1 //===-- ELFWriter.cpp - Target-independent ELF Writer code ----------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // 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 "ELFWriter.h"
35 #include "llvm/Module.h"
36 #include "llvm/PassManager.h"
37 #include "llvm/DerivedTypes.h"
38 #include "llvm/CodeGen/FileWriters.h"
39 #include "llvm/CodeGen/MachineCodeEmitter.h"
40 #include "llvm/CodeGen/MachineConstantPool.h"
41 #include "llvm/CodeGen/MachineFunctionPass.h"
42 #include "llvm/Target/TargetData.h"
43 #include "llvm/Target/TargetELFWriterInfo.h"
44 #include "llvm/Target/TargetMachine.h"
45 #include "llvm/Support/Mangler.h"
46 #include "llvm/Support/OutputBuffer.h"
47 #include "llvm/Support/Streams.h"
48 #include "llvm/Support/raw_ostream.h"
52 char ELFWriter::ID
= 0;
53 /// AddELFWriter - Concrete function to add the ELF writer to the function pass
55 MachineCodeEmitter
*llvm::AddELFWriter(PassManagerBase
&PM
,
58 ELFWriter
*EW
= new ELFWriter(O
, TM
);
60 return &EW
->getMachineCodeEmitter();
63 //===----------------------------------------------------------------------===//
64 // ELFCodeEmitter Implementation
65 //===----------------------------------------------------------------------===//
68 /// ELFCodeEmitter - This class is used by the ELFWriter to emit the code for
69 /// functions to the ELF file.
70 class ELFCodeEmitter
: public MachineCodeEmitter
{
73 ELFWriter::ELFSection
*ES
; // Section to write to.
74 std::vector
<unsigned char> *OutBuffer
;
77 explicit ELFCodeEmitter(ELFWriter
&ew
) : EW(ew
), TM(EW
.TM
), OutBuffer(0) {}
79 void startFunction(MachineFunction
&F
);
80 bool finishFunction(MachineFunction
&F
);
82 void addRelocation(const MachineRelocation
&MR
) {
83 assert(0 && "relo not handled yet!");
86 virtual void StartMachineBasicBlock(MachineBasicBlock
*MBB
) {
89 virtual uintptr_t getConstantPoolEntryAddress(unsigned Index
) const {
90 assert(0 && "CP not implementated yet!");
93 virtual uintptr_t getJumpTableEntryAddress(unsigned Index
) const {
94 assert(0 && "JT not implementated yet!");
98 virtual uintptr_t getMachineBasicBlockAddress(MachineBasicBlock
*MBB
) const {
99 assert(0 && "JT not implementated yet!");
103 virtual uintptr_t getLabelAddress(uint64_t Label
) const {
104 assert(0 && "Label address not implementated yet!");
109 virtual void emitLabel(uint64_t LabelID
) {
110 assert(0 && "emit Label not implementated yet!");
115 virtual void setModuleInfo(llvm::MachineModuleInfo
* MMI
) { }
118 /// JIT SPECIFIC FUNCTIONS - DO NOT IMPLEMENT THESE HERE!
119 void startGVStub(const GlobalValue
* F
, unsigned StubSize
,
120 unsigned Alignment
= 1) {
121 assert(0 && "JIT specific function called!");
124 void startGVStub(const GlobalValue
* F
, void *Buffer
, unsigned StubSize
) {
125 assert(0 && "JIT specific function called!");
128 void *finishGVStub(const GlobalValue
*F
) {
129 assert(0 && "JIT specific function called!");
136 /// startFunction - This callback is invoked when a new machine function is
137 /// about to be emitted.
138 void ELFCodeEmitter::startFunction(MachineFunction
&F
) {
139 // Align the output buffer to the appropriate alignment.
140 unsigned Align
= 16; // FIXME: GENERICIZE!!
141 // Get the ELF Section that this function belongs in.
142 ES
= &EW
.getSection(".text", ELFWriter::ELFSection::SHT_PROGBITS
,
143 ELFWriter::ELFSection::SHF_EXECINSTR
|
144 ELFWriter::ELFSection::SHF_ALLOC
);
145 OutBuffer
= &ES
->SectionData
;
146 cerr
<< "FIXME: This code needs to be updated for changes in the "
147 << "CodeEmitter interfaces. In particular, this should set "
148 << "BufferBegin/BufferEnd/CurBufferPtr, not deal with OutBuffer!";
151 // Upgrade the section alignment if required.
152 if (ES
->Align
< Align
) ES
->Align
= Align
;
154 // Add padding zeros to the end of the buffer to make sure that the
155 // function will start on the correct byte alignment within the section.
156 OutputBuffer
OB(*OutBuffer
,
157 TM
.getTargetData()->getPointerSizeInBits() == 64,
158 TM
.getTargetData()->isLittleEndian());
160 FnStart
= OutBuffer
->size();
163 /// finishFunction - This callback is invoked after the function is completely
165 bool ELFCodeEmitter::finishFunction(MachineFunction
&F
) {
166 // We now know the size of the function, add a symbol to represent it.
167 ELFWriter::ELFSym
FnSym(F
.getFunction());
169 // Figure out the binding (linkage) of the symbol.
170 switch (F
.getFunction()->getLinkage()) {
172 // appending linkage is illegal for functions.
173 assert(0 && "Unknown linkage type!");
174 case GlobalValue::ExternalLinkage
:
175 FnSym
.SetBind(ELFWriter::ELFSym::STB_GLOBAL
);
177 case GlobalValue::LinkOnceAnyLinkage
:
178 case GlobalValue::LinkOnceODRLinkage
:
179 case GlobalValue::WeakAnyLinkage
:
180 case GlobalValue::WeakODRLinkage
:
181 FnSym
.SetBind(ELFWriter::ELFSym::STB_WEAK
);
183 case GlobalValue::PrivateLinkage
:
184 assert (0 && "PrivateLinkage should not be in the symbol table.");
185 case GlobalValue::InternalLinkage
:
186 FnSym
.SetBind(ELFWriter::ELFSym::STB_LOCAL
);
190 ES
->Size
= OutBuffer
->size();
192 FnSym
.SetType(ELFWriter::ELFSym::STT_FUNC
);
193 FnSym
.SectionIdx
= ES
->SectionIdx
;
194 FnSym
.Value
= FnStart
; // Value = Offset from start of Section.
195 FnSym
.Size
= OutBuffer
->size()-FnStart
;
197 // Finally, add it to the symtab.
198 EW
.SymbolTable
.push_back(FnSym
);
202 //===----------------------------------------------------------------------===//
203 // ELFWriter Implementation
204 //===----------------------------------------------------------------------===//
206 ELFWriter::ELFWriter(raw_ostream
&o
, TargetMachine
&tm
)
207 : MachineFunctionPass(&ID
), O(o
), TM(tm
) {
208 e_flags
= 0; // e_flags defaults to 0, no flags.
210 is64Bit
= TM
.getTargetData()->getPointerSizeInBits() == 64;
211 isLittleEndian
= TM
.getTargetData()->isLittleEndian();
213 // Create the machine code emitter object for this target.
214 MCE
= new ELFCodeEmitter(*this);
218 ELFWriter::~ELFWriter() {
222 // doInitialization - Emit the file header and all of the global variables for
223 // the module to the ELF file.
224 bool ELFWriter::doInitialization(Module
&M
) {
225 Mang
= new Mangler(M
);
227 // Local alias to shortenify coming code.
228 std::vector
<unsigned char> &FH
= FileHeader
;
229 OutputBuffer
FHOut(FH
, is64Bit
, isLittleEndian
);
231 FHOut
.outbyte(0x7F); // EI_MAG0
232 FHOut
.outbyte('E'); // EI_MAG1
233 FHOut
.outbyte('L'); // EI_MAG2
234 FHOut
.outbyte('F'); // EI_MAG3
235 FHOut
.outbyte(is64Bit
? 2 : 1); // EI_CLASS
236 FHOut
.outbyte(isLittleEndian
? 1 : 2); // EI_DATA
237 FHOut
.outbyte(1); // EI_VERSION
238 FH
.resize(16); // EI_PAD up to 16 bytes.
240 // This should change for shared objects.
241 FHOut
.outhalf(1); // e_type = ET_REL
242 FHOut
.outhalf(TM
.getELFWriterInfo()->getEMachine()); // target-defined
243 FHOut
.outword(1); // e_version = 1
244 FHOut
.outaddr(0); // e_entry = 0 -> no entry point in .o file
245 FHOut
.outaddr(0); // e_phoff = 0 -> no program header for .o
247 ELFHeader_e_shoff_Offset
= FH
.size();
248 FHOut
.outaddr(0); // e_shoff
249 FHOut
.outword(e_flags
); // e_flags = whatever the target wants
251 FHOut
.outhalf(is64Bit
? 64 : 52); // e_ehsize = ELF header size
252 FHOut
.outhalf(0); // e_phentsize = prog header entry size
253 FHOut
.outhalf(0); // e_phnum = # prog header entries = 0
254 FHOut
.outhalf(is64Bit
? 64 : 40); // e_shentsize = sect hdr entry size
257 ELFHeader_e_shnum_Offset
= FH
.size();
258 FHOut
.outhalf(0); // e_shnum = # of section header ents
259 ELFHeader_e_shstrndx_Offset
= FH
.size();
260 FHOut
.outhalf(0); // e_shstrndx = Section # of '.shstrtab'
262 // Add the null section, which is required to be first in the file.
263 getSection("", 0, 0);
265 // Start up the symbol table. The first entry in the symtab is the null
267 SymbolTable
.push_back(ELFSym(0));
272 void ELFWriter::EmitGlobal(GlobalVariable
*GV
) {
273 // If this is an external global, emit it now. TODO: Note that it would be
274 // better to ignore the symbol here and only add it to the symbol table if
276 if (!GV
->hasInitializer()) {
277 ELFSym
ExternalSym(GV
);
278 ExternalSym
.SetBind(ELFSym::STB_GLOBAL
);
279 ExternalSym
.SetType(ELFSym::STT_NOTYPE
);
280 ExternalSym
.SectionIdx
= ELFSection::SHN_UNDEF
;
281 SymbolTable
.push_back(ExternalSym
);
285 unsigned Align
= TM
.getTargetData()->getPreferredAlignment(GV
);
287 TM
.getTargetData()->getTypePaddedSize(GV
->getType()->getElementType());
289 // If this global has a zero initializer, it is part of the .bss or common
291 if (GV
->getInitializer()->isNullValue()) {
292 // If this global is part of the common block, add it now. Variables are
293 // part of the common block if they are zero initialized and allowed to be
294 // merged with other symbols.
295 if (GV
->hasLinkOnceLinkage() || GV
->hasWeakLinkage() ||
296 GV
->hasCommonLinkage()) {
297 ELFSym
CommonSym(GV
);
298 // Value for common symbols is the alignment required.
299 CommonSym
.Value
= Align
;
300 CommonSym
.Size
= Size
;
301 CommonSym
.SetBind(ELFSym::STB_GLOBAL
);
302 CommonSym
.SetType(ELFSym::STT_OBJECT
);
303 // TODO SOMEDAY: add ELF visibility.
304 CommonSym
.SectionIdx
= ELFSection::SHN_COMMON
;
305 SymbolTable
.push_back(CommonSym
);
309 // Otherwise, this symbol is part of the .bss section. Emit it now.
311 // Handle alignment. Ensure section is aligned at least as much as required
313 ELFSection
&BSSSection
= getBSSSection();
314 BSSSection
.Align
= std::max(BSSSection
.Align
, Align
);
316 // Within the section, emit enough virtual padding to get us to an alignment
319 BSSSection
.Size
= (BSSSection
.Size
+ Align
- 1) & ~(Align
-1);
322 BSSSym
.Value
= BSSSection
.Size
;
324 BSSSym
.SetType(ELFSym::STT_OBJECT
);
326 switch (GV
->getLinkage()) {
327 default: // weak/linkonce/common handled above
328 assert(0 && "Unexpected linkage type!");
329 case GlobalValue::AppendingLinkage
: // FIXME: This should be improved!
330 case GlobalValue::ExternalLinkage
:
331 BSSSym
.SetBind(ELFSym::STB_GLOBAL
);
333 case GlobalValue::InternalLinkage
:
334 BSSSym
.SetBind(ELFSym::STB_LOCAL
);
338 // Set the idx of the .bss section
339 BSSSym
.SectionIdx
= BSSSection
.SectionIdx
;
340 if (!GV
->hasPrivateLinkage())
341 SymbolTable
.push_back(BSSSym
);
343 // Reserve space in the .bss section for this symbol.
344 BSSSection
.Size
+= Size
;
348 // FIXME: handle .rodata
349 //assert(!GV->isConstant() && "unimp");
351 // FIXME: handle .data
352 //assert(0 && "unimp");
356 bool ELFWriter::runOnMachineFunction(MachineFunction
&MF
) {
357 // Nothing to do here, this is all done through the MCE object above.
361 /// doFinalization - Now that the module has been completely processed, emit
362 /// the ELF file to 'O'.
363 bool ELFWriter::doFinalization(Module
&M
) {
364 // Okay, the ELF header and .text sections have been completed, build the
365 // .data, .bss, and "common" sections next.
366 for (Module::global_iterator I
= M
.global_begin(), E
= M
.global_end();
370 // Emit the symbol table now, if non-empty.
373 // FIXME: Emit the relocations now.
375 // Emit the string table for the sections in the ELF file we have.
376 EmitSectionTableStringTable();
378 // Emit the sections to the .o file, and emit the section table for the file.
379 OutputSectionsAndSectionTable();
381 // We are done with the abstract symbols.
385 // Release the name mangler object.
386 delete Mang
; Mang
= 0;
390 /// EmitSymbolTable - If the current symbol table is non-empty, emit the string
391 /// table for it and then the symbol table itself.
392 void ELFWriter::EmitSymbolTable() {
393 if (SymbolTable
.size() == 1) return; // Only the null entry.
395 // FIXME: compact all local symbols to the start of the symtab.
396 unsigned FirstNonLocalSymbol
= 1;
398 ELFSection
&StrTab
= getSection(".strtab", ELFSection::SHT_STRTAB
, 0);
401 DataBuffer
&StrTabBuf
= StrTab
.SectionData
;
402 OutputBuffer
StrTabOut(StrTabBuf
, is64Bit
, isLittleEndian
);
404 // Set the zero'th symbol to a null byte, as required.
405 StrTabOut
.outbyte(0);
406 SymbolTable
[0].NameIdx
= 0;
408 for (unsigned i
= 1, e
= SymbolTable
.size(); i
!= e
; ++i
) {
409 // Use the name mangler to uniquify the LLVM symbol.
410 std::string Name
= Mang
->getValueName(SymbolTable
[i
].GV
);
413 SymbolTable
[i
].NameIdx
= 0;
415 SymbolTable
[i
].NameIdx
= Index
;
417 // Add the name to the output buffer, including the null terminator.
418 StrTabBuf
.insert(StrTabBuf
.end(), Name
.begin(), Name
.end());
420 // Add a null terminator.
421 StrTabBuf
.push_back(0);
423 // Keep track of the number of bytes emitted to this section.
424 Index
+= Name
.size()+1;
427 assert(Index
== StrTabBuf
.size());
430 // Now that we have emitted the string table and know the offset into the
431 // string table of each symbol, emit the symbol table itself.
432 ELFSection
&SymTab
= getSection(".symtab", ELFSection::SHT_SYMTAB
, 0);
433 SymTab
.Align
= is64Bit
? 8 : 4;
434 SymTab
.Link
= SymTab
.SectionIdx
; // Section Index of .strtab.
435 SymTab
.Info
= FirstNonLocalSymbol
; // First non-STB_LOCAL symbol.
436 SymTab
.EntSize
= 16; // Size of each symtab entry. FIXME: wrong for ELF64
437 DataBuffer
&SymTabBuf
= SymTab
.SectionData
;
438 OutputBuffer
SymTabOut(SymTabBuf
, is64Bit
, isLittleEndian
);
440 if (!is64Bit
) { // 32-bit and 64-bit formats are shuffled a bit.
441 for (unsigned i
= 0, e
= SymbolTable
.size(); i
!= e
; ++i
) {
442 ELFSym
&Sym
= SymbolTable
[i
];
443 SymTabOut
.outword(Sym
.NameIdx
);
444 SymTabOut
.outaddr32(Sym
.Value
);
445 SymTabOut
.outword(Sym
.Size
);
446 SymTabOut
.outbyte(Sym
.Info
);
447 SymTabOut
.outbyte(Sym
.Other
);
448 SymTabOut
.outhalf(Sym
.SectionIdx
);
451 for (unsigned i
= 0, e
= SymbolTable
.size(); i
!= e
; ++i
) {
452 ELFSym
&Sym
= SymbolTable
[i
];
453 SymTabOut
.outword(Sym
.NameIdx
);
454 SymTabOut
.outbyte(Sym
.Info
);
455 SymTabOut
.outbyte(Sym
.Other
);
456 SymTabOut
.outhalf(Sym
.SectionIdx
);
457 SymTabOut
.outaddr64(Sym
.Value
);
458 SymTabOut
.outxword(Sym
.Size
);
462 SymTab
.Size
= SymTabBuf
.size();
465 /// EmitSectionTableStringTable - This method adds and emits a section for the
466 /// ELF Section Table string table: the string table that holds all of the
468 void ELFWriter::EmitSectionTableStringTable() {
469 // First step: add the section for the string table to the list of sections:
470 ELFSection
&SHStrTab
= getSection(".shstrtab", ELFSection::SHT_STRTAB
, 0);
472 // Now that we know which section number is the .shstrtab section, update the
473 // e_shstrndx entry in the ELF header.
474 OutputBuffer
FHOut(FileHeader
, is64Bit
, isLittleEndian
);
475 FHOut
.fixhalf(SHStrTab
.SectionIdx
, ELFHeader_e_shstrndx_Offset
);
477 // Set the NameIdx of each section in the string table and emit the bytes for
480 DataBuffer
&Buf
= SHStrTab
.SectionData
;
482 for (std::list
<ELFSection
>::iterator I
= SectionList
.begin(),
483 E
= SectionList
.end(); I
!= E
; ++I
) {
484 // Set the index into the table. Note if we have lots of entries with
485 // common suffixes, we could memoize them here if we cared.
488 // Add the name to the output buffer, including the null terminator.
489 Buf
.insert(Buf
.end(), I
->Name
.begin(), I
->Name
.end());
491 // Add a null terminator.
494 // Keep track of the number of bytes emitted to this section.
495 Index
+= I
->Name
.size()+1;
498 // Set the size of .shstrtab now that we know what it is.
499 assert(Index
== Buf
.size());
500 SHStrTab
.Size
= Index
;
503 /// OutputSectionsAndSectionTable - Now that we have constructed the file header
504 /// and all of the sections, emit these to the ostream destination and emit the
506 void ELFWriter::OutputSectionsAndSectionTable() {
507 // Pass #1: Compute the file offset for each section.
508 size_t FileOff
= FileHeader
.size(); // File header first.
510 // Emit all of the section data in order.
511 for (std::list
<ELFSection
>::iterator I
= SectionList
.begin(),
512 E
= SectionList
.end(); I
!= E
; ++I
) {
513 // Align FileOff to whatever the alignment restrictions of the section are.
515 FileOff
= (FileOff
+I
->Align
-1) & ~(I
->Align
-1);
517 FileOff
+= I
->SectionData
.size();
520 // Align Section Header.
521 unsigned TableAlign
= is64Bit
? 8 : 4;
522 FileOff
= (FileOff
+TableAlign
-1) & ~(TableAlign
-1);
524 // Now that we know where all of the sections will be emitted, set the e_shnum
525 // entry in the ELF header.
526 OutputBuffer
FHOut(FileHeader
, is64Bit
, isLittleEndian
);
527 FHOut
.fixhalf(NumSections
, ELFHeader_e_shnum_Offset
);
529 // Now that we know the offset in the file of the section table, update the
530 // e_shoff address in the ELF header.
531 FHOut
.fixaddr(FileOff
, ELFHeader_e_shoff_Offset
);
533 // Now that we know all of the data in the file header, emit it and all of the
535 O
.write((char*)&FileHeader
[0], FileHeader
.size());
536 FileOff
= FileHeader
.size();
537 DataBuffer().swap(FileHeader
);
540 OutputBuffer
TableOut(Table
, is64Bit
, isLittleEndian
);
542 // Emit all of the section data and build the section table itself.
543 while (!SectionList
.empty()) {
544 const ELFSection
&S
= *SectionList
.begin();
546 // Align FileOff to whatever the alignment restrictions of the section are.
548 for (size_t NewFileOff
= (FileOff
+S
.Align
-1) & ~(S
.Align
-1);
549 FileOff
!= NewFileOff
; ++FileOff
)
551 O
.write((char*)&S
.SectionData
[0], S
.SectionData
.size());
552 FileOff
+= S
.SectionData
.size();
554 TableOut
.outword(S
.NameIdx
); // sh_name - Symbol table name idx
555 TableOut
.outword(S
.Type
); // sh_type - Section contents & semantics
556 TableOut
.outword(S
.Flags
); // sh_flags - Section flags.
557 TableOut
.outaddr(S
.Addr
); // sh_addr - The mem addr this section is in.
558 TableOut
.outaddr(S
.Offset
); // sh_offset - Offset from the file start.
559 TableOut
.outword(S
.Size
); // sh_size - The section size.
560 TableOut
.outword(S
.Link
); // sh_link - Section header table index link.
561 TableOut
.outword(S
.Info
); // sh_info - Auxillary information.
562 TableOut
.outword(S
.Align
); // sh_addralign - Alignment of section.
563 TableOut
.outword(S
.EntSize
); // sh_entsize - Size of entries in the section
565 SectionList
.pop_front();
568 // Align output for the section table.
569 for (size_t NewFileOff
= (FileOff
+TableAlign
-1) & ~(TableAlign
-1);
570 FileOff
!= NewFileOff
; ++FileOff
)
573 // Emit the section table itself.
574 O
.write((char*)&Table
[0], Table
.size());