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"
42 //===----------------------------------------------------------------------===//
43 // ELFCodeEmitter Implementation
44 //===----------------------------------------------------------------------===//
47 /// ELFCodeEmitter - This class is used by the ELFWriter to emit the code for
48 /// functions to the ELF file.
49 class ELFCodeEmitter : public MachineCodeEmitter {
51 ELFWriter::ELFSection *ES; // Section to write to.
52 std::vector<unsigned char> *OutBuffer;
55 ELFCodeEmitter(ELFWriter &ew) : EW(ew), OutBuffer(0) {}
57 void startFunction(MachineFunction &F);
58 void finishFunction(MachineFunction &F);
60 void emitConstantPool(MachineConstantPool *MCP) {
61 if (MCP->isEmpty()) return;
64 virtual void emitByte(unsigned char B) {
65 OutBuffer->push_back(B);
67 virtual void emitWordAt(unsigned W, unsigned *Ptr) {
70 virtual void emitWord(unsigned W) {
73 virtual uint64_t getCurrentPCValue() {
74 return OutBuffer->size();
76 virtual uint64_t getCurrentPCOffset() {
77 return OutBuffer->size()-FnStart;
79 void addRelocation(const MachineRelocation &MR) {
80 assert(0 && "relo not handled yet!");
82 virtual uint64_t getConstantPoolEntryAddress(unsigned Index) {
83 assert(0 && "CP not implementated yet!");
87 virtual unsigned char* allocateGlobal(unsigned size, unsigned alignment) {
88 assert(0 && "Globals not implemented yet!");
92 /// JIT SPECIFIC FUNCTIONS - DO NOT IMPLEMENT THESE HERE!
93 void startFunctionStub(unsigned StubSize) {
94 assert(0 && "JIT specific function called!");
97 void *finishFunctionStub(const Function *F) {
98 assert(0 && "JIT specific function called!");
105 /// startFunction - This callback is invoked when a new machine function is
106 /// about to be emitted.
107 void ELFCodeEmitter::startFunction(MachineFunction &F) {
108 // Align the output buffer to the appropriate alignment.
109 unsigned Align = 16; // FIXME: GENERICIZE!!
110 // Get the ELF Section that this function belongs in.
111 ES = &EW.getSection(".text", ELFWriter::ELFSection::SHT_PROGBITS,
112 ELFWriter::ELFSection::SHF_EXECINSTR |
113 ELFWriter::ELFSection::SHF_ALLOC);
114 OutBuffer = &ES->SectionData;
116 // Upgrade the section alignment if required.
117 if (ES->Align < Align) ES->Align = Align;
119 // Add padding zeros to the end of the buffer to make sure that the
120 // function will start on the correct byte alignment within the section.
121 size_t SectionOff = OutBuffer->size();
122 ELFWriter::align(*OutBuffer, Align);
124 FnStart = OutBuffer->size();
127 /// finishFunction - This callback is invoked after the function is completely
129 void ELFCodeEmitter::finishFunction(MachineFunction &F) {
130 // We now know the size of the function, add a symbol to represent it.
131 ELFWriter::ELFSym FnSym(F.getFunction());
133 // Figure out the binding (linkage) of the symbol.
134 switch (F.getFunction()->getLinkage()) {
136 // appending linkage is illegal for functions.
137 assert(0 && "Unknown linkage type!");
138 case GlobalValue::ExternalLinkage:
139 FnSym.SetBind(ELFWriter::ELFSym::STB_GLOBAL);
141 case GlobalValue::LinkOnceLinkage:
142 case GlobalValue::WeakLinkage:
143 FnSym.SetBind(ELFWriter::ELFSym::STB_WEAK);
145 case GlobalValue::InternalLinkage:
146 FnSym.SetBind(ELFWriter::ELFSym::STB_LOCAL);
150 ES->Size = OutBuffer->size();
152 FnSym.SetType(ELFWriter::ELFSym::STT_FUNC);
153 FnSym.SectionIdx = ES->SectionIdx;
154 FnSym.Value = FnStart; // Value = Offset from start of Section.
155 FnSym.Size = OutBuffer->size()-FnStart;
157 // Finally, add it to the symtab.
158 EW.SymbolTable.push_back(FnSym);
161 //===----------------------------------------------------------------------===//
162 // ELFWriter Implementation
163 //===----------------------------------------------------------------------===//
165 ELFWriter::ELFWriter(std::ostream &o, TargetMachine &tm) : O(o), TM(tm) {
166 e_machine = 0; // e_machine defaults to 'No Machine'
167 e_flags = 0; // e_flags defaults to 0, no flags.
169 is64Bit = TM.getTargetData().getPointerSizeInBits() == 64;
170 isLittleEndian = TM.getTargetData().isLittleEndian();
172 // Create the machine code emitter object for this target.
173 MCE = new ELFCodeEmitter(*this);
177 ELFWriter::~ELFWriter() {
181 // doInitialization - Emit the file header and all of the global variables for
182 // the module to the ELF file.
183 bool ELFWriter::doInitialization(Module &M) {
184 Mang = new Mangler(M);
186 // Local alias to shortenify coming code.
187 std::vector<unsigned char> &FH = FileHeader;
189 outbyte(FH, 0x7F); // EI_MAG0
190 outbyte(FH, 'E'); // EI_MAG1
191 outbyte(FH, 'L'); // EI_MAG2
192 outbyte(FH, 'F'); // EI_MAG3
193 outbyte(FH, is64Bit ? 2 : 1); // EI_CLASS
194 outbyte(FH, isLittleEndian ? 1 : 2); // EI_DATA
195 outbyte(FH, 1); // EI_VERSION
196 FH.resize(16); // EI_PAD up to 16 bytes.
198 // This should change for shared objects.
199 outhalf(FH, 1); // e_type = ET_REL
200 outhalf(FH, e_machine); // e_machine = whatever the target wants
201 outword(FH, 1); // e_version = 1
202 outaddr(FH, 0); // e_entry = 0 -> no entry point in .o file
203 outaddr(FH, 0); // e_phoff = 0 -> no program header for .o
205 ELFHeader_e_shoff_Offset = FH.size();
206 outaddr(FH, 0); // e_shoff
207 outword(FH, e_flags); // e_flags = whatever the target wants
209 outhalf(FH, is64Bit ? 64 : 52); // e_ehsize = ELF header size
210 outhalf(FH, 0); // e_phentsize = prog header entry size
211 outhalf(FH, 0); // e_phnum = # prog header entries = 0
212 outhalf(FH, is64Bit ? 64 : 40); // e_shentsize = sect hdr entry size
215 ELFHeader_e_shnum_Offset = FH.size();
216 outhalf(FH, 0); // e_shnum = # of section header ents
217 ELFHeader_e_shstrndx_Offset = FH.size();
218 outhalf(FH, 0); // e_shstrndx = Section # of '.shstrtab'
220 // Add the null section, which is required to be first in the file.
221 getSection("", 0, 0);
223 // Start up the symbol table. The first entry in the symtab is the null
225 SymbolTable.push_back(ELFSym(0));
230 void ELFWriter::EmitGlobal(GlobalVariable *GV) {
231 // If this is an external global, emit it now. TODO: Note that it would be
232 // better to ignore the symbol here and only add it to the symbol table if
234 if (!GV->hasInitializer()) {
235 ELFSym ExternalSym(GV);
236 ExternalSym.SetBind(ELFSym::STB_GLOBAL);
237 ExternalSym.SetType(ELFSym::STT_NOTYPE);
238 ExternalSym.SectionIdx = ELFSection::SHN_UNDEF;
239 SymbolTable.push_back(ExternalSym);
243 const Type *GVType = (const Type*)GV->getType();
244 unsigned Align = TM.getTargetData().getTypeAlignment(GVType);
245 unsigned Size = TM.getTargetData().getTypeSize(GVType);
247 // If this global has a zero initializer, it is part of the .bss or common
249 if (GV->getInitializer()->isNullValue()) {
250 // If this global is part of the common block, add it now. Variables are
251 // part of the common block if they are zero initialized and allowed to be
252 // merged with other symbols.
253 if (GV->hasLinkOnceLinkage() || GV->hasWeakLinkage()) {
254 ELFSym CommonSym(GV);
255 // Value for common symbols is the alignment required.
256 CommonSym.Value = Align;
257 CommonSym.Size = Size;
258 CommonSym.SetBind(ELFSym::STB_GLOBAL);
259 CommonSym.SetType(ELFSym::STT_OBJECT);
260 // TODO SOMEDAY: add ELF visibility.
261 CommonSym.SectionIdx = ELFSection::SHN_COMMON;
262 SymbolTable.push_back(CommonSym);
266 // Otherwise, this symbol is part of the .bss section. Emit it now.
268 // Handle alignment. Ensure section is aligned at least as much as required
270 ELFSection &BSSSection = getBSSSection();
271 BSSSection.Align = std::max(BSSSection.Align, Align);
273 // Within the section, emit enough virtual padding to get us to an alignment
276 BSSSection.Size = (BSSSection.Size + Align - 1) & ~(Align-1);
279 BSSSym.Value = BSSSection.Size;
281 BSSSym.SetType(ELFSym::STT_OBJECT);
283 switch (GV->getLinkage()) {
284 default: // weak/linkonce handled above
285 assert(0 && "Unexpected linkage type!");
286 case GlobalValue::AppendingLinkage: // FIXME: This should be improved!
287 case GlobalValue::ExternalLinkage:
288 BSSSym.SetBind(ELFSym::STB_GLOBAL);
290 case GlobalValue::InternalLinkage:
291 BSSSym.SetBind(ELFSym::STB_LOCAL);
295 // Set the idx of the .bss section
296 BSSSym.SectionIdx = BSSSection.SectionIdx;
297 SymbolTable.push_back(BSSSym);
299 // Reserve space in the .bss section for this symbol.
300 BSSSection.Size += Size;
304 // FIXME: handle .rodata
305 //assert(!GV->isConstant() && "unimp");
307 // FIXME: handle .data
308 //assert(0 && "unimp");
312 bool ELFWriter::runOnMachineFunction(MachineFunction &MF) {
313 // Nothing to do here, this is all done through the MCE object above.
317 /// doFinalization - Now that the module has been completely processed, emit
318 /// the ELF file to 'O'.
319 bool ELFWriter::doFinalization(Module &M) {
320 // Okay, the ELF header and .text sections have been completed, build the
321 // .data, .bss, and "common" sections next.
322 for (Module::global_iterator I = M.global_begin(), E = M.global_end();
326 // Emit the symbol table now, if non-empty.
329 // FIXME: Emit the relocations now.
331 // Emit the string table for the sections in the ELF file we have.
332 EmitSectionTableStringTable();
334 // Emit the sections to the .o file, and emit the section table for the file.
335 OutputSectionsAndSectionTable();
337 // We are done with the abstract symbols.
341 // Release the name mangler object.
342 delete Mang; Mang = 0;
346 /// EmitSymbolTable - If the current symbol table is non-empty, emit the string
347 /// table for it and then the symbol table itself.
348 void ELFWriter::EmitSymbolTable() {
349 if (SymbolTable.size() == 1) return; // Only the null entry.
351 // FIXME: compact all local symbols to the start of the symtab.
352 unsigned FirstNonLocalSymbol = 1;
354 ELFSection &StrTab = getSection(".strtab", ELFSection::SHT_STRTAB, 0);
357 DataBuffer &StrTabBuf = StrTab.SectionData;
359 // Set the zero'th symbol to a null byte, as required.
360 outbyte(StrTabBuf, 0);
361 SymbolTable[0].NameIdx = 0;
363 for (unsigned i = 1, e = SymbolTable.size(); i != e; ++i) {
364 // Use the name mangler to uniquify the LLVM symbol.
365 std::string Name = Mang->getValueName(SymbolTable[i].GV);
368 SymbolTable[i].NameIdx = 0;
370 SymbolTable[i].NameIdx = Index;
372 // Add the name to the output buffer, including the null terminator.
373 StrTabBuf.insert(StrTabBuf.end(), Name.begin(), Name.end());
375 // Add a null terminator.
376 StrTabBuf.push_back(0);
378 // Keep track of the number of bytes emitted to this section.
379 Index += Name.size()+1;
382 assert(Index == StrTabBuf.size());
385 // Now that we have emitted the string table and know the offset into the
386 // string table of each symbol, emit the symbol table itself.
387 ELFSection &SymTab = getSection(".symtab", ELFSection::SHT_SYMTAB, 0);
388 SymTab.Align = is64Bit ? 8 : 4;
389 SymTab.Link = SymTab.SectionIdx; // Section Index of .strtab.
390 SymTab.Info = FirstNonLocalSymbol; // First non-STB_LOCAL symbol.
391 SymTab.EntSize = 16; // Size of each symtab entry. FIXME: wrong for ELF64
392 DataBuffer &SymTabBuf = SymTab.SectionData;
394 if (!is64Bit) { // 32-bit and 64-bit formats are shuffled a bit.
395 for (unsigned i = 0, e = SymbolTable.size(); i != e; ++i) {
396 ELFSym &Sym = SymbolTable[i];
397 outword(SymTabBuf, Sym.NameIdx);
398 outaddr32(SymTabBuf, Sym.Value);
399 outword(SymTabBuf, Sym.Size);
400 outbyte(SymTabBuf, Sym.Info);
401 outbyte(SymTabBuf, Sym.Other);
402 outhalf(SymTabBuf, Sym.SectionIdx);
405 for (unsigned i = 0, e = SymbolTable.size(); i != e; ++i) {
406 ELFSym &Sym = SymbolTable[i];
407 outword(SymTabBuf, Sym.NameIdx);
408 outbyte(SymTabBuf, Sym.Info);
409 outbyte(SymTabBuf, Sym.Other);
410 outhalf(SymTabBuf, Sym.SectionIdx);
411 outaddr64(SymTabBuf, Sym.Value);
412 outxword(SymTabBuf, Sym.Size);
416 SymTab.Size = SymTabBuf.size();
419 /// EmitSectionTableStringTable - This method adds and emits a section for the
420 /// ELF Section Table string table: the string table that holds all of the
422 void ELFWriter::EmitSectionTableStringTable() {
423 // First step: add the section for the string table to the list of sections:
424 ELFSection &SHStrTab = getSection(".shstrtab", ELFSection::SHT_STRTAB, 0);
426 // Now that we know which section number is the .shstrtab section, update the
427 // e_shstrndx entry in the ELF header.
428 fixhalf(FileHeader, SHStrTab.SectionIdx, ELFHeader_e_shstrndx_Offset);
430 // Set the NameIdx of each section in the string table and emit the bytes for
433 DataBuffer &Buf = SHStrTab.SectionData;
435 for (std::list<ELFSection>::iterator I = SectionList.begin(),
436 E = SectionList.end(); I != E; ++I) {
437 // Set the index into the table. Note if we have lots of entries with
438 // common suffixes, we could memoize them here if we cared.
441 // Add the name to the output buffer, including the null terminator.
442 Buf.insert(Buf.end(), I->Name.begin(), I->Name.end());
444 // Add a null terminator.
447 // Keep track of the number of bytes emitted to this section.
448 Index += I->Name.size()+1;
451 // Set the size of .shstrtab now that we know what it is.
452 assert(Index == Buf.size());
453 SHStrTab.Size = Index;
456 /// OutputSectionsAndSectionTable - Now that we have constructed the file header
457 /// and all of the sections, emit these to the ostream destination and emit the
459 void ELFWriter::OutputSectionsAndSectionTable() {
460 // Pass #1: Compute the file offset for each section.
461 size_t FileOff = FileHeader.size(); // File header first.
463 // Emit all of the section data in order.
464 for (std::list<ELFSection>::iterator I = SectionList.begin(),
465 E = SectionList.end(); I != E; ++I) {
466 // Align FileOff to whatever the alignment restrictions of the section are.
468 FileOff = (FileOff+I->Align-1) & ~(I->Align-1);
470 FileOff += I->SectionData.size();
473 // Align Section Header.
474 unsigned TableAlign = is64Bit ? 8 : 4;
475 FileOff = (FileOff+TableAlign-1) & ~(TableAlign-1);
477 // Now that we know where all of the sections will be emitted, set the e_shnum
478 // entry in the ELF header.
479 fixhalf(FileHeader, NumSections, ELFHeader_e_shnum_Offset);
481 // Now that we know the offset in the file of the section table, update the
482 // e_shoff address in the ELF header.
483 fixaddr(FileHeader, FileOff, ELFHeader_e_shoff_Offset);
485 // Now that we know all of the data in the file header, emit it and all of the
487 O.write((char*)&FileHeader[0], FileHeader.size());
488 FileOff = FileHeader.size();
489 DataBuffer().swap(FileHeader);
493 // Emit all of the section data and build the section table itself.
494 while (!SectionList.empty()) {
495 const ELFSection &S = *SectionList.begin();
497 // Align FileOff to whatever the alignment restrictions of the section are.
499 for (size_t NewFileOff = (FileOff+S.Align-1) & ~(S.Align-1);
500 FileOff != NewFileOff; ++FileOff)
502 O.write((char*)&S.SectionData[0], S.SectionData.size());
503 FileOff += S.SectionData.size();
505 outword(Table, S.NameIdx); // sh_name - Symbol table name idx
506 outword(Table, S.Type); // sh_type - Section contents & semantics
507 outword(Table, S.Flags); // sh_flags - Section flags.
508 outaddr(Table, S.Addr); // sh_addr - The mem addr this section is in.
509 outaddr(Table, S.Offset); // sh_offset - Offset from the file start.
510 outword(Table, S.Size); // sh_size - The section size.
511 outword(Table, S.Link); // sh_link - Section header table index link.
512 outword(Table, S.Info); // sh_info - Auxillary information.
513 outword(Table, S.Align); // sh_addralign - Alignment of section.
514 outword(Table, S.EntSize); // sh_entsize - Size of entries in the section.
516 SectionList.pop_front();
519 // Align output for the section table.
520 for (size_t NewFileOff = (FileOff+TableAlign-1) & ~(TableAlign-1);
521 FileOff != NewFileOff; ++FileOff)
524 // Emit the section table itself.
525 O.write((char*)&Table[0], Table.size());