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 class="layout" style="width:100%">
16 <li><a href="#introduction">Introduction</a>
18 <li><a href="#phil">Philosophy behind LLVM debugging information</a></li>
19 <li><a href="#debugopt">Debugging optimized code</a></li>
20 <li><a href="#future">Future work</a></li>
22 <li><a href="#llvm-db">Using the <tt>llvm-db</tt> tool</a>
24 <li><a href="#limitations">Limitations of <tt>llvm-db</tt></a></li>
25 <li><a href="#sample">A sample <tt>llvm-db</tt> session</a></li>
26 <li><a href="#startup">Starting the debugger</a></li>
27 <li><a href="#commands">Commands recognized by the debugger</a></li>
30 <li><a href="#architecture">Architecture of the LLVM debugger</a>
32 <li><a href="#arch_debugger">The Debugger and InferiorProcess classes</a></li>
33 <li><a href="#arch_info">The RuntimeInfo, ProgramInfo, and SourceLanguage classes</a></li>
34 <li><a href="#arch_llvm-db">The <tt>llvm-db</tt> tool</a></li>
35 <li><a href="#arch_todo">Short-term TODO list</a></li>
38 <li><a href="#format">Debugging information format</a>
40 <li><a href="#format_common_anchors">Anchors for global objects</a></li>
41 <li><a href="#format_common_stoppoint">Representing stopping points in the source program</a></li>
42 <li><a href="#format_common_lifetime">Object lifetimes and scoping</a></li>
43 <li><a href="#format_common_descriptors">Object descriptor formats</a>
45 <li><a href="#format_common_source_files">Representation of source files</a></li>
46 <li><a href="#format_common_program_objects">Representation of program objects</a></li>
47 <li><a href="#format_common_object_contexts">Program object contexts</a></li>
49 <li><a href="#format_common_intrinsics">Debugger intrinsic functions</a></li>
50 <li><a href="#format_common_tags">Values for debugger tags</a></li>
52 <li><a href="#ccxx_frontend">C/C++ front-end specific debug information</a>
54 <li><a href="#ccxx_pse">Program Scope Entries</a>
56 <li><a href="#ccxx_compilation_units">Compilation unit entries</a></li>
57 <li><a href="#ccxx_modules">Module, namespace, and importing entries</a></li>
59 <li><a href="#ccxx_dataobjects">Data objects (program variables)</a></li>
64 <img src="img/venusflytrap.jpg" alt="A leafy and green bug eater" width="247"
69 <div class="doc_author">
70 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
74 <!-- *********************************************************************** -->
75 <div class="doc_section"><a name="introduction">Introduction</a></div> <!--
76 *********************************************************************** -->
78 <div class="doc_text">
80 <p>This document is the central repository for all information pertaining to
81 debug information in LLVM. It describes the <a href="#llvm-db">user
82 interface</a> for the <tt>llvm-db</tt> tool, which provides a
83 powerful <a href="#llvm-db">source-level debugger</a>
84 to users of LLVM-based compilers. It then describes the <a
85 href="#architecture">various components</a> that make up the debugger and the
86 libraries which future clients may use. Finally, it describes the <a
87 href="#format">actual format that the LLVM debug information</a> takes,
88 which is useful for those interested in creating front-ends or dealing directly
89 with the information.</p>
93 <!-- ======================================================================= -->
94 <div class="doc_subsection">
95 <a name="phil">Philosophy behind LLVM debugging information</a>
98 <div class="doc_text">
100 <p>The idea of the LLVM debugging information is to capture how the important
101 pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
102 Several design aspects have shaped the solution that appears here. The
103 important ones are:</p>
106 <li>Debugging information should have very little impact on the rest of the
107 compiler. No transformations, analyses, or code generators should need to be
108 modified because of debugging information.</li>
110 <li>LLVM optimizations should interact in <a href="#debugopt">well-defined and
111 easily described ways</a> with the debugging information.</li>
113 <li>Because LLVM is designed to support arbitrary programming languages,
114 LLVM-to-LLVM tools should not need to know anything about the semantics of the
115 source-level-language.</li>
117 <li>Source-level languages are often <b>widely</b> different from one another.
118 LLVM should not put any restrictions of the flavor of the source-language, and
119 the debugging information should work with any language.</li>
121 <li>With code generator support, it should be possible to use an LLVM compiler
122 to compile a program to native machine code and standard debugging formats.
123 This allows compatibility with traditional machine-code level debuggers, like
128 <p>The approach used by the LLVM implementation is to use a small set of <a
129 href="#format_common_intrinsics">intrinsic functions</a> to define a mapping
130 between LLVM program objects and the source-level objects. The description of
131 the source-level program is maintained in LLVM global variables in an <a
132 href="#ccxx_frontend">implementation-defined format</a> (the C/C++ front-end
133 currently uses working draft 7 of the <a
134 href="http://www.eagercon.com/dwarf/dwarf3std.htm">Dwarf 3 standard</a>).</p>
136 <p>When a program is debugged, the debugger interacts with the user and turns
137 the stored debug information into source-language specific information. As
138 such, the debugger must be aware of the source-language, and is thus tied to a
139 specific language of family of languages. The <a href="#llvm-db">LLVM
140 debugger</a> is designed to be modular in its support for source-languages.</p>
145 <!-- ======================================================================= -->
146 <div class="doc_subsection">
147 <a name="debugopt">Debugging optimized code</a>
150 <div class="doc_text">
152 <p>An extremely high priority of LLVM debugging information is to make it
153 interact well with optimizations and analysis. In particular, the LLVM debug
154 information provides the following guarantees:</p>
158 <li>LLVM debug information <b>always provides information to accurately read the
159 source-level state of the program</b>, regardless of which LLVM optimizations
160 have been run, and without any modification to the optimizations themselves.
161 However, some optimizations may impact the ability to modify the current state
162 of the program with a debugger, such as setting program variables, or calling
163 function that have been deleted.</li>
165 <li>LLVM optimizations gracefully interact with debugging information. If they
166 are not aware of debug information, they are automatically disabled as necessary
167 in the cases that would invalidate the debug info. This retains the LLVM
168 features making it easy to write new transformations.</li>
170 <li>As desired, LLVM optimizations can be upgraded to be aware of the LLVM
171 debugging information, allowing them to update the debugging information as they
172 perform aggressive optimizations. This means that, with effort, the LLVM
173 optimizers could optimize debug code just as well as non-debug code.</li>
175 <li>LLVM debug information does not prevent many important optimizations from
176 happening (for example inlining, basic block reordering/merging/cleanup, tail
177 duplication, etc), further reducing the amount of the compiler that eventually
178 is "aware" of debugging information.</li>
180 <li>LLVM debug information is automatically optimized along with the rest of the
181 program, using existing facilities. For example, duplicate information is
182 automatically merged by the linker, and unused information is automatically
187 <p>Basically, the debug information allows you to compile a program with
188 "<tt>-O0 -g</tt>" and get full debug information, allowing you to arbitrarily
189 modify the program as it executes from the debugger. Compiling a program with
190 "<tt>-O3 -g</tt>" gives you full debug information that is always available and
191 accurate for reading (e.g., you get accurate stack traces despite tail call
192 elimination and inlining), but you might lose the ability to modify the program
193 and call functions where were optimized out of the program, or inlined away
198 <!-- ======================================================================= -->
199 <div class="doc_subsection">
200 <a name="future">Future work</a>
203 <div class="doc_text">
204 <p>There are several important extensions that could be eventually added to the
205 LLVM debugger. The most important extension would be to upgrade the LLVM code
206 generators to support debugging information. This would also allow, for
207 example, the X86 code generator to emit native objects that contain debugging
208 information consumable by traditional source-level debuggers like GDB or
211 <p>Additionally, LLVM optimizations can be upgraded to incrementally update the
212 debugging information, <a href="#commands">new commands</a> can be added to the
213 debugger, and thread support could be added to the debugger.</p>
215 <p>The "SourceLanguage" modules provided by <tt>llvm-db</tt> could be
216 substantially improved to provide good support for C++ language features like
217 namespaces and scoping rules.</p>
219 <p>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>
223 <p>For someone so inclined, it should be straight-forward to write different
224 front-ends for the LLVM debugger, as the LLVM debugging engine is cleanly
225 separated from the <tt>llvm-db</tt> front-end. A new LLVM GUI debugger or IDE
230 <!-- *********************************************************************** -->
231 <div class="doc_section">
232 <a name="llvm-db">Using the <tt>llvm-db</tt> tool</a>
234 <!-- *********************************************************************** -->
236 <div class="doc_text">
238 <p>The <tt>llvm-db</tt> tool provides a GDB-like interface for source-level
239 debugging of programs. This tool provides many standard commands for inspecting
240 and modifying the program as it executes, loading new programs, single stepping,
241 placing breakpoints, etc. This section describes how to use the debugger.</p>
243 <p><tt>llvm-db</tt> has been designed to be as similar to GDB in its user
244 interface as possible. This should make it extremely easy to learn
245 <tt>llvm-db</tt> if you already know <tt>GDB</tt>. In general, <tt>llvm-db</tt>
246 provides the subset of GDB commands that are applicable to LLVM debugging users.
247 If there is a command missing that make a reasonable amount of sense within the
248 <a href="#limitations">limitations of <tt>llvm-db</tt></a>, please report it as
249 a bug or, better yet, submit a patch to add it.</p>
253 <!-- ======================================================================= -->
254 <div class="doc_subsection">
255 <a name="limitations">Limitations of <tt>llvm-db</tt></a>
258 <div class="doc_text">
260 <p><tt>llvm-db</tt> is designed to be modular and easy to extend. This
261 extensibility was key to getting the debugger up-and-running quickly, because we
262 can start with simple-but-unsophisicated implementations of various components.
263 Because of this, it is currently missing many features, though they should be
264 easy to add over time (patches welcomed!). The biggest inherent limitations of
265 <tt>llvm-db</tt> are currently due to extremely simple <a
266 href="#arch_debugger">debugger backend</a> (implemented in
267 "lib/Debugger/UnixLocalInferiorProcess.cpp") which is designed to work without
268 any cooperation from the code generators. Because it is so simple, it suffers
269 from the following inherent limitations:</p>
273 <li>Running a program in <tt>llvm-db</tt> is a bit slower than running it with
274 <tt>lli</tt> (i.e., in the JIT).</li>
276 <li>Inspection of the target hardware is not supported. This means that you
277 cannot, for example, print the contents of X86 registers.</li>
279 <li>Inspection of LLVM code is not supported. This means that you cannot print
280 the contents of arbitrary LLVM values, or use commands such as <tt>stepi</tt>.
281 This also means that you cannot debug code without debug information.</li>
283 <li>Portions of the debugger run in the same address space as the program being
284 debugged. This means that memory corruption by the program could trample on
285 portions of the debugger.</li>
287 <li>Attaching to existing processes and core files is not currently
292 <p>That said, the debugger is still quite useful, and all of these limitations
293 can be eliminated by integrating support for the debugger into the code
294 generators, and writing a new <a href="#arch_debugger">InferiorProcess</a>
295 subclass to use it. See the <a href="#future">future work</a> section for ideas
296 of how to extend the LLVM debugger despite these limitations.</p>
301 <!-- ======================================================================= -->
302 <div class="doc_subsection">
303 <a name="sample">A sample <tt>llvm-db</tt> session</a>
306 <div class="doc_text">
308 <p>TODO: this is obviously lame, when more is implemented, this can be much
312 $ <b>llvm-db funccall</b>
313 llvm-db: The LLVM source-level debugger
314 Loading program... successfully loaded 'funccall.bc'!
315 (llvm-db) <b>create</b>
316 Starting program: funccall.bc
317 main at funccall.c:9:2
319 (llvm-db) <b>list main</b>
330 (llvm-db) <b>list</b>
332 (llvm-db) <b>step</b>
335 foo at funccall.c:5:2
338 #0 -> 0x85ffba0 in foo at funccall.c:5:2
339 #1 0x85ffd98 in main at funccall.c:10:2
340 (llvm-db) <b>finish</b>
341 main at funccall.c:11:2
346 The program stopped with exit code 0
347 (llvm-db) <b>quit</b>
355 <!-- ======================================================================= -->
356 <div class="doc_subsection">
357 <a name="startup">Starting the debugger</a>
360 <div class="doc_text">
362 <p>There are three ways to start up the <tt>llvm-db</tt> debugger:</p>
364 <p>When run with no options, just <tt>llvm-db</tt>, the debugger starts up
365 without a program loaded at all. You must use the <a
366 href="#c_file"><tt>file</tt> command</a> to load a program, and the <a
367 href="#c_set_args"><tt>set args</tt></a> or <a href="#c_run"><tt>run</tt></a>
368 commands to specify the arguments for the program.</p>
370 <p>If you start the debugger with one argument, as <tt>llvm-db
371 <program></tt>, the debugger will start up and load in the specified
372 program. You can then optionally specify arguments to the program with the <a
373 href="#c_set_args"><tt>set args</tt></a> or <a href="#c_run"><tt>run</tt></a>
376 <p>The third way to start the program is with the <tt>--args</tt> option. This
377 option allows you to specify the program to load and the arguments to start out
378 with. <!-- No options to <tt>llvm-db</tt> may be specified after the
379 <tt>-args</tt> option. --> Example use: <tt>llvm-db --args ls /home</tt></p>
383 <!-- ======================================================================= -->
384 <div class="doc_subsection">
385 <a name="commands">Commands recognized by the debugger</a>
388 <div class="doc_text">
390 <p>FIXME: this needs work obviously. See the <a
391 href="http://sources.redhat.com/gdb/documentation/">GDB documentation</a> for
392 information about what these do, or try '<tt>help [command]</tt>' within
393 <tt>llvm-db</tt> to get information.</p>
396 <h2>General usage:</h2>
398 <li>help [command]</li>
400 <li><a name="c_file">file</a> [program]</li>
403 <h2>Program inspection and interaction:</h2>
405 <li>create (start the program, stopping it ASAP in <tt>main</tt>)</li>
413 <li>list [start[, end]]</li>
415 <li>info sources</li>
416 <li>info functions</li>
419 <h2>Call stack inspection:</h2>
428 <h2>Debugger inspection and interaction:</h2>
433 <li>show listsize</li>
434 <li>set listsize</li>
435 <li>show language</li>
436 <li>set language</li>
438 <li>set args [args]</li>
449 <li>info variables</li>
450 <li>info program</li>
455 <li>... many others</li>
460 <!-- *********************************************************************** -->
461 <div class="doc_section">
462 <a name="architecture">Architecture of the LLVM debugger</a>
464 <!-- *********************************************************************** -->
466 <div class="doc_text">
467 <p>The LLVM debugger is built out of three distinct layers of software. These
468 layers provide clients with different interface options depending on what pieces
469 of they want to implement themselves, and it also promotes code modularity and
470 good design. The three layers are the <a href="#arch_debugger">Debugger
471 interface</a>, the <a href="#arch_info">"info" interfaces</a>, and the <a
472 href="#arch_llvm-db"><tt>llvm-db</tt> tool</a> itself.</p>
475 <!-- ======================================================================= -->
476 <div class="doc_subsection">
477 <a name="arch_debugger">The Debugger and InferiorProcess classes</a>
480 <div class="doc_text">
481 <p>The Debugger class (defined in the <tt>include/llvm/Debugger/</tt> directory)
482 is a low-level class which is used to maintain information about the loaded
483 program, as well as start and stop the program running as necessary. This class
484 does not provide any high-level analysis or control over the program, only
485 exposing simple interfaces like <tt>load/unloadProgram</tt>,
486 <tt>create/killProgram</tt>, <tt>step/next/finish/contProgram</tt>, and
487 low-level methods for installing breakpoints.</p>
490 The Debugger class is itself a wrapper around the lowest-level InferiorProcess
491 class. This class is used to represent an instance of the program running under
492 debugger control. The InferiorProcess class can be implemented in different
493 ways for different targets and execution scenarios (e.g., remote debugging).
494 The InferiorProcess class exposes a small and simple collection of interfaces
495 which are useful for inspecting the current state of the program (such as
496 collecting stack trace information, reading the memory image of the process,
497 etc). The interfaces in this class are designed to be as low-level and simple
498 as possible, to make it easy to create new instances of the class.
502 The Debugger class exposes the currently active instance of InferiorProcess
503 through the <tt>Debugger::getRunningProcess</tt> method, which returns a
504 <tt>const</tt> reference to the class. This means that clients of the Debugger
505 class can only <b>inspect</b> the running instance of the program directly. To
506 change the executing process in some way, they must use the interces exposed by
511 <!-- ======================================================================= -->
512 <div class="doc_subsection">
513 <a name="arch_info">The RuntimeInfo, ProgramInfo, and SourceLanguage classes</a>
516 <div class="doc_text">
518 The next-highest level of debugger abstraction is provided through the
519 ProgramInfo, RuntimeInfo, SourceLanguage and related classes (also defined in
520 the <tt>include/llvm/Debugger/</tt> directory). These classes efficiently
521 decode the debugging information and low-level interfaces exposed by
522 InferiorProcess into a higher-level representation, suitable for analysis by the
527 The ProgramInfo class exposes a variety of different kinds of information about
528 the program objects in the source-level-language. The SourceFileInfo class
529 represents a source-file in the program (e.g. a .cpp or .h file). The
530 SourceFileInfo class captures information such as which SourceLanguage was used
531 to compile the file, where the debugger can get access to the actual file text
532 (which is lazily loaded on demand), etc. The SourceFunctionInfo class
533 represents a... <b>FIXME: finish</b>. The ProgramInfo class provides interfaces
534 to lazily find and decode the information needed to create the Source*Info
535 classes requested by the debugger.
539 The RuntimeInfo class exposes information about the currently executed program,
540 by decoding information from the InferiorProcess and ProgramInfo classes. It
541 provides a StackFrame class which provides an easy-to-use interface for
542 inspecting the current and suspended stack frames in the program.
546 The SourceLanguage class is an abstract interface used by the debugger to
547 perform all source-language-specific tasks. For example, this interface is used
548 by the ProgramInfo class to decode language-specific types and functions and by
549 the debugger front-end (such as <a href="#arch_llvm-db"><tt>llvm-db</tt></a> to
550 evaluate source-langauge expressions typed into the debugger. This class uses
551 the RuntimeInfo & ProgramInfo classes to get information about the current
552 execution context and the loaded program, respectively.
557 <!-- ======================================================================= -->
558 <div class="doc_subsection">
559 <a name="arch_llvm-db">The <tt>llvm-db</tt> tool</a>
562 <div class="doc_text">
564 The <tt>llvm-db</tt> is designed to be a debugger providing an interface as <a
565 href="#llvm-db">similar to GDB</a> as reasonable, but no more so than that.
566 Because the <a href="#arch_debugger">Debugger</a> and <a
567 href="#arch_info">info</a> classes implement all of the heavy lifting and
568 analysis, <tt>llvm-db</tt> (which lives in <tt>llvm/tools/llvm-db</tt>) consists
569 mainly of of code to interact with the user and parse commands. The CLIDebugger
570 constructor registers all of the builtin commands for the debugger, and each
571 command is implemented as a CLIDebugger::[name]Command method.
576 <!-- ======================================================================= -->
577 <div class="doc_subsection">
578 <a name="arch_todo">Short-term TODO list</a>
581 <div class="doc_text">
584 FIXME: this section will eventually go away. These are notes to myself of
585 things that should be implemented, but haven't yet.
589 <b>Breakpoints:</b> Support is already implemented in the 'InferiorProcess'
590 class, though it hasn't been tested yet. To finish breakpoint support, we need
591 to implement breakCommand (which should reuse the linespec parser from the list
592 command), and handle the fact that 'break foo' or 'break file.c:53' may insert
593 multiple breakpoints. Also, if you say 'break file.c:53' and there is no
594 stoppoint on line 53, the breakpoint should go on the next available line. My
595 idea was to have the Debugger class provide a "Breakpoint" class which
596 encapsulated this messiness, giving the debugger front-end a simple interface.
597 The debugger front-end would have to map the really complex semantics of
598 temporary breakpoints and 'conditional' breakpoints onto this intermediate
599 level. Also, breakpoints should survive as much as possible across program
604 <b>UnixLocalInferiorProcess.cpp speedup</b>: There is no reason for the debugged
605 process to code gen the globals corresponding to debug information. The
606 IntrinsicLowering object could instead change descriptors into constant expr
607 casts of the constant address of the LLVM objects for the descriptors. This
608 would also allow us to eliminate the mapping back and forth between physical
609 addresses that must be done.</p>
612 <b>Process deaths</b>: The InferiorProcessDead exception should be extended to
613 know "how" a process died, i.e., it was killed by a signal. This is easy to
614 collect in the UnixLocalInferiorProcess, we just need to represent it.</p>
618 <!-- *********************************************************************** -->
619 <div class="doc_section">
620 <a name="format">Debugging information format</a>
622 <!-- *********************************************************************** -->
624 <div class="doc_text">
626 <p>LLVM debugging information has been carefully designed to make it possible
627 for the optimizer to optimize the program and debugging information without
628 necessarily having to know anything about debugging information. In particular,
629 the global constant merging pass automatically eliminates duplicated debugging
630 information (often caused by header files), the global dead code elimination
631 pass automatically deletes debugging information for a function if it decides to
632 delete the function, and the linker eliminates debug information when it merges
633 <tt>linkonce</tt> functions.</p>
635 <p>To do this, most of the debugging information (descriptors for types,
636 variables, functions, source files, etc) is inserted by the language front-end
637 in the form of LLVM global variables. These LLVM global variables are no
638 different from any other global variables, except that they have a web of LLVM
639 intrinsic functions that point to them. If the last references to a particular
640 piece of debugging information are deleted (for example, by the
641 <tt>-globaldce</tt> pass), the extraneous debug information will automatically
642 become dead and be removed by the optimizer.</p>
644 <p>The debugger is designed to be agnostic about the contents of most of the
645 debugging information. It uses a <a href="#arch_info">source-language-specific
646 module</a> to decode the information that represents variables, types,
647 functions, namespaces, etc: this allows for arbitrary source-language semantics
648 and type-systems to be used, as long as there is a module written for the
649 debugger to interpret the information.</p>
651 <p>To provide basic functionality, the LLVM debugger does have to make some
652 assumptions about the source-level language being debugged, though it keeps
653 these to a minimum. The only common features that the LLVM debugger assumes
654 exist are <a href="#format_common_source_files">source files</a>, and <a
655 href="#format_program_objects">program objects</a>. These abstract objects are
656 used by the debugger to form stack traces, show information about local
659 <p>This section of the documentation first describes the representation aspects
660 common to any source-language. The <a href="#ccxx_frontend">next section</a>
661 describes the data layout conventions used by the C and C++ front-ends.</p>
665 <!-- ======================================================================= -->
666 <div class="doc_subsection">
667 <a name="format_common_anchors">Anchors for global objects</a>
670 <div class="doc_text">
671 <p>One important aspect of the LLVM debug representation is that it allows the
672 LLVM debugger to efficiently index all of the global objects without having the
673 scan the program. To do this, all of the global objects use "anchor" globals of
674 type "<tt>{}</tt>", with designated names. These anchor objects obviously do
675 not contain any content or meaning by themselves, but all of the global objects
676 of a particular type (e.g., source file descriptors) contain a pointer to the
677 anchor. This pointer allows the debugger to use def-use chains to find all
678 global objects of that type.</p>
680 <p>So far, the following names are recognized as anchors by the LLVM
684 %<a href="#format_common_source_files">llvm.dbg.translation_units</a> = linkonce global {} {}
685 %<a href="#format_program_objects">llvm.dbg.globals</a> = linkonce global {} {}
688 <p>Using anchors in this way (where the source file descriptor points to the
689 anchors, as opposed to having a list of source file descriptors) allows for the
690 standard dead global elimination and merging passes to automatically remove
691 unused debugging information. If the globals were kept track of through lists,
692 there would always be an object pointing to the descriptors, thus would never be
697 <!-- ======================================================================= -->
698 <div class="doc_subsection">
699 <a name="format_common_stoppoint">
700 Representing stopping points in the source program
704 <div class="doc_text">
706 <p>LLVM debugger "stop points" are a key part of the debugging representation
707 that allows the LLVM to maintain simple semantics for <a
708 href="#debugopt">debugging optimized code</a>. The basic idea is that the
709 front-end inserts calls to the <tt>%llvm.dbg.stoppoint</tt> intrinsic function
710 at every point in the program where the debugger should be able to inspect the
711 program (these correspond to places the debugger stops when you "<tt>step</tt>"
712 through it). The front-end can choose to place these as fine-grained as it
713 would like (for example, before every subexpression evaluated), but it is
714 recommended to only put them after every source statement that includes
717 <p>Using calls to this intrinsic function to demark legal points for the
718 debugger to inspect the program automatically disables any optimizations that
719 could potentially confuse debugging information. To non-debug-information-aware
720 transformations, these calls simply look like calls to an external function,
721 which they must assume to do anything (including reading or writing to any part
722 of reachable memory). On the other hand, it does not impact many optimizations,
723 such as code motion of non-trapping instructions, nor does it impact
724 optimization of subexpressions, code duplication transformations, or basic-block
725 reordering transformations.</p>
727 <p>An important aspect of the calls to the <tt>%llvm.dbg.stoppoint</tt>
728 intrinsic is that the function-local debugging information is woven together
729 with use-def chains. This makes it easy for the debugger to, for example,
730 locate the 'next' stop point. For a concrete example of stop points, see the
731 example in <a href="#format_common_lifetime">the next section</a>.</p>
736 <!-- ======================================================================= -->
737 <div class="doc_subsection">
738 <a name="format_common_lifetime">Object lifetimes and scoping</a>
741 <div class="doc_text">
742 <p>In many languages, the local variables in functions can have their lifetime
743 or scope limited to a subset of a function. In the C family of languages, for
744 example, variables are only live (readable and writable) within the source block
745 that they are defined in. In functional languages, values are only readable
746 after they have been defined. Though this is a very obvious concept, it is also
747 non-trivial to model in LLVM, because it has no notion of scoping in this sense,
748 and does not want to be tied to a language's scoping rules.</p>
750 <p>In order to handle this, the LLVM debug format uses the notion of "regions"
751 of a function, delineated by calls to intrinsic functions. These intrinsic
752 functions define new regions of the program and indicate when the region
753 lifetime expires. Consider the following C fragment, for example:</p>
767 <p>Compiled to LLVM, this function would be represented like this (FIXME: CHECK
768 AND UPDATE THIS):</p>
775 <a name="#icl_ex_D1">%D1</a> = call {}* %llvm.dbg.func.start(<a href="#format_program_objects">%lldb.global</a>* %d.foo)
776 %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)
778 %D3 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D2, ...)
779 <i>;; Evaluate expression on line 2, assigning to X.</i>
780 %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)
782 %D5 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D4, ...)
783 <i>;; Evaluate expression on line 3, assigning to Y.</i>
784 %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)
786 <a name="#icl_ex_D1">%D7</a> = call {}* %llvm.region.start({}* %D6)
787 %D8 = call {}* %llvm.dbg.DEFINEVARIABLE({}* %D7, ...)
788 <i>;; Evaluate expression on line 5, assigning to Z.</i>
789 %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)
791 <i>;; Code for line 6.</i>
792 %D10 = call {}* %llvm.region.end({}* %D9)
793 %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)
795 <i>;; Code for line 8.</i>
796 <a name="#icl_ex_D1">%D12</a> = call {}* %llvm.region.end({}* %D11)
801 <p>This example illustrates a few important details about the LLVM debugging
802 information. In particular, it shows how the various intrinsics used are woven
803 together with def-use and use-def chains, similar to how <a
804 href="#format_common_anchors">anchors</a> are used with globals. This allows
805 the debugger to analyze the relationship between statements, variable
806 definitions, and the code used to implement the function.</p>
808 <p>In this example, two explicit regions are defined, one with the <a
809 href="#icl_ex_D1">definition of the <tt>%D1</tt> variable</a> and one with the
810 <a href="#icl_ex_D7">definition of <tt>%D7</tt></a>. In the case of
811 <tt>%D1</tt>, the debug information indicates that the function whose <a
812 href="#format_program_objects">descriptor</a> is specified as an argument to the
813 intrinsic. This defines a new stack frame whose lifetime ends when the region
814 is ended by <a href="#icl_ex_D12">the <tt>%D12</tt> call</a>.</p>
816 <p>Using regions to represent the boundaries of source-level functions allow
817 LLVM interprocedural optimizations to arbitrarily modify LLVM functions without
818 having to worry about breaking mapping information between the LLVM code and the
819 and source-level program. In particular, the inliner requires no modification
820 to support inlining with debugging information: there is no explicit correlation
821 drawn between LLVM functions and their source-level counterparts (note however,
822 that if the inliner inlines all instances of a non-strong-linkage function into
823 its caller that it will not be possible for the user to manually invoke the
824 inlined function from the debugger).</p>
826 <p>Once the function has been defined, the <a
827 href="#format_common_stoppoint">stopping point</a> corresponding to line #2 of
828 the function is encountered. At this point in the function, <b>no</b> local
829 variables are live. As lines 2 and 3 of the example are executed, their
830 variable definitions are automatically introduced into the program, without the
831 need to specify a new region. These variables do not require new regions to be
832 introduced because they go out of scope at the same point in the program: line
835 <p>In contrast, the <tt>Z</tt> variable goes out of scope at a different time,
836 on line 7. For this reason, it is defined within <a href="#icl_ex_D7">the
837 <tt>%D7</tt> region</a>, which kills the availability of <tt>Z</tt> before the
838 code for line 8 is executed. In this way, regions can support arbitrary
839 source-language scoping rules, as long as they can only be nested (ie, one scope
840 cannot partially overlap with a part of another scope).</p>
842 <p>It is worth noting that this scoping mechanism is used to control scoping of
843 all declarations, not just variable declarations. For example, the scope of a
844 C++ using declaration is controlled with this, and the <tt>llvm-db</tt> C++
845 support routines could use this to change how name lookup is performed (though
846 this is not implemented yet).</p>
850 <!-- ======================================================================= -->
851 <div class="doc_subsection">
852 <a name="format_common_descriptors">Object descriptor formats</a>
855 <div class="doc_text">
856 <p>The LLVM debugger expects the descriptors for program objects to start in a
857 canonical format, but the descriptors can include additional information
858 appended at the end that is source-language specific. All LLVM debugging
859 information is versioned, allowing backwards compatibility in the case that the
860 core structures need to change in some way. Also, all debugging information
861 objects start with a <a href="#format_common_tags">tag</a> to indicate what type
862 of object it is. The source-language is allows to define its own objects, by
863 using unreserved tag numbers.</p>
865 <p>The lowest-level descriptor are those describing <a
866 href="#format_common_source_files">the files containing the program source
867 code</a>, as most other descriptors (sometimes indirectly) refer to them.
872 <!-- ------------------------------------------------------------------------ ->
873 <div class="doc_subsubsection">
874 <a name="format_common_source_files">Representation of source files</a>
877 <div class="doc_text">
879 Source file descriptors are patterned after the Dwarf "compile_unit" object.
880 The descriptor currently is defined to have at least the following LLVM
884 %lldb.compile_unit = type {
885 uint, <i>;; Tag: <a href="#tag_compile_unit">LLVM_COMPILE_UNIT</a></i>
886 ushort, <i>;; LLVM debug version number</i>
887 ushort, <i>;; Dwarf language identifier</i>
888 sbyte*, <i>;; Filename</i>
889 sbyte*, <i>;; Working directory when compiled</i>
890 sbyte* <i>;; Producer of the debug information</i>
895 These descriptors contain the version number for the debug info, a source
896 language ID for the file (we use the Dwarf 3.0 ID numbers, such as
897 <tt>DW_LANG_C89</tt>, <tt>DW_LANG_C_plus_plus</tt>, <tt>DW_LANG_Cobol74</tt>,
898 etc), three strings describing the filename, working directory of the compiler,
899 and an identifier string for the compiler that produced it. Note that actual
900 compile_unit declarations must also include an <a
901 href="#format_common_anchors">anchor</a> to <tt>llvm.dbg.translation_units</tt>,
902 but it is not specified where the anchor is to be located. Here is an example
907 %arraytest_source_file = internal constant %lldb.compile_unit {
908 <a href="#tag_compile_unit">uint 17</a>, ; Tag value
909 ushort 0, ; Version #0
910 ushort 1, ; DW_LANG_C89
911 sbyte* getelementptr ([12 x sbyte]* %.str_1, long 0, long 0), ; filename
912 sbyte* getelementptr ([12 x sbyte]* %.str_2, long 0, long 0), ; working dir
913 sbyte* getelementptr ([12 x sbyte]* %.str_3, long 0, long 0), ; producer
914 {}* %llvm.dbg.translation_units ; Anchor
916 %.str_1 = internal constant [12 x sbyte] c"arraytest.c\00"
917 %.str_2 = internal constant [12 x sbyte] c"/home/sabre\00"
918 %.str_3 = internal constant [12 x sbyte] c"llvmgcc 3.4\00"
922 Note that the LLVM constant merging pass should eliminate duplicate copies of
923 the strings that get emitted to each translation unit, such as the producer.
929 <!-- ----------------------------------------------------------------------- -->
930 <div class="doc_subsubsection">
931 <a name="format_program_objects">Representation of program objects</a>
934 <div class="doc_text">
936 The LLVM debugger needs to know about some source-language program objects, in
937 order to build stack traces, print information about local variables, and other
938 related activities. The LLVM debugger differentiates between three different
939 types of program objects: subprograms (functions, messages, methods, etc),
940 variables (locals and globals), and others. Because source-languages have
941 widely varying forms of these objects, the LLVM debugger expects only a few
942 fields in the descriptor for each object:
946 %lldb.object = type {
947 uint, <i>;; <a href="#format_common_tag">A tag</a></i>
948 <i>any</i>*, <i>;; The <a href="#format_common_object_contexts">context</a> for the object</i>
949 sbyte* <i>;; The object 'name'</i>
953 <p>The first field contains a tag for the descriptor. The second field contains
954 either a pointer to the descriptor for the containing <a
955 href="#format_common_source_files">source file</a>, or it contains a pointer to
956 another program object whose context pointer eventually reaches a source file.
957 Through this <a href="#format_common_object_contexts">context</a> pointer, the
958 LLVM debugger can establish the debug version number of the object.</p>
960 <p>The third field contains a string that the debugger can use to identify the
961 object if it does not contain explicit support for the source-language in use
962 (ie, the 'unknown' source language handler uses this string). This should be
963 some sort of unmangled string that corresponds to the object, but it is a
964 quality of implementation issue what exactly it contains (it is legal, though
965 not useful, for all of these strings to be null).</p>
967 <p>Note again that descriptors can be extended to include
968 source-language-specific information in addition to the fields required by the
969 LLVM debugger. See the <a href="#ccxx_descriptors">section on the C/C++
970 front-end</a> for more information. Also remember that global objects
971 (functions, selectors, global variables, etc) must contain an <a
972 href="#format_common_anchors">anchor</a> to the <tt>llvm.dbg.globals</tt>
977 <!-- ======================================================================= -->
978 <div class="doc_subsection">
979 <a name="format_common_object_contexts">Program object contexts</a>
982 <div class="doc_text">
984 Allow source-language specific contexts, use to identify namespaces etc
985 Must end up in a source file descriptor.
986 Debugger core ignores all unknown context objects.
990 <!-- ======================================================================= -->
991 <div class="doc_subsection">
992 <a name="format_common_intrinsics">Debugger intrinsic functions</a>
995 <div class="doc_text">
997 Define each intrinsics, as an extension of the language reference manual.
1000 llvm.dbg.region.start
1002 llvm.dbg.function.start
1007 <!-- ======================================================================= -->
1008 <div class="doc_subsection">
1009 <a name="format_common_tags">Values for debugger tags</a>
1012 <div class="doc_text">
1014 <p>Happen to be the same value as the similarly named Dwarf-3 tags, this may
1015 change in the future.</p>
1018 <a name="tag_compile_unit">LLVM_COMPILE_UNIT</a> : 17
1019 <a name="tag_subprogram">LLVM_SUBPROGRAM</a> : 46
1020 <a name="tag_variable">LLVM_VARIABLE</a> : 52
1021 <!-- <a name="tag_formal_parameter">LLVM_FORMAL_PARAMETER : 5-->
1027 <!-- *********************************************************************** -->
1028 <div class="doc_section">
1029 <a name="ccxx_frontend">C/C++ front-end specific debug information</a>
1032 <div class="doc_text">
1034 <p>The C and C++ front-ends represent information about the program in a format
1035 that is effectively identical to <a
1036 href="http://www.eagercon.com/dwarf/dwarf3std.htm">Dwarf 3.0</a> in terms of
1037 information content. This allows code generators to trivially support native
1038 debuggers by generating standard dwarf information, and contains enough
1039 information for non-dwarf targets to translate it as needed.</p>
1041 <p>The basic debug information required by the debugger is (intentionally)
1042 designed to be as minimal as possible. This basic information is so minimal
1043 that it is unlikely that <b>any</b> source-language could be adequately
1044 described by it. Because of this, the debugger format was designed for
1045 extension to support source-language-specific information. The extended
1046 descriptors are read and interpreted by the <a
1047 href="#arch_info">language-specific</a> modules in the debugger if there is
1048 support available, otherwise it is ignored.</p>
1050 <p>This section describes the extensions used to represent C and C++ programs.
1051 Other languages could pattern themselves after this (which itself is tuned to
1052 representing programs in the same way that Dwarf 3 does), or they could choose
1053 to provide completely different extensions if they don't fit into the Dwarf
1054 model. As support for debugging information gets added to the various LLVM
1055 source-language front-ends, the information used should be documented here.</p>
1059 <!-- ======================================================================= -->
1060 <div class="doc_subsection">
1061 <a name="ccxx_pse">Program Scope Entries</a>
1064 <div class="doc_text">
1068 <!-- -------------------------------------------------------------------------->
1069 <div class="doc_subsubsection">
1070 <a name="ccxx_compilation_units">Compilation unit entries</a>
1073 <div class="doc_text">
1075 Translation units do not add any information over the standard <a
1076 href="#format_common_source_files">source file representation</a> already
1077 expected by the debugger. As such, it uses descriptors of the type specified,
1078 with a trailing <a href="#format_common_anchors">anchor</a>.
1082 <!-- -------------------------------------------------------------------------->
1083 <div class="doc_subsubsection">
1084 <a name="ccxx_modules">Module, namespace, and importing entries</a>
1087 <div class="doc_text">
1091 <!-- ======================================================================= -->
1092 <div class="doc_subsection">
1093 <a name="ccxx_dataobjects">Data objects (program variables)</a>
1096 <div class="doc_text">
1101 <!-- *********************************************************************** -->
1105 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1106 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1107 <a href="http://validator.w3.org/check/referer"><img
1108 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1110 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1111 <a href="http://llvm.cs.uiuc.edu">LLVM Compiler Infrastructure</a><br>
1112 Last modified: $Date$