1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>Source Level Debugging with LLVM</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
10 <div class="doc_title">Source Level Debugging with LLVM</div>
14 <img src="venusflytrap.jpg" alt="A leafy and green bug eater"
15 width=247 height=369 align=right>
17 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#phil">Philosophy behind LLVM debugging information</a></li>
20 <li><a href="#debugopt">Debugging optimized code</a></li>
21 <li><a href="#future">Future work</a></li>
23 <li><a href="#llvm-db">Using the <tt>llvm-db</tt> tool</a>
25 <li><a href="#limitations">Limitations of <tt>llvm-db</tt></a></li>
26 <li><a href="#sample">A sample <tt>llvm-db</tt> session</a></li>
27 <li><a href="#startup">Starting the debugger</a></li>
28 <li><a href="#commands">Commands recognized by the debugger</a></li>
31 <li><a href="#architecture">Architecture of the LLVM debugger</a></li>
33 <li><a href="#arch_debugger">The Debugger and InferiorProcess classes</a></li>
34 <li><a href="#arch_info">The RuntimeInfo, ProgramInfo, and SourceLanguage classes</a></li>
35 <li><a href="#arch_llvm-db">The <tt>llvm-db</tt> tool</a></li>
36 <li><a href="#arch_todo">Short-term TODO list</a></li>
39 <li><a href="#format">Debugging information format</a></li>
41 <li><a href="#format_common_anchors">Anchors for global objects</a></li>
42 <li><a href="#format_common_stoppoint">Representing stopping points in the source program</a></li>
43 <li><a href="#format_common_lifetime">Object lifetimes and scoping</a></li>
44 <li><a href="#format_common_descriptors">Object descriptor formats</a></li>
46 <li><a href="#format_common_source_files">Representation of source files</a></li>
47 <li><a href="#format_common_program_objects">Representation of program objects</a></li>
48 <li><a href="#format_common_object_contexts">Program object contexts</a></li>
50 <li><a href="#format_common_intrinsics">Debugger intrinsic functions</a></li>
51 <li><a href="#format_common_tags">Values for debugger tags</a></li>
53 <li><a href="#ccxx_frontend">C/C++ front-end specific debug information</a></li>
55 <li><a href="#ccxx_pse">Program Scope Entries</a></li>
57 <li><a href="#ccxx_compilation_units">Compilation unit entries</a></li>
58 <li><a href="#ccxx_modules">Module, namespace, and importing entries</a></li>
60 <li><a href="#ccxx_dataobjects">Data objects (program variables)</a></li>
64 <!-- *********************************************************************** -->
65 <div class="doc_section"><a name="introduction">Introduction</a></div>
66 <!-- *********************************************************************** -->
68 <div class="doc_text">
70 <p>This document is the central repository for all information pertaining to
71 debug information in LLVM. It describes the <a href="#llvm-db">user
72 interface</a> for the <a href="CommandGuide/llvm-db.html"><tt>llvm-db</tt>
73 tool</a>, which provides a powerful <a href="#llvm-db">source-level debugger</a>
74 to users of LLVM-based compilers. It then describes the <a
75 href="#architecture">various components</a> that make up the debugger and the
76 libraries which future clients may use. Finally, it describes the <a
77 href="#format">actual format that the LLVM debug information</a> takes,
78 which is useful for those interested in creating front-ends or dealing directly
79 with the information.</p>
83 <!-- ======================================================================= -->
84 <div class="doc_subsection">
85 <a name="phil">Philosophy behind LLVM debugging information</a>
88 <div class="doc_text">
91 The idea of the LLVM debugging information is to capture how the important
92 pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
93 Several design aspects have shaped the solution that appears here. The
94 important ones are:</p>
97 <li>Debugging information should have very little impact on the rest of the
98 compiler. No transformations, analyses, or code generators should need to be
99 modified because of debugging information.</li>
101 <li>LLVM optimizations should interact in <a href="#debugopt">well-defined and
102 easily described ways</a> with the debugging information.</li>
104 <li>Because LLVM is designed to support arbitrary programming languages,
105 LLVM-to-LLVM tools should not need to know anything about the semantics of the
106 source-level-language.</li>
108 <li>Source-level languages are often <b>widely</b> different from one another.
109 LLVM should not put any restrictions of the flavor of the source-language, and
110 the debugging information should work with any language.</li>
112 <li>With code generator support, it should be possible to use an LLVM compiler
113 to compile a program to native machine code and standard debugging formats.
114 This allows compatibility with traditional machine-code level debuggers, like
120 The approach used by the LLVM implementation is to use a small set of <a
121 href="#format_common_intrinsics">intrinsic functions</a> to define a mapping
122 between LLVM program objects and the source-level objects. The description of
123 the source-level program is maintained in LLVM global variables in an <a
124 href="#ccxx_frontend">implementation-defined format</a> (the C/C++ front-end
125 currently uses working draft 7 of the <a
126 href="http://www.eagercon.com/dwarf/dwarf3std.htm">Dwarf 3 standard</a>).</p>
129 When a program is debugged, the debugger interacts with the user and turns the
130 stored debug information into source-language specific information. As such,
131 the debugger must be aware of the source-language, and is thus tied to a
132 specific language of family of languages. The <a href="#llvm-db">LLVM
133 debugger</a> is designed to be modular in its support for source-languages.
139 <!-- ======================================================================= -->
140 <div class="doc_subsection">
141 <a name="debugopt">Debugging optimized code</a>
144 <div class="doc_text">
146 An extremely high priority of LLVM debugging information is to make it interact
147 well with optimizations and analysis. In particular, the LLVM debug information
148 provides the following guarantees:</p>
152 <li>LLVM debug information <b>always provides information to accurately read the
153 source-level state of the program</b>, regardless of which LLVM optimizations
154 have been run, and without any modification to the optimizations themselves.
155 However, some optimizations may impact the ability to modify the current state
156 of the program with a debugger, such as setting program variables, or calling
157 function that have been deleted.</li>
159 <li>LLVM optimizations gracefully interact with debugging information. If they
160 are not aware of debug information, they are automatically disabled as necessary
161 in the cases that would invalidate the debug info. This retains the LLVM
162 features making it easy to write new transformations.</li>
164 <li>As desired, LLVM optimizations can be upgraded to be aware of the LLVM
165 debugging information, allowing them to update the debugging information as they
166 perform aggressive optimizations. This means that, with effort, the LLVM
167 optimizers could optimize debug code just as well as non-debug code.</li>
169 <li>LLVM debug information does not prevent many important optimizations from
170 happening (for example inlining, basic block reordering/merging/cleanup, tail
171 duplication, etc), further reducing the amount of the compiler that eventually
172 is "aware" of debugging information.</li>
174 <li>LLVM debug information is automatically optimized along with the rest of the
175 program, using existing facilities. For example, duplicate information is
176 automatically merged by the linker, and unused information is automatically
182 Basically, the debug information allows you to compile a program with "<tt>-O0
183 -g</tt>" and get full debug information, allowing you to arbitrarily modify the
184 program as it executes from the debugger. Compiling a program with "<tt>-O3
185 -g</tt>" gives you full debug information that is always available and accurate
186 for reading (e.g., you get accurate stack traces despite tail call elimination
187 and inlining), but you might lose the ability to modify the program and call
188 functions where were optimized out of the program, or inlined away completely.
194 <!-- ======================================================================= -->
195 <div class="doc_subsection">
196 <a name="future">Future work</a>
199 <div class="doc_text">
201 There are several important extensions that could be eventually added to the
202 LLVM debugger. The most important extension would be to upgrade the LLVM code
203 generators to support debugging information. This would also allow, for
204 example, the X86 code generator to emit native objects that contain debugging
205 information consumable by traditional source-level debuggers like GDB or
209 Additionally, LLVM optimizations can be upgraded to incrementally update the
210 debugging information, <a href="#commands">new commands</a> can be added to the
211 debugger, and thread support could be added to the debugger.</p>
214 The "SourceLanguage" modules provided by <tt>llvm-db</tt> could be substantially
215 improved to provide good support for C++ language features like namespaces and
219 After working with the debugger for a while, perhaps the nicest improvement
220 would be to add some sort of line editor, such as GNU readline (but one that is
221 compatible with the LLVM license).</p>
224 For someone so inclined, it should be straight-forward to write different
225 front-ends for the LLVM debugger, as the LLVM debugging engine is cleanly
226 separated from the <tt>llvm-db</tt> front-end. A new LLVM GUI debugger or IDE
233 <!-- *********************************************************************** -->
234 <div class="doc_section">
235 <a name="llvm-db">Using the <tt>llvm-db</tt> tool</a>
237 <!-- *********************************************************************** -->
239 <div class="doc_text">
242 The <tt>llvm-db</tt> tool provides a GDB-like interface for source-level
243 debugging of programs. This tool provides many standard commands for inspecting
244 and modifying the program as it executes, loading new programs, single stepping,
245 placing breakpoints, etc. This section describes how to use the debugger.
248 <p><tt>llvm-db</tt> has been designed to be as similar to GDB in its user
249 interface as possible. This should make it extremely easy to learn
250 <tt>llvm-db</tt> if you already know <tt>GDB</tt>. In general, <tt>llvm-db</tt>
251 provides the subset of GDB commands that are applicable to LLVM debugging users.
252 If there is a command missing that make a reasonable amount of sense within the
253 <a href="#limitations">limitations of <tt>llvm-db</tt></a>, please report it as
254 a bug or, better yet, submit a patch to add it. :)</p>
258 <!-- ======================================================================= -->
259 <div class="doc_subsection">
260 <a name="limitations">Limitations of <tt>llvm-db</tt></a>
263 <div class="doc_text">
265 <p><tt>llvm-db</tt> is designed to be modular and easy to extend. This
266 extensibility was key to getting the debugger up-and-running quickly, because we
267 can start with simple-but-unsophisicated implementations of various components.
268 Because of this, it is currently missing many features, though they should be
269 easy to add over time (patches welcomed!). The biggest inherent limitations of
270 <tt>llvm-db</tt> are currently due to extremely simple <a
271 href="#arch_debugger">debugger backend</a> (implemented in
272 "lib/Debugger/UnixLocalInferiorProcess.cpp") which is designed to work without
273 any cooperation from the code generators. Because it is so simple, it suffers
274 from the following inherent limitations:</p>
278 <li>Running a program in <tt>llvm-db</tt> is a bit slower than running it with
279 <tt>lli</tt> (i.e., in the JIT).</li>
281 <li>Inspection of the target hardware is not supported. This means that you
282 cannot, for example, print the contents of X86 registers.</li>
284 <li>Inspection of LLVM code is not supported. This means that you cannot print
285 the contents of arbitrary LLVM values, or use commands such as <tt>stepi</tt>.
286 This also means that you cannot debug code without debug information.</li>
288 <li>Portions of the debugger run in the same address space as the program being
289 debugged. This means that memory corruption by the program could trample on
290 portions of the debugger.</li>
292 <li>Attaching to existing processes and core files is not currently
297 <p>That said, the debugger is still quite useful, and all of these limitations
298 can be eliminated by integrating support for the debugger into the code
299 generators, and writing a new <a href="#arch_debugger">InferiorProcess</a>
300 subclass to use it. See the <a href="#future">future work</a> section for ideas
301 of how to extend the LLVM debugger despite these limitations.</p>
306 <!-- ======================================================================= -->
307 <div class="doc_subsection">
308 <a name="sample">A sample <tt>llvm-db</tt> session</a>
311 <div class="doc_text">
313 <p>TODO: this is obviously lame, when more is implemented, this can be much
317 $ <b>llvm-db funccall</b>
318 llvm-db: The LLVM source-level debugger
319 Loading program... successfully loaded 'funccall.bc'!
320 (llvm-db) <b>create</b>
321 Starting program: funccall.bc
322 main at funccall.c:9:2
324 (llvm-db) <b>list main</b>
335 (llvm-db) <b>list</b>
337 (llvm-db) <b>step</b>
340 foo at funccall.c:5:2
343 #0 -> 0x85ffba0 in foo at funccall.c:5:2
344 #1 0x85ffd98 in main at funccall.c:10:2
345 (llvm-db) <b>finish</b>
346 main at funccall.c:11:2
351 The program stopped with exit code 0
352 (llvm-db) <b>quit</b>
360 <!-- ======================================================================= -->
361 <div class="doc_subsection">
362 <a name="startup">Starting the debugger</a>
365 <div class="doc_text">
367 <p>There are three ways to start up the <tt>llvm-db</tt> debugger:</p>
369 <p>When run with no options, just <tt>llvm-db</tt>, the debugger starts up
370 without a program loaded at all. You must use the <a
371 href="#c_file"><tt>file</tt> command</a> to load a program, and the <a
372 href="c_set_args"><tt>set args</tt></a> or <a href="#c_run"><tt>run</tt></a>
373 commands to specify the arguments for the program.</p>
375 <p>If you start the debugger with one argument, as <tt>llvm-db
376 <program></tt>, the debugger will start up and load in the specified
377 program. You can then optionally specify arguments to the program with the <a
378 href="c_set_args"><tt>set args</tt></a> or <a href="#c_run"><tt>run</tt></a>
381 <p>The third way to start the program is with the <tt>--args</tt> option. This
382 option allows you to specify the program to load and the arguments to start out
383 with. <!-- No options to <tt>llvm-db</tt> may be specified after the
384 <tt>-args</tt> option. --> Example use: <tt>llvm-db --args ls /home</tt></p>
388 <!-- ======================================================================= -->
389 <div class="doc_subsection">
390 <a name="commands">Commands recognized by the debugger</a>
393 <div class="doc_text">
395 <p>FIXME: this needs work obviously. See the <a
396 href="http://sources.redhat.com/gdb/documentation/">GDB documentation</a> for
397 information about what these do, or try '<tt>help [command]</tt>' within
398 <tt>llvm-db</tt> to get information.</p>
401 <h2>General usage:</h2>
403 <li>help [command]</li>
405 <li><a name="c_file">file</a> [program]</li>
408 <h2>Program inspection and interaction:</h2>
410 <li>create (start the program, stopping it ASAP in <tt>main</tt>)</li>
418 <li>list [start[, end]]</li>
420 <li>info sources</li>
421 <li>info functions</li>
424 <h2>Call stack inspection:</h2>
433 <h2>Debugger inspection and interaction:</h2>
438 <li>show listsize</li>
439 <li>set listsize</li>
440 <li>show language</li>
441 <li>set language</li>
443 <li>set args [args]</li>
454 <li>info variables</li>
455 <li>info program</li>
460 <li>... many others</li>
465 <!-- *********************************************************************** -->
466 <div class="doc_section">
467 <a name="architecture">Architecture of the LLVM debugger</a>
469 <!-- *********************************************************************** -->
471 <div class="doc_text">
474 The LLVM debugger is built out of three distinct layers of software. These
475 layers provide clients with different interface options depending on what pieces
476 of they want to implement themselves, and it also promotes code modularity and
477 good design. The three layers are the <a href="#arch_debugger">Debugger
478 interface</a>, the <a href="#arch_info">"info" interfaces</a>, and the
479 <a href="#arch_llvm-db"><tt>llvm-db</tt> tool</a> itself.
483 <!-- ======================================================================= -->
484 <div class="doc_subsection">
485 <a name="arch_debugger">The Debugger and InferiorProcess classes</a>
488 <div class="doc_text">
490 The Debugger class (defined in the <tt>include/llvm/Debugger/</tt> directory) is
491 a low-level class which is used to maintain information about the loaded
492 program, as well as start and stop the program running as necessary. This class
493 does not provide any high-level analysis or control over the program, only
494 exposing simple interfaces like <tt>load/unloadProgram</tt>,
495 <tt>create/killProgram</tt>, <tt>step/next/finish/contProgram</tt>, and
496 low-level methods for installing breakpoints.
500 The Debugger class is itself a wrapper around the lowest-level InferiorProcess
501 class. This class is used to represent an instance of the program running under
502 debugger control. The InferiorProcess class can be implemented in different
503 ways for different targets and execution scenarios (e.g., remote debugging).
504 The InferiorProcess class exposes a small and simple collection of interfaces
505 which are useful for inspecting the current state of the program (such as
506 collecting stack trace information, reading the memory image of the process,
507 etc). The interfaces in this class are designed to be as low-level and simple
508 as possible, to make it easy to create new instances of the class.
512 The Debugger class exposes the currently active instance of InferiorProcess
513 through the <tt>Debugger::getRunningProcess</tt> method, which returns a
514 <tt>const</tt> reference to the class. This means that clients of the Debugger
515 class can only <b>inspect</b> the running instance of the program directly. To
516 change the executing process in some way, they must use the interces exposed by
521 <!-- ======================================================================= -->
522 <div class="doc_subsection">
523 <a name="arch_info">The RuntimeInfo, ProgramInfo, and SourceLanguage classes</a>
526 <div class="doc_text">
528 The next-highest level of debugger abstraction is provided through the
529 ProgramInfo, RuntimeInfo, SourceLanguage and related classes (also defined in
530 the <tt>include/llvm/Debugger/</tt> directory). These classes efficiently
531 decode the debugging information and low-level interfaces exposed by
532 InferiorProcess into a higher-level representation, suitable for analysis by the
537 The ProgramInfo class exposes a variety of different kinds of information about
538 the program objects in the source-level-language. The SourceFileInfo class
539 represents a source-file in the program (e.g. a .cpp or .h file). The
540 SourceFileInfo class captures information such as which SourceLanguage was used
541 to compile the file, where the debugger can get access to the actual file text
542 (which is lazily loaded on demand), etc. The SourceFunctionInfo class
543 represents a... <b>FIXME: finish</b>. The ProgramInfo class provides interfaces
544 to lazily find and decode the information needed to create the Source*Info
545 classes requested by the debugger.
549 The RuntimeInfo class exposes information about the currently executed program,
550 by decoding information from the InferiorProcess and ProgramInfo classes. It
551 provides a StackFrame class which provides an easy-to-use interface for
552 inspecting the current and suspended stack frames in the program.
556 The SourceLanguage class is an abstract interface used by the debugger to
557 perform all source-language-specific tasks. For example, this interface is used
558 by the ProgramInfo class to decode language-specific types and functions and by
559 the debugger front-end (such as <a href="#arch_llvm-db"><tt>llvm-db</tt></a> to
560 evaluate source-langauge expressions typed into the debugger. This class uses
561 the RuntimeInfo & ProgramInfo classes to get information about the current
562 execution context and the loaded program, respectively.
567 <!-- ======================================================================= -->
568 <div class="doc_subsection">
569 <a name="arch_llvm-db">The <tt>llvm-db</tt> tool</a>
572 <div class="doc_text">
574 The <tt>llvm-db</tt> is designed to be a debugger providing an interface as <a
575 href="#llvm-db">similar to GDB</a> as reasonable, but no more so than that.
576 Because the <a href="#arch_debugger">Debugger</a> and <a
577 href="#arch_info">info</a> classes implement all of the heavy lifting and
578 analysis, <tt>llvm-db</tt> (which lives in <tt>llvm/tools/llvm-db</tt>) consists
579 mainly of of code to interact with the user and parse commands. The CLIDebugger
580 constructor registers all of the builtin commands for the debugger, and each
581 command is implemented as a CLIDebugger::[name]Command method.
586 <!-- ======================================================================= -->
587 <div class="doc_subsection">
588 <a name="arch_todo">Short-term TODO list</a>
591 <div class="doc_text">
594 FIXME: this section will eventually go away. These are notes to myself of
595 things that should be implemented, but haven't yet.
599 <b>Breakpoints:</b> Support is already implemented in the 'InferiorProcess'
600 class, though it hasn't been tested yet. To finish breakpoint support, we need
601 to implement breakCommand (which should reuse the linespec parser from the list
602 command), and handle the fact that 'break foo' or 'break file.c:53' may insert
603 multiple breakpoints. Also, if you say 'break file.c:53' and there is no
604 stoppoint on line 53, the breakpoint should go on the next available line. My
605 idea was to have the Debugger class provide a "Breakpoint" class which
606 encapsulated this messiness, giving the debugger front-end a simple interface.
607 The debugger front-end would have to map the really complex semantics of
608 temporary breakpoints and 'conditional' breakpoints onto this intermediate
609 level. Also, breakpoints should survive as much as possible across program
614 <b>UnixLocalInferiorProcess.cpp speedup</b>: There is no reason for the debugged
615 process to code gen the globals corresponding to debug information. The
616 IntrinsicLowering object could instead change descriptors into constant expr
617 casts of the constant address of the LLVM objects for the descriptors. This
618 would also allow us to eliminate the mapping back and forth between physical
619 addresses that must be done.</p>
622 <b>Process deaths</b>: The InferiorProcessDead exception should be extended to
623 know "how" a process died, i.e., it was killed by a signal. This is easy to
624 collect in the UnixLocalInferiorProcess, we just need to represent it.</p>
628 <!-- *********************************************************************** -->
629 <div class="doc_section">
630 <a name="format">Debugging information format</a>
632 <!-- *********************************************************************** -->
634 <div class="doc_text">
636 <p>LLVM debugging information has been carefully designed to make it possible
637 for the optimizer to optimize the program and debugging information without
638 necessarily having to know anything about debugging information. In particular,
639 the global constant merging pass automatically eliminates duplicated debugging
640 information (often caused by header files), the global dead code elimination
641 pass automatically deletes debugging information for a function if it decides to
642 delete the function, and the linker eliminates debug information when it merges
643 <tt>linkonce</tt> functions.</p>
645 <p>To do this, most of the debugging information (descriptors for types,
646 variables, functions, source files, etc) is inserted by the language front-end
647 in the form of LLVM global variables. These LLVM global variables are no
648 different from any other global variables, except that they have a web of LLVM
649 intrinsic functions that point to them. If the last references to a particular
650 piece of debugging information are deleted (for example, by the
651 <tt>-globaldce</tt> pass), the extraneous debug information will automatically
652 become dead and be removed by the optimizer.</p>
654 <p>The debugger is designed to be agnostic about the contents of most of the
655 debugging information. It uses a <a href="#arch_info">source-language-specific
656 module</a> to decode the information that represents variables, types,
657 functions, namespaces, etc: this allows for arbitrary source-language semantics
658 and type-systems to be used, as long as there is a module written for the
659 debugger to interpret the information.
663 To provide basic functionality, the LLVM debugger does have to make some
664 assumptions about the source-level language being debugged, though it keeps
665 these to a minimum. The only common features that the LLVM debugger assumes
666 exist are <a href="#format_common_source_files">source files</a>, and <a
667 href="#format_program_objects">program objects</a>. These abstract objects are
668 used by the debugger to form stack traces, show information about local
671 <p>This section of the documentation first describes the representation aspects
672 common to any source-language. The <a href="#ccxx_frontend">next section</a>
673 describes the data layout conventions used by the C and C++ front-ends.</p>
677 <!-- ======================================================================= -->
678 <div class="doc_subsection">
679 <a name="format_common_anchors">Anchors for global objects</a>
682 <div class="doc_text">
684 One important aspect of the LLVM debug representation is that it allows the LLVM
685 debugger to efficiently index all of the global objects without having the scan
686 the program. To do this, all of the global objects use "anchor" globals of type
687 "<tt>{}</tt>", with designated names. These anchor objects obviously do not
688 contain any content or meaning by themselves, but all of the global objects of a
689 particular type (e.g., source file descriptors) contain a pointer to the anchor.
690 This pointer allows the debugger to use def-use chains to find all global
691 objects of that type.
695 So far, the following names are recognized as anchors by the LLVM debugger:
699 %<a href="#format_common_source_files">llvm.dbg.translation_units</a> = linkonce global {} {}
700 %<a href="#format_program_objects">llvm.dbg.globals</a> = linkonce global {} {}
704 Using anchors in this way (where the source file descriptor points to the
705 anchors, as opposed to having a list of source file descriptors) allows for the
706 standard dead global elimination and merging passes to automatically remove
707 unused debugging information. If the globals were kept track of through lists,
708 there would always be an object pointing to the descriptors, thus would never be
715 <!-- ======================================================================= -->
716 <div class="doc_subsection">
717 <a name="format_common_stoppoint">
718 Representing stopping points in the source program
722 <div class="doc_text">
724 <p>LLVM debugger "stop points" are a key part of the debugging representation
725 that allows the LLVM to maintain simple semantics for <a
726 href="#debugopt">debugging optimized code</a>. The basic idea is that the
727 front-end inserts calls to the <tt>%llvm.dbg.stoppoint</tt> intrinsic function
728 at every point in the program where the debugger should be able to inspect the
729 program (these correspond to places the debugger stops when you "<tt>step</tt>"
730 through it). The front-end can choose to place these as fine-grained as it
731 would like (for example, before every subexpression evaluated), but it is
732 recommended to only put them after every source statement that includes
736 Using calls to this intrinsic function to demark legal points for the debugger
737 to inspect the program automatically disables any optimizations that could
738 potentially confuse debugging information. To non-debug-information-aware
739 transformations, these calls simply look like calls to an external function,
740 which they must assume to do anything (including reading or writing to any part
741 of reachable memory). On the other hand, it does not impact many optimizations,
742 such as code motion of non-trapping instructions, nor does it impact
743 optimization of subexpressions, code duplication transformations, or basic-block
744 reordering transformations.</p>
747 An important aspect of the calls to the <tt>%llvm.dbg.stoppoint</tt> intrinsic
748 is that the function-local debugging information is woven together with use-def
749 chains. This makes it easy for the debugger to, for example, locate the 'next'
750 stop point. For a concrete example of stop points, see the example in <a
751 href="#format_common_lifetime">the next section</a>.</p>
756 <!-- ======================================================================= -->
757 <div class="doc_subsection">
758 <a name="format_common_lifetime">Object lifetimes and scoping</a>
761 <div class="doc_text">
763 In many languages, the local variables in functions can have their lifetime or
764 scope limited to a subset of a function. In the C family of languages, for
765 example, variables are only live (readable and writable) within the source block
766 that they are defined in. In functional languages, values are only readable
767 after they have been defined. Though this is a very obvious concept, it is also
768 non-trivial to model in LLVM, because it has no notion of scoping in this sense,
769 and does not want to be tied to a language's scoping rules.
773 In order to handle this, the LLVM debug format uses the notion of "regions" of a
774 function, delineated by calls to intrinsic functions. These intrinsic functions
775 define new regions of the program and indicate when the region lifetime expires.
776 Consider the following C fragment, for example:
792 Compiled to LLVM, this function would be represented like this (FIXME: CHECK AND
801 <a name="#icl_ex_D1">%D1</a> = call {}* %llvm.dbg.func.start(<a href="#format_program_objects">%lldb.global</a>* %d.foo)
802 %D2 = call {}* <a href="#format_common_stoppoint">%llvm.dbg.stoppoint</a>({}* %D1, uint 2, uint 2, <a href="#format_common_source_files">%lldb.compile_unit</a>* %file)
804 %D3 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D2, ...)
805 <i>;; Evaluate expression on line 2, assigning to X.</i>
806 %D4 = call {}* <a href="#format_common_stoppoint">%llvm.dbg.stoppoint</a>({}* %D3, uint 3, uint 2, <a href="#format_common_source_files">%lldb.compile_unit</a>* %file)
808 %D5 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D4, ...)
809 <i>;; Evaluate expression on line 3, assigning to Y.</i>
810 %D6 = call {}* <a href="#format_common_stoppoint">%llvm.dbg.stoppoint</a>({}* %D5, uint 5, uint 4, <a href="#format_common_source_files">%lldb.compile_unit</a>* %file)
812 <a name="#icl_ex_D1">%D7</a> = call {}* %llvm.region.start({}* %D6)
813 %D8 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D7, ...)
814 <i>;; Evaluate expression on line 5, assigning to Z.</i>
815 %D9 = call {}* <a href="#format_common_stoppoint">%llvm.dbg.stoppoint</a>({}* %D8, uint 6, uint 4, <a href="#format_common_source_files">%lldb.compile_unit</a>* %file)
817 <i>;; Code for line 6.</i>
818 %D10 = call {}* %llvm.region.end({}* %D9)
819 %D11 = call {}* <a href="#format_common_stoppoint">%llvm.dbg.stoppoint</a>({}* %D10, uint 8, uint 2, <a href="#format_common_source_files">%lldb.compile_unit</a>* %file)
821 <i>;; Code for line 8.</i>
822 <a name="#icl_ex_D1">%D12</a> = call {}* %llvm.region.end({}* %D11)
828 This example illustrates a few important details about the LLVM debugging
829 information. In particular, it shows how the various intrinsics used are woven
830 together with def-use and use-def chains, similar to how <a
831 href="#format_common_anchors">anchors</a> are used with globals. This allows the
832 debugger to analyze the relationship between statements, variable definitions,
833 and the code used to implement the function.</p>
836 In this example, two explicit regions are defined, one with the <a
837 href="#icl_ex_D1">definition of the <tt>%D1</tt> variable</a> and one with the
838 <a href="#icl_ex_D7">definition of <tt>%D7</tt></a>. In the case of
839 <tt>%D1</tt>, the debug information indicates that the function whose <a
840 href="#format_program_objects">descriptor</a> is specified as an argument to the
841 intrinsic. This defines a new stack frame whose lifetime ends when the region
842 is ended by <a href="#icl_ex_D12">the <tt>%D12</tt> call</a>.</p>
845 Using regions to represent the boundaries of source-level functions allow LLVM
846 interprocedural optimizations to arbitrarily modify LLVM functions without
847 having to worry about breaking mapping information between the LLVM code and the
848 and source-level program. In particular, the inliner requires no modification
849 to support inlining with debugging information: there is no explicit correlation
850 drawn between LLVM functions and their source-level counterparts (note however,
851 that if the inliner inlines all instances of a non-strong-linkage function into
852 its caller that it will not be possible for the user to manually invoke the
853 inlined function from the debugger).</p>
856 Once the function has been defined, the <a
857 href="#format_common_stoppoint">stopping point</a> corresponding to line #2 of the
858 function is encountered. At this point in the function, <b>no</b> local
859 variables are live. As lines 2 and 3 of the example are executed, their
860 variable definitions are automatically introduced into the program, without the
861 need to specify a new region. These variables do not require new regions to be
862 introduced because they go out of scope at the same point in the program: line
867 In contrast, the <tt>Z</tt> variable goes out of scope at a different time, on
868 line 7. For this reason, it is defined within <a href="#icl_ex_D7">the
869 <tt>%D7</tt> region</a>, which kills the availability of <tt>Z</tt> before the
870 code for line 8 is executed. In this way, regions can support arbitrary
871 source-language scoping rules, as long as they can only be nested (ie, one scope
872 cannot partially overlap with a part of another scope).
876 It is worth noting that this scoping mechanism is used to control scoping of all
877 declarations, not just variable declarations. For example, the scope of a C++
878 using declaration is controlled with this, and the <tt>llvm-db</tt> C++ support
879 routines could use this to change how name lookup is performed (though this is
880 not implemented yet).
886 <!-- ======================================================================= -->
887 <div class="doc_subsection">
888 <a name="format_common_descriptors">Object descriptor formats</a>
891 <div class="doc_text">
893 The LLVM debugger expects the descriptors for program objects to start in a
894 canonical format, but the descriptors can include additional information
895 appended at the end that is source-language specific. All LLVM debugging
896 information is versioned, allowing backwards compatibility in the case that the
897 core structures need to change in some way. Also, all debugging information
898 objects start with a <a href="#format_common_tags">tag</a> to indicate what type
899 of object it is. The source-language is allows to define its own objects, by
900 using unreserved tag numbers.</p>
902 <p>The lowest-level descriptor are those describing <a
903 href="#format_common_source_files">the files containing the program source
904 code</a>, as most other descriptors (sometimes indirectly) refer to them.
909 <!----------------------------------------------------------------------------->
910 <div class="doc_subsubsection">
911 <a name="format_common_source_files">Representation of source files</a>
914 <div class="doc_text">
916 Source file descriptors are patterned after the Dwarf "compile_unit" object.
917 The descriptor currently is defined to have at least the following LLVM
921 %lldb.compile_unit = type {
922 uint, <i>;; Tag: <a href="#tag_compile_unit">LLVM_COMPILE_UNIT</a></i>
923 ushort, <i>;; LLVM debug version number</i>
924 ushort, <i>;; Dwarf language identifier</i>
925 sbyte*, <i>;; Filename</i>
926 sbyte*, <i>;; Working directory when compiled</i>
927 sbyte* <i>;; Producer of the debug information</i>
932 These descriptors contain the version number for the debug info, a source
933 language ID for the file (we use the Dwarf 3.0 ID numbers, such as
934 <tt>DW_LANG_C89</tt>, <tt>DW_LANG_C_plus_plus</tt>, <tt>DW_LANG_Cobol74</tt>,
935 etc), three strings describing the filename, working directory of the compiler,
936 and an identifier string for the compiler that produced it. Note that actual
937 compile_unit declarations must also include an <a
938 href="#format_common_anchors">anchor</a> to <tt>llvm.dbg.translation_units</tt>,
939 but it is not specified where the anchor is to be located. Here is an example
944 %arraytest_source_file = internal constant %lldb.compile_unit {
945 <a href="#tag_compile_unit">uint 17</a>, ; Tag value
946 ushort 0, ; Version #0
947 ushort 1, ; DW_LANG_C89
948 sbyte* getelementptr ([12 x sbyte]* %.str_1, long 0, long 0), ; filename
949 sbyte* getelementptr ([12 x sbyte]* %.str_2, long 0, long 0), ; working dir
950 sbyte* getelementptr ([12 x sbyte]* %.str_3, long 0, long 0), ; producer
951 {}* %llvm.dbg.translation_units ; Anchor
953 %.str_1 = internal constant [12 x sbyte] c"arraytest.c\00"
954 %.str_2 = internal constant [12 x sbyte] c"/home/sabre\00"
955 %.str_3 = internal constant [12 x sbyte] c"llvmgcc 3.4\00"
959 Note that the LLVM constant merging pass should eliminate duplicate copies of
960 the strings that get emitted to each translation unit, such as the producer.
966 <!----------------------------------------------------------------------------->
967 <div class="doc_subsubsection">
968 <a name="format_program_objects">Representation of program objects</a>
971 <div class="doc_text">
973 The LLVM debugger needs to know about some source-language program objects, in
974 order to build stack traces, print information about local variables, and other
975 related activities. The LLVM debugger differentiates between three different
976 types of program objects: subprograms (functions, messages, methods, etc),
977 variables (locals and globals), and others. Because source-languages have
978 widely varying forms of these objects, the LLVM debugger expects only a few
979 fields in the descriptor for each object:
983 %lldb.object = type {
984 uint, <i>;; <a href="#format_common_tag">A tag</a></i>
985 <i>any</i>*, <i>;; The <a href="#format_common_object_contexts">context</a> for the object</i>
986 sbyte* <i>;; The object 'name'</i>
991 The first field contains a tag for the descriptor. The second field contains
992 either a pointer to the descriptor for the containing <a
993 href="#format_common_source_files">source file</a>, or it contains a pointer to
994 another program object whose context pointer eventually reaches a source file.
995 Through this <a href="#format_common_object_contexts">context</a> pointer, the
996 LLVM debugger can establish the debug version number of the object.</p>
999 The third field contains a string that the debugger can use to identify the
1000 object if it does not contain explicit support for the source-language in use
1001 (ie, the 'unknown' source language handler uses this string). This should be
1002 some sort of unmangled string that corresponds to the object, but it is a
1003 quality of implementation issue what exactly it contains (it is legal, though
1004 not useful, for all of these strings to be null).
1008 Note again that descriptors can be extended to include source-language-specific
1009 information in addition to the fields required by the LLVM debugger. See the <a
1010 href="#ccxx_descriptors">section on the C/C++ front-end</a> for more
1011 information. Also remember that global objects (functions, selectors, global
1012 variables, etc) must contain an <a href="format_common_anchors">anchor</a> to
1013 the <tt>llvm.dbg.globals</tt> variable.
1018 <!-- ======================================================================= -->
1019 <div class="doc_subsection">
1020 <a name="format_common_object_contexts">Program object contexts</a>
1023 <div class="doc_text">
1025 Allow source-language specific contexts, use to identify namespaces etc
1026 Must end up in a source file descriptor.
1027 Debugger core ignores all unknown context objects.
1033 <!-- ======================================================================= -->
1034 <div class="doc_subsection">
1035 <a name="format_common_intrinsics">Debugger intrinsic functions</a>
1038 <div class="doc_text">
1040 Define each intrinsics, as an extension of the language reference manual.
1043 llvm.dbg.region.start
1045 llvm.dbg.function.start
1052 <!-- ======================================================================= -->
1053 <div class="doc_subsection">
1054 <a name="format_common_tags">Values for debugger tags</a>
1057 <div class="doc_text">
1060 Happen to be the same value as the similarly named Dwarf-3 tags, this may change
1066 <a name="tag_compile_unit">LLVM_COMPILE_UNIT</a> : 17
1067 <a name="tag_subprogram">LLVM_SUBPROGRAM</a> : 46
1068 <a name="tag_variable">LLVM_VARIABLE</a> : 52
1069 <!-- <a name="tag_formal_parameter">LLVM_FORMAL_PARAMETER : 5-->
1075 <!-- *********************************************************************** -->
1076 <div class="doc_section">
1077 <a name="ccxx_frontend">C/C++ front-end specific debug information</a>
1080 <div class="doc_text">
1083 The C and C++ front-ends represent information about the program in a format
1084 that is effectively identical to <a
1085 href="http://www.eagercon.com/dwarf/dwarf3std.htm">Dwarf 3.0</a> in terms of
1086 information content. This allows code generators to trivially support native
1087 debuggers by generating standard dwarf information, and contains enough
1088 information for non-dwarf targets to translate it as needed.</p>
1091 The basic debug information required by the debugger is (intentionally) designed
1092 to be as minimal as possible. This basic information is so minimal that it is
1093 unlikely that <b>any</b> source-language could be adequately described by it.
1094 Because of this, the debugger format was designed for extension to support
1095 source-language-specific information. The extended descriptors are read and
1096 interpreted by the <a href="#arch_info">language-specific</a> modules in the
1097 debugger if there is support available, otherwise it is ignored.
1101 This section describes the extensions used to represent C and C++ programs.
1102 Other languages could pattern themselves after this (which itself is tuned to
1103 representing programs in the same way that Dwarf 3 does), or they could choose
1104 to provide completely different extensions if they don't fit into the Dwarf
1105 model. As support for debugging information gets added to the various LLVM
1106 source-language front-ends, the information used should be documented here.
1111 <!-- ======================================================================= -->
1112 <div class="doc_subsection">
1113 <a name="ccxx_pse">Program Scope Entries</a>
1116 <div class="doc_text">
1122 <!----------------------------------------------------------------------------->
1123 <div class="doc_subsubsection">
1124 <a name="ccxx_compilation_units">Compilation unit entries</a>
1127 <div class="doc_text">
1129 Translation units do not add any information over the standard <a
1130 href="#format_common_source_files">source file representation</a> already
1131 expected by the debugger. As such, it uses descriptors of the type specified,
1132 with a trailing <a href="#format_common_anchors">anchor</a>.
1136 <!----------------------------------------------------------------------------->
1137 <div class="doc_subsubsection">
1138 <a name="ccxx_modules">Module, namespace, and importing entries</a>
1141 <div class="doc_text">
1147 <!-- ======================================================================= -->
1148 <div class="doc_subsection">
1149 <a name="ccxx_dataobjects">Data objects (program variables)</a>
1152 <div class="doc_text">
1159 <!-- *********************************************************************** -->
1161 <div class="doc_footer">
1162 <address><a href="mailto:sabre@nondot.org">Chris Lattner</a></address>
1163 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a>
1165 Last modified: $Date$