1 //=== MachOWriter.h - Target-independent Mach-O writer support --*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by Nate Begeman and is distributed under the
6 // University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the MachOWriter class.
12 //===----------------------------------------------------------------------===//
14 #ifndef LLVM_CODEGEN_MACHOWRITER_H
15 #define LLVM_CODEGEN_MACHOWRITER_H
17 #include "llvm/CodeGen/MachineRelocation.h"
18 #include "llvm/CodeGen/MachineFunctionPass.h"
24 class MachineCodeEmitter;
25 class MachOCodeEmitter;
27 /// MachOWriter - This class implements the common target-independent code for
28 /// writing Mach-O files. Targets should derive a class from this to
29 /// parameterize the output format.
31 class MachOWriter : public MachineFunctionPass {
32 friend class MachOCodeEmitter;
34 MachineCodeEmitter &getMachineCodeEmitter() const {
35 return *(MachineCodeEmitter*)MCE;
40 typedef std::vector<unsigned char> DataBuffer;
43 MachOWriter(std::ostream &O, TargetMachine &TM);
45 /// Output stream to send the resultant object file to.
49 /// Target machine description.
53 /// Mang - The object used to perform name mangling for this module.
57 /// MCE - The MachineCodeEmitter object that we are exposing to emit machine
58 /// code for functions to the .o file.
59 MachOCodeEmitter *MCE;
61 /// is64Bit/isLittleEndian - This information is inferred from the target
62 /// machine directly, indicating what header values and flags to set.
63 bool is64Bit, isLittleEndian;
65 /// doInitialization - Emit the file header and all of the global variables
66 /// for the module to the Mach-O file.
67 bool doInitialization(Module &M);
69 bool runOnMachineFunction(MachineFunction &MF);
71 /// doFinalization - Now that the module has been completely processed, emit
72 /// the Mach-O file to 'O'.
73 bool doFinalization(Module &M);
75 /// MachOHeader - This struct contains the header information about a
76 /// specific architecture type/subtype pair that is emitted to the file.
78 uint32_t magic; // mach magic number identifier
79 uint32_t cputype; // cpu specifier
80 uint32_t cpusubtype; // machine specifier
81 uint32_t filetype; // type of file
82 uint32_t ncmds; // number of load commands
83 uint32_t sizeofcmds; // the size of all the load commands
84 uint32_t flags; // flags
85 uint32_t reserved; // 64-bit only
87 /// HeaderData - The actual data for the header which we are building
88 /// up for emission to the file.
89 DataBuffer HeaderData;
91 // Constants for the cputype field
92 // see <mach/machine.h>
93 enum { CPU_TYPE_I386 = 7,
94 CPU_TYPE_X86_64 = 7 | 0x1000000,
97 CPU_TYPE_POWERPC = 18,
98 CPU_TYPE_POWERPC64 = 18 | 0x1000000
101 // Constants for the cpusubtype field
102 // see <mach/machine.h>
103 enum { CPU_SUBTYPE_I386_ALL = 3,
104 CPU_SUBTYPE_X86_64_ALL = 3,
105 CPU_SUBTYPE_ARM_ALL = 0,
106 CPU_SUBTYPE_SPARC_ALL = 0,
107 CPU_SUBTYPE_POWERPC_ALL = 0
110 // Constants for the filetype field
111 // see <mach-o/loader.h> for additional info on the various types
112 enum { MH_OBJECT = 1, // relocatable object file
113 MH_EXECUTE = 2, // demand paged executable file
114 MH_FVMLIB = 3, // fixed VM shared library file
115 MH_CORE = 4, // core file
116 MH_PRELOAD = 5, // preloaded executable file
117 MH_DYLIB = 6, // dynamically bound shared library
118 MH_DYLINKER = 7, // dynamic link editor
119 MH_BUNDLE = 8, // dynamically bound bundle file
120 MH_DYLIB_STUB = 9, // shared library stub for static linking only
121 MH_DSYM = 10 // companion file wiht only debug sections
124 // Constants for the flags field
125 enum { MH_NOUNDEFS = 1 << 0,
126 // the object file has no undefined references
127 MH_INCRLINK = 1 << 1,
128 // the object file is the output of an incremental link against
129 // a base file and cannot be link edited again
130 MH_DYLDLINK = 1 << 2,
131 // the object file is input for the dynamic linker and cannot be
132 // statically link edited again.
133 MH_BINDATLOAD = 1 << 3,
134 // the object file's undefined references are bound by the
135 // dynamic linker when loaded.
136 MH_PREBOUND = 1 << 4,
137 // the file has its dynamic undefined references prebound
138 MH_SPLIT_SEGS = 1 << 5,
139 // the file has its read-only and read-write segments split
140 // see <mach/shared_memory_server.h>
141 MH_LAZY_INIT = 1 << 6,
142 // the shared library init routine is to be run lazily via
143 // catching memory faults to its writable segments (obsolete)
144 MH_TWOLEVEL = 1 << 7,
145 // the image is using two-level namespace bindings
146 MH_FORCE_FLAT = 1 << 8,
147 // the executable is forcing all images to use flat namespace
149 MH_NOMULTIDEFS = 1 << 8,
150 // this umbrella guarantees no multiple definitions of symbols
151 // in its sub-images so the two-level namespace hints can
153 MH_NOFIXPREBINDING = 1 << 10,
154 // do not have dyld notify the prebidning agent about this
156 MH_PREBINDABLE = 1 << 11,
157 // the binary is not prebound but can have its prebinding
158 // redone. only used when MH_PREBOUND is not set.
159 MH_ALLMODSBOUND = 1 << 12,
160 // indicates that this binary binds to all two-level namespace
161 // modules of its dependent libraries. Only used when
162 // MH_PREBINDABLE and MH_TWOLEVEL are both set.
163 MH_SUBSECTIONS_VIA_SYMBOLS = 1 << 13,
164 // safe to divide up the sections into sub-sections via symbols
165 // for dead code stripping.
166 MH_CANONICAL = 1 << 14,
167 // the binary has been canonicalized via the unprebind operation
168 MH_WEAK_DEFINES = 1 << 15,
169 // the final linked image contains external weak symbols
170 MH_BINDS_TO_WEAK = 1 << 16,
171 // the final linked image uses weak symbols
172 MH_ALLOW_STACK_EXECUTION = 1 << 17
173 // When this bit is set, all stacks in the task will be given
174 // stack execution privilege. Only used in MH_EXECUTE filetype
177 MachOHeader() : magic(0), cputype(0), cpusubtype(0), filetype(0),
178 ncmds(0), sizeofcmds(0), flags(0), reserved(0) { }
180 /// cmdSize - This routine returns the size of the MachOSection as written
181 /// to disk, depending on whether the destination is a 64 bit Mach-O file.
182 unsigned cmdSize(bool is64Bit) const {
184 return 8 * sizeof(uint32_t);
186 return 7 * sizeof(uint32_t);
189 /// setMagic - This routine sets the appropriate value for the 'magic'
190 /// field based on pointer size and endianness.
191 void setMagic(bool isLittleEndian, bool is64Bit) {
193 if (is64Bit) magic = 0xcffaedfe;
194 else magic = 0xcefaedfe;
196 if (is64Bit) magic = 0xfeedfacf;
197 else magic = 0xfeedface;
201 /// Header - An instance of MachOHeader that we will update while we build
202 /// the file, and then emit during finalization.
207 /// MachOSegment - This struct contains the necessary information to
208 /// emit the load commands for each section in the file.
209 struct MachOSegment {
210 uint32_t cmd; // LC_SEGMENT or LC_SEGMENT_64
211 uint32_t cmdsize; // Total size of this struct and section commands
212 std::string segname; // segment name
213 uint64_t vmaddr; // address of this segment
214 uint64_t vmsize; // size of this segment, may be larger than filesize
215 uint64_t fileoff; // offset in file
216 uint64_t filesize; // amount to read from file
217 uint32_t maxprot; // maximum VM protection
218 uint32_t initprot; // initial VM protection
219 uint32_t nsects; // number of sections in this segment
220 uint32_t flags; // flags
222 // Constants for the vm protection fields
223 // see <mach-o/vm_prot.h>
224 enum { VM_PROT_NONE = 0x00,
225 VM_PROT_READ = 0x01, // read permission
226 VM_PROT_WRITE = 0x02, // write permission
227 VM_PROT_EXECUTE = 0x04, // execute permission,
231 // Constants for the cmd field
232 // see <mach-o/loader.h>
233 enum { LC_SEGMENT = 0x01, // segment of this file to be mapped
234 LC_SEGMENT_64 = 0x19 // 64-bit segment of this file to be mapped
237 /// cmdSize - This routine returns the size of the MachOSection as written
238 /// to disk, depending on whether the destination is a 64 bit Mach-O file.
239 unsigned cmdSize(bool is64Bit) const {
241 return 6 * sizeof(uint32_t) + 4 * sizeof(uint64_t) + 16;
243 return 10 * sizeof(uint32_t) + 16; // addresses only 32 bits
246 MachOSegment(const std::string &seg, bool is64Bit)
247 : cmd(is64Bit ? LC_SEGMENT_64 : LC_SEGMENT), cmdsize(0), segname(seg),
248 vmaddr(0), vmsize(0), fileoff(0), filesize(0), maxprot(VM_PROT_ALL),
249 initprot(VM_PROT_ALL), nsects(0), flags(0) { }
252 /// MachOSection - This struct contains information about each section in a
253 /// particular segment that is emitted to the file. This is eventually
254 /// turned into the SectionCommand in the load command for a particlar
256 struct MachOSection {
257 std::string sectname; // name of this section,
258 std::string segname; // segment this section goes in
259 uint64_t addr; // memory address of this section
260 uint64_t size; // size in bytes of this section
261 uint32_t offset; // file offset of this section
262 uint32_t align; // section alignment (power of 2)
263 uint32_t reloff; // file offset of relocation entries
264 uint32_t nreloc; // number of relocation entries
265 uint32_t flags; // flags (section type and attributes)
266 uint32_t reserved1; // reserved (for offset or index)
267 uint32_t reserved2; // reserved (for count or sizeof)
268 uint32_t reserved3; // reserved (64 bit only)
270 /// A unique number for this section, which will be used to match symbols
271 /// to the correct section.
274 /// SectionData - The actual data for this section which we are building
275 /// up for emission to the file.
276 DataBuffer SectionData;
278 // Constants for the section types (low 8 bits of flags field)
279 // see <mach-o/loader.h>
280 enum { S_REGULAR = 0,
283 // zero fill on demand section
284 S_CSTRING_LITERALS = 2,
285 // section with only literal C strings
286 S_4BYTE_LITERALS = 3,
287 // section with only 4 byte literals
288 S_8BYTE_LITERALS = 4,
289 // section with only 8 byte literals
290 S_LITERAL_POINTERS = 5,
291 // section with only pointers to literals
292 S_NON_LAZY_SYMBOL_POINTERS = 6,
293 // section with only non-lazy symbol pointers
294 S_LAZY_SYMBOL_POINTERS = 7,
295 // section with only lazy symbol pointers
297 // section with only symbol stubs
298 // byte size of stub in the reserved2 field
299 S_MOD_INIT_FUNC_POINTERS = 9,
300 // section with only function pointers for initialization
301 S_MOD_TERM_FUNC_POINTERS = 10,
302 // section with only function pointers for termination
304 // section contains symbols that are coalesced
306 // zero fill on demand section (that can be larger than 4GB)
308 // section with only pairs of function pointers for interposing
309 S_16BYTE_LITERALS = 14
310 // section with only 16 byte literals
313 // Constants for the section flags (high 24 bits of flags field)
314 // see <mach-o/loader.h>
315 enum { S_ATTR_PURE_INSTRUCTIONS = 1 << 31,
316 // section contains only true machine instructions
317 S_ATTR_NO_TOC = 1 << 30,
318 // section contains coalesced symbols that are not to be in a
319 // ranlib table of contents
320 S_ATTR_STRIP_STATIC_SYMS = 1 << 29,
321 // ok to strip static symbols in this section in files with the
323 S_ATTR_NO_DEAD_STRIP = 1 << 28,
325 S_ATTR_LIVE_SUPPORT = 1 << 27,
326 // blocks are live if they reference live blocks
327 S_ATTR_SELF_MODIFYING_CODE = 1 << 26,
328 // used with i386 code stubs written on by dyld
329 S_ATTR_DEBUG = 1 << 25,
331 S_ATTR_SOME_INSTRUCTIONS = 1 << 10,
332 // section contains some machine instructions
333 S_ATTR_EXT_RELOC = 1 << 9,
334 // section has external relocation entries
335 S_ATTR_LOC_RELOC = 1 << 8
336 // section has local relocation entries
339 /// cmdSize - This routine returns the size of the MachOSection as written
340 /// to disk, depending on whether the destination is a 64 bit Mach-O file.
341 unsigned cmdSize(bool is64Bit) const {
343 return 7 * sizeof(uint32_t) + 2 * sizeof(uint64_t) + 32;
345 return 9 * sizeof(uint32_t) + 32; // addresses only 32 bits
348 MachOSection(const std::string &seg, const std::string §)
349 : sectname(sect), segname(seg), addr(0), size(0), offset(0), align(0),
350 reloff(0), nreloc(0), flags(0), reserved1(0), reserved2(0),
354 /// SectionList - This is the list of sections that we have emitted to the
355 /// file. Once the file has been completely built, the segment load command
356 /// SectionCommands are constructed from this info.
357 std::list<MachOSection> SectionList;
359 /// SectionLookup - This is a mapping from section name to SectionList entry
360 std::map<std::string, MachOSection*> SectionLookup;
362 /// getSection - Return the section with the specified name, creating a new
363 /// section if one does not already exist.
364 MachOSection &getSection(const std::string &seg, const std::string §,
365 unsigned Flags = 0) {
366 MachOSection *&SN = SectionLookup[seg+sect];
369 SectionList.push_back(MachOSection(seg, sect));
370 SN = &SectionList.back();
371 SN->Index = SectionList.size();
372 SN->flags = MachOSection::S_REGULAR | Flags;
375 MachOSection &getTextSection() {
376 return getSection("__TEXT", "__text",
377 MachOSection::S_ATTR_PURE_INSTRUCTIONS |
378 MachOSection::S_ATTR_SOME_INSTRUCTIONS);
381 /// MachOSymTab - This struct contains information about the offsets and
382 /// size of symbol table information.
385 uint32_t cmd; // LC_SYMTAB
386 uint32_t cmdsize; // sizeof( MachOSymTab )
387 uint32_t symoff; // symbol table offset
388 uint32_t nsyms; // number of symbol table entries
389 uint32_t stroff; // string table offset
390 uint32_t strsize; // string table size in bytes
392 // Constants for the cmd field
393 // see <mach-o/loader.h>
394 enum { LC_SYMTAB = 0x02 // link-edit stab symbol table info
397 MachOSymTab() : cmd(LC_SYMTAB), cmdsize(6 * sizeof(uint32_t)), symoff(0),
398 nsyms(0), stroff(0), strsize(0) { }
401 /// MachOSymTab - This struct contains information about the offsets and
402 /// size of symbol table information.
404 struct MachODySymTab {
405 uint32_t cmd; // LC_DYSYMTAB
406 uint32_t cmdsize; // sizeof( MachODySymTab )
407 uint32_t ilocalsym; // index to local symbols
408 uint32_t nlocalsym; // number of local symbols
409 uint32_t iextdefsym; // index to externally defined symbols
410 uint32_t nextdefsym; // number of externally defined symbols
411 uint32_t iundefsym; // index to undefined symbols
412 uint32_t nundefsym; // number of undefined symbols
413 uint32_t tocoff; // file offset to table of contents
414 uint32_t ntoc; // number of entries in table of contents
415 uint32_t modtaboff; // file offset to module table
416 uint32_t nmodtab; // number of module table entries
417 uint32_t extrefsymoff; // offset to referenced symbol table
418 uint32_t nextrefsyms; // number of referenced symbol table entries
419 uint32_t indirectsymoff; // file offset to the indirect symbol table
420 uint32_t nindirectsyms; // number of indirect symbol table entries
421 uint32_t extreloff; // offset to external relocation entries
422 uint32_t nextrel; // number of external relocation entries
423 uint32_t locreloff; // offset to local relocation entries
424 uint32_t nlocrel; // number of local relocation entries
426 // Constants for the cmd field
427 // see <mach-o/loader.h>
428 enum { LC_DYSYMTAB = 0x0B // dynamic link-edit symbol table info
431 MachODySymTab() : cmd(LC_DYSYMTAB), cmdsize(20 * sizeof(uint32_t)),
432 ilocalsym(0), nlocalsym(0), iextdefsym(0), nextdefsym(0),
433 iundefsym(0), nundefsym(0), tocoff(0), ntoc(0), modtaboff(0),
434 nmodtab(0), extrefsymoff(0), nextrefsyms(0), indirectsymoff(0),
435 nindirectsyms(0), extreloff(0), nextrel(0), locreloff(0), nlocrel(0) { }
438 /// SymTab - The "stab" style symbol table information
440 /// DySymTab - symbol table info for the dynamic link editor
441 MachODySymTab DySymTab;
443 /// MachOSym - This struct contains information about each symbol that is
444 /// added to logical symbol table for the module. This is eventually
445 /// turned into a real symbol table in the file.
447 const GlobalValue *GV; // The global value this corresponds to.
448 uint32_t n_strx; // index into the string table
449 uint8_t n_type; // type flag
450 uint8_t n_sect; // section number or NO_SECT
451 int16_t n_desc; // see <mach-o/stab.h>
452 uint64_t n_value; // value for this symbol (or stab offset)
454 // Constants for the n_sect field
455 // see <mach-o/nlist.h>
456 enum { NO_SECT = 0 }; // symbol is not in any section
458 // Constants for the n_type field
459 // see <mach-o/nlist.h>
460 enum { N_UNDF = 0x0, // undefined, n_sect == NO_SECT
461 N_ABS = 0x2, // absolute, n_sect == NO_SECT
462 N_SECT = 0xe, // defined in section number n_sect
463 N_PBUD = 0xc, // prebound undefined (defined in a dylib)
464 N_INDR = 0xa // indirect
466 // The following bits are OR'd into the types above. For example, a type
467 // of 0x0f would be an external N_SECT symbol (0x0e | 0x01).
468 enum { N_EXT = 0x01, // external symbol bit
469 N_PEXT = 0x10 // private external symbol bit
472 // Constants for the n_desc field
473 // see <mach-o/loader.h>
474 enum { REFERENCE_FLAG_UNDEFINED_NON_LAZY = 0,
475 REFERENCE_FLAG_UNDEFINED_LAZY = 1,
476 REFERENCE_FLAG_DEFINED = 2,
477 REFERENCE_FLAG_PRIVATE_DEFINED = 3,
478 REFERENCE_FLAG_PRIVATE_UNDEFINED_NON_LAZY = 4,
479 REFERENCE_FLAG_PRIVATE_UNDEFINED_LAZY = 5
481 enum { N_NO_DEAD_STRIP = 0x0020, // symbol is not to be dead stripped
482 N_WEAK_REF = 0x0040, // symbol is weak referenced
483 N_WEAK_DEF = 0x0080 // coalesced symbol is a weak definition
486 /// entrySize - This routine returns the size of a symbol table entry as
488 static unsigned entrySize() { return 12; }
490 MachOSym(const GlobalValue *gv, uint8_t sect) : GV(gv), n_strx(0),
491 n_type(N_UNDF), n_sect(sect), n_desc(0), n_value(0) {}
494 /// SymbolTable - This is the list of symbols we have emitted to the file.
495 /// This actually gets rearranged before emission to the file (to put the
496 /// local symbols first in the list).
497 std::vector<MachOSym> SymbolTable;
499 /// DynamicSymbolTable - This is just a vector of indices into
500 /// SymbolTable to aid in emitting the DYSYMTAB load command.
501 std::vector<unsigned> DynamicSymbolTable;
503 /// StringTable - The table of strings referenced by SymbolTable entries
504 std::vector<std::string> StringTable;
506 // align - Emit padding into the file until the current output position is
507 // aligned to the specified power of two boundary.
508 static void align(DataBuffer &Output, unsigned Boundary) {
509 assert(Boundary && (Boundary & (Boundary-1)) == 0 &&
510 "Must align to 2^k boundary");
511 size_t Size = Output.size();
512 if (Size & (Boundary-1)) {
513 // Add padding to get alignment to the correct place.
514 size_t Pad = Boundary-(Size & (Boundary-1));
515 Output.resize(Size+Pad);
519 void outbyte(DataBuffer &Output, unsigned char X) {
522 void outhalf(DataBuffer &Output, unsigned short X) {
523 if (isLittleEndian) {
524 Output.push_back(X&255);
525 Output.push_back(X >> 8);
527 Output.push_back(X >> 8);
528 Output.push_back(X&255);
531 void outword(DataBuffer &Output, unsigned X) {
532 if (isLittleEndian) {
533 Output.push_back((X >> 0) & 255);
534 Output.push_back((X >> 8) & 255);
535 Output.push_back((X >> 16) & 255);
536 Output.push_back((X >> 24) & 255);
538 Output.push_back((X >> 24) & 255);
539 Output.push_back((X >> 16) & 255);
540 Output.push_back((X >> 8) & 255);
541 Output.push_back((X >> 0) & 255);
544 void outxword(DataBuffer &Output, uint64_t X) {
545 if (isLittleEndian) {
546 Output.push_back(unsigned(X >> 0) & 255);
547 Output.push_back(unsigned(X >> 8) & 255);
548 Output.push_back(unsigned(X >> 16) & 255);
549 Output.push_back(unsigned(X >> 24) & 255);
550 Output.push_back(unsigned(X >> 32) & 255);
551 Output.push_back(unsigned(X >> 40) & 255);
552 Output.push_back(unsigned(X >> 48) & 255);
553 Output.push_back(unsigned(X >> 56) & 255);
555 Output.push_back(unsigned(X >> 56) & 255);
556 Output.push_back(unsigned(X >> 48) & 255);
557 Output.push_back(unsigned(X >> 40) & 255);
558 Output.push_back(unsigned(X >> 32) & 255);
559 Output.push_back(unsigned(X >> 24) & 255);
560 Output.push_back(unsigned(X >> 16) & 255);
561 Output.push_back(unsigned(X >> 8) & 255);
562 Output.push_back(unsigned(X >> 0) & 255);
565 void outaddr32(DataBuffer &Output, unsigned X) {
568 void outaddr64(DataBuffer &Output, uint64_t X) {
571 void outaddr(DataBuffer &Output, uint64_t X) {
573 outword(Output, (unsigned)X);
577 void outstring(DataBuffer &Output, std::string &S, unsigned Length) {
578 char *buffer = (char *)calloc(1, Length);
580 // FIXME: it is unclear if mach-o requires null terminated strings, or
581 // if a string of 16 bytes with no null terminator is ok. If so,
582 // we should switch to strncpy.
583 strlcpy(buffer, S.c_str(), Length);
585 for (i = 0; i < Length; ++i)
586 outbyte(Output, buffer[i]);
591 void EmitGlobal(GlobalVariable *GV);
592 void EmitHeaderAndLoadCommands();
594 void EmitRelocations();
595 void EmitSymbolTable();
596 void EmitStringTable();