blob: 5c67f1b70a9a20569c039e61f8511c67d20d34cb [file] [log] [blame]
Benjamin Kramer685a2502011-07-20 19:37:35 +00001//===-- MCFunction.cpp ----------------------------------------------------===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file defines the algorithm to break down a region of machine code
11// into basic blocks and try to reconstruct a CFG from it.
12//
13//===----------------------------------------------------------------------===//
14
15#include "MCFunction.h"
16#include "llvm/ADT/STLExtras.h"
17#include "llvm/MC/MCDisassembler.h"
18#include "llvm/MC/MCInst.h"
19#include "llvm/MC/MCInstPrinter.h"
Benjamin Kramer41ab14b2011-08-08 18:56:44 +000020#include "llvm/MC/MCInstrAnalysis.h"
Benjamin Kramer685a2502011-07-20 19:37:35 +000021#include "llvm/MC/MCInstrDesc.h"
22#include "llvm/MC/MCInstrInfo.h"
23#include "llvm/Support/MemoryObject.h"
24#include "llvm/Support/raw_ostream.h"
25#include "llvm/Support/system_error.h"
26#include <set>
27using namespace llvm;
28
29MCFunction
30MCFunction::createFunctionFromMC(StringRef Name, const MCDisassembler *DisAsm,
31 const MemoryObject &Region, uint64_t Start,
Benjamin Kramer41ab14b2011-08-08 18:56:44 +000032 uint64_t End, const MCInstrAnalysis *Ana,
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000033 raw_ostream &DebugOut,
34 SmallVectorImpl<uint64_t> &Calls) {
35 std::vector<MCDecodedInst> Instructions;
Benjamin Kramer685a2502011-07-20 19:37:35 +000036 std::set<uint64_t> Splits;
37 Splits.insert(Start);
Benjamin Kramer685a2502011-07-20 19:37:35 +000038 uint64_t Size;
39
Benjamin Kramer685a2502011-07-20 19:37:35 +000040 MCFunction f(Name);
41
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000042 {
43 DenseSet<uint64_t> VisitedInsts;
44 SmallVector<uint64_t, 16> WorkList;
45 WorkList.push_back(Start);
46 // Disassemble code and gather basic block split points.
47 while (!WorkList.empty()) {
48 uint64_t Index = WorkList.pop_back_val();
49 if (VisitedInsts.find(Index) != VisitedInsts.end())
Benjamin Kramera894c8e2011-09-20 17:53:01 +000050 continue; // Already visited this location.
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000051
52 for (;Index < End; Index += Size) {
Benjamin Kramera894c8e2011-09-20 17:53:01 +000053 VisitedInsts.insert(Index);
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000054
Benjamin Kramera894c8e2011-09-20 17:53:01 +000055 MCInst Inst;
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000056 if (DisAsm->getInstruction(Inst, Size, Region, Index, DebugOut, nulls())){
Benjamin Kramera894c8e2011-09-20 17:53:01 +000057 Instructions.push_back(MCDecodedInst(Index, Size, Inst));
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000058 if (Ana->isBranch(Inst)) {
59 uint64_t targ = Ana->evaluateBranch(Inst, Index, Size);
Benjamin Kramera894c8e2011-09-20 17:53:01 +000060 if (targ != -1ULL && targ == Index+Size)
61 continue; // Skip nop jumps.
62
63 // If we could determine the branch target, make a note to start a
64 // new basic block there and add the target to the worklist.
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000065 if (targ != -1ULL) {
66 Splits.insert(targ);
67 WorkList.push_back(targ);
68 WorkList.push_back(Index+Size);
69 }
70 Splits.insert(Index+Size);
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000071 break;
72 } else if (Ana->isReturn(Inst)) {
Benjamin Kramera894c8e2011-09-20 17:53:01 +000073 // Return instruction. This basic block ends here.
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000074 Splits.insert(Index+Size);
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000075 break;
76 } else if (Ana->isCall(Inst)) {
77 uint64_t targ = Ana->evaluateBranch(Inst, Index, Size);
Benjamin Kramera894c8e2011-09-20 17:53:01 +000078 // Add the call to the call list if the destination is known.
79 if (targ != -1ULL && targ != Index+Size)
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000080 Calls.push_back(targ);
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000081 }
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000082 } else {
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000083 errs().write_hex(Index) << ": warning: invalid instruction encoding\n";
84 if (Size == 0)
85 Size = 1; // skip illegible bytes
86 }
87 }
88 }
89 }
90
Benjamin Kramera894c8e2011-09-20 17:53:01 +000091 // Make sure the instruction list is sorted.
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000092 std::sort(Instructions.begin(), Instructions.end());
93
Benjamin Kramera894c8e2011-09-20 17:53:01 +000094 // Create basic blocks.
Benjamin Kramer685a2502011-07-20 19:37:35 +000095 unsigned ii = 0, ie = Instructions.size();
96 for (std::set<uint64_t>::iterator spi = Splits.begin(),
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000097 spe = llvm::prior(Splits.end()); spi != spe; ++spi) {
Benjamin Kramer685a2502011-07-20 19:37:35 +000098 MCBasicBlock BB;
Benjamin Kramer0b8b7712011-09-19 17:56:04 +000099 uint64_t BlockEnd = *llvm::next(spi);
Benjamin Kramer685a2502011-07-20 19:37:35 +0000100 // Add instructions to the BB.
101 for (; ii != ie; ++ii) {
102 if (Instructions[ii].Address < *spi ||
103 Instructions[ii].Address >= BlockEnd)
104 break;
105 BB.addInst(Instructions[ii]);
106 }
107 f.addBlock(*spi, BB);
108 }
109
Benjamin Kramer0b8b7712011-09-19 17:56:04 +0000110 std::sort(f.Blocks.begin(), f.Blocks.end());
111
Benjamin Kramer685a2502011-07-20 19:37:35 +0000112 // Calculate successors of each block.
113 for (MCFunction::iterator i = f.begin(), e = f.end(); i != e; ++i) {
Benjamin Kramera894c8e2011-09-20 17:53:01 +0000114 MCBasicBlock &BB = const_cast<MCBasicBlock&>(i->second);
Benjamin Kramer685a2502011-07-20 19:37:35 +0000115 if (BB.getInsts().empty()) continue;
116 const MCDecodedInst &Inst = BB.getInsts().back();
Benjamin Kramer685a2502011-07-20 19:37:35 +0000117
Benjamin Kramer41ab14b2011-08-08 18:56:44 +0000118 if (Ana->isBranch(Inst.Inst)) {
119 uint64_t targ = Ana->evaluateBranch(Inst.Inst, Inst.Address, Inst.Size);
120 if (targ == -1ULL) {
Benjamin Kramer685a2502011-07-20 19:37:35 +0000121 // Indirect branch. Bail and add all blocks of the function as a
122 // successor.
123 for (MCFunction::iterator i = f.begin(), e = f.end(); i != e; ++i)
Benjamin Kramer0b8b7712011-09-19 17:56:04 +0000124 BB.addSucc(i->first);
Benjamin Kramer41ab14b2011-08-08 18:56:44 +0000125 } else if (targ != Inst.Address+Inst.Size)
Benjamin Kramer0b8b7712011-09-19 17:56:04 +0000126 BB.addSucc(targ);
Benjamin Kramer41ab14b2011-08-08 18:56:44 +0000127 // Conditional branches can also fall through to the next block.
128 if (Ana->isConditionalBranch(Inst.Inst) && llvm::next(i) != e)
Benjamin Kramer0b8b7712011-09-19 17:56:04 +0000129 BB.addSucc(llvm::next(i)->first);
Benjamin Kramer685a2502011-07-20 19:37:35 +0000130 } else {
131 // No branch. Fall through to the next block.
Benjamin Kramer41ab14b2011-08-08 18:56:44 +0000132 if (!Ana->isReturn(Inst.Inst) && llvm::next(i) != e)
Benjamin Kramer0b8b7712011-09-19 17:56:04 +0000133 BB.addSucc(llvm::next(i)->first);
Benjamin Kramer685a2502011-07-20 19:37:35 +0000134 }
135 }
136
137 return f;
138}