1 //===- BugDriver.cpp - Top-Level BugPoint class implementation ------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This class contains all of the shared state and information that is used by
10 // the BugPoint tool to track down errors in optimizations.  This class is the
11 // main driver class that invokes all sub-functionality.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "BugDriver.h"
16 #include "ToolRunner.h"
17 #include "llvm/IR/Module.h"
18 #include "llvm/IR/Verifier.h"
19 #include "llvm/IRReader/IRReader.h"
20 #include "llvm/Linker/Linker.h"
21 #include "llvm/Pass.h"
22 #include "llvm/Support/CommandLine.h"
23 #include "llvm/Support/FileUtilities.h"
24 #include "llvm/Support/Host.h"
25 #include "llvm/Support/SourceMgr.h"
26 #include "llvm/Support/raw_ostream.h"
27 #include <memory>
28 using namespace llvm;
29 
30 namespace llvm {
31 Triple TargetTriple;
32 }
33 
34 DiscardTemp::~DiscardTemp() {
35   if (SaveTemps) {
36     if (Error E = File.keep())
37       errs() << "Failed to keep temp file " << toString(std::move(E)) << '\n';
38     return;
39   }
40   if (Error E = File.discard())
41     errs() << "Failed to delete temp file " << toString(std::move(E)) << '\n';
42 }
43 
44 // Anonymous namespace to define command line options for debugging.
45 //
46 namespace {
47 // Output - The user can specify a file containing the expected output of the
48 // program.  If this filename is set, it is used as the reference diff source,
49 // otherwise the raw input run through an interpreter is used as the reference
50 // source.
51 //
52 cl::opt<std::string> OutputFile("output",
53                                 cl::desc("Specify a reference program output "
54                                          "(for miscompilation detection)"));
55 }
56 
57 /// If we reduce or update the program somehow, call this method to update
58 /// bugdriver with it.  This deletes the old module and sets the specified one
59 /// as the current program.
60 void BugDriver::setNewProgram(std::unique_ptr<Module> M) {
61   Program = std::move(M);
62 }
63 
64 /// getPassesString - Turn a list of passes into a string which indicates the
65 /// command line options that must be passed to add the passes.
66 ///
67 std::string llvm::getPassesString(const std::vector<std::string> &Passes) {
68   std::string Result;
69   for (unsigned i = 0, e = Passes.size(); i != e; ++i) {
70     if (i)
71       Result += " ";
72     Result += "-";
73     Result += Passes[i];
74   }
75   return Result;
76 }
77 
78 BugDriver::BugDriver(const char *toolname, bool find_bugs, unsigned timeout,
79                      unsigned memlimit, bool use_valgrind, LLVMContext &ctxt)
80     : Context(ctxt), ToolName(toolname), ReferenceOutputFile(OutputFile),
81       Program(nullptr), Interpreter(nullptr), SafeInterpreter(nullptr),
82       cc(nullptr), run_find_bugs(find_bugs), Timeout(timeout),
83       MemoryLimit(memlimit), UseValgrind(use_valgrind) {}
84 
85 BugDriver::~BugDriver() {
86   if (Interpreter != SafeInterpreter)
87     delete Interpreter;
88   delete SafeInterpreter;
89   delete cc;
90 }
91 
92 std::unique_ptr<Module> llvm::parseInputFile(StringRef Filename,
93                                              LLVMContext &Ctxt) {
94   SMDiagnostic Err;
95   std::unique_ptr<Module> Result = parseIRFile(Filename, Err, Ctxt);
96   if (!Result) {
97     Err.print("bugpoint", errs());
98     return Result;
99   }
100 
101   if (verifyModule(*Result, &errs())) {
102     errs() << "bugpoint: " << Filename << ": error: input module is broken!\n";
103     return std::unique_ptr<Module>();
104   }
105 
106   // If we don't have an override triple, use the first one to configure
107   // bugpoint, or use the host triple if none provided.
108   if (TargetTriple.getTriple().empty()) {
109     Triple TheTriple(Result->getTargetTriple());
110 
111     if (TheTriple.getTriple().empty())
112       TheTriple.setTriple(sys::getDefaultTargetTriple());
113 
114     TargetTriple.setTriple(TheTriple.getTriple());
115   }
116 
117   Result->setTargetTriple(TargetTriple.getTriple()); // override the triple
118   return Result;
119 }
120 
121 std::unique_ptr<Module> BugDriver::swapProgramIn(std::unique_ptr<Module> M) {
122   std::unique_ptr<Module> OldProgram = std::move(Program);
123   Program = std::move(M);
124   return OldProgram;
125 }
126 
127 // This method takes the specified list of LLVM input files, attempts to load
128 // them, either as assembly or bitcode, then link them together. It returns
129 // true on failure (if, for example, an input bitcode file could not be
130 // parsed), and false on success.
131 //
132 bool BugDriver::addSources(const std::vector<std::string> &Filenames) {
133   assert(!Program && "Cannot call addSources multiple times!");
134   assert(!Filenames.empty() && "Must specify at least on input filename!");
135 
136   // Load the first input file.
137   Program = parseInputFile(Filenames[0], Context);
138   if (!Program)
139     return true;
140 
141   outs() << "Read input file      : '" << Filenames[0] << "'\n";
142 
143   for (unsigned i = 1, e = Filenames.size(); i != e; ++i) {
144     std::unique_ptr<Module> M = parseInputFile(Filenames[i], Context);
145     if (!M.get())
146       return true;
147 
148     outs() << "Linking in input file: '" << Filenames[i] << "'\n";
149     if (Linker::linkModules(*Program, std::move(M)))
150       return true;
151   }
152 
153   outs() << "*** All input ok\n";
154 
155   // All input files read successfully!
156   return false;
157 }
158 
159 /// run - The top level method that is invoked after all of the instance
160 /// variables are set up from command line arguments.
161 ///
162 Error BugDriver::run() {
163   if (run_find_bugs) {
164     // Rearrange the passes and apply them to the program. Repeat this process
165     // until the user kills the program or we find a bug.
166     return runManyPasses(PassesToRun);
167   }
168 
169   // If we're not running as a child, the first thing that we must do is
170   // determine what the problem is. Does the optimization series crash the
171   // compiler, or does it produce illegal code?  We make the top-level
172   // decision by trying to run all of the passes on the input program,
173   // which should generate a bitcode file.  If it does generate a bitcode
174   // file, then we know the compiler didn't crash, so try to diagnose a
175   // miscompilation.
176   if (!PassesToRun.empty()) {
177     outs() << "Running selected passes on program to test for crash: ";
178     if (runPasses(*Program, PassesToRun))
179       return debugOptimizerCrash();
180   }
181 
182   // Set up the execution environment, selecting a method to run LLVM bitcode.
183   if (Error E = initializeExecutionEnvironment())
184     return E;
185 
186   // Test to see if we have a code generator crash.
187   outs() << "Running the code generator to test for a crash: ";
188   if (Error E = compileProgram(*Program)) {
189     outs() << toString(std::move(E));
190     return debugCodeGeneratorCrash();
191   }
192   outs() << '\n';
193 
194   // Run the raw input to see where we are coming from.  If a reference output
195   // was specified, make sure that the raw output matches it.  If not, it's a
196   // problem in the front-end or the code generator.
197   //
198   bool CreatedOutput = false;
199   if (ReferenceOutputFile.empty()) {
200     outs() << "Generating reference output from raw program: ";
201     if (Error E = createReferenceFile(*Program)) {
202       errs() << toString(std::move(E));
203       return debugCodeGeneratorCrash();
204     }
205     CreatedOutput = true;
206   }
207 
208   // Make sure the reference output file gets deleted on exit from this
209   // function, if appropriate.
210   std::string ROF(ReferenceOutputFile);
211   FileRemover RemoverInstance(ROF, CreatedOutput && !SaveTemps);
212 
213   // Diff the output of the raw program against the reference output.  If it
214   // matches, then we assume there is a miscompilation bug and try to
215   // diagnose it.
216   outs() << "*** Checking the code generator...\n";
217   Expected<bool> Diff = diffProgram(*Program, "", "", false);
218   if (Error E = Diff.takeError()) {
219     errs() << toString(std::move(E));
220     return debugCodeGeneratorCrash();
221   }
222   if (!*Diff) {
223     outs() << "\n*** Output matches: Debugging miscompilation!\n";
224     if (Error E = debugMiscompilation()) {
225       errs() << toString(std::move(E));
226       return debugCodeGeneratorCrash();
227     }
228     return Error::success();
229   }
230 
231   outs() << "\n*** Input program does not match reference diff!\n";
232   outs() << "Debugging code generator problem!\n";
233   if (Error E = debugCodeGenerator()) {
234     errs() << toString(std::move(E));
235     return debugCodeGeneratorCrash();
236   }
237   return Error::success();
238 }
239 
240 void llvm::PrintFunctionList(const std::vector<Function *> &Funcs) {
241   unsigned NumPrint = Funcs.size();
242   if (NumPrint > 10)
243     NumPrint = 10;
244   for (unsigned i = 0; i != NumPrint; ++i)
245     outs() << " " << Funcs[i]->getName();
246   if (NumPrint < Funcs.size())
247     outs() << "... <" << Funcs.size() << " total>";
248   outs().flush();
249 }
250 
251 void llvm::PrintGlobalVariableList(const std::vector<GlobalVariable *> &GVs) {
252   unsigned NumPrint = GVs.size();
253   if (NumPrint > 10)
254     NumPrint = 10;
255   for (unsigned i = 0; i != NumPrint; ++i)
256     outs() << " " << GVs[i]->getName();
257   if (NumPrint < GVs.size())
258     outs() << "... <" << GVs.size() << " total>";
259   outs().flush();
260 }
261