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 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="description"
8 content="Exception Handling in LLVM.">
9 <link rel="stylesheet" href="llvm.css" type="text/css">
14 <h1>Exception Handling in LLVM</h1>
16 <table class="layout" style="width:100%">
20 <li><a href="#introduction">Introduction</a>
22 <li><a href="#itanium">Itanium ABI Zero-cost Exception Handling</a></li>
23 <li><a href="#sjlj">Setjmp/Longjmp Exception Handling</a></li>
24 <li><a href="#overview">Overview</a></li>
26 <li><a href="#codegen">LLVM Code Generation</a>
28 <li><a href="#throw">Throw</a></li>
29 <li><a href="#try_catch">Try/Catch</a></li>
30 <li><a href="#cleanups">Cleanups</a></li>
31 <li><a href="#throw_filters">Throw Filters</a></li>
32 <li><a href="#restrictions">Restrictions</a></li>
34 <li><a href="#format_common_intrinsics">Exception Handling Intrinsics</a>
36 <li><a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a></li>
37 <li><a href="#llvm_eh_sjlj_setjmp"><tt>llvm.eh.sjlj.setjmp</tt></a></li>
38 <li><a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a></li>
39 <li><a href="#llvm_eh_sjlj_lsda"><tt>llvm.eh.sjlj.lsda</tt></a></li>
40 <li><a href="#llvm_eh_sjlj_callsite"><tt>llvm.eh.sjlj.callsite</tt></a></li>
42 <li><a href="#asm">Asm Table Formats</a>
44 <li><a href="#unwind_tables">Exception Handling Frame</a></li>
45 <li><a href="#exception_tables">Exception Tables</a></li>
51 <div class="doc_author">
52 <p>Written by the <a href="http://llvm.org/">LLVM Team</a></p>
56 <!-- *********************************************************************** -->
57 <h2><a name="introduction">Introduction</a></h2>
58 <!-- *********************************************************************** -->
62 <p>This document is the central repository for all information pertaining to
63 exception handling in LLVM. It describes the format that LLVM exception
64 handling information takes, which is useful for those interested in creating
65 front-ends or dealing directly with the information. Further, this document
66 provides specific examples of what exception handling information is used for
69 <!-- ======================================================================= -->
71 <a name="itanium">Itanium ABI Zero-cost Exception Handling</a>
76 <p>Exception handling for most programming languages is designed to recover from
77 conditions that rarely occur during general use of an application. To that
78 end, exception handling should not interfere with the main flow of an
79 application's algorithm by performing checkpointing tasks, such as saving the
80 current pc or register state.</p>
82 <p>The Itanium ABI Exception Handling Specification defines a methodology for
83 providing outlying data in the form of exception tables without inlining
84 speculative exception handling code in the flow of an application's main
85 algorithm. Thus, the specification is said to add "zero-cost" to the normal
86 execution of an application.</p>
88 <p>A more complete description of the Itanium ABI exception handling runtime
89 support of can be found at
90 <a href="http://www.codesourcery.com/cxx-abi/abi-eh.html">Itanium C++ ABI:
91 Exception Handling</a>. A description of the exception frame format can be
93 <a href="http://refspecs.freestandards.org/LSB_3.0.0/LSB-Core-generic/LSB-Core-generic/ehframechpt.html">Exception
94 Frames</a>, with details of the DWARF 4 specification at
95 <a href="http://dwarfstd.org/Dwarf4Std.php">DWARF 4 Standard</a>.
96 A description for the C++ exception table formats can be found at
97 <a href="http://www.codesourcery.com/cxx-abi/exceptions.pdf">Exception Handling
102 <!-- ======================================================================= -->
104 <a name="sjlj">Setjmp/Longjmp Exception Handling</a>
109 <p>Setjmp/Longjmp (SJLJ) based exception handling uses LLVM intrinsics
110 <a href="#llvm_eh_sjlj_setjmp"><tt>llvm.eh.sjlj.setjmp</tt></a> and
111 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a> to
112 handle control flow for exception handling.</p>
114 <p>For each function which does exception processing — be
115 it <tt>try</tt>/<tt>catch</tt> blocks or cleanups — that function
116 registers itself on a global frame list. When exceptions are unwinding, the
117 runtime uses this list to identify which functions need processing.<p>
119 <p>Landing pad selection is encoded in the call site entry of the function
120 context. The runtime returns to the function via
121 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a>, where
122 a switch table transfers control to the appropriate landing pad based on
123 the index stored in the function context.</p>
125 <p>In contrast to DWARF exception handling, which encodes exception regions
126 and frame information in out-of-line tables, SJLJ exception handling
127 builds and removes the unwind frame context at runtime. This results in
128 faster exception handling at the expense of slower execution when no
129 exceptions are thrown. As exceptions are, by their nature, intended for
130 uncommon code paths, DWARF exception handling is generally preferred to
135 <!-- ======================================================================= -->
137 <a name="overview">Overview</a>
142 <p>When an exception is thrown in LLVM code, the runtime does its best to find a
143 handler suited to processing the circumstance.</p>
145 <p>The runtime first attempts to find an <i>exception frame</i> corresponding to
146 the function where the exception was thrown. If the programming language
147 supports exception handling (e.g. C++), the exception frame contains a
148 reference to an exception table describing how to process the exception. If
149 the language does not support exception handling (e.g. C), or if the
150 exception needs to be forwarded to a prior activation, the exception frame
151 contains information about how to unwind the current activation and restore
152 the state of the prior activation. This process is repeated until the
153 exception is handled. If the exception is not handled and no activations
154 remain, then the application is terminated with an appropriate error
157 <p>Because different programming languages have different behaviors when
158 handling exceptions, the exception handling ABI provides a mechanism for
159 supplying <i>personalities</i>. An exception handling personality is defined
160 by way of a <i>personality function</i> (e.g. <tt>__gxx_personality_v0</tt>
161 in C++), which receives the context of the exception, an <i>exception
162 structure</i> containing the exception object type and value, and a reference
163 to the exception table for the current function. The personality function
164 for the current compile unit is specified in a <i>common exception
167 <p>The organization of an exception table is language dependent. For C++, an
168 exception table is organized as a series of code ranges defining what to do
169 if an exception occurs in that range. Typically, the information associated
170 with a range defines which types of exception objects (using C++ <i>type
171 info</i>) that are handled in that range, and an associated action that
172 should take place. Actions typically pass control to a <i>landing
175 <p>A landing pad corresponds roughly to the code found in the <tt>catch</tt>
176 portion of a <tt>try</tt>/<tt>catch</tt> sequence. When execution resumes at
177 a landing pad, it receives an <i>exception structure</i> and a
178 <i>selector value</i> corresponding to the <i>type</i> of exception
179 thrown. The selector is then used to determine which <i>catch</i> should
180 actually process the exception.</p>
186 <!-- ======================================================================= -->
188 <a name="codegen">LLVM Code Generation</a>
193 <p>From a C++ developer's perspective, exceptions are defined in terms of the
194 <tt>throw</tt> and <tt>try</tt>/<tt>catch</tt> statements. In this section
195 we will describe the implementation of LLVM exception handling in terms of
198 <!-- ======================================================================= -->
200 <a name="throw">Throw</a>
205 <p>Languages that support exception handling typically provide a <tt>throw</tt>
206 operation to initiate the exception process. Internally, a <tt>throw</tt>
207 operation breaks down into two steps.</p>
210 <li>A request is made to allocate exception space for an exception structure.
211 This structure needs to survive beyond the current activation. This
212 structure will contain the type and value of the object being thrown.</li>
214 <li>A call is made to the runtime to raise the exception, passing the
215 exception structure as an argument.</li>
218 <p>In C++, the allocation of the exception structure is done by the
219 <tt>__cxa_allocate_exception</tt> runtime function. The exception raising is
220 handled by <tt>__cxa_throw</tt>. The type of the exception is represented
221 using a C++ RTTI structure.</p>
225 <!-- ======================================================================= -->
227 <a name="try_catch">Try/Catch</a>
232 <p>A call within the scope of a <i>try</i> statement can potentially raise an
233 exception. In those circumstances, the LLVM C++ front-end replaces the call
234 with an <tt>invoke</tt> instruction. Unlike a call, the <tt>invoke</tt> has
235 two potential continuation points:</p>
238 <li>where to continue when the call succeeds as per normal, and</li>
240 <li>where to continue if the call raises an exception, either by a throw or
241 the unwinding of a throw</li>
244 <p>The term used to define a the place where an <tt>invoke</tt> continues after
245 an exception is called a <i>landing pad</i>. LLVM landing pads are
246 conceptually alternative function entry points where an exception structure
247 reference and a type info index are passed in as arguments. The landing pad
248 saves the exception structure reference and then proceeds to select the catch
249 block that corresponds to the type info of the exception object.</p>
251 <p>The LLVM <a href="LangRef.html#i_landingpad"><tt>landingpad</tt>
252 instruction</a> is used to convey information about the landing pad to the
253 back end. For C++, the <tt>landingpad</tt> instruction returns a pointer and
254 integer pair corresponding to the pointer to the <i>exception structure</i>
255 and the <i>selector value</i> respectively.</p>
257 <p>The <tt>landingpad</tt> instruction takes a reference to the personality
258 function to be used for this <tt>try</tt>/<tt>catch</tt> sequence. The
259 remainder of the instruction is a list of <i>cleanup</i>, <i>catch</i>,
260 and <i>filter</i> clauses. The exception is tested against the clauses
261 sequentially from first to last. The selector value is a positive number if
262 the exception matched a type info, a negative number if it matched a filter,
263 and zero if it matched a cleanup. If nothing is matched, the behavior of
264 the program is <a href="#restrictions">undefined</a>. If a type info matched,
265 then the selector value is the index of the type info in the exception table,
266 which can be obtained using the
267 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic.</p>
269 <p>Once the landing pad has the type info selector, the code branches to the
270 code for the first catch. The catch then checks the value of the type info
271 selector against the index of type info for that catch. Since the type info
272 index is not known until all the type infos have been gathered in the
273 backend, the catch code must call the
274 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic to
275 determine the index for a given type info. If the catch fails to match the
276 selector then control is passed on to the next catch.</p>
278 <p>Finally, the entry and exit of catch code is bracketed with calls to
279 <tt>__cxa_begin_catch</tt> and <tt>__cxa_end_catch</tt>.</p>
282 <li><tt>__cxa_begin_catch</tt> takes an exception structure reference as an
283 argument and returns the value of the exception object.</li>
285 <li><tt>__cxa_end_catch</tt> takes no arguments. This function:<br><br>
287 <li>Locates the most recently caught exception and decrements its handler
289 <li>Removes the exception from the <i>caught</i> stack if the handler
290 count goes to zero, and</li>
291 <li>Destroys the exception if the handler count goes to zero and the
292 exception was not re-thrown by throw.</li>
294 <p><b>Note:</b> a rethrow from within the catch may replace this call with
295 a <tt>__cxa_rethrow</tt>.</p></li>
300 <!-- ======================================================================= -->
302 <a name="cleanups">Cleanups</a>
307 <p>A cleanup is extra code which needs to be run as part of unwinding a scope.
308 C++ destructors are a typical example, but other languages and language
309 extensions provide a variety of different kinds of cleanups. In general, a
310 landing pad may need to run arbitrary amounts of cleanup code before actually
311 entering a catch block. To indicate the presence of cleanups, a
312 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instruction</a>
313 should have a <i>cleanup</i> clause. Otherwise, the unwinder will not stop at
314 the landing pad if there are no catches or filters that require it to.</p>
316 <p><b>Note:</b> Do not allow a new exception to propagate out of the execution
317 of a cleanup. This can corrupt the internal state of the unwinder.
318 Different languages describe different high-level semantics for these
319 situations: for example, C++ requires that the process be terminated, whereas
320 Ada cancels both exceptions and throws a third.</p>
322 <p>When all cleanups are finished, if the exception is not handled by the
323 current function, resume unwinding by calling the
324 <a href="LangRef.html#i_resume"><tt>resume</tt> instruction</a>, passing in
325 the result of the <tt>landingpad</tt> instruction for the original landing
330 <!-- ======================================================================= -->
332 <a name="throw_filters">Throw Filters</a>
337 <p>C++ allows the specification of which exception types may be thrown from a
338 function. To represent this, a top level landing pad may exist to filter out
339 invalid types. To express this in LLVM code the
340 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instruction</a> will
341 have a filter clause. The clause consists of an array of type infos.
342 <tt>landingpad</tt> will return a negative value if the exception does not
343 match any of the type infos. If no match is found then a call
344 to <tt>__cxa_call_unexpected</tt> should be made, otherwise
345 <tt>_Unwind_Resume</tt>. Each of these functions requires a reference to the
346 exception structure. Note that the most general form of a
347 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instruction</a> can
348 have any number of catch, cleanup, and filter clauses (though having more
349 than one cleanup is pointless). The LLVM C++ front-end can generate such
350 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instructions</a> due
351 to inlining creating nested exception handling scopes.</p>
355 <!-- ======================================================================= -->
357 <a name="restrictions">Restrictions</a>
362 <p>The unwinder delegates the decision of whether to stop in a call frame to
363 that call frame's language-specific personality function. Not all unwinders
364 guarantee that they will stop to perform cleanups. For example, the GNU C++
365 unwinder doesn't do so unless the exception is actually caught somewhere
366 further up the stack.</p>
368 <p>In order for inlining to behave correctly, landing pads must be prepared to
369 handle selector results that they did not originally advertise. Suppose that
370 a function catches exceptions of type <tt>A</tt>, and it's inlined into a
371 function that catches exceptions of type <tt>B</tt>. The inliner will update
372 the <tt>landingpad</tt> instruction for the inlined landing pad to include
373 the fact that <tt>B</tt> is also caught. If that landing pad assumes that it
374 will only be entered to catch an <tt>A</tt>, it's in for a rude awakening.
375 Consequently, landing pads must test for the selector results they understand
376 and then resume exception propagation with the
377 <a href="LangRef.html#i_resume"><tt>resume</tt> instruction</a> if none of
378 the conditions match.</p>
384 <!-- ======================================================================= -->
386 <a name="format_common_intrinsics">Exception Handling Intrinsics</a>
391 <p>In addition to the
392 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt></a> and
393 <a href="LangRef.html#i_resume"><tt>resume</tt></a> instructions, LLVM uses
394 several intrinsic functions (name prefixed with <i><tt>llvm.eh</tt></i>) to
395 provide exception handling information at various points in generated
398 <!-- ======================================================================= -->
400 <a name="llvm_eh_typeid_for">llvm.eh.typeid.for</a>
406 i32 @llvm.eh.typeid.for(i8* %type_info)
409 <p>This intrinsic returns the type info index in the exception table of the
410 current function. This value can be used to compare against the result
411 of <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instruction</a>.
412 The single argument is a reference to a type info.</p>
416 <!-- ======================================================================= -->
418 <a name="llvm_eh_sjlj_setjmp">llvm.eh.sjlj.setjmp</a>
424 i32 @llvm.eh.sjlj.setjmp(i8* %setjmp_buf)
427 <p>For SJLJ based exception handling, this intrinsic forces register saving for
428 the current function and stores the address of the following instruction for
429 use as a destination address
430 by <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a>. The
431 buffer format and the overall functioning of this intrinsic is compatible
432 with the GCC <tt>__builtin_setjmp</tt> implementation allowing code built
433 with the clang and GCC to interoperate.</p>
435 <p>The single parameter is a pointer to a five word buffer in which the calling
436 context is saved. The front end places the frame pointer in the first word,
437 and the target implementation of this intrinsic should place the destination
439 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a> in the
440 second word. The following three words are available for use in a
441 target-specific manner.</p>
445 <!-- ======================================================================= -->
447 <a name="llvm_eh_sjlj_longjmp">llvm.eh.sjlj.longjmp</a>
453 void @llvm.eh.sjlj.longjmp(i8* %setjmp_buf)
456 <p>For SJLJ based exception handling, the <tt>llvm.eh.sjlj.longjmp</tt>
457 intrinsic is used to implement <tt>__builtin_longjmp()</tt>. The single
458 parameter is a pointer to a buffer populated
459 by <a href="#llvm_eh_sjlj_setjmp"><tt>llvm.eh.sjlj.setjmp</tt></a>. The frame
460 pointer and stack pointer are restored from the buffer, then control is
461 transferred to the destination address.</p>
464 <!-- ======================================================================= -->
466 <a name="llvm_eh_sjlj_lsda">llvm.eh.sjlj.lsda</a>
472 i8* @llvm.eh.sjlj.lsda()
475 <p>For SJLJ based exception handling, the <tt>llvm.eh.sjlj.lsda</tt> intrinsic
476 returns the address of the Language Specific Data Area (LSDA) for the current
477 function. The SJLJ front-end code stores this address in the exception
478 handling function context for use by the runtime.</p>
482 <!-- ======================================================================= -->
484 <a name="llvm_eh_sjlj_callsite">llvm.eh.sjlj.callsite</a>
490 void @llvm.eh.sjlj.callsite(i32 %call_site_num)
493 <p>For SJLJ based exception handling, the <tt>llvm.eh.sjlj.callsite</tt>
494 intrinsic identifies the callsite value associated with the
495 following <tt>invoke</tt> instruction. This is used to ensure that landing
496 pad entries in the LSDA are generated in matching order.</p>
500 <!-- ======================================================================= -->
502 <a name="asm">Asm Table Formats</a>
507 <p>There are two tables that are used by the exception handling runtime to
508 determine which actions should be taken when an exception is thrown.</p>
510 <!-- ======================================================================= -->
512 <a name="unwind_tables">Exception Handling Frame</a>
517 <p>An exception handling frame <tt>eh_frame</tt> is very similar to the unwind
518 frame used by DWARF debug info. The frame contains all the information
519 necessary to tear down the current frame and restore the state of the prior
520 frame. There is an exception handling frame for each function in a compile
521 unit, plus a common exception handling frame that defines information common
522 to all functions in the unit.</p>
524 <!-- Todo - Table details here. -->
528 <!-- ======================================================================= -->
530 <a name="exception_tables">Exception Tables</a>
535 <p>An exception table contains information about what actions to take when an
536 exception is thrown in a particular part of a function's code. There is one
537 exception table per function, except leaf functions and functions that have
538 calls only to non-throwing functions. They do not need an exception
541 <!-- Todo - Table details here. -->
547 <!-- *********************************************************************** -->
551 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
552 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
553 <a href="http://validator.w3.org/check/referer"><img
554 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
556 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
557 Last modified: $Date$