blob: 572a5a57c6767d9723edee678e64025a11835cea [file] [log] [blame]
Nick Lewycky833a0032011-09-01 21:09:04 +00001//===-- lib/MC/Disassembler.cpp - Disassembler Public C Interface ---------===//
Kevin Enderby93f79362011-03-28 18:25:07 +00002//
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//===----------------------------------------------------------------------===//
Chris Lattner97ff42d2011-05-22 04:52:24 +00009
Kevin Enderby93f79362011-03-28 18:25:07 +000010#include "Disassembler.h"
Kevin Enderby93f79362011-03-28 18:25:07 +000011#include "llvm-c/Disassembler.h"
12
Kevin Enderby93f79362011-03-28 18:25:07 +000013#include "llvm/MC/MCAsmInfo.h"
Evan Cheng4c816482011-07-20 06:54:19 +000014#include "llvm/MC/MCContext.h"
Kevin Enderby93f79362011-03-28 18:25:07 +000015#include "llvm/MC/MCDisassembler.h"
16#include "llvm/MC/MCInst.h"
17#include "llvm/MC/MCInstPrinter.h"
Evan Cheng4c816482011-07-20 06:54:19 +000018#include "llvm/MC/MCRegisterInfo.h"
Kevin Enderby93f79362011-03-28 18:25:07 +000019#include "llvm/Support/MemoryObject.h"
Evan Cheng3e74d6f2011-08-24 18:08:43 +000020#include "llvm/Support/TargetRegistry.h"
Kevin Enderbyfc0d7402012-02-17 19:18:29 +000021#include "llvm/Support/TargetSelect.h"
David Blaikie4d6ccb52012-01-20 21:51:11 +000022#include "llvm/Support/ErrorHandling.h"
Kevin Enderby93f79362011-03-28 18:25:07 +000023
24namespace llvm {
25class Target;
26} // namespace llvm
27using namespace llvm;
28
Kevin Enderby93f79362011-03-28 18:25:07 +000029// LLVMCreateDisasm() creates a disassembler for the TripleName. Symbolic
30// disassembly is supported by passing a block of information in the DisInfo
Chris Lattner97ff42d2011-05-22 04:52:24 +000031// parameter and specifying the TagType and callback functions as described in
Kevin Enderby93f79362011-03-28 18:25:07 +000032// the header llvm-c/Disassembler.h . The pointer to the block and the
Chris Lattner97ff42d2011-05-22 04:52:24 +000033// functions can all be passed as NULL. If successful, this returns a
34// disassembler context. If not, it returns NULL.
Kevin Enderby93f79362011-03-28 18:25:07 +000035//
36LLVMDisasmContextRef LLVMCreateDisasm(const char *TripleName, void *DisInfo,
37 int TagType, LLVMOpInfoCallback GetOpInfo,
38 LLVMSymbolLookupCallback SymbolLookUp) {
Kevin Enderbyfc0d7402012-02-17 19:18:29 +000039 // Initialize targets and assembly printers/parsers.
40 // FIXME: Clients are responsible for initializing the targets. And this
41 // would be done by calling routines in "llvm-c/Target.h" which are static
42 // line functions. But the current use of LLVMCreateDisasm() is to dynamically
Kevin Enderbyd9165eb2012-02-17 19:26:00 +000043 // load libLTO with dlopen() and then lookup the symbols using dlsym().
Kevin Enderbyfc0d7402012-02-17 19:18:29 +000044 // And since these initialize routines are static that does not work which
45 // is why the call to them in this 'C' library API was added back.
46 llvm::InitializeAllTargetInfos();
47 llvm::InitializeAllTargetMCs();
48 llvm::InitializeAllAsmParsers();
49 llvm::InitializeAllDisassemblers();
50
Kevin Enderby93f79362011-03-28 18:25:07 +000051 // Get the target.
52 std::string Error;
53 const Target *TheTarget = TargetRegistry::lookupTarget(TripleName, Error);
54 assert(TheTarget && "Unable to create target!");
55
56 // Get the assembler info needed to setup the MCContext.
Evan Cheng1abf2cb2011-07-14 23:50:31 +000057 const MCAsmInfo *MAI = TheTarget->createMCAsmInfo(TripleName);
Kevin Enderby93f79362011-03-28 18:25:07 +000058 assert(MAI && "Unable to create target asm info!");
59
Evan Cheng0e6a0522011-07-18 20:57:22 +000060 const MCRegisterInfo *MRI = TheTarget->createMCRegInfo(TripleName);
61 assert(MRI && "Unable to create target register info!");
62
Kevin Enderby93f79362011-03-28 18:25:07 +000063 // Package up features to be passed to target/subtarget
64 std::string FeaturesStr;
Evan Cheng276365d2011-06-30 01:53:36 +000065 std::string CPU;
Kevin Enderby93f79362011-03-28 18:25:07 +000066
James Molloyb9505852011-09-07 17:24:38 +000067 const MCSubtargetInfo *STI = TheTarget->createMCSubtargetInfo(TripleName, CPU,
68 FeaturesStr);
69 assert(STI && "Unable to create subtarget info!");
70
Kevin Enderby93f79362011-03-28 18:25:07 +000071 // Set up the MCContext for creating symbols and MCExpr's.
Evan Cheng203576a2011-07-20 19:50:42 +000072 MCContext *Ctx = new MCContext(*MAI, *MRI, 0);
Kevin Enderby93f79362011-03-28 18:25:07 +000073 assert(Ctx && "Unable to create MCContext!");
74
75 // Set up disassembler.
James Molloyb9505852011-09-07 17:24:38 +000076 MCDisassembler *DisAsm = TheTarget->createMCDisassembler(*STI);
Kevin Enderby93f79362011-03-28 18:25:07 +000077 assert(DisAsm && "Unable to create disassembler!");
Kevin Enderby9e5887b2011-10-04 22:44:48 +000078 DisAsm->setupForSymbolicDisassembly(GetOpInfo, SymbolLookUp, DisInfo, Ctx);
Kevin Enderby93f79362011-03-28 18:25:07 +000079
80 // Set up the instruction printer.
81 int AsmPrinterVariant = MAI->getAssemblerDialect();
Evan Chengb2627992011-07-06 19:45:42 +000082 MCInstPrinter *IP = TheTarget->createMCInstPrinter(AsmPrinterVariant,
Jim Grosbachc6449b62012-03-05 19:33:20 +000083 *MAI, *MRI, *STI);
Kevin Enderby93f79362011-03-28 18:25:07 +000084 assert(IP && "Unable to create instruction printer!");
85
86 LLVMDisasmContext *DC = new LLVMDisasmContext(TripleName, DisInfo, TagType,
87 GetOpInfo, SymbolLookUp,
Evan Cheng203576a2011-07-20 19:50:42 +000088 TheTarget, MAI, MRI,
Evan Cheng0e6a0522011-07-18 20:57:22 +000089 Ctx, DisAsm, IP);
Kevin Enderby93f79362011-03-28 18:25:07 +000090 assert(DC && "Allocation failure!");
Owen Anderson8f29e692011-09-15 18:37:20 +000091
Kevin Enderby93f79362011-03-28 18:25:07 +000092 return DC;
93}
94
95//
96// LLVMDisasmDispose() disposes of the disassembler specified by the context.
97//
98void LLVMDisasmDispose(LLVMDisasmContextRef DCR){
99 LLVMDisasmContext *DC = (LLVMDisasmContext *)DCR;
100 delete DC;
101}
102
103namespace {
104//
105// The memory object created by LLVMDisasmInstruction().
106//
107class DisasmMemoryObject : public MemoryObject {
Kevin Enderby93f79362011-03-28 18:25:07 +0000108 uint8_t *Bytes;
109 uint64_t Size;
110 uint64_t BasePC;
111public:
112 DisasmMemoryObject(uint8_t *bytes, uint64_t size, uint64_t basePC) :
113 Bytes(bytes), Size(size), BasePC(basePC) {}
114
115 uint64_t getBase() const { return BasePC; }
Derek Schuffadef06a2012-02-29 01:09:06 +0000116 uint64_t getExtent() const { return Size; }
Kevin Enderby93f79362011-03-28 18:25:07 +0000117
Derek Schuffadef06a2012-02-29 01:09:06 +0000118 int readByte(uint64_t Addr, uint8_t *Byte) const {
Kevin Enderby93f79362011-03-28 18:25:07 +0000119 if (Addr - BasePC >= Size)
120 return -1;
121 *Byte = Bytes[Addr - BasePC];
122 return 0;
123 }
124};
Chris Lattner97ff42d2011-05-22 04:52:24 +0000125} // end anonymous namespace
Kevin Enderby93f79362011-03-28 18:25:07 +0000126
127//
Benjamin Kramer5731ff62011-04-09 14:06:12 +0000128// LLVMDisasmInstruction() disassembles a single instruction using the
Kevin Enderby93f79362011-03-28 18:25:07 +0000129// disassembler context specified in the parameter DC. The bytes of the
Benjamin Kramer5731ff62011-04-09 14:06:12 +0000130// instruction are specified in the parameter Bytes, and contains at least
Kevin Enderby93f79362011-03-28 18:25:07 +0000131// BytesSize number of bytes. The instruction is at the address specified by
132// the PC parameter. If a valid instruction can be disassembled its string is
133// returned indirectly in OutString which whos size is specified in the
134// parameter OutStringSize. This function returns the number of bytes in the
135// instruction or zero if there was no valid instruction. If this function
136// returns zero the caller will have to pick how many bytes they want to step
137// over by printing a .byte, .long etc. to continue.
138//
139size_t LLVMDisasmInstruction(LLVMDisasmContextRef DCR, uint8_t *Bytes,
140 uint64_t BytesSize, uint64_t PC, char *OutString,
141 size_t OutStringSize){
142 LLVMDisasmContext *DC = (LLVMDisasmContext *)DCR;
143 // Wrap the pointer to the Bytes, BytesSize and PC in a MemoryObject.
144 DisasmMemoryObject MemoryObject(Bytes, BytesSize, PC);
145
146 uint64_t Size;
147 MCInst Inst;
148 const MCDisassembler *DisAsm = DC->getDisAsm();
149 MCInstPrinter *IP = DC->getIP();
James Molloyee064432011-09-01 22:01:14 +0000150 MCDisassembler::DecodeStatus S;
151 S = DisAsm->getInstruction(Inst, Size, MemoryObject, PC,
Owen Anderson98c5dda2011-09-15 23:38:46 +0000152 /*REMOVE*/ nulls(), DC->CommentStream);
James Molloyee064432011-09-01 22:01:14 +0000153 switch (S) {
154 case MCDisassembler::Fail:
155 case MCDisassembler::SoftFail:
James Molloyc047dca2011-09-01 18:02:14 +0000156 // FIXME: Do something different for soft failure modes?
Kevin Enderby93f79362011-03-28 18:25:07 +0000157 return 0;
James Molloyee064432011-09-01 22:01:14 +0000158
159 case MCDisassembler::Success: {
Owen Anderson8f29e692011-09-15 18:37:20 +0000160 DC->CommentStream.flush();
Owen Anderson8f29e692011-09-15 18:37:20 +0000161 StringRef Comments = DC->CommentsToEmit.str();
162
Owen Anderson98c5dda2011-09-15 23:38:46 +0000163 SmallVector<char, 64> InsnStr;
164 raw_svector_ostream OS(InsnStr);
165 IP->printInst(&Inst, OS, Comments);
166 OS.flush();
Owen Anderson8f29e692011-09-15 18:37:20 +0000167
Owen Anderson8f29e692011-09-15 18:37:20 +0000168 // Tell the comment stream that the vector changed underneath it.
Owen Anderson98c5dda2011-09-15 23:38:46 +0000169 DC->CommentsToEmit.clear();
Owen Anderson8f29e692011-09-15 18:37:20 +0000170 DC->CommentStream.resync();
171
James Molloyee064432011-09-01 22:01:14 +0000172 assert(OutStringSize != 0 && "Output buffer cannot be zero size");
173 size_t OutputSize = std::min(OutStringSize-1, InsnStr.size());
174 std::memcpy(OutString, InsnStr.data(), OutputSize);
175 OutString[OutputSize] = '\0'; // Terminate string.
176
177 return Size;
James Molloyc047dca2011-09-01 18:02:14 +0000178 }
James Molloyee064432011-09-01 22:01:14 +0000179 }
David Blaikie4d6ccb52012-01-20 21:51:11 +0000180 llvm_unreachable("Invalid DecodeStatus!");
Kevin Enderby93f79362011-03-28 18:25:07 +0000181}