1 //===-- ELFWriter.cpp - Target-independent ELF Writer code ----------------===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by Chris Lattner and is distributed under the
6 // University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the target-independent ELF writer. This file writes out
11 // the ELF file in the following order:
14 // #2. '.text' section
15 // #3. '.data' section
16 // #4. '.bss' section (conceptual position in file)
18 // #X. '.shstrtab' section
21 // The entries in the section table are laid out as:
22 // #0. Null entry [required]
23 // #1. ".text" entry - the program code
24 // #2. ".data" entry - global variables with initializers. [ if needed ]
25 // #3. ".bss" entry - global variables without initializers. [ if needed ]
27 // #N. ".shstrtab" entry - String table for the section names.
29 // NOTE: This code should eventually be extended to support 64-bit ELF (this
30 // won't be hard), but we haven't done so yet!
32 //===----------------------------------------------------------------------===//
34 #include "llvm/CodeGen/ELFWriter.h"
35 #include "llvm/Module.h"
36 #include "llvm/CodeGen/MachineCodeEmitter.h"
37 #include "llvm/CodeGen/MachineConstantPool.h"
38 #include "llvm/Target/TargetMachine.h"
39 #include "llvm/Support/Mangler.h"
43 //===----------------------------------------------------------------------===//
44 // ELFCodeEmitter Implementation
45 //===----------------------------------------------------------------------===//
48 /// ELFCodeEmitter - This class is used by the ELFWriter to emit the code for
49 /// functions to the ELF file.
50 class ELFCodeEmitter : public MachineCodeEmitter {
52 ELFWriter::ELFSection *ES; // Section to write to.
53 std::vector<unsigned char> *OutBuffer;
56 ELFCodeEmitter(ELFWriter &ew) : EW(ew), OutBuffer(0) {}
58 void startFunction(MachineFunction &F);
59 bool finishFunction(MachineFunction &F);
61 void addRelocation(const MachineRelocation &MR) {
62 assert(0 && "relo not handled yet!");
65 virtual void StartMachineBasicBlock(MachineBasicBlock *MBB) {
68 virtual intptr_t getConstantPoolEntryAddress(unsigned Index) const {
69 assert(0 && "CP not implementated yet!");
72 virtual intptr_t getJumpTableEntryAddress(unsigned Index) const {
73 assert(0 && "JT not implementated yet!");
77 virtual intptr_t getMachineBasicBlockAddress(MachineBasicBlock *MBB) const {
78 assert(0 && "JT not implementated yet!");
82 /// JIT SPECIFIC FUNCTIONS - DO NOT IMPLEMENT THESE HERE!
83 void startFunctionStub(unsigned StubSize) {
84 assert(0 && "JIT specific function called!");
87 void *finishFunctionStub(const Function *F) {
88 assert(0 && "JIT specific function called!");
95 /// startFunction - This callback is invoked when a new machine function is
96 /// about to be emitted.
97 void ELFCodeEmitter::startFunction(MachineFunction &F) {
98 // Align the output buffer to the appropriate alignment.
99 unsigned Align = 16; // FIXME: GENERICIZE!!
100 // Get the ELF Section that this function belongs in.
101 ES = &EW.getSection(".text", ELFWriter::ELFSection::SHT_PROGBITS,
102 ELFWriter::ELFSection::SHF_EXECINSTR |
103 ELFWriter::ELFSection::SHF_ALLOC);
104 OutBuffer = &ES->SectionData;
105 std::cerr << "FIXME: This code needs to be updated for changes in the"
106 << " CodeEmitter interfaces. In particular, this should set "
107 << "BufferBegin/BufferEnd/CurBufferPtr, not deal with OutBuffer!";
110 // Upgrade the section alignment if required.
111 if (ES->Align < Align) ES->Align = Align;
113 // Add padding zeros to the end of the buffer to make sure that the
114 // function will start on the correct byte alignment within the section.
115 size_t SectionOff = OutBuffer->size();
116 ELFWriter::align(*OutBuffer, Align);
118 FnStart = OutBuffer->size();
121 /// finishFunction - This callback is invoked after the function is completely
123 bool ELFCodeEmitter::finishFunction(MachineFunction &F) {
124 // We now know the size of the function, add a symbol to represent it.
125 ELFWriter::ELFSym FnSym(F.getFunction());
127 // Figure out the binding (linkage) of the symbol.
128 switch (F.getFunction()->getLinkage()) {
130 // appending linkage is illegal for functions.
131 assert(0 && "Unknown linkage type!");
132 case GlobalValue::ExternalLinkage:
133 FnSym.SetBind(ELFWriter::ELFSym::STB_GLOBAL);
135 case GlobalValue::LinkOnceLinkage:
136 case GlobalValue::WeakLinkage:
137 FnSym.SetBind(ELFWriter::ELFSym::STB_WEAK);
139 case GlobalValue::InternalLinkage:
140 FnSym.SetBind(ELFWriter::ELFSym::STB_LOCAL);
144 ES->Size = OutBuffer->size();
146 FnSym.SetType(ELFWriter::ELFSym::STT_FUNC);
147 FnSym.SectionIdx = ES->SectionIdx;
148 FnSym.Value = FnStart; // Value = Offset from start of Section.
149 FnSym.Size = OutBuffer->size()-FnStart;
151 // Finally, add it to the symtab.
152 EW.SymbolTable.push_back(FnSym);
156 //===----------------------------------------------------------------------===//
157 // ELFWriter Implementation
158 //===----------------------------------------------------------------------===//
160 ELFWriter::ELFWriter(std::ostream &o, TargetMachine &tm) : O(o), TM(tm) {
161 e_machine = 0; // e_machine defaults to 'No Machine'
162 e_flags = 0; // e_flags defaults to 0, no flags.
164 is64Bit = TM.getTargetData()->getPointerSizeInBits() == 64;
165 isLittleEndian = TM.getTargetData()->isLittleEndian();
167 // Create the machine code emitter object for this target.
168 MCE = new ELFCodeEmitter(*this);
172 ELFWriter::~ELFWriter() {
176 // doInitialization - Emit the file header and all of the global variables for
177 // the module to the ELF file.
178 bool ELFWriter::doInitialization(Module &M) {
179 Mang = new Mangler(M);
181 // Local alias to shortenify coming code.
182 std::vector<unsigned char> &FH = FileHeader;
184 outbyte(FH, 0x7F); // EI_MAG0
185 outbyte(FH, 'E'); // EI_MAG1
186 outbyte(FH, 'L'); // EI_MAG2
187 outbyte(FH, 'F'); // EI_MAG3
188 outbyte(FH, is64Bit ? 2 : 1); // EI_CLASS
189 outbyte(FH, isLittleEndian ? 1 : 2); // EI_DATA
190 outbyte(FH, 1); // EI_VERSION
191 FH.resize(16); // EI_PAD up to 16 bytes.
193 // This should change for shared objects.
194 outhalf(FH, 1); // e_type = ET_REL
195 outhalf(FH, e_machine); // e_machine = whatever the target wants
196 outword(FH, 1); // e_version = 1
197 outaddr(FH, 0); // e_entry = 0 -> no entry point in .o file
198 outaddr(FH, 0); // e_phoff = 0 -> no program header for .o
200 ELFHeader_e_shoff_Offset = FH.size();
201 outaddr(FH, 0); // e_shoff
202 outword(FH, e_flags); // e_flags = whatever the target wants
204 outhalf(FH, is64Bit ? 64 : 52); // e_ehsize = ELF header size
205 outhalf(FH, 0); // e_phentsize = prog header entry size
206 outhalf(FH, 0); // e_phnum = # prog header entries = 0
207 outhalf(FH, is64Bit ? 64 : 40); // e_shentsize = sect hdr entry size
210 ELFHeader_e_shnum_Offset = FH.size();
211 outhalf(FH, 0); // e_shnum = # of section header ents
212 ELFHeader_e_shstrndx_Offset = FH.size();
213 outhalf(FH, 0); // e_shstrndx = Section # of '.shstrtab'
215 // Add the null section, which is required to be first in the file.
216 getSection("", 0, 0);
218 // Start up the symbol table. The first entry in the symtab is the null
220 SymbolTable.push_back(ELFSym(0));
225 void ELFWriter::EmitGlobal(GlobalVariable *GV) {
226 // If this is an external global, emit it now. TODO: Note that it would be
227 // better to ignore the symbol here and only add it to the symbol table if
229 if (!GV->hasInitializer()) {
230 ELFSym ExternalSym(GV);
231 ExternalSym.SetBind(ELFSym::STB_GLOBAL);
232 ExternalSym.SetType(ELFSym::STT_NOTYPE);
233 ExternalSym.SectionIdx = ELFSection::SHN_UNDEF;
234 SymbolTable.push_back(ExternalSym);
238 const Type *GVType = (const Type*)GV->getType();
239 unsigned Align = TM.getTargetData()->getTypeAlignment(GVType);
240 unsigned Size = TM.getTargetData()->getTypeSize(GVType);
242 // If this global has a zero initializer, it is part of the .bss or common
244 if (GV->getInitializer()->isNullValue()) {
245 // If this global is part of the common block, add it now. Variables are
246 // part of the common block if they are zero initialized and allowed to be
247 // merged with other symbols.
248 if (GV->hasLinkOnceLinkage() || GV->hasWeakLinkage()) {
249 ELFSym CommonSym(GV);
250 // Value for common symbols is the alignment required.
251 CommonSym.Value = Align;
252 CommonSym.Size = Size;
253 CommonSym.SetBind(ELFSym::STB_GLOBAL);
254 CommonSym.SetType(ELFSym::STT_OBJECT);
255 // TODO SOMEDAY: add ELF visibility.
256 CommonSym.SectionIdx = ELFSection::SHN_COMMON;
257 SymbolTable.push_back(CommonSym);
261 // Otherwise, this symbol is part of the .bss section. Emit it now.
263 // Handle alignment. Ensure section is aligned at least as much as required
265 ELFSection &BSSSection = getBSSSection();
266 BSSSection.Align = std::max(BSSSection.Align, Align);
268 // Within the section, emit enough virtual padding to get us to an alignment
271 BSSSection.Size = (BSSSection.Size + Align - 1) & ~(Align-1);
274 BSSSym.Value = BSSSection.Size;
276 BSSSym.SetType(ELFSym::STT_OBJECT);
278 switch (GV->getLinkage()) {
279 default: // weak/linkonce handled above
280 assert(0 && "Unexpected linkage type!");
281 case GlobalValue::AppendingLinkage: // FIXME: This should be improved!
282 case GlobalValue::ExternalLinkage:
283 BSSSym.SetBind(ELFSym::STB_GLOBAL);
285 case GlobalValue::InternalLinkage:
286 BSSSym.SetBind(ELFSym::STB_LOCAL);
290 // Set the idx of the .bss section
291 BSSSym.SectionIdx = BSSSection.SectionIdx;
292 SymbolTable.push_back(BSSSym);
294 // Reserve space in the .bss section for this symbol.
295 BSSSection.Size += Size;
299 // FIXME: handle .rodata
300 //assert(!GV->isConstant() && "unimp");
302 // FIXME: handle .data
303 //assert(0 && "unimp");
307 bool ELFWriter::runOnMachineFunction(MachineFunction &MF) {
308 // Nothing to do here, this is all done through the MCE object above.
312 /// doFinalization - Now that the module has been completely processed, emit
313 /// the ELF file to 'O'.
314 bool ELFWriter::doFinalization(Module &M) {
315 // Okay, the ELF header and .text sections have been completed, build the
316 // .data, .bss, and "common" sections next.
317 for (Module::global_iterator I = M.global_begin(), E = M.global_end();
321 // Emit the symbol table now, if non-empty.
324 // FIXME: Emit the relocations now.
326 // Emit the string table for the sections in the ELF file we have.
327 EmitSectionTableStringTable();
329 // Emit the sections to the .o file, and emit the section table for the file.
330 OutputSectionsAndSectionTable();
332 // We are done with the abstract symbols.
336 // Release the name mangler object.
337 delete Mang; Mang = 0;
341 /// EmitSymbolTable - If the current symbol table is non-empty, emit the string
342 /// table for it and then the symbol table itself.
343 void ELFWriter::EmitSymbolTable() {
344 if (SymbolTable.size() == 1) return; // Only the null entry.
346 // FIXME: compact all local symbols to the start of the symtab.
347 unsigned FirstNonLocalSymbol = 1;
349 ELFSection &StrTab = getSection(".strtab", ELFSection::SHT_STRTAB, 0);
352 DataBuffer &StrTabBuf = StrTab.SectionData;
354 // Set the zero'th symbol to a null byte, as required.
355 outbyte(StrTabBuf, 0);
356 SymbolTable[0].NameIdx = 0;
358 for (unsigned i = 1, e = SymbolTable.size(); i != e; ++i) {
359 // Use the name mangler to uniquify the LLVM symbol.
360 std::string Name = Mang->getValueName(SymbolTable[i].GV);
363 SymbolTable[i].NameIdx = 0;
365 SymbolTable[i].NameIdx = Index;
367 // Add the name to the output buffer, including the null terminator.
368 StrTabBuf.insert(StrTabBuf.end(), Name.begin(), Name.end());
370 // Add a null terminator.
371 StrTabBuf.push_back(0);
373 // Keep track of the number of bytes emitted to this section.
374 Index += Name.size()+1;
377 assert(Index == StrTabBuf.size());
380 // Now that we have emitted the string table and know the offset into the
381 // string table of each symbol, emit the symbol table itself.
382 ELFSection &SymTab = getSection(".symtab", ELFSection::SHT_SYMTAB, 0);
383 SymTab.Align = is64Bit ? 8 : 4;
384 SymTab.Link = SymTab.SectionIdx; // Section Index of .strtab.
385 SymTab.Info = FirstNonLocalSymbol; // First non-STB_LOCAL symbol.
386 SymTab.EntSize = 16; // Size of each symtab entry. FIXME: wrong for ELF64
387 DataBuffer &SymTabBuf = SymTab.SectionData;
389 if (!is64Bit) { // 32-bit and 64-bit formats are shuffled a bit.
390 for (unsigned i = 0, e = SymbolTable.size(); i != e; ++i) {
391 ELFSym &Sym = SymbolTable[i];
392 outword(SymTabBuf, Sym.NameIdx);
393 outaddr32(SymTabBuf, Sym.Value);
394 outword(SymTabBuf, Sym.Size);
395 outbyte(SymTabBuf, Sym.Info);
396 outbyte(SymTabBuf, Sym.Other);
397 outhalf(SymTabBuf, Sym.SectionIdx);
400 for (unsigned i = 0, e = SymbolTable.size(); i != e; ++i) {
401 ELFSym &Sym = SymbolTable[i];
402 outword(SymTabBuf, Sym.NameIdx);
403 outbyte(SymTabBuf, Sym.Info);
404 outbyte(SymTabBuf, Sym.Other);
405 outhalf(SymTabBuf, Sym.SectionIdx);
406 outaddr64(SymTabBuf, Sym.Value);
407 outxword(SymTabBuf, Sym.Size);
411 SymTab.Size = SymTabBuf.size();
414 /// EmitSectionTableStringTable - This method adds and emits a section for the
415 /// ELF Section Table string table: the string table that holds all of the
417 void ELFWriter::EmitSectionTableStringTable() {
418 // First step: add the section for the string table to the list of sections:
419 ELFSection &SHStrTab = getSection(".shstrtab", ELFSection::SHT_STRTAB, 0);
421 // Now that we know which section number is the .shstrtab section, update the
422 // e_shstrndx entry in the ELF header.
423 fixhalf(FileHeader, SHStrTab.SectionIdx, ELFHeader_e_shstrndx_Offset);
425 // Set the NameIdx of each section in the string table and emit the bytes for
428 DataBuffer &Buf = SHStrTab.SectionData;
430 for (std::list<ELFSection>::iterator I = SectionList.begin(),
431 E = SectionList.end(); I != E; ++I) {
432 // Set the index into the table. Note if we have lots of entries with
433 // common suffixes, we could memoize them here if we cared.
436 // Add the name to the output buffer, including the null terminator.
437 Buf.insert(Buf.end(), I->Name.begin(), I->Name.end());
439 // Add a null terminator.
442 // Keep track of the number of bytes emitted to this section.
443 Index += I->Name.size()+1;
446 // Set the size of .shstrtab now that we know what it is.
447 assert(Index == Buf.size());
448 SHStrTab.Size = Index;
451 /// OutputSectionsAndSectionTable - Now that we have constructed the file header
452 /// and all of the sections, emit these to the ostream destination and emit the
454 void ELFWriter::OutputSectionsAndSectionTable() {
455 // Pass #1: Compute the file offset for each section.
456 size_t FileOff = FileHeader.size(); // File header first.
458 // Emit all of the section data in order.
459 for (std::list<ELFSection>::iterator I = SectionList.begin(),
460 E = SectionList.end(); I != E; ++I) {
461 // Align FileOff to whatever the alignment restrictions of the section are.
463 FileOff = (FileOff+I->Align-1) & ~(I->Align-1);
465 FileOff += I->SectionData.size();
468 // Align Section Header.
469 unsigned TableAlign = is64Bit ? 8 : 4;
470 FileOff = (FileOff+TableAlign-1) & ~(TableAlign-1);
472 // Now that we know where all of the sections will be emitted, set the e_shnum
473 // entry in the ELF header.
474 fixhalf(FileHeader, NumSections, ELFHeader_e_shnum_Offset);
476 // Now that we know the offset in the file of the section table, update the
477 // e_shoff address in the ELF header.
478 fixaddr(FileHeader, FileOff, ELFHeader_e_shoff_Offset);
480 // Now that we know all of the data in the file header, emit it and all of the
482 O.write((char*)&FileHeader[0], FileHeader.size());
483 FileOff = FileHeader.size();
484 DataBuffer().swap(FileHeader);
488 // Emit all of the section data and build the section table itself.
489 while (!SectionList.empty()) {
490 const ELFSection &S = *SectionList.begin();
492 // Align FileOff to whatever the alignment restrictions of the section are.
494 for (size_t NewFileOff = (FileOff+S.Align-1) & ~(S.Align-1);
495 FileOff != NewFileOff; ++FileOff)
497 O.write((char*)&S.SectionData[0], S.SectionData.size());
498 FileOff += S.SectionData.size();
500 outword(Table, S.NameIdx); // sh_name - Symbol table name idx
501 outword(Table, S.Type); // sh_type - Section contents & semantics
502 outword(Table, S.Flags); // sh_flags - Section flags.
503 outaddr(Table, S.Addr); // sh_addr - The mem addr this section is in.
504 outaddr(Table, S.Offset); // sh_offset - Offset from the file start.
505 outword(Table, S.Size); // sh_size - The section size.
506 outword(Table, S.Link); // sh_link - Section header table index link.
507 outword(Table, S.Info); // sh_info - Auxillary information.
508 outword(Table, S.Align); // sh_addralign - Alignment of section.
509 outword(Table, S.EntSize); // sh_entsize - Size of entries in the section.
511 SectionList.pop_front();
514 // Align output for the section table.
515 for (size_t NewFileOff = (FileOff+TableAlign-1) & ~(TableAlign-1);
516 FileOff != NewFileOff; ++FileOff)
519 // Emit the section table itself.
520 O.write((char*)&Table[0], Table.size());