1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>Exception Handling in LLVM</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
10 <div class="doc_title">Exception Handling in LLVM</div>
12 <table class="layout" style="width:100%">
16 <li><a href="#introduction">Introduction</a>
18 <li><a href="#itanium">Itanium ABI Zero-cost Exception Handling</a></li>
19 <li><a href="#overview">Overview</a></li>
21 <li><a href="#codegen">LLVM Code Generation</a>
23 <li><a href="#throw">Throw</a></li>
24 <li><a href="#try_catch">Try/Catch</a></li>
25 <li><a href="#cleanups">Cleanups</a></li>
26 <li><a href="#throw_filters">Throw Filters</a></li>
27 <li><a href="#restrictions">Restrictions</a></li>
29 <li><a href="#format_common_intrinsics">Exception Handling Intrinsics</a>
31 <li><a href="#llvm_eh_exception"><tt>llvm.eh.exception</tt></a></li>
32 <li><a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a></li>
33 <li><a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a></li>
35 <li><a href="#asm">Asm Table Formats</a>
37 <li><a href="#unwind_tables">Exception Handling Frame</a></li>
38 <li><a href="#exception_tables">Exception Tables</a></li>
40 <li><a href="#todo">ToDo</a></li>
45 <div class="doc_author">
46 <p>Written by <a href="mailto:jlaskey@mac.com">Jim Laskey</a></p>
50 <!-- *********************************************************************** -->
51 <div class="doc_section"><a name="introduction">Introduction</a></div>
52 <!-- *********************************************************************** -->
54 <div class="doc_text">
56 <p>This document is the central repository for all information pertaining to
57 exception handling in LLVM. It describes the format that LLVM exception
58 handling information takes, which is useful for those interested in creating
59 front-ends or dealing directly with the information. Further, this document
60 provides specific examples of what exception handling information is used for
65 <!-- ======================================================================= -->
66 <div class="doc_subsection">
67 <a name="itanium">Itanium ABI Zero-cost Exception Handling</a>
70 <div class="doc_text">
72 <p>Exception handling for most programming languages is designed to recover from
73 conditions that rarely occur during general use of an application. To that end,
74 exception handling should not interfere with the main flow of an
75 application's algorithm by performing checkpointing tasks such as saving
76 the current pc or register state.</p>
78 <p>The Itanium ABI Exception Handling Specification defines a methodology for
79 providing outlying data in the form of exception tables without inlining
80 speculative exception handling code in the flow of an application's main
81 algorithm. Thus, the specification is said to add "zero-cost" to the normal
82 execution of an application.</p>
84 <p>A more complete description of the Itanium ABI exception handling runtime
85 support of can be found at <a
86 href="http://www.codesourcery.com/cxx-abi/abi-eh.html">Itanium C++ ABI:
87 Exception Handling.</a> A description of the exception frame format can be
89 href="http://refspecs.freestandards.org/LSB_3.0.0/LSB-Core-generic/LSB-
90 Core-generic/ehframechpt.html">Exception Frames</a>, with details of the Dwarf
91 specification at <a href="http://www.eagercon.com/dwarf/dwarf3std.htm">Dwarf 3
92 Standard.</a> A description for the C++ exception table formats can be found at
93 <a href="http://www.codesourcery.com/cxx-abi/exceptions.pdf">Exception Handling
98 <!-- ======================================================================= -->
99 <div class="doc_subsection">
100 <a name="overview">Overview</a>
103 <div class="doc_text">
105 <p>When an exception is thrown in llvm code, the runtime does a best effort to
106 find a handler suited to process the circumstance.</p>
108 <p>The runtime first attempts to find an <i>exception frame</i> corresponding to
109 the function where the exception was thrown. If the programming language (ex.
110 C++) supports exception handling, the exception frame contains a reference to an
111 exception table describing how to process the exception. If the language (ex.
112 C) does not support exception handling or if the exception needs to be forwarded
113 to a prior activation, the exception frame contains information about how to
114 unwind the current activation and restore the state of the prior activation.
115 This process is repeated until the exception is handled. If the exception is
116 not handled and no activations remain, then the application is terminated with
117 an appropriate error message.</p>
119 <p>Since different programming languages have different behaviors when handling
120 exceptions, the exception handling ABI provides a mechanism for supplying
121 <i>personalities.</i> An exception handling personality is defined by way of a
122 <i>personality function</i> (ex. for C++ <tt>__gxx_personality_v0</tt>) which
123 receives the context of the exception, an <i>exception structure</i> containing
124 the exception object type and value, and a reference to the exception table for
125 the current function. The personality function for the current compile unit is
126 specified in a <i>common exception frame</i>.</p>
128 <p>The organization of an exception table is language dependent. For C++, an
129 exception table is organized as a series of code ranges defining what to do if
130 an exception occurs in that range. Typically, the information associated with a
131 range defines which types of exception objects (using C++ <i>type info</i>) that
132 are handled in that range, and an associated action that should take place.
133 Actions typically pass control to a <i>landing pad</i>.</p>
135 <p>A landing pad corresponds to the code found in the catch portion of a
136 try/catch sequence. When execution resumes at a landing pad, it receives the
137 exception structure and a selector corresponding to the <i>type</i> of exception
138 thrown. The selector is then used to determine which catch should actually
139 process the exception.</p>
143 <!-- ======================================================================= -->
144 <div class="doc_section">
145 <a name="codegen">LLVM Code Generation</a>
148 <div class="doc_text">
150 <p>At the time of this writing, only C++ exception handling support is available
151 in LLVM. So the remainder of this document will be somewhat C++-centric.</p>
153 <p>From the C++ developers perspective, exceptions are defined in terms of the
154 <tt>throw</tt> and <tt>try/catch</tt> statements. In this section we will
155 describe the implementation of llvm exception handling in terms of C++
160 <!-- ======================================================================= -->
161 <div class="doc_subsection">
162 <a name="throw">Throw</a>
165 <div class="doc_text">
167 <p>Languages that support exception handling typically provide a <tt>throw</tt>
168 operation to initiate the exception process. Internally, a throw operation
169 breaks down into two steps. First, a request is made to allocate exception
170 space for an exception structure. This structure needs to survive beyond the
171 current activation. This structure will contain the type and value of the
172 object being thrown. Second, a call is made to the runtime to raise the
173 exception, passing the exception structure as an argument.</p>
175 <p>In C++, the allocation of the exception structure is done by the
176 <tt>__cxa_allocate_exception</tt> runtime function. The exception raising is
177 handled by <tt>__cxa_throw</tt>. The type of the exception is represented using
178 a C++ RTTI type info structure.</p>
182 <!-- ======================================================================= -->
183 <div class="doc_subsection">
184 <a name="try_catch">Try/Catch</a>
187 <div class="doc_text">
189 <p>A call within the scope of a try statement can potentially raise an exception.
190 In those circumstances, the LLVM C++ front-end replaces the call with an
191 <tt>invoke</tt> instruction. Unlike a call, the invoke has two potential
192 continuation points; where to continue when the call succeeds as per normal, and
193 where to continue if the call raises an exception, either by a throw or the
194 unwinding of a throw.</p>
196 <p>The term used to define a the place where an invoke continues after an
197 exception is called a <i>landing pad</i>. LLVM landing pads are conceptually
198 alternative function entry points where a exception structure reference and a type
199 info index are passed in as arguments. The landing pad saves the exception
200 structure reference and then proceeds to select the catch block that corresponds
201 to the type info of the exception object.</p>
203 <p>Two llvm intrinsic functions are used convey information about the landing
204 pad to the back end.</p>
206 <p><a href="#llvm_eh_exception"><tt>llvm.eh.exception</tt></a> takes no
207 arguments and returns the exception structure reference. The backend replaces
208 this intrinsic with the code that accesses the first argument of a call. The
209 LLVM C++ front end generates code to save this value in an alloca location for
210 further use in the landing pad and catch code.</p>
212 <p><a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> takes a minimum of
213 three arguments. The first argument is the reference to the exception
214 structure. The second argument is a reference to the personality function to be
215 used for this try catch sequence. Each of the remaining arguments is either a
216 reference to the type info for a catch statement,
217 a <a href="#throw_filters">filter</a> expression,
218 or the number zero representing a <a href="#cleanups">cleanup</a>.
219 The exception is tested against the arguments sequentially from first to last.
220 The result of the <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> is a
221 positive number if the exception matched a type info, a negative number if it matched
222 a filter, and zero if it matched a cleanup. If nothing is matched, the behaviour of
223 the program is <a href="#restrictions">undefined</a>.
224 The LLVM C++ front end generates code to save the selector value in an alloca
225 location for further use in the landing pad and catch code.
226 If a type info matched then the selector value is the index of the type info in
227 the exception table, which can be obtained using the
228 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic.</p>
230 <p>Once the landing pad has the type info selector, the code branches to the
231 code for the first catch. The catch then checks the value of the type info
232 selector against the index of type info for that catch. Since the type info
233 index is not known until all the type info have been gathered in the backend,
234 the catch code will call the <a
235 href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic to
236 determine the index for a given type info. If the catch fails to match the
237 selector then control is passed on to the next catch. Note: Since the landing
238 pad will not be used if there is no match in the list of type info on the call
239 to <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a>, then neither the
240 last catch nor <i>catch all</i> need to perform the the check against the
243 <p>Finally, the entry and exit of catch code is bracketed with calls to
244 <tt>__cxa_begin_catch</tt> and <tt>__cxa_end_catch</tt>.
245 <tt>__cxa_begin_catch</tt> takes a exception structure reference as an argument
246 and returns the value of the exception object.</tt> <tt>__cxa_end_catch</tt>
247 takes a exception structure reference as an argument. This function clears the
248 exception from the exception space. Note: a rethrow from within the catch may
249 replace this call with a <tt>__cxa_rethrow</tt>.</p>
253 <!-- ======================================================================= -->
254 <div class="doc_subsection">
255 <a name="cleanups">Cleanups</a>
258 <div class="doc_text">
260 <p>To handle destructors and cleanups in try code, control may not run directly
261 from a landing pad to the first catch. Control may actually flow from the
262 landing pad to clean up code and then to the first catch. Since the required
263 clean up for each invoke in a try may be different (ex., intervening
264 constructor), there may be several landing pads for a given try. If cleanups
265 need to be run, the number zero should be passed as the last
266 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> argument.
267 However for C++ a <tt>null i8*</tt> <a href="#restrictions">must</a> be passed
273 <!-- ======================================================================= -->
274 <div class="doc_subsection">
275 <a name="throw_filters">Throw Filters</a>
278 <div class="doc_text">
280 <p>C++ allows the specification of which exception types that can be thrown from
281 a function. To represent this a top level landing pad may exist to filter out
282 invalid types. To express this in LLVM code the landing pad will call <a
283 href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a>. The arguments are the
284 length of the filter expression (the number of type infos plus one), followed by
285 the type infos themselves.
286 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> will return a negative
287 value if the exception does not match any of the type infos. If no match is
288 found then a call to <tt>__cxa_call_unexpected</tt> should be made, otherwise
289 <tt>_Unwind_Resume</tt>. Each of these functions require a reference to the
290 exception structure.</p>
294 <!-- ======================================================================= -->
295 <div class="doc_subsection">
296 <a name="restrictions">Restrictions</a>
299 <div class="doc_text">
301 <p>The semantics of the invoke instruction require that any exception that
302 unwinds through an invoke call should result in a branch to the invoke's unwind
303 label. However such a branch will only happen if the
304 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> matches.
305 Thus in order to ensure correct operation, the front-end must only generate
306 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> calls that are
307 guaranteed to always match whatever exception unwinds through the invoke.
308 For most languages it is enough to pass zero, indicating the presence of
309 a <a href="#cleanups">cleanup</a>, as the last
310 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> argument.
311 However for C++ this is not sufficient, because the C++ personality function
312 will terminate the program if it detects that unwinding the exception only
313 results in matches with cleanups. For C++ a <tt>null i8*</tt> should
314 be passed as the last
315 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> argument instead.
316 This is interpreted as a catch-all by the C++ personality function, and will
322 <!-- ======================================================================= -->
323 <div class="doc_section">
324 <a name="format_common_intrinsics">Exception Handling Intrinsics</a>
327 <div class="doc_text">
329 <p>LLVM uses several intrinsic functions (name prefixed with "llvm.eh") to
330 provide exception handling information at various points in generated code.</p>
334 <!-- ======================================================================= -->
335 <div class="doc_subsubsection">
336 <a name="llvm_eh_exception">llvm.eh.exception</a>
339 <div class="doc_text">
341 i8* %<a href="#llvm_eh_exception">llvm.eh.exception</a>( )
344 <p>This intrinsic indicates that the exception structure is available at this
345 point in the code. The backend will replace this intrinsic with code to fetch
346 the first argument of a call. The effect is that the intrinsic result is the
347 exception structure reference.</p>
351 <!-- ======================================================================= -->
352 <div class="doc_subsubsection">
353 <a name="llvm_eh_selector">llvm.eh.selector</a>
356 <div class="doc_text">
358 i32 %<a href="#llvm_eh_selector">llvm.eh.selector</a>(i8*, i8*, i8*, ...)
361 <p>This intrinsic indicates that the exception selector is available at this
362 point in the code. The backend will replace this intrinsic with code to fetch
363 the second argument of a call. The effect is that the intrinsic result is the
364 exception selector.</p>
366 <p><a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> takes a minimum of
367 three arguments. The first argument is the reference to the exception
368 structure. The second argument is a reference to the personality function to be
369 used for this try catch sequence. Each of the remaining arguments is either a
370 reference to the type info for a catch statement,
371 a <a href="#throw_filters">filter</a> expression,
372 or the number zero representing a <a href="#cleanups">cleanup</a>.
373 The exception is tested against the arguments sequentially from first to last.
374 The result of the <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> is a
375 positive number if the exception matched a type info, a negative number if it matched
376 a filter, and zero if it matched a cleanup. If nothing is matched, the behaviour of
377 the program is <a href="#restrictions">undefined</a>.
378 If a type info matched then the selector value is the index of the type info in
379 the exception table, which can be obtained using the
380 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic.</p>
384 <!-- ======================================================================= -->
385 <div class="doc_subsubsection">
386 <a name="llvm_eh_typeid_for">llvm.eh.typeid.for</a>
389 <div class="doc_text">
391 i32 %<a href="#llvm_eh_typeid_for">llvm.eh.typeid.for</a>(i8*)
394 <p>This intrinsic returns the type info index in the exception table of the
395 current function. This value can be used to compare against the result of <a
396 href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a>. The single argument is
397 a reference to a type info.</p>
401 <!-- ======================================================================= -->
402 <div class="doc_section">
403 <a name="asm">Asm Table Formats</a>
406 <div class="doc_text">
408 <p>There are two tables that are used by the exception handling runtime to
409 determine which actions should take place when an exception is thrown.</p>
413 <!-- ======================================================================= -->
414 <div class="doc_subsection">
415 <a name="unwind_tables">Exception Handling Frame</a>
418 <div class="doc_text">
420 <p>An exception handling frame <tt>eh_frame</tt> is very similar to the unwind
421 frame used by dwarf debug info. The frame contains all the information
422 necessary to tear down the current frame and restore the state of the prior
423 frame. There is an exception handling frame for each function in a compile
424 unit, plus a common exception handling frame that defines information common to
425 all functions in the unit.</p>
427 <p>Todo - Table details here.</p>
431 <!-- ======================================================================= -->
432 <div class="doc_subsection">
433 <a name="exception_tables">Exception Tables</a>
436 <div class="doc_text">
438 <p>An exception table contains information about what actions to take when an
439 exception is thrown in a particular part of a function's code. There is
440 one exception table per function except leaf routines and functions that have
441 only calls to non-throwing functions will not need an exception table.</p>
443 <p>Todo - Table details here.</p>
447 <!-- ======================================================================= -->
448 <div class="doc_section">
449 <a name="todo">ToDo</a>
452 <div class="doc_text">
456 <li><p>Testing/Testing/Testing.</li></p>
462 <!-- *********************************************************************** -->
466 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
467 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
468 <a href="http://validator.w3.org/check/referer"><img
469 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
471 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
472 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
473 Last modified: $Date$