basic data structures for the disassembler

include/fxos/disassembly.h

@@ -6,10 +6,12 @@
 #define LIBFXOS_DISASSEMBLY_H
 
 #include <fxos/lang.h>
+#include <fxos/target.h>
 #include <fxos/semantics.h>
 
 #include <vector>
 #include <optional>
+#include <map>
 
 namespace FxOS {
 
@@ -20,15 +22,46 @@ namespace FxOS {
    instructions with parameters, not manually. See <fxos/load.h>. */
 void register_instruction(Instruction ins);
 
-/* A loaded instruction with all relevant information, and more. */
-class LoadedInstruction
+/* An argument for a concrete instruction. */
+struct ConcreteInstructionArg
 {
-private:
-	/* What instruction it is */
-	Instruction &m_inst;
+	ConcreteInstructionArg();
 
-	/* Operands for arguments, if they have been determined */
-//	std::vector<std::optional<Operand>> args;
+	//---
+	// Data set by the abstract interpretation passes
+	//---
+
+	/* Location in CPU or memory, if that can be determined */
+	std::optional<Location> loc;
+	/* Alternatively, data type, which can sometimes be determined uniquely
+	   even if the location is not constant */
+	std::optional<DataType> type;
+
+	//---
+	// Data set by the syscall and regs passes
+	//---
+
+	/* If the value is a syscall address, the syscall's id */
+	int syscall_id;
+	/* If the value is a peripheral register, its address */
+	uint32_t reg_address;
+};
+
+/* A loaded and annotated instruction. */
+struct ConcreteInstruction
+{
+	ConcreteInstruction(Instruction &inst);
+	ConcreteInstruction(ConcreteInstruction const &other) = default;
+
+	/* What instruction it is */
+	Instruction &inst;
+
+	/* Argument information (contains data set by several passes) */
+	ConcreteInstructionArg args[2];
+
+	//---
+	// Data set by the pcrel pass
+	//---
 
 	/* Jump targets, used for jump instructions only. The first jmp is for
 	   unconditional jumps; jmpt and jmpf are for conditional jumps. In
@@ -36,6 +69,34 @@ private:
 	   preceding branch due to the delay slot mechanism. */
 	union { uint32_t jmp, jmpt; };
  	uint32_t jmpf;
+
+ 	//---
+ 	// Data set by the cfg pass
+ 	//---
+
+ 	/* Whether this instruction is a basic block leader */
+ 	bool leader;
+};
+
+/* Short aliases */
+using CI    = ConcreteInstruction;
+using CIArg = ConcreteInstructionArg;
+
+/* Disassembly interface that automatically loads code from a target */
+class Disassembly
+{
+public:
+	Disassembly(Target &target);
+
+	/* Get the storage to any concrete instruction. The instruction will be
+	   loaded and initialized if it had not been read before. */
+	ConcreteInstruction &readins(uint32_t pc);
+
+private:
+	/* Underlying target */
+	Target &m_target;
+	/* Loaded instructions by address */
+	std::map<uint32_t, ConcreteInstruction> m_instructions;
 };
 
 } /* namespace FxOS */

include/fxos/semantics.h

@@ -24,6 +24,13 @@ namespace FxOS {
 class DataType
 {
 public:
+	/* Copy constructor */
+	DataType(DataType const &other);
+	DataType & operator = (DataType other);
+
+	/* Destructor that takes into account the non-trivial union */
+	~DataType();
+
 	enum DataKind {
 		/* Base types */
 		Integral,
@@ -44,7 +51,6 @@ public:
 	/* Type alignment, can only be 1, 2 or 4 */
 	uint16_t align;
 
-private:
 	/* Type of fields in bit fields */
 	using Field = std::pair<int, std::string>;
 
@@ -52,14 +58,18 @@ private:
 		/* For integer types of size 1, whether to display as char
 		   (might be extended to more attributes later) */
 		bool ischar;
-		/* For bit field types */
-		std::vector<Field> fields;
-		/* For struct types */
-		std::vector<DataType> attributes;
 		/* For array types, number of elements (0 if unknown or
 		   variable-size NUL-terminated strings) */
 		int elements;
 	};
+
+	/* The following members are not in the union because they have non-
+	   trivial destructors/copy and I don't want to care. */
+
+	/* For bit field types */
+	std::vector<Field> fields;
+	/* For struct types */
+	std::vector<DataType> attributes;
 };
 
 //---

lib/disassembly.cpp

@@ -21,4 +21,53 @@ void register_instruction(Instruction ins)
 	insmap[opcode] = ins;
 }
 
+//---
+// Concrete (instanciated) arguments and instructions
+//---
+
+ConcreteInstructionArg::ConcreteInstructionArg():
+	loc {}, type {}, syscall_id {-1}
+{
+	reg_address = -1;
+}
+
+ConcreteInstruction::ConcreteInstruction(Instruction &inst):
+	inst {inst}, jmpt {}, jmpf {}, leader {false}
+{
+}
+
+//---
+// Disassembler interface
+//---
+
+Disassembly::Disassembly(Target &target):
+	m_target {target}, m_instructions {}
+{
+}
+
+ConcreteInstruction &Disassembly::readins(uint32_t pc)
+{
+	if(pc & 1) throw std::runtime_error("Disassembly::readins at odd PC");
+
+	try
+	{
+		return m_instructions.at(pc);
+	}
+	catch(std::out_of_range &e)
+	{
+		uint16_t opcode = m_target.read_u16(pc);
+		if(!insmap[opcode])
+		{
+			throw std::runtime_error("No instruction for opcode");
+		}
+
+		Instruction &inst = *insmap[opcode];
+
+		ConcreteInstruction ci(inst);
+//		std::pair<uint32_t, ConcreteInstruction> p(pc, ci);
+//		m_instructions.emplace(std::make_pair(pc, ci));
+		return m_instructions.at(pc);
+	}
+}
+
 } /* namespace FxOS */

lib/semantics.cpp

@@ -0,0 +1,5 @@
+#include <fxos/semantics.h>
+
+namespace FxOS {
+
+} /* namespace FxOS */