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>
41 <li><a href="#llvm_eh_sjlj_dispatchsetup"><tt>llvm.eh.sjlj.dispatchsetup</tt></a></li>
43 <li><a href="#asm">Asm Table Formats</a>
45 <li><a href="#unwind_tables">Exception Handling Frame</a></li>
46 <li><a href="#exception_tables">Exception Tables</a></li>
52 <div class="doc_author">
53 <p>Written by <a href="mailto:jlaskey@mac.com">Jim Laskey</a></p>
57 <!-- *********************************************************************** -->
58 <h2><a name="introduction">Introduction</a></h2>
59 <!-- *********************************************************************** -->
63 <p>This document is the central repository for all information pertaining to
64 exception handling in LLVM. It describes the format that LLVM exception
65 handling information takes, which is useful for those interested in creating
66 front-ends or dealing directly with the information. Further, this document
67 provides specific examples of what exception handling information is used for
70 <!-- ======================================================================= -->
72 <a name="itanium">Itanium ABI Zero-cost Exception Handling</a>
77 <p>Exception handling for most programming languages is designed to recover from
78 conditions that rarely occur during general use of an application. To that
79 end, exception handling should not interfere with the main flow of an
80 application's algorithm by performing checkpointing tasks, such as saving the
81 current pc or register state.</p>
83 <p>The Itanium ABI Exception Handling Specification defines a methodology for
84 providing outlying data in the form of exception tables without inlining
85 speculative exception handling code in the flow of an application's main
86 algorithm. Thus, the specification is said to add "zero-cost" to the normal
87 execution of an application.</p>
89 <p>A more complete description of the Itanium ABI exception handling runtime
90 support of can be found at
91 <a href="http://www.codesourcery.com/cxx-abi/abi-eh.html">Itanium C++ ABI:
92 Exception Handling</a>. A description of the exception frame format can be
94 <a href="http://refspecs.freestandards.org/LSB_3.0.0/LSB-Core-generic/LSB-Core-generic/ehframechpt.html">Exception
95 Frames</a>, with details of the DWARF 4 specification at
96 <a href="http://dwarfstd.org/Dwarf4Std.php">DWARF 4 Standard</a>.
97 A description for the C++ exception table formats can be found at
98 <a href="http://www.codesourcery.com/cxx-abi/exceptions.pdf">Exception Handling
103 <!-- ======================================================================= -->
105 <a name="sjlj">Setjmp/Longjmp Exception Handling</a>
110 <p>Setjmp/Longjmp (SJLJ) based exception handling uses LLVM intrinsics
111 <a href="#llvm_eh_sjlj_setjmp"><tt>llvm.eh.sjlj.setjmp</tt></a> and
112 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a> to
113 handle control flow for exception handling.</p>
115 <p>For each function which does exception processing — be
116 it <tt>try</tt>/<tt>catch</tt> blocks or cleanups — that function
117 registers itself on a global frame list. When exceptions are unwinding, the
118 runtime uses this list to identify which functions need processing.<p>
120 <p>Landing pad selection is encoded in the call site entry of the function
121 context. The runtime returns to the function via
122 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a>, where
123 a switch table transfers control to the appropriate landing pad based on
124 the index stored in the function context.</p>
126 <p>In contrast to DWARF exception handling, which encodes exception regions
127 and frame information in out-of-line tables, SJLJ exception handling
128 builds and removes the unwind frame context at runtime. This results in
129 faster exception handling at the expense of slower execution when no
130 exceptions are thrown. As exceptions are, by their nature, intended for
131 uncommon code paths, DWARF exception handling is generally preferred to
136 <!-- ======================================================================= -->
138 <a name="overview">Overview</a>
143 <p>When an exception is thrown in LLVM code, the runtime does its best to find a
144 handler suited to processing the circumstance.</p>
146 <p>The runtime first attempts to find an <i>exception frame</i> corresponding to
147 the function where the exception was thrown. If the programming language
148 supports exception handling (e.g. C++), the exception frame contains a
149 reference to an exception table describing how to process the exception. If
150 the language does not support exception handling (e.g. C), or if the
151 exception needs to be forwarded to a prior activation, the exception frame
152 contains information about how to unwind the current activation and restore
153 the state of the prior activation. This process is repeated until the
154 exception is handled. If the exception is not handled and no activations
155 remain, then the application is terminated with an appropriate error
158 <p>Because different programming languages have different behaviors when
159 handling exceptions, the exception handling ABI provides a mechanism for
160 supplying <i>personalities</i>. An exception handling personality is defined
161 by way of a <i>personality function</i> (e.g. <tt>__gxx_personality_v0</tt>
162 in C++), which receives the context of the exception, an <i>exception
163 structure</i> containing the exception object type and value, and a reference
164 to the exception table for the current function. The personality function
165 for the current compile unit is specified in a <i>common exception
168 <p>The organization of an exception table is language dependent. For C++, an
169 exception table is organized as a series of code ranges defining what to do
170 if an exception occurs in that range. Typically, the information associated
171 with a range defines which types of exception objects (using C++ <i>type
172 info</i>) that are handled in that range, and an associated action that
173 should take place. Actions typically pass control to a <i>landing
176 <p>A landing pad corresponds roughly to the code found in the <tt>catch</tt>
177 portion of a <tt>try</tt>/<tt>catch</tt> sequence. When execution resumes at
178 a landing pad, it receives an <i>exception structure</i> and a
179 <i>selector value</i> corresponding to the <i>type</i> of exception
180 thrown. The selector is then used to determine which <i>catch</i> should
181 actually process the exception.</p>
187 <!-- ======================================================================= -->
189 <a name="codegen">LLVM Code Generation</a>
194 <p>From a C++ developer's perspective, exceptions are defined in terms of the
195 <tt>throw</tt> and <tt>try</tt>/<tt>catch</tt> statements. In this section
196 we will describe the implementation of LLVM exception handling in terms of
199 <!-- ======================================================================= -->
201 <a name="throw">Throw</a>
206 <p>Languages that support exception handling typically provide a <tt>throw</tt>
207 operation to initiate the exception process. Internally, a <tt>throw</tt>
208 operation breaks down into two steps.</p>
211 <li>A request is made to allocate exception space for an exception structure.
212 This structure needs to survive beyond the current activation. This
213 structure will contain the type and value of the object being thrown.</li>
215 <li>A call is made to the runtime to raise the exception, passing the
216 exception structure as an argument.</li>
219 <p>In C++, the allocation of the exception structure is done by the
220 <tt>__cxa_allocate_exception</tt> runtime function. The exception raising is
221 handled by <tt>__cxa_throw</tt>. The type of the exception is represented
222 using a C++ RTTI structure.</p>
226 <!-- ======================================================================= -->
228 <a name="try_catch">Try/Catch</a>
233 <p>A call within the scope of a <i>try</i> statement can potentially raise an
234 exception. In those circumstances, the LLVM C++ front-end replaces the call
235 with an <tt>invoke</tt> instruction. Unlike a call, the <tt>invoke</tt> has
236 two potential continuation points:</p>
239 <li>where to continue when the call succeeds as per normal, and</li>
241 <li>where to continue if the call raises an exception, either by a throw or
242 the unwinding of a throw</li>
245 <p>The term used to define a the place where an <tt>invoke</tt> continues after
246 an exception is called a <i>landing pad</i>. LLVM landing pads are
247 conceptually alternative function entry points where an exception structure
248 reference and a type info index are passed in as arguments. The landing pad
249 saves the exception structure reference and then proceeds to select the catch
250 block that corresponds to the type info of the exception object.</p>
252 <p>The LLVM <a href="LangRef.html#i_landingpad"><tt>landingpad</tt>
253 instruction</a> is used to convey information about the landing pad to the
254 back end. For C++, the <tt>landingpad</tt> instruction returns a pointer and
255 integer pair corresponding to the pointer to the <i>exception structure</i>
256 and the <i>selector value</i> respectively.</p>
258 <p>The <tt>landingpad</tt> instruction takes a reference to the personality
259 function to be used for this <tt>try</tt>/<tt>catch</tt> sequence. The
260 remainder of the instruction is a list of <i>cleanup</i>, <i>catch</i>,
261 and <i>filter</i> clauses. The exception is tested against the clauses
262 sequentially from first to last. The selector value is a positive number if
263 the exception matched a type info, a negative number if it matched a filter,
264 and zero if it matched a cleanup. If nothing is matched, the behavior of
265 the program is <a href="#restrictions">undefined</a>. If a type info matched,
266 then the selector value is the index of the type info in the exception table,
267 which can be obtained using the
268 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic.</p>
270 <p>Once the landing pad has the type info selector, the code branches to the
271 code for the first catch. The catch then checks the value of the type info
272 selector against the index of type info for that catch. Since the type info
273 index is not known until all the type infos have been gathered in the
274 backend, the catch code must call the
275 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic to
276 determine the index for a given type info. If the catch fails to match the
277 selector then control is passed on to the next catch.</p>
279 <p>Finally, the entry and exit of catch code is bracketed with calls to
280 <tt>__cxa_begin_catch</tt> and <tt>__cxa_end_catch</tt>.</p>
283 <li><tt>__cxa_begin_catch</tt> takes an exception structure reference as an
284 argument and returns the value of the exception object.</li>
286 <li><tt>__cxa_end_catch</tt> takes no arguments. This function:<br><br>
288 <li>Locates the most recently caught exception and decrements its handler
290 <li>Removes the exception from the <i>caught</i> stack if the handler
291 count goes to zero, and</li>
292 <li>Destroys the exception if the handler count goes to zero and the
293 exception was not re-thrown by throw.</li>
295 <p><b>Note:</b> a rethrow from within the catch may replace this call with
296 a <tt>__cxa_rethrow</tt>.</p></li>
301 <!-- ======================================================================= -->
303 <a name="cleanups">Cleanups</a>
308 <p>A cleanup is extra code which needs to be run as part of unwinding a scope.
309 C++ destructors are a typical example, but other languages and language
310 extensions provide a variety of different kinds of cleanups. In general, a
311 landing pad may need to run arbitrary amounts of cleanup code before actually
312 entering a catch block. To indicate the presence of cleanups, a
313 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instruction</a>
314 should have a <i>cleanup</i> clause. Otherwise, the unwinder will not stop at
315 the landing pad if there are no catches or filters that require it to.</p>
317 <p><b>Note:</b> Do not allow a new exception to propagate out of the execution
318 of a cleanup. This can corrupt the internal state of the unwinder.
319 Different languages describe different high-level semantics for these
320 situations: for example, C++ requires that the process be terminated, whereas
321 Ada cancels both exceptions and throws a third.</p>
323 <p>When all cleanups are finished, if the exception is not handled by the
324 current function, resume unwinding by calling the
325 <a href="LangRef.html#i_resume"><tt>resume</tt> instruction</a>, passing in
326 the result of the <tt>landingpad</tt> instruction for the original landing
331 <!-- ======================================================================= -->
333 <a name="throw_filters">Throw Filters</a>
338 <p>C++ allows the specification of which exception types may be thrown from a
339 function. To represent this, a top level landing pad may exist to filter out
340 invalid types. To express this in LLVM code the
341 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instruction</a> will
342 have a filter clause. The clause consists of an array of type infos.
343 <tt>landingpad</tt> will return a negative value if the exception does not
344 match any of the type infos. If no match is found then a call
345 to <tt>__cxa_call_unexpected</tt> should be made, otherwise
346 <tt>_Unwind_Resume</tt>. Each of these functions requires a reference to the
347 exception structure. Note that the most general form of a
348 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instruction</a> can
349 have any number of catch, cleanup, and filter clauses (though having more
350 than one cleanup is pointless). The LLVM C++ front-end can generate such
351 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instructions</a> due
352 to inlining creating nested exception handling scopes.</p>
356 <!-- ======================================================================= -->
358 <a name="restrictions">Restrictions</a>
363 <p>The unwinder delegates the decision of whether to stop in a call frame to
364 that call frame's language-specific personality function. Not all unwinders
365 guarantee that they will stop to perform cleanups. For example, the GNU C++
366 unwinder doesn't do so unless the exception is actually caught somewhere
367 further up the stack.</p>
369 <p>In order for inlining to behave correctly, landing pads must be prepared to
370 handle selector results that they did not originally advertise. Suppose that
371 a function catches exceptions of type <tt>A</tt>, and it's inlined into a
372 function that catches exceptions of type <tt>B</tt>. The inliner will update
373 the <tt>landingpad</tt> instruction for the inlined landing pad to include
374 the fact that <tt>B</tt> is also caught. If that landing pad assumes that it
375 will only be entered to catch an <tt>A</tt>, it's in for a rude awakening.
376 Consequently, landing pads must test for the selector results they understand
377 and then resume exception propagation with the
378 <a href="LangRef.html#i_resume"><tt>resume</tt> instruction</a> if none of
379 the conditions match.</p>
385 <!-- ======================================================================= -->
387 <a name="format_common_intrinsics">Exception Handling Intrinsics</a>
392 <p>In addition to the
393 <a href="LangRef.html#i_landingpad"><tt>landingpad</tt></a> and
394 <a href="LangRef.html#i_resume"><tt>resume</tt></a> instructions, LLVM uses
395 several intrinsic functions (name prefixed with <i><tt>llvm.eh</tt></i>) to
396 provide exception handling information at various points in generated
399 <!-- ======================================================================= -->
401 <a name="llvm_eh_typeid_for">llvm.eh.typeid.for</a>
407 i32 @llvm.eh.typeid.for(i8* %type_info)
410 <p>This intrinsic returns the type info index in the exception table of the
411 current function. This value can be used to compare against the result
412 of <a href="LangRef.html#i_landingpad"><tt>landingpad</tt> instruction</a>.
413 The single argument is a reference to a type info.</p>
417 <!-- ======================================================================= -->
419 <a name="llvm_eh_sjlj_setjmp">llvm.eh.sjlj.setjmp</a>
425 i32 @llvm.eh.sjlj.setjmp(i8* %setjmp_buf)
428 <p>For SJLJ based exception handling, this intrinsic forces register saving for
429 the current function and stores the address of the following instruction for
430 use as a destination address
431 by <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a>. The
432 buffer format and the overall functioning of this intrinsic is compatible
433 with the GCC <tt>__builtin_setjmp</tt> implementation allowing code built
434 with the clang and GCC to interoperate.</p>
436 <p>The single parameter is a pointer to a five word buffer in which the calling
437 context is saved. The front end places the frame pointer in the first word,
438 and the target implementation of this intrinsic should place the destination
440 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a> in the
441 second word. The following three words are available for use in a
442 target-specific manner.</p>
446 <!-- ======================================================================= -->
448 <a name="llvm_eh_sjlj_longjmp">llvm.eh.sjlj.longjmp</a>
454 void @llvm.eh.sjlj.longjmp(i8* %setjmp_buf)
457 <p>For SJLJ based exception handling, the <tt>llvm.eh.sjlj.longjmp</tt>
458 intrinsic is used to implement <tt>__builtin_longjmp()</tt>. The single
459 parameter is a pointer to a buffer populated
460 by <a href="#llvm_eh_sjlj_setjmp"><tt>llvm.eh.sjlj.setjmp</tt></a>. The frame
461 pointer and stack pointer are restored from the buffer, then control is
462 transferred to the destination address.</p>
465 <!-- ======================================================================= -->
467 <a name="llvm_eh_sjlj_lsda">llvm.eh.sjlj.lsda</a>
473 i8* @llvm.eh.sjlj.lsda()
476 <p>For SJLJ based exception handling, the <tt>llvm.eh.sjlj.lsda</tt> intrinsic
477 returns the address of the Language Specific Data Area (LSDA) for the current
478 function. The SJLJ front-end code stores this address in the exception
479 handling function context for use by the runtime.</p>
483 <!-- ======================================================================= -->
485 <a name="llvm_eh_sjlj_callsite">llvm.eh.sjlj.callsite</a>
491 void @llvm.eh.sjlj.callsite(i32 %call_site_num)
494 <p>For SJLJ based exception handling, the <tt>llvm.eh.sjlj.callsite</tt>
495 intrinsic identifies the callsite value associated with the
496 following <tt>invoke</tt> instruction. This is used to ensure that landing
497 pad entries in the LSDA are generated in matching order.</p>
501 <!-- ======================================================================= -->
503 <a name="llvm_eh_sjlj_dispatchsetup">llvm.eh.sjlj.dispatchsetup</a>
509 void @llvm.eh.sjlj.dispatchsetup(i32 %dispatch_value)
512 <p>For SJLJ based exception handling, the <tt>llvm.eh.sjlj.dispatchsetup</tt>
513 intrinsic is used by targets to do any unwind edge setup they need. By
514 default, no action is taken.</p>
520 <!-- ======================================================================= -->
522 <a name="asm">Asm Table Formats</a>
527 <p>There are two tables that are used by the exception handling runtime to
528 determine which actions should be taken when an exception is thrown.</p>
530 <!-- ======================================================================= -->
532 <a name="unwind_tables">Exception Handling Frame</a>
537 <p>An exception handling frame <tt>eh_frame</tt> is very similar to the unwind
538 frame used by DWARF debug info. The frame contains all the information
539 necessary to tear down the current frame and restore the state of the prior
540 frame. There is an exception handling frame for each function in a compile
541 unit, plus a common exception handling frame that defines information common
542 to all functions in the unit.</p>
544 <!-- Todo - Table details here. -->
548 <!-- ======================================================================= -->
550 <a name="exception_tables">Exception Tables</a>
555 <p>An exception table contains information about what actions to take when an
556 exception is thrown in a particular part of a function's code. There is one
557 exception table per function, except leaf functions and functions that have
558 calls only to non-throwing functions. They do not need an exception
561 <!-- Todo - Table details here. -->
567 <!-- *********************************************************************** -->
571 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
572 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
573 <a href="http://validator.w3.org/check/referer"><img
574 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
576 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
577 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
578 Last modified: $Date$