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>
12 <table border="0" width="100%">
17 <li><a href="#introduction">Introduction</a>
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>
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>
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>
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>
55 <li><a href="#ccxx_pse">Program Scope Entries</a>
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 <td align="right" valign="top">
65 <img src="img/venusflytrap.jpg" alt="A leafy and green bug eater" width="247"
72 <!-- *********************************************************************** -->
73 <div class="doc_section"><a name="introduction">Introduction</a></div> <!--
74 *********************************************************************** -->
76 <div class="doc_text">
78 <p>This document is the central repository for all information pertaining to
79 debug information in LLVM. It describes the <a href="#llvm-db">user
80 interface</a> for the <a href="CommandGuide/llvm-db.html"><tt>llvm-db</tt>
81 tool</a>, which provides a powerful <a href="#llvm-db">source-level debugger</a>
82 to users of LLVM-based compilers. It then describes the <a
83 href="#architecture">various components</a> that make up the debugger and the
84 libraries which future clients may use. Finally, it describes the <a
85 href="#format">actual format that the LLVM debug information</a> takes,
86 which is useful for those interested in creating front-ends or dealing directly
87 with the information.</p>
91 <!-- ======================================================================= -->
92 <div class="doc_subsection">
93 <a name="phil">Philosophy behind LLVM debugging information</a>
96 <div class="doc_text">
98 <p>The idea of the LLVM debugging information is to capture how the important
99 pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
100 Several design aspects have shaped the solution that appears here. The
101 important ones are:</p>
104 <li>Debugging information should have very little impact on the rest of the
105 compiler. No transformations, analyses, or code generators should need to be
106 modified because of debugging information.</li>
108 <li>LLVM optimizations should interact in <a href="#debugopt">well-defined and
109 easily described ways</a> with the debugging information.</li>
111 <li>Because LLVM is designed to support arbitrary programming languages,
112 LLVM-to-LLVM tools should not need to know anything about the semantics of the
113 source-level-language.</li>
115 <li>Source-level languages are often <b>widely</b> different from one another.
116 LLVM should not put any restrictions of the flavor of the source-language, and
117 the debugging information should work with any language.</li>
119 <li>With code generator support, it should be possible to use an LLVM compiler
120 to compile a program to native machine code and standard debugging formats.
121 This allows compatibility with traditional machine-code level debuggers, like
126 <p>The approach used by the LLVM implementation is to use a small set of <a
127 href="#format_common_intrinsics">intrinsic functions</a> to define a mapping
128 between LLVM program objects and the source-level objects. The description of
129 the source-level program is maintained in LLVM global variables in an <a
130 href="#ccxx_frontend">implementation-defined format</a> (the C/C++ front-end
131 currently uses working draft 7 of the <a
132 href="http://www.eagercon.com/dwarf/dwarf3std.htm">Dwarf 3 standard</a>).</p>
134 <p>When a program is debugged, the debugger interacts with the user and turns
135 the stored debug information into source-language specific information. As
136 such, the debugger must be aware of the source-language, and is thus tied to a
137 specific language of family of languages. The <a href="#llvm-db">LLVM
138 debugger</a> is designed to be modular in its support for source-languages.</p>
143 <!-- ======================================================================= -->
144 <div class="doc_subsection">
145 <a name="debugopt">Debugging optimized code</a>
148 <div class="doc_text">
150 <p>An extremely high priority of LLVM debugging information is to make it
151 interact well with optimizations and analysis. In particular, the LLVM debug
152 information provides the following guarantees:</p>
156 <li>LLVM debug information <b>always provides information to accurately read the
157 source-level state of the program</b>, regardless of which LLVM optimizations
158 have been run, and without any modification to the optimizations themselves.
159 However, some optimizations may impact the ability to modify the current state
160 of the program with a debugger, such as setting program variables, or calling
161 function that have been deleted.</li>
163 <li>LLVM optimizations gracefully interact with debugging information. If they
164 are not aware of debug information, they are automatically disabled as necessary
165 in the cases that would invalidate the debug info. This retains the LLVM
166 features making it easy to write new transformations.</li>
168 <li>As desired, LLVM optimizations can be upgraded to be aware of the LLVM
169 debugging information, allowing them to update the debugging information as they
170 perform aggressive optimizations. This means that, with effort, the LLVM
171 optimizers could optimize debug code just as well as non-debug code.</li>
173 <li>LLVM debug information does not prevent many important optimizations from
174 happening (for example inlining, basic block reordering/merging/cleanup, tail
175 duplication, etc), further reducing the amount of the compiler that eventually
176 is "aware" of debugging information.</li>
178 <li>LLVM debug information is automatically optimized along with the rest of the
179 program, using existing facilities. For example, duplicate information is
180 automatically merged by the linker, and unused information is automatically
185 <p>Basically, the debug information allows you to compile a program with
186 "<tt>-O0 -g</tt>" and get full debug information, allowing you to arbitrarily
187 modify the program as it executes from the debugger. Compiling a program with
188 "<tt>-O3 -g</tt>" gives you full debug information that is always available and
189 accurate for reading (e.g., you get accurate stack traces despite tail call
190 elimination and inlining), but you might lose the ability to modify the program
191 and call functions where were optimized out of the program, or inlined away
196 <!-- ======================================================================= -->
197 <div class="doc_subsection">
198 <a name="future">Future work</a>
201 <div class="doc_text">
202 <p>There are several important extensions that could be eventually added to the
203 LLVM debugger. The most important extension would be to upgrade the LLVM code
204 generators to support debugging information. This would also allow, for
205 example, the X86 code generator to emit native objects that contain debugging
206 information consumable by traditional source-level debuggers like GDB or
209 <p>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>
213 <p>The "SourceLanguage" modules provided by <tt>llvm-db</tt> could be
214 substantially improved to provide good support for C++ language features like
215 namespaces and scoping rules.</p>
217 <p>After working with the debugger for a while, perhaps the nicest improvement
218 would be to add some sort of line editor, such as GNU readline (but one that is
219 compatible with the LLVM license).</p>
221 <p>For someone so inclined, it should be straight-forward to write different
222 front-ends for the LLVM debugger, as the LLVM debugging engine is cleanly
223 separated from the <tt>llvm-db</tt> front-end. A new LLVM GUI debugger or IDE
224 would be nice. :)</p>
228 <!-- *********************************************************************** -->
229 <div class="doc_section">
230 <a name="llvm-db">Using the <tt>llvm-db</tt> tool</a>
232 <!-- *********************************************************************** -->
234 <div class="doc_text">
236 <p>The <tt>llvm-db</tt> tool provides a GDB-like interface for source-level
237 debugging of programs. This tool provides many standard commands for inspecting
238 and modifying the program as it executes, loading new programs, single stepping,
239 placing breakpoints, etc. This section describes how to use the debugger.</p>
241 <p><tt>llvm-db</tt> has been designed to be as similar to GDB in its user
242 interface as possible. This should make it extremely easy to learn
243 <tt>llvm-db</tt> if you already know <tt>GDB</tt>. In general, <tt>llvm-db</tt>
244 provides the subset of GDB commands that are applicable to LLVM debugging users.
245 If there is a command missing that make a reasonable amount of sense within the
246 <a href="#limitations">limitations of <tt>llvm-db</tt></a>, please report it as
247 a bug or, better yet, submit a patch to add it. :)</p>
251 <!-- ======================================================================= -->
252 <div class="doc_subsection">
253 <a name="limitations">Limitations of <tt>llvm-db</tt></a>
256 <div class="doc_text">
258 <p><tt>llvm-db</tt> is designed to be modular and easy to extend. This
259 extensibility was key to getting the debugger up-and-running quickly, because we
260 can start with simple-but-unsophisicated implementations of various components.
261 Because of this, it is currently missing many features, though they should be
262 easy to add over time (patches welcomed!). The biggest inherent limitations of
263 <tt>llvm-db</tt> are currently due to extremely simple <a
264 href="#arch_debugger">debugger backend</a> (implemented in
265 "lib/Debugger/UnixLocalInferiorProcess.cpp") which is designed to work without
266 any cooperation from the code generators. Because it is so simple, it suffers
267 from the following inherent limitations:</p>
271 <li>Running a program in <tt>llvm-db</tt> is a bit slower than running it with
272 <tt>lli</tt> (i.e., in the JIT).</li>
274 <li>Inspection of the target hardware is not supported. This means that you
275 cannot, for example, print the contents of X86 registers.</li>
277 <li>Inspection of LLVM code is not supported. This means that you cannot print
278 the contents of arbitrary LLVM values, or use commands such as <tt>stepi</tt>.
279 This also means that you cannot debug code without debug information.</li>
281 <li>Portions of the debugger run in the same address space as the program being
282 debugged. This means that memory corruption by the program could trample on
283 portions of the debugger.</li>
285 <li>Attaching to existing processes and core files is not currently
290 <p>That said, the debugger is still quite useful, and all of these limitations
291 can be eliminated by integrating support for the debugger into the code
292 generators, and writing a new <a href="#arch_debugger">InferiorProcess</a>
293 subclass to use it. See the <a href="#future">future work</a> section for ideas
294 of how to extend the LLVM debugger despite these limitations.</p>
299 <!-- ======================================================================= -->
300 <div class="doc_subsection">
301 <a name="sample">A sample <tt>llvm-db</tt> session</a>
304 <div class="doc_text">
306 <p>TODO: this is obviously lame, when more is implemented, this can be much
310 $ <b>llvm-db funccall</b>
311 llvm-db: The LLVM source-level debugger
312 Loading program... successfully loaded 'funccall.bc'!
313 (llvm-db) <b>create</b>
314 Starting program: funccall.bc
315 main at funccall.c:9:2
317 (llvm-db) <b>list main</b>
328 (llvm-db) <b>list</b>
330 (llvm-db) <b>step</b>
333 foo at funccall.c:5:2
336 #0 -> 0x85ffba0 in foo at funccall.c:5:2
337 #1 0x85ffd98 in main at funccall.c:10:2
338 (llvm-db) <b>finish</b>
339 main at funccall.c:11:2
344 The program stopped with exit code 0
345 (llvm-db) <b>quit</b>
353 <!-- ======================================================================= -->
354 <div class="doc_subsection">
355 <a name="startup">Starting the debugger</a>
358 <div class="doc_text">
360 <p>There are three ways to start up the <tt>llvm-db</tt> debugger:</p>
362 <p>When run with no options, just <tt>llvm-db</tt>, the debugger starts up
363 without a program loaded at all. You must use the <a
364 href="#c_file"><tt>file</tt> command</a> to load a program, and the <a
365 href="c_set_args"><tt>set args</tt></a> or <a href="#c_run"><tt>run</tt></a>
366 commands to specify the arguments for the program.</p>
368 <p>If you start the debugger with one argument, as <tt>llvm-db
369 <program></tt>, the debugger will start up and load in the specified
370 program. You can then optionally specify arguments to the program with the <a
371 href="c_set_args"><tt>set args</tt></a> or <a href="#c_run"><tt>run</tt></a>
374 <p>The third way to start the program is with the <tt>--args</tt> option. This
375 option allows you to specify the program to load and the arguments to start out
376 with. <!-- No options to <tt>llvm-db</tt> may be specified after the
377 <tt>-args</tt> option. --> Example use: <tt>llvm-db --args ls /home</tt></p>
381 <!-- ======================================================================= -->
382 <div class="doc_subsection">
383 <a name="commands">Commands recognized by the debugger</a>
386 <div class="doc_text">
388 <p>FIXME: this needs work obviously. See the <a
389 href="http://sources.redhat.com/gdb/documentation/">GDB documentation</a> for
390 information about what these do, or try '<tt>help [command]</tt>' within
391 <tt>llvm-db</tt> to get information.</p>
394 <h2>General usage:</h2>
396 <li>help [command]</li>
398 <li><a name="c_file">file</a> [program]</li>
401 <h2>Program inspection and interaction:</h2>
403 <li>create (start the program, stopping it ASAP in <tt>main</tt>)</li>
411 <li>list [start[, end]]</li>
413 <li>info sources</li>
414 <li>info functions</li>
417 <h2>Call stack inspection:</h2>
426 <h2>Debugger inspection and interaction:</h2>
431 <li>show listsize</li>
432 <li>set listsize</li>
433 <li>show language</li>
434 <li>set language</li>
436 <li>set args [args]</li>
447 <li>info variables</li>
448 <li>info program</li>
453 <li>... many others</li>
458 <!-- *********************************************************************** -->
459 <div class="doc_section">
460 <a name="architecture">Architecture of the LLVM debugger</a>
462 <!-- *********************************************************************** -->
464 <div class="doc_text">
465 <p>The LLVM debugger is built out of three distinct layers of software. These
466 layers provide clients with different interface options depending on what pieces
467 of they want to implement themselves, and it also promotes code modularity and
468 good design. The three layers are the <a href="#arch_debugger">Debugger
469 interface</a>, the <a href="#arch_info">"info" interfaces</a>, and the <a
470 href="#arch_llvm-db"><tt>llvm-db</tt> tool</a> itself.</p>
473 <!-- ======================================================================= -->
474 <div class="doc_subsection">
475 <a name="arch_debugger">The Debugger and InferiorProcess classes</a>
478 <div class="doc_text">
479 <p>The Debugger class (defined in the <tt>include/llvm/Debugger/</tt> directory)
480 is a low-level class which is used to maintain information about the loaded
481 program, as well as start and stop the program running as necessary. This class
482 does not provide any high-level analysis or control over the program, only
483 exposing simple interfaces like <tt>load/unloadProgram</tt>,
484 <tt>create/killProgram</tt>, <tt>step/next/finish/contProgram</tt>, and
485 low-level methods for installing breakpoints.</p>
488 The Debugger class is itself a wrapper around the lowest-level InferiorProcess
489 class. This class is used to represent an instance of the program running under
490 debugger control. The InferiorProcess class can be implemented in different
491 ways for different targets and execution scenarios (e.g., remote debugging).
492 The InferiorProcess class exposes a small and simple collection of interfaces
493 which are useful for inspecting the current state of the program (such as
494 collecting stack trace information, reading the memory image of the process,
495 etc). The interfaces in this class are designed to be as low-level and simple
496 as possible, to make it easy to create new instances of the class.
500 The Debugger class exposes the currently active instance of InferiorProcess
501 through the <tt>Debugger::getRunningProcess</tt> method, which returns a
502 <tt>const</tt> reference to the class. This means that clients of the Debugger
503 class can only <b>inspect</b> the running instance of the program directly. To
504 change the executing process in some way, they must use the interces exposed by
509 <!-- ======================================================================= -->
510 <div class="doc_subsection">
511 <a name="arch_info">The RuntimeInfo, ProgramInfo, and SourceLanguage classes</a>
514 <div class="doc_text">
516 The next-highest level of debugger abstraction is provided through the
517 ProgramInfo, RuntimeInfo, SourceLanguage and related classes (also defined in
518 the <tt>include/llvm/Debugger/</tt> directory). These classes efficiently
519 decode the debugging information and low-level interfaces exposed by
520 InferiorProcess into a higher-level representation, suitable for analysis by the
525 The ProgramInfo class exposes a variety of different kinds of information about
526 the program objects in the source-level-language. The SourceFileInfo class
527 represents a source-file in the program (e.g. a .cpp or .h file). The
528 SourceFileInfo class captures information such as which SourceLanguage was used
529 to compile the file, where the debugger can get access to the actual file text
530 (which is lazily loaded on demand), etc. The SourceFunctionInfo class
531 represents a... <b>FIXME: finish</b>. The ProgramInfo class provides interfaces
532 to lazily find and decode the information needed to create the Source*Info
533 classes requested by the debugger.
537 The RuntimeInfo class exposes information about the currently executed program,
538 by decoding information from the InferiorProcess and ProgramInfo classes. It
539 provides a StackFrame class which provides an easy-to-use interface for
540 inspecting the current and suspended stack frames in the program.
544 The SourceLanguage class is an abstract interface used by the debugger to
545 perform all source-language-specific tasks. For example, this interface is used
546 by the ProgramInfo class to decode language-specific types and functions and by
547 the debugger front-end (such as <a href="#arch_llvm-db"><tt>llvm-db</tt></a> to
548 evaluate source-langauge expressions typed into the debugger. This class uses
549 the RuntimeInfo & ProgramInfo classes to get information about the current
550 execution context and the loaded program, respectively.
555 <!-- ======================================================================= -->
556 <div class="doc_subsection">
557 <a name="arch_llvm-db">The <tt>llvm-db</tt> tool</a>
560 <div class="doc_text">
562 The <tt>llvm-db</tt> is designed to be a debugger providing an interface as <a
563 href="#llvm-db">similar to GDB</a> as reasonable, but no more so than that.
564 Because the <a href="#arch_debugger">Debugger</a> and <a
565 href="#arch_info">info</a> classes implement all of the heavy lifting and
566 analysis, <tt>llvm-db</tt> (which lives in <tt>llvm/tools/llvm-db</tt>) consists
567 mainly of of code to interact with the user and parse commands. The CLIDebugger
568 constructor registers all of the builtin commands for the debugger, and each
569 command is implemented as a CLIDebugger::[name]Command method.
574 <!-- ======================================================================= -->
575 <div class="doc_subsection">
576 <a name="arch_todo">Short-term TODO list</a>
579 <div class="doc_text">
582 FIXME: this section will eventually go away. These are notes to myself of
583 things that should be implemented, but haven't yet.
587 <b>Breakpoints:</b> Support is already implemented in the 'InferiorProcess'
588 class, though it hasn't been tested yet. To finish breakpoint support, we need
589 to implement breakCommand (which should reuse the linespec parser from the list
590 command), and handle the fact that 'break foo' or 'break file.c:53' may insert
591 multiple breakpoints. Also, if you say 'break file.c:53' and there is no
592 stoppoint on line 53, the breakpoint should go on the next available line. My
593 idea was to have the Debugger class provide a "Breakpoint" class which
594 encapsulated this messiness, giving the debugger front-end a simple interface.
595 The debugger front-end would have to map the really complex semantics of
596 temporary breakpoints and 'conditional' breakpoints onto this intermediate
597 level. Also, breakpoints should survive as much as possible across program
602 <b>UnixLocalInferiorProcess.cpp speedup</b>: There is no reason for the debugged
603 process to code gen the globals corresponding to debug information. The
604 IntrinsicLowering object could instead change descriptors into constant expr
605 casts of the constant address of the LLVM objects for the descriptors. This
606 would also allow us to eliminate the mapping back and forth between physical
607 addresses that must be done.</p>
610 <b>Process deaths</b>: The InferiorProcessDead exception should be extended to
611 know "how" a process died, i.e., it was killed by a signal. This is easy to
612 collect in the UnixLocalInferiorProcess, we just need to represent it.</p>
616 <!-- *********************************************************************** -->
617 <div class="doc_section">
618 <a name="format">Debugging information format</a>
620 <!-- *********************************************************************** -->
622 <div class="doc_text">
624 <p>LLVM debugging information has been carefully designed to make it possible
625 for the optimizer to optimize the program and debugging information without
626 necessarily having to know anything about debugging information. In particular,
627 the global constant merging pass automatically eliminates duplicated debugging
628 information (often caused by header files), the global dead code elimination
629 pass automatically deletes debugging information for a function if it decides to
630 delete the function, and the linker eliminates debug information when it merges
631 <tt>linkonce</tt> functions.</p>
633 <p>To do this, most of the debugging information (descriptors for types,
634 variables, functions, source files, etc) is inserted by the language front-end
635 in the form of LLVM global variables. These LLVM global variables are no
636 different from any other global variables, except that they have a web of LLVM
637 intrinsic functions that point to them. If the last references to a particular
638 piece of debugging information are deleted (for example, by the
639 <tt>-globaldce</tt> pass), the extraneous debug information will automatically
640 become dead and be removed by the optimizer.</p>
642 <p>The debugger is designed to be agnostic about the contents of most of the
643 debugging information. It uses a <a href="#arch_info">source-language-specific
644 module</a> to decode the information that represents variables, types,
645 functions, namespaces, etc: this allows for arbitrary source-language semantics
646 and type-systems to be used, as long as there is a module written for the
647 debugger to interpret the information.</p>
649 <p>To provide basic functionality, the LLVM debugger does have to make some
650 assumptions about the source-level language being debugged, though it keeps
651 these to a minimum. The only common features that the LLVM debugger assumes
652 exist are <a href="#format_common_source_files">source files</a>, and <a
653 href="#format_program_objects">program objects</a>. These abstract objects are
654 used by the debugger to form stack traces, show information about local
657 <p>This section of the documentation first describes the representation aspects
658 common to any source-language. The <a href="#ccxx_frontend">next section</a>
659 describes the data layout conventions used by the C and C++ front-ends.</p>
663 <!-- ======================================================================= -->
664 <div class="doc_subsection">
665 <a name="format_common_anchors">Anchors for global objects</a>
668 <div class="doc_text">
669 <p>One important aspect of the LLVM debug representation is that it allows the
670 LLVM debugger to efficiently index all of the global objects without having the
671 scan the program. To do this, all of the global objects use "anchor" globals of
672 type "<tt>{}</tt>", with designated names. These anchor objects obviously do
673 not contain any content or meaning by themselves, but all of the global objects
674 of a particular type (e.g., source file descriptors) contain a pointer to the
675 anchor. This pointer allows the debugger to use def-use chains to find all
676 global objects of that type.</p>
678 <p>So far, the following names are recognized as anchors by the LLVM
682 %<a href="#format_common_source_files">llvm.dbg.translation_units</a> = linkonce global {} {}
683 %<a href="#format_program_objects">llvm.dbg.globals</a> = linkonce global {} {}
686 <p>Using anchors in this way (where the source file descriptor points to the
687 anchors, as opposed to having a list of source file descriptors) allows for the
688 standard dead global elimination and merging passes to automatically remove
689 unused debugging information. If the globals were kept track of through lists,
690 there would always be an object pointing to the descriptors, thus would never be
695 <!-- ======================================================================= -->
696 <div class="doc_subsection">
697 <a name="format_common_stoppoint">
698 Representing stopping points in the source program
702 <div class="doc_text">
704 <p>LLVM debugger "stop points" are a key part of the debugging representation
705 that allows the LLVM to maintain simple semantics for <a
706 href="#debugopt">debugging optimized code</a>. The basic idea is that the
707 front-end inserts calls to the <tt>%llvm.dbg.stoppoint</tt> intrinsic function
708 at every point in the program where the debugger should be able to inspect the
709 program (these correspond to places the debugger stops when you "<tt>step</tt>"
710 through it). The front-end can choose to place these as fine-grained as it
711 would like (for example, before every subexpression evaluated), but it is
712 recommended to only put them after every source statement that includes
715 <p>Using calls to this intrinsic function to demark legal points for the
716 debugger to inspect the program automatically disables any optimizations that
717 could potentially confuse debugging information. To non-debug-information-aware
718 transformations, these calls simply look like calls to an external function,
719 which they must assume to do anything (including reading or writing to any part
720 of reachable memory). On the other hand, it does not impact many optimizations,
721 such as code motion of non-trapping instructions, nor does it impact
722 optimization of subexpressions, code duplication transformations, or basic-block
723 reordering transformations.</p>
725 <p>An important aspect of the calls to the <tt>%llvm.dbg.stoppoint</tt>
726 intrinsic is that the function-local debugging information is woven together
727 with use-def chains. This makes it easy for the debugger to, for example,
728 locate the 'next' stop point. For a concrete example of stop points, see the
729 example in <a href="#format_common_lifetime">the next section</a>.</p>
734 <!-- ======================================================================= -->
735 <div class="doc_subsection">
736 <a name="format_common_lifetime">Object lifetimes and scoping</a>
739 <div class="doc_text">
740 <p>In many languages, the local variables in functions can have their lifetime
741 or scope limited to a subset of a function. In the C family of languages, for
742 example, variables are only live (readable and writable) within the source block
743 that they are defined in. In functional languages, values are only readable
744 after they have been defined. Though this is a very obvious concept, it is also
745 non-trivial to model in LLVM, because it has no notion of scoping in this sense,
746 and does not want to be tied to a language's scoping rules.</p>
748 <p>In order to handle this, the LLVM debug format uses the notion of "regions"
749 of a function, delineated by calls to intrinsic functions. These intrinsic
750 functions define new regions of the program and indicate when the region
751 lifetime expires. Consider the following C fragment, for example:</p>
765 <p>Compiled to LLVM, this function would be represented like this (FIXME: CHECK
766 AND UPDATE THIS):</p>
773 <a name="#icl_ex_D1">%D1</a> = call {}* %llvm.dbg.func.start(<a href="#format_program_objects">%lldb.global</a>* %d.foo)
774 %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)
776 %D3 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D2, ...)
777 <i>;; Evaluate expression on line 2, assigning to X.</i>
778 %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)
780 %D5 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D4, ...)
781 <i>;; Evaluate expression on line 3, assigning to Y.</i>
782 %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)
784 <a name="#icl_ex_D1">%D7</a> = call {}* %llvm.region.start({}* %D6)
785 %D8 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D7, ...)
786 <i>;; Evaluate expression on line 5, assigning to Z.</i>
787 %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)
789 <i>;; Code for line 6.</i>
790 %D10 = call {}* %llvm.region.end({}* %D9)
791 %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)
793 <i>;; Code for line 8.</i>
794 <a name="#icl_ex_D1">%D12</a> = call {}* %llvm.region.end({}* %D11)
799 <p>This example illustrates a few important details about the LLVM debugging
800 information. In particular, it shows how the various intrinsics used are woven
801 together with def-use and use-def chains, similar to how <a
802 href="#format_common_anchors">anchors</a> are used with globals. This allows
803 the debugger to analyze the relationship between statements, variable
804 definitions, and the code used to implement the function.</p>
806 <p>In this example, two explicit regions are defined, one with the <a
807 href="#icl_ex_D1">definition of the <tt>%D1</tt> variable</a> and one with the
808 <a href="#icl_ex_D7">definition of <tt>%D7</tt></a>. In the case of
809 <tt>%D1</tt>, the debug information indicates that the function whose <a
810 href="#format_program_objects">descriptor</a> is specified as an argument to the
811 intrinsic. This defines a new stack frame whose lifetime ends when the region
812 is ended by <a href="#icl_ex_D12">the <tt>%D12</tt> call</a>.</p>
814 <p>Using regions to represent the boundaries of source-level functions allow
815 LLVM interprocedural optimizations to arbitrarily modify LLVM functions without
816 having to worry about breaking mapping information between the LLVM code and the
817 and source-level program. In particular, the inliner requires no modification
818 to support inlining with debugging information: there is no explicit correlation
819 drawn between LLVM functions and their source-level counterparts (note however,
820 that if the inliner inlines all instances of a non-strong-linkage function into
821 its caller that it will not be possible for the user to manually invoke the
822 inlined function from the debugger).</p>
824 <p>Once the function has been defined, the <a
825 href="#format_common_stoppoint">stopping point</a> corresponding to line #2 of
826 the function is encountered. At this point in the function, <b>no</b> local
827 variables are live. As lines 2 and 3 of the example are executed, their
828 variable definitions are automatically introduced into the program, without the
829 need to specify a new region. These variables do not require new regions to be
830 introduced because they go out of scope at the same point in the program: line
833 <p>In contrast, the <tt>Z</tt> variable goes out of scope at a different time,
834 on line 7. For this reason, it is defined within <a href="#icl_ex_D7">the
835 <tt>%D7</tt> region</a>, which kills the availability of <tt>Z</tt> before the
836 code for line 8 is executed. In this way, regions can support arbitrary
837 source-language scoping rules, as long as they can only be nested (ie, one scope
838 cannot partially overlap with a part of another scope).</p>
840 <p>It is worth noting that this scoping mechanism is used to control scoping of
841 all declarations, not just variable declarations. For example, the scope of a
842 C++ using declaration is controlled with this, and the <tt>llvm-db</tt> C++
843 support routines could use this to change how name lookup is performed (though
844 this is not implemented yet).</p>
848 <!-- ======================================================================= -->
849 <div class="doc_subsection">
850 <a name="format_common_descriptors">Object descriptor formats</a>
853 <div class="doc_text">
854 <p>The LLVM debugger expects the descriptors for program objects to start in a
855 canonical format, but the descriptors can include additional information
856 appended at the end that is source-language specific. All LLVM debugging
857 information is versioned, allowing backwards compatibility in the case that the
858 core structures need to change in some way. Also, all debugging information
859 objects start with a <a href="#format_common_tags">tag</a> to indicate what type
860 of object it is. The source-language is allows to define its own objects, by
861 using unreserved tag numbers.</p>
863 <p>The lowest-level descriptor are those describing <a
864 href="#format_common_source_files">the files containing the program source
865 code</a>, as most other descriptors (sometimes indirectly) refer to them.
870 <!-- ------------------------------------------------------------------------ ->
871 <div class="doc_subsubsection">
872 <a name="format_common_source_files">Representation of source files</a>
875 <div class="doc_text">
877 Source file descriptors are patterned after the Dwarf "compile_unit" object.
878 The descriptor currently is defined to have at least the following LLVM
882 %lldb.compile_unit = type {
883 uint, <i>;; Tag: <a href="#tag_compile_unit">LLVM_COMPILE_UNIT</a></i>
884 ushort, <i>;; LLVM debug version number</i>
885 ushort, <i>;; Dwarf language identifier</i>
886 sbyte*, <i>;; Filename</i>
887 sbyte*, <i>;; Working directory when compiled</i>
888 sbyte* <i>;; Producer of the debug information</i>
893 These descriptors contain the version number for the debug info, a source
894 language ID for the file (we use the Dwarf 3.0 ID numbers, such as
895 <tt>DW_LANG_C89</tt>, <tt>DW_LANG_C_plus_plus</tt>, <tt>DW_LANG_Cobol74</tt>,
896 etc), three strings describing the filename, working directory of the compiler,
897 and an identifier string for the compiler that produced it. Note that actual
898 compile_unit declarations must also include an <a
899 href="#format_common_anchors">anchor</a> to <tt>llvm.dbg.translation_units</tt>,
900 but it is not specified where the anchor is to be located. Here is an example
905 %arraytest_source_file = internal constant %lldb.compile_unit {
906 <a href="#tag_compile_unit">uint 17</a>, ; Tag value
907 ushort 0, ; Version #0
908 ushort 1, ; DW_LANG_C89
909 sbyte* getelementptr ([12 x sbyte]* %.str_1, long 0, long 0), ; filename
910 sbyte* getelementptr ([12 x sbyte]* %.str_2, long 0, long 0), ; working dir
911 sbyte* getelementptr ([12 x sbyte]* %.str_3, long 0, long 0), ; producer
912 {}* %llvm.dbg.translation_units ; Anchor
914 %.str_1 = internal constant [12 x sbyte] c"arraytest.c\00"
915 %.str_2 = internal constant [12 x sbyte] c"/home/sabre\00"
916 %.str_3 = internal constant [12 x sbyte] c"llvmgcc 3.4\00"
920 Note that the LLVM constant merging pass should eliminate duplicate copies of
921 the strings that get emitted to each translation unit, such as the producer.
927 <!-- ----------------------------------------------------------------------- -->
928 <div class="doc_subsubsection">
929 <a name="format_program_objects">Representation of program objects</a>
932 <div class="doc_text">
934 The LLVM debugger needs to know about some source-language program objects, in
935 order to build stack traces, print information about local variables, and other
936 related activities. The LLVM debugger differentiates between three different
937 types of program objects: subprograms (functions, messages, methods, etc),
938 variables (locals and globals), and others. Because source-languages have
939 widely varying forms of these objects, the LLVM debugger expects only a few
940 fields in the descriptor for each object:
944 %lldb.object = type {
945 uint, <i>;; <a href="#format_common_tag">A tag</a></i>
946 <i>any</i>*, <i>;; The <a href="#format_common_object_contexts">context</a> for the object</i>
947 sbyte* <i>;; The object 'name'</i>
951 <p>The first field contains a tag for the descriptor. The second field contains
952 either a pointer to the descriptor for the containing <a
953 href="#format_common_source_files">source file</a>, or it contains a pointer to
954 another program object whose context pointer eventually reaches a source file.
955 Through this <a href="#format_common_object_contexts">context</a> pointer, the
956 LLVM debugger can establish the debug version number of the object.</p>
958 <p>The third field contains a string that the debugger can use to identify the
959 object if it does not contain explicit support for the source-language in use
960 (ie, the 'unknown' source language handler uses this string). This should be
961 some sort of unmangled string that corresponds to the object, but it is a
962 quality of implementation issue what exactly it contains (it is legal, though
963 not useful, for all of these strings to be null).</p>
965 <p>Note again that descriptors can be extended to include
966 source-language-specific information in addition to the fields required by the
967 LLVM debugger. See the <a href="#ccxx_descriptors">section on the C/C++
968 front-end</a> for more information. Also remember that global objects
969 (functions, selectors, global variables, etc) must contain an <a
970 href="format_common_anchors">anchor</a> to the <tt>llvm.dbg.globals</tt>
975 <!-- ======================================================================= -->
976 <div class="doc_subsection">
977 <a name="format_common_object_contexts">Program object contexts</a>
980 <div class="doc_text">
982 Allow source-language specific contexts, use to identify namespaces etc
983 Must end up in a source file descriptor.
984 Debugger core ignores all unknown context objects.
988 <!-- ======================================================================= -->
989 <div class="doc_subsection">
990 <a name="format_common_intrinsics">Debugger intrinsic functions</a>
993 <div class="doc_text">
995 Define each intrinsics, as an extension of the language reference manual.
998 llvm.dbg.region.start
1000 llvm.dbg.function.start
1005 <!-- ======================================================================= -->
1006 <div class="doc_subsection">
1007 <a name="format_common_tags">Values for debugger tags</a>
1010 <div class="doc_text">
1012 <p>Happen to be the same value as the similarly named Dwarf-3 tags, this may
1013 change in the future.</p>
1016 <a name="tag_compile_unit">LLVM_COMPILE_UNIT</a> : 17
1017 <a name="tag_subprogram">LLVM_SUBPROGRAM</a> : 46
1018 <a name="tag_variable">LLVM_VARIABLE</a> : 52
1019 <!-- <a name="tag_formal_parameter">LLVM_FORMAL_PARAMETER : 5-->
1025 <!-- *********************************************************************** -->
1026 <div class="doc_section">
1027 <a name="ccxx_frontend">C/C++ front-end specific debug information</a>
1030 <div class="doc_text">
1032 <p>The C and C++ front-ends represent information about the program in a format
1033 that is effectively identical to <a
1034 href="http://www.eagercon.com/dwarf/dwarf3std.htm">Dwarf 3.0</a> in terms of
1035 information content. This allows code generators to trivially support native
1036 debuggers by generating standard dwarf information, and contains enough
1037 information for non-dwarf targets to translate it as needed.</p>
1039 <p>The basic debug information required by the debugger is (intentionally)
1040 designed to be as minimal as possible. This basic information is so minimal
1041 that it is unlikely that <b>any</b> source-language could be adequately
1042 described by it. Because of this, the debugger format was designed for
1043 extension to support source-language-specific information. The extended
1044 descriptors are read and interpreted by the <a
1045 href="#arch_info">language-specific</a> modules in the debugger if there is
1046 support available, otherwise it is ignored.</p>
1048 <p>This section describes the extensions used to represent C and C++ programs.
1049 Other languages could pattern themselves after this (which itself is tuned to
1050 representing programs in the same way that Dwarf 3 does), or they could choose
1051 to provide completely different extensions if they don't fit into the Dwarf
1052 model. As support for debugging information gets added to the various LLVM
1053 source-language front-ends, the information used should be documented here.</p>
1057 <!-- ======================================================================= -->
1058 <div class="doc_subsection">
1059 <a name="ccxx_pse">Program Scope Entries</a>
1062 <div class="doc_text">
1066 <!-- -------------------------------------------------------------------------->
1067 <div class="doc_subsubsection">
1068 <a name="ccxx_compilation_units">Compilation unit entries</a>
1071 <div class="doc_text">
1073 Translation units do not add any information over the standard <a
1074 href="#format_common_source_files">source file representation</a> already
1075 expected by the debugger. As such, it uses descriptors of the type specified,
1076 with a trailing <a href="#format_common_anchors">anchor</a>.
1080 <!-- -------------------------------------------------------------------------->
1081 <div class="doc_subsubsection">
1082 <a name="ccxx_modules">Module, namespace, and importing entries</a>
1085 <div class="doc_text">
1089 <!-- ======================================================================= -->
1090 <div class="doc_subsection">
1091 <a name="ccxx_dataobjects">Data objects (program variables)</a>
1094 <div class="doc_text">
1099 <!-- *********************************************************************** -->
1103 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1104 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1105 <a href="http://validator.w3.org/check/referer"><img
1106 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1108 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1109 <a href="http://llvm.cs.uiuc.edu">LLVM Compiler Infrastructure</a><br>
1110 Last modified: $Date$