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