1 ================================
2 Source Level Debugging with LLVM
3 ================================
11 This document is the central repository for all information pertaining to debug
12 information in LLVM. It describes the :ref:`actual format that the LLVM debug
13 information takes <format>`, which is useful for those interested in creating
14 front-ends or dealing directly with the information. Further, this document
15 provides specific examples of what debug information for C/C++ looks like.
17 Philosophy behind LLVM debugging information
18 --------------------------------------------
20 The idea of the LLVM debugging information is to capture how the important
21 pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
22 Several design aspects have shaped the solution that appears here. The
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.
29 * LLVM optimizations should interact in :ref:`well-defined and easily described
30 ways <intro_debugopt>` with the debugging information.
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.
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.
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.
45 The approach used by the LLVM implementation is to use a small set of
46 :ref:`intrinsic functions <format_common_intrinsics>` to define a mapping
47 between LLVM program objects and the source-level objects. The description of
48 the source-level program is maintained in LLVM metadata in an
49 :ref:`implementation-defined format <ccxx_frontend>` (the C/C++ front-end
50 currently uses working draft 7 of the `DWARF 3 standard
51 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_).
53 When a program is being debugged, a debugger interacts with the user and turns
54 the stored debug information into source-language specific information. As
55 such, a debugger must be aware of the source-language, and is thus tied to a
56 specific language or family of languages.
58 Debug information consumers
59 ---------------------------
61 The role of debug information is to provide meta information normally stripped
62 away during the compilation process. This meta information provides an LLVM
63 user a relationship between generated code and the original program source
66 Currently, debug information is consumed by DwarfDebug to produce dwarf
67 information used by the gdb debugger. Other targets could use the same
68 information to produce stabs or other debug forms.
70 It would also be reasonable to use debug information to feed profiling tools
71 for analysis of generated code, or, tools for reconstructing the original
72 source from generated code.
74 TODO - expound a bit more.
78 Debugging optimized code
79 ------------------------
81 An extremely high priority of LLVM debugging information is to make it interact
82 well with optimizations and analysis. In particular, the LLVM debug
83 information provides the following guarantees:
85 * LLVM debug information **always provides information to accurately read
86 the source-level state of the program**, regardless of which LLVM
87 optimizations have been run, and without any modification to the
88 optimizations themselves. However, some optimizations may impact the
89 ability to modify the current state of the program with a debugger, such
90 as setting program variables, or calling functions that have been
93 * As desired, LLVM optimizations can be upgraded to be aware of the LLVM
94 debugging information, allowing them to update the debugging information
95 as they perform aggressive optimizations. This means that, with effort,
96 the LLVM optimizers could optimize debug code just as well as non-debug
99 * LLVM debug information does not prevent optimizations from
100 happening (for example inlining, basic block reordering/merging/cleanup,
101 tail duplication, etc).
103 * LLVM debug information is automatically optimized along with the rest of
104 the program, using existing facilities. For example, duplicate
105 information is automatically merged by the linker, and unused information
106 is automatically removed.
108 Basically, the debug information allows you to compile a program with
109 "``-O0 -g``" and get full debug information, allowing you to arbitrarily modify
110 the program as it executes from a debugger. Compiling a program with
111 "``-O3 -g``" gives you full debug information that is always available and
112 accurate for reading (e.g., you get accurate stack traces despite tail call
113 elimination and inlining), but you might lose the ability to modify the program
114 and call functions where were optimized out of the program, or inlined away
117 :ref:`LLVM test suite <test-suite-quickstart>` provides a framework to test
118 optimizer's handling of debugging information. It can be run like this:
122 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level
125 This will test impact of debugging information on optimization passes. If
126 debugging information influences optimization passes then it will be reported
127 as a failure. See :doc:`TestingGuide` for more information on LLVM test
128 infrastructure and how to run various tests.
132 Debugging information format
133 ============================
135 LLVM debugging information has been carefully designed to make it possible for
136 the optimizer to optimize the program and debugging information without
137 necessarily having to know anything about debugging information. In
138 particular, the use of metadata avoids duplicated debugging information from
139 the beginning, and the global dead code elimination pass automatically deletes
140 debugging information for a function if it decides to delete the function.
142 To do this, most of the debugging information (descriptors for types,
143 variables, functions, source files, etc) is inserted by the language front-end
144 in the form of LLVM metadata.
146 Debug information is designed to be agnostic about the target debugger and
147 debugging information representation (e.g. DWARF/Stabs/etc). It uses a generic
148 pass to decode the information that represents variables, types, functions,
149 namespaces, etc: this allows for arbitrary source-language semantics and
150 type-systems to be used, as long as there is a module written for the target
151 debugger to interpret the information.
153 To provide basic functionality, the LLVM debugger does have to make some
154 assumptions about the source-level language being debugged, though it keeps
155 these to a minimum. The only common features that the LLVM debugger assumes
156 exist are `source files <LangRef.html#difile>`_, and `program objects
157 <LangRef.html#diglobalvariable>`_. These abstract objects are used by a
158 debugger to form stack traces, show information about local variables, etc.
160 This section of the documentation first describes the representation aspects
161 common to any source-language. :ref:`ccxx_frontend` describes the data layout
162 conventions used by the C and C++ front-ends.
164 Debug information descriptors are `specialized metadata nodes
165 <LangRef.html#specialized-metadata>`_, first-class subclasses of ``Metadata``.
167 .. _format_common_intrinsics:
169 Debugger intrinsic functions
170 ----------------------------
172 LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to
173 provide debug information at various points in generated code.
180 void @llvm.dbg.declare(metadata, metadata, metadata)
182 This intrinsic provides information about a local element (e.g., variable).
183 The first argument is metadata holding the alloca for the variable. The second
184 argument is a `local variable <LangRef.html#dilocalvariable>`_ containing a
185 description of the variable. The third argument is a `complex expression
186 <LangRef.html#diexpression>`_.
193 void @llvm.dbg.value(metadata, i64, metadata, metadata)
195 This intrinsic provides information when a user source variable is set to a new
196 value. The first argument is the new value (wrapped as metadata). The second
197 argument is the offset in the user source variable where the new value is
198 written. The third argument is a `local variable
199 <LangRef.html#dilocalvariable>`_ containing a description of the variable. The
200 third argument is a `complex expression <LangRef.html#diexpression>`_.
202 Object lifetimes and scoping
203 ============================
205 In many languages, the local variables in functions can have their lifetimes or
206 scopes limited to a subset of a function. In the C family of languages, for
207 example, variables are only live (readable and writable) within the source
208 block that they are defined in. In functional languages, values are only
209 readable after they have been defined. Though this is a very obvious concept,
210 it is non-trivial to model in LLVM, because it has no notion of scoping in this
211 sense, and does not want to be tied to a language's scoping rules.
213 In order to handle this, the LLVM debug format uses the metadata attached to
214 llvm instructions to encode line number and scoping information. Consider the
215 following C fragment, for example:
229 Compiled to LLVM, this function would be represented like this:
233 ; Function Attrs: nounwind ssp uwtable
234 define void @foo() #0 {
236 %X = alloca i32, align 4
237 %Y = alloca i32, align 4
238 %Z = alloca i32, align 4
239 call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !13), !dbg !14
240 store i32 21, i32* %X, align 4, !dbg !14
241 call void @llvm.dbg.declare(metadata i32* %Y, metadata !15, metadata !13), !dbg !16
242 store i32 22, i32* %Y, align 4, !dbg !16
243 call void @llvm.dbg.declare(metadata i32* %Z, metadata !17, metadata !13), !dbg !19
244 store i32 23, i32* %Z, align 4, !dbg !19
245 %0 = load i32, i32* %X, align 4, !dbg !20
246 store i32 %0, i32* %Z, align 4, !dbg !21
247 %1 = load i32, i32* %Y, align 4, !dbg !22
248 store i32 %1, i32* %X, align 4, !dbg !23
252 ; Function Attrs: nounwind readnone
253 declare void @llvm.dbg.declare(metadata, metadata, metadata) #1
255 attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
256 attributes #1 = { nounwind readnone }
259 !llvm.module.flags = !{!7, !8, !9}
262 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)", isOptimized: false, runtimeVersion: 0, emissionKind: 1, enums: !2, retainedTypes: !2, subprograms: !3, globals: !2, imports: !2)
263 !1 = !DIFile(filename: "/dev/stdin", directory: "/Users/dexonsmith/data/llvm/debug-info")
266 !4 = !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5, isLocal: false, isDefinition: true, scopeLine: 1, isOptimized: false, function: void ()* @foo, variables: !2)
267 !5 = !DISubroutineType(types: !6)
269 !7 = !{i32 2, !"Dwarf Version", i32 2}
270 !8 = !{i32 2, !"Debug Info Version", i32 3}
271 !9 = !{i32 1, !"PIC Level", i32 2}
272 !10 = !{!"clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)"}
273 !11 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "X", scope: !4, file: !1, line: 2, type: !12)
274 !12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
275 !13 = !DIExpression()
276 !14 = !DILocation(line: 2, column: 9, scope: !4)
277 !15 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "Y", scope: !4, file: !1, line: 3, type: !12)
278 !16 = !DILocation(line: 3, column: 9, scope: !4)
279 !17 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "Z", scope: !18, file: !1, line: 5, type: !12)
280 !18 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5)
281 !19 = !DILocation(line: 5, column: 11, scope: !18)
282 !20 = !DILocation(line: 6, column: 11, scope: !18)
283 !21 = !DILocation(line: 6, column: 9, scope: !18)
284 !22 = !DILocation(line: 8, column: 9, scope: !4)
285 !23 = !DILocation(line: 8, column: 7, scope: !4)
286 !24 = !DILocation(line: 9, column: 3, scope: !4)
289 This example illustrates a few important details about LLVM debugging
290 information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
291 location information, which are attached to an instruction, are applied
292 together to allow a debugger to analyze the relationship between statements,
293 variable definitions, and the code used to implement the function.
297 call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !13), !dbg !14
298 ; [debug line = 2:7] [debug variable = X]
300 The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
301 variable ``X``. The metadata ``!dbg !14`` attached to the intrinsic provides
302 scope information for the variable ``X``.
306 !14 = !DILocation(line: 2, column: 9, scope: !4)
307 !4 = !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5,
308 isLocal: false, isDefinition: true, scopeLine: 1,
309 isOptimized: false, function: void ()* @foo,
312 Here ``!14`` is metadata providing `location information
313 <LangRef.html#dilocation>`_. In this example, scope is encoded by ``!4``, a
314 `subprogram descriptor <LangRef.html#disubprogram>`_. This way the location
315 information attached to the intrinsics indicates that the variable ``X`` is
316 declared at line number 2 at a function level scope in function ``foo``.
318 Now lets take another example.
322 call void @llvm.dbg.declare(metadata i32* %Z, metadata !17, metadata !13), !dbg !19
323 ; [debug line = 5:9] [debug variable = Z]
325 The third intrinsic ``%llvm.dbg.declare`` encodes debugging information for
326 variable ``Z``. The metadata ``!dbg !19`` attached to the intrinsic provides
327 scope information for the variable ``Z``.
331 !18 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5)
332 !19 = !DILocation(line: 5, column: 11, scope: !18)
334 Here ``!19`` indicates that ``Z`` is declared at line number 5 and column
335 number 0 inside of lexical scope ``!18``. The lexical scope itself resides
336 inside of subprogram ``!4`` described above.
338 The scope information attached with each instruction provides a straightforward
339 way to find instructions covered by a scope.
343 C/C++ front-end specific debug information
344 ==========================================
346 The C and C++ front-ends represent information about the program in a format
347 that is effectively identical to `DWARF 3.0
348 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_ in terms of information
349 content. This allows code generators to trivially support native debuggers by
350 generating standard dwarf information, and contains enough information for
351 non-dwarf targets to translate it as needed.
353 This section describes the forms used to represent C and C++ programs. Other
354 languages could pattern themselves after this (which itself is tuned to
355 representing programs in the same way that DWARF 3 does), or they could choose
356 to provide completely different forms if they don't fit into the DWARF model.
357 As support for debugging information gets added to the various LLVM
358 source-language front-ends, the information used should be documented here.
360 The following sections provide examples of a few C/C++ constructs and the debug
361 information that would best describe those constructs. The canonical
362 references are the ``DIDescriptor`` classes defined in
363 ``include/llvm/IR/DebugInfo.h`` and the implementations of the helper functions
364 in ``lib/IR/DIBuilder.cpp``.
366 C/C++ source file information
367 -----------------------------
369 ``llvm::Instruction`` provides easy access to metadata attached with an
370 instruction. One can extract line number information encoded in LLVM IR using
371 ``Instruction::getMetadata()`` and ``DILocation::getLineNumber()``.
375 if (MDNode *N = I->getMetadata("dbg")) { // Here I is an LLVM instruction
376 DILocation Loc(N); // DILocation is in DebugInfo.h
377 unsigned Line = Loc.getLineNumber();
378 StringRef File = Loc.getFilename();
379 StringRef Dir = Loc.getDirectory();
382 C/C++ global variable information
383 ---------------------------------
385 Given an integer global variable declared as follows:
391 a C/C++ front-end would generate the following descriptors:
396 ;; Define the global itself.
398 @MyGlobal = global i32 100, align 4
401 ;; List of debug info of globals
405 ;; Some unrelated metadata.
406 !llvm.module.flags = !{!6, !7}
408 ;; Define the compile unit.
409 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1,
411 "clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)",
412 isOptimized: false, runtimeVersion: 0, emissionKind: 1,
413 enums: !2, retainedTypes: !2, subprograms: !2, globals:
419 !1 = !DIFile(filename: "/dev/stdin",
420 directory: "/Users/dexonsmith/data/llvm/debug-info")
425 ;; The Array of Global Variables
429 ;; Define the global variable itself.
431 !4 = !DIGlobalVariable(name: "MyGlobal", scope: !0, file: !1, line: 1,
432 type: !5, isLocal: false, isDefinition: true,
433 variable: i32* @MyGlobal)
438 !5 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
440 ;; Dwarf version to output.
441 !6 = !{i32 2, !"Dwarf Version", i32 2}
443 ;; Debug info schema version.
444 !7 = !{i32 2, !"Debug Info Version", i32 3}
446 C/C++ function information
447 --------------------------
449 Given a function declared as follows:
453 int main(int argc, char *argv[]) {
457 a C/C++ front-end would generate the following descriptors:
462 ;; Define the anchor for subprograms.
464 !4 = !DISubprogram(name: "main", scope: !1, file: !1, line: 1, type: !5,
465 isLocal: false, isDefinition: true, scopeLine: 1,
466 flags: DIFlagPrototyped, isOptimized: false,
467 function: i32 (i32, i8**)* @main, variables: !2)
470 ;; Define the subprogram itself.
472 define i32 @main(i32 %argc, i8** %argv) {
476 Debugging information format
477 ============================
479 Debugging Information Extension for Objective C Properties
480 ----------------------------------------------------------
485 Objective C provides a simpler way to declare and define accessor methods using
486 declared properties. The language provides features to declare a property and
487 to let compiler synthesize accessor methods.
489 The debugger lets developer inspect Objective C interfaces and their instance
490 variables and class variables. However, the debugger does not know anything
491 about the properties defined in Objective C interfaces. The debugger consumes
492 information generated by compiler in DWARF format. The format does not support
493 encoding of Objective C properties. This proposal describes DWARF extensions to
494 encode Objective C properties, which the debugger can use to let developers
495 inspect Objective C properties.
500 Objective C properties exist separately from class members. A property can be
501 defined only by "setter" and "getter" selectors, and be calculated anew on each
502 access. Or a property can just be a direct access to some declared ivar.
503 Finally it can have an ivar "automatically synthesized" for it by the compiler,
504 in which case the property can be referred to in user code directly using the
505 standard C dereference syntax as well as through the property "dot" syntax, but
506 there is no entry in the ``@interface`` declaration corresponding to this ivar.
508 To facilitate debugging, these properties we will add a new DWARF TAG into the
509 ``DW_TAG_structure_type`` definition for the class to hold the description of a
510 given property, and a set of DWARF attributes that provide said description.
511 The property tag will also contain the name and declared type of the property.
513 If there is a related ivar, there will also be a DWARF property attribute placed
514 in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
515 for that property. And in the case where the compiler synthesizes the ivar
516 directly, the compiler is expected to generate a ``DW_TAG_member`` for that
517 ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
518 to access this ivar directly in code, and with the property attribute pointing
519 back to the property it is backing.
521 The following examples will serve as illustration for our discussion:
538 This produces the following DWARF (this is a "pseudo dwarfdump" output):
542 0x00000100: TAG_structure_type [7] *
543 AT_APPLE_runtime_class( 0x10 )
545 AT_decl_file( "Objc_Property.m" )
548 0x00000110 TAG_APPLE_property
550 AT_type ( {0x00000150} ( int ) )
552 0x00000120: TAG_APPLE_property
554 AT_type ( {0x00000150} ( int ) )
556 0x00000130: TAG_member [8]
558 AT_APPLE_property ( {0x00000110} "p1" )
559 AT_type( {0x00000150} ( int ) )
560 AT_artificial ( 0x1 )
562 0x00000140: TAG_member [8]
564 AT_APPLE_property ( {0x00000120} "p2" )
565 AT_type( {0x00000150} ( int ) )
567 0x00000150: AT_type( ( int ) )
569 Note, the current convention is that the name of the ivar for an
570 auto-synthesized property is the name of the property from which it derives
571 with an underscore prepended, as is shown in the example. But we actually
572 don't need to know this convention, since we are given the name of the ivar
575 Also, it is common practice in ObjC to have different property declarations in
576 the @interface and @implementation - e.g. to provide a read-only property in
577 the interface,and a read-write interface in the implementation. In that case,
578 the compiler should emit whichever property declaration will be in force in the
579 current translation unit.
581 Developers can decorate a property with attributes which are encoded using
582 ``DW_AT_APPLE_property_attribute``.
586 @property (readonly, nonatomic) int pr;
590 TAG_APPLE_property [8]
592 AT_type ( {0x00000147} (int) )
593 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
595 The setter and getter method names are attached to the property using
596 ``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
601 @property (setter=myOwnP3Setter:) int p3;
602 -(void)myOwnP3Setter:(int)a;
607 -(void)myOwnP3Setter:(int)a{ }
610 The DWARF for this would be:
614 0x000003bd: TAG_structure_type [7] *
615 AT_APPLE_runtime_class( 0x10 )
617 AT_decl_file( "Objc_Property.m" )
620 0x000003cd TAG_APPLE_property
622 AT_APPLE_property_setter ( "myOwnP3Setter:" )
623 AT_type( {0x00000147} ( int ) )
625 0x000003f3: TAG_member [8]
627 AT_type ( {0x00000147} ( int ) )
628 AT_APPLE_property ( {0x000003cd} )
629 AT_artificial ( 0x1 )
634 +-----------------------+--------+
636 +=======================+========+
637 | DW_TAG_APPLE_property | 0x4200 |
638 +-----------------------+--------+
643 +--------------------------------+--------+-----------+
644 | Attribute | Value | Classes |
645 +================================+========+===========+
646 | DW_AT_APPLE_property | 0x3fed | Reference |
647 +--------------------------------+--------+-----------+
648 | DW_AT_APPLE_property_getter | 0x3fe9 | String |
649 +--------------------------------+--------+-----------+
650 | DW_AT_APPLE_property_setter | 0x3fea | String |
651 +--------------------------------+--------+-----------+
652 | DW_AT_APPLE_property_attribute | 0x3feb | Constant |
653 +--------------------------------+--------+-----------+
658 +--------------------------------------+-------+
660 +======================================+=======+
661 | DW_APPLE_PROPERTY_readonly | 0x01 |
662 +--------------------------------------+-------+
663 | DW_APPLE_PROPERTY_getter | 0x02 |
664 +--------------------------------------+-------+
665 | DW_APPLE_PROPERTY_assign | 0x04 |
666 +--------------------------------------+-------+
667 | DW_APPLE_PROPERTY_readwrite | 0x08 |
668 +--------------------------------------+-------+
669 | DW_APPLE_PROPERTY_retain | 0x10 |
670 +--------------------------------------+-------+
671 | DW_APPLE_PROPERTY_copy | 0x20 |
672 +--------------------------------------+-------+
673 | DW_APPLE_PROPERTY_nonatomic | 0x40 |
674 +--------------------------------------+-------+
675 | DW_APPLE_PROPERTY_setter | 0x80 |
676 +--------------------------------------+-------+
677 | DW_APPLE_PROPERTY_atomic | 0x100 |
678 +--------------------------------------+-------+
679 | DW_APPLE_PROPERTY_weak | 0x200 |
680 +--------------------------------------+-------+
681 | DW_APPLE_PROPERTY_strong | 0x400 |
682 +--------------------------------------+-------+
683 | DW_APPLE_PROPERTY_unsafe_unretained | 0x800 |
684 +--------------------------------+-----+-------+
686 Name Accelerator Tables
687 -----------------------
692 The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
693 debugger needs. The "``pub``" in the section name indicates that the entries
694 in the table are publicly visible names only. This means no static or hidden
695 functions show up in the "``.debug_pubnames``". No static variables or private
696 class variables are in the "``.debug_pubtypes``". Many compilers add different
697 things to these tables, so we can't rely upon the contents between gcc, icc, or
700 The typical query given by users tends not to match up with the contents of
701 these tables. For example, the DWARF spec states that "In the case of the name
702 of a function member or static data member of a C++ structure, class or union,
703 the name presented in the "``.debug_pubnames``" section is not the simple name
704 given by the ``DW_AT_name attribute`` of the referenced debugging information
705 entry, but rather the fully qualified name of the data or function member."
706 So the only names in these tables for complex C++ entries is a fully
707 qualified name. Debugger users tend not to enter their search strings as
708 "``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
709 "``a::b::c``". So the name entered in the name table must be demangled in
710 order to chop it up appropriately and additional names must be manually entered
711 into the table to make it effective as a name lookup table for debuggers to
714 All debuggers currently ignore the "``.debug_pubnames``" table as a result of
715 its inconsistent and useless public-only name content making it a waste of
716 space in the object file. These tables, when they are written to disk, are not
717 sorted in any way, leaving every debugger to do its own parsing and sorting.
718 These tables also include an inlined copy of the string values in the table
719 itself making the tables much larger than they need to be on disk, especially
720 for large C++ programs.
722 Can't we just fix the sections by adding all of the names we need to this
723 table? No, because that is not what the tables are defined to contain and we
724 won't know the difference between the old bad tables and the new good tables.
725 At best we could make our own renamed sections that contain all of the data we
728 These tables are also insufficient for what a debugger like LLDB needs. LLDB
729 uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then
730 often asked to look for type "``foo``" or namespace "``bar``", or list items in
731 namespace "``baz``". Namespaces are not included in the pubnames or pubtypes
732 tables. Since clang asks a lot of questions when it is parsing an expression,
733 we need to be very fast when looking up names, as it happens a lot. Having new
734 accelerator tables that are optimized for very quick lookups will benefit this
735 type of debugging experience greatly.
737 We would like to generate name lookup tables that can be mapped into memory
738 from disk, and used as is, with little or no up-front parsing. We would also
739 be able to control the exact content of these different tables so they contain
740 exactly what we need. The Name Accelerator Tables were designed to fix these
741 issues. In order to solve these issues we need to:
743 * Have a format that can be mapped into memory from disk and used as is
744 * Lookups should be very fast
745 * Extensible table format so these tables can be made by many producers
746 * Contain all of the names needed for typical lookups out of the box
747 * Strict rules for the contents of tables
749 Table size is important and the accelerator table format should allow the reuse
750 of strings from common string tables so the strings for the names are not
751 duplicated. We also want to make sure the table is ready to be used as-is by
752 simply mapping the table into memory with minimal header parsing.
754 The name lookups need to be fast and optimized for the kinds of lookups that
755 debuggers tend to do. Optimally we would like to touch as few parts of the
756 mapped table as possible when doing a name lookup and be able to quickly find
757 the name entry we are looking for, or discover there are no matches. In the
758 case of debuggers we optimized for lookups that fail most of the time.
760 Each table that is defined should have strict rules on exactly what is in the
761 accelerator tables and documented so clients can rely on the content.
769 Typical hash tables have a header, buckets, and each bucket points to the
782 The BUCKETS are an array of offsets to DATA for each hash:
787 | 0x00001000 | BUCKETS[0]
788 | 0x00002000 | BUCKETS[1]
789 | 0x00002200 | BUCKETS[2]
790 | 0x000034f0 | BUCKETS[3]
792 | 0xXXXXXXXX | BUCKETS[n_buckets]
795 So for ``bucket[3]`` in the example above, we have an offset into the table
796 0x000034f0 which points to a chain of entries for the bucket. Each bucket must
797 contain a next pointer, full 32 bit hash value, the string itself, and the data
798 for the current string value.
803 0x000034f0: | 0x00003500 | next pointer
804 | 0x12345678 | 32 bit hash
805 | "erase" | string value
806 | data[n] | HashData for this bucket
808 0x00003500: | 0x00003550 | next pointer
809 | 0x29273623 | 32 bit hash
810 | "dump" | string value
811 | data[n] | HashData for this bucket
813 0x00003550: | 0x00000000 | next pointer
814 | 0x82638293 | 32 bit hash
815 | "main" | string value
816 | data[n] | HashData for this bucket
819 The problem with this layout for debuggers is that we need to optimize for the
820 negative lookup case where the symbol we're searching for is not present. So
821 if we were to lookup "``printf``" in the table above, we would make a 32 hash
822 for "``printf``", it might match ``bucket[3]``. We would need to go to the
823 offset 0x000034f0 and start looking to see if our 32 bit hash matches. To do
824 so, we need to read the next pointer, then read the hash, compare it, and skip
825 to the next bucket. Each time we are skipping many bytes in memory and
826 touching new cache pages just to do the compare on the full 32 bit hash. All
827 of these accesses then tell us that we didn't have a match.
832 To solve the issues mentioned above we have structured the hash tables a bit
833 differently: a header, buckets, an array of all unique 32 bit hash values,
834 followed by an array of hash value data offsets, one for each hash value, then
835 the data for all hash values:
851 The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By
852 making all of the full 32 bit hash values contiguous in memory, we allow
853 ourselves to efficiently check for a match while touching as little memory as
854 possible. Most often checking the 32 bit hash values is as far as the lookup
855 goes. If it does match, it usually is a match with no collisions. So for a
856 table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
857 values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
862 .-------------------------.
863 | HEADER.magic | uint32_t
864 | HEADER.version | uint16_t
865 | HEADER.hash_function | uint16_t
866 | HEADER.bucket_count | uint32_t
867 | HEADER.hashes_count | uint32_t
868 | HEADER.header_data_len | uint32_t
869 | HEADER_DATA | HeaderData
870 |-------------------------|
871 | BUCKETS | uint32_t[n_buckets] // 32 bit hash indexes
872 |-------------------------|
873 | HASHES | uint32_t[n_hashes] // 32 bit hash values
874 |-------------------------|
875 | OFFSETS | uint32_t[n_hashes] // 32 bit offsets to hash value data
876 |-------------------------|
878 `-------------------------'
880 So taking the exact same data from the standard hash example above we end up
893 | ... | BUCKETS[n_buckets]
895 | 0x........ | HASHES[0]
896 | 0x........ | HASHES[1]
897 | 0x........ | HASHES[2]
898 | 0x........ | HASHES[3]
899 | 0x........ | HASHES[4]
900 | 0x........ | HASHES[5]
901 | 0x12345678 | HASHES[6] hash for BUCKETS[3]
902 | 0x29273623 | HASHES[7] hash for BUCKETS[3]
903 | 0x82638293 | HASHES[8] hash for BUCKETS[3]
904 | 0x........ | HASHES[9]
905 | 0x........ | HASHES[10]
906 | 0x........ | HASHES[11]
907 | 0x........ | HASHES[12]
908 | 0x........ | HASHES[13]
909 | 0x........ | HASHES[n_hashes]
911 | 0x........ | OFFSETS[0]
912 | 0x........ | OFFSETS[1]
913 | 0x........ | OFFSETS[2]
914 | 0x........ | OFFSETS[3]
915 | 0x........ | OFFSETS[4]
916 | 0x........ | OFFSETS[5]
917 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3]
918 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3]
919 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3]
920 | 0x........ | OFFSETS[9]
921 | 0x........ | OFFSETS[10]
922 | 0x........ | OFFSETS[11]
923 | 0x........ | OFFSETS[12]
924 | 0x........ | OFFSETS[13]
925 | 0x........ | OFFSETS[n_hashes]
933 0x000034f0: | 0x00001203 | .debug_str ("erase")
934 | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
935 | 0x........ | HashData[0]
936 | 0x........ | HashData[1]
937 | 0x........ | HashData[2]
938 | 0x........ | HashData[3]
939 | 0x00000000 | String offset into .debug_str (terminate data for hash)
941 0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
942 | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
943 | 0x........ | HashData[0]
944 | 0x........ | HashData[1]
945 | 0x00001203 | String offset into .debug_str ("dump")
946 | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
947 | 0x........ | HashData[0]
948 | 0x........ | HashData[1]
949 | 0x........ | HashData[2]
950 | 0x00000000 | String offset into .debug_str (terminate data for hash)
952 0x00003550: | 0x00001203 | String offset into .debug_str ("main")
953 | 0x00000009 | A 32 bit array count - number of HashData with name "main"
954 | 0x........ | HashData[0]
955 | 0x........ | HashData[1]
956 | 0x........ | HashData[2]
957 | 0x........ | HashData[3]
958 | 0x........ | HashData[4]
959 | 0x........ | HashData[5]
960 | 0x........ | HashData[6]
961 | 0x........ | HashData[7]
962 | 0x........ | HashData[8]
963 | 0x00000000 | String offset into .debug_str (terminate data for hash)
966 So we still have all of the same data, we just organize it more efficiently for
967 debugger lookup. If we repeat the same "``printf``" lookup from above, we
968 would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
969 hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which
970 is the index into the ``HASHES`` table. We would then compare any consecutive
971 32 bit hashes values in the ``HASHES`` array as long as the hashes would be in
972 ``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo
973 ``n_buckets`` is still 3. In the case of a failed lookup we would access the
974 memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
975 before we know that we have no match. We don't end up marching through
976 multiple words of memory and we really keep the number of processor data cache
977 lines being accessed as small as possible.
979 The string hash that is used for these lookup tables is the Daniel J.
980 Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a
981 very good hash for all kinds of names in programs with very few hash
984 Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
989 These name hash tables are designed to be generic where specializations of the
990 table get to define additional data that goes into the header ("``HeaderData``"),
991 how the string value is stored ("``KeyType``") and the content of the data for each
997 The header has a fixed part, and the specialized part. The exact format of the
1004 uint32_t magic; // 'HASH' magic value to allow endian detection
1005 uint16_t version; // Version number
1006 uint16_t hash_function; // The hash function enumeration that was used
1007 uint32_t bucket_count; // The number of buckets in this hash table
1008 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table
1009 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
1010 // Specifically the length of the following HeaderData field - this does not
1011 // include the size of the preceding fields
1012 HeaderData header_data; // Implementation specific header data
1015 The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
1016 encoded as an ASCII integer. This allows the detection of the start of the
1017 hash table and also allows the table's byte order to be determined so the table
1018 can be correctly extracted. The "``magic``" value is followed by a 16 bit
1019 ``version`` number which allows the table to be revised and modified in the
1020 future. The current version number is 1. ``hash_function`` is a ``uint16_t``
1021 enumeration that specifies which hash function was used to produce this table.
1022 The current values for the hash function enumerations include:
1026 enum HashFunctionType
1028 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
1031 ``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
1032 are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit
1033 hash values that are in the ``HASHES`` array, and is the same number of offsets
1034 are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size
1035 in bytes of the ``HeaderData`` that is filled in by specialized versions of
1041 The header is followed by the buckets, hashes, offsets, and hash value data.
1047 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below
1048 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table
1049 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above
1052 ``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The
1053 ``hashes`` array contains all of the 32 bit hash values for all names in the
1054 hash table. Each hash in the ``hashes`` table has an offset in the ``offsets``
1055 array that points to the data for the hash value.
1057 This table setup makes it very easy to repurpose these tables to contain
1058 different data, while keeping the lookup mechanism the same for all tables.
1059 This layout also makes it possible to save the table to disk and map it in
1060 later and do very efficient name lookups with little or no parsing.
1062 DWARF lookup tables can be implemented in a variety of ways and can store a lot
1063 of information for each name. We want to make the DWARF tables extensible and
1064 able to store the data efficiently so we have used some of the DWARF features
1065 that enable efficient data storage to define exactly what kind of data we store
1068 The ``HeaderData`` contains a definition of the contents of each HashData chunk.
1069 We might want to store an offset to all of the debug information entries (DIEs)
1070 for each name. To keep things extensible, we create a list of items, or
1071 Atoms, that are contained in the data for each name. First comes the type of
1072 the data in each atom:
1079 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
1080 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question
1081 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
1082 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
1083 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags
1086 The enumeration values and their meanings are:
1088 .. code-block:: none
1090 eAtomTypeNULL - a termination atom that specifies the end of the atom list
1091 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name
1092 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE
1093 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
1094 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...)
1095 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...)
1097 Then we allow each atom type to define the atom type and how the data for each
1098 atom type data is encoded:
1104 uint16_t type; // AtomType enum value
1105 uint16_t form; // DWARF DW_FORM_XXX defines
1108 The ``form`` type above is from the DWARF specification and defines the exact
1109 encoding of the data for the Atom type. See the DWARF specification for the
1110 ``DW_FORM_`` definitions.
1116 uint32_t die_offset_base;
1117 uint32_t atom_count;
1118 Atoms atoms[atom_count0];
1121 ``HeaderData`` defines the base DIE offset that should be added to any atoms
1122 that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
1123 ``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines
1124 what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
1125 each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
1126 should be interpreted.
1128 For the current implementations of the "``.apple_names``" (all functions +
1129 globals), the "``.apple_types``" (names of all types that are defined), and
1130 the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
1135 HeaderData.atom_count = 1;
1136 HeaderData.atoms[0].type = eAtomTypeDIEOffset;
1137 HeaderData.atoms[0].form = DW_FORM_data4;
1139 This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
1140 encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have
1141 multiple matching DIEs in a single file, which could come up with an inlined
1142 function for instance. Future tables could include more information about the
1143 DIE such as flags indicating if the DIE is a function, method, block,
1146 The KeyType for the DWARF table is a 32 bit string table offset into the
1147 ".debug_str" table. The ".debug_str" is the string table for the DWARF which
1148 may already contain copies of all of the strings. This helps make sure, with
1149 help from the compiler, that we reuse the strings between all of the DWARF
1150 sections and keeps the hash table size down. Another benefit to having the
1151 compiler generate all strings as DW_FORM_strp in the debug info, is that
1152 DWARF parsing can be made much faster.
1154 After a lookup is made, we get an offset into the hash data. The hash data
1155 needs to be able to deal with 32 bit hash collisions, so the chunk of data
1156 at the offset in the hash data consists of a triple:
1161 uint32_t hash_data_count
1162 HashData[hash_data_count]
1164 If "str_offset" is zero, then the bucket contents are done. 99.9% of the
1165 hash data chunks contain a single item (no 32 bit hash collision):
1167 .. code-block:: none
1170 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
1171 | 0x00000004 | uint32_t HashData count
1172 | 0x........ | uint32_t HashData[0] DIE offset
1173 | 0x........ | uint32_t HashData[1] DIE offset
1174 | 0x........ | uint32_t HashData[2] DIE offset
1175 | 0x........ | uint32_t HashData[3] DIE offset
1176 | 0x00000000 | uint32_t KeyType (end of hash chain)
1179 If there are collisions, you will have multiple valid string offsets:
1181 .. code-block:: none
1184 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
1185 | 0x00000004 | uint32_t HashData count
1186 | 0x........ | uint32_t HashData[0] DIE offset
1187 | 0x........ | uint32_t HashData[1] DIE offset
1188 | 0x........ | uint32_t HashData[2] DIE offset
1189 | 0x........ | uint32_t HashData[3] DIE offset
1190 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
1191 | 0x00000002 | uint32_t HashData count
1192 | 0x........ | uint32_t HashData[0] DIE offset
1193 | 0x........ | uint32_t HashData[1] DIE offset
1194 | 0x00000000 | uint32_t KeyType (end of hash chain)
1197 Current testing with real world C++ binaries has shown that there is around 1
1198 32 bit hash collision per 100,000 name entries.
1203 As we said, we want to strictly define exactly what is included in the
1204 different tables. For DWARF, we have 3 tables: "``.apple_names``",
1205 "``.apple_types``", and "``.apple_namespaces``".
1207 "``.apple_names``" sections should contain an entry for each DWARF DIE whose
1208 ``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
1209 ``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
1210 ``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains
1211 ``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
1212 static variables). All global and static variables should be included,
1213 including those scoped within functions and classes. For example using the
1225 Both of the static ``var`` variables would be included in the table. All
1226 functions should emit both their full names and their basenames. For C or C++,
1227 the full name is the mangled name (if available) which is usually in the
1228 ``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
1229 function basename. If global or static variables have a mangled name in a
1230 ``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
1231 simple name found in the ``DW_AT_name`` attribute.
1233 "``.apple_types``" sections should contain an entry for each DWARF DIE whose
1238 * DW_TAG_enumeration_type
1239 * DW_TAG_pointer_type
1240 * DW_TAG_reference_type
1241 * DW_TAG_string_type
1242 * DW_TAG_structure_type
1243 * DW_TAG_subroutine_type
1246 * DW_TAG_ptr_to_member_type
1248 * DW_TAG_subrange_type
1253 * DW_TAG_packed_type
1254 * DW_TAG_volatile_type
1255 * DW_TAG_restrict_type
1256 * DW_TAG_interface_type
1257 * DW_TAG_unspecified_type
1258 * DW_TAG_shared_type
1260 Only entries with a ``DW_AT_name`` attribute are included, and the entry must
1261 not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
1262 value). For example, using the following code:
1272 We get a few type DIEs:
1274 .. code-block:: none
1276 0x00000067: TAG_base_type [5]
1277 AT_encoding( DW_ATE_signed )
1279 AT_byte_size( 0x04 )
1281 0x0000006e: TAG_pointer_type [6]
1282 AT_type( {0x00000067} ( int ) )
1283 AT_byte_size( 0x08 )
1285 The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
1287 "``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
1288 If we run into a namespace that has no name this is an anonymous namespace, and
1289 the name should be output as "``(anonymous namespace)``" (without the quotes).
1290 Why? This matches the output of the ``abi::cxa_demangle()`` that is in the
1291 standard C++ library that demangles mangled names.
1294 Language Extensions and File Format Changes
1295 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1297 Objective-C Extensions
1298 """"""""""""""""""""""
1300 "``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
1301 Objective-C class. The name used in the hash table is the name of the
1302 Objective-C class itself. If the Objective-C class has a category, then an
1303 entry is made for both the class name without the category, and for the class
1304 name with the category. So if we have a DIE at offset 0x1234 with a name of
1305 method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
1306 an entry for "``NSString``" that points to DIE 0x1234, and an entry for
1307 "``NSString(my_additions)``" that points to 0x1234. This allows us to quickly
1308 track down all Objective-C methods for an Objective-C class when doing
1309 expressions. It is needed because of the dynamic nature of Objective-C where
1310 anyone can add methods to a class. The DWARF for Objective-C methods is also
1311 emitted differently from C++ classes where the methods are not usually
1312 contained in the class definition, they are scattered about across one or more
1313 compile units. Categories can also be defined in different shared libraries.
1314 So we need to be able to quickly find all of the methods and class functions
1315 given the Objective-C class name, or quickly find all methods and class
1316 functions for a class + category name. This table does not contain any
1317 selector names, it just maps Objective-C class names (or class names +
1318 category) to all of the methods and class functions. The selectors are added
1319 as function basenames in the "``.debug_names``" section.
1321 In the "``.apple_names``" section for Objective-C functions, the full name is
1322 the entire function name with the brackets ("``-[NSString
1323 stringWithCString:]``") and the basename is the selector only
1324 ("``stringWithCString:``").
1329 The sections names for the apple hash tables are for non-mach-o files. For
1330 mach-o files, the sections should be contained in the ``__DWARF`` segment with
1333 * "``.apple_names``" -> "``__apple_names``"
1334 * "``.apple_types``" -> "``__apple_types``"
1335 * "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
1336 * "``.apple_objc``" -> "``__apple_objc``"