1 ================================
2 Source Level Debugging with LLVM
3 ================================
5 .. sectionauthor:: Chris Lattner <sabre@nondot.org> and Jim Laskey <jlaskey@mac.com>
13 This document is the central repository for all information pertaining to debug
14 information in LLVM. It describes the :ref:`actual format that the LLVM debug
15 information takes <format>`, which is useful for those interested in creating
16 front-ends or dealing directly with the information. Further, this document
17 provides specific examples of what debug information for C/C++ looks like.
19 Philosophy behind LLVM debugging information
20 --------------------------------------------
22 The idea of the LLVM debugging information is to capture how the important
23 pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
24 Several design aspects have shaped the solution that appears here. The
27 * Debugging information should have very little impact on the rest of the
28 compiler. No transformations, analyses, or code generators should need to
29 be modified because of debugging information.
31 * LLVM optimizations should interact in :ref:`well-defined and easily described
32 ways <intro_debugopt>` with the debugging information.
34 * Because LLVM is designed to support arbitrary programming languages,
35 LLVM-to-LLVM tools should not need to know anything about the semantics of
36 the source-level-language.
38 * Source-level languages are often **widely** different from one another.
39 LLVM should not put any restrictions of the flavor of the source-language,
40 and the debugging information should work with any language.
42 * With code generator support, it should be possible to use an LLVM compiler
43 to compile a program to native machine code and standard debugging
44 formats. This allows compatibility with traditional machine-code level
45 debuggers, like GDB or DBX.
47 The approach used by the LLVM implementation is to use a small set of
48 :ref:`intrinsic functions <format_common_intrinsics>` to define a mapping
49 between LLVM program objects and the source-level objects. The description of
50 the source-level program is maintained in LLVM metadata in an
51 :ref:`implementation-defined format <ccxx_frontend>` (the C/C++ front-end
52 currently uses working draft 7 of the `DWARF 3 standard
53 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_).
55 When a program is being debugged, a debugger interacts with the user and turns
56 the stored debug information into source-language specific information. As
57 such, a debugger must be aware of the source-language, and is thus tied to a
58 specific language or family of languages.
60 Debug information consumers
61 ---------------------------
63 The role of debug information is to provide meta information normally stripped
64 away during the compilation process. This meta information provides an LLVM
65 user a relationship between generated code and the original program source
68 Currently, debug information is consumed by DwarfDebug to produce dwarf
69 information used by the gdb debugger. Other targets could use the same
70 information to produce stabs or other debug forms.
72 It would also be reasonable to use debug information to feed profiling tools
73 for analysis of generated code, or, tools for reconstructing the original
74 source from generated code.
76 TODO - expound a bit more.
80 Debugging optimized code
81 ------------------------
83 An extremely high priority of LLVM debugging information is to make it interact
84 well with optimizations and analysis. In particular, the LLVM debug
85 information provides the following guarantees:
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, and without any modification to the
90 optimizations themselves. However, some optimizations may impact the
91 ability to modify the current state of the program with a debugger, such
92 as setting program variables, or calling functions that have been
95 * As desired, LLVM optimizations can be upgraded to be aware of the LLVM
96 debugging information, allowing them to update the debugging information
97 as they perform aggressive optimizations. This means that, with effort,
98 the LLVM optimizers could optimize debug code just as well as non-debug
101 * LLVM debug information does not prevent optimizations from
102 happening (for example inlining, basic block reordering/merging/cleanup,
103 tail duplication, etc).
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.
110 Basically, the debug information allows you to compile a program with
111 "``-O0 -g``" and get full debug information, allowing you to arbitrarily modify
112 the program as it executes from a debugger. Compiling a program with
113 "``-O3 -g``" gives you full debug information that is always available and
114 accurate for reading (e.g., you get accurate stack traces despite tail call
115 elimination and inlining), but you might lose the ability to modify the program
116 and call functions where were optimized out of the program, or inlined away
119 :ref:`LLVM test suite <test-suite-quickstart>` provides a framework to test
120 optimizer's handling of debugging information. It can be run like this:
124 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level
127 This will test impact of debugging information on optimization passes. If
128 debugging information influences optimization passes then it will be reported
129 as a failure. See :doc:`TestingGuide` for more information on LLVM test
130 infrastructure and how to run various tests.
134 Debugging information format
135 ============================
137 LLVM debugging information has been carefully designed to make it possible for
138 the optimizer to optimize the program and debugging information without
139 necessarily having to know anything about debugging information. In
140 particular, the use of metadata avoids duplicated debugging information from
141 the beginning, and the global dead code elimination pass automatically deletes
142 debugging information for a function if it decides to delete the function.
144 To do this, most of the debugging information (descriptors for types,
145 variables, functions, source files, etc) is inserted by the language front-end
146 in the form of LLVM metadata.
148 Debug information is designed to be agnostic about the target debugger and
149 debugging information representation (e.g. DWARF/Stabs/etc). It uses a generic
150 pass to decode the information that represents variables, types, functions,
151 namespaces, etc: this allows for arbitrary source-language semantics and
152 type-systems to be used, as long as there is a module written for the target
153 debugger to interpret the information.
155 To provide basic functionality, the LLVM debugger does have to make some
156 assumptions about the source-level language being debugged, though it keeps
157 these to a minimum. The only common features that the LLVM debugger assumes
158 exist are :ref:`source files <format_files>`, and :ref:`program objects
159 <format_global_variables>`. These abstract objects are used by a debugger to
160 form stack traces, show information about local variables, etc.
162 This section of the documentation first describes the representation aspects
163 common to any source-language. :ref:`ccxx_frontend` describes the data layout
164 conventions used by the C and C++ front-ends.
166 Debug information descriptors
167 -----------------------------
169 In consideration of the complexity and volume of debug information, LLVM
170 provides a specification for well formed debug descriptors.
172 Consumers of LLVM debug information expect the descriptors for program objects
173 to start in a canonical format, but the descriptors can include additional
174 information appended at the end that is source-language specific. All LLVM
175 debugging information is versioned, allowing backwards compatibility in the
176 case that the core structures need to change in some way. Also, all debugging
177 information objects start with a tag to indicate what type of object it is.
178 The source-language is allowed to define its own objects, by using unreserved
179 tag numbers. We recommend using with tags in the range 0x1000 through 0x2000
180 (there is a defined ``enum DW_TAG_user_base = 0x1000``.)
182 The fields of debug descriptors used internally by LLVM are restricted to only
183 the simple data types ``i32``, ``i1``, ``float``, ``double``, ``mdstring`` and
193 <a name="LLVMDebugVersion">The first field of a descriptor is always an
194 ``i32`` containing a tag value identifying the content of the descriptor.
195 The remaining fields are specific to the descriptor. The values of tags are
196 loosely bound to the tag values of DWARF information entries. However, that
197 does not restrict the use of the information supplied to DWARF targets. To
198 facilitate versioning of debug information, the tag is augmented with the
199 current debug version (``LLVMDebugVersion = 8 << 16`` or 0x80000 or
202 The details of the various descriptors follow.
204 Compile unit descriptors
205 ^^^^^^^^^^^^^^^^^^^^^^^^
210 i32, ;; Tag = 17 + LLVMDebugVersion (DW_TAG_compile_unit)
211 i32, ;; Unused field.
212 i32, ;; DWARF language identifier (ex. DW_LANG_C89)
213 metadata, ;; Source file name
214 metadata, ;; Source file directory (includes trailing slash)
215 metadata ;; Producer (ex. "4.0.1 LLVM (LLVM research group)")
216 i1, ;; True if this is a main compile unit.
217 i1, ;; True if this is optimized.
219 i32 ;; Runtime version
220 metadata ;; List of enums types
221 metadata ;; List of retained types
222 metadata ;; List of subprograms
223 metadata ;; List of global variables
226 These descriptors contain a source language ID for the file (we use the DWARF
227 3.0 ID numbers, such as ``DW_LANG_C89``, ``DW_LANG_C_plus_plus``,
228 ``DW_LANG_Cobol74``, etc), three strings describing the filename, working
229 directory of the compiler, and an identifier string for the compiler that
232 Compile unit descriptors provide the root context for objects declared in a
233 specific compilation unit. File descriptors are defined using this context.
234 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
235 keep track of subprograms, global variables and type information.
245 i32, ;; Tag = 41 + LLVMDebugVersion (DW_TAG_file_type)
246 metadata, ;; Source file name
247 metadata, ;; Source file directory (includes trailing slash)
251 These descriptors contain information for a file. Global variables and top
252 level functions would be defined using this context. File descriptors also
253 provide context for source line correspondence.
255 Each input file is encoded as a separate file descriptor in LLVM debugging
258 .. _format_global_variables:
260 Global variable descriptors
261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
266 i32, ;; Tag = 52 + LLVMDebugVersion (DW_TAG_variable)
267 i32, ;; Unused field.
268 metadata, ;; Reference to context descriptor
270 metadata, ;; Display name (fully qualified C++ name)
271 metadata, ;; MIPS linkage name (for C++)
272 metadata, ;; Reference to file where defined
273 i32, ;; Line number where defined
274 metadata, ;; Reference to type descriptor
275 i1, ;; True if the global is local to compile unit (static)
276 i1, ;; True if the global is defined in the compile unit (not extern)
277 {}* ;; Reference to the global variable
280 These descriptors provide debug information about globals variables. They
281 provide details such as name, type and where the variable is defined. All
282 global variables are collected inside the named metadata ``!llvm.dbg.cu``.
284 .. _format_subprograms:
286 Subprogram descriptors
287 ^^^^^^^^^^^^^^^^^^^^^^
292 i32, ;; Tag = 46 + LLVMDebugVersion (DW_TAG_subprogram)
293 i32, ;; Unused field.
294 metadata, ;; Reference to context descriptor
296 metadata, ;; Display name (fully qualified C++ name)
297 metadata, ;; MIPS linkage name (for C++)
298 metadata, ;; Reference to file where defined
299 i32, ;; Line number where defined
300 metadata, ;; Reference to type descriptor
301 i1, ;; True if the global is local to compile unit (static)
302 i1, ;; True if the global is defined in the compile unit (not extern)
303 i32, ;; Line number where the scope of the subprogram begins
304 i32, ;; Virtuality, e.g. dwarf::DW_VIRTUALITY__virtual
305 i32, ;; Index into a virtual function
306 metadata, ;; indicates which base type contains the vtable pointer for the
308 i32, ;; Flags - Artifical, Private, Protected, Explicit, Prototyped.
310 Function * , ;; Pointer to LLVM function
311 metadata, ;; Lists function template parameters
312 metadata, ;; Function declaration descriptor
313 metadata ;; List of function variables
316 These descriptors provide debug information about functions, methods and
317 subprograms. They provide details such as name, return types and the source
318 location where the subprogram is defined.
326 i32, ;; Tag = 11 + LLVMDebugVersion (DW_TAG_lexical_block)
327 metadata,;; Reference to context descriptor
329 i32, ;; Column number
330 metadata,;; Reference to source file
331 i32 ;; Unique ID to identify blocks from a template function
334 This descriptor provides debug information about nested blocks within a
335 subprogram. The line number and column numbers are used to dinstinguish two
336 lexical blocks at same depth.
341 i32, ;; Tag = 11 + LLVMDebugVersion (DW_TAG_lexical_block)
342 metadata ;; Reference to the scope we're annotating with a file change
343 metadata,;; Reference to the file the scope is enclosed in.
346 This descriptor provides a wrapper around a lexical scope to handle file
347 changes in the middle of a lexical block.
349 .. _format_basic_type:
351 Basic type descriptors
352 ^^^^^^^^^^^^^^^^^^^^^^
357 i32, ;; Tag = 36 + LLVMDebugVersion (DW_TAG_base_type)
358 metadata, ;; Reference to context
359 metadata, ;; Name (may be "" for anonymous types)
360 metadata, ;; Reference to file where defined (may be NULL)
361 i32, ;; Line number where defined (may be 0)
363 i64, ;; Alignment in bits
364 i64, ;; Offset in bits
366 i32 ;; DWARF type encoding
369 These descriptors define primitive types used in the code. Example ``int``,
370 ``bool`` and ``float``. The context provides the scope of the type, which is
371 usually the top level. Since basic types are not usually user defined the
372 context and line number can be left as NULL and 0. The size, alignment and
373 offset are expressed in bits and can be 64 bit values. The alignment is used
374 to round the offset when embedded in a :ref:`composite type
375 <format_composite_type>` (example to keep float doubles on 64 bit boundaries).
376 The offset is the bit offset if embedded in a :ref:`composite type
377 <format_composite_type>`.
379 The type encoding provides the details of the type. The values are typically
380 one of the following:
388 DW_ATE_signed_char = 6
390 DW_ATE_unsigned_char = 8
392 .. _format_derived_type:
394 Derived type descriptors
395 ^^^^^^^^^^^^^^^^^^^^^^^^
400 i32, ;; Tag (see below)
401 metadata, ;; Reference to context
402 metadata, ;; Name (may be "" for anonymous types)
403 metadata, ;; Reference to file where defined (may be NULL)
404 i32, ;; Line number where defined (may be 0)
406 i64, ;; Alignment in bits
407 i64, ;; Offset in bits
408 i32, ;; Flags to encode attributes, e.g. private
409 metadata, ;; Reference to type derived from
410 metadata, ;; (optional) Name of the Objective C property associated with
411 ;; Objective-C an ivar, or the type of which this
412 ;; pointer-to-member is pointing to members of.
413 metadata, ;; (optional) Name of the Objective C property getter selector.
414 metadata, ;; (optional) Name of the Objective C property setter selector.
415 i32 ;; (optional) Objective C property attributes.
418 These descriptors are used to define types derived from other types. The value
419 of the tag varies depending on the meaning. The following are possible tag
424 DW_TAG_formal_parameter = 5
426 DW_TAG_pointer_type = 15
427 DW_TAG_reference_type = 16
429 DW_TAG_ptr_to_member_type = 31
430 DW_TAG_const_type = 38
431 DW_TAG_volatile_type = 53
432 DW_TAG_restrict_type = 55
434 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
435 <format_composite_type>` or :ref:`subprogram <format_subprograms>`. The type
436 of the member is the :ref:`derived type <format_derived_type>`.
437 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
438 argument of a subprogram.
440 ``DW_TAG_typedef`` is used to provide a name for the derived type.
442 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
443 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
444 :ref:`derived type <format_derived_type>`.
446 :ref:`Derived type <format_derived_type>` location can be determined from the
447 context and line number. The size, alignment and offset are expressed in bits
448 and can be 64 bit values. The alignment is used to round the offset when
449 embedded in a :ref:`composite type <format_composite_type>` (example to keep
450 float doubles on 64 bit boundaries.) The offset is the bit offset if embedded
451 in a :ref:`composite type <format_composite_type>`.
453 Note that the ``void *`` type is expressed as a type derived from NULL.
455 .. _format_composite_type:
457 Composite type descriptors
458 ^^^^^^^^^^^^^^^^^^^^^^^^^^
463 i32, ;; Tag (see below)
464 metadata, ;; Reference to context
465 metadata, ;; Name (may be "" for anonymous types)
466 metadata, ;; Reference to file where defined (may be NULL)
467 i32, ;; Line number where defined (may be 0)
469 i64, ;; Alignment in bits
470 i64, ;; Offset in bits
472 metadata, ;; Reference to type derived from
473 metadata, ;; Reference to array of member descriptors
474 i32 ;; Runtime languages
477 These descriptors are used to define types that are composed of 0 or more
478 elements. The value of the tag varies depending on the meaning. The following
479 are possible tag values:
483 DW_TAG_array_type = 1
484 DW_TAG_enumeration_type = 4
485 DW_TAG_structure_type = 19
486 DW_TAG_union_type = 23
487 DW_TAG_vector_type = 259
488 DW_TAG_subroutine_type = 21
489 DW_TAG_inheritance = 28
491 The vector flag indicates that an array type is a native packed vector.
493 The members of array types (tag = ``DW_TAG_array_type``) or vector types (tag =
494 ``DW_TAG_vector_type``) are :ref:`subrange descriptors <format_subrange>`, each
495 representing the range of subscripts at that level of indexing.
497 The members of enumeration types (tag = ``DW_TAG_enumeration_type``) are
498 :ref:`enumerator descriptors <format_enumerator>`, each representing the
499 definition of enumeration value for the set. All enumeration type descriptors
500 are collected inside the named metadata ``!llvm.dbg.cu``.
502 The members of structure (tag = ``DW_TAG_structure_type``) or union (tag =
503 ``DW_TAG_union_type``) types are any one of the :ref:`basic
504 <format_basic_type>`, :ref:`derived <format_derived_type>` or :ref:`composite
505 <format_composite_type>` type descriptors, each representing a field member of
506 the structure or union.
508 For C++ classes (tag = ``DW_TAG_structure_type``), member descriptors provide
509 information about base classes, static members and member functions. If a
510 member is a :ref:`derived type descriptor <format_derived_type>` and has a tag
511 of ``DW_TAG_inheritance``, then the type represents a base class. If the member
512 of is a :ref:`global variable descriptor <format_global_variables>` then it
513 represents a static member. And, if the member is a :ref:`subprogram
514 descriptor <format_subprograms>` then it represents a member function. For
515 static members and member functions, ``getName()`` returns the members link or
516 the C++ mangled name. ``getDisplayName()`` the simplied version of the name.
518 The first member of subroutine (tag = ``DW_TAG_subroutine_type``) type elements
519 is the return type for the subroutine. The remaining elements are the formal
520 arguments to the subroutine.
522 :ref:`Composite type <format_composite_type>` location can be determined from
523 the context and line number. The size, alignment and offset are expressed in
524 bits and can be 64 bit values. The alignment is used to round the offset when
525 embedded in a :ref:`composite type <format_composite_type>` (as an example, to
526 keep float doubles on 64 bit boundaries). The offset is the bit offset if
527 embedded in a :ref:`composite type <format_composite_type>`.
537 i32, ;; Tag = 33 + LLVMDebugVersion (DW_TAG_subrange_type)
542 These descriptors are used to define ranges of array subscripts for an array
543 :ref:`composite type <format_composite_type>`. The low value defines the lower
544 bounds typically zero for C/C++. The high value is the upper bounds. Values
545 are 64 bit. ``High - Low + 1`` is the size of the array. If ``Low > High``
546 the array bounds are not included in generated debugging information.
548 .. _format_enumerator:
550 Enumerator descriptors
551 ^^^^^^^^^^^^^^^^^^^^^^
556 i32, ;; Tag = 40 + LLVMDebugVersion (DW_TAG_enumerator)
561 These descriptors are used to define members of an enumeration :ref:`composite
562 type <format_composite_type>`, it associates the name to the value.
570 i32, ;; Tag (see below)
573 metadata, ;; Reference to file where defined
574 i32, ;; 24 bit - Line number where defined
575 ;; 8 bit - Argument number. 1 indicates 1st argument.
576 metadata, ;; Type descriptor
578 metadata ;; (optional) Reference to inline location
581 These descriptors are used to define variables local to a sub program. The
582 value of the tag depends on the usage of the variable:
586 DW_TAG_auto_variable = 256
587 DW_TAG_arg_variable = 257
588 DW_TAG_return_variable = 258
590 An auto variable is any variable declared in the body of the function. An
591 argument variable is any variable that appears as a formal argument to the
592 function. A return variable is used to track the result of a function and has
593 no source correspondent.
595 The context is either the subprogram or block where the variable is defined.
596 Name the source variable name. Context and line indicate where the variable
597 was defined. Type descriptor defines the declared type of the variable.
599 .. _format_common_intrinsics:
601 Debugger intrinsic functions
602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
604 LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to
605 provide debug information at various points in generated code.
612 void %llvm.dbg.declare(metadata, metadata)
614 This intrinsic provides information about a local element (e.g., variable).
615 The first argument is metadata holding the alloca for the variable. The second
616 argument is metadata containing a description of the variable.
623 void %llvm.dbg.value(metadata, i64, metadata)
625 This intrinsic provides information when a user source variable is set to a new
626 value. The first argument is the new value (wrapped as metadata). The second
627 argument is the offset in the user source variable where the new value is
628 written. The third argument is metadata containing a description of the user
631 Object lifetimes and scoping
632 ============================
634 In many languages, the local variables in functions can have their lifetimes or
635 scopes limited to a subset of a function. In the C family of languages, for
636 example, variables are only live (readable and writable) within the source
637 block that they are defined in. In functional languages, values are only
638 readable after they have been defined. Though this is a very obvious concept,
639 it is non-trivial to model in LLVM, because it has no notion of scoping in this
640 sense, and does not want to be tied to a language's scoping rules.
642 In order to handle this, the LLVM debug format uses the metadata attached to
643 llvm instructions to encode line number and scoping information. Consider the
644 following C fragment, for example:
658 Compiled to LLVM, this function would be represented like this:
662 define void @foo() nounwind ssp {
664 %X = alloca i32, align 4 ; <i32*> [#uses=4]
665 %Y = alloca i32, align 4 ; <i32*> [#uses=4]
666 %Z = alloca i32, align 4 ; <i32*> [#uses=3]
667 %0 = bitcast i32* %X to {}* ; <{}*> [#uses=1]
668 call void @llvm.dbg.declare(metadata !{i32 * %X}, metadata !0), !dbg !7
669 store i32 21, i32* %X, !dbg !8
670 %1 = bitcast i32* %Y to {}* ; <{}*> [#uses=1]
671 call void @llvm.dbg.declare(metadata !{i32 * %Y}, metadata !9), !dbg !10
672 store i32 22, i32* %Y, !dbg !11
673 %2 = bitcast i32* %Z to {}* ; <{}*> [#uses=1]
674 call void @llvm.dbg.declare(metadata !{i32 * %Z}, metadata !12), !dbg !14
675 store i32 23, i32* %Z, !dbg !15
676 %tmp = load i32* %X, !dbg !16 ; <i32> [#uses=1]
677 %tmp1 = load i32* %Y, !dbg !16 ; <i32> [#uses=1]
678 %add = add nsw i32 %tmp, %tmp1, !dbg !16 ; <i32> [#uses=1]
679 store i32 %add, i32* %Z, !dbg !16
680 %tmp2 = load i32* %Y, !dbg !17 ; <i32> [#uses=1]
681 store i32 %tmp2, i32* %X, !dbg !17
685 declare void @llvm.dbg.declare(metadata, metadata) nounwind readnone
687 !0 = metadata !{i32 459008, metadata !1, metadata !"X",
688 metadata !3, i32 2, metadata !6}; [ DW_TAG_auto_variable ]
689 !1 = metadata !{i32 458763, metadata !2}; [DW_TAG_lexical_block ]
690 !2 = metadata !{i32 458798, i32 0, metadata !3, metadata !"foo", metadata !"foo",
691 metadata !"foo", metadata !3, i32 1, metadata !4,
692 i1 false, i1 true}; [DW_TAG_subprogram ]
693 !3 = metadata !{i32 458769, i32 0, i32 12, metadata !"foo.c",
694 metadata !"/private/tmp", metadata !"clang 1.1", i1 true,
695 i1 false, metadata !"", i32 0}; [DW_TAG_compile_unit ]
696 !4 = metadata !{i32 458773, metadata !3, metadata !"", null, i32 0, i64 0, i64 0,
697 i64 0, i32 0, null, metadata !5, i32 0}; [DW_TAG_subroutine_type ]
698 !5 = metadata !{null}
699 !6 = metadata !{i32 458788, metadata !3, metadata !"int", metadata !3, i32 0,
700 i64 32, i64 32, i64 0, i32 0, i32 5}; [DW_TAG_base_type ]
701 !7 = metadata !{i32 2, i32 7, metadata !1, null}
702 !8 = metadata !{i32 2, i32 3, metadata !1, null}
703 !9 = metadata !{i32 459008, metadata !1, metadata !"Y", metadata !3, i32 3,
704 metadata !6}; [ DW_TAG_auto_variable ]
705 !10 = metadata !{i32 3, i32 7, metadata !1, null}
706 !11 = metadata !{i32 3, i32 3, metadata !1, null}
707 !12 = metadata !{i32 459008, metadata !13, metadata !"Z", metadata !3, i32 5,
708 metadata !6}; [ DW_TAG_auto_variable ]
709 !13 = metadata !{i32 458763, metadata !1}; [DW_TAG_lexical_block ]
710 !14 = metadata !{i32 5, i32 9, metadata !13, null}
711 !15 = metadata !{i32 5, i32 5, metadata !13, null}
712 !16 = metadata !{i32 6, i32 5, metadata !13, null}
713 !17 = metadata !{i32 8, i32 3, metadata !1, null}
714 !18 = metadata !{i32 9, i32 1, metadata !2, null}
716 This example illustrates a few important details about LLVM debugging
717 information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
718 location information, which are attached to an instruction, are applied
719 together to allow a debugger to analyze the relationship between statements,
720 variable definitions, and the code used to implement the function.
724 call void @llvm.dbg.declare(metadata, metadata !0), !dbg !7
726 The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
727 variable ``X``. The metadata ``!dbg !7`` attached to the intrinsic provides
728 scope information for the variable ``X``.
732 !7 = metadata !{i32 2, i32 7, metadata !1, null}
733 !1 = metadata !{i32 458763, metadata !2}; [DW_TAG_lexical_block ]
734 !2 = metadata !{i32 458798, i32 0, metadata !3, metadata !"foo",
735 metadata !"foo", metadata !"foo", metadata !3, i32 1,
736 metadata !4, i1 false, i1 true}; [DW_TAG_subprogram ]
738 Here ``!7`` is metadata providing location information. It has four fields:
739 line number, column number, scope, and original scope. The original scope
740 represents inline location if this instruction is inlined inside a caller, and
741 is null otherwise. In this example, scope is encoded by ``!1``. ``!1``
742 represents a lexical block inside the scope ``!2``, where ``!2`` is a
743 :ref:`subprogram descriptor <format_subprograms>`. This way the location
744 information attached to the intrinsics indicates that the variable ``X`` is
745 declared at line number 2 at a function level scope in function ``foo``.
747 Now lets take another example.
751 call void @llvm.dbg.declare(metadata, metadata !12), !dbg !14
753 The second intrinsic ``%llvm.dbg.declare`` encodes debugging information for
754 variable ``Z``. The metadata ``!dbg !14`` attached to the intrinsic provides
755 scope information for the variable ``Z``.
759 !13 = metadata !{i32 458763, metadata !1}; [DW_TAG_lexical_block ]
760 !14 = metadata !{i32 5, i32 9, metadata !13, null}
762 Here ``!14`` indicates that ``Z`` is declared at line number 5 and
763 column number 9 inside of lexical scope ``!13``. The lexical scope itself
764 resides inside of lexical scope ``!1`` described above.
766 The scope information attached with each instruction provides a straightforward
767 way to find instructions covered by a scope.
771 C/C++ front-end specific debug information
772 ==========================================
774 The C and C++ front-ends represent information about the program in a format
775 that is effectively identical to `DWARF 3.0
776 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_ in terms of information
777 content. This allows code generators to trivially support native debuggers by
778 generating standard dwarf information, and contains enough information for
779 non-dwarf targets to translate it as needed.
781 This section describes the forms used to represent C and C++ programs. Other
782 languages could pattern themselves after this (which itself is tuned to
783 representing programs in the same way that DWARF 3 does), or they could choose
784 to provide completely different forms if they don't fit into the DWARF model.
785 As support for debugging information gets added to the various LLVM
786 source-language front-ends, the information used should be documented here.
788 The following sections provide examples of various C/C++ constructs and the
789 debug information that would best describe those constructs.
791 C/C++ source file information
792 -----------------------------
794 Given the source files ``MySource.cpp`` and ``MyHeader.h`` located in the
795 directory ``/Users/mine/sources``, the following code:
799 #include "MyHeader.h"
801 int main(int argc, char *argv[]) {
805 a C/C++ front-end would generate the following descriptors:
811 ;; Define the compile unit for the main source file "/Users/mine/sources/MySource.cpp".
816 i32 4, ;; Language Id
817 metadata !"MySource.cpp",
818 metadata !"/Users/mine/sources",
819 metadata !"4.2.1 (Based on Apple Inc. build 5649) (LLVM build 00)",
820 i1 true, ;; Main Compile Unit
821 i1 false, ;; Optimized compile unit
822 metadata !"", ;; Compiler flags
823 i32 0} ;; Runtime version
826 ;; Define the file for the file "/Users/mine/sources/MySource.cpp".
830 metadata !"MySource.cpp",
831 metadata !"/Users/mine/sources",
832 metadata !2 ;; Compile unit
836 ;; Define the file for the file "/Users/mine/sources/Myheader.h"
840 metadata !"Myheader.h"
841 metadata !"/Users/mine/sources",
842 metadata !2 ;; Compile unit
847 ``llvm::Instruction`` provides easy access to metadata attached with an
848 instruction. One can extract line number information encoded in LLVM IR using
849 ``Instruction::getMetadata()`` and ``DILocation::getLineNumber()``.
853 if (MDNode *N = I->getMetadata("dbg")) { // Here I is an LLVM instruction
854 DILocation Loc(N); // DILocation is in DebugInfo.h
855 unsigned Line = Loc.getLineNumber();
856 StringRef File = Loc.getFilename();
857 StringRef Dir = Loc.getDirectory();
860 C/C++ global variable information
861 ---------------------------------
863 Given an integer global variable declared as follows:
869 a C/C++ front-end would generate the following descriptors:
874 ;; Define the global itself.
876 %MyGlobal = global int 100
879 ;; List of debug info of globals
883 ;; Define the compile unit.
888 metadata !"foo.cpp", ;; File
889 metadata !"/Volumes/Data/tmp", ;; Directory
890 metadata !"clang version 3.1 ", ;; Producer
891 i1 true, ;; Deprecated field
892 i1 false, ;; "isOptimized"?
893 metadata !"", ;; Flags
894 i32 0, ;; Runtime Version
895 metadata !1, ;; Enum Types
896 metadata !1, ;; Retained Types
897 metadata !1, ;; Subprograms
898 metadata !3 ;; Global Variables
899 } ; [ DW_TAG_compile_unit ]
901 ;; The Array of Global Variables
911 ;; Define the global variable itself.
917 metadata !"MyGlobal", ;; Name
918 metadata !"MyGlobal", ;; Display Name
919 metadata !"", ;; Linkage Name
923 i32 0, ;; IsLocalToUnit
924 i32 1, ;; IsDefinition
925 i32* @MyGlobal ;; LLVM-IR Value
926 } ; [ DW_TAG_variable ]
933 metadata !"foo.cpp", ;; File
934 metadata !"/Volumes/Data/tmp", ;; Directory
936 } ; [ DW_TAG_file_type ]
944 metadata !"int", ;; Name
947 i64 32, ;; Size in Bits
948 i64 32, ;; Align in Bits
952 } ; [ DW_TAG_base_type ]
954 C/C++ function information
955 --------------------------
957 Given a function declared as follows:
961 int main(int argc, char *argv[]) {
965 a C/C++ front-end would generate the following descriptors:
970 ;; Define the anchor for subprograms. Note that the second field of the
971 ;; anchor is 46, which is the same as the tag for subprograms
972 ;; (46 = DW_TAG_subprogram.)
977 metadata !1, ;; Context
978 metadata !"main", ;; Name
979 metadata !"main", ;; Display name
980 metadata !"main", ;; Linkage name
982 i32 1, ;; Line number
984 i1 false, ;; Is local
985 i1 true, ;; Is definition
986 i32 0, ;; Virtuality attribute, e.g. pure virtual function
987 i32 0, ;; Index into virtual table for C++ methods
988 i32 0, ;; Type that holds virtual table.
990 i1 false, ;; True if this function is optimized
991 Function *, ;; Pointer to llvm::Function
992 null ;; Function template parameters
995 ;; Define the subprogram itself.
997 define i32 @main(i32 %argc, i8** %argv) {
1004 The following are the basic type descriptors for C/C++ core types:
1009 .. code-block:: llvm
1013 metadata !1, ;; Context
1014 metadata !"bool", ;; Name
1015 metadata !1, ;; File
1016 i32 0, ;; Line number
1017 i64 8, ;; Size in Bits
1018 i64 8, ;; Align in Bits
1019 i64 0, ;; Offset in Bits
1027 .. code-block:: llvm
1031 metadata !1, ;; Context
1032 metadata !"char", ;; Name
1033 metadata !1, ;; File
1034 i32 0, ;; Line number
1035 i64 8, ;; Size in Bits
1036 i64 8, ;; Align in Bits
1037 i64 0, ;; Offset in Bits
1045 .. code-block:: llvm
1049 metadata !1, ;; Context
1050 metadata !"unsigned char",
1051 metadata !1, ;; File
1052 i32 0, ;; Line number
1053 i64 8, ;; Size in Bits
1054 i64 8, ;; Align in Bits
1055 i64 0, ;; Offset in Bits
1063 .. code-block:: llvm
1067 metadata !1, ;; Context
1068 metadata !"short int",
1069 metadata !1, ;; File
1070 i32 0, ;; Line number
1071 i64 16, ;; Size in Bits
1072 i64 16, ;; Align in Bits
1073 i64 0, ;; Offset in Bits
1081 .. code-block:: llvm
1085 metadata !1, ;; Context
1086 metadata !"short unsigned int",
1087 metadata !1, ;; File
1088 i32 0, ;; Line number
1089 i64 16, ;; Size in Bits
1090 i64 16, ;; Align in Bits
1091 i64 0, ;; Offset in Bits
1099 .. code-block:: llvm
1103 metadata !1, ;; Context
1104 metadata !"int", ;; Name
1105 metadata !1, ;; File
1106 i32 0, ;; Line number
1107 i64 32, ;; Size in Bits
1108 i64 32, ;; Align in Bits
1109 i64 0, ;; Offset in Bits
1117 .. code-block:: llvm
1121 metadata !1, ;; Context
1122 metadata !"unsigned int",
1123 metadata !1, ;; File
1124 i32 0, ;; Line number
1125 i64 32, ;; Size in Bits
1126 i64 32, ;; Align in Bits
1127 i64 0, ;; Offset in Bits
1135 .. code-block:: llvm
1139 metadata !1, ;; Context
1140 metadata !"long long int",
1141 metadata !1, ;; File
1142 i32 0, ;; Line number
1143 i64 64, ;; Size in Bits
1144 i64 64, ;; Align in Bits
1145 i64 0, ;; Offset in Bits
1153 .. code-block:: llvm
1157 metadata !1, ;; Context
1158 metadata !"long long unsigned int",
1159 metadata !1, ;; File
1160 i32 0, ;; Line number
1161 i64 64, ;; Size in Bits
1162 i64 64, ;; Align in Bits
1163 i64 0, ;; Offset in Bits
1171 .. code-block:: llvm
1175 metadata !1, ;; Context
1177 metadata !1, ;; File
1178 i32 0, ;; Line number
1179 i64 32, ;; Size in Bits
1180 i64 32, ;; Align in Bits
1181 i64 0, ;; Offset in Bits
1189 .. code-block:: llvm
1193 metadata !1, ;; Context
1194 metadata !"double",;; Name
1195 metadata !1, ;; File
1196 i32 0, ;; Line number
1197 i64 64, ;; Size in Bits
1198 i64 64, ;; Align in Bits
1199 i64 0, ;; Offset in Bits
1207 Given the following as an example of C/C++ derived type:
1211 typedef const int *IntPtr;
1213 a C/C++ front-end would generate the following descriptors:
1215 .. code-block:: llvm
1218 ;; Define the typedef "IntPtr".
1222 metadata !1, ;; Context
1223 metadata !"IntPtr", ;; Name
1224 metadata !3, ;; File
1225 i32 0, ;; Line number
1226 i64 0, ;; Size in bits
1227 i64 0, ;; Align in bits
1228 i64 0, ;; Offset in bits
1230 metadata !4 ;; Derived From type
1233 ;; Define the pointer type.
1237 metadata !1, ;; Context
1238 metadata !"", ;; Name
1239 metadata !1, ;; File
1240 i32 0, ;; Line number
1241 i64 64, ;; Size in bits
1242 i64 64, ;; Align in bits
1243 i64 0, ;; Offset in bits
1245 metadata !5 ;; Derived From type
1248 ;; Define the const type.
1252 metadata !1, ;; Context
1253 metadata !"", ;; Name
1254 metadata !1, ;; File
1255 i32 0, ;; Line number
1256 i64 32, ;; Size in bits
1257 i64 32, ;; Align in bits
1258 i64 0, ;; Offset in bits
1260 metadata !6 ;; Derived From type
1263 ;; Define the int type.
1267 metadata !1, ;; Context
1268 metadata !"int", ;; Name
1269 metadata !1, ;; File
1270 i32 0, ;; Line number
1271 i64 32, ;; Size in bits
1272 i64 32, ;; Align in bits
1273 i64 0, ;; Offset in bits
1278 C/C++ struct/union types
1279 ------------------------
1281 Given the following as an example of C/C++ struct type:
1291 a C/C++ front-end would generate the following descriptors:
1293 .. code-block:: llvm
1296 ;; Define basic type for unsigned int.
1300 metadata !1, ;; Context
1301 metadata !"unsigned int",
1302 metadata !1, ;; File
1303 i32 0, ;; Line number
1304 i64 32, ;; Size in Bits
1305 i64 32, ;; Align in Bits
1306 i64 0, ;; Offset in Bits
1311 ;; Define composite type for struct Color.
1315 metadata !1, ;; Context
1316 metadata !"Color", ;; Name
1317 metadata !1, ;; Compile unit
1318 i32 1, ;; Line number
1319 i64 96, ;; Size in bits
1320 i64 32, ;; Align in bits
1321 i64 0, ;; Offset in bits
1323 null, ;; Derived From
1324 metadata !3, ;; Elements
1325 i32 0 ;; Runtime Language
1329 ;; Define the Red field.
1333 metadata !1, ;; Context
1334 metadata !"Red", ;; Name
1335 metadata !1, ;; File
1336 i32 2, ;; Line number
1337 i64 32, ;; Size in bits
1338 i64 32, ;; Align in bits
1339 i64 0, ;; Offset in bits
1341 metadata !5 ;; Derived From type
1345 ;; Define the Green field.
1349 metadata !1, ;; Context
1350 metadata !"Green", ;; Name
1351 metadata !1, ;; File
1352 i32 3, ;; Line number
1353 i64 32, ;; Size in bits
1354 i64 32, ;; Align in bits
1355 i64 32, ;; Offset in bits
1357 metadata !5 ;; Derived From type
1361 ;; Define the Blue field.
1365 metadata !1, ;; Context
1366 metadata !"Blue", ;; Name
1367 metadata !1, ;; File
1368 i32 4, ;; Line number
1369 i64 32, ;; Size in bits
1370 i64 32, ;; Align in bits
1371 i64 64, ;; Offset in bits
1373 metadata !5 ;; Derived From type
1377 ;; Define the array of fields used by the composite type Color.
1379 !3 = metadata !{metadata !4, metadata !6, metadata !7}
1381 C/C++ enumeration types
1382 -----------------------
1384 Given the following as an example of C/C++ enumeration type:
1394 a C/C++ front-end would generate the following descriptors:
1396 .. code-block:: llvm
1399 ;; Define composite type for enum Trees
1403 metadata !1, ;; Context
1404 metadata !"Trees", ;; Name
1405 metadata !1, ;; File
1406 i32 1, ;; Line number
1407 i64 32, ;; Size in bits
1408 i64 32, ;; Align in bits
1409 i64 0, ;; Offset in bits
1411 null, ;; Derived From type
1412 metadata !3, ;; Elements
1413 i32 0 ;; Runtime language
1417 ;; Define the array of enumerators used by composite type Trees.
1419 !3 = metadata !{metadata !4, metadata !5, metadata !6}
1422 ;; Define Spruce enumerator.
1424 !4 = metadata !{i32 524328, metadata !"Spruce", i64 100}
1427 ;; Define Oak enumerator.
1429 !5 = metadata !{i32 524328, metadata !"Oak", i64 200}
1432 ;; Define Maple enumerator.
1434 !6 = metadata !{i32 524328, metadata !"Maple", i64 300}
1436 Debugging information format
1437 ============================
1439 Debugging Information Extension for Objective C Properties
1440 ----------------------------------------------------------
1445 Objective C provides a simpler way to declare and define accessor methods using
1446 declared properties. The language provides features to declare a property and
1447 to let compiler synthesize accessor methods.
1449 The debugger lets developer inspect Objective C interfaces and their instance
1450 variables and class variables. However, the debugger does not know anything
1451 about the properties defined in Objective C interfaces. The debugger consumes
1452 information generated by compiler in DWARF format. The format does not support
1453 encoding of Objective C properties. This proposal describes DWARF extensions to
1454 encode Objective C properties, which the debugger can use to let developers
1455 inspect Objective C properties.
1460 Objective C properties exist separately from class members. A property can be
1461 defined only by "setter" and "getter" selectors, and be calculated anew on each
1462 access. Or a property can just be a direct access to some declared ivar.
1463 Finally it can have an ivar "automatically synthesized" for it by the compiler,
1464 in which case the property can be referred to in user code directly using the
1465 standard C dereference syntax as well as through the property "dot" syntax, but
1466 there is no entry in the ``@interface`` declaration corresponding to this ivar.
1468 To facilitate debugging, these properties we will add a new DWARF TAG into the
1469 ``DW_TAG_structure_type`` definition for the class to hold the description of a
1470 given property, and a set of DWARF attributes that provide said description.
1471 The property tag will also contain the name and declared type of the property.
1473 If there is a related ivar, there will also be a DWARF property attribute placed
1474 in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
1475 for that property. And in the case where the compiler synthesizes the ivar
1476 directly, the compiler is expected to generate a ``DW_TAG_member`` for that
1477 ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
1478 to access this ivar directly in code, and with the property attribute pointing
1479 back to the property it is backing.
1481 The following examples will serve as illustration for our discussion:
1483 .. code-block:: objc
1495 @synthesize p2 = n2;
1498 This produces the following DWARF (this is a "pseudo dwarfdump" output):
1500 .. code-block:: none
1502 0x00000100: TAG_structure_type [7] *
1503 AT_APPLE_runtime_class( 0x10 )
1505 AT_decl_file( "Objc_Property.m" )
1508 0x00000110 TAG_APPLE_property
1510 AT_type ( {0x00000150} ( int ) )
1512 0x00000120: TAG_APPLE_property
1514 AT_type ( {0x00000150} ( int ) )
1516 0x00000130: TAG_member [8]
1518 AT_APPLE_property ( {0x00000110} "p1" )
1519 AT_type( {0x00000150} ( int ) )
1520 AT_artificial ( 0x1 )
1522 0x00000140: TAG_member [8]
1524 AT_APPLE_property ( {0x00000120} "p2" )
1525 AT_type( {0x00000150} ( int ) )
1527 0x00000150: AT_type( ( int ) )
1529 Note, the current convention is that the name of the ivar for an
1530 auto-synthesized property is the name of the property from which it derives
1531 with an underscore prepended, as is shown in the example. But we actually
1532 don't need to know this convention, since we are given the name of the ivar
1535 Also, it is common practice in ObjC to have different property declarations in
1536 the @interface and @implementation - e.g. to provide a read-only property in
1537 the interface,and a read-write interface in the implementation. In that case,
1538 the compiler should emit whichever property declaration will be in force in the
1539 current translation unit.
1541 Developers can decorate a property with attributes which are encoded using
1542 ``DW_AT_APPLE_property_attribute``.
1544 .. code-block:: objc
1546 @property (readonly, nonatomic) int pr;
1548 .. code-block:: none
1550 TAG_APPLE_property [8]
1552 AT_type ( {0x00000147} (int) )
1553 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
1555 The setter and getter method names are attached to the property using
1556 ``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
1558 .. code-block:: objc
1561 @property (setter=myOwnP3Setter:) int p3;
1562 -(void)myOwnP3Setter:(int)a;
1567 -(void)myOwnP3Setter:(int)a{ }
1570 The DWARF for this would be:
1572 .. code-block:: none
1574 0x000003bd: TAG_structure_type [7] *
1575 AT_APPLE_runtime_class( 0x10 )
1577 AT_decl_file( "Objc_Property.m" )
1580 0x000003cd TAG_APPLE_property
1582 AT_APPLE_property_setter ( "myOwnP3Setter:" )
1583 AT_type( {0x00000147} ( int ) )
1585 0x000003f3: TAG_member [8]
1587 AT_type ( {0x00000147} ( int ) )
1588 AT_APPLE_property ( {0x000003cd} )
1589 AT_artificial ( 0x1 )
1594 +-----------------------+--------+
1596 +=======================+========+
1597 | DW_TAG_APPLE_property | 0x4200 |
1598 +-----------------------+--------+
1600 New DWARF Attributes
1601 ^^^^^^^^^^^^^^^^^^^^
1603 +--------------------------------+--------+-----------+
1604 | Attribute | Value | Classes |
1605 +================================+========+===========+
1606 | DW_AT_APPLE_property | 0x3fed | Reference |
1607 +--------------------------------+--------+-----------+
1608 | DW_AT_APPLE_property_getter | 0x3fe9 | String |
1609 +--------------------------------+--------+-----------+
1610 | DW_AT_APPLE_property_setter | 0x3fea | String |
1611 +--------------------------------+--------+-----------+
1612 | DW_AT_APPLE_property_attribute | 0x3feb | Constant |
1613 +--------------------------------+--------+-----------+
1618 +--------------------------------+-------+
1620 +================================+=======+
1621 | DW_AT_APPLE_PROPERTY_readonly | 0x1 |
1622 +--------------------------------+-------+
1623 | DW_AT_APPLE_PROPERTY_readwrite | 0x2 |
1624 +--------------------------------+-------+
1625 | DW_AT_APPLE_PROPERTY_assign | 0x4 |
1626 +--------------------------------+-------+
1627 | DW_AT_APPLE_PROPERTY_retain | 0x8 |
1628 +--------------------------------+-------+
1629 | DW_AT_APPLE_PROPERTY_copy | 0x10 |
1630 +--------------------------------+-------+
1631 | DW_AT_APPLE_PROPERTY_nonatomic | 0x20 |
1632 +--------------------------------+-------+
1634 Name Accelerator Tables
1635 -----------------------
1640 The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
1641 debugger needs. The "``pub``" in the section name indicates that the entries
1642 in the table are publicly visible names only. This means no static or hidden
1643 functions show up in the "``.debug_pubnames``". No static variables or private
1644 class variables are in the "``.debug_pubtypes``". Many compilers add different
1645 things to these tables, so we can't rely upon the contents between gcc, icc, or
1648 The typical query given by users tends not to match up with the contents of
1649 these tables. For example, the DWARF spec states that "In the case of the name
1650 of a function member or static data member of a C++ structure, class or union,
1651 the name presented in the "``.debug_pubnames``" section is not the simple name
1652 given by the ``DW_AT_name attribute`` of the referenced debugging information
1653 entry, but rather the fully qualified name of the data or function member."
1654 So the only names in these tables for complex C++ entries is a fully
1655 qualified name. Debugger users tend not to enter their search strings as
1656 "``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
1657 "``a::b::c``". So the name entered in the name table must be demangled in
1658 order to chop it up appropriately and additional names must be manually entered
1659 into the table to make it effective as a name lookup table for debuggers to
1662 All debuggers currently ignore the "``.debug_pubnames``" table as a result of
1663 its inconsistent and useless public-only name content making it a waste of
1664 space in the object file. These tables, when they are written to disk, are not
1665 sorted in any way, leaving every debugger to do its own parsing and sorting.
1666 These tables also include an inlined copy of the string values in the table
1667 itself making the tables much larger than they need to be on disk, especially
1668 for large C++ programs.
1670 Can't we just fix the sections by adding all of the names we need to this
1671 table? No, because that is not what the tables are defined to contain and we
1672 won't know the difference between the old bad tables and the new good tables.
1673 At best we could make our own renamed sections that contain all of the data we
1676 These tables are also insufficient for what a debugger like LLDB needs. LLDB
1677 uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then
1678 often asked to look for type "``foo``" or namespace "``bar``", or list items in
1679 namespace "``baz``". Namespaces are not included in the pubnames or pubtypes
1680 tables. Since clang asks a lot of questions when it is parsing an expression,
1681 we need to be very fast when looking up names, as it happens a lot. Having new
1682 accelerator tables that are optimized for very quick lookups will benefit this
1683 type of debugging experience greatly.
1685 We would like to generate name lookup tables that can be mapped into memory
1686 from disk, and used as is, with little or no up-front parsing. We would also
1687 be able to control the exact content of these different tables so they contain
1688 exactly what we need. The Name Accelerator Tables were designed to fix these
1689 issues. In order to solve these issues we need to:
1691 * Have a format that can be mapped into memory from disk and used as is
1692 * Lookups should be very fast
1693 * Extensible table format so these tables can be made by many producers
1694 * Contain all of the names needed for typical lookups out of the box
1695 * Strict rules for the contents of tables
1697 Table size is important and the accelerator table format should allow the reuse
1698 of strings from common string tables so the strings for the names are not
1699 duplicated. We also want to make sure the table is ready to be used as-is by
1700 simply mapping the table into memory with minimal header parsing.
1702 The name lookups need to be fast and optimized for the kinds of lookups that
1703 debuggers tend to do. Optimally we would like to touch as few parts of the
1704 mapped table as possible when doing a name lookup and be able to quickly find
1705 the name entry we are looking for, or discover there are no matches. In the
1706 case of debuggers we optimized for lookups that fail most of the time.
1708 Each table that is defined should have strict rules on exactly what is in the
1709 accelerator tables and documented so clients can rely on the content.
1714 Standard Hash Tables
1715 """"""""""""""""""""
1717 Typical hash tables have a header, buckets, and each bucket points to the
1720 .. code-block:: none
1730 The BUCKETS are an array of offsets to DATA for each hash:
1732 .. code-block:: none
1735 | 0x00001000 | BUCKETS[0]
1736 | 0x00002000 | BUCKETS[1]
1737 | 0x00002200 | BUCKETS[2]
1738 | 0x000034f0 | BUCKETS[3]
1740 | 0xXXXXXXXX | BUCKETS[n_buckets]
1743 So for ``bucket[3]`` in the example above, we have an offset into the table
1744 0x000034f0 which points to a chain of entries for the bucket. Each bucket must
1745 contain a next pointer, full 32 bit hash value, the string itself, and the data
1746 for the current string value.
1748 .. code-block:: none
1751 0x000034f0: | 0x00003500 | next pointer
1752 | 0x12345678 | 32 bit hash
1753 | "erase" | string value
1754 | data[n] | HashData for this bucket
1756 0x00003500: | 0x00003550 | next pointer
1757 | 0x29273623 | 32 bit hash
1758 | "dump" | string value
1759 | data[n] | HashData for this bucket
1761 0x00003550: | 0x00000000 | next pointer
1762 | 0x82638293 | 32 bit hash
1763 | "main" | string value
1764 | data[n] | HashData for this bucket
1767 The problem with this layout for debuggers is that we need to optimize for the
1768 negative lookup case where the symbol we're searching for is not present. So
1769 if we were to lookup "``printf``" in the table above, we would make a 32 hash
1770 for "``printf``", it might match ``bucket[3]``. We would need to go to the
1771 offset 0x000034f0 and start looking to see if our 32 bit hash matches. To do
1772 so, we need to read the next pointer, then read the hash, compare it, and skip
1773 to the next bucket. Each time we are skipping many bytes in memory and
1774 touching new cache pages just to do the compare on the full 32 bit hash. All
1775 of these accesses then tell us that we didn't have a match.
1780 To solve the issues mentioned above we have structured the hash tables a bit
1781 differently: a header, buckets, an array of all unique 32 bit hash values,
1782 followed by an array of hash value data offsets, one for each hash value, then
1783 the data for all hash values:
1785 .. code-block:: none
1799 The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By
1800 making all of the full 32 bit hash values contiguous in memory, we allow
1801 ourselves to efficiently check for a match while touching as little memory as
1802 possible. Most often checking the 32 bit hash values is as far as the lookup
1803 goes. If it does match, it usually is a match with no collisions. So for a
1804 table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
1805 values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
1808 .. code-block:: none
1810 .-------------------------.
1811 | HEADER.magic | uint32_t
1812 | HEADER.version | uint16_t
1813 | HEADER.hash_function | uint16_t
1814 | HEADER.bucket_count | uint32_t
1815 | HEADER.hashes_count | uint32_t
1816 | HEADER.header_data_len | uint32_t
1817 | HEADER_DATA | HeaderData
1818 |-------------------------|
1819 | BUCKETS | uint32_t[bucket_count] // 32 bit hash indexes
1820 |-------------------------|
1821 | HASHES | uint32_t[hashes_count] // 32 bit hash values
1822 |-------------------------|
1823 | OFFSETS | uint32_t[hashes_count] // 32 bit offsets to hash value data
1824 |-------------------------|
1826 `-------------------------'
1828 So taking the exact same data from the standard hash example above we end up
1831 .. code-block:: none
1841 | ... | BUCKETS[n_buckets]
1843 | 0x........ | HASHES[0]
1844 | 0x........ | HASHES[1]
1845 | 0x........ | HASHES[2]
1846 | 0x........ | HASHES[3]
1847 | 0x........ | HASHES[4]
1848 | 0x........ | HASHES[5]
1849 | 0x12345678 | HASHES[6] hash for BUCKETS[3]
1850 | 0x29273623 | HASHES[7] hash for BUCKETS[3]
1851 | 0x82638293 | HASHES[8] hash for BUCKETS[3]
1852 | 0x........ | HASHES[9]
1853 | 0x........ | HASHES[10]
1854 | 0x........ | HASHES[11]
1855 | 0x........ | HASHES[12]
1856 | 0x........ | HASHES[13]
1857 | 0x........ | HASHES[n_hashes]
1859 | 0x........ | OFFSETS[0]
1860 | 0x........ | OFFSETS[1]
1861 | 0x........ | OFFSETS[2]
1862 | 0x........ | OFFSETS[3]
1863 | 0x........ | OFFSETS[4]
1864 | 0x........ | OFFSETS[5]
1865 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3]
1866 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3]
1867 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3]
1868 | 0x........ | OFFSETS[9]
1869 | 0x........ | OFFSETS[10]
1870 | 0x........ | OFFSETS[11]
1871 | 0x........ | OFFSETS[12]
1872 | 0x........ | OFFSETS[13]
1873 | 0x........ | OFFSETS[n_hashes]
1881 0x000034f0: | 0x00001203 | .debug_str ("erase")
1882 | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
1883 | 0x........ | HashData[0]
1884 | 0x........ | HashData[1]
1885 | 0x........ | HashData[2]
1886 | 0x........ | HashData[3]
1887 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1889 0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
1890 | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
1891 | 0x........ | HashData[0]
1892 | 0x........ | HashData[1]
1893 | 0x00001203 | String offset into .debug_str ("dump")
1894 | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
1895 | 0x........ | HashData[0]
1896 | 0x........ | HashData[1]
1897 | 0x........ | HashData[2]
1898 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1900 0x00003550: | 0x00001203 | String offset into .debug_str ("main")
1901 | 0x00000009 | A 32 bit array count - number of HashData with name "main"
1902 | 0x........ | HashData[0]
1903 | 0x........ | HashData[1]
1904 | 0x........ | HashData[2]
1905 | 0x........ | HashData[3]
1906 | 0x........ | HashData[4]
1907 | 0x........ | HashData[5]
1908 | 0x........ | HashData[6]
1909 | 0x........ | HashData[7]
1910 | 0x........ | HashData[8]
1911 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1914 So we still have all of the same data, we just organize it more efficiently for
1915 debugger lookup. If we repeat the same "``printf``" lookup from above, we
1916 would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
1917 hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which
1918 is the index into the ``HASHES`` table. We would then compare any consecutive
1919 32 bit hashes values in the ``HASHES`` array as long as the hashes would be in
1920 ``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo
1921 ``n_buckets`` is still 3. In the case of a failed lookup we would access the
1922 memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
1923 before we know that we have no match. We don't end up marching through
1924 multiple words of memory and we really keep the number of processor data cache
1925 lines being accessed as small as possible.
1927 The string hash that is used for these lookup tables is the Daniel J.
1928 Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a
1929 very good hash for all kinds of names in programs with very few hash
1932 Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
1937 These name hash tables are designed to be generic where specializations of the
1938 table get to define additional data that goes into the header ("``HeaderData``"),
1939 how the string value is stored ("``KeyType``") and the content of the data for each
1945 The header has a fixed part, and the specialized part. The exact format of the
1952 uint32_t magic; // 'HASH' magic value to allow endian detection
1953 uint16_t version; // Version number
1954 uint16_t hash_function; // The hash function enumeration that was used
1955 uint32_t bucket_count; // The number of buckets in this hash table
1956 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table
1957 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
1958 // Specifically the length of the following HeaderData field - this does not
1959 // include the size of the preceding fields
1960 HeaderData header_data; // Implementation specific header data
1963 The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
1964 encoded as an ASCII integer. This allows the detection of the start of the
1965 hash table and also allows the table's byte order to be determined so the table
1966 can be correctly extracted. The "``magic``" value is followed by a 16 bit
1967 ``version`` number which allows the table to be revised and modified in the
1968 future. The current version number is 1. ``hash_function`` is a ``uint16_t``
1969 enumeration that specifies which hash function was used to produce this table.
1970 The current values for the hash function enumerations include:
1974 enum HashFunctionType
1976 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
1979 ``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
1980 are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit
1981 hash values that are in the ``HASHES`` array, and is the same number of offsets
1982 are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size
1983 in bytes of the ``HeaderData`` that is filled in by specialized versions of
1989 The header is followed by the buckets, hashes, offsets, and hash value data.
1995 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below
1996 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table
1997 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above
2000 ``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The
2001 ``hashes`` array contains all of the 32 bit hash values for all names in the
2002 hash table. Each hash in the ``hashes`` table has an offset in the ``offsets``
2003 array that points to the data for the hash value.
2005 This table setup makes it very easy to repurpose these tables to contain
2006 different data, while keeping the lookup mechanism the same for all tables.
2007 This layout also makes it possible to save the table to disk and map it in
2008 later and do very efficient name lookups with little or no parsing.
2010 DWARF lookup tables can be implemented in a variety of ways and can store a lot
2011 of information for each name. We want to make the DWARF tables extensible and
2012 able to store the data efficiently so we have used some of the DWARF features
2013 that enable efficient data storage to define exactly what kind of data we store
2016 The ``HeaderData`` contains a definition of the contents of each HashData chunk.
2017 We might want to store an offset to all of the debug information entries (DIEs)
2018 for each name. To keep things extensible, we create a list of items, or
2019 Atoms, that are contained in the data for each name. First comes the type of
2020 the data in each atom:
2027 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
2028 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question
2029 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
2030 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
2031 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags
2034 The enumeration values and their meanings are:
2036 .. code-block:: none
2038 eAtomTypeNULL - a termination atom that specifies the end of the atom list
2039 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name
2040 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE
2041 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
2042 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...)
2043 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...)
2045 Then we allow each atom type to define the atom type and how the data for each
2046 atom type data is encoded:
2052 uint16_t type; // AtomType enum value
2053 uint16_t form; // DWARF DW_FORM_XXX defines
2056 The ``form`` type above is from the DWARF specification and defines the exact
2057 encoding of the data for the Atom type. See the DWARF specification for the
2058 ``DW_FORM_`` definitions.
2064 uint32_t die_offset_base;
2065 uint32_t atom_count;
2066 Atoms atoms[atom_count0];
2069 ``HeaderData`` defines the base DIE offset that should be added to any atoms
2070 that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
2071 ``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines
2072 what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
2073 each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
2074 should be interpreted.
2076 For the current implementations of the "``.apple_names``" (all functions +
2077 globals), the "``.apple_types``" (names of all types that are defined), and
2078 the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
2083 HeaderData.atom_count = 1;
2084 HeaderData.atoms[0].type = eAtomTypeDIEOffset;
2085 HeaderData.atoms[0].form = DW_FORM_data4;
2087 This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
2088 encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have
2089 multiple matching DIEs in a single file, which could come up with an inlined
2090 function for instance. Future tables could include more information about the
2091 DIE such as flags indicating if the DIE is a function, method, block,
2094 The KeyType for the DWARF table is a 32 bit string table offset into the
2095 ".debug_str" table. The ".debug_str" is the string table for the DWARF which
2096 may already contain copies of all of the strings. This helps make sure, with
2097 help from the compiler, that we reuse the strings between all of the DWARF
2098 sections and keeps the hash table size down. Another benefit to having the
2099 compiler generate all strings as DW_FORM_strp in the debug info, is that
2100 DWARF parsing can be made much faster.
2102 After a lookup is made, we get an offset into the hash data. The hash data
2103 needs to be able to deal with 32 bit hash collisions, so the chunk of data
2104 at the offset in the hash data consists of a triple:
2109 uint32_t hash_data_count
2110 HashData[hash_data_count]
2112 If "str_offset" is zero, then the bucket contents are done. 99.9% of the
2113 hash data chunks contain a single item (no 32 bit hash collision):
2115 .. code-block:: none
2118 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2119 | 0x00000004 | uint32_t HashData count
2120 | 0x........ | uint32_t HashData[0] DIE offset
2121 | 0x........ | uint32_t HashData[1] DIE offset
2122 | 0x........ | uint32_t HashData[2] DIE offset
2123 | 0x........ | uint32_t HashData[3] DIE offset
2124 | 0x00000000 | uint32_t KeyType (end of hash chain)
2127 If there are collisions, you will have multiple valid string offsets:
2129 .. code-block:: none
2132 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2133 | 0x00000004 | uint32_t HashData count
2134 | 0x........ | uint32_t HashData[0] DIE offset
2135 | 0x........ | uint32_t HashData[1] DIE offset
2136 | 0x........ | uint32_t HashData[2] DIE offset
2137 | 0x........ | uint32_t HashData[3] DIE offset
2138 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
2139 | 0x00000002 | uint32_t HashData count
2140 | 0x........ | uint32_t HashData[0] DIE offset
2141 | 0x........ | uint32_t HashData[1] DIE offset
2142 | 0x00000000 | uint32_t KeyType (end of hash chain)
2145 Current testing with real world C++ binaries has shown that there is around 1
2146 32 bit hash collision per 100,000 name entries.
2151 As we said, we want to strictly define exactly what is included in the
2152 different tables. For DWARF, we have 3 tables: "``.apple_names``",
2153 "``.apple_types``", and "``.apple_namespaces``".
2155 "``.apple_names``" sections should contain an entry for each DWARF DIE whose
2156 ``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
2157 ``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
2158 ``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains
2159 ``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
2160 static variables). All global and static variables should be included,
2161 including those scoped within functions and classes. For example using the
2173 Both of the static ``var`` variables would be included in the table. All
2174 functions should emit both their full names and their basenames. For C or C++,
2175 the full name is the mangled name (if available) which is usually in the
2176 ``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
2177 function basename. If global or static variables have a mangled name in a
2178 ``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
2179 simple name found in the ``DW_AT_name`` attribute.
2181 "``.apple_types``" sections should contain an entry for each DWARF DIE whose
2186 * DW_TAG_enumeration_type
2187 * DW_TAG_pointer_type
2188 * DW_TAG_reference_type
2189 * DW_TAG_string_type
2190 * DW_TAG_structure_type
2191 * DW_TAG_subroutine_type
2194 * DW_TAG_ptr_to_member_type
2196 * DW_TAG_subrange_type
2202 * DW_TAG_packed_type
2203 * DW_TAG_volatile_type
2204 * DW_TAG_restrict_type
2205 * DW_TAG_interface_type
2206 * DW_TAG_unspecified_type
2207 * DW_TAG_shared_type
2209 Only entries with a ``DW_AT_name`` attribute are included, and the entry must
2210 not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
2211 value). For example, using the following code:
2221 We get a few type DIEs:
2223 .. code-block:: none
2225 0x00000067: TAG_base_type [5]
2226 AT_encoding( DW_ATE_signed )
2228 AT_byte_size( 0x04 )
2230 0x0000006e: TAG_pointer_type [6]
2231 AT_type( {0x00000067} ( int ) )
2232 AT_byte_size( 0x08 )
2234 The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
2236 "``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
2237 If we run into a namespace that has no name this is an anonymous namespace, and
2238 the name should be output as "``(anonymous namespace)``" (without the quotes).
2239 Why? This matches the output of the ``abi::cxa_demangle()`` that is in the
2240 standard C++ library that demangles mangled names.
2243 Language Extensions and File Format Changes
2244 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2246 Objective-C Extensions
2247 """"""""""""""""""""""
2249 "``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
2250 Objective-C class. The name used in the hash table is the name of the
2251 Objective-C class itself. If the Objective-C class has a category, then an
2252 entry is made for both the class name without the category, and for the class
2253 name with the category. So if we have a DIE at offset 0x1234 with a name of
2254 method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
2255 an entry for "``NSString``" that points to DIE 0x1234, and an entry for
2256 "``NSString(my_additions)``" that points to 0x1234. This allows us to quickly
2257 track down all Objective-C methods for an Objective-C class when doing
2258 expressions. It is needed because of the dynamic nature of Objective-C where
2259 anyone can add methods to a class. The DWARF for Objective-C methods is also
2260 emitted differently from C++ classes where the methods are not usually
2261 contained in the class definition, they are scattered about across one or more
2262 compile units. Categories can also be defined in different shared libraries.
2263 So we need to be able to quickly find all of the methods and class functions
2264 given the Objective-C class name, or quickly find all methods and class
2265 functions for a class + category name. This table does not contain any
2266 selector names, it just maps Objective-C class names (or class names +
2267 category) to all of the methods and class functions. The selectors are added
2268 as function basenames in the "``.debug_names``" section.
2270 In the "``.apple_names``" section for Objective-C functions, the full name is
2271 the entire function name with the brackets ("``-[NSString
2272 stringWithCString:]``") and the basename is the selector only
2273 ("``stringWithCString:``").
2278 The sections names for the apple hash tables are for non mach-o files. For
2279 mach-o files, the sections should be contained in the ``__DWARF`` segment with
2282 * "``.apple_names``" -> "``__apple_names``"
2283 * "``.apple_types``" -> "``__apple_types``"
2284 * "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
2285 * "``.apple_objc``" -> "``__apple_objc``"