStart);
 	}
 
+	for (auto const& [refPos, tagId]: tagRef)
+	{
+		solAssert(tagId < m_tagPositionsInBytecode.size(), "Reference to non-existing tag.");
+		size_t tagPos = m_tagPositionsInBytecode[tagId];
+		solAssert(tagPos != std::numeric_limits<size_t>::max(), "Reference to tag without position.");
+
+		ptrdiff_t const relativeJumpOffset = static_cast<ptrdiff_t>(tagPos) - (static_cast<ptrdiff_t>(refPos) + 2);
+		solRequire(-0x8000 <= relativeJumpOffset && relativeJumpOffset <= 0x7FFF, AssemblyException, "Relative jump too far");
+		solAssert(relativeJumpOffset < -2 || 0 <= relativeJumpOffset, "Relative jump offset into immediate argument.");
+		setBigEndianUint16(ret.bytecode, refPos, static_cast<size_t>(static_cast<uint16_t>(relativeJumpOffset)));
+	}
+
 	for (auto i: referencedSubIds)
 		ret.bytecode += m_subs[i]->assemble().bytecode;
 

libevmasm/Assembly.h

@@ -65,6 +65,7 @@ class Assembly
 	}
 
 	std::optional<uint8_t> eofVersion() const { return m_eofVersion; }
+	bool supportsRelativeJumps() const { return m_eofVersion.has_value(); }
 	AssemblyItem newTag() { assertThrow(m_usedTags < 0xffffffff, AssemblyException, ""); return AssemblyItem(Tag, m_usedTags++); }
 	AssemblyItem newPushTag() { assertThrow(m_usedTags < 0xffffffff, AssemblyException, ""); return AssemblyItem(PushTag, m_usedTags++); }
 	/// Returns a tag identified by the given name. Creates it if it does not yet exist.

libevmasm/AssemblyItem.cpp

@@ -62,20 +62,30 @@ AssemblyItem AssemblyItem::toSubAssemblyTag(size_t _subId) const
 
 std::pair<size_t, size_t> AssemblyItem::splitForeignPushTag() const
 {
-	assertThrow(m_type == PushTag || m_type == Tag, util::Exception, "");
+	solAssert(m_type == PushTag || m_type == Tag || m_type == RelativeJump || m_type == ConditionalRelativeJump);
 	u256 combined = u256(data());
 	size_t subId = static_cast<size_t>((combined >> 64) - 1);
 	size_t tag = static_cast<size_t>(combined & 0xffffffffffffffffULL);
 	return std::make_pair(subId, tag);
 }
 
+size_t AssemblyItem::relativeJumpTagID() const
+{
+	solAssert(m_type == RelativeJump || m_type == ConditionalRelativeJump);
+	auto const [subId, tagId] = splitForeignPushTag();
+	solAssert(subId == std::numeric_limits<size_t>::max(), "Relative jump to sub");
+	return tagId;
+}
+
 std::pair<std::string, std::string> AssemblyItem::nameAndData(langutil::EVMVersion _evmVersion) const
 {
 	switch (type())
 	{
 	case Operation:
 	case EOFCreate:
 	case ReturnContract:
+	case RelativeJump:
+	case ConditionalRelativeJump:
 		return {instructionInfo(instruction(), _evmVersion).name, m_data != nullptr ? toStringInHex(*m_data) : ""};
 	case Push:
 		return {"PUSH", toStringInHex(data())};
@@ -115,7 +125,8 @@ std::pair<std::string, std::string> AssemblyItem::nameAndData(langutil::EVMVersi
 
 void AssemblyItem::setPushTagSubIdAndTag(size_t _subId, size_t _tag)
 {
-	assertThrow(m_type == PushTag || m_type == Tag, util::Exception, "");
+	solAssert(m_type == PushTag || m_type == Tag || m_type == RelativeJump || m_type == ConditionalRelativeJump);
+	solAssert(!(m_type == RelativeJump || m_type == ConditionalRelativeJump) || _subId == std::numeric_limits<size_t>::max());
 	u256 data = _tag;
 	if (_subId != std::numeric_limits<size_t>::max())
 		data |= (u256(_subId) + 1) << 64;
@@ -167,6 +178,8 @@ size_t AssemblyItem::bytesRequired(size_t _addressLength, langutil::EVMVersion _
 	}
 	case VerbatimBytecode:
 		return std::get<2>(*m_verbatimBytecode).size();
+	case RelativeJump:
+	case ConditionalRelativeJump:
 	case AuxDataLoadN:
 		return 1 + 2;
 	case EOFCreate:
@@ -201,6 +214,8 @@ size_t AssemblyItem::returnValues() const
 	case Operation:
 	case EOFCreate:
 	case ReturnContract:
+	case RelativeJump:
+	case ConditionalRelativeJump:
 		// The latest EVMVersion is used here, since the InstructionInfo is assumed to be
 		// the same across all EVM versions except for the instruction name.
 		return static_cast<size_t>(instructionInfo(instruction(), EVMVersion()).ret);
@@ -236,6 +251,8 @@ bool AssemblyItem::canBeFunctional() const
 	case Operation:
 	case EOFCreate:
 	case ReturnContract:
+	case RelativeJump:
+	case ConditionalRelativeJump:
 		return !isDupInstruction(instruction()) && !isSwapInstruction(instruction());
 	case Push:
 	case PushTag:
@@ -360,6 +377,12 @@ std::string AssemblyItem::toAssemblyText(Assembly const& _assembly) const
 	case ReturnContract:
 		text = "returcontract{" +  std::to_string(static_cast<size_t>(data())) + "}";
 		break;
+	case RelativeJump:
+		text = "rjump{" + std::string("tag_") + std::to_string(relativeJumpTagID()) + "}";
+		break;
+	case ConditionalRelativeJump:
+		text = "rjumpi{" + std::string("tag_") + std::to_string(relativeJumpTagID()) + "}";
+		break;
 	}
 	if (m_jumpType == JumpType::IntoFunction || m_jumpType == JumpType::OutOfFunction)
 	{
@@ -380,6 +403,8 @@ std::ostream& solidity::evmasm::operator<<(std::ostream& _out, AssemblyItem cons
 	case Operation:
 	case EOFCreate:
 	case ReturnContract:
+	case RelativeJump:
+	case ConditionalRelativeJump:
 		_out << " " << instructionInfo(_item.instruction(), EVMVersion()).name;
 		if (_item.instruction() == Instruction::JUMP || _item.instruction() == Instruction::JUMPI)
 			_out << "\t" << _item.getJumpTypeAsString();

libevmasm/AssemblyItem.h

@@ -57,6 +57,8 @@ enum AssemblyItemType
 	AuxDataLoadN,
 	EOFCreate, ///< Creates new contract using subcontainer as initcode
 	ReturnContract, ///< Returns new container (with auxiliary data filled in) to be deployed
+	RelativeJump, ///< Jumps to relative position accordingly to its argument
+	ConditionalRelativeJump, ///< Same as RelativeJump but takes condition from the stack
 	VerbatimBytecode ///< Contains data that is inserted into the bytecode code section without modification.
 };
 
@@ -111,20 +113,32 @@ class AssemblyItem
 	{
 		return AssemblyItem(ReturnContract, Instruction::RETURNCONTRACT, _containerID, std::move(_debugData));
 	}
+	static AssemblyItem relativeJumpTo(AssemblyItem _tag, langutil::DebugData::ConstPtr _debugData = langutil::DebugData::create())
+	{
+		solAssert(_tag.type() == Tag);
+		return AssemblyItem(RelativeJump, Instruction::RJUMP, _tag.data(), _debugData);
+	}
+	static AssemblyItem conditionalRelativeJumpTo(AssemblyItem _tag, langutil::DebugData::ConstPtr _debugData = langutil::DebugData::create())
+	{
+		solAssert(_tag.type() == Tag);
+		return AssemblyItem(ConditionalRelativeJump, Instruction::RJUMPI, _tag.data(), _debugData);
+	}
 
 	AssemblyItem(AssemblyItem const&) = default;
 	AssemblyItem(AssemblyItem&&) = default;
 	AssemblyItem& operator=(AssemblyItem const&) = default;
 	AssemblyItem& operator=(AssemblyItem&&) = default;
 
-	AssemblyItem tag() const { assertThrow(m_type == PushTag || m_type == Tag, util::Exception, ""); return AssemblyItem(Tag, data()); }
-	AssemblyItem pushTag() const { assertThrow(m_type == PushTag || m_type == Tag, util::Exception, ""); return AssemblyItem(PushTag, data()); }
+	AssemblyItem tag() const { solAssert(m_type == PushTag || m_type == Tag || m_type == RelativeJump || m_type == ConditionalRelativeJump); return AssemblyItem(Tag, data()); }
+	AssemblyItem pushTag() const { solAssert(m_type == PushTag || m_type == Tag || m_type == RelativeJump || m_type == ConditionalRelativeJump); return AssemblyItem(PushTag, data()); }
 	/// Converts the tag to a subassembly tag. This has to be called in order to move a tag across assemblies.
 	/// @param _subId the identifier of the subassembly the tag is taken from.
 	AssemblyItem toSubAssemblyTag(size_t _subId) const;
 	/// @returns splits the data of the push tag into sub assembly id and actual tag id.
 	/// The sub assembly id of non-foreign push tags is -1.
 	std::pair<size_t, size_t> splitForeignPushTag() const;
+	/// @returns relative jump target tag ID. Asserts that it is not foreign tag.
+	size_t relativeJumpTagID() const;
 	/// Sets sub-assembly part and tag for a push tag.
 	void setPushTagSubIdAndTag(size_t _subId, size_t _tag);
 
@@ -145,9 +159,14 @@ class AssemblyItem
 	/// @returns true if the item has m_instruction properly set.
 	bool hasInstruction() const
 	{
-		return m_type == Operation || m_type == EOFCreate || m_type == ReturnContract;
+		return
+			m_type == Operation ||
+			m_type == EOFCreate ||
+			m_type == ReturnContract ||
+			m_type == RelativeJump ||
+			m_type == ConditionalRelativeJump;
 	}
-	/// @returns the instruction of this item (only valid if type() == Operation || EOFCreate || ReturnContract)
+	/// @returns the instruction of this item (only valid if hasInstruction returns true)
 	Instruction instruction() const
 	{
 		solAssert(hasInstruction());

libevmasm/BlockDeduplicator.cpp

@@ -106,7 +106,7 @@ bool BlockDeduplicator::applyTagReplacement(
 {
 	bool changed = false;
 	for (AssemblyItem& item: _items)
-		if (item.type() == PushTag)
+		if (item.type() == PushTag || item.type() == RelativeJump || item.type() == ConditionalRelativeJump)
 		{
 			size_t subId;
 			size_t tagId;
@@ -131,7 +131,7 @@ BlockDeduplicator::BlockIterator& BlockDeduplicator::BlockIterator::operator++()
 {
 	if (it == end)
 		return *this;
-	if (SemanticInformation::altersControlFlow(*it) && *it != AssemblyItem{Instruction::JUMPI})
+	if (SemanticInformation::altersControlFlow(*it) && *it != AssemblyItem{Instruction::JUMPI} && it->type() != ConditionalRelativeJump)
 		it = end;
 	else
 	{

libevmasm/GasMeter.cpp

@@ -272,6 +272,8 @@ unsigned GasMeter::runGas(Instruction _instruction, langutil::EVMVersion _evmVer
 	{
 	case Tier::Zero:        return GasCosts::tier0Gas;
 	case Tier::Base:        return GasCosts::tier1Gas;
+	case Tier::RJump:       return GasCosts::tier1Gas;
+	case Tier::RJumpI:      return GasCosts::rjumpiGas;
 	case Tier::VeryLow:     return GasCosts::tier2Gas;
 	case Tier::Low:         return GasCosts::tier3Gas;
 	case Tier::Mid:         return GasCosts::tier4Gas;

libevmasm/GasMeter.h

@@ -177,6 +177,7 @@ namespace GasCosts
 		return _evmVersion >= langutil::EVMVersion::istanbul() ? 16 : 68;
 	}
 	static unsigned const copyGas = 3;
+	static unsigned const rjumpiGas = 4;
 }
 
 /**

libevmasm/Instruction.cpp

@@ -169,6 +169,8 @@ std::map<std::string, Instruction> const solidity::evmasm::c_instructions =
 	{ "LOG3", Instruction::LOG3 },
 	{ "LOG4", Instruction::LOG4 },
 	{ "DATALOADN", Instruction::DATALOADN },
+	{ "RJUMP", Instruction::RJUMP },
+	{ "RJUMPI", Instruction::RJUMPI },
 	{ "EOFCREATE", Instruction::EOFCREATE },
 	{ "RETURNCONTRACT", Instruction::RETURNCONTRACT },
 	{ "CREATE", Instruction::CREATE },
@@ -256,6 +258,8 @@ static std::map<Instruction, InstructionInfo> const c_instructionInfo =
 	{Instruction::GAS,            {"GAS",             0,  0,   1,  false,      Tier::Base}},
 	{Instruction::JUMPDEST,       {"JUMPDEST",        0,  0,   0,  true,       Tier::Special}},
 	{Instruction::DATALOADN,      {"DATALOADN",       2,  0,   1,  true,       Tier::Low}},
+	{Instruction::RJUMP,          {"RJUMP",           2,  0,   0,  true,       Tier::RJump}},
+	{Instruction::RJUMPI,         {"RJUMPI",          2,  1,   0,  true,       Tier::RJumpI}},
 	{Instruction::PUSH0,          {"PUSH0",           0,  0,   1,  false,      Tier::Base}},
 	{Instruction::PUSH1,          {"PUSH1",           1,  0,   1,  false,      Tier::VeryLow}},
 	{Instruction::PUSH2,          {"PUSH2",           2,  0,   1,  false,      Tier::VeryLow}},

libevmasm/Instruction.h

@@ -183,6 +183,9 @@ enum class Instruction: uint8_t
 	LOG4,                     ///< Makes a log entry; 4 topics.
 
 	DATALOADN = 0xd1,         ///< load data from EOF data section
+
+	RJUMP = 0xe0,             ///< relative jump
+	RJUMPI = 0xe1,            ///< conditional relative jump
 	EOFCREATE = 0xec,         ///< create a new account with associated container code.
 	RETURNCONTRACT = 0xee,    ///< return container to be deployed with axiliary data filled in.
 	CREATE = 0xf0,            ///< create a new account with associated code
@@ -298,7 +301,9 @@ enum class Tier
 	// NOTE: Tiers should be ordered by cost, since we sometimes perform comparisons between them.
 	Zero = 0,   // 0, Zero
 	Base,       // 2, Quick
+	RJump,      // 2, RJump
 	VeryLow,    // 3, Fastest
+	RJumpI,     // 4,
 	Low,        // 5, Fast
 	Mid,        // 8, Mid
 	High,       // 10, Slow

libevmasm/JumpdestRemover.cpp

@@ -58,7 +58,7 @@ std::set<size_t> JumpdestRemover::referencedTags(AssemblyItems const& _items, si
 {
 	std::set<size_t> ret;
 	for (auto const& item: _items)
-		if (item.type() == PushTag)
+		if (item.type() == PushTag || item.type() == RelativeJump || item.type() == ConditionalRelativeJump)
 		{
 			auto subAndTag = item.splitForeignPushTag();
 			if (subAndTag.first == _subId)