1 //=== MachOWriter.h - Target-independent Mach-O writer support --*- C++ -*-===//
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 defines the MachOWriter class.
12 //===----------------------------------------------------------------------===//
17 #include "llvm/Constants.h"
18 #include "llvm/DerivedTypes.h"
19 #include "llvm/CodeGen/MachineFunctionPass.h"
20 #include "llvm/CodeGen/MachineRelocation.h"
21 #include "llvm/Target/TargetData.h"
22 #include "llvm/Target/TargetMachine.h"
23 #include "llvm/Target/TargetMachOWriterInfo.h"
29 class MachineCodeEmitter;
30 class MachOCodeEmitter;
33 /// MachOSym - This struct contains information about each symbol that is
34 /// added to logical symbol table for the module. This is eventually
35 /// turned into a real symbol table in the file.
37 const GlobalValue *GV; // The global value this corresponds to.
38 std::string GVName; // The mangled name of the global value.
39 uint32_t n_strx; // index into the string table
40 uint8_t n_type; // type flag
41 uint8_t n_sect; // section number or NO_SECT
42 int16_t n_desc; // see <mach-o/stab.h>
43 uint64_t n_value; // value for this symbol (or stab offset)
45 // Constants for the n_sect field
46 // see <mach-o/nlist.h>
47 enum { NO_SECT = 0 }; // symbol is not in any section
49 // Constants for the n_type field
50 // see <mach-o/nlist.h>
51 enum { N_UNDF = 0x0, // undefined, n_sect == NO_SECT
52 N_ABS = 0x2, // absolute, n_sect == NO_SECT
53 N_SECT = 0xe, // defined in section number n_sect
54 N_PBUD = 0xc, // prebound undefined (defined in a dylib)
55 N_INDR = 0xa // indirect
57 // The following bits are OR'd into the types above. For example, a type
58 // of 0x0f would be an external N_SECT symbol (0x0e | 0x01).
59 enum { N_EXT = 0x01, // external symbol bit
60 N_PEXT = 0x10 // private external symbol bit
63 // Constants for the n_desc field
64 // see <mach-o/loader.h>
65 enum { REFERENCE_FLAG_UNDEFINED_NON_LAZY = 0,
66 REFERENCE_FLAG_UNDEFINED_LAZY = 1,
67 REFERENCE_FLAG_DEFINED = 2,
68 REFERENCE_FLAG_PRIVATE_DEFINED = 3,
69 REFERENCE_FLAG_PRIVATE_UNDEFINED_NON_LAZY = 4,
70 REFERENCE_FLAG_PRIVATE_UNDEFINED_LAZY = 5
72 enum { N_NO_DEAD_STRIP = 0x0020, // symbol is not to be dead stripped
73 N_WEAK_REF = 0x0040, // symbol is weak referenced
74 N_WEAK_DEF = 0x0080 // coalesced symbol is a weak definition
77 MachOSym(const GlobalValue *gv, std::string name, uint8_t sect,
81 /// MachOWriter - This class implements the common target-independent code for
82 /// writing Mach-O files. Targets should derive a class from this to
83 /// parameterize the output format.
85 class MachOWriter : public MachineFunctionPass {
86 friend class MachOCodeEmitter;
89 MachineCodeEmitter &getMachineCodeEmitter() const {
90 return *(MachineCodeEmitter*)MCE;
93 MachOWriter(std::ostream &O, TargetMachine &TM);
94 virtual ~MachOWriter();
96 virtual const char *getPassName() const {
97 return "Mach-O Writer";
100 typedef std::vector<unsigned char> DataBuffer;
102 /// Output stream to send the resultant object file to.
106 /// Target machine description.
110 /// Mang - The object used to perform name mangling for this module.
114 /// MCE - The MachineCodeEmitter object that we are exposing to emit machine
115 /// code for functions to the .o file.
116 MachOCodeEmitter *MCE;
118 /// is64Bit/isLittleEndian - This information is inferred from the target
119 /// machine directly, indicating what header values and flags to set.
120 bool is64Bit, isLittleEndian;
122 /// doInitialization - Emit the file header and all of the global variables
123 /// for the module to the Mach-O file.
124 bool doInitialization(Module &M);
126 bool runOnMachineFunction(MachineFunction &MF);
128 /// doFinalization - Now that the module has been completely processed, emit
129 /// the Mach-O file to 'O'.
130 bool doFinalization(Module &M);
132 /// MachOHeader - This struct contains the header information about a
133 /// specific architecture type/subtype pair that is emitted to the file.
135 uint32_t magic; // mach magic number identifier
136 uint32_t filetype; // type of file
137 uint32_t ncmds; // number of load commands
138 uint32_t sizeofcmds; // the size of all the load commands
139 uint32_t flags; // flags
140 uint32_t reserved; // 64-bit only
142 /// HeaderData - The actual data for the header which we are building
143 /// up for emission to the file.
144 DataBuffer HeaderData;
146 // Constants for the filetype field
147 // see <mach-o/loader.h> for additional info on the various types
148 enum { MH_OBJECT = 1, // relocatable object file
149 MH_EXECUTE = 2, // demand paged executable file
150 MH_FVMLIB = 3, // fixed VM shared library file
151 MH_CORE = 4, // core file
152 MH_PRELOAD = 5, // preloaded executable file
153 MH_DYLIB = 6, // dynamically bound shared library
154 MH_DYLINKER = 7, // dynamic link editor
155 MH_BUNDLE = 8, // dynamically bound bundle file
156 MH_DYLIB_STUB = 9, // shared library stub for static linking only
157 MH_DSYM = 10 // companion file wiht only debug sections
160 // Constants for the flags field
161 enum { MH_NOUNDEFS = 1 << 0,
162 // the object file has no undefined references
163 MH_INCRLINK = 1 << 1,
164 // the object file is the output of an incremental link against
165 // a base file and cannot be link edited again
166 MH_DYLDLINK = 1 << 2,
167 // the object file is input for the dynamic linker and cannot be
168 // statically link edited again.
169 MH_BINDATLOAD = 1 << 3,
170 // the object file's undefined references are bound by the
171 // dynamic linker when loaded.
172 MH_PREBOUND = 1 << 4,
173 // the file has its dynamic undefined references prebound
174 MH_SPLIT_SEGS = 1 << 5,
175 // the file has its read-only and read-write segments split
176 // see <mach/shared_memory_server.h>
177 MH_LAZY_INIT = 1 << 6,
178 // the shared library init routine is to be run lazily via
179 // catching memory faults to its writable segments (obsolete)
180 MH_TWOLEVEL = 1 << 7,
181 // the image is using two-level namespace bindings
182 MH_FORCE_FLAT = 1 << 8,
183 // the executable is forcing all images to use flat namespace
185 MH_NOMULTIDEFS = 1 << 8,
186 // this umbrella guarantees no multiple definitions of symbols
187 // in its sub-images so the two-level namespace hints can
189 MH_NOFIXPREBINDING = 1 << 10,
190 // do not have dyld notify the prebidning agent about this
192 MH_PREBINDABLE = 1 << 11,
193 // the binary is not prebound but can have its prebinding
194 // redone. only used when MH_PREBOUND is not set.
195 MH_ALLMODSBOUND = 1 << 12,
196 // indicates that this binary binds to all two-level namespace
197 // modules of its dependent libraries. Only used when
198 // MH_PREBINDABLE and MH_TWOLEVEL are both set.
199 MH_SUBSECTIONS_VIA_SYMBOLS = 1 << 13,
200 // safe to divide up the sections into sub-sections via symbols
201 // for dead code stripping.
202 MH_CANONICAL = 1 << 14,
203 // the binary has been canonicalized via the unprebind operation
204 MH_WEAK_DEFINES = 1 << 15,
205 // the final linked image contains external weak symbols
206 MH_BINDS_TO_WEAK = 1 << 16,
207 // the final linked image uses weak symbols
208 MH_ALLOW_STACK_EXECUTION = 1 << 17
209 // When this bit is set, all stacks in the task will be given
210 // stack execution privilege. Only used in MH_EXECUTE filetype
213 MachOHeader() : magic(0), filetype(0), ncmds(0), sizeofcmds(0), flags(0),
216 /// cmdSize - This routine returns the size of the MachOSection as written
217 /// to disk, depending on whether the destination is a 64 bit Mach-O file.
218 unsigned cmdSize(bool is64Bit) const {
220 return 8 * sizeof(uint32_t);
222 return 7 * sizeof(uint32_t);
225 /// setMagic - This routine sets the appropriate value for the 'magic'
226 /// field based on pointer size and endianness.
227 void setMagic(bool isLittleEndian, bool is64Bit) {
229 if (is64Bit) magic = 0xcffaedfe;
230 else magic = 0xcefaedfe;
232 if (is64Bit) magic = 0xfeedfacf;
233 else magic = 0xfeedface;
237 /// Header - An instance of MachOHeader that we will update while we build
238 /// the file, and then emit during finalization.
241 /// MachOSegment - This struct contains the necessary information to
242 /// emit the load commands for each section in the file.
243 struct MachOSegment {
244 uint32_t cmd; // LC_SEGMENT or LC_SEGMENT_64
245 uint32_t cmdsize; // Total size of this struct and section commands
246 std::string segname; // segment name
247 uint64_t vmaddr; // address of this segment
248 uint64_t vmsize; // size of this segment, may be larger than filesize
249 uint64_t fileoff; // offset in file
250 uint64_t filesize; // amount to read from file
251 uint32_t maxprot; // maximum VM protection
252 uint32_t initprot; // initial VM protection
253 uint32_t nsects; // number of sections in this segment
254 uint32_t flags; // flags
256 // The following constants are getting pulled in by one of the
257 // system headers, which creates a neat clash with the enum.
258 #if !defined(VM_PROT_NONE)
259 #define VM_PROT_NONE 0x00
261 #if !defined(VM_PROT_READ)
262 #define VM_PROT_READ 0x01
264 #if !defined(VM_PROT_WRITE)
265 #define VM_PROT_WRITE 0x02
267 #if !defined(VM_PROT_EXECUTE)
268 #define VM_PROT_EXECUTE 0x04
270 #if !defined(VM_PROT_ALL)
271 #define VM_PROT_ALL 0x07
274 // Constants for the vm protection fields
275 // see <mach-o/vm_prot.h>
276 enum { SEG_VM_PROT_NONE = VM_PROT_NONE,
277 SEG_VM_PROT_READ = VM_PROT_READ, // read permission
278 SEG_VM_PROT_WRITE = VM_PROT_WRITE, // write permission
279 SEG_VM_PROT_EXECUTE = VM_PROT_EXECUTE,
280 SEG_VM_PROT_ALL = VM_PROT_ALL
283 // Constants for the cmd field
284 // see <mach-o/loader.h>
285 enum { LC_SEGMENT = 0x01, // segment of this file to be mapped
286 LC_SEGMENT_64 = 0x19 // 64-bit segment of this file to be mapped
289 /// cmdSize - This routine returns the size of the MachOSection as written
290 /// to disk, depending on whether the destination is a 64 bit Mach-O file.
291 unsigned cmdSize(bool is64Bit) const {
293 return 6 * sizeof(uint32_t) + 4 * sizeof(uint64_t) + 16;
295 return 10 * sizeof(uint32_t) + 16; // addresses only 32 bits
298 MachOSegment(const std::string &seg, bool is64Bit)
299 : cmd(is64Bit ? LC_SEGMENT_64 : LC_SEGMENT), cmdsize(0), segname(seg),
300 vmaddr(0), vmsize(0), fileoff(0), filesize(0), maxprot(VM_PROT_ALL),
301 initprot(VM_PROT_ALL), nsects(0), flags(0) { }
304 /// MachOSection - This struct contains information about each section in a
305 /// particular segment that is emitted to the file. This is eventually
306 /// turned into the SectionCommand in the load command for a particlar
308 struct MachOSection {
309 std::string sectname; // name of this section,
310 std::string segname; // segment this section goes in
311 uint64_t addr; // memory address of this section
312 uint64_t size; // size in bytes of this section
313 uint32_t offset; // file offset of this section
314 uint32_t align; // section alignment (power of 2)
315 uint32_t reloff; // file offset of relocation entries
316 uint32_t nreloc; // number of relocation entries
317 uint32_t flags; // flags (section type and attributes)
318 uint32_t reserved1; // reserved (for offset or index)
319 uint32_t reserved2; // reserved (for count or sizeof)
320 uint32_t reserved3; // reserved (64 bit only)
322 /// A unique number for this section, which will be used to match symbols
323 /// to the correct section.
326 /// SectionData - The actual data for this section which we are building
327 /// up for emission to the file.
328 DataBuffer SectionData;
330 /// RelocBuffer - A buffer to hold the mach-o relocations before we write
331 /// them out at the appropriate location in the file.
332 DataBuffer RelocBuffer;
334 /// Relocations - The relocations that we have encountered so far in this
335 /// section that we will need to convert to MachORelocation entries when
336 /// the file is written.
337 std::vector<MachineRelocation> Relocations;
339 // Constants for the section types (low 8 bits of flags field)
340 // see <mach-o/loader.h>
341 enum { S_REGULAR = 0,
344 // zero fill on demand section
345 S_CSTRING_LITERALS = 2,
346 // section with only literal C strings
347 S_4BYTE_LITERALS = 3,
348 // section with only 4 byte literals
349 S_8BYTE_LITERALS = 4,
350 // section with only 8 byte literals
351 S_LITERAL_POINTERS = 5,
352 // section with only pointers to literals
353 S_NON_LAZY_SYMBOL_POINTERS = 6,
354 // section with only non-lazy symbol pointers
355 S_LAZY_SYMBOL_POINTERS = 7,
356 // section with only lazy symbol pointers
358 // section with only symbol stubs
359 // byte size of stub in the reserved2 field
360 S_MOD_INIT_FUNC_POINTERS = 9,
361 // section with only function pointers for initialization
362 S_MOD_TERM_FUNC_POINTERS = 10,
363 // section with only function pointers for termination
365 // section contains symbols that are coalesced
367 // zero fill on demand section (that can be larger than 4GB)
369 // section with only pairs of function pointers for interposing
370 S_16BYTE_LITERALS = 14
371 // section with only 16 byte literals
374 // Constants for the section flags (high 24 bits of flags field)
375 // see <mach-o/loader.h>
376 enum { S_ATTR_PURE_INSTRUCTIONS = 1 << 31,
377 // section contains only true machine instructions
378 S_ATTR_NO_TOC = 1 << 30,
379 // section contains coalesced symbols that are not to be in a
380 // ranlib table of contents
381 S_ATTR_STRIP_STATIC_SYMS = 1 << 29,
382 // ok to strip static symbols in this section in files with the
384 S_ATTR_NO_DEAD_STRIP = 1 << 28,
386 S_ATTR_LIVE_SUPPORT = 1 << 27,
387 // blocks are live if they reference live blocks
388 S_ATTR_SELF_MODIFYING_CODE = 1 << 26,
389 // used with i386 code stubs written on by dyld
390 S_ATTR_DEBUG = 1 << 25,
392 S_ATTR_SOME_INSTRUCTIONS = 1 << 10,
393 // section contains some machine instructions
394 S_ATTR_EXT_RELOC = 1 << 9,
395 // section has external relocation entries
396 S_ATTR_LOC_RELOC = 1 << 8
397 // section has local relocation entries
400 /// cmdSize - This routine returns the size of the MachOSection as written
401 /// to disk, depending on whether the destination is a 64 bit Mach-O file.
402 unsigned cmdSize(bool is64Bit) const {
404 return 7 * sizeof(uint32_t) + 2 * sizeof(uint64_t) + 32;
406 return 9 * sizeof(uint32_t) + 32; // addresses only 32 bits
409 MachOSection(const std::string &seg, const std::string §)
410 : sectname(sect), segname(seg), addr(0), size(0), offset(0), align(2),
411 reloff(0), nreloc(0), flags(0), reserved1(0), reserved2(0),
417 /// SectionList - This is the list of sections that we have emitted to the
418 /// file. Once the file has been completely built, the segment load command
419 /// SectionCommands are constructed from this info.
420 std::vector<MachOSection*> SectionList;
422 /// SectionLookup - This is a mapping from section name to SectionList entry
423 std::map<std::string, MachOSection*> SectionLookup;
425 /// GVSection - This is a mapping from a GlobalValue to a MachOSection,
426 /// to aid in emitting relocations.
427 std::map<GlobalValue*, MachOSection*> GVSection;
429 /// GVOffset - This is a mapping from a GlobalValue to an offset from the
430 /// start of the section in which the GV resides, to aid in emitting
432 std::map<GlobalValue*, intptr_t> GVOffset;
434 /// getSection - Return the section with the specified name, creating a new
435 /// section if one does not already exist.
436 MachOSection *getSection(const std::string &seg, const std::string §,
437 unsigned Flags = 0) {
438 MachOSection *MOS = SectionLookup[seg+sect];
441 MOS = new MachOSection(seg, sect);
442 SectionList.push_back(MOS);
443 MOS->Index = SectionList.size();
444 MOS->flags = MachOSection::S_REGULAR | Flags;
445 SectionLookup[seg+sect] = MOS;
448 MachOSection *getTextSection(bool isCode = true) {
450 return getSection("__TEXT", "__text",
451 MachOSection::S_ATTR_PURE_INSTRUCTIONS |
452 MachOSection::S_ATTR_SOME_INSTRUCTIONS);
454 return getSection("__TEXT", "__text");
456 MachOSection *getBSSSection() {
457 return getSection("__DATA", "__bss", MachOSection::S_ZEROFILL);
459 MachOSection *getDataSection() {
460 return getSection("__DATA", "__data");
462 MachOSection *getConstSection(Constant *C) {
463 const ConstantArray *CVA = dyn_cast<ConstantArray>(C);
464 if (CVA && CVA->isCString())
465 return getSection("__TEXT", "__cstring",
466 MachOSection::S_CSTRING_LITERALS);
468 const Type *Ty = C->getType();
469 if (Ty->isPrimitiveType() || Ty->isInteger()) {
470 unsigned Size = TM.getTargetData()->getABITypeSize(Ty);
472 default: break; // Fall through to __TEXT,__const
474 return getSection("__TEXT", "__literal4",
475 MachOSection::S_4BYTE_LITERALS);
477 return getSection("__TEXT", "__literal8",
478 MachOSection::S_8BYTE_LITERALS);
480 return getSection("__TEXT", "__literal16",
481 MachOSection::S_16BYTE_LITERALS);
484 return getSection("__TEXT", "__const");
486 MachOSection *getJumpTableSection() {
487 if (TM.getRelocationModel() == Reloc::PIC_)
488 return getTextSection(false);
490 return getSection("__TEXT", "__const");
493 /// MachOSymTab - This struct contains information about the offsets and
494 /// size of symbol table information.
497 uint32_t cmd; // LC_SYMTAB
498 uint32_t cmdsize; // sizeof( MachOSymTab )
499 uint32_t symoff; // symbol table offset
500 uint32_t nsyms; // number of symbol table entries
501 uint32_t stroff; // string table offset
502 uint32_t strsize; // string table size in bytes
504 // Constants for the cmd field
505 // see <mach-o/loader.h>
506 enum { LC_SYMTAB = 0x02 // link-edit stab symbol table info
509 MachOSymTab() : cmd(LC_SYMTAB), cmdsize(6 * sizeof(uint32_t)), symoff(0),
510 nsyms(0), stroff(0), strsize(0) { }
513 /// MachOSymTab - This struct contains information about the offsets and
514 /// size of symbol table information.
516 struct MachODySymTab {
517 uint32_t cmd; // LC_DYSYMTAB
518 uint32_t cmdsize; // sizeof( MachODySymTab )
519 uint32_t ilocalsym; // index to local symbols
520 uint32_t nlocalsym; // number of local symbols
521 uint32_t iextdefsym; // index to externally defined symbols
522 uint32_t nextdefsym; // number of externally defined symbols
523 uint32_t iundefsym; // index to undefined symbols
524 uint32_t nundefsym; // number of undefined symbols
525 uint32_t tocoff; // file offset to table of contents
526 uint32_t ntoc; // number of entries in table of contents
527 uint32_t modtaboff; // file offset to module table
528 uint32_t nmodtab; // number of module table entries
529 uint32_t extrefsymoff; // offset to referenced symbol table
530 uint32_t nextrefsyms; // number of referenced symbol table entries
531 uint32_t indirectsymoff; // file offset to the indirect symbol table
532 uint32_t nindirectsyms; // number of indirect symbol table entries
533 uint32_t extreloff; // offset to external relocation entries
534 uint32_t nextrel; // number of external relocation entries
535 uint32_t locreloff; // offset to local relocation entries
536 uint32_t nlocrel; // number of local relocation entries
538 // Constants for the cmd field
539 // see <mach-o/loader.h>
540 enum { LC_DYSYMTAB = 0x0B // dynamic link-edit symbol table info
543 MachODySymTab() : cmd(LC_DYSYMTAB), cmdsize(20 * sizeof(uint32_t)),
544 ilocalsym(0), nlocalsym(0), iextdefsym(0), nextdefsym(0),
545 iundefsym(0), nundefsym(0), tocoff(0), ntoc(0), modtaboff(0),
546 nmodtab(0), extrefsymoff(0), nextrefsyms(0), indirectsymoff(0),
547 nindirectsyms(0), extreloff(0), nextrel(0), locreloff(0), nlocrel(0) { }
550 /// SymTab - The "stab" style symbol table information
552 /// DySymTab - symbol table info for the dynamic link editor
553 MachODySymTab DySymTab;
556 // FIXME: this does not appear to be sorting 'f' after 'F'
557 bool operator()(const MachOSym &LHS, const MachOSym &RHS) {
558 return LHS.GVName < RHS.GVName;
562 /// PartitionByLocal - Simple boolean predicate that returns true if Sym is
563 /// a local symbol rather than an external symbol.
564 static bool PartitionByLocal(const MachOSym &Sym);
566 /// PartitionByDefined - Simple boolean predicate that returns true if Sym
567 /// is defined in this module.
568 static bool PartitionByDefined(const MachOSym &Sym);
572 /// SymbolTable - This is the list of symbols we have emitted to the file.
573 /// This actually gets rearranged before emission to the file (to put the
574 /// local symbols first in the list).
575 std::vector<MachOSym> SymbolTable;
577 /// SymT - A buffer to hold the symbol table before we write it out at the
578 /// appropriate location in the file.
581 /// StrT - A buffer to hold the string table before we write it out at the
582 /// appropriate location in the file.
585 /// PendingSyms - This is a list of externally defined symbols that we have
586 /// been asked to emit, but have not seen a reference to. When a reference
587 /// is seen, the symbol will move from this list to the SymbolTable.
588 std::vector<GlobalValue*> PendingGlobals;
590 /// DynamicSymbolTable - This is just a vector of indices into
591 /// SymbolTable to aid in emitting the DYSYMTAB load command.
592 std::vector<unsigned> DynamicSymbolTable;
594 static void InitMem(const Constant *C, void *Addr, intptr_t Offset,
595 const TargetData *TD,
596 std::vector<MachineRelocation> &MRs);
599 void AddSymbolToSection(MachOSection *MOS, GlobalVariable *GV);
600 void EmitGlobal(GlobalVariable *GV);
601 void EmitHeaderAndLoadCommands();
603 void BufferSymbolAndStringTable();
604 void CalculateRelocations(MachOSection &MOS);
606 MachineRelocation GetJTRelocation(unsigned Offset,
607 MachineBasicBlock *MBB) const {
608 return TM.getMachOWriterInfo()->GetJTRelocation(Offset, MBB);
611 /// GetTargetRelocation - Returns the number of relocations.
612 unsigned GetTargetRelocation(MachineRelocation &MR,
616 OutputBuffer &RelocOut,
617 OutputBuffer &SecOut,
620 return TM.getMachOWriterInfo()->GetTargetRelocation(MR, FromIdx, ToAddr,