private void RunTest(string sExp, Action<ProcedureBuilder> builder)
{
var pb = new ProcedureBuilder(this.pb.Program.Architecture);
builder(pb);
var proc = pb.Procedure;
var dg = new DominatorGraph<Block>(proc.ControlGraph, proc.EntryBlock);
// Perform the initial transformation
var ssa = new SsaTransform(programFlow, proc, dg);
// Propagate values and simplify the results.
// We hope the the sequence
// esp = fp - 4
// mov [esp-4],eax
// will become
// esp_2 = fp - 4
// mov [fp - 8],eax
var vp = new ValuePropagator(ssa.SsaState.Identifiers, proc);
vp.Transform();
ssa.RenameFrameAccesses = true;
ssa.AddUseInstructions = true;
ssa.Transform();
var writer = new StringWriter();
proc.Write(false, writer);
var sActual = writer.ToString();
if (sActual != sExp)
Debug.Print(sActual);
Assert.AreEqual(sExp, sActual);
}
private Procedure RunTest(string sExp, IProcessorArchitecture arch, Func<Procedure> mkProc)
{
var proc = mkProc();
progBuilder.ResolveUnresolved();
var ssa = new SsaTransform(pf, proc, proc.CreateBlockDominatorGraph());
var vp = new ValuePropagator(arch, ssa.SsaState.Identifiers, proc);
vp.Transform();
ssa.RenameFrameAccesses = true;
ssa.AddUseInstructions = true;
ssa.Transform();
vp.Transform();
var trf = new TrashedRegisterFinder2(
arch,
pf,
proc,
ssa.SsaState.Identifiers,
NullDecompilerEventListener.Instance);
var flow = trf.Compute();
var sw = new StringWriter();
sw.Write("Preserved: ");
sw.WriteLine(string.Join(",", flow.Preserved.OrderBy(p => p.ToString())));
sw.Write("Trashed: ");
sw.WriteLine(string.Join(",", flow.Trashed.OrderBy(p => p.ToString())));
if (flow.Constants.Count > 0)
{
sw.Write("Constants: ");
sw.WriteLine(string.Join(
",",
flow.Constants
.OrderBy(kv => kv.Key.ToString())
.Select(kv => string.Format(
"{0}:{1}", kv.Key, kv.Value))));
}
var sActual = sw.ToString();
if (sActual != sExp)
{
proc.Dump(true);
Debug.WriteLine(sActual);
Assert.AreEqual(sExp, sActual);
}
pf.ProcedureFlows2.Add(proc, flow);
return proc;
}
private void UntangleProcedureScc(IList <Procedure> procs)
{
if (procs.Count == 1)
{
var proc = procs[0];
Aliases alias = new Aliases(proc, program.Architecture, flow);
alias.Transform();
// Transform the procedure to SSA state. When encountering 'call' instructions,
// they can be to functions already visited. If so, they have a "ProcedureFlow"
// associated with them. If they have not been visited, or are computed destinations
// (e.g. vtables) they will have no "ProcedureFlow" associated with them yet, in
// which case the the SSA treats the call as a "hell node".
var doms = proc.CreateBlockDominatorGraph();
var sst = new SsaTransform(flow, proc, importResolver, doms);
var ssa = sst.SsaState;
// Propagate condition codes and registers. At the end, the hope is that
// all statements like (x86) mem[esp_42+4] will have been converted to
// mem[fp - 30]. We also hope that procedure constants kept in registers
// are propagated to the corresponding call sites.
var cce = new ConditionCodeEliminator(ssa.Identifiers, program.Platform);
cce.Transform();
var vp = new ValuePropagator(program.Architecture, ssa.Identifiers, proc);
vp.Transform();
// Now compute SSA for the stack-based variables as well. That is:
// mem[fp - 30] becomes wLoc30, while
// mem[fp + 30] becomes wArg30.
// This allows us to compute the dataflow of this procedure.
sst.RenameFrameAccesses = true;
sst.AddUseInstructions = true;
sst.Transform();
// Propagate those newly discovered identifiers.
vp.Transform();
// At this point, the computation of _actual_ ProcedureFlow should be possible.
var tid = new TrashedRegisterFinder2(program.Architecture, flow, proc, ssa.Identifiers, this.eventListener);
tid.Compute();
DeadCode.Eliminate(proc, ssa);
// Build expressions. A definition with a single use can be subsumed
// into the using expression.
var coa = new Coalescer(proc, ssa);
coa.Transform();
DeadCode.Eliminate(proc, ssa);
var liv = new LinearInductionVariableFinder(
proc,
ssa.Identifiers,
new BlockDominatorGraph(proc.ControlGraph, proc.EntryBlock));
liv.Find();
foreach (var de in liv.Contexts)
{
var str = new StrengthReduction(ssa, de.Key, de.Value);
str.ClassifyUses();
str.ModifyUses();
}
//var opt = new OutParameterTransformer(proc, ssa.Identifiers);
//opt.Transform();
DeadCode.Eliminate(proc, ssa);
// Definitions with multiple uses and variables joined by PHI functions become webs.
var web = new WebBuilder(proc, ssa.Identifiers, program.InductionVariables);
web.Transform();
ssa.ConvertBack(false);
}
else
{
throw new NotImplementedException();
}
}
private void UntangleProcedureScc(IList<Procedure> procs)
{
if (procs.Count == 1)
{
var proc = procs[0];
Aliases alias = new Aliases(proc, program.Architecture, flow);
alias.Transform();
// Transform the procedure to SSA state. When encountering 'call' instructions,
// they can be to functions already visited. If so, they have a "ProcedureFlow"
// associated with them. If they have not been visited, or are computed destinations
// (e.g. vtables) they will have no "ProcedureFlow" associated with them yet, in
// which case the the SSA treats the call as a "hell node".
var doms = proc.CreateBlockDominatorGraph();
var sst = new SsaTransform(flow, proc, doms);
var ssa = sst.SsaState;
// Propagate condition codes and registers. At the end, the hope is that
// all statements like (x86) mem[esp_42+4] will have been converted to
// mem[fp - 30]. We also hope that procedure constants kept in registers
// are propagated to the corresponding call sites.
var cce = new ConditionCodeEliminator(ssa.Identifiers, program.Platform);
cce.Transform();
var vp = new ValuePropagator(program.Architecture, ssa.Identifiers, proc);
vp.Transform();
// Now compute SSA for the stack-based variables as well. That is:
// mem[fp - 30] becomes wLoc30, while
// mem[fp + 30] becomes wArg30.
// This allows us to compute the dataflow of this procedure.
sst.RenameFrameAccesses = true;
sst.AddUseInstructions = true;
sst.Transform();
// Propagate those newly discovered identifiers.
vp.Transform();
// At this point, the computation of _actual_ ProcedureFlow should be possible.
var tid = new TrashedRegisterFinder2(program.Architecture, flow, proc, ssa.Identifiers, this.eventListener);
tid.Compute();
DeadCode.Eliminate(proc, ssa);
// Build expressions. A definition with a single use can be subsumed
// into the using expression.
var coa = new Coalescer(proc, ssa);
coa.Transform();
DeadCode.Eliminate(proc, ssa);
var liv = new LinearInductionVariableFinder(
proc,
ssa.Identifiers,
new BlockDominatorGraph(proc.ControlGraph, proc.EntryBlock));
liv.Find();
foreach (var de in liv.Contexts)
{
var str = new StrengthReduction(ssa, de.Key, de.Value);
str.ClassifyUses();
str.ModifyUses();
}
//var opt = new OutParameterTransformer(proc, ssa.Identifiers);
//opt.Transform();
DeadCode.Eliminate(proc, ssa);
// Definitions with multiple uses and variables joined by PHI functions become webs.
var web = new WebBuilder(proc, ssa.Identifiers, program.InductionVariables);
web.Transform();
ssa.ConvertBack(false);
}
else
{
throw new NotImplementedException();
}
}
/// <summary>
/// Converts all registers and stack accesses to SSA variables.
/// </summary>
/// <param name="proc"></param>
/// <returns>The SsaTransform for the procedure.</returns>
public SsaTransform ConvertToSsa(Procedure proc)
{
if (program.NeedsSsaTransform)
{
// Transform the procedure to SSA state. When encountering 'call'
// instructions, they can be to functions already visited. If so,
// they have a "ProcedureFlow" associated with them. If they have
// not been visited, or are computed destinations (e.g. vtables)
// they will have no "ProcedureFlow" associated with them yet, in
// which case the the SSA treats the call as a "hell node".
var sst = new SsaTransform(program, proc, sccProcs, dynamicLinker, this.ProgramDataFlow);
var ssa = sst.Transform();
DumpWatchedProcedure("After SSA", ssa.Procedure);
// Merge unaligned memory accesses.
var fuser = new UnalignedMemoryAccessFuser(ssa);
fuser.Transform();
// After value propagation expressions like (x86)
// mem[esp_42+4] will have been converted to mem[fp - 30].
// We also hope that procedure constants
// kept in registers are propagated to the corresponding call
// sites.
var vp = new ValuePropagator(program.SegmentMap, ssa, program.CallGraph, dynamicLinker, eventListener);
vp.Transform();
DumpWatchedProcedure("After first VP", ssa.Procedure);
// Fuse additions and subtractions that are linked by the carry flag.
var larw = new LongAddRewriter(ssa);
larw.Transform();
// Propagate condition codes and registers.
var cce = new ConditionCodeEliminator(ssa, program.Platform);
cce.Transform();
vp.Transform();
DumpWatchedProcedure("After CCE", ssa.Procedure);
// Now compute SSA for the stack-based variables as well. That is:
// mem[fp - 30] becomes wLoc30, while
// mem[fp + 30] becomes wArg30.
// This allows us to compute the dataflow of this procedure.
sst.RenameFrameAccesses = true;
sst.Transform();
DumpWatchedProcedure("After SSA frame accesses", ssa.Procedure);
var icrw = new IndirectCallRewriter(program, ssa, eventListener);
while (!eventListener.IsCanceled() && icrw.Rewrite())
{
vp.Transform();
sst.RenameFrameAccesses = true;
sst.Transform();
}
var fpuGuesser = new FpuStackReturnGuesser(ssa);
fpuGuesser.Rewrite();
// By placing use statements in the exit block, we will collect
// reaching definitions in the use statements.
sst.AddUsesToExitBlock();
sst.RemoveDeadSsaIdentifiers();
// Backpropagate stack pointer from procedure return.
var spBackpropagator = new StackPointerBackpropagator(ssa);
spBackpropagator.BackpropagateStackPointer();
DumpWatchedProcedure("After SP BP", ssa.Procedure);
// Propagate those newly created stack-based identifiers.
vp.Transform();
DumpWatchedProcedure("After VP2", ssa.Procedure);
return(sst);
}
else
{
// We are assuming phi functions are already generated.
var sst = new SsaTransform(program, proc, sccProcs, dynamicLinker, this.ProgramDataFlow);
return(sst);
}
}
/// <summary>
/// Processes procedures individually, building complex expression trees out
/// of the simple, close-to-the-machine code generated by the disassembly.
/// </summary>
/// <param name="rl"></param>
public void BuildExpressionTrees()
{
int i = 0;
foreach (Procedure proc in program.Procedures.Values)
{
if (eventListener.IsCanceled())
{
break;
}
eventListener.ShowProgress("Building complex expressions.", i, program.Procedures.Values.Count);
++i;
try
{
var larw = new LongAddRewriter(proc, program.Architecture);
larw.Transform();
Aliases alias = new Aliases(proc, program.Architecture, flow);
alias.Transform();
var doms = new DominatorGraph <Block>(proc.ControlGraph, proc.EntryBlock);
var sst = new SsaTransform(flow, proc, importResolver, doms, new HashSet <RegisterStorage>());
var ssa = sst.SsaState;
var vp = new ValuePropagator(program.Architecture, ssa, eventListener);
sst.RenameFrameAccesses = true;
var icrw = new IndirectCallRewriter(program, ssa, eventListener);
while (!eventListener.IsCanceled() && icrw.Rewrite())
{
vp.Transform();
sst.Transform();
}
var cce = new ConditionCodeEliminator(ssa, program.Platform);
cce.Transform();
//var cd = new ConstDivisionImplementedByMultiplication(ssa);
//cd.Transform();
DeadCode.Eliminate(proc, ssa);
vp.Transform();
DeadCode.Eliminate(proc, ssa);
// Build expressions. A definition with a single use can be subsumed
// into the using expression.
var coa = new Coalescer(proc, ssa);
coa.Transform();
DeadCode.Eliminate(proc, ssa);
vp.Transform();
var liv = new LinearInductionVariableFinder(
proc,
ssa.Identifiers,
new BlockDominatorGraph(proc.ControlGraph, proc.EntryBlock));
liv.Find();
foreach (KeyValuePair <LinearInductionVariable, LinearInductionVariableContext> de in liv.Contexts)
{
var str = new StrengthReduction(ssa, de.Key, de.Value);
str.ClassifyUses();
str.ModifyUses();
}
var opt = new OutParameterTransformer(proc, ssa.Identifiers);
opt.Transform();
DeadCode.Eliminate(proc, ssa);
// Definitions with multiple uses and variables joined by PHI functions become webs.
var web = new WebBuilder(proc, ssa.Identifiers, program.InductionVariables);
web.Transform();
ssa.ConvertBack(false);
}
catch (StatementCorrelatedException stex)
{
eventListener.Error(
eventListener.CreateStatementNavigator(program, stex.Statement),
stex,
"An error occurred during data flow analysis.");
}
catch (Exception ex)
{
eventListener.Error(
new NullCodeLocation(proc.Name),
ex,
"An error occurred during data flow analysis.");
}
}
}
/// <summary>
/// Processes procedures individually, building complex expression trees out
/// of the simple, close-to-the-machine code generated by the disassembly.
/// </summary>
/// <param name="rl"></param>
public void BuildExpressionTrees()
{
int i = 0;
foreach (Procedure proc in program.Procedures.Values)
{
if (eventListener.IsCanceled())
break;
eventListener.ShowProgress("Building complex expressions.", i, program.Procedures.Values.Count);
++i;
try
{
var larw = new LongAddRewriter(proc, program.Architecture);
larw.Transform();
Aliases alias = new Aliases(proc, program.Architecture, flow);
alias.Transform();
var doms = new DominatorGraph<Block>(proc.ControlGraph, proc.EntryBlock);
var sst = new SsaTransform(flow, proc, importResolver, doms, new HashSet<RegisterStorage>());
var ssa = sst.SsaState;
var vp = new ValuePropagator(program.Architecture, ssa);
sst.RenameFrameAccesses = true;
var icrw = new IndirectCallRewriter(program, ssa, eventListener);
while (!eventListener.IsCanceled() && icrw.Rewrite())
{
vp.Transform();
sst.Transform();
}
var cce = new ConditionCodeEliminator(ssa, program.Platform);
cce.Transform();
//var cd = new ConstDivisionImplementedByMultiplication(ssa);
//cd.Transform();
DeadCode.Eliminate(proc, ssa);
vp.Transform();
DeadCode.Eliminate(proc, ssa);
// Build expressions. A definition with a single use can be subsumed
// into the using expression.
var coa = new Coalescer(proc, ssa);
coa.Transform();
DeadCode.Eliminate(proc, ssa);
vp.Transform();
var liv = new LinearInductionVariableFinder(
proc,
ssa.Identifiers,
new BlockDominatorGraph(proc.ControlGraph, proc.EntryBlock));
liv.Find();
foreach (KeyValuePair<LinearInductionVariable, LinearInductionVariableContext> de in liv.Contexts)
{
var str = new StrengthReduction(ssa, de.Key, de.Value);
str.ClassifyUses();
str.ModifyUses();
}
var opt = new OutParameterTransformer(proc, ssa.Identifiers);
opt.Transform();
DeadCode.Eliminate(proc, ssa);
// Definitions with multiple uses and variables joined by PHI functions become webs.
var web = new WebBuilder(proc, ssa.Identifiers, program.InductionVariables);
web.Transform();
ssa.ConvertBack(false);
}
catch (StatementCorrelatedException stex)
{
eventListener.Error(
eventListener.CreateStatementNavigator(program, stex.Statement),
stex,
"An error occurred during data flow analysis.");
}
catch (Exception ex)
{
eventListener.Error(
new NullCodeLocation(proc.Name),
ex,
"An error occurred during data flow analysis.");
}
}
}
private void RunTest(string sExp, Action<ProcedureBuilder> builder)
{
var pb = new ProcedureBuilder(this.pb.Program.Architecture);
builder(pb);
var proc = pb.Procedure;
var dg = new DominatorGraph<Block>(proc.ControlGraph, proc.EntryBlock);
var project = new Project
{
Programs = { this.pb.Program }
};
var importResolver = new ImportResolver(
project,
this.pb.Program,
new FakeDecompilerEventListener());
var arch = new FakeArchitecture();
var platform = new FakePlatform(null, arch);
// Register r1 is assumed to always be implicit when calling
// another procedure.
var implicitRegs = new HashSet<RegisterStorage>
{
arch.GetRegister(1)
};
Debug.Print("GetRegister(1) {0}", arch.GetRegister(1));
this.pb.Program.Platform = platform;
this.pb.Program.Platform = new FakePlatform(null, new FakeArchitecture());
this.pb.Program.SegmentMap = new SegmentMap(
Address.Ptr32(0x0000),
new ImageSegment(
".text",
Address.Ptr32(0),
0x40000,
AccessMode.ReadWriteExecute));
// Perform the initial transformation
var ssa = new SsaTransform(programFlow, proc, importResolver, dg, implicitRegs);
// Propagate values and simplify the results.
// We hope the the sequence
// esp = fp - 4
// mov [esp-4],eax
// will become
// esp_2 = fp - 4
// mov [fp - 8],eax
var vp = new ValuePropagator(this.pb.Program.Architecture, ssa.SsaState);
vp.Transform();
ssa.RenameFrameAccesses = true;
ssa.AddUseInstructions = true;
ssa.Transform();
var writer = new StringWriter();
proc.Write(false, writer);
var sActual = writer.ToString();
if (sActual != sExp)
Debug.Print(sActual);
Assert.AreEqual(sExp, sActual);
}
