1================================ 2Source Level Debugging with LLVM 3================================ 4 5.. contents:: 6 :local: 7 8Introduction 9============ 10 11This document is the central repository for all information pertaining to debug 12information in LLVM. It describes the :ref:`actual format that the LLVM debug 13information takes <format>`, which is useful for those interested in creating 14front-ends or dealing directly with the information. Further, this document 15provides specific examples of what debug information for C/C++ looks like. 16 17Philosophy behind LLVM debugging information 18-------------------------------------------- 19 20The idea of the LLVM debugging information is to capture how the important 21pieces of the source-language's Abstract Syntax Tree map onto LLVM code. 22Several design aspects have shaped the solution that appears here. The 23important ones are: 24 25* Debugging information should have very little impact on the rest of the 26 compiler. No transformations, analyses, or code generators should need to 27 be modified because of debugging information. 28 29* LLVM optimizations should interact in :ref:`well-defined and easily described 30 ways <intro_debugopt>` with the debugging information. 31 32* Because LLVM is designed to support arbitrary programming languages, 33 LLVM-to-LLVM tools should not need to know anything about the semantics of 34 the source-level-language. 35 36* Source-level languages are often **widely** different from one another. 37 LLVM should not put any restrictions of the flavor of the source-language, 38 and the debugging information should work with any language. 39 40* With code generator support, it should be possible to use an LLVM compiler 41 to compile a program to native machine code and standard debugging 42 formats. This allows compatibility with traditional machine-code level 43 debuggers, like GDB or DBX. 44 45The approach used by the LLVM implementation is to use a small set of 46:ref:`intrinsic functions <format_common_intrinsics>` to define a mapping 47between LLVM program objects and the source-level objects. The description of 48the source-level program is maintained in LLVM metadata in an 49:ref:`implementation-defined format <ccxx_frontend>` (the C/C++ front-end 50currently uses working draft 7 of the `DWARF 3 standard 51<http://www.eagercon.com/dwarf/dwarf3std.htm>`_). 52 53When a program is being debugged, a debugger interacts with the user and turns 54the stored debug information into source-language specific information. As 55such, a debugger must be aware of the source-language, and is thus tied to a 56specific language or family of languages. 57 58Debug information consumers 59--------------------------- 60 61The role of debug information is to provide meta information normally stripped 62away during the compilation process. This meta information provides an LLVM 63user a relationship between generated code and the original program source 64code. 65 66Currently, there are two backend consumers of debug info: DwarfDebug and 67CodeViewDebug. DwarfDebug produces DWARF suitable for use with GDB, LLDB, and 68other DWARF-based debuggers. :ref:`CodeViewDebug <codeview>` produces CodeView, 69the Microsoft debug info format, which is usable with Microsoft debuggers such 70as Visual Studio and WinDBG. LLVM's debug information format is mostly derived 71from and inspired by DWARF, but it is feasible to translate into other target 72debug info formats such as STABS. 73 74It would also be reasonable to use debug information to feed profiling tools 75for analysis of generated code, or, tools for reconstructing the original 76source from generated code. 77 78.. _intro_debugopt: 79 80Debug information and optimizations 81----------------------------------- 82 83An extremely high priority of LLVM debugging information is to make it interact 84well with optimizations and analysis. In particular, the LLVM debug 85information provides the following guarantees: 86 87* LLVM debug information **always provides information to accurately read 88 the source-level state of the program**, regardless of which LLVM 89 optimizations have been run. :doc:`HowToUpdateDebugInfo` specifies how debug 90 info should be updated in various kinds of code transformations to avoid 91 breaking this guarantee, and how to preserve as much useful debug info as 92 possible. Note that some optimizations may impact the ability to modify the 93 current state of the program with a debugger, such as setting program 94 variables, or calling functions that have been deleted. 95 96* As desired, LLVM optimizations can be upgraded to be aware of debugging 97 information, allowing them to update the debugging information as they 98 perform aggressive optimizations. This means that, with effort, the LLVM 99 optimizers could optimize debug code just as well as non-debug code. 100 101* LLVM debug information does not prevent optimizations from 102 happening (for example inlining, basic block reordering/merging/cleanup, 103 tail duplication, etc). 104 105* LLVM debug information is automatically optimized along with the rest of 106 the program, using existing facilities. For example, duplicate 107 information is automatically merged by the linker, and unused information 108 is automatically removed. 109 110Basically, the debug information allows you to compile a program with 111"``-O0 -g``" and get full debug information, allowing you to arbitrarily modify 112the program as it executes from a debugger. Compiling a program with 113"``-O3 -g``" gives you full debug information that is always available and 114accurate for reading (e.g., you get accurate stack traces despite tail call 115elimination and inlining), but you might lose the ability to modify the program 116and call functions which were optimized out of the program, or inlined away 117completely. 118 119The :doc:`LLVM test-suite <TestSuiteMakefileGuide>` provides a framework to 120test the optimizer's handling of debugging information. It can be run like 121this: 122 123.. code-block:: bash 124 125 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level 126 % make TEST=dbgopt 127 128This will test impact of debugging information on optimization passes. If 129debugging information influences optimization passes then it will be reported 130as a failure. See :doc:`TestingGuide` for more information on LLVM test 131infrastructure and how to run various tests. 132 133.. _format: 134 135Debugging information format 136============================ 137 138LLVM debugging information has been carefully designed to make it possible for 139the optimizer to optimize the program and debugging information without 140necessarily having to know anything about debugging information. In 141particular, the use of metadata avoids duplicated debugging information from 142the beginning, and the global dead code elimination pass automatically deletes 143debugging information for a function if it decides to delete the function. 144 145To do this, most of the debugging information (descriptors for types, 146variables, functions, source files, etc) is inserted by the language front-end 147in the form of LLVM metadata. 148 149Debug information is designed to be agnostic about the target debugger and 150debugging information representation (e.g. DWARF/Stabs/etc). It uses a generic 151pass to decode the information that represents variables, types, functions, 152namespaces, etc: this allows for arbitrary source-language semantics and 153type-systems to be used, as long as there is a module written for the target 154debugger to interpret the information. 155 156To provide basic functionality, the LLVM debugger does have to make some 157assumptions about the source-level language being debugged, though it keeps 158these to a minimum. The only common features that the LLVM debugger assumes 159exist are `source files <LangRef.html#difile>`_, and `program objects 160<LangRef.html#diglobalvariable>`_. These abstract objects are used by a 161debugger to form stack traces, show information about local variables, etc. 162 163This section of the documentation first describes the representation aspects 164common to any source-language. :ref:`ccxx_frontend` describes the data layout 165conventions used by the C and C++ front-ends. 166 167Debug information descriptors are `specialized metadata nodes 168<LangRef.html#specialized-metadata>`_, first-class subclasses of ``Metadata``. 169 170.. _format_common_intrinsics: 171 172Debugger intrinsic functions 173---------------------------- 174 175LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to 176track source local variables through optimization and code generation. 177 178``llvm.dbg.addr`` 179^^^^^^^^^^^^^^^^^^^^ 180 181.. code-block:: llvm 182 183 void @llvm.dbg.addr(metadata, metadata, metadata) 184 185This intrinsic provides information about a local element (e.g., variable). 186The first argument is metadata holding the address of variable, typically a 187static alloca in the function entry block. The second argument is a 188`local variable <LangRef.html#dilocalvariable>`_ containing a description of 189the variable. The third argument is a `complex expression 190<LangRef.html#diexpression>`_. An `llvm.dbg.addr` intrinsic describes the 191*address* of a source variable. 192 193.. code-block:: text 194 195 %i.addr = alloca i32, align 4 196 call void @llvm.dbg.addr(metadata i32* %i.addr, metadata !1, 197 metadata !DIExpression()), !dbg !2 198 !1 = !DILocalVariable(name: "i", ...) ; int i 199 !2 = !DILocation(...) 200 ... 201 %buffer = alloca [256 x i8], align 8 202 ; The address of i is buffer+64. 203 call void @llvm.dbg.addr(metadata [256 x i8]* %buffer, metadata !3, 204 metadata !DIExpression(DW_OP_plus, 64)), !dbg !4 205 !3 = !DILocalVariable(name: "i", ...) ; int i 206 !4 = !DILocation(...) 207 208A frontend should generate exactly one call to ``llvm.dbg.addr`` at the point 209of declaration of a source variable. Optimization passes that fully promote the 210variable from memory to SSA values will replace this call with possibly 211multiple calls to `llvm.dbg.value`. Passes that delete stores are effectively 212partial promotion, and they will insert a mix of calls to ``llvm.dbg.value`` 213and ``llvm.dbg.addr`` to track the source variable value when it is available. 214After optimization, there may be multiple calls to ``llvm.dbg.addr`` describing 215the program points where the variables lives in memory. All calls for the same 216concrete source variable must agree on the memory location. 217 218 219``llvm.dbg.declare`` 220^^^^^^^^^^^^^^^^^^^^ 221 222.. code-block:: llvm 223 224 void @llvm.dbg.declare(metadata, metadata, metadata) 225 226This intrinsic is identical to `llvm.dbg.addr`, except that there can only be 227one call to `llvm.dbg.declare` for a given concrete `local variable 228<LangRef.html#dilocalvariable>`_. It is not control-dependent, meaning that if 229a call to `llvm.dbg.declare` exists and has a valid location argument, that 230address is considered to be the true home of the variable across its entire 231lifetime. This makes it hard for optimizations to preserve accurate debug info 232in the presence of ``llvm.dbg.declare``, so we are transitioning away from it, 233and we plan to deprecate it in future LLVM releases. 234 235 236``llvm.dbg.value`` 237^^^^^^^^^^^^^^^^^^ 238 239.. code-block:: llvm 240 241 void @llvm.dbg.value(metadata, metadata, metadata) 242 243This intrinsic provides information when a user source variable is set to a new 244value. The first argument is the new value (wrapped as metadata). The second 245argument is a `local variable <LangRef.html#dilocalvariable>`_ containing a 246description of the variable. The third argument is a `complex expression 247<LangRef.html#diexpression>`_. 248 249An `llvm.dbg.value` intrinsic describes the *value* of a source variable 250directly, not its address. Note that the value operand of this intrinsic may 251be indirect (i.e, a pointer to the source variable), provided that interpreting 252the complex expression derives the direct value. 253 254Object lifetimes and scoping 255============================ 256 257In many languages, the local variables in functions can have their lifetimes or 258scopes limited to a subset of a function. In the C family of languages, for 259example, variables are only live (readable and writable) within the source 260block that they are defined in. In functional languages, values are only 261readable after they have been defined. Though this is a very obvious concept, 262it is non-trivial to model in LLVM, because it has no notion of scoping in this 263sense, and does not want to be tied to a language's scoping rules. 264 265In order to handle this, the LLVM debug format uses the metadata attached to 266llvm instructions to encode line number and scoping information. Consider the 267following C fragment, for example: 268 269.. code-block:: c 270 271 1. void foo() { 272 2. int X = 21; 273 3. int Y = 22; 274 4. { 275 5. int Z = 23; 276 6. Z = X; 277 7. } 278 8. X = Y; 279 9. } 280 281.. FIXME: Update the following example to use llvm.dbg.addr once that is the 282 default in clang. 283 284Compiled to LLVM, this function would be represented like this: 285 286.. code-block:: text 287 288 ; Function Attrs: nounwind ssp uwtable 289 define void @foo() #0 !dbg !4 { 290 entry: 291 %X = alloca i32, align 4 292 %Y = alloca i32, align 4 293 %Z = alloca i32, align 4 294 call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !13), !dbg !14 295 store i32 21, i32* %X, align 4, !dbg !14 296 call void @llvm.dbg.declare(metadata i32* %Y, metadata !15, metadata !13), !dbg !16 297 store i32 22, i32* %Y, align 4, !dbg !16 298 call void @llvm.dbg.declare(metadata i32* %Z, metadata !17, metadata !13), !dbg !19 299 store i32 23, i32* %Z, align 4, !dbg !19 300 %0 = load i32, i32* %X, align 4, !dbg !20 301 store i32 %0, i32* %Z, align 4, !dbg !21 302 %1 = load i32, i32* %Y, align 4, !dbg !22 303 store i32 %1, i32* %X, align 4, !dbg !23 304 ret void, !dbg !24 305 } 306 307 ; Function Attrs: nounwind readnone 308 declare void @llvm.dbg.declare(metadata, metadata, metadata) #1 309 310 attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } 311 attributes #1 = { nounwind readnone } 312 313 !llvm.dbg.cu = !{!0} 314 !llvm.module.flags = !{!7, !8, !9} 315 !llvm.ident = !{!10} 316 317 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !2, subprograms: !3, globals: !2, imports: !2) 318 !1 = !DIFile(filename: "/dev/stdin", directory: "/Users/dexonsmith/data/llvm/debug-info") 319 !2 = !{} 320 !3 = !{!4} 321 !4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5, isLocal: false, isDefinition: true, scopeLine: 1, isOptimized: false, variables: !2) 322 !5 = !DISubroutineType(types: !6) 323 !6 = !{null} 324 !7 = !{i32 2, !"Dwarf Version", i32 2} 325 !8 = !{i32 2, !"Debug Info Version", i32 3} 326 !9 = !{i32 1, !"PIC Level", i32 2} 327 !10 = !{!"clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)"} 328 !11 = !DILocalVariable(name: "X", scope: !4, file: !1, line: 2, type: !12) 329 !12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed) 330 !13 = !DIExpression() 331 !14 = !DILocation(line: 2, column: 9, scope: !4) 332 !15 = !DILocalVariable(name: "Y", scope: !4, file: !1, line: 3, type: !12) 333 !16 = !DILocation(line: 3, column: 9, scope: !4) 334 !17 = !DILocalVariable(name: "Z", scope: !18, file: !1, line: 5, type: !12) 335 !18 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5) 336 !19 = !DILocation(line: 5, column: 11, scope: !18) 337 !20 = !DILocation(line: 6, column: 11, scope: !18) 338 !21 = !DILocation(line: 6, column: 9, scope: !18) 339 !22 = !DILocation(line: 8, column: 9, scope: !4) 340 !23 = !DILocation(line: 8, column: 7, scope: !4) 341 !24 = !DILocation(line: 9, column: 3, scope: !4) 342 343 344This example illustrates a few important details about LLVM debugging 345information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and 346location information, which are attached to an instruction, are applied 347together to allow a debugger to analyze the relationship between statements, 348variable definitions, and the code used to implement the function. 349 350.. code-block:: llvm 351 352 call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !13), !dbg !14 353 ; [debug line = 2:7] [debug variable = X] 354 355The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the 356variable ``X``. The metadata ``!dbg !14`` attached to the intrinsic provides 357scope information for the variable ``X``. 358 359.. code-block:: text 360 361 !14 = !DILocation(line: 2, column: 9, scope: !4) 362 !4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5, 363 isLocal: false, isDefinition: true, scopeLine: 1, 364 isOptimized: false, variables: !2) 365 366Here ``!14`` is metadata providing `location information 367<LangRef.html#dilocation>`_. In this example, scope is encoded by ``!4``, a 368`subprogram descriptor <LangRef.html#disubprogram>`_. This way the location 369information attached to the intrinsics indicates that the variable ``X`` is 370declared at line number 2 at a function level scope in function ``foo``. 371 372Now lets take another example. 373 374.. code-block:: llvm 375 376 call void @llvm.dbg.declare(metadata i32* %Z, metadata !17, metadata !13), !dbg !19 377 ; [debug line = 5:9] [debug variable = Z] 378 379The third intrinsic ``%llvm.dbg.declare`` encodes debugging information for 380variable ``Z``. The metadata ``!dbg !19`` attached to the intrinsic provides 381scope information for the variable ``Z``. 382 383.. code-block:: text 384 385 !18 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5) 386 !19 = !DILocation(line: 5, column: 11, scope: !18) 387 388Here ``!19`` indicates that ``Z`` is declared at line number 5 and column 389number 11 inside of lexical scope ``!18``. The lexical scope itself resides 390inside of subprogram ``!4`` described above. 391 392The scope information attached with each instruction provides a straightforward 393way to find instructions covered by a scope. 394 395Object lifetime in optimized code 396================================= 397 398In the example above, every variable assignment uniquely corresponds to a 399memory store to the variable's position on the stack. However in heavily 400optimized code LLVM promotes most variables into SSA values, which can 401eventually be placed in physical registers or memory locations. To track SSA 402values through compilation, when objects are promoted to SSA values an 403``llvm.dbg.value`` intrinsic is created for each assignment, recording the 404variable's new location. Compared with the ``llvm.dbg.declare`` intrinsic: 405 406* A dbg.value terminates the effect of any preceding dbg.values for (any 407 overlapping fragments of) the specified variable. 408* The dbg.value's position in the IR defines where in the instruction stream 409 the variable's value changes. 410* Operands can be constants, indicating the variable is assigned a 411 constant value. 412 413Care must be taken to update ``llvm.dbg.value`` intrinsics when optimization 414passes alter or move instructions and blocks -- the developer could observe such 415changes reflected in the value of variables when debugging the program. For any 416execution of the optimized program, the set of variable values presented to the 417developer by the debugger should not show a state that would never have existed 418in the execution of the unoptimized program, given the same input. Doing so 419risks misleading the developer by reporting a state that does not exist, 420damaging their understanding of the optimized program and undermining their 421trust in the debugger. 422 423Sometimes perfectly preserving variable locations is not possible, often when a 424redundant calculation is optimized out. In such cases, a ``llvm.dbg.value`` 425with operand ``undef`` should be used, to terminate earlier variable locations 426and let the debugger present ``optimized out`` to the developer. Withholding 427these potentially stale variable values from the developer diminishes the 428amount of available debug information, but increases the reliability of the 429remaining information. 430 431To illustrate some potential issues, consider the following example: 432 433.. code-block:: llvm 434 435 define i32 @foo(i32 %bar, i1 %cond) { 436 entry: 437 call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2) 438 br i1 %cond, label %truebr, label %falsebr 439 truebr: 440 %tval = add i32 %bar, 1 441 call @llvm.dbg.value(metadata i32 %tval, metadata !1, metadata !2) 442 %g1 = call i32 @gazonk() 443 br label %exit 444 falsebr: 445 %fval = add i32 %bar, 2 446 call @llvm.dbg.value(metadata i32 %fval, metadata !1, metadata !2) 447 %g2 = call i32 @gazonk() 448 br label %exit 449 exit: 450 %merge = phi [ %tval, %truebr ], [ %fval, %falsebr ] 451 %g = phi [ %g1, %truebr ], [ %g2, %falsebr ] 452 call @llvm.dbg.value(metadata i32 %merge, metadata !1, metadata !2) 453 call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2) 454 %plusten = add i32 %merge, 10 455 %toret = add i32 %plusten, %g 456 call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2) 457 ret i32 %toret 458 } 459 460Containing two source-level variables in ``!1`` and ``!3``. The function could, 461perhaps, be optimized into the following code: 462 463.. code-block:: llvm 464 465 define i32 @foo(i32 %bar, i1 %cond) { 466 entry: 467 %g = call i32 @gazonk() 468 %addoper = select i1 %cond, i32 11, i32 12 469 %plusten = add i32 %bar, %addoper 470 %toret = add i32 %plusten, %g 471 ret i32 %toret 472 } 473 474What ``llvm.dbg.value`` intrinsics should be placed to represent the original variable 475locations in this code? Unfortunately the second, third and fourth 476dbg.values for ``!1`` in the source function have had their operands 477(%tval, %fval, %merge) optimized out. Assuming we cannot recover them, we 478might consider this placement of dbg.values: 479 480.. code-block:: llvm 481 482 define i32 @foo(i32 %bar, i1 %cond) { 483 entry: 484 call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2) 485 %g = call i32 @gazonk() 486 call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2) 487 %addoper = select i1 %cond, i32 11, i32 12 488 %plusten = add i32 %bar, %addoper 489 %toret = add i32 %plusten, %g 490 call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2) 491 ret i32 %toret 492 } 493 494However, this will cause ``!3`` to have the return value of ``@gazonk()`` at 495the same time as ``!1`` has the constant value zero -- a pair of assignments 496that never occurred in the unoptimized program. To avoid this, we must terminate 497the range that ``!1`` has the constant value assignment by inserting an undef 498dbg.value before the dbg.value for ``!3``: 499 500.. code-block:: llvm 501 502 define i32 @foo(i32 %bar, i1 %cond) { 503 entry: 504 call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2) 505 %g = call i32 @gazonk() 506 call @llvm.dbg.value(metadata i32 undef, metadata !1, metadata !2) 507 call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2) 508 %addoper = select i1 %cond, i32 11, i32 12 509 %plusten = add i32 %bar, %addoper 510 %toret = add i32 %plusten, %g 511 call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2) 512 ret i32 %toret 513 } 514 515In general, if any dbg.value has its operand optimized out and cannot be 516recovered, then an undef dbg.value is necessary to terminate earlier variable 517locations. Additional undef dbg.values may be necessary when the debugger can 518observe re-ordering of assignments. 519 520How variable location metadata is transformed during CodeGen 521============================================================ 522 523LLVM preserves debug information throughout mid-level and backend passes, 524ultimately producing a mapping between source-level information and 525instruction ranges. This 526is relatively straightforwards for line number information, as mapping 527instructions to line numbers is a simple association. For variable locations 528however the story is more complex. As each ``llvm.dbg.value`` intrinsic 529represents a source-level assignment of a value to a source variable, the 530variable location intrinsics effectively embed a small imperative program 531within the LLVM IR. By the end of CodeGen, this becomes a mapping from each 532variable to their machine locations over ranges of instructions. 533From IR to object emission, the major transformations which affect variable 534location fidelity are: 535 5361. Instruction Selection 5372. Register allocation 5383. Block layout 539 540each of which are discussed below. In addition, instruction scheduling can 541significantly change the ordering of the program, and occurs in a number of 542different passes. 543 544Some variable locations are not transformed during CodeGen. Stack locations 545specified by ``llvm.dbg.declare`` are valid and unchanging for the entire 546duration of the function, and are recorded in a simple MachineFunction table. 547Location changes in the prologue and epilogue of a function are also ignored: 548frame setup and destruction may take several instructions, require a 549disproportionate amount of debugging information in the output binary to 550describe, and should be stepped over by debuggers anyway. 551 552Variable locations in Instruction Selection and MIR 553--------------------------------------------------- 554 555Instruction selection creates a MIR function from an IR function, and just as 556it transforms ``intermediate`` instructions into machine instructions, so must 557``intermediate`` variable locations become machine variable locations. 558Within IR, variable locations are always identified by a Value, but in MIR 559there can be different types of variable locations. In addition, some IR 560locations become unavailable, for example if the operation of multiple IR 561instructions are combined into one machine instruction (such as 562multiply-and-accumulate) then intermediate Values are lost. To track variable 563locations through instruction selection, they are first separated into 564locations that do not depend on code generation (constants, stack locations, 565allocated virtual registers) and those that do. For those that do, debug 566metadata is attached to SDNodes in SelectionDAGs. After instruction selection 567has occurred and a MIR function is created, if the SDNode associated with debug 568metadata is allocated a virtual register, that virtual register is used as the 569variable location. If the SDNode is folded into a machine instruction or 570otherwise transformed into a non-register, the variable location becomes 571unavailable. 572 573Locations that are unavailable are treated as if they have been optimized out: 574in IR the location would be assigned ``undef`` by a debug intrinsic, and in MIR 575the equivalent location is used. 576 577After MIR locations are assigned to each variable, machine pseudo-instructions 578corresponding to each ``llvm.dbg.value`` and ``llvm.dbg.addr`` intrinsic are 579inserted. These ``DBG_VALUE`` instructions appear thus: 580 581.. code-block:: text 582 583 DBG_VALUE %1, $noreg, !123, !DIExpression() 584 585And have the following operands: 586 * The first operand can record the variable location as a register, 587 a frame index, an immediate, or the base address register if the original 588 debug intrinsic referred to memory. ``$noreg`` indicates the variable 589 location is undefined, equivalent to an ``undef`` dbg.value operand. 590 * The type of the second operand indicates whether the variable location is 591 directly referred to by the DBG_VALUE, or whether it is indirect. The 592 ``$noreg`` register signifies the former, an immediate operand (0) the 593 latter. 594 * Operand 3 is the Variable field of the original debug intrinsic. 595 * Operand 4 is the Expression field of the original debug intrinsic. 596 597The position at which the DBG_VALUEs are inserted should correspond to the 598positions of their matching ``llvm.dbg.value`` intrinsics in the IR block. As 599with optimization, LLVM aims to preserve the order in which variable 600assignments occurred in the source program. However SelectionDAG performs some 601instruction scheduling, which can reorder assignments (discussed below). 602Function parameter locations are moved to the beginning of the function if 603they're not already, to ensure they're immediately available on function entry. 604 605To demonstrate variable locations during instruction selection, consider 606the following example: 607 608.. code-block:: llvm 609 610 define i32 @foo(i32* %addr) { 611 entry: 612 call void @llvm.dbg.value(metadata i32 0, metadata !3, metadata !DIExpression()), !dbg !5 613 br label %bb1, !dbg !5 614 615 bb1: ; preds = %bb1, %entry 616 %bar.0 = phi i32 [ 0, %entry ], [ %add, %bb1 ] 617 call void @llvm.dbg.value(metadata i32 %bar.0, metadata !3, metadata !DIExpression()), !dbg !5 618 %addr1 = getelementptr i32, i32 *%addr, i32 1, !dbg !5 619 call void @llvm.dbg.value(metadata i32 *%addr1, metadata !3, metadata !DIExpression()), !dbg !5 620 %loaded1 = load i32, i32* %addr1, !dbg !5 621 %addr2 = getelementptr i32, i32 *%addr, i32 %bar.0, !dbg !5 622 call void @llvm.dbg.value(metadata i32 *%addr2, metadata !3, metadata !DIExpression()), !dbg !5 623 %loaded2 = load i32, i32* %addr2, !dbg !5 624 %add = add i32 %bar.0, 1, !dbg !5 625 call void @llvm.dbg.value(metadata i32 %add, metadata !3, metadata !DIExpression()), !dbg !5 626 %added = add i32 %loaded1, %loaded2 627 %cond = icmp ult i32 %added, %bar.0, !dbg !5 628 br i1 %cond, label %bb1, label %bb2, !dbg !5 629 630 bb2: ; preds = %bb1 631 ret i32 0, !dbg !5 632 } 633 634If one compiles this IR with ``llc -o - -start-after=codegen-prepare -stop-after=expand-isel-pseudos -mtriple=x86_64--``, the following MIR is produced: 635 636.. code-block:: text 637 638 bb.0.entry: 639 successors: %bb.1(0x80000000) 640 liveins: $rdi 641 642 %2:gr64 = COPY $rdi 643 %3:gr32 = MOV32r0 implicit-def dead $eflags 644 DBG_VALUE 0, $noreg, !3, !DIExpression(), debug-location !5 645 646 bb.1.bb1: 647 successors: %bb.1(0x7c000000), %bb.2(0x04000000) 648 649 %0:gr32 = PHI %3, %bb.0, %1, %bb.1 650 DBG_VALUE %0, $noreg, !3, !DIExpression(), debug-location !5 651 DBG_VALUE %2, $noreg, !3, !DIExpression(DW_OP_plus_uconst, 4, DW_OP_stack_value), debug-location !5 652 %4:gr32 = MOV32rm %2, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1) 653 %5:gr64_nosp = MOVSX64rr32 %0, debug-location !5 654 DBG_VALUE $noreg, $noreg, !3, !DIExpression(), debug-location !5 655 %1:gr32 = INC32r %0, implicit-def dead $eflags, debug-location !5 656 DBG_VALUE %1, $noreg, !3, !DIExpression(), debug-location !5 657 %6:gr32 = ADD32rm %4, %2, 4, killed %5, 0, $noreg, implicit-def dead $eflags :: (load 4 from %ir.addr2) 658 %7:gr32 = SUB32rr %6, %0, implicit-def $eflags, debug-location !5 659 JB_1 %bb.1, implicit $eflags, debug-location !5 660 JMP_1 %bb.2, debug-location !5 661 662 bb.2.bb2: 663 %8:gr32 = MOV32r0 implicit-def dead $eflags 664 $eax = COPY %8, debug-location !5 665 RET 0, $eax, debug-location !5 666 667Observe first that there is a DBG_VALUE instruction for every ``llvm.dbg.value`` 668intrinsic in the source IR, ensuring no source level assignments go missing. 669Then consider the different ways in which variable locations have been recorded: 670 671* For the first dbg.value an immediate operand is used to record a zero value. 672* The dbg.value of the PHI instruction leads to a DBG_VALUE of virtual register 673 ``%0``. 674* The first GEP has its effect folded into the first load instruction 675 (as a 4-byte offset), but the variable location is salvaged by folding 676 the GEPs effect into the DIExpression. 677* The second GEP is also folded into the corresponding load. However, it is 678 insufficiently simple to be salvaged, and is emitted as a ``$noreg`` 679 DBG_VALUE, indicating that the variable takes on an undefined location. 680* The final dbg.value has its Value placed in virtual register ``%1``. 681 682Instruction Scheduling 683---------------------- 684 685A number of passes can reschedule instructions, notably instruction selection 686and the pre-and-post RA machine schedulers. Instruction scheduling can 687significantly change the nature of the program -- in the (very unlikely) worst 688case the instruction sequence could be completely reversed. In such 689circumstances LLVM follows the principle applied to optimizations, that it is 690better for the debugger not to display any state than a misleading state. 691Thus, whenever instructions are advanced in order of execution, any 692corresponding DBG_VALUE is kept in its original position, and if an instruction 693is delayed then the variable is given an undefined location for the duration 694of the delay. To illustrate, consider this pseudo-MIR: 695 696.. code-block:: text 697 698 %1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1) 699 DBG_VALUE %1, $noreg, !1, !2 700 %4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags 701 DBG_VALUE %4, $noreg, !3, !4 702 %7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags 703 DBG_VALUE %7, $noreg, !5, !6 704 705Imagine that the SUB32rr were moved forward to give us the following MIR: 706 707.. code-block:: text 708 709 %7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags 710 %1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1) 711 DBG_VALUE %1, $noreg, !1, !2 712 %4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags 713 DBG_VALUE %4, $noreg, !3, !4 714 DBG_VALUE %7, $noreg, !5, !6 715 716In this circumstance LLVM would leave the MIR as shown above. Were we to move 717the DBG_VALUE of virtual register %7 upwards with the SUB32rr, we would re-order 718assignments and introduce a new state of the program. Whereas with the solution 719above, the debugger will see one fewer combination of variable values, because 720``!3`` and ``!5`` will change value at the same time. This is preferred over 721misrepresenting the original program. 722 723In comparison, if one sunk the MOV32rm, LLVM would produce the following: 724 725.. code-block:: text 726 727 DBG_VALUE $noreg, $noreg, !1, !2 728 %4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags 729 DBG_VALUE %4, $noreg, !3, !4 730 %7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags 731 DBG_VALUE %7, $noreg, !5, !6 732 %1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1) 733 DBG_VALUE %1, $noreg, !1, !2 734 735Here, to avoid presenting a state in which the first assignment to ``!1`` 736disappears, the DBG_VALUE at the top of the block assigns the variable the 737undefined location, until its value is available at the end of the block where 738an additional DBG_VALUE is added. Were any other DBG_VALUE for ``!1`` to occur 739in the instructions that the MOV32rm was sunk past, the DBG_VALUE for ``%1`` 740would be dropped and the debugger would never observe it in the variable. This 741accurately reflects that the value is not available during the corresponding 742portion of the original program. 743 744Variable locations during Register Allocation 745--------------------------------------------- 746 747To avoid debug instructions interfering with the register allocator, the 748LiveDebugVariables pass extracts variable locations from a MIR function and 749deletes the corresponding DBG_VALUE instructions. Some localized copy 750propagation is performed within blocks. After register allocation, the 751VirtRegRewriter pass re-inserts DBG_VALUE instructions in their original 752positions, translating virtual register references into their physical 753machine locations. To avoid encoding incorrect variable locations, in this 754pass any DBG_VALUE of a virtual register that is not live, is replaced by 755the undefined location. 756 757LiveDebugValues expansion of variable locations 758----------------------------------------------- 759 760After all optimizations have run and shortly before emission, the 761LiveDebugValues pass runs to achieve two aims: 762 763* To propagate the location of variables through copies and register spills, 764* For every block, to record every valid variable location in that block. 765 766After this pass the DBG_VALUE instruction changes meaning: rather than 767corresponding to a source-level assignment where the variable may change value, 768it asserts the location of a variable in a block, and loses effect outside the 769block. Propagating variable locations through copies and spills is 770straightforwards: determining the variable location in every basic block 771requires the consideration of control flow. Consider the following IR, which 772presents several difficulties: 773 774.. code-block:: text 775 776 define dso_local i32 @foo(i1 %cond, i32 %input) !dbg !12 { 777 entry: 778 br i1 %cond, label %truebr, label %falsebr 779 780 bb1: 781 %value = phi i32 [ %value1, %truebr ], [ %value2, %falsebr ] 782 br label %exit, !dbg !26 783 784 truebr: 785 call void @llvm.dbg.value(metadata i32 %input, metadata !30, metadata !DIExpression()), !dbg !24 786 call void @llvm.dbg.value(metadata i32 1, metadata !23, metadata !DIExpression()), !dbg !24 787 %value1 = add i32 %input, 1 788 br label %bb1 789 790 falsebr: 791 call void @llvm.dbg.value(metadata i32 %input, metadata !30, metadata !DIExpression()), !dbg !24 792 call void @llvm.dbg.value(metadata i32 2, metadata !23, metadata !DIExpression()), !dbg !24 793 %value = add i32 %input, 2 794 br label %bb1 795 796 exit: 797 ret i32 %value, !dbg !30 798 } 799 800Here the difficulties are: 801 802* The control flow is roughly the opposite of basic block order 803* The value of the ``!23`` variable merges into ``%bb1``, but there is no PHI 804 node 805 806As mentioned above, the ``llvm.dbg.value`` intrinsics essentially form an 807imperative program embedded in the IR, with each intrinsic defining a variable 808location. This *could* be converted to an SSA form by mem2reg, in the same way 809that it uses use-def chains to identify control flow merges and insert phi 810nodes for IR Values. However, because debug variable locations are defined for 811every machine instruction, in effect every IR instruction uses every variable 812location, which would lead to a large number of debugging intrinsics being 813generated. 814 815Examining the example above, variable ``!30`` is assigned ``%input`` on both 816conditional paths through the function, while ``!23`` is assigned differing 817constant values on either path. Where control flow merges in ``%bb1`` we would 818want ``!30`` to keep its location (``%input``), but ``!23`` to become undefined 819as we cannot determine at runtime what value it should have in %bb1 without 820inserting a PHI node. mem2reg does not insert the PHI node to avoid changing 821codegen when debugging is enabled, and does not insert the other dbg.values 822to avoid adding very large numbers of intrinsics. 823 824Instead, LiveDebugValues determines variable locations when control 825flow merges. A dataflow analysis is used to propagate locations between blocks: 826when control flow merges, if a variable has the same location in all 827predecessors then that location is propagated into the successor. If the 828predecessor locations disagree, the location becomes undefined. 829 830Once LiveDebugValues has run, every block should have all valid variable 831locations described by DBG_VALUE instructions within the block. Very little 832effort is then required by supporting classes (such as 833DbgEntityHistoryCalculator) to build a map of each instruction to every 834valid variable location, without the need to consider control flow. From 835the example above, it is otherwise difficult to determine that the location 836of variable ``!30`` should flow "up" into block ``%bb1``, but that the location 837of variable ``!23`` should not flow "down" into the ``%exit`` block. 838 839.. _ccxx_frontend: 840 841C/C++ front-end specific debug information 842========================================== 843 844The C and C++ front-ends represent information about the program in a 845format that is effectively identical to `DWARF <http://www.dwarfstd.org/>`_ 846in terms of information content. This allows code generators to 847trivially support native debuggers by generating standard dwarf 848information, and contains enough information for non-dwarf targets to 849translate it as needed. 850 851This section describes the forms used to represent C and C++ programs. Other 852languages could pattern themselves after this (which itself is tuned to 853representing programs in the same way that DWARF does), or they could choose 854to provide completely different forms if they don't fit into the DWARF model. 855As support for debugging information gets added to the various LLVM 856source-language front-ends, the information used should be documented here. 857 858The following sections provide examples of a few C/C++ constructs and 859the debug information that would best describe those constructs. The 860canonical references are the ``DINode`` classes defined in 861``include/llvm/IR/DebugInfoMetadata.h`` and the implementations of the 862helper functions in ``lib/IR/DIBuilder.cpp``. 863 864C/C++ source file information 865----------------------------- 866 867``llvm::Instruction`` provides easy access to metadata attached with an 868instruction. One can extract line number information encoded in LLVM IR using 869``Instruction::getDebugLoc()`` and ``DILocation::getLine()``. 870 871.. code-block:: c++ 872 873 if (DILocation *Loc = I->getDebugLoc()) { // Here I is an LLVM instruction 874 unsigned Line = Loc->getLine(); 875 StringRef File = Loc->getFilename(); 876 StringRef Dir = Loc->getDirectory(); 877 bool ImplicitCode = Loc->isImplicitCode(); 878 } 879 880When the flag ImplicitCode is true then it means that the Instruction has been 881added by the front-end but doesn't correspond to source code written by the user. For example 882 883.. code-block:: c++ 884 885 if (MyBoolean) { 886 MyObject MO; 887 ... 888 } 889 890At the end of the scope the MyObject's destructor is called but it isn't written 891explicitly. This information is useful to avoid to have counters on brackets when 892making code coverage. 893 894C/C++ global variable information 895--------------------------------- 896 897Given an integer global variable declared as follows: 898 899.. code-block:: c 900 901 _Alignas(8) int MyGlobal = 100; 902 903a C/C++ front-end would generate the following descriptors: 904 905.. code-block:: text 906 907 ;; 908 ;; Define the global itself. 909 ;; 910 @MyGlobal = global i32 100, align 8, !dbg !0 911 912 ;; 913 ;; List of debug info of globals 914 ;; 915 !llvm.dbg.cu = !{!1} 916 917 ;; Some unrelated metadata. 918 !llvm.module.flags = !{!6, !7} 919 !llvm.ident = !{!8} 920 921 ;; Define the global variable itself 922 !0 = distinct !DIGlobalVariable(name: "MyGlobal", scope: !1, file: !2, line: 1, type: !5, isLocal: false, isDefinition: true, align: 64) 923 924 ;; Define the compile unit. 925 !1 = distinct !DICompileUnit(language: DW_LANG_C99, file: !2, 926 producer: "clang version 4.0.0", 927 isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, 928 enums: !3, globals: !4) 929 930 ;; 931 ;; Define the file 932 ;; 933 !2 = !DIFile(filename: "/dev/stdin", 934 directory: "/Users/dexonsmith/data/llvm/debug-info") 935 936 ;; An empty array. 937 !3 = !{} 938 939 ;; The Array of Global Variables 940 !4 = !{!0} 941 942 ;; 943 ;; Define the type 944 ;; 945 !5 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) 946 947 ;; Dwarf version to output. 948 !6 = !{i32 2, !"Dwarf Version", i32 4} 949 950 ;; Debug info schema version. 951 !7 = !{i32 2, !"Debug Info Version", i32 3} 952 953 ;; Compiler identification 954 !8 = !{!"clang version 4.0.0"} 955 956 957The align value in DIGlobalVariable description specifies variable alignment in 958case it was forced by C11 _Alignas(), C++11 alignas() keywords or compiler 959attribute __attribute__((aligned ())). In other case (when this field is missing) 960alignment is considered default. This is used when producing DWARF output 961for DW_AT_alignment value. 962 963C/C++ function information 964-------------------------- 965 966Given a function declared as follows: 967 968.. code-block:: c 969 970 int main(int argc, char *argv[]) { 971 return 0; 972 } 973 974a C/C++ front-end would generate the following descriptors: 975 976.. code-block:: text 977 978 ;; 979 ;; Define the anchor for subprograms. 980 ;; 981 !4 = !DISubprogram(name: "main", scope: !1, file: !1, line: 1, type: !5, 982 isLocal: false, isDefinition: true, scopeLine: 1, 983 flags: DIFlagPrototyped, isOptimized: false, 984 variables: !2) 985 986 ;; 987 ;; Define the subprogram itself. 988 ;; 989 define i32 @main(i32 %argc, i8** %argv) !dbg !4 { 990 ... 991 } 992 993C++ specific debug information 994============================== 995 996C++ special member functions information 997---------------------------------------- 998 999DWARF v5 introduces attributes defined to enhance debugging information of C++ programs. LLVM can generate (or omit) these appropriate DWARF attributes. In C++ a special member function Ctors, Dtors, Copy/Move Ctors, assignment operators can be declared with C++11 keyword deleted. This is represented in LLVM using spFlags value DISPFlagDeleted. 1000 1001Given a class declaration with copy constructor declared as deleted: 1002 1003.. code-block:: c 1004 1005 class foo { 1006 public: 1007 foo(const foo&) = deleted; 1008 }; 1009 1010A C++ frontend would generate following: 1011 1012.. code-block:: text 1013 1014 !17 = !DISubprogram(name: "foo", scope: !11, file: !1, line: 5, type: !18, scopeLine: 5, flags: DIFlagPublic | DIFlagPrototyped, spFlags: DISPFlagDeleted) 1015 1016and this will produce an additional DWARF attribute as: 1017 1018.. code-block:: text 1019 1020 DW_TAG_subprogram [7] * 1021 DW_AT_name [DW_FORM_strx1] (indexed (00000006) string = "foo") 1022 DW_AT_decl_line [DW_FORM_data1] (5) 1023 ... 1024 DW_AT_deleted [DW_FORM_flag_present] (true) 1025 1026Fortran specific debug information 1027================================== 1028 1029Fortran function information 1030---------------------------- 1031 1032There are a few DWARF attributes defined to support client debugging of Fortran programs. LLVM can generate (or omit) the appropriate DWARF attributes for the prefix-specs of ELEMENTAL, PURE, IMPURE, RECURSIVE, and NON_RECURSIVE. This is done by using the spFlags values: DISPFlagElemental, DISPFlagPure, and DISPFlagRecursive. 1033 1034.. code-block:: fortran 1035 1036 elemental function elem_func(a) 1037 1038a Fortran front-end would generate the following descriptors: 1039 1040.. code-block:: text 1041 1042 !11 = distinct !DISubprogram(name: "subroutine2", scope: !1, file: !1, 1043 line: 5, type: !8, scopeLine: 6, 1044 spFlags: DISPFlagDefinition | DISPFlagElemental, unit: !0, 1045 retainedNodes: !2) 1046 1047and this will materialize an additional DWARF attribute as: 1048 1049.. code-block:: text 1050 1051 DW_TAG_subprogram [3] 1052 DW_AT_low_pc [DW_FORM_addr] (0x0000000000000010 ".text") 1053 DW_AT_high_pc [DW_FORM_data4] (0x00000001) 1054 ... 1055 DW_AT_elemental [DW_FORM_flag_present] (true) 1056 1057There are a few DWARF tags defined to represent Fortran specific constructs i.e DW_TAG_string_type for representing Fortran character(n). In LLVM this is represented as DIStringType. 1058 1059.. code-block:: fortran 1060 1061 character(len=*), intent(in) :: string 1062 1063a Fortran front-end would generate the following descriptors: 1064 1065.. code-block:: text 1066 1067 !DILocalVariable(name: "string", arg: 1, scope: !10, file: !3, line: 4, type: !15) 1068 !DIStringType(name: "character(*)!2", stringLength: !16, stringLengthExpression: !DIExpression(), size: 32) 1069 1070and this will materialize in DWARF tags as: 1071 1072.. code-block:: text 1073 1074 DW_TAG_string_type 1075 DW_AT_name ("character(*)!2") 1076 DW_AT_string_length (0x00000064) 1077 0x00000064: DW_TAG_variable 1078 DW_AT_location (DW_OP_fbreg +16) 1079 DW_AT_type (0x00000083 "integer*8") 1080 ... 1081 DW_AT_artificial (true) 1082 1083Debugging information format 1084============================ 1085 1086Debugging Information Extension for Objective C Properties 1087---------------------------------------------------------- 1088 1089Introduction 1090^^^^^^^^^^^^ 1091 1092Objective C provides a simpler way to declare and define accessor methods using 1093declared properties. The language provides features to declare a property and 1094to let compiler synthesize accessor methods. 1095 1096The debugger lets developer inspect Objective C interfaces and their instance 1097variables and class variables. However, the debugger does not know anything 1098about the properties defined in Objective C interfaces. The debugger consumes 1099information generated by compiler in DWARF format. The format does not support 1100encoding of Objective C properties. This proposal describes DWARF extensions to 1101encode Objective C properties, which the debugger can use to let developers 1102inspect Objective C properties. 1103 1104Proposal 1105^^^^^^^^ 1106 1107Objective C properties exist separately from class members. A property can be 1108defined only by "setter" and "getter" selectors, and be calculated anew on each 1109access. Or a property can just be a direct access to some declared ivar. 1110Finally it can have an ivar "automatically synthesized" for it by the compiler, 1111in which case the property can be referred to in user code directly using the 1112standard C dereference syntax as well as through the property "dot" syntax, but 1113there is no entry in the ``@interface`` declaration corresponding to this ivar. 1114 1115To facilitate debugging, these properties we will add a new DWARF TAG into the 1116``DW_TAG_structure_type`` definition for the class to hold the description of a 1117given property, and a set of DWARF attributes that provide said description. 1118The property tag will also contain the name and declared type of the property. 1119 1120If there is a related ivar, there will also be a DWARF property attribute placed 1121in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG 1122for that property. And in the case where the compiler synthesizes the ivar 1123directly, the compiler is expected to generate a ``DW_TAG_member`` for that 1124ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used 1125to access this ivar directly in code, and with the property attribute pointing 1126back to the property it is backing. 1127 1128The following examples will serve as illustration for our discussion: 1129 1130.. code-block:: objc 1131 1132 @interface I1 { 1133 int n2; 1134 } 1135 1136 @property int p1; 1137 @property int p2; 1138 @end 1139 1140 @implementation I1 1141 @synthesize p1; 1142 @synthesize p2 = n2; 1143 @end 1144 1145This produces the following DWARF (this is a "pseudo dwarfdump" output): 1146 1147.. code-block:: none 1148 1149 0x00000100: TAG_structure_type [7] * 1150 AT_APPLE_runtime_class( 0x10 ) 1151 AT_name( "I1" ) 1152 AT_decl_file( "Objc_Property.m" ) 1153 AT_decl_line( 3 ) 1154 1155 0x00000110 TAG_APPLE_property 1156 AT_name ( "p1" ) 1157 AT_type ( {0x00000150} ( int ) ) 1158 1159 0x00000120: TAG_APPLE_property 1160 AT_name ( "p2" ) 1161 AT_type ( {0x00000150} ( int ) ) 1162 1163 0x00000130: TAG_member [8] 1164 AT_name( "_p1" ) 1165 AT_APPLE_property ( {0x00000110} "p1" ) 1166 AT_type( {0x00000150} ( int ) ) 1167 AT_artificial ( 0x1 ) 1168 1169 0x00000140: TAG_member [8] 1170 AT_name( "n2" ) 1171 AT_APPLE_property ( {0x00000120} "p2" ) 1172 AT_type( {0x00000150} ( int ) ) 1173 1174 0x00000150: AT_type( ( int ) ) 1175 1176Note, the current convention is that the name of the ivar for an 1177auto-synthesized property is the name of the property from which it derives 1178with an underscore prepended, as is shown in the example. But we actually 1179don't need to know this convention, since we are given the name of the ivar 1180directly. 1181 1182Also, it is common practice in ObjC to have different property declarations in 1183the @interface and @implementation - e.g. to provide a read-only property in 1184the interface, and a read-write interface in the implementation. In that case, 1185the compiler should emit whichever property declaration will be in force in the 1186current translation unit. 1187 1188Developers can decorate a property with attributes which are encoded using 1189``DW_AT_APPLE_property_attribute``. 1190 1191.. code-block:: objc 1192 1193 @property (readonly, nonatomic) int pr; 1194 1195.. code-block:: none 1196 1197 TAG_APPLE_property [8] 1198 AT_name( "pr" ) 1199 AT_type ( {0x00000147} (int) ) 1200 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic) 1201 1202The setter and getter method names are attached to the property using 1203``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes. 1204 1205.. code-block:: objc 1206 1207 @interface I1 1208 @property (setter=myOwnP3Setter:) int p3; 1209 -(void)myOwnP3Setter:(int)a; 1210 @end 1211 1212 @implementation I1 1213 @synthesize p3; 1214 -(void)myOwnP3Setter:(int)a{ } 1215 @end 1216 1217The DWARF for this would be: 1218 1219.. code-block:: none 1220 1221 0x000003bd: TAG_structure_type [7] * 1222 AT_APPLE_runtime_class( 0x10 ) 1223 AT_name( "I1" ) 1224 AT_decl_file( "Objc_Property.m" ) 1225 AT_decl_line( 3 ) 1226 1227 0x000003cd TAG_APPLE_property 1228 AT_name ( "p3" ) 1229 AT_APPLE_property_setter ( "myOwnP3Setter:" ) 1230 AT_type( {0x00000147} ( int ) ) 1231 1232 0x000003f3: TAG_member [8] 1233 AT_name( "_p3" ) 1234 AT_type ( {0x00000147} ( int ) ) 1235 AT_APPLE_property ( {0x000003cd} ) 1236 AT_artificial ( 0x1 ) 1237 1238New DWARF Tags 1239^^^^^^^^^^^^^^ 1240 1241+-----------------------+--------+ 1242| TAG | Value | 1243+=======================+========+ 1244| DW_TAG_APPLE_property | 0x4200 | 1245+-----------------------+--------+ 1246 1247New DWARF Attributes 1248^^^^^^^^^^^^^^^^^^^^ 1249 1250+--------------------------------+--------+-----------+ 1251| Attribute | Value | Classes | 1252+================================+========+===========+ 1253| DW_AT_APPLE_property | 0x3fed | Reference | 1254+--------------------------------+--------+-----------+ 1255| DW_AT_APPLE_property_getter | 0x3fe9 | String | 1256+--------------------------------+--------+-----------+ 1257| DW_AT_APPLE_property_setter | 0x3fea | String | 1258+--------------------------------+--------+-----------+ 1259| DW_AT_APPLE_property_attribute | 0x3feb | Constant | 1260+--------------------------------+--------+-----------+ 1261 1262New DWARF Constants 1263^^^^^^^^^^^^^^^^^^^ 1264 1265+--------------------------------------+-------+ 1266| Name | Value | 1267+======================================+=======+ 1268| DW_APPLE_PROPERTY_readonly | 0x01 | 1269+--------------------------------------+-------+ 1270| DW_APPLE_PROPERTY_getter | 0x02 | 1271+--------------------------------------+-------+ 1272| DW_APPLE_PROPERTY_assign | 0x04 | 1273+--------------------------------------+-------+ 1274| DW_APPLE_PROPERTY_readwrite | 0x08 | 1275+--------------------------------------+-------+ 1276| DW_APPLE_PROPERTY_retain | 0x10 | 1277+--------------------------------------+-------+ 1278| DW_APPLE_PROPERTY_copy | 0x20 | 1279+--------------------------------------+-------+ 1280| DW_APPLE_PROPERTY_nonatomic | 0x40 | 1281+--------------------------------------+-------+ 1282| DW_APPLE_PROPERTY_setter | 0x80 | 1283+--------------------------------------+-------+ 1284| DW_APPLE_PROPERTY_atomic | 0x100 | 1285+--------------------------------------+-------+ 1286| DW_APPLE_PROPERTY_weak | 0x200 | 1287+--------------------------------------+-------+ 1288| DW_APPLE_PROPERTY_strong | 0x400 | 1289+--------------------------------------+-------+ 1290| DW_APPLE_PROPERTY_unsafe_unretained | 0x800 | 1291+--------------------------------------+-------+ 1292| DW_APPLE_PROPERTY_nullability | 0x1000| 1293+--------------------------------------+-------+ 1294| DW_APPLE_PROPERTY_null_resettable | 0x2000| 1295+--------------------------------------+-------+ 1296| DW_APPLE_PROPERTY_class | 0x4000| 1297+--------------------------------------+-------+ 1298 1299Name Accelerator Tables 1300----------------------- 1301 1302Introduction 1303^^^^^^^^^^^^ 1304 1305The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a 1306debugger needs. The "``pub``" in the section name indicates that the entries 1307in the table are publicly visible names only. This means no static or hidden 1308functions show up in the "``.debug_pubnames``". No static variables or private 1309class variables are in the "``.debug_pubtypes``". Many compilers add different 1310things to these tables, so we can't rely upon the contents between gcc, icc, or 1311clang. 1312 1313The typical query given by users tends not to match up with the contents of 1314these tables. For example, the DWARF spec states that "In the case of the name 1315of a function member or static data member of a C++ structure, class or union, 1316the name presented in the "``.debug_pubnames``" section is not the simple name 1317given by the ``DW_AT_name attribute`` of the referenced debugging information 1318entry, but rather the fully qualified name of the data or function member." 1319So the only names in these tables for complex C++ entries is a fully 1320qualified name. Debugger users tend not to enter their search strings as 1321"``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or 1322"``a::b::c``". So the name entered in the name table must be demangled in 1323order to chop it up appropriately and additional names must be manually entered 1324into the table to make it effective as a name lookup table for debuggers to 1325use. 1326 1327All debuggers currently ignore the "``.debug_pubnames``" table as a result of 1328its inconsistent and useless public-only name content making it a waste of 1329space in the object file. These tables, when they are written to disk, are not 1330sorted in any way, leaving every debugger to do its own parsing and sorting. 1331These tables also include an inlined copy of the string values in the table 1332itself making the tables much larger than they need to be on disk, especially 1333for large C++ programs. 1334 1335Can't we just fix the sections by adding all of the names we need to this 1336table? No, because that is not what the tables are defined to contain and we 1337won't know the difference between the old bad tables and the new good tables. 1338At best we could make our own renamed sections that contain all of the data we 1339need. 1340 1341These tables are also insufficient for what a debugger like LLDB needs. LLDB 1342uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then 1343often asked to look for type "``foo``" or namespace "``bar``", or list items in 1344namespace "``baz``". Namespaces are not included in the pubnames or pubtypes 1345tables. Since clang asks a lot of questions when it is parsing an expression, 1346we need to be very fast when looking up names, as it happens a lot. Having new 1347accelerator tables that are optimized for very quick lookups will benefit this 1348type of debugging experience greatly. 1349 1350We would like to generate name lookup tables that can be mapped into memory 1351from disk, and used as is, with little or no up-front parsing. We would also 1352be able to control the exact content of these different tables so they contain 1353exactly what we need. The Name Accelerator Tables were designed to fix these 1354issues. In order to solve these issues we need to: 1355 1356* Have a format that can be mapped into memory from disk and used as is 1357* Lookups should be very fast 1358* Extensible table format so these tables can be made by many producers 1359* Contain all of the names needed for typical lookups out of the box 1360* Strict rules for the contents of tables 1361 1362Table size is important and the accelerator table format should allow the reuse 1363of strings from common string tables so the strings for the names are not 1364duplicated. We also want to make sure the table is ready to be used as-is by 1365simply mapping the table into memory with minimal header parsing. 1366 1367The name lookups need to be fast and optimized for the kinds of lookups that 1368debuggers tend to do. Optimally we would like to touch as few parts of the 1369mapped table as possible when doing a name lookup and be able to quickly find 1370the name entry we are looking for, or discover there are no matches. In the 1371case of debuggers we optimized for lookups that fail most of the time. 1372 1373Each table that is defined should have strict rules on exactly what is in the 1374accelerator tables and documented so clients can rely on the content. 1375 1376Hash Tables 1377^^^^^^^^^^^ 1378 1379Standard Hash Tables 1380"""""""""""""""""""" 1381 1382Typical hash tables have a header, buckets, and each bucket points to the 1383bucket contents: 1384 1385.. code-block:: none 1386 1387 .------------. 1388 | HEADER | 1389 |------------| 1390 | BUCKETS | 1391 |------------| 1392 | DATA | 1393 `------------' 1394 1395The BUCKETS are an array of offsets to DATA for each hash: 1396 1397.. code-block:: none 1398 1399 .------------. 1400 | 0x00001000 | BUCKETS[0] 1401 | 0x00002000 | BUCKETS[1] 1402 | 0x00002200 | BUCKETS[2] 1403 | 0x000034f0 | BUCKETS[3] 1404 | | ... 1405 | 0xXXXXXXXX | BUCKETS[n_buckets] 1406 '------------' 1407 1408So for ``bucket[3]`` in the example above, we have an offset into the table 14090x000034f0 which points to a chain of entries for the bucket. Each bucket must 1410contain a next pointer, full 32 bit hash value, the string itself, and the data 1411for the current string value. 1412 1413.. code-block:: none 1414 1415 .------------. 1416 0x000034f0: | 0x00003500 | next pointer 1417 | 0x12345678 | 32 bit hash 1418 | "erase" | string value 1419 | data[n] | HashData for this bucket 1420 |------------| 1421 0x00003500: | 0x00003550 | next pointer 1422 | 0x29273623 | 32 bit hash 1423 | "dump" | string value 1424 | data[n] | HashData for this bucket 1425 |------------| 1426 0x00003550: | 0x00000000 | next pointer 1427 | 0x82638293 | 32 bit hash 1428 | "main" | string value 1429 | data[n] | HashData for this bucket 1430 `------------' 1431 1432The problem with this layout for debuggers is that we need to optimize for the 1433negative lookup case where the symbol we're searching for is not present. So 1434if we were to lookup "``printf``" in the table above, we would make a 32-bit 1435hash for "``printf``", it might match ``bucket[3]``. We would need to go to 1436the offset 0x000034f0 and start looking to see if our 32 bit hash matches. To 1437do so, we need to read the next pointer, then read the hash, compare it, and 1438skip to the next bucket. Each time we are skipping many bytes in memory and 1439touching new pages just to do the compare on the full 32 bit hash. All of 1440these accesses then tell us that we didn't have a match. 1441 1442Name Hash Tables 1443"""""""""""""""" 1444 1445To solve the issues mentioned above we have structured the hash tables a bit 1446differently: a header, buckets, an array of all unique 32 bit hash values, 1447followed by an array of hash value data offsets, one for each hash value, then 1448the data for all hash values: 1449 1450.. code-block:: none 1451 1452 .-------------. 1453 | HEADER | 1454 |-------------| 1455 | BUCKETS | 1456 |-------------| 1457 | HASHES | 1458 |-------------| 1459 | OFFSETS | 1460 |-------------| 1461 | DATA | 1462 `-------------' 1463 1464The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By 1465making all of the full 32 bit hash values contiguous in memory, we allow 1466ourselves to efficiently check for a match while touching as little memory as 1467possible. Most often checking the 32 bit hash values is as far as the lookup 1468goes. If it does match, it usually is a match with no collisions. So for a 1469table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash 1470values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and 1471``OFFSETS`` as: 1472 1473.. code-block:: none 1474 1475 .-------------------------. 1476 | HEADER.magic | uint32_t 1477 | HEADER.version | uint16_t 1478 | HEADER.hash_function | uint16_t 1479 | HEADER.bucket_count | uint32_t 1480 | HEADER.hashes_count | uint32_t 1481 | HEADER.header_data_len | uint32_t 1482 | HEADER_DATA | HeaderData 1483 |-------------------------| 1484 | BUCKETS | uint32_t[n_buckets] // 32 bit hash indexes 1485 |-------------------------| 1486 | HASHES | uint32_t[n_hashes] // 32 bit hash values 1487 |-------------------------| 1488 | OFFSETS | uint32_t[n_hashes] // 32 bit offsets to hash value data 1489 |-------------------------| 1490 | ALL HASH DATA | 1491 `-------------------------' 1492 1493So taking the exact same data from the standard hash example above we end up 1494with: 1495 1496.. code-block:: none 1497 1498 .------------. 1499 | HEADER | 1500 |------------| 1501 | 0 | BUCKETS[0] 1502 | 2 | BUCKETS[1] 1503 | 5 | BUCKETS[2] 1504 | 6 | BUCKETS[3] 1505 | | ... 1506 | ... | BUCKETS[n_buckets] 1507 |------------| 1508 | 0x........ | HASHES[0] 1509 | 0x........ | HASHES[1] 1510 | 0x........ | HASHES[2] 1511 | 0x........ | HASHES[3] 1512 | 0x........ | HASHES[4] 1513 | 0x........ | HASHES[5] 1514 | 0x12345678 | HASHES[6] hash for BUCKETS[3] 1515 | 0x29273623 | HASHES[7] hash for BUCKETS[3] 1516 | 0x82638293 | HASHES[8] hash for BUCKETS[3] 1517 | 0x........ | HASHES[9] 1518 | 0x........ | HASHES[10] 1519 | 0x........ | HASHES[11] 1520 | 0x........ | HASHES[12] 1521 | 0x........ | HASHES[13] 1522 | 0x........ | HASHES[n_hashes] 1523 |------------| 1524 | 0x........ | OFFSETS[0] 1525 | 0x........ | OFFSETS[1] 1526 | 0x........ | OFFSETS[2] 1527 | 0x........ | OFFSETS[3] 1528 | 0x........ | OFFSETS[4] 1529 | 0x........ | OFFSETS[5] 1530 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3] 1531 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3] 1532 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3] 1533 | 0x........ | OFFSETS[9] 1534 | 0x........ | OFFSETS[10] 1535 | 0x........ | OFFSETS[11] 1536 | 0x........ | OFFSETS[12] 1537 | 0x........ | OFFSETS[13] 1538 | 0x........ | OFFSETS[n_hashes] 1539 |------------| 1540 | | 1541 | | 1542 | | 1543 | | 1544 | | 1545 |------------| 1546 0x000034f0: | 0x00001203 | .debug_str ("erase") 1547 | 0x00000004 | A 32 bit array count - number of HashData with name "erase" 1548 | 0x........ | HashData[0] 1549 | 0x........ | HashData[1] 1550 | 0x........ | HashData[2] 1551 | 0x........ | HashData[3] 1552 | 0x00000000 | String offset into .debug_str (terminate data for hash) 1553 |------------| 1554 0x00003500: | 0x00001203 | String offset into .debug_str ("collision") 1555 | 0x00000002 | A 32 bit array count - number of HashData with name "collision" 1556 | 0x........ | HashData[0] 1557 | 0x........ | HashData[1] 1558 | 0x00001203 | String offset into .debug_str ("dump") 1559 | 0x00000003 | A 32 bit array count - number of HashData with name "dump" 1560 | 0x........ | HashData[0] 1561 | 0x........ | HashData[1] 1562 | 0x........ | HashData[2] 1563 | 0x00000000 | String offset into .debug_str (terminate data for hash) 1564 |------------| 1565 0x00003550: | 0x00001203 | String offset into .debug_str ("main") 1566 | 0x00000009 | A 32 bit array count - number of HashData with name "main" 1567 | 0x........ | HashData[0] 1568 | 0x........ | HashData[1] 1569 | 0x........ | HashData[2] 1570 | 0x........ | HashData[3] 1571 | 0x........ | HashData[4] 1572 | 0x........ | HashData[5] 1573 | 0x........ | HashData[6] 1574 | 0x........ | HashData[7] 1575 | 0x........ | HashData[8] 1576 | 0x00000000 | String offset into .debug_str (terminate data for hash) 1577 `------------' 1578 1579So we still have all of the same data, we just organize it more efficiently for 1580debugger lookup. If we repeat the same "``printf``" lookup from above, we 1581would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit 1582hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which 1583is the index into the ``HASHES`` table. We would then compare any consecutive 158432 bit hashes values in the ``HASHES`` array as long as the hashes would be in 1585``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo 1586``n_buckets`` is still 3. In the case of a failed lookup we would access the 1587memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes 1588before we know that we have no match. We don't end up marching through 1589multiple words of memory and we really keep the number of processor data cache 1590lines being accessed as small as possible. 1591 1592The string hash that is used for these lookup tables is the Daniel J. 1593Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a 1594very good hash for all kinds of names in programs with very few hash 1595collisions. 1596 1597Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``. 1598 1599Details 1600^^^^^^^ 1601 1602These name hash tables are designed to be generic where specializations of the 1603table get to define additional data that goes into the header ("``HeaderData``"), 1604how the string value is stored ("``KeyType``") and the content of the data for each 1605hash value. 1606 1607Header Layout 1608""""""""""""" 1609 1610The header has a fixed part, and the specialized part. The exact format of the 1611header is: 1612 1613.. code-block:: c 1614 1615 struct Header 1616 { 1617 uint32_t magic; // 'HASH' magic value to allow endian detection 1618 uint16_t version; // Version number 1619 uint16_t hash_function; // The hash function enumeration that was used 1620 uint32_t bucket_count; // The number of buckets in this hash table 1621 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table 1622 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment 1623 // Specifically the length of the following HeaderData field - this does not 1624 // include the size of the preceding fields 1625 HeaderData header_data; // Implementation specific header data 1626 }; 1627 1628The header starts with a 32 bit "``magic``" value which must be ``'HASH'`` 1629encoded as an ASCII integer. This allows the detection of the start of the 1630hash table and also allows the table's byte order to be determined so the table 1631can be correctly extracted. The "``magic``" value is followed by a 16 bit 1632``version`` number which allows the table to be revised and modified in the 1633future. The current version number is 1. ``hash_function`` is a ``uint16_t`` 1634enumeration that specifies which hash function was used to produce this table. 1635The current values for the hash function enumerations include: 1636 1637.. code-block:: c 1638 1639 enum HashFunctionType 1640 { 1641 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function 1642 }; 1643 1644``bucket_count`` is a 32 bit unsigned integer that represents how many buckets 1645are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit 1646hash values that are in the ``HASHES`` array, and is the same number of offsets 1647are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size 1648in bytes of the ``HeaderData`` that is filled in by specialized versions of 1649this table. 1650 1651Fixed Lookup 1652"""""""""""" 1653 1654The header is followed by the buckets, hashes, offsets, and hash value data. 1655 1656.. code-block:: c 1657 1658 struct FixedTable 1659 { 1660 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below 1661 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table 1662 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above 1663 }; 1664 1665``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The 1666``hashes`` array contains all of the 32 bit hash values for all names in the 1667hash table. Each hash in the ``hashes`` table has an offset in the ``offsets`` 1668array that points to the data for the hash value. 1669 1670This table setup makes it very easy to repurpose these tables to contain 1671different data, while keeping the lookup mechanism the same for all tables. 1672This layout also makes it possible to save the table to disk and map it in 1673later and do very efficient name lookups with little or no parsing. 1674 1675DWARF lookup tables can be implemented in a variety of ways and can store a lot 1676of information for each name. We want to make the DWARF tables extensible and 1677able to store the data efficiently so we have used some of the DWARF features 1678that enable efficient data storage to define exactly what kind of data we store 1679for each name. 1680 1681The ``HeaderData`` contains a definition of the contents of each HashData chunk. 1682We might want to store an offset to all of the debug information entries (DIEs) 1683for each name. To keep things extensible, we create a list of items, or 1684Atoms, that are contained in the data for each name. First comes the type of 1685the data in each atom: 1686 1687.. code-block:: c 1688 1689 enum AtomType 1690 { 1691 eAtomTypeNULL = 0u, 1692 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding 1693 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question 1694 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2 1695 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags 1696 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags 1697 }; 1698 1699The enumeration values and their meanings are: 1700 1701.. code-block:: none 1702 1703 eAtomTypeNULL - a termination atom that specifies the end of the atom list 1704 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name 1705 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE 1706 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is 1707 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...) 1708 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...) 1709 1710Then we allow each atom type to define the atom type and how the data for each 1711atom type data is encoded: 1712 1713.. code-block:: c 1714 1715 struct Atom 1716 { 1717 uint16_t type; // AtomType enum value 1718 uint16_t form; // DWARF DW_FORM_XXX defines 1719 }; 1720 1721The ``form`` type above is from the DWARF specification and defines the exact 1722encoding of the data for the Atom type. See the DWARF specification for the 1723``DW_FORM_`` definitions. 1724 1725.. code-block:: c 1726 1727 struct HeaderData 1728 { 1729 uint32_t die_offset_base; 1730 uint32_t atom_count; 1731 Atoms atoms[atom_count0]; 1732 }; 1733 1734``HeaderData`` defines the base DIE offset that should be added to any atoms 1735that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``, 1736``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines 1737what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large 1738each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data 1739should be interpreted. 1740 1741For the current implementations of the "``.apple_names``" (all functions + 1742globals), the "``.apple_types``" (names of all types that are defined), and 1743the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom`` 1744array to be: 1745 1746.. code-block:: c 1747 1748 HeaderData.atom_count = 1; 1749 HeaderData.atoms[0].type = eAtomTypeDIEOffset; 1750 HeaderData.atoms[0].form = DW_FORM_data4; 1751 1752This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is 1753encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have 1754multiple matching DIEs in a single file, which could come up with an inlined 1755function for instance. Future tables could include more information about the 1756DIE such as flags indicating if the DIE is a function, method, block, 1757or inlined. 1758 1759The KeyType for the DWARF table is a 32 bit string table offset into the 1760".debug_str" table. The ".debug_str" is the string table for the DWARF which 1761may already contain copies of all of the strings. This helps make sure, with 1762help from the compiler, that we reuse the strings between all of the DWARF 1763sections and keeps the hash table size down. Another benefit to having the 1764compiler generate all strings as DW_FORM_strp in the debug info, is that 1765DWARF parsing can be made much faster. 1766 1767After a lookup is made, we get an offset into the hash data. The hash data 1768needs to be able to deal with 32 bit hash collisions, so the chunk of data 1769at the offset in the hash data consists of a triple: 1770 1771.. code-block:: c 1772 1773 uint32_t str_offset 1774 uint32_t hash_data_count 1775 HashData[hash_data_count] 1776 1777If "str_offset" is zero, then the bucket contents are done. 99.9% of the 1778hash data chunks contain a single item (no 32 bit hash collision): 1779 1780.. code-block:: none 1781 1782 .------------. 1783 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main") 1784 | 0x00000004 | uint32_t HashData count 1785 | 0x........ | uint32_t HashData[0] DIE offset 1786 | 0x........ | uint32_t HashData[1] DIE offset 1787 | 0x........ | uint32_t HashData[2] DIE offset 1788 | 0x........ | uint32_t HashData[3] DIE offset 1789 | 0x00000000 | uint32_t KeyType (end of hash chain) 1790 `------------' 1791 1792If there are collisions, you will have multiple valid string offsets: 1793 1794.. code-block:: none 1795 1796 .------------. 1797 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main") 1798 | 0x00000004 | uint32_t HashData count 1799 | 0x........ | uint32_t HashData[0] DIE offset 1800 | 0x........ | uint32_t HashData[1] DIE offset 1801 | 0x........ | uint32_t HashData[2] DIE offset 1802 | 0x........ | uint32_t HashData[3] DIE offset 1803 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print") 1804 | 0x00000002 | uint32_t HashData count 1805 | 0x........ | uint32_t HashData[0] DIE offset 1806 | 0x........ | uint32_t HashData[1] DIE offset 1807 | 0x00000000 | uint32_t KeyType (end of hash chain) 1808 `------------' 1809 1810Current testing with real world C++ binaries has shown that there is around 1 181132 bit hash collision per 100,000 name entries. 1812 1813Contents 1814^^^^^^^^ 1815 1816As we said, we want to strictly define exactly what is included in the 1817different tables. For DWARF, we have 3 tables: "``.apple_names``", 1818"``.apple_types``", and "``.apple_namespaces``". 1819 1820"``.apple_names``" sections should contain an entry for each DWARF DIE whose 1821``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or 1822``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``, 1823``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains 1824``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and 1825static variables). All global and static variables should be included, 1826including those scoped within functions and classes. For example using the 1827following code: 1828 1829.. code-block:: c 1830 1831 static int var = 0; 1832 1833 void f () 1834 { 1835 static int var = 0; 1836 } 1837 1838Both of the static ``var`` variables would be included in the table. All 1839functions should emit both their full names and their basenames. For C or C++, 1840the full name is the mangled name (if available) which is usually in the 1841``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the 1842function basename. If global or static variables have a mangled name in a 1843``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the 1844simple name found in the ``DW_AT_name`` attribute. 1845 1846"``.apple_types``" sections should contain an entry for each DWARF DIE whose 1847tag is one of: 1848 1849* DW_TAG_array_type 1850* DW_TAG_class_type 1851* DW_TAG_enumeration_type 1852* DW_TAG_pointer_type 1853* DW_TAG_reference_type 1854* DW_TAG_string_type 1855* DW_TAG_structure_type 1856* DW_TAG_subroutine_type 1857* DW_TAG_typedef 1858* DW_TAG_union_type 1859* DW_TAG_ptr_to_member_type 1860* DW_TAG_set_type 1861* DW_TAG_subrange_type 1862* DW_TAG_base_type 1863* DW_TAG_const_type 1864* DW_TAG_file_type 1865* DW_TAG_namelist 1866* DW_TAG_packed_type 1867* DW_TAG_volatile_type 1868* DW_TAG_restrict_type 1869* DW_TAG_atomic_type 1870* DW_TAG_interface_type 1871* DW_TAG_unspecified_type 1872* DW_TAG_shared_type 1873 1874Only entries with a ``DW_AT_name`` attribute are included, and the entry must 1875not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero 1876value). For example, using the following code: 1877 1878.. code-block:: c 1879 1880 int main () 1881 { 1882 int *b = 0; 1883 return *b; 1884 } 1885 1886We get a few type DIEs: 1887 1888.. code-block:: none 1889 1890 0x00000067: TAG_base_type [5] 1891 AT_encoding( DW_ATE_signed ) 1892 AT_name( "int" ) 1893 AT_byte_size( 0x04 ) 1894 1895 0x0000006e: TAG_pointer_type [6] 1896 AT_type( {0x00000067} ( int ) ) 1897 AT_byte_size( 0x08 ) 1898 1899The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``. 1900 1901"``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs. 1902If we run into a namespace that has no name this is an anonymous namespace, and 1903the name should be output as "``(anonymous namespace)``" (without the quotes). 1904Why? This matches the output of the ``abi::cxa_demangle()`` that is in the 1905standard C++ library that demangles mangled names. 1906 1907 1908Language Extensions and File Format Changes 1909^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1910 1911Objective-C Extensions 1912"""""""""""""""""""""" 1913 1914"``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an 1915Objective-C class. The name used in the hash table is the name of the 1916Objective-C class itself. If the Objective-C class has a category, then an 1917entry is made for both the class name without the category, and for the class 1918name with the category. So if we have a DIE at offset 0x1234 with a name of 1919method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add 1920an entry for "``NSString``" that points to DIE 0x1234, and an entry for 1921"``NSString(my_additions)``" that points to 0x1234. This allows us to quickly 1922track down all Objective-C methods for an Objective-C class when doing 1923expressions. It is needed because of the dynamic nature of Objective-C where 1924anyone can add methods to a class. The DWARF for Objective-C methods is also 1925emitted differently from C++ classes where the methods are not usually 1926contained in the class definition, they are scattered about across one or more 1927compile units. Categories can also be defined in different shared libraries. 1928So we need to be able to quickly find all of the methods and class functions 1929given the Objective-C class name, or quickly find all methods and class 1930functions for a class + category name. This table does not contain any 1931selector names, it just maps Objective-C class names (or class names + 1932category) to all of the methods and class functions. The selectors are added 1933as function basenames in the "``.debug_names``" section. 1934 1935In the "``.apple_names``" section for Objective-C functions, the full name is 1936the entire function name with the brackets ("``-[NSString 1937stringWithCString:]``") and the basename is the selector only 1938("``stringWithCString:``"). 1939 1940Mach-O Changes 1941"""""""""""""" 1942 1943The sections names for the apple hash tables are for non-mach-o files. For 1944mach-o files, the sections should be contained in the ``__DWARF`` segment with 1945names as follows: 1946 1947* "``.apple_names``" -> "``__apple_names``" 1948* "``.apple_types``" -> "``__apple_types``" 1949* "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit) 1950* "``.apple_objc``" -> "``__apple_objc``" 1951 1952.. _codeview: 1953 1954CodeView Debug Info Format 1955========================== 1956 1957LLVM supports emitting CodeView, the Microsoft debug info format, and this 1958section describes the design and implementation of that support. 1959 1960Format Background 1961----------------- 1962 1963CodeView as a format is clearly oriented around C++ debugging, and in C++, the 1964majority of debug information tends to be type information. Therefore, the 1965overriding design constraint of CodeView is the separation of type information 1966from other "symbol" information so that type information can be efficiently 1967merged across translation units. Both type information and symbol information is 1968generally stored as a sequence of records, where each record begins with a 196916-bit record size and a 16-bit record kind. 1970 1971Type information is usually stored in the ``.debug$T`` section of the object 1972file. All other debug info, such as line info, string table, symbol info, and 1973inlinee info, is stored in one or more ``.debug$S`` sections. There may only be 1974one ``.debug$T`` section per object file, since all other debug info refers to 1975it. If a PDB (enabled by the ``/Zi`` MSVC option) was used during compilation, 1976the ``.debug$T`` section will contain only an ``LF_TYPESERVER2`` record pointing 1977to the PDB. When using PDBs, symbol information appears to remain in the object 1978file ``.debug$S`` sections. 1979 1980Type records are referred to by their index, which is the number of records in 1981the stream before a given record plus ``0x1000``. Many common basic types, such 1982as the basic integral types and unqualified pointers to them, are represented 1983using type indices less than ``0x1000``. Such basic types are built in to 1984CodeView consumers and do not require type records. 1985 1986Each type record may only contain type indices that are less than its own type 1987index. This ensures that the graph of type stream references is acyclic. While 1988the source-level type graph may contain cycles through pointer types (consider a 1989linked list struct), these cycles are removed from the type stream by always 1990referring to the forward declaration record of user-defined record types. Only 1991"symbol" records in the ``.debug$S`` streams may refer to complete, 1992non-forward-declaration type records. 1993 1994Working with CodeView 1995--------------------- 1996 1997These are instructions for some common tasks for developers working to improve 1998LLVM's CodeView support. Most of them revolve around using the CodeView dumper 1999embedded in ``llvm-readobj``. 2000 2001* Testing MSVC's output:: 2002 2003 $ cl -c -Z7 foo.cpp # Use /Z7 to keep types in the object file 2004 $ llvm-readobj --codeview foo.obj 2005 2006* Getting LLVM IR debug info out of Clang:: 2007 2008 $ clang -g -gcodeview --target=x86_64-windows-msvc foo.cpp -S -emit-llvm 2009 2010 Use this to generate LLVM IR for LLVM test cases. 2011 2012* Generate and dump CodeView from LLVM IR metadata:: 2013 2014 $ llc foo.ll -filetype=obj -o foo.obj 2015 $ llvm-readobj --codeview foo.obj > foo.txt 2016 2017 Use this pattern in lit test cases and FileCheck the output of llvm-readobj 2018 2019Improving LLVM's CodeView support is a process of finding interesting type 2020records, constructing a C++ test case that makes MSVC emit those records, 2021dumping the records, understanding them, and then generating equivalent records 2022in LLVM's backend. 2023